[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <7D77A452-E61E-4B8B-B49C-949E1C8E257C@vmware.com>
Date: Mon, 16 Oct 2023 20:35:24 +0000
From: Nadav Amit <namit@...are.com>
To: Linus Torvalds <torvalds@...ux-foundation.org>,
Uros Bizjak <ubizjak@...il.com>
CC: the arch/x86 maintainers <x86@...nel.org>,
Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
Andy Lutomirski <luto@...nel.org>,
Brian Gerst <brgerst@...il.com>,
Denys Vlasenko <dvlasenk@...hat.com>,
"H . Peter Anvin" <hpa@...or.com>,
Peter Zijlstra <peterz@...radead.org>,
Thomas Gleixner <tglx@...utronix.de>,
Josh Poimboeuf <jpoimboe@...hat.com>,
Nick Desaulniers <ndesaulniers@...gle.com>
Subject: Re: [PATCH v2 -tip] x86/percpu: Use C for arch_raw_cpu_ptr()
> On Oct 16, 2023, at 10:24 PM, Linus Torvalds <torvalds@...ux-foundation.org> wrote:
>
> So there's some serious corruption there, but from the oops itself I
> can't tell the source. I guess if we get 'current' wrong anywhere, all
> bets are off.
I don’t think it means that it the aliasing does not work; I think it all means that
it actually works *too well*.
I have encountered several such issues before [1], and while some have been fixed,
some have not (I looked at switch_fpu_finish()), and might under the right/wrong
circumstances use the wrongly-“cached” current. Moreover, perhaps new problems
have been added since my old patch.
Perhaps the whack-a-mole approach that I took in [1] is wrong. Instead, perhaps it
is better to use an uncached version of current when compiling
arch/x86/kernel/process_64.c and arch/x86/kernel/process_32.c , so they would not
use the const alias, but instead use the original (non-const) one. Some macro
magic and an additional "-D” build flag can do that.
But first, there is a need to confirm that’s actually the problem. I’ll try to do
it tomorrow.
Regards,
Nadav
[1] https://lore.kernel.org/all/20190823224424.15296-5-namit@vmware.com/
Powered by blists - more mailing lists