Re: [PATCH net-next v2] net: skbuff: optimization of net_zcopy_get() call in pskb_carve helpers

在 2026/7/21 17:39, Paolo Abeni 写道:

On 7/8/26 7:54 AM, Yun Lu wrote:

quoted

From: Yun Lu <redacted>

Commit 98d0912e9f84 ("net: skbuff: fix missing zerocopy reference in
pskb_carve helpers") introduced two calls of net_zcopy_get(skb_zcopy(skb)).
In fact, skb_zcopy() has already been executed once before. When calling
net_zcopy_get(), skb_zcopy() always returns skb_uarg(skb), which results
in adding some unnecessary instructions in skb_zcopy.

I'm quite surprised the compiler can't generate the same binary with the
old code. blot-o-meter seams to agree with me.

Thanks for checking with bloat-o-meter — you’re right. I did the measurement

I should have done earlier: I compiled net/core/skbuff.o with and without

this patch (gcc 9.3.0, x86-64 defconfig-based build) and diffed the

disassembly of pskb_expand_head() and pskb_carve() (into which the two

static helpers are inlined). The generated code is byte-for-byte identical,

so the compiler does fold skb_zcopy() into skb_uarg(skb) on its own.

I think that the consistency argument is enough here to justify the
patch, but it would have been better to double the above guess with some
actual measurement.

/P

This patch is only a code-consistency cleanup, with no instruction-level 
optimization.

Thank you for pointing that out — I’ll be more rigorous next time.