diff mbox series

[net-next] gve: Reduce alloc and copy costs in the GQ rx path

Message ID 20221029165322.1294983-1-shailend@google.com (mailing list archive)
State Accepted
Commit 82fd151d38d9fda714c5bb2e9e79ecd6bdc72da6
Delegated to: Netdev Maintainers
Headers show
Series [net-next] gve: Reduce alloc and copy costs in the GQ rx path | expand

Checks

Context Check Description
netdev/tree_selection success Clearly marked for net-next
netdev/fixes_present success Fixes tag not required for -next series
netdev/subject_prefix success Link
netdev/cover_letter success Single patches do not need cover letters
netdev/patch_count success Link
netdev/header_inline success No static functions without inline keyword in header files
netdev/build_32bit success Errors and warnings before: 0 this patch: 0
netdev/cc_maintainers warning 10 maintainers not CCed: jrkim@google.com tglx@linutronix.de bigeasy@linutronix.de jeroendb@google.com awogbemila@google.com amhamza.mgc@gmail.com csully@google.com xliutaox@google.com edumazet@google.com pabeni@redhat.com
netdev/build_clang success Errors and warnings before: 3 this patch: 3
netdev/module_param success Was 0 now: 0
netdev/verify_signedoff success Signed-off-by tag matches author and committer
netdev/check_selftest success No net selftest shell script
netdev/verify_fixes success No Fixes tag
netdev/build_allmodconfig_warn success Errors and warnings before: 0 this patch: 0
netdev/checkpatch warning WARNING: line length of 82 exceeds 80 columns WARNING: line length of 83 exceeds 80 columns WARNING: line length of 84 exceeds 80 columns WARNING: line length of 98 exceeds 80 columns
netdev/kdoc success Errors and warnings before: 0 this patch: 0
netdev/source_inline success Was 0 now: 0

Commit Message

Shailend Chand Oct. 29, 2022, 4:53 p.m. UTC
Previously, even if just one of the many fragments of a 9k packet
required a copy, we'd copy the whole packet into a freshly-allocated
9k-sized linear SKB, and this led to performance issues.

By having a pool of pages to copy into, each fragment can be
independently handled, leading to a reduced incidence of
allocation and copy.

Signed-off-by: Shailend Chand <shailend@google.com>
---
 drivers/net/ethernet/google/gve/gve.h         |  26 +-
 drivers/net/ethernet/google/gve/gve_ethtool.c |   2 +
 drivers/net/ethernet/google/gve/gve_rx.c      | 542 ++++++++++--------
 drivers/net/ethernet/google/gve/gve_rx_dqo.c  |   2 +-
 drivers/net/ethernet/google/gve/gve_utils.c   |  30 +-
 drivers/net/ethernet/google/gve/gve_utils.h   |   2 +-
 6 files changed, 338 insertions(+), 266 deletions(-)

Comments

patchwork-bot+netdevbpf@kernel.org Nov. 2, 2022, noon UTC | #1
Hello:

This patch was applied to netdev/net-next.git (master)
by David S. Miller <davem@davemloft.net>:

On Sat, 29 Oct 2022 09:53:22 -0700 you wrote:
> Previously, even if just one of the many fragments of a 9k packet
> required a copy, we'd copy the whole packet into a freshly-allocated
> 9k-sized linear SKB, and this led to performance issues.
> 
> By having a pool of pages to copy into, each fragment can be
> independently handled, leading to a reduced incidence of
> allocation and copy.
> 
> [...]

Here is the summary with links:
  - [net-next] gve: Reduce alloc and copy costs in the GQ rx path
    https://git.kernel.org/netdev/net-next/c/82fd151d38d9

You are awesome, thank you!
diff mbox series

Patch

diff --git a/drivers/net/ethernet/google/gve/gve.h b/drivers/net/ethernet/google/gve/gve.h
index 160735484465..5655da9cd236 100644
--- a/drivers/net/ethernet/google/gve/gve.h
+++ b/drivers/net/ethernet/google/gve/gve.h
@@ -60,7 +60,8 @@  struct gve_rx_slot_page_info {
 	void *page_address;
 	u32 page_offset; /* offset to write to in page */
 	int pagecnt_bias; /* expected pagecnt if only the driver has a ref */
-	u8 can_flip;
+	u16 pad; /* adjustment for rx padding */
+	u8 can_flip; /* tracks if the networking stack is using the page */
 };
 
 /* A list of pages registered with the device during setup and used by a queue
@@ -149,10 +150,17 @@  struct gve_rx_ctx {
 	/* head and tail of skb chain for the current packet or NULL if none */
 	struct sk_buff *skb_head;
 	struct sk_buff *skb_tail;
-	u16 total_expected_size;
-	u8 expected_frag_cnt;
-	u8 curr_frag_cnt;
-	u8 reuse_frags;
+	u32 total_size;
+	u8 frag_cnt;
+	bool drop_pkt;
+};
+
+struct gve_rx_cnts {
+	u32 ok_pkt_bytes;
+	u16 ok_pkt_cnt;
+	u16 total_pkt_cnt;
+	u16 cont_pkt_cnt;
+	u16 desc_err_pkt_cnt;
 };
 
 /* Contains datapath state used to represent an RX queue. */
@@ -167,6 +175,10 @@  struct gve_rx_ring {
 			/* threshold for posting new buffs and descs */
 			u32 db_threshold;
 			u16 packet_buffer_size;
+
+			u32 qpl_copy_pool_mask;
+			u32 qpl_copy_pool_head;
+			struct gve_rx_slot_page_info *qpl_copy_pool;
 		};
 
 		/* DQO fields. */
@@ -216,7 +228,9 @@  struct gve_rx_ring {
 	u64 rx_desc_err_dropped_pkt; /* free-running count of packets dropped by descriptor error */
 	u64 rx_cont_packet_cnt; /* free-running multi-fragment packets received */
 	u64 rx_frag_flip_cnt; /* free-running count of rx segments where page_flip was used */
-	u64 rx_frag_copy_cnt; /* free-running count of rx segments copied into skb linear portion */
+	u64 rx_frag_copy_cnt; /* free-running count of rx segments copied */
+	u64 rx_frag_alloc_cnt; /* free-running count of rx page allocations */
+
 	u32 q_num; /* queue index */
 	u32 ntfy_id; /* notification block index */
 	struct gve_queue_resources *q_resources; /* head and tail pointer idx */
diff --git a/drivers/net/ethernet/google/gve/gve_ethtool.c b/drivers/net/ethernet/google/gve/gve_ethtool.c
index 7b9a2d9d9624..e7e4ba99c20e 100644
--- a/drivers/net/ethernet/google/gve/gve_ethtool.c
+++ b/drivers/net/ethernet/google/gve/gve_ethtool.c
@@ -45,6 +45,7 @@  static const char gve_gstrings_main_stats[][ETH_GSTRING_LEN] = {
 static const char gve_gstrings_rx_stats[][ETH_GSTRING_LEN] = {
 	"rx_posted_desc[%u]", "rx_completed_desc[%u]", "rx_consumed_desc[%u]", "rx_bytes[%u]",
 	"rx_cont_packet_cnt[%u]", "rx_frag_flip_cnt[%u]", "rx_frag_copy_cnt[%u]",
+	"rx_frag_alloc_cnt[%u]",
 	"rx_dropped_pkt[%u]", "rx_copybreak_pkt[%u]", "rx_copied_pkt[%u]",
 	"rx_queue_drop_cnt[%u]", "rx_no_buffers_posted[%u]",
 	"rx_drops_packet_over_mru[%u]", "rx_drops_invalid_checksum[%u]",
@@ -271,6 +272,7 @@  gve_get_ethtool_stats(struct net_device *netdev,
 			data[i++] = rx->rx_cont_packet_cnt;
 			data[i++] = rx->rx_frag_flip_cnt;
 			data[i++] = rx->rx_frag_copy_cnt;
+			data[i++] = rx->rx_frag_alloc_cnt;
 			/* rx dropped packets */
 			data[i++] = tmp_rx_skb_alloc_fail +
 				tmp_rx_buf_alloc_fail +
diff --git a/drivers/net/ethernet/google/gve/gve_rx.c b/drivers/net/ethernet/google/gve/gve_rx.c
index 021bbf308d68..e2f4494c65fb 100644
--- a/drivers/net/ethernet/google/gve/gve_rx.c
+++ b/drivers/net/ethernet/google/gve/gve_rx.c
@@ -35,6 +35,12 @@  static void gve_rx_unfill_pages(struct gve_priv *priv, struct gve_rx_ring *rx)
 				     rx->data.page_info[i].pagecnt_bias - 1);
 		gve_unassign_qpl(priv, rx->data.qpl->id);
 		rx->data.qpl = NULL;
+
+		for (i = 0; i < rx->qpl_copy_pool_mask + 1; i++) {
+			page_ref_sub(rx->qpl_copy_pool[i].page,
+				     rx->qpl_copy_pool[i].pagecnt_bias - 1);
+			put_page(rx->qpl_copy_pool[i].page);
+		}
 	}
 	kvfree(rx->data.page_info);
 	rx->data.page_info = NULL;
@@ -63,6 +69,10 @@  static void gve_rx_free_ring(struct gve_priv *priv, int idx)
 	dma_free_coherent(dev, bytes, rx->data.data_ring,
 			  rx->data.data_bus);
 	rx->data.data_ring = NULL;
+
+	kvfree(rx->qpl_copy_pool);
+	rx->qpl_copy_pool = NULL;
+
 	netif_dbg(priv, drv, priv->dev, "freed rx ring %d\n", idx);
 }
 
@@ -101,6 +111,7 @@  static int gve_prefill_rx_pages(struct gve_rx_ring *rx)
 	u32 slots;
 	int err;
 	int i;
+	int j;
 
 	/* Allocate one page per Rx queue slot. Each page is split into two
 	 * packet buffers, when possible we "page flip" between the two.
@@ -135,7 +146,31 @@  static int gve_prefill_rx_pages(struct gve_rx_ring *rx)
 			goto alloc_err;
 	}
 
+	if (!rx->data.raw_addressing) {
+		for (j = 0; j < rx->qpl_copy_pool_mask + 1; j++) {
+			struct page *page = alloc_page(GFP_KERNEL);
+
+			if (!page)
+				goto alloc_err_qpl;
+
+			rx->qpl_copy_pool[j].page = page;
+			rx->qpl_copy_pool[j].page_offset = 0;
+			rx->qpl_copy_pool[j].page_address = page_address(page);
+
+			/* The page already has 1 ref. */
+			page_ref_add(page, INT_MAX - 1);
+			rx->qpl_copy_pool[j].pagecnt_bias = INT_MAX;
+		}
+	}
+
 	return slots;
+
+alloc_err_qpl:
+	while (j--) {
+		page_ref_sub(rx->qpl_copy_pool[j].page,
+			     rx->qpl_copy_pool[j].pagecnt_bias - 1);
+		put_page(rx->qpl_copy_pool[j].page);
+	}
 alloc_err:
 	while (i--)
 		gve_rx_free_buffer(&priv->pdev->dev,
@@ -146,12 +181,11 @@  static int gve_prefill_rx_pages(struct gve_rx_ring *rx)
 
 static void gve_rx_ctx_clear(struct gve_rx_ctx *ctx)
 {
-	ctx->curr_frag_cnt = 0;
-	ctx->total_expected_size = 0;
-	ctx->expected_frag_cnt = 0;
 	ctx->skb_head = NULL;
 	ctx->skb_tail = NULL;
-	ctx->reuse_frags = false;
+	ctx->total_size = 0;
+	ctx->frag_cnt = 0;
+	ctx->drop_pkt = false;
 }
 
 static int gve_rx_alloc_ring(struct gve_priv *priv, int idx)
@@ -181,10 +215,22 @@  static int gve_rx_alloc_ring(struct gve_priv *priv, int idx)
 						GFP_KERNEL);
 	if (!rx->data.data_ring)
 		return -ENOMEM;
+
+	rx->qpl_copy_pool_mask = min_t(u32, U32_MAX, slots * 2) - 1;
+	rx->qpl_copy_pool_head = 0;
+	rx->qpl_copy_pool = kvcalloc(rx->qpl_copy_pool_mask + 1,
+				     sizeof(rx->qpl_copy_pool[0]),
+				     GFP_KERNEL);
+
+	if (!rx->qpl_copy_pool) {
+		err = -ENOMEM;
+		goto abort_with_slots;
+	}
+
 	filled_pages = gve_prefill_rx_pages(rx);
 	if (filled_pages < 0) {
 		err = -ENOMEM;
-		goto abort_with_slots;
+		goto abort_with_copy_pool;
 	}
 	rx->fill_cnt = filled_pages;
 	/* Ensure data ring slots (packet buffers) are visible. */
@@ -236,6 +282,9 @@  static int gve_rx_alloc_ring(struct gve_priv *priv, int idx)
 	rx->q_resources = NULL;
 abort_filled:
 	gve_rx_unfill_pages(priv, rx);
+abort_with_copy_pool:
+	kvfree(rx->qpl_copy_pool);
+	rx->qpl_copy_pool = NULL;
 abort_with_slots:
 	bytes = sizeof(*rx->data.data_ring) * slots;
 	dma_free_coherent(hdev, bytes, rx->data.data_ring, rx->data.data_bus);
@@ -292,30 +341,47 @@  static enum pkt_hash_types gve_rss_type(__be16 pkt_flags)
 	return PKT_HASH_TYPE_L2;
 }
 
-static u16 gve_rx_ctx_padding(struct gve_rx_ctx *ctx)
-{
-	return (ctx->curr_frag_cnt == 0) ? GVE_RX_PAD : 0;
-}
-
 static struct sk_buff *gve_rx_add_frags(struct napi_struct *napi,
 					struct gve_rx_slot_page_info *page_info,
 					u16 packet_buffer_size, u16 len,
 					struct gve_rx_ctx *ctx)
 {
-	u32 offset = page_info->page_offset +  gve_rx_ctx_padding(ctx);
-	struct sk_buff *skb;
+	u32 offset = page_info->page_offset + page_info->pad;
+	struct sk_buff *skb = ctx->skb_tail;
+	int num_frags = 0;
 
-	if (!ctx->skb_head)
-		ctx->skb_head = napi_get_frags(napi);
+	if (!skb) {
+		skb = napi_get_frags(napi);
+		if (unlikely(!skb))
+			return NULL;
 
-	if (unlikely(!ctx->skb_head))
-		return NULL;
+		ctx->skb_head = skb;
+		ctx->skb_tail = skb;
+	} else {
+		num_frags = skb_shinfo(ctx->skb_tail)->nr_frags;
+		if (num_frags == MAX_SKB_FRAGS) {
+			skb = napi_alloc_skb(napi, 0);
+			if (!skb)
+				return NULL;
+
+			// We will never chain more than two SKBs: 2 * 16 * 2k > 64k
+			// which is why we do not need to chain by using skb->next
+			skb_shinfo(ctx->skb_tail)->frag_list = skb;
 
-	skb = ctx->skb_head;
-	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page_info->page,
+			ctx->skb_tail = skb;
+			num_frags = 0;
+		}
+	}
+
+	if (skb != ctx->skb_head) {
+		ctx->skb_head->len += len;
+		ctx->skb_head->data_len += len;
+		ctx->skb_head->truesize += packet_buffer_size;
+	}
+	skb_add_rx_frag(skb, num_frags, page_info->page,
 			offset, len, packet_buffer_size);
 
-	return skb;
+	return ctx->skb_head;
 }
 
 static void gve_rx_flip_buff(struct gve_rx_slot_page_info *page_info, __be64 *slot_addr)
@@ -363,6 +429,92 @@  gve_rx_raw_addressing(struct device *dev, struct net_device *netdev,
 	return skb;
 }
 
+static struct sk_buff *gve_rx_copy_to_pool(struct gve_rx_ring *rx,
+					   struct gve_rx_slot_page_info *page_info,
+					   u16 len, struct napi_struct *napi)
+{
+	u32 pool_idx = rx->qpl_copy_pool_head & rx->qpl_copy_pool_mask;
+	void *src = page_info->page_address + page_info->page_offset;
+	struct gve_rx_slot_page_info *copy_page_info;
+	struct gve_rx_ctx *ctx = &rx->ctx;
+	bool alloc_page = false;
+	struct sk_buff *skb;
+	void *dst;
+
+	copy_page_info = &rx->qpl_copy_pool[pool_idx];
+	if (!copy_page_info->can_flip) {
+		int recycle = gve_rx_can_recycle_buffer(copy_page_info);
+
+		if (unlikely(recycle < 0)) {
+			gve_schedule_reset(rx->gve);
+			return NULL;
+		}
+		alloc_page = !recycle;
+	}
+
+	if (alloc_page) {
+		struct gve_rx_slot_page_info alloc_page_info;
+		struct page *page;
+
+		/* The least recently used page turned out to be
+		 * still in use by the kernel. Ignoring it and moving
+		 * on alleviates head-of-line blocking.
+		 */
+		rx->qpl_copy_pool_head++;
+
+		page = alloc_page(GFP_ATOMIC);
+		if (!page)
+			return NULL;
+
+		alloc_page_info.page = page;
+		alloc_page_info.page_offset = 0;
+		alloc_page_info.page_address = page_address(page);
+		alloc_page_info.pad = page_info->pad;
+
+		memcpy(alloc_page_info.page_address, src, page_info->pad + len);
+		skb = gve_rx_add_frags(napi, &alloc_page_info,
+				       rx->packet_buffer_size,
+				       len, ctx);
+
+		u64_stats_update_begin(&rx->statss);
+		rx->rx_frag_copy_cnt++;
+		rx->rx_frag_alloc_cnt++;
+		u64_stats_update_end(&rx->statss);
+
+		return skb;
+	}
+
+	dst = copy_page_info->page_address + copy_page_info->page_offset;
+	memcpy(dst, src, page_info->pad + len);
+	copy_page_info->pad = page_info->pad;
+
+	skb = gve_rx_add_frags(napi, copy_page_info,
+			       rx->packet_buffer_size, len, ctx);
+	if (unlikely(!skb))
+		return NULL;
+
+	gve_dec_pagecnt_bias(copy_page_info);
+	copy_page_info->page_offset += rx->packet_buffer_size;
+	copy_page_info->page_offset &= (PAGE_SIZE - 1);
+
+	if (copy_page_info->can_flip) {
+		/* We have used both halves of this copy page, it
+		 * is time for it to go to the back of the queue.
+		 */
+		copy_page_info->can_flip = false;
+		rx->qpl_copy_pool_head++;
+		prefetch(rx->qpl_copy_pool[rx->qpl_copy_pool_head & rx->qpl_copy_pool_mask].page);
+	} else {
+		copy_page_info->can_flip = true;
+	}
+
+	u64_stats_update_begin(&rx->statss);
+	rx->rx_frag_copy_cnt++;
+	u64_stats_update_end(&rx->statss);
+
+	return skb;
+}
+
 static struct sk_buff *
 gve_rx_qpl(struct device *dev, struct net_device *netdev,
 	   struct gve_rx_ring *rx, struct gve_rx_slot_page_info *page_info,
@@ -377,7 +529,7 @@  gve_rx_qpl(struct device *dev, struct net_device *netdev,
 	 * choice is to copy the data out of it so that we can return it to the
 	 * device.
 	 */
-	if (ctx->reuse_frags) {
+	if (page_info->can_flip) {
 		skb = gve_rx_add_frags(napi, page_info, rx->packet_buffer_size, len, ctx);
 		/* No point in recycling if we didn't get the skb */
 		if (skb) {
@@ -386,116 +538,23 @@  gve_rx_qpl(struct device *dev, struct net_device *netdev,
 			gve_rx_flip_buff(page_info, &data_slot->qpl_offset);
 		}
 	} else {
-		const u16 padding = gve_rx_ctx_padding(ctx);
-
-		skb = gve_rx_copy(netdev, napi, page_info, len, padding, ctx);
-		if (skb) {
-			u64_stats_update_begin(&rx->statss);
-			rx->rx_frag_copy_cnt++;
-			u64_stats_update_end(&rx->statss);
-		}
+		skb = gve_rx_copy_to_pool(rx, page_info, len, napi);
 	}
 	return skb;
 }
 
-#define GVE_PKTCONT_BIT_IS_SET(x) (GVE_RXF_PKT_CONT & (x))
-static u16 gve_rx_get_fragment_size(struct gve_rx_ctx *ctx, struct gve_rx_desc *desc)
-{
-	return be16_to_cpu(desc->len) - gve_rx_ctx_padding(ctx);
-}
-
-static bool gve_rx_ctx_init(struct gve_rx_ctx *ctx, struct gve_rx_ring *rx)
-{
-	bool qpl_mode = !rx->data.raw_addressing, packet_size_error = false;
-	bool buffer_error = false, desc_error = false, seqno_error = false;
-	struct gve_rx_slot_page_info *page_info;
-	struct gve_priv *priv = rx->gve;
-	u32 idx = rx->cnt & rx->mask;
-	bool reuse_frags, can_flip;
-	struct gve_rx_desc *desc;
-	u16 packet_size = 0;
-	u16 n_frags = 0;
-	int recycle;
-
-	/** In QPL mode, we only flip buffers when all buffers containing the packet
-	 * can be flipped. RDA can_flip decisions will be made later, per frag.
-	 */
-	can_flip = qpl_mode;
-	reuse_frags = can_flip;
-	do {
-		u16 frag_size;
-
-		n_frags++;
-		desc = &rx->desc.desc_ring[idx];
-		desc_error = unlikely(desc->flags_seq & GVE_RXF_ERR) || desc_error;
-		if (GVE_SEQNO(desc->flags_seq) != rx->desc.seqno) {
-			seqno_error = true;
-			netdev_warn(priv->dev,
-				    "RX seqno error: want=%d, got=%d, dropping packet and scheduling reset.",
-				    rx->desc.seqno, GVE_SEQNO(desc->flags_seq));
-		}
-		frag_size = be16_to_cpu(desc->len);
-		packet_size += frag_size;
-		if (frag_size > rx->packet_buffer_size) {
-			packet_size_error = true;
-			netdev_warn(priv->dev,
-				    "RX fragment error: packet_buffer_size=%d, frag_size=%d, dropping packet.",
-				    rx->packet_buffer_size, be16_to_cpu(desc->len));
-		}
-		page_info = &rx->data.page_info[idx];
-		if (can_flip) {
-			recycle = gve_rx_can_recycle_buffer(page_info);
-			reuse_frags = reuse_frags && recycle > 0;
-			buffer_error = buffer_error || unlikely(recycle < 0);
-		}
-		idx = (idx + 1) & rx->mask;
-		rx->desc.seqno = gve_next_seqno(rx->desc.seqno);
-	} while (GVE_PKTCONT_BIT_IS_SET(desc->flags_seq));
-
-	prefetch(rx->desc.desc_ring + idx);
-
-	ctx->curr_frag_cnt = 0;
-	ctx->total_expected_size = packet_size - GVE_RX_PAD;
-	ctx->expected_frag_cnt = n_frags;
-	ctx->skb_head = NULL;
-	ctx->reuse_frags = reuse_frags;
-
-	if (ctx->expected_frag_cnt > 1) {
-		u64_stats_update_begin(&rx->statss);
-		rx->rx_cont_packet_cnt++;
-		u64_stats_update_end(&rx->statss);
-	}
-	if (ctx->total_expected_size > priv->rx_copybreak && !ctx->reuse_frags && qpl_mode) {
-		u64_stats_update_begin(&rx->statss);
-		rx->rx_copied_pkt++;
-		u64_stats_update_end(&rx->statss);
-	}
-
-	if (unlikely(buffer_error || seqno_error || packet_size_error)) {
-		gve_schedule_reset(priv);
-		return false;
-	}
-
-	if (unlikely(desc_error)) {
-		u64_stats_update_begin(&rx->statss);
-		rx->rx_desc_err_dropped_pkt++;
-		u64_stats_update_end(&rx->statss);
-		return false;
-	}
-	return true;
-}
-
 static struct sk_buff *gve_rx_skb(struct gve_priv *priv, struct gve_rx_ring *rx,
 				  struct gve_rx_slot_page_info *page_info, struct napi_struct *napi,
-				  u16 len, union gve_rx_data_slot *data_slot)
+				  u16 len, union gve_rx_data_slot *data_slot,
+				  bool is_only_frag)
 {
 	struct net_device *netdev = priv->dev;
 	struct gve_rx_ctx *ctx = &rx->ctx;
 	struct sk_buff *skb = NULL;
 
-	if (len <= priv->rx_copybreak && ctx->expected_frag_cnt == 1) {
+	if (len <= priv->rx_copybreak && is_only_frag)  {
 		/* Just copy small packets */
-		skb = gve_rx_copy(netdev, napi, page_info, len, GVE_RX_PAD, ctx);
+		skb = gve_rx_copy(netdev, napi, page_info, len, GVE_RX_PAD);
 		if (skb) {
 			u64_stats_update_begin(&rx->statss);
 			rx->rx_copied_pkt++;
@@ -504,29 +563,25 @@  static struct sk_buff *gve_rx_skb(struct gve_priv *priv, struct gve_rx_ring *rx,
 			u64_stats_update_end(&rx->statss);
 		}
 	} else {
-		if (rx->data.raw_addressing) {
-			int recycle = gve_rx_can_recycle_buffer(page_info);
+		int recycle = gve_rx_can_recycle_buffer(page_info);
 
-			if (unlikely(recycle < 0)) {
-				gve_schedule_reset(priv);
-				return NULL;
-			}
-			page_info->can_flip = recycle;
-			if (page_info->can_flip) {
-				u64_stats_update_begin(&rx->statss);
-				rx->rx_frag_flip_cnt++;
-				u64_stats_update_end(&rx->statss);
-			}
+		if (unlikely(recycle < 0)) {
+			gve_schedule_reset(priv);
+			return NULL;
+		}
+		page_info->can_flip = recycle;
+		if (page_info->can_flip) {
+			u64_stats_update_begin(&rx->statss);
+			rx->rx_frag_flip_cnt++;
+			u64_stats_update_end(&rx->statss);
+		}
+
+		if (rx->data.raw_addressing) {
 			skb = gve_rx_raw_addressing(&priv->pdev->dev, netdev,
 						    page_info, len, napi,
 						    data_slot,
 						    rx->packet_buffer_size, ctx);
 		} else {
-			if (ctx->reuse_frags) {
-				u64_stats_update_begin(&rx->statss);
-				rx->rx_frag_flip_cnt++;
-				u64_stats_update_end(&rx->statss);
-			}
 			skb = gve_rx_qpl(&priv->pdev->dev, netdev, rx,
 					 page_info, len, napi, data_slot);
 		}
@@ -534,101 +589,113 @@  static struct sk_buff *gve_rx_skb(struct gve_priv *priv, struct gve_rx_ring *rx,
 	return skb;
 }
 
