Skip to contentSkip to conversation

smpboot: Check for successfull allocation of cpumask vars

RFC PATCH Superseded Pranith Kumar 2 participants 4 messages Dec 9, 2014 x86

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 smpboot: Check for successfull allocation of cpumask vars

zalloc_cpumask_var() can return 0 on allocation failure when CONFIG_CPUMASK_OFFSTACK is set. Check for the return value and WARN() on failure of an allocation in such cases.

---
 arch/x86/kernel/smpboot.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)
arch/x86/kernel/smpboot.c +7 -3
index 7a8f584..9b1df21 100644
@@ -1083,6 +1083,7 @@ static void __init smp_cpu_index_default(void)
 void __init native_smp_prepare_cpus(unsigned int max_cpus)
 {
 	unsigned int i;
+	bool ret = true;
 
 	preempt_disable();
 	smp_cpu_index_default();
@@ -1096,9 +1097,12 @@ void __init native_smp_prepare_cpus(unsigned int max_cpus)
 
 	current_thread_info()->cpu = 0;  /* needed? */
 	for_each_possible_cpu(i) {
-		zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
-		zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
-		zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
+		ret &= zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
+		ret &= zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
+		ret &= zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
+
+		if (WARN(!ret, "cpumask alloc for cpu %d failed!", i))
+			break;
 	}
 	set_cpu_sibling_map(0);
 
Signature · 1 line
1.9.1
#1 lore raw
Quoted 3 lines · Pranith Kumar

zalloc_cpumask_var() can return 0 on allocation failure when CONFIG_CPUMASK_OFFSTACK is set. Check for the return value and WARN() on failure of an allocation in such cases.

And that warning helps in which way?

It just prints a completely useless backtrace and breaks out of the loop, but it does not prevent that later on code will trip over the non allocated per cpu data.

Thanks,

tglx

#2 lore raw
Quoted 11 lines · Thomas Gleixner
Quoted 3 lines · Pranith Kumar

zalloc_cpumask_var() can return 0 on allocation failure when CONFIG_CPUMASK_OFFSTACK is set. Check for the return value and WARN() on failure of an allocation in such cases.

And that warning helps in which way?

It just prints a completely useless backtrace and breaks out of the loop, but it does not prevent that later on code will trip over the non allocated per cpu data.

I agree. May be just a pr_warn() saying that an allocation failed perhaps?

To prevent further accesses, we can clear the cpu bit from the cpu masks(online/possible/present) for the failed cpu and continue trying to allocate for other cpus. We don't break out of the loop. Removing the cpu from the cpu masks will disable accesses of the non allocated per cpu data.

What do you suggest we do in such cases?

Signature · 1 line
Pranith
#3 lore raw
Quoted 6 lines · Pranith Kumar
Quoted 5 lines · Thomas Gleixner
Quoted 3 lines · Pranith Kumar

zalloc_cpumask_var() can return 0 on allocation failure when CONFIG_CPUMASK_OFFSTACK is set. Check for the return value and WARN() on failure of an allocation in such cases.

Quoted 8 lines
Quoted 6 lines

And that warning helps in which way?

It just prints a completely useless backtrace and breaks out of the loop, but it does not prevent that later on code will trip over the non allocated per cpu data.

I agree. May be just a pr_warn() saying that an allocation failed perhaps?

Yep.

Quoted 7 lines

To prevent further accesses, we can clear the cpu bit from the cpu masks(online/possible/present) for the failed cpu and continue trying to allocate for other cpus. We don't break out of the loop. Removing the cpu from the cpu masks will disable accesses of the non allocated per cpu data.

What do you suggest we do in such cases?

Pretty much what you said, but we should definitely break out of the loop. There is no point to try more allocations if we failed the first one.

Thanks,

tglx