[Qemu-devel] sparc-linux-usermode broken after latest pull

[Qemu-devel] sparc-linux-usermode broken after latest pull

[David Munday]

Hello,
Our lab just did a pull from the qemu repository and now we are unable to run a 32-bit sparc target on a 64-bit host. This functionality was working before we did the pull. The error message we get is:mmap: No such device or address

And then the process dies. We are not sure why this is occuring but have verified against our earlier qemu copy that the behavior did not exist before our pull on Sunday.

We appreciate any insights anyone might have.

Thanks,
David

Re: [Qemu-devel] sparc-linux-usermode broken after latest pull

[Blue Swirl]

On 4/15/10, David Munday <cromom@soe.ucsc.edu> wrote:
> Hello,
>  Our lab just did a pull from the qemu repository and now we are unable to run a 32-bit sparc target on a 64-bit host. This functionality was working before we did the pull. The error message we get is:mmap: No such device or address
>
>  And then the process dies. We are not sure why this is occuring but have verified against our earlier qemu copy that the behavior did not exist before our pull on Sunday.
>
>  We appreciate any insights anyone might have.

I can't replicate this, sparc-linux-user, sparc32plus-linux-user and
sparc64-linux-user still work like before on amd64 Linux host.

What is the host and which program? Could you bisect the problem,
there haven't been that many commits in last few days?

Re: [Qemu-devel] sparc-linux-usermode broken after latest pull

[Gabriel Southern]

On Thu, Apr 15, 2010 at 9:33 AM, Blue Swirl <blauwirbel@gmail.com> wrote:
> On 4/15/10, David Munday <cromom@soe.ucsc.edu> wrote:
>> Hello,
>>  Our lab just did a pull from the qemu repository and now we are unable to run a 32-bit sparc target on a 64-bit host. This functionality was working before we did the pull. The error message we get is:mmap: No such device or address
>>
>>  And then the process dies. We are not sure why this is occuring but have verified against our earlier qemu copy that the behavior did not exist before our pull on Sunday.
>>
>>  We appreciate any insights anyone might have.
>
> I can't replicate this, sparc-linux-user, sparc32plus-linux-user and
> sparc64-linux-user still work like before on amd64 Linux host.
>
> What is the host and which program? Could you bisect the problem,
> there haven't been that many commits in last few days?
>

I have also noticed that sparc-linux-user does not compile on an AMD64
host (CentOS 5.4):

cc1: warnings being treated as errors
/mada/users/gsouther/projs/qemu/linux-user/mmap.c: In function ‘mmap_find_vma’:
/mada/users/gsouther/projs/qemu/linux-user/mmap.c:367: warning:
comparison is always false due to limited range of data type
make[1]: *** [mmap.o] Error 1
gmake: *** [subdir-sparc-linux-user] Error 2

The compilation warning can be fixed by removing the ul specifier from
-1ul, but I'm wondering if you had noticed the same problem.

-Gabriel

Re: [Qemu-devel] sparc-linux-usermode broken after latest pull

[Blue Swirl]

On 4/15/10, Gabriel Southern <gabriel.southern@gmail.com> wrote:
> On Thu, Apr 15, 2010 at 9:33 AM, Blue Swirl <blauwirbel@gmail.com> wrote:
>  > On 4/15/10, David Munday <cromom@soe.ucsc.edu> wrote:
>  >> Hello,
>  >>  Our lab just did a pull from the qemu repository and now we are unable to run a 32-bit sparc target on a 64-bit host. This functionality was working before we did the pull. The error message we get is:mmap: No such device or address
>  >>
>  >>  And then the process dies. We are not sure why this is occuring but have verified against our earlier qemu copy that the behavior did not exist before our pull on Sunday.
>  >>
>  >>  We appreciate any insights anyone might have.
>  >
>  > I can't replicate this, sparc-linux-user, sparc32plus-linux-user and
>  > sparc64-linux-user still work like before on amd64 Linux host.
>  >
>  > What is the host and which program? Could you bisect the problem,
>  > there haven't been that many commits in last few days?
>  >
>
>
> I have also noticed that sparc-linux-user does not compile on an AMD64
>  host (CentOS 5.4):
>
>  cc1: warnings being treated as errors
>  /mada/users/gsouther/projs/qemu/linux-user/mmap.c: In function ‘mmap_find_vma’:
>  /mada/users/gsouther/projs/qemu/linux-user/mmap.c:367: warning:
>  comparison is always false due to limited range of data type
>  make[1]: *** [mmap.o] Error 1
>  gmake: *** [subdir-sparc-linux-user] Error 2
>
>  The compilation warning can be fixed by removing the ul specifier from
>  -1ul, but I'm wondering if you had noticed the same problem.

