Question regarding userspace tracing memcpy in glibc

[Holger Hans Peter Freyther]

On 12/10/2010 06:49 PM, Frank Ch. Eigler wrote:
> "Ariel Shaqed (Scolnicov)" <ariels@correlix.com> writes:
> 
>> [...]
>> I believe the problem is much worse.  Usually no functions are called
>> for memcpy (GCC implements them as intrinsics).  [...]
> 
> You're right, and moreover, gcc doesn't emit any metadata.  I'll ask
> our compiler folks to investigate the possibility of producing dwarf
> data for a pretend-inlined-memcpy.

Hi again,

if we leave out the case where the compiler knows the amount to copy, what
would be the way forward to be able to intercept ASM stubs? Get hand written
DWARF2 symbols into the glibc code? Do you normally discuss these things in
bug reports or the mailinglist?

regards
	holger