[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAHk-=wixtChqkJ_NmORGMnq1pmD0x45HFgML54G2Dm-bRYAyWg@mail.gmail.com>
Date: Thu, 3 Apr 2025 12:12:44 -0700
From: Linus Torvalds <torvalds@...ux-foundation.org>
To: Josh Poimboeuf <jpoimboe@...nel.org>
Cc: Ingo Molnar <mingo@...nel.org>, linux-kernel@...r.kernel.org,
Peter Zijlstra <a.p.zijlstra@...llo.nl>, Thomas Gleixner <tglx@...utronix.de>,
Borislav Petkov <bp@...en8.de>, Andrew Morton <akpm@...ux-foundation.org>
Subject: Re: [GIT PULL] objtool fixes
On Thu, 3 Apr 2025 at 11:24, Josh Poimboeuf <jpoimboe@...nel.org> wrote:
>
> There's also smap_save() / smap_restore(). For the latter we'd need to
> look for alternatives with "push reg; popf", which is definitely not
> SMAP-specific. So we'd need to start reading feature bits again, which
> ends up even worse than what we had before.
Now, I agree that smap_save / smap_restore might be worthy of an
annotation, to show that "this is just a push/pop, but the intent is
to save AC".
Would that be ok?
Because I want to have *less* noise for clac/stac, not more.
The reason I noticed this immediately is that clac/stac are literally
one of my "top 10" things I ever look at.
The optimized user copies (so the masked_user_access_begin() ->
unsafe_get_user() -> user_read_access_end() path) and the dcache are
the two places I look at almost constantly, because it turns out that
those two things are the ones that show up the most on my machines.
The dcache doesn't do the clac/stac part (well, 'strncpy_from_user()'
does, but while that's related to, it precedes the actual pathname
lookup), so it's not like clac/stac is *all* I look at, but it really
is important to me.
Linus
Powered by blists - more mailing lists