[net-next v3 PATCH] octeontx2-pf: Use hardware register for CQE count

[net-next v3 PATCH] octeontx2-pf: Use hardware register for CQE count

[Geetha sowjanya]

Current driver uses software CQ head pointer to poll on CQE
header in memory to determine if CQE is valid. Software needs
to make sure, that the reads of the CQE do not get re-ordered
so much that it ends up with an inconsistent view of the CQE.
To ensure that DMB barrier after read to first CQE cacheline
and before reading of the rest of the CQE is needed.
But having barrier for every CQE read will impact the performance,
instead use hardware CQ head and tail pointers to find the
valid number of CQEs.

Signed-off-by: Geetha sowjanya <gakula@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
---
v1->v2:
-Fixed compilation warnings.

v2->v3:
-Fixed changelog.

 .../marvell/octeontx2/nic/otx2_common.c       |  3 +
 .../marvell/octeontx2/nic/otx2_common.h       |  1 +
 .../marvell/octeontx2/nic/otx2_txrx.c         | 69 +++++++++++++++++--
 .../marvell/octeontx2/nic/otx2_txrx.h         |  5 ++
 include/linux/soc/marvell/octeontx2/asm.h     | 14 ++++
 5 files changed, 85 insertions(+), 7 deletions(-)

diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c
index 78df173e6df2..4c3dbade8cfb 100644
--- a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c
+++ b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c
@@ -1006,6 +1006,9 @@ int otx2_config_nix_queues(struct otx2_nic *pfvf)
 			return err;
 	}
 
+	pfvf->cq_op_addr = (__force u64 *)otx2_get_regaddr(pfvf,
+							   NIX_LF_CQ_OP_STATUS);
+
 	/* Initialize work queue for receive buffer refill */
 	pfvf->refill_wrk = devm_kcalloc(pfvf->dev, pfvf->qset.cq_cnt,
 					sizeof(struct refill_work), GFP_KERNEL);
diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.h b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.h
index 8e51a1db7e29..ef855dc4123a 100644
--- a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.h
+++ b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.h
@@ -337,6 +337,7 @@ struct otx2_nic {
 #define OTX2_FLAG_TC_MATCHALL_INGRESS_ENABLED	BIT_ULL(13)
 #define OTX2_FLAG_DMACFLTR_SUPPORT		BIT_ULL(14)
 	u64			flags;
+	u64			*cq_op_addr;
 
 	struct otx2_qset	qset;
 	struct otx2_hw		hw;
diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.c b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.c
index f42b1d4e0c67..3f3ec8ffc4dd 100644
--- a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.c
+++ b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.c
@@ -18,6 +18,31 @@
 
 #define CQE_ADDR(CQ, idx) ((CQ)->cqe_base + ((CQ)->cqe_size * (idx)))
 
+static int otx2_nix_cq_op_status(struct otx2_nic *pfvf,
+				 struct otx2_cq_queue *cq)
+{
+	u64 incr = (u64)(cq->cq_idx) << 32;
+	u64 status;
+
+	status = otx2_atomic64_fetch_add(incr, pfvf->cq_op_addr);
+
+	if (unlikely(status & BIT_ULL(CQ_OP_STAT_OP_ERR) ||
+		     status & BIT_ULL(CQ_OP_STAT_CQ_ERR))) {
+		dev_err(pfvf->dev, "CQ stopped due to error");
+		return -EINVAL;
+	}
+
+	cq->cq_tail = status & 0xFFFFF;
+	cq->cq_head = (status >> 20) & 0xFFFFF;
+	if (cq->cq_tail < cq->cq_head)
+		cq->pend_cqe = (cq->cqe_cnt - cq->cq_head) +
+				cq->cq_tail;
+	else
+		cq->pend_cqe = cq->cq_tail - cq->cq_head;
+
+	return 0;
+}
+
 static struct nix_cqe_hdr_s *otx2_get_next_cqe(struct otx2_cq_queue *cq)
 {
 	struct nix_cqe_hdr_s *cqe_hdr;
@@ -318,7 +343,14 @@ static int otx2_rx_napi_handler(struct otx2_nic *pfvf,
 	struct nix_cqe_rx_s *cqe;
 	int processed_cqe = 0;
 
-	while (likely(processed_cqe < budget)) {
+	if (cq->pend_cqe >= budget)
+		goto process_cqe;
+
+	if (otx2_nix_cq_op_status(pfvf, cq) || !cq->pend_cqe)
+		return 0;
+
+process_cqe:
+	while (likely(processed_cqe < budget) && cq->pend_cqe) {
 		cqe = (struct nix_cqe_rx_s *)CQE_ADDR(cq, cq->cq_head);
 		if (cqe->hdr.cqe_type == NIX_XQE_TYPE_INVALID ||
 		    !cqe->sg.seg_addr) {
@@ -334,6 +366,7 @@ static int otx2_rx_napi_handler(struct otx2_nic *pfvf,
 		cqe->hdr.cqe_type = NIX_XQE_TYPE_INVALID;
 		cqe->sg.seg_addr = 0x00;
 		processed_cqe++;
+		cq->pend_cqe--;
 	}
 
 	/* Free CQEs to HW */
@@ -368,7 +401,14 @@ static int otx2_tx_napi_handler(struct otx2_nic *pfvf,
 	struct nix_cqe_tx_s *cqe;
 	int processed_cqe = 0;
 