-static bool gve_rx(struct gve_rx_ring *rx, netdev_features_t feat,
-		   u64 *packet_size_bytes, u32 *work_done)
+#define GVE_PKTCONT_BIT_IS_SET(x) (GVE_RXF_PKT_CONT & (x))
+static void gve_rx(struct gve_rx_ring *rx, netdev_features_t feat,
+		   struct gve_rx_desc *desc, u32 idx,
+		   struct gve_rx_cnts *cnts)
 {
+	bool is_last_frag = !GVE_PKTCONT_BIT_IS_SET(desc->flags_seq);
 	struct gve_rx_slot_page_info *page_info;
+	u16 frag_size = be16_to_cpu(desc->len);
 	struct gve_rx_ctx *ctx = &rx->ctx;
 	union gve_rx_data_slot *data_slot;
 	struct gve_priv *priv = rx->gve;
-	struct gve_rx_desc *first_desc;
 	struct sk_buff *skb = NULL;
-	struct gve_rx_desc *desc;
-	struct napi_struct *napi;
 	dma_addr_t page_bus;
-	u32 work_cnt = 0;
 	void *va;
-	u32 idx;
-	u16 len;
 
-	idx = rx->cnt & rx->mask;
-	first_desc = &rx->desc.desc_ring[idx];
-	desc = first_desc;
-	napi = &priv->ntfy_blocks[rx->ntfy_id].napi;
+	struct napi_struct *napi = &priv->ntfy_blocks[rx->ntfy_id].napi;
+	bool is_first_frag = ctx->frag_cnt == 0;
 
-	if (unlikely(!gve_rx_ctx_init(ctx, rx)))
-		goto skb_alloc_fail;
+	bool is_only_frag = is_first_frag && is_last_frag;
 
-	while (ctx->curr_frag_cnt < ctx->expected_frag_cnt) {
-		/* Prefetch two packet buffers ahead, we will need it soon. */
-		page_info = &rx->data.page_info[(idx + 2) & rx->mask];
-		va = page_info->page_address + page_info->page_offset;
+	if (unlikely(ctx->drop_pkt))
+		goto finish_frag;
 
-		prefetch(page_info->page); /* Kernel page struct. */
-		prefetch(va);              /* Packet header. */
-		prefetch(va + 64);         /* Next cacheline too. */
+	if (desc->flags_seq & GVE_RXF_ERR) {
+		ctx->drop_pkt = true;
+		cnts->desc_err_pkt_cnt++;
+		napi_free_frags(napi);
+		goto finish_frag;
+	}
 
-		len = gve_rx_get_fragment_size(ctx, desc);
+	if (unlikely(frag_size > rx->packet_buffer_size)) {
+		netdev_warn(priv->dev, "Unexpected frag size %d, can't exceed %d, scheduling reset",
+			    frag_size, rx->packet_buffer_size);
+		ctx->drop_pkt = true;
+		napi_free_frags(napi);
+		gve_schedule_reset(rx->gve);
+		goto finish_frag;
+	}
 
-		page_info = &rx->data.page_info[idx];
-		data_slot = &rx->data.data_ring[idx];
-		page_bus = rx->data.raw_addressing ?
-			   be64_to_cpu(data_slot->addr) - page_info->page_offset :
-			   rx->data.qpl->page_buses[idx];
-		dma_sync_single_for_cpu(&priv->pdev->dev, page_bus, PAGE_SIZE, DMA_FROM_DEVICE);
-
-		skb = gve_rx_skb(priv, rx, page_info, napi, len, data_slot);
-		if (!skb) {
-			u64_stats_update_begin(&rx->statss);
-			rx->rx_skb_alloc_fail++;
-			u64_stats_update_end(&rx->statss);
-			goto skb_alloc_fail;
+	/* Prefetch two packet buffers ahead, we will need it soon. */
+	page_info = &rx->data.page_info[(idx + 2) & rx->mask];
+	va = page_info->page_address + page_info->page_offset;
+	prefetch(page_info->page); /* Kernel page struct. */
+	prefetch(va);              /* Packet header. */
+	prefetch(va + 64);         /* Next cacheline too. */
+
+	page_info = &rx->data.page_info[idx];
+	data_slot = &rx->data.data_ring[idx];
+	page_bus = (rx->data.raw_addressing) ?
+		be64_to_cpu(data_slot->addr) - page_info->page_offset :
+		rx->data.qpl->page_buses[idx];
+	dma_sync_single_for_cpu(&priv->pdev->dev, page_bus,
+				PAGE_SIZE, DMA_FROM_DEVICE);
+	page_info->pad = is_first_frag ? GVE_RX_PAD : 0;
+	frag_size -= page_info->pad;
+
+	skb = gve_rx_skb(priv, rx, page_info, napi, frag_size,
+			 data_slot, is_only_frag);
+	if (!skb) {
+		u64_stats_update_begin(&rx->statss);
+		rx->rx_skb_alloc_fail++;
+		u64_stats_update_end(&rx->statss);
+
+		napi_free_frags(napi);
+		ctx->drop_pkt = true;
+		goto finish_frag;
+	}
+	ctx->total_size += frag_size;
+
+	if (is_first_frag) {
+		if (likely(feat & NETIF_F_RXCSUM)) {
+			/* NIC passes up the partial sum */
+			if (desc->csum)
+				skb->ip_summed = CHECKSUM_COMPLETE;
+			else
+				skb->ip_summed = CHECKSUM_NONE;
+			skb->csum = csum_unfold(desc->csum);
 		}
 
-		ctx->curr_frag_cnt++;
-		rx->cnt++;
-		idx = rx->cnt & rx->mask;
-		work_cnt++;
-		desc = &rx->desc.desc_ring[idx];
+		/* parse flags & pass relevant info up */
+		if (likely(feat & NETIF_F_RXHASH) &&
+		    gve_needs_rss(desc->flags_seq))
+			skb_set_hash(skb, be32_to_cpu(desc->rss_hash),
+				     gve_rss_type(desc->flags_seq));
 	}
 
-	if (likely(feat & NETIF_F_RXCSUM)) {
-		/* NIC passes up the partial sum */
-		if (first_desc->csum)
-			skb->ip_summed = CHECKSUM_COMPLETE;
+	if (is_last_frag) {
+		skb_record_rx_queue(skb, rx->q_num);
+		if (skb_is_nonlinear(skb))
+			napi_gro_frags(napi);
 		else
-			skb->ip_summed = CHECKSUM_NONE;
-		skb->csum = csum_unfold(first_desc->csum);
+			napi_gro_receive(napi, skb);
+		goto finish_ok_pkt;
 	}
 
-	/* parse flags & pass relevant info up */
-	if (likely(feat & NETIF_F_RXHASH) &&
-	    gve_needs_rss(first_desc->flags_seq))
-		skb_set_hash(skb, be32_to_cpu(first_desc->rss_hash),
-			     gve_rss_type(first_desc->flags_seq));
-
-	*packet_size_bytes = skb->len + (skb->protocol ? ETH_HLEN : 0);
-	*work_done = work_cnt;
-	skb_record_rx_queue(skb, rx->q_num);
-	if (skb_is_nonlinear(skb))
-		napi_gro_frags(napi);
-	else
-		napi_gro_receive(napi, skb);
-
-	gve_rx_ctx_clear(ctx);
-	return true;
-
-skb_alloc_fail:
-	if (napi->skb)
-		napi_free_frags(napi);
-	*packet_size_bytes = 0;
-	*work_done = ctx->expected_frag_cnt;
-	while (ctx->curr_frag_cnt < ctx->expected_frag_cnt) {
-		rx->cnt++;
-		ctx->curr_frag_cnt++;
+	goto finish_frag;
+
+finish_ok_pkt:
+	cnts->ok_pkt_bytes += ctx->total_size;
+	cnts->ok_pkt_cnt++;
+finish_frag:
+	ctx->frag_cnt++;
+	if (is_last_frag) {
+		cnts->total_pkt_cnt++;
+		cnts->cont_pkt_cnt += (ctx->frag_cnt > 1);
+		gve_rx_ctx_clear(ctx);
 	}
-	gve_rx_ctx_clear(ctx);
-	return false;
 }
 
 bool gve_rx_work_pending(struct gve_rx_ring *rx)
@@ -704,36 +771,39 @@  static bool gve_rx_refill_buffers(struct gve_priv *priv, struct gve_rx_ring *rx)
 static int gve_clean_rx_done(struct gve_rx_ring *rx, int budget,
 			     netdev_features_t feat)
 {
-	u32 work_done = 0, total_packet_cnt = 0, ok_packet_cnt = 0;
+	struct gve_rx_ctx *ctx = &rx->ctx;
 	struct gve_priv *priv = rx->gve;
+	struct gve_rx_cnts cnts = {0};
+	struct gve_rx_desc *next_desc;
 	u32 idx = rx->cnt & rx->mask;
-	struct gve_rx_desc *desc;
-	u64 bytes = 0;
+	u32 work_done = 0;
+
+	struct gve_rx_desc *desc = &rx->desc.desc_ring[idx];
 
-	desc = &rx->desc.desc_ring[idx];
+	// Exceed budget only if (and till) the inflight packet is consumed.
 	while ((GVE_SEQNO(desc->flags_seq) == rx->desc.seqno) &&
-	       work_done < budget) {
-		u64 packet_size_bytes = 0;
-		u32 work_cnt = 0;
-		bool dropped;
-
-		netif_info(priv, rx_status, priv->dev,
-			   "[%d] idx=%d desc=%p desc->flags_seq=0x%x\n",
-			   rx->q_num, idx, desc, desc->flags_seq);
-		netif_info(priv, rx_status, priv->dev,
-			   "[%d] seqno=%d rx->desc.seqno=%d\n",
-			   rx->q_num, GVE_SEQNO(desc->flags_seq),
-			   rx->desc.seqno);
-
-		dropped = !gve_rx(rx, feat, &packet_size_bytes, &work_cnt);
-		if (!dropped) {
-			bytes += packet_size_bytes;
-			ok_packet_cnt++;
-		}
-		total_packet_cnt++;
+	       (work_done < budget || ctx->frag_cnt)) {
+		next_desc = &rx->desc.desc_ring[(idx + 1) & rx->mask];
+		prefetch(next_desc);
+
+		gve_rx(rx, feat, desc, idx, &cnts);
+
+		rx->cnt++;
 		idx = rx->cnt & rx->mask;
 		desc = &rx->desc.desc_ring[idx];
-		work_done += work_cnt;
+		rx->desc.seqno = gve_next_seqno(rx->desc.seqno);
+		work_done++;
+	}
+
+	// The device will only send whole packets.
+	if (unlikely(ctx->frag_cnt)) {
+		struct napi_struct *napi = &priv->ntfy_blocks[rx->ntfy_id].napi;
+
+		napi_free_frags(napi);
+		gve_rx_ctx_clear(&rx->ctx);
+		netdev_warn(priv->dev, "Unexpected seq number %d with incomplete packet, expected %d, scheduling reset",
+			    GVE_SEQNO(desc->flags_seq), rx->desc.seqno);
+		gve_schedule_reset(rx->gve);
 	}
 
