Re: [RFC5 PATCH v6 00/21] ILP32 for ARM64

[Yury Norov]

On Sat, Mar 26, 2016 at 09:45:53PM +0800, Zhangjian (Bamvor) wrote:
> Hi, guys

Hi,
> 
> Does any body test the bigendian? We found lots of failures in be in
> our arm64 hardware. E.g. the signal issue.

I'm afraid, nobody yet. Thank you for work on it.

> 
> IIUC, the signal of struct in ILP32 is align with the aarch32. If so,
> we need to revert the following patch wrote by Andrew in 2014 which
> align the kernel_sigaction of ilp32 to lp64:
> Revert "[AARCH64] Add kernel_sigaction.h for AARCH64 ILP32"
> 
> And we also need to handle the uc_stack properly in kernel. After
> apply these two patches, we could fix lots of failure in bigendian.
> 
> Regards
> 
> Bamvor

Andrew sent me similar patches yesterday. See 
https://github.com/apinski-cavium/linux (branch pin1)
https://github.com/norov/glibc.git (branch new-api)

There are also some other fixes, so my fail list is like this:
[Float tests skipped but fail too]
pipeio_1                       FAIL       5
pipeio_6                       FAIL       1
abort01                        FAIL       2
clone02                        FAIL       4
execve03                       FAIL       4
fcntl17                        FAIL       4
fcntl21                        FAIL       4
kill11                         FAIL       2
mmap16                         FAIL       6
open12                         FAIL       2
rename11                       FAIL       2
rmdir02                        FAIL       2
umount2_01                     FAIL       2
umount2_02                     FAIL       2
umount2_03                     FAIL       2
utime06                        FAIL       2
mtest06                        FAIL       11

Some tests fail both on lp64 and ilp32, so it seems, it's not a
problem of ilp32 itself.

Some tests fail only when run in scenario, so I cannot reproduce it.
I suspect it's because core dump fails. 

Also, Andrew told, there's an issue in unwind, and it is caused by GCC
bug, not kernel or glibc. To reproduce, run unwind from signal
handler (trigo does it). I think, at least float tests fail due to it.
Andrew is in vacation now, so he may answer longer than usual.

See Andrew's kernel signal patch: