On (22/11/21 22:16), Johannes Weiner wrote:
> On Tue, Nov 22, 2022 at 10:33:38AM +0900, Sergey Senozhatsky wrote:
> > zswap_frontswap_load() should be called from preemptible
> > context (we even call mutex_lock() there) and it does not
> > look like we need to do GFP_ATOMIC allocaion for temp
> > buffer there. Use GFP_KERNEL instead.
> >
> > Signed-off-by: Sergey Senozhatsky <senozhatsky@xxxxxxxxxxxx>
> > ---
> > mm/zswap.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/mm/zswap.c b/mm/zswap.c
> > index 2d69c1d678fe..f6c89049cf70 100644
> > --- a/mm/zswap.c
> > +++ b/mm/zswap.c
> > @@ -1314,7 +1314,7 @@ static int zswap_frontswap_load(unsigned type, pgoff_t offset,
> > }
> >
> > if (!zpool_can_sleep_mapped(entry->pool->zpool)) {
> > - tmp = kmalloc(entry->length, GFP_ATOMIC);
> > + tmp = kmalloc(entry->length, GFP_KERNEL);
>
> There is another one in zswap_writeback_entry() that seems equally
> arbitrary. They came in through the same commit, with no further
> explanation as to this choice. Do you want to pick that up too?
Yup, you patch it in https://lore.kernel.org/lkml/20221119001536.2086599-2-nphamcs@xxxxxxxxx/
and I guess it's there just by accident. We probably want a separate
patch instead that touches those GFP_ATOMIC allocations in both places.
So I have it like this at present
---
>From 66e4acffc0926498f818d11f2f57b9c772131f6e Mon Sep 17 00:00:00 2001
From: Sergey Senozhatsky <senozhatsky@xxxxxxxxxxxx>
Date: Tue, 22 Nov 2022 10:26:13 +0900
Subject: [PATCH] zswap: do not allocate from atomic pool
zswap_frontswap_load() should be called from preemptible
context (we even call mutex_lock() there) and it does not
look like we need to do GFP_ATOMIC allocaion for temp
buffer. The same applies to zswap_writeback_entry().
Use GFP_KERNEL for temporary buffer allocation in both
cases.
Signed-off-by: Johannes Weiner <hannes@xxxxxxxxxxx>
Signed-off-by: Nhat Pham <nphamcs@xxxxxxxxx>
Signed-off-by: Sergey Senozhatsky <senozhatsky@xxxxxxxxxxxx>
---
mm/zpool.c | 7 +++++++
mm/zswap.c | 4 ++--
2 files changed, 9 insertions(+), 2 deletions(-)
diff --git a/mm/zpool.c b/mm/zpool.c
index 68facc193496..f46c0d5e766c 100644
--- a/mm/zpool.c
+++ b/mm/zpool.c
@@ -387,6 +387,13 @@ bool zpool_evictable(struct zpool *zpool)
* zpool_can_sleep_mapped - Test if zpool can sleep when do mapped.
* @zpool: The zpool to test
*
+ * Some allocators enter non-preemptible context in ->map() callback (e.g.
+ * disable pagefaults) and exit that context in ->unmap(), which limits what
+ * we can do with the mapped object. For instance, we cannot wait for
+ * asynchronous crypto API to decompress such an object or take mutexes
+ * since those will call into the scheduler. This function tells us whether
+ * we use such an allocator.
+ *
* Returns: true if zpool can sleep; false otherwise.
*/
bool zpool_can_sleep_mapped(struct zpool *zpool)
diff --git a/mm/zswap.c b/mm/zswap.c
index 2d48fd59cc7a..3019f0bde194 100644
--- a/mm/zswap.c
+++ b/mm/zswap.c
@@ -958,7 +958,7 @@ static int zswap_writeback_entry(struct zpool *pool, unsigned long handle)
};
if (!zpool_can_sleep_mapped(pool)) {
- tmp = kmalloc(PAGE_SIZE, GFP_ATOMIC);
+ tmp = kmalloc(PAGE_SIZE, GFP_KERNEL);
if (!tmp)
return -ENOMEM;
}
@@ -1311,7 +1311,7 @@ static int zswap_frontswap_load(unsigned type, pgoff_t offset,
}
if (!zpool_can_sleep_mapped(entry->pool->zpool)) {
- tmp = kmalloc(entry->length, GFP_ATOMIC);
+ tmp = kmalloc(entry->length, GFP_KERNEL);
if (!tmp) {
ret = -ENOMEM;
goto freeentry;
--
2.38.1.584.g0f3c55d4c2-goog
