Skip to contentSkip to conversation

bus: arm-ccn: cpumask attribute

PATCH In review Dan Carpenter 3 participants 4 messages May 14, 2015 Kernel janitors

lore
Share & tools
lore
#0 lore raw

Hello Pawel Moll,

The patch ffa415245b86: "bus: arm-ccn: cpumask attribute" from Apr 16, 2015, leads to the following static checker warning:

	drivers/bus/arm-ccn.c:1188 arm_ccn_pmu_cpu_notifier()
	warn: unsigned 'target' is never less than zero.

drivers/bus/arm-ccn.c
1175 static int arm_ccn_pmu_cpu_notifier(struct notifier_block *nb,
1176 unsigned long action, void *hcpu)
1177 {

  1178          struct arm_ccn_dt *dt = container_of(nb, struct arm_ccn_dt, cpu_nb);
  1179          struct arm_ccn *ccn = container_of(dt, struct arm_ccn, dt);
  1180          unsigned int cpu = (long)hcpu; /* for (long) see kernel/cpu.c */
  1181          unsigned int target;
  1182  
  1183          switch (action & ~CPU_TASKS_FROZEN) {
  1184          case CPU_DOWN_PREPARE:
  1185                  if (!cpumask_test_and_clear_cpu(cpu, &dt->cpu))
  1186                          break;
  1187                  target = cpumask_any_but(cpu_online_mask, cpu);
  1188                  if (target < 0)
                            ^^^^^^^^^^

target is unsigned and cpumask_any_but() returns unsigned.

  1189                          break;
  1190                  perf_pmu_migrate_context(&dt->pmu, cpu, target);
  1191                  cpumask_set_cpu(target, &dt->cpu);
  1192                  WARN_ON(irq_set_affinity(ccn->irq, &dt->cpu) != 0);
  1193          default:
  1194                  break;
  1195          }
  1196  
  1197          return NOTIFY_OK;
  1198  }

regards,
dan carpenter

#1 lore raw

PATCH 1/1 bus: arm-ccn: Handle correctly no-more-cpus case

When migrating events the driver picks another cpu using cpumask_any_but() function, which returns value >= nr_cpu_ids when there is none available, not a negative value as the code assumed. Fixed now.

---
Another day, another arm-ccn.c update...

This time Dan's static checker spotted unsigned int target being expected to carry negative values. Fixed now.

Interestingly enough, cpumask_any_but() implementation (and its normal prototype) returns int, but version for NR_CPUS = 1 case, inlined in linux/cpumask.h returns unsigned int...

 drivers/bus/arm-ccn.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/bus/arm-ccn.c +1 -1
index 7d9879e..cc322fb 100644
@@ -1184,7 +1184,7 @@ static int arm_ccn_pmu_cpu_notifier(struct notifier_block *nb,
 		if (!cpumask_test_and_clear_cpu(cpu, &dt->cpu))
 			break;
 		target = cpumask_any_but(cpu_online_mask, cpu);
-		if (target < 0)
+		if (target >= nr_cpu_ids)
 			break;
 		perf_pmu_migrate_context(&dt->pmu, cpu, target);
 		cpumask_set_cpu(target, &dt->cpu);
Signature · 1 line
2.1.0
Acked-by on v1 1/1
#2 lore raw

Re: [PATCH] bus: arm-ccn: Handle correctly no-more-cpus case

Quoted 4 lines · Pawel Moll

When migrating events the driver picks another cpu using cpumask_any_but() function, which returns value >= nr_cpu_ids when there is none available, not a negative value as the code assumed. Fixed now.

The fix looks good to me:

Does this need to be CC'd to stable? What does perf_pmu_migrate_context do when passed a target >= nr_cpus?

The original bug seems to have been copied over from arm-cci.c, which will need the same fix. That appears to be my fault -- I'd mostly been following the x86 uncore PMU drivers, but they figure out the target CPU in a different way for which -1 is a sane error case.

I'll spin a patch for arm-cci.c momentarily.

Thanks,
Mark.

Quoted 32 lines

---
Another day, another arm-ccn.c update...

This time Dan's static checker spotted unsigned int target being expected to carry negative values. Fixed now.

Interestingly enough, cpumask_any_but() implementation (and its normal prototype) returns int, but version for NR_CPUS = 1 case, inlined in linux/cpumask.h returns unsigned int...

 drivers/bus/arm-ccn.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/bus/arm-ccn.c +1 -1
index 7d9879e..cc322fb 100644
@@ -1184,7 +1184,7 @@ static int arm_ccn_pmu_cpu_notifier(struct notifier_block *nb,
 		if (!cpumask_test_and_clear_cpu(cpu, &dt->cpu))
 			break;
 		target = cpumask_any_but(cpu_online_mask, cpu);
-		if (target < 0)
+		if (target >= nr_cpu_ids)
 			break;
 		perf_pmu_migrate_context(&dt->pmu, cpu, target);
 		cpumask_set_cpu(target, &dt->cpu);
Signature · 1 line
2.1.0
#3 lore raw

Re: [PATCH] bus: arm-ccn: Handle correctly no-more-cpus case

Quoted 6 lines · Mark Rutland
Quoted 4 lines · Pawel Moll

When migrating events the driver picks another cpu using cpumask_any_but() function, which returns value >= nr_cpu_ids when there is none available, not a negative value as the code assumed. Fixed now.

Quoted 6 lines

The fix looks good to me:

Does this need to be CC'd to stable? What does perf_pmu_migrate_context do when passed a target >= nr_cpus?

Never mind, it looks like we can never encounter that case anyway. In the case of kexec or reset we won't notify CPU_DOWN_PREPARE on the final CPU, and we can't hotplug the final CPU.

Which means that the target check is irrelevant as we should always get a valid cpu back from cpumask_any_but in the cases we'll call it. So we could just delete it entirely, assuming I haven't missed a CPU_DOWN_PREPARE notification path...

Mark.

Quoted 22 lines
Quoted 16 lines
drivers/bus/arm-ccn.c +1 -1
index 7d9879e..cc322fb 100644
@@ -1184,7 +1184,7 @@ static int arm_ccn_pmu_cpu_notifier(struct notifier_block *nb,
 		if (!cpumask_test_and_clear_cpu(cpu, &dt->cpu))
 			break;
 		target = cpumask_any_but(cpu_online_mask, cpu);
-		if (target < 0)
+		if (target >= nr_cpu_ids)
 			break;
 		perf_pmu_migrate_context(&dt->pmu, cpu, target);
 		cpumask_set_cpu(target, &dt->cpu);
Signature · 1 line
2.1.0

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel