by Colin Foster

[permalink] [raw]

Subject: Re: [RFC v5 net-next 08/13] mfd: add interface to check whether a device is mfd

On Thu, Dec 30, 2021 at 01:43:53PM +0000, Lee Jones wrote:
> On Wed, 29 Dec 2021, Colin Foster wrote:
>
> > On Wed, Dec 29, 2021 at 03:25:55PM +0000, Lee Jones wrote:
> > > On Sat, 18 Dec 2021, Colin Foster wrote:
> > >
> > > > Some drivers will need to create regmaps differently based on whether they
> > > > are a child of an MFD or a standalone device. An example of this would be
> > > > if a regmap were directly memory-mapped or an external bus. In the
> > > > memory-mapped case a call to devm_regmap_init_mmio would return the correct
> > > > regmap. In the case of an MFD, the regmap would need to be requested from
> > > > the parent device.
> > > >
> > > > This addition allows the driver to correctly reason about these scenarios.
> > > >
> > > > Signed-off-by: Colin Foster <[email protected]>
> > > > ---
> > > > drivers/mfd/mfd-core.c | 5 +++++
> > > > include/linux/mfd/core.h | 10 ++++++++++
> > > > 2 files changed, 15 insertions(+)
> > > >
> > > > diff --git a/drivers/mfd/mfd-core.c b/drivers/mfd/mfd-core.c
> > > > index 684a011a6396..905f508a31b4 100644
> > > > --- a/drivers/mfd/mfd-core.c
> > > > +++ b/drivers/mfd/mfd-core.c
> > > > @@ -33,6 +33,11 @@ static struct device_type mfd_dev_type = {
> > > > .name = "mfd_device",
> > > > };
> > > >
> > > > +int device_is_mfd(struct platform_device *pdev)
> > > > +{
> > > > + return (!strcmp(pdev->dev.type->name, mfd_dev_type.name));
> > > > +}
> > > > +
> > >
> > > Why is this device different to any other that has ever been
> > > mainlined?
> >
> > Hi Lee,
> >
> > First, let me apologize for not responding to your response from the
> > related RFC from earlier this month. It had been blocked by my spam
> > filter and I had not seen it until just now. I'll have to check that
> > more diligently now.
> >
> > Moving on...
> >
> > That's a question I keep asking myself. Either there's something I'm
> > missing, or there's something new I'm doing.
> >
> > This is taking existing drivers that work via MMIO regmaps and making
> > them interface-independent. As Vladimir pointed out here:
> > https://lore.kernel.org/all/20211204022037.dkipkk42qet4u7go@skbuf/T/
> > device_is_mfd could be dropped in lieu of an mfd-specific probe
> > function.
> >
> > If there's something I'm missing, please let me know. But it feels like
> > devm_get_regmap_from_resource at the end of the day would be the best
> > solution to the design, and that doesn't exist. And implementing
> > something like that is a task that I feel I'm not capable of tackling at
> > this time.
>
> I'm really not a fan of leaking any MFD API outside of drivers/mfd.
> MFD isn't a tangible thing. It's a Linuxiusm, something we made up, a
> figment of your imagination.
>
> What happens if you were to all dev_get_regmap() in the non-MFD case
> or when you call devm_regmap_init_mmio() when the driver was
> registered via the MFD framework?

I'd imagine dev_get_regmap in a non-MFD case would be the same as
dev_get_and_ioremap_resource() followed by devm_regmap_init_mmio().

In the MFD case it would possibly request the regmap from the parent,
which could reason about how to create the regmap. As you understand,
this is exactly the behavior I created in this patch set. I did it by
way of ocelot_get_regmap_from_resource, and admit it isn't the best way.
But it certainly seems there isn't an existing method that I'm missing.

I'm coming from a pretty narrow field of view, but believe my use-case
is a valid one. If that is true, and there isn't another design I should
use... this is the opportunity to create it. Implementing
ocelot_get_regmap_from_resource is a way to achieve my needs without
affecting anyone else.

Going one step further and implementing mfd_get_regmap_from_parent (or
similar) would creep into the design of MFD. I don't know enough about
MFD and the users to suggest this. I wouldn't want to start venturing
down that path without blessing from the community. And this would
indirectly affect every MFD driver.

