Skip to contentSkip to conversation

ashmem: Fix ashmem shrinker nr_to_scan

RFC In review Joel Fernandes (Google) 2 participants 2 messages Mar 6, 2020 – Mar 10, 2020 Staging

lore
Share & tools
lore
#0 lore raw

RFC 1/1 ashmem: Fix ashmem shrinker nr_to_scan

nr_to_scan is the number of pages to be freed however ashmem doesn't discount nr_to_scan correctly as it frees ranges. It should be discounting them by pages than by ranges. Correct the issue.

---
 drivers/staging/android/ashmem.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/staging/android/ashmem.c +3 -1
index 5891d0744a760..cb525ea6db98a 100644
@@ -458,6 +458,8 @@ ashmem_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
 		lru_del(range);
 
 		freed += range_size(range);
+		sc->nr_to_scan -=  range_size(range);
+
 		mutex_unlock(&ashmem_mutex);
 		f->f_op->fallocate(f,
 				   FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
@@ -467,7 +469,7 @@ ashmem_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
 			wake_up_all(&ashmem_shrink_wait);
 		if (!mutex_trylock(&ashmem_mutex))
 			goto out;
-		if (--sc->nr_to_scan <= 0)
+		if (sc->nr_to_scan <= 0)
 			break;
 	}
 	mutex_unlock(&ashmem_mutex);
Signature · 1 line
2.25.0.265.gbab2e86ba0-goog
#1 lore raw
Quoted 15 lines · Joel Fernandes (Google)

nr_to_scan is the number of pages to be freed however ashmem doesn't discount nr_to_scan correctly as it frees ranges. It should be discounting them by pages than by ranges. Correct the issue.

---
 drivers/staging/android/ashmem.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/staging/android/ashmem.c +0 -0
index 5891d0744a760..cb525ea6db98a 100644

drivers/staging/android/ashmem.c ashmem_shrink_scan(struct shrinker *shrink, struct shrink_control *sc) · line 461

Quoted 4 lines
drivers/staging/android/ashmem.c +1 -0
@@ -458,6 +458,8 @@ ashmem_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
 		lru_del(range);
 
 		freed += range_size(range);
+		sc->nr_to_scan -=  range_size(range);
                                 ^^

Two space characters.

In the old code we didn't *really* have to worry about sc->nr_to_scan dropping to negative, but now we do. ->nr_to_scan is unsigned so it would be a high positive value now so it's maybe a forever loop? I'm too lazy to verify...

regards,
dan carpenter