On 12/04/18 00:46, Stefano Stabellini wrote:
On Wed, 11 Apr 2018, Julien Grall wrote:
On 11/04/18 14:19, Mirela Simonovic wrote:
Freeing percpu area is done when a non-boot CPU is disabled upon suspend.
This use to be scheduled for execution after a period of time, what caused
the following racing issues. If CPU is enabled after it is disabled and
before the freeing of percpu area is performed, Xen would crash upon
initializing percpu area because per cpu offset is not marked as
INVALID_PERCPU_AREA (this suppose to happen when cpu area is freed).
To resolve the racing issue, free percpu area right away instead
scheduling it for later.
The reason of using the RCU is you want to make sure that none of the other
CPUs will access that percpu data before freeing it. So I don't think this
patch is valid.
It looks like to me a rcu barrier is missing after calling cpu_down somewhere
in the CPU off path. I am not entirely sure where.
We need a rcu_barrier(). Perhaps, it could be added on cpu_on before
initializing the percpu area?
Do you mind giving a bit more details on your thought? cpu_up looks a
strange place as no one should access the percpu area after the CPU is
down. So it feels the rcu_barrier should be somewhere before
PSCI_cpu_off is called.
Signed-off-by: Mirela Simonovic <mirela.simono...@aggios.com>
xen/arch/arm/percpu.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/xen/arch/arm/percpu.c b/xen/arch/arm/percpu.c
index 25442c48fe..e4e8405f43 100644
@@ -46,7 +46,7 @@ static void free_percpu_area(unsigned int cpu)
struct free_info *info = &per_cpu(free_info, cpu);
info->cpu = cpu;
- call_rcu(&info->rcu, _free_percpu_area);
static int cpu_percpu_callback(
Xen-devel mailing list