ath10k: split ce initialization and allocation

Definitions by which copy engine structure are
allocated do not change so it doesn't make much
sense to re-create those structures each time
device is booted (e.g. due to firmware recovery).

This should decrease chance of memory allocation
failures.

While at it remove per_transfer_context pointer
indirection. The array has been trailing the copy
engine ringbuffer structure anyway. This also
saves pointer size worth of bytes for each copy
engine ringbuffer.

Reported-By: Avery Pennarun <apenwarr@gmail.com>
Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>

1 files changed, 178 insertions, 129 deletions

diff --git a/drivers/net/wireless/ath/ath10k/ce.c b/drivers/net/wireless/ath/ath10k/ce.c
index 653a240142e5..1e4cad8632b5 100644
--- a/drivers/net/wireless/ath/ath10k/ce.c
+++ b/drivers/net/wireless/ath/ath10k/ce.c
@@ -840,34 +840,17 @@ void ath10k_ce_recv_cb_register(struct ath10k_ce_pipe *ce_state,
 
 static int ath10k_ce_init_src_ring(struct ath10k *ar,
 				   unsigned int ce_id,
-				   struct ath10k_ce_pipe *ce_state,
 				   const struct ce_attr *attr)
 {
-	struct ath10k_ce_ring *src_ring;
-	unsigned int nentries = attr->src_nentries;
-	unsigned int ce_nbytes;
-	u32 ctrl_addr = ath10k_ce_base_address(ce_id);
-	dma_addr_t base_addr;
-	char *ptr;
-
-	nentries = roundup_pow_of_two(nentries);
-
-	if (ce_state->src_ring) {
-		WARN_ON(ce_state->src_ring->nentries != nentries);
-		return 0;
-	}
+	struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
+	struct ath10k_ce_pipe *ce_state = &ar_pci->ce_states[ce_id];
+	struct ath10k_ce_ring *src_ring = ce_state->src_ring;
+	u32 nentries, ctrl_addr = ath10k_ce_base_address(ce_id);
 
-	ce_nbytes = sizeof(struct ath10k_ce_ring) + (nentries * sizeof(void *));
-	ptr = kzalloc(ce_nbytes, GFP_KERNEL);
-	if (ptr == NULL)
-		return -ENOMEM;
+	nentries = roundup_pow_of_two(attr->src_nentries);
 
-	ce_state->src_ring = (struct ath10k_ce_ring *)ptr;
-	src_ring = ce_state->src_ring;
-
-	ptr += sizeof(struct ath10k_ce_ring);
-	src_ring->nentries = nentries;
-	src_ring->nentries_mask = nentries - 1;
+	memset(src_ring->per_transfer_context, 0,
+	       nentries * sizeof(*src_ring->per_transfer_context));
 
 	src_ring->sw_index = ath10k_ce_src_ring_read_index_get(ar, ctrl_addr);
 	src_ring->sw_index &= src_ring->nentries_mask;
@@ -877,7 +860,74 @@ static int ath10k_ce_init_src_ring(struct ath10k *ar,
 		ath10k_ce_src_ring_write_index_get(ar, ctrl_addr);
 	src_ring->write_index &= src_ring->nentries_mask;
 
