Skip to contentSkip to conversation

tip/master merge problem?

DISCUSSION Mike Travis 2 participants 2 messages Dec 31, 2008 – Jan 2, 2009

lore
Share & tools
lore
#0 lore raw

Hi Ingo,

I found this code snippet in kernel/sched.c in the tip/master tree. It appears to be a merge mistake: (cut and pasted - no tabs)

292 /*

293                  * If this cpu is the one which updates jiffies, then
294                  * give up the assignment and let it be taken by the
295                  * cpu which runs the tick timer next, which might be
296                  * this cpu as well. If we don't drop this here the
297                  * jiffies might be stale and do_timer() never
298                  * invoked.
299                  */

300 if (cpu == tick_do_timer_cpu)
301 tick_do_timer_cpu = TICK_DO_TIMER_NONE;
302
303 if (delta_jiffies > 1)
304
305 /* Skip reprogram of event if its not changed */
306 if (ts->tick_stopped && ktime_equal(expires, dev->next_event))
307 goto out;
308
309 cpumask_set_cpu(cpu, nohz_cpu_mask);

I think the "Skip reprogram" was inserted into the middle of a previous if-then statement.

git blames it on:
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 292) /*

00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 293)               * If this cpu is the one which updates jiffies, then
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 294)               * give up the assignment and let it be taken by the
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 295)               * cpu which runs the tick timer next, which might be
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 296)               * this cpu as well. If we don't drop this here the
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 297)               * jiffies might be stale and do_timer() never
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 298)               * invoked.
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 299)               */

00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 300) if (cpu == tick_do_timer_cpu)
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 301) tick_do_timer_cpu = TICK_DO_TIMER_NONE;
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 302)
6ba9b346 (Ingo Molnar 2007-02-19 18:11:56 +0000 303) if (delta_jiffies > 1)
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 304)
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 305) /* Skip reprogram of event if its not changed */
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 306) if (ts->tick_stopped && ktime_equal(expires, dev->next_event))
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 307) goto out;
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 308)
6a7b3dc3 (Rusty Russell 2008-11-25 02:35:04 +1030 309) cpumask_set_cpu(cpu, nohz_cpu_mask);

I've Cc'd both interested parties.

Thanks,
Mike

#1 lore raw
Quoted 25 lines · Mike Travis

Hi Ingo,

I found this code snippet in kernel/sched.c in the tip/master tree. It appears to be a merge mistake: (cut and pasted - no tabs)

292 /*

293                  * If this cpu is the one which updates jiffies, then
294                  * give up the assignment and let it be taken by the
295                  * cpu which runs the tick timer next, which might be
296                  * this cpu as well. If we don't drop this here the
297                  * jiffies might be stale and do_timer() never
298                  * invoked.
299                  */

300 if (cpu == tick_do_timer_cpu)
301 tick_do_timer_cpu = TICK_DO_TIMER_NONE;
302
303 if (delta_jiffies > 1)
304
305 /* Skip reprogram of event if its not changed */
306 if (ts->tick_stopped && ktime_equal(expires, dev->next_event))
307 goto out;
308
309 cpumask_set_cpu(cpu, nohz_cpu_mask);

Quoted 2 lines

I think the "Skip reprogram" was inserted into the middle of a previous if-then statement.

indeed, this is plain broken. It happened to build and boot but still it's broken.

Quoted 20 lines

git blames it on:
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 292) /*

00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 293)               * If this cpu is the one which updates jiffies, then
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 294)               * give up the assignment and let it be taken by the
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 295)               * cpu which runs the tick timer next, which might be
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 296)               * this cpu as well. If we don't drop this here the
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 297)               * jiffies might be stale and do_timer() never
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 298)               * invoked.
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 299)               */

00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 300) if (cpu == tick_do_timer_cpu)
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 301) tick_do_timer_cpu = TICK_DO_TIMER_NONE;
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 302)
6ba9b346 (Ingo Molnar 2007-02-19 18:11:56 +0000 303) if (delta_jiffies > 1)
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 304)
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 305) /* Skip reprogram of event if its not changed */
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 306) if (ts->tick_stopped && ktime_equal(expires, dev->next_event))
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 307) goto out;
00147449 (Woodruff, Richard 2008-12-01 14:18:11 -0800 308)
6a7b3dc3 (Rusty Russell 2008-11-25 02:35:04 +1030 309) cpumask_set_cpu(cpu, nohz_cpu_mask);

Quoted 1 line

I've Cc'd both interested parties.

purely my fault. I've zapped that conflict resolution and picked up yours.

Ingo