scx_cosmos: Add null check for scx_bpf_cpu_rq() #2587
Conversation
scx_bpf_cpu_rq() may return NULL so check for it. Additionally also add a NULL check for ->curr. Fixes: a937a24 ("scx_cosmos: Introduce flat idle CPU scan") Reported-by: Jake Hillion <[email protected]> Closes: https://lore.kernel.org/lkml/y23etey3foin5nrxgj6e4g373b3ap6oxqa5rrvuvwyus3umw5s@bgh3d6uuga5t/ Signed-off-by: Christian Loehle <[email protected]>
cloehle
force-pushed
the
cloehle/scx_cosmos-null-idle-check
branch
from
1559a4b to
c0d1e10
Compare
There was a problem hiding this comment.
This will unblock #2584 to update the sched_ext/for-next kernel, and this does return NULL sometimes so should be checked. As mentioned in https://lore.kernel.org/sched-ext/aJIWHk_4IuOxyh5d@gpd4/T/#m8e663f569b713377557976be11c4ce20ff88be07 though, I think we should implement these changes in a way that doesn't break existing scheduler binaries for a couple of kernel versions.
|
@cloehle does this change still make sense with the differences in the kernel patches? Happy to merge if so. |
|
Hm.. this would still break the logic in the scheduler, because it would consider CPUs as busy (in the NULL case) while they're not. What we should do instead is introduce a compat helper to check if the new API to get the curr task is available and rely on that. And if we get NULL I'd rather error the scheduler or drop this logic completely. |
scx_bpf_cpu_rq() may return NULL so check for it.
Additionally also add a NULL check for ->curr.