-	src_ring->per_transfer_context = (void **)ptr;
+	ath10k_ce_src_ring_base_addr_set(ar, ctrl_addr,
+					 src_ring->base_addr_ce_space);
+	ath10k_ce_src_ring_size_set(ar, ctrl_addr, nentries);
+	ath10k_ce_src_ring_dmax_set(ar, ctrl_addr, attr->src_sz_max);
+	ath10k_ce_src_ring_byte_swap_set(ar, ctrl_addr, 0);
+	ath10k_ce_src_ring_lowmark_set(ar, ctrl_addr, 0);
+	ath10k_ce_src_ring_highmark_set(ar, ctrl_addr, nentries);
+
+	ath10k_dbg(ATH10K_DBG_BOOT,
+		   "boot init ce src ring id %d entries %d base_addr %p\n",
+		   ce_id, nentries, src_ring->base_addr_owner_space);
+
+	return 0;
+}
+
+static int ath10k_ce_init_dest_ring(struct ath10k *ar,
+				    unsigned int ce_id,
+				    const struct ce_attr *attr)
+{
+	struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
+	struct ath10k_ce_pipe *ce_state = &ar_pci->ce_states[ce_id];
+	struct ath10k_ce_ring *dest_ring = ce_state->dest_ring;
+	u32 nentries, ctrl_addr = ath10k_ce_base_address(ce_id);
+
+	nentries = roundup_pow_of_two(attr->dest_nentries);
+
+	memset(dest_ring->per_transfer_context, 0,
+	       nentries * sizeof(*dest_ring->per_transfer_context));
+
+	dest_ring->sw_index = ath10k_ce_dest_ring_read_index_get(ar, ctrl_addr);
+	dest_ring->sw_index &= dest_ring->nentries_mask;
+	dest_ring->write_index =
+		ath10k_ce_dest_ring_write_index_get(ar, ctrl_addr);
+	dest_ring->write_index &= dest_ring->nentries_mask;
+
+	ath10k_ce_dest_ring_base_addr_set(ar, ctrl_addr,
+					  dest_ring->base_addr_ce_space);
+	ath10k_ce_dest_ring_size_set(ar, ctrl_addr, nentries);
+	ath10k_ce_dest_ring_byte_swap_set(ar, ctrl_addr, 0);
+	ath10k_ce_dest_ring_lowmark_set(ar, ctrl_addr, 0);
+	ath10k_ce_dest_ring_highmark_set(ar, ctrl_addr, nentries);
+
+	ath10k_dbg(ATH10K_DBG_BOOT,
+		   "boot ce dest ring id %d entries %d base_addr %p\n",
+		   ce_id, nentries, dest_ring->base_addr_owner_space);
+
+	return 0;
+}
+
+static struct ath10k_ce_ring *
+ath10k_ce_alloc_src_ring(struct ath10k *ar, unsigned int ce_id,
+			 const struct ce_attr *attr)
+{
+	struct ath10k_ce_ring *src_ring;
+	u32 nentries = attr->src_nentries;
+	dma_addr_t base_addr;
+
+	nentries = roundup_pow_of_two(nentries);
+
+	src_ring = kzalloc(sizeof(*src_ring) +
+			   (nentries *
+			    sizeof(*src_ring->per_transfer_context)),
+			   GFP_KERNEL);
+	if (src_ring == NULL)
+		return ERR_PTR(-ENOMEM);
+
+	src_ring->nentries = nentries;
+	src_ring->nentries_mask = nentries - 1;
 
 	/*
 	 * Legacy platforms that do not support cache
@@ -889,9 +939,8 @@ static int ath10k_ce_init_src_ring(struct ath10k *ar,
 				    CE_DESC_RING_ALIGN),
 				   &base_addr, GFP_KERNEL);
 	if (!src_ring->base_addr_owner_space_unaligned) {
-		kfree(ce_state->src_ring);
-		ce_state->src_ring = NULL;
-		return -ENOMEM;
+		kfree(src_ring);
+		return ERR_PTR(-ENOMEM);
 	}
 
 	src_ring->base_addr_ce_space_unaligned = base_addr;
@@ -916,69 +965,37 @@ static int ath10k_ce_init_src_ring(struct ath10k *ar,
 				   CE_DESC_RING_ALIGN),
 				  src_ring->base_addr_owner_space,
 				  src_ring->base_addr_ce_space);
-		kfree(ce_state->src_ring);
-		ce_state->src_ring = NULL;
-		return -ENOMEM;
+		kfree(src_ring);
+		return ERR_PTR(-ENOMEM);
 	}
 
 	src_ring->shadow_base = PTR_ALIGN(
 			src_ring->shadow_base_unaligned,
 			CE_DESC_RING_ALIGN);
 
-	ath10k_ce_src_ring_base_addr_set(ar, ctrl_addr,
-					 src_ring->base_addr_ce_space);
-	ath10k_ce_src_ring_size_set(ar, ctrl_addr, nentries);
-	ath10k_ce_src_ring_dmax_set(ar, ctrl_addr, attr->src_sz_max);
-	ath10k_ce_src_ring_byte_swap_set(ar, ctrl_addr, 0);
-	ath10k_ce_src_ring_lowmark_set(ar, ctrl_addr, 0);
-	ath10k_ce_src_ring_highmark_set(ar, ctrl_addr, nentries);
-
-	ath10k_dbg(ATH10K_DBG_BOOT,
-		   "boot ce src ring id %d entries %d base_addr %p\n",
-		   ce_id, nentries, src_ring->base_addr_owner_space);
-
-	return 0;
+	return src_ring;
 }
 
-static int ath10k_ce_init_dest_ring(struct ath10k *ar,
-				    unsigned int ce_id,
-				    struct ath10k_ce_pipe *ce_state,
-				    const struct ce_attr *attr)
+static struct ath10k_ce_ring *
+ath10k_ce_alloc_dest_ring(struct ath10k *ar, unsigned int ce_id,
+			  const struct ce_attr *attr)
 {
 	struct ath10k_ce_ring *dest_ring;
-	unsigned int nentries = attr->dest_nentries;
-	unsigned int ce_nbytes;
-	u32 ctrl_addr = ath10k_ce_base_address(ce_id);
+	u32 nentries;
 	dma_addr_t base_addr;
-	char *ptr;
 
-	nentries = roundup_pow_of_two(nentries);
+	nentries = roundup_pow_of_two(attr->dest_nentries);
 
-	if (ce_state->dest_ring) {
-		WARN_ON(ce_state->dest_ring->nentries != nentries);
-		return 0;
-	}
-
-	ce_nbytes = sizeof(struct ath10k_ce_ring) + (nentries * sizeof(void *));
-	ptr = kzalloc(ce_nbytes, GFP_KERNEL);
-	if (ptr == NULL)
-		return -ENOMEM;
-
-	ce_state->dest_ring = (struct ath10k_ce_ring *)ptr;
-	dest_ring = ce_state->dest_ring;
+	dest_ring = kzalloc(sizeof(*dest_ring) +
+			    (nentries *
+			     sizeof(*dest_ring->per_transfer_context)),
+			    GFP_KERNEL);
+	if (dest_ring == NULL)
+		return ERR_PTR(-ENOMEM);
 
-	ptr += sizeof(struct ath10k_ce_ring);
 	dest_ring->nentries = nentries;
 	dest_ring->nentries_mask = nentries - 1;
 
-	dest_ring->sw_index = ath10k_ce_dest_ring_read_index_get(ar, ctrl_addr);
-	dest_ring->sw_index &= dest_ring->nentries_mask;
-	dest_ring->write_index =
-		ath10k_ce_dest_ring_write_index_get(ar, ctrl_addr);
-	dest_ring->write_index &= dest_ring->nentries_mask;
-
-	dest_ring->per_transfer_context = (void **)ptr;
-
 	/*
 	 * Legacy platforms that do not support cache
 	 * coherent DMA are unsupported
@@ -989,9 +1006,8 @@ static int ath10k_ce_init_dest_ring(struct ath10k *ar,
 				    CE_DESC_RING_ALIGN),
 				   &base_addr, GFP_KERNEL);
 	if (!dest_ring->base_addr_owner_space_unaligned) {
-		kfree(ce_state->dest_ring);
-		ce_state->dest_ring = NULL;
-		return -ENOMEM;
+		kfree(dest_ring);
+		return ERR_PTR(-ENOMEM);
 	}
 
 	dest_ring->base_addr_ce_space_unaligned = base_addr;
@@ -1010,39 +1026,7 @@ static int ath10k_ce_init_dest_ring(struct ath10k *ar,
 			dest_ring->base_addr_ce_space_unaligned,
 			CE_DESC_RING_ALIGN);
 
