linux-kernel - Re: [PATCH] ARM64: perf: support dwarf unwinding in compat mode

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]

Message-ID: <CAFrcx1=QWPFeuNDc0JatiykGnRuvPyOgwwCxnW7pSAiNa9RsxA@mail.gmail.com>
Date:	Fri, 17 Jan 2014 10:00:09 +0100
From:	Jean Pihet <jean.pihet@...aro.org>
To:	Will Deacon <will.deacon@....com>
Cc:	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	"linaro-kernel@...ts.linaro.org" <linaro-kernel@...ts.linaro.org>,
	"linux-arm-kernel@...ts.infradead.org" 
	<linux-arm-kernel@...ts.infradead.org>,
	Arnaldo <acme@...stprotocols.net>,
	"patches@...aro.org" <patches@...aro.org>,
	Jiri Olsa <jolsa@...hat.com>, Ingo Molnar <mingo@...nel.org>
Subject: Re: [PATCH] ARM64: perf: support dwarf unwinding in compat mode

Hi Will,

Some more thoughts below

On 16 January 2014 14:47, Jean Pihet <jean.pihet@...aro.org> wrote:
> Will,
>
> On 16 January 2014 13:57, Will Deacon <will.deacon@....com> wrote:
>> On Thu, Jan 16, 2014 at 12:26:53PM +0000, Jean Pihet wrote:
>>> On 16 January 2014 12:56, Will Deacon <will.deacon@....com> wrote:
>>> > In your previous series, compat backtracing is actually split out into a
>>> > separate function (compat_user_backtrace), so it would be more consistent to
>>> > have a compat_user_stack_pointer macro, rather than add this check here.
> The compat_user_backtrace function is used to unwind using the frame
> pointer, it is not used to unwind using the dwarf info (which uses the
> user stack pointer).
>
>>>
>>> Do you mean this change instead?
>>
>> I don't think so...
>>
>>> diff --git a/kernel/events/internal.h b/kernel/events/internal.h
>>> index 569b2187..9b88d2e 100644
>>> --- a/kernel/events/internal.h
>>> +++ b/kernel/events/internal.h
>>> @@ -185,7 +185,8 @@ static inline bool arch_perf_have_user_stack_dump(void)
>>>         return true;
>>>  }
>>>
>>> -#define perf_user_stack_pointer(regs) user_stack_pointer(regs)
>>> +#define perf_user_stack_pointer(regs) \
>>> +       (!compat_user_mode(regs)) ? ((regs)->sp) : ((regs)->compat_sp)
>>
>> This doesn't belong in core code; compat_user_mode and the fields of regs
>> are arm64-specific.
> Right.
>
>> So I suppose you need to rework your original patch to
>> call compat_user_stack_pointer (which we already define in compat.h for
>> arm64) if compat_user_mode(regs)).
> The perf core code calls perf_user_stack_pointer(regs) to retrieve the
> stack pointer, with perf_user_stack_pointer(regs) defined as
> user_stack_pointer(regs).
> The problem is that perf is not aware of the compat mode, so every
> arch has to implement user_stack_pointer(regs) correctly.
>
> For this reason I think the first patch proposal is the right one
> unless the perf core code is redesigned to handle different ABIs. Do
> you see a better implementation?
>
>>
>> The problem there is the inconsistency with respect to the regs argument:
>>
>>   user_stack_pointer(regs)      // Returns user stack pointer for regs
>>   current_user_stack_pointer()  // Returns current user stack pointer
>>   compat_user_stack_pointer()   // Doesn't take a regs argument!
>>
>> On top of that, x86 treats those last two functions differently when current
>> is a compat task.
>>
>> So the simplest thing would be to make compat_user_stack_pointer expand to
>> user_stack_pointer(current_pt_regs()) on arm64 and merge that in with your
>> original patch fixing user_stack_pointer.

I see 2 issues in your proposal:

1) user_stack_pointer(regs) calls compat_user_stack_pointer if
compat_user_mode(regs)) and compat_user_stack_pointer expands to
user_stack_pointer. I see a circular dependency in the macros.

2) current_pt_regs() returns the current task regs although perf
passes a regs struct that had been recorded previously.

Am I missing something?

Thx,
Jean

>>
>> Will
>
> Thx!
> Jean
--
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/