Commit 762f4c57 authored by Jesse Brandeburg's avatar Jesse Brandeburg Committed by Jeff Garzik

ixgbe: recycle pages in packet split mode

most of the time we only need 1500 bytes for a packet which means
we don't need a whole 4k page for each packet.  Share the allocation
by using a reference count to the page and giving half to two
receive descriptors.  This can enable us to use packet split mode
all the time due to the performance increase of allocating half
the pages.
Signed-off-by: default avatarJesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: default avatarJeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: default avatarJeff Garzik <jgarzik@redhat.com>
parent ff819cfb
......@@ -114,6 +114,7 @@ struct ixgbe_rx_buffer {
dma_addr_t dma;
struct page *page;
dma_addr_t page_dma;
unsigned int page_offset;
};
struct ixgbe_queue_stats {
......
......@@ -493,15 +493,23 @@ static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
while (cleaned_count--) {
rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
if (!bi->page &&
if (!bi->page_dma &&
(adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)) {
if (!bi->page) {
bi->page = alloc_page(GFP_ATOMIC);
if (!bi->page) {
adapter->alloc_rx_page_failed++;
goto no_buffers;
}
bi->page_dma = pci_map_page(pdev, bi->page, 0,
PAGE_SIZE,
bi->page_offset = 0;
} else {
/* use a half page if we're re-using */
bi->page_offset ^= (PAGE_SIZE / 2);
}
bi->page_dma = pci_map_page(pdev, bi->page,
bi->page_offset,
(PAGE_SIZE / 2),
PCI_DMA_FROMDEVICE);
}
......@@ -620,11 +628,18 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_adapter *adapter,
if (upper_len) {
pci_unmap_page(pdev, rx_buffer_info->page_dma,
PAGE_SIZE, PCI_DMA_FROMDEVICE);
PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
rx_buffer_info->page_dma = 0;
skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
rx_buffer_info->page, 0, upper_len);
rx_buffer_info->page,
rx_buffer_info->page_offset,
upper_len);
if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
(page_count(rx_buffer_info->page) != 1))
rx_buffer_info->page = NULL;
else
get_page(rx_buffer_info->page);
skb->len += upper_len;
skb->data_len += upper_len;
......@@ -647,6 +662,7 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_adapter *adapter,
rx_buffer_info->skb = next_buffer->skb;
rx_buffer_info->dma = next_buffer->dma;
next_buffer->skb = skb;
next_buffer->dma = 0;
adapter->non_eop_descs++;
goto next_desc;
}
......@@ -1534,10 +1550,7 @@ static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
int rx_buf_len;
/* Decide whether to use packet split mode or not */
if (netdev->mtu > ETH_DATA_LEN)
adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
else
adapter->flags &= ~IXGBE_FLAG_RX_PS_ENABLED;
/* Set the RX buffer length according to the mode */
if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
......@@ -2018,12 +2031,12 @@ static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
}
if (!rx_buffer_info->page)
continue;
pci_unmap_page(pdev, rx_buffer_info->page_dma, PAGE_SIZE,
pci_unmap_page(pdev, rx_buffer_info->page_dma, PAGE_SIZE / 2,
PCI_DMA_FROMDEVICE);
rx_buffer_info->page_dma = 0;
put_page(rx_buffer_info->page);
rx_buffer_info->page = NULL;
rx_buffer_info->page_offset = 0;
}
size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment