linux-kernel - Re: [tip:perf/core] perf/x86: Fix USER/KERNEL tagging of samples

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Mon, 9 Jul 2012 10:55:49 -0700
From:	Linus Torvalds <torvalds@...ux-foundation.org>
To:	Peter Zijlstra <a.p.zijlstra@...llo.nl>
Cc:	mingo@...nel.org, hpa@...or.com, eranian@...gle.com,
	linux-kernel@...r.kernel.org, fweisbec@...il.com,
	akpm@...ux-foundation.org, tglx@...utronix.de,
	linux-tip-commits@...r.kernel.org,
	Robert Richter <robert.richter@....com>
Subject: Re: [tip:perf/core] perf/x86: Fix USER/KERNEL tagging of samples

On Mon, Jul 9, 2012 at 4:23 AM, Peter Zijlstra <a.p.zijlstra@...llo.nl> wrote:
>
> How about something like the below?
>
> I've also modified perf_instruction_pointer() to account for the VM86
> and IA32 non-zero segment base cases. At least, I tried to do so, I've
> never had the 'pleasure' of poking at this segment descriptor stuff
> before.

The code base calculations look correct to me per se, although I'd put
the VM86 check into code_segment_base(). But whatever.

> Ingo didn't really like doing that though, his suggestion was to kill
> all those IPs by mapping them to a special value (~0UL or so).

Hmm. I don't care much one way or another, but I don't really see the
advantage of doing that.

The expensive part is the checking - since in all normal cases we'll
be using the standard CS and not be in vm86 mode. And you have to do
the checking whether you then map it to ~0ul or do a proper segment
base calculation. So why not just do it right? Your code seems sane
enough to me.

Now "regs" always contains a proper register state, and when the perf
code wants a linear address, perf_instruction_pointer() seems to do
the right translation. Looks good to me, although I didn't check all
the uses.

However, it is worth pointing out that sp/bp have exactly the same
segment base issue. So if you do stack tracing into user mode, you
should really do the same thing for those. And quite frankly, at that
point vm86 mode and the stack segment matters in other ways than just
the base pointer: a 16-bit stack segment acts fundamentally
differently from a 32-bit one. So at that point it may well make much
more sense to take the approach Ingo suggests, and simply not follow
stack frames at all.

Anyway, ACK for that patch from me. It may not solve all the problems,
but it looks ok.

                    Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/