Merge remote-tracking branch 'airlied/drm-next' into topic/drm-misc

Backmerge drm-next to be able to apply Chris' connector_unregister_all
cleanup (need latest i915 and sun4i state for that).

Also there's a trivial conflict in ttm_bo.c that git rerere fails to
remember.

Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>

141 files changed, 3846 insertions, 641 deletions

diff --git a/drivers/gpu/drm/nouveau/include/nvif/cl0080.h b/drivers/gpu/drm/nouveau/include/nvif/cl0080.h
index 331620a52afa..287a7d6fa480 100644
--- a/drivers/gpu/drm/nouveau/include/nvif/cl0080.h
+++ b/drivers/gpu/drm/nouveau/include/nvif/cl0080.h
@@ -29,6 +29,7 @@ struct nv_device_info_v0 {
 #define NV_DEVICE_INFO_V0_FERMI                                            0x07
 #define NV_DEVICE_INFO_V0_KEPLER                                           0x08
 #define NV_DEVICE_INFO_V0_MAXWELL                                          0x09
+#define NV_DEVICE_INFO_V0_PASCAL                                           0x0a
 	__u8  family;
 	__u8  pad06[2];
 	__u64 ram_size;
diff --git a/drivers/gpu/drm/nouveau/include/nvif/class.h b/drivers/gpu/drm/nouveau/include/nvif/class.h
index 982aad8fa645..e6e9537537cf 100644
--- a/drivers/gpu/drm/nouveau/include/nvif/class.h
+++ b/drivers/gpu/drm/nouveau/include/nvif/class.h
@@ -39,6 +39,7 @@
 #define KEPLER_CHANNEL_GPFIFO_A                       /* cla06f.h */ 0x0000a06f
 #define KEPLER_CHANNEL_GPFIFO_B                       /* cla06f.h */ 0x0000a16f
 #define MAXWELL_CHANNEL_GPFIFO_A                      /* cla06f.h */ 0x0000b06f
+#define PASCAL_CHANNEL_GPFIFO_A                       /* cla06f.h */ 0x0000c06f
 
 #define NV50_DISP                                     /* cl5070.h */ 0x00005070
 #define G82_DISP                                      /* cl5070.h */ 0x00008270
@@ -50,6 +51,8 @@
 #define GK110_DISP                                    /* cl5070.h */ 0x00009270
 #define GM107_DISP                                    /* cl5070.h */ 0x00009470
 #define GM200_DISP                                    /* cl5070.h */ 0x00009570
+#define GP100_DISP                                    /* cl5070.h */ 0x00009770
+#define GP104_DISP                                    /* cl5070.h */ 0x00009870
 
 #define NV31_MPEG                                                    0x00003174
 #define G82_MPEG                                                     0x00008274
@@ -86,6 +89,8 @@
 #define GK110_DISP_CORE_CHANNEL_DMA                   /* cl507d.h */ 0x0000927d
 #define GM107_DISP_CORE_CHANNEL_DMA                   /* cl507d.h */ 0x0000947d
 #define GM200_DISP_CORE_CHANNEL_DMA                   /* cl507d.h */ 0x0000957d
+#define GP100_DISP_CORE_CHANNEL_DMA                   /* cl507d.h */ 0x0000977d
+#define GP104_DISP_CORE_CHANNEL_DMA                   /* cl507d.h */ 0x0000987d
 
 #define NV50_DISP_OVERLAY_CHANNEL_DMA                 /* cl507e.h */ 0x0000507e
 #define G82_DISP_OVERLAY_CHANNEL_DMA                  /* cl507e.h */ 0x0000827e
@@ -105,6 +110,8 @@
 #define MAXWELL_A                                     /* cl9097.h */ 0x0000b097
 #define MAXWELL_B                                     /* cl9097.h */ 0x0000b197
 
+#define PASCAL_A                                      /* cl9097.h */ 0x0000c097
+
 #define NV74_BSP                                                     0x000074b0
 
 #define GT212_MSVLD                                                  0x000085b1
@@ -128,6 +135,8 @@
 #define FERMI_DMA                                                    0x000090b5
 #define KEPLER_DMA_COPY_A                                            0x0000a0b5
 #define MAXWELL_DMA_COPY_A                                           0x0000b0b5
+#define PASCAL_DMA_COPY_A                                            0x0000c0b5
+#define PASCAL_DMA_COPY_B                                            0x0000c1b5
 
 #define FERMI_DECOMPRESS                                             0x000090b8
 
@@ -137,6 +146,7 @@
 #define KEPLER_COMPUTE_B                                             0x0000a1c0
 #define MAXWELL_COMPUTE_A                                            0x0000b0c0
 #define MAXWELL_COMPUTE_B                                            0x0000b1c0
+#define PASCAL_COMPUTE_A                                             0x0000c0c0
 
 #define NV74_CIPHER                                                  0x000074c1
 #endif
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/core/device.h b/drivers/gpu/drm/nouveau/include/nvkm/core/device.h
index c612dc1f1eb4..7ea8aa7ca408 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/core/device.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/core/device.h
@@ -16,9 +16,9 @@ enum nvkm_devidx {
 	NVKM_SUBDEV_MC,
 	NVKM_SUBDEV_BUS,
 	NVKM_SUBDEV_TIMER,
+	NVKM_SUBDEV_INSTMEM,
 	NVKM_SUBDEV_FB,
 	NVKM_SUBDEV_LTC,
-	NVKM_SUBDEV_INSTMEM,
 	NVKM_SUBDEV_MMU,
 	NVKM_SUBDEV_BAR,
 	NVKM_SUBDEV_PMU,
@@ -33,7 +33,10 @@ enum nvkm_devidx {
 	NVKM_ENGINE_CE0,
 	NVKM_ENGINE_CE1,
 	NVKM_ENGINE_CE2,
-	NVKM_ENGINE_CE_LAST = NVKM_ENGINE_CE2,
+	NVKM_ENGINE_CE3,
+	NVKM_ENGINE_CE4,
+	NVKM_ENGINE_CE5,
+	NVKM_ENGINE_CE_LAST = NVKM_ENGINE_CE5,
 
 	NVKM_ENGINE_CIPHER,
 	NVKM_ENGINE_DISP,
@@ -50,7 +53,8 @@ enum nvkm_devidx {
 
 	NVKM_ENGINE_NVENC0,
 	NVKM_ENGINE_NVENC1,
-	NVKM_ENGINE_NVENC_LAST = NVKM_ENGINE_NVENC1,
+	NVKM_ENGINE_NVENC2,
+	NVKM_ENGINE_NVENC_LAST = NVKM_ENGINE_NVENC2,
 
 	NVKM_ENGINE_NVDEC,
 	NVKM_ENGINE_PM,
@@ -102,6 +106,7 @@ struct nvkm_device {
 		NV_C0    = 0xc0,
 		NV_E0    = 0xe0,
 		GM100    = 0x110,
+		GP100    = 0x130,
 	} card_type;
 	u32 chipset;
 	u8  chiprev;
@@ -136,7 +141,7 @@ struct nvkm_device {
 	struct nvkm_volt *volt;
 
 	struct nvkm_engine *bsp;
-	struct nvkm_engine *ce[3];
+	struct nvkm_engine *ce[6];
 	struct nvkm_engine *cipher;
 	struct nvkm_disp *disp;
 	struct nvkm_dma *dma;
@@ -149,7 +154,7 @@ struct nvkm_device {
 	struct nvkm_engine *mspdec;
 	struct nvkm_engine *msppp;
 	struct nvkm_engine *msvld;
-	struct nvkm_engine *nvenc[2];
+	struct nvkm_engine *nvenc[3];
 	struct nvkm_engine *nvdec;
 	struct nvkm_pm *pm;
 	struct nvkm_engine *sec;
@@ -170,7 +175,6 @@ struct nvkm_device_func {
 	void (*fini)(struct nvkm_device *, bool suspend);
 	resource_size_t (*resource_addr)(struct nvkm_device *, unsigned bar);
 	resource_size_t (*resource_size)(struct nvkm_device *, unsigned bar);
-	bool cpu_coherent;
 };
 
 struct nvkm_device_quirk {
@@ -206,7 +210,7 @@ struct nvkm_device_chip {
 	int (*volt    )(struct nvkm_device *, int idx, struct nvkm_volt **);
 
 	int (*bsp     )(struct nvkm_device *, int idx, struct nvkm_engine **);
-	int (*ce[3]   )(struct nvkm_device *, int idx, struct nvkm_engine **);
+	int (*ce[6]   )(struct nvkm_device *, int idx, struct nvkm_engine **);
 	int (*cipher  )(struct nvkm_device *, int idx, struct nvkm_engine **);
 	int (*disp    )(struct nvkm_device *, int idx, struct nvkm_disp **);
 	int (*dma     )(struct nvkm_device *, int idx, struct nvkm_dma **);
@@ -219,7 +223,7 @@ struct nvkm_device_chip {
 	int (*mspdec  )(struct nvkm_device *, int idx, struct nvkm_engine **);
 	int (*msppp   )(struct nvkm_device *, int idx, struct nvkm_engine **);
 	int (*msvld   )(struct nvkm_device *, int idx, struct nvkm_engine **);
-	int (*nvenc[2])(struct nvkm_device *, int idx, struct nvkm_engine **);
+	int (*nvenc[3])(struct nvkm_device *, int idx, struct nvkm_engine **);
 	int (*nvdec   )(struct nvkm_device *, int idx, struct nvkm_engine **);
 	int (*pm      )(struct nvkm_device *, int idx, struct nvkm_pm **);
 	int (*sec     )(struct nvkm_device *, int idx, struct nvkm_engine **);
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/core/tegra.h b/drivers/gpu/drm/nouveau/include/nvkm/core/tegra.h
index b5370cb56e3c..e5c9b6268dcc 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/core/tegra.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/core/tegra.h
@@ -28,6 +28,7 @@ struct nvkm_device_tegra {
 	} iommu;
 
 	int gpu_speedo;
+	int gpu_speedo_id;
 };
 
 struct nvkm_device_tegra_func {
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/engine/ce.h b/drivers/gpu/drm/nouveau/include/nvkm/engine/ce.h
index 594d719ba41e..d3d26a1e215d 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/engine/ce.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/engine/ce.h
@@ -7,4 +7,6 @@ int gf100_ce_new(struct nvkm_device *, int, struct nvkm_engine **);
 int gk104_ce_new(struct nvkm_device *, int, struct nvkm_engine **);
 int gm107_ce_new(struct nvkm_device *, int, struct nvkm_engine **);
 int gm200_ce_new(struct nvkm_device *, int, struct nvkm_engine **);
+int gp100_ce_new(struct nvkm_device *, int, struct nvkm_engine **);
+int gp104_ce_new(struct nvkm_device *, int, struct nvkm_engine **);
 #endif
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/engine/disp.h b/drivers/gpu/drm/nouveau/include/nvkm/engine/disp.h
index d4fdce27b297..e82049667ce4 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/engine/disp.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/engine/disp.h
@@ -32,4 +32,6 @@ int gk104_disp_new(struct nvkm_device *, int, struct nvkm_disp **);
 int gk110_disp_new(struct nvkm_device *, int, struct nvkm_disp **);
 int gm107_disp_new(struct nvkm_device *, int, struct nvkm_disp **);
 int gm200_disp_new(struct nvkm_device *, int, struct nvkm_disp **);
+int gp100_disp_new(struct nvkm_device *, int, struct nvkm_disp **);
+int gp104_disp_new(struct nvkm_device *, int, struct nvkm_disp **);
 #endif
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/engine/fifo.h b/drivers/gpu/drm/nouveau/include/nvkm/engine/fifo.h
index 15ddfcf5e8db..ed92fec5292c 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/engine/fifo.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/engine/fifo.h
@@ -66,4 +66,5 @@ int gk20a_fifo_new(struct nvkm_device *, int, struct nvkm_fifo **);
 int gm107_fifo_new(struct nvkm_device *, int, struct nvkm_fifo **);
 int gm200_fifo_new(struct nvkm_device *, int, struct nvkm_fifo **);
 int gm20b_fifo_new(struct nvkm_device *, int, struct nvkm_fifo **);
+int gp100_fifo_new(struct nvkm_device *, int, struct nvkm_fifo **);
 #endif
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/engine/gr.h b/drivers/gpu/drm/nouveau/include/nvkm/engine/gr.h
index 6515f5810a26..89cf99307828 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/engine/gr.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/engine/gr.h
@@ -42,4 +42,5 @@ int gk20a_gr_new(struct nvkm_device *, int, struct nvkm_gr **);
 int gm107_gr_new(struct nvkm_device *, int, struct nvkm_gr **);
 int gm200_gr_new(struct nvkm_device *, int, struct nvkm_gr **);
 int gm20b_gr_new(struct nvkm_device *, int, struct nvkm_gr **);
+int gp100_gr_new(struct nvkm_device *, int, struct nvkm_gr **);
 #endif
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/bios.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/bios.h
index e39a1fea930b..a72f3290528a 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/bios.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/bios.h
@@ -7,6 +7,9 @@ struct nvkm_bios {
 	u32 size;
 	u8 *data;
 
+	u32 image0_size;
+	u32 imaged_addr;
+
 	u32 bmp_offset;
 	u32 bit_offset;
 
@@ -22,10 +25,9 @@ struct nvkm_bios {
 u8  nvbios_checksum(const u8 *data, int size);
 u16 nvbios_findstr(const u8 *data, int size, const char *str, int len);
 int nvbios_memcmp(struct nvkm_bios *, u32 addr, const char *, u32 len);
-
-#define nvbios_rd08(b,o) (b)->data[(o)]
-#define nvbios_rd16(b,o) get_unaligned_le16(&(b)->data[(o)])
-#define nvbios_rd32(b,o) get_unaligned_le32(&(b)->data[(o)])
+u8  nvbios_rd08(struct nvkm_bios *, u32 addr);
+u16 nvbios_rd16(struct nvkm_bios *, u32 addr);
+u32 nvbios_rd32(struct nvkm_bios *, u32 addr);
 
 int nvkm_bios_new(struct nvkm_device *, int, struct nvkm_bios **);
 #endif
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/bios/disp.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/bios/disp.h
index db10c11f0595..c5a6ebd5a478 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/bios/disp.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/bios/disp.h
@@ -25,7 +25,8 @@ u16 nvbios_outp_match(struct nvkm_bios *, u16 type, u16 mask,
 		      u8 *ver, u8 *hdr, u8 *cnt, u8 *len, struct nvbios_outp *);
 
 struct nvbios_ocfg {
-	u16 match;
+	u8  proto;
+	u8  flags;
 	u16 clkcmp[2];
 };
 
@@ -33,7 +34,7 @@ u16 nvbios_ocfg_entry(struct nvkm_bios *, u16 outp, u8 idx,
 		      u8 *ver, u8 *hdr, u8 *cnt, u8 *len);
 u16 nvbios_ocfg_parse(struct nvkm_bios *, u16 outp, u8 idx,
 		      u8 *ver, u8 *hdr, u8 *cnt, u8 *len, struct nvbios_ocfg *);
-u16 nvbios_ocfg_match(struct nvkm_bios *, u16 outp, u16 type,
+u16 nvbios_ocfg_match(struct nvkm_bios *, u16 outp, u8 proto, u8 flags,
 		      u8 *ver, u8 *hdr, u8 *cnt, u8 *len, struct nvbios_ocfg *);
 u16 nvbios_oclk_match(struct nvkm_bios *, u16 cmp, u32 khz);
 #endif
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/fb.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/fb.h
index 0a734fd06acf..3a410275fa71 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/fb.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/fb.h
@@ -56,6 +56,8 @@ struct nvkm_fb {
 		int regions;
 	} tile;
 
+	u8 page;
+
 	struct nvkm_memory *mmu_rd;
 	struct nvkm_memory *mmu_wr;
 };
@@ -91,6 +93,8 @@ int gk104_fb_new(struct nvkm_device *, int, struct nvkm_fb **);
 int gk20a_fb_new(struct nvkm_device *, int, struct nvkm_fb **);
 int gm107_fb_new(struct nvkm_device *, int, struct nvkm_fb **);
 int gm200_fb_new(struct nvkm_device *, int, struct nvkm_fb **);
+int gp100_fb_new(struct nvkm_device *, int, struct nvkm_fb **);
+int gp104_fb_new(struct nvkm_device *, int, struct nvkm_fb **);
 
 #include <subdev/bios.h>
 #include <subdev/bios/ramcfg.h>
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/ltc.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/ltc.h
index c6b90b6543b3..cd755baf9cab 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/ltc.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/ltc.h
@@ -38,4 +38,5 @@ int gk104_ltc_new(struct nvkm_device *, int, struct nvkm_ltc **);
 int gk20a_ltc_new(struct nvkm_device *, int, struct nvkm_ltc **);
 int gm107_ltc_new(struct nvkm_device *, int, struct nvkm_ltc **);
 int gm200_ltc_new(struct nvkm_device *, int, struct nvkm_ltc **);
+int gp100_ltc_new(struct nvkm_device *, int, struct nvkm_ltc **);
 #endif
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/mc.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/mc.h
index 2e80682b2da1..27d25b18d85c 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/mc.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/mc.h
@@ -7,11 +7,14 @@ struct nvkm_mc {
 	struct nvkm_subdev subdev;
 };
 
-void nvkm_mc_intr(struct nvkm_mc *, bool *handled);
-void nvkm_mc_intr_unarm(struct nvkm_mc *);
-void nvkm_mc_intr_rearm(struct nvkm_mc *);
-void nvkm_mc_reset(struct nvkm_mc *, enum nvkm_devidx);
-void nvkm_mc_unk260(struct nvkm_mc *, u32 data);
+void nvkm_mc_enable(struct nvkm_device *, enum nvkm_devidx);
+void nvkm_mc_disable(struct nvkm_device *, enum nvkm_devidx);
+void nvkm_mc_reset(struct nvkm_device *, enum nvkm_devidx);
+void nvkm_mc_intr(struct nvkm_device *, bool *handled);
+void nvkm_mc_intr_unarm(struct nvkm_device *);
+void nvkm_mc_intr_rearm(struct nvkm_device *);
+void nvkm_mc_intr_mask(struct nvkm_device *, enum nvkm_devidx, bool enable);
+void nvkm_mc_unk260(struct nvkm_device *, u32 data);
 
 int nv04_mc_new(struct nvkm_device *, int, struct nvkm_mc **);
 int nv11_mc_new(struct nvkm_device *, int, struct nvkm_mc **);
@@ -24,4 +27,5 @@ int gt215_mc_new(struct nvkm_device *, int, struct nvkm_mc **);
 int gf100_mc_new(struct nvkm_device *, int, struct nvkm_mc **);
 int gk104_mc_new(struct nvkm_device *, int, struct nvkm_mc **);
 int gk20a_mc_new(struct nvkm_device *, int, struct nvkm_mc **);
+int gp100_mc_new(struct nvkm_device *, int, struct nvkm_mc **);
 #endif
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/pci.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/pci.h
index ddb913889d7e..e6523e2cea9f 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/pci.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/pci.h
@@ -47,6 +47,7 @@ int g94_pci_new(struct nvkm_device *, int, struct nvkm_pci **);
 int gf100_pci_new(struct nvkm_device *, int, struct nvkm_pci **);
 int gf106_pci_new(struct nvkm_device *, int, struct nvkm_pci **);
 int gk104_pci_new(struct nvkm_device *, int, struct nvkm_pci **);
+int gp100_pci_new(struct nvkm_device *, int, struct nvkm_pci **);
 
 /* pcie functions */
 int nvkm_pcie_set_link(struct nvkm_pci *, enum nvkm_pcie_speed, u8 width);
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/secboot.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/secboot.h
index c6edd95a5b69..b04c38c07761 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/secboot.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/secboot.h
@@ -43,9 +43,8 @@ struct nvkm_secboot {
 	const struct nvkm_secboot_func *func;
 	struct nvkm_subdev subdev;
 
+	enum nvkm_devidx devidx;
 	u32 base;
-	u32 irq_mask;
-	u32 enable_mask;
 };
 #define nvkm_secboot(p) container_of((p), struct nvkm_secboot, subdev)
 
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/top.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/top.h
index 8fb575a92c48..71ebbfd4484f 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/top.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/top.h
@@ -8,10 +8,11 @@ struct nvkm_top {
 	struct list_head device;
 };
 
-u32 nvkm_top_reset(struct nvkm_top *, enum nvkm_devidx);
-u32 nvkm_top_intr(struct nvkm_top *, u32 intr, u64 *subdevs);
-enum nvkm_devidx nvkm_top_fault(struct nvkm_top *, int fault);
-enum nvkm_devidx nvkm_top_engine(struct nvkm_top *, int, int *runl, int *engn);
+u32 nvkm_top_reset(struct nvkm_device *, enum nvkm_devidx);
+u32 nvkm_top_intr(struct nvkm_device *, u32 intr, u64 *subdevs);
+u32 nvkm_top_intr_mask(struct nvkm_device *, enum nvkm_devidx);
+enum nvkm_devidx nvkm_top_fault(struct nvkm_device *, int fault);
+enum nvkm_devidx nvkm_top_engine(struct nvkm_device *, int, int *runl, int *engn);
 
 int gk104_top_new(struct nvkm_device *, int, struct nvkm_top **);
 #endif
diff --git a/drivers/gpu/drm/nouveau/include/nvkm/subdev/volt.h b/drivers/gpu/drm/nouveau/include/nvkm/subdev/volt.h
index feff55cff05b..b765f4ffcde6 100644
--- a/drivers/gpu/drm/nouveau/include/nvkm/subdev/volt.h
+++ b/drivers/gpu/drm/nouveau/include/nvkm/subdev/volt.h
@@ -12,6 +12,9 @@ struct nvkm_volt {
 		u32 uv;
 		u8 vid;
 	} vid[256];
+
+	u32 max_uv;
+	u32 min_uv;
 };
 
 int nvkm_volt_get(struct nvkm_volt *);
diff --git a/drivers/gpu/drm/nouveau/nouveau_abi16.c b/drivers/gpu/drm/nouveau/nouveau_abi16.c
index eb7de487a2b3..7bd4683216d0 100644
--- a/drivers/gpu/drm/nouveau/nouveau_abi16.c
+++ b/drivers/gpu/drm/nouveau/nouveau_abi16.c
@@ -100,6 +100,7 @@ nouveau_abi16_swclass(struct nouveau_drm *drm)
 	case NV_DEVICE_INFO_V0_FERMI:
 	case NV_DEVICE_INFO_V0_KEPLER:
 	case NV_DEVICE_INFO_V0_MAXWELL:
+	case NV_DEVICE_INFO_V0_PASCAL:
 		return NVIF_CLASS_SW_GF100;
 	}
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index 5e3f3e826476..528bdeffb339 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -209,8 +209,7 @@ nouveau_bo_new(struct drm_device *dev, int size, int align,
 	nvbo->tile_flags = tile_flags;
 	nvbo->bo.bdev = &drm->ttm.bdev;
 
-	if (!nvxx_device(&drm->device)->func->cpu_coherent)
-		nvbo->force_coherent = flags & TTM_PL_FLAG_UNCACHED;
+	nvbo->force_coherent = flags & TTM_PL_FLAG_UNCACHED;
 
 	nvbo->page_shift = 12;
 	if (drm->client.vm) {
@@ -424,13 +423,7 @@ nouveau_bo_map(struct nouveau_bo *nvbo)
 	if (ret)
 		return ret;
 
-	/*
-	 * TTM buffers allocated using the DMA API already have a mapping, let's
-	 * use it instead.
-	 */
-	if (!nvbo->force_coherent)
-		ret = ttm_bo_kmap(&nvbo->bo, 0, nvbo->bo.mem.num_pages,
-				  &nvbo->kmap);
+	ret = ttm_bo_kmap(&nvbo->bo, 0, nvbo->bo.mem.num_pages, &nvbo->kmap);
 
 	ttm_bo_unreserve(&nvbo->bo);
 	return ret;
@@ -442,12 +435,7 @@ nouveau_bo_unmap(struct nouveau_bo *nvbo)
 	if (!nvbo)
 		return;
 
-	/*
-	 * TTM buffers allocated using the DMA API already had a coherent
-	 * mapping which we used, no need to unmap.
-	 */
-	if (!nvbo->force_coherent)
-		ttm_bo_kunmap(&nvbo->kmap);
+	ttm_bo_kunmap(&nvbo->kmap);
 }
 
 void
@@ -506,35 +494,13 @@ nouveau_bo_validate(struct nouveau_bo *nvbo, bool interruptible,
 	return 0;
 }
 
-static inline void *
-_nouveau_bo_mem_index(struct nouveau_bo *nvbo, unsigned index, void *mem, u8 sz)
-{
-	struct ttm_dma_tt *dma_tt;
-	u8 *m = mem;
-
-	index *= sz;
-
-	if (m) {
-		/* kmap'd address, return the corresponding offset */
-		m += index;
-	} else {
-		/* DMA-API mapping, lookup the right address */
-		dma_tt = (struct ttm_dma_tt *)nvbo->bo.ttm;
-		m = dma_tt->cpu_address[index / PAGE_SIZE];
-		m += index % PAGE_SIZE;
-	}
-
-	return m;
-}
-#define nouveau_bo_mem_index(o, i, m) _nouveau_bo_mem_index(o, i, m, sizeof(*m))
-
 void
 nouveau_bo_wr16(struct nouveau_bo *nvbo, unsigned index, u16 val)
 {
 	bool is_iomem;
 	u16 *mem = ttm_kmap_obj_virtual(&nvbo->kmap, &is_iomem);
 
-	mem = nouveau_bo_mem_index(nvbo, index, mem);
+	mem += index;
 
 	if (is_iomem)
 		iowrite16_native(val, (void __force __iomem *)mem);
@@ -548,7 +514,7 @@ nouveau_bo_rd32(struct nouveau_bo *nvbo, unsigned index)
 	bool is_iomem;
 	u32 *mem = ttm_kmap_obj_virtual(&nvbo->kmap, &is_iomem);
 
-	mem = nouveau_bo_mem_index(nvbo, index, mem);
+	mem += index;
 
 	if (is_iomem)
 		return ioread32_native((void __force __iomem *)mem);
@@ -562,7 +528,7 @@ nouveau_bo_wr32(struct nouveau_bo *nvbo, unsigned index, u32 val)
 	bool is_iomem;
 	u32 *mem = ttm_kmap_obj_virtual(&nvbo->kmap, &is_iomem);
 
-	mem = nouveau_bo_mem_index(nvbo, index, mem);
+	mem += index;
 
 	if (is_iomem)
 		iowrite32_native(val, (void __force __iomem *)mem);
@@ -1082,7 +1048,6 @@ nouveau_bo_move_m2mf(struct ttm_buffer_object *bo, int evict, bool intr,
 				ret = ttm_bo_move_accel_cleanup(bo,
 								&fence->base,
 								evict,
-								no_wait_gpu,
 								new_mem);
 				nouveau_fence_unref(&fence);
 			}
@@ -1104,6 +1069,10 @@ nouveau_bo_move_init(struct nouveau_drm *drm)
 			    struct ttm_mem_reg *, struct ttm_mem_reg *);
 		int (*init)(struct nouveau_channel *, u32 handle);
 	} _methods[] = {
+		{  "COPY", 4, 0xc1b5, nve0_bo_move_copy, nve0_bo_move_init },
+		{  "GRCE", 0, 0xc1b5, nve0_bo_move_copy, nvc0_bo_move_init },
+		{  "COPY", 4, 0xc0b5, nve0_bo_move_copy, nve0_bo_move_init },
+		{  "GRCE", 0, 0xc0b5, nve0_bo_move_copy, nvc0_bo_move_init },
 		{  "COPY", 4, 0xb0b5, nve0_bo_move_copy, nve0_bo_move_init },
 		{  "GRCE", 0, 0xb0b5, nve0_bo_move_copy, nvc0_bo_move_init },
 		{  "COPY", 4, 0xa0b5, nve0_bo_move_copy, nve0_bo_move_init },
@@ -1289,6 +1258,10 @@ nouveau_bo_move(struct ttm_buffer_object *bo, bool evict, bool intr,
 	struct nouveau_drm_tile *new_tile = NULL;
 	int ret = 0;
 
+	ret = ttm_bo_wait(bo, intr, no_wait_gpu);
+	if (ret)
+		return ret;
+
 	if (nvbo->pin_refcnt)
 		NV_WARN(drm, "Moving pinned object %p!\n", nvbo);
 
@@ -1324,7 +1297,7 @@ nouveau_bo_move(struct ttm_buffer_object *bo, bool evict, bool intr,
 	/* Fallback to software copy. */
 	ret = ttm_bo_wait(bo, intr, no_wait_gpu);
 	if (ret == 0)
-		ret = ttm_bo_move_memcpy(bo, evict, no_wait_gpu, new_mem);
+		ret = ttm_bo_move_memcpy(bo, evict, intr, no_wait_gpu, new_mem);
 
 out:
 	if (drm->device.info.family < NV_DEVICE_INFO_V0_TESLA) {
@@ -1488,14 +1461,6 @@ nouveau_ttm_tt_populate(struct ttm_tt *ttm)
 	dev = drm->dev;
 	pdev = device->dev;
 
-	/*
-	 * Objects matching this condition have been marked as force_coherent,
-	 * so use the DMA API for them.
-	 */
-	if (!nvxx_device(&drm->device)->func->cpu_coherent &&
-	    ttm->caching_state == tt_uncached)
-		return ttm_dma_populate(ttm_dma, dev->dev);
-
 #if IS_ENABLED(CONFIG_AGP)
 	if (drm->agp.bridge) {
 		return ttm_agp_tt_populate(ttm);
@@ -1553,16 +1518,6 @@ nouveau_ttm_tt_unpopulate(struct ttm_tt *ttm)
 	dev = drm->dev;
 	pdev = device->dev;
 
-	/*
-	 * Objects matching this condition have been marked as force_coherent,
-	 * so use the DMA API for them.
-	 */
-	if (!nvxx_device(&drm->device)->func->cpu_coherent &&
-	    ttm->caching_state == tt_uncached) {
-		ttm_dma_unpopulate(ttm_dma, dev->dev);
-		return;
-	}
-
 #if IS_ENABLED(CONFIG_AGP)
 	if (drm->agp.bridge) {
 		ttm_agp_tt_unpopulate(ttm);
diff --git a/drivers/gpu/drm/nouveau/nouveau_chan.c b/drivers/gpu/drm/nouveau/nouveau_chan.c
index b1d2527c5625..f9b3c811187e 100644
--- a/drivers/gpu/drm/nouveau/nouveau_chan.c
+++ b/drivers/gpu/drm/nouveau/nouveau_chan.c
@@ -191,7 +191,8 @@ static int
 nouveau_channel_ind(struct nouveau_drm *drm, struct nvif_device *device,
 		    u32 engine, struct nouveau_channel **pchan)
 {
-	static const u16 oclasses[] = { MAXWELL_CHANNEL_GPFIFO_A,
+	static const u16 oclasses[] = { PASCAL_CHANNEL_GPFIFO_A,
+					MAXWELL_CHANNEL_GPFIFO_A,
 					KEPLER_CHANNEL_GPFIFO_B,
 					KEPLER_CHANNEL_GPFIFO_A,
 					FERMI_CHANNEL_GPFIFO,
diff --git a/drivers/gpu/drm/nouveau/nouveau_display.c b/drivers/gpu/drm/nouveau/nouveau_display.c
index 6e97862d31a1..afbf557b23d4 100644
--- a/drivers/gpu/drm/nouveau/nouveau_display.c
+++ b/drivers/gpu/drm/nouveau/nouveau_display.c
@@ -495,6 +495,8 @@ nouveau_display_create(struct drm_device *dev)
 
 	if (nouveau_modeset != 2 && drm->vbios.dcb.entries) {
 		static const u16 oclass[] = {
+			GP104_DISP,
+			GP100_DISP,
 			GM200_DISP,
 			GM107_DISP,
 			GK110_DISP,
diff --git a/drivers/gpu/drm/nouveau/nouveau_drm.c b/drivers/gpu/drm/nouveau/nouveau_drm.c
index 1161f8b452f2..66c1280c0f1f 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_drm.c
@@ -198,6 +198,7 @@ nouveau_accel_init(struct nouveau_drm *drm)
 		case KEPLER_CHANNEL_GPFIFO_A:
 		case KEPLER_CHANNEL_GPFIFO_B:
 		case MAXWELL_CHANNEL_GPFIFO_A:
+		case PASCAL_CHANNEL_GPFIFO_A:
 			ret = nvc0_fence_create(drm);
 			break;
 		default:
@@ -316,7 +317,16 @@ static int nouveau_drm_probe(struct pci_dev *pdev,
 	if (vga_switcheroo_client_probe_defer(pdev))
 		return -EPROBE_DEFER;
 
-	/* remove conflicting drivers (vesafb, efifb etc) */
+	/* We need to check that the chipset is supported before booting
+	 * fbdev off the hardware, as there's no way to put it back.
+	 */
+	ret = nvkm_device_pci_new(pdev, NULL, "error", true, false, 0, &device);
+	if (ret)
+		return ret;
+
+	nvkm_device_del(&device);
+
+	/* Remove conflicting drivers (vesafb, efifb etc). */
 	aper = alloc_apertures(3);
 	if (!aper)
 		return -ENOMEM;
@@ -430,6 +440,11 @@ nouveau_drm_load(struct drm_device *dev, unsigned long flags)
 	nouveau_vga_init(drm);
 
 	if (drm->device.info.family >= NV_DEVICE_INFO_V0_TESLA) {
+		if (!nvxx_device(&drm->device)->mmu) {
+			ret = -ENOSYS;
+			goto fail_device;
+		}
+
 		ret = nvkm_vm_new(nvxx_device(&drm->device), 0, (1ULL << 40),
 				  0x1000, NULL, &drm->client.vm);
 		if (ret)
diff --git a/drivers/gpu/drm/nouveau/nouveau_fbcon.c b/drivers/gpu/drm/nouveau/nouveau_fbcon.c
index 57aaf98a26f9..d1f248fd3506 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fbcon.c
+++ b/drivers/gpu/drm/nouveau/nouveau_fbcon.c
@@ -552,6 +552,8 @@ nouveau_fbcon_init(struct drm_device *dev)
 	if (ret)
 		goto fini;
 
+	if (fbcon->helper.fbdev)
+		fbcon->helper.fbdev->pixmap.buf_align = 4;
 	return 0;
 
 fini:
diff --git a/drivers/gpu/drm/nouveau/nouveau_hwmon.c b/drivers/gpu/drm/nouveau/nouveau_hwmon.c
index 1ff4166af26e..71f764bf4cc6 100644
--- a/drivers/gpu/drm/nouveau/nouveau_hwmon.c
+++ b/drivers/gpu/drm/nouveau/nouveau_hwmon.c
@@ -535,6 +535,40 @@ static SENSOR_DEVICE_ATTR(in0_input, S_IRUGO,
 			  nouveau_hwmon_get_in0_input, NULL, 0);
 
 static ssize_t
+nouveau_hwmon_get_in0_min(struct device *d,
+			    struct device_attribute *a, char *buf)
+{
+	struct drm_device *dev = dev_get_drvdata(d);
+	struct nouveau_drm *drm = nouveau_drm(dev);
+	struct nvkm_volt *volt = nvxx_volt(&drm->device);
+
+	if (!volt || !volt->min_uv)
+		return -ENODEV;
+
+	return sprintf(buf, "%i\n", volt->min_uv / 1000);
+}
+
+static SENSOR_DEVICE_ATTR(in0_min, S_IRUGO,
+			  nouveau_hwmon_get_in0_min, NULL, 0);
+
+static ssize_t
+nouveau_hwmon_get_in0_max(struct device *d,
+			    struct device_attribute *a, char *buf)
+{
+	struct drm_device *dev = dev_get_drvdata(d);
+	struct nouveau_drm *drm = nouveau_drm(dev);
+	struct nvkm_volt *volt = nvxx_volt(&drm->device);
+
+	if (!volt || !volt->max_uv)
+		return -ENODEV;
+
+	return sprintf(buf, "%i\n", volt->max_uv / 1000);
+}
+
+static SENSOR_DEVICE_ATTR(in0_max, S_IRUGO,
+			  nouveau_hwmon_get_in0_max, NULL, 0);
+
+static ssize_t
 nouveau_hwmon_get_in0_label(struct device *d,
 			    struct device_attribute *a, char *buf)
 {
@@ -594,6 +628,8 @@ static struct attribute *hwmon_pwm_fan_attributes[] = {
 
 static struct attribute *hwmon_in0_attributes[] = {
 	&sensor_dev_attr_in0_input.dev_attr.attr,
+	&sensor_dev_attr_in0_min.dev_attr.attr,
+	&sensor_dev_attr_in0_max.dev_attr.attr,
 	&sensor_dev_attr_in0_label.dev_attr.attr,
 	NULL
 };
diff --git a/drivers/gpu/drm/nouveau/nouveau_ttm.c b/drivers/gpu/drm/nouveau/nouveau_ttm.c
index bcee91497eb9..1825dbc33192 100644
--- a/drivers/gpu/drm/nouveau/nouveau_ttm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_ttm.c
@@ -164,6 +164,7 @@ nouveau_gart_manager_new(struct ttm_mem_type_manager *man,
 	case NV_DEVICE_INFO_V0_FERMI:
 	case NV_DEVICE_INFO_V0_KEPLER:
 	case NV_DEVICE_INFO_V0_MAXWELL:
+	case NV_DEVICE_INFO_V0_PASCAL:
 		node->memtype = (nvbo->tile_flags & 0xff00) >> 8;
 		break;
 	default:
diff --git a/drivers/gpu/drm/nouveau/nv04_fbcon.c b/drivers/gpu/drm/nouveau/nv04_fbcon.c
index 0f3e4bb411cc..7d9248b8c664 100644
--- a/drivers/gpu/drm/nouveau/nv04_fbcon.c
+++ b/drivers/gpu/drm/nouveau/nv04_fbcon.c
@@ -82,7 +82,6 @@ nv04_fbcon_imageblit(struct fb_info *info, const struct fb_image *image)
 	uint32_t fg;
 	uint32_t bg;
 	uint32_t dsize;
-	uint32_t width;
 	uint32_t *data = (uint32_t *)image->data;
 	int ret;
 
@@ -93,9 +92,6 @@ nv04_fbcon_imageblit(struct fb_info *info, const struct fb_image *image)
 	if (ret)
 		return ret;
 
-	width = ALIGN(image->width, 8);
-	dsize = ALIGN(width * image->height, 32) >> 5;
-
 	if (info->fix.visual == FB_VISUAL_TRUECOLOR ||
 	    info->fix.visual == FB_VISUAL_DIRECTCOLOR) {
 		fg = ((uint32_t *) info->pseudo_palette)[image->fg_color];
@@ -111,10 +107,11 @@ nv04_fbcon_imageblit(struct fb_info *info, const struct fb_image *image)
 			 ((image->dx + image->width) & 0xffff));
 	OUT_RING(chan, bg);
 	OUT_RING(chan, fg);
-	OUT_RING(chan, (image->height << 16) | width);
+	OUT_RING(chan, (image->height << 16) | image->width);
 	OUT_RING(chan, (image->height << 16) | image->width);
 	OUT_RING(chan, (image->dy << 16) | (image->dx & 0xffff));
 
+	dsize = ALIGN(image->width * image->height, 32) >> 5;
 	while (dsize) {
 		int iter_len = dsize > 128 ? 128 : dsize;
 
diff --git a/drivers/gpu/drm/nouveau/nv50_display.c b/drivers/gpu/drm/nouveau/nv50_display.c
index 7a7788212df7..7d0edcbcfca7 100644
--- a/drivers/gpu/drm/nouveau/nv50_display.c
+++ b/drivers/gpu/drm/nouveau/nv50_display.c
@@ -297,6 +297,8 @@ nv50_core_create(struct nvif_device *device, struct nvif_object *disp,
 		.pushbuf = 0xb0007d00,
 	};
 	static const s32 oclass[] = {
+		GP104_DISP_CORE_CHANNEL_DMA,
+		GP100_DISP_CORE_CHANNEL_DMA,
 		GM200_DISP_CORE_CHANNEL_DMA,
 		GM107_DISP_CORE_CHANNEL_DMA,
 		GK110_DISP_CORE_CHANNEL_DMA,
diff --git a/drivers/gpu/drm/nouveau/nv50_fbcon.c b/drivers/gpu/drm/nouveau/nv50_fbcon.c
index 33d9ee0fac40..1aeb698e9707 100644
--- a/drivers/gpu/drm/nouveau/nv50_fbcon.c
+++ b/drivers/gpu/drm/nouveau/nv50_fbcon.c
@@ -95,7 +95,7 @@ nv50_fbcon_imageblit(struct fb_info *info, const struct fb_image *image)
 	struct nouveau_fbdev *nfbdev = info->par;
 	struct nouveau_drm *drm = nouveau_drm(nfbdev->dev);
 	struct nouveau_channel *chan = drm->channel;
-	uint32_t width, dwords, *data = (uint32_t *)image->data;
+	uint32_t dwords, *data = (uint32_t *)image->data;
 	uint32_t mask = ~(~0 >> (32 - info->var.bits_per_pixel));
 	uint32_t *palette = info->pseudo_palette;
 	int ret;
@@ -107,9 +107,6 @@ nv50_fbcon_imageblit(struct fb_info *info, const struct fb_image *image)
 	if (ret)
 		return ret;
 
-	width = ALIGN(image->width, 32);
-	dwords = (width * image->height) >> 5;
-
 	BEGIN_NV04(chan, NvSub2D, 0x0814, 2);
 	if (info->fix.visual == FB_VISUAL_TRUECOLOR ||
 	    info->fix.visual == FB_VISUAL_DIRECTCOLOR) {
@@ -128,6 +125,7 @@ nv50_fbcon_imageblit(struct fb_info *info, const struct fb_image *image)
 	OUT_RING(chan, 0);
 	OUT_RING(chan, image->dy);
 
+	dwords = ALIGN(image->width * image->height, 32) >> 5;
 	while (dwords) {
 		int push = dwords > 2047 ? 2047 : dwords;
 
diff --git a/drivers/gpu/drm/nouveau/nvc0_fbcon.c b/drivers/gpu/drm/nouveau/nvc0_fbcon.c
index a0913359ac05..839f4c8c1805 100644
--- a/drivers/gpu/drm/nouveau/nvc0_fbcon.c
+++ b/drivers/gpu/drm/nouveau/nvc0_fbcon.c
@@ -95,7 +95,7 @@ nvc0_fbcon_imageblit(struct fb_info *info, const struct fb_image *image)
 	struct nouveau_fbdev *nfbdev = info->par;
 	struct nouveau_drm *drm = nouveau_drm(nfbdev->dev);
 	struct nouveau_channel *chan = drm->channel;
-	uint32_t width, dwords, *data = (uint32_t *)image->data;
+	uint32_t dwords, *data = (uint32_t *)image->data;
 	uint32_t mask = ~(~0 >> (32 - info->var.bits_per_pixel));
 	uint32_t *palette = info->pseudo_palette;
 	int ret;
@@ -107,9 +107,6 @@ nvc0_fbcon_imageblit(struct fb_info *info, const struct fb_image *image)
 	if (ret)
 		return ret;
 
-	width = ALIGN(image->width, 32);
-	dwords = (width * image->height) >> 5;
-
 	BEGIN_NVC0(chan, NvSub2D, 0x0814, 2);
 	if (info->fix.visual == FB_VISUAL_TRUECOLOR ||
 	    info->fix.visual == FB_VISUAL_DIRECTCOLOR) {
@@ -128,6 +125,7 @@ nvc0_fbcon_imageblit(struct fb_info *info, const struct fb_image *image)
 	OUT_RING  (chan, 0);
 	OUT_RING  (chan, image->dy);
 
+	dwords = ALIGN(image->width * image->height, 32) >> 5;
 	while (dwords) {
 		int push = dwords > 2047 ? 2047 : dwords;
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/core/subdev.c b/drivers/gpu/drm/nouveau/nvkm/core/subdev.c
index b18557858f19..19044aba265e 100644
--- a/drivers/gpu/drm/nouveau/nvkm/core/subdev.c
+++ b/drivers/gpu/drm/nouveau/nvkm/core/subdev.c
@@ -57,6 +57,9 @@ nvkm_subdev_name[NVKM_SUBDEV_NR] = {
 	[NVKM_ENGINE_CE0     ] = "ce0",
 	[NVKM_ENGINE_CE1     ] = "ce1",
 	[NVKM_ENGINE_CE2     ] = "ce2",
+	[NVKM_ENGINE_CE3     ] = "ce3",
+	[NVKM_ENGINE_CE4     ] = "ce4",
+	[NVKM_ENGINE_CE5     ] = "ce5",
 	[NVKM_ENGINE_CIPHER  ] = "cipher",
 	[NVKM_ENGINE_DISP    ] = "disp",
 	[NVKM_ENGINE_DMAOBJ  ] = "dma",
@@ -71,6 +74,7 @@ nvkm_subdev_name[NVKM_SUBDEV_NR] = {
 	[NVKM_ENGINE_MSVLD   ] = "msvld",
 	[NVKM_ENGINE_NVENC0  ] = "nvenc0",
 	[NVKM_ENGINE_NVENC1  ] = "nvenc1",
+	[NVKM_ENGINE_NVENC2  ] = "nvenc2",
 	[NVKM_ENGINE_NVDEC   ] = "nvdec",
 	[NVKM_ENGINE_PM      ] = "pm",
 	[NVKM_ENGINE_SEC     ] = "sec",
@@ -105,7 +109,7 @@ nvkm_subdev_fini(struct nvkm_subdev *subdev, bool suspend)
 		}
 	}
 
-	nvkm_mc_reset(device->mc, subdev->index);
+	nvkm_mc_reset(device, subdev->index);
 
 	time = ktime_to_us(ktime_get()) - time;
 	nvkm_trace(subdev, "%s completed in %lldus\n", action, time);
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/ce/Kbuild b/drivers/gpu/drm/nouveau/nvkm/engine/ce/Kbuild
index 9c19d59b47df..a4458a8eb30a 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/ce/Kbuild
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/ce/Kbuild
@@ -3,3 +3,5 @@ nvkm-y += nvkm/engine/ce/gf100.o
 nvkm-y += nvkm/engine/ce/gk104.o
 nvkm-y += nvkm/engine/ce/gm107.o
 nvkm-y += nvkm/engine/ce/gm200.o
+nvkm-y += nvkm/engine/ce/gp100.o
+nvkm-y += nvkm/engine/ce/gp104.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/ce/gp100.c b/drivers/gpu/drm/nouveau/nvkm/engine/ce/gp100.c
new file mode 100644
index 000000000000..c7710456bc30
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/ce/gp100.c
@@ -0,0 +1,102 @@
+/*
+ * Copyright 2015 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "priv.h"
+#include <core/enum.h>
+
+#include <nvif/class.h>
+
+static const struct nvkm_enum
+gp100_ce_launcherr_report[] = {
+	{ 0x0, "NO_ERR" },
+	{ 0x1, "2D_LAYER_EXCEEDS_DEPTH" },
+	{ 0x2, "INVALID_ALIGNMENT" },
+	{ 0x3, "MEM2MEM_RECT_OUT_OF_BOUNDS" },
+	{ 0x4, "SRC_LINE_EXCEEDS_PITCH" },
+	{ 0x5, "SRC_LINE_EXCEEDS_NEG_PITCH" },
+	{ 0x6, "DST_LINE_EXCEEDS_PITCH" },
+	{ 0x7, "DST_LINE_EXCEEDS_NEG_PITCH" },
+	{ 0x8, "BAD_SRC_PIXEL_COMP_REF" },
+	{ 0x9, "INVALID_VALUE" },
+	{ 0xa, "UNUSED_FIELD" },
+	{ 0xb, "INVALID_OPERATION" },
+	{ 0xc, "NO_RESOURCES" },
+	{ 0xd, "INVALID_CONFIG" },
+	{}
+};
+
+static void
+gp100_ce_intr_launcherr(struct nvkm_engine *ce, const u32 base)
+{
+	struct nvkm_subdev *subdev = &ce->subdev;
+	struct nvkm_device *device = subdev->device;
+	u32 stat = nvkm_rd32(device, 0x104418 + base);
+	const struct nvkm_enum *en =
+		nvkm_enum_find(gp100_ce_launcherr_report, stat & 0x0000000f);
+	nvkm_warn(subdev, "LAUNCHERR %08x [%s]\n", stat, en ? en->name : "");
+}
+
+void
+gp100_ce_intr(struct nvkm_engine *ce)
+{
+	const u32 base = (ce->subdev.index - NVKM_ENGINE_CE0) * 0x80;
+	struct nvkm_subdev *subdev = &ce->subdev;
+	struct nvkm_device *device = subdev->device;
+	u32 mask = nvkm_rd32(device, 0x10440c + base);
+	u32 intr = nvkm_rd32(device, 0x104410 + base) & mask;
+	if (intr & 0x00000001) { //XXX: guess
+		nvkm_warn(subdev, "BLOCKPIPE\n");
+		nvkm_wr32(device, 0x104410 + base, 0x00000001);
+		intr &= ~0x00000001;
+	}
+	if (intr & 0x00000002) { //XXX: guess
+		nvkm_warn(subdev, "NONBLOCKPIPE\n");
+		nvkm_wr32(device, 0x104410 + base, 0x00000002);
+		intr &= ~0x00000002;
+	}
+	if (intr & 0x00000004) {
+		gp100_ce_intr_launcherr(ce, base);
+		nvkm_wr32(device, 0x104410 + base, 0x00000004);
+		intr &= ~0x00000004;
+	}
+	if (intr) {
+		nvkm_warn(subdev, "intr %08x\n", intr);
+		nvkm_wr32(device, 0x104410 + base, intr);
+	}
+}
+
+static const struct nvkm_engine_func
+gp100_ce = {
+	.intr = gp100_ce_intr,
+	.sclass = {
+		{ -1, -1, PASCAL_DMA_COPY_A },
+		{}
+	}
+};
+
+int
+gp100_ce_new(struct nvkm_device *device, int index,
+	     struct nvkm_engine **pengine)
+{
+	return nvkm_engine_new_(&gp100_ce, device, index, true, pengine);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/ce/gp104.c b/drivers/gpu/drm/nouveau/nvkm/engine/ce/gp104.c
new file mode 100644
index 000000000000..20e019788a53
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/ce/gp104.c
@@ -0,0 +1,44 @@
+/*
+ * Copyright 2015 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "priv.h"
+#include <core/enum.h>
+
+#include <nvif/class.h>
+
+static const struct nvkm_engine_func
+gp104_ce = {
+	.intr = gp100_ce_intr,
+	.sclass = {
+		{ -1, -1, PASCAL_DMA_COPY_B },
+		{ -1, -1, PASCAL_DMA_COPY_A },
+		{}
+	}
+};
+
+int
+gp104_ce_new(struct nvkm_device *device, int index,
+	     struct nvkm_engine **pengine)
+{
+	return nvkm_engine_new_(&gp104_ce, device, index, true, pengine);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/ce/priv.h b/drivers/gpu/drm/nouveau/nvkm/engine/ce/priv.h
index e2fa8b161943..2dce405976ad 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/ce/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/ce/priv.h
@@ -4,4 +4,5 @@
 
 void gt215_ce_intr(struct nvkm_falcon *, struct nvkm_fifo_chan *);
 void gk104_ce_intr(struct nvkm_engine *);
+void gp100_ce_intr(struct nvkm_engine *);
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c b/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c
index 4572debcb0c9..7218a067a6c5 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c
@@ -2148,6 +2148,67 @@ nv12b_chipset = {
 	.sw = gf100_sw_new,
 };
 
+static const struct nvkm_device_chip
+nv130_chipset = {
+	.name = "GP100",
+	.bar = gf100_bar_new,
+	.bios = nvkm_bios_new,
+	.bus = gf100_bus_new,
+	.devinit = gm200_devinit_new,
+	.fb = gp100_fb_new,
+	.fuse = gm107_fuse_new,
+	.gpio = gk104_gpio_new,
+	.i2c = gm200_i2c_new,
+	.ibus = gm200_ibus_new,
+	.imem = nv50_instmem_new,
+	.ltc = gp100_ltc_new,
+	.mc = gp100_mc_new,
+	.mmu = gf100_mmu_new,
+	.secboot = gm200_secboot_new,
+	.pci = gp100_pci_new,
+	.timer = gk20a_timer_new,
+	.top = gk104_top_new,
+	.ce[0] = gp100_ce_new,
+	.ce[1] = gp100_ce_new,
+	.ce[2] = gp100_ce_new,
+	.ce[3] = gp100_ce_new,
+	.ce[4] = gp100_ce_new,
+	.ce[5] = gp100_ce_new,
+	.dma = gf119_dma_new,
+	.disp = gp100_disp_new,
+	.fifo = gp100_fifo_new,
+	.gr = gp100_gr_new,
+	.sw = gf100_sw_new,
+};
+
+static const struct nvkm_device_chip
+nv134_chipset = {
+	.name = "GP104",
+	.bar = gf100_bar_new,
+	.bios = nvkm_bios_new,
+	.bus = gf100_bus_new,
+	.devinit = gm200_devinit_new,
+	.fb = gp104_fb_new,
+	.fuse = gm107_fuse_new,
+	.gpio = gk104_gpio_new,
+	.i2c = gm200_i2c_new,
+	.ibus = gm200_ibus_new,
+	.imem = nv50_instmem_new,
+	.ltc = gp100_ltc_new,
+	.mc = gp100_mc_new,
+	.mmu = gf100_mmu_new,
+	.pci = gp100_pci_new,
+	.timer = gk20a_timer_new,
+	.top = gk104_top_new,
+	.ce[0] = gp104_ce_new,
+	.ce[1] = gp104_ce_new,
+	.ce[2] = gp104_ce_new,
+	.ce[3] = gp104_ce_new,
+	.disp = gp104_disp_new,
+	.dma = gf119_dma_new,
+	.fifo = gp100_fifo_new,
+};
+
 static int
 nvkm_device_event_ctor(struct nvkm_object *object, void *data, u32 size,
 		       struct nvkm_notify *notify)
@@ -2221,6 +2282,9 @@ nvkm_device_engine(struct nvkm_device *device, int index)
 	_(CE0    , device->ce[0]   ,  device->ce[0]);
 	_(CE1    , device->ce[1]   ,  device->ce[1]);
 	_(CE2    , device->ce[2]   ,  device->ce[2]);
+	_(CE3    , device->ce[3]   ,  device->ce[3]);
+	_(CE4    , device->ce[4]   ,  device->ce[4]);
+	_(CE5    , device->ce[5]   ,  device->ce[5]);
 	_(CIPHER , device->cipher  ,  device->cipher);
 	_(DISP   , device->disp    , &device->disp->engine);
 	_(DMAOBJ , device->dma     , &device->dma->engine);
@@ -2235,6 +2299,7 @@ nvkm_device_engine(struct nvkm_device *device, int index)
 	_(MSVLD  , device->msvld   ,  device->msvld);
 	_(NVENC0 , device->nvenc[0],  device->nvenc[0]);
 	_(NVENC1 , device->nvenc[1],  device->nvenc[1]);
+	_(NVENC2 , device->nvenc[2],  device->nvenc[2]);
 	_(NVDEC  , device->nvdec   ,  device->nvdec);
 	_(PM     , device->pm      , &device->pm->engine);
 	_(SEC    , device->sec     ,  device->sec);
@@ -2492,6 +2557,7 @@ nvkm_device_ctor(const struct nvkm_device_func *func,
 			case 0x100: device->card_type = NV_E0; break;
 			case 0x110:
 			case 0x120: device->card_type = GM100; break;
+			case 0x130: device->card_type = GP100; break;
 			default:
 				break;
 			}
@@ -2576,6 +2642,8 @@ nvkm_device_ctor(const struct nvkm_device_func *func,
 		case 0x124: device->chip = &nv124_chipset; break;
 		case 0x126: device->chip = &nv126_chipset; break;
 		case 0x12b: device->chip = &nv12b_chipset; break;
+		case 0x130: device->chip = &nv130_chipset; break;
+		case 0x134: device->chip = &nv134_chipset; break;
 		default:
 			nvdev_error(device, "unknown chipset (%08x)\n", boot0);
 			goto done;
@@ -2659,6 +2727,9 @@ nvkm_device_ctor(const struct nvkm_device_func *func,
 		_(NVKM_ENGINE_CE0     ,    ce[0]);
 		_(NVKM_ENGINE_CE1     ,    ce[1]);
 		_(NVKM_ENGINE_CE2     ,    ce[2]);
+		_(NVKM_ENGINE_CE3     ,    ce[3]);
+		_(NVKM_ENGINE_CE4     ,    ce[4]);
+		_(NVKM_ENGINE_CE5     ,    ce[5]);
 		_(NVKM_ENGINE_CIPHER  ,   cipher);
 		_(NVKM_ENGINE_DISP    ,     disp);
 		_(NVKM_ENGINE_DMAOBJ  ,      dma);
@@ -2673,6 +2744,7 @@ nvkm_device_ctor(const struct nvkm_device_func *func,
 		_(NVKM_ENGINE_MSVLD   ,    msvld);
 		_(NVKM_ENGINE_NVENC0  , nvenc[0]);
 		_(NVKM_ENGINE_NVENC1  , nvenc[1]);
+		_(NVKM_ENGINE_NVENC2  , nvenc[2]);
 		_(NVKM_ENGINE_NVDEC   ,    nvdec);
 		_(NVKM_ENGINE_PM      ,       pm);
 		_(NVKM_ENGINE_SEC     ,      sec);
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/device/pci.c b/drivers/gpu/drm/nouveau/nvkm/engine/device/pci.c
index 18fab3973ce5..b1b693219db3 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/device/pci.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/device/pci.c
@@ -1614,7 +1614,6 @@ nvkm_device_pci_func = {
 	.fini = nvkm_device_pci_fini,
 	.resource_addr = nvkm_device_pci_resource_addr,
 	.resource_size = nvkm_device_pci_resource_size,
-	.cpu_coherent = !IS_ENABLED(CONFIG_ARM) && !IS_ENABLED(CONFIG_ARM64),
 };
 
 int
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/device/tegra.c b/drivers/gpu/drm/nouveau/nvkm/engine/device/tegra.c
index ec12efb4689a..939682f18788 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/device/tegra.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/device/tegra.c
@@ -191,13 +191,11 @@ static irqreturn_t
 nvkm_device_tegra_intr(int irq, void *arg)
 {
 	struct nvkm_device_tegra *tdev = arg;
-	struct nvkm_mc *mc = tdev->device.mc;
+	struct nvkm_device *device = &tdev->device;
 	bool handled = false;
-	if (likely(mc)) {
-		nvkm_mc_intr_unarm(mc);
-		nvkm_mc_intr(mc, &handled);
-		nvkm_mc_intr_rearm(mc);
-	}
+	nvkm_mc_intr_unarm(device);
+	nvkm_mc_intr(device, &handled);
+	nvkm_mc_intr_rearm(device);
 	return handled ? IRQ_HANDLED : IRQ_NONE;
 }
 
@@ -247,7 +245,6 @@ nvkm_device_tegra_func = {
 	.fini = nvkm_device_tegra_fini,
 	.resource_addr = nvkm_device_tegra_resource_addr,
 	.resource_size = nvkm_device_tegra_resource_size,
-	.cpu_coherent = false,
 };
 
 int
@@ -313,6 +310,7 @@ nvkm_device_tegra_new(const struct nvkm_device_tegra_func *func,
 		goto remove;
 
 	tdev->gpu_speedo = tegra_sku_info.gpu_speedo_value;
+	tdev->gpu_speedo_id = tegra_sku_info.gpu_speedo_id;
 	ret = nvkm_device_ctor(&nvkm_device_tegra_func, NULL, &pdev->dev,
 			       NVKM_DEVICE_TEGRA, pdev->id, NULL,
 			       cfg, dbg, detect, mmio, subdev_mask,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/device/user.c b/drivers/gpu/drm/nouveau/nvkm/engine/device/user.c
index 137066426ed7..79a8f71cf788 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/device/user.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/device/user.c
@@ -102,6 +102,7 @@ nvkm_udevice_info(struct nvkm_udevice *udev, void *data, u32 size)
 	case NV_C0: args->v0.family = NV_DEVICE_INFO_V0_FERMI; break;
 	case NV_E0: args->v0.family = NV_DEVICE_INFO_V0_KEPLER; break;
 	case GM100: args->v0.family = NV_DEVICE_INFO_V0_MAXWELL; break;
+	case GP100: args->v0.family = NV_DEVICE_INFO_V0_PASCAL; break;
 	default:
 		args->v0.family = 0;
 		break;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/Kbuild b/drivers/gpu/drm/nouveau/nvkm/engine/disp/Kbuild
index a74c5dd27dc0..77a52b54a31e 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/Kbuild
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/Kbuild
@@ -10,6 +10,8 @@ nvkm-y += nvkm/engine/disp/gk104.o
 nvkm-y += nvkm/engine/disp/gk110.o
 nvkm-y += nvkm/engine/disp/gm107.o
 nvkm-y += nvkm/engine/disp/gm200.o
+nvkm-y += nvkm/engine/disp/gp100.o
+nvkm-y += nvkm/engine/disp/gp104.o
 
 nvkm-y += nvkm/engine/disp/outp.o
 nvkm-y += nvkm/engine/disp/outpdp.o
@@ -18,6 +20,7 @@ nvkm-y += nvkm/engine/disp/piornv50.o
 nvkm-y += nvkm/engine/disp/sornv50.o
 nvkm-y += nvkm/engine/disp/sorg94.o
 nvkm-y += nvkm/engine/disp/sorgf119.o
+nvkm-y += nvkm/engine/disp/sorgm107.o
 nvkm-y += nvkm/engine/disp/sorgm200.o
 nvkm-y += nvkm/engine/disp/dport.o
 
@@ -44,12 +47,15 @@ nvkm-y += nvkm/engine/disp/rootgk104.o
 nvkm-y += nvkm/engine/disp/rootgk110.o
 nvkm-y += nvkm/engine/disp/rootgm107.o
 nvkm-y += nvkm/engine/disp/rootgm200.o
+nvkm-y += nvkm/engine/disp/rootgp100.o
+nvkm-y += nvkm/engine/disp/rootgp104.o
 
 nvkm-y += nvkm/engine/disp/channv50.o
 nvkm-y += nvkm/engine/disp/changf119.o
 
 nvkm-y += nvkm/engine/disp/dmacnv50.o
 nvkm-y += nvkm/engine/disp/dmacgf119.o
+nvkm-y += nvkm/engine/disp/dmacgp104.o
 
 nvkm-y += nvkm/engine/disp/basenv50.o
 nvkm-y += nvkm/engine/disp/baseg84.o
@@ -58,6 +64,7 @@ nvkm-y += nvkm/engine/disp/basegt215.o
 nvkm-y += nvkm/engine/disp/basegf119.o
 nvkm-y += nvkm/engine/disp/basegk104.o
 nvkm-y += nvkm/engine/disp/basegk110.o
+nvkm-y += nvkm/engine/disp/basegp104.o
 
 nvkm-y += nvkm/engine/disp/corenv50.o
 nvkm-y += nvkm/engine/disp/coreg84.o
@@ -69,6 +76,8 @@ nvkm-y += nvkm/engine/disp/coregk104.o
 nvkm-y += nvkm/engine/disp/coregk110.o
 nvkm-y += nvkm/engine/disp/coregm107.o
 nvkm-y += nvkm/engine/disp/coregm200.o
+nvkm-y += nvkm/engine/disp/coregp100.o
+nvkm-y += nvkm/engine/disp/coregp104.o
 
 nvkm-y += nvkm/engine/disp/ovlynv50.o
 nvkm-y += nvkm/engine/disp/ovlyg84.o
@@ -76,6 +85,7 @@ nvkm-y += nvkm/engine/disp/ovlygt200.o
 nvkm-y += nvkm/engine/disp/ovlygt215.o
 nvkm-y += nvkm/engine/disp/ovlygf119.o
 nvkm-y += nvkm/engine/disp/ovlygk104.o
+nvkm-y += nvkm/engine/disp/ovlygp104.o
 
 nvkm-y += nvkm/engine/disp/piocnv50.o
 nvkm-y += nvkm/engine/disp/piocgf119.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/basegp104.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/basegp104.c
new file mode 100644
index 000000000000..51688e37c54e
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/basegp104.c
@@ -0,0 +1,38 @@
+/*
+ * Copyright 2016 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs <bskeggs@redhat.com>
+ */
+#include "dmacnv50.h"
+#include "rootnv50.h"
+
+#include <nvif/class.h>
+
+const struct nv50_disp_dmac_oclass
+gp104_disp_base_oclass = {
+	.base.oclass = GK110_DISP_BASE_CHANNEL_DMA,
+	.base.minver = 0,
+	.base.maxver = 0,
+	.ctor = nv50_disp_base_new,
+	.func = &gp104_disp_dmac_func,
+	.mthd = &gf119_disp_base_chan_mthd,
+	.chid = 1,
+};
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/channv50.h b/drivers/gpu/drm/nouveau/nvkm/engine/disp/channv50.h
index aee374884c96..f5f683d9fd20 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/channv50.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/channv50.h
@@ -85,6 +85,7 @@ extern const struct nv50_disp_mthd_list gf119_disp_core_mthd_pior;
 extern const struct nv50_disp_chan_mthd gf119_disp_base_chan_mthd;
 
 extern const struct nv50_disp_chan_mthd gk104_disp_core_chan_mthd;
+extern const struct nv50_disp_chan_mthd gk104_disp_ovly_chan_mthd;
 
 struct nv50_disp_pioc_oclass {
 	int (*ctor)(const struct nv50_disp_chan_func *,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/coregf119.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/coregf119.c
index 6b1dc703dac7..21fbf89b6319 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/coregf119.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/coregf119.c
@@ -171,7 +171,7 @@ gf119_disp_core_chan_mthd = {
 	}
 };
 
-static void
+void
 gf119_disp_core_fini(struct nv50_disp_dmac *chan)
 {
 	struct nv50_disp *disp = chan->base.root->disp;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/coregp100.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/coregp100.c
new file mode 100644
index 000000000000..d5dff6619d4d
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/coregp100.c
@@ -0,0 +1,38 @@
+/*
+ * Copyright 2015 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs <bskeggs@redhat.com>
+ */
+#include "dmacnv50.h"
+#include "rootnv50.h"
+
+#include <nvif/class.h>
+
+const struct nv50_disp_dmac_oclass
+gp100_disp_core_oclass = {
+	.base.oclass = GP100_DISP_CORE_CHANNEL_DMA,
+	.base.minver = 0,
+	.base.maxver = 0,
+	.ctor = nv50_disp_core_new,
+	.func = &gf119_disp_core_func,
+	.mthd = &gk104_disp_core_chan_mthd,
+	.chid = 0,
+};
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/coregp104.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/coregp104.c
new file mode 100644
index 000000000000..6922f4007b61
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/coregp104.c
@@ -0,0 +1,78 @@
+/*
+ * Copyright 2016 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs <bskeggs@redhat.com>
+ */
+#include "dmacnv50.h"
+#include "rootnv50.h"
+
+#include <subdev/timer.h>
+
+#include <nvif/class.h>
+
+static int
+gp104_disp_core_init(struct nv50_disp_dmac *chan)
+{
+	struct nv50_disp *disp = chan->base.root->disp;
+	struct nvkm_subdev *subdev = &disp->base.engine.subdev;
+	struct nvkm_device *device = subdev->device;
+
+	/* enable error reporting */
+	nvkm_mask(device, 0x6100a0, 0x00000001, 0x00000001);
+
+	/* initialise channel for dma command submission */
+	nvkm_wr32(device, 0x611494, chan->push);
+	nvkm_wr32(device, 0x611498, 0x00010000);
+	nvkm_wr32(device, 0x61149c, 0x00000001);
+	nvkm_mask(device, 0x610490, 0x00000010, 0x00000010);
+	nvkm_wr32(device, 0x640000, 0x00000000);
+	nvkm_wr32(device, 0x610490, 0x01000013);
+
+	/* wait for it to go inactive */
+	if (nvkm_msec(device, 2000,
+		if (!(nvkm_rd32(device, 0x610490) & 0x80000000))
+			break;
+	) < 0) {
+		nvkm_error(subdev, "core init: %08x\n",
+			   nvkm_rd32(device, 0x610490));
+		return -EBUSY;
+	}
+
+	return 0;
+}
+
+const struct nv50_disp_dmac_func
+gp104_disp_core_func = {
+	.init = gp104_disp_core_init,
+	.fini = gf119_disp_core_fini,
+	.bind = gf119_disp_dmac_bind,
+};
+
+const struct nv50_disp_dmac_oclass
+gp104_disp_core_oclass = {
+	.base.oclass = GP104_DISP_CORE_CHANNEL_DMA,
+	.base.minver = 0,
+	.base.maxver = 0,
+	.ctor = nv50_disp_core_new,
+	.func = &gp104_disp_core_func,
+	.mthd = &gk104_disp_core_chan_mthd,
+	.chid = 0,
+};
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/dmacgf119.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/dmacgf119.c
index 876b14549a58..a57f7cef307a 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/dmacgf119.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/dmacgf119.c
@@ -36,7 +36,7 @@ gf119_disp_dmac_bind(struct nv50_disp_dmac *chan,
 				 chan->base.chid << 27 | 0x00000001);
 }
 
-static void
+void
 gf119_disp_dmac_fini(struct nv50_disp_dmac *chan)
 {
 	struct nv50_disp *disp = chan->base.root->disp;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/dmacgp104.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/dmacgp104.c
new file mode 100644
index 000000000000..ad24c2c57696
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/dmacgp104.c
@@ -0,0 +1,66 @@
+/*
+ * Copyright 2016 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs <bskeggs@redhat.com>
+ */
+#include "dmacnv50.h"
+#include "rootnv50.h"
+
+#include <subdev/timer.h>
+
+static int
+gp104_disp_dmac_init(struct nv50_disp_dmac *chan)
+{
+	struct nv50_disp *disp = chan->base.root->disp;
+	struct nvkm_subdev *subdev = &disp->base.engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	int chid = chan->base.chid;
+
+	/* enable error reporting */
+	nvkm_mask(device, 0x6100a0, 0x00000001 << chid, 0x00000001 << chid);
+
+	/* initialise channel for dma command submission */
+	nvkm_wr32(device, 0x611494 + (chid * 0x0010), chan->push);
+	nvkm_wr32(device, 0x611498 + (chid * 0x0010), 0x00010000);
+	nvkm_wr32(device, 0x61149c + (chid * 0x0010), 0x00000001);
+	nvkm_mask(device, 0x610490 + (chid * 0x0010), 0x00000010, 0x00000010);
+	nvkm_wr32(device, 0x640000 + (chid * 0x1000), 0x00000000);
+	nvkm_wr32(device, 0x610490 + (chid * 0x0010), 0x00000013);
+
+	/* wait for it to go inactive */
+	if (nvkm_msec(device, 2000,
+		if (!(nvkm_rd32(device, 0x610490 + (chid * 0x10)) & 0x80000000))
+			break;
+	) < 0) {
+		nvkm_error(subdev, "ch %d init: %08x\n", chid,
+			   nvkm_rd32(device, 0x610490 + (chid * 0x10)));
+		return -EBUSY;
+	}
+
+	return 0;
+}
+
+const struct nv50_disp_dmac_func
+gp104_disp_dmac_func = {
+	.init = gp104_disp_dmac_init,
+	.fini = gf119_disp_dmac_fini,
+	.bind = gf119_disp_dmac_bind,
+};
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/dmacnv50.h b/drivers/gpu/drm/nouveau/nvkm/engine/disp/dmacnv50.h
index fc84eb8b5c45..43ac05857853 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/dmacnv50.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/dmacnv50.h
@@ -25,8 +25,12 @@ int nv50_disp_dmac_bind(struct nv50_disp_dmac *, struct nvkm_object *, u32);
 extern const struct nv50_disp_dmac_func nv50_disp_core_func;
 
 extern const struct nv50_disp_dmac_func gf119_disp_dmac_func;
+void gf119_disp_dmac_fini(struct nv50_disp_dmac *);
 int gf119_disp_dmac_bind(struct nv50_disp_dmac *, struct nvkm_object *, u32);
 extern const struct nv50_disp_dmac_func gf119_disp_core_func;
+void gf119_disp_core_fini(struct nv50_disp_dmac *);
+
+extern const struct nv50_disp_dmac_func gp104_disp_dmac_func;
 
 struct nv50_disp_dmac_oclass {
 	int (*ctor)(const struct nv50_disp_dmac_func *,
@@ -88,4 +92,10 @@ extern const struct nv50_disp_dmac_oclass gk110_disp_base_oclass;
 extern const struct nv50_disp_dmac_oclass gm107_disp_core_oclass;
 
 extern const struct nv50_disp_dmac_oclass gm200_disp_core_oclass;
+
+extern const struct nv50_disp_dmac_oclass gp100_disp_core_oclass;
+
+extern const struct nv50_disp_dmac_oclass gp104_disp_core_oclass;
+extern const struct nv50_disp_dmac_oclass gp104_disp_base_oclass;
+extern const struct nv50_disp_dmac_oclass gp104_disp_ovly_oclass;
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/gf119.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/gf119.c
index f0314664349c..29e84b241cca 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/gf119.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/gf119.c
@@ -79,8 +79,7 @@ exec_lookup(struct nv50_disp *disp, int head, int or, u32 ctrl,
 	list_for_each_entry(outp, &disp->base.outp, head) {
 		if ((outp->info.hasht & 0xff) == type &&
 		    (outp->info.hashm & mask) == mask) {
-			*data = nvbios_outp_match(bios, outp->info.hasht,
-							outp->info.hashm,
+			*data = nvbios_outp_match(bios, outp->info.hasht, mask,
 						  ver, hdr, cnt, len, info);
 			if (!*data)
 				return NULL;
@@ -155,25 +154,21 @@ exec_clkcmp(struct nv50_disp *disp, int head, int id, u32 pclk, u32 *conf)
 	if (!outp)
 		return NULL;
 
+	*conf = (ctrl & 0x00000f00) >> 8;
 	switch (outp->info.type) {
 	case DCB_OUTPUT_TMDS:
-		*conf = (ctrl & 0x00000f00) >> 8;
 		if (*conf == 5)
 			*conf |= 0x0100;
 		break;
 	case DCB_OUTPUT_LVDS:
-		*conf = disp->sor.lvdsconf;
+		*conf |= disp->sor.lvdsconf;
 		break;
-	case DCB_OUTPUT_DP:
-		*conf = (ctrl & 0x00000f00) >> 8;
-		break;
-	case DCB_OUTPUT_ANALOG:
 	default:
-		*conf = 0x00ff;
 		break;
 	}
 
-	data = nvbios_ocfg_match(bios, data, *conf, &ver, &hdr, &cnt, &len, &info2);
+	data = nvbios_ocfg_match(bios, data, *conf & 0xff, *conf >> 8,
+				 &ver, &hdr, &cnt, &len, &info2);
 	if (data && id < 0xff) {
 		data = nvbios_oclk_match(bios, info2.clkcmp[id], pclk);
 		if (data) {
@@ -418,7 +413,7 @@ gf119_disp_intr_supervisor(struct work_struct *work)
 	nvkm_wr32(device, 0x6101d0, 0x80000000);
 }
 
-static void
+void
 gf119_disp_intr_error(struct nv50_disp *disp, int chid)
 {
 	struct nvkm_subdev *subdev = &disp->base.engine.subdev;
@@ -466,7 +461,7 @@ gf119_disp_intr(struct nv50_disp *disp)
 		u32 stat = nvkm_rd32(device, 0x61009c);
 		int chid = ffs(stat) - 1;
 		if (chid >= 0)
-			gf119_disp_intr_error(disp, chid);
+			disp->func->intr_error(disp, chid);
 		intr &= ~0x00000002;
 	}
 