-	ath10k_ce_dest_ring_base_addr_set(ar, ctrl_addr,
-					  dest_ring->base_addr_ce_space);
-	ath10k_ce_dest_ring_size_set(ar, ctrl_addr, nentries);
-	ath10k_ce_dest_ring_byte_swap_set(ar, ctrl_addr, 0);
-	ath10k_ce_dest_ring_lowmark_set(ar, ctrl_addr, 0);
-	ath10k_ce_dest_ring_highmark_set(ar, ctrl_addr, nentries);
-
-	ath10k_dbg(ATH10K_DBG_BOOT,
-		   "boot ce dest ring id %d entries %d base_addr %p\n",
-		   ce_id, nentries, dest_ring->base_addr_owner_space);
-
-	return 0;
-}
-
-static struct ath10k_ce_pipe *ath10k_ce_init_state(struct ath10k *ar,
-					     unsigned int ce_id,
-					     const struct ce_attr *attr)
-{
-	struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
-	struct ath10k_ce_pipe *ce_state = &ar_pci->ce_states[ce_id];
-	u32 ctrl_addr = ath10k_ce_base_address(ce_id);
-
-	spin_lock_bh(&ar_pci->ce_lock);
-
-	ce_state->ar = ar;
-	ce_state->id = ce_id;
-	ce_state->ctrl_addr = ctrl_addr;
-	ce_state->attr_flags = attr->flags;
-	ce_state->src_sz_max = attr->src_sz_max;
-
-	spin_unlock_bh(&ar_pci->ce_lock);
-
-	return ce_state;
+	return dest_ring;
 }
 
 /*
@@ -1052,11 +1036,11 @@ static struct ath10k_ce_pipe *ath10k_ce_init_state(struct ath10k *ar,
  * initialization. It may be that only one side or the other is
  * initialized by software/firmware.
  */
-struct ath10k_ce_pipe *ath10k_ce_init(struct ath10k *ar,
-				unsigned int ce_id,
-				const struct ce_attr *attr)
+int ath10k_ce_init_pipe(struct ath10k *ar, unsigned int ce_id,
+			const struct ce_attr *attr)
 {
-	struct ath10k_ce_pipe *ce_state;
+	struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
+	struct ath10k_ce_pipe *ce_state = &ar_pci->ce_states[ce_id];
 	int ret;
 
 	/*
@@ -1072,44 +1056,109 @@ struct ath10k_ce_pipe *ath10k_ce_init(struct ath10k *ar,
 
 	ret = ath10k_pci_wake(ar);
 	if (ret)
-		return NULL;
+		return ret;
 
-	ce_state = ath10k_ce_init_state(ar, ce_id, attr);
-	if (!ce_state) {
-		ath10k_err("Failed to initialize CE state for ID: %d\n", ce_id);
-		goto out;
-	}
+	spin_lock_bh(&ar_pci->ce_lock);
+	ce_state->ar = ar;
+	ce_state->id = ce_id;
+	ce_state->ctrl_addr = ath10k_ce_base_address(ce_id);
+	ce_state->attr_flags = attr->flags;
+	ce_state->src_sz_max = attr->src_sz_max;
+	spin_unlock_bh(&ar_pci->ce_lock);
 
 	if (attr->src_nentries) {
-		ret = ath10k_ce_init_src_ring(ar, ce_id, ce_state, attr);
+		ret = ath10k_ce_init_src_ring(ar, ce_id, attr);
 		if (ret) {
 			ath10k_err("Failed to initialize CE src ring for ID: %d (%d)\n",
 				   ce_id, ret);
-			ath10k_ce_deinit(ce_state);
-			ce_state = NULL;
 			goto out;
 		}
 	}
 
 	if (attr->dest_nentries) {
-		ret = ath10k_ce_init_dest_ring(ar, ce_id, ce_state, attr);
+		ret = ath10k_ce_init_dest_ring(ar, ce_id, attr);
 		if (ret) {
 			ath10k_err("Failed to initialize CE dest ring for ID: %d (%d)\n",
 				   ce_id, ret);
-			ath10k_ce_deinit(ce_state);
-			ce_state = NULL;
 			goto out;
 		}
 	}
 
 out:
 	ath10k_pci_sleep(ar);
-	return ce_state;
+	return ret;
 }
 
-void ath10k_ce_deinit(struct ath10k_ce_pipe *ce_state)
+static void ath10k_ce_deinit_src_ring(struct ath10k *ar, unsigned int ce_id)
 {
-	struct ath10k *ar = ce_state->ar;
+	u32 ctrl_addr = ath10k_ce_base_address(ce_id);
+
+	ath10k_ce_src_ring_base_addr_set(ar, ctrl_addr, 0);
+	ath10k_ce_src_ring_size_set(ar, ctrl_addr, 0);
+	ath10k_ce_src_ring_dmax_set(ar, ctrl_addr, 0);
+	ath10k_ce_src_ring_highmark_set(ar, ctrl_addr, 0);
+}
+
+static void ath10k_ce_deinit_dest_ring(struct ath10k *ar, unsigned int ce_id)
+{
+	u32 ctrl_addr = ath10k_ce_base_address(ce_id);
+
+	ath10k_ce_dest_ring_base_addr_set(ar, ctrl_addr, 0);
+	ath10k_ce_dest_ring_size_set(ar, ctrl_addr, 0);
+	ath10k_ce_dest_ring_highmark_set(ar, ctrl_addr, 0);
+}
+
+void ath10k_ce_deinit_pipe(struct ath10k *ar, unsigned int ce_id)
+{
+	int ret;
+
+	ret = ath10k_pci_wake(ar);
+	if (ret)
+		return;
+
+	ath10k_ce_deinit_src_ring(ar, ce_id);
+	ath10k_ce_deinit_dest_ring(ar, ce_id);
+
+	ath10k_pci_sleep(ar);
+}
+
+int ath10k_ce_alloc_pipe(struct ath10k *ar, int ce_id,
+			 const struct ce_attr *attr)
+{
+	struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
+	struct ath10k_ce_pipe *ce_state = &ar_pci->ce_states[ce_id];
+	int ret;
+
+	if (attr->src_nentries) {
+		ce_state->src_ring = ath10k_ce_alloc_src_ring(ar, ce_id, attr);
+		if (IS_ERR(ce_state->src_ring)) {
+			ret = PTR_ERR(ce_state->src_ring);
+			ath10k_err("failed to allocate copy engine source ring %d: %d\n",
+				   ce_id, ret);
+			ce_state->src_ring = NULL;
+			return ret;
+		}
+	}
+
+	if (attr->dest_nentries) {
+		ce_state->dest_ring = ath10k_ce_alloc_dest_ring(ar, ce_id,
+								attr);
+		if (IS_ERR(ce_state->dest_ring)) {
+			ret = PTR_ERR(ce_state->dest_ring);
+			ath10k_err("failed to allocate copy engine destination ring %d: %d\n",
+				   ce_id, ret);
+			ce_state->dest_ring = NULL;
+			return ret;
+		}
+	}
+
+	return 0;
+}
+
+void ath10k_ce_free_pipe(struct ath10k *ar, int ce_id)
+{
+	struct ath10k_pci *ar_pci = ath10k_pci_priv(ar);
+	struct ath10k_ce_pipe *ce_state = &ar_pci->ce_states[ce_id];
 
 	if (ce_state->src_ring) {
 		kfree(ce_state->src_ring->shadow_base_unaligned);