Re: [PATCH v1 04/12] x86/mce: Remove noinstr annotation from mce_setup()

From: Borislav Petkov <bp@alien8.de>
To: Peter Zijlstra <peterz@infradead.org>
Cc: Tony Luck <tony.luck@intel.com>, X86 ML <x86@kernel.org>,
	LKML <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH v1 04/12] x86/mce: Remove noinstr annotation from mce_setup()
Date: Fri, 10 Dec 2021 15:18:33 +0100	[thread overview]
Message-ID: <YbNhuXeZm7x3bwcp@zn.tnic> (raw)
In-Reply-To: <YbHzuJmDXJ/ryWR6@hirez.programming.kicks-ass.net>

On Thu, Dec 09, 2021 at 01:16:56PM +0100, Peter Zijlstra wrote:
> Yeah, that's what it does; but *why* is that correct?

Correct, shmorect - it is the #MC handler nasty.

> I'm thinking we're well past the exception entry code and are only
> using noinstr as a means to limit the amount of code in the MCE
> handler?

Well, one of the calls to mce_gather_info() happen in #MC context. That
one calls mce_setup() and that thing calls out to

# ./arch/x86/include/asm/paravirt.h:116:        PVOP_VCALL4(cpu.cpuid, eax, ebx, ecx, edx);
        cmpq    $0, pv_ops+176(%rip)    #, pv_ops.cpu.cpuid

and you get this:

vmlinux.o: warning: objtool: pv_ops[22]: xen_cpuid
vmlinux.o: warning: objtool: pv_ops[22]: native_cpuid
vmlinux.o: warning: objtool: mce_setup()+0xa0: call to pv_ops[22]() leaves .noinstr.text section

I think this is too much and too specific text to stick in the code as a
comment.

I can stick it in the commit message if you prefer that but frankly,
seeing those instrumentation_begin/_end() sandwiches are already hints
enough in my head to read "TODO" there...

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette