This patch set cleans up some of the handling of page frags used in the skb
allocation. The issue was we were having to use a number of calls to
virt_to_head_page in a number of places and then following that up with
put_page. Both calls end up being expensive, the first due to size, and
the second due to the fact that we end up having to call a number of other
functions before we finally see the page freed in the case of compound
pages.
The skb_free_frag function is meant to resolve that by providing a
centralized location for the virt_to_head_page call and by coalesing
several checks such as the check for PageHead into a single check so that
we can keep the instruction cound minimal when freeing the page frag.
With this change I am seeing an improvement of about 5% in a simple
receive/drop test.
---
Alexander Duyck (6):
net: Add skb_free_frag to replace use of put_page in freeing skb->head
netcp: Replace put_page(virt_to_head_page(ptr)) w/ skb_free_frag
mvneta: Replace put_page(virt_to_head_page(ptr)) w/ skb_free_frag
e1000: Replace e1000_free_frag with skb_free_frag
hisilicon: Replace put_page(virt_to_head_page()) with skb_free_frag()
bnx2x, tg3: Replace put_page(virt_to_head_page()) with skb_free_frag()
drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c | 2 +-
drivers/net/ethernet/broadcom/tg3.c | 2 +-
drivers/net/ethernet/hisilicon/hip04_eth.c | 2 +-
drivers/net/ethernet/intel/e1000/e1000_main.c | 19 ++++++---------
drivers/net/ethernet/marvell/mvneta.c | 2 +-
drivers/net/ethernet/ti/netcp_core.c | 2 +-
include/linux/gfp.h | 1 +
include/linux/skbuff.h | 1 +
mm/page_alloc.c | 4 +--
net/core/skbuff.c | 29 +++++++++++++++++++++--
10 files changed, 41 insertions(+), 23 deletions(-)
--
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>