On 04/22/2015 01:14 PM, Cong Wang wrote: > On Wed, Apr 22, 2015 at 12:43 PM, Alexander Duyck > <alexander.du...@gmail.com> wrote: >> On 04/22/2015 10:45 AM, Cong Wang wrote: >>> The second parameter of eth_get_headlen() is the length of >>> the frame buffer, not the header length of skb. >>> >>> Cc: Jeff Kirsher <jeffrey.t.kirs...@intel.com> >>> Signed-off-by: Cong Wang <xiyou.wangc...@gmail.com> >>> --- >>> drivers/net/ethernet/intel/igb/igb_main.c | 4 +++- >>> 1 file changed, 3 insertions(+), 1 deletion(-) >>> >>> diff --git a/drivers/net/ethernet/intel/igb/igb_main.c >>> b/drivers/net/ethernet/intel/igb/igb_main.c >>> index a0a9b1f..7b3a370 100644 >>> --- a/drivers/net/ethernet/intel/igb/igb_main.c >>> +++ b/drivers/net/ethernet/intel/igb/igb_main.c >>> @@ -6852,7 +6852,9 @@ static void igb_pull_tail(struct igb_ring *rx_ring, >>> /* we need the header to contain the greater of either ETH_HLEN or >>> * 60 bytes if the skb->len is less than 60 for skb_pad. >>> */ >>> - pull_len = eth_get_headlen(va, IGB_RX_HDR_LEN); >>> + pull_len = eth_get_headlen(va, skb_frag_size(frag)); >>> + if (unlikely(pull_len > IGB_RX_HDR_LEN)) >>> + pull_len = IGB_RX_HDR_LEN; >>> >>> /* align pull length to size of long to optimize memcpy performance */ >>> skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long))); >> You have this part right. The length represents the maximum length we >> are willing to traverse in the buffer. So if we 100% want to get the >> entire header regardless of what we can copy into then we could follow >> your approach. However, since the allocated space in the skb is only >> IGB_RX_HDR_LEN we only really want to traverse up to that length. Then >> that is all we copy out of the header. > But the frag size could be smaller than IGB_RX_HDR_LEN which is > the case this patch tries to fix.
No it can't. We only add frags if the first frag is larger than IGB_RX_HDR_LEN. Go look at the code where this driver calls add_rx_frag. You should find there is copybreak code in there that kicks in for anything less than or equal to IGB_RX_HDR_LEN in size on the first frag. It is there to allow us to avoid having to perform an atomic inc/dec on the page. >> As a result we don't need the extra code for putting the upper limit on >> pull_len since that is factored in by passing IGB_RX_HDR_LEN as the >> maximum traversal length. > That is why I put unlikely() there, even with tunnel headers we unlikely > pull header longer than IGB_RX_HDR_LEN, the code is just for > completeness. The code is fine as is. I suggest you go look at the code where frags are attached to the sk_buff. - Alex -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