@@ -510,6 +505,7 @@ gf119_disp_new_(const struct nv50_disp_func *func, struct nvkm_device *device,
 static const struct nv50_disp_func
 gf119_disp = {
 	.intr = gf119_disp_intr,
+	.intr_error = gf119_disp_intr_error,
 	.uevent = &gf119_disp_chan_uevent,
 	.super = gf119_disp_intr_supervisor,
 	.root = &gf119_disp_root_oclass,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/gk104.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/gk104.c
index a86384b8e388..37f145cf30d7 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/gk104.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/gk104.c
@@ -27,6 +27,7 @@
 static const struct nv50_disp_func
 gk104_disp = {
 	.intr = gf119_disp_intr,
+	.intr_error = gf119_disp_intr_error,
 	.uevent = &gf119_disp_chan_uevent,
 	.super = gf119_disp_intr_supervisor,
 	.root = &gk104_disp_root_oclass,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/gk110.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/gk110.c
index 0d574c7e594a..e14ac946608c 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/gk110.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/gk110.c
@@ -27,6 +27,7 @@
 static const struct nv50_disp_func
 gk110_disp = {
 	.intr = gf119_disp_intr,
+	.intr_error = gf119_disp_intr_error,
 	.uevent = &gf119_disp_chan_uevent,
 	.super = gf119_disp_intr_supervisor,
 	.root = &gk110_disp_root_oclass,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/gm107.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/gm107.c
index b6944142d616..2f2437cc5891 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/gm107.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/gm107.c
@@ -27,6 +27,7 @@
 static const struct nv50_disp_func
 gm107_disp = {
 	.intr = gf119_disp_intr,
+	.intr_error = gf119_disp_intr_error,
 	.uevent = &gf119_disp_chan_uevent,
 	.super = gf119_disp_intr_supervisor,
 	.root = &gm107_disp_root_oclass,
@@ -36,7 +37,7 @@ gm107_disp = {
 	.outp.internal.crt = nv50_dac_output_new,
 	.outp.internal.tmds = nv50_sor_output_new,
 	.outp.internal.lvds = nv50_sor_output_new,
-	.outp.internal.dp = gf119_sor_dp_new,
+	.outp.internal.dp = gm107_sor_dp_new,
 	.dac.nr = 3,
 	.dac.power = nv50_dac_power,
 	.dac.sense = nv50_dac_sense,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/gm200.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/gm200.c
index 67eec8620719..9f368d4ee61e 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/gm200.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/gm200.c
@@ -27,6 +27,7 @@
 static const struct nv50_disp_func
 gm200_disp = {
 	.intr = gf119_disp_intr,
+	.intr_error = gf119_disp_intr_error,
 	.uevent = &gf119_disp_chan_uevent,
 	.super = gf119_disp_intr_supervisor,
 	.root = &gm200_disp_root_oclass,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/gp100.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/gp100.c
new file mode 100644
index 000000000000..4f81bf31435e
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/gp100.c
@@ -0,0 +1,55 @@
+/*
+ * Copyright 2015 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs <bskeggs@redhat.com>
+ */
+#include "nv50.h"
+#include "rootnv50.h"
+
+static const struct nv50_disp_func
+gp100_disp = {
+	.intr = gf119_disp_intr,
+	.intr_error = gf119_disp_intr_error,
+	.uevent = &gf119_disp_chan_uevent,
+	.super = gf119_disp_intr_supervisor,
+	.root = &gp100_disp_root_oclass,
+	.head.vblank_init = gf119_disp_vblank_init,
+	.head.vblank_fini = gf119_disp_vblank_fini,
+	.head.scanoutpos = gf119_disp_root_scanoutpos,
+	.outp.internal.crt = nv50_dac_output_new,
+	.outp.internal.tmds = nv50_sor_output_new,
+	.outp.internal.lvds = nv50_sor_output_new,
+	.outp.internal.dp = gm200_sor_dp_new,
+	.dac.nr = 3,
+	.dac.power = nv50_dac_power,
+	.dac.sense = nv50_dac_sense,
+	.sor.nr = 4,
+	.sor.power = nv50_sor_power,
+	.sor.hda_eld = gf119_hda_eld,
+	.sor.hdmi = gk104_hdmi_ctrl,
+	.sor.magic = gm200_sor_magic,
+};
+
+int
+gp100_disp_new(struct nvkm_device *device, int index, struct nvkm_disp **pdisp)
+{
+	return gf119_disp_new_(&gp100_disp, device, index, pdisp);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/gp104.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/gp104.c
new file mode 100644
index 000000000000..3bf3380336e4
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/gp104.c
@@ -0,0 +1,81 @@
+/*
+ * Copyright 2016 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs <bskeggs@redhat.com>
+ */
+#include "nv50.h"
+#include "rootnv50.h"
+
+static void
+gp104_disp_intr_error(struct nv50_disp *disp, int chid)
+{
+	struct nvkm_subdev *subdev = &disp->base.engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	u32 mthd = nvkm_rd32(device, 0x6111f0 + (chid * 12));
+	u32 data = nvkm_rd32(device, 0x6111f4 + (chid * 12));
+	u32 unkn = nvkm_rd32(device, 0x6111f8 + (chid * 12));
+
+	nvkm_error(subdev, "chid %d mthd %04x data %08x %08x %08x\n",
+		   chid, (mthd & 0x0000ffc), data, mthd, unkn);
+
+	if (chid < ARRAY_SIZE(disp->chan)) {
+		switch (mthd & 0xffc) {
+		case 0x0080:
+			nv50_disp_chan_mthd(disp->chan[chid], NV_DBG_ERROR);
+			break;
+		default:
+			break;
+		}
+	}
+
+	nvkm_wr32(device, 0x61009c, (1 << chid));
+	nvkm_wr32(device, 0x6111f0 + (chid * 12), 0x90000000);
+}
+
+static const struct nv50_disp_func
+gp104_disp = {
+	.intr = gf119_disp_intr,
+	.intr_error = gp104_disp_intr_error,
+	.uevent = &gf119_disp_chan_uevent,
+	.super = gf119_disp_intr_supervisor,
+	.root = &gp104_disp_root_oclass,
+	.head.vblank_init = gf119_disp_vblank_init,
+	.head.vblank_fini = gf119_disp_vblank_fini,
+	.head.scanoutpos = gf119_disp_root_scanoutpos,
+	.outp.internal.crt = nv50_dac_output_new,
+	.outp.internal.tmds = nv50_sor_output_new,
+	.outp.internal.lvds = nv50_sor_output_new,
+	.outp.internal.dp = gm200_sor_dp_new,
+	.dac.nr = 3,
+	.dac.power = nv50_dac_power,
+	.dac.sense = nv50_dac_sense,
+	.sor.nr = 4,
+	.sor.power = nv50_sor_power,
+	.sor.hda_eld = gf119_hda_eld,
+	.sor.hdmi = gk104_hdmi_ctrl,
+	.sor.magic = gm200_sor_magic,
+};
+
+int
+gp104_disp_new(struct nvkm_device *device, int index, struct nvkm_disp **pdisp)
+{
+	return gf119_disp_new_(&gp104_disp, device, index, pdisp);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/nv50.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/nv50.c
index 4226d2153b9c..fbb8c7dc18fd 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/nv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/nv50.c
@@ -32,6 +32,7 @@
 #include <subdev/bios/init.h>
 #include <subdev/bios/pll.h>
 #include <subdev/devinit.h>
+#include <subdev/timer.h>
 
 static const struct nvkm_disp_oclass *
 nv50_disp_root_(struct nvkm_disp *base)
@@ -269,8 +270,7 @@ exec_lookup(struct nv50_disp *disp, int head, int or, u32 ctrl,
 	list_for_each_entry(outp, &disp->base.outp, head) {
 		if ((outp->info.hasht & 0xff) == type &&
 		    (outp->info.hashm & mask) == mask) {
-			*data = nvbios_outp_match(bios, outp->info.hasht,
-							outp->info.hashm,
+			*data = nvbios_outp_match(bios, outp->info.hasht, mask,
 						  ver, hdr, cnt, len, info);
 			if (!*data)
 				return NULL;
@@ -387,22 +387,17 @@ exec_clkcmp(struct nv50_disp *disp, int head, int id, u32 pclk, u32 *conf)
 	if (!outp)
 		return NULL;
 
+	*conf = (ctrl & 0x00000f00) >> 8;
 	if (outp->info.location == 0) {
 		switch (outp->info.type) {
 		case DCB_OUTPUT_TMDS:
-			*conf = (ctrl & 0x00000f00) >> 8;
 			if (*conf == 5)
 				*conf |= 0x0100;
 			break;
 		case DCB_OUTPUT_LVDS:
-			*conf = disp->sor.lvdsconf;
+			*conf |= disp->sor.lvdsconf;
 			break;
-		case DCB_OUTPUT_DP:
-			*conf = (ctrl & 0x00000f00) >> 8;
-			break;
-		case DCB_OUTPUT_ANALOG:
 		default:
-			*conf = 0x00ff;
 			break;
 		}
 	} else {
@@ -410,7 +405,8 @@ exec_clkcmp(struct nv50_disp *disp, int head, int id, u32 pclk, u32 *conf)
 		pclk = pclk / 2;
 	}
 
-	data = nvbios_ocfg_match(bios, data, *conf, &ver, &hdr, &cnt, &len, &info2);
+	data = nvbios_ocfg_match(bios, data, *conf & 0xff, *conf >> 8,
+				 &ver, &hdr, &cnt, &len, &info2);
 	if (data && id < 0xff) {
 		data = nvbios_oclk_match(bios, info2.clkcmp[id], pclk);
 		if (data) {
@@ -430,6 +426,134 @@ exec_clkcmp(struct nv50_disp *disp, int head, int id, u32 pclk, u32 *conf)
 	return outp;
 }
 
+static bool
+nv50_disp_dptmds_war(struct nvkm_device *device)
+{
+	switch (device->chipset) {
+	case 0x94:
+	case 0x96:
+	case 0x98:
+	case 0xaa:
+	case 0xac:
+		return true;
+	default:
+		break;
+	}
+	return false;
+}
+
+static bool
+nv50_disp_dptmds_war_needed(struct nv50_disp *disp, struct dcb_output *outp)
+{
+	struct nvkm_device *device = disp->base.engine.subdev.device;
+	const u32 soff = __ffs(outp->or) * 0x800;
+	if (nv50_disp_dptmds_war(device) && outp->type == DCB_OUTPUT_TMDS) {
+		switch (nvkm_rd32(device, 0x614300 + soff) & 0x00030000) {
+		case 0x00000000:
+		case 0x00030000:
+			return true;
+		default:
+			break;
+		}
+	}
+	return false;
+
+}
+
+static void
+nv50_disp_dptmds_war_2(struct nv50_disp *disp, struct dcb_output *outp)
+{
+	struct nvkm_device *device = disp->base.engine.subdev.device;
+	const u32 soff = __ffs(outp->or) * 0x800;
+
+	if (!nv50_disp_dptmds_war_needed(disp, outp))
+		return;
+
+	nvkm_mask(device, 0x00e840, 0x80000000, 0x80000000);
+	nvkm_mask(device, 0x614300 + soff, 0x03000000, 0x03000000);
+	nvkm_mask(device, 0x61c10c + soff, 0x00000001, 0x00000001);
+
+	nvkm_mask(device, 0x61c00c + soff, 0x0f000000, 0x00000000);
+	nvkm_mask(device, 0x61c008 + soff, 0xff000000, 0x14000000);
+	nvkm_usec(device, 400, NVKM_DELAY);
+	nvkm_mask(device, 0x61c008 + soff, 0xff000000, 0x00000000);
+	nvkm_mask(device, 0x61c00c + soff, 0x0f000000, 0x01000000);
+
+	if (nvkm_rd32(device, 0x61c004 + soff) & 0x00000001) {
+		u32 seqctl = nvkm_rd32(device, 0x61c030 + soff);
+		u32  pu_pc = seqctl & 0x0000000f;
+		nvkm_wr32(device, 0x61c040 + soff + pu_pc * 4, 0x1f008000);
+	}
+}
+
+static void
+nv50_disp_dptmds_war_3(struct nv50_disp *disp, struct dcb_output *outp)
+{
+	struct nvkm_device *device = disp->base.engine.subdev.device;
+	const u32 soff = __ffs(outp->or) * 0x800;
+	u32 sorpwr;
+
+	if (!nv50_disp_dptmds_war_needed(disp, outp))
+		return;
+
+	sorpwr = nvkm_rd32(device, 0x61c004 + soff);
+	if (sorpwr & 0x00000001) {
+		u32 seqctl = nvkm_rd32(device, 0x61c030 + soff);
+		u32  pd_pc = (seqctl & 0x00000f00) >> 8;
+		u32  pu_pc =  seqctl & 0x0000000f;
+
+		nvkm_wr32(device, 0x61c040 + soff + pd_pc * 4, 0x1f008000);
+
+		nvkm_msec(device, 2000,
+			if (!(nvkm_rd32(device, 0x61c030 + soff) & 0x10000000))
+				break;
+		);
+		nvkm_mask(device, 0x61c004 + soff, 0x80000001, 0x80000000);
+		nvkm_msec(device, 2000,
+			if (!(nvkm_rd32(device, 0x61c030 + soff) & 0x10000000))
+				break;
+		);
+
+		nvkm_wr32(device, 0x61c040 + soff + pd_pc * 4, 0x00002000);
+		nvkm_wr32(device, 0x61c040 + soff + pu_pc * 4, 0x1f000000);
+	}
+
+	nvkm_mask(device, 0x61c10c + soff, 0x00000001, 0x00000000);
+	nvkm_mask(device, 0x614300 + soff, 0x03000000, 0x00000000);
+
+	if (sorpwr & 0x00000001) {
+		nvkm_mask(device, 0x61c004 + soff, 0x80000001, 0x80000001);
+	}
+}
+
+static void
+nv50_disp_update_sppll1(struct nv50_disp *disp)
+{
+	struct nvkm_device *device = disp->base.engine.subdev.device;
+	bool used = false;
+	int sor;
+
+	if (!nv50_disp_dptmds_war(device))
+		return;
+
+	for (sor = 0; sor < disp->func->sor.nr; sor++) {
+		u32 clksor = nvkm_rd32(device, 0x614300 + (sor * 0x800));
+		switch (clksor & 0x03000000) {
+		case 0x02000000:
+		case 0x03000000:
+			used = true;
+			break;
+		default:
+			break;
+		}
+	}
+
+	if (used)
+		return;
+
+	nvkm_mask(device, 0x00e840, 0x80000000, 0x00000000);
+}
+
 static void
 nv50_disp_intr_unk10_0(struct nv50_disp *disp, int head)
 {
@@ -683,6 +807,8 @@ nv50_disp_intr_unk20_2(struct nv50_disp *disp, int head)
 
 	nvkm_mask(device, hreg, 0x0000000f, hval);
 	nvkm_mask(device, oreg, mask, oval);
+
+	nv50_disp_dptmds_war_2(disp, &outp->info);
 }
 
 /* If programming a TMDS output on a SOR that can also be configured for
@@ -724,6 +850,7 @@ nv50_disp_intr_unk40_0(struct nv50_disp *disp, int head)
 
 	if (outp->info.location == 0 && outp->info.type == DCB_OUTPUT_TMDS)
 		nv50_disp_intr_unk40_0_tmds(disp, &outp->info);
+	nv50_disp_dptmds_war_3(disp, &outp->info);
 }
 
 void
@@ -771,6 +898,7 @@ nv50_disp_intr_supervisor(struct work_struct *work)
 				continue;
 			nv50_disp_intr_unk40_0(disp, head);
 		}
+		nv50_disp_update_sppll1(disp);
 	}
 
 	nvkm_wr32(device, 0x610030, 0x80000000);
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/nv50.h b/drivers/gpu/drm/nouveau/nvkm/engine/disp/nv50.h
index aecebd8717e5..1e1de6bfe85a 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/nv50.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/nv50.h
@@ -68,6 +68,7 @@ struct nv50_disp_func_outp {
 
 struct nv50_disp_func {
 	void (*intr)(struct nv50_disp *);
+	void (*intr_error)(struct nv50_disp *, int chid);
 
 	const struct nvkm_event_func *uevent;
 	void (*super)(struct work_struct *);
@@ -114,4 +115,5 @@ void gf119_disp_vblank_init(struct nv50_disp *, int);
 void gf119_disp_vblank_fini(struct nv50_disp *, int);
 void gf119_disp_intr(struct nv50_disp *);
 void gf119_disp_intr_supervisor(struct work_struct *);
+void gf119_disp_intr_error(struct nv50_disp *, int);
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/outpdp.h b/drivers/gpu/drm/nouveau/nvkm/engine/disp/outpdp.h
index e9067ba4e179..4e983f6d7032 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/outpdp.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/outpdp.h
@@ -62,7 +62,12 @@ int g94_sor_dp_lnk_pwr(struct nvkm_output_dp *, int);
 int gf119_sor_dp_new(struct nvkm_disp *, int, struct dcb_output *,
 		     struct nvkm_output **);
 int gf119_sor_dp_lnk_ctl(struct nvkm_output_dp *, int, int, bool);
+int gf119_sor_dp_drv_ctl(struct nvkm_output_dp *, int, int, int, int);
 
-int  gm200_sor_dp_new(struct nvkm_disp *, int, struct dcb_output *,
-		      struct nvkm_output **);
+int gm107_sor_dp_new(struct nvkm_disp *, int, struct dcb_output *,
+		     struct nvkm_output **);
+int gm107_sor_dp_pattern(struct nvkm_output_dp *, int);
+
+int gm200_sor_dp_new(struct nvkm_disp *, int, struct dcb_output *,
+		     struct nvkm_output **);
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/ovlygk104.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/ovlygk104.c
index 2e2dc0641ef2..2f0220b39f34 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/ovlygk104.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/ovlygk104.c
@@ -80,7 +80,7 @@ gk104_disp_ovly_mthd_base = {
 	}
 };
 
-static const struct nv50_disp_chan_mthd
+const struct nv50_disp_chan_mthd
 gk104_disp_ovly_chan_mthd = {
 	.name = "Overlay",
 	.addr = 0x001000,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/ovlygp104.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/ovlygp104.c
new file mode 100644
index 000000000000..97e2dd2d908e
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/ovlygp104.c
@@ -0,0 +1,38 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "dmacnv50.h"
+#include "rootnv50.h"
+
+#include <nvif/class.h>
+
+const struct nv50_disp_dmac_oclass
+gp104_disp_ovly_oclass = {
+	.base.oclass = GK104_DISP_OVERLAY_CONTROL_DMA,
+	.base.minver = 0,
+	.base.maxver = 0,
+	.ctor = nv50_disp_ovly_new,
+	.func = &gp104_disp_dmac_func,
+	.mthd = &gk104_disp_ovly_chan_mthd,
+	.chid = 5,
+};
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/rootgp100.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/rootgp100.c
new file mode 100644
index 000000000000..ac8fdd728ec6
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/rootgp100.c
@@ -0,0 +1,58 @@
+/*
+ * Copyright 2015 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs <bskeggs@redhat.com>
+ */
+#include "rootnv50.h"
+#include "dmacnv50.h"
+
+#include <nvif/class.h>
+
+static const struct nv50_disp_root_func
+gp100_disp_root = {
+	.init = gf119_disp_root_init,
+	.fini = gf119_disp_root_fini,
+	.dmac = {
+		&gp100_disp_core_oclass,
+		&gk110_disp_base_oclass,
+		&gk104_disp_ovly_oclass,
+	},
+	.pioc = {
+		&gk104_disp_oimm_oclass,
+		&gk104_disp_curs_oclass,
+	},
+};
+
+static int
+gp100_disp_root_new(struct nvkm_disp *disp, const struct nvkm_oclass *oclass,
+		    void *data, u32 size, struct nvkm_object **pobject)
+{
+	return nv50_disp_root_new_(&gp100_disp_root, disp, oclass,
+				   data, size, pobject);
+}
+
+const struct nvkm_disp_oclass
+gp100_disp_root_oclass = {
+	.base.oclass = GP100_DISP,
+	.base.minver = -1,
+	.base.maxver = -1,
+	.ctor = gp100_disp_root_new,
+};
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/rootgp104.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/rootgp104.c
new file mode 100644
index 000000000000..8443e04dc626
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/rootgp104.c
@@ -0,0 +1,58 @@
+/*
+ * Copyright 2016 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs <bskeggs@redhat.com>
+ */
+#include "rootnv50.h"
+#include "dmacnv50.h"
+
+#include <nvif/class.h>
+
+static const struct nv50_disp_root_func
+gp104_disp_root = {
+	.init = gf119_disp_root_init,
+	.fini = gf119_disp_root_fini,
+	.dmac = {
+		&gp104_disp_core_oclass,
+		&gp104_disp_base_oclass,
+		&gp104_disp_ovly_oclass,
+	},
+	.pioc = {
+		&gk104_disp_oimm_oclass,
+		&gk104_disp_curs_oclass,
+	},
+};
+
+static int
+gp104_disp_root_new(struct nvkm_disp *disp, const struct nvkm_oclass *oclass,
+		    void *data, u32 size, struct nvkm_object **pobject)
+{
+	return nv50_disp_root_new_(&gp104_disp_root, disp, oclass,
+				   data, size, pobject);
+}
+
+const struct nvkm_disp_oclass
+gp104_disp_root_oclass = {
+	.base.oclass = GP104_DISP,
+	.base.minver = -1,
+	.base.maxver = -1,
+	.ctor = gp104_disp_root_new,
+};
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/rootnv50.h b/drivers/gpu/drm/nouveau/nvkm/engine/disp/rootnv50.h
index cb449ed8d92c..ad00f1724b72 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/rootnv50.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/rootnv50.h
@@ -40,4 +40,6 @@ extern const struct nvkm_disp_oclass gk104_disp_root_oclass;
 extern const struct nvkm_disp_oclass gk110_disp_root_oclass;
 extern const struct nvkm_disp_oclass gm107_disp_root_oclass;
 extern const struct nvkm_disp_oclass gm200_disp_root_oclass;
+extern const struct nvkm_disp_oclass gp100_disp_root_oclass;
+extern const struct nvkm_disp_oclass gp104_disp_root_oclass;
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/sorgf119.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/sorgf119.c
index b4b41b135643..22706c0a54b5 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/sorgf119.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/sorgf119.c
@@ -40,8 +40,7 @@ static int
 gf119_sor_dp_pattern(struct nvkm_output_dp *outp, int pattern)
 {
 	struct nvkm_device *device = outp->base.disp->engine.subdev.device;
-	const u32 loff = gf119_sor_loff(outp);
-	nvkm_mask(device, 0x61c110 + loff, 0x0f0f0f0f, 0x01010101 * pattern);
+	nvkm_mask(device, 0x61c110, 0x0f0f0f0f, 0x01010101 * pattern);
 	return 0;
 }
 
@@ -64,7 +63,7 @@ gf119_sor_dp_lnk_ctl(struct nvkm_output_dp *outp, int nr, int bw, bool ef)
 	return 0;
 }
 
-static int
+int
 gf119_sor_dp_drv_ctl(struct nvkm_output_dp *outp,
 		     int ln, int vs, int pe, int pc)
 {
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/sorgm107.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/sorgm107.c
new file mode 100644
index 000000000000..37790b2617c5
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/sorgm107.c
@@ -0,0 +1,53 @@
+/*
+ * Copyright 2016 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs <bskeggs@redhat.com>
+ */
+#include "nv50.h"
+#include "outpdp.h"
+
+int
+gm107_sor_dp_pattern(struct nvkm_output_dp *outp, int pattern)
+{
+	struct nvkm_device *device = outp->base.disp->engine.subdev.device;
+	const u32 soff = outp->base.or * 0x800;
+	const u32 data = 0x01010101 * pattern;
+	if (outp->base.info.sorconf.link & 1)
+		nvkm_mask(device, 0x61c110 + soff, 0x0f0f0f0f, data);
+	else
+		nvkm_mask(device, 0x61c12c + soff, 0x0f0f0f0f, data);
+	return 0;
+}
+
+static const struct nvkm_output_dp_func
+gm107_sor_dp_func = {
+	.pattern = gm107_sor_dp_pattern,
+	.lnk_pwr = g94_sor_dp_lnk_pwr,
+	.lnk_ctl = gf119_sor_dp_lnk_ctl,
+	.drv_ctl = gf119_sor_dp_drv_ctl,
+};
+
+int
+gm107_sor_dp_new(struct nvkm_disp *disp, int index,
+		 struct dcb_output *dcbE, struct nvkm_output **poutp)
+{
+	return nvkm_output_dp_new_(&gm107_sor_dp_func, disp, index, dcbE, poutp);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/disp/sorgm200.c b/drivers/gpu/drm/nouveau/nvkm/engine/disp/sorgm200.c
index 2cfbef9c344f..c44fa7ea672a 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/disp/sorgm200.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/disp/sorgm200.c
@@ -57,19 +57,6 @@ gm200_sor_dp_lane_map(struct nvkm_device *device, u8 lane)
 }
 
 static int
-gm200_sor_dp_pattern(struct nvkm_output_dp *outp, int pattern)
-{
-	struct nvkm_device *device = outp->base.disp->engine.subdev.device;
-	const u32 soff = gm200_sor_soff(outp);
-	const u32 data = 0x01010101 * pattern;
-	if (outp->base.info.sorconf.link & 1)
-		nvkm_mask(device, 0x61c110 + soff, 0x0f0f0f0f, data);
-	else
-		nvkm_mask(device, 0x61c12c + soff, 0x0f0f0f0f, data);
-	return 0;
-}
-
-static int
 gm200_sor_dp_lnk_pwr(struct nvkm_output_dp *outp, int nr)
 {
 	struct nvkm_device *device = outp->base.disp->engine.subdev.device;
@@ -129,7 +116,7 @@ gm200_sor_dp_drv_ctl(struct nvkm_output_dp *outp,
 
 static const struct nvkm_output_dp_func
 gm200_sor_dp_func = {
-	.pattern = gm200_sor_dp_pattern,
+	.pattern = gm107_sor_dp_pattern,
 	.lnk_pwr = gm200_sor_dp_lnk_pwr,
 	.lnk_ctl = gf119_sor_dp_lnk_ctl,
 	.drv_ctl = gm200_sor_dp_drv_ctl,
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/Kbuild b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/Kbuild
index 65e5d291ecda..98651a43bc12 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/Kbuild
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/Kbuild
@@ -13,6 +13,7 @@ nvkm-y += nvkm/engine/fifo/gk20a.o
 nvkm-y += nvkm/engine/fifo/gm107.o
 nvkm-y += nvkm/engine/fifo/gm200.o
 nvkm-y += nvkm/engine/fifo/gm20b.o
+nvkm-y += nvkm/engine/fifo/gp100.o
 
 nvkm-y += nvkm/engine/fifo/chan.o
 nvkm-y += nvkm/engine/fifo/channv50.o
@@ -31,3 +32,4 @@ nvkm-y += nvkm/engine/fifo/gpfifogf100.o
 nvkm-y += nvkm/engine/fifo/gpfifogk104.o
 nvkm-y += nvkm/engine/fifo/gpfifogk110.o
 nvkm-y += nvkm/engine/fifo/gpfifogm200.o
+nvkm-y += nvkm/engine/fifo/gpfifogp100.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/changk104.h b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/changk104.h
index e06f4d46f802..230f64e5f731 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/changk104.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/changk104.h
@@ -27,4 +27,5 @@ int gk104_fifo_gpfifo_new(struct nvkm_fifo *, const struct nvkm_oclass *,
 extern const struct nvkm_fifo_chan_oclass gk104_fifo_gpfifo_oclass;
 extern const struct nvkm_fifo_chan_oclass gk110_fifo_gpfifo_oclass;
 extern const struct nvkm_fifo_chan_oclass gm200_fifo_gpfifo_oclass;
+extern const struct nvkm_fifo_chan_oclass gp100_fifo_gpfifo_oclass;
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
index 743f3a189f28..103c0afaaa6d 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gk104.c
@@ -329,7 +329,7 @@ gk104_fifo_intr_fault(struct gk104_fifo *fifo, int unit)
 	}
 
 	if (eu == NULL) {
-		enum nvkm_devidx engidx = nvkm_top_fault(device->top, unit);
+		enum nvkm_devidx engidx = nvkm_top_fault(device, unit);
 		if (engidx < NVKM_SUBDEV_NR) {
 			const char *src = nvkm_subdev_name[engidx];
 			char *dst = en;
@@ -589,7 +589,6 @@ gk104_fifo_oneinit(struct nvkm_fifo *base)
 	struct gk104_fifo *fifo = gk104_fifo(base);
 	struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
 	struct nvkm_device *device = subdev->device;
-	struct nvkm_top *top = device->top;
 	int engn, runl, pbid, ret, i, j;
 	enum nvkm_devidx engidx;
 	u32 *map;
@@ -608,7 +607,7 @@ gk104_fifo_oneinit(struct nvkm_fifo *base)
 
 	/* Determine runlist configuration from topology device info. */
 	i = 0;
-	while ((int)(engidx = nvkm_top_engine(top, i++, &runl, &engn)) >= 0) {
+	while ((int)(engidx = nvkm_top_engine(device, i++, &runl, &engn)) >= 0) {
 		/* Determine which PBDMA handles requests for this engine. */
 		for (j = 0, pbid = -1; j < fifo->pbdma_nr; j++) {
 			if (map[j] & (1 << runl)) {
@@ -617,8 +616,8 @@ gk104_fifo_oneinit(struct nvkm_fifo *base)
 			}
 		}
 
-		nvkm_debug(subdev, "engine %2d: runlist %2d pbdma %2d\n",
-			   engn, runl, pbid);
+		nvkm_debug(subdev, "engine %2d: runlist %2d pbdma %2d (%s)\n",
+			   engn, runl, pbid, nvkm_subdev_name[engidx]);
 
 		fifo->engine[engn].engine = nvkm_device_engine(device, engidx);
 		fifo->engine[engn].runl = runl;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gp100.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gp100.c
new file mode 100644
index 000000000000..eff83f7fb705
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gp100.c
@@ -0,0 +1,67 @@
+/*
+ * Copyright 2016 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "gk104.h"
+#include "changk104.h"
+
+static const struct nvkm_enum
+gp100_fifo_fault_engine[] = {
+	{ 0x01, "DISPLAY" },
+	{ 0x03, "IFB", NULL, NVKM_ENGINE_IFB },
+	{ 0x04, "BAR1", NULL, NVKM_SUBDEV_BAR },
+	{ 0x05, "BAR2", NULL, NVKM_SUBDEV_INSTMEM },
+	{ 0x06, "HOST0" },
+	{ 0x07, "HOST1" },
+	{ 0x08, "HOST2" },
+	{ 0x09, "HOST3" },
+	{ 0x0a, "HOST4" },
+	{ 0x0b, "HOST5" },
+	{ 0x0c, "HOST6" },
+	{ 0x0d, "HOST7" },
+	{ 0x0e, "HOST8" },
+	{ 0x0f, "HOST9" },
+	{ 0x10, "HOST10" },
+	{ 0x13, "PERF" },
+	{ 0x17, "PMU" },
+	{ 0x18, "PTP" },
+	{ 0x1f, "PHYSICAL" },
+	{}
+};
+
+static const struct gk104_fifo_func
+gp100_fifo = {
+	.fault.engine = gp100_fifo_fault_engine,
+	.fault.reason = gk104_fifo_fault_reason,
+	.fault.hubclient = gk104_fifo_fault_hubclient,
+	.fault.gpcclient = gk104_fifo_fault_gpcclient,
+	.chan = {
+		&gp100_fifo_gpfifo_oclass,
+		NULL
+	},
+};
+
+int
+gp100_fifo_new(struct nvkm_device *device, int index, struct nvkm_fifo **pfifo)
+{
+	return gk104_fifo_new_(&gp100_fifo, device, index, 4096, pfifo);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogp100.c b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogp100.c
new file mode 100644
index 000000000000..1530a9217aea
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/fifo/gpfifogp100.c
@@ -0,0 +1,34 @@
+/*
+ * Copyright 2016 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "changk104.h"
+
+#include <nvif/class.h>
+
+const struct nvkm_fifo_chan_oclass
+gp100_fifo_gpfifo_oclass = {
+	.base.oclass = PASCAL_CHANNEL_GPFIFO_A,
+	.base.minver = 0,
+	.base.maxver = 0,
+	.ctor = gk104_fifo_gpfifo_new,
+};
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/Kbuild b/drivers/gpu/drm/nouveau/nvkm/engine/gr/Kbuild
index 290ed0db8047..f1c494182248 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/Kbuild
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/Kbuild
@@ -31,6 +31,7 @@ nvkm-y += nvkm/engine/gr/gk20a.o
 nvkm-y += nvkm/engine/gr/gm107.o
 nvkm-y += nvkm/engine/gr/gm200.o
 nvkm-y += nvkm/engine/gr/gm20b.o
+nvkm-y += nvkm/engine/gr/gp100.o
 
 nvkm-y += nvkm/engine/gr/ctxnv40.o
 nvkm-y += nvkm/engine/gr/ctxnv50.o
@@ -48,3 +49,4 @@ nvkm-y += nvkm/engine/gr/ctxgk20a.o
 nvkm-y += nvkm/engine/gr/ctxgm107.o
 nvkm-y += nvkm/engine/gr/ctxgm200.o
 nvkm-y += nvkm/engine/gr/ctxgm20b.o
+nvkm-y += nvkm/engine/gr/ctxgp100.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgf100.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgf100.c
index b02d8f50ea6a..bc77eea351a5 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgf100.c
@@ -1240,7 +1240,7 @@ gf100_grctx_generate_main(struct gf100_gr *gr, struct gf100_grctx *info)
 	const struct gf100_grctx_func *grctx = gr->func->grctx;
 	u32 idle_timeout;
 
-	nvkm_mc_unk260(device->mc, 0);
+	nvkm_mc_unk260(device, 0);
 
 	gf100_gr_mmio(gr, grctx->hub);
 	gf100_gr_mmio(gr, grctx->gpc);
@@ -1264,7 +1264,7 @@ gf100_grctx_generate_main(struct gf100_gr *gr, struct gf100_grctx *info)
 	gf100_gr_icmd(gr, grctx->icmd);
 	nvkm_wr32(device, 0x404154, idle_timeout);
 	gf100_gr_mthd(gr, grctx->mthd);
-	nvkm_mc_unk260(device->mc, 1);
+	nvkm_mc_unk260(device, 1);
 }
 
 int
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgf100.h b/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgf100.h
index ac895edce164..52048b5a5274 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgf100.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgf100.h
@@ -101,6 +101,8 @@ void gm200_grctx_generate_405b60(struct gf100_gr *);
 
 extern const struct gf100_grctx_func gm20b_grctx;
 
+extern const struct gf100_grctx_func gp100_grctx;
+
 /* context init value lists */
 
 extern const struct gf100_gr_pack gf100_grctx_pack_icmd[];
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgf117.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgf117.c
index f521de11a299..c925ade5880e 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgf117.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgf117.c
@@ -226,7 +226,7 @@ gf117_grctx_generate_main(struct gf100_gr *gr, struct gf100_grctx *info)
 	u32 idle_timeout;
 	int i;
 
-	nvkm_mc_unk260(device->mc, 0);
+	nvkm_mc_unk260(device, 0);
 
 	gf100_gr_mmio(gr, grctx->hub);
 	gf100_gr_mmio(gr, grctx->gpc);
@@ -253,7 +253,7 @@ gf117_grctx_generate_main(struct gf100_gr *gr, struct gf100_grctx *info)
 	gf100_gr_icmd(gr, grctx->icmd);
 	nvkm_wr32(device, 0x404154, idle_timeout);
 	gf100_gr_mthd(gr, grctx->mthd);
-	nvkm_mc_unk260(device->mc, 1);
+	nvkm_mc_unk260(device, 1);
 }
 
 const struct gf100_grctx_func
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgk104.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgk104.c
index 9ba337778ef5..c46b3fdf7203 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgk104.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgk104.c
@@ -950,7 +950,7 @@ gk104_grctx_generate_main(struct gf100_gr *gr, struct gf100_grctx *info)
 	u32 idle_timeout;
 	int i;
 
