Eric Dumazet wrote:
On Thu, Feb 29, 2024 at 2:22 PM Richard Gobert richardbgobert@gmail.com wrote:
Eric Dumazet wrote:
My intuition is that this patch has a high cost for normal GRO processing. SW-GRO is already a bottleneck on ARM cores in smart NICS.
I would suggest instead using parameters to give both the nhoff and thoff values this would avoid many conditionals in the fast path.
->
INDIRECT_CALLABLE_SCOPE int udp6_gro_complete(struct sk_buff *skb, int nhoff, int thoff) { const struct ipv6hdr *ipv6h = (const struct ipv6hdr *)(skb->data + nhoff); struct udphdr *uh = (struct udphdr *)(skb->data + thoff); ... }
INDIRECT_CALLABLE_SCOPE int tcp6_gro_complete(struct sk_buff *skb, int nhoff, int thoff) { const struct ipv6hdr *iph = (const struct ipv6hdr *)(skb->data + nhoff); struct tcphdr *th = (struct tcphdr *)(skb->data + thoff);
Why storing in skb fields things that really could be propagated more efficiently as function parameters ?
Hi Eric, Thanks for the review!
I agree, the conditionals could be a problem and are actually not needed. The third commit in this patch series introduces an optimisation for ipv6/ipv4 using the correct {inner_}network_header. We can remove the conditionals; I thought about multiple ways to do so. First, remove the conditional in skb_gro_network_offset:
static inline int skb_gro_network_offset(const struct sk_buff *skb) { const u32 mask = NAPI_GRO_CB(skb)->encap_mark - 1; return (skb_network_offset(skb) & mask) | (skb_inner_network_offset(skb) & ~mask); }I was trying to say that we do not need all these helpers, storing state in NAPI_GRO_CB(skb), dirtying cache lines...
Ideally, the skb network/transport/... headers could be set at the last stage, in gro_complete(big_gro_skb), instead of doing this for each segment.
All the gro_receive() could be much faster by using additional parameters (nhoff, thoff)
skb_gro_offset() could be replaced by the current offset (nhoff or other name), passed as a parameter.
Here is a WIP for gro_complete() step, this looks large but this is only adding a 2nd 'offset' parameter
Prior offset (typically network offset), called p_off Old argument nhoff, (renamed thoff if that makes sense), pointing to the current offset.
You're right, it seemed to me like a broad change but it is mainly cosmetic. I'll finish your version and submit it to fix the bug.
I still believe that setting inner_network_header is a valuable change. For example, although skb_gro_network_offset is used - setting it in encapsulation protocol functions (such as ipip_gro_receive) allow us to remove conditionals from {ipv6,inet}_gro_receive gro_list loop and remove flush_id from napi_gro_cb as written in the 3rd commit. What are your thoughts about it as a separate patch?