[Intel-wired-lan] [PATCH 0/9] ixgbevf: update Rx/Tx code path for build_skb

[Intel-wired-lan] [PATCH 0/9] ixgbevf: update Rx/Tx code path for build_skb

[Emil Tantilov]

This series includes changes that were previously introduced in igb and
ixgbe. Mainly the changes from Alexander Duyck prior to the new Rx page
and build_skb support. The goal is to bring the Tx/Rx code path closer
to igb/ixgbe and eventually introduce XDP support as well.

---

Emil Tantilov (9):
      ixgbevf: add function for checking if we can reuse page
      ixgbevf: only DMA sync frame length
      ixgbevf: use length to determine if descriptor is done
      ixgbevf: add support for DMA_ATTR_SKIP_CPU_SYNC/WEAK_ORDERING
      ixgbevf: update code to better handle incrementing page count
      ixgbevf: add counters for Rx page allocations
      ixgbevf: clear rx_buffer_info in configure instead of clean
      ixgbevf: improve performance and reduce size of ixgbevf_tx_map()
      ixgbevf: don't bother clearing tx_buffer_info in ixgbevf_clean_tx_ring()


 drivers/net/ethernet/intel/ixgbevf/ethtool.c      |    3 
 drivers/net/ethernet/intel/ixgbevf/ixgbevf.h      |   16 +
 drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |  362 +++++++++++++--------
 3 files changed, 234 insertions(+), 147 deletions(-)

--
Signature

[Intel-wired-lan] [PATCH 1/9] ixgbevf: add function for checking if we can reuse page

[Emil Tantilov]

Introduce ixgbevf_can_reuse_page() similar to the change in ixgbe from
commit af43da0dba0b
("ixgbe: Add function for checking to see if we can reuse page")

Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
---
 drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |   59 ++++++++++++---------
 1 file changed, 33 insertions(+), 26 deletions(-)

diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
index ed5c3ae..7ffd429 100644
--- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
+++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
@@ -754,6 +754,38 @@ static inline bool ixgbevf_page_is_reserved(struct page *page)
 	return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
 }
 
+static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer,
+				      struct page *page,
+				      const unsigned int truesize)
+{
+	/* avoid re-using remote pages */
+	if (unlikely(ixgbevf_page_is_reserved(page)))
+		return false;
+
+#if (PAGE_SIZE < 8192)
+	/* if we are only owner of page we can reuse it */
+	if (unlikely(page_count(page) != 1))
+		return false;
+
+	/* flip page offset to other buffer */
+	rx_buffer->page_offset ^= IXGBEVF_RX_BUFSZ;
+
+#else
+	/* move offset up to the next cache line */
+	rx_buffer->page_offset += truesize;
+
+	if (rx_buffer->page_offset > (PAGE_SIZE - IXGBEVF_RX_BUFSZ))
+		return false;
+
+#endif
+	/* Even if we own the page, we are not allowed to use atomic_set()
+	 * This would break get_page_unless_zero() users.
+	 */
+	page_ref_inc(page);
+
+	return true;
+}
+
 /**
  * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
  * @rx_ring: rx descriptor ring to transact packets on
@@ -815,32 +847,7 @@ static bool ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
 	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
 			(unsigned long)va & ~PAGE_MASK, size, truesize);
 
-	/* avoid re-using remote pages */
-	if (unlikely(ixgbevf_page_is_reserved(page)))
-		return false;
-
-#if (PAGE_SIZE < 8192)
-	/* if we are only owner of page we can reuse it */
-	if (unlikely(page_count(page) != 1))
-		return false;
-
-	/* flip page offset to other buffer */
-	rx_buffer->page_offset ^= IXGBEVF_RX_BUFSZ;
-
-#else
-	/* move offset up to the next cache line */
-	rx_buffer->page_offset += truesize;
-
-	if (rx_buffer->page_offset > (PAGE_SIZE - IXGBEVF_RX_BUFSZ))
-		return false;
-
-#endif
-	/* Even if we own the page, we are not allowed to use atomic_set()
-	 * This would break get_page_unless_zero() users.
-	 */
-	page_ref_inc(page);
-
-	return true;
+	return ixgbevf_can_reuse_rx_page(rx_buffer, page, truesize);
 }
 
 static struct sk_buff *ixgbevf_fetch_rx_buffer(struct ixgbevf_ring *rx_ring,

[Intel-wired-lan] [PATCH 2/9] ixgbevf: only DMA sync frame length

[Emil Tantilov]

Based on commit 64f2525ca4e7 ("igb: Only DMA sync frame length")

On some architectures synching a buffer for DMA may be expensive.
Instead of the entire 2K receive buffer only synchronize the length of
the frame, which will typically be the MTU or smaller.

Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
---
 drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |    7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
index 7ffd429..0cc2688 100644
--- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
+++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
@@ -803,12 +803,12 @@ static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer,
  **/
 static bool ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
 				struct ixgbevf_rx_buffer *rx_buffer,
+				u16 size,
 				union ixgbe_adv_rx_desc *rx_desc,
 				struct sk_buff *skb)
 {
 	struct page *page = rx_buffer->page;
 	unsigned char *va = page_address(page) + rx_buffer->page_offset;
-	unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
 #if (PAGE_SIZE < 8192)
 	unsigned int truesize = IXGBEVF_RX_BUFSZ;
 #else
@@ -856,6 +856,7 @@ static struct sk_buff *ixgbevf_fetch_rx_buffer(struct ixgbevf_ring *rx_ring,
 {
 	struct ixgbevf_rx_buffer *rx_buffer;
 	struct page *page;
+	u16 size = le16_to_cpu(rx_desc->wb.upper.length);
 
 	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
 	page = rx_buffer->page;
@@ -890,11 +891,11 @@ static struct sk_buff *ixgbevf_fetch_rx_buffer(struct ixgbevf_ring *rx_ring,
 	dma_sync_single_range_for_cpu(rx_ring->dev,
 				      rx_buffer->dma,
 				      rx_buffer->page_offset,
-				      IXGBEVF_RX_BUFSZ,
+				      size,
 				      DMA_FROM_DEVICE);
 
 	/* pull page into skb */
-	if (ixgbevf_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
+	if (ixgbevf_add_rx_frag(rx_ring, rx_buffer, size, rx_desc, skb)) {
 		/* hand second half of page back to the ring */
 		ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
 	} else {

[Intel-wired-lan] [PATCH 3/9] ixgbevf: use length to determine if descriptor is done

[Emil Tantilov]

Based on:
commit 7ec0116c9131 ("igb: Use length to determine if descriptor is done")

This change makes it so that we use the length of the packet instead of the
DD status bit to determine if a new descriptor is ready to be processed.
The obvious advantage is that it cuts down on reads as we don't really even
need the DD bit if going from a 0 to a non-zero value on size is enough to
inform us that the packet has been completed.

In addition we only reset the Rx descriptor length for descriptor zero when
resetting a ring instead of having to do a memset with 0 over the entire
ring. By doing this we can save some time on initialization.

Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
---
 drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |   14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
index 0cc2688..725fe2d 100644
--- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
+++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
@@ -653,8 +653,8 @@ static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
 			i -= rx_ring->count;
 		}
 
-		/* clear the hdr_addr for the next_to_use descriptor */
-		rx_desc->read.hdr_addr = 0;
+		/* clear the length for the next_to_use descriptor */
+		rx_desc->wb.upper.length = 0;
 
 		cleaned_count--;
 	} while (cleaned_count);
@@ -938,7 +938,7 @@ static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
 
 		rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
 
-		if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
+		if (!rx_desc->wb.upper.length)
 			break;
 
 		/* This memory barrier is needed to keep us from reading
@@ -1729,6 +1729,7 @@ static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
 				      struct ixgbevf_ring *ring)
 {
 	struct ixgbe_hw *hw = &adapter->hw;
+	union ixgbe_adv_rx_desc *rx_desc;
 	u64 rdba = ring->dma;
 	u32 rxdctl;
 	u8 reg_idx = ring->reg_idx;
@@ -1757,6 +1758,10 @@ static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
 	IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
 	ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
 
+	/* initialize Rx descriptor 0 */
+	rx_desc = IXGBEVF_RX_DESC(ring, 0);
+	rx_desc->wb.upper.length = 0;
+
 	/* reset ntu and ntc to place SW in sync with hardwdare */
 	ring->next_to_clean = 0;
 	ring->next_to_use = 0;
@@ -2141,9 +2146,6 @@ static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
 
 	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
 	memset(rx_ring->rx_buffer_info, 0, size);
-
-	/* Zero out the descriptor ring */
-	memset(rx_ring->desc, 0, rx_ring->size);
 }
 
 /**

[Intel-wired-lan] [PATCH 4/9] ixgbevf: add support for DMA_ATTR_SKIP_CPU_SYNC/WEAK_ORDERING

[Emil Tantilov]

Based on commit 5be5955425c2
("igb: update driver to make use of DMA_ATTR_SKIP_CPU_SYNC")

and

commit 7bd175928280 ("igb: Add support for DMA_ATTR_WEAK_ORDERING")

Convert the calls to dma_map/unmap_page() to the attributes version
and add DMA_ATTR_SKIP_CPU_SYNC/WEAK_ORDERING which should help
improve performance on some platforms.

Move sync_for_cpu call before we perform a prefetch to avoid
invalidating the first 128 bytes of the packet on architectures where
that call may invalidate the cache.

Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
---
 drivers/net/ethernet/intel/ixgbevf/ixgbevf.h      |    3 +
 drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |   57 +++++++++++++--------
 2 files changed, 38 insertions(+), 22 deletions(-)

diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h b/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
index 581f44b..b1da9f4 100644
--- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
+++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
@@ -260,6 +260,9 @@ static inline void ixgbevf_write_tail(struct ixgbevf_ring *ring, u32 value)
 #define MIN_MSIX_Q_VECTORS	1
 #define MIN_MSIX_COUNT		(MIN_MSIX_Q_VECTORS + NON_Q_VECTORS)
 
+#define IXGBEVF_RX_DMA_ATTR \
+	(DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING)
+
 /* board specific private data structure */
 struct ixgbevf_adapter {
 	/* this field must be first, see ixgbevf_process_skb_fields */
diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
index 725fe2d..fbd493e 100644
--- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
+++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
@@ -595,8 +595,8 @@ static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
 	}
 
 	/* map page for use */
-	dma = dma_map_page(rx_ring->dev, page, 0,
-			   PAGE_SIZE, DMA_FROM_DEVICE);
+	dma = dma_map_page_attrs(rx_ring->dev, page, 0, PAGE_SIZE,
+				 DMA_FROM_DEVICE, IXGBEVF_RX_DMA_ATTR);
 
 	/* if mapping failed free memory back to system since
 	 * there isn't much point in holding memory we can't use
@@ -639,6 +639,12 @@ static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
 		if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
 			break;
 
+		/* sync the buffer for use by the device */
+		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
+						 bi->page_offset,
+						 IXGBEVF_RX_BUFSZ,
+						 DMA_FROM_DEVICE);
+
 		/* Refresh the desc even if pkt_addr didn't change
 		 * because each write-back erases this info.
 		 */
@@ -741,12 +747,6 @@ static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
 	new_buff->page = old_buff->page;
 	new_buff->dma = old_buff->dma;
 	new_buff->page_offset = old_buff->page_offset;
-
-	/* sync the buffer for use by the device */
-	dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
-					 new_buff->page_offset,
-					 IXGBEVF_RX_BUFSZ,
-					 DMA_FROM_DEVICE);
 }
 
 static inline bool ixgbevf_page_is_reserved(struct page *page)
@@ -862,6 +862,13 @@ static struct sk_buff *ixgbevf_fetch_rx_buffer(struct ixgbevf_ring *rx_ring,
 	page = rx_buffer->page;
 	prefetchw(page);
 
+	/* we are reusing so sync this buffer for CPU use */
+	dma_sync_single_range_for_cpu(rx_ring->dev,
+				      rx_buffer->dma,
+				      rx_buffer->page_offset,
+				      size,
+				      DMA_FROM_DEVICE);
+
 	if (likely(!skb)) {
 		void *page_addr = page_address(page) +
 				  rx_buffer->page_offset;
@@ -887,21 +894,15 @@ static struct sk_buff *ixgbevf_fetch_rx_buffer(struct ixgbevf_ring *rx_ring,
 		prefetchw(skb->data);
 	}
 
