From: Adhemerval Zanella Netto <adhemerval.zanella@linaro.org>
To: Mateusz Guzik <mjguzik@gmail.com>
Cc: libc-alpha@sourceware.org,
Linus Torvalds <torvalds@linux-foundation.org>
Subject: Re: [PATCH] io: Do not implement fstat with fstatat
Date: Tue, 5 Sep 2023 17:56:51 -0300 [thread overview]
Message-ID: <5b4cb21e-5bb0-5015-4c4b-f4fb4b9d2892@linaro.org> (raw)
In-Reply-To: <CAGudoHFpiedB7uUUB1C7J3TxxxeMjB8XxFHixDe48jBy4ZZ0Fg@mail.gmail.com>
On 05/09/23 17:48, Mateusz Guzik wrote:
> On 9/5/23, Adhemerval Zanella <adhemerval.zanella@linaro.org> wrote:
>> AT_EMPTY_PATH is a requirement to implement fstat over fstatat,
>> however it does not prevent the kernel to read the path argument.
>> It is not an issue, but it has performance implications on x86_64
>> with SMAP enable, since each newfstatat syscall generates a pagefault
>> even with an empty path.
>>
>
> This does not generate page faults in the usual case.
>
> How about:
>
> On x86-64 with SMAP-capable CPUs (vast majority today) the kernel is
> forced to perform expensive user memory access. After that regular
> lookup is performed which adds even more overhead.
Alright, I have changed to:
AT_EMPTY_PATH is a requirement to implement fstat over fstatat,
however it does not prevent the kernel to read the path argument.
It is not an issue, but on x86-64 with SMAP-capable CPUs the kernel is
forced to perform expensive user memory access. After that regular
lookup is performed which adds even more overhead.
>
> I have no commentso n the patch itself.
>
> Thanks for quick turn around.
>
>
>> Instead, issue the fstat syscall directly on LFS fstat implementation
>> (32 bit architectures will still continue to use statx, which is
>> required to have 64 bit time_t support). it should be even a
>> small performance gain on non x86_64, since there is no need
>> to handle the path argument.
>>
>> Checked on x86_64-linux-gnu.
>> ---
>> sysdeps/unix/sysv/linux/fstat64.c | 37 +++++++++++++++++++++++--
>> sysdeps/unix/sysv/linux/fstatat64.c | 12 ++------
>> sysdeps/unix/sysv/linux/internal-stat.h | 31 +++++++++++++++++++++
>> 3 files changed, 68 insertions(+), 12 deletions(-)
>> create mode 100644 sysdeps/unix/sysv/linux/internal-stat.h
>>
>> diff --git a/sysdeps/unix/sysv/linux/fstat64.c
>> b/sysdeps/unix/sysv/linux/fstat64.c
>> index 124384e57f..a7f4935a01 100644
>> --- a/sysdeps/unix/sysv/linux/fstat64.c
>> +++ b/sysdeps/unix/sysv/linux/fstat64.c
>> @@ -19,20 +19,53 @@
>> #define __fstat __redirect___fstat
>> #define fstat __redirect_fstat
>> #include <sys/stat.h>
>> +#undef __fstat
>> +#undef fstat
>> #include <fcntl.h>
>> -#include <kernel_stat.h>
>> -#include <stat_t64_cp.h>
>> +#include <internal-stat.h>
>> #include <errno.h>
>>
>> int
>> __fstat64_time64 (int fd, struct __stat64_t64 *buf)
>> {
>> +#if !FSTATAT_USE_STATX
>> +# if XSTAT_IS_XSTAT64
>> +# ifdef __NR_newfstatat
>> + /* 64-bit kABI, e.g. aarch64, ia64, powerpc64*, s390x, riscv64, and
>> + x86_64. */
>> + return INLINE_SYSCALL_CALL (fstat, fd, buf);
>> +# elif defined __NR_fstatat64
>> +# if STAT64_IS_KERNEL_STAT64
>> + /* 64-bit kABI outlier, e.g. alpha */
>> + return INLINE_SYSCALL_CALL (fstat64, fd, buf);
>> +# else
>> + /* 64-bit kABI outlier, e.g. sparc64. */
>> + struct kernel_stat64 kst64;
>> + int r = INLINE_SYSCALL_CALL (fstat64, fd, &kst64);
>> + if (r == 0)
>> + __cp_stat64_kstat64 (buf, &kst64);
>> + return r;
>> +# endif /* STAT64_IS_KERNEL_STAT64 */
>> +# endif
>> +# else /* XSTAT_IS_XSTAT64 */
>> + /* 64-bit kabi outlier, e.g. mips64 and mips64-n32. */
>> + struct kernel_stat kst;
>> + int r = INLINE_SYSCALL_CALL (fstat, fd, &kst);
>> + if (r == 0)
>> + __cp_kstat_stat64_t64 (&kst, buf);
>> + return r;
>> +# endif
>> +#else /* !FSTATAT_USE_STATX */
>> + /* All kABIs with non-LFS support and with old 32-bit time_t support
>> + e.g. arm, csky, i386, hppa, m68k, microblaze, nios2, sh, powerpc32,
>> + and sparc32. */
>> if (fd < 0)
>> {
>> __set_errno (EBADF);
>> return -1;
>> }
>> return __fstatat64_time64 (fd, "", buf, AT_EMPTY_PATH);
>> +#endif
>> }
>> #if __TIMESIZE != 64
>> hidden_def (__fstat64_time64)
>> diff --git a/sysdeps/unix/sysv/linux/fstatat64.c
>> b/sysdeps/unix/sysv/linux/fstatat64.c
>> index 3509d3ca6d..127c6ff601 100644
>> --- a/sysdeps/unix/sysv/linux/fstatat64.c
>> +++ b/sysdeps/unix/sysv/linux/fstatat64.c
>> @@ -21,12 +21,10 @@
>> #include <sys/stat.h>
>> #include <fcntl.h>
>> #include <string.h>
>> -#include <kernel_stat.h>
>> #include <sysdep.h>
>> #include <time.h>
>> -#include <kstat_cp.h>
>> -#include <stat_t64_cp.h>
>> #include <sys/sysmacros.h>
>> +#include <internal-stat.h>
>>
>> #if __TIMESIZE == 64 \
>> && (__WORDSIZE == 32 \
>> @@ -40,11 +38,7 @@ _Static_assert (sizeof (__blkcnt_t) == sizeof
>> (__blkcnt64_t),
>> "__blkcnt_t and __blkcnt64_t must match");
>> #endif
>>
>> -#if (__WORDSIZE == 32 \
>> - && (!defined __SYSCALL_WORDSIZE || __SYSCALL_WORDSIZE == 32)) \
>> - || defined STAT_HAS_TIME32 \
>> - || (!defined __NR_newfstatat && !defined __NR_fstatat64)
>> -# define FSTATAT_USE_STATX 1
>> +#if FSTATAT_USE_STATX
>>
>> static inline int
>> fstatat64_time64_statx (int fd, const char *file, struct __stat64_t64
>> *buf,
>> @@ -79,8 +73,6 @@ fstatat64_time64_statx (int fd, const char *file, struct
>> __stat64_t64 *buf,
>>
>> return r;
>> }
>> -#else
>> -# define FSTATAT_USE_STATX 0
>> #endif
>>
>> /* Only statx supports 64-bit timestamps for 32-bit architectures with
>> diff --git a/sysdeps/unix/sysv/linux/internal-stat.h
>> b/sysdeps/unix/sysv/linux/internal-stat.h
>> new file mode 100644
>> index 0000000000..e3b0569853
>> --- /dev/null
>> +++ b/sysdeps/unix/sysv/linux/internal-stat.h
>> @@ -0,0 +1,31 @@
>> +/* Internal stat definitions.
>> + Copyright (C) 2023 Free Software Foundation, Inc.
>> + This file is part of the GNU C Library.
>> +
>> + The GNU C Library is free software; you can redistribute it and/or
>> + modify it under the terms of the GNU Lesser General Public
>> + License as published by the Free Software Foundation; either
>> + version 2.1 of the License, or (at your option) any later version.
>> +
>> + The GNU C Library is distributed in the hope that it will be useful,
>> + but WITHOUT ANY WARRANTY; without even the implied warranty of
>> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
>> + Lesser General Public License for more details.
>> +
>> + You should have received a copy of the GNU Lesser General Public
>> + License along with the GNU C Library; if not, see
>> + <https://www.gnu.org/licenses/>. */
>> +
>> +#include <sysdep.h>
>> +#include <stat_t64_cp.h>
>> +#include <kernel_stat.h>
>> +#include <kstat_cp.h>
>> +
>> +#if (__WORDSIZE == 32 \
>> + && (!defined __SYSCALL_WORDSIZE || __SYSCALL_WORDSIZE == 32)) \
>> + || defined STAT_HAS_TIME32 \
>> + || (!defined __NR_newfstatat && !defined __NR_fstatat64)
>> +# define FSTATAT_USE_STATX 1
>> +#else
>> +# define FSTATAT_USE_STATX 0
>> +#endif
>> --
>> 2.34.1
>>
>>
>
>
next prev parent reply other threads:[~2023-09-05 20:56 UTC|newest]
Thread overview: 10+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-09-05 20:34 Adhemerval Zanella
2023-09-05 20:48 ` Mateusz Guzik
2023-09-05 20:56 ` Adhemerval Zanella Netto [this message]
2023-09-05 21:03 ` Mateusz Guzik
2023-09-06 20:48 ` Florian Weimer
2023-09-11 13:11 ` Adhemerval Zanella Netto
2023-09-11 19:32 ` Linus Torvalds
2023-09-11 22:10 ` Adhemerval Zanella Netto
2023-09-12 5:31 ` Florian Weimer
2023-09-12 13:13 ` Adhemerval Zanella Netto
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=5b4cb21e-5bb0-5015-4c4b-f4fb4b9d2892@linaro.org \
--to=adhemerval.zanella@linaro.org \
--cc=libc-alpha@sourceware.org \
--cc=mjguzik@gmail.com \
--cc=torvalds@linux-foundation.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).