[net-next PATCH 0/4] Clean up intel driver page reuse Rx code

[net-next PATCH 0/4] Clean up intel driver page reuse Rx code

[Alexander Duyck]

This patch series cleans up the page reuse and Rx path so that all of the
fixes that have been applied to any one driver are now in place for igb,
fm10k, and ixgbe.  It occured to me that fm10k was missing the pfmemalloc
bits, and ixgbe was missing some logic that reduced the number of writes
needed as well as the pfmemalloc fix.

In addition ixgbe was carrying around some mostly-dead code that was
wrapping a call to writel.  I removed it since it masked the fact that
ixgbe was missing the mmiowb it was supposed to have between the tail write
and the Tx queue lock release to prevent the MMIO access from racing
between CPUs.

Also one change made to all 3 drivers is that we now only overwrite 3 of
the 4 DWORDs in the Rx descriptor after allocation.  The last DWORD
contains the upper 32 bits of the header address on fetch, and the length
and vlan on writeback.  Since we are no longer using header split we don't
need to clear it prior to descriptor fetch as it is unused so we can save
ourselves a cycle on 32b systems by just leaving it untouched.

---

Alexander Duyck (4):
      igb: Clean-up page reuse code
      fm10k: Clean-up page reuse code
      ixgbe: Clean-up page reuse code
      ixgbe: Remove tail write abstraction and add missing barrier


 drivers/net/ethernet/intel/fm10k/fm10k_main.c |   34 ++++---
 drivers/net/ethernet/intel/igb/igb_main.c     |   35 +++----
 drivers/net/ethernet/intel/ixgbe/ixgbe.h      |    5 -
 drivers/net/ethernet/intel/ixgbe/ixgbe_main.c |  118 ++++++++++++-------------
 4 files changed, 89 insertions(+), 103 deletions(-)

--

[net-next PATCH 1/4] igb: Clean-up page reuse code

[Alexander Duyck]

This patch cleans up the page reuse code getting it into a state where all
the workarounds needed are in place as well as cleaning up a few minor
oversights such as using __free_pages instead of put_page to drop a locally
allocated page.

It also cleans up how we clear the descriptor status bits.  Previously they
were zeroed as a part of clearing the hdr_addr.  However the hdr_addr is a
64 bit field and 64 bit writes can be a bit more expensive on on 32 bit
systems.  Since we are no longer using the header split feature the upper
32 bits of the address no longer need to be cleared.  As a result we can
just clear the status bits and leave the length and VLAN fields as-is which
should provide more information in debugging.

Cc: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
---
 drivers/net/ethernet/intel/igb/igb_main.c |   35 +++++++++++++----------------
 1 file changed, 16 insertions(+), 19 deletions(-)

diff --git a/drivers/net/ethernet/intel/igb/igb_main.c b/drivers/net/ethernet/intel/igb/igb_main.c
index 1e35fae..ade207d 100644
--- a/drivers/net/ethernet/intel/igb/igb_main.c
+++ b/drivers/net/ethernet/intel/igb/igb_main.c
@@ -6529,15 +6529,17 @@ static void igb_reuse_rx_page(struct igb_ring *rx_ring,
 					 DMA_FROM_DEVICE);
 }
 
+static inline bool igb_page_is_reserved(struct page *page)
+{
+	return (page_to_nid(page) != numa_mem_id()) || page->pfmemalloc;
+}
+
 static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer,
 				  struct page *page,
 				  unsigned int truesize)
 {
 	/* avoid re-using remote pages */
-	if (unlikely(page_to_nid(page) != numa_node_id()))
-		return false;
-
-	if (unlikely(page->pfmemalloc))
+	if (unlikely(igb_page_is_reserved(page)))
 		return false;
 
 #if (PAGE_SIZE < 8192)
@@ -6547,22 +6549,19 @@ static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer,
 
