Re: eglibc / FP / Movidis
On Wed, May 5, 2010 at 11:38 AM, Shane McDonald
<mcdonald.shane@gmail.com> wrote:
> On Wed, May 5, 2010 at 8:04 AM, Clint Adams <schizo@debian.org> wrote:
>> eglibc is failing to build on both gabrielli and corelli.
>>
>> One of the issues is floating-point math (invalid operations,
>> divide by zero, etc.). Does anyone have any insight?
>
> There is currently a thread [1] on the linux-mips mailing list regarding a bug
> in the in-kernel floating point emulator. A proposed patch [2] is available,
> but there are concerns that it may cause problems in an SMP system.
> This bug was stumbled across when the "yorick" package was buildd'ed
> on lucatelli.
>
> I'm guessing that this bug is the same one that's causing eglibc to fail
> its regression tests. I've got a non-FPU machine running a kernel
> with the proposed patch; I can try building eglibc and see if it's
> the same problem. It's a slow machine, and I'll have to wait until
> I get home to start it up, and it looks like eglibc takes a long time
> to compile, so I won't have an immediate answer.
>
> Shane McDonald
>
> [1] http://www.linux-mips.org/archives/linux-mips/2010-05/msg00005.html
> [2] http://www.linux-mips.org/archives/linux-mips/2010-05/msg00048.html
I tried building eglibc_2.10.2-7 on my FPU-less mips machine with
a kernel patched for the identified FPU emulator bug. Although I still
see some unexpected failures, the floating point failures you reported
seem to be resolved with this kernel patch.
Encountered regressions that don't match expected failures:
tst-cancel4.out, Error 1
tst-cancel5.out, Error 1
tst-mmap-eofsync.out, Error 1
tst-mmap-fflushsync.out, Error 1
My system has an MSP7120 processor and is running a 2.6.28
kernel patched for the FPU bug I mentioned. It is running
etch, but I built eglibc using pbuilder in a sid chroot.
Shane
Reply to: