Re: nop5-optimized USDTs WAS: Re: [PATCHv6 perf/core 09/22] uprobes/x86: Add uprobe syscall to speed up uprobe
flat view
From: Andrii Nakryiko <hidden>
Date: 2025-09-10 14:30:53
Also in:
bpf, lkml
On Fri, Sep 5, 2025 at 4:24 AM Peter Zijlstra [off-list ref] wrote:
On Thu, Sep 04, 2025 at 11:58:26PM +0200, Peter Zijlstra wrote:quoted
On Thu, Sep 04, 2025 at 11:56:17PM +0200, Peter Zijlstra wrote:quoted
Ooh, that suggests we do something like so:N/m, I need to go sleep, that doesn't work right for the 32bit nops that use lea instead of nopl. I'll see if I can come up with something more sensible.Something like this. Can someone please look very critical at this fancy insn_is_nop()?
Can't truly review that low-level instruction decoding logic (and you seem to have found an issue yourself), but superficially the cases that are claimed to be handled seem like legit no-op instructions. And the overall logic of nop handling in can_optimize and emulation seems to be intact as well. Thanks for generalizing all this! To the extent that this means anything: Acked-by: Andrii Nakryiko <andrii@kernel.org>
quoted hunk ↗ jump to hunk
--- arch/x86/include/asm/insn-eval.h | 2 + arch/x86/kernel/alternative.c | 20 +-------- arch/x86/kernel/uprobes.c | 32 ++------------ arch/x86/lib/insn-eval.c | 92 ++++++++++++++++++++++++++++++++++++++++ 4 files changed, 98 insertions(+), 48 deletions(-)
[...]