On 25-04-18, 15:32, Shilpasri G Bhat wrote:

Hi,

On 04/25/2018 02:47 PM, Viresh Kumar wrote:

quoted

On 25-04-18, 14:32, Shilpasri G Bhat wrote:

quoted

gpstate_timer_handler() uses synchronous smp_call to set the pstate
on the requested core. This causes the below hard lockup:

[c000003fe566b320] [c0000000001d5340] smp_call_function_single+0x110/0x180 (unreliable)
[c000003fe566b390] [c0000000001d55e0] smp_call_function_any+0x180/0x250
[c000003fe566b3f0] [c000000000acd3e8] gpstate_timer_handler+0x1e8/0x580
[c000003fe566b4a0] [c0000000001b46b0] call_timer_fn+0x50/0x1c0
[c000003fe566b520] [c0000000001b4958] expire_timers+0x138/0x1f0
[c000003fe566b590] [c0000000001b4bf8] run_timer_softirq+0x1e8/0x270
[c000003fe566b630] [c000000000d0d6c8] __do_softirq+0x158/0x3e4
[c000003fe566b710] [c000000000114be8] irq_exit+0xe8/0x120
[c000003fe566b730] [c000000000024d0c] timer_interrupt+0x9c/0xe0
[c000003fe566b760] [c000000000009014] decrementer_common+0x114/0x120
-- interrupt: 901 at doorbell_global_ipi+0x34/0x50
LR = arch_send_call_function_ipi_mask+0x120/0x130
[c000003fe566ba50] [c00000000004876c]
arch_send_call_function_ipi_mask+0x4c/0x130
[c000003fe566ba90] [c0000000001d59f0] smp_call_function_many+0x340/0x450
[c000003fe566bb00] [c000000000075f18] pmdp_invalidate+0x98/0xe0
[c000003fe566bb30] [c0000000003a1120] change_huge_pmd+0xe0/0x270
[c000003fe566bba0] [c000000000349278] change_protection_range+0xb88/0xe40
[c000003fe566bcf0] [c0000000003496c0] mprotect_fixup+0x140/0x340
[c000003fe566bdb0] [c000000000349a74] SyS_mprotect+0x1b4/0x350
[c000003fe566be30] [c00000000000b184] system_call+0x58/0x6c

One way to avoid this is removing the smp-call. We can ensure that the timer
always runs on one of the policy-cpus. If the timer gets migrated to a
cpu outside the policy then re-queue it back on the policy->cpus. This way
we can get rid of the smp-call which was being used to set the pstate
on the policy->cpus.

Fixes: 7bc54b652f13 (timers, cpufreq/powernv: Initialize the gpstate timer as pinned)
Cc: <redacted>        [4.8+]
Reported-by: Nicholas Piggin <npiggin@gmail.com>
Reported-by: Pridhiviraj Paidipeddi <redacted>
Signed-off-by: Shilpasri G Bhat <redacted>
---
Changes from V1:
- Remove smp_call in the pstate handler.

 drivers/cpufreq/powernv-cpufreq.c | 23 ++++++++++++++++++++---
 1 file changed, 20 insertions(+), 3 deletions(-)

diff --git a/drivers/cpufreq/powernv-cpufreq.c b/drivers/cpufreq/powernv-cpufreq.c
index 71f8682..dc8ffb5 100644
--- a/drivers/cpufreq/powernv-cpufreq.c
+++ b/drivers/cpufreq/powernv-cpufreq.c

@@ -679,6 +679,25 @@ void gpstate_timer_handler(struct timer_list *t)
 
 	if (!spin_trylock(&gpstates->gpstate_lock))
 		return;
+	/*
+	 * If the timer has migrated to the different cpu then bring
+	 * it back to one of the policy->cpus
+	 */
+	if (!cpumask_test_cpu(raw_smp_processor_id(), policy->cpus)) {
+		/*
+		 * Timer should be deleted if policy is inactive.
+		 * If policy is active then re-queue on one of the
+		 * policy->cpus.
+		 */

This looks racy. Shouldn't you guarantee that the timer is already
removed in a synchronous way before de-activating the policy ?

The timer is deleted in driver->stop_cpu(). So we ensure to remove the timer
before de-activating the policy.

quoted

quoted

+		if (!cpumask_empty(policy->cpus)) {

So are you suggesting to remove ^^ the check for active policy here?
(I put that as a safety check.)

Either you are sure or you are not, and you don't need a safety check
if you are sure :)

-- 
viresh