time/sched_clock: Use raw_read_seqcount_latch() during suspend
authorAhmed S. Darwish <a.darwish@linutronix.de>
Thu, 27 Aug 2020 11:40:37 +0000 (13:40 +0200)
committerPeter Zijlstra <peterz@infradead.org>
Thu, 10 Sep 2020 09:19:28 +0000 (11:19 +0200)
sched_clock uses seqcount_t latching to switch between two storage
places protected by the sequence counter. This allows it to have
interruptible, NMI-safe, seqcount_t write side critical sections.

Since 7fc26327b756 ("seqlock: Introduce raw_read_seqcount_latch()"),
raw_read_seqcount_latch() became the standardized way for seqcount_t
latch read paths. Due to the dependent load, it has one read memory
barrier less than the currently used raw_read_seqcount() API.

Use raw_read_seqcount_latch() for the suspend path.

Commit aadd6e5caaac ("time/sched_clock: Use raw_read_seqcount_latch()")
missed changing that instance of raw_read_seqcount().

References: 1809bfa44e10 ("timers, sched/clock: Avoid deadlock during read from NMI")
Signed-off-by: Ahmed S. Darwish <a.darwish@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200715092345.GA231464@debian-buster-darwi.lab.linutronix.de
kernel/time/sched_clock.c

index 1c03eec..8c6b5fe 100644 (file)
@@ -258,7 +258,7 @@ void __init generic_sched_clock_init(void)
  */
 static u64 notrace suspended_sched_clock_read(void)
 {
-       unsigned int seq = raw_read_seqcount(&cd.seq);
+       unsigned int seq = raw_read_seqcount_latch(&cd.seq);
 
        return cd.read_data[seq & 1].epoch_cyc;
 }