 	/* flip page offset to other buffer */
 	rx_buffer->page_offset ^= IGB_RX_BUFSZ;
-
-	/* Even if we own the page, we are not allowed to use atomic_set()
-	 * This would break get_page_unless_zero() users.
-	 */
-	atomic_inc(&page->_count);
 #else
 	/* move offset up to the next cache line */
 	rx_buffer->page_offset += truesize;
 
 	if (rx_buffer->page_offset > (PAGE_SIZE - IGB_RX_BUFSZ))
 		return false;
-
-	/* bump ref count on page before it is given to the stack */
-	get_page(page);
 #endif
 
+	/* Even if we own the page, we are not allowed to use atomic_set()
+	 * This would break get_page_unless_zero() users.
+	 */
+	atomic_inc(&page->_count);
+
 	return true;
 }
 
@@ -6605,13 +6604,12 @@ static bool igb_add_rx_frag(struct igb_ring *rx_ring,
 
 		memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
 
-		/* we can reuse buffer as-is, just make sure it is local */
-		if (likely((page_to_nid(page) == numa_node_id()) &&
-			   !page->pfmemalloc))
+		/* page is not reserved, we can reuse buffer as-is */
+		if (likely(!igb_page_is_reserved(page)))
 			return true;
 
 		/* this page cannot be reused so discard it */
-		put_page(page);
+		__free_page(page);
 		return false;
 	}
 
@@ -6629,7 +6627,6 @@ static struct sk_buff *igb_fetch_rx_buffer(struct igb_ring *rx_ring,
 	struct page *page;
 
 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
-
 	page = rx_buffer->page;
 	prefetchw(page);
 
@@ -7050,8 +7047,8 @@ void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
 			i -= rx_ring->count;
 		}
 
-		/* clear the hdr_addr for the next_to_use descriptor */
-		rx_desc->read.hdr_addr = 0;
+		/* clear the status bits for the next_to_use descriptor */
+		rx_desc->wb.upper.status_error = 0;
 
 		cleaned_count--;
 	} while (cleaned_count);

[net-next PATCH 2/4] fm10k: Clean-up page reuse code

[Alexander Duyck]

This patch cleans up the page reuse code getting it into a state where all
the workarounds needed are in place as well as cleaning up a few minor
oversights such as using __free_pages instead of put_page to drop a locally
allocated page.

It also cleans up how we clear the descriptor status bits.  Previously they
were zeroed as a part of clearing the hdr_addr.  However the hdr_addr is a
64 bit field and 64 bit writes can be a bit more expensive on on 32 bit
systems.  Since we are no longer using the header split feature the upper
32 bits of the address no longer need to be cleared.  As a result we can
just clear the status bits and leave the length and VLAN fields as-is which
should provide more information in debugging.

Cc: Matthew Vick <matthew.vick@intel.com>
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
---
 drivers/net/ethernet/intel/fm10k/fm10k_main.c |   34 +++++++++++++------------
 1 file changed, 17 insertions(+), 17 deletions(-)

diff --git a/drivers/net/ethernet/intel/fm10k/fm10k_main.c b/drivers/net/ethernet/intel/fm10k/fm10k_main.c
index 73457ed..8c8deec 100644
--- a/drivers/net/ethernet/intel/fm10k/fm10k_main.c
+++ b/drivers/net/ethernet/intel/fm10k/fm10k_main.c
@@ -97,7 +97,6 @@ static bool fm10k_alloc_mapped_page(struct fm10k_ring *rx_ring,
 	 */
 	if (dma_mapping_error(rx_ring->dev, dma)) {
 		__free_page(page);
-		bi->page = NULL;
 
 		rx_ring->rx_stats.alloc_failed++;
 		return false;
@@ -147,8 +146,8 @@ void fm10k_alloc_rx_buffers(struct fm10k_ring *rx_ring, u16 cleaned_count)
 			i -= rx_ring->count;
 		}
 
-		/* clear the hdr_addr for the next_to_use descriptor */
-		rx_desc->q.hdr_addr = 0;
+		/* clear the status bits for the next_to_use descriptor */
+		rx_desc->d.staterr = 0;
 
 		cleaned_count--;
 	} while (cleaned_count);
@@ -194,7 +193,7 @@ static void fm10k_reuse_rx_page(struct fm10k_ring *rx_ring,
 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
 
 	/* transfer page from old buffer to new buffer */
-	memcpy(new_buff, old_buff, sizeof(struct fm10k_rx_buffer));
+	*new_buff = *old_buff;
 
 	/* sync the buffer for use by the device */
 	dma_sync_single_range_for_device(rx_ring->dev, old_buff->dma,
@@ -203,12 +202,17 @@ static void fm10k_reuse_rx_page(struct fm10k_ring *rx_ring,
 					 DMA_FROM_DEVICE);
 }
 
+static inline bool fm10k_page_is_reserved(struct page *page)
+{
+	return (page_to_nid(page) != numa_mem_id()) || page->pfmemalloc;
+}
+
 static bool fm10k_can_reuse_rx_page(struct fm10k_rx_buffer *rx_buffer,
 				    struct page *page,
 				    unsigned int truesize)
 {
 	/* avoid re-using remote pages */
-	if (unlikely(page_to_nid(page) != numa_mem_id()))
+	if (unlikely(fm10k_page_is_reserved(page)))
 		return false;
 
 #if (PAGE_SIZE < 8192)
@@ -218,22 +222,19 @@ static bool fm10k_can_reuse_rx_page(struct fm10k_rx_buffer *rx_buffer,
 
 	/* flip page offset to other buffer */
 	rx_buffer->page_offset ^= FM10K_RX_BUFSZ;
-
-	/* Even if we own the page, we are not allowed to use atomic_set()
-	 * This would break get_page_unless_zero() users.
-	 */
-	atomic_inc(&page->_count);
 #else
 	/* move offset up to the next cache line */
 	rx_buffer->page_offset += truesize;
 
 	if (rx_buffer->page_offset > (PAGE_SIZE - FM10K_RX_BUFSZ))
 		return false;
-
-	/* bump ref count on page before it is given to the stack */
-	get_page(page);
 #endif
 
+	/* Even if we own the page, we are not allowed to use atomic_set()
+	 * This would break get_page_unless_zero() users.
+	 */
+	atomic_inc(&page->_count);
+
 	return true;
 }
 
@@ -270,12 +271,12 @@ static bool fm10k_add_rx_frag(struct fm10k_ring *rx_ring,
 
 		memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
 
-		/* we can reuse buffer as-is, just make sure it is local */
-		if (likely(page_to_nid(page) == numa_mem_id()))
+		/* page is not reserved, we can reuse buffer as-is */
+		if (likely(!fm10k_page_is_reserved(page)))
 			return true;
 
 		/* this page cannot be reused so discard it */
-		put_page(page);
+		__free_page(page);
 		return false;
 	}
 
@@ -293,7 +294,6 @@ static struct sk_buff *fm10k_fetch_rx_buffer(struct fm10k_ring *rx_ring,
 	struct page *page;
 
 	rx_buffer = &rx_ring->rx_buffer[rx_ring->next_to_clean];
-
 	page = rx_buffer->page;
 	prefetchw(page);
 

[net-next PATCH 3/4] ixgbe: Clean-up page reuse code

[Alexander Duyck]

This patch cleans up the page reuse code getting it into a state where all
the workarounds needed are in place as well as cleaning up a few minor
oversights such as using __free_pages instead of put_page to drop a locally
allocated page.

It also cleans up how we clear the descriptor status bits.  Previously they
were zeroed as a part of clearing the hdr_addr.  However the hdr_addr is a
64 bit field and 64 bit writes can be a bit more expensive on on 32 bit
systems.  Since we are no longer using the header split feature the upper
32 bits of the address no longer need to be cleared.  As a result we can
just clear the status bits and leave the length and VLAN fields as-is which
should provide more information in debugging.


Cc: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Cc: Don Skidmore <donald.c.skidmore@intel.com>
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
---
 drivers/net/ethernet/intel/ixgbe/ixgbe_main.c |   78 ++++++++++++-------------
 1 file changed, 36 insertions(+), 42 deletions(-)

diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c b/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c
index f5fcba4..a72c852 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c
+++ b/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c
@@ -1436,20 +1436,17 @@ static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
 				    struct ixgbe_rx_buffer *bi)
 {
 	struct page *page = bi->page;
-	dma_addr_t dma = bi->dma;
+	dma_addr_t dma;
 
 	/* since we are recycling buffers we should seldom need to alloc */
-	if (likely(dma))
+	if (likely(page))
 		return true;
 
 	/* alloc new page for storage */
-	if (likely(!page)) {
-		page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
-		if (unlikely(!page)) {
-			rx_ring->rx_stats.alloc_rx_page_failed++;
-			return false;
-		}
-		bi->page = page;
+	page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
+	if (unlikely(!page)) {
+		rx_ring->rx_stats.alloc_rx_page_failed++;
+		return false;
 	}
 
 	/* map page for use */
@@ -1462,13 +1459,13 @@ static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
 	 */
 	if (dma_mapping_error(rx_ring->dev, dma)) {
 		__free_pages(page, ixgbe_rx_pg_order(rx_ring));
-		bi->page = NULL;
 
 		rx_ring->rx_stats.alloc_rx_page_failed++;
 		return false;
 	}
 
 	bi->dma = dma;
+	bi->page = page;
 	bi->page_offset = 0;
 
 	return true;
@@ -1512,8 +1509,8 @@ void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
 			i -= rx_ring->count;
 		}
 
-		/* clear the hdr_addr for the next_to_use descriptor */
-		rx_desc->read.hdr_addr = 0;
+		/* clear the status bits for the next_to_use descriptor */
+		rx_desc->wb.upper.status_error = 0;
 
 		cleaned_count--;
 	} while (cleaned_count);
@@ -1798,9 +1795,7 @@ static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
 	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
 
 	/* transfer page from old buffer to new buffer */
-	new_buff->page = old_buff->page;
-	new_buff->dma = old_buff->dma;
-	new_buff->page_offset = old_buff->page_offset;
+	*new_buff = *old_buff;
 
 	/* sync the buffer for use by the device */
 	dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
@@ -1809,6 +1804,11 @@ static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
 					 DMA_FROM_DEVICE);
 }
 
+static inline bool ixgbe_page_is_reserved(struct page *page)
+{
+	return (page_to_nid(page) != numa_mem_id()) || page->pfmemalloc;
+}
+
 /**
  * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
  * @rx_ring: rx descriptor ring to transact packets on
@@ -1844,12 +1844,12 @@ static bool ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
 
 		memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
 
-		/* we can reuse buffer as-is, just make sure it is local */
-		if (likely(page_to_nid(page) == numa_node_id()))
+		/* page is not reserved, we can reuse buffer as-is */
+		if (likely(!ixgbe_page_is_reserved(page)))
 			return true;
 
 		/* this page cannot be reused so discard it */
-		put_page(page);
+		__free_pages(page, ixgbe_rx_pg_order(rx_ring));
 		return false;
 	}
 
@@ -1857,7 +1857,7 @@ static bool ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
 			rx_buffer->page_offset, size, truesize);
 
 	/* avoid re-using remote pages */
-	if (unlikely(page_to_nid(page) != numa_node_id()))
+	if (unlikely(ixgbe_page_is_reserved(page)))
 		return false;
 
 #if (PAGE_SIZE < 8192)
@@ -1867,22 +1867,19 @@ static bool ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
 
 	/* flip page offset to other buffer */
 	rx_buffer->page_offset ^= truesize;
-
-	/* Even if we own the page, we are not allowed to use atomic_set()
-	 * This would break get_page_unless_zero() users.
-	 */
-	atomic_inc(&page->_count);
 #else
 	/* move offset up to the next cache line */
 	rx_buffer->page_offset += truesize;
 
 	if (rx_buffer->page_offset > last_offset)
 		return false;
-
-	/* bump ref count on page before it is given to the stack */
-	get_page(page);
 #endif
 
+	/* Even if we own the page, we are not allowed to use atomic_set()
+	 * This would break get_page_unless_zero() users.
+	 */
+	atomic_inc(&page->_count);
+
 	return true;
 }
 
@@ -1945,6 +1942,8 @@ dma_sync:
 					      rx_buffer->page_offset,
 					      ixgbe_rx_bufsz(rx_ring),
 					      DMA_FROM_DEVICE);
+
+		rx_buffer->skb = NULL;
 	}
 
 	/* pull page into skb */
@@ -1962,8 +1961,6 @@ dma_sync:
 	}
 
 	/* clear contents of buffer_info */
-	rx_buffer->skb = NULL;
-	rx_buffer->dma = 0;
 	rx_buffer->page = NULL;
 
 	return skb;
@@ -4345,29 +4342,26 @@ static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
 
 	/* Free all the Rx ring sk_buffs */
 	for (i = 0; i < rx_ring->count; i++) {
-		struct ixgbe_rx_buffer *rx_buffer;
+		struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
 
-		rx_buffer = &rx_ring->rx_buffer_info[i];
 		if (rx_buffer->skb) {
 			struct sk_buff *skb = rx_buffer->skb;
-			if (IXGBE_CB(skb)->page_released) {
+			if (IXGBE_CB(skb)->page_released)
 				dma_unmap_page(dev,
 					       IXGBE_CB(skb)->dma,
 					       ixgbe_rx_bufsz(rx_ring),
 					       DMA_FROM_DEVICE);
-				IXGBE_CB(skb)->page_released = false;
-			}
 			dev_kfree_skb(skb);
 			rx_buffer->skb = NULL;
 		}
-		if (rx_buffer->dma)
-			dma_unmap_page(dev, rx_buffer->dma,
-				       ixgbe_rx_pg_size(rx_ring),
-				       DMA_FROM_DEVICE);
-		rx_buffer->dma = 0;
-		if (rx_buffer->page)
-			__free_pages(rx_buffer->page,
-				     ixgbe_rx_pg_order(rx_ring));
+
+		if (!rx_buffer->page)
+			continue;
+
+		dma_unmap_page(dev, rx_buffer->dma,
+			       ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
+		__free_pages(rx_buffer->page, ixgbe_rx_pg_order(rx_ring));
+
 		rx_buffer->page = NULL;
 	}
 

[net-next PATCH 4/4] ixgbe: Remove tail write abstraction and add missing barrier

[Alexander Duyck]

This change cleans up the tail writes for the ixgbe descriptor queues.  The
current implementation had me confused as I wasn't sure if it was still
making use of the suprise remove logic or not.

It also adds the mmiowb which is needed on ia64, mips, and a couple other
architectures in order to synchronize the MMIO writes with the Tx queue
_xmit_lock spinlock.

Cc: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Cc: Don Skidmore <donald.c.skidmore@intel.com>
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
---
 drivers/net/ethernet/intel/ixgbe/ixgbe.h      |    5 ---
 drivers/net/ethernet/intel/ixgbe/ixgbe_main.c |   40 +++++++++++++------------
 2 files changed, 20 insertions(+), 25 deletions(-)

diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe.h b/drivers/net/ethernet/intel/ixgbe/ixgbe.h
index 5032a60..86fa607 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe.h
+++ b/drivers/net/ethernet/intel/ixgbe/ixgbe.h
@@ -553,11 +553,6 @@ static inline u16 ixgbe_desc_unused(struct ixgbe_ring *ring)
 	return ((ntc > ntu) ? 0 : ring->count) + ntc - ntu - 1;
 }
 
-static inline void ixgbe_write_tail(struct ixgbe_ring *ring, u32 value)
-{
-	writel(value, ring->tail);
-}
-
 #define IXGBE_RX_DESC(R, i)	    \
 	(&(((union ixgbe_adv_rx_desc *)((R)->desc))[i]))
 #define IXGBE_TX_DESC(R, i)	    \
diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c b/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c
index a72c852..6b29dfe 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c
+++ b/drivers/net/ethernet/intel/ixgbe/ixgbe_main.c
@@ -1416,22 +1416,6 @@ static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
 	skb->ip_summed = CHECKSUM_UNNECESSARY;
 }
 
-static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
-{
-	rx_ring->next_to_use = val;
-
-	/* update next to alloc since we have filled the ring */
-	rx_ring->next_to_alloc = val;
-	/*
-	 * Force memory writes to complete before letting h/w
-	 * know there are new descriptors to fetch.  (Only
-	 * applicable for weak-ordered memory model archs,
-	 * such as IA-64).
-	 */
-	wmb();
-	ixgbe_write_tail(rx_ring, val);
-}
-
 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
 				    struct ixgbe_rx_buffer *bi)
 {
@@ -1517,8 +1501,20 @@ void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
 
 	i += rx_ring->count;
 
-	if (rx_ring->next_to_use != i)
-		ixgbe_release_rx_desc(rx_ring, i);
+	if (rx_ring->next_to_use != i) {
+		rx_ring->next_to_use = i;
+
+		/* update next to alloc since we have filled the ring */
+		rx_ring->next_to_alloc = i;
+
+		/* Force memory writes to complete before letting h/w
+		 * know there are new descriptors to fetch.  (Only
+		 * applicable for weak-ordered memory model archs,
+		 * such as IA-64).
+		 */
+		wmb();
+		writel(i, rx_ring->tail);
+	}
 }
 
 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
@@ -6955,8 +6951,12 @@ static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
 	ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
 
 	if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) {
-		/* notify HW of packet */
-		ixgbe_write_tail(tx_ring, i);
+		writel(i, tx_ring->tail);
+
+		/* we need this if more than one processor can write to our tail
+		 * at a time, it synchronizes IO on IA64/Altix systems
+		 */
+		mmiowb();
 	}
 
 	return;

Re: [net-next PATCH 0/4] Clean up intel driver page reuse Rx code

[Jeff Kirsher]

[-- Attachment #1: Type: text/plain, Size: 1801 bytes --]

On Thu, 2014-11-13 at 08:18 -0800, Alexander Duyck wrote:
> This patch series cleans up the page reuse and Rx path so that all of the
> fixes that have been applied to any one driver are now in place for igb,
> fm10k, and ixgbe.  It occured to me that fm10k was missing the pfmemalloc
> bits, and ixgbe was missing some logic that reduced the number of writes
> needed as well as the pfmemalloc fix.
> 
> In addition ixgbe was carrying around some mostly-dead code that was
> wrapping a call to writel.  I removed it since it masked the fact that
> ixgbe was missing the mmiowb it was supposed to have between the tail write
> and the Tx queue lock release to prevent the MMIO access from racing
> between CPUs.
> 
> Also one change made to all 3 drivers is that we now only overwrite 3 of
> the 4 DWORDs in the Rx descriptor after allocation.  The last DWORD
> contains the upper 32 bits of the header address on fetch, and the length
> and vlan on writeback.  Since we are no longer using header split we don't
> need to clear it prior to descriptor fetch as it is unused so we can save
> ourselves a cycle on 32b systems by just leaving it untouched.
> 
> ---
> 
> Alexander Duyck (4):
>       igb: Clean-up page reuse code
>       fm10k: Clean-up page reuse code
>       ixgbe: Clean-up page reuse code
>       ixgbe: Remove tail write abstraction and add missing barrier
> 
> 
>  drivers/net/ethernet/intel/fm10k/fm10k_main.c |   34 ++++---
>  drivers/net/ethernet/intel/igb/igb_main.c     |   35 +++----
>  drivers/net/ethernet/intel/ixgbe/ixgbe.h      |    5 -
>  drivers/net/ethernet/intel/ixgbe/ixgbe_main.c |  118 ++++++++++++-------------
>  4 files changed, 89 insertions(+), 103 deletions(-)
> 
> --

Thanks Alex, I will add your series to my queue.

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 819 bytes --]