From: Catherine Sullivan <csully@google.com>
Date: Mon, 11 Oct 2021 08:36:45 -0700
Subject: gve: Add rx buffer pagecnt bias
Patch-mainline: v5.16-rc1
Git-commit: 58401b2a46e7d5090873e7725ac786685475d638
References: bsc#1191655

Add a pagecnt bias field to rx buffer info struct to eliminate
needing to increment the atomic page ref count on every pass in the
rx hotpath.

Also prefetch two packet pages ahead.

Fixes: ede3fcf5ec67f ("gve: Add support for raw addressing to the rx path")
Signed-off-by: Yanchun Fu <yangchun@google.com>
Signed-off-by: Nathan Lewis <npl@google.com>
Signed-off-by: Catherine Sullivan <csully@google.com>
Signed-off-by: David Awogbemila <awogbemila@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Acked-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
---
 drivers/net/ethernet/google/gve/gve_rx.c |   52 ++++++++++++++++++++++---------
 1 file changed, 37 insertions(+), 15 deletions(-)

--- a/drivers/net/ethernet/google/gve/gve_rx.c
+++ b/drivers/net/ethernet/google/gve/gve_rx.c
@@ -16,19 +16,23 @@ static void gve_rx_free_buffer(struct de
 	dma_addr_t dma = (dma_addr_t)(be64_to_cpu(data_slot->addr) &
 				      GVE_DATA_SLOT_ADDR_PAGE_MASK);
 
+	page_ref_sub(page_info->page, page_info->pagecnt_bias - 1);
 	gve_free_page(dev, page_info->page, dma, DMA_FROM_DEVICE);
 }
 
 static void gve_rx_unfill_pages(struct gve_priv *priv, struct gve_rx_ring *rx)
 {
-	if (rx->data.raw_addressing) {
-		u32 slots = rx->mask + 1;
-		int i;
+	u32 slots = rx->mask + 1;
+	int i;
 
+	if (rx->data.raw_addressing) {
 		for (i = 0; i < slots; i++)
 			gve_rx_free_buffer(&priv->pdev->dev, &rx->data.page_info[i],
 					   &rx->data.data_ring[i]);
 	} else {
+		for (i = 0; i < slots; i++)
+			page_ref_sub(rx->data.page_info[i].page,
+				     rx->data.page_info[i].pagecnt_bias - 1);
 		gve_unassign_qpl(priv, rx->data.qpl->id);
 		rx->data.qpl = NULL;
 	}
@@ -69,6 +73,9 @@ static void gve_setup_rx_buffer(struct g
 	page_info->page_offset = 0;
 	page_info->page_address = page_address(page);
 	*slot_addr = cpu_to_be64(addr);
+	/* The page already has 1 ref */
+	page_ref_add(page, INT_MAX - 1);
+	page_info->pagecnt_bias = INT_MAX;
 }
 
 static int gve_rx_alloc_buffer(struct gve_priv *priv, struct device *dev,
@@ -299,17 +306,18 @@ static bool gve_rx_can_flip_buffers(stru
 		? netdev->mtu + GVE_RX_PAD + ETH_HLEN <= PAGE_SIZE / 2 : false;
 }
 
-static int gve_rx_can_recycle_buffer(struct page *page)
+static int gve_rx_can_recycle_buffer(struct gve_rx_slot_page_info *page_info)
 {
-	int pagecount = page_count(page);
+	int pagecount = page_count(page_info->page);
 
 	/* This page is not being used by any SKBs - reuse */
-	if (pagecount == 1)
+	if (pagecount == page_info->pagecnt_bias)
 		return 1;
 	/* This page is still being used by an SKB - we can't reuse */
-	else if (pagecount >= 2)
+	else if (pagecount > page_info->pagecnt_bias)
 		return 0;
-	WARN(pagecount < 1, "Pagecount should never be < 1");
+	WARN(pagecount < page_info->pagecnt_bias,
+	     "Pagecount should never be less than the bias.");
 	return -1;
 }
 
@@ -325,11 +333,11 @@ gve_rx_raw_addressing(struct device *dev
 	if (!skb)
 		return NULL;
 
-	/* Optimistically stop the kernel from freeing the page by increasing
-	 * the page bias. We will check the refcount in refill to determine if
-	 * we need to alloc a new page.
+	/* Optimistically stop the kernel from freeing the page.
+	 * We will check again in refill to determine if we need to alloc a
+	 * new page.
 	 */
-	get_page(page_info->page);
+	gve_dec_pagecnt_bias(page_info);
 
 	return skb;
 }
@@ -352,7 +360,7 @@ gve_rx_qpl(struct device *dev, struct ne
 		/* No point in recycling if we didn't get the skb */
 		if (skb) {
 			/* Make sure that the page isn't freed. */
-			get_page(page_info->page);
+			gve_dec_pagecnt_bias(page_info);
 			gve_rx_flip_buff(page_info, &data_slot->qpl_offset);
 		}
 	} else {
@@ -376,8 +384,18 @@ static bool gve_rx(struct gve_rx_ring *r
 	union gve_rx_data_slot *data_slot;
 	struct sk_buff *skb = NULL;
 	dma_addr_t page_bus;
+	void *va;
 	u16 len;
 
+	/* Prefetch two packet pages ahead, we will need it soon. */
+	page_info = &rx->data.page_info[(idx + 2) & rx->mask];
+	va = page_info->page_address + GVE_RX_PAD +
+		page_info->page_offset;
+
+	prefetch(page_info->page); /* Kernel page struct. */
+	prefetch(va);              /* Packet header. */
+	prefetch(va + 64);         /* Next cacheline too. */
+
 	/* drop this packet */
 	if (unlikely(rx_desc->flags_seq & GVE_RXF_ERR)) {
 		u64_stats_update_begin(&rx->statss);
@@ -408,7 +426,7 @@ static bool gve_rx(struct gve_rx_ring *r
 		int recycle = 0;
 
 		if (can_flip) {
-			recycle = gve_rx_can_recycle_buffer(page_info->page);
+			recycle = gve_rx_can_recycle_buffer(page_info);
 			if (recycle < 0) {
 				if (!rx->data.raw_addressing)
 					gve_schedule_reset(priv);
@@ -499,7 +517,7 @@ static bool gve_rx_refill_buffers(struct
 			 * owns half the page it is impossible to tell which half. Either
 			 * the whole page is free or it needs to be replaced.
 			 */
-			int recycle = gve_rx_can_recycle_buffer(page_info->page);
+			int recycle = gve_rx_can_recycle_buffer(page_info);
 
 			if (recycle < 0) {
 				if (!rx->data.raw_addressing)
@@ -546,6 +564,10 @@ static int gve_clean_rx_done(struct gve_
 			   "[%d] seqno=%d rx->desc.seqno=%d\n",
 			   rx->q_num, GVE_SEQNO(desc->flags_seq),
 			   rx->desc.seqno);
+
+		/* prefetch two descriptors ahead */
+		prefetch(rx->desc.desc_ring + ((cnt + 2) & rx->mask));
+
 		dropped = !gve_rx(rx, desc, feat, idx);
 		if (!dropped) {
 			bytes += be16_to_cpu(desc->len) - GVE_RX_PAD;