2021-10-05 13:36:02

by Jürgen Groß

Subject: [PATCH] xen/balloon: fix cancelled balloon action

In case a ballooning action is cancelled the new kernel thread handling
the ballooning might end up in a busy loop.

Fix that by handling the cancelled action gracefully.

While at it introduce a short wait for the BP_WAIT case.

Cc: [email protected]
Fixes: 8480ed9c2bbd56 ("xen/balloon: use a kernel thread instead a workqueue")
Reported-by: Marek Marczykowski-Górecki <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
---
drivers/xen/balloon.c | 21 +++++++++++++++------
1 file changed, 15 insertions(+), 6 deletions(-)

diff --git a/drivers/xen/balloon.c b/drivers/xen/balloon.c
index 43ebfe36ac27..3a50f097ed3e 100644
--- a/drivers/xen/balloon.c
+++ b/drivers/xen/balloon.c
@@ -491,12 +491,12 @@ static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp)
}

/*
- * Stop waiting if either state is not BP_EAGAIN and ballooning action is
- * needed, or if the credit has changed while state is BP_EAGAIN.
+ * Stop waiting if either state is BP_DONE and ballooning action is
+ * needed, or if the credit has changed while state is not BP_DONE.
*/
static bool balloon_thread_cond(enum bp_state state, long credit)
{
- if (state != BP_EAGAIN)
+ if (state == BP_DONE)
credit = 0;

return current_credit() != credit || kthread_should_stop();
@@ -516,10 +516,19 @@ static int balloon_thread(void *unused)

set_freezable();
for (;;) {
- if (state == BP_EAGAIN)
- timeout = balloon_stats.schedule_delay * HZ;
- else
+ switch (state) {
+ case BP_DONE:
+ case BP_ECANCELED:
timeout = 3600 * HZ;
+ break;
+ case BP_EAGAIN:
+ timeout = balloon_stats.schedule_delay * HZ;
+ break;
+ case BP_WAIT:
+ timeout = HZ;
+ break;
+ }
+
credit = current_credit();

wait_event_freezable_timeout(balloon_thread_wq,
--
2.26.2

2021-10-06 01:13:44

by Boris Ostrovsky

Subject: Re: [PATCH] xen/balloon: fix cancelled balloon action

On 10/5/21 9:34 AM, Juergen Gross wrote:
> In case a ballooning action is cancelled the new kernel thread handling
> the ballooning might end up in a busy loop.
>
> Fix that by handling the cancelled action gracefully.
>
> While at it introduce a short wait for the BP_WAIT case.
>
> Cc: [email protected]
> Fixes: 8480ed9c2bbd56 ("xen/balloon: use a kernel thread instead a workqueue")
> Reported-by: Marek Marczykowski-Górecki <[email protected]>
> Signed-off-by: Juergen Gross <[email protected]>


Reviewed-by: Boris Ostrovsky <[email protected]>