Fix memory leaks with zephyr native drivers #6540
Conversation
Free the memory allocated for DMA channels when the ref count is 0 when using native zephyr drivers. Signed-off-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
ranj063
requested review from
dbaluta,
lbetlej,
lgirdwood,
mmaka1 and
plbossart
as code owners
ranj063
changed the title
Return NULL when dma_init() fails. Signed-off-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
This prevents memory leaks during repeated PCM start/stop tests. Signed-off-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>
| if (--dma->sref == 0) { | ||
| rfree(dma->chan); | ||
| dma->chan = NULL; | ||
| } |
There was a problem hiding this comment.
Without this, we lose one "dma->chan = rzalloc(SOF_MEM_ZONE_RUNTIME_SHARED" worth of memory whenever sref goes to 1->0 and back.
aborisovich
left a comment
aborisovich
left a comment
There was a problem hiding this comment.
Great find but is really using goto statement mandatory?
It is commonly known fact that this is not a clean code solution...
In fact, it is one of the classical examples of bad code layout. There are too many books written about it,
|
@aborisovich wrote:
Goto usage is pretty widespread in Linux kernel, but only for very specific usage/patterns. I think this falls to one of the well accepted patterns where goto is used to jump to error handler (kind of poor man's exception handler) to make it more explicit that locks are released correctly. |
|
Still 01.org trouble, checked the pr-device-test results offline and they are good (a few DUTs unavailable for the run, but we have test results with similar configurations that pass). Proceeding with merge. |
6 participants
Free the memory allocated for DMA channels when the ref count is 0 and the DMA block config during DAI reset when using natize zephyr drivers to prevent memory leaks seen during stress tests
Fixes #6535