usb: mtu3: support 36-bit DMA address

add support for 36-bit DMA address

[ Felipe Balbi: fix printk format for dma_addr_t ]

Signed-off-by: Chunfeng Yun <chunfeng.yun@mediatek.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

4 files changed, 142 insertions, 21 deletions

diff --git a/drivers/usb/mtu3/mtu3.h b/drivers/usb/mtu3/mtu3.h
index 2795294ec92a..ef2dc92a2109 100644
--- a/drivers/usb/mtu3/mtu3.h
+++ b/drivers/usb/mtu3/mtu3.h
@@ -46,6 +46,9 @@ struct mtu3_request;
 #define	MU3D_EP_RXCR1(epnum)	(U3D_RX1CSR1 + (((epnum) - 1) * 0x10))
 #define	MU3D_EP_RXCR2(epnum)	(U3D_RX1CSR2 + (((epnum) - 1) * 0x10))
 
+#define USB_QMU_TQHIAR(epnum)	(U3D_TXQHIAR1 + (((epnum) - 1) * 0x4))
+#define USB_QMU_RQHIAR(epnum)	(U3D_RXQHIAR1 + (((epnum) - 1) * 0x4))
+
 #define USB_QMU_RQCSR(epnum)	(U3D_RXQCSR1 + (((epnum) - 1) * 0x10))
 #define USB_QMU_RQSAR(epnum)	(U3D_RXQSAR1 + (((epnum) - 1) * 0x10))
 #define USB_QMU_RQCPR(epnum)	(U3D_RXQCPR1 + (((epnum) - 1) * 0x10))
@@ -138,23 +141,33 @@ struct mtu3_fifo_info {
  *	Checksum value is calculated over the 16 bytes of the GPD by default;
  * @data_buf_len (RX ONLY): This value indicates the length of
  *	the assigned data buffer
+ * @tx_ext_addr (TX ONLY): [3:0] are 4 extension bits of @buffer,
+ *	[7:4] are 4 extension bits of @next_gpd
  * @next_gpd: Physical address of the next GPD
  * @buffer: Physical address of the data buffer
  * @buf_len:
  *	(TX): This value indicates the length of the assigned data buffer
  *	(RX): The total length of data received
  * @ext_len: reserved
+ * @rx_ext_addr(RX ONLY): [3:0] are 4 extension bits of @buffer,
+ *	[7:4] are 4 extension bits of @next_gpd
  * @ext_flag:
  *	bit5 (TX ONLY): Zero Length Packet (ZLP),
  */
 struct qmu_gpd {
 	__u8 flag;
 	__u8 chksum;
-	__le16 data_buf_len;
+	union {
+		__le16 data_buf_len;
+		__le16 tx_ext_addr;
+	};
 	__le32 next_gpd;
 	__le32 buffer;
 	__le16 buf_len;
-	__u8 ext_len;
+	union {
+		__u8 ext_len;
+		__u8 rx_ext_addr;
+	};
 	__u8 ext_flag;
 } __packed;
 
diff --git a/drivers/usb/mtu3/mtu3_core.c b/drivers/usb/mtu3/mtu3_core.c
index 947579842ad7..cd4528f5f337 100644
--- a/drivers/usb/mtu3/mtu3_core.c
+++ b/drivers/usb/mtu3/mtu3_core.c
@@ -17,6 +17,7 @@
  *
  */
 
+#include <linux/dma-mapping.h>
 #include <linux/kernel.h>
 #include <linux/module.h>
 #include <linux/of_address.h>
@@ -759,7 +760,31 @@ static void mtu3_hw_exit(struct mtu3 *mtu)
 	mtu3_mem_free(mtu);
 }
 
