Messages in this thread | | | From | David Howells <> | Subject | Re: [PATCH 14/14] workqueue: use mod_delayed_work() instead of cancel + queue | Date | Mon, 13 Aug 2012 14:27:57 +0100 |
| |
Tejun Heo <tj@kernel.org> wrote:
> Convert delayed_work users doing cancel_delayed_work() followed by > queue_delayed_work() to mod_delayed_work(). > > Most conversions are straight-forward. Ones worth mentioning are, > > * drivers/edac/edac_mc.c: edac_mc_workq_setup() converted to always > use mod_delayed_work() and cancel loop in > edac_mc_reset_delay_period() is dropped. > > * drivers/platform/x86/thinkpad_acpi.c: No need to remember whether > watchdog is active or not. @fan_watchdog_active and related code > dropped. > > * drivers/power/charger-manager.c: Seemingly a lot of > delayed_work_pending() abuse going on here. > [delayed_]work_pending() are unsynchronized and racy when used like > this. I converted one instance in fullbatt_handler(). Please > conver the rest so that it invokes workqueue APIs for the intended > target state rather than trying to game work item pending state > transitions. e.g. if timer should be modified - call > mod_delayed_work(), canceled - call cancel_delayed_work[_sync](). > > * drivers/thermal/thermal_sys.c: thermal_zone_device_set_polling() > simplified. Note that round_jiffies() calls in this function are > meaningless. round_jiffies() work on absolute jiffies not delta > delay used by delayed_work. > > v2: Tomi pointed out that __cancel_delayed_work() users can't be > safely converted to mod_delayed_work(). They could be calling it > from irq context and if that happens while delayed_work_timer_fn() > is running, it could deadlock. __cancel_delayed_work() users are > dropped. > > Signed-off-by: Tejun Heo <tj@kernel.org> > Acked-by: Henrique de Moraes Holschuh <hmh@hmh.eng.br> > Acked-by: Dmitry Torokhov <dmitry.torokhov@gmail.com> > Acked-by: Anton Vorontsov <cbouatmailru@gmail.com>
Acked-by: David Howells <dhowells@redhat.com>
| |