From: Vincent Fu <vincentfu@gmail.com>
To: axboe@kernel.dk, fio@vger.kernel.org
Cc: Vincent Fu <vincent.fu@samsung.com>
Subject: [PATCH] gettime: fix cpuclock-test on AMD platforms
Date: Tue, 27 Feb 2024 10:26:00 -0500 [thread overview]
Message-ID: <20240227155856.5012-1-vincent.fu@samsung.com> (raw)
Starting with gcc 11 __sync_synchronize() compiles to
lock or QWORD PTR [rsp], 0
on x86_64 platforms. Previously it compiled to an mfence instruction.
See line 47 of https://godbolt.org/z/xfE18K7b4 for an example.
On Intel platforms this change does not affect the result of fio's CPU
clock test. But on AMD platforms, this change causes fio's CPU clock
test to fail and fio to fall back to clock_gettime() instead of using
the CPU clock for timing.
This patch has fio explicitly use an mfence instruction instead of
__sync_synchornize() in the CPU clock test code on x86_64 platforms in
order to allow the CPU clock test to pass on AMD platforms.
Signed-off-by: Vincent Fu <vincent.fu@samsung.com>
---
arch/arch-x86_64.h | 5 +++++
arch/arch.h | 7 +++++++
gettime.c | 2 +-
3 files changed, 13 insertions(+), 1 deletion(-)
diff --git a/arch/arch-x86_64.h b/arch/arch-x86_64.h
index 86ce1b7e..b402dc6d 100644
--- a/arch/arch-x86_64.h
+++ b/arch/arch-x86_64.h
@@ -26,6 +26,11 @@ static inline unsigned long arch_ffz(unsigned long bitmask)
return bitmask;
}
+static inline void tsc_barrier(void)
+{
+ __asm__ __volatile__("mfence":::"memory");
+}
+
static inline unsigned long long get_cpu_clock(void)
{
unsigned int lo, hi;
diff --git a/arch/arch.h b/arch/arch.h
index 3ee9b053..7e294ddf 100644
--- a/arch/arch.h
+++ b/arch/arch.h
@@ -108,6 +108,13 @@ extern unsigned long arch_flags;
#include "arch-generic.h"
#endif
+#if !defined(__x86_64__) && defined(CONFIG_SYNC_SYNC)
+static inline void tsc_barrier(void)
+{
+ __sync_synchronize();
+}
+#endif
+
#include "../lib/ffz.h"
/* IWYU pragma: end_exports */
diff --git a/gettime.c b/gettime.c
index bc66a3ac..5ca31206 100644
--- a/gettime.c
+++ b/gettime.c
@@ -623,7 +623,7 @@ static void *clock_thread_fn(void *data)
seq = *t->seq;
if (seq == UINT_MAX)
break;
- __sync_synchronize();
+ tsc_barrier();
tsc = get_cpu_clock();
} while (seq != atomic32_compare_and_swap(t->seq, seq, seq + 1));
--
2.43.0
next reply other threads:[~2024-02-27 15:59 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-02-27 15:26 Vincent Fu [this message]
2024-02-27 16:34 ` [PATCH] gettime: fix cpuclock-test on AMD platforms Jens Axboe
2024-02-27 17:38 ` Vincent Fu
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=20240227155856.5012-1-vincent.fu@samsung.com \
--to=vincentfu@gmail.com \
--cc=axboe@kernel.dk \
--cc=fio@vger.kernel.org \
--cc=vincent.fu@samsung.com \
/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).