2020-12-02 15:19:43

by Qian Cai

Subject: [PATCH] mm/swapfile: Do not sleep with a spin lock held

We can't call kvfree() with a spin lock held, so defer it.

Signed-off-by: Qian Cai <[email protected]>
---
mm/swapfile.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/mm/swapfile.c b/mm/swapfile.c
index c4a613688a17..d58361109066 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -2867,6 +2867,7 @@ late_initcall(max_swapfiles_check);
static struct swap_info_struct *alloc_swap_info(void)
{
struct swap_info_struct *p;
+ struct swap_info_struct *defer = NULL;
unsigned int type;
int i;

@@ -2895,7 +2896,7 @@ static struct swap_info_struct *alloc_swap_info(void)
smp_wmb();
WRITE_ONCE(nr_swapfiles, nr_swapfiles + 1);
} else {
- kvfree(p);
+ defer = p;
p = swap_info[type];
/*
* Do not memset this entry: a racing procfs swap_next()
@@ -2908,6 +2909,7 @@ static struct swap_info_struct *alloc_swap_info(void)
plist_node_init(&p->avail_lists[i], 0);
p->flags = SWP_USED;
spin_unlock(&swap_lock);
+ kvfree(defer);
spin_lock_init(&p->lock);
spin_lock_init(&p->cont_lock);

--
2.28.0

2020-12-02 23:19:57

by Andrew Morton

Subject: Re: [PATCH] mm/swapfile: Do not sleep with a spin lock held

On Wed, 2 Dec 2020 10:15:49 -0500 Qian Cai <[email protected]> wrote:

> We can't call kvfree() with a spin lock held, so defer it.
>

Fixes: 873d7bcfd066 ("mm/swapfile.c: use kvzalloc for swap_info_struct allocatio
n")

Do you think it's worth a cc:stable? IOW, is this known to ever
produce a warning?

(If not, why did we need 873d7bcfd066??)

2020-12-02 23:44:26

by Qian Cai

Subject: Re: [PATCH] mm/swapfile: Do not sleep with a spin lock held

On Wed, 2020-12-02 at 15:15 -0800, Andrew Morton wrote:
> On Wed, 2 Dec 2020 10:15:49 -0500 Qian Cai <[email protected]> wrote:
>
> > We can't call kvfree() with a spin lock held, so defer it.
> >
>
> Fixes: 873d7bcfd066 ("mm/swapfile.c: use kvzalloc for swap_info_struct
> allocatio
> n")
>
> Do you think it's worth a cc:stable? IOW, is this known to ever
> produce a warning?

Yes, it did trigger a might_sleep() warning.