CMA Regression issue - [PATCH v2] of: reserved_mem: Restructure code to call reserved mem init functions earlier

[Oreoluwa Babatunde]

On 7/4/2024 4:41 AM, Aisheng Dong wrote:
> Hi Oreoluwa,
>
> We observed a CMA regression issue on MX8MN that the CMA region failed to be
> setup by rmem_cma_setup() since the following patch:
> f2a524d9ef5b ("of: reserved_mem: Restructure code to call reserved mem init functions earlier") [1]
> And the CMA node was incorrect parsed (e.g. non-reuseable) due to NULL node pointer.
> Below is one line boot log tested with the tag next-20240703 of linux-next tree.
> OF: reserved mem: 0x0000000058000000..0x000000007fffffff (655360 KiB) map non-reusable linux,cma
> NOTE: actually there is a reusable property:
> linux,cma {
>         compatible = "shared-dma-pool";
>         reusable;
>         size = <0 0x28000000>;
>         alloc-ranges = <0 0x40000000 0 0x40000000>;
>         linux,cma-default;
> };
>
> The root cause was that the CMA region was initialized at a very early stage
> since above commit in question.
> The call flow became:
> early_init_fdt_scan_reserved_mem()->fdt_scan_reserved_mem()->
> __reserved_mem_alloc_size()->of_reserved_mem_save_node(NULL, xxx)->
> of_init_reserved_mem_node()->__reserved_mem_init_node().
>
> The rmem->dev_node is still NULL when calling __reserved_mem_init_node()
> which fails to call the rmem_cma_setup().
>
> Please let me know if there's already a fix patch.
>
> [1] https://lore.kernel.org/all/20240620001027.2326275-1-quic_obabatun@quicinc.com/
>
> Regards
> Aisheng
Hi Aisheng,

I have uploaded a fix here:
https://lore.kernel.org/all/20240708230613.448846-1-quic_obabatun@quicinc.com/

Please test and see if this fixes the issue.

Thank you!
Oreoluwa