Re: fw_devlink on will break all snps,dw-apb-gpio users

[Robin Murphy <robin.murphy@arm.com>]

On 2020-10-15 10:52, Jisheng Zhang wrote:
> On Thu, 15 Oct 2020 01:48:13 -0700
> Saravana Kannan <saravanak@google.com> wrote:
> 
>> On Thu, Oct 15, 2020 at 1:15 AM Jisheng Zhang
>> <Jisheng.Zhang@synaptics.com> wrote:
>>>
>>> On Wed, 14 Oct 2020 22:04:24 -0700 Saravana Kannan wrote:
>>>   
>>>>
>>>>
>>>> On Wed, Oct 14, 2020 at 9:02 PM Jisheng Zhang
>>>> <Jisheng.Zhang@synaptics.com> wrote:
>>>>>
>>>>> On Wed, 14 Oct 2020 10:29:36 -0700
>>>>> Saravana Kannan <saravanak@google.com> wrote:
>>>>>   
>>>>>>
>>>>>>
>>>>>> On Wed, Oct 14, 2020 at 4:12 AM Jisheng Zhang
>>>>>> <Jisheng.Zhang@synaptics.com> wrote:
>>>>>>>
>>>>>>> Hi,
>>>>>>>
>>>>>>> If set fw_devlink as on, any consumers of dw apb gpio won't probe.
>>>>>>>
>>>>>>> The related dts looks like:
>>>>>>>
>>>>>>> gpio0: gpio@2400 {
>>>>>>>         compatible = "snps,dw-apb-gpio";
>>>>>>>         #address-cells = <1>;
>>>>>>>         #size-cells = <0>;
>>>>>>>
>>>>>>>         porta: gpio-port@0 {
>>>>>>>                compatible = "snps,dw-apb-gpio-port";
>>>>>>>                gpio-controller;
>>>>>>>                #gpio-cells = <2>;
>>>>>>>                ngpios = <32>;
>>>>>>>                reg = <0>;
>>>>>>>         };
>>>>>>> };
>>>>>>>
>>>>>>> device_foo {
>>>>>>>          status = "okay"
>>>>>>>          ...;
>>>>>>>          reset-gpio = <&porta, 0, GPIO_ACTIVE_HIGH>;
>>>>>>> };
>>>>>>>
>>>>>>> If I change the reset-gpio property to use another kind of gpio phandle,
>>>>>>> e.g gpio expander, then device_foo can be probed successfully.
>>>>>>>
>>>>>>> The gpio expander dt node looks like:
>>>>>>>
>>>>>>>          expander3: gpio@44 {
>>>>>>>                  compatible = "fcs,fxl6408";
>>>>>>>                  pinctrl-names = "default";
>>>>>>>                  pinctrl-0 = <&expander3_pmux>;
>>>>>>>                  reg = <0x44>;
>>>>>>>                  gpio-controller;
>>>>>>>                  #gpio-cells = <2>;
>>>>>>>                  interrupt-parent = <&portb>;
>>>>>>>                  interrupts = <23 IRQ_TYPE_NONE>;
>>>>>>>                  interrupt-controller;
>>>>>>>                  #interrupt-cells = <2>;
>>>>>>>          };
>>>>>>>
>>>>>>> The common pattern looks like the devlink can't cope with suppliers from
>>>>>>> child dt node.
>>>>>>
>>>>>> fw_devlink doesn't have any problem dealing with child devices being
>>>>>> suppliers. The problem with your case is that the
>>>>>> drivers/gpio/gpio-dwapb.c driver directly parses the child nodes and
>>>>>> never creates struct devices for them. If you have a node with
>>>>>> compatible string, fw_devlink expects you to create and probe a struct
>>>>>> device for it. So change your driver to add the child devices as
>>>>>> devices instead of just parsing the node directly and doing stuff with
>>>>>> it.
>>>>>>
>>>>>> Either that, or stop putting "compatible" string in a node if you
>>>>>> don't plan to actually treat it as a device -- but that's too late for
>>>>>> this driver (it needs to be backward compatible). So change the driver
>>>>>> to add of_platform_populate() and write a driver that probes
>>>>>> "snps,dw-apb-gpio-port".
>>>>>>   
>>>>>
>>>>> Thanks for the information. The "snps,dw-apb-gpio-port" is never used,
>>>>> so I just sent out a series to remove it.
>>>>
>>>> I'd actually prefer that you fix the kernel code to actually use it.
>>>> So that fw_devlink can be backward compatible (Older DT + new kernel).
>>>> The change is pretty trivial (I just have time to do it for you).
>>>>   
>>>
>>> I agree the change is trivial, but it will add some useless LoCs like below.
>>
>> It's not useless if it preserves backward compatibility with DT.
>>
>>> I'm not sure whether this is acceptable.So add GPIO and DT maintainers to comment.
>>>
>>> Hi Linus, Rob,
>>>
>>> Could you please comment? A simple introduction of the problem:
