[Qemu-devel] User mode emulation's clone does not seem to work with glibc >= 2.4's fork

[Qemu-devel] User mode emulation's clone does not seem to work with glibc >= 2.4's fork

[Lubomir Kundrak]

I tried to run SH4 binaries linked with glibc-2.8 with qemu-sh4. fork()
system call was always returning with "Invalid argument".

Post glibc 2.3 fork() on Linux seems to call clone():

glibc-2.4/nptl/sysdeps/unix/sysv/linux/sh/fork.c:
 24 /* TLS pointer argument is passed as the 5-th argument.  */
 25 #define ARCH_FORK() \
 26   INLINE_SYSCALL (clone, 5,                                                   \
 27                   CLONE_CHILD_SETTID | CLONE_CHILD_CLEARTID | SIGCHLD, 0,     \
 28                   NULL, &THREAD_SELF->tid, NULL)

qemu/linux-user/syscall.c:

2713 int do_fork(CPUState *env, unsigned int flags, abi_ulong newsp)
...
2720     if (flags & CLONE_VM) {
...
2794     } else {
2795         /* if no CLONE_VM, we consider it is a fork */
2796         if ((flags & ~CSIGNAL) != 0)
2797             return -EINVAL;
2798         ret = fork();
2799     }
2800     return ret;

Here CLONE_CHILD_SETTID and CLONE_CHILD_CLEARTID being set cause the
~CSIGNAL conditional to fail. If the test is omited, some assertion fail
in glibc's fork() implementation is triggered.

-- 
Lubomir Kundrak (Red Hat Security Response Team)

Re: [Qemu-devel] User mode emulation's clone does not seem to work with glibc >= 2.4's fork

[Paul Brook]

On Tuesday 29 April 2008, Lubomir Kundrak wrote:
> I tried to run SH4 binaries linked with glibc-2.8 with qemu-sh4. fork()
> system call was always returning with "Invalid argument".

qemu doesn't currently support NPTL. I'm working on it.

Paul

[Qemu-devel] Re: User mode emulation's clone does not seem to work with glibc >= 2.4's fork

[michael]

Hi,

Paul Brook wrote:
> On Tuesday 29 April 2008, Lubomir Kundrak wrote:
>> I tried to run SH4 binaries linked with glibc-2.8 with qemu-sh4. fork()
>> system call was always returning with "Invalid argument".
> 
> qemu doesn't currently support NPTL. I'm working on it.
> 
> Paul
> 
> 
>
I'm working on sh4 to provide an NPTL support. I just force the
NPTL support of sh4 in the configuration and add the cpu_set_tls
using the env->gbr. I have sh: 
../nptl/sysdeps/unix/sysv/linux/sh/../fork.c:138: __libc_fork: Assertion 
`(self->tid) != ppid this assertion in tforhe libc code. The 
child_tidptr is nil, so I can't set the tid for the child. I don't want 
to fix this issue but just understand what happen.

Regards Michael

[Qemu-devel] Re: User mode emulation's clone does not seem to work with glibc >= 2.4's fork

[michael]

Hi,

Paul Brook wrote:
> On Tuesday 29 April 2008, Lubomir Kundrak wrote:
>> I tried to run SH4 binaries linked with glibc-2.8 with qemu-sh4. fork()
>> system call was always returning with "Invalid argument".
> 
> qemu doesn't currently support NPTL. I'm working on it.
> 
> Paul
> 
> 
>
I'm working on sh4 to provide an NPTL support. I just force the
NPTL support of sh4 in the configuration and add the cpu_set_tls
using the env->gbr. I have sh:
../nptl/sysdeps/unix/sysv/linux/sh/../fork.c:138: __libc_fork: Assertion
`(self->tid) != ppid this assertion in tforhe libc code. The
child_tidptr is nil, so I can't set the tid for the child. I don't want
to fix this issue but just understand what happen.

Regards Michael

[Qemu-devel] Re: User mode emulation's clone does not seem to work with glibc >= 2.4's fork

[michael]

Hi,

Paul Brook wrote:
> On Tuesday 29 April 2008, Lubomir Kundrak wrote:
>> I tried to run SH4 binaries linked with glibc-2.8 with qemu-sh4. fork()
>> system call was always returning with "Invalid argument".
> 
> qemu doesn't currently support NPTL. I'm working on it.
> 
> Paul
> 
> 
>
I'm working on sh4 to provide an NPTL support. I just force the
NPTL support of sh4 in the configuration and add the cpu_set_tls
using the env->gbr. I have sh:
../nptl/sysdeps/unix/sysv/linux/sh/../fork.c:138: __libc_fork: Assertion
`(self->tid) != ppid this assertion in tforhe libc code. The
child_tidptr is nil, so I can't set the tid for the child. I change
the code of the fork like this:

#if !defined(TARGET_SH4)
         ret = get_errno(do_fork(cpu_env, arg1, arg2, arg3, arg4, arg5));
#else
         ret = get_errno(do_fork(cpu_env, arg1, arg2, arg3, arg5, arg4));
#endif

An know seem to work the sh4 target. I think that is not the only 
change. Is it right?

Regards Michael