mtk: fix memory allocation problem #8385
Conversation
When I was using xtensa to build mt8188 platform, I encountered an error: 'redefinition of i'. To fix the problem, I removed the declaration in the for loop. Signed-off-by: Trevor Wu <[email protected]>
Because the size of struct comp_buffer has increased, the current heap layout for the basic use case is not enough. Some use cases are experiencing out of memory issues, c0 memory src/lib/alloc.c:765 ERROR failed to alloc 0x180 bytes zone 0x4 caps 0x1 flags 0x0 To fix this issue, we increase the number of 512-byte heap blocks to 32. Signed-off-by: Trevor Wu <[email protected]>
wenliangwu
requested review from
yaochunhung,
kuanhsuncheng,
ranj063,
jxstelter,
lgirdwood,
plbossart,
mmaka1,
lbetlej,
dbaluta and
kv2019i
as code owners
| @@ -1450,9 +1450,9 @@ int module_adapter_reset(struct comp_dev *dev) | |||
| #endif /* CONFIG_ZEPHYR_DP_SCHEDULER */ | |||
| if (IS_PROCESSING_MODE_SINK_SOURCE(mod)) { | |||
| /* for both LL and DP processing */ | |||
| for (int i = 0; i < mod->num_of_sources; i++) | |||
| for (i = 0; i < mod->num_of_sources; i++) | |||
There was a problem hiding this comment.
@singalsu @iuliana-prodan did we not recently change this in another place ? I think the newer xtensa compilers are possibly stricter here and hence we are hitting this now.
| @@ -115,7 +115,7 @@ | |||
| #define HEAP_RT_COUNT64 32 | |||
| #define HEAP_RT_COUNT128 32 | |||
| #define HEAP_RT_COUNT256 32 | |||
| #define HEAP_RT_COUNT512 4 | |||
| #define HEAP_RT_COUNT512 32 | |||
There was a problem hiding this comment.
Fwiw, Intel has not had to make a change like this since switching RTOS to Zephyr.
There was a problem hiding this comment.
https://github.com/thesofproject/sof/blob/main/src/platform/library/include/platform/lib/memory.h#L97
is this intel's definition? still used by ipc4 build?
Btw, 4-->32 too big? do you observe around 15k bytes needed to enlarge?
There was a problem hiding this comment.
After confirming that 15k is only a small portion of the heap buffer, I am uncertain whether there may be other use cases that require a larger buffer size. Therefore, I decided to follow the solution in #7982.
|
@wszypelt There is one Intel test case failing (test_003_01_kd_topology_host_gw), but as this PR doesn't really touch any Intel code, I can't see how this is related. So can I proceed to merge, or can you rerun the test just to be sure? |
|
@kv2019i I'll try to test PR again, results should be available within 30 minutes |
Similar to the issue described in #7982, MediaTek is also facing out-of-memory problems in our use cases. We have implemented the same solution to address this issue.
Additionally, I fixed the compilation error "redefinition of i" in module adaptor while using xtensa toolchain.