xen: release lock occasionally during ballooning

When dom0 is being ballooned balloon_process() will hold the balloon
mutex until it is finished. This will block e.g. creation of new
domains as the device backends for the new domain need some
autoballooned pages for the ring buffers.

Avoid this by releasing the balloon mutex from time to time during
ballooning. Adjust the comment above balloon_process() regarding
multiple instances of balloon_process().

Instead of open coding it, just use cond_resched().

Signed-off-by: Juergen Gross <jgross@suse.com>
Signed-off-by: David Vrabel <david.vrabel@citrix.com>
This commit is contained in:
Juergen Gross 2015-07-20 13:49:39 +02:00 коммит произвёл David Vrabel
Родитель 30b03d05e0
Коммит 929423fa83
1 изменённых файлов: 7 добавлений и 8 удалений

Просмотреть файл

@ -472,7 +472,7 @@ static enum bp_state decrease_reservation(unsigned long nr_pages, gfp_t gfp)
}
/*
* We avoid multiple worker processes conflicting via the balloon mutex.
* As this is a work item it is guaranteed to run as a single instance only.
* We may of course race updates of the target counts (which are protected
* by the balloon lock), or with changes to the Xen hard limit, but we will
* recover from these in time.
@ -482,9 +482,10 @@ static void balloon_process(struct work_struct *work)
enum bp_state state = BP_DONE;
long credit;
mutex_lock(&balloon_mutex);
do {
mutex_lock(&balloon_mutex);
credit = current_credit();
if (credit > 0) {
@ -499,17 +500,15 @@ static void balloon_process(struct work_struct *work)
state = update_schedule(state);
#ifndef CONFIG_PREEMPT
if (need_resched())
schedule();
#endif
mutex_unlock(&balloon_mutex);
cond_resched();
} while (credit && state == BP_DONE);
/* Schedule more work if there is some still to be done. */
if (state == BP_EAGAIN)
schedule_delayed_work(&balloon_worker, balloon_stats.schedule_delay * HZ);
mutex_unlock(&balloon_mutex);
}
/* Resets the Xen limit, sets new target, and kicks off processing. */