Re: [PATCH] x86: make PTRACE_GETREGSET return 32-bit regs if 64-bit process entered kernel with int 80

[Denys Vlasenko <dvlasenk@redhat.com>]

On 02/14/2013 09:55 PM, Cyrill Gorcunov wrote:
> On Thu, Feb 14, 2013 at 11:21:12AM -0800, H. Peter Anvin wrote:
>> On 02/14/2013 11:18 AM, Oleg Nesterov wrote:
>>> On 02/14, H. Peter Anvin wrote:
>>>>
>>>> On 02/14/2013 07:00 AM, Oleg Nesterov wrote:
>>>>> On 02/14, Denys Vlasenko wrote:
>>>>>>
>>>>>> Determining personality of a ptraced process is a murky area.
>>>>>> On x86, for years strace was looking at segment selectors,
>>>>>> which is conceptually wrong: see, for example,
>>>>>> https://lkml.org/lkml/2012/1/18/320
>>>>>>
>>>>
>>>> One proposal that keeps being on the table is to export a regset with
>>>> metadatam, including process mode at launch (i386, x86-64, x32).
>>>
>>> Yes... but if this metadata includes TS_COMPAT-is-set, then strace should
>>> do PTRACE_GETREGSET(REGSET_META) + PTRACE_GETREGSET(REGSET_GENERAL) every
>>> time. Or REGSET_META should include META+GENERAL.
>>>
>>> IOW, it is not clear to me what this "meta" should actually report.
>>
>> That is one of the things that needs to be nailed down.  In particular,
>> what are the things people need.
> 
> Indeed, having some "official" way for compat bit retrieval would be
> a great thing for us (c/r camp) since we've just met the same problem.
> And at moment I sticked for the same trick as gdb does (cs/ds test).
> 
> But, guys, if only I'm not missing something completely obvious,
> can't we simply provide task-compat bit to userspace in say /proc/pid/stat
> or something? Then the strace/gdb would be able to always know if
> the tracee is actually in compat mode. Or I miss something fundamental
> here?

strace needs to get that data on every syscall entry in the traced process.
Doing open/read/close on every syscall entry is going to slow it down a lot.

-- 
vda