-	/* we are reusing so sync this buffer for CPU use */
-	dma_sync_single_range_for_cpu(rx_ring->dev,
-				      rx_buffer->dma,
-				      rx_buffer->page_offset,
-				      size,
-				      DMA_FROM_DEVICE);
-
 	/* pull page into skb */
 	if (ixgbevf_add_rx_frag(rx_ring, rx_buffer, size, rx_desc, skb)) {
 		/* hand second half of page back to the ring */
 		ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
 	} else {
 		/* we are not reusing the buffer so unmap it */
-		dma_unmap_page(rx_ring->dev, rx_buffer->dma,
-			       PAGE_SIZE, DMA_FROM_DEVICE);
+		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
+				     PAGE_SIZE, DMA_FROM_DEVICE,
+				     IXGBEVF_RX_DMA_ATTR);
 	}
 
 	/* clear contents of buffer_info */
@@ -2116,7 +2117,6 @@ void ixgbevf_up(struct ixgbevf_adapter *adapter)
  **/
 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
 {
-	struct device *dev = rx_ring->dev;
 	unsigned long size;
 	unsigned int i;
 
@@ -2135,10 +2135,23 @@ static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
 		struct ixgbevf_rx_buffer *rx_buffer;
 
 		rx_buffer = &rx_ring->rx_buffer_info[i];
-		if (rx_buffer->dma)
-			dma_unmap_page(dev, rx_buffer->dma,
-				       PAGE_SIZE, DMA_FROM_DEVICE);
-		rx_buffer->dma = 0;
+
+		/* Invalidate cache lines that may have been written to by
+		 * device so that we avoid corrupting memory.
+		 */
+		dma_sync_single_range_for_cpu(rx_ring->dev,
+					      rx_buffer->dma,
+					      rx_buffer->page_offset,
+					      IXGBEVF_RX_BUFSZ,
+					      DMA_FROM_DEVICE);
+
+		/* free resources associated with mapping */
+		dma_unmap_page_attrs(rx_ring->dev,
+				     rx_buffer->dma,
+				     PAGE_SIZE,
+				     DMA_FROM_DEVICE,
+				     IXGBEVF_RX_DMA_ATTR);
+
 		if (rx_buffer->page)
 			__free_page(rx_buffer->page);
 		rx_buffer->page = NULL;

[Intel-wired-lan] [PATCH 5/9] ixgbevf: update code to better handle incrementing page count

[Emil Tantilov]

Based on commit bd4171a5d4c2
("igb: update code to better handle incrementing page count")

Update the driver code so that we do bulk updates of the page reference
count instead of just incrementing it by one reference at a time.  The
advantage to doing this is that we cut down on atomic operations and
this in turn should give us a slight improvement in cycles per packet.
In addition if we eventually move this over to using build_skb the gains
will be more noticeable.

Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
---
 drivers/net/ethernet/intel/ixgbevf/ixgbevf.h      |    7 ++++-
 drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |   30 +++++++++++++++------
 2 files changed, 28 insertions(+), 9 deletions(-)

diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h b/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
index b1da9f4..c70a789 100644
--- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
+++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
@@ -62,7 +62,12 @@ struct ixgbevf_tx_buffer {
 struct ixgbevf_rx_buffer {
 	dma_addr_t dma;
 	struct page *page;
-	unsigned int page_offset;
+#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
+	__u32 page_offset;
+#else
+	__u16 page_offset;
+#endif
+	__u16 pagecnt_bias;
 };
 
 struct ixgbevf_stats {
diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
index fbd493e..ae2402d 100644
--- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
+++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
@@ -611,6 +611,7 @@ static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
 	bi->dma = dma;
 	bi->page = page;
 	bi->page_offset = 0;
+	bi->pagecnt_bias = 1;
 
 	return true;
 }
@@ -747,6 +748,7 @@ static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
 	new_buff->page = old_buff->page;
 	new_buff->dma = old_buff->dma;
 	new_buff->page_offset = old_buff->page_offset;
+	new_buff->pagecnt_bias = old_buff->pagecnt_bias;
 }
 
 static inline bool ixgbevf_page_is_reserved(struct page *page)
@@ -758,13 +760,15 @@ static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer,
 				      struct page *page,
 				      const unsigned int truesize)
 {
+	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias--;
+
 	/* avoid re-using remote pages */
 	if (unlikely(ixgbevf_page_is_reserved(page)))
 		return false;
 
 #if (PAGE_SIZE < 8192)
 	/* if we are only owner of page we can reuse it */
-	if (unlikely(page_count(page) != 1))
+	if (unlikely(page_ref_count(page) != pagecnt_bias))
 		return false;
 
 	/* flip page offset to other buffer */
@@ -778,10 +782,15 @@ static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer,
 		return false;
 
 #endif
-	/* Even if we own the page, we are not allowed to use atomic_set()
-	 * This would break get_page_unless_zero() users.
+
+	/* If we have drained the page fragment pool we need to update
+	 * the pagecnt_bias and page count so that we fully restock the
+	 * number of references the driver holds.
 	 */
-	page_ref_inc(page);
+	if (unlikely(pagecnt_bias == 1)) {
+		page_ref_add(page, USHRT_MAX);
+		rx_buffer->pagecnt_bias = USHRT_MAX;
+	}
 
 	return true;
 }
@@ -827,7 +836,6 @@ static bool ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
 			return true;
 
 		/* this page cannot be reused so discard it */
-		put_page(page);
 		return false;
 	}
 
@@ -899,10 +907,13 @@ static struct sk_buff *ixgbevf_fetch_rx_buffer(struct ixgbevf_ring *rx_ring,
 		/* hand second half of page back to the ring */
 		ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
 	} else {
-		/* we are not reusing the buffer so unmap it */
+		/* We are not reusing the buffer so unmap it and free
+		 * any references we are holding to it
+		 */
 		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
 				     PAGE_SIZE, DMA_FROM_DEVICE,
 				     IXGBEVF_RX_DMA_ATTR);
+		__page_frag_cache_drain(page, rx_buffer->pagecnt_bias);
 	}
 
 	/* clear contents of buffer_info */
@@ -2135,6 +2146,8 @@ static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
 		struct ixgbevf_rx_buffer *rx_buffer;
 
 		rx_buffer = &rx_ring->rx_buffer_info[i];
+		if (!rx_buffer->page)
+			continue;
 
 		/* Invalidate cache lines that may have been written to by
 		 * device so that we avoid corrupting memory.
@@ -2152,8 +2165,9 @@ static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
 				     DMA_FROM_DEVICE,
 				     IXGBEVF_RX_DMA_ATTR);
 
-		if (rx_buffer->page)
-			__free_page(rx_buffer->page);
+		__page_frag_cache_drain(rx_buffer->page,
+					rx_buffer->pagecnt_bias);
+
 		rx_buffer->page = NULL;
 	}
 

[Intel-wired-lan] [PATCH 6/9] ixgbevf: add counters for Rx page allocations

[Emil Tantilov]

We already had placehloders for failed page and buffer allocations.
Added alloc_rx_page and made sure the stats are properly updated and
exposed in ethtool.

Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
---
 drivers/net/ethernet/intel/ixgbevf/ethtool.c      |    3 +++
 drivers/net/ethernet/intel/ixgbevf/ixgbevf.h      |    6 ++++-
 drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |   23 ++++++++++++++++-----
 3 files changed, 25 insertions(+), 7 deletions(-)

diff --git a/drivers/net/ethernet/intel/ixgbevf/ethtool.c b/drivers/net/ethernet/intel/ixgbevf/ethtool.c
index ff9d05f..4400e49 100644
--- a/drivers/net/ethernet/intel/ixgbevf/ethtool.c
+++ b/drivers/net/ethernet/intel/ixgbevf/ethtool.c
@@ -75,6 +75,9 @@ struct ixgbe_stats {
 	IXGBEVF_STAT("tx_timeout_count", tx_timeout_count),
 	IXGBEVF_NETDEV_STAT(multicast),
 	IXGBEVF_STAT("rx_csum_offload_errors", hw_csum_rx_error),
+	IXGBEVF_STAT("alloc_rx_page", alloc_rx_page),
+	IXGBEVF_STAT("alloc_rx_page_failed", alloc_rx_page_failed),
+	IXGBEVF_STAT("alloc_rx_buff_failed", alloc_rx_buff_failed),
 };
 
 #define IXGBEVF_QUEUE_STATS_LEN ( \
diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h b/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
index c70a789..f695242 100644
--- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
+++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
@@ -84,6 +84,7 @@ struct ixgbevf_tx_queue_stats {
 struct ixgbevf_rx_queue_stats {
 	u64 alloc_rx_page_failed;
 	u64 alloc_rx_buff_failed;
+	u64 alloc_rx_page;
 	u64 csum_err;
 };
 
@@ -295,8 +296,9 @@ struct ixgbevf_adapter {
 	u64 hw_csum_rx_error;
 	u64 hw_rx_no_dma_resources;
 	int num_msix_vectors;
-	u32 alloc_rx_page_failed;
-	u32 alloc_rx_buff_failed;
+	u64 alloc_rx_page_failed;
+	u64 alloc_rx_buff_failed;
+	u64 alloc_rx_page;
 
 	struct msix_entry *msix_entries;
 
diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
index ae2402d..350afec 100644
--- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
+++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
@@ -604,7 +604,7 @@ static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
 	if (dma_mapping_error(rx_ring->dev, dma)) {
 		__free_page(page);
 
-		rx_ring->rx_stats.alloc_rx_buff_failed++;
+		rx_ring->rx_stats.alloc_rx_page_failed++;
 		return false;
 	}
 
@@ -612,6 +612,7 @@ static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
 	bi->page = page;
 	bi->page_offset = 0;
 	bi->pagecnt_bias = 1;
+	rx_ring->rx_stats.alloc_rx_page++;
 
 	return true;
 }
@@ -963,8 +964,10 @@ static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
 		skb = ixgbevf_fetch_rx_buffer(rx_ring, rx_desc, skb);
 
 		/* exit if we failed to retrieve a buffer */
-		if (!skb)
+		if (!skb) {
+			rx_ring->rx_stats.alloc_rx_buff_failed++;
 			break;
+		}
 
 		cleaned_count++;
 
@@ -2749,6 +2752,8 @@ static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
 {
 	struct ixgbe_hw *hw = &adapter->hw;
+	u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
+	u64 alloc_rx_page = 0, hw_csum_rx_error = 0;
 	int i;
 
 	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
@@ -2769,10 +2774,18 @@ void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
 				adapter->stats.vfmprc);
 
 	for (i = 0;  i  < adapter->num_rx_queues;  i++) {
-		adapter->hw_csum_rx_error +=
-			adapter->rx_ring[i]->hw_csum_rx_error;
-		adapter->rx_ring[i]->hw_csum_rx_error = 0;
+		struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
+
+		hw_csum_rx_error += rx_ring->rx_stats.csum_err;
+		alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
+		alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
+		alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
 	}
+
+	adapter->hw_csum_rx_error = hw_csum_rx_error;
+	adapter->alloc_rx_page_failed = alloc_rx_page_failed;
+	adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
+	adapter->alloc_rx_page = alloc_rx_page;
 }
 
 /**

[Intel-wired-lan] [PATCH 7/9] ixgbevf: clear rx_buffer_info in configure instead of clean

[Emil Tantilov]

Based on commit d2bead576e67
("igb: Clear Rx buffer_info in configure instead of clean")

This change makes it so that instead of going through the entire ring on Rx
cleanup we only go through the region that was designated to be cleaned up
and stop when we reach the region where new allocations should start.

In addition we can avoid having to perform a memset on the Rx buffer_info
structures until we are about to start using the ring again.

Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
---
 drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |   26 +++++++++++----------
 1 file changed, 13 insertions(+), 13 deletions(-)

diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
index 350afec..a793f9e 100644
--- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
+++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
@@ -1773,6 +1773,10 @@ static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
 	IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
 	ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
 
+	/* initialize rx_buffer_info */
+	memset(ring->rx_buffer_info, 0,
+	       sizeof(struct ixgbevf_rx_buffer) * ring->count);
+
 	/* initialize Rx descriptor 0 */
 	rx_desc = IXGBEVF_RX_DESC(ring, 0);
 	rx_desc->wb.upper.length = 0;
@@ -2131,8 +2135,7 @@ void ixgbevf_up(struct ixgbevf_adapter *adapter)
  **/
 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
 {
-	unsigned long size;
-	unsigned int i;
+	u16 i = rx_ring->next_to_clean;
 
 	/* Free Rx ring sk_buff */
 	if (rx_ring->skb) {
@@ -2140,17 +2143,11 @@ static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
 		rx_ring->skb = NULL;
 	}
 
-	/* ring already cleared, nothing to do */
-	if (!rx_ring->rx_buffer_info)
-		return;
-
 	/* Free all the Rx ring pages */
-	for (i = 0; i < rx_ring->count; i++) {
+	while (i != rx_ring->next_to_alloc) {
 		struct ixgbevf_rx_buffer *rx_buffer;
 
 		rx_buffer = &rx_ring->rx_buffer_info[i];
-		if (!rx_buffer->page)
-			continue;
 
 		/* Invalidate cache lines that may have been written to by
 		 * device so that we avoid corrupting memory.
@@ -2171,11 +2168,14 @@ static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
 		__page_frag_cache_drain(rx_buffer->page,
 					rx_buffer->pagecnt_bias);
 
-		rx_buffer->page = NULL;
+		i++;
+		if (i == rx_ring->count)
+			i = 0;
 	}
 
-	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
-	memset(rx_ring->rx_buffer_info, 0, size);
+	rx_ring->next_to_alloc = 0;
+	rx_ring->next_to_clean = 0;
+	rx_ring->next_to_use = 0;
 }
 
 /**
@@ -3090,7 +3090,7 @@ int ixgbevf_setup_rx_resources(struct ixgbevf_ring *rx_ring)
 	int size;
 
 	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
-	rx_ring->rx_buffer_info = vzalloc(size);
+	rx_ring->rx_buffer_info = vmalloc(size);
 	if (!rx_ring->rx_buffer_info)
 		goto err;
 

[Intel-wired-lan] [PATCH 8/9] ixgbevf: improve performance and reduce size of ixgbevf_tx_map()

[Emil Tantilov]

Based on commit ec718254cbfe
("ixgbe: Improve performance and reduce size of ixgbe_tx_map")

This change is meant to both improve the performance and reduce the size of
ixgbevf_tx_map().

Expand the work done in the main loop by pushing first into tx_buffer.
This allows us to pull in the dma_mapping_error check, the tx_buffer value
assignment, and the initial DMA value assignment to the Tx descriptor.

Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
---
 drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |   45 +++++++++------------
 1 file changed, 20 insertions(+), 25 deletions(-)

diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
index a793f9e..d3415ee 100644
--- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
+++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
@@ -3532,34 +3532,37 @@ static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
 			   struct ixgbevf_tx_buffer *first,
 			   const u8 hdr_len)
 {
-	dma_addr_t dma;
 	struct sk_buff *skb = first->skb;
 	struct ixgbevf_tx_buffer *tx_buffer;
 	union ixgbe_adv_tx_desc *tx_desc;
-	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
-	unsigned int data_len = skb->data_len;
-	unsigned int size = skb_headlen(skb);
-	unsigned int paylen = skb->len - hdr_len;
+	struct skb_frag_struct *frag;
+	dma_addr_t dma;
+	unsigned int data_len, size;
 	u32 tx_flags = first->tx_flags;
-	__le32 cmd_type;
+	__le32 cmd_type = ixgbevf_tx_cmd_type(tx_flags);
 	u16 i = tx_ring->next_to_use;
 
 	tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
 
-	ixgbevf_tx_olinfo_status(tx_desc, tx_flags, paylen);
-	cmd_type = ixgbevf_tx_cmd_type(tx_flags);
+	ixgbevf_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
+
+	size = skb_headlen(skb);
+	data_len = skb->data_len;
 
 	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
-	if (dma_mapping_error(tx_ring->dev, dma))
-		goto dma_error;
 
-	/* record length, and DMA address */
-	dma_unmap_len_set(first, len, size);
-	dma_unmap_addr_set(first, dma, dma);
+	tx_buffer = first;
 
-	tx_desc->read.buffer_addr = cpu_to_le64(dma);
+	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
+		if (dma_mapping_error(tx_ring->dev, dma))
+			goto dma_error;
+
+		/* record length, and DMA address */
+		dma_unmap_len_set(tx_buffer, len, size);
+		dma_unmap_addr_set(tx_buffer, dma, dma);
+
+		tx_desc->read.buffer_addr = cpu_to_le64(dma);
 
-	for (;;) {
 		while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
 			tx_desc->read.cmd_type_len =
 				cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
@@ -3570,12 +3573,12 @@ static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
 				tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 				i = 0;
 			}
+			tx_desc->read.olinfo_status = 0;
 
 			dma += IXGBE_MAX_DATA_PER_TXD;
 			size -= IXGBE_MAX_DATA_PER_TXD;
 
 			tx_desc->read.buffer_addr = cpu_to_le64(dma);
-			tx_desc->read.olinfo_status = 0;
 		}
 
 		if (likely(!data_len))
@@ -3589,23 +3592,15 @@ static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
 			tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 			i = 0;
 		}
+		tx_desc->read.olinfo_status = 0;
 
 		size = skb_frag_size(frag);
 		data_len -= size;
 
 		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
 				       DMA_TO_DEVICE);
-		if (dma_mapping_error(tx_ring->dev, dma))
-			goto dma_error;
 
 		tx_buffer = &tx_ring->tx_buffer_info[i];
-		dma_unmap_len_set(tx_buffer, len, size);
-		dma_unmap_addr_set(tx_buffer, dma, dma);
-
-		tx_desc->read.buffer_addr = cpu_to_le64(dma);
-		tx_desc->read.olinfo_status = 0;
-
-		frag++;
 	}
 
 	/* write last descriptor with RS and EOP bits */

[Intel-wired-lan] [PATCH 9/9] ixgbevf: don't bother clearing tx_buffer_info in ixgbevf_clean_tx_ring()

[Emil Tantilov]

Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
---
 drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |  115 +++++++++++++--------
 1 file changed, 72 insertions(+), 43 deletions(-)

diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
index d3415ee..9b3d43d 100644
--- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
+++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
@@ -206,28 +206,6 @@ static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
 	}
 }
 
-static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
-					struct ixgbevf_tx_buffer *tx_buffer)
-{
-	if (tx_buffer->skb) {
-		dev_kfree_skb_any(tx_buffer->skb);
-		if (dma_unmap_len(tx_buffer, len))
-			dma_unmap_single(tx_ring->dev,
-					 dma_unmap_addr(tx_buffer, dma),
-					 dma_unmap_len(tx_buffer, len),
-					 DMA_TO_DEVICE);
-	} else if (dma_unmap_len(tx_buffer, len)) {
-		dma_unmap_page(tx_ring->dev,
-			       dma_unmap_addr(tx_buffer, dma),
-			       dma_unmap_len(tx_buffer, len),
-			       DMA_TO_DEVICE);
-	}
-	tx_buffer->next_to_watch = NULL;
-	tx_buffer->skb = NULL;
-	dma_unmap_len_set(tx_buffer, len, 0);
-	/* tx_buffer must be completely set up in the transmit path */
-}
-
 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
 {
 	return ring->stats.packets;
@@ -349,7 +327,6 @@ static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
 				 DMA_TO_DEVICE);
 
 		/* clear tx_buffer data */
-		tx_buffer->skb = NULL;
 		dma_unmap_len_set(tx_buffer, len, 0);
 
 		/* unmap remaining buffers */
@@ -1576,6 +1553,10 @@ static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
 	txdctl |= (1u << 8) |    /* HTHRESH = 1 */
 		   32;           /* PTHRESH = 32 */
 
+	/* reinitialize tx_buffer_info */
+	memset(ring->tx_buffer_info, 0,
+	       sizeof(struct ixgbevf_tx_buffer) * ring->count);
+
 	clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
 
 	IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
@@ -2184,23 +2165,57 @@ static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
  **/
 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
 {
-	struct ixgbevf_tx_buffer *tx_buffer_info;
-	unsigned long size;
-	unsigned int i;
+	u16 i = tx_ring->next_to_clean;
+	struct ixgbevf_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
 
-	if (!tx_ring->tx_buffer_info)
-		return;
+	while (i != tx_ring->next_to_use) {
+		union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
+
+		/* Free all the Tx ring sk_buffs */
+		dev_kfree_skb_any(tx_buffer->skb);
+
+		/* unmap skb header data */
+		dma_unmap_single(tx_ring->dev,
+				 dma_unmap_addr(tx_buffer, dma),
+				 dma_unmap_len(tx_buffer, len),
+				 DMA_TO_DEVICE);
+
+		/* check for eop_desc to determine the end of the packet */
+		eop_desc = tx_buffer->next_to_watch;
+		tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
+
+		/* unmap remaining buffers */
+		while (tx_desc != eop_desc) {
+			tx_buffer++;
+			tx_desc++;
+			i++;
+			if (unlikely(i == tx_ring->count)) {
+				i = 0;
+				tx_buffer = tx_ring->tx_buffer_info;
+				tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
+			}
+
+			/* unmap any remaining paged data */
+			if (dma_unmap_len(tx_buffer, len))
+				dma_unmap_page(tx_ring->dev,
+					       dma_unmap_addr(tx_buffer, dma),
+					       dma_unmap_len(tx_buffer, len),
+					       DMA_TO_DEVICE);
+		}
 
-	/* Free all the Tx ring sk_buffs */
-	for (i = 0; i < tx_ring->count; i++) {
-		tx_buffer_info = &tx_ring->tx_buffer_info[i];
-		ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
+		/* move us one more past the eop_desc for start of next pkt */
+		tx_buffer++;
+		i++;
+		if (unlikely(i == tx_ring->count)) {
+			i = 0;
+			tx_buffer = tx_ring->tx_buffer_info;
+		}
 	}
 
-	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
-	memset(tx_ring->tx_buffer_info, 0, size);
+	/* reset next_to_use and next_to_clean */
+	tx_ring->next_to_use = 0;
+	tx_ring->next_to_clean = 0;
 
-	memset(tx_ring->desc, 0, tx_ring->size);
 }
 
 /**
@@ -3030,7 +3045,7 @@ int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
 	int size;
 
 	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
-	tx_ring->tx_buffer_info = vzalloc(size);
+	tx_ring->tx_buffer_info = vmalloc(size);
 	if (!tx_ring->tx_buffer_info)
 		goto err;
 
@@ -3634,18 +3649,32 @@ static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
 	return;
 dma_error:
 	dev_err(tx_ring->dev, "TX DMA map failed\n");
+	tx_buffer = &tx_ring->tx_buffer_info[i];
 
 	/* clear dma mappings for failed tx_buffer_info map */
-	for (;;) {
+	while (tx_buffer != first) {
+		if (dma_unmap_len(tx_buffer, len))
+			dma_unmap_page(tx_ring->dev,
+				       dma_unmap_addr(tx_buffer, dma),
+				       dma_unmap_len(tx_buffer, len),
+				       DMA_TO_DEVICE);
+		dma_unmap_len_set(tx_buffer, len, 0);
+
+		if (i-- == 0)
+			i += tx_ring->count;
 		tx_buffer = &tx_ring->tx_buffer_info[i];
-		ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer);
-		if (tx_buffer == first)
-			break;
-		if (i == 0)
-			i = tx_ring->count;
-		i--;
 	}
 
+	if (dma_unmap_len(tx_buffer, len))
+		dma_unmap_single(tx_ring->dev,
+				 dma_unmap_addr(tx_buffer, dma),
+				 dma_unmap_len(tx_buffer, len),
+				 DMA_TO_DEVICE);
+	dma_unmap_len_set(tx_buffer, len, 0);
+
+	dev_kfree_skb_any(tx_buffer->skb);
+	tx_buffer->skb = NULL;
+
 	tx_ring->next_to_use = i;
 }
 