-	nvkm_mc_unk260(device->mc, 0);
+	nvkm_mc_unk260(device, 0);
 
 	gf100_gr_mmio(gr, grctx->hub);
 	gf100_gr_mmio(gr, grctx->gpc);
@@ -979,7 +979,7 @@ gk104_grctx_generate_main(struct gf100_gr *gr, struct gf100_grctx *info)
 	gf100_gr_icmd(gr, grctx->icmd);
 	nvkm_wr32(device, 0x404154, idle_timeout);
 	gf100_gr_mthd(gr, grctx->mthd);
-	nvkm_mc_unk260(device->mc, 1);
+	nvkm_mc_unk260(device, 1);
 
 	nvkm_mask(device, 0x418800, 0x00200000, 0x00200000);
 	nvkm_mask(device, 0x41be10, 0x00800000, 0x00800000);
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgp100.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgp100.c
new file mode 100644
index 000000000000..3d1ae7ddf7dd
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/ctxgp100.c
@@ -0,0 +1,179 @@
+/*
+ * Copyright 2016 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs <bskeggs@redhat.com>
+ */
+#include "ctxgf100.h"
+
+#include <subdev/fb.h>
+
+/*******************************************************************************
+ * PGRAPH context implementation
+ ******************************************************************************/
+
+static void
+gp100_grctx_generate_pagepool(struct gf100_grctx *info)
+{
+	const struct gf100_grctx_func *grctx = info->gr->func->grctx;
+	const u32 access = NV_MEM_ACCESS_RW | NV_MEM_ACCESS_SYS;
+	const int s = 8;
+	const int b = mmio_vram(info, grctx->pagepool_size, (1 << s), access);
+	mmio_refn(info, 0x40800c, 0x00000000, s, b);
+	mmio_wr32(info, 0x408010, 0x80000000);
+	mmio_refn(info, 0x419004, 0x00000000, s, b);
+	mmio_wr32(info, 0x419008, 0x00000000);
+}
+
+static void
+gp100_grctx_generate_attrib(struct gf100_grctx *info)
+{
+	struct gf100_gr *gr = info->gr;
+	const struct gf100_grctx_func *grctx = gr->func->grctx;
+	const u32  alpha = grctx->alpha_nr;
+	const u32 attrib = grctx->attrib_nr;
+	const u32 pertpc = 0x20 * (grctx->attrib_nr_max + grctx->alpha_nr_max);
+	const u32   size = roundup(gr->tpc_total * pertpc, 0x80);
+	const u32 access = NV_MEM_ACCESS_RW;
+	const int s = 12;
+	const int b = mmio_vram(info, size, (1 << s), access);
+	const int max_batches = 0xffff;
+	u32 ao = 0;
+	u32 bo = ao + grctx->alpha_nr_max * gr->tpc_total;
+	int gpc, ppc, n = 0;
+
+	mmio_refn(info, 0x418810, 0x80000000, s, b);
+	mmio_refn(info, 0x419848, 0x10000000, s, b);
+	mmio_refn(info, 0x419c2c, 0x10000000, s, b);
+	mmio_refn(info, 0x419b00, 0x00000000, s, b);
+	mmio_wr32(info, 0x419b04, 0x80000000 | size >> 7);
+	mmio_wr32(info, 0x405830, attrib);
+	mmio_wr32(info, 0x40585c, alpha);
+	mmio_wr32(info, 0x4064c4, ((alpha / 4) << 16) | max_batches);
+
+	for (gpc = 0; gpc < gr->gpc_nr; gpc++) {
+		for (ppc = 0; ppc < gr->ppc_nr[gpc]; ppc++, n++) {
+			const u32 as =  alpha * gr->ppc_tpc_nr[gpc][ppc];
+			const u32 bs = attrib * gr->ppc_tpc_nr[gpc][ppc];
+			const u32 u = 0x418ea0 + (n * 0x04);
+			const u32 o = PPC_UNIT(gpc, ppc, 0);
+			if (!(gr->ppc_mask[gpc] & (1 << ppc)))
+				continue;
+			mmio_wr32(info, o + 0xc0, bs);
+			mmio_wr32(info, o + 0xf4, bo);
+			mmio_wr32(info, o + 0xf0, bs);
+			bo += grctx->attrib_nr_max * gr->ppc_tpc_nr[gpc][ppc];
+			mmio_wr32(info, o + 0xe4, as);
+			mmio_wr32(info, o + 0xf8, ao);
+			ao += grctx->alpha_nr_max * gr->ppc_tpc_nr[gpc][ppc];
+			mmio_wr32(info, u, bs);
+		}
+	}
+
+	mmio_wr32(info, 0x418eec, 0x00000000);
+	mmio_wr32(info, 0x41befc, 0x00000000);
+}
+
+static void
+gp100_grctx_generate_405b60(struct gf100_gr *gr)
+{
+	struct nvkm_device *device = gr->base.engine.subdev.device;
+	const u32 dist_nr = DIV_ROUND_UP(gr->tpc_total, 4);
+	u32 dist[TPC_MAX / 4] = {};
+	u32 gpcs[GPC_MAX * 2] = {};
+	u8  tpcnr[GPC_MAX];
+	int tpc, gpc, i;
+
+	memcpy(tpcnr, gr->tpc_nr, sizeof(gr->tpc_nr));
+
+	/* won't result in the same distribution as the binary driver where
+	 * some of the gpcs have more tpcs than others, but this shall do
+	 * for the moment.  the code for earlier gpus has this issue too.
+	 */
+	for (gpc = -1, i = 0; i < gr->tpc_total; i++) {
+		do {
+			gpc = (gpc + 1) % gr->gpc_nr;
+		} while(!tpcnr[gpc]);
+		tpc = gr->tpc_nr[gpc] - tpcnr[gpc]--;
+
+		dist[i / 4] |= ((gpc << 4) | tpc) << ((i % 4) * 8);
+		gpcs[gpc + (gr->gpc_nr * (tpc / 4))] |= i << (tpc * 8);
+	}
+
+	for (i = 0; i < dist_nr; i++)
+		nvkm_wr32(device, 0x405b60 + (i * 4), dist[i]);
+	for (i = 0; i < gr->gpc_nr * 2; i++)
+		nvkm_wr32(device, 0x405ba0 + (i * 4), gpcs[i]);
+}
+
+static void
+gp100_grctx_generate_main(struct gf100_gr *gr, struct gf100_grctx *info)
+{
+	struct nvkm_device *device = gr->base.engine.subdev.device;
+	const struct gf100_grctx_func *grctx = gr->func->grctx;
+	u32 idle_timeout, tmp;
+	int i;
+
+	gf100_gr_mmio(gr, gr->fuc_sw_ctx);
+
+	idle_timeout = nvkm_mask(device, 0x404154, 0xffffffff, 0x00000000);
+
+	grctx->pagepool(info);
+	grctx->bundle(info);
+	grctx->attrib(info);
+	grctx->unkn(gr);
+
+	gm200_grctx_generate_tpcid(gr);
+	gf100_grctx_generate_r406028(gr);
+	gk104_grctx_generate_r418bb8(gr);
+
+	for (i = 0; i < 8; i++)
+		nvkm_wr32(device, 0x4064d0 + (i * 0x04), 0x00000000);
+	nvkm_wr32(device, 0x406500, 0x00000000);
+
+	nvkm_wr32(device, 0x405b00, (gr->tpc_total << 8) | gr->gpc_nr);
+
+	for (tmp = 0, i = 0; i < gr->gpc_nr; i++)
+		tmp |= ((1 << gr->tpc_nr[i]) - 1) << (i * 5);
+	nvkm_wr32(device, 0x4041c4, tmp);
+
+	gp100_grctx_generate_405b60(gr);
+
+	gf100_gr_icmd(gr, gr->fuc_bundle);
+	nvkm_wr32(device, 0x404154, idle_timeout);
+	gf100_gr_mthd(gr, gr->fuc_method);
+}
+
+const struct gf100_grctx_func
+gp100_grctx = {
+	.main  = gp100_grctx_generate_main,
+	.unkn  = gk104_grctx_generate_unkn,
+	.bundle = gm107_grctx_generate_bundle,
+	.bundle_size = 0x3000,
+	.bundle_min_gpm_fifo_depth = 0x180,
+	.bundle_token_limit = 0x1080,
+	.pagepool = gp100_grctx_generate_pagepool,
+	.pagepool_size = 0x20000,
+	.attrib = gp100_grctx_generate_attrib,
+	.attrib_nr_max = 0x660,
+	.attrib_nr = 0x440,
+	.alpha_nr_max = 0xc00,
+	.alpha_nr = 0x800,
+};
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.c
index 9513badb8220..157919c788e6 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.c
@@ -949,22 +949,41 @@ gf100_gr_trap_gpc_rop(struct gf100_gr *gr, int gpc)
 }
 
 static const struct nvkm_enum gf100_mp_warp_error[] = {
-	{ 0x00, "NO_ERROR" },
-	{ 0x01, "STACK_MISMATCH" },
+	{ 0x01, "STACK_ERROR" },
+	{ 0x02, "API_STACK_ERROR" },
+	{ 0x03, "RET_EMPTY_STACK_ERROR" },
+	{ 0x04, "PC_WRAP" },
 	{ 0x05, "MISALIGNED_PC" },
-	{ 0x08, "MISALIGNED_GPR" },
-	{ 0x09, "INVALID_OPCODE" },
-	{ 0x0d, "GPR_OUT_OF_BOUNDS" },
-	{ 0x0e, "MEM_OUT_OF_BOUNDS" },
-	{ 0x0f, "UNALIGNED_MEM_ACCESS" },
+	{ 0x06, "PC_OVERFLOW" },
+	{ 0x07, "MISALIGNED_IMMC_ADDR" },
+	{ 0x08, "MISALIGNED_REG" },
+	{ 0x09, "ILLEGAL_INSTR_ENCODING" },
+	{ 0x0a, "ILLEGAL_SPH_INSTR_COMBO" },
+	{ 0x0b, "ILLEGAL_INSTR_PARAM" },
+	{ 0x0c, "INVALID_CONST_ADDR" },
+	{ 0x0d, "OOR_REG" },
+	{ 0x0e, "OOR_ADDR" },
+	{ 0x0f, "MISALIGNED_ADDR" },
 	{ 0x10, "INVALID_ADDR_SPACE" },
-	{ 0x11, "INVALID_PARAM" },
+	{ 0x11, "ILLEGAL_INSTR_PARAM2" },
+	{ 0x12, "INVALID_CONST_ADDR_LDC" },
+	{ 0x13, "GEOMETRY_SM_ERROR" },
+	{ 0x14, "DIVERGENT" },
+	{ 0x15, "WARP_EXIT" },
 	{}
 };
 
 static const struct nvkm_bitfield gf100_mp_global_error[] = {
+	{ 0x00000001, "SM_TO_SM_FAULT" },
+	{ 0x00000002, "L1_ERROR" },
 	{ 0x00000004, "MULTIPLE_WARP_ERRORS" },
-	{ 0x00000008, "OUT_OF_STACK_SPACE" },
+	{ 0x00000008, "PHYSICAL_STACK_OVERFLOW" },
+	{ 0x00000010, "BPT_INT" },
+	{ 0x00000020, "BPT_PAUSE" },
+	{ 0x00000040, "SINGLE_STEP_COMPLETE" },
+	{ 0x20000000, "ECC_SEC_ERROR" },
+	{ 0x40000000, "ECC_DED_ERROR" },
+	{ 0x80000000, "TIMEOUT" },
 	{}
 };
 
@@ -1438,24 +1457,30 @@ gf100_gr_init_ctxctl(struct gf100_gr *gr)
 	struct nvkm_device *device = subdev->device;
 	struct nvkm_secboot *sb = device->secboot;
 	int i;
+	int ret = 0;
 
 	if (gr->firmware) {
 		/* load fuc microcode */
-		nvkm_mc_unk260(device->mc, 0);
+		nvkm_mc_unk260(device, 0);
 
 		/* securely-managed falcons must be reset using secure boot */
 		if (nvkm_secboot_is_managed(sb, NVKM_SECBOOT_FALCON_FECS))
-			nvkm_secboot_reset(sb, NVKM_SECBOOT_FALCON_FECS);
+			ret = nvkm_secboot_reset(sb, NVKM_SECBOOT_FALCON_FECS);
 		else
 			gf100_gr_init_fw(gr, 0x409000, &gr->fuc409c,
 					 &gr->fuc409d);
+		if (ret)
+			return ret;
+
 		if (nvkm_secboot_is_managed(sb, NVKM_SECBOOT_FALCON_GPCCS))
-			nvkm_secboot_reset(sb, NVKM_SECBOOT_FALCON_GPCCS);
+			ret = nvkm_secboot_reset(sb, NVKM_SECBOOT_FALCON_GPCCS);
 		else
 			gf100_gr_init_fw(gr, 0x41a000, &gr->fuc41ac,
 					 &gr->fuc41ad);
+		if (ret)
+			return ret;
 
-		nvkm_mc_unk260(device->mc, 1);
+		nvkm_mc_unk260(device, 1);
 
 		/* start both of them running */
 		nvkm_wr32(device, 0x409840, 0xffffffff);
@@ -1557,7 +1582,7 @@ gf100_gr_init_ctxctl(struct gf100_gr *gr)
 	}
 
 	/* load HUB microcode */
-	nvkm_mc_unk260(device->mc, 0);
+	nvkm_mc_unk260(device, 0);
 	nvkm_wr32(device, 0x4091c0, 0x01000000);
 	for (i = 0; i < gr->func->fecs.ucode->data.size / 4; i++)
 		nvkm_wr32(device, 0x4091c4, gr->func->fecs.ucode->data.data[i]);
@@ -1580,7 +1605,7 @@ gf100_gr_init_ctxctl(struct gf100_gr *gr)
 			nvkm_wr32(device, 0x41a188, i >> 6);
 		nvkm_wr32(device, 0x41a184, gr->func->gpccs.ucode->code.data[i]);
 	}
-	nvkm_mc_unk260(device->mc, 1);
+	nvkm_mc_unk260(device, 1);
 
 	/* load register lists */
 	gf100_gr_init_csdata(gr, grctx->hub, 0x409000, 0x000, 0x000000);
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.h b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.h
index 2b98abdb9270..268b8d60ff73 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.h
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.h
@@ -292,4 +292,6 @@ extern const struct gf100_gr_init gm107_gr_init_l1c_0[];
 extern const struct gf100_gr_init gm107_gr_init_wwdx_0[];
 extern const struct gf100_gr_init gm107_gr_init_cbm_0[];
 void gm107_gr_init_bios(struct gf100_gr *);
+
+void gm200_gr_init_gpc_mmu(struct gf100_gr *);
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gk20a.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gk20a.c
index 4ca8ed15191c..de8b806b88fd 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gk20a.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gk20a.c
@@ -361,6 +361,5 @@ gk20a_gr_new(struct nvkm_device *device, int index, struct nvkm_gr **pgr)
 	if (ret)
 		return ret;
 
-
 	return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gm200.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gm200.c
index 4dfa4513bb6c..6435f1257572 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gm200.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gm200.c
@@ -38,7 +38,7 @@ gm200_gr_rops(struct gf100_gr *gr)
 	return nvkm_rd32(gr->base.engine.subdev.device, 0x12006c);
 }
 
-static void
+void
 gm200_gr_init_gpc_mmu(struct gf100_gr *gr)
 {
 	struct nvkm_device *device = gr->base.engine.subdev.device;
diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gp100.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gp100.c
new file mode 100644
index 000000000000..26ad79def0ff
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gp100.c
@@ -0,0 +1,171 @@
+/*
+ * Copyright 2016 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs <bskeggs@redhat.com>
+ */
+#include "gf100.h"
+#include "ctxgf100.h"
+
+#include <nvif/class.h>
+
+/*******************************************************************************
+ * PGRAPH engine/subdev functions
+ ******************************************************************************/
+
+static void
+gp100_gr_init_rop_active_fbps(struct gf100_gr *gr)
+{
+	struct nvkm_device *device = gr->base.engine.subdev.device;
+	/*XXX: otherwise identical to gm200 aside from mask.. do everywhere? */
+	const u32 fbp_count = nvkm_rd32(device, 0x12006c) & 0x0000000f;
+	nvkm_mask(device, 0x408850, 0x0000000f, fbp_count); /* zrop */
+	nvkm_mask(device, 0x408958, 0x0000000f, fbp_count); /* crop */
+}
+
+static int
+gp100_gr_init(struct gf100_gr *gr)
+{
+	struct nvkm_device *device = gr->base.engine.subdev.device;
+	const u32 magicgpc918 = DIV_ROUND_UP(0x00800000, gr->tpc_total);
+	u32 data[TPC_MAX / 8] = {};
+	u8  tpcnr[GPC_MAX];
+	int gpc, tpc, rop;
+	int i;
+
+	gr->func->init_gpc_mmu(gr);
+
+	gf100_gr_mmio(gr, gr->fuc_sw_nonctx);
+
+	nvkm_wr32(device, GPC_UNIT(0, 0x3018), 0x00000001);
+
+	memset(data, 0x00, sizeof(data));
+	memcpy(tpcnr, gr->tpc_nr, sizeof(gr->tpc_nr));
+	for (i = 0, gpc = -1; i < gr->tpc_total; i++) {
+		do {
+			gpc = (gpc + 1) % gr->gpc_nr;
+		} while (!tpcnr[gpc]);
+		tpc = gr->tpc_nr[gpc] - tpcnr[gpc]--;
+
+		data[i / 8] |= tpc << ((i % 8) * 4);
+	}
+
+	nvkm_wr32(device, GPC_BCAST(0x0980), data[0]);
+	nvkm_wr32(device, GPC_BCAST(0x0984), data[1]);
+	nvkm_wr32(device, GPC_BCAST(0x0988), data[2]);
+	nvkm_wr32(device, GPC_BCAST(0x098c), data[3]);
+
+	for (gpc = 0; gpc < gr->gpc_nr; gpc++) {
+		nvkm_wr32(device, GPC_UNIT(gpc, 0x0914),
+			  gr->screen_tile_row_offset << 8 | gr->tpc_nr[gpc]);
+		nvkm_wr32(device, GPC_UNIT(gpc, 0x0910), 0x00040000 |
+							 gr->tpc_total);
+		nvkm_wr32(device, GPC_UNIT(gpc, 0x0918), magicgpc918);
+	}
+
+	nvkm_wr32(device, GPC_BCAST(0x3fd4), magicgpc918);
+	nvkm_wr32(device, GPC_BCAST(0x08ac), nvkm_rd32(device, 0x100800));
+	nvkm_wr32(device, GPC_BCAST(0x033c), nvkm_rd32(device, 0x100804));
+
+	gr->func->init_rop_active_fbps(gr);
+
+	nvkm_wr32(device, 0x400500, 0x00010001);
+	nvkm_wr32(device, 0x400100, 0xffffffff);
+	nvkm_wr32(device, 0x40013c, 0xffffffff);
+	nvkm_wr32(device, 0x400124, 0x00000002);
+	nvkm_wr32(device, 0x409c24, 0x000f0002);
+	nvkm_wr32(device, 0x405848, 0xc0000000);
+	nvkm_mask(device, 0x40584c, 0x00000000, 0x00000001);
+	nvkm_wr32(device, 0x404000, 0xc0000000);
+	nvkm_wr32(device, 0x404600, 0xc0000000);
+	nvkm_wr32(device, 0x408030, 0xc0000000);
+	nvkm_wr32(device, 0x404490, 0xc0000000);
+	nvkm_wr32(device, 0x406018, 0xc0000000);
+	nvkm_wr32(device, 0x407020, 0x40000000);
+	nvkm_wr32(device, 0x405840, 0xc0000000);
+	nvkm_wr32(device, 0x405844, 0x00ffffff);
+	nvkm_mask(device, 0x419cc0, 0x00000008, 0x00000008);
+
+	nvkm_mask(device, 0x419c9c, 0x00010000, 0x00010000);
+	nvkm_mask(device, 0x419c9c, 0x00020000, 0x00020000);
+
+	gr->func->init_ppc_exceptions(gr);
+
+	for (gpc = 0; gpc < gr->gpc_nr; gpc++) {
+		nvkm_wr32(device, GPC_UNIT(gpc, 0x0420), 0xc0000000);
+		nvkm_wr32(device, GPC_UNIT(gpc, 0x0900), 0xc0000000);
+		nvkm_wr32(device, GPC_UNIT(gpc, 0x1028), 0xc0000000);
+		nvkm_wr32(device, GPC_UNIT(gpc, 0x0824), 0xc0000000);
+		for (tpc = 0; tpc < gr->tpc_nr[gpc]; tpc++) {
+			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x508), 0xffffffff);
+			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x50c), 0xffffffff);
+			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x224), 0xc0000000);
+			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x48c), 0xc0000000);
+			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x084), 0xc0000000);
+			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x430), 0xc0000000);
+			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x644), 0x00dffffe);
+			nvkm_wr32(device, TPC_UNIT(gpc, tpc, 0x64c), 0x00000105);
+		}
+		nvkm_wr32(device, GPC_UNIT(gpc, 0x2c90), 0xffffffff);
+		nvkm_wr32(device, GPC_UNIT(gpc, 0x2c94), 0xffffffff);
+	}
+
+	for (rop = 0; rop < gr->rop_nr; rop++) {
+		nvkm_wr32(device, ROP_UNIT(rop, 0x144), 0x40000000);
+		nvkm_wr32(device, ROP_UNIT(rop, 0x070), 0x40000000);
+		nvkm_wr32(device, ROP_UNIT(rop, 0x204), 0xffffffff);
+		nvkm_wr32(device, ROP_UNIT(rop, 0x208), 0xffffffff);
+	}
+
+	nvkm_wr32(device, 0x400108, 0xffffffff);
+	nvkm_wr32(device, 0x400138, 0xffffffff);
+	nvkm_wr32(device, 0x400118, 0xffffffff);
+	nvkm_wr32(device, 0x400130, 0xffffffff);
+	nvkm_wr32(device, 0x40011c, 0xffffffff);
+	nvkm_wr32(device, 0x400134, 0xffffffff);
+
+	gf100_gr_zbc_init(gr);
+
+	return gf100_gr_init_ctxctl(gr);
+}
+
+static const struct gf100_gr_func
+gp100_gr = {
+	.init = gp100_gr_init,
+	.init_gpc_mmu = gm200_gr_init_gpc_mmu,
+	.init_rop_active_fbps = gp100_gr_init_rop_active_fbps,
+	.init_ppc_exceptions = gk104_gr_init_ppc_exceptions,
+	.rops = gm200_gr_rops,
+	.ppc_nr = 2,
+	.grctx = &gp100_grctx,
+	.sclass = {
+		{ -1, -1, FERMI_TWOD_A },
+		{ -1, -1, KEPLER_INLINE_TO_MEMORY_B },
+		{ -1, -1, PASCAL_A, &gf100_fermi },
+		{ -1, -1, PASCAL_COMPUTE_A },
+		{}
+	}
+};
+
+int
+gp100_gr_new(struct nvkm_device *device, int index, struct nvkm_gr **pgr)
+{
+	return gm200_gr_new_(&gp100_gr, device, index, pgr);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/bios/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/bios/base.c
index e15b9627b07e..f3c30b2a788e 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/bios/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/bios/base.c
@@ -26,6 +26,49 @@
 #include <subdev/bios.h>
 #include <subdev/bios/bmp.h>
 #include <subdev/bios/bit.h>
+#include <subdev/bios/image.h>
+
+static bool
+nvbios_addr(struct nvkm_bios *bios, u32 *addr, u8 size)
+{
+	u32 p = *addr;
+
+	if (*addr > bios->image0_size && bios->imaged_addr) {
+		*addr -= bios->image0_size;
+		*addr += bios->imaged_addr;
+	}
+
+	if (unlikely(*addr + size >= bios->size)) {
+		nvkm_error(&bios->subdev, "OOB %d %08x %08x\n", size, p, *addr);
+		return false;
+	}
+
+	return true;
+}
+
+u8
+nvbios_rd08(struct nvkm_bios *bios, u32 addr)
+{
+	if (likely(nvbios_addr(bios, &addr, 1)))
+		return bios->data[addr];
+	return 0x00;
+}
+
+u16
+nvbios_rd16(struct nvkm_bios *bios, u32 addr)
+{
+	if (likely(nvbios_addr(bios, &addr, 2)))
+		return get_unaligned_le16(&bios->data[addr]);
+	return 0x0000;
+}
+
+u32
+nvbios_rd32(struct nvkm_bios *bios, u32 addr)
+{
+	if (likely(nvbios_addr(bios, &addr, 4)))
+		return get_unaligned_le32(&bios->data[addr]);
+	return 0x00000000;
+}
 
 u8
 nvbios_checksum(const u8 *data, int size)
@@ -100,8 +143,9 @@ int
 nvkm_bios_new(struct nvkm_device *device, int index, struct nvkm_bios **pbios)
 {
 	struct nvkm_bios *bios;
+	struct nvbios_image image;
 	struct bit_entry bit_i;
-	int ret;
+	int ret, idx = 0;
 
 	if (!(bios = *pbios = kzalloc(sizeof(*bios), GFP_KERNEL)))
 		return -ENOMEM;
@@ -111,6 +155,19 @@ nvkm_bios_new(struct nvkm_device *device, int index, struct nvkm_bios **pbios)
 	if (ret)
 		return ret;
 
+	/* Some tables have weird pointers that need adjustment before
+	 * they're dereferenced.  I'm not entirely sure why...
+	 */
+	if (nvbios_image(bios, idx++, &image)) {
+		bios->image0_size = image.size;
+		while (nvbios_image(bios, idx++, &image)) {
+			if (image.type == 0xe0) {
+				bios->imaged_addr = image.base;
+				break;
+			}
+		}
+	}
+
 	/* detect type of vbios we're dealing with */
 	bios->bmp_offset = nvbios_findstr(bios->data, bios->size,
 					  "\xff\x7f""NV\0", 5);
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/bios/disp.c b/drivers/gpu/drm/nouveau/nvkm/subdev/bios/disp.c
index a5e92135cd77..9efb1b48cd54 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/bios/disp.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/bios/disp.c
@@ -141,7 +141,8 @@ nvbios_ocfg_parse(struct nvkm_bios *bios, u16 outp, u8 idx,
 {
 	u16 data = nvbios_ocfg_entry(bios, outp, idx, ver, hdr, cnt, len);
 	if (data) {
-		info->match     = nvbios_rd16(bios, data + 0x00);
+		info->proto     = nvbios_rd08(bios, data + 0x00);
+		info->flags     = nvbios_rd16(bios, data + 0x01);
 		info->clkcmp[0] = nvbios_rd16(bios, data + 0x02);
 		info->clkcmp[1] = nvbios_rd16(bios, data + 0x04);
 	}
@@ -149,12 +150,13 @@ nvbios_ocfg_parse(struct nvkm_bios *bios, u16 outp, u8 idx,
 }
 
 u16
-nvbios_ocfg_match(struct nvkm_bios *bios, u16 outp, u16 type,
+nvbios_ocfg_match(struct nvkm_bios *bios, u16 outp, u8 proto, u8 flags,
 		  u8 *ver, u8 *hdr, u8 *cnt, u8 *len, struct nvbios_ocfg *info)
 {
 	u16 data, idx = 0;
 	while ((data = nvbios_ocfg_parse(bios, outp, idx++, ver, hdr, cnt, len, info))) {
-		if (info->match == type)
+		if ((info->proto == proto || info->proto == 0xff) &&
+		    (info->flags == flags))
 			break;
 	}
 	return data;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/bios/dp.c b/drivers/gpu/drm/nouveau/nvkm/subdev/bios/dp.c
index 05332476354a..d89e78c4e689 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/bios/dp.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/bios/dp.c
@@ -40,6 +40,7 @@ nvbios_dp_table(struct nvkm_bios *bios, u8 *ver, u8 *hdr, u8 *cnt, u8 *len)
 				case 0x30:
 				case 0x40:
 				case 0x41:
+				case 0x42:
 					*hdr = nvbios_rd08(bios, data + 0x01);
 					*len = nvbios_rd08(bios, data + 0x02);
 					*cnt = nvbios_rd08(bios, data + 0x03);
@@ -70,6 +71,7 @@ nvbios_dpout_entry(struct nvkm_bios *bios, u8 idx,
 			break;
 		case 0x40:
 		case 0x41:
+		case 0x42:
 			*hdr = nvbios_rd08(bios, data + 0x04);
 			*cnt = 0;
 			*len = 0;
@@ -109,6 +111,7 @@ nvbios_dpout_parse(struct nvkm_bios *bios, u8 idx,
 			break;
 		case 0x40:
 		case 0x41:
+		case 0x42:
 			info->flags     = nvbios_rd08(bios, data + 0x04);
 			info->script[0] = nvbios_rd16(bios, data + 0x05);
 			info->script[1] = nvbios_rd16(bios, data + 0x07);
@@ -180,6 +183,11 @@ nvbios_dpcfg_parse(struct nvkm_bios *bios, u16 outp, u8 idx,
 			info->pe    = nvbios_rd08(bios, data + 0x02);
 			info->tx_pu = nvbios_rd08(bios, data + 0x03);
 			break;
+		case 0x42:
+			info->dc    = nvbios_rd08(bios, data + 0x00);
+			info->pe    = nvbios_rd08(bios, data + 0x01);
+			info->tx_pu = nvbios_rd08(bios, data + 0x02);
+			break;
 		default:
 			data = 0x0000;
 			break;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/bios/image.c b/drivers/gpu/drm/nouveau/nvkm/subdev/bios/image.c
index 74b14cf09308..1dbff7aeafec 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/bios/image.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/bios/image.c
@@ -68,11 +68,16 @@ nvbios_imagen(struct nvkm_bios *bios, struct nvbios_image *image)
 bool
 nvbios_image(struct nvkm_bios *bios, int idx, struct nvbios_image *image)
 {
+	u32 imaged_addr = bios->imaged_addr;
 	memset(image, 0x00, sizeof(*image));
+	bios->imaged_addr = 0;
 	do {
 		image->base += image->size;
-		if (image->last || !nvbios_imagen(bios, image))
+		if (image->last || !nvbios_imagen(bios, image)) {
+			bios->imaged_addr = imaged_addr;
 			return false;
+		}
 	} while(idx--);
+	bios->imaged_addr = imaged_addr;
 	return true;
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/bios/pll.c b/drivers/gpu/drm/nouveau/nvkm/subdev/bios/pll.c
index 91a7dc56e406..2ca23a9157ab 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/bios/pll.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/bios/pll.c
@@ -77,15 +77,17 @@ g84_pll_mapping[] = {
 	{}
 };
 
-static u16
+static u32
 pll_limits_table(struct nvkm_bios *bios, u8 *ver, u8 *hdr, u8 *cnt, u8 *len)
 {
 	struct bit_entry bit_C;
-	u16 data = 0x0000;
+	u32 data = 0x0000;
 
 	if (!bit_entry(bios, 'C', &bit_C)) {
 		if (bit_C.version == 1 && bit_C.length >= 10)
 			data = nvbios_rd16(bios, bit_C.offset + 8);
+		if (bit_C.version == 2 && bit_C.length >= 4)
+			data = nvbios_rd32(bios, bit_C.offset + 0);
 		if (data) {
 			*ver = nvbios_rd08(bios, data + 0);
 			*hdr = nvbios_rd08(bios, data + 1);
@@ -137,12 +139,12 @@ pll_map(struct nvkm_bios *bios)
 	}
 }
 
-static u16
+static u32
 pll_map_reg(struct nvkm_bios *bios, u32 reg, u32 *type, u8 *ver, u8 *len)
 {
 	struct pll_mapping *map;
 	u8  hdr, cnt;
-	u16 data;
+	u32 data;
 
 	data = pll_limits_table(bios, ver, &hdr, &cnt, len);
 	if (data && *ver >= 0x30) {
@@ -160,7 +162,7 @@ pll_map_reg(struct nvkm_bios *bios, u32 reg, u32 *type, u8 *ver, u8 *len)
 	map = pll_map(bios);
 	while (map && map->reg) {
 		if (map->reg == reg && *ver >= 0x20) {
-			u16 addr = (data += hdr);
+			u32 addr = (data += hdr);
 			*type = map->type;
 			while (cnt--) {
 				if (nvbios_rd32(bios, data) == map->reg)
@@ -179,12 +181,12 @@ pll_map_reg(struct nvkm_bios *bios, u32 reg, u32 *type, u8 *ver, u8 *len)
 	return 0x0000;
 }
 
-static u16
+static u32
 pll_map_type(struct nvkm_bios *bios, u8 type, u32 *reg, u8 *ver, u8 *len)
 {
 	struct pll_mapping *map;
 	u8  hdr, cnt;
-	u16 data;
+	u32 data;
 
 	data = pll_limits_table(bios, ver, &hdr, &cnt, len);
 	if (data && *ver >= 0x30) {
@@ -202,7 +204,7 @@ pll_map_type(struct nvkm_bios *bios, u8 type, u32 *reg, u8 *ver, u8 *len)
 	map = pll_map(bios);
 	while (map && map->reg) {
 		if (map->type == type && *ver >= 0x20) {
-			u16 addr = (data += hdr);
+			u32 addr = (data += hdr);
 			*reg = map->reg;
 			while (cnt--) {
 				if (nvbios_rd32(bios, data) == map->reg)
@@ -228,7 +230,7 @@ nvbios_pll_parse(struct nvkm_bios *bios, u32 type, struct nvbios_pll *info)
 	struct nvkm_device *device = subdev->device;
 	u8  ver, len;
 	u32 reg = type;
-	u16 data;
+	u32 data;
 
 	if (type > PLL_MAX) {
 		reg  = type;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/bios/pmu.c b/drivers/gpu/drm/nouveau/nvkm/subdev/bios/pmu.c
index c268e5afe852..b4a308f3cf7b 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/bios/pmu.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/bios/pmu.c
@@ -26,21 +26,6 @@
 #include <subdev/bios/image.h>
 #include <subdev/bios/pmu.h>
 
-static u32
-weirdo_pointer(struct nvkm_bios *bios, u32 data)
-{
-	struct nvbios_image image;
-	int idx = 0;
-	if (nvbios_image(bios, idx++, &image)) {
-		data -= image.size;
-		while (nvbios_image(bios, idx++, &image)) {
-			if (image.type == 0xe0)
-				return image.base + data;
-		}
-	}
-	return 0;
-}
-
 u32
 nvbios_pmuTe(struct nvkm_bios *bios, u8 *ver, u8 *hdr, u8 *cnt, u8 *len)
 {
@@ -50,7 +35,7 @@ nvbios_pmuTe(struct nvkm_bios *bios, u8 *ver, u8 *hdr, u8 *cnt, u8 *len)
 	if (!bit_entry(bios, 'p', &bit_p)) {
 		if (bit_p.version == 2 && bit_p.length >= 4)
 			data = nvbios_rd32(bios, bit_p.offset + 0x00);
-		if ((data = weirdo_pointer(bios, data))) {
+		if (data) {
 			*ver = nvbios_rd08(bios, data + 0x00); /* maybe? */
 			*hdr = nvbios_rd08(bios, data + 0x01);
 			*len = nvbios_rd08(bios, data + 0x02);
@@ -97,8 +82,7 @@ nvbios_pmuRm(struct nvkm_bios *bios, u8 type, struct nvbios_pmuR *info)
 	u32 data;
 	memset(info, 0x00, sizeof(*info));
 	while ((data = nvbios_pmuEp(bios, idx++, &ver, &hdr, &pmuE))) {
-		if ( pmuE.type == type &&
-		    (data = weirdo_pointer(bios, pmuE.data))) {
+		if (pmuE.type == type && (data = pmuE.data)) {
 			info->init_addr_pmu = nvbios_rd32(bios, data + 0x08);
 			info->args_addr_pmu = nvbios_rd32(bios, data + 0x0c);
 			info->boot_addr     = data + 0x30;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/bios/rammap.c b/drivers/gpu/drm/nouveau/nvkm/subdev/bios/rammap.c
index d0ae7454764e..b57c370c725d 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/bios/rammap.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/bios/rammap.c
@@ -30,11 +30,11 @@ nvbios_rammapTe(struct nvkm_bios *bios, u8 *ver, u8 *hdr,
 		u8 *cnt, u8 *len, u8 *snr, u8 *ssz)
 {
 	struct bit_entry bit_P;
-	u16 rammap = 0x0000;
+	u32 rammap = 0x0000;
 
 	if (!bit_entry(bios, 'P', &bit_P)) {
 		if (bit_P.version == 2)
-			rammap = nvbios_rd16(bios, bit_P.offset + 4);
+			rammap = nvbios_rd32(bios, bit_P.offset + 4);
 
 		if (rammap) {
 			*ver = nvbios_rd08(bios, rammap + 0);
@@ -61,7 +61,7 @@ nvbios_rammapEe(struct nvkm_bios *bios, int idx,
 		u8 *ver, u8 *hdr, u8 *cnt, u8 *len)
 {
 	u8  snr, ssz;
-	u16 rammap = nvbios_rammapTe(bios, ver, hdr, cnt, len, &snr, &ssz);
+	u32 rammap = nvbios_rammapTe(bios, ver, hdr, cnt, len, &snr, &ssz);
 	if (rammap && idx < *cnt) {
 		rammap = rammap + *hdr + (idx * (*len + (snr * ssz)));
 		*hdr = *len;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/clk/gf100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/clk/gf100.c
index 78c449b417b7..89d5543118cf 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/clk/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/clk/gf100.c
@@ -99,7 +99,7 @@ read_div(struct gf100_clk *clk, int doff, u32 dsrc, u32 dctl)
 {
 	struct nvkm_device *device = clk->base.subdev.device;
 	u32 ssrc = nvkm_rd32(device, dsrc + (doff * 4));
-	u32 sctl = nvkm_rd32(device, dctl + (doff * 4));
+	u32 sclk, sctl, sdiv = 2;
 
 	switch (ssrc & 0x00000003) {
 	case 0:
@@ -109,13 +109,21 @@ read_div(struct gf100_clk *clk, int doff, u32 dsrc, u32 dctl)
 	case 2:
 		return 100000;
 	case 3:
-		if (sctl & 0x80000000) {
-			u32 sclk = read_vco(clk, dsrc + (doff * 4));
-			u32 sdiv = (sctl & 0x0000003f) + 2;
-			return (sclk * 2) / sdiv;
+		sclk = read_vco(clk, dsrc + (doff * 4));
+
+		/* Memclk has doff of 0 despite its alt. location */
+		if (doff <= 2) {
+			sctl = nvkm_rd32(device, dctl + (doff * 4));
+
+			if (sctl & 0x80000000) {
+				if (ssrc & 0x100)
+					sctl >>= 8;
+
+				sdiv = (sctl & 0x3f) + 2;
+			}
 		}
 
-		return read_vco(clk, dsrc + (doff * 4));
+		return (sclk * 2) / sdiv;
 	default:
 		return 0;
 	}
@@ -366,11 +374,17 @@ gf100_clk_prog_2(struct gf100_clk *clk, int idx)
 		if (info->coef) {
 			nvkm_wr32(device, addr + 0x04, info->coef);
 			nvkm_mask(device, addr + 0x00, 0x00000001, 0x00000001);
+
+			/* Test PLL lock */
+			nvkm_mask(device, addr + 0x00, 0x00000010, 0x00000000);
 			nvkm_msec(device, 2000,
 				if (nvkm_rd32(device, addr + 0x00) & 0x00020000)
 					break;
 			);
-			nvkm_mask(device, addr + 0x00, 0x00020004, 0x00000004);
+			nvkm_mask(device, addr + 0x00, 0x00000010, 0x00000010);
+
+			/* Enable sync mode */
+			nvkm_mask(device, addr + 0x00, 0x00000004, 0x00000004);
 		}
 	}
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/clk/gk104.c b/drivers/gpu/drm/nouveau/nvkm/subdev/clk/gk104.c
index 975c401bccab..06bc0d2d6ae1 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/clk/gk104.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/clk/gk104.c
@@ -393,11 +393,17 @@ gk104_clk_prog_2(struct gk104_clk *clk, int idx)
 	if (info->coef) {
 		nvkm_wr32(device, addr + 0x04, info->coef);
 		nvkm_mask(device, addr + 0x00, 0x00000001, 0x00000001);
+
+		/* Test PLL lock */
+		nvkm_mask(device, addr + 0x00, 0x00000010, 0x00000000);
 		nvkm_msec(device, 2000,
 			if (nvkm_rd32(device, addr + 0x00) & 0x00020000)
 				break;
 		);
-		nvkm_mask(device, addr + 0x00, 0x00020004, 0x00000004);
+		nvkm_mask(device, addr + 0x00, 0x00000010, 0x00000010);
+
+		/* Enable sync mode */
+		nvkm_mask(device, addr + 0x00, 0x00000004, 0x00000004);
 	}
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/clk/gk20a.c b/drivers/gpu/drm/nouveau/nvkm/subdev/clk/gk20a.c
index 5f0ee24e31b8..218893e3e5f9 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/clk/gk20a.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/clk/gk20a.c
@@ -28,69 +28,6 @@
 #include <core/tegra.h>
 #include <subdev/timer.h>
 
-#define KHZ (1000)
-#define MHZ (KHZ * 1000)
-
-#define MASK(w)	((1 << w) - 1)
-
-#define GPCPLL_CFG		(SYS_GPCPLL_CFG_BASE + 0)
-#define GPCPLL_CFG_ENABLE	BIT(0)
-#define GPCPLL_CFG_IDDQ		BIT(1)
-#define GPCPLL_CFG_LOCK_DET_OFF	BIT(4)
-#define GPCPLL_CFG_LOCK		BIT(17)
-
-#define GPCPLL_COEFF		(SYS_GPCPLL_CFG_BASE + 4)
-#define GPCPLL_COEFF_M_SHIFT	0
-#define GPCPLL_COEFF_M_WIDTH	8
-#define GPCPLL_COEFF_N_SHIFT	8
-#define GPCPLL_COEFF_N_WIDTH	8
-#define GPCPLL_COEFF_P_SHIFT	16
-#define GPCPLL_COEFF_P_WIDTH	6
-
-#define GPCPLL_CFG2			(SYS_GPCPLL_CFG_BASE + 0xc)
-#define GPCPLL_CFG2_SETUP2_SHIFT	16
-#define GPCPLL_CFG2_PLL_STEPA_SHIFT	24
-
-#define GPCPLL_CFG3			(SYS_GPCPLL_CFG_BASE + 0x18)
-#define GPCPLL_CFG3_PLL_STEPB_SHIFT	16
-
-#define GPC_BCASE_GPCPLL_CFG_BASE		0x00132800
-#define GPCPLL_NDIV_SLOWDOWN			(SYS_GPCPLL_CFG_BASE + 0x1c)
-#define GPCPLL_NDIV_SLOWDOWN_NDIV_LO_SHIFT	0
-#define GPCPLL_NDIV_SLOWDOWN_NDIV_MID_SHIFT	8
-#define GPCPLL_NDIV_SLOWDOWN_STEP_SIZE_LO2MID_SHIFT	16
-#define GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT	22
-#define GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT	31
-
-#define SEL_VCO				(SYS_GPCPLL_CFG_BASE + 0x100)
-#define SEL_VCO_GPC2CLK_OUT_SHIFT	0
-
-#define GPC2CLK_OUT			(SYS_GPCPLL_CFG_BASE + 0x250)
-#define GPC2CLK_OUT_SDIV14_INDIV4_WIDTH	1
-#define GPC2CLK_OUT_SDIV14_INDIV4_SHIFT	31
-#define GPC2CLK_OUT_SDIV14_INDIV4_MODE	1
-#define GPC2CLK_OUT_VCODIV_WIDTH	6
-#define GPC2CLK_OUT_VCODIV_SHIFT	8
-#define GPC2CLK_OUT_VCODIV1		0
-#define GPC2CLK_OUT_VCODIV_MASK		(MASK(GPC2CLK_OUT_VCODIV_WIDTH) << \
-					GPC2CLK_OUT_VCODIV_SHIFT)
-#define GPC2CLK_OUT_BYPDIV_WIDTH	6
-#define GPC2CLK_OUT_BYPDIV_SHIFT	0
-#define GPC2CLK_OUT_BYPDIV31		0x3c
-#define GPC2CLK_OUT_INIT_MASK	((MASK(GPC2CLK_OUT_SDIV14_INDIV4_WIDTH) << \
-		GPC2CLK_OUT_SDIV14_INDIV4_SHIFT)\
-		| (MASK(GPC2CLK_OUT_VCODIV_WIDTH) << GPC2CLK_OUT_VCODIV_SHIFT)\
-		| (MASK(GPC2CLK_OUT_BYPDIV_WIDTH) << GPC2CLK_OUT_BYPDIV_SHIFT))
-#define GPC2CLK_OUT_INIT_VAL	((GPC2CLK_OUT_SDIV14_INDIV4_MODE << \
-		GPC2CLK_OUT_SDIV14_INDIV4_SHIFT) \
-		| (GPC2CLK_OUT_VCODIV1 << GPC2CLK_OUT_VCODIV_SHIFT) \
-		| (GPC2CLK_OUT_BYPDIV31 << GPC2CLK_OUT_BYPDIV_SHIFT))
-
-#define GPC_BCAST_NDIV_SLOWDOWN_DEBUG	(GPC_BCASE_GPCPLL_CFG_BASE + 0xa0)
-#define GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_SHIFT	24
-#define GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_MASK \
-	    (0x1 << GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_SHIFT)
-
 static const u8 _pl_to_div[] = {
 /* PL:   0, 1, 2, 3, 4, 5, 6,  7,  8,  9, 10, 11, 12, 13, 14 */
 /* p: */ 1, 2, 3, 4, 5, 6, 8, 10, 12, 16, 12, 16, 20, 24, 32,
@@ -124,7 +61,7 @@ static const struct gk20a_clk_pllg_params gk20a_pllg_params = {
 	.min_pl = 1, .max_pl = 32,
 };
 
-static void
+void
 gk20a_pllg_read_mnp(struct gk20a_clk *clk, struct gk20a_pll *pll)
 {
 	struct nvkm_device *device = clk->base.subdev.device;
@@ -136,20 +73,33 @@ gk20a_pllg_read_mnp(struct gk20a_clk *clk, struct gk20a_pll *pll)
 	pll->pl = (val >> GPCPLL_COEFF_P_SHIFT) & MASK(GPCPLL_COEFF_P_WIDTH);
 }
 
-static u32
-gk20a_pllg_calc_rate(struct gk20a_clk *clk)
+void
+gk20a_pllg_write_mnp(struct gk20a_clk *clk, const struct gk20a_pll *pll)
+{
+	struct nvkm_device *device = clk->base.subdev.device;
+	u32 val;
+
+	val = (pll->m & MASK(GPCPLL_COEFF_M_WIDTH)) << GPCPLL_COEFF_M_SHIFT;
+	val |= (pll->n & MASK(GPCPLL_COEFF_N_WIDTH)) << GPCPLL_COEFF_N_SHIFT;
+	val |= (pll->pl & MASK(GPCPLL_COEFF_P_WIDTH)) << GPCPLL_COEFF_P_SHIFT;
+	nvkm_wr32(device, GPCPLL_COEFF, val);
+}
+
+u32
+gk20a_pllg_calc_rate(struct gk20a_clk *clk, struct gk20a_pll *pll)
 {
 	u32 rate;
 	u32 divider;
 
-	rate = clk->parent_rate * clk->pll.n;
-	divider = clk->pll.m * clk->pl_to_div(clk->pll.pl);
+	rate = clk->parent_rate * pll->n;
+	divider = pll->m * clk->pl_to_div(pll->pl);
 
 	return rate / divider / 2;
 }
 
-static int
-gk20a_pllg_calc_mnp(struct gk20a_clk *clk, unsigned long rate)
+int
+gk20a_pllg_calc_mnp(struct gk20a_clk *clk, unsigned long rate,
+		    struct gk20a_pll *pll)
 {
 	struct nvkm_subdev *subdev = &clk->base.subdev;
 	u32 target_clk_f, ref_clk_f, target_freq;
@@ -163,16 +113,13 @@ gk20a_pllg_calc_mnp(struct gk20a_clk *clk, unsigned long rate)
 	target_clk_f = rate * 2 / KHZ;
 	ref_clk_f = clk->parent_rate / KHZ;
 
-	max_vco_f = clk->params->max_vco;
+	target_vco_f = target_clk_f + target_clk_f / 50;
+	max_vco_f = max(clk->params->max_vco, target_vco_f);
 	min_vco_f = clk->params->min_vco;
 	best_m = clk->params->max_m;
 	best_n = clk->params->min_n;
 	best_pl = clk->params->min_pl;
 
-	target_vco_f = target_clk_f + target_clk_f / 50;
-	if (max_vco_f < target_vco_f)
-		max_vco_f = target_vco_f;
-
 	/* min_pl <= high_pl <= max_pl */
 	high_pl = (max_vco_f + target_vco_f - 1) / target_vco_f;
 	high_pl = min(high_pl, clk->params->max_pl);
@@ -195,9 +142,7 @@ gk20a_pllg_calc_mnp(struct gk20a_clk *clk, unsigned long rate)
 		target_vco_f = target_clk_f * clk->pl_to_div(pl);
 
 		for (m = clk->params->min_m; m <= clk->params->max_m; m++) {
-			u32 u_f, vco_f;
-
-			u_f = ref_clk_f / m;
+			u32 u_f = ref_clk_f / m;
 
 			if (u_f < clk->params->min_u)
 				break;
@@ -211,6 +156,8 @@ gk20a_pllg_calc_mnp(struct gk20a_clk *clk, unsigned long rate)
 				break;
 
 			for (; n <= n2; n++) {
+				u32 vco_f;
+
 				if (n < clk->params->min_n)
 					continue;
 				if (n > clk->params->max_n)
@@ -247,16 +194,16 @@ found_match:
 			   "no best match for target @ %dMHz on gpc_pll",
 			   target_clk_f / KHZ);
 
-	clk->pll.m = best_m;
-	clk->pll.n = best_n;
-	clk->pll.pl = best_pl;
+	pll->m = best_m;
+	pll->n = best_n;
+	pll->pl = best_pl;
 
-	target_freq = gk20a_pllg_calc_rate(clk);
+	target_freq = gk20a_pllg_calc_rate(clk, pll);
 
 	nvkm_debug(subdev,
-		   "actual target freq %d MHz, M %d, N %d, PL %d(div%d)\n",
-		   target_freq / MHZ, clk->pll.m, clk->pll.n, clk->pll.pl,
-		   clk->pl_to_div(clk->pll.pl));
+		   "actual target freq %d KHz, M %d, N %d, PL %d(div%d)\n",
+		   target_freq / KHZ, pll->m, pll->n, pll->pl,
+		   clk->pl_to_div(pll->pl));
 	return 0;
 }
 
@@ -265,45 +212,36 @@ gk20a_pllg_slide(struct gk20a_clk *clk, u32 n)
 {
 	struct nvkm_subdev *subdev = &clk->base.subdev;
 	struct nvkm_device *device = subdev->device;
-	u32 val;
-	int ramp_timeout;
+	struct gk20a_pll pll;
+	int ret = 0;
 
 	/* get old coefficients */
-	val = nvkm_rd32(device, GPCPLL_COEFF);
+	gk20a_pllg_read_mnp(clk, &pll);
 	/* do nothing if NDIV is the same */
-	if (n == ((val >> GPCPLL_COEFF_N_SHIFT) & MASK(GPCPLL_COEFF_N_WIDTH)))
+	if (n == pll.n)
 		return 0;
 
-	/* setup */
-	nvkm_mask(device, GPCPLL_CFG2, 0xff << GPCPLL_CFG2_PLL_STEPA_SHIFT,
-		0x2b << GPCPLL_CFG2_PLL_STEPA_SHIFT);
-	nvkm_mask(device, GPCPLL_CFG3, 0xff << GPCPLL_CFG3_PLL_STEPB_SHIFT,
-		0xb << GPCPLL_CFG3_PLL_STEPB_SHIFT);
-
 	/* pll slowdown mode */
 	nvkm_mask(device, GPCPLL_NDIV_SLOWDOWN,
 		BIT(GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT),
 		BIT(GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT));
 
 	/* new ndiv ready for ramp */
-	val = nvkm_rd32(device, GPCPLL_COEFF);
-	val &= ~(MASK(GPCPLL_COEFF_N_WIDTH) << GPCPLL_COEFF_N_SHIFT);
-	val |= (n & MASK(GPCPLL_COEFF_N_WIDTH)) << GPCPLL_COEFF_N_SHIFT;
+	pll.n = n;
 	udelay(1);
-	nvkm_wr32(device, GPCPLL_COEFF, val);
+	gk20a_pllg_write_mnp(clk, &pll);
 
 	/* dynamic ramp to new ndiv */
-	val = nvkm_rd32(device, GPCPLL_NDIV_SLOWDOWN);
-	val |= 0x1 << GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT;
 	udelay(1);
-	nvkm_wr32(device, GPCPLL_NDIV_SLOWDOWN, val);
+	nvkm_mask(device, GPCPLL_NDIV_SLOWDOWN,
+		  BIT(GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT),
+		  BIT(GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT));
 
-	for (ramp_timeout = 500; ramp_timeout > 0; ramp_timeout--) {
-		udelay(1);
-		val = nvkm_rd32(device, GPC_BCAST_NDIV_SLOWDOWN_DEBUG);
-		if (val & GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_MASK)
-			break;
-	}
+	/* wait for ramping to complete */
+	if (nvkm_wait_usec(device, 500, GPC_BCAST_NDIV_SLOWDOWN_DEBUG,
+		GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_MASK,
+		GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_MASK) < 0)
+		ret = -ETIMEDOUT;
 
 	/* exit slowdown mode */
 	nvkm_mask(device, GPCPLL_NDIV_SLOWDOWN,
@@ -311,21 +249,35 @@ gk20a_pllg_slide(struct gk20a_clk *clk, u32 n)
 		BIT(GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT), 0);
 	nvkm_rd32(device, GPCPLL_NDIV_SLOWDOWN);
 
-	if (ramp_timeout <= 0) {
-		nvkm_error(subdev, "gpcpll dynamic ramp timeout\n");
-		return -ETIMEDOUT;
-	}
-
-	return 0;
+	return ret;
 }
 
-static void
+static int
 gk20a_pllg_enable(struct gk20a_clk *clk)
 {
 	struct nvkm_device *device = clk->base.subdev.device;
+	u32 val;
 
 	nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_ENABLE, GPCPLL_CFG_ENABLE);
 	nvkm_rd32(device, GPCPLL_CFG);
+
+	/* enable lock detection */
+	val = nvkm_rd32(device, GPCPLL_CFG);
+	if (val & GPCPLL_CFG_LOCK_DET_OFF) {
+		val &= ~GPCPLL_CFG_LOCK_DET_OFF;
+		nvkm_wr32(device, GPCPLL_CFG, val);
+	}
+
+	/* wait for lock */
+	if (nvkm_wait_usec(device, 300, GPCPLL_CFG, GPCPLL_CFG_LOCK,
+			   GPCPLL_CFG_LOCK) < 0)
+		return -ETIMEDOUT;
+
+	/* switch to VCO mode */
+	nvkm_mask(device, SEL_VCO, BIT(SEL_VCO_GPC2CLK_OUT_SHIFT),
+		BIT(SEL_VCO_GPC2CLK_OUT_SHIFT));
+
+	return 0;
 }
 
 static void
@@ -333,117 +285,81 @@ gk20a_pllg_disable(struct gk20a_clk *clk)
 {
 	struct nvkm_device *device = clk->base.subdev.device;
 
+	/* put PLL in bypass before disabling it */
+	nvkm_mask(device, SEL_VCO, BIT(SEL_VCO_GPC2CLK_OUT_SHIFT), 0);
+
 	nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_ENABLE, 0);
 	nvkm_rd32(device, GPCPLL_CFG);
 }
 
 static int
-_gk20a_pllg_program_mnp(struct gk20a_clk *clk, bool allow_slide)
+gk20a_pllg_program_mnp(struct gk20a_clk *clk, const struct gk20a_pll *pll)
 {
 	struct nvkm_subdev *subdev = &clk->base.subdev;
 	struct nvkm_device *device = subdev->device;
-	u32 val, cfg;
-	struct gk20a_pll old_pll;
-	u32 n_lo;
-
-	/* get old coefficients */
-	gk20a_pllg_read_mnp(clk, &old_pll);
-
-	/* do NDIV slide if there is no change in M and PL */
-	cfg = nvkm_rd32(device, GPCPLL_CFG);
-	if (allow_slide && clk->pll.m == old_pll.m &&
-	    clk->pll.pl == old_pll.pl && (cfg & GPCPLL_CFG_ENABLE)) {
-		return gk20a_pllg_slide(clk, clk->pll.n);
-	}
-
-	/* slide down to NDIV_LO */
-	if (allow_slide && (cfg & GPCPLL_CFG_ENABLE)) {
-		int ret;
-
-		n_lo = DIV_ROUND_UP(old_pll.m * clk->params->min_vco,
-				    clk->parent_rate / KHZ);
-		ret = gk20a_pllg_slide(clk, n_lo);
+	struct gk20a_pll cur_pll;
+	int ret;
 
-		if (ret)
-			return ret;
-	}
+	gk20a_pllg_read_mnp(clk, &cur_pll);
 
-	/* split FO-to-bypass jump in halfs by setting out divider 1:2 */
+	/* split VCO-to-bypass jump in half by setting out divider 1:2 */
 	nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_VCODIV_MASK,
-		0x2 << GPC2CLK_OUT_VCODIV_SHIFT);
-
-	/* put PLL in bypass before programming it */
-	val = nvkm_rd32(device, SEL_VCO);
-	val &= ~(BIT(SEL_VCO_GPC2CLK_OUT_SHIFT));
+		  GPC2CLK_OUT_VCODIV2 << GPC2CLK_OUT_VCODIV_SHIFT);
+	/* Intentional 2nd write to assure linear divider operation */
+	nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_VCODIV_MASK,
+		  GPC2CLK_OUT_VCODIV2 << GPC2CLK_OUT_VCODIV_SHIFT);
+	nvkm_rd32(device, GPC2CLK_OUT);
 	udelay(2);
-	nvkm_wr32(device, SEL_VCO, val);
-
-	/* get out from IDDQ */
-	val = nvkm_rd32(device, GPCPLL_CFG);
-	if (val & GPCPLL_CFG_IDDQ) {
-		val &= ~GPCPLL_CFG_IDDQ;
-		nvkm_wr32(device, GPCPLL_CFG, val);
-		nvkm_rd32(device, GPCPLL_CFG);
-		udelay(2);
-	}
 
 	gk20a_pllg_disable(clk);
 
-	nvkm_debug(subdev, "%s: m=%d n=%d pl=%d\n", __func__,
-		   clk->pll.m, clk->pll.n, clk->pll.pl);
-
-	n_lo = DIV_ROUND_UP(clk->pll.m * clk->params->min_vco,
-			    clk->parent_rate / KHZ);
-	val = clk->pll.m << GPCPLL_COEFF_M_SHIFT;
-	val |= (allow_slide ? n_lo : clk->pll.n) << GPCPLL_COEFF_N_SHIFT;
-	val |= clk->pll.pl << GPCPLL_COEFF_P_SHIFT;
-	nvkm_wr32(device, GPCPLL_COEFF, val);
+	gk20a_pllg_write_mnp(clk, pll);
 
-	gk20a_pllg_enable(clk);
-
-	val = nvkm_rd32(device, GPCPLL_CFG);
-	if (val & GPCPLL_CFG_LOCK_DET_OFF) {
-		val &= ~GPCPLL_CFG_LOCK_DET_OFF;
-		nvkm_wr32(device, GPCPLL_CFG, val);
-	}
-
-	if (nvkm_usec(device, 300,
-		if (nvkm_rd32(device, GPCPLL_CFG) & GPCPLL_CFG_LOCK)
-			break;
-	) < 0)
-		return -ETIMEDOUT;
-
-	/* switch to VCO mode */
-	nvkm_mask(device, SEL_VCO, BIT(SEL_VCO_GPC2CLK_OUT_SHIFT),
-		  BIT(SEL_VCO_GPC2CLK_OUT_SHIFT));
+	ret = gk20a_pllg_enable(clk);
+	if (ret)
+		return ret;
 
 	/* restore out divider 1:1 */
-	val = nvkm_rd32(device, GPC2CLK_OUT);
-	if ((val & GPC2CLK_OUT_VCODIV_MASK) !=
-	    (GPC2CLK_OUT_VCODIV1 << GPC2CLK_OUT_VCODIV_SHIFT)) {
-		val &= ~GPC2CLK_OUT_VCODIV_MASK;
-		val |= GPC2CLK_OUT_VCODIV1 << GPC2CLK_OUT_VCODIV_SHIFT;
-		udelay(2);
-		nvkm_wr32(device, GPC2CLK_OUT, val);
-		/* Intentional 2nd write to assure linear divider operation */
-		nvkm_wr32(device, GPC2CLK_OUT, val);
-		nvkm_rd32(device, GPC2CLK_OUT);
-	}
+	udelay(2);
+	nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_VCODIV_MASK,
+		  GPC2CLK_OUT_VCODIV1 << GPC2CLK_OUT_VCODIV_SHIFT);
+	/* Intentional 2nd write to assure linear divider operation */
+	nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_VCODIV_MASK,
+		  GPC2CLK_OUT_VCODIV1 << GPC2CLK_OUT_VCODIV_SHIFT);
+	nvkm_rd32(device, GPC2CLK_OUT);
 
-	/* slide up to new NDIV */
-	return allow_slide ? gk20a_pllg_slide(clk, clk->pll.n) : 0;
+	return 0;
 }
 
 static int
-gk20a_pllg_program_mnp(struct gk20a_clk *clk)
+gk20a_pllg_program_mnp_slide(struct gk20a_clk *clk, const struct gk20a_pll *pll)
 {
-	int err;
+	struct gk20a_pll cur_pll;
+	int ret;
 
-	err = _gk20a_pllg_program_mnp(clk, true);
-	if (err)
-		err = _gk20a_pllg_program_mnp(clk, false);
+	if (gk20a_pllg_is_enabled(clk)) {
+		gk20a_pllg_read_mnp(clk, &cur_pll);
+
+		/* just do NDIV slide if there is no change to M and PL */
+		if (pll->m == cur_pll.m && pll->pl == cur_pll.pl)
+			return gk20a_pllg_slide(clk, pll->n);
+
+		/* slide down to current NDIV_LO */
+		cur_pll.n = gk20a_pllg_n_lo(clk, &cur_pll);
+		ret = gk20a_pllg_slide(clk, cur_pll.n);
+		if (ret)
+			return ret;
+	}
+
+	/* program MNP with the new clock parameters and new NDIV_LO */
+	cur_pll = *pll;
+	cur_pll.n = gk20a_pllg_n_lo(clk, &cur_pll);
+	ret = gk20a_pllg_program_mnp(clk, &cur_pll);
+	if (ret)
+		return ret;
 
-	return err;
+	/* slide up to new NDIV */
+	return gk20a_pllg_slide(clk, pll->n);
 }
 
 static struct nvkm_pstate
@@ -546,13 +462,14 @@ gk20a_clk_read(struct nvkm_clk *base, enum nv_clk_src src)
 	struct gk20a_clk *clk = gk20a_clk(base);
 	struct nvkm_subdev *subdev = &clk->base.subdev;
 	struct nvkm_device *device = subdev->device;
+	struct gk20a_pll pll;
 
 	switch (src) {
 	case nv_clk_src_crystal:
 		return device->crystal;
 	case nv_clk_src_gpc:
-		gk20a_pllg_read_mnp(clk, &clk->pll);
-		return gk20a_pllg_calc_rate(clk) / GK20A_CLK_GPC_MDIV;
+		gk20a_pllg_read_mnp(clk, &pll);
+		return gk20a_pllg_calc_rate(clk, &pll) / GK20A_CLK_GPC_MDIV;
 	default:
 		nvkm_error(subdev, "invalid clock source %d\n", src);
 		return -EINVAL;
@@ -565,15 +482,20 @@ gk20a_clk_calc(struct nvkm_clk *base, struct nvkm_cstate *cstate)
 	struct gk20a_clk *clk = gk20a_clk(base);
 
 	return gk20a_pllg_calc_mnp(clk, cstate->domain[nv_clk_src_gpc] *
-					 GK20A_CLK_GPC_MDIV);
+					 GK20A_CLK_GPC_MDIV, &clk->pll);
 }
 
 int
 gk20a_clk_prog(struct nvkm_clk *base)
 {
 	struct gk20a_clk *clk = gk20a_clk(base);
+	int ret;
+
+	ret = gk20a_pllg_program_mnp_slide(clk, &clk->pll);
+	if (ret)
+		ret = gk20a_pllg_program_mnp(clk, &clk->pll);
 
-	return gk20a_pllg_program_mnp(clk);
+	return ret;
 }
 
 void
@@ -581,29 +503,62 @@ gk20a_clk_tidy(struct nvkm_clk *base)
 {
 }
 
+int
+gk20a_clk_setup_slide(struct gk20a_clk *clk)
+{
+	struct nvkm_subdev *subdev = &clk->base.subdev;
+	struct nvkm_device *device = subdev->device;
+	u32 step_a, step_b;
+
+	switch (clk->parent_rate) {
+	case 12000000:
+	case 12800000:
+	case 13000000:
+		step_a = 0x2b;
+		step_b = 0x0b;
+		break;
+	case 19200000:
+		step_a = 0x12;
+		step_b = 0x08;
+		break;
+	case 38400000:
+		step_a = 0x04;
+		step_b = 0x05;
+		break;
+	default:
+		nvkm_error(subdev, "invalid parent clock rate %u KHz",
+			   clk->parent_rate / KHZ);
+		return -EINVAL;
+	}
+
+	nvkm_mask(device, GPCPLL_CFG2, 0xff << GPCPLL_CFG2_PLL_STEPA_SHIFT,
+		step_a << GPCPLL_CFG2_PLL_STEPA_SHIFT);
+	nvkm_mask(device, GPCPLL_CFG3, 0xff << GPCPLL_CFG3_PLL_STEPB_SHIFT,
+		step_b << GPCPLL_CFG3_PLL_STEPB_SHIFT);
+
+	return 0;
+}
+
 void
 gk20a_clk_fini(struct nvkm_clk *base)
 {
 	struct nvkm_device *device = base->subdev.device;
 	struct gk20a_clk *clk = gk20a_clk(base);
-	u32 val;
 
 	/* slide to VCO min */
-	val = nvkm_rd32(device, GPCPLL_CFG);
-	if (val & GPCPLL_CFG_ENABLE) {
+	if (gk20a_pllg_is_enabled(clk)) {
 		struct gk20a_pll pll;
 		u32 n_lo;
 
 		gk20a_pllg_read_mnp(clk, &pll);
-		n_lo = DIV_ROUND_UP(pll.m * clk->params->min_vco,
-				    clk->parent_rate / KHZ);
+		n_lo = gk20a_pllg_n_lo(clk, &pll);
 		gk20a_pllg_slide(clk, n_lo);
 	}
 
-	/* put PLL in bypass before disabling it */
-	nvkm_mask(device, SEL_VCO, BIT(SEL_VCO_GPC2CLK_OUT_SHIFT), 0);
-
 	gk20a_pllg_disable(clk);
+
+	/* set IDDQ */
+	nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_IDDQ, 1);
 }
 
 static int
@@ -614,9 +569,18 @@ gk20a_clk_init(struct nvkm_clk *base)
 	struct nvkm_device *device = subdev->device;
 	int ret;
 
+	/* get out from IDDQ */
+	nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_IDDQ, 0);
+	nvkm_rd32(device, GPCPLL_CFG);
+	udelay(5);
+
 	nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_INIT_MASK,
 		  GPC2CLK_OUT_INIT_VAL);
 
+	ret = gk20a_clk_setup_slide(clk);
+	if (ret)
+		return ret;
+
 	/* Start with lowest frequency */
 	base->func->calc(base, &base->func->pstates[0].base);
 	ret = base->func->prog(&clk->base);
@@ -646,7 +610,7 @@ gk20a_clk = {
 };
 
 int
-_gk20a_clk_ctor(struct nvkm_device *device, int index,
+gk20a_clk_ctor(struct nvkm_device *device, int index,
 		const struct nvkm_clk_func *func,
 		const struct gk20a_clk_pllg_params *params,
 		struct gk20a_clk *clk)
@@ -685,7 +649,7 @@ gk20a_clk_new(struct nvkm_device *device, int index, struct nvkm_clk **pclk)
 		return -ENOMEM;
 	*pclk = &clk->base;
 
-	ret = _gk20a_clk_ctor(device, index, &gk20a_clk, &gk20a_pllg_params,
+	ret = gk20a_clk_ctor(device, index, &gk20a_clk, &gk20a_pllg_params,
 			      clk);
 
 	clk->pl_to_div = pl_to_div;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/clk/gk20a.h b/drivers/gpu/drm/nouveau/nvkm/subdev/clk/gk20a.h
index 13c46740197d..0d1450972162 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/clk/gk20a.h
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/clk/gk20a.h
@@ -24,9 +24,79 @@
 #ifndef __NVKM_CLK_GK20A_H__
 #define __NVKM_CLK_GK20A_H__
 
+#define KHZ (1000)
+#define MHZ (KHZ * 1000)
+
+#define MASK(w)	((1 << (w)) - 1)
+
 #define GK20A_CLK_GPC_MDIV 1000
 
 #define SYS_GPCPLL_CFG_BASE	0x00137000
+#define GPCPLL_CFG		(SYS_GPCPLL_CFG_BASE + 0)
+#define GPCPLL_CFG_ENABLE	BIT(0)
+#define GPCPLL_CFG_IDDQ		BIT(1)
+#define GPCPLL_CFG_LOCK_DET_OFF	BIT(4)
+#define GPCPLL_CFG_LOCK		BIT(17)
+
+#define GPCPLL_CFG2		(SYS_GPCPLL_CFG_BASE + 0xc)
+#define GPCPLL_CFG2_SETUP2_SHIFT	16
+#define GPCPLL_CFG2_PLL_STEPA_SHIFT	24
+
+#define GPCPLL_CFG3			(SYS_GPCPLL_CFG_BASE + 0x18)
+#define GPCPLL_CFG3_VCO_CTRL_SHIFT		0
+#define GPCPLL_CFG3_VCO_CTRL_WIDTH		9
+#define GPCPLL_CFG3_VCO_CTRL_MASK		\
+	(MASK(GPCPLL_CFG3_VCO_CTRL_WIDTH) << GPCPLL_CFG3_VCO_CTRL_SHIFT)
+#define GPCPLL_CFG3_PLL_STEPB_SHIFT		16
+#define GPCPLL_CFG3_PLL_STEPB_WIDTH		8
+
+#define GPCPLL_COEFF		(SYS_GPCPLL_CFG_BASE + 4)
+#define GPCPLL_COEFF_M_SHIFT	0
+#define GPCPLL_COEFF_M_WIDTH	8
+#define GPCPLL_COEFF_N_SHIFT	8
+#define GPCPLL_COEFF_N_WIDTH	8
+#define GPCPLL_COEFF_N_MASK	\
+	(MASK(GPCPLL_COEFF_N_WIDTH) << GPCPLL_COEFF_N_SHIFT)
+#define GPCPLL_COEFF_P_SHIFT	16
+#define GPCPLL_COEFF_P_WIDTH	6
+
+#define GPCPLL_NDIV_SLOWDOWN			(SYS_GPCPLL_CFG_BASE + 0x1c)
+#define GPCPLL_NDIV_SLOWDOWN_NDIV_LO_SHIFT	0
+#define GPCPLL_NDIV_SLOWDOWN_NDIV_MID_SHIFT	8
+#define GPCPLL_NDIV_SLOWDOWN_STEP_SIZE_LO2MID_SHIFT	16
+#define GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT	22
+#define GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT	31
+
+#define GPC_BCAST_GPCPLL_CFG_BASE		0x00132800
+#define GPC_BCAST_NDIV_SLOWDOWN_DEBUG	(GPC_BCAST_GPCPLL_CFG_BASE + 0xa0)
+#define GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_SHIFT	24
+#define GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_MASK \
+	(0x1 << GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_SHIFT)
+
+#define SEL_VCO				(SYS_GPCPLL_CFG_BASE + 0x100)
+#define SEL_VCO_GPC2CLK_OUT_SHIFT	0
+
+#define GPC2CLK_OUT			(SYS_GPCPLL_CFG_BASE + 0x250)
+#define GPC2CLK_OUT_SDIV14_INDIV4_WIDTH	1
+#define GPC2CLK_OUT_SDIV14_INDIV4_SHIFT	31
+#define GPC2CLK_OUT_SDIV14_INDIV4_MODE	1
+#define GPC2CLK_OUT_VCODIV_WIDTH	6
+#define GPC2CLK_OUT_VCODIV_SHIFT	8
+#define GPC2CLK_OUT_VCODIV1		0
+#define GPC2CLK_OUT_VCODIV2		2
+#define GPC2CLK_OUT_VCODIV_MASK		(MASK(GPC2CLK_OUT_VCODIV_WIDTH) << \
+					GPC2CLK_OUT_VCODIV_SHIFT)
+#define GPC2CLK_OUT_BYPDIV_WIDTH	6
+#define GPC2CLK_OUT_BYPDIV_SHIFT	0
+#define GPC2CLK_OUT_BYPDIV31		0x3c
+#define GPC2CLK_OUT_INIT_MASK	((MASK(GPC2CLK_OUT_SDIV14_INDIV4_WIDTH) << \
+		GPC2CLK_OUT_SDIV14_INDIV4_SHIFT)\
+		| (MASK(GPC2CLK_OUT_VCODIV_WIDTH) << GPC2CLK_OUT_VCODIV_SHIFT)\
+		| (MASK(GPC2CLK_OUT_BYPDIV_WIDTH) << GPC2CLK_OUT_BYPDIV_SHIFT))
+#define GPC2CLK_OUT_INIT_VAL	((GPC2CLK_OUT_SDIV14_INDIV4_MODE << \
+		GPC2CLK_OUT_SDIV14_INDIV4_SHIFT) \
+		| (GPC2CLK_OUT_VCODIV1 << GPC2CLK_OUT_VCODIV_SHIFT) \
+		| (GPC2CLK_OUT_BYPDIV31 << GPC2CLK_OUT_BYPDIV_SHIFT))
 
 /* All frequencies in Khz */
 struct gk20a_clk_pllg_params {
@@ -54,7 +124,29 @@ struct gk20a_clk {
 };
 #define gk20a_clk(p) container_of((p), struct gk20a_clk, base)
 
-int _gk20a_clk_ctor(struct nvkm_device *, int, const struct nvkm_clk_func *,
+u32 gk20a_pllg_calc_rate(struct gk20a_clk *, struct gk20a_pll *);
+int gk20a_pllg_calc_mnp(struct gk20a_clk *, unsigned long, struct gk20a_pll *);
+void gk20a_pllg_read_mnp(struct gk20a_clk *, struct gk20a_pll *);
+void gk20a_pllg_write_mnp(struct gk20a_clk *, const struct gk20a_pll *);
+
+static inline bool
+gk20a_pllg_is_enabled(struct gk20a_clk *clk)
+{
+	struct nvkm_device *device = clk->base.subdev.device;
+	u32 val;
+
+	val = nvkm_rd32(device, GPCPLL_CFG);
+	return val & GPCPLL_CFG_ENABLE;
+}
+
+static inline u32
+gk20a_pllg_n_lo(struct gk20a_clk *clk, struct gk20a_pll *pll)
+{
+	return DIV_ROUND_UP(pll->m * clk->params->min_vco,
+			    clk->parent_rate / KHZ);
+}
+
+int gk20a_clk_ctor(struct nvkm_device *, int, const struct nvkm_clk_func *,
 		    const struct gk20a_clk_pllg_params *, struct gk20a_clk *);
 void gk20a_clk_fini(struct nvkm_clk *);
 int gk20a_clk_read(struct nvkm_clk *, enum nv_clk_src);
@@ -62,4 +154,6 @@ int gk20a_clk_calc(struct nvkm_clk *, struct nvkm_cstate *);
 int gk20a_clk_prog(struct nvkm_clk *);
 void gk20a_clk_tidy(struct nvkm_clk *);
 
+int gk20a_clk_setup_slide(struct gk20a_clk *);
+
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/clk/gm20b.c b/drivers/gpu/drm/nouveau/nvkm/subdev/clk/gm20b.c
index 71b2bbb61973..b284e949f732 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/clk/gm20b.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/clk/gm20b.c
@@ -21,20 +21,123 @@
  */
 
 #include <subdev/clk.h>
+#include <subdev/volt.h>
+#include <subdev/timer.h>
 #include <core/device.h>
+#include <core/tegra.h>
 
 #include "priv.h"
 #include "gk20a.h"
 
-#define KHZ (1000)
-#define MHZ (KHZ * 1000)
-
-#define MASK(w)	((1 << w) - 1)
+#define GPCPLL_CFG_SYNC_MODE	BIT(2)
 
 #define BYPASSCTRL_SYS	(SYS_GPCPLL_CFG_BASE + 0x340)
 #define BYPASSCTRL_SYS_GPCPLL_SHIFT	0
 #define BYPASSCTRL_SYS_GPCPLL_WIDTH	1
 
+#define GPCPLL_CFG2_SDM_DIN_SHIFT	0
+#define GPCPLL_CFG2_SDM_DIN_WIDTH	8
+#define GPCPLL_CFG2_SDM_DIN_MASK	\
+	(MASK(GPCPLL_CFG2_SDM_DIN_WIDTH) << GPCPLL_CFG2_SDM_DIN_SHIFT)
+#define GPCPLL_CFG2_SDM_DIN_NEW_SHIFT	8
+#define GPCPLL_CFG2_SDM_DIN_NEW_WIDTH	15
+#define GPCPLL_CFG2_SDM_DIN_NEW_MASK	\
+	(MASK(GPCPLL_CFG2_SDM_DIN_NEW_WIDTH) << GPCPLL_CFG2_SDM_DIN_NEW_SHIFT)
+#define GPCPLL_CFG2_SETUP2_SHIFT	16
+#define GPCPLL_CFG2_PLL_STEPA_SHIFT	24
+
+#define GPCPLL_DVFS0	(SYS_GPCPLL_CFG_BASE + 0x10)
+#define GPCPLL_DVFS0_DFS_COEFF_SHIFT	0
+#define GPCPLL_DVFS0_DFS_COEFF_WIDTH	7
+#define GPCPLL_DVFS0_DFS_COEFF_MASK	\
+	(MASK(GPCPLL_DVFS0_DFS_COEFF_WIDTH) << GPCPLL_DVFS0_DFS_COEFF_SHIFT)
+#define GPCPLL_DVFS0_DFS_DET_MAX_SHIFT	8
+#define GPCPLL_DVFS0_DFS_DET_MAX_WIDTH	7
+#define GPCPLL_DVFS0_DFS_DET_MAX_MASK	\
+	(MASK(GPCPLL_DVFS0_DFS_DET_MAX_WIDTH) << GPCPLL_DVFS0_DFS_DET_MAX_SHIFT)
+
+#define GPCPLL_DVFS1		(SYS_GPCPLL_CFG_BASE + 0x14)
+#define GPCPLL_DVFS1_DFS_EXT_DET_SHIFT		0
+#define GPCPLL_DVFS1_DFS_EXT_DET_WIDTH		7
+#define GPCPLL_DVFS1_DFS_EXT_STRB_SHIFT		7
+#define GPCPLL_DVFS1_DFS_EXT_STRB_WIDTH		1
+#define GPCPLL_DVFS1_DFS_EXT_CAL_SHIFT		8
+#define GPCPLL_DVFS1_DFS_EXT_CAL_WIDTH		7
+#define GPCPLL_DVFS1_DFS_EXT_SEL_SHIFT		15
+#define GPCPLL_DVFS1_DFS_EXT_SEL_WIDTH		1
+#define GPCPLL_DVFS1_DFS_CTRL_SHIFT		16
+#define GPCPLL_DVFS1_DFS_CTRL_WIDTH		12
+#define GPCPLL_DVFS1_EN_SDM_SHIFT		28
+#define GPCPLL_DVFS1_EN_SDM_WIDTH		1
+#define GPCPLL_DVFS1_EN_SDM_BIT			BIT(28)
+#define GPCPLL_DVFS1_EN_DFS_SHIFT		29
+#define GPCPLL_DVFS1_EN_DFS_WIDTH		1
+#define GPCPLL_DVFS1_EN_DFS_BIT			BIT(29)
+#define GPCPLL_DVFS1_EN_DFS_CAL_SHIFT		30
+#define GPCPLL_DVFS1_EN_DFS_CAL_WIDTH		1
+#define GPCPLL_DVFS1_EN_DFS_CAL_BIT		BIT(30)
+#define GPCPLL_DVFS1_DFS_CAL_DONE_SHIFT		31
+#define GPCPLL_DVFS1_DFS_CAL_DONE_WIDTH		1
+#define GPCPLL_DVFS1_DFS_CAL_DONE_BIT		BIT(31)
+
+#define GPC_BCAST_GPCPLL_DVFS2	(GPC_BCAST_GPCPLL_CFG_BASE + 0x20)
+#define GPC_BCAST_GPCPLL_DVFS2_DFS_EXT_STROBE_BIT	BIT(16)
+
+#define GPCPLL_CFG3_PLL_DFS_TESTOUT_SHIFT	24
+#define GPCPLL_CFG3_PLL_DFS_TESTOUT_WIDTH	7
+
+#define DFS_DET_RANGE	6	/* -2^6 ... 2^6-1 */
+#define SDM_DIN_RANGE	12	/* -2^12 ... 2^12-1 */
+
+struct gm20b_clk_dvfs_params {
+	s32 coeff_slope;
+	s32 coeff_offs;
+	u32 vco_ctrl;
+};
+
+static const struct gm20b_clk_dvfs_params gm20b_dvfs_params = {
+	.coeff_slope = -165230,
+	.coeff_offs = 214007,
+	.vco_ctrl = 0x7 << 3,
+};
+
+/*
+ * base.n is now the *integer* part of the N factor.
+ * sdm_din contains n's decimal part.
+ */
+struct gm20b_pll {
+	struct gk20a_pll base;
+	u32 sdm_din;
+};
+
+struct gm20b_clk_dvfs {
+	u32 dfs_coeff;
+	s32 dfs_det_max;
+	s32 dfs_ext_cal;
+};
+
+struct gm20b_clk {
+	/* currently applied parameters */
+	struct gk20a_clk base;
+	struct gm20b_clk_dvfs dvfs;
+	u32 uv;
+
+	/* new parameters to apply */
+	struct gk20a_pll new_pll;
+	struct gm20b_clk_dvfs new_dvfs;
+	u32 new_uv;
+
+	const struct gm20b_clk_dvfs_params *dvfs_params;
+
+	/* fused parameters */
+	s32 uvdet_slope;
+	s32 uvdet_offs;
+
+	/* safe frequency we can use at minimum voltage */
+	u32 safe_fmax_vmin;
+};
+#define gm20b_clk(p) container_of((gk20a_clk(p)), struct gm20b_clk, base)
+
 static u32 pl_to_div(u32 pl)
 {
 	return pl;
@@ -53,6 +156,484 @@ static const struct gk20a_clk_pllg_params gm20b_pllg_params = {
 	.min_pl = 1, .max_pl = 31,
 };
 
+static void
+gm20b_pllg_read_mnp(struct gm20b_clk *clk, struct gm20b_pll *pll)
+{
+	struct nvkm_subdev *subdev = &clk->base.base.subdev;
+	struct nvkm_device *device = subdev->device;
+	u32 val;
+
+	gk20a_pllg_read_mnp(&clk->base, &pll->base);
+	val = nvkm_rd32(device, GPCPLL_CFG2);
+	pll->sdm_din = (val >> GPCPLL_CFG2_SDM_DIN_SHIFT) &
+		       MASK(GPCPLL_CFG2_SDM_DIN_WIDTH);
+}
+
+static void
+gm20b_pllg_write_mnp(struct gm20b_clk *clk, const struct gm20b_pll *pll)
+{
+	struct nvkm_device *device = clk->base.base.subdev.device;
+
+	nvkm_mask(device, GPCPLL_CFG2, GPCPLL_CFG2_SDM_DIN_MASK,
+		  pll->sdm_din << GPCPLL_CFG2_SDM_DIN_SHIFT);
+	gk20a_pllg_write_mnp(&clk->base, &pll->base);
+}
+
+/*
+ * Determine DFS_COEFF for the requested voltage. Always select external
+ * calibration override equal to the voltage, and set maximum detection
+ * limit "0" (to make sure that PLL output remains under F/V curve when
+ * voltage increases).
+ */
+static void
+gm20b_dvfs_calc_det_coeff(struct gm20b_clk *clk, s32 uv,
+			  struct gm20b_clk_dvfs *dvfs)
+{
+	struct nvkm_subdev *subdev = &clk->base.base.subdev;
+	const struct gm20b_clk_dvfs_params *p = clk->dvfs_params;
+	u32 coeff;
+	/* Work with mv as uv would likely trigger an overflow */
+	s32 mv = DIV_ROUND_CLOSEST(uv, 1000);
+
+	/* coeff = slope * voltage + offset */
+	coeff = DIV_ROUND_CLOSEST(mv * p->coeff_slope, 1000) + p->coeff_offs;
+	coeff = DIV_ROUND_CLOSEST(coeff, 1000);
+	dvfs->dfs_coeff = min_t(u32, coeff, MASK(GPCPLL_DVFS0_DFS_COEFF_WIDTH));
+
+	dvfs->dfs_ext_cal = DIV_ROUND_CLOSEST(uv - clk->uvdet_offs,
+					     clk->uvdet_slope);
+	/* should never happen */
+	if (abs(dvfs->dfs_ext_cal) >= BIT(DFS_DET_RANGE))
+		nvkm_error(subdev, "dfs_ext_cal overflow!\n");
+
+	dvfs->dfs_det_max = 0;
+
+	nvkm_debug(subdev, "%s uv: %d coeff: %x, ext_cal: %d, det_max: %d\n",
+		   __func__, uv, dvfs->dfs_coeff, dvfs->dfs_ext_cal,
+		   dvfs->dfs_det_max);
+}
+
+/*
+ * Solve equation for integer and fractional part of the effective NDIV:
+ *
+ * n_eff = n_int + 1/2 + (SDM_DIN / 2^(SDM_DIN_RANGE + 1)) +
+ *         (DVFS_COEFF * DVFS_DET_DELTA) / 2^DFS_DET_RANGE
+ *
+ * The SDM_DIN LSB is finally shifted out, since it is not accessible by sw.
+ */
+static void
+gm20b_dvfs_calc_ndiv(struct gm20b_clk *clk, u32 n_eff, u32 *n_int, u32 *sdm_din)
+{
+	struct nvkm_subdev *subdev = &clk->base.base.subdev;
+	const struct gk20a_clk_pllg_params *p = clk->base.params;
+	u32 n;
+	s32 det_delta;
+	u32 rem, rem_range;
+
+	/* calculate current ext_cal and subtract previous one */
+	det_delta = DIV_ROUND_CLOSEST(((s32)clk->uv) - clk->uvdet_offs,
+				      clk->uvdet_slope);
+	det_delta -= clk->dvfs.dfs_ext_cal;
+	det_delta = min(det_delta, clk->dvfs.dfs_det_max);
+	det_delta *= clk->dvfs.dfs_coeff;
+
+	/* integer part of n */
+	n = (n_eff << DFS_DET_RANGE) - det_delta;
+	/* should never happen! */
+	if (n <= 0) {
+		nvkm_error(subdev, "ndiv <= 0 - setting to 1...\n");
+		n = 1 << DFS_DET_RANGE;
+	}
+	if (n >> DFS_DET_RANGE > p->max_n) {
+		nvkm_error(subdev, "ndiv > max_n - setting to max_n...\n");
+		n = p->max_n << DFS_DET_RANGE;
+	}
+	*n_int = n >> DFS_DET_RANGE;
+
+	/* fractional part of n */
+	rem = ((u32)n) & MASK(DFS_DET_RANGE);
+	rem_range = SDM_DIN_RANGE + 1 - DFS_DET_RANGE;
+	/* subtract 2^SDM_DIN_RANGE to account for the 1/2 of the equation */
+	rem = (rem << rem_range) - BIT(SDM_DIN_RANGE);
+	/* lose 8 LSB and clip - sdm_din only keeps the most significant byte */
+	*sdm_din = (rem >> BITS_PER_BYTE) & MASK(GPCPLL_CFG2_SDM_DIN_WIDTH);
+
+	nvkm_debug(subdev, "%s n_eff: %d, n_int: %d, sdm_din: %d\n", __func__,
+		   n_eff, *n_int, *sdm_din);
+}
+
+static int
+gm20b_pllg_slide(struct gm20b_clk *clk, u32 n)
+{
+	struct nvkm_subdev *subdev = &clk->base.base.subdev;
+	struct nvkm_device *device = subdev->device;
+	struct gm20b_pll pll;
+	u32 n_int, sdm_din;
+	int ret = 0;
+
+	/* calculate the new n_int/sdm_din for this n/uv */
+	gm20b_dvfs_calc_ndiv(clk, n, &n_int, &sdm_din);
+
+	/* get old coefficients */
+	gm20b_pllg_read_mnp(clk, &pll);
+	/* do nothing if NDIV is the same */
+	if (n_int == pll.base.n && sdm_din == pll.sdm_din)
+		return 0;
+
+	/* pll slowdown mode */
+	nvkm_mask(device, GPCPLL_NDIV_SLOWDOWN,
+		BIT(GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT),
+		BIT(GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT));
+
+	/* new ndiv ready for ramp */
+	/* in DVFS mode SDM is updated via "new" field */
+	nvkm_mask(device, GPCPLL_CFG2, GPCPLL_CFG2_SDM_DIN_NEW_MASK,
+		  sdm_din << GPCPLL_CFG2_SDM_DIN_NEW_SHIFT);
+	pll.base.n = n_int;
+	udelay(1);
+	gk20a_pllg_write_mnp(&clk->base, &pll.base);
+
+	/* dynamic ramp to new ndiv */
+	udelay(1);
+	nvkm_mask(device, GPCPLL_NDIV_SLOWDOWN,
+		  BIT(GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT),
+		  BIT(GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT));
+
+	/* wait for ramping to complete */
+	if (nvkm_wait_usec(device, 500, GPC_BCAST_NDIV_SLOWDOWN_DEBUG,
+		GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_MASK,
+		GPC_BCAST_NDIV_SLOWDOWN_DEBUG_PLL_DYNRAMP_DONE_SYNCED_MASK) < 0)
+		ret = -ETIMEDOUT;
+
+	/* in DVFS mode complete SDM update */
+	nvkm_mask(device, GPCPLL_CFG2, GPCPLL_CFG2_SDM_DIN_MASK,
+		  sdm_din << GPCPLL_CFG2_SDM_DIN_SHIFT);
+
+	/* exit slowdown mode */
+	nvkm_mask(device, GPCPLL_NDIV_SLOWDOWN,
+		BIT(GPCPLL_NDIV_SLOWDOWN_SLOWDOWN_USING_PLL_SHIFT) |
+		BIT(GPCPLL_NDIV_SLOWDOWN_EN_DYNRAMP_SHIFT), 0);
+	nvkm_rd32(device, GPCPLL_NDIV_SLOWDOWN);
+
+	return ret;
+}
+
+static int
+gm20b_pllg_enable(struct gm20b_clk *clk)
+{
+	struct nvkm_device *device = clk->base.base.subdev.device;
+
+	nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_ENABLE, GPCPLL_CFG_ENABLE);
+	nvkm_rd32(device, GPCPLL_CFG);
+
+	/* In DVFS mode lock cannot be used - so just delay */
+	udelay(40);
+
+	/* set SYNC_MODE for glitchless switch out of bypass */
+	nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_SYNC_MODE,
+		       GPCPLL_CFG_SYNC_MODE);
+	nvkm_rd32(device, GPCPLL_CFG);
+
+	/* switch to VCO mode */
+	nvkm_mask(device, SEL_VCO, BIT(SEL_VCO_GPC2CLK_OUT_SHIFT),
+		  BIT(SEL_VCO_GPC2CLK_OUT_SHIFT));
+
+	return 0;
+}
+
+static void
+gm20b_pllg_disable(struct gm20b_clk *clk)
+{
+	struct nvkm_device *device = clk->base.base.subdev.device;
+
+	/* put PLL in bypass before disabling it */
+	nvkm_mask(device, SEL_VCO, BIT(SEL_VCO_GPC2CLK_OUT_SHIFT), 0);
+
+	/* clear SYNC_MODE before disabling PLL */
+	nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_SYNC_MODE, 0);
+
+	nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_ENABLE, 0);
+	nvkm_rd32(device, GPCPLL_CFG);
+}
+
+static int
+gm20b_pllg_program_mnp(struct gm20b_clk *clk, const struct gk20a_pll *pll)
+{
+	struct nvkm_subdev *subdev = &clk->base.base.subdev;
+	struct nvkm_device *device = subdev->device;
+	struct gm20b_pll cur_pll;
+	u32 n_int, sdm_din;
+	/* if we only change pdiv, we can do a glitchless transition */
+	bool pdiv_only;
+	int ret;
+
+	gm20b_dvfs_calc_ndiv(clk, pll->n, &n_int, &sdm_din);
+	gm20b_pllg_read_mnp(clk, &cur_pll);
+	pdiv_only = cur_pll.base.n == n_int && cur_pll.sdm_din == sdm_din &&
+		    cur_pll.base.m == pll->m;
+
+	/* need full sequence if clock not enabled yet */
+	if (!gk20a_pllg_is_enabled(&clk->base))
+		pdiv_only = false;
+
+	/* split VCO-to-bypass jump in half by setting out divider 1:2 */
+	nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_VCODIV_MASK,
+		  GPC2CLK_OUT_VCODIV2 << GPC2CLK_OUT_VCODIV_SHIFT);
+	/* Intentional 2nd write to assure linear divider operation */
+	nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_VCODIV_MASK,
+		  GPC2CLK_OUT_VCODIV2 << GPC2CLK_OUT_VCODIV_SHIFT);
+	nvkm_rd32(device, GPC2CLK_OUT);
+	udelay(2);
+
+	if (pdiv_only) {
+		u32 old = cur_pll.base.pl;
+		u32 new = pll->pl;
+
+		/*
+		 * we can do a glitchless transition only if the old and new PL
+		 * parameters share at least one bit set to 1. If this is not
+		 * the case, calculate and program an interim PL that will allow
+		 * us to respect that rule.
+		 */
+		if ((old & new) == 0) {
+			cur_pll.base.pl = min(old | BIT(ffs(new) - 1),
+					      new | BIT(ffs(old) - 1));
+			gk20a_pllg_write_mnp(&clk->base, &cur_pll.base);
+		}
+
+		cur_pll.base.pl = new;
+		gk20a_pllg_write_mnp(&clk->base, &cur_pll.base);
+	} else {
+		/* disable before programming if more than pdiv changes */
+		gm20b_pllg_disable(clk);
+
+		cur_pll.base = *pll;
+		cur_pll.base.n = n_int;
+		cur_pll.sdm_din = sdm_din;
+		gm20b_pllg_write_mnp(clk, &cur_pll);
+
+		ret = gm20b_pllg_enable(clk);
+		if (ret)
+			return ret;
+	}
+
+	/* restore out divider 1:1 */
+	udelay(2);
+	nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_VCODIV_MASK,
+		  GPC2CLK_OUT_VCODIV1 << GPC2CLK_OUT_VCODIV_SHIFT);
+	/* Intentional 2nd write to assure linear divider operation */
+	nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_VCODIV_MASK,
+		  GPC2CLK_OUT_VCODIV1 << GPC2CLK_OUT_VCODIV_SHIFT);
+	nvkm_rd32(device, GPC2CLK_OUT);
+
+	return 0;
+}
+
+static int
+gm20b_pllg_program_mnp_slide(struct gm20b_clk *clk, const struct gk20a_pll *pll)
+{
+	struct gk20a_pll cur_pll;
+	int ret;
+
+	if (gk20a_pllg_is_enabled(&clk->base)) {
+		gk20a_pllg_read_mnp(&clk->base, &cur_pll);
+
+		/* just do NDIV slide if there is no change to M and PL */
+		if (pll->m == cur_pll.m && pll->pl == cur_pll.pl)
+			return gm20b_pllg_slide(clk, pll->n);
+
+		/* slide down to current NDIV_LO */
+		cur_pll.n = gk20a_pllg_n_lo(&clk->base, &cur_pll);
+		ret = gm20b_pllg_slide(clk, cur_pll.n);
+		if (ret)
+			return ret;
+	}
+
+	/* program MNP with the new clock parameters and new NDIV_LO */
+	cur_pll = *pll;
+	cur_pll.n = gk20a_pllg_n_lo(&clk->base, &cur_pll);
+	ret = gm20b_pllg_program_mnp(clk, &cur_pll);
+	if (ret)
+		return ret;
+
+	/* slide up to new NDIV */
+	return gm20b_pllg_slide(clk, pll->n);
+}
+
+static int
+gm20b_clk_calc(struct nvkm_clk *base, struct nvkm_cstate *cstate)
+{
+	struct gm20b_clk *clk = gm20b_clk(base);
+	struct nvkm_subdev *subdev = &base->subdev;
+	struct nvkm_volt *volt = base->subdev.device->volt;
+	int ret;
+
+	ret = gk20a_pllg_calc_mnp(&clk->base, cstate->domain[nv_clk_src_gpc] *
+					     GK20A_CLK_GPC_MDIV, &clk->new_pll);
+	if (ret)
+		return ret;
+
+	clk->new_uv = volt->vid[cstate->voltage].uv;
+	gm20b_dvfs_calc_det_coeff(clk, clk->new_uv, &clk->new_dvfs);
+
+	nvkm_debug(subdev, "%s uv: %d uv\n", __func__, clk->new_uv);
+
+	return 0;
+}
+
+/*
+ * Compute PLL parameters that are always safe for the current voltage
+ */
+static void
+gm20b_dvfs_calc_safe_pll(struct gm20b_clk *clk, struct gk20a_pll *pll)
+{
+	u32 rate = gk20a_pllg_calc_rate(&clk->base, pll) / KHZ;
+	u32 parent_rate = clk->base.parent_rate / KHZ;
+	u32 nmin, nsafe;
+
+	/* remove a safe margin of 10% */
+	if (rate > clk->safe_fmax_vmin)
+		rate = rate * (100 - 10) / 100;
+
+	/* gpc2clk */
+	rate *= 2;
+
+	nmin = DIV_ROUND_UP(pll->m * clk->base.params->min_vco, parent_rate);
+	nsafe = pll->m * rate / (clk->base.parent_rate);
+
+	if (nsafe < nmin) {
+		pll->pl = DIV_ROUND_UP(nmin * parent_rate, pll->m * rate);
+		nsafe = nmin;
+	}
+
+	pll->n = nsafe;
+}
+
+static void
+gm20b_dvfs_program_coeff(struct gm20b_clk *clk, u32 coeff)
+{
+	struct nvkm_device *device = clk->base.base.subdev.device;
+
+	/* strobe to read external DFS coefficient */
+	nvkm_mask(device, GPC_BCAST_GPCPLL_DVFS2,
+		  GPC_BCAST_GPCPLL_DVFS2_DFS_EXT_STROBE_BIT,
+		  GPC_BCAST_GPCPLL_DVFS2_DFS_EXT_STROBE_BIT);
+
+	nvkm_mask(device, GPCPLL_DVFS0, GPCPLL_DVFS0_DFS_COEFF_MASK,
+		  coeff << GPCPLL_DVFS0_DFS_COEFF_SHIFT);
+
+	udelay(1);
+	nvkm_mask(device, GPC_BCAST_GPCPLL_DVFS2,
+		  GPC_BCAST_GPCPLL_DVFS2_DFS_EXT_STROBE_BIT, 0);
+}
+
+static void
+gm20b_dvfs_program_ext_cal(struct gm20b_clk *clk, u32 dfs_det_cal)
+{
+	struct nvkm_device *device = clk->base.base.subdev.device;
+	u32 val;
+
+	nvkm_mask(device, GPC_BCAST_GPCPLL_DVFS2, MASK(DFS_DET_RANGE + 1),
+		  dfs_det_cal);
+	udelay(1);
+
+	val = nvkm_rd32(device, GPCPLL_DVFS1);
+	if (!(val & BIT(25))) {
+		/* Use external value to overwrite calibration value */
+		val |= BIT(25) | BIT(16);
+		nvkm_wr32(device, GPCPLL_DVFS1, val);
+	}
+}
+
+static void
+gm20b_dvfs_program_dfs_detection(struct gm20b_clk *clk,
+				 struct gm20b_clk_dvfs *dvfs)
+{
+	struct nvkm_device *device = clk->base.base.subdev.device;
+
+	/* strobe to read external DFS coefficient */
+	nvkm_mask(device, GPC_BCAST_GPCPLL_DVFS2,
+		  GPC_BCAST_GPCPLL_DVFS2_DFS_EXT_STROBE_BIT,
+		  GPC_BCAST_GPCPLL_DVFS2_DFS_EXT_STROBE_BIT);
+
+	nvkm_mask(device, GPCPLL_DVFS0,
+		  GPCPLL_DVFS0_DFS_COEFF_MASK | GPCPLL_DVFS0_DFS_DET_MAX_MASK,
+		  dvfs->dfs_coeff << GPCPLL_DVFS0_DFS_COEFF_SHIFT |
+		  dvfs->dfs_det_max << GPCPLL_DVFS0_DFS_DET_MAX_SHIFT);
+
+	udelay(1);
+	nvkm_mask(device, GPC_BCAST_GPCPLL_DVFS2,
+		  GPC_BCAST_GPCPLL_DVFS2_DFS_EXT_STROBE_BIT, 0);
+
+	gm20b_dvfs_program_ext_cal(clk, dvfs->dfs_ext_cal);
+}
+
+static int
+gm20b_clk_prog(struct nvkm_clk *base)
+{
+	struct gm20b_clk *clk = gm20b_clk(base);
+	u32 cur_freq;
+	int ret;
+
+	/* No change in DVFS settings? */
+	if (clk->uv == clk->new_uv)
+		goto prog;
+
+	/*
+	 * Interim step for changing DVFS detection settings: low enough
+	 * frequency to be safe at at DVFS coeff = 0.
+	 *
+	 * 1. If voltage is increasing:
+	 * - safe frequency target matches the lowest - old - frequency
+	 * - DVFS settings are still old
+	 * - Voltage already increased to new level by volt, but maximum
+	 *   detection limit assures PLL output remains under F/V curve
+	 *
+	 * 2. If voltage is decreasing:
+	 * - safe frequency target matches the lowest - new - frequency
+	 * - DVFS settings are still old
+	 * - Voltage is also old, it will be lowered by volt afterwards
+	 *
+	 * Interim step can be skipped if old frequency is below safe minimum,
+	 * i.e., it is low enough to be safe at any voltage in operating range
+	 * with zero DVFS coefficient.
+	 */
+	cur_freq = nvkm_clk_read(&clk->base.base, nv_clk_src_gpc);
+	if (cur_freq > clk->safe_fmax_vmin) {
+		struct gk20a_pll pll_safe;
+
+		if (clk->uv < clk->new_uv)
+			/* voltage will raise: safe frequency is current one */
+			pll_safe = clk->base.pll;
+		else
+			/* voltage will drop: safe frequency is new one */
+			pll_safe = clk->new_pll;
+
+		gm20b_dvfs_calc_safe_pll(clk, &pll_safe);
+		ret = gm20b_pllg_program_mnp_slide(clk, &pll_safe);
+		if (ret)
+			return ret;
+	}
+
+	/*
+	 * DVFS detection settings transition:
+	 * - Set DVFS coefficient zero
+	 * - Set calibration level to new voltage
+	 * - Set DVFS coefficient to match new voltage
+	 */
+	gm20b_dvfs_program_coeff(clk, 0);
+	gm20b_dvfs_program_ext_cal(clk, clk->new_dvfs.dfs_ext_cal);
+	gm20b_dvfs_program_coeff(clk, clk->new_dvfs.dfs_coeff);
+	gm20b_dvfs_program_dfs_detection(clk, &clk->new_dvfs);
+
+prog:
+	clk->uv = clk->new_uv;
+	clk->dvfs = clk->new_dvfs;
+	clk->base.pll = clk->new_pll;
+
+	return gm20b_pllg_program_mnp_slide(clk, &clk->base.pll);
+}
+
 static struct nvkm_pstate
 gm20b_pstates[] = {
 	{
@@ -133,9 +714,99 @@ gm20b_pstates[] = {
 			.voltage = 12,
 		},
 	},
-
 };
 
+static void
+gm20b_clk_fini(struct nvkm_clk *base)
+{
+	struct nvkm_device *device = base->subdev.device;
+	struct gm20b_clk *clk = gm20b_clk(base);
+
+	/* slide to VCO min */
+	if (gk20a_pllg_is_enabled(&clk->base)) {
+		struct gk20a_pll pll;
+		u32 n_lo;
+
+		gk20a_pllg_read_mnp(&clk->base, &pll);
+		n_lo = gk20a_pllg_n_lo(&clk->base, &pll);
+		gm20b_pllg_slide(clk, n_lo);
+	}
+
+	gm20b_pllg_disable(clk);
+
+	/* set IDDQ */
+	nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_IDDQ, 1);
+}
+
+static int
+gm20b_clk_init_dvfs(struct gm20b_clk *clk)
+{
+	struct nvkm_subdev *subdev = &clk->base.base.subdev;
+	struct nvkm_device *device = subdev->device;
+	bool fused = clk->uvdet_offs && clk->uvdet_slope;
+	static const s32 ADC_SLOPE_UV = 10000; /* default ADC detection slope */
+	u32 data;
+	int ret;
+
+	/* Enable NA DVFS */
+	nvkm_mask(device, GPCPLL_DVFS1, GPCPLL_DVFS1_EN_DFS_BIT,
+		  GPCPLL_DVFS1_EN_DFS_BIT);
+
+	/* Set VCO_CTRL */
+	if (clk->dvfs_params->vco_ctrl)
+		nvkm_mask(device, GPCPLL_CFG3, GPCPLL_CFG3_VCO_CTRL_MASK,
+		      clk->dvfs_params->vco_ctrl << GPCPLL_CFG3_VCO_CTRL_SHIFT);
+
+	if (fused) {
+		/* Start internal calibration, but ignore results */
+		nvkm_mask(device, GPCPLL_DVFS1, GPCPLL_DVFS1_EN_DFS_CAL_BIT,
+			  GPCPLL_DVFS1_EN_DFS_CAL_BIT);
+
+		/* got uvdev parameters from fuse, skip calibration */
+		goto calibrated;
+	}
+
+	/*
+	 * If calibration parameters are not fused, start internal calibration,
+	 * wait for completion, and use results along with default slope to
+	 * calculate ADC offset during boot.
+	 */
+	nvkm_mask(device, GPCPLL_DVFS1, GPCPLL_DVFS1_EN_DFS_CAL_BIT,
+			  GPCPLL_DVFS1_EN_DFS_CAL_BIT);
+
+	/* Wait for internal calibration done (spec < 2us). */
+	ret = nvkm_wait_usec(device, 10, GPCPLL_DVFS1,
+			     GPCPLL_DVFS1_DFS_CAL_DONE_BIT,
+			     GPCPLL_DVFS1_DFS_CAL_DONE_BIT);
+	if (ret < 0) {
+		nvkm_error(subdev, "GPCPLL calibration timeout\n");
+		return -ETIMEDOUT;
+	}
+
+	data = nvkm_rd32(device, GPCPLL_CFG3) >>
+			 GPCPLL_CFG3_PLL_DFS_TESTOUT_SHIFT;
+	data &= MASK(GPCPLL_CFG3_PLL_DFS_TESTOUT_WIDTH);
+
+	clk->uvdet_slope = ADC_SLOPE_UV;
+	clk->uvdet_offs = ((s32)clk->uv) - data * ADC_SLOPE_UV;
+
+	nvkm_debug(subdev, "calibrated DVFS parameters: offs %d, slope %d\n",
+		   clk->uvdet_offs, clk->uvdet_slope);
+
+calibrated:
+	/* Compute and apply initial DVFS parameters */
+	gm20b_dvfs_calc_det_coeff(clk, clk->uv, &clk->dvfs);
+	gm20b_dvfs_program_coeff(clk, 0);
+	gm20b_dvfs_program_ext_cal(clk, clk->dvfs.dfs_ext_cal);
+	gm20b_dvfs_program_coeff(clk, clk->dvfs.dfs_coeff);
+	gm20b_dvfs_program_dfs_detection(clk, &clk->new_dvfs);
+
+	return 0;
+}
+
+/* Forward declaration to detect speedo >=1 in gm20b_clk_init() */
+static const struct nvkm_clk_func gm20b_clk;
+
 static int
 gm20b_clk_init(struct nvkm_clk *base)
 {
@@ -143,15 +814,56 @@ gm20b_clk_init(struct nvkm_clk *base)
 	struct nvkm_subdev *subdev = &clk->base.subdev;
 	struct nvkm_device *device = subdev->device;
 	int ret;
+	u32 data;
+
+	/* get out from IDDQ */
+	nvkm_mask(device, GPCPLL_CFG, GPCPLL_CFG_IDDQ, 0);
+	nvkm_rd32(device, GPCPLL_CFG);
+	udelay(5);
+
+	nvkm_mask(device, GPC2CLK_OUT, GPC2CLK_OUT_INIT_MASK,
+		  GPC2CLK_OUT_INIT_VAL);
 
 	/* Set the global bypass control to VCO */
 	nvkm_mask(device, BYPASSCTRL_SYS,
 	       MASK(BYPASSCTRL_SYS_GPCPLL_WIDTH) << BYPASSCTRL_SYS_GPCPLL_SHIFT,
 	       0);
 
+	ret = gk20a_clk_setup_slide(clk);
+	if (ret)
+		return ret;
+
+	/* If not fused, set RAM SVOP PDP data 0x2, and enable fuse override */
+	data = nvkm_rd32(device, 0x021944);
+	if (!(data & 0x3)) {
+		data |= 0x2;
+		nvkm_wr32(device, 0x021944, data);
+
+		data = nvkm_rd32(device, 0x021948);
+		data |=  0x1;
+		nvkm_wr32(device, 0x021948, data);
+	}
+
+	/* Disable idle slow down  */
+	nvkm_mask(device, 0x20160, 0x003f0000, 0x0);
+
+	/* speedo >= 1? */
+	if (clk->base.func == &gm20b_clk) {
+		struct gm20b_clk *_clk = gm20b_clk(base);
+		struct nvkm_volt *volt = device->volt;
+
+		/* Get current voltage */
+		_clk->uv = nvkm_volt_get(volt);
+
+		/* Initialize DVFS */
+		ret = gm20b_clk_init_dvfs(_clk);
+		if (ret)
+			return ret;
+	}
+
 	/* Start with lowest frequency */
 	base->func->calc(base, &base->func->pstates[0].base);
-	ret = base->func->prog(&clk->base);
+	ret = base->func->prog(base);
 	if (ret) {
 		nvkm_error(subdev, "cannot initialize clock\n");
 		return ret;
@@ -169,6 +881,7 @@ gm20b_clk_speedo0 = {
 	.prog = gk20a_clk_prog,
 	.tidy = gk20a_clk_tidy,
 	.pstates = gm20b_pstates,
+	/* Speedo 0 only supports 12 voltages */
 	.nr_pstates = ARRAY_SIZE(gm20b_pstates) - 1,
 	.domains = {
 		{ nv_clk_src_crystal, 0xff },
@@ -177,8 +890,26 @@ gm20b_clk_speedo0 = {
 	},
 };
 
-int
-gm20b_clk_new(struct nvkm_device *device, int index, struct nvkm_clk **pclk)
+static const struct nvkm_clk_func
+gm20b_clk = {
+	.init = gm20b_clk_init,
+	.fini = gm20b_clk_fini,
+	.read = gk20a_clk_read,
+	.calc = gm20b_clk_calc,
+	.prog = gm20b_clk_prog,
+	.tidy = gk20a_clk_tidy,
+	.pstates = gm20b_pstates,
+	.nr_pstates = ARRAY_SIZE(gm20b_pstates),
+	.domains = {
+		{ nv_clk_src_crystal, 0xff },
+		{ nv_clk_src_gpc, 0xff, 0, "core", GK20A_CLK_GPC_MDIV },
+		{ nv_clk_src_max },
+	},
+};
+
+static int
+gm20b_clk_new_speedo0(struct nvkm_device *device, int index,
+		      struct nvkm_clk **pclk)
 {
 	struct gk20a_clk *clk;
 	int ret;
@@ -188,11 +919,156 @@ gm20b_clk_new(struct nvkm_device *device, int index, struct nvkm_clk **pclk)
 		return -ENOMEM;
 	*pclk = &clk->base;
 
-	ret = _gk20a_clk_ctor(device, index, &gm20b_clk_speedo0,
-			      &gm20b_pllg_params, clk);
+	ret = gk20a_clk_ctor(device, index, &gm20b_clk_speedo0,
+			     &gm20b_pllg_params, clk);
 
 	clk->pl_to_div = pl_to_div;
 	clk->div_to_pl = div_to_pl;
 
 	return ret;
 }
+
+/* FUSE register */
+#define FUSE_RESERVED_CALIB0	0x204
+#define FUSE_RESERVED_CALIB0_INTERCEPT_FRAC_SHIFT	0
+#define FUSE_RESERVED_CALIB0_INTERCEPT_FRAC_WIDTH	4
+#define FUSE_RESERVED_CALIB0_INTERCEPT_INT_SHIFT	4
+#define FUSE_RESERVED_CALIB0_INTERCEPT_INT_WIDTH	10
+#define FUSE_RESERVED_CALIB0_SLOPE_FRAC_SHIFT		14
+#define FUSE_RESERVED_CALIB0_SLOPE_FRAC_WIDTH		10
+#define FUSE_RESERVED_CALIB0_SLOPE_INT_SHIFT		24
+#define FUSE_RESERVED_CALIB0_SLOPE_INT_WIDTH		6
+#define FUSE_RESERVED_CALIB0_FUSE_REV_SHIFT		30
+#define FUSE_RESERVED_CALIB0_FUSE_REV_WIDTH		2
+
+static int
+gm20b_clk_init_fused_params(struct gm20b_clk *clk)
+{
+	struct nvkm_subdev *subdev = &clk->base.base.subdev;
+	u32 val = 0;
+	u32 rev = 0;
+
+#if IS_ENABLED(CONFIG_ARCH_TEGRA)
+	tegra_fuse_readl(FUSE_RESERVED_CALIB0, &val);
+	rev = (val >> FUSE_RESERVED_CALIB0_FUSE_REV_SHIFT) &
+	      MASK(FUSE_RESERVED_CALIB0_FUSE_REV_WIDTH);
+#endif
+
+	/* No fused parameters, we will calibrate later */
+	if (rev == 0)
+		return -EINVAL;
+
+	/* Integer part in mV + fractional part in uV */
+	clk->uvdet_slope = ((val >> FUSE_RESERVED_CALIB0_SLOPE_INT_SHIFT) &
+			MASK(FUSE_RESERVED_CALIB0_SLOPE_INT_WIDTH)) * 1000 +
+			((val >> FUSE_RESERVED_CALIB0_SLOPE_FRAC_SHIFT) &
+			MASK(FUSE_RESERVED_CALIB0_SLOPE_FRAC_WIDTH));
+
+	/* Integer part in mV + fractional part in 100uV */
+	clk->uvdet_offs = ((val >> FUSE_RESERVED_CALIB0_INTERCEPT_INT_SHIFT) &
+			MASK(FUSE_RESERVED_CALIB0_INTERCEPT_INT_WIDTH)) * 1000 +
+			((val >> FUSE_RESERVED_CALIB0_INTERCEPT_FRAC_SHIFT) &
+			 MASK(FUSE_RESERVED_CALIB0_INTERCEPT_FRAC_WIDTH)) * 100;
+
+	nvkm_debug(subdev, "fused calibration data: slope %d, offs %d\n",
+		   clk->uvdet_slope, clk->uvdet_offs);
+	return 0;
+}
+
+static int
+gm20b_clk_init_safe_fmax(struct gm20b_clk *clk)
+{
+	struct nvkm_subdev *subdev = &clk->base.base.subdev;
+	struct nvkm_volt *volt = subdev->device->volt;
+	struct nvkm_pstate *pstates = clk->base.base.func->pstates;
+	int nr_pstates = clk->base.base.func->nr_pstates;
+	int vmin, id = 0;
+	u32 fmax = 0;
+	int i;
+
+	/* find lowest voltage we can use */
+	vmin = volt->vid[0].uv;
+	for (i = 1; i < volt->vid_nr; i++) {
+		if (volt->vid[i].uv <= vmin) {
+			vmin = volt->vid[i].uv;
+			id = volt->vid[i].vid;
+		}
+	}
+
+	/* find max frequency at this voltage */
+	for (i = 0; i < nr_pstates; i++)
+		if (pstates[i].base.voltage == id)
+			fmax = max(fmax,
+				   pstates[i].base.domain[nv_clk_src_gpc]);
+
+	if (!fmax) {
+		nvkm_error(subdev, "failed to evaluate safe fmax\n");
+		return -EINVAL;
+	}
+
+	/* we are safe at 90% of the max frequency */
+	clk->safe_fmax_vmin = fmax * (100 - 10) / 100;
+	nvkm_debug(subdev, "safe fmax @ vmin = %u Khz\n", clk->safe_fmax_vmin);
+
+	return 0;
+}
+
+int
+gm20b_clk_new(struct nvkm_device *device, int index, struct nvkm_clk **pclk)
+{
+	struct nvkm_device_tegra *tdev = device->func->tegra(device);
+	struct gm20b_clk *clk;
+	struct nvkm_subdev *subdev;
+	struct gk20a_clk_pllg_params *clk_params;
+	int ret;
+
+	/* Speedo 0 GPUs cannot use noise-aware PLL */
+	if (tdev->gpu_speedo_id == 0)
+		return gm20b_clk_new_speedo0(device, index, pclk);
+
+	/* Speedo >= 1, use NAPLL */
+	clk = kzalloc(sizeof(*clk) + sizeof(*clk_params), GFP_KERNEL);
+	if (!clk)
+		return -ENOMEM;
+	*pclk = &clk->base.base;
+	subdev = &clk->base.base.subdev;
+
+	/* duplicate the clock parameters since we will patch them below */
+	clk_params = (void *) (clk + 1);
+	*clk_params = gm20b_pllg_params;
+	ret = gk20a_clk_ctor(device, index, &gm20b_clk, clk_params,
+			     &clk->base);
+	if (ret)
+		return ret;
+
+	/*
+	 * NAPLL can only work with max_u, clamp the m range so
+	 * gk20a_pllg_calc_mnp always uses it
+	 */
+	clk_params->max_m = clk_params->min_m = DIV_ROUND_UP(clk_params->max_u,
+						(clk->base.parent_rate / KHZ));
+	if (clk_params->max_m == 0) {
+		nvkm_warn(subdev, "cannot use NAPLL, using legacy clock...\n");
+		kfree(clk);
+		return gm20b_clk_new_speedo0(device, index, pclk);
+	}
+
+	clk->base.pl_to_div = pl_to_div;
+	clk->base.div_to_pl = div_to_pl;
+
+	clk->dvfs_params = &gm20b_dvfs_params;
+
+	ret = gm20b_clk_init_fused_params(clk);
+	/*
+	 * we will calibrate during init - should never happen on
+	 * prod parts
+	 */
+	if (ret)
+		nvkm_warn(subdev, "no fused calibration parameters\n");
+
+	ret = gm20b_clk_init_safe_fmax(clk);
+	if (ret)
+		return ret;
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/Kbuild b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/Kbuild
index 842d5de96d73..edcc157e6ac8 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/Kbuild
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/Kbuild
@@ -24,6 +24,8 @@ nvkm-y += nvkm/subdev/fb/gk104.o
 nvkm-y += nvkm/subdev/fb/gk20a.o
 nvkm-y += nvkm/subdev/fb/gm107.o
 nvkm-y += nvkm/subdev/fb/gm200.o
+nvkm-y += nvkm/subdev/fb/gp100.o
+nvkm-y += nvkm/subdev/fb/gp104.o
 
 nvkm-y += nvkm/subdev/fb/ram.o
 nvkm-y += nvkm/subdev/fb/ramnv04.o
@@ -41,6 +43,7 @@ nvkm-y += nvkm/subdev/fb/rammcp77.o
 nvkm-y += nvkm/subdev/fb/ramgf100.o
 nvkm-y += nvkm/subdev/fb/ramgk104.o
 nvkm-y += nvkm/subdev/fb/ramgm107.o
+nvkm-y += nvkm/subdev/fb/ramgp100.o
 nvkm-y += nvkm/subdev/fb/sddr2.o
 nvkm-y += nvkm/subdev/fb/sddr3.o
 nvkm-y += nvkm/subdev/fb/gddr3.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/base.c
index ce90242b8cce..a7049c041594 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/base.c
@@ -25,6 +25,7 @@
 #include "ram.h"
 
 #include <core/memory.h>
+#include <core/option.h>
 #include <subdev/bios.h>
 #include <subdev/bios/M0203.h>
 #include <engine/gr.h>
@@ -134,6 +135,10 @@ nvkm_fb_init(struct nvkm_subdev *subdev)
 
 	if (fb->func->init)
 		fb->func->init(fb);
+	if (fb->func->init_page)
+		fb->func->init_page(fb);
+	if (fb->func->init_unkn)
+		fb->func->init_unkn(fb);
 	return 0;
 }
 
