Skip to contentSkip to conversation

percpu: fixes for v2.6.36-rc5

GIT PULL Tejun Heo 1 participant 1 message Sep 23, 2010

Subsystems (2)
lore
Share & tools
lore

Tejun Heo

author maintainer
#0 lore raw

Hello, Linus.

Please pull from the following branch to receive a percpu fix for v2.6.36-rc5.

git://git.kernel.org/pub/scm/linux/kernel/git/tj/percpu.git for-linus

It contains only one patch to fix incorrect pcpu_last_unit_cpu initialization. When the number of possible CPUs isn't power of two, there can be unused holes at the end of each percpu chunk. Percpu init code uses NR_CPUS to mark these holes. pcpu_last_unit_cpu should contain the CPU which maps to the last unit in the chunk but when the above condition occurs, it contained NR_CPUS instead, which makes percpu code miscalculate the address range covered by a chunk for per_cpu_ptr_to_phys() and tlb and v[un]map flushses and has potential to cause weird issues especially on virtual cache architectures.

This problem was discovered by CAI Qian during kvm kdump testing. Kudos to him for finding this subtle problem.

Thanks.

Tejun Heo (1):
percpu: fix pcpu_last_unit_cpu

 mm/percpu.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
mm/percpu.c +1 -1
index 58c572b..c76ef38 100644
@@ -1401,9 +1401,9 @@ int __init pcpu_setup_first_chunk(const struct pcpu_alloc_info *ai,
 
 			if (pcpu_first_unit_cpu == NR_CPUS)
 				pcpu_first_unit_cpu = cpu;
+			pcpu_last_unit_cpu = cpu;
 		}
 	}
-	pcpu_last_unit_cpu = cpu;
 	pcpu_nr_units = unit;
 
 	for_each_possible_cpu(cpu)