zephyr: lib: alloc: k_panic is not required in failure scenario #8853
Conversation
For rballoc_allign() call when caps are not correct it is enough to return error. k_panic() call is not required here. Previous change exposed this issue: thesofproject#8832, but it is sufficient to log error and return NULL at this point. Signed-off-by: Jaroslaw Stelter <Jaroslaw.Stelter@intel.com>
jxstelter
force-pushed
the
prv-jstelter-rballoc-fix
branch
from
5f267da
to
6fc162f
Compare
| @@ -388,7 +388,8 @@ void *rballoc_align(uint32_t flags, uint32_t caps, size_t bytes, | |||
| heap = &l3_heap; | |||
| return (__sparse_force void *)l3_heap_alloc_aligned(heap, align, bytes); | |||
| #else | |||
| k_panic(); | |||
| tr_err(&zephyr_tr, "L3_HEAP not available."); | |||
| return NULL; | |||
There was a problem hiding this comment.
will the propagate all the way up to the host?
There was a problem hiding this comment.
The default result of this call was to return NULL on allocation failure. I added k_panic() when L3_HEAP support was added to rballoc_allign(). Although this change exposed a bug in the IPC3 handling code, it seems that returning an allocation error is sufficient here.
There was a problem hiding this comment.
Sorry, not what I meant, I skipped the word "error". I mean't will the host properly be notified of this new allocation error path should it be hit.
|
I just spotted this only by chance while looking for something else. I could have lost a fair amount of time trying to do something else in the same area. Please mention this PR in the bug: #8832 |
@marc-hb it is mentioned in the bug, it was crosslinked when @jxstelter wrote the bug number in the first comment, it just doesn't generate a notification |
My point exactly. I know a lot about this issue right now, the context is still fresh in my mind so I wouldn't need to read that bug again. |
For rballoc_allign() call when caps are not correct it is enough to return error. k_panic() call is not required here. Previous change exposed this issue: #8832, but it is sufficient to log error and return NULL at this point.
Signed-off-by: Jaroslaw Stelter Jaroslaw.Stelter@intel.com