@@ -171,6 +176,7 @@ nvkm_fb_ctor(const struct nvkm_fb_func *func, struct nvkm_device *device,
 	nvkm_subdev_ctor(&nvkm_fb, device, index, &fb->subdev);
 	fb->func = func;
 	fb->tile.regions = fb->func->tile.regions;
+	fb->page = nvkm_longopt(device->cfgopt, "NvFbBigPage", 0);
 }
 
 int
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gf100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gf100.c
index e649ead5ccfc..76433cc66fff 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gf100.c
@@ -72,6 +72,22 @@ gf100_fb_oneinit(struct nvkm_fb *fb)
 }
 
 void
+gf100_fb_init_page(struct nvkm_fb *fb)
+{
+	struct nvkm_device *device = fb->subdev.device;
+	switch (fb->page) {
+	case 16:
+		nvkm_mask(device, 0x100c80, 0x00000001, 0x00000001);
+		break;
+	case 17:
+	default:
+		nvkm_mask(device, 0x100c80, 0x00000001, 0x00000000);
+		fb->page = 17;
+		break;
+	}
+}
+
+void
 gf100_fb_init(struct nvkm_fb *base)
 {
 	struct gf100_fb *fb = gf100_fb(base);
@@ -79,8 +95,6 @@ gf100_fb_init(struct nvkm_fb *base)
 
 	if (fb->r100c10_page)
 		nvkm_wr32(device, 0x100c10, fb->r100c10 >> 8);
-
-	nvkm_mask(device, 0x100c80, 0x00000001, 0x00000000); /* 128KiB lpg */
 }
 
 void *
@@ -125,6 +139,7 @@ gf100_fb = {
 	.dtor = gf100_fb_dtor,
 	.oneinit = gf100_fb_oneinit,
 	.init = gf100_fb_init,
+	.init_page = gf100_fb_init_page,
 	.intr = gf100_fb_intr,
 	.ram_new = gf100_ram_new,
 	.memtype_valid = gf100_fb_memtype_valid,
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gf100.h b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gf100.h
index 2160e5a39c9a..449f431644b3 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gf100.h
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gf100.h
@@ -14,4 +14,6 @@ int gf100_fb_new_(const struct nvkm_fb_func *, struct nvkm_device *,
 void *gf100_fb_dtor(struct nvkm_fb *);
 void gf100_fb_init(struct nvkm_fb *);
 void gf100_fb_intr(struct nvkm_fb *);
+
+void gp100_fb_init(struct nvkm_fb *);
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gk104.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gk104.c
index b41f0f70038c..4245e2e6e604 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gk104.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gk104.c
@@ -29,6 +29,7 @@ gk104_fb = {
 	.dtor = gf100_fb_dtor,
 	.oneinit = gf100_fb_oneinit,
 	.init = gf100_fb_init,
+	.init_page = gf100_fb_init_page,
 	.intr = gf100_fb_intr,
 	.ram_new = gk104_ram_new,
 	.memtype_valid = gf100_fb_memtype_valid,
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gk20a.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gk20a.c
index 7306f7dfc3b9..f815fe2bbf08 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gk20a.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gk20a.c
@@ -27,7 +27,6 @@ static void
 gk20a_fb_init(struct nvkm_fb *fb)
 {
 	struct nvkm_device *device = fb->subdev.device;
-	nvkm_mask(device, 0x100c80, 0x00000001, 0x00000000); /* 128KiB lpg */
 	nvkm_wr32(device, 0x100cc8, nvkm_memory_addr(fb->mmu_wr) >> 8);
 	nvkm_wr32(device, 0x100ccc, nvkm_memory_addr(fb->mmu_rd) >> 8);
 }
@@ -36,6 +35,7 @@ static const struct nvkm_fb_func
 gk20a_fb = {
 	.oneinit = gf100_fb_oneinit,
 	.init = gk20a_fb_init,
+	.init_page = gf100_fb_init_page,
 	.memtype_valid = gf100_fb_memtype_valid,
 };
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gm107.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gm107.c
index 4869fdb753c9..db699025f546 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gm107.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gm107.c
@@ -29,6 +29,7 @@ gm107_fb = {
 	.dtor = gf100_fb_dtor,
 	.oneinit = gf100_fb_oneinit,
 	.init = gf100_fb_init,
+	.init_page = gf100_fb_init_page,
 	.intr = gf100_fb_intr,
 	.ram_new = gm107_ram_new,
 	.memtype_valid = gf100_fb_memtype_valid,
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gm200.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gm200.c
index 44f5716f64d8..62f653240be3 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gm200.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gm200.c
@@ -26,6 +26,24 @@
 
 #include <core/memory.h>
 
+void
+gm200_fb_init_page(struct nvkm_fb *fb)
+{
+	struct nvkm_device *device = fb->subdev.device;
+	switch (fb->page) {
+	case 16:
+		nvkm_mask(device, 0x100c80, 0x00000801, 0x00000001);
+		break;
+	case 17:
+		nvkm_mask(device, 0x100c80, 0x00000801, 0x00000000);
+		break;
+	default:
+		nvkm_mask(device, 0x100c80, 0x00000800, 0x00000800);
+		fb->page = 0;
+		break;
+	}
+}
+
 static void
 gm200_fb_init(struct nvkm_fb *base)
 {
@@ -48,6 +66,7 @@ gm200_fb = {
 	.dtor = gf100_fb_dtor,
 	.oneinit = gf100_fb_oneinit,
 	.init = gm200_fb_init,
+	.init_page = gm200_fb_init_page,
 	.intr = gf100_fb_intr,
 	.ram_new = gm107_ram_new,
 	.memtype_valid = gf100_fb_memtype_valid,
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gp100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gp100.c
new file mode 100644
index 000000000000..98474aec1921
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gp100.c
@@ -0,0 +1,69 @@
+/*
+ * Copyright 2016 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs <bskeggs@redhat.com>
+ */
+#include "gf100.h"
+#include "ram.h"
+
+#include <core/memory.h>
+
+static void
+gp100_fb_init_unkn(struct nvkm_fb *base)
+{
+	struct nvkm_device *device = gf100_fb(base)->base.subdev.device;
+	nvkm_wr32(device, 0x1fac80, nvkm_rd32(device, 0x100c80));
+	nvkm_wr32(device, 0x1facc4, nvkm_rd32(device, 0x100cc4));
+	nvkm_wr32(device, 0x1facc8, nvkm_rd32(device, 0x100cc8));
+	nvkm_wr32(device, 0x1faccc, nvkm_rd32(device, 0x100ccc));
+}
+
+void
+gp100_fb_init(struct nvkm_fb *base)
+{
+	struct gf100_fb *fb = gf100_fb(base);
+	struct nvkm_device *device = fb->base.subdev.device;
+
+	if (fb->r100c10_page)
+		nvkm_wr32(device, 0x100c10, fb->r100c10 >> 8);
+
+	nvkm_wr32(device, 0x100cc8, nvkm_memory_addr(fb->base.mmu_wr) >> 8);
+	nvkm_wr32(device, 0x100ccc, nvkm_memory_addr(fb->base.mmu_rd) >> 8);
+	nvkm_mask(device, 0x100cc4, 0x00060000,
+		  max(nvkm_memory_size(fb->base.mmu_rd) >> 16, (u64)2) << 17);
+}
+
+static const struct nvkm_fb_func
+gp100_fb = {
+	.dtor = gf100_fb_dtor,
+	.oneinit = gf100_fb_oneinit,
+	.init = gp100_fb_init,
+	.init_page = gm200_fb_init_page,
+	.init_unkn = gp100_fb_init_unkn,
+	.ram_new = gp100_ram_new,
+	.memtype_valid = gf100_fb_memtype_valid,
+};
+
+int
+gp100_fb_new(struct nvkm_device *device, int index, struct nvkm_fb **pfb)
+{
+	return gf100_fb_new_(&gp100_fb, device, index, pfb);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gp104.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gp104.c
new file mode 100644
index 000000000000..92cb71861bec
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/gp104.c
@@ -0,0 +1,43 @@
+/*
+ * Copyright 2016 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs <bskeggs@redhat.com>
+ */
+#include "gf100.h"
+#include "ram.h"
+
+#include <core/memory.h>
+
+static const struct nvkm_fb_func
+gp104_fb = {
+	.dtor = gf100_fb_dtor,
+	.oneinit = gf100_fb_oneinit,
+	.init = gp100_fb_init,
+	.init_page = gm200_fb_init_page,
+	.ram_new = gp100_ram_new,
+	.memtype_valid = gf100_fb_memtype_valid,
+};
+
+int
+gp104_fb_new(struct nvkm_device *device, int index, struct nvkm_fb **pfb)
+{
+	return gf100_fb_new_(&gp104_fb, device, index, pfb);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/priv.h b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/priv.h
index d97d640e60a0..e905d44fa1d5 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/priv.h
@@ -8,6 +8,8 @@ struct nvkm_fb_func {
 	void *(*dtor)(struct nvkm_fb *);
 	int (*oneinit)(struct nvkm_fb *);
 	void (*init)(struct nvkm_fb *);
+	void (*init_page)(struct nvkm_fb *);
+	void (*init_unkn)(struct nvkm_fb *);
 	void (*intr)(struct nvkm_fb *);
 
 	struct {
@@ -60,5 +62,8 @@ void nv46_fb_tile_init(struct nvkm_fb *, int i, u32 addr, u32 size,
 		       u32 pitch, u32 flags, struct nvkm_fb_tile *);
 
 int gf100_fb_oneinit(struct nvkm_fb *);
+void gf100_fb_init_page(struct nvkm_fb *);
 bool gf100_fb_memtype_valid(struct nvkm_fb *, u32);
+
+void gm200_fb_init_page(struct nvkm_fb *);
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ram.h b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ram.h
index f816cbf2ced3..b9ec0ae6723a 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ram.h
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ram.h
@@ -47,4 +47,5 @@ int mcp77_ram_new(struct nvkm_fb *, struct nvkm_ram **);
 int gf100_ram_new(struct nvkm_fb *, struct nvkm_ram **);
 int gk104_ram_new(struct nvkm_fb *, struct nvkm_ram **);
 int gm107_ram_new(struct nvkm_fb *, struct nvkm_ram **);
+int gp100_ram_new(struct nvkm_fb *, struct nvkm_ram **);
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ramgp100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ramgp100.c
new file mode 100644
index 000000000000..f3be408b5e5e
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/fb/ramgp100.c
@@ -0,0 +1,146 @@
+/*
+ * Copyright 2013 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "ram.h"
+
+#include <subdev/bios.h>
+#include <subdev/bios/init.h>
+#include <subdev/bios/rammap.h>
+
+static int
+gp100_ram_init(struct nvkm_ram *ram)
+{
+	struct nvkm_subdev *subdev = &ram->fb->subdev;
+	struct nvkm_device *device = subdev->device;
+	struct nvkm_bios *bios = device->bios;
+	u8  ver, hdr, cnt, len, snr, ssz;
+	u32 data;
+	int i;
+
+	/* run a bunch of tables from rammap table.  there's actually
+	 * individual pointers for each rammap entry too, but, nvidia
+	 * seem to just run the last two entries' scripts early on in
+	 * their init, and never again.. we'll just run 'em all once
+	 * for now.
+	 *
+	 * i strongly suspect that each script is for a separate mode
+	 * (likely selected by 0x9a065c's lower bits?), and the
+	 * binary driver skips the one that's already been setup by
+	 * the init tables.
+	 */
+	data = nvbios_rammapTe(bios, &ver, &hdr, &cnt, &len, &snr, &ssz);
+	if (!data || hdr < 0x15)
+		return -EINVAL;
+
+	cnt  = nvbios_rd08(bios, data + 0x14); /* guess at count */
+	data = nvbios_rd32(bios, data + 0x10); /* guess u32... */
+	if (cnt) {
+		u32 save = nvkm_rd32(device, 0x9a065c) & 0x000000f0;
+		for (i = 0; i < cnt; i++, data += 4) {
+			if (i != save >> 4) {
+				nvkm_mask(device, 0x9a065c, 0x000000f0, i << 4);
+				nvbios_exec(&(struct nvbios_init) {
+						.subdev = subdev,
+						.bios = bios,
+						.offset = nvbios_rd32(bios, data),
+						.execute = 1,
+					    });
+			}
+		}
+		nvkm_mask(device, 0x9a065c, 0x000000f0, save);
+	}
+
+	nvkm_mask(device, 0x9a0584, 0x11000000, 0x00000000);
+	nvkm_wr32(device, 0x10ecc0, 0xffffffff);
+	nvkm_mask(device, 0x9a0160, 0x00000010, 0x00000010);
+	return 0;
+}
+
+static const struct nvkm_ram_func
+gp100_ram_func = {
+	.init = gp100_ram_init,
+	.get = gf100_ram_get,
+	.put = gf100_ram_put,
+};
+
+int
+gp100_ram_new(struct nvkm_fb *fb, struct nvkm_ram **pram)
+{
+	struct nvkm_ram *ram;
+	struct nvkm_subdev *subdev = &fb->subdev;
+	struct nvkm_device *device = subdev->device;
+	enum nvkm_ram_type type = nvkm_fb_bios_memtype(device->bios);
+	const u32 rsvd_head = ( 256 * 1024); /* vga memory */
+	const u32 rsvd_tail = (1024 * 1024); /* vbios etc */
+	u32 fbpa_num = nvkm_rd32(device, 0x022438), fbpa;
+	u32 fbio_opt = nvkm_rd32(device, 0x021c14);
+	u64 part, size = 0, comm = ~0ULL;
+	bool mixed = false;
+	int ret;
+
+	nvkm_debug(subdev, "022438: %08x\n", fbpa_num);
+	nvkm_debug(subdev, "021c14: %08x\n", fbio_opt);
+	for (fbpa = 0; fbpa < fbpa_num; fbpa++) {
+		if (!(fbio_opt & (1 << fbpa))) {
+			part = nvkm_rd32(device, 0x90020c + (fbpa * 0x4000));
+			nvkm_debug(subdev, "fbpa %02x: %lld MiB\n", fbpa, part);
+			part = part << 20;
+			if (part != comm) {
+				if (comm != ~0ULL)
+					mixed = true;
+				comm = min(comm, part);
+			}
+			size = size + part;
+		}
+	}
+
+	ret = nvkm_ram_new_(&gp100_ram_func, fb, type, size, 0, &ram);
+	*pram = ram;
+	if (ret)
+		return ret;
+
+	nvkm_mm_fini(&ram->vram);
+
+	if (mixed) {
+		ret = nvkm_mm_init(&ram->vram, rsvd_head >> NVKM_RAM_MM_SHIFT,
+				   ((comm * fbpa_num) - rsvd_head) >>
+				   NVKM_RAM_MM_SHIFT, 1);
+		if (ret)
+			return ret;
+
+		ret = nvkm_mm_init(&ram->vram, (0x1000000000ULL + comm) >>
+				   NVKM_RAM_MM_SHIFT,
+				   (size - (comm * fbpa_num) - rsvd_tail) >>
+				   NVKM_RAM_MM_SHIFT, 1);
+		if (ret)
+			return ret;
+	} else {
+		ret = nvkm_mm_init(&ram->vram, rsvd_head >> NVKM_RAM_MM_SHIFT,
+				   (size - rsvd_head - rsvd_tail) >>
+				   NVKM_RAM_MM_SHIFT, 1);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/iccsense/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/iccsense/base.c
index 323c79abe468..41bd5d0f7692 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/iccsense/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/iccsense/base.c
@@ -276,6 +276,8 @@ nvkm_iccsense_oneinit(struct nvkm_subdev *subdev)
 		struct pwr_rail_t *r = &stbl.rail[i];
 		struct nvkm_iccsense_rail *rail;
 		struct nvkm_iccsense_sensor *sensor;
+		int (*read)(struct nvkm_iccsense *,
+			    struct nvkm_iccsense_rail *);
 
 		if (!r->mode || r->resistor_mohm == 0)
 			continue;
@@ -284,31 +286,31 @@ nvkm_iccsense_oneinit(struct nvkm_subdev *subdev)
 		if (!sensor)
 			continue;
 
-		rail = kmalloc(sizeof(*rail), GFP_KERNEL);
-		if (!rail)
-			return -ENOMEM;
-
 		switch (sensor->type) {
 		case NVBIOS_EXTDEV_INA209:
 			if (r->rail != 0)
 				continue;
-			rail->read = nvkm_iccsense_ina209_read;
+			read = nvkm_iccsense_ina209_read;
 			break;
 		case NVBIOS_EXTDEV_INA219:
 			if (r->rail != 0)
 				continue;
-			rail->read = nvkm_iccsense_ina219_read;
+			read = nvkm_iccsense_ina219_read;
 			break;
 		case NVBIOS_EXTDEV_INA3221:
 			if (r->rail >= 3)
 				continue;
-			rail->read = nvkm_iccsense_ina3221_read;
+			read = nvkm_iccsense_ina3221_read;
 			break;
 		default:
 			continue;
 		}
 
+		rail = kmalloc(sizeof(*rail), GFP_KERNEL);
+		if (!rail)
+			return -ENOMEM;
 		sensor->rail_mask |= 1 << r->rail;
+		rail->read = read;
 		rail->sensor = sensor;
 		rail->idx = r->rail;
 		rail->mohm = r->resistor_mohm;
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/Kbuild b/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/Kbuild
index 932b366598aa..12d6f4f102cb 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/Kbuild
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/Kbuild
@@ -3,3 +3,4 @@ nvkm-y += nvkm/subdev/ltc/gf100.o
 nvkm-y += nvkm/subdev/ltc/gk104.o
 nvkm-y += nvkm/subdev/ltc/gm107.o
 nvkm-y += nvkm/subdev/ltc/gm200.o
+nvkm-y += nvkm/subdev/ltc/gp100.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/gf100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/gf100.c
index c9eb677967a8..4a0fa0a9b802 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/gf100.c
@@ -23,7 +23,6 @@
  */
 #include "priv.h"
 
-#include <core/enum.h>
 #include <subdev/fb.h>
 #include <subdev/timer.h>
 
@@ -71,7 +70,7 @@ gf100_ltc_zbc_clear_depth(struct nvkm_ltc *ltc, int i, const u32 depth)
 	nvkm_wr32(device, 0x17ea58, depth);
 }
 
-static const struct nvkm_bitfield
+const struct nvkm_bitfield
 gf100_ltc_lts_intr_name[] = {
 	{ 0x00000001, "IDLE_ERROR_IQ" },
 	{ 0x00000002, "IDLE_ERROR_CBC" },
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/gm107.c b/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/gm107.c
index e292f5679418..ec0a3844b2d1 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/gm107.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/gm107.c
@@ -68,18 +68,22 @@ gm107_ltc_zbc_clear_depth(struct nvkm_ltc *ltc, int i, const u32 depth)
 	nvkm_wr32(device, 0x17e34c, depth);
 }
 
-static void
-gm107_ltc_lts_isr(struct nvkm_ltc *ltc, int c, int s)
+void
+gm107_ltc_intr_lts(struct nvkm_ltc *ltc, int c, int s)
 {
 	struct nvkm_subdev *subdev = &ltc->subdev;
 	struct nvkm_device *device = subdev->device;
-	u32 base = 0x140000 + (c * 0x2000) + (s * 0x200);
-	u32 stat = nvkm_rd32(device, base + 0x00c);
+	u32 base = 0x140400 + (c * 0x2000) + (s * 0x200);
+	u32 intr = nvkm_rd32(device, base + 0x00c);
+	u16 stat = intr & 0x0000ffff;
+	char msg[128];
 
 	if (stat) {
-		nvkm_error(subdev, "LTC%d_LTS%d: %08x\n", c, s, stat);
-		nvkm_wr32(device, base + 0x00c, stat);
+		nvkm_snprintbf(msg, sizeof(msg), gf100_ltc_lts_intr_name, stat);
+		nvkm_error(subdev, "LTC%d_LTS%d: %08x [%s]\n", c, s, intr, msg);
 	}
+
+	nvkm_wr32(device, base + 0x00c, intr);
 }
 
 void
@@ -92,7 +96,7 @@ gm107_ltc_intr(struct nvkm_ltc *ltc)
 	while (mask) {
 		u32 s, c = __ffs(mask);
 		for (s = 0; s < ltc->lts_nr; s++)
-			gm107_ltc_lts_isr(ltc, c, s);
+			gm107_ltc_intr_lts(ltc, c, s);
 		mask &= ~(1 << c);
 	}
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/gm200.c b/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/gm200.c
index 2a29bfd5125a..e18e0dc19ec8 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/gm200.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/gm200.c
@@ -46,7 +46,7 @@ static const struct nvkm_ltc_func
 gm200_ltc = {
 	.oneinit = gm200_ltc_oneinit,
 	.init = gm200_ltc_init,
-	.intr = gm107_ltc_intr, /*XXX: not validated */
+	.intr = gm107_ltc_intr,
 	.cbc_clear = gm107_ltc_cbc_clear,
 	.cbc_wait = gm107_ltc_cbc_wait,
 	.zbc = 16,
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/gp100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/gp100.c
new file mode 100644
index 000000000000..0bdfb2f40266
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/gp100.c
@@ -0,0 +1,75 @@
+/*
+ * Copyright 2016 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#include "priv.h"
+
+static void
+gp100_ltc_intr(struct nvkm_ltc *ltc)
+{
+	struct nvkm_device *device = ltc->subdev.device;
+	u32 mask;
+
+	mask = nvkm_rd32(device, 0x0001c0);
+	while (mask) {
+		u32 s, c = __ffs(mask);
+		for (s = 0; s < ltc->lts_nr; s++)
+			gm107_ltc_intr_lts(ltc, c, s);
+		mask &= ~(1 << c);
+	}
+}
+
+static int
+gp100_ltc_oneinit(struct nvkm_ltc *ltc)
+{
+	struct nvkm_device *device = ltc->subdev.device;
+	ltc->ltc_nr = nvkm_rd32(device, 0x12006c);
+	ltc->lts_nr = nvkm_rd32(device, 0x17e280) >> 28;
+	/*XXX: tagram allocation - TBD */
+	return nvkm_mm_init(&ltc->tags, 0, 0, 1);
+}
+
+static void
+gp100_ltc_init(struct nvkm_ltc *ltc)
+{
+	/*XXX: PMU LS call to setup tagram address */
+}
+
+static const struct nvkm_ltc_func
+gp100_ltc = {
+	.oneinit = gp100_ltc_oneinit,
+	.init = gp100_ltc_init,
+	.intr = gp100_ltc_intr,
+	.cbc_clear = gm107_ltc_cbc_clear,
+	.cbc_wait = gm107_ltc_cbc_wait,
+	.zbc = 16,
+	.zbc_clear_color = gm107_ltc_zbc_clear_color,
+	.zbc_clear_depth = gm107_ltc_zbc_clear_depth,
+	.invalidate = gf100_ltc_invalidate,
+	.flush = gf100_ltc_flush,
+};
+
+int
+gp100_ltc_new(struct nvkm_device *device, int index, struct nvkm_ltc **pltc)
+{
+	return nvkm_ltc_new_(&gp100_ltc, device, index, pltc);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/priv.h b/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/priv.h
index 6d81c695ed0d..8b95f96e3ffa 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/ltc/priv.h
@@ -2,6 +2,7 @@
 #define __NVKM_LTC_PRIV_H__
 #define nvkm_ltc(p) container_of((p), struct nvkm_ltc, subdev)
 #include <subdev/ltc.h>
+#include <core/enum.h>
 
 int nvkm_ltc_new_(const struct nvkm_ltc_func *, struct nvkm_device *,
 		  int index, struct nvkm_ltc **);
@@ -31,8 +32,10 @@ void gf100_ltc_zbc_clear_color(struct nvkm_ltc *, int, const u32[4]);
 void gf100_ltc_zbc_clear_depth(struct nvkm_ltc *, int, const u32);
 void gf100_ltc_invalidate(struct nvkm_ltc *);
 void gf100_ltc_flush(struct nvkm_ltc *);
+extern const struct nvkm_bitfield gf100_ltc_lts_intr_name[];
 
 void gm107_ltc_intr(struct nvkm_ltc *);
+void gm107_ltc_intr_lts(struct nvkm_ltc *, int ltc, int lts);
 void gm107_ltc_cbc_clear(struct nvkm_ltc *, u32, u32);
 void gm107_ltc_cbc_wait(struct nvkm_ltc *);
 void gm107_ltc_zbc_clear_color(struct nvkm_ltc *, int, const u32[4]);
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/Kbuild b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/Kbuild
index 49695ac7be2e..12943f92c206 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/Kbuild
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/Kbuild
@@ -10,3 +10,4 @@ nvkm-y += nvkm/subdev/mc/gt215.o
 nvkm-y += nvkm/subdev/mc/gf100.o
 nvkm-y += nvkm/subdev/mc/gk104.o
 nvkm-y += nvkm/subdev/mc/gk20a.o
+nvkm-y += nvkm/subdev/mc/gp100.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/base.c
index 350a8caa84c8..6b25e25f9eba 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/base.c
@@ -27,43 +27,67 @@
 #include <subdev/top.h>
 
 void
-nvkm_mc_unk260(struct nvkm_mc *mc, u32 data)
+nvkm_mc_unk260(struct nvkm_device *device, u32 data)
 {
-	if (mc->func->unk260)
+	struct nvkm_mc *mc = device->mc;
+	if (likely(mc) && mc->func->unk260)
 		mc->func->unk260(mc, data);
 }
 
 void
-nvkm_mc_intr_unarm(struct nvkm_mc *mc)
+nvkm_mc_intr_mask(struct nvkm_device *device, enum nvkm_devidx devidx, bool en)
 {
-	return mc->func->intr_unarm(mc);
+	struct nvkm_mc *mc = device->mc;
+	const struct nvkm_mc_map *map;
+	if (likely(mc) && mc->func->intr_mask) {
+		u32 mask = nvkm_top_intr_mask(device, devidx);
+		for (map = mc->func->intr; !mask && map->stat; map++) {
+			if (map->unit == devidx)
+				mask = map->stat;
+		}
+		mc->func->intr_mask(mc, mask, en ? mask : 0);
+	}
+}
+
+void
+nvkm_mc_intr_unarm(struct nvkm_device *device)
+{
+	struct nvkm_mc *mc = device->mc;
+	if (likely(mc))
+		mc->func->intr_unarm(mc);
 }
 
 void
-nvkm_mc_intr_rearm(struct nvkm_mc *mc)
+nvkm_mc_intr_rearm(struct nvkm_device *device)
 {
-	return mc->func->intr_rearm(mc);
+	struct nvkm_mc *mc = device->mc;
+	if (likely(mc))
+		mc->func->intr_rearm(mc);
 }
 
 static u32
-nvkm_mc_intr_mask(struct nvkm_mc *mc)
+nvkm_mc_intr_stat(struct nvkm_mc *mc)
 {
-	u32 intr = mc->func->intr_mask(mc);
+	u32 intr = mc->func->intr_stat(mc);
 	if (WARN_ON_ONCE(intr == 0xffffffff))
 		intr = 0; /* likely fallen off the bus */
 	return intr;
 }
 
 void
-nvkm_mc_intr(struct nvkm_mc *mc, bool *handled)
+nvkm_mc_intr(struct nvkm_device *device, bool *handled)
 {
-	struct nvkm_device *device = mc->subdev.device;
+	struct nvkm_mc *mc = device->mc;
 	struct nvkm_subdev *subdev;
-	const struct nvkm_mc_map *map = mc->func->intr;
-	u32 stat, intr = nvkm_mc_intr_mask(mc);
+	const struct nvkm_mc_map *map;
+	u32 stat, intr;
 	u64 subdevs;
 
-	stat = nvkm_top_intr(device->top, intr, &subdevs);
+	if (unlikely(!mc))
+		return;
+
+	intr = nvkm_mc_intr_stat(mc);
+	stat = nvkm_top_intr(device, intr, &subdevs);
 	while (subdevs) {
 		enum nvkm_devidx subidx = __ffs64(subdevs);
 		subdev = nvkm_device_subdev(device, subidx);
@@ -72,14 +96,13 @@ nvkm_mc_intr(struct nvkm_mc *mc, bool *handled)
 		subdevs &= ~BIT_ULL(subidx);
 	}
 
-	while (map->stat) {
+	for (map = mc->func->intr; map->stat; map++) {
 		if (intr & map->stat) {
 			subdev = nvkm_device_subdev(device, map->unit);
 			if (subdev)
 				nvkm_subdev_intr(subdev);
 			stat &= ~map->stat;
 		}
-		map++;
 	}
 
 	if (stat)
@@ -87,22 +110,32 @@ nvkm_mc_intr(struct nvkm_mc *mc, bool *handled)
 	*handled = intr != 0;
 }
 
-static void
-nvkm_mc_reset_(struct nvkm_mc *mc, enum nvkm_devidx devidx)
+static u32
+nvkm_mc_reset_mask(struct nvkm_device *device, bool isauto,
+		   enum nvkm_devidx devidx)
 {
-	struct nvkm_device *device = mc->subdev.device;
+	struct nvkm_mc *mc = device->mc;
 	const struct nvkm_mc_map *map;
-	u64 pmc_enable;
-
-	if (!(pmc_enable = nvkm_top_reset(device->top, devidx))) {
-		for (map = mc->func->reset; map && map->stat; map++) {
-			if (map->unit == devidx) {
-				pmc_enable = map->stat;
-				break;
+	u64 pmc_enable = 0;
+	if (likely(mc)) {
+		if (!(pmc_enable = nvkm_top_reset(device, devidx))) {
+			for (map = mc->func->reset; map && map->stat; map++) {
+				if (!isauto || !map->noauto) {
+					if (map->unit == devidx) {
+						pmc_enable = map->stat;
+						break;
+					}
+				}
 			}
 		}
 	}
+	return pmc_enable;
+}
 
+void
+nvkm_mc_reset(struct nvkm_device *device, enum nvkm_devidx devidx)
+{
+	u64 pmc_enable = nvkm_mc_reset_mask(device, true, devidx);
 	if (pmc_enable) {
 		nvkm_mask(device, 0x000200, pmc_enable, 0x00000000);
 		nvkm_mask(device, 0x000200, pmc_enable, pmc_enable);
@@ -111,17 +144,27 @@ nvkm_mc_reset_(struct nvkm_mc *mc, enum nvkm_devidx devidx)
 }
 
 void
-nvkm_mc_reset(struct nvkm_mc *mc, enum nvkm_devidx devidx)
+nvkm_mc_disable(struct nvkm_device *device, enum nvkm_devidx devidx)
 {
-	if (likely(mc))
-		nvkm_mc_reset_(mc, devidx);
+	u64 pmc_enable = nvkm_mc_reset_mask(device, false, devidx);
+	if (pmc_enable)
+		nvkm_mask(device, 0x000200, pmc_enable, 0x00000000);
+}
+
+void
+nvkm_mc_enable(struct nvkm_device *device, enum nvkm_devidx devidx)
+{
+	u64 pmc_enable = nvkm_mc_reset_mask(device, false, devidx);
+	if (pmc_enable) {
+		nvkm_mask(device, 0x000200, pmc_enable, pmc_enable);
+		nvkm_rd32(device, 0x000200);
+	}
 }
 
 static int
 nvkm_mc_fini(struct nvkm_subdev *subdev, bool suspend)
 {
-	struct nvkm_mc *mc = nvkm_mc(subdev);
-	nvkm_mc_intr_unarm(mc);
+	nvkm_mc_intr_unarm(subdev->device);
 	return 0;
 }
 
@@ -131,7 +174,7 @@ nvkm_mc_init(struct nvkm_subdev *subdev)
 	struct nvkm_mc *mc = nvkm_mc(subdev);
 	if (mc->func->init)
 		mc->func->init(mc);
-	nvkm_mc_intr_rearm(mc);
+	nvkm_mc_intr_rearm(subdev->device);
 	return 0;
 }
 
@@ -148,16 +191,21 @@ nvkm_mc = {
 	.fini = nvkm_mc_fini,
 };
 
+void
+nvkm_mc_ctor(const struct nvkm_mc_func *func, struct nvkm_device *device,
+	     int index, struct nvkm_mc *mc)
+{
+	nvkm_subdev_ctor(&nvkm_mc, device, index, &mc->subdev);
+	mc->func = func;
+}
+
 int
 nvkm_mc_new_(const struct nvkm_mc_func *func, struct nvkm_device *device,
 	     int index, struct nvkm_mc **pmc)
 {
 	struct nvkm_mc *mc;
-
 	if (!(mc = *pmc = kzalloc(sizeof(*mc), GFP_KERNEL)))
 		return -ENOMEM;
-
-	nvkm_subdev_ctor(&nvkm_mc, device, index, &mc->subdev);
-	mc->func = func;
+	nvkm_mc_ctor(func, device, index, *pmc);
 	return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/g84.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/g84.c
index 5c85b47f071d..c3d66ef5dc12 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/g84.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/g84.c
@@ -57,7 +57,7 @@ g84_mc = {
 	.intr = g84_mc_intr,
 	.intr_unarm = nv04_mc_intr_unarm,
 	.intr_rearm = nv04_mc_intr_rearm,
-	.intr_mask = nv04_mc_intr_mask,
+	.intr_stat = nv04_mc_intr_stat,
 	.reset = g84_mc_reset,
 };
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/g98.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/g98.c
index 0280b43cc10c..93ad4982ce5f 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/g98.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/g98.c
@@ -57,7 +57,7 @@ g98_mc = {
 	.intr = g98_mc_intr,
 	.intr_unarm = nv04_mc_intr_unarm,
 	.intr_rearm = nv04_mc_intr_rearm,
-	.intr_mask = nv04_mc_intr_mask,
+	.intr_stat = nv04_mc_intr_stat,
 	.reset = g98_mc_reset,
 };
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/gf100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/gf100.c
index 8397e223bd43..d2c4d6033abb 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/gf100.c
@@ -76,7 +76,7 @@ gf100_mc_intr_rearm(struct nvkm_mc *mc)
 }
 
 u32
-gf100_mc_intr_mask(struct nvkm_mc *mc)
+gf100_mc_intr_stat(struct nvkm_mc *mc)
 {
 	struct nvkm_device *device = mc->subdev.device;
 	u32 intr0 = nvkm_rd32(device, 0x000100);
@@ -85,6 +85,14 @@ gf100_mc_intr_mask(struct nvkm_mc *mc)
 }
 
 void
+gf100_mc_intr_mask(struct nvkm_mc *mc, u32 mask, u32 stat)
+{
+	struct nvkm_device *device = mc->subdev.device;
+	nvkm_mask(device, 0x000640, mask, stat);
+	nvkm_mask(device, 0x000644, mask, stat);
+}
+
+void
 gf100_mc_unk260(struct nvkm_mc *mc, u32 data)
 {
 	nvkm_wr32(mc->subdev.device, 0x000260, data);
@@ -97,6 +105,7 @@ gf100_mc = {
 	.intr_unarm = gf100_mc_intr_unarm,
 	.intr_rearm = gf100_mc_intr_rearm,
 	.intr_mask = gf100_mc_intr_mask,
+	.intr_stat = gf100_mc_intr_stat,
 	.reset = gf100_mc_reset,
 	.unk260 = gf100_mc_unk260,
 };
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/gk104.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/gk104.c
index 317464212c7d..7b8c6ecad1a5 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/gk104.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/gk104.c
@@ -26,6 +26,7 @@
 const struct nvkm_mc_map
 gk104_mc_reset[] = {
 	{ 0x00000100, NVKM_ENGINE_FIFO },
+	{ 0x00002000, NVKM_SUBDEV_PMU, true },
 	{}
 };
 
@@ -53,6 +54,7 @@ gk104_mc = {
 	.intr_unarm = gf100_mc_intr_unarm,
 	.intr_rearm = gf100_mc_intr_rearm,
 	.intr_mask = gf100_mc_intr_mask,
+	.intr_stat = gf100_mc_intr_stat,
 	.reset = gk104_mc_reset,
 	.unk260 = gf100_mc_unk260,
 };
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/gk20a.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/gk20a.c
index 60b044f517ed..ca1bf3279dbe 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/gk20a.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/gk20a.c
@@ -30,6 +30,7 @@ gk20a_mc = {
 	.intr_unarm = gf100_mc_intr_unarm,
 	.intr_rearm = gf100_mc_intr_rearm,
 	.intr_mask = gf100_mc_intr_mask,
+	.intr_stat = gf100_mc_intr_stat,
 	.reset = gk104_mc_reset,
 };
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/gp100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/gp100.c
new file mode 100644
index 000000000000..4d22f4abd6de
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/gp100.c
@@ -0,0 +1,103 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+#define gp100_mc(p) container_of((p), struct gp100_mc, base)
+#include "priv.h"
+
+struct gp100_mc {
+	struct nvkm_mc base;
+	spinlock_t lock;
+	bool intr;
+	u32 mask;
+};
+
+static void
+gp100_mc_intr_update(struct gp100_mc *mc)
+{
+	struct nvkm_device *device = mc->base.subdev.device;
+	u32 mask = mc->intr ? mc->mask : 0, i;
+	for (i = 0; i < 2; i++) {
+		nvkm_wr32(device, 0x000180 + (i * 0x04), ~mask);
+		nvkm_wr32(device, 0x000160 + (i * 0x04),  mask);
+	}
+}
+
+static void
+gp100_mc_intr_unarm(struct nvkm_mc *base)
+{
+	struct gp100_mc *mc = gp100_mc(base);
+	unsigned long flags;
+	spin_lock_irqsave(&mc->lock, flags);
+	mc->intr = false;
+	gp100_mc_intr_update(mc);
+	spin_unlock_irqrestore(&mc->lock, flags);
+}
+
+static void
+gp100_mc_intr_rearm(struct nvkm_mc *base)
+{
+	struct gp100_mc *mc = gp100_mc(base);
+	unsigned long flags;
+	spin_lock_irqsave(&mc->lock, flags);
+	mc->intr = true;
+	gp100_mc_intr_update(mc);
+	spin_unlock_irqrestore(&mc->lock, flags);
+}
+
+static void
+gp100_mc_intr_mask(struct nvkm_mc *base, u32 mask, u32 intr)
+{
+	struct gp100_mc *mc = gp100_mc(base);
+	unsigned long flags;
+	spin_lock_irqsave(&mc->lock, flags);
+	mc->mask = (mc->mask & ~mask) | intr;
+	gp100_mc_intr_update(mc);
+	spin_unlock_irqrestore(&mc->lock, flags);
+}
+
+static const struct nvkm_mc_func
+gp100_mc = {
+	.init = nv50_mc_init,
+	.intr = gk104_mc_intr,
+	.intr_unarm = gp100_mc_intr_unarm,
+	.intr_rearm = gp100_mc_intr_rearm,
+	.intr_mask = gp100_mc_intr_mask,
+	.intr_stat = gf100_mc_intr_stat,
+	.reset = gk104_mc_reset,
+};
+
+int
+gp100_mc_new(struct nvkm_device *device, int index, struct nvkm_mc **pmc)
+{
+	struct gp100_mc *mc;
+
+	if (!(mc = kzalloc(sizeof(*mc), GFP_KERNEL)))
+		return -ENOMEM;
+	nvkm_mc_ctor(&gp100_mc, device, index, &mc->base);
+	*pmc = &mc->base;
+
+	spin_lock_init(&mc->lock);
+	mc->intr = false;
+	mc->mask = 0x7fffffff;
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/gt215.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/gt215.c
index aad0ba95bf18..99d50a3d956f 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/gt215.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/gt215.c
@@ -53,13 +53,20 @@ gt215_mc_intr[] = {
 	{},
 };
 
