Re: [PATCH v3 2/2] remoteproc: meson-mx-ao-arc: Add a driver for the AO ARC remote procesor

[Martin Blumenstingl]

Hi Mathieu,

On Thu, Aug 5, 2021 at 6:15 PM Mathieu Poirier
<mathieu.poirier@xxxxxxxxxx> wrote:
>
> On Wed, Aug 04, 2021 at 11:03:57PM +0200, Martin Blumenstingl wrote:
> > Hi Mathieu,
> >
> > thanks for taking the time to look into this!
> >
> > (I will address any of your comments that I am not mentioning in this
> > email anymore. Thanks a lot for the suggestions!)
> >
> > On Wed, Jul 28, 2021 at 7:58 PM Mathieu Poirier
> > <mathieu.poirier@xxxxxxxxxx> wrote:
> > [...]
> > > > +     writel(FIELD_PREP(AO_REMAP_REG0_REMAP_AHB_SRAM_BITS_17_14_FOR_ARM_CPU,
> > > > +                            priv->sram_pa >> 14),
> > > Indentation problem
> > The idea here is to align priv->sram_pa with AO_REMAP_REG0... which
> > are both arguments to FIELD_PREP
>
> Right, this is what I would have expected.  When I applied the patch on my side
> "priv->sram_pa ..." was aligned wiht the 'M' of "AO_REMAP_ ...".
>
> > Maybe using something like this will make that easier to read:
> >     tmp = FIELD_PREP(AO_REMAP_REG0_REMAP_AHB_SRAM_BITS_17_14_FOR_ARM_CPU,
> >                                      priv->sram_pa >> 14);
> >     writel(tmp, priv->remap_base + AO_REMAP_REG0);
>
> I think the main problem is that
> AO_REMAP_REG0_REMAP_AHB_SRAM_BITS_17_14_FOR_ARM_CPU is simply too long.  I
> suggest making is shorter and add a comment to describe exactly what it does.
AO_CPU_CNTL_AHB_SRAM_BITS_31_20 is used below and when looking at it
now I think the alignment is also strange.
For the next version I'll go with the tmp variable as I think it
improves readability, even with the long(er) macro names.

[...]
> > > > +     priv->arc_reset = devm_reset_control_get_exclusive(dev, NULL);
> > > > +     if (IS_ERR(priv->arc_reset)) {
> > >
> > > Function __reset_control_get() in __devm_reset_control_get() can return NULL so
> > > this should be IS_ERR_OR_NULL().
> > The logic in there is: return optional ? NULL : ERR_PTR(-...);
>
> Ok, so you meant to do that.  And I just checked reset_control_reset() and it does
> account for a NULL parameter.  I'm good with this one but add a comment to
> make sure future readers don't think you've omitted to properly deal with the
> NULL return value.
>
> > I am requesting a mandatory reset line here, so reset core will never
> > return NULL
> > See also [0]
>
> Indeed, I've read that too.  Nonetheless __reset_control_get() can return NULL
> by way of __reset_control_get_from_lookup().
I could not find where __reset_control_get_from_lookup returns NULL in
case optional is false (which it is in this case because
devm_reset_control_get_exclusive requests a "mandatory" reset line).
Can you please point me to the problematic line(s) as I'd like to send
a patch (which fixes this) to the reset subsystem maintainers

$ git grep -A1 devm_reset_control_get_exclusive | grep IS_ERR_OR_NULL
drivers/remoteproc/ti_k3_r5_remoteproc.c-       if
(IS_ERR_OR_NULL(core->reset)) {
$

I suspect that this can be simplified then as well.


Best regards,
Martin


[0] https://elixir.bootlin.com/linux/v5.14-rc4/source/include/linux/reset.h#L227
[1] https://elixir.bootlin.com/linux/v5.14-rc4/source/drivers/reset/core.c#L932