-/*-------------------------------------------------------------------------*/
+/**
+ * we set 32-bit DMA mask by default, here check whether the controller
+ * supports 36-bit DMA or not, if it does, set 36-bit DMA mask.
+ */
+static int mtu3_set_dma_mask(struct mtu3 *mtu)
+{
+	struct device *dev = mtu->dev;
+	bool is_36bit = false;
+	int ret = 0;
+	u32 value;
+
+	value = mtu3_readl(mtu->mac_base, U3D_MISC_CTRL);
+	if (value & DMA_ADDR_36BIT) {
+		is_36bit = true;
+		ret = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(36));
+		/* If set 36-bit DMA mask fails, fall back to 32-bit DMA mask */
+		if (ret) {
+			is_36bit = false;
+			ret = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32));
+		}
+	}
+	dev_info(dev, "dma mask: %s bits\n", is_36bit ? "36" : "32");
+
+	return ret;
+}
 
 int ssusb_gadget_init(struct ssusb_mtk *ssusb)
 {
@@ -820,6 +845,12 @@ int ssusb_gadget_init(struct ssusb_mtk *ssusb)
 		return ret;
 	}
 
+	ret = mtu3_set_dma_mask(mtu);
+	if (ret) {
+		dev_err(dev, "mtu3 set dma_mask failed:%d\n", ret);
+		goto dma_mask_err;
+	}
+
 	ret = devm_request_irq(dev, mtu->irq, mtu3_irq, 0, dev_name(dev), mtu);
 	if (ret) {
 		dev_err(dev, "request irq %d failed!\n", mtu->irq);
@@ -845,6 +876,7 @@ int ssusb_gadget_init(struct ssusb_mtk *ssusb)
 gadget_err:
 	device_init_wakeup(dev, false);
 
+dma_mask_err:
 irq_err:
 	mtu3_hw_exit(mtu);
 	ssusb->u3d = NULL;
diff --git a/drivers/usb/mtu3/mtu3_hw_regs.h b/drivers/usb/mtu3/mtu3_hw_regs.h
index 06b29664470f..b6059752dc12 100644
--- a/drivers/usb/mtu3/mtu3_hw_regs.h
+++ b/drivers/usb/mtu3/mtu3_hw_regs.h
@@ -58,6 +58,8 @@
 #define U3D_QCR1		(SSUSB_DEV_BASE + 0x0404)
 #define U3D_QCR2		(SSUSB_DEV_BASE + 0x0408)
 #define U3D_QCR3		(SSUSB_DEV_BASE + 0x040C)
+#define U3D_TXQHIAR1		(SSUSB_DEV_BASE + 0x0484)
+#define U3D_RXQHIAR1		(SSUSB_DEV_BASE + 0x04C4)
 
 #define U3D_TXQCSR1		(SSUSB_DEV_BASE + 0x0510)
 #define U3D_TXQSAR1		(SSUSB_DEV_BASE + 0x0514)
@@ -189,6 +191,13 @@
 #define QMU_RX_COZ(x)		(BIT(16) << (x))
 #define QMU_RX_ZLP(x)		(BIT(0) << (x))
 
+/* U3D_TXQHIAR1 */
+/* U3D_RXQHIAR1 */
+#define QMU_LAST_DONE_PTR_HI(x)	(((x) >> 16) & 0xf)
+#define QMU_CUR_GPD_ADDR_HI(x)	(((x) >> 8) & 0xf)
+#define QMU_START_ADDR_HI_MSK	GENMASK(3, 0)
+#define QMU_START_ADDR_HI(x)	(((x) & 0xf) << 0)
+
 /* U3D_TXQCSR1 */
 /* U3D_RXQCSR1 */
 #define QMU_Q_ACTIVE		BIT(15)
@@ -225,6 +234,7 @@
 #define CAP_TX_EP_NUM(x)	((x) & 0x1f)
 
 /* U3D_MISC_CTRL */
+#define DMA_ADDR_36BIT		BIT(31)
 #define VBUS_ON			BIT(1)
 #define VBUS_FRC_EN		BIT(0)
 
diff --git a/drivers/usb/mtu3/mtu3_qmu.c b/drivers/usb/mtu3/mtu3_qmu.c
index 7d9ba8a52368..42145a3f1422 100644
--- a/drivers/usb/mtu3/mtu3_qmu.c
+++ b/drivers/usb/mtu3/mtu3_qmu.c
@@ -40,7 +40,58 @@
 #define GPD_FLAGS_IOC	BIT(7)
 
