]> www.infradead.org Git - users/jedix/linux-maple.git/commitdiff
i40e/i40evf: Add support for mapping pages with DMA attributes
authorAlexander Duyck <alexander.h.duyck@intel.com>
Mon, 30 Jan 2017 20:29:35 +0000 (12:29 -0800)
committerJack Vogel <jack.vogel@oracle.com>
Wed, 5 Jul 2017 20:20:48 +0000 (13:20 -0700)
This patch adds support for DMA_ATTR_SKIP_CPU_SYNC and
DMA_ATTR_WEAK_ORDERING. By enabling both of these for the Rx path we
are able to see performance improvements on architectures that implement
either one due to the fact that page mapping and unmapping only has to
sync what is actually being used instead of the entire buffer. In addition
by enabling the weak ordering attribute enables a performance improvement
for architectures that can associate a memory ordering with a DMA buffer
such as Sparc.

Change-ID: If176824e8231c5b24b8a5d55b339a6026738fc75
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Orabug: 26396243
Local modifications to account for dma_attr data type difference.
(cherry picked from commit 59605bc09630c2b577858c371edf89c099b5f925)
Signed-off-by: Jack Vogel <jack.vogel@oracle.com>
Reviewed-by: Shannon Nelson <shannon.nelson@oracle.com>
drivers/net/ethernet/intel/i40e/i40e_txrx.c
drivers/net/ethernet/intel/i40e/i40e_txrx.h
drivers/net/ethernet/intel/i40evf/i40e_txrx.c
drivers/net/ethernet/intel/i40evf/i40e_txrx.h

index db5f9056114291d93311563bdfd1869c51795380..1db8e00945a9e798ac1e7eeb1833459584dd762b 100644 (file)
@@ -1120,8 +1120,8 @@ err:
  **/
 void i40e_clean_rx_ring(struct i40e_ring *rx_ring)
 {
-       struct device *dev = rx_ring->dev;
        unsigned long bi_size;
+       DEFINE_DMA_ATTRS(dattrs);
        u16 i;
 
        /* ring already cleared, nothing to do */
@@ -1140,7 +1140,21 @@ void i40e_clean_rx_ring(struct i40e_ring *rx_ring)
                if (!rx_bi->page)
                        continue;
 
-               dma_unmap_page(dev, rx_bi->dma, PAGE_SIZE, DMA_FROM_DEVICE);
+               /* 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_bi->dma,
+                                             rx_bi->page_offset,
+                                             I40E_RXBUFFER_2048,
+                                             DMA_FROM_DEVICE);
+
+               /* free resources associated with mapping */
+               I40E_DMA_ATTRS(&dattrs);
+               dma_unmap_page_attrs(rx_ring->dev, rx_bi->dma,
+                                    PAGE_SIZE,
+                                    DMA_FROM_DEVICE,
+                                    &dattrs);
                __free_pages(rx_bi->page, 0);
 
                rx_bi->page = NULL;
@@ -1253,6 +1267,7 @@ static bool i40e_alloc_mapped_page(struct i40e_ring *rx_ring,
                                   struct i40e_rx_buffer *bi)
 {
        struct page *page = bi->page;
+       DEFINE_DMA_ATTRS(dattrs);
        dma_addr_t dma;
 
        /* since we are recycling buffers we should seldom need to alloc */
@@ -1269,7 +1284,11 @@ static bool i40e_alloc_mapped_page(struct i40e_ring *rx_ring,
        }
 
        /* map page for use */
-       dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE, DMA_FROM_DEVICE);
+       I40E_DMA_ATTRS(&dattrs);
+       dma = dma_map_page_attrs(rx_ring->dev, page, 0,
+                                PAGE_SIZE,
+                                DMA_FROM_DEVICE,
+                                &dattrs);
 
        /* if mapping failed free memory back to system since
         * there isn't much point in holding memory we can't use
@@ -1329,6 +1348,12 @@ bool i40e_alloc_rx_buffers(struct i40e_ring *rx_ring, u16 cleaned_count)
                if (!i40e_alloc_mapped_page(rx_ring, bi))
                        goto no_buffers;
 
+               /* sync the buffer for use by the device */
+               dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
+                                                bi->page_offset,
+                                                I40E_RXBUFFER_2048,
+                                                DMA_FROM_DEVICE);
+
                /* Refresh the desc even if buffer_addrs didn't change
                 * because each write-back erases this info.
                 */
@@ -1743,6 +1768,7 @@ struct sk_buff *i40e_fetch_rx_buffer(struct i40e_ring *rx_ring,
        unsigned int size =
                (local_status_error_len & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
                I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
+       DEFINE_DMA_ATTRS(dattrs);
        struct i40e_rx_buffer *rx_buffer;
        struct page *page;
 
@@ -1789,8 +1815,9 @@ struct sk_buff *i40e_fetch_rx_buffer(struct i40e_ring *rx_ring,
                rx_ring->rx_stats.page_reuse_count++;
        } else {
                /* we are not reusing the buffer so unmap it */
-               dma_unmap_page(rx_ring->dev, rx_buffer->dma, PAGE_SIZE,
-                              DMA_FROM_DEVICE);
+               I40E_DMA_ATTRS(&dattrs);
+               dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma, PAGE_SIZE,
+                                    DMA_FROM_DEVICE, &dattrs);
        }
 
        /* clear contents of buffer_info */
index f80979025c0131a07e7b956826f17877b8b7081a..a3511bd822c13f8a74c65d65a2f981b807957308 100644 (file)
@@ -133,6 +133,10 @@ enum i40e_dyn_idx_t {
 #define I40E_RX_HDR_SIZE I40E_RXBUFFER_256
 #define i40e_rx_desc i40e_32byte_rx_desc
 
+#define I40E_RX_DMA_ATTR \
+       (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING)
+#define I40E_DMA_ATTRS(x) dma_set_attr(I40E_RX_DMA_ATTR, x)
+
 /**
  * i40e_test_staterr - tests bits in Rx descriptor status and error fields
  * @rx_desc: pointer to receive descriptor (in le64 format)
index 790fdfe42f012c115c0cde0fcc6d91f7245e9fc4..1745611695e00f5c89990a275ca4a140e10f8ca1 100644 (file)
@@ -483,8 +483,8 @@ err:
  **/
 void i40evf_clean_rx_ring(struct i40e_ring *rx_ring)
 {
-       struct device *dev = rx_ring->dev;
        unsigned long bi_size;
+       DEFINE_DMA_ATTRS(dattrs);
        u16 i;
 
        /* ring already cleared, nothing to do */
@@ -503,7 +503,21 @@ void i40evf_clean_rx_ring(struct i40e_ring *rx_ring)
                if (!rx_bi->page)
                        continue;
 
-               dma_unmap_page(dev, rx_bi->dma, PAGE_SIZE, DMA_FROM_DEVICE);
+               /* 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_bi->dma,
+                                             rx_bi->page_offset,
+                                             I40E_RXBUFFER_2048,
+                                             DMA_FROM_DEVICE);
+
+               /* free resources associated with mapping */
+               I40E_DMA_ATTRS(&dattrs);
+               dma_unmap_page_attrs(rx_ring->dev, rx_bi->dma,
+                                    PAGE_SIZE,
+                                    DMA_FROM_DEVICE,
+                                    &dattrs);
                __free_pages(rx_bi->page, 0);
 
                rx_bi->page = NULL;
@@ -616,6 +630,7 @@ static bool i40e_alloc_mapped_page(struct i40e_ring *rx_ring,
                                   struct i40e_rx_buffer *bi)
 {
        struct page *page = bi->page;
+       DEFINE_DMA_ATTRS(dattrs);
        dma_addr_t dma;
 
        /* since we are recycling buffers we should seldom need to alloc */
@@ -632,7 +647,11 @@ static bool i40e_alloc_mapped_page(struct i40e_ring *rx_ring,
        }
 
        /* map page for use */
-       dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE, DMA_FROM_DEVICE);
+       I40E_DMA_ATTRS(&dattrs);
+       dma = dma_map_page_attrs(rx_ring->dev, page, 0,
+                                PAGE_SIZE,
+                                DMA_FROM_DEVICE,
+                                &dattrs);
 
        /* if mapping failed free memory back to system since
         * there isn't much point in holding memory we can't use
@@ -692,6 +711,12 @@ bool i40evf_alloc_rx_buffers(struct i40e_ring *rx_ring, u16 cleaned_count)
                if (!i40e_alloc_mapped_page(rx_ring, bi))
                        goto no_buffers;
 
+               /* sync the buffer for use by the device */
+               dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
+                                                bi->page_offset,
+                                                I40E_RXBUFFER_2048,
+                                                DMA_FROM_DEVICE);
+
                /* Refresh the desc even if buffer_addrs didn't change
                 * because each write-back erases this info.
                 */
@@ -1096,6 +1121,7 @@ struct sk_buff *i40evf_fetch_rx_buffer(struct i40e_ring *rx_ring,
        unsigned int size =
                (local_status_error_len & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
                I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
+       DEFINE_DMA_ATTRS(dattrs);
        struct i40e_rx_buffer *rx_buffer;
        struct page *page;
 
@@ -1142,8 +1168,9 @@ struct sk_buff *i40evf_fetch_rx_buffer(struct i40e_ring *rx_ring,
                rx_ring->rx_stats.page_reuse_count++;
        } else {
                /* we are not reusing the buffer so unmap it */
-               dma_unmap_page(rx_ring->dev, rx_buffer->dma, PAGE_SIZE,
-                              DMA_FROM_DEVICE);
+               I40E_DMA_ATTRS(&dattrs);
+               dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma, PAGE_SIZE,
+                                    DMA_FROM_DEVICE, &dattrs);
        }
 
        /* clear contents of buffer_info */
index 17c84930f216193fce819acc6cd7996121e6a895..8ad98fb708de358d83a00c39a9d343777cbffb5d 100644 (file)
@@ -120,6 +120,10 @@ enum i40e_dyn_idx_t {
 #define I40E_RX_HDR_SIZE I40E_RXBUFFER_256
 #define i40e_rx_desc i40e_32byte_rx_desc
 
+#define I40E_RX_DMA_ATTR \
+       (DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING)
+#define I40E_DMA_ATTRS(x) dma_set_attr(I40E_RX_DMA_ATTR, x)
+
 /**
  * i40e_test_staterr - tests bits in Rx descriptor status and error fields
  * @rx_desc: pointer to receive descriptor (in le64 format)