-	while (likely(processed_cqe < budget)) {
+	if (cq->pend_cqe >= budget)
+		goto process_cqe;
+
+	if (otx2_nix_cq_op_status(pfvf, cq) || !cq->pend_cqe)
+		return 0;
+
+process_cqe:
+	while (likely(processed_cqe < budget) && cq->pend_cqe) {
 		cqe = (struct nix_cqe_tx_s *)otx2_get_next_cqe(cq);
 		if (unlikely(!cqe)) {
 			if (!processed_cqe)
@@ -380,6 +420,7 @@ static int otx2_tx_napi_handler(struct otx2_nic *pfvf,
 
 		cqe->hdr.cqe_type = NIX_XQE_TYPE_INVALID;
 		processed_cqe++;
+		cq->pend_cqe--;
 	}
 
 	/* Free CQEs to HW */
@@ -936,10 +977,16 @@ void otx2_cleanup_rx_cqes(struct otx2_nic *pfvf, struct otx2_cq_queue *cq)
 	int processed_cqe = 0;
 	u64 iova, pa;
 
-	while ((cqe = (struct nix_cqe_rx_s *)otx2_get_next_cqe(cq))) {
-		if (!cqe->sg.subdc)
-			continue;
+	if (otx2_nix_cq_op_status(pfvf, cq) || !cq->pend_cqe)
+		return;
+
+	while (cq->pend_cqe) {
+		cqe = (struct nix_cqe_rx_s *)otx2_get_next_cqe(cq);
 		processed_cqe++;
+		cq->pend_cqe--;
+
+		if (!cqe)
+			continue;
 		if (cqe->sg.segs > 1) {
 			otx2_free_rcv_seg(pfvf, cqe, cq->cq_idx);
 			continue;
@@ -965,7 +1012,16 @@ void otx2_cleanup_tx_cqes(struct otx2_nic *pfvf, struct otx2_cq_queue *cq)
 
 	sq = &pfvf->qset.sq[cq->cint_idx];
 
-	while ((cqe = (struct nix_cqe_tx_s *)otx2_get_next_cqe(cq))) {
+	if (otx2_nix_cq_op_status(pfvf, cq) || !cq->pend_cqe)
+		return;
+
+	while (cq->pend_cqe) {
+		cqe = (struct nix_cqe_tx_s *)otx2_get_next_cqe(cq);
+		processed_cqe++;
+		cq->pend_cqe--;
+
+		if (!cqe)
+			continue;
 		sg = &sq->sg[cqe->comp.sqe_id];
 		skb = (struct sk_buff *)sg->skb;
 		if (skb) {
@@ -973,7 +1029,6 @@ void otx2_cleanup_tx_cqes(struct otx2_nic *pfvf, struct otx2_cq_queue *cq)
 			dev_kfree_skb_any(skb);
 			sg->skb = (u64)NULL;
 		}
-		processed_cqe++;
 	}
 
 	/* Free CQEs to HW */
diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.h b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.h
index 3ff1ad79c001..6a97631ff226 100644
--- a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.h
+++ b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_txrx.h
@@ -56,6 +56,9 @@
  */
 #define CQ_QCOUNT_DEFAULT	1
 
+#define CQ_OP_STAT_OP_ERR       63
+#define CQ_OP_STAT_CQ_ERR       46
+
 struct queue_stats {
 	u64	bytes;
 	u64	pkts;
@@ -122,6 +125,8 @@ struct otx2_cq_queue {
 	u16			pool_ptrs;
 	u32			cqe_cnt;
 	u32			cq_head;
+	u32			cq_tail;
+	u32			pend_cqe;
 	void			*cqe_base;
 	struct qmem		*cqe;
 	struct otx2_pool	*rbpool;
diff --git a/include/linux/soc/marvell/octeontx2/asm.h b/include/linux/soc/marvell/octeontx2/asm.h
index fa1d6af0164e..0f79fd7f81a1 100644
--- a/include/linux/soc/marvell/octeontx2/asm.h
+++ b/include/linux/soc/marvell/octeontx2/asm.h
@@ -34,9 +34,23 @@
 			 : [rf] "+r"(val)		\
 			 : [rs] "r"(addr));		\
 })
+
+static inline u64 otx2_atomic64_fetch_add(u64 incr, u64 *ptr)
+{
+	u64 result;
+
+	asm volatile (".cpu  generic+lse\n"
+		      "ldadda %x[i], %x[r], [%[b]]"
+		      : [r] "=r" (result), "+m" (*ptr)
+		      : [i] "r" (incr), [b] "r" (ptr)
+		      : "memory");
+	return result;
+}
+
 #else
 #define otx2_lmt_flush(ioaddr)          ({ 0; })
 #define cn10k_lmt_flush(val, addr)	({ addr = val; })
+#define otx2_atomic64_fetch_add(incr, ptr)	({ incr; })
 #endif
 
 #endif /* __SOC_OTX2_ASM_H */
-- 
2.17.1

Re: [net-next v3 PATCH] octeontx2-pf: Use hardware register for CQE count

[patchwork-bot+netdevbpf]

Hello:

This patch was applied to netdev/net-next.git (refs/heads/master):

On Tue, 28 Sep 2021 11:25:26 +0530 you wrote:
> Current driver uses software CQ head pointer to poll on CQE
> header in memory to determine if CQE is valid. Software needs
> to make sure, that the reads of the CQE do not get re-ordered
> so much that it ends up with an inconsistent view of the CQE.
> To ensure that DMB barrier after read to first CQE cacheline
> and before reading of the rest of the CQE is needed.
> But having barrier for every CQE read will impact the performance,
> instead use hardware CQ head and tail pointers to find the
> valid number of CQEs.
> 
> [...]

Here is the summary with links:
  - [net-next,v3] octeontx2-pf: Use hardware register for CQE count
    https://git.kernel.org/netdev/net-next/c/af3826db74d1

You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html