 #define GPD_EXT_FLAG_ZLP	BIT(5)
+#define GPD_EXT_NGP(x)		(((x) & 0xf) << 4)
+#define GPD_EXT_BUF(x)		(((x) & 0xf) << 0)
 
+#define HILO_GEN64(hi, lo) (((u64)(hi) << 32) + (lo))
+#define HILO_DMA(hi, lo)	\
+	((dma_addr_t)HILO_GEN64((le32_to_cpu(hi)), (le32_to_cpu(lo))))
+
+static dma_addr_t read_txq_cur_addr(void __iomem *mbase, u8 epnum)
+{
+	u32 txcpr;
+	u32 txhiar;
+
+	txcpr = mtu3_readl(mbase, USB_QMU_TQCPR(epnum));
+	txhiar = mtu3_readl(mbase, USB_QMU_TQHIAR(epnum));
+
+	return HILO_DMA(QMU_CUR_GPD_ADDR_HI(txhiar), txcpr);
+}
+
+static dma_addr_t read_rxq_cur_addr(void __iomem *mbase, u8 epnum)
+{
+	u32 rxcpr;
+	u32 rxhiar;
+
+	rxcpr = mtu3_readl(mbase, USB_QMU_RQCPR(epnum));
+	rxhiar = mtu3_readl(mbase, USB_QMU_RQHIAR(epnum));
+
+	return HILO_DMA(QMU_CUR_GPD_ADDR_HI(rxhiar), rxcpr);
+}
+
+static void write_txq_start_addr(void __iomem *mbase, u8 epnum, dma_addr_t dma)
+{
+	u32 tqhiar;
+
+	mtu3_writel(mbase, USB_QMU_TQSAR(epnum),
+		    cpu_to_le32(lower_32_bits(dma)));
+	tqhiar = mtu3_readl(mbase, USB_QMU_TQHIAR(epnum));
+	tqhiar &= ~QMU_START_ADDR_HI_MSK;
+	tqhiar |= QMU_START_ADDR_HI(upper_32_bits(dma));
+	mtu3_writel(mbase, USB_QMU_TQHIAR(epnum), tqhiar);
+}
+
+static void write_rxq_start_addr(void __iomem *mbase, u8 epnum, dma_addr_t dma)
+{
+	u32 rqhiar;
+
+	mtu3_writel(mbase, USB_QMU_RQSAR(epnum),
+		    cpu_to_le32(lower_32_bits(dma)));
+	rqhiar = mtu3_readl(mbase, USB_QMU_RQHIAR(epnum));
+	rqhiar &= ~QMU_START_ADDR_HI_MSK;
+	rqhiar |= QMU_START_ADDR_HI(upper_32_bits(dma));
+	mtu3_writel(mbase, USB_QMU_RQHIAR(epnum), rqhiar);
+}
 
 static struct qmu_gpd *gpd_dma_to_virt(struct mtu3_gpd_ring *ring,
 		dma_addr_t dma_addr)
@@ -193,21 +244,27 @@ static int mtu3_prepare_tx_gpd(struct mtu3_ep *mep, struct mtu3_request *mreq)
 	struct mtu3_gpd_ring *ring = &mep->gpd_ring;
 	struct qmu_gpd *gpd = ring->enqueue;
 	struct usb_request *req = &mreq->request;
+	dma_addr_t enq_dma;
+	u16 ext_addr;
 
 	/* set all fields to zero as default value */
 	memset(gpd, 0, sizeof(*gpd));
 
-	gpd->buffer = cpu_to_le32((u32)req->dma);
+	gpd->buffer = cpu_to_le32(lower_32_bits(req->dma));
+	ext_addr = GPD_EXT_BUF(upper_32_bits(req->dma));
 	gpd->buf_len = cpu_to_le16(req->length);
 	gpd->flag |= GPD_FLAGS_IOC;
 
 	/* get the next GPD */
 	enq = advance_enq_gpd(ring);
-	dev_dbg(mep->mtu->dev, "TX-EP%d queue gpd=%p, enq=%p\n",
-		mep->epnum, gpd, enq);
+	enq_dma = gpd_virt_to_dma(ring, enq);
+	dev_dbg(mep->mtu->dev, "TX-EP%d queue gpd=%p, enq=%p, qdma=%pad\n",
+		mep->epnum, gpd, enq, enq_dma);
 
 	enq->flag &= ~GPD_FLAGS_HWO;
-	gpd->next_gpd = cpu_to_le32((u32)gpd_virt_to_dma(ring, enq));
+	gpd->next_gpd = cpu_to_le32(lower_32_bits(enq_dma));
+	ext_addr |= GPD_EXT_NGP(upper_32_bits(enq_dma));
+	gpd->tx_ext_addr = cpu_to_le16(ext_addr);
 
 	if (req->zero)
 		gpd->ext_flag |= GPD_EXT_FLAG_ZLP;
@@ -226,21 +283,27 @@ static int mtu3_prepare_rx_gpd(struct mtu3_ep *mep, struct mtu3_request *mreq)
 	struct mtu3_gpd_ring *ring = &mep->gpd_ring;
 	struct qmu_gpd *gpd = ring->enqueue;
 	struct usb_request *req = &mreq->request;
+	dma_addr_t enq_dma;
+	u16 ext_addr;
 
 	/* set all fields to zero as default value */
 	memset(gpd, 0, sizeof(*gpd));
 
-	gpd->buffer = cpu_to_le32((u32)req->dma);
+	gpd->buffer = cpu_to_le32(lower_32_bits(req->dma));
+	ext_addr = GPD_EXT_BUF(upper_32_bits(req->dma));
 	gpd->data_buf_len = cpu_to_le16(req->length);
 	gpd->flag |= GPD_FLAGS_IOC;
 
 	/* get the next GPD */
 	enq = advance_enq_gpd(ring);
-	dev_dbg(mep->mtu->dev, "RX-EP%d queue gpd=%p, enq=%p\n",
-		mep->epnum, gpd, enq);
+	enq_dma = gpd_virt_to_dma(ring, enq);
+	dev_dbg(mep->mtu->dev, "RX-EP%d queue gpd=%p, enq=%p, qdma=%pad\n",
+		mep->epnum, gpd, enq, enq_dma);
 
 	enq->flag &= ~GPD_FLAGS_HWO;
-	gpd->next_gpd = cpu_to_le32((u32)gpd_virt_to_dma(ring, enq));
+	gpd->next_gpd = cpu_to_le32(lower_32_bits(enq_dma));
+	ext_addr |= GPD_EXT_NGP(upper_32_bits(enq_dma));
+	gpd->rx_ext_addr = cpu_to_le16(ext_addr);
 	gpd->chksum = qmu_calc_checksum((u8 *)gpd);
 	gpd->flag |= GPD_FLAGS_HWO;
 
@@ -267,8 +330,8 @@ int mtu3_qmu_start(struct mtu3_ep *mep)
 
 	if (mep->is_in) {
 		/* set QMU start address */
-		mtu3_writel(mbase, USB_QMU_TQSAR(mep->epnum), ring->dma);
-		mtu3_setbits(mbase, MU3D_EP_TXCR0(mep->epnum), TX_DMAREQEN);
+		write_txq_start_addr(mbase, epnum, ring->dma);
+		mtu3_setbits(mbase, MU3D_EP_TXCR0(epnum), TX_DMAREQEN);
 		mtu3_setbits(mbase, U3D_QCR0, QMU_TX_CS_EN(epnum));
 		/* send zero length packet according to ZLP flag in GPD */
 		mtu3_setbits(mbase, U3D_QCR1, QMU_TX_ZLP(epnum));
@@ -282,8 +345,8 @@ int mtu3_qmu_start(struct mtu3_ep *mep)
 		mtu3_writel(mbase, USB_QMU_TQCSR(epnum), QMU_Q_START);
 
