From: Matheus Castanho <msc@linux.ibm.com>
To: Adhemerval Zanella <adhemerval.zanella@linaro.org>,
GNU C Library <libc-alpha@sourceware.org>
Subject: Re: semtimedop, powerpc, time64 and older kernels
Date: Wed, 30 Sep 2020 17:45:19 -0300 [thread overview]
Message-ID: <3e02f4be-92af-5b81-b207-e30308a68b80@linux.ibm.com> (raw)
In-Reply-To: <7b907046-152a-d057-c01d-2d0a80f83931@linaro.org>
On 9/30/20 4:12 PM, Adhemerval Zanella wrote:
>
>
> On 30/09/2020 15:29, Matheus Castanho wrote:
>> Also, looks like my email client messed up the diff *sigh*. I'm sending
>> a proper patch attached this time.
>>
>> --
>> Matheus Castanho
>>
>
>> From 1c0a497a3f986bc6980581c9eab482ccf7bb190f Mon Sep 17 00:00:00 2001
>> From: Matheus Castanho <msc@linux.ibm.com>
>> Date: Wed, 30 Sep 2020 14:22:18 -0300
>> Subject: [PATCH] sysvipc: Fix semtimedop for Linux < 5.1
>>
>> Kernels older than 5.1 will fail with ENOSYS when calling
>> semtimedop_time64 syscall in __semtimedop_time64. Just like for
>> !__ASSUME_TIME64_SYSCALLS, we should fallback to using the old mechanism
>> in such cases.
>> ---
>> sysdeps/unix/sysv/linux/semtimedop.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/sysdeps/unix/sysv/linux/semtimedop.c b/sysdeps/unix/sysv/linux/semtimedop.c
>> index a9ad922ee2..510fea1852 100644
>> --- a/sysdeps/unix/sysv/linux/semtimedop.c
>> +++ b/sysdeps/unix/sysv/linux/semtimedop.c
>> @@ -32,7 +32,7 @@ __semtimedop64 (int semid, struct sembuf *sops, size_t nsops,
>> int r = INLINE_SYSCALL_CALL (semtimedop_time64, semid, sops, nsops,
>> timeout);
>>
>> -#ifndef __ASSUME_TIME64_SYSCALLS
>> +#if !(defined __ASSUME_TIME64_SYSCALLS) || __LINUX_KERNEL_VERSION < 0x050100
>> if (r == 0 || errno != ENOSYS)
>> return r;
>>
>> --
>> 2.26.2
>
> Thanks for catching it and although it fixes the regression, we have
> kernel-features.h exactly to avoid using __LINUX_KERNEL_VERSION through the
> implementations. Also this is sub-optimal since it forces semtimeopd issues
> __NR_semtimeop and then __NR_ipc on powerpc64 and we have
> __ASSUME_DIRECT_SYSVIPC_SYSCALLS exactly to avoid this strategy of handling
> ENOSYS for newer syscalls and thus slowing it down the implementation on
> older kernels (--enable-kernel exists exactly to get rid of this older kernel
> support).
>
Thanks for the explanation.
> I forgot that powerpc64 and s390x used the older multiplexed __NR_ipc and
> kernel v5.1 decided to add proper __NR_semtimedop (and it was in fact handled
> by 720e9541f5d919). I think a better fix is the one below, since it:
>
> 1. Issues __NR_semtimeop_time64 iff it is defined (32-bit architectures).
> 2. Issues __NR_semtimeop otherwise iff glibc is configured for a kernel that
> supports it (for powerpc64 it will only for --enable-kernel=5.1).
> Otherwise it will use only 3.
> 3. Issues __NR_ipc with IPCOP_semtimedop.
>
> For powerpc64 it will issue either __NR_ipc (default) or __NR_semtimeop
> (--enable-kernel=5.1), while for powerpc it will use either
> __NR_semtimeop_time64 and fallback to __NR_ipc or just issue
> __NR_semtimeop_time64.
Sounds good to me. That is indeed a better solution. Thanks!
>
> I am running some regressions before commit it.
>
> ---
>
> diff --git a/sysdeps/unix/sysv/linux/semtimedop.c b/sysdeps/unix/sysv/linux/semtimedop.c
> index a9ad922ee2..29647f8ccd 100644
> --- a/sysdeps/unix/sysv/linux/semtimedop.c
> +++ b/sysdeps/unix/sysv/linux/semtimedop.c
> @@ -26,11 +26,15 @@ int
> __semtimedop64 (int semid, struct sembuf *sops, size_t nsops,
> const struct __timespec64 *timeout)
> {
> -#ifndef __NR_semtimedop_time64
> -# define __NR_semtimedop_time64 __NR_semtimedop
> + int r;
> +#if defined __NR_semtimedop_time64
> + r = INLINE_SYSCALL_CALL (semtimedop_time64, semid, sops, nsops, timeout);
> +#elif defined __ASSUME_DIRECT_SYSVIPC_SYSCALLS && defined __NR_semtimedop
> + r = INLINE_SYSCALL_CALL (semtimedop, semid, sops, nsops, timeout);
> +#else
> + r = INLINE_SYSCALL_CALL (ipc, IPCOP_semtimedop, semid,
> + SEMTIMEDOP_IPC_ARGS (nsops, sops, timeout));
> #endif
> - int r = INLINE_SYSCALL_CALL (semtimedop_time64, semid, sops, nsops,
> - timeout);
>
> #ifndef __ASSUME_TIME64_SYSCALLS
> if (r == 0 || errno != ENOSYS)
>
The change looks good and fixes the issue. Tested on ppc64le.
Reviewed-by: Matheus Castanho <msc@linux.ibm.com>
--
Matheus Castanho
prev parent reply other threads:[~2020-09-30 20:45 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-09-30 18:01 Matheus Castanho
2020-09-30 18:29 ` Matheus Castanho
2020-09-30 18:45 ` Andreas Schwab
2020-09-30 19:12 ` Adhemerval Zanella
2020-09-30 20:45 ` Matheus Castanho [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=3e02f4be-92af-5b81-b207-e30308a68b80@linux.ibm.com \
--to=msc@linux.ibm.com \
--cc=adhemerval.zanella@linaro.org \
--cc=libc-alpha@sourceware.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).