Re: [patch] Fix CLONE_VM vs. TLS [Re: Is CLONE_VM really needed in gdbserver?]

["H.J. Lu" <hjl.tools@gmail.com>]

On Wed, Jan 27, 2010 at 2:12 PM, Jan Kratochvil
<jan.kratochvil@redhat.com> wrote:
> On Wed, 27 Jan 2010 01:08:26 +0100, Daniel Jacobowitz wrote:
>> On Tue, Jan 26, 2010 at 03:51:38PM -0800, H.J. Lu wrote:
>> > Hi,
>> >
>> > There is a race condition between gdbserver and ld.so on Linux/x86-64:
>> >
>> > http://www.sourceware.org/bugzilla/show_bug.cgi?id=11214
>> >
>> > Is CLONE_VM really needed? In general, CLONE_VM is a very bad
>> > idea if there is any symbol lookup in both parent and child processes.
>>
>> It is necessary because gdbserver supports uClinux.  However, on Linux
>> we might be able to get away with fork (see linux_tracefork_child in
>> gdb/linux-nat.c).
>
> Coded it as suggested.
>
> No regressions on {x86_64,x86_64-m32,i686}-fedora12-linux-gnu and on
> x86_64-fedora12-linux-gnu via gdbserver.
>
> Verified linux_supports_tracefork_flag gets still set with the patch.
>
> Verified unsetting HAVE_FORK for gdb/gdbserver/ still works the same.
>
> gdb/ already tests HAVE_FORK in config.in but it is brought in by other
> macros, therefore rather added an explicit configure.ac test for it.
>
> Have not found an easy enough uClinux disk image of some arch for qemu-*.
>
>
>

I think it is a glibc regression. I will try to fix it in glibc.


-- 
H.J.