[RFC PATCH] mfd: mfd-core: inherit only valid dma_masks/flags from parent

[RFC PATCH] mfd: mfd-core: inherit only valid dma_masks/flags from parent

[Michael Walle]

Only copy the dma_masks and flags from the parent device, if the parent
has a valid dma_mask/flags. Commit cdfee5623290 ("driver core:
initialize a default DMA mask for platform device") initialize the DMA
masks of a platform device. But if the parent doesn't have a dma_mask
set, for example if it's an I2C device, the dma_mask of the child
platform device will be set to zero again. Which leads to many "DMA mask
not set" warnings, if the MFD cell has the of_compatible property set.

[    1.877937] sl28cpld-pwm sl28cpld-pwm: DMA mask not set
[    1.883282] sl28cpld-pwm sl28cpld-pwm.0: DMA mask not set
[    1.888795] sl28cpld-gpio sl28cpld-gpio: DMA mask not set

Thus a MFD child should just inherit valid dma_masks and keep the
platform default otherwise.

Cc: Christoph Hellwig <hch@lst.de>
Cc: Rob Herring <robh@kernel.org>
Cc: Robin Murphy <robin.murphy@arm.com>
Cc: Russell King <linux@armlinux.org.uk>
Signed-off-by: Michael Walle <michael@walle.cc>
---

Hi,

I don't know if that is the correct way of handling things. Maybe I'm
also doing something wrong in my driver, I had a look at other I2C MFD
drivers but couldn't find a clue why they shouldn't have the same
problem.

There was also a thread [1] about this topic, but there seems to be no
conclusion.

[1] https://www.spinics.net/lists/linux-renesas-soc/msg31581.html

 drivers/mfd/mfd-core.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/drivers/mfd/mfd-core.c b/drivers/mfd/mfd-core.c
index b9eb8f40c073..5d8ea5e8e93c 100644
--- a/drivers/mfd/mfd-core.c
+++ b/drivers/mfd/mfd-core.c
@@ -139,9 +139,12 @@ static int mfd_add_device(struct device *parent, int id,
 
 	pdev->dev.parent = parent;
 	pdev->dev.type = &mfd_dev_type;
-	pdev->dev.dma_mask = parent->dma_mask;
-	pdev->dev.dma_parms = parent->dma_parms;
-	pdev->dev.coherent_dma_mask = parent->coherent_dma_mask;
+	if (parent->dma_mask)
+		pdev->dev.dma_mask = parent->dma_mask;
+	if (parent->dma_parms)
+		pdev->dev.dma_parms = parent->dma_parms;
+	if (parent->coherent_dma_mask)
+		pdev->dev.coherent_dma_mask = parent->coherent_dma_mask;
 
 	ret = regulator_bulk_register_supply_alias(
 			&pdev->dev, cell->parent_supplies,
-- 
2.20.1

Re: [RFC PATCH] mfd: mfd-core: inherit only valid dma_masks/flags from parent

[Christoph Hellwig]

On Wed, Mar 11, 2020 at 12:09:35AM +0100, Michael Walle wrote:
> Only copy the dma_masks and flags from the parent device, if the parent
> has a valid dma_mask/flags. Commit cdfee5623290 ("driver core:
> initialize a default DMA mask for platform device") initialize the DMA
> masks of a platform device. But if the parent doesn't have a dma_mask
> set, for example if it's an I2C device, the dma_mask of the child
> platform device will be set to zero again. Which leads to many "DMA mask
> not set" warnings, if the MFD cell has the of_compatible property set.
> 
> [    1.877937] sl28cpld-pwm sl28cpld-pwm: DMA mask not set
> [    1.883282] sl28cpld-pwm sl28cpld-pwm.0: DMA mask not set
> [    1.888795] sl28cpld-gpio sl28cpld-gpio: DMA mask not set
> 
> Thus a MFD child should just inherit valid dma_masks and keep the
> platform default otherwise.
> 
> Cc: Christoph Hellwig <hch@lst.de>
> Cc: Rob Herring <robh@kernel.org>
> Cc: Robin Murphy <robin.murphy@arm.com>
> Cc: Russell King <linux@armlinux.org.uk>
> Signed-off-by: Michael Walle <michael@walle.cc>
> ---
> 
> Hi,
> 
> I don't know if that is the correct way of handling things. Maybe I'm
> also doing something wrong in my driver, I had a look at other I2C MFD
> drivers but couldn't find a clue why they shouldn't have the same
> problem.
> 
> There was also a thread [1] about this topic, but there seems to be no
> conclusion.
> 
> [1] https://www.spinics.net/lists/linux-renesas-soc/msg31581.html
> 
>  drivers/mfd/mfd-core.c | 9 ++++++---
>  1 file changed, 6 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/mfd/mfd-core.c b/drivers/mfd/mfd-core.c
> index b9eb8f40c073..5d8ea5e8e93c 100644
> --- a/drivers/mfd/mfd-core.c
> +++ b/drivers/mfd/mfd-core.c
> @@ -139,9 +139,12 @@ static int mfd_add_device(struct device *parent, int id,
>  
>  	pdev->dev.parent = parent;
>  	pdev->dev.type = &mfd_dev_type;
> -	pdev->dev.dma_mask = parent->dma_mask;
> -	pdev->dev.dma_parms = parent->dma_parms;
> -	pdev->dev.coherent_dma_mask = parent->coherent_dma_mask;
> +	if (parent->dma_mask)
> +		pdev->dev.dma_mask = parent->dma_mask;
> +	if (parent->dma_parms)
> +		pdev->dev.dma_parms = parent->dma_parms;

Both of these are pointers, and we can't just share them.  You need
to allocate storage for them and the allocate the values over.

