If the kernel is compiled with CONFIG_PREEMPT_VOLUNTARY and OP-TEE is
executing a long running workload, the following errors are raised:
[ 1705.971228] rcu: INFO: rcu_sched detected stalls on CPUs/tasks:
[ 1705.977195] rcu: (detected by 0, t=2102 jiffies, g=51977, q=3)
[ 1705.983152] rcu: All QSes seen, last rcu_sched kthread activity 2102 (140596-138494), jiffies_till_next_fqs=1, root ->qsmask 0x0
[ 1705.994729] optee-xtest R running task 0 169 157 0x00000002
While OP-TEE is returning regulary to the kernel due to timer
interrrupts, the OPTEE_SMC_FUNC_FOREIGN_INTR case does not contain an
explicit rescheduling point. Add a might_sleep() to the RPC request case
to ensure that the kernel can reschedule another task if OP-TEE requests
RPC handling.
Signed-off-by: Rouven Czerwinski <[email protected]>
---
drivers/tee/optee/call.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/drivers/tee/optee/call.c b/drivers/tee/optee/call.c
index aa942703ae65..13b0269a0abc 100644
--- a/drivers/tee/optee/call.c
+++ b/drivers/tee/optee/call.c
@@ -148,6 +148,7 @@ u32 optee_do_call_with_arg(struct tee_context *ctx, phys_addr_t parg)
*/
optee_cq_wait_for_completion(&optee->call_queue, &w);
} else if (OPTEE_SMC_RETURN_IS_RPC(res.a0)) {
+ might_sleep();
param.a0 = res.a0;
param.a1 = res.a1;
param.a2 = res.a2;
--
2.20.1
Hi Rouven,
On Wed, Jul 3, 2019 at 12:04 PM Rouven Czerwinski
<[email protected]> wrote:
>
> If the kernel is compiled with CONFIG_PREEMPT_VOLUNTARY and OP-TEE is
> executing a long running workload, the following errors are raised:
>
> [ 1705.971228] rcu: INFO: rcu_sched detected stalls on CPUs/tasks:
> [ 1705.977195] rcu: (detected by 0, t=2102 jiffies, g=51977, q=3)
> [ 1705.983152] rcu: All QSes seen, last rcu_sched kthread activity 2102 (140596-138494), jiffies_till_next_fqs=1, root ->qsmask 0x0
> [ 1705.994729] optee-xtest R running task 0 169 157 0x00000002
>
> While OP-TEE is returning regulary to the kernel due to timer
> interrrupts, the OPTEE_SMC_FUNC_FOREIGN_INTR case does not contain an
> explicit rescheduling point. Add a might_sleep() to the RPC request case
> to ensure that the kernel can reschedule another task if OP-TEE requests
> RPC handling.
>
> Signed-off-by: Rouven Czerwinski <[email protected]>
> ---
> drivers/tee/optee/call.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/drivers/tee/optee/call.c b/drivers/tee/optee/call.c
> index aa942703ae65..13b0269a0abc 100644
> --- a/drivers/tee/optee/call.c
> +++ b/drivers/tee/optee/call.c
> @@ -148,6 +148,7 @@ u32 optee_do_call_with_arg(struct tee_context *ctx, phys_addr_t parg)
> */
> optee_cq_wait_for_completion(&optee->call_queue, &w);
> } else if (OPTEE_SMC_RETURN_IS_RPC(res.a0)) {
> + might_sleep();
> param.a0 = res.a0;
> param.a1 = res.a1;
> param.a2 = res.a2;
> --
> 2.20.1
>
Looks good. I'll pick this up and fix the spell errors in the commit message.
Thanks,
Jens