| Message ID | 20231024180718.626055-1-kaleshsingh@google.com |
|---|---|
| State | New |
| Headers | show |
| Series | PM/sleep: Instrument start of filesystems sync | expand |
diff --git a/kernel/power/main.c b/kernel/power/main.c index f6425ae3e8b0..b6a88b43b6da 100644 --- a/kernel/power/main.c +++ b/kernel/power/main.c @@ -87,6 +87,7 @@ void ksys_sync_helper(void) ktime_t start; long elapsed_msecs; + pr_info("Syncing filesystems ... "); start = ktime_get(); ksys_sync(); elapsed_msecs = ktime_to_ms(ktime_sub(ktime_get(), start));
Systems, such as Android, that implement opportunistic sleep from a userspace power manager service need to guard the acquisition of new wakelocks (incrementing a counter in userspace) while attempting a suspend (write to /sys/power/state). While attempting to suspend, processes attempting to acquire a userspace wakelock are also blocked. If the suspend is blocked for too long it can cause a watchdog bite in android. Some device vendors reported this issue in cases where the filesystem sync is not fast enough (sometimes > 100s): [ 2742.247785] [07-10 15:08:02.247] Filesystems sync: 107.127 seconds Commit b5dee3130bb4 ("PM / sleep: Refactor filesystems sync to reduce duplication") introduced ksys_sync_helper() which logged both the start and end of the filesystems sync. Commit c64546b17bc9 ("PM / sleep: Measure the time of filesystems syncing") improved on this by adding the elapsed time, but also removed instrumentation indicating the start of filesystems sync. Re-add the log indicating the start of filesystem sync. This helps to identify when the above scenario has occurred in production devices in the field. Cc: Rafael J. Wysocki <rafael@kernel.org> Cc: Harry Pan <harry.pan@intel.com> Signed-off-by: Kalesh Singh <kaleshsingh@google.com> --- kernel/power/main.c | 1 + 1 file changed, 1 insertion(+) base-commit: 05d3ef8bba77c1b5f98d941d8b2d4aeab8118ef1