 	if (!work_done && rx->fill_cnt - rx->cnt > rx->db_threshold)
@@ -741,8 +811,10 @@  static int gve_clean_rx_done(struct gve_rx_ring *rx, int budget,
 
 	if (work_done) {
 		u64_stats_update_begin(&rx->statss);
-		rx->rpackets += ok_packet_cnt;
-		rx->rbytes += bytes;
+		rx->rpackets += cnts.ok_pkt_cnt;
+		rx->rbytes += cnts.ok_pkt_bytes;
+		rx->rx_cont_packet_cnt += cnts.cont_pkt_cnt;
+		rx->rx_desc_err_dropped_pkt += cnts.desc_err_pkt_cnt;
 		u64_stats_update_end(&rx->statss);
 	}
 
@@ -767,7 +839,7 @@  static int gve_clean_rx_done(struct gve_rx_ring *rx, int budget,
 	}
 
 	gve_rx_write_doorbell(priv, rx);
-	return total_packet_cnt;
+	return cnts.total_pkt_cnt;
 }
 
 int gve_rx_poll(struct gve_notify_block *block, int budget)
diff --git a/drivers/net/ethernet/google/gve/gve_rx_dqo.c b/drivers/net/ethernet/google/gve/gve_rx_dqo.c
index 2e6461b0ea8b..630f42a3037b 100644
--- a/drivers/net/ethernet/google/gve/gve_rx_dqo.c
+++ b/drivers/net/ethernet/google/gve/gve_rx_dqo.c
@@ -568,7 +568,7 @@  static int gve_rx_dqo(struct napi_struct *napi, struct gve_rx_ring *rx,
 
 	if (eop && buf_len <= priv->rx_copybreak) {
 		rx->ctx.skb_head = gve_rx_copy(priv->dev, napi,
-					       &buf_state->page_info, buf_len, 0, NULL);
+					       &buf_state->page_info, buf_len, 0);
 		if (unlikely(!rx->ctx.skb_head))
 			goto error;
 		rx->ctx.skb_tail = rx->ctx.skb_head;
diff --git a/drivers/net/ethernet/google/gve/gve_utils.c b/drivers/net/ethernet/google/gve/gve_utils.c
index d57508bc4307..6ba46adaaee3 100644
--- a/drivers/net/ethernet/google/gve/gve_utils.c
+++ b/drivers/net/ethernet/google/gve/gve_utils.c
@@ -50,34 +50,18 @@  void gve_rx_add_to_block(struct gve_priv *priv, int queue_idx)
 
 struct sk_buff *gve_rx_copy(struct net_device *dev, struct napi_struct *napi,
 			    struct gve_rx_slot_page_info *page_info, u16 len,
-			    u16 padding, struct gve_rx_ctx *ctx)
+			    u16 padding)
 {
 	void *va = page_info->page_address + padding + page_info->page_offset;
-	int skb_linear_offset = 0;
-	bool set_protocol = false;
 	struct sk_buff *skb;
 
-	if (ctx) {
-		if (!ctx->skb_head)
-			ctx->skb_head = napi_alloc_skb(napi, ctx->total_expected_size);
-
-		if (unlikely(!ctx->skb_head))
-			return NULL;
-		skb = ctx->skb_head;
-		skb_linear_offset = skb->len;
-		set_protocol = ctx->curr_frag_cnt == ctx->expected_frag_cnt - 1;
-	} else {
-		skb = napi_alloc_skb(napi, len);
-
-		if (unlikely(!skb))
-			return NULL;
-		set_protocol = true;
-	}
-	__skb_put(skb, len);
-	skb_copy_to_linear_data_offset(skb, skb_linear_offset, va, len);
+	skb = napi_alloc_skb(napi, len);
+	if (unlikely(!skb))
+		return NULL;
 
-	if (set_protocol)
-		skb->protocol = eth_type_trans(skb, dev);
+	__skb_put(skb, len);
+	skb_copy_to_linear_data_offset(skb, 0, va, len);
+	skb->protocol = eth_type_trans(skb, dev);
 
 	return skb;
 }
diff --git a/drivers/net/ethernet/google/gve/gve_utils.h b/drivers/net/ethernet/google/gve/gve_utils.h
index 6d98e69fd3b8..79595940b351 100644
--- a/drivers/net/ethernet/google/gve/gve_utils.h
+++ b/drivers/net/ethernet/google/gve/gve_utils.h
@@ -19,7 +19,7 @@  void gve_rx_add_to_block(struct gve_priv *priv, int queue_idx);
 
 struct sk_buff *gve_rx_copy(struct net_device *dev, struct napi_struct *napi,
 			    struct gve_rx_slot_page_info *page_info, u16 len,
-			    u16 pad, struct gve_rx_ctx *ctx);
+			    u16 pad);
 
 /* Decrement pagecnt_bias. Set it back to INT_MAX if it reached zero. */
 void gve_dec_pagecnt_bias(struct gve_rx_slot_page_info *page_info);