octeontx2-pf: Add support for page pool

Page pool for each rx queue enhance rx side performance
by reclaiming buffers back to each queue specific pool. DMA
mapping is done only for first allocation of buffers.
As subsequent buffers allocation avoid DMA mapping,
it results in performance improvement.

Image        |  Performance
------------ | ------------
Vannila      |   3Mpps
             |
with this    |   42Mpps
change	     |
---------------------------

Signed-off-by: Ratheesh Kannoth <rkannoth@marvell.com>
Link: https://lore.kernel.org/r/20230522020404.152020-1-rkannoth@marvell.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

1 files changed, 68 insertions, 10 deletions

diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c
index f9286648e45c..a79cb680bb23 100644
--- a/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c
+++ b/drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c
@@ -518,11 +518,32 @@ void otx2_config_irq_coalescing(struct otx2_nic *pfvf, int qidx)
 		     (pfvf->hw.cq_ecount_wait - 1));
 }
 
+static int otx2_alloc_pool_buf(struct otx2_nic *pfvf, struct otx2_pool *pool,
+			       dma_addr_t *dma)
+{
+	unsigned int offset = 0;
+	struct page *page;
+	size_t sz;
+
+	sz = SKB_DATA_ALIGN(pool->rbsize);
+	sz = ALIGN(sz, OTX2_ALIGN);
+
+	page = page_pool_alloc_frag(pool->page_pool, &offset, sz, GFP_ATOMIC);
+	if (unlikely(!page))
+		return -ENOMEM;
+
+	*dma = page_pool_get_dma_addr(page) + offset;
+	return 0;
+}
+
 static int __otx2_alloc_rbuf(struct otx2_nic *pfvf, struct otx2_pool *pool,
 			     dma_addr_t *dma)
 {
 	u8 *buf;
 
+	if (pool->page_pool)
+		return otx2_alloc_pool_buf(pfvf, pool, dma);
+
 	buf = napi_alloc_frag_align(pool->rbsize, OTX2_ALIGN);
 	if (unlikely(!buf))
 		return -ENOMEM;
@@ -1205,10 +1226,31 @@ void otx2_sq_free_sqbs(struct otx2_nic *pfvf)
 	}
 }
 
+void otx2_free_bufs(struct otx2_nic *pfvf, struct otx2_pool *pool,
+		    u64 iova, int size)
+{
+	struct page *page;
+	u64 pa;
+
+	pa = otx2_iova_to_phys(pfvf->iommu_domain, iova);
+	page = virt_to_head_page(phys_to_virt(pa));
+
+	if (pool->page_pool) {
+		page_pool_put_full_page(pool->page_pool, page, true);
+	} else {
+		dma_unmap_page_attrs(pfvf->dev, iova, size,
+				     DMA_FROM_DEVICE,
+				     DMA_ATTR_SKIP_CPU_SYNC);
+
+		put_page(page);
+	}
+}
+
 void otx2_free_aura_ptr(struct otx2_nic *pfvf, int type)
 {
 	int pool_id, pool_start = 0, pool_end = 0, size = 0;
-	u64 iova, pa;
+	struct otx2_pool *pool;
+	u64 iova;
 
 	if (type == AURA_NIX_SQ) {
 		pool_start = otx2_get_pool_idx(pfvf, type, 0);
@@ -1224,15 +1266,13 @@ void otx2_free_aura_ptr(struct otx2_nic *pfvf, int type)
 	/* Free SQB and RQB pointers from the aura pool */
 	for (pool_id = pool_start; pool_id < pool_end; pool_id++) {
 		iova = otx2_aura_allocptr(pfvf, pool_id);
+		pool = &pfvf->qset.pool[pool_id];
 		while (iova) {
 			if (type == AURA_NIX_RQ)
 				iova -= OTX2_HEAD_ROOM;
 
-			pa = otx2_iova_to_phys(pfvf->iommu_domain, iova);
-			dma_unmap_page_attrs(pfvf->dev, iova, size,
-					     DMA_FROM_DEVICE,
-					     DMA_ATTR_SKIP_CPU_SYNC);
-			put_page(virt_to_page(phys_to_virt(pa)));
+			otx2_free_bufs(pfvf, pool, iova, size);
+
 			iova = otx2_aura_allocptr(pfvf, pool_id);
 		}
 	}
@@ -1250,6 +1290,8 @@ void otx2_aura_pool_free(struct otx2_nic *pfvf)
 		pool = &pfvf->qset.pool[pool_id];
 		qmem_free(pfvf->dev, pool->stack);
 		qmem_free(pfvf->dev, pool->fc_addr);
+		page_pool_destroy(pool->page_pool);
+		pool->page_pool = NULL;
 	}
 	devm_kfree(pfvf->dev, pfvf->qset.pool);
 	pfvf->qset.pool = NULL;
@@ -1333,8 +1375,9 @@ int otx2_aura_init(struct otx2_nic *pfvf, int aura_id,
 }
 
 int otx2_pool_init(struct otx2_nic *pfvf, u16 pool_id,
-		   int stack_pages, int numptrs, int buf_size)
+		   int stack_pages, int numptrs, int buf_size, int type)
 {
+	struct page_pool_params pp_params = { 0 };
 	struct npa_aq_enq_req *aq;
 	struct otx2_pool *pool;
 	int err;
@@ -1378,6 +1421,22 @@ int otx2_pool_init(struct otx2_nic *pfvf, u16 pool_id,
 	aq->ctype = NPA_AQ_CTYPE_POOL;
 	aq->op = NPA_AQ_INSTOP_INIT;
 
+	if (type != AURA_NIX_RQ) {
+		pool->page_pool = NULL;
+		return 0;
+	}
+
+	pp_params.flags = PP_FLAG_PAGE_FRAG | PP_FLAG_DMA_MAP;
+	pp_params.pool_size = numptrs;
+	pp_params.nid = NUMA_NO_NODE;
+	pp_params.dev = pfvf->dev;
+	pp_params.dma_dir = DMA_FROM_DEVICE;
+	pool->page_pool = page_pool_create(&pp_params);
+	if (IS_ERR(pool->page_pool)) {
+		netdev_err(pfvf->netdev, "Creation of page pool failed\n");
+		return PTR_ERR(pool->page_pool);
+	}
+
 	return 0;
 }
 
@@ -1412,7 +1471,7 @@ int otx2_sq_aura_pool_init(struct otx2_nic *pfvf)
 
 		/* Initialize pool context */
 		err = otx2_pool_init(pfvf, pool_id, stack_pages,
-				     num_sqbs, hw->sqb_size);
+				     num_sqbs, hw->sqb_size, AURA_NIX_SQ);
 		if (err)
 			goto fail;
 	}
@@ -1475,7 +1534,7 @@ int otx2_rq_aura_pool_init(struct otx2_nic *pfvf)
 	}
 	for (pool_id = 0; pool_id < hw->rqpool_cnt; pool_id++) {
 		err = otx2_pool_init(pfvf, pool_id, stack_pages,
-				     num_ptrs, pfvf->rbsize);
+				     num_ptrs, pfvf->rbsize, AURA_NIX_RQ);
 		if (err)
 			goto fail;
 	}
@@ -1659,7 +1718,6 @@ int otx2_nix_config_bp(struct otx2_nic *pfvf, bool enable)
 	req->bpid_per_chan = 0;
 #endif
 
-
 	return otx2_sync_mbox_msg(&pfvf->mbox);
 }
 EXPORT_SYMBOL(otx2_nix_config_bp);