[v2,6/9] x86/pvclock: improve atomic update of last_value in pvclock_clocksource_read

Commit Message

Peter Zijlstra Jan. 26, 2023, 3:08 p.m. UTC

From: Uros Bizjak <[e-mail]>

Improve atomic update of last_value in pvclock_clocksource_read:

- Atomic update can be skipped if the "last_value" is already
  equal to "ret".

- The detection of atomic update failure is not correct. The value,
  returned by atomic64_cmpxchg should be compared to the old value
  from the location to be updated. If these two are the same, then
  atomic update succeeded and "last_value" location is updated to
  "ret" in an atomic way. Otherwise, the atomic update failed and
  it should be retried with the value from "last_value" - exactly
  what atomic64_try_cmpxchg does in a correct and more optimal way.

Signed-off-by: Uros Bizjak <[e-mail]>
Signed-off-by: Peter Zijlstra (Intel) <[e-mail]>
Link: https://lkml.kernel.org/r/[e-mail]
---
 arch/x86/kernel/pvclock.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

Patch

diff --git a/arch/x86/kernel/pvclock.c b/arch/x86/kernel/pvclock.c
index eda37df016f0..5a2a517dd61b 100644
--- a/arch/x86/kernel/pvclock.c
+++ b/arch/x86/kernel/pvclock.c
@@ -102,10 +102,9 @@  u64 pvclock_clocksource_read(struct pvclock_vcpu_time_info *src)
 	 */
 	last = atomic64_read(&last_value);
 	do {
-		if (ret < last)
+		if (ret <= last)
 			return last;
-		last = atomic64_cmpxchg(&last_value, last, ret);
-	} while (unlikely(last != ret));
+	} while (!atomic64_try_cmpxchg(&last_value, &last, ret));
 
 	return ret;
 }