stop_machine: Mark per cpu stopper enabled early

commit 14e568e78 (stop_machine: Use smpboot threads) introduced the following regression: Before this commit the stopper enabled bit was set in the online notifier. CPU0 CPU1 cpu_up cpu online hotplug_notifier(ONLINE) stopper(CPU1)->enabled = true; ... stop_machine() The conversion to smpboot threads moved the enablement to the wakeup path of the parked thread. The majority of users seem to have the following working order: CPU0 CPU1 cpu_up cpu online unpark_threads() wakeup(stopper[CPU1]) .... stopper thread runs stopper(CPU1)->enabled = true; stop_machine() But Konrad and Sander have observed: CPU0 CPU1 cpu_up cpu online unpark_threads() wakeup(stopper[CPU1]) .... stop_machine() stopper thread runs stopper(CPU1)->enabled = true; Now the stop machinery kicks CPU0 into the stop loop, where it gets stuck forever because the queue code saw stopper(CPU1)->enabled == false, so CPU0 waits for CPU1 to enter stomp_machine, but the CPU1 stopper work got discarded due to enabled == false. Add a pre_unpark function to the smpboot thread descriptor and call it before waking the thread. This fixes the problem at hand, but the stop_machine code should be more robust. The stopper->enabled flag smells fishy at best. Thanks to Konrad for going through a loop of debug patches and providing the information to decode this issue.

Thomas Gleixner committed 13 years ago

Commit 46c498c1 parent 1a13c0b3 files · +7 −1Browse files

Unified · Split

3 files

include/linux/smpboot.h +4 −0
kernel/smpboot.c +2 −0
kernel/stop_machine.c +1 −1

include/linux/smpboot.h+4−0

@@ -24,6 +24,9 @@ struct smpboot_thread_data; · 23 unmodified lines
24	24	* parked (cpu offline)
25	25	* @unpark: Optional unpark function, called when the thread is
26	26	* unparked (cpu online)
	27	+ * @pre_unpark: Optional unpark function, called before the thread is
	28	+ * unparked (cpu online). This is not guaranteed to be
	29	+ * called on the target cpu of the thread. Careful!
27	30	* @selfparking: Thread is not parked by the park function.
28	31	* @thread_comm: The base name of the thread
29	32	*/
@@ -37,6 +40,7 @@ struct smp_hotplug_thread { · 7 unmodified lines
37	40	void (*cleanup)(unsigned int cpu, bool online);
38	41	void (*park)(unsigned int cpu);
39	42	void (*unpark)(unsigned int cpu);
	43	+ void (*pre_unpark)(unsigned int cpu);
40	44	bool selfparking;
41	45	const char *thread_comm;
42	46	};

kernel/smpboot.c+2−0

@@ -209,6 +209,8 @@ static void smpboot_unpark_thread(struct smp_hotplug_thread *ht, unsigned int cp · 208 unmodified lines
209	209	{
210	210	struct task_struct tsk = per_cpu_ptr(ht->store, cpu);
211	211
	212	+ if (ht->pre_unpark)
	213	+ ht->pre_unpark(cpu);
212	214	kthread_unpark(tsk);
213	215	}
214	216

kernel/stop_machine.c+1−1

@@ -336,7 +336,7 @@ static struct smp_hotplug_thread cpu_stop_threads = { · 335 unmodified lines
336	336	.create = cpu_stop_create,
337	337	.setup = cpu_stop_unpark,
338	338	.park = cpu_stop_park,
339		− .unpark = cpu_stop_unpark,
	339	+ .pre_unpark = cpu_stop_unpark,
340	340	.selfparking = true,
341	341	};
342	342