Going all in and implementing device_get_regmap_from_resource... I don't
know that I'd be comfortable even starting down that path knowing that
it would affect every device. Perhaps it would have to utilize something
like IORESOURCE_REG that seems to only get utilized in a handful of
files:
https://elixir.bootlin.com/linux/v5.16-rc7/C/ident/IORESOURCE_REG

>
> --
> Lee Jones [李琼斯]
> Senior Technical Lead - Developer Services
> Linaro.org │ Open source software for Arm SoCs
> Follow Linaro: Facebook | Twitter | Blog

2022-01-10 12:23:23

by Lee Jones

[permalink] [raw]

Subject: Re: [RFC v5 net-next 08/13] mfd: add interface to check whether a device is mfd

On Thu, 30 Dec 2021, Colin Foster wrote:

> On Thu, Dec 30, 2021 at 01:43:53PM +0000, Lee Jones wrote:
> > On Wed, 29 Dec 2021, Colin Foster wrote:
> >
> > > On Wed, Dec 29, 2021 at 03:25:55PM +0000, Lee Jones wrote:
> > > > On Sat, 18 Dec 2021, Colin Foster wrote:
> > > >
> > > > > Some drivers will need to create regmaps differently based on whether they
> > > > > are a child of an MFD or a standalone device. An example of this would be
> > > > > if a regmap were directly memory-mapped or an external bus. In the
> > > > > memory-mapped case a call to devm_regmap_init_mmio would return the correct
> > > > > regmap. In the case of an MFD, the regmap would need to be requested from
> > > > > the parent device.
> > > > >
> > > > > This addition allows the driver to correctly reason about these scenarios.
> > > > >
> > > > > Signed-off-by: Colin Foster <[email protected]>
> > > > > ---
> > > > > drivers/mfd/mfd-core.c | 5 +++++
> > > > > include/linux/mfd/core.h | 10 ++++++++++
> > > > > 2 files changed, 15 insertions(+)
> > > > >
> > > > > diff --git a/drivers/mfd/mfd-core.c b/drivers/mfd/mfd-core.c
> > > > > index 684a011a6396..905f508a31b4 100644
> > > > > --- a/drivers/mfd/mfd-core.c
> > > > > +++ b/drivers/mfd/mfd-core.c
> > > > > @@ -33,6 +33,11 @@ static struct device_type mfd_dev_type = {
> > > > > .name = "mfd_device",
> > > > > };
> > > > >
> > > > > +int device_is_mfd(struct platform_device *pdev)
> > > > > +{
> > > > > + return (!strcmp(pdev->dev.type->name, mfd_dev_type.name));
> > > > > +}
> > > > > +
> > > >
> > > > Why is this device different to any other that has ever been
> > > > mainlined?
> > >
> > > Hi Lee,
> > >
> > > First, let me apologize for not responding to your response from the
> > > related RFC from earlier this month. It had been blocked by my spam
> > > filter and I had not seen it until just now. I'll have to check that
> > > more diligently now.
> > >
> > > Moving on...
> > >
> > > That's a question I keep asking myself. Either there's something I'm
> > > missing, or there's something new I'm doing.
> > >
> > > This is taking existing drivers that work via MMIO regmaps and making
> > > them interface-independent. As Vladimir pointed out here:
> > > https://lore.kernel.org/all/20211204022037.dkipkk42qet4u7go@skbuf/T/
> > > device_is_mfd could be dropped in lieu of an mfd-specific probe
> > > function.
> > >
> > > If there's something I'm missing, please let me know. But it feels like
> > > devm_get_regmap_from_resource at the end of the day would be the best
> > > solution to the design, and that doesn't exist. And implementing
> > > something like that is a task that I feel I'm not capable of tackling at
> > > this time.
> >
> > I'm really not a fan of leaking any MFD API outside of drivers/mfd.
> > MFD isn't a tangible thing. It's a Linuxiusm, something we made up, a
> > figment of your imagination.
> >
> > What happens if you were to all dev_get_regmap() in the non-MFD case
> > or when you call devm_regmap_init_mmio() when the driver was
> > registered via the MFD framework?
>
> I'd imagine dev_get_regmap in a non-MFD case would be the same as
> dev_get_and_ioremap_resource() followed by devm_regmap_init_mmio().
>
> In the MFD case it would possibly request the regmap from the parent,
> which could reason about how to create the regmap. As you understand,
> this is exactly the behavior I created in this patch set. I did it by
> way of ocelot_get_regmap_from_resource, and admit it isn't the best way.
> But it certainly seems there isn't an existing method that I'm missing.
>
> I'm coming from a pretty narrow field of view, but believe my use-case
> is a valid one. If that is true, and there isn't another design I should
> use... this is the opportunity to create it. Implementing
> ocelot_get_regmap_from_resource is a way to achieve my needs without
> affecting anyone else.
>
> Going one step further and implementing mfd_get_regmap_from_parent (or
> similar) would creep into the design of MFD. I don't know enough about
> MFD and the users to suggest this. I wouldn't want to start venturing
> down that path without blessing from the community. And this would
> indirectly affect every MFD driver.
>
> Going all in and implementing device_get_regmap_from_resource... I don't
> know that I'd be comfortable even starting down that path knowing that
> it would affect every device. Perhaps it would have to utilize something
> like IORESOURCE_REG that seems to only get utilized in a handful of
> files:
> https://elixir.bootlin.com/linux/v5.16-rc7/C/ident/IORESOURCE_REG

Let's speak to Mark and see if he can provide any insight.

--
Lee Jones [李琼斯]
Principal Technical Lead - Developer Services
Linaro.org │ Open source software for Arm SoCs
Follow Linaro: Facebook | Twitter | Blog