+static void
+gt215_mc_intr_mask(struct nvkm_mc *mc, u32 mask, u32 stat)
+{
+	nvkm_mask(mc->subdev.device, 0x000640, mask, stat);
+}
+
 static const struct nvkm_mc_func
 gt215_mc = {
 	.init = nv50_mc_init,
 	.intr = gt215_mc_intr,
 	.intr_unarm = nv04_mc_intr_unarm,
 	.intr_rearm = nv04_mc_intr_rearm,
-	.intr_mask = nv04_mc_intr_mask,
+	.intr_mask = gt215_mc_intr_mask,
+	.intr_stat = nv04_mc_intr_stat,
 	.reset = gt215_mc_reset,
 };
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/nv04.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/nv04.c
index a062624e906b..6509defd1460 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/nv04.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/nv04.c
@@ -56,7 +56,7 @@ nv04_mc_intr_rearm(struct nvkm_mc *mc)
 }
 
 u32
-nv04_mc_intr_mask(struct nvkm_mc *mc)
+nv04_mc_intr_stat(struct nvkm_mc *mc)
 {
 	return nvkm_rd32(mc->subdev.device, 0x000100);
 }
@@ -75,7 +75,7 @@ nv04_mc = {
 	.intr = nv04_mc_intr,
 	.intr_unarm = nv04_mc_intr_unarm,
 	.intr_rearm = nv04_mc_intr_rearm,
-	.intr_mask = nv04_mc_intr_mask,
+	.intr_stat = nv04_mc_intr_stat,
 	.reset = nv04_mc_reset,
 };
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/nv11.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/nv11.c
index 55f0b9166b52..9213107901e6 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/nv11.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/nv11.c
@@ -39,7 +39,7 @@ nv11_mc = {
 	.intr = nv11_mc_intr,
 	.intr_unarm = nv04_mc_intr_unarm,
 	.intr_rearm = nv04_mc_intr_rearm,
-	.intr_mask = nv04_mc_intr_mask,
+	.intr_stat = nv04_mc_intr_stat,
 	.reset = nv04_mc_reset,
 };
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/nv17.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/nv17.c
index c40fa67f79a5..64bf5bbf8146 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/nv17.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/nv17.c
@@ -48,7 +48,7 @@ nv17_mc = {
 	.intr = nv17_mc_intr,
 	.intr_unarm = nv04_mc_intr_unarm,
 	.intr_rearm = nv04_mc_intr_rearm,
-	.intr_mask = nv04_mc_intr_mask,
+	.intr_stat = nv04_mc_intr_stat,
 	.reset = nv17_mc_reset,
 };
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/nv44.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/nv44.c
index cc56271db564..65fa44a64b98 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/nv44.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/nv44.c
@@ -43,7 +43,7 @@ nv44_mc = {
 	.intr = nv17_mc_intr,
 	.intr_unarm = nv04_mc_intr_unarm,
 	.intr_rearm = nv04_mc_intr_rearm,
-	.intr_mask = nv04_mc_intr_mask,
+	.intr_stat = nv04_mc_intr_stat,
 	.reset = nv17_mc_reset,
 };
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/nv50.c b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/nv50.c
index 343b6078580d..fe93b4fd7100 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/nv50.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/nv50.c
@@ -50,7 +50,7 @@ nv50_mc = {
 	.intr = nv50_mc_intr,
 	.intr_unarm = nv04_mc_intr_unarm,
 	.intr_rearm = nv04_mc_intr_rearm,
-	.intr_mask = nv04_mc_intr_mask,
+	.intr_stat = nv04_mc_intr_stat,
 	.reset = nv17_mc_reset,
 };
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/priv.h b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/priv.h
index a12038118512..4f0576a06d24 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mc/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mc/priv.h
@@ -3,12 +3,15 @@
 #define nvkm_mc(p) container_of((p), struct nvkm_mc, subdev)
 #include <subdev/mc.h>
 
+void nvkm_mc_ctor(const struct nvkm_mc_func *, struct nvkm_device *,
+		  int index, struct nvkm_mc *);
 int nvkm_mc_new_(const struct nvkm_mc_func *, struct nvkm_device *,
 		 int index, struct nvkm_mc **);
 
 struct nvkm_mc_map {
 	u32 stat;
 	u32 unit;
+	bool noauto;
 };
 
 struct nvkm_mc_func {
@@ -18,8 +21,10 @@ struct nvkm_mc_func {
 	void (*intr_unarm)(struct nvkm_mc *);
 	/* enable reporting of interrupts to host */
 	void (*intr_rearm)(struct nvkm_mc *);
+	/* (un)mask delivery of specific interrupts */
+	void (*intr_mask)(struct nvkm_mc *, u32 mask, u32 stat);
 	/* retrieve pending interrupt mask (NV_PMC_INTR) */
-	u32 (*intr_mask)(struct nvkm_mc *);
+	u32 (*intr_stat)(struct nvkm_mc *);
 	const struct nvkm_mc_map *reset;
 	void (*unk260)(struct nvkm_mc *, u32);
 };
@@ -27,7 +32,7 @@ struct nvkm_mc_func {
 void nv04_mc_init(struct nvkm_mc *);
 void nv04_mc_intr_unarm(struct nvkm_mc *);
 void nv04_mc_intr_rearm(struct nvkm_mc *);
-u32 nv04_mc_intr_mask(struct nvkm_mc *);
+u32 nv04_mc_intr_stat(struct nvkm_mc *);
 extern const struct nvkm_mc_map nv04_mc_reset[];
 
 extern const struct nvkm_mc_map nv17_mc_intr[];
@@ -39,7 +44,8 @@ void nv50_mc_init(struct nvkm_mc *);
 
 void gf100_mc_intr_unarm(struct nvkm_mc *);
 void gf100_mc_intr_rearm(struct nvkm_mc *);
-u32 gf100_mc_intr_mask(struct nvkm_mc *);
+void gf100_mc_intr_mask(struct nvkm_mc *, u32, u32);
+u32 gf100_mc_intr_stat(struct nvkm_mc *);
 void gf100_mc_unk260(struct nvkm_mc *, u32);
 
 extern const struct nvkm_mc_map gk104_mc_intr[];
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pci/Kbuild b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/Kbuild
index 3c2519fdeb81..2a31b7d66a6d 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pci/Kbuild
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/Kbuild
@@ -10,3 +10,4 @@ nvkm-y += nvkm/subdev/pci/g94.o
 nvkm-y += nvkm/subdev/pci/gf100.o
 nvkm-y += nvkm/subdev/pci/gf106.o
 nvkm-y += nvkm/subdev/pci/gk104.o
+nvkm-y += nvkm/subdev/pci/gp100.o
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pci/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/base.c
index 6b0328bd7eed..eb9b278198b2 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/pci/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/base.c
@@ -69,15 +69,13 @@ static irqreturn_t
 nvkm_pci_intr(int irq, void *arg)
 {
 	struct nvkm_pci *pci = arg;
-	struct nvkm_mc *mc = pci->subdev.device->mc;
+	struct nvkm_device *device = pci->subdev.device;
 	bool handled = false;
-	if (likely(mc)) {
-		nvkm_mc_intr_unarm(mc);
-		if (pci->msi)
-			pci->func->msi_rearm(pci);
-		nvkm_mc_intr(mc, &handled);
-		nvkm_mc_intr_rearm(mc);
-	}
+	nvkm_mc_intr_unarm(device);
+	if (pci->msi)
+		pci->func->msi_rearm(pci);
+	nvkm_mc_intr(device, &handled);
+	nvkm_mc_intr_rearm(device);
 	return handled ? IRQ_HANDLED : IRQ_NONE;
 }
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/pci/gp100.c b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/gp100.c
new file mode 100644
index 000000000000..82c5234a06ff
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/pci/gp100.c
@@ -0,0 +1,44 @@
+/*
+ * Copyright 2015 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs <bskeggs@redhat.com>
+ */
+#include "priv.h"
+
+static void
+gp100_pci_msi_rearm(struct nvkm_pci *pci)
+{
+	nvkm_pci_wr32(pci, 0x0704, 0x00000000);
+}
+
+static const struct nvkm_pci_func
+gp100_pci_func = {
+	.rd32 = nv40_pci_rd32,
+	.wr08 = nv40_pci_wr08,
+	.wr32 = nv40_pci_wr32,
+	.msi_rearm = gp100_pci_msi_rearm,
+};
+
+int
+gp100_pci_new(struct nvkm_device *device, int index, struct nvkm_pci **ppci)
+{
+	return nvkm_pci_new_(&gp100_pci_func, device, index, ppci);
+}
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/base.c
index 213fdba6cfa0..314be2192b7d 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/base.c
@@ -19,8 +19,9 @@
  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  * DEALINGS IN THE SOFTWARE.
  */
-
 #include "priv.h"
+
+#include <subdev/mc.h>
 #include <subdev/timer.h>
 
 static const char *
@@ -70,12 +71,11 @@ nvkm_secboot_falcon_enable(struct nvkm_secboot *sb)
 	int ret;
 
 	/* enable engine */
-	nvkm_mask(device, 0x200, sb->enable_mask, sb->enable_mask);
-	nvkm_rd32(device, 0x200);
+	nvkm_mc_enable(device, sb->devidx);
 	ret = nvkm_wait_msec(device, 10, sb->base + 0x10c, 0x6, 0x0);
 	if (ret < 0) {
-		nvkm_mask(device, 0x200, sb->enable_mask, 0x0);
 		nvkm_error(&sb->subdev, "Falcon mem scrubbing timeout\n");
+		nvkm_mc_disable(device, sb->devidx);
 		return ret;
 	}
 
@@ -85,8 +85,7 @@ nvkm_secboot_falcon_enable(struct nvkm_secboot *sb)
 
 	/* enable IRQs */
 	nvkm_wr32(device, sb->base + 0x010, 0xff);
-	nvkm_mask(device, 0x640, sb->irq_mask, sb->irq_mask);
-	nvkm_mask(device, 0x644, sb->irq_mask, sb->irq_mask);
+	nvkm_mc_intr_mask(device, sb->devidx, true);
 
 	return 0;
 }
@@ -97,14 +96,13 @@ nvkm_secboot_falcon_disable(struct nvkm_secboot *sb)
 	struct nvkm_device *device = sb->subdev.device;
 
 	/* disable IRQs and wait for any previous code to complete */
-	nvkm_mask(device, 0x644, sb->irq_mask, 0x0);
-	nvkm_mask(device, 0x640, sb->irq_mask, 0x0);
+	nvkm_mc_intr_mask(device, sb->devidx, false);
 	nvkm_wr32(device, sb->base + 0x014, 0xff);
 
 	falcon_wait_idle(device, sb->base);
 
 	/* disable engine */
-	nvkm_mask(device, 0x200, sb->enable_mask, 0x0);
+	nvkm_mc_disable(device, sb->devidx);
 
 	return 0;
 }
@@ -216,14 +214,7 @@ nvkm_secboot_oneinit(struct nvkm_subdev *subdev)
 		return ret;
 	}
 
-	/*
-	 * Build all blobs - the same blobs can be used to perform secure boot
-	 * multiple times
-	 */
-	if (sb->func->prepare_blobs)
-		ret = sb->func->prepare_blobs(sb);
-
-	return ret;
+	return 0;
 }
 
 static int
@@ -270,9 +261,8 @@ nvkm_secboot_ctor(const struct nvkm_secboot_func *func,
 	/* setup the performing falcon's base address and masks */
 	switch (func->boot_falcon) {
 	case NVKM_SECBOOT_FALCON_PMU:
+		sb->devidx = NVKM_SUBDEV_PMU;
 		sb->base = 0x10a000;
-		sb->irq_mask = 0x1000000;
-		sb->enable_mask = 0x2000;
 		break;
 	default:
 		nvkm_error(&sb->subdev, "invalid secure boot falcon\n");
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/gm200.c b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/gm200.c
index cc100dc940ea..f1e2dc914366 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/gm200.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/gm200.c
@@ -860,6 +860,8 @@ gm200_secboot_prepare_ls_blob(struct gm200_secboot *gsb)
 
 	/* Write LS blob */
 	ret = ls_ucode_mgr_write_wpr(gsb, &mgr, gsb->ls_blob);
+	if (ret)
+		nvkm_gpuobj_del(&gsb->ls_blob);
 
 cleanup:
 	ls_ucode_mgr_cleanup(&mgr);
@@ -1023,29 +1025,34 @@ gm20x_secboot_prepare_blobs(struct gm200_secboot *gsb)
 	int ret;
 
 	/* Load and prepare the managed falcon's firmwares */
-	ret = gm200_secboot_prepare_ls_blob(gsb);
-	if (ret)
-		return ret;
+	if (!gsb->ls_blob) {
+		ret = gm200_secboot_prepare_ls_blob(gsb);
+		if (ret)
+			return ret;
+	}
 
 	/* Load the HS firmware that will load the LS firmwares */
-	ret = gm200_secboot_prepare_hs_blob(gsb, "acr/ucode_load",
-					    &gsb->acr_load_blob,
-					    &gsb->acr_load_bl_desc, true);
-	if (ret)
-		return ret;
+	if (!gsb->acr_load_blob) {
+		ret = gm200_secboot_prepare_hs_blob(gsb, "acr/ucode_load",
+						&gsb->acr_load_blob,
+						&gsb->acr_load_bl_desc, true);
+		if (ret)
+			return ret;
+	}
 
 	/* Load the HS firmware bootloader */
-	ret = gm200_secboot_prepare_hsbl_blob(gsb);
-	if (ret)
-		return ret;
+	if (!gsb->hsbl_blob) {
+		ret = gm200_secboot_prepare_hsbl_blob(gsb);
+		if (ret)
+			return ret;
+	}
 
 	return 0;
 }
 
 static int
-gm200_secboot_prepare_blobs(struct nvkm_secboot *sb)
+gm200_secboot_prepare_blobs(struct gm200_secboot *gsb)
 {
-	struct gm200_secboot *gsb = gm200_secboot(sb);
 	int ret;
 
 	ret = gm20x_secboot_prepare_blobs(gsb);
@@ -1053,15 +1060,37 @@ gm200_secboot_prepare_blobs(struct nvkm_secboot *sb)
 		return ret;
 
 	/* dGPU only: load the HS firmware that unprotects the WPR region */
-	ret = gm200_secboot_prepare_hs_blob(gsb, "acr/ucode_unload",
-					    &gsb->acr_unload_blob,
-					    &gsb->acr_unload_bl_desc, false);
-	if (ret)
-		return ret;
+	if (!gsb->acr_unload_blob) {
+		ret = gm200_secboot_prepare_hs_blob(gsb, "acr/ucode_unload",
+					       &gsb->acr_unload_blob,
+					       &gsb->acr_unload_bl_desc, false);
+		if (ret)
+			return ret;
+	}
 
 	return 0;
 }
 
+static int
+gm200_secboot_blobs_ready(struct gm200_secboot *gsb)
+{
+	struct nvkm_subdev *subdev = &gsb->base.subdev;
+	int ret;
+
+	/* firmware already loaded, nothing to do... */
+	if (gsb->firmware_ok)
+		return 0;
+
+	ret = gsb->func->prepare_blobs(gsb);
+	if (ret) {
+		nvkm_error(subdev, "failed to load secure firmware\n");
+		return ret;
+	}
+
+	gsb->firmware_ok = true;
+
+	return 0;
+}
 
 
 /*
@@ -1234,6 +1263,11 @@ gm200_secboot_reset(struct nvkm_secboot *sb, enum nvkm_secboot_falcon falcon)
 	struct gm200_secboot *gsb = gm200_secboot(sb);
 	int ret;
 
+	/* Make sure all blobs are ready */
+	ret = gm200_secboot_blobs_ready(gsb);
+	if (ret)
+		return ret;
+
 	/*
 	 * Dummy GM200 implementation: perform secure boot each time we are
 	 * called on FECS. Since only FECS and GPCCS are managed and started
@@ -1373,7 +1407,6 @@ gm200_secboot = {
 	.dtor = gm200_secboot_dtor,
 	.init = gm200_secboot_init,
 	.fini = gm200_secboot_fini,
-	.prepare_blobs = gm200_secboot_prepare_blobs,
 	.reset = gm200_secboot_reset,
 	.start = gm200_secboot_start,
 	.managed_falcons = BIT(NVKM_SECBOOT_FALCON_FECS) |
@@ -1415,6 +1448,7 @@ gm200_secboot_func = {
 	.bl_desc_size = sizeof(struct gm200_flcn_bl_desc),
 	.fixup_bl_desc = gm200_secboot_fixup_bl_desc,
 	.fixup_hs_desc = gm200_secboot_fixup_hs_desc,
+	.prepare_blobs = gm200_secboot_prepare_blobs,
 };
 
 int
@@ -1487,3 +1521,19 @@ MODULE_FIRMWARE("nvidia/gm206/gr/sw_ctx.bin");
 MODULE_FIRMWARE("nvidia/gm206/gr/sw_nonctx.bin");
 MODULE_FIRMWARE("nvidia/gm206/gr/sw_bundle_init.bin");
 MODULE_FIRMWARE("nvidia/gm206/gr/sw_method_init.bin");
+
+MODULE_FIRMWARE("nvidia/gp100/acr/bl.bin");
+MODULE_FIRMWARE("nvidia/gp100/acr/ucode_load.bin");
+MODULE_FIRMWARE("nvidia/gp100/acr/ucode_unload.bin");
+MODULE_FIRMWARE("nvidia/gp100/gr/fecs_bl.bin");
+MODULE_FIRMWARE("nvidia/gp100/gr/fecs_inst.bin");
+MODULE_FIRMWARE("nvidia/gp100/gr/fecs_data.bin");
+MODULE_FIRMWARE("nvidia/gp100/gr/fecs_sig.bin");
+MODULE_FIRMWARE("nvidia/gp100/gr/gpccs_bl.bin");
+MODULE_FIRMWARE("nvidia/gp100/gr/gpccs_inst.bin");
+MODULE_FIRMWARE("nvidia/gp100/gr/gpccs_data.bin");
+MODULE_FIRMWARE("nvidia/gp100/gr/gpccs_sig.bin");
+MODULE_FIRMWARE("nvidia/gp100/gr/sw_ctx.bin");
+MODULE_FIRMWARE("nvidia/gp100/gr/sw_nonctx.bin");
+MODULE_FIRMWARE("nvidia/gp100/gr/sw_bundle_init.bin");
+MODULE_FIRMWARE("nvidia/gp100/gr/sw_method_init.bin");
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/gm20b.c b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/gm20b.c
index 684320484b70..d5395ebfe8d3 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/gm20b.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/gm20b.c
@@ -42,6 +42,32 @@ struct gm20b_flcn_bl_desc {
 	u32 data_size;
 };
 
+static int
+gm20b_secboot_prepare_blobs(struct gm200_secboot *gsb)
+{
+	struct nvkm_subdev *subdev = &gsb->base.subdev;
+	int acr_size;
+	int ret;
+
+	ret = gm20x_secboot_prepare_blobs(gsb);
+	if (ret)
+		return ret;
+
+	acr_size = gsb->acr_load_blob->size;
+	/*
+	 * On Tegra the WPR region is set by the bootloader. It is illegal for
+	 * the HS blob to be larger than this region.
+	 */
+	if (acr_size > gsb->wpr_size) {
+		nvkm_error(subdev, "WPR region too small for FW blob!\n");
+		nvkm_error(subdev, "required: %dB\n", acr_size);
+		nvkm_error(subdev, "WPR size: %dB\n", gsb->wpr_size);
+		return -ENOSPC;
+	}
+
+	return 0;
+}
+
 /**
  * gm20b_secboot_fixup_bl_desc - adapt BL descriptor to format used by GM20B FW
  *
@@ -88,6 +114,7 @@ gm20b_secboot_func = {
 	.bl_desc_size = sizeof(struct gm20b_flcn_bl_desc),
 	.fixup_bl_desc = gm20b_secboot_fixup_bl_desc,
 	.fixup_hs_desc = gm20b_secboot_fixup_hs_desc,
+	.prepare_blobs = gm20b_secboot_prepare_blobs,
 };
 
 
@@ -147,32 +174,6 @@ gm20b_tegra_read_wpr(struct gm200_secboot *gsb)
 #endif
 
 static int
-gm20b_secboot_prepare_blobs(struct nvkm_secboot *sb)
-{
-	struct gm200_secboot *gsb = gm200_secboot(sb);
-	int acr_size;
-	int ret;
-
-	ret = gm20x_secboot_prepare_blobs(gsb);
-	if (ret)
-		return ret;
-
-	acr_size = gsb->acr_load_blob->size;
-	/*
-	 * On Tegra the WPR region is set by the bootloader. It is illegal for
-	 * the HS blob to be larger than this region.
-	 */
-	if (acr_size > gsb->wpr_size) {
-		nvkm_error(&sb->subdev, "WPR region too small for FW blob!\n");
-		nvkm_error(&sb->subdev, "required: %dB\n", acr_size);
-		nvkm_error(&sb->subdev, "WPR size: %dB\n", gsb->wpr_size);
-		return -ENOSPC;
-	}
-
-	return 0;
-}
-
-static int
 gm20b_secboot_init(struct nvkm_secboot *sb)
 {
 	struct gm200_secboot *gsb = gm200_secboot(sb);
@@ -189,7 +190,6 @@ static const struct nvkm_secboot_func
 gm20b_secboot = {
 	.dtor = gm200_secboot_dtor,
 	.init = gm20b_secboot_init,
-	.prepare_blobs = gm20b_secboot_prepare_blobs,
 	.reset = gm200_secboot_reset,
 	.start = gm200_secboot_start,
 	.managed_falcons = BIT(NVKM_SECBOOT_FALCON_FECS),
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/priv.h b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/priv.h
index f2b09dee7c5d..a9a8a0e1017e 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/priv.h
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/secboot/priv.h
@@ -30,7 +30,6 @@ struct nvkm_secboot_func {
 	int (*init)(struct nvkm_secboot *);
 	int (*fini)(struct nvkm_secboot *, bool suspend);
 	void *(*dtor)(struct nvkm_secboot *);
-	int (*prepare_blobs)(struct nvkm_secboot *);
 	int (*reset)(struct nvkm_secboot *, enum nvkm_secboot_falcon);
 	int (*start)(struct nvkm_secboot *, enum nvkm_secboot_falcon);
 
@@ -147,10 +146,8 @@ struct hsflcn_acr_desc {
  * @inst:		instance block for HS falcon
  * @pgd:		page directory for the HS falcon
  * @vm:			address space used by the HS falcon
- * @bl_desc_size:	size of the BL descriptor used by this chip.
- * @fixup_bl_desc:	hook that generates the proper BL descriptor format from
- *			the generic GM200 format into a data array of size
- *			bl_desc_size
+ * @falcon_state:	current state of the managed falcons
+ * @firmware_ok:	whether the firmware blobs have been created
  */
 struct gm200_secboot {
 	struct nvkm_secboot base;
@@ -196,9 +193,19 @@ struct gm200_secboot {
 		RUNNING,
 	} falcon_state[NVKM_SECBOOT_FALCON_END];
 
+	bool firmware_ok;
 };
 #define gm200_secboot(sb) container_of(sb, struct gm200_secboot, base)
 
+/**
+ * Contains functions we wish to abstract between GM200-like implementations
+ * @bl_desc_size:	size of the BL descriptor used by this chip.
+ * @fixup_bl_desc:	hook that generates the proper BL descriptor format from
+ *			the generic GM200 format into a data array of size
+ *			bl_desc_size
+ * @fixup_hs_desc:	hook that twiddles the HS descriptor before it is used
+ * @prepare_blobs:	prepares the various blobs needed for secure booting
+ */
 struct gm200_secboot_func {
 	/*
 	 * Size of the bootloader descriptor for this chip. A block of this
@@ -214,6 +221,7 @@ struct gm200_secboot_func {
 	 * we want the HS FW to set up.
 	 */
 	void (*fixup_hs_desc)(struct gm200_secboot *, struct hsflcn_acr_desc *);
+	int (*prepare_blobs)(struct gm200_secboot *);
 };
 
 int gm200_secboot_init(struct nvkm_secboot *);
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/top/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/top/base.c
index a1b264664aad..fe063d5728e2 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/top/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/top/base.c
@@ -41,8 +41,9 @@ nvkm_top_device_new(struct nvkm_top *top)
 }
 
 u32
-nvkm_top_reset(struct nvkm_top *top, enum nvkm_devidx index)
+nvkm_top_reset(struct nvkm_device *device, enum nvkm_devidx index)
 {
+	struct nvkm_top *top = device->top;
 	struct nvkm_top_device *info;
 
 	if (top) {
@@ -56,8 +57,25 @@ nvkm_top_reset(struct nvkm_top *top, enum nvkm_devidx index)
 }
 
 u32
-nvkm_top_intr(struct nvkm_top *top, u32 intr, u64 *psubdevs)
+nvkm_top_intr_mask(struct nvkm_device *device, enum nvkm_devidx devidx)
 {
+	struct nvkm_top *top = device->top;
+	struct nvkm_top_device *info;
+
+	if (top) {
+		list_for_each_entry(info, &top->device, head) {
+			if (info->index == devidx && info->intr >= 0)
+				return BIT(info->intr);
+		}
+	}
+
+	return 0;
+}
+
+u32
+nvkm_top_intr(struct nvkm_device *device, u32 intr, u64 *psubdevs)
+{
+	struct nvkm_top *top = device->top;
 	struct nvkm_top_device *info;
 	u64 subdevs = 0;
 	u32 handled = 0;
@@ -78,8 +96,9 @@ nvkm_top_intr(struct nvkm_top *top, u32 intr, u64 *psubdevs)
 }
 
 enum nvkm_devidx
-nvkm_top_fault(struct nvkm_top *top, int fault)
+nvkm_top_fault(struct nvkm_device *device, int fault)
 {
+	struct nvkm_top *top = device->top;
 	struct nvkm_top_device *info;
 
 	list_for_each_entry(info, &top->device, head) {
@@ -91,8 +110,9 @@ nvkm_top_fault(struct nvkm_top *top, int fault)
 }
 
 enum nvkm_devidx
-nvkm_top_engine(struct nvkm_top *top, int index, int *runl, int *engn)
+nvkm_top_engine(struct nvkm_device *device, int index, int *runl, int *engn)
 {
+	struct nvkm_top *top = device->top;
 	struct nvkm_top_device *info;
 	int n = 0;
 
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/top/gk104.c b/drivers/gpu/drm/nouveau/nvkm/subdev/top/gk104.c
index e06acc340e99..efac3402f9dd 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/top/gk104.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/top/gk104.c
@@ -29,7 +29,7 @@ gk104_top_oneinit(struct nvkm_top *top)
 	struct nvkm_subdev *subdev = &top->subdev;
 	struct nvkm_device *device = subdev->device;
 	struct nvkm_top_device *info = NULL;
-	u32 data, type;
+	u32 data, type, inst;
 	int i;
 
 	for (i = 0; i < 64; i++) {
@@ -37,6 +37,7 @@ gk104_top_oneinit(struct nvkm_top *top)
 			if (!(info = nvkm_top_device_new(top)))
 				return -ENOMEM;
 			type = ~0;
+			inst = 0;
 		}
 
 		data = nvkm_rd32(device, 0x022700 + (i * 0x04));
@@ -45,6 +46,7 @@ gk104_top_oneinit(struct nvkm_top *top)
 		case 0x00000000: /* NOT_VALID */
 			continue;
 		case 0x00000001: /* DATA */
+			inst        = (data & 0x3c000000) >> 26;
 			info->addr  = (data & 0x00fff000);
 			info->fault = (data & 0x000000f8) >> 3;
 			break;
@@ -67,27 +69,32 @@ gk104_top_oneinit(struct nvkm_top *top)
 			continue;
 
 		/* Translate engine type to NVKM engine identifier. */
+#define A_(A) if (inst == 0) info->index = NVKM_ENGINE_##A
+#define B_(A) if (inst + NVKM_ENGINE_##A##0 < NVKM_ENGINE_##A##_LAST + 1)      \
+		info->index = NVKM_ENGINE_##A##0 + inst
 		switch (type) {
-		case 0x00000000: info->index = NVKM_ENGINE_GR; break;
-		case 0x00000001: info->index = NVKM_ENGINE_CE0; break;
-		case 0x00000002: info->index = NVKM_ENGINE_CE1; break;
-		case 0x00000003: info->index = NVKM_ENGINE_CE2; break;
-		case 0x00000008: info->index = NVKM_ENGINE_MSPDEC; break;
-		case 0x00000009: info->index = NVKM_ENGINE_MSPPP; break;
-		case 0x0000000a: info->index = NVKM_ENGINE_MSVLD; break;
-		case 0x0000000b: info->index = NVKM_ENGINE_MSENC; break;
-		case 0x0000000c: info->index = NVKM_ENGINE_VIC; break;
-		case 0x0000000d: info->index = NVKM_ENGINE_SEC; break;
-		case 0x0000000e: info->index = NVKM_ENGINE_NVENC0; break;
-		case 0x0000000f: info->index = NVKM_ENGINE_NVENC1; break;
-		case 0x00000010: info->index = NVKM_ENGINE_NVDEC; break;
+		case 0x00000000: A_(GR    ); break;
+		case 0x00000001: A_(CE0   ); break;
+		case 0x00000002: A_(CE1   ); break;
+		case 0x00000003: A_(CE2   ); break;
+		case 0x00000008: A_(MSPDEC); break;
+		case 0x00000009: A_(MSPPP ); break;
+		case 0x0000000a: A_(MSVLD ); break;
+		case 0x0000000b: A_(MSENC ); break;
+		case 0x0000000c: A_(VIC   ); break;
+		case 0x0000000d: A_(SEC   ); break;
+		case 0x0000000e: B_(NVENC ); break;
+		case 0x0000000f: A_(NVENC1); break;
+		case 0x00000010: A_(NVDEC ); break;
+		case 0x00000013: B_(CE    ); break;
 			break;
 		default:
 			break;
 		}
 
-		nvkm_debug(subdev, "%02x (%8s): addr %06x fault %2d engine %2d "
-				   "runlist %2d intr %2d reset %2d\n", type,
+		nvkm_debug(subdev, "%02x.%d (%8s): addr %06x fault %2d "
+				   "engine %2d runlist %2d intr %2d "
+				   "reset %2d\n", type, inst,
 			   info->index == NVKM_SUBDEV_NR ? NULL :
 					  nvkm_subdev_name[info->index],
 			   info->addr, info->fault, info->engine, info->runlist,
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/volt/base.c b/drivers/gpu/drm/nouveau/nvkm/subdev/volt/base.c
index 6b2d7531a7ff..1c3d23b0e84a 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/volt/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/volt/base.c
@@ -120,6 +120,8 @@ nvkm_volt_parse_bios(struct nvkm_bios *bios, struct nvkm_volt *volt)
 
 	data = nvbios_volt_parse(bios, &ver, &hdr, &cnt, &len, &info);
 	if (data && info.vidmask && info.base && info.step) {
+		volt->min_uv = info.min;
+		volt->max_uv = info.max;
 		for (i = 0; i < info.vidmask + 1; i++) {
 			if (info.base >= info.min &&
 				info.base <= info.max) {
@@ -131,6 +133,8 @@ nvkm_volt_parse_bios(struct nvkm_bios *bios, struct nvkm_volt *volt)
 		}
 		volt->vid_mask = info.vidmask;
 	} else if (data && info.vidmask) {
+		volt->min_uv = 0xffffffff;
+		volt->max_uv = 0;
 		for (i = 0; i < cnt; i++) {
 			data = nvbios_volt_entry_parse(bios, i, &ver, &hdr,
 						       &ivid);
@@ -138,9 +142,14 @@ nvkm_volt_parse_bios(struct nvkm_bios *bios, struct nvkm_volt *volt)
 				volt->vid[volt->vid_nr].uv = ivid.voltage;
 				volt->vid[volt->vid_nr].vid = ivid.vid;
 				volt->vid_nr++;
+				volt->min_uv = min(volt->min_uv, ivid.voltage);
+				volt->max_uv = max(volt->max_uv, ivid.voltage);
 			}
 		}
 		volt->vid_mask = info.vidmask;
+	} else if (data && info.type == NVBIOS_VOLT_PWM) {
+		volt->min_uv = info.base;
+		volt->max_uv = info.base + info.pwm_range;
 	}
 }
 
@@ -181,8 +190,11 @@ nvkm_volt_ctor(const struct nvkm_volt_func *func, struct nvkm_device *device,
 	volt->func = func;
 
 	/* Assuming the non-bios device should build the voltage table later */
-	if (bios)
+	if (bios) {
 		nvkm_volt_parse_bios(bios, volt);
+		nvkm_debug(&volt->subdev, "min: %iuv max: %iuv\n",
+			   volt->min_uv, volt->max_uv);
+	}
 
 	if (volt->vid_nr) {
 		for (i = 0; i < volt->vid_nr; i++) {
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/volt/gk20a.c b/drivers/gpu/drm/nouveau/nvkm/subdev/volt/gk20a.c
index d554455326da..ce5d83cdc7cf 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/volt/gk20a.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/volt/gk20a.c
@@ -77,18 +77,19 @@ gk20a_volt_get_cvb_t_voltage(int speedo, int temp, int s_scale, int t_scale,
 	return mv;
 }
 
-int
+static int
 gk20a_volt_calc_voltage(const struct cvb_coef *coef, int speedo)
 {
+	static const int v_scale = 1000;
 	int mv;
 
 	mv = gk20a_volt_get_cvb_t_voltage(speedo, -10, 100, 10, coef);
-	mv = DIV_ROUND_UP(mv, 1000);
+	mv = DIV_ROUND_UP(mv, v_scale);
 
 	return mv * 1000;
 }
 
-int
+static int
 gk20a_volt_vid_get(struct nvkm_volt *base)
 {
 	struct gk20a_volt *volt = gk20a_volt(base);
@@ -103,7 +104,7 @@ gk20a_volt_vid_get(struct nvkm_volt *base)
 	return -EINVAL;
 }
 
-int
+static int
 gk20a_volt_vid_set(struct nvkm_volt *base, u8 vid)
 {
 	struct gk20a_volt *volt = gk20a_volt(base);
@@ -113,7 +114,7 @@ gk20a_volt_vid_set(struct nvkm_volt *base, u8 vid)
 	return regulator_set_voltage(volt->vdd, volt->base.vid[vid].uv, 1200000);
 }
 
-int
+static int
 gk20a_volt_set_id(struct nvkm_volt *base, u8 id, int condition)
 {
 	struct gk20a_volt *volt = gk20a_volt(base);
@@ -143,9 +144,9 @@ gk20a_volt = {
 };
 
 int
-_gk20a_volt_ctor(struct nvkm_device *device, int index,
-		 const struct cvb_coef *coefs, int nb_coefs,
-		 struct gk20a_volt *volt)
+gk20a_volt_ctor(struct nvkm_device *device, int index,
+		const struct cvb_coef *coefs, int nb_coefs,
+		int vmin, struct gk20a_volt *volt)
 {
 	struct nvkm_device_tegra *tdev = device->func->tegra(device);
 	int i, uv;
@@ -160,9 +161,9 @@ _gk20a_volt_ctor(struct nvkm_device *device, int index,
 	volt->base.vid_nr = nb_coefs;
 	for (i = 0; i < volt->base.vid_nr; i++) {
 		volt->base.vid[i].vid = i;
-		volt->base.vid[i].uv =
-			gk20a_volt_calc_voltage(&coefs[i],
-						tdev->gpu_speedo);
+		volt->base.vid[i].uv = max(
+			gk20a_volt_calc_voltage(&coefs[i], tdev->gpu_speedo),
+			vmin);
 		nvkm_debug(&volt->base.subdev, "%2d: vid=%d, uv=%d\n", i,
 			   volt->base.vid[i].vid, volt->base.vid[i].uv);
 	}
@@ -180,6 +181,6 @@ gk20a_volt_new(struct nvkm_device *device, int index, struct nvkm_volt **pvolt)
 		return -ENOMEM;
 	*pvolt = &volt->base;
 
-	return _gk20a_volt_ctor(device, index, gk20a_cvb_coef,
-				ARRAY_SIZE(gk20a_cvb_coef), volt);
+	return gk20a_volt_ctor(device, index, gk20a_cvb_coef,
+			       ARRAY_SIZE(gk20a_cvb_coef), 0, volt);
 }
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/volt/gk20a.h b/drivers/gpu/drm/nouveau/nvkm/subdev/volt/gk20a.h
index 0fa3b502bcf8..6a6c97f9684e 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/volt/gk20a.h
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/volt/gk20a.h
@@ -37,13 +37,8 @@ struct gk20a_volt {
 	struct regulator *vdd;
 };
 
-int _gk20a_volt_ctor(struct nvkm_device *device, int index,
-		     const struct cvb_coef *coefs, int nb_coefs,
-		     struct gk20a_volt *volt);
-
-int gk20a_volt_calc_voltage(const struct cvb_coef *coef, int speedo);
-int gk20a_volt_vid_get(struct nvkm_volt *volt);
-int gk20a_volt_vid_set(struct nvkm_volt *volt, u8 vid);
-int gk20a_volt_set_id(struct nvkm_volt *volt, u8 id, int condition);
+int gk20a_volt_ctor(struct nvkm_device *device, int index,
+		    const struct cvb_coef *coefs, int nb_coefs,
+		    int vmin, struct gk20a_volt *volt);
 
 #endif
diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/volt/gm20b.c b/drivers/gpu/drm/nouveau/nvkm/subdev/volt/gm20b.c
index 49b5ecb701e4..74db4d28930f 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/volt/gm20b.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/volt/gm20b.c
@@ -41,16 +41,52 @@ const struct cvb_coef gm20b_cvb_coef[] = {
 	/* 921600 */ { 2647676, -106455, 1632 },
 };
 
+static const struct cvb_coef gm20b_na_cvb_coef[] = {
+	/* KHz,         c0,     c1,   c2,    c3,     c4,   c5 */
+	/*  76800 */ {  814294, 8144, -940, 808, -21583, 226 },
+	/* 153600 */ {  856185, 8144, -940, 808, -21583, 226 },
+	/* 230400 */ {  898077, 8144, -940, 808, -21583, 226 },
+	/* 307200 */ {  939968, 8144, -940, 808, -21583, 226 },
+	/* 384000 */ {  981860, 8144, -940, 808, -21583, 226 },
+	/* 460800 */ { 1023751, 8144, -940, 808, -21583, 226 },
+	/* 537600 */ { 1065642, 8144, -940, 808, -21583, 226 },
+	/* 614400 */ { 1107534, 8144, -940, 808, -21583, 226 },
+	/* 691200 */ { 1149425, 8144, -940, 808, -21583, 226 },
+	/* 768000 */ { 1191317, 8144, -940, 808, -21583, 226 },
+	/* 844800 */ { 1233208, 8144, -940, 808, -21583, 226 },
+	/* 921600 */ { 1275100, 8144, -940, 808, -21583, 226 },
+	/* 998400 */ { 1316991, 8144, -940, 808, -21583, 226 },
+};
+
+const u32 speedo_to_vmin[] = {
+	/*   0,      1,      2,      3,      4, */
+	950000, 840000, 818750, 840000, 810000,
+};
+
 int
 gm20b_volt_new(struct nvkm_device *device, int index, struct nvkm_volt **pvolt)
 {
+	struct nvkm_device_tegra *tdev = device->func->tegra(device);
 	struct gk20a_volt *volt;
+	u32 vmin;
+
+	if (tdev->gpu_speedo_id >= ARRAY_SIZE(speedo_to_vmin)) {
+		nvdev_error(device, "unsupported speedo %d\n",
+			    tdev->gpu_speedo_id);
+		return -EINVAL;
+	}
 
 	volt = kzalloc(sizeof(*volt), GFP_KERNEL);
 	if (!volt)
 		return -ENOMEM;
 	*pvolt = &volt->base;
 
-	return _gk20a_volt_ctor(device, index, gm20b_cvb_coef,
-				ARRAY_SIZE(gm20b_cvb_coef), volt);
+	vmin = speedo_to_vmin[tdev->gpu_speedo_id];
+
+	if (tdev->gpu_speedo_id >= 1)
+		return gk20a_volt_ctor(device, index, gm20b_na_cvb_coef,
+				     ARRAY_SIZE(gm20b_na_cvb_coef), vmin, volt);
+	else
+		return gk20a_volt_ctor(device, index, gm20b_cvb_coef,
+					ARRAY_SIZE(gm20b_cvb_coef), vmin, volt);
 }