[stack] segment only corefiles, and [vdso]

[stack] segment only corefiles, and [vdso]

[Phil Muldoon]

When I implemented that experimental feature, I only included the 
[stack] segment. However on IA32, there is always a visit to the [vdso] 
segment which I did not include. Should --stackonly include the [vdso] 
segment, and implement a do what I mean, not what I say approach? How 
big can the [vdso] segment get? Is it a fixed size?

Regards

Re: [stack] segment only corefiles, and [vdso]

[Mark Wielaard]

Hi Phil,

On Tue, 2008-02-26 at 15:19 +0000, Phil Muldoon wrote:
> When I implemented that experimental feature, I only included the 
> [stack] segment. However on IA32, there is always a visit to the [vdso] 
> segment which I did not include.

What is the visit from/for?

Cheers,

Mark

Re: [stack] segment only corefiles, and [vdso]

[Phil Muldoon]

Mark Wielaard wrote:
> Hi Phil,
>
> On Tue, 2008-02-26 at 15:19 +0000, Phil Muldoon wrote:
>   
>> When I implemented that experimental feature, I only included the 
>> [stack] segment. However on IA32, there is always a visit to the [vdso] 
>> segment which I did not include.
>>     
>
> What is the visit from/for?
>   

Well it appears to happen on every IA32 process. Why/to ... not terribly 
sure other than every process/corefile seems to include a visit there. 
just fstack a 32 bit process.

Regards

Phil
> Cheers,
>
> Mark
>
>   

Re: [stack] segment only corefiles, and [vdso]

[Mark Wielaard]

Hi Phil,

On Tue, 2008-02-26 at 16:00 +0000, Phil Muldoon wrote:
> Mark Wielaard wrote:
> > On Tue, 2008-02-26 at 15:19 +0000, Phil Muldoon wrote:
> >> When I implemented that experimental feature, I only included the 
> >> [stack] segment. However on IA32, there is always a visit to the [vdso] 
> >> segment which I did not include.
> >
> > What is the visit from/for? 
> 
> Well it appears to happen on every IA32 process. Why/to ... not terribly 
> sure other than every process/corefile seems to include a visit there. 
> just fstack a 32 bit process.

I assume this is when you fstack a process that was stuck inside a
system call and it tries to do a symbol lookup? I don't know why that
would be x86 specific. But I assume you must include [vdso] because it
isn't available any other way (and even if it was on the system, then
you might be fstacking a core file that comes from a different system
that has a different vsdo layout).

Cheers,

Mark

Re: [stack] segment only corefiles, and [vdso]

[Phil Muldoon]

Mark Wielaard wrote:
> Hi Phil,
>
>   
> I assume this is when you fstack a process that was stuck inside a
> system call and it tries to do a symbol lookup? I don't know why that
> would be x86 specific. 

It's not a very scientific statement on my behalf, just that I only 
observe it on IA32 not on x8664. Sorry should have qualified that up front.

> But I assume you must include [vdso] because it
> isn't available any other way (and even if it was on the system, then
> you might be fstacking a core file that comes from a different system
> that has a different vsdo layout).
>
>   
Right so --stackonly should include [vdso] then to make the corefile 
back trace workable. Strictly speaking we're breaking the contract 
there, but oh well.

I think I will replace that (--stack only) option with a regex option 
dump instead. For example:

--regex="(heap\|stack\|vdso)"

or something.

Phil

> Cheers,
>
> Mark
>
>