[Qemu-devel] IPC syscall emulation and glibc 2.3.x fix

[Qemu-devel] IPC syscall emulation and glibc 2.3.x fix

[Derrik Pates]

I'm working on adding emulation of the ipc syscall to qemu. I've managed 
to get SHM (mostly) working - at least, to the point that some X apps 
like it alright. I've run xwf (a simple GTK+-based file manager) and I 
managed to (with a lot of debug options) get Loki's port of SimCity 3000 
started (though it hangs if you try to launch a game - it appears to get 
a SIGRTMIN and become confused by that, also the colors are obviously 
ordered wrong in the image, but the speed in the menu is pretty good). I 
intend to add support for semaphores and message queues as well, once I 
get the shmctl() stuff finished - still working on a few issues with 
getting stats on SHM segments.

Also, I've found a solution to some glibc changes that caused a build 
error on line 536 of cpu-exec.c (at least on PPC). However, to 
successfully #ifdef that so that it can still work with older glibc 
versions, I need to be able to fix the original configure.in that the 
configure script is generated from. Can that be added to CVS, so that it 
can be fixed as well?

Once I have this stuff ready (as well as possibly wrapping the _sysctl 
syscall), I'll submit a patch with all my changes.

--
Derrik Pates
dpates@dsdk12.net

Re: [Qemu-devel] IPC syscall emulation and glibc 2.3.x fix

[Fabrice Bellard]

Derrik Pates wrote:
> I'm working on adding emulation of the ipc syscall to qemu. I've managed 
> to get SHM (mostly) working - at least, to the point that some X apps 
> like it alright. I've run xwf (a simple GTK+-based file manager) and I 
> managed to (with a lot of debug options) get Loki's port of SimCity 3000 
> started (though it hangs if you try to launch a game - it appears to get 
> a SIGRTMIN and become confused by that, also the colors are obviously 
> ordered wrong in the image, but the speed in the menu is pretty good). I 
> intend to add support for semaphores and message queues as well, once I 
> get the shmctl() stuff finished - still working on a few issues with 
> getting stats on SHM segments.
> 
> Also, I've found a solution to some glibc changes that caused a build 
> error on line 536 of cpu-exec.c (at least on PPC). However, to 
> successfully #ifdef that so that it can still work with older glibc 
> versions, I need to be able to fix the original configure.in that the 
> configure script is generated from. Can that be added to CVS, so that it 
> can be fixed as well?
> 
> Once I have this stuff ready (as well as possibly wrapping the _sysctl 
> syscall), I'll submit a patch with all my changes.

Thank you for the work ! Don't forget to patch 'linux-test.c' so that it 
is possible to test automatically the ipc syscalls implementation.

Fabrice.

Re: [Qemu-devel] IPC syscall emulation and glibc 2.3.x fix

[Derrik Pates]

Fabrice Bellard wrote:
> Thank you for the work ! Don't forget to patch 'linux-test.c' so that it 
> is possible to test automatically the ipc syscalls implementation.

I'll have to check that out. Will you be including the configure.in that 
is used to generate the qemu configure script in CVS, so that I can add 
the necessary configure-time check for the 'gregs' member in the 
uc->uc_mcontext structure? I definitely would like to see qemu built 
out-of-the-box on glibc 2.3.x systems, and I've seen at least one other 
message to the mailing list from someone who ran into the same problem, 
but wasn't able to find a solution.

--
Derrik Pates
dpates@dsdk12.net

Re: [Qemu-devel] IPC syscall emulation and glibc 2.3.x fix

[Daniel Jacobowitz]

On Tue, Oct 28, 2003 at 07:38:07PM -0500, Derrik Pates wrote:
> Fabrice Bellard wrote:
> >Thank you for the work ! Don't forget to patch 'linux-test.c' so that it 
> >is possible to test automatically the ipc syscalls implementation.
> 
> I'll have to check that out. Will you be including the configure.in that 
> is used to generate the qemu configure script in CVS, so that I can add 
> the necessary configure-time check for the 'gregs' member in the 
> uc->uc_mcontext structure? I definitely would like to see qemu built 
> out-of-the-box on glibc 2.3.x systems, and I've seen at least one other 
> message to the mailing list from someone who ran into the same problem, 
> but wasn't able to find a solution.

This is a bug in some versions of glibc 2.3.x, and fixed in the most
recent ones (it's still broken in Debian, I need to fix that).  i.e.
the old form is correct.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer