Re: [PATCH v2 2/2] misc: sram: Generate unique names for subpools

From: Dmitry Osipenko
Date: Tue Jun 20 2023 - 04:23:56 EST


19.06.2023 10:11, Linus Walleij пишет:
> On Sun, Jun 18, 2023 at 11:33 PM Dmitry Osipenko <digetx@xxxxxxxxx> wrote:
>
>>> if (!label)
>>> - label = child->name;
>>> -
>>> - block->label = devm_kstrdup(sram->dev,
>>> - label, GFP_KERNEL);
>>> + block->label = devm_kasprintf(sram->dev, GFP_KERNEL,
>>> + "%s", dev_name(sram->dev));
>>
>> This broke device-trees that have no label property.
>
> Which system is affected? Asking so I can inspect the DTS file
> and figure out how this needs to work.

NVIDIA Tegra2/3 video decoder driver fails to probe with this change.

https://elixir.bootlin.com/linux/latest/source/drivers/media/platform/nvidia/tegra-vde/vde.c#L312

https://elixir.bootlin.com/linux/latest/source/arch/arm/boot/dts/tegra20.dtsi#L347

>> The SRAM DT binding says:
>>
>> "
>> label:
>> description:
>> The name for the reserved partition, if omitted, the label is taken
>> from the node name excluding the unit address.
>> "
>>
>> Not sure whether breakage was on purpose, otherwise doc needs to be
>> updated or there should be explicit check for the duplicated node names.
>>
>> Secondly, AFAICS, the dev_name(sram->dev) is the name of the parent SRAM
>> device and not of the children sub-nodes, hence it's now always the same
>> dev_name(sram->dev) for all sub-nodes.
>
> Sounds like I should go back to the original approach in patch v1:
> https://lore.kernel.org/linux-devicetree/20230417-ux500-sram-v1-2-5924988bb835@xxxxxxxxxx/
>
> and also augment the DTS binding text to say it uses the full node name
> including the address.
>
> Does that look OK to you, or will this regress your system as well?

That may work, but then seems you'll also need to update
of_gen_pool_get() to use np_pool->full_name instead of np_pool->name.

https://elixir.bootlin.com/linux/latest/source/lib/genalloc.c#L898