Message-Id: <1680901196-20643-3-git-send-email-haiyangz@microsoft.com>
Date:   Fri,  7 Apr 2023 13:59:55 -0700
From:   Haiyang Zhang <haiyangz@...rosoft.com>
To:     linux-hyperv@...r.kernel.org, netdev@...r.kernel.org
Cc:     haiyangz@...rosoft.com, decui@...rosoft.com, kys@...rosoft.com,
        paulros@...rosoft.com, olaf@...fle.de, vkuznets@...hat.com,
        davem@...emloft.net, wei.liu@...nel.org, edumazet@...gle.com,
        kuba@...nel.org, pabeni@...hat.com, leon@...nel.org,
        longli@...rosoft.com, ssengar@...ux.microsoft.com,
        linux-rdma@...r.kernel.org, daniel@...earbox.net,
        john.fastabend@...il.com, bpf@...r.kernel.org, ast@...nel.org,
        sharmaajay@...rosoft.com, hawk@...nel.org,
        linux-kernel@...r.kernel.org
Subject: [PATCH V2,net-next, 2/3] net: mana: Enable RX path to handle various MTU sizes

Update RX data path to allocate and use RX queue DMA buffers with
proper size based on potentially various MTU sizes.

Signed-off-by: Haiyang Zhang <haiyangz@...rosoft.com>

---
V2:
Refectored to multiple patches for readability. Suggested by Yunsheng Lin.

---
 drivers/net/ethernet/microsoft/mana/mana_en.c | 188 +++++++++++-------
 include/net/mana/mana.h                       |  13 +-
 2 files changed, 124 insertions(+), 77 deletions(-)

diff --git a/drivers/net/ethernet/microsoft/mana/mana_en.c b/drivers/net/ethernet/microsoft/mana/mana_en.c
index 112c642dc89b..e5d5dea763f2 100644
--- a/drivers/net/ethernet/microsoft/mana/mana_en.c
+++ b/drivers/net/ethernet/microsoft/mana/mana_en.c
@@ -1185,10 +1185,10 @@ static void mana_post_pkt_rxq(struct mana_rxq *rxq)
 	WARN_ON_ONCE(recv_buf_oob->wqe_inf.wqe_size_in_bu != 1);
 }
 
-static struct sk_buff *mana_build_skb(void *buf_va, uint pkt_len,
-				      struct xdp_buff *xdp)
+static struct sk_buff *mana_build_skb(struct mana_rxq *rxq, void *buf_va,
+				      uint pkt_len, struct xdp_buff *xdp)
 {
-	struct sk_buff *skb = napi_build_skb(buf_va, PAGE_SIZE);
+	struct sk_buff *skb = napi_build_skb(buf_va, rxq->alloc_size);
 
 	if (!skb)
 		return NULL;
@@ -1196,11 +1196,12 @@ static struct sk_buff *mana_build_skb(void *buf_va, uint pkt_len,
 	if (xdp->data_hard_start) {
 		skb_reserve(skb, xdp->data - xdp->data_hard_start);
 		skb_put(skb, xdp->data_end - xdp->data);
-	} else {
-		skb_reserve(skb, XDP_PACKET_HEADROOM);
-		skb_put(skb, pkt_len);
+		return skb;
 	}
 
+	skb_reserve(skb, rxq->headroom);
+	skb_put(skb, pkt_len);
+
 	return skb;
 }
 
@@ -1233,7 +1234,7 @@ static void mana_rx_skb(void *buf_va, struct mana_rxcomp_oob *cqe,
 	if (act != XDP_PASS && act != XDP_TX)
 		goto drop_xdp;
 
-	skb = mana_build_skb(buf_va, pkt_len, &xdp);
+	skb = mana_build_skb(rxq, buf_va, pkt_len, &xdp);
 
 	if (!skb)
 		goto drop;
@@ -1282,14 +1283,72 @@ static void mana_rx_skb(void *buf_va, struct mana_rxcomp_oob *cqe,
 	u64_stats_update_end(&rx_stats->syncp);
 
 drop:
-	WARN_ON_ONCE(rxq->xdp_save_page);
-	rxq->xdp_save_page = virt_to_page(buf_va);
+	WARN_ON_ONCE(rxq->xdp_save_va);
+	/* Save for reuse */
+	rxq->xdp_save_va = buf_va;
 
 	++ndev->stats.rx_dropped;
 
 	return;
 }
 
+static void *mana_get_rxfrag(struct mana_rxq *rxq, struct device *dev,
+			     dma_addr_t *da, bool is_napi)
+{
+	struct page *page;
+	void *va;
+
+	/* Reuse XDP dropped page if available */
+	if (rxq->xdp_save_va) {
+		va = rxq->xdp_save_va;
+		rxq->xdp_save_va = NULL;
+	} else if (rxq->alloc_size > PAGE_SIZE) {
+		if (is_napi)
+			va = napi_alloc_frag(rxq->alloc_size);
+		else
+			va = netdev_alloc_frag(rxq->alloc_size);
+
+		if (!va)
+			return NULL;
+	} else {
+		page = dev_alloc_page();
+		if (!page)
+			return NULL;
+
+		va = page_to_virt(page);
+	}
+
+	*da = dma_map_single(dev, va + rxq->headroom, rxq->datasize,
+			     DMA_FROM_DEVICE);
+
+	if (dma_mapping_error(dev, *da)) {
+		put_page(virt_to_head_page(va));
+		return NULL;
+	}
+
+	return va;
+}
+
+/* Allocate frag for rx buffer, and save the old buf */
+static void mana_refill_rxoob(struct device *dev, struct mana_rxq *rxq,
+			      struct mana_recv_buf_oob *rxoob, void **old_buf)
+{
+	dma_addr_t da;
+	void *va;
+
+	va = mana_get_rxfrag(rxq, dev, &da, true);
+
+	if (!va)
+		return;
+
+	dma_unmap_single(dev, rxoob->sgl[0].address, rxq->datasize,
+			 DMA_FROM_DEVICE);
+	*old_buf = rxoob->buf_va;
+
+	rxoob->buf_va = va;
+	rxoob->sgl[0].address = da;
+}
+
 static void mana_process_rx_cqe(struct mana_rxq *rxq, struct mana_cq *cq,
 				struct gdma_comp *cqe)
 {
@@ -1299,10 +1358,8 @@ static void mana_process_rx_cqe(struct mana_rxq *rxq, struct mana_cq *cq,
 	struct mana_recv_buf_oob *rxbuf_oob;
 	struct mana_port_context *apc;
 	struct device *dev = gc->dev;
-	void *new_buf, *old_buf;
-	struct page *new_page;
+	void *old_buf = NULL;
 	u32 curr, pktlen;
-	dma_addr_t da;
 
 	apc = netdev_priv(ndev);
 
@@ -1345,40 +1402,11 @@ static void mana_process_rx_cqe(struct mana_rxq *rxq, struct mana_cq *cq,
 	rxbuf_oob = &rxq->rx_oobs[curr];
 	WARN_ON_ONCE(rxbuf_oob->wqe_inf.wqe_size_in_bu != 1);
 
-	/* Reuse XDP dropped page if available */
-	if (rxq->xdp_save_page) {
-		new_page = rxq->xdp_save_page;
-		rxq->xdp_save_page = NULL;
-	} else {
-		new_page = alloc_page(GFP_ATOMIC);
-	}
-
-	if (new_page) {
-		da = dma_map_page(dev, new_page, XDP_PACKET_HEADROOM, rxq->datasize,
-				  DMA_FROM_DEVICE);
-
-		if (dma_mapping_error(dev, da)) {
-			__free_page(new_page);
-			new_page = NULL;
-		}
-	}
-
-	new_buf = new_page ? page_to_virt(new_page) : NULL;
-
-	if (new_buf) {
-		dma_unmap_page(dev, rxbuf_oob->buf_dma_addr, rxq->datasize,
-			       DMA_FROM_DEVICE);
-
-		old_buf = rxbuf_oob->buf_va;
-
-		/* refresh the rxbuf_oob with the new page */
-		rxbuf_oob->buf_va = new_buf;
-		rxbuf_oob->buf_dma_addr = da;
-		rxbuf_oob->sgl[0].address = rxbuf_oob->buf_dma_addr;
-	} else {
-		old_buf = NULL; /* drop the packet if no memory */
-	}
+	mana_refill_rxoob(dev, rxq, rxbuf_oob, &old_buf);
 
