On Tue, Mar 4, 2014 at 12:34 PM, Yuri Gribov <[email protected]> wrote:

> > 2) Currently the top frame in the allocation stack is a function in ASan
> > runtime, so I assume you mean malloc_context_size=2.
>
> Right.
>
> > I agree it's not wise to call _UnwindBacktrace if we're interested only
> in the top frame.
> > ...
> > I don't like the idea of
> > introducing a special case for malloc_context_size=1 in our malloc/new
> interceptors
>
> So just to clarify: you are ok with special-casing
> malloc_context_size=2 but not malloc_context_size=1 ?
>

I don't like the idea of special-casing malloc_context_size=2 (fetching
both current and caller PC
in the malloc/new interceptor, and passing them both to function/macros
doing the unwinding,
so that it wouldn't do the actual unwinding if malloc_context_size=2). We
can go this path, but only
if there's a performance benefit at the end.


>
> > Do you have
> > any stats/justifications for that proposal? E.g. is ASan unwinder
> visible in
> > profile for malloc-intensive application even for small values of
> > malloc_context_size?
>
> I did run malloc-intensive code a couple of months ago with malloc
> contexts turned on and it was horribly slow.


Is there are a reason why you can't compile your code/libraries with
-fno-omit-frame-pointer and use a "fast" unwinder?
We run very malloc-intensive applications, and fast unwinder (with enabled
contexts) doesn't cause any significant slowdown.


> Unfortunately I don't
> remember whether I tried to reduce default size or just turned it off
> alltogether...
>
> -Y
>
> --
> You received this message because you are subscribed to the Google Groups
> "address-sanitizer" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to [email protected].
> For more options, visit https://groups.google.com/groups/opt_out.
>



-- 
Alexey Samsonov, MSK

-- 
You received this message because you are subscribed to the Google Groups 
"address-sanitizer" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/groups/opt_out.

Reply via email to