[Intel-wired-lan] [PATCH 1/9] ixgbevf: add function for checking if we can reuse page

[Singh, Krishneil K]

> -----Original Message-----
> From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf
> Of Emil Tantilov
> Sent: Monday, December 11, 2017 10:37 AM
> To: intel-wired-lan at lists.osuosl.org
> Subject: [Intel-wired-lan] [PATCH 1/9] ixgbevf: add function for checking if we
> can reuse page
> 
> Introduce ixgbevf_can_reuse_page() similar to the change in ixgbe from
> commit af43da0dba0b
> ("ixgbe: Add function for checking to see if we can reuse page")
> 
> Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
> ---
>  drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |   59 ++++++++++++---------
>  1 file changed, 33 insertions(+), 26 deletions(-)

Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>

[Intel-wired-lan] [PATCH 2/9] ixgbevf: only DMA sync frame length

[Singh, Krishneil K]

> -----Original Message-----
> From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf
> Of Emil Tantilov
> Sent: Monday, December 11, 2017 10:37 AM
> To: intel-wired-lan at lists.osuosl.org
> Subject: [Intel-wired-lan] [PATCH 2/9] ixgbevf: only DMA sync frame length
> 
> Based on commit 64f2525ca4e7 ("igb: Only DMA sync frame length")
> 
> On some architectures synching a buffer for DMA may be expensive.
> Instead of the entire 2K receive buffer only synchronize the length of
> the frame, which will typically be the MTU or smaller.
> 
> Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
> ---

Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>

[Intel-wired-lan] [PATCH 3/9] ixgbevf: use length to determine if descriptor is done

[Singh, Krishneil K]

> -----Original Message-----
> From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf
> Of Emil Tantilov
> Sent: Monday, December 11, 2017 10:37 AM
> To: intel-wired-lan at lists.osuosl.org
> Subject: [Intel-wired-lan] [PATCH 3/9] ixgbevf: use length to determine if
> descriptor is done
> 
> Based on:
> commit 7ec0116c9131 ("igb: Use length to determine if descriptor is done")
> 
> This change makes it so that we use the length of the packet instead of the
> DD status bit to determine if a new descriptor is ready to be processed.
> The obvious advantage is that it cuts down on reads as we don't really even
> need the DD bit if going from a 0 to a non-zero value on size is enough to
> inform us that the packet has been completed.
> 
> In addition we only reset the Rx descriptor length for descriptor zero when
> resetting a ring instead of having to do a memset with 0 over the entire
> ring. By doing this we can save some time on initialization.
> 
> Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
> ---

Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>

[Intel-wired-lan] [PATCH 4/9] ixgbevf: add support for DMA_ATTR_SKIP_CPU_SYNC/WEAK_ORDERING

[Singh, Krishneil K]

> -----Original Message-----
> From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf
> Of Emil Tantilov
> Sent: Monday, December 11, 2017 10:37 AM
> To: intel-wired-lan at lists.osuosl.org
> Subject: [Intel-wired-lan] [PATCH 4/9] ixgbevf: add support for
> DMA_ATTR_SKIP_CPU_SYNC/WEAK_ORDERING
> 
> Based on commit 5be5955425c2
> ("igb: update driver to make use of DMA_ATTR_SKIP_CPU_SYNC")
> 
> and
> 
> commit 7bd175928280 ("igb: Add support for DMA_ATTR_WEAK_ORDERING")
> 
> Convert the calls to dma_map/unmap_page() to the attributes version
> and add DMA_ATTR_SKIP_CPU_SYNC/WEAK_ORDERING which should help
> improve performance on some platforms.
> 
> Move sync_for_cpu call before we perform a prefetch to avoid
> invalidating the first 128 bytes of the packet on architectures where
> that call may invalidate the cache.
> 
> Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
> ---

Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>








	

[Intel-wired-lan] [PATCH 5/9] ixgbevf: update code to better handle incrementing page count

[Singh, Krishneil K]

> -----Original Message-----
> From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf
> Of Emil Tantilov
> Sent: Monday, December 11, 2017 10:37 AM
> To: intel-wired-lan at lists.osuosl.org
> Subject: [Intel-wired-lan] [PATCH 5/9] ixgbevf: update code to better handle
> incrementing page count
> 
> Based on commit bd4171a5d4c2
> ("igb: update code to better handle incrementing page count")
> 
> Update the driver code so that we do bulk updates of the page reference
> count instead of just incrementing it by one reference at a time.  The
> advantage to doing this is that we cut down on atomic operations and
> this in turn should give us a slight improvement in cycles per packet.
> In addition if we eventually move this over to using build_skb the gains
> will be more noticeable.
> 
> Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
> ---

Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>

[Intel-wired-lan] [PATCH 6/9] ixgbevf: add counters for Rx page allocations

[Singh, Krishneil K]

> -----Original Message-----
> From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf
> Of Emil Tantilov
> Sent: Monday, December 11, 2017 10:37 AM
> To: intel-wired-lan at lists.osuosl.org
> Subject: [Intel-wired-lan] [PATCH 6/9] ixgbevf: add counters for Rx page
> allocations
> 
> We already had placehloders for failed page and buffer allocations.
> Added alloc_rx_page and made sure the stats are properly updated and
> exposed in ethtool.
> 
> Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
> ---

Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>

[Intel-wired-lan] [PATCH 7/9] ixgbevf: clear rx_buffer_info in configure instead of clean

[Singh, Krishneil K]

> -----Original Message-----
> From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf
> Of Emil Tantilov
> Sent: Monday, December 11, 2017 10:37 AM
> To: intel-wired-lan at lists.osuosl.org
> Subject: [Intel-wired-lan] [PATCH 7/9] ixgbevf: clear rx_buffer_info in configure
> instead of clean
> 
> Based on commit d2bead576e67
> ("igb: Clear Rx buffer_info in configure instead of clean")
> 
> This change makes it so that instead of going through the entire ring on Rx
> cleanup we only go through the region that was designated to be cleaned up
> and stop when we reach the region where new allocations should start.
> 
> In addition we can avoid having to perform a memset on the Rx buffer_info
> structures until we are about to start using the ring again.
> 
> Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
> ---

Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>

[Intel-wired-lan] [PATCH 8/9] ixgbevf: improve performance and reduce size of ixgbevf_tx_map()

[Singh, Krishneil K]

> -----Original Message-----
> From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf
> Of Emil Tantilov
> Sent: Monday, December 11, 2017 10:37 AM
> To: intel-wired-lan at lists.osuosl.org
> Subject: [Intel-wired-lan] [PATCH 8/9] ixgbevf: improve performance and
> reduce size of ixgbevf_tx_map()
> 
> Based on commit ec718254cbfe
> ("ixgbe: Improve performance and reduce size of ixgbe_tx_map")
> 
> This change is meant to both improve the performance and reduce the size of
> ixgbevf_tx_map().
> 
> Expand the work done in the main loop by pushing first into tx_buffer.
> This allows us to pull in the dma_mapping_error check, the tx_buffer value
> assignment, and the initial DMA value assignment to the Tx descriptor.
> 
> Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
> ---
>  drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |   45 +++++++++------------
>  1 file changed, 20 insertions(+), 25 deletions(-)
> 
> diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> index a793f9e..d3415ee 100644
> --- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> +++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> @@ -3532,34 +3532,37 @@ static void ixgbevf_tx_map(struct ixgbevf_ring
> *tx_ring,
>  			   struct ixgbevf_tx_buffer *first,
>  			   const u8 hdr_len)
>  {
> -	dma_addr_t dma;
>  	struct sk_buff *skb = first->skb;
>  	struct ixgbevf_tx_buffer *tx_buffer;
>  	union ixgbe_adv_tx_desc *tx_desc;
> -	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
> -	unsigned int data_len = skb->data_len;
> -	unsigned int size = skb_headlen(skb);
> -	unsigned int paylen = skb->len - hdr_len;
> +	struct skb_frag_struct *frag;
> +	dma_addr_t dma;
> +	unsigned int data_len, size;
>  	u32 tx_flags = first->tx_flags;
> -	__le32 cmd_type;
> +	__le32 cmd_type = ixgbevf_tx_cmd_type(tx_flags);
>  	u16 i = tx_ring->next_to_use;
> 
>  	tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
> 
> -	ixgbevf_tx_olinfo_status(tx_desc, tx_flags, paylen);
> -	cmd_type = ixgbevf_tx_cmd_type(tx_flags);
> +	ixgbevf_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
> +
> +	size = skb_headlen(skb);
> +	data_len = skb->data_len;
> 
>  	dma = dma_map_single(tx_ring->dev, skb->data, size,
> DMA_TO_DEVICE);
> -	if (dma_mapping_error(tx_ring->dev, dma))
> -		goto dma_error;
> 
> -	/* record length, and DMA address */
> -	dma_unmap_len_set(first, len, size);
> -	dma_unmap_addr_set(first, dma, dma);
> +	tx_buffer = first;
> 
> -	tx_desc->read.buffer_addr = cpu_to_le64(dma);
> +	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
> +		if (dma_mapping_error(tx_ring->dev, dma))
> +			goto dma_error;
> +
> +		/* record length, and DMA address */
> +		dma_unmap_len_set(tx_buffer, len, size);
> +		dma_unmap_addr_set(tx_buffer, dma, dma);
> +
> +		tx_desc->read.buffer_addr = cpu_to_le64(dma);
> 
> -	for (;;) {
>  		while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
>  			tx_desc->read.cmd_type_len =
>  				cmd_type |
> cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
> @@ -3570,12 +3573,12 @@ static void ixgbevf_tx_map(struct ixgbevf_ring
> *tx_ring,
>  				tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
>  				i = 0;
>  			}
> +			tx_desc->read.olinfo_status = 0;
> 
>  			dma += IXGBE_MAX_DATA_PER_TXD;
>  			size -= IXGBE_MAX_DATA_PER_TXD;
> 
>  			tx_desc->read.buffer_addr = cpu_to_le64(dma);
> -			tx_desc->read.olinfo_status = 0;
>  		}
> 
>  		if (likely(!data_len))
> @@ -3589,23 +3592,15 @@ static void ixgbevf_tx_map(struct ixgbevf_ring
> *tx_ring,
>  			tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
>  			i = 0;
>  		}
> +		tx_desc->read.olinfo_status = 0;
> 
>  		size = skb_frag_size(frag);
>  		data_len -= size;
> 
>  		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
>  				       DMA_TO_DEVICE);
> -		if (dma_mapping_error(tx_ring->dev, dma))
> -			goto dma_error;
> 
>  		tx_buffer = &tx_ring->tx_buffer_info[i];
> -		dma_unmap_len_set(tx_buffer, len, size);
> -		dma_unmap_addr_set(tx_buffer, dma, dma);
> -
> -		tx_desc->read.buffer_addr = cpu_to_le64(dma);
> -		tx_desc->read.olinfo_status = 0;
> -
> -		frag++;
>  	}
> 
>  	/* write last descriptor with RS and EOP bits */
> 
> _______________________________________________
> Intel-wired-lan mailing list
> Intel-wired-lan at osuosl.org
> https://lists.osuosl.org/mailman/listinfo/intel-wired-lan

Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>

[Intel-wired-lan] [PATCH 9/9] ixgbevf: don't bother clearing tx_buffer_info in ixgbevf_clean_tx_ring()

[Singh, Krishneil K]

