On 19.10.2023 03:06, Stefano Stabellini wrote:
> On Wed, 18 Oct 2023, Simone Ballarin wrote:
>> --- a/xen/common/sched/core.c
>> +++ b/xen/common/sched/core.c
>> @@ -1504,6 +1504,8 @@ long vcpu_yield(void)
>> {
>> struct vcpu * v=current;
>> spinlock_t *lock;
>> + domid_t domain_id;
>> + int vcpu_id;
>>
>> rcu_read_lock(&sched_res_rculock);
>>
>> @@ -1515,7 +1517,9 @@ long vcpu_yield(void)
>>
>> SCHED_STAT_CRANK(vcpu_yield);
>>
>> - TRACE_2D(TRC_SCHED_YIELD, current->domain->domain_id, current->vcpu_id);
>> + domain_id = current->domain->domain_id;
>> + vcpu_id = current->vcpu_id;
>> + TRACE_2D(TRC_SCHED_YIELD, domain_id, vcpu_id);
>
> Also here it looks like accessing the current pointer is considered a
> side effect. Why? This is a just a global variable that is only accessed
> for reading.
Not exactly. It's a per-CPU variable access on Arm, but involves a
function call on x86. Still that function call has no other side
effects, but I guess Misra wouldn't honor this.
I'm afraid though that the suggested change violates rule 2.2, as
the two new assignments are dead code when !CONFIG_TRACEBUFFER.
Jan