+	/* Unsuccessful refill will have old_buf == NULL.
+	 * In this case, mana_rx_skb() will drop the packet.
+	 */
 	mana_rx_skb(old_buf, oob, rxq);
 
 drop:
@@ -1659,8 +1687,8 @@ static void mana_destroy_rxq(struct mana_port_context *apc,
 
 	mana_deinit_cq(apc, &rxq->rx_cq);
 
-	if (rxq->xdp_save_page)
-		__free_page(rxq->xdp_save_page);
+	if (rxq->xdp_save_va)
+		put_page(virt_to_head_page(rxq->xdp_save_va));
 
 	for (i = 0; i < rxq->num_rx_buf; i++) {
 		rx_oob = &rxq->rx_oobs[i];
@@ -1668,10 +1696,10 @@ static void mana_destroy_rxq(struct mana_port_context *apc,
 		if (!rx_oob->buf_va)
 			continue;
 
-		dma_unmap_page(dev, rx_oob->buf_dma_addr, rxq->datasize,
-			       DMA_FROM_DEVICE);
+		dma_unmap_single(dev, rx_oob->sgl[0].address,
+				 rx_oob->sgl[0].size, DMA_FROM_DEVICE);
 
-		free_page((unsigned long)rx_oob->buf_va);
+		put_page(virt_to_head_page(rx_oob->buf_va));
 		rx_oob->buf_va = NULL;
 	}
 
@@ -1681,6 +1709,26 @@ static void mana_destroy_rxq(struct mana_port_context *apc,
 	kfree(rxq);
 }
 
+static int mana_fill_rx_oob(struct mana_recv_buf_oob *rx_oob, u32 mem_key,
+			    struct mana_rxq *rxq, struct device *dev)
+{
+	dma_addr_t da;
+	void *va;
+
+	va = mana_get_rxfrag(rxq, dev, &da, false);
+
+	if (!va)
+		return -ENOMEM;
+
+	rx_oob->buf_va = va;
+
+	rx_oob->sgl[0].address = da;
+	rx_oob->sgl[0].size = rxq->datasize;
+	rx_oob->sgl[0].mem_key = mem_key;
+
+	return 0;
+}
+
 #define MANA_WQE_HEADER_SIZE 16
 #define MANA_WQE_SGE_SIZE 16
 
@@ -1690,11 +1738,10 @@ static int mana_alloc_rx_wqe(struct mana_port_context *apc,
 	struct gdma_context *gc = apc->ac->gdma_dev->gdma_context;
 	struct mana_recv_buf_oob *rx_oob;
 	struct device *dev = gc->dev;
-	struct page *page;
-	dma_addr_t da;
 	u32 buf_idx;
+	int ret;
 
-	WARN_ON(rxq->datasize == 0 || rxq->datasize > PAGE_SIZE);
+	WARN_ON(rxq->datasize == 0);
 
 	*rxq_size = 0;
 	*cq_size = 0;
@@ -1703,25 +1750,12 @@ static int mana_alloc_rx_wqe(struct mana_port_context *apc,
 		rx_oob = &rxq->rx_oobs[buf_idx];
 		memset(rx_oob, 0, sizeof(*rx_oob));
 
-		page = alloc_page(GFP_KERNEL);
-		if (!page)
-			return -ENOMEM;
-
-		da = dma_map_page(dev, page, XDP_PACKET_HEADROOM, rxq->datasize,
-				  DMA_FROM_DEVICE);
-
-		if (dma_mapping_error(dev, da)) {
-			__free_page(page);
-			return -ENOMEM;
-		}
-
-		rx_oob->buf_va = page_to_virt(page);
-		rx_oob->buf_dma_addr = da;
-
 		rx_oob->num_sge = 1;
-		rx_oob->sgl[0].address = rx_oob->buf_dma_addr;
-		rx_oob->sgl[0].size = rxq->datasize;
-		rx_oob->sgl[0].mem_key = apc->ac->gdma_dev->gpa_mkey;
+
+		ret = mana_fill_rx_oob(rx_oob, apc->ac->gdma_dev->gpa_mkey, rxq,
+				       dev);
+		if (ret)
+			return ret;
 
 		rx_oob->wqe_req.sgl = rx_oob->sgl;
 		rx_oob->wqe_req.num_sge = rx_oob->num_sge;
@@ -1764,6 +1798,7 @@ static struct mana_rxq *mana_create_rxq(struct mana_port_context *apc,
 	struct mana_obj_spec wq_spec;
 	struct mana_obj_spec cq_spec;
 	struct gdma_queue_spec spec;
+	unsigned int mtu = ndev->mtu;
 	struct mana_cq *cq = NULL;
 	struct gdma_context *gc;
 	u32 cq_size, rq_size;
@@ -1780,9 +1815,18 @@ static struct mana_rxq *mana_create_rxq(struct mana_port_context *apc,
 	rxq->ndev = ndev;
 	rxq->num_rx_buf = RX_BUFFERS_PER_QUEUE;
 	rxq->rxq_idx = rxq_idx;
-	rxq->datasize = ALIGN(MAX_FRAME_SIZE, 64);
 	rxq->rxobj = INVALID_MANA_HANDLE;
 
+	rxq->datasize = ALIGN(mtu + ETH_HLEN, 64);
+
+	if (mtu > MANA_XDP_MTU_MAX) {
+		rxq->alloc_size = mtu + MANA_RXBUF_PAD;
+		rxq->headroom = 0;
+	} else {
+		rxq->alloc_size = mtu + MANA_RXBUF_PAD + XDP_PACKET_HEADROOM;
+		rxq->headroom = XDP_PACKET_HEADROOM;
+	}
+
 	err = mana_alloc_rx_wqe(apc, rxq, &rq_size, &cq_size);
 	if (err)
 		goto out;
diff --git a/include/net/mana/mana.h b/include/net/mana/mana.h
index bb11a6535d80..fee99d704281 100644
--- a/include/net/mana/mana.h
+++ b/include/net/mana/mana.h
@@ -36,9 +36,6 @@ enum TRI_STATE {
 
 #define COMP_ENTRY_SIZE 64
 
-#define ADAPTER_MTU_SIZE 1500
-#define MAX_FRAME_SIZE (ADAPTER_MTU_SIZE + 14)
-
 #define RX_BUFFERS_PER_QUEUE 512
 
 #define MAX_SEND_BUFFERS_PER_QUEUE 256
@@ -282,7 +279,6 @@ struct mana_recv_buf_oob {
 	struct gdma_wqe_request wqe_req;
 
 	void *buf_va;
-	dma_addr_t buf_dma_addr;
 
 	/* SGL of the buffer going to be sent has part of the work request. */
 	u32 num_sge;
@@ -295,6 +291,11 @@ struct mana_recv_buf_oob {
 	struct gdma_posted_wqe_info wqe_inf;
 };
 
+#define MANA_RXBUF_PAD (SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) \
+			+ ETH_HLEN)
+
+#define MANA_XDP_MTU_MAX (PAGE_SIZE - MANA_RXBUF_PAD - XDP_PACKET_HEADROOM)
+
 struct mana_rxq {
 	struct gdma_queue *gdma_rq;
 	/* Cache the gdma receive queue id */
@@ -304,6 +305,8 @@ struct mana_rxq {
 	u32 rxq_idx;
 
 	u32 datasize;
+	u32 alloc_size;
+	u32 headroom;
 
 	mana_handle_t rxobj;
 
@@ -322,7 +325,7 @@ struct mana_rxq {
 
 	struct bpf_prog __rcu *bpf_prog;
 	struct xdp_rxq_info xdp_rxq;
-	struct page *xdp_save_page;
+	void *xdp_save_va; /* for reusing */
 	bool xdp_flush;
 	int xdp_rc; /* XDP redirect return code */
 
-- 
2.25.1

Powered by blists - more mailing lists