> -----Original Message-----
> From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf
> Of Emil Tantilov
> Sent: Monday, December 11, 2017 10:38 AM
> To: intel-wired-lan at lists.osuosl.org
> Subject: [Intel-wired-lan] [PATCH 9/9] ixgbevf: don't bother clearing
> tx_buffer_info in ixgbevf_clean_tx_ring()
> 
> Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
> ---
>  drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |  115 +++++++++++++-------
> -
>  1 file changed, 72 insertions(+), 43 deletions(-)
> 
> diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> index d3415ee..9b3d43d 100644
> --- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> +++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> @@ -206,28 +206,6 @@ static void ixgbevf_set_ivar(struct ixgbevf_adapter
> *adapter, s8 direction,
>  	}
>  }
> 
> -static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
> -					struct ixgbevf_tx_buffer *tx_buffer)
> -{
> -	if (tx_buffer->skb) {
> -		dev_kfree_skb_any(tx_buffer->skb);
> -		if (dma_unmap_len(tx_buffer, len))
> -			dma_unmap_single(tx_ring->dev,
> -					 dma_unmap_addr(tx_buffer, dma),
> -					 dma_unmap_len(tx_buffer, len),
> -					 DMA_TO_DEVICE);
> -	} else if (dma_unmap_len(tx_buffer, len)) {
> -		dma_unmap_page(tx_ring->dev,
> -			       dma_unmap_addr(tx_buffer, dma),
> -			       dma_unmap_len(tx_buffer, len),
> -			       DMA_TO_DEVICE);
> -	}
> -	tx_buffer->next_to_watch = NULL;
> -	tx_buffer->skb = NULL;
> -	dma_unmap_len_set(tx_buffer, len, 0);
> -	/* tx_buffer must be completely set up in the transmit path */
> -}
> -
>  static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
>  {
>  	return ring->stats.packets;
> @@ -349,7 +327,6 @@ static bool ixgbevf_clean_tx_irq(struct
> ixgbevf_q_vector *q_vector,
>  				 DMA_TO_DEVICE);
> 
>  		/* clear tx_buffer data */
> -		tx_buffer->skb = NULL;
>  		dma_unmap_len_set(tx_buffer, len, 0);
> 
>  		/* unmap remaining buffers */
> @@ -1576,6 +1553,10 @@ static void ixgbevf_configure_tx_ring(struct
> ixgbevf_adapter *adapter,
>  	txdctl |= (1u << 8) |    /* HTHRESH = 1 */
>  		   32;           /* PTHRESH = 32 */
> 
> +	/* reinitialize tx_buffer_info */
> +	memset(ring->tx_buffer_info, 0,
> +	       sizeof(struct ixgbevf_tx_buffer) * ring->count);
> +
>  	clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
> 
>  	IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
> @@ -2184,23 +2165,57 @@ static void ixgbevf_clean_rx_ring(struct
> ixgbevf_ring *rx_ring)
>   **/
>  static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
>  {
> -	struct ixgbevf_tx_buffer *tx_buffer_info;
> -	unsigned long size;
> -	unsigned int i;
> +	u16 i = tx_ring->next_to_clean;
> +	struct ixgbevf_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
> 
> -	if (!tx_ring->tx_buffer_info)
> -		return;
> +	while (i != tx_ring->next_to_use) {
> +		union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
> +
> +		/* Free all the Tx ring sk_buffs */
> +		dev_kfree_skb_any(tx_buffer->skb);
> +
> +		/* unmap skb header data */
> +		dma_unmap_single(tx_ring->dev,
> +				 dma_unmap_addr(tx_buffer, dma),
> +				 dma_unmap_len(tx_buffer, len),
> +				 DMA_TO_DEVICE);
> +
> +		/* check for eop_desc to determine the end of the packet */
> +		eop_desc = tx_buffer->next_to_watch;
> +		tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
> +
> +		/* unmap remaining buffers */
> +		while (tx_desc != eop_desc) {
> +			tx_buffer++;
> +			tx_desc++;
> +			i++;
> +			if (unlikely(i == tx_ring->count)) {
> +				i = 0;
> +				tx_buffer = tx_ring->tx_buffer_info;
> +				tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
> +			}
> +
> +			/* unmap any remaining paged data */
> +			if (dma_unmap_len(tx_buffer, len))
> +				dma_unmap_page(tx_ring->dev,
> +					       dma_unmap_addr(tx_buffer,
> dma),
> +					       dma_unmap_len(tx_buffer, len),
> +					       DMA_TO_DEVICE);
> +		}
> 
> -	/* Free all the Tx ring sk_buffs */
> -	for (i = 0; i < tx_ring->count; i++) {
> -		tx_buffer_info = &tx_ring->tx_buffer_info[i];
> -		ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
> +		/* move us one more past the eop_desc for start of next pkt */
> +		tx_buffer++;
> +		i++;
> +		if (unlikely(i == tx_ring->count)) {
> +			i = 0;
> +			tx_buffer = tx_ring->tx_buffer_info;
> +		}
>  	}
> 
> -	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
> -	memset(tx_ring->tx_buffer_info, 0, size);
> +	/* reset next_to_use and next_to_clean */
> +	tx_ring->next_to_use = 0;
> +	tx_ring->next_to_clean = 0;
> 
> -	memset(tx_ring->desc, 0, tx_ring->size);
>  }
> 
>  /**
> @@ -3030,7 +3045,7 @@ int ixgbevf_setup_tx_resources(struct ixgbevf_ring
> *tx_ring)
>  	int size;
> 
>  	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
> -	tx_ring->tx_buffer_info = vzalloc(size);
> +	tx_ring->tx_buffer_info = vmalloc(size);
>  	if (!tx_ring->tx_buffer_info)
>  		goto err;
> 
> @@ -3634,18 +3649,32 @@ static void ixgbevf_tx_map(struct ixgbevf_ring
> *tx_ring,
>  	return;
>  dma_error:
>  	dev_err(tx_ring->dev, "TX DMA map failed\n");
> +	tx_buffer = &tx_ring->tx_buffer_info[i];
> 
>  	/* clear dma mappings for failed tx_buffer_info map */
> -	for (;;) {
> +	while (tx_buffer != first) {
> +		if (dma_unmap_len(tx_buffer, len))
> +			dma_unmap_page(tx_ring->dev,
> +				       dma_unmap_addr(tx_buffer, dma),
> +				       dma_unmap_len(tx_buffer, len),
> +				       DMA_TO_DEVICE);
> +		dma_unmap_len_set(tx_buffer, len, 0);
> +
> +		if (i-- == 0)
> +			i += tx_ring->count;
>  		tx_buffer = &tx_ring->tx_buffer_info[i];
> -		ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer);
> -		if (tx_buffer == first)
> -			break;
> -		if (i == 0)
> -			i = tx_ring->count;
> -		i--;
>  	}
> 
> +	if (dma_unmap_len(tx_buffer, len))
> +		dma_unmap_single(tx_ring->dev,
> +				 dma_unmap_addr(tx_buffer, dma),
> +				 dma_unmap_len(tx_buffer, len),
> +				 DMA_TO_DEVICE);
> +	dma_unmap_len_set(tx_buffer, len, 0);
> +
> +	dev_kfree_skb_any(tx_buffer->skb);
> +	tx_buffer->skb = NULL;
> +
>  	tx_ring->next_to_use = i;
>  }
> 
> 
> _______________________________________________
> Intel-wired-lan mailing list
> Intel-wired-lan at osuosl.org
> https://lists.osuosl.org/mailman/listinfo/intel-wired-lan

Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>

[Intel-wired-lan] [PATCH 4/9] ixgbevf: add support for DMA_ATTR_SKIP_CPU_SYNC/WEAK_ORDERING

[Singh, Krishneil K]

> -----Original Message-----
> From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf Of
> Emil Tantilov
> Sent: Monday, December 11, 2017 10:37 AM
> To: intel-wired-lan at lists.osuosl.org
> Subject: [Intel-wired-lan] [PATCH 4/9] ixgbevf: add support for
> DMA_ATTR_SKIP_CPU_SYNC/WEAK_ORDERING
> 
> Based on commit 5be5955425c2
> ("igb: update driver to make use of DMA_ATTR_SKIP_CPU_SYNC")
> 
> and
> 
> commit 7bd175928280 ("igb: Add support for DMA_ATTR_WEAK_ORDERING")
> 
> Convert the calls to dma_map/unmap_page() to the attributes version
> and add DMA_ATTR_SKIP_CPU_SYNC/WEAK_ORDERING which should help
> improve performance on some platforms.
> 
> Move sync_for_cpu call before we perform a prefetch to avoid
> invalidating the first 128 bytes of the packet on architectures where
> that call may invalidate the cache.
> 
> Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
> ---
>  drivers/net/ethernet/intel/ixgbevf/ixgbevf.h      |    3 +
>  drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |   57 +++++++++++++-------
> -
>  2 files changed, 38 insertions(+), 22 deletions(-)
> 
> diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
> b/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
> index 581f44b..b1da9f4 100644
> --- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
> +++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
> @@ -260,6 +260,9 @@ static inline void ixgbevf_write_tail(struct ixgbevf_ring
> *ring, u32 value)
>  #define MIN_MSIX_Q_VECTORS	1
>  #define MIN_MSIX_COUNT		(MIN_MSIX_Q_VECTORS +
> NON_Q_VECTORS)
> 
> +#define IXGBEVF_RX_DMA_ATTR \
> +	(DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING)
> +
>  /* board specific private data structure */
>  struct ixgbevf_adapter {
>  	/* this field must be first, see ixgbevf_process_skb_fields */
> diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> index 725fe2d..fbd493e 100644
> --- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> +++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> @@ -595,8 +595,8 @@ static bool ixgbevf_alloc_mapped_page(struct
> ixgbevf_ring *rx_ring,
>  	}
> 
>  	/* map page for use */
> -	dma = dma_map_page(rx_ring->dev, page, 0,
> -			   PAGE_SIZE, DMA_FROM_DEVICE);
> +	dma = dma_map_page_attrs(rx_ring->dev, page, 0, PAGE_SIZE,
> +				 DMA_FROM_DEVICE,
> IXGBEVF_RX_DMA_ATTR);
> 
>  	/* if mapping failed free memory back to system since
>  	 * there isn't much point in holding memory we can't use
> @@ -639,6 +639,12 @@ static void ixgbevf_alloc_rx_buffers(struct
> ixgbevf_ring *rx_ring,
>  		if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
>  			break;
> 
> +		/* sync the buffer for use by the device */
> +		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
> +						 bi->page_offset,
> +						 IXGBEVF_RX_BUFSZ,
> +						 DMA_FROM_DEVICE);
> +
>  		/* Refresh the desc even if pkt_addr didn't change
>  		 * because each write-back erases this info.
>  		 */
> @@ -741,12 +747,6 @@ static void ixgbevf_reuse_rx_page(struct ixgbevf_ring
> *rx_ring,
>  	new_buff->page = old_buff->page;
>  	new_buff->dma = old_buff->dma;
>  	new_buff->page_offset = old_buff->page_offset;
> -
> -	/* sync the buffer for use by the device */
> -	dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
> -					 new_buff->page_offset,
> -					 IXGBEVF_RX_BUFSZ,
> -					 DMA_FROM_DEVICE);
>  }
> 
>  static inline bool ixgbevf_page_is_reserved(struct page *page)
> @@ -862,6 +862,13 @@ static struct sk_buff *ixgbevf_fetch_rx_buffer(struct
> ixgbevf_ring *rx_ring,
>  	page = rx_buffer->page;
>  	prefetchw(page);
> 
> +	/* we are reusing so sync this buffer for CPU use */
> +	dma_sync_single_range_for_cpu(rx_ring->dev,
> +				      rx_buffer->dma,
> +				      rx_buffer->page_offset,
> +				      size,
> +				      DMA_FROM_DEVICE);
> +
>  	if (likely(!skb)) {
>  		void *page_addr = page_address(page) +
>  				  rx_buffer->page_offset;
> @@ -887,21 +894,15 @@ static struct sk_buff *ixgbevf_fetch_rx_buffer(struct
> ixgbevf_ring *rx_ring,
>  		prefetchw(skb->data);
>  	}
> 
> -	/* we are reusing so sync this buffer for CPU use */
> -	dma_sync_single_range_for_cpu(rx_ring->dev,
> -				      rx_buffer->dma,
> -				      rx_buffer->page_offset,
> -				      size,
> -				      DMA_FROM_DEVICE);
> -
>  	/* pull page into skb */
>  	if (ixgbevf_add_rx_frag(rx_ring, rx_buffer, size, rx_desc, skb)) {
>  		/* hand second half of page back to the ring */
>  		ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
>  	} else {
>  		/* we are not reusing the buffer so unmap it */
> -		dma_unmap_page(rx_ring->dev, rx_buffer->dma,
> -			       PAGE_SIZE, DMA_FROM_DEVICE);
> +		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
> +				     PAGE_SIZE, DMA_FROM_DEVICE,
> +				     IXGBEVF_RX_DMA_ATTR);
>  	}
> 
>  	/* clear contents of buffer_info */
> @@ -2116,7 +2117,6 @@ void ixgbevf_up(struct ixgbevf_adapter *adapter)
>   **/
>  static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
>  {
> -	struct device *dev = rx_ring->dev;
>  	unsigned long size;
>  	unsigned int i;
> 
> @@ -2135,10 +2135,23 @@ static void ixgbevf_clean_rx_ring(struct
> ixgbevf_ring *rx_ring)
>  		struct ixgbevf_rx_buffer *rx_buffer;
> 
>  		rx_buffer = &rx_ring->rx_buffer_info[i];
> -		if (rx_buffer->dma)
> -			dma_unmap_page(dev, rx_buffer->dma,
> -				       PAGE_SIZE, DMA_FROM_DEVICE);
> -		rx_buffer->dma = 0;
> +
> +		/* Invalidate cache lines that may have been written to by
> +		 * device so that we avoid corrupting memory.
> +		 */
> +		dma_sync_single_range_for_cpu(rx_ring->dev,
> +					      rx_buffer->dma,
> +					      rx_buffer->page_offset,
> +					      IXGBEVF_RX_BUFSZ,
> +					      DMA_FROM_DEVICE);
> +
> +		/* free resources associated with mapping */
> +		dma_unmap_page_attrs(rx_ring->dev,
> +				     rx_buffer->dma,
> +				     PAGE_SIZE,
> +				     DMA_FROM_DEVICE,
> +				     IXGBEVF_RX_DMA_ATTR);
> +
>  		if (rx_buffer->page)
>  			__free_page(rx_buffer->page);
>  		rx_buffer->page = NULL;
> 
> _______________________________________________
> Intel-wired-lan mailing list
> Intel-wired-lan at osuosl.org
> https://lists.osuosl.org/mailman/listinfo/intel-wired-lan

Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>

[Intel-wired-lan] [PATCH 8/9] ixgbevf: improve performance and reduce size of ixgbevf_tx_map()

[Singh, Krishneil K]

> -----Original Message-----
> From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf Of
> Emil Tantilov
> Sent: Monday, December 11, 2017 10:37 AM
> To: intel-wired-lan at lists.osuosl.org
> Subject: [Intel-wired-lan] [PATCH 8/9] ixgbevf: improve performance and
> reduce size of ixgbevf_tx_map()
> 
> Based on commit ec718254cbfe
> ("ixgbe: Improve performance and reduce size of ixgbe_tx_map")
> 
> This change is meant to both improve the performance and reduce the size of
> ixgbevf_tx_map().
> 
> Expand the work done in the main loop by pushing first into tx_buffer.
> This allows us to pull in the dma_mapping_error check, the tx_buffer value
> assignment, and the initial DMA value assignment to the Tx descriptor.
> 
> Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
> ---
>  drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |   45 +++++++++------------
>  1 file changed, 20 insertions(+), 25 deletions(-)
> 
> diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> index a793f9e..d3415ee 100644
> --- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> +++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> @@ -3532,34 +3532,37 @@ static void ixgbevf_tx_map(struct ixgbevf_ring
> *tx_ring,
>  			   struct ixgbevf_tx_buffer *first,
>  			   const u8 hdr_len)
>  {
> -	dma_addr_t dma;
>  	struct sk_buff *skb = first->skb;
>  	struct ixgbevf_tx_buffer *tx_buffer;
>  	union ixgbe_adv_tx_desc *tx_desc;
> -	struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
> -	unsigned int data_len = skb->data_len;
> -	unsigned int size = skb_headlen(skb);
> -	unsigned int paylen = skb->len - hdr_len;
> +	struct skb_frag_struct *frag;
> +	dma_addr_t dma;
> +	unsigned int data_len, size;
>  	u32 tx_flags = first->tx_flags;
> -	__le32 cmd_type;
> +	__le32 cmd_type = ixgbevf_tx_cmd_type(tx_flags);
>  	u16 i = tx_ring->next_to_use;
> 
>  	tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
> 
> -	ixgbevf_tx_olinfo_status(tx_desc, tx_flags, paylen);
> -	cmd_type = ixgbevf_tx_cmd_type(tx_flags);
> +	ixgbevf_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
> +
> +	size = skb_headlen(skb);
> +	data_len = skb->data_len;
> 
>  	dma = dma_map_single(tx_ring->dev, skb->data, size,
> DMA_TO_DEVICE);
> -	if (dma_mapping_error(tx_ring->dev, dma))
> -		goto dma_error;
> 
> -	/* record length, and DMA address */
> -	dma_unmap_len_set(first, len, size);
> -	dma_unmap_addr_set(first, dma, dma);
> +	tx_buffer = first;
> 
> -	tx_desc->read.buffer_addr = cpu_to_le64(dma);
> +	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
> +		if (dma_mapping_error(tx_ring->dev, dma))
> +			goto dma_error;
> +
> +		/* record length, and DMA address */
> +		dma_unmap_len_set(tx_buffer, len, size);
> +		dma_unmap_addr_set(tx_buffer, dma, dma);
> +
> +		tx_desc->read.buffer_addr = cpu_to_le64(dma);
> 
> -	for (;;) {
>  		while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
>  			tx_desc->read.cmd_type_len =
>  				cmd_type |
> cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
> @@ -3570,12 +3573,12 @@ static void ixgbevf_tx_map(struct ixgbevf_ring
> *tx_ring,
>  				tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
>  				i = 0;
>  			}
> +			tx_desc->read.olinfo_status = 0;
> 
>  			dma += IXGBE_MAX_DATA_PER_TXD;
>  			size -= IXGBE_MAX_DATA_PER_TXD;
> 
>  			tx_desc->read.buffer_addr = cpu_to_le64(dma);
> -			tx_desc->read.olinfo_status = 0;
>  		}
> 
>  		if (likely(!data_len))
> @@ -3589,23 +3592,15 @@ static void ixgbevf_tx_map(struct ixgbevf_ring
> *tx_ring,
>  			tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
>  			i = 0;
>  		}
> +		tx_desc->read.olinfo_status = 0;
> 
>  		size = skb_frag_size(frag);
>  		data_len -= size;
> 
>  		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
>  				       DMA_TO_DEVICE);
> -		if (dma_mapping_error(tx_ring->dev, dma))
> -			goto dma_error;
> 
>  		tx_buffer = &tx_ring->tx_buffer_info[i];
> -		dma_unmap_len_set(tx_buffer, len, size);
> -		dma_unmap_addr_set(tx_buffer, dma, dma);
> -
> -		tx_desc->read.buffer_addr = cpu_to_le64(dma);
> -		tx_desc->read.olinfo_status = 0;
> -
> -		frag++;
>  	}
> 
>  	/* write last descriptor with RS and EOP bits */
> 
> _______________________________________________
> Intel-wired-lan mailing list
> Intel-wired-lan at osuosl.org
> https://lists.osuosl.org/mailman/listinfo/intel-wired-lan

Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>

[Intel-wired-lan] [PATCH 6/9] ixgbevf: add counters for Rx page allocations

[Singh, Krishneil K]

> -----Original Message-----
> From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf Of
> Emil Tantilov
> Sent: Monday, December 11, 2017 10:37 AM
> To: intel-wired-lan at lists.osuosl.org
> Subject: [Intel-wired-lan] [PATCH 6/9] ixgbevf: add counters for Rx page
> allocations
> 
> We already had placehloders for failed page and buffer allocations.
> Added alloc_rx_page and made sure the stats are properly updated and
> exposed in ethtool.
> 
> Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
> ---
>  drivers/net/ethernet/intel/ixgbevf/ethtool.c      |    3 +++
>  drivers/net/ethernet/intel/ixgbevf/ixgbevf.h      |    6 ++++-
>  drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |   23 ++++++++++++++++--
> ---
>  3 files changed, 25 insertions(+), 7 deletions(-)
> 
> diff --git a/drivers/net/ethernet/intel/ixgbevf/ethtool.c
> b/drivers/net/ethernet/intel/ixgbevf/ethtool.c
> index ff9d05f..4400e49 100644
> --- a/drivers/net/ethernet/intel/ixgbevf/ethtool.c
> +++ b/drivers/net/ethernet/intel/ixgbevf/ethtool.c
> @@ -75,6 +75,9 @@ struct ixgbe_stats {
>  	IXGBEVF_STAT("tx_timeout_count", tx_timeout_count),
>  	IXGBEVF_NETDEV_STAT(multicast),
>  	IXGBEVF_STAT("rx_csum_offload_errors", hw_csum_rx_error),
> +	IXGBEVF_STAT("alloc_rx_page", alloc_rx_page),
> +	IXGBEVF_STAT("alloc_rx_page_failed", alloc_rx_page_failed),
> +	IXGBEVF_STAT("alloc_rx_buff_failed", alloc_rx_buff_failed),
>  };
> 
>  #define IXGBEVF_QUEUE_STATS_LEN ( \
> diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
> b/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
> index c70a789..f695242 100644
> --- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
> +++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
> @@ -84,6 +84,7 @@ struct ixgbevf_tx_queue_stats {
>  struct ixgbevf_rx_queue_stats {
>  	u64 alloc_rx_page_failed;
>  	u64 alloc_rx_buff_failed;
> +	u64 alloc_rx_page;
>  	u64 csum_err;
>  };
> 
> @@ -295,8 +296,9 @@ struct ixgbevf_adapter {
>  	u64 hw_csum_rx_error;
>  	u64 hw_rx_no_dma_resources;
>  	int num_msix_vectors;
> -	u32 alloc_rx_page_failed;
> -	u32 alloc_rx_buff_failed;
> +	u64 alloc_rx_page_failed;
> +	u64 alloc_rx_buff_failed;
> +	u64 alloc_rx_page;
> 
>  	struct msix_entry *msix_entries;
> 
> diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> index ae2402d..350afec 100644
> --- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> +++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> @@ -604,7 +604,7 @@ static bool ixgbevf_alloc_mapped_page(struct
> ixgbevf_ring *rx_ring,
>  	if (dma_mapping_error(rx_ring->dev, dma)) {
>  		__free_page(page);
> 
> -		rx_ring->rx_stats.alloc_rx_buff_failed++;
> +		rx_ring->rx_stats.alloc_rx_page_failed++;
>  		return false;
>  	}
> 
> @@ -612,6 +612,7 @@ static bool ixgbevf_alloc_mapped_page(struct
> ixgbevf_ring *rx_ring,
>  	bi->page = page;
>  	bi->page_offset = 0;
>  	bi->pagecnt_bias = 1;
> +	rx_ring->rx_stats.alloc_rx_page++;
> 
>  	return true;
>  }
> @@ -963,8 +964,10 @@ static int ixgbevf_clean_rx_irq(struct
> ixgbevf_q_vector *q_vector,
>  		skb = ixgbevf_fetch_rx_buffer(rx_ring, rx_desc, skb);
> 
>  		/* exit if we failed to retrieve a buffer */
> -		if (!skb)
> +		if (!skb) {
> +			rx_ring->rx_stats.alloc_rx_buff_failed++;
>  			break;
> +		}
> 
>  		cleaned_count++;
> 
> @@ -2749,6 +2752,8 @@ static int ixgbevf_sw_init(struct ixgbevf_adapter
> *adapter)
>  void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
>  {
>  	struct ixgbe_hw *hw = &adapter->hw;
> +	u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
> +	u64 alloc_rx_page = 0, hw_csum_rx_error = 0;
>  	int i;
> 
>  	if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
> @@ -2769,10 +2774,18 @@ void ixgbevf_update_stats(struct ixgbevf_adapter
> *adapter)
>  				adapter->stats.vfmprc);
> 
>  	for (i = 0;  i  < adapter->num_rx_queues;  i++) {
> -		adapter->hw_csum_rx_error +=
> -			adapter->rx_ring[i]->hw_csum_rx_error;
> -		adapter->rx_ring[i]->hw_csum_rx_error = 0;
> +		struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
> +
> +		hw_csum_rx_error += rx_ring->rx_stats.csum_err;
> +		alloc_rx_page_failed += rx_ring-
> >rx_stats.alloc_rx_page_failed;
> +		alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
> +		alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
>  	}
> +
> +	adapter->hw_csum_rx_error = hw_csum_rx_error;
> +	adapter->alloc_rx_page_failed = alloc_rx_page_failed;
> +	adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
> +	adapter->alloc_rx_page = alloc_rx_page;
>  }
> 
>  /**
> 
> _______________________________________________
> Intel-wired-lan mailing list
> Intel-wired-lan at osuosl.org
> https://lists.osuosl.org/mailman/listinfo/intel-wired-lan

Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>

[Intel-wired-lan] [PATCH 7/9] ixgbevf: clear rx_buffer_info in configure instead of clean

[Singh, Krishneil K]

> -----Original Message-----
> From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf Of
> Emil Tantilov
> Sent: Monday, December 11, 2017 10:37 AM
> To: intel-wired-lan at lists.osuosl.org
> Subject: [Intel-wired-lan] [PATCH 7/9] ixgbevf: clear rx_buffer_info in configure
> instead of clean
> 
> Based on commit d2bead576e67
> ("igb: Clear Rx buffer_info in configure instead of clean")
> 
> This change makes it so that instead of going through the entire ring on Rx
> cleanup we only go through the region that was designated to be cleaned up
> and stop when we reach the region where new allocations should start.
> 
> In addition we can avoid having to perform a memset on the Rx buffer_info
> structures until we are about to start using the ring again.
> 
> Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
> ---
>  drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |   26 +++++++++++----------
>  1 file changed, 13 insertions(+), 13 deletions(-)
> 
> diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> index 350afec..a793f9e 100644
> --- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> +++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> @@ -1773,6 +1773,10 @@ static void ixgbevf_configure_rx_ring(struct
> ixgbevf_adapter *adapter,
>  	IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
>  	ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
> 
> +	/* initialize rx_buffer_info */
> +	memset(ring->rx_buffer_info, 0,
> +	       sizeof(struct ixgbevf_rx_buffer) * ring->count);
> +
>  	/* initialize Rx descriptor 0 */
>  	rx_desc = IXGBEVF_RX_DESC(ring, 0);
>  	rx_desc->wb.upper.length = 0;
> @@ -2131,8 +2135,7 @@ void ixgbevf_up(struct ixgbevf_adapter *adapter)
>   **/
>  static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
>  {
> -	unsigned long size;
> -	unsigned int i;
> +	u16 i = rx_ring->next_to_clean;
> 
>  	/* Free Rx ring sk_buff */
>  	if (rx_ring->skb) {
> @@ -2140,17 +2143,11 @@ static void ixgbevf_clean_rx_ring(struct
> ixgbevf_ring *rx_ring)
>  		rx_ring->skb = NULL;
>  	}
> 
> -	/* ring already cleared, nothing to do */
> -	if (!rx_ring->rx_buffer_info)
> -		return;
> -
>  	/* Free all the Rx ring pages */
> -	for (i = 0; i < rx_ring->count; i++) {
> +	while (i != rx_ring->next_to_alloc) {
>  		struct ixgbevf_rx_buffer *rx_buffer;
> 
>  		rx_buffer = &rx_ring->rx_buffer_info[i];
> -		if (!rx_buffer->page)
> -			continue;
> 
>  		/* Invalidate cache lines that may have been written to by
>  		 * device so that we avoid corrupting memory.
> @@ -2171,11 +2168,14 @@ static void ixgbevf_clean_rx_ring(struct
> ixgbevf_ring *rx_ring)
>  		__page_frag_cache_drain(rx_buffer->page,
>  					rx_buffer->pagecnt_bias);
> 
> -		rx_buffer->page = NULL;
> +		i++;
> +		if (i == rx_ring->count)
> +			i = 0;
>  	}
> 
> -	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
> -	memset(rx_ring->rx_buffer_info, 0, size);
> +	rx_ring->next_to_alloc = 0;
> +	rx_ring->next_to_clean = 0;
> +	rx_ring->next_to_use = 0;
>  }
> 
>  /**
> @@ -3090,7 +3090,7 @@ int ixgbevf_setup_rx_resources(struct ixgbevf_ring
> *rx_ring)
>  	int size;
> 
>  	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
> -	rx_ring->rx_buffer_info = vzalloc(size);
> +	rx_ring->rx_buffer_info = vmalloc(size);
>  	if (!rx_ring->rx_buffer_info)
>  		goto err;
> 
> 
> _______________________________________________
> Intel-wired-lan mailing list
> Intel-wired-lan at osuosl.org
> https://lists.osuosl.org/mailman/listinfo/intel-wired-lan

Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>

[Intel-wired-lan] [PATCH 2/9] ixgbevf: only DMA sync frame length

[Singh, Krishneil K]

> -----Original Message-----
> From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf Of
> Emil Tantilov
> Sent: Monday, December 11, 2017 10:37 AM
> To: intel-wired-lan at lists.osuosl.org
> Subject: [Intel-wired-lan] [PATCH 2/9] ixgbevf: only DMA sync frame length
> 
> Based on commit 64f2525ca4e7 ("igb: Only DMA sync frame length")
> 
> On some architectures synching a buffer for DMA may be expensive.
> Instead of the entire 2K receive buffer only synchronize the length of
> the frame, which will typically be the MTU or smaller.
> 
> Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
> ---
>  drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |    7 ++++---
>  1 file changed, 4 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> index 7ffd429..0cc2688 100644
> --- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> +++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> @@ -803,12 +803,12 @@ static bool ixgbevf_can_reuse_rx_page(struct
> ixgbevf_rx_buffer *rx_buffer,
>   **/
>  static bool ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
>  				struct ixgbevf_rx_buffer *rx_buffer,
> +				u16 size,
>  				union ixgbe_adv_rx_desc *rx_desc,
>  				struct sk_buff *skb)
>  {
>  	struct page *page = rx_buffer->page;
>  	unsigned char *va = page_address(page) + rx_buffer->page_offset;
> -	unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
>  #if (PAGE_SIZE < 8192)
>  	unsigned int truesize = IXGBEVF_RX_BUFSZ;
>  #else
> @@ -856,6 +856,7 @@ static struct sk_buff *ixgbevf_fetch_rx_buffer(struct
> ixgbevf_ring *rx_ring,
>  {
>  	struct ixgbevf_rx_buffer *rx_buffer;
>  	struct page *page;
> +	u16 size = le16_to_cpu(rx_desc->wb.upper.length);
> 
>  	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
>  	page = rx_buffer->page;
> @@ -890,11 +891,11 @@ static struct sk_buff *ixgbevf_fetch_rx_buffer(struct
> ixgbevf_ring *rx_ring,
>  	dma_sync_single_range_for_cpu(rx_ring->dev,
>  				      rx_buffer->dma,
>  				      rx_buffer->page_offset,
> -				      IXGBEVF_RX_BUFSZ,
> +				      size,
>  				      DMA_FROM_DEVICE);
> 
>  	/* pull page into skb */
> -	if (ixgbevf_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
> +	if (ixgbevf_add_rx_frag(rx_ring, rx_buffer, size, rx_desc, skb)) {
>  		/* hand second half of page back to the ring */
>  		ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
>  	} else {
> 
> _______________________________________________
> Intel-wired-lan mailing list
> Intel-wired-lan at osuosl.org
> https://lists.osuosl.org/mailman/listinfo/intel-wired-lan

Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>

[Intel-wired-lan] [PATCH 5/9] ixgbevf: update code to better handle incrementing page count

[Singh, Krishneil K]

> -----Original Message-----
> From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf Of
> Emil Tantilov
> Sent: Monday, December 11, 2017 10:37 AM
> To: intel-wired-lan at lists.osuosl.org
> Subject: [Intel-wired-lan] [PATCH 5/9] ixgbevf: update code to better handle
> incrementing page count
> 
> Based on commit bd4171a5d4c2
> ("igb: update code to better handle incrementing page count")
> 
> Update the driver code so that we do bulk updates of the page reference
> count instead of just incrementing it by one reference at a time.  The
> advantage to doing this is that we cut down on atomic operations and
> this in turn should give us a slight improvement in cycles per packet.
> In addition if we eventually move this over to using build_skb the gains
> will be more noticeable.
> 
> Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
> ---
>  drivers/net/ethernet/intel/ixgbevf/ixgbevf.h      |    7 ++++-
>  drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |   30 +++++++++++++++----
> --
>  2 files changed, 28 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
> b/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
> index b1da9f4..c70a789 100644
> --- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
> +++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf.h
> @@ -62,7 +62,12 @@ struct ixgbevf_tx_buffer {
>  struct ixgbevf_rx_buffer {
>  	dma_addr_t dma;
>  	struct page *page;
> -	unsigned int page_offset;
> +#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
> +	__u32 page_offset;
> +#else
> +	__u16 page_offset;
> +#endif
> +	__u16 pagecnt_bias;
>  };
> 
>  struct ixgbevf_stats {
> diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> index fbd493e..ae2402d 100644
> --- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> +++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> @@ -611,6 +611,7 @@ static bool ixgbevf_alloc_mapped_page(struct
> ixgbevf_ring *rx_ring,
>  	bi->dma = dma;
>  	bi->page = page;
>  	bi->page_offset = 0;
> +	bi->pagecnt_bias = 1;
> 
>  	return true;
>  }
> @@ -747,6 +748,7 @@ static void ixgbevf_reuse_rx_page(struct ixgbevf_ring
> *rx_ring,
>  	new_buff->page = old_buff->page;
>  	new_buff->dma = old_buff->dma;
>  	new_buff->page_offset = old_buff->page_offset;
> +	new_buff->pagecnt_bias = old_buff->pagecnt_bias;
>  }
> 
>  static inline bool ixgbevf_page_is_reserved(struct page *page)
> @@ -758,13 +760,15 @@ static bool ixgbevf_can_reuse_rx_page(struct
> ixgbevf_rx_buffer *rx_buffer,
>  				      struct page *page,
>  				      const unsigned int truesize)
>  {
> +	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias--;
> +
>  	/* avoid re-using remote pages */
>  	if (unlikely(ixgbevf_page_is_reserved(page)))
>  		return false;
> 
>  #if (PAGE_SIZE < 8192)
>  	/* if we are only owner of page we can reuse it */
> -	if (unlikely(page_count(page) != 1))
> +	if (unlikely(page_ref_count(page) != pagecnt_bias))
>  		return false;
> 
>  	/* flip page offset to other buffer */
> @@ -778,10 +782,15 @@ static bool ixgbevf_can_reuse_rx_page(struct
> ixgbevf_rx_buffer *rx_buffer,
>  		return false;
> 
>  #endif
> -	/* Even if we own the page, we are not allowed to use atomic_set()
> -	 * This would break get_page_unless_zero() users.
> +
> +	/* If we have drained the page fragment pool we need to update
> +	 * the pagecnt_bias and page count so that we fully restock the
> +	 * number of references the driver holds.
>  	 */
> -	page_ref_inc(page);
> +	if (unlikely(pagecnt_bias == 1)) {
> +		page_ref_add(page, USHRT_MAX);
> +		rx_buffer->pagecnt_bias = USHRT_MAX;
> +	}
> 
>  	return true;
>  }
> @@ -827,7 +836,6 @@ static bool ixgbevf_add_rx_frag(struct ixgbevf_ring
> *rx_ring,
>  			return true;
> 
>  		/* this page cannot be reused so discard it */
> -		put_page(page);
>  		return false;
>  	}
> 
> @@ -899,10 +907,13 @@ static struct sk_buff *ixgbevf_fetch_rx_buffer(struct
> ixgbevf_ring *rx_ring,
>  		/* hand second half of page back to the ring */
>  		ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
>  	} else {
> -		/* we are not reusing the buffer so unmap it */
> +		/* We are not reusing the buffer so unmap it and free
> +		 * any references we are holding to it
> +		 */
>  		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
>  				     PAGE_SIZE, DMA_FROM_DEVICE,
>  				     IXGBEVF_RX_DMA_ATTR);
> +		__page_frag_cache_drain(page, rx_buffer->pagecnt_bias);
>  	}
> 
>  	/* clear contents of buffer_info */
> @@ -2135,6 +2146,8 @@ static void ixgbevf_clean_rx_ring(struct ixgbevf_ring
> *rx_ring)
>  		struct ixgbevf_rx_buffer *rx_buffer;
> 
>  		rx_buffer = &rx_ring->rx_buffer_info[i];
> +		if (!rx_buffer->page)
> +			continue;
> 
>  		/* Invalidate cache lines that may have been written to by
>  		 * device so that we avoid corrupting memory.
> @@ -2152,8 +2165,9 @@ static void ixgbevf_clean_rx_ring(struct ixgbevf_ring
> *rx_ring)
>  				     DMA_FROM_DEVICE,
>  				     IXGBEVF_RX_DMA_ATTR);
> 
> -		if (rx_buffer->page)
> -			__free_page(rx_buffer->page);
> +		__page_frag_cache_drain(rx_buffer->page,
> +					rx_buffer->pagecnt_bias);
> +
>  		rx_buffer->page = NULL;
>  	}
> 
> 
> _______________________________________________
> Intel-wired-lan mailing list
> Intel-wired-lan at osuosl.org
> https://lists.osuosl.org/mailman/listinfo/intel-wired-lan

Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>

[Intel-wired-lan] [PATCH 9/9] ixgbevf: don't bother clearing tx_buffer_info in ixgbevf_clean_tx_ring()

[Singh, Krishneil K]

> -----Original Message-----
> From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf Of
> Emil Tantilov
> Sent: Monday, December 11, 2017 10:38 AM
> To: intel-wired-lan at lists.osuosl.org
> Subject: [Intel-wired-lan] [PATCH 9/9] ixgbevf: don't bother clearing
> tx_buffer_info in ixgbevf_clean_tx_ring()
> 
> Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
> ---
>  drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |  115 +++++++++++++------
> --
>  1 file changed, 72 insertions(+), 43 deletions(-)
> 
> diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> index d3415ee..9b3d43d 100644
> --- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> +++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> @@ -206,28 +206,6 @@ static void ixgbevf_set_ivar(struct ixgbevf_adapter
> *adapter, s8 direction,
>  	}
>  }
> 
> -static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
> -					struct ixgbevf_tx_buffer *tx_buffer)
> -{
> -	if (tx_buffer->skb) {
> -		dev_kfree_skb_any(tx_buffer->skb);
> -		if (dma_unmap_len(tx_buffer, len))
> -			dma_unmap_single(tx_ring->dev,
> -					 dma_unmap_addr(tx_buffer, dma),
> -					 dma_unmap_len(tx_buffer, len),
> -					 DMA_TO_DEVICE);
> -	} else if (dma_unmap_len(tx_buffer, len)) {
> -		dma_unmap_page(tx_ring->dev,
> -			       dma_unmap_addr(tx_buffer, dma),
> -			       dma_unmap_len(tx_buffer, len),
> -			       DMA_TO_DEVICE);
> -	}
> -	tx_buffer->next_to_watch = NULL;
> -	tx_buffer->skb = NULL;
> -	dma_unmap_len_set(tx_buffer, len, 0);
> -	/* tx_buffer must be completely set up in the transmit path */
> -}
> -
>  static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
>  {
>  	return ring->stats.packets;
> @@ -349,7 +327,6 @@ static bool ixgbevf_clean_tx_irq(struct
> ixgbevf_q_vector *q_vector,
>  				 DMA_TO_DEVICE);
> 
>  		/* clear tx_buffer data */
> -		tx_buffer->skb = NULL;
>  		dma_unmap_len_set(tx_buffer, len, 0);
> 
>  		/* unmap remaining buffers */
> @@ -1576,6 +1553,10 @@ static void ixgbevf_configure_tx_ring(struct
> ixgbevf_adapter *adapter,
>  	txdctl |= (1u << 8) |    /* HTHRESH = 1 */
>  		   32;           /* PTHRESH = 32 */
> 
> +	/* reinitialize tx_buffer_info */
> +	memset(ring->tx_buffer_info, 0,
> +	       sizeof(struct ixgbevf_tx_buffer) * ring->count);
> +
>  	clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
> 
>  	IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
> @@ -2184,23 +2165,57 @@ static void ixgbevf_clean_rx_ring(struct
> ixgbevf_ring *rx_ring)
>   **/
>  static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
>  {
> -	struct ixgbevf_tx_buffer *tx_buffer_info;
> -	unsigned long size;
> -	unsigned int i;
> +	u16 i = tx_ring->next_to_clean;
> +	struct ixgbevf_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
> 
> -	if (!tx_ring->tx_buffer_info)
> -		return;
> +	while (i != tx_ring->next_to_use) {
> +		union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
> +
> +		/* Free all the Tx ring sk_buffs */
> +		dev_kfree_skb_any(tx_buffer->skb);
> +
> +		/* unmap skb header data */
> +		dma_unmap_single(tx_ring->dev,
> +				 dma_unmap_addr(tx_buffer, dma),
> +				 dma_unmap_len(tx_buffer, len),
> +				 DMA_TO_DEVICE);
> +
> +		/* check for eop_desc to determine the end of the packet */
> +		eop_desc = tx_buffer->next_to_watch;
> +		tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
> +
> +		/* unmap remaining buffers */
> +		while (tx_desc != eop_desc) {
> +			tx_buffer++;
> +			tx_desc++;
> +			i++;
> +			if (unlikely(i == tx_ring->count)) {
> +				i = 0;
> +				tx_buffer = tx_ring->tx_buffer_info;
> +				tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
> +			}
> +
> +			/* unmap any remaining paged data */
> +			if (dma_unmap_len(tx_buffer, len))
> +				dma_unmap_page(tx_ring->dev,
> +					       dma_unmap_addr(tx_buffer,
> dma),
> +					       dma_unmap_len(tx_buffer, len),
> +					       DMA_TO_DEVICE);
> +		}
> 
> -	/* Free all the Tx ring sk_buffs */
> -	for (i = 0; i < tx_ring->count; i++) {
> -		tx_buffer_info = &tx_ring->tx_buffer_info[i];
> -		ixgbevf_unmap_and_free_tx_resource(tx_ring,
> tx_buffer_info);
> +		/* move us one more past the eop_desc for start of next pkt */
> +		tx_buffer++;
> +		i++;
> +		if (unlikely(i == tx_ring->count)) {
> +			i = 0;
> +			tx_buffer = tx_ring->tx_buffer_info;
> +		}
>  	}
> 
> -	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
> -	memset(tx_ring->tx_buffer_info, 0, size);
> +	/* reset next_to_use and next_to_clean */
> +	tx_ring->next_to_use = 0;
> +	tx_ring->next_to_clean = 0;
> 
> -	memset(tx_ring->desc, 0, tx_ring->size);
>  }
> 
>  /**
> @@ -3030,7 +3045,7 @@ int ixgbevf_setup_tx_resources(struct ixgbevf_ring
> *tx_ring)
>  	int size;
> 
>  	size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
> -	tx_ring->tx_buffer_info = vzalloc(size);
> +	tx_ring->tx_buffer_info = vmalloc(size);
>  	if (!tx_ring->tx_buffer_info)
>  		goto err;
> 
> @@ -3634,18 +3649,32 @@ static void ixgbevf_tx_map(struct ixgbevf_ring
> *tx_ring,
>  	return;
>  dma_error:
>  	dev_err(tx_ring->dev, "TX DMA map failed\n");
> +	tx_buffer = &tx_ring->tx_buffer_info[i];
> 
>  	/* clear dma mappings for failed tx_buffer_info map */
> -	for (;;) {
> +	while (tx_buffer != first) {
> +		if (dma_unmap_len(tx_buffer, len))
> +			dma_unmap_page(tx_ring->dev,
> +				       dma_unmap_addr(tx_buffer, dma),
> +				       dma_unmap_len(tx_buffer, len),
> +				       DMA_TO_DEVICE);
> +		dma_unmap_len_set(tx_buffer, len, 0);
> +
> +		if (i-- == 0)
> +			i += tx_ring->count;
>  		tx_buffer = &tx_ring->tx_buffer_info[i];
> -		ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer);
> -		if (tx_buffer == first)
> -			break;
> -		if (i == 0)
> -			i = tx_ring->count;
> -		i--;
>  	}
> 
> +	if (dma_unmap_len(tx_buffer, len))
> +		dma_unmap_single(tx_ring->dev,
> +				 dma_unmap_addr(tx_buffer, dma),
> +				 dma_unmap_len(tx_buffer, len),
> +				 DMA_TO_DEVICE);
> +	dma_unmap_len_set(tx_buffer, len, 0);
> +
> +	dev_kfree_skb_any(tx_buffer->skb);
> +	tx_buffer->skb = NULL;
> +
>  	tx_ring->next_to_use = i;
>  }
> 
> 
> _______________________________________________
> Intel-wired-lan mailing list
> Intel-wired-lan at osuosl.org
> https://lists.osuosl.org/mailman/listinfo/intel-wired-lan

Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>

[Intel-wired-lan] [PATCH 3/9] ixgbevf: use length to determine if descriptor is done

[Singh, Krishneil K]

> -----Original Message-----
> From: Intel-wired-lan [mailto:intel-wired-lan-bounces at osuosl.org] On Behalf Of
> Emil Tantilov
> Sent: Monday, December 11, 2017 10:37 AM
> To: intel-wired-lan at lists.osuosl.org
> Subject: [Intel-wired-lan] [PATCH 3/9] ixgbevf: use length to determine if
> descriptor is done
> 
> Based on:
> commit 7ec0116c9131 ("igb: Use length to determine if descriptor is done")
> 
> This change makes it so that we use the length of the packet instead of the
> DD status bit to determine if a new descriptor is ready to be processed.
> The obvious advantage is that it cuts down on reads as we don't really even
> need the DD bit if going from a 0 to a non-zero value on size is enough to
> inform us that the packet has been completed.
> 
> In addition we only reset the Rx descriptor length for descriptor zero when
> resetting a ring instead of having to do a memset with 0 over the entire
> ring. By doing this we can save some time on initialization.
> 
> Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
> ---
>  drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c |   14 ++++++++------
>  1 file changed, 8 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> index 0cc2688..725fe2d 100644
> --- a/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> +++ b/drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c
> @@ -653,8 +653,8 @@ static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring
> *rx_ring,
>  			i -= rx_ring->count;
>  		}
> 
> -		/* clear the hdr_addr for the next_to_use descriptor */
> -		rx_desc->read.hdr_addr = 0;
> +		/* clear the length for the next_to_use descriptor */
> +		rx_desc->wb.upper.length = 0;
> 
>  		cleaned_count--;
>  	} while (cleaned_count);
> @@ -938,7 +938,7 @@ static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector
> *q_vector,
> 
>  		rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
> 
> -		if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
> +		if (!rx_desc->wb.upper.length)
>  			break;
> 
>  		/* This memory barrier is needed to keep us from reading
> @@ -1729,6 +1729,7 @@ static void ixgbevf_configure_rx_ring(struct
> ixgbevf_adapter *adapter,
>  				      struct ixgbevf_ring *ring)
>  {
>  	struct ixgbe_hw *hw = &adapter->hw;
> +	union ixgbe_adv_rx_desc *rx_desc;
>  	u64 rdba = ring->dma;
>  	u32 rxdctl;
>  	u8 reg_idx = ring->reg_idx;
> @@ -1757,6 +1758,10 @@ static void ixgbevf_configure_rx_ring(struct
> ixgbevf_adapter *adapter,
>  	IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
>  	ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
> 
> +	/* initialize Rx descriptor 0 */
> +	rx_desc = IXGBEVF_RX_DESC(ring, 0);
> +	rx_desc->wb.upper.length = 0;
> +
>  	/* reset ntu and ntc to place SW in sync with hardwdare */
>  	ring->next_to_clean = 0;
>  	ring->next_to_use = 0;
> @@ -2141,9 +2146,6 @@ static void ixgbevf_clean_rx_ring(struct ixgbevf_ring
> *rx_ring)
> 
>  	size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
>  	memset(rx_ring->rx_buffer_info, 0, size);
> -
> -	/* Zero out the descriptor ring */
> -	memset(rx_ring->desc, 0, rx_ring->size);
>  }
> 
>  /**
> 
> _______________________________________________
> Intel-wired-lan mailing list
> Intel-wired-lan at osuosl.org
> https://lists.osuosl.org/mailman/listinfo/intel-wired-lan

Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>