 	} else {
-		mtu3_writel(mbase, USB_QMU_RQSAR(mep->epnum), ring->dma);
-		mtu3_setbits(mbase, MU3D_EP_RXCR0(mep->epnum), RX_DMAREQEN);
+		write_rxq_start_addr(mbase, epnum, ring->dma);
+		mtu3_setbits(mbase, MU3D_EP_RXCR0(epnum), RX_DMAREQEN);
 		mtu3_setbits(mbase, U3D_QCR0, QMU_RX_CS_EN(epnum));
 		/* don't expect ZLP */
 		mtu3_clrbits(mbase, U3D_QCR3, QMU_RX_ZLP(epnum));
@@ -353,9 +416,9 @@ static void qmu_tx_zlp_error_handler(struct mtu3 *mtu, u8 epnum)
 	struct mtu3_gpd_ring *ring = &mep->gpd_ring;
 	void __iomem *mbase = mtu->mac_base;
 	struct qmu_gpd *gpd_current = NULL;
-	dma_addr_t gpd_dma = mtu3_readl(mbase, USB_QMU_TQCPR(epnum));
 	struct usb_request *req = NULL;
 	struct mtu3_request *mreq;
+	dma_addr_t cur_gpd_dma;
 	u32 txcsr = 0;
 	int ret;
 
@@ -365,7 +428,8 @@ static void qmu_tx_zlp_error_handler(struct mtu3 *mtu, u8 epnum)
 	else
 		return;
 
-	gpd_current = gpd_dma_to_virt(ring, gpd_dma);
+	cur_gpd_dma = read_txq_cur_addr(mbase, epnum);
+	gpd_current = gpd_dma_to_virt(ring, cur_gpd_dma);
 
 	if (le16_to_cpu(gpd_current->buf_len) != 0) {
 		dev_err(mtu->dev, "TX EP%d buffer length error(!=0)\n", epnum);
@@ -408,12 +472,13 @@ static void qmu_done_tx(struct mtu3 *mtu, u8 epnum)
 	void __iomem *mbase = mtu->mac_base;
 	struct qmu_gpd *gpd = ring->dequeue;
 	struct qmu_gpd *gpd_current = NULL;
-	dma_addr_t gpd_dma = mtu3_readl(mbase, USB_QMU_TQCPR(epnum));
 	struct usb_request *request = NULL;
 	struct mtu3_request *mreq;
+	dma_addr_t cur_gpd_dma;
 
 	/*transfer phy address got from QMU register to virtual address */
-	gpd_current = gpd_dma_to_virt(ring, gpd_dma);
+	cur_gpd_dma = read_txq_cur_addr(mbase, epnum);
+	gpd_current = gpd_dma_to_virt(ring, cur_gpd_dma);
 
 	dev_dbg(mtu->dev, "%s EP%d, last=%p, current=%p, enq=%p\n",
 		__func__, epnum, gpd, gpd_current, ring->enqueue);
@@ -446,11 +511,12 @@ static void qmu_done_rx(struct mtu3 *mtu, u8 epnum)
 	void __iomem *mbase = mtu->mac_base;
 	struct qmu_gpd *gpd = ring->dequeue;
 	struct qmu_gpd *gpd_current = NULL;
-	dma_addr_t gpd_dma = mtu3_readl(mbase, USB_QMU_RQCPR(epnum));
 	struct usb_request *req = NULL;
 	struct mtu3_request *mreq;
+	dma_addr_t cur_gpd_dma;
 
-	gpd_current = gpd_dma_to_virt(ring, gpd_dma);
+	cur_gpd_dma = read_rxq_cur_addr(mbase, epnum);
+	gpd_current = gpd_dma_to_virt(ring, cur_gpd_dma);
 
 	dev_dbg(mtu->dev, "%s EP%d, last=%p, current=%p, enq=%p\n",
 		__func__, epnum, gpd, gpd_current, ring->enqueue);