Interrupt latency benchmark #44
Conversation
rmn30
force-pushed
the
interrupt_latency
branch
3 times, most recently
from
b2963fc
to
6b73198
Compare
sdk/core/scheduler/main.cc
Outdated
| @@ -413,19 +413,19 @@ int __cheri_compartment("sched") | |||
| } | |||
| return typed_op<Event>(evt, [&](Event &event) { | |||
| return event.bits_wait( | |||
| retBits, bitsToWait, clearOnExit, waitAll, timeout); | |||
| retBits, bitsToWait, flags & EventWaitClearOnExit, flags & EventWaitAll, timeout); | |||
There was a problem hiding this comment.
It would be nice to wrap this in the same kind of template that I did for PAL features in snmalloc. Something like:
template<EventWaitFlags Desired>
bool has_event_flag(int eventFlags)
{
return eventFlags & int(Desired);
}
I am always somewhat nervous of raw & for checking bitfields because you're relying on the implicit conversion to bool, which can happen in a different place if you don't put enough brackets around the expression. Putting it in a separate function forces it to happen at the evaluation of the return and better expresses intentionality.
rmn30
force-pushed
the
interrupt_latency
branch
from
6b73198
to
1e7761b
Compare
sdk/include/event.h
Outdated
| @@ -98,3 +121,24 @@ int __cheri_compartment("sched") | |||
| event_delete(struct SObjStruct *heapCapability, void *evt); | |||
|
|
|||
| __END_DECLS | |||
|
|
|||
| #ifdef __cplusplus | |||
There was a problem hiding this comment.
Not sure these should go in the public header: I don't see them as being useful for anything other than the scheduler (a helper for constructing the flags would be).
|
Marking as draft due to unresolved issues especially around |
rmn30
force-pushed
the
interrupt_latency
branch
from
5a529b6
to
7ef216b
Compare
|
After some consideration have decided that we don't need the |
This can mess up the interrupt latency benchmark because an interrupt can fire in call to event_wait_bits even though we called from an interrupt disabled context. We need to evaluate whether we really need to enable intererupts here.
davidchisnall
force-pushed
the
interrupt_latency
branch
from
7ef216b
to
0022710
Compare
No description provided.