Skip to contentSkip to conversation

deadlock between cpu_stopper & native_flush_tlb_others()->smp_call_function_many()

DISCUSSION Igor Mammedov 2 participants 3 messages Mar 3, 2014 – Mar 5, 2014

lore
Share & tools
lore
#0 lore raw

It looks like I hit a deadlock between smp_call_function_many() and cpu_stopper threads.

Where smp_call_function_many() on CPU1 called from native_flush_tlb_others() waits on call to be complete on CPU2 while CPU2 waits on state synchronization in multi_cpu_stop() which can't be completed until stop work queued on CPU1 is completed, which can't be done since CPU1 is busy looping in smp_call_function_many().

CPU1 CPU2
stop_machine()
queue stop work on cpu 1&2

native_flush_tlb_others()
smp_call_function_many()

      ...
---------------------------------------------------------
                                    cpu_stopper_thread()
                                       multi_cpu_stop()
                                          do {
                                             ...
                                             msdata->state == MULTI_STOP_PREPARE
                                             msdata->active_cpus == 0110
                                             msdata->thread_ack == 1
                                          } while (curstate != MULTI_STOP_EXIT)
                                    waiting until CPU1 ACKs state, i.e. thread_ack == 0
---------------------------------------------------------                                    
      ...
      if (wait) {
         for_cpu(0110) {
           csd_lock_wait(csd);

waiting until call on CPU2 is completed

Are there any suggestions on how to fix this nicely?

#1 lore raw
Quoted 4 lines · Igor Mammedov

It looks like I hit a deadlock between smp_call_function_many() and cpu_stopper threads.

Are there any suggestions on how to fix this nicely?

This patch should fix it:

https://lkml.org/lkml/2014/2/28/194

#2 lore raw
Quoted 9 lines · On Mon, 03 Mar 2014 11:45:04 -0500 Rik van Riel <[email protected]> wrote:
Quoted 4 lines · Igor Mammedov

It looks like I hit a deadlock between smp_call_function_many() and cpu_stopper threads.

Are there any suggestions on how to fix this nicely?

This patch should fix it:

https://lkml.org/lkml/2014/2/28/194

I've just tried it and it doesn't fix issue.