>>>
>>> As pointed out by Saravana, "gpio-dwapb.c driver directly parses the child
>>> nodes and never creates struct devices for them. If you have a node with
>>> compatible string, fw_devlink expects you to create and probe a struct
>>> device for it", so once we set fw_devlink=on, then any users of gpio-dwapb
>>> as below won't be probed.
>>>
>>> device_foo {
>>>           status = "okay"
>>>           ...;
>>>           reset-gpio = <&porta, 0, GPIO_ACTIVE_HIGH>;
>>> };
>>>
>>> The compatible string "snps,dw-apb-gpio-port" is never used, but it's in
>>> the dt-binding since the dw gpio mainlined. I believe the every dw apb
>>> users just copy the compatible string in to soc dtsi. So I submit a series
>>> to remove the unused "snps,dw-apb-gpio-port" https://lkml.org/lkml/2020/10/14/1186
>>> But this will break Older DT + new kernel with fw_devlink on. Which solution
>>> is better?
>>>
>>> If the following patch is acceptable, I can submit it once 5.10-rc1 is out.
>>>
>>> thanks
>>>
>>> diff --git a/drivers/gpio/gpio-dwapb.c b/drivers/gpio/gpio-dwapb.c
>>> index 1d8d55bd63aa..b8e012e48b59 100644
>>> --- a/drivers/gpio/gpio-dwapb.c
>>> +++ b/drivers/gpio/gpio-dwapb.c
>>> @@ -19,6 +19,7 @@
>>>   #include <linux/of_address.h>
>>>   #include <linux/of_device.h>
>>>   #include <linux/of_irq.h>
>>> +#include <linux/of_platform.h>
>>>   #include <linux/platform_device.h>
>>>   #include <linux/property.h>
>>>   #include <linux/reset.h>
>>> @@ -694,6 +695,10 @@ static int dwapb_gpio_probe(struct platform_device *pdev)
>>>          }
>>>          platform_set_drvdata(pdev, gpio);
>>>
>>> +       err = devm_of_platform_populate(dev);
>>> +       if (err)
>>> +               goto out_unregister;
>>> +
>>>          return 0;
>>>
>>>   out_unregister:
>>> @@ -820,6 +825,25 @@ static struct platform_driver dwapb_gpio_driver = {
>>>
>>>   module_platform_driver(dwapb_gpio_driver);
>>>
>>> +static const struct of_device_id dwapb_port_of_match[] = {
>>> +       { .compatible = "snps,dw-apb-gpio-port" },
>>> +       { /* Sentinel */ }
>>> +};
>>> +
>>> +static int dwapb_gpio_port_probe(struct platform_device *pdev)
>>> +{
>>> +       return 0;
>>
>> No, I'm not asking to do a stub/dummy probe. Move the stuff you do
>> inside device_for_each_child_node{} and dwapb_gpio_add_port() into
>> this probe function. Those two pieces of code together are effectively
>> "probing" a separate gpio controller for each of the child nodes. So
>> just create a real struct device (like we do for every other
>> "compatible" DT node) and probe each of them properly using the device
>> driver core.
> 
> Then I believe the modifications are non-trivial. Maybe Linus and Rob
> can comment which way is better, fix the dts or modify the gpio-dwapb.c.
> Personally, I prefer fixing dts, because this doesn't remove or modify
> any used properties or compatible string, it just removes the unused
> compatible string.

You appear to be assuming that:

A) There a no consumers of DTBs and DT bindings other than Linux.
B) No Linux user ever updates their kernel image without also updating 
their DTB.

I can assure you that, in general, neither of those hold true. Hacking 
DTs to work around internal implementation details in Linux is rarely if 
ever a good or even viable idea.

Robin.

> 
> Thanks
> 
> 
>>
>>> +}
>>> +
>>> +static struct platform_driver dwapb_gpio_port_driver = {
>>> +       .driver         = {
>>> +               .name   = "gpio-dwapb-port",
>>> +               .of_match_table = dwapb_port_of_match,
>>> +       },
>>> +       .probe          = dwapb_gpio_port_probe,
>>> +};
>>> +module_platform_driver(dwapb_gpio_port_driver);
>>> +
>>>   MODULE_LICENSE("GPL");
>>>   MODULE_AUTHOR("Jamie Iles");
>>>   MODULE_DESCRIPTION("Synopsys DesignWare APB GPIO driver");
>>>   
> 
> 
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
>