Does not happen here for some reason. I think removing the ul would be
correct since that is what the code above uses, but even better would
be to use (abi_ulong)-1 consistently.

RE: [Qemu-devel] sparc-linux-usermode broken after latest pull

[David Munday]

Hi,
I will bisect the code and reply back with the last working patch shortly. In the meantime, I'm working on a different machine where the latest pull is working, but am having trouble with pthreads on a sparc32 target. It seems that forking is supported, but when running a pthread based program on sparc32 user mode, when I run with gdb, there is a core dump in static_code_gen_buffer(). 
I've put together a small test suite that exposes this problem. The tar at his link contains the following: http://users.soe.ucsc.edu/~cromom/mt_test.tar.gz

fork_test.out : test program with forking (works with qemu user mode)
ThreadSafeFIFO.out: test program with pthreads (has core dump with qemu user mode)

>From gdb I can see that when running ThreadSafeFIFO, do_fork is called twice, but there is only one call to pthread_create().

Thanks for the help,
David



-----Original Message-----
From: Blue Swirl [mailto:blauwirbel@gmail.com] 
Sent: Thursday, April 15, 2010 12:16 PM
To: Gabriel Southern
Cc: David Munday; qemu-devel@nongnu.org
Subject: Re: [Qemu-devel] sparc-linux-usermode broken after latest pull

On 4/15/10, Gabriel Southern <gabriel.southern@gmail.com> wrote:
> On Thu, Apr 15, 2010 at 9:33 AM, Blue Swirl <blauwirbel@gmail.com> wrote:
>  > On 4/15/10, David Munday <cromom@soe.ucsc.edu> wrote:
>  >> Hello,
>  >>  Our lab just did a pull from the qemu repository and now we are unable to run a 32-bit sparc target on a 64-bit host. This functionality was working before we did the pull. The error message we get is:mmap: No such device or address
>  >>
>  >>  And then the process dies. We are not sure why this is occuring but have verified against our earlier qemu copy that the behavior did not exist before our pull on Sunday.
>  >>
>  >>  We appreciate any insights anyone might have.
>  >
>  > I can't replicate this, sparc-linux-user, sparc32plus-linux-user and
>  > sparc64-linux-user still work like before on amd64 Linux host.
>  >
>  > What is the host and which program? Could you bisect the problem,
>  > there haven't been that many commits in last few days?
>  >
>
>
> I have also noticed that sparc-linux-user does not compile on an AMD64
>  host (CentOS 5.4):
>
>  cc1: warnings being treated as errors
>  /mada/users/gsouther/projs/qemu/linux-user/mmap.c: In function ‘mmap_find_vma’:
>  /mada/users/gsouther/projs/qemu/linux-user/mmap.c:367: warning:
>  comparison is always false due to limited range of data type
>  make[1]: *** [mmap.o] Error 1
>  gmake: *** [subdir-sparc-linux-user] Error 2
>
>  The compilation warning can be fixed by removing the ul specifier from
>  -1ul, but I'm wondering if you had noticed the same problem.

Does not happen here for some reason. I think removing the ul would be
correct since that is what the code above uses, but even better would
be to use (abi_ulong)-1 consistently.