Re: [RFC PATCH] mfd: mfd-core: inherit only valid dma_masks/flags from parent

[Michael Walle]

Am 2020-03-11 07:18, schrieb Christoph Hellwig:
> On Wed, Mar 11, 2020 at 12:09:35AM +0100, Michael Walle wrote:
>> Only copy the dma_masks and flags from the parent device, if the 
>> parent
>> has a valid dma_mask/flags. Commit cdfee5623290 ("driver core:
>> initialize a default DMA mask for platform device") initialize the DMA
>> masks of a platform device. But if the parent doesn't have a dma_mask
>> set, for example if it's an I2C device, the dma_mask of the child
>> platform device will be set to zero again. Which leads to many "DMA 
>> mask
>> not set" warnings, if the MFD cell has the of_compatible property set.
>> 
>> [    1.877937] sl28cpld-pwm sl28cpld-pwm: DMA mask not set
>> [    1.883282] sl28cpld-pwm sl28cpld-pwm.0: DMA mask not set
>> [    1.888795] sl28cpld-gpio sl28cpld-gpio: DMA mask not set
>> 
>> Thus a MFD child should just inherit valid dma_masks and keep the
>> platform default otherwise.
>> 
>> Cc: Christoph Hellwig <hch@lst.de>
>> Cc: Rob Herring <robh@kernel.org>
>> Cc: Robin Murphy <robin.murphy@arm.com>
>> Cc: Russell King <linux@armlinux.org.uk>
>> Signed-off-by: Michael Walle <michael@walle.cc>
>> ---
>> 
>> Hi,
>> 
>> I don't know if that is the correct way of handling things. Maybe I'm
>> also doing something wrong in my driver, I had a look at other I2C MFD
>> drivers but couldn't find a clue why they shouldn't have the same
>> problem.
>> 
>> There was also a thread [1] about this topic, but there seems to be no
>> conclusion.
>> 
>> [1] https://www.spinics.net/lists/linux-renesas-soc/msg31581.html
>> 
>>  drivers/mfd/mfd-core.c | 9 ++++++---
>>  1 file changed, 6 insertions(+), 3 deletions(-)
>> 
>> diff --git a/drivers/mfd/mfd-core.c b/drivers/mfd/mfd-core.c
>> index b9eb8f40c073..5d8ea5e8e93c 100644
>> --- a/drivers/mfd/mfd-core.c
>> +++ b/drivers/mfd/mfd-core.c
>> @@ -139,9 +139,12 @@ static int mfd_add_device(struct device *parent, 
>> int id,
>> 
>>  	pdev->dev.parent = parent;
>>  	pdev->dev.type = &mfd_dev_type;
>> -	pdev->dev.dma_mask = parent->dma_mask;
>> -	pdev->dev.dma_parms = parent->dma_parms;
>> -	pdev->dev.coherent_dma_mask = parent->coherent_dma_mask;
>> +	if (parent->dma_mask)
>> +		pdev->dev.dma_mask = parent->dma_mask;
>> +	if (parent->dma_parms)
>> +		pdev->dev.dma_parms = parent->dma_parms;
> 
> Both of these are pointers, and we can't just share them.  You need
> to allocate storage for them and the allocate the values over.

So they were alreay copied wrong before this patch? If so, should
there be a fixes patch for it? The commit who introduced the copy
dates back to 2013:
   b018e1361bad3 mfd: core: Copy DMA mask and params from parent

-michael

Re: [RFC PATCH] mfd: mfd-core: inherit only valid dma_masks/flags from parent

[Robin Murphy]

On 2020-03-10 11:09 pm, Michael Walle wrote:
> Only copy the dma_masks and flags from the parent device, if the parent
> has a valid dma_mask/flags. Commit cdfee5623290 ("driver core:
> initialize a default DMA mask for platform device") initialize the DMA
> masks of a platform device. But if the parent doesn't have a dma_mask
> set, for example if it's an I2C device, the dma_mask of the child
> platform device will be set to zero again. Which leads to many "DMA mask
> not set" warnings, if the MFD cell has the of_compatible property set.
> 
> [    1.877937] sl28cpld-pwm sl28cpld-pwm: DMA mask not set
> [    1.883282] sl28cpld-pwm sl28cpld-pwm.0: DMA mask not set
> [    1.888795] sl28cpld-gpio sl28cpld-gpio: DMA mask not set
> 
> Thus a MFD child should just inherit valid dma_masks and keep the
> platform default otherwise.
> 
> Cc: Christoph Hellwig <hch@lst.de>
> Cc: Rob Herring <robh@kernel.org>
> Cc: Robin Murphy <robin.murphy@arm.com>
> Cc: Russell King <linux@armlinux.org.uk>
> Signed-off-by: Michael Walle <michael@walle.cc>
> ---
> 
> Hi,
> 
> I don't know if that is the correct way of handling things. Maybe I'm
> also doing something wrong in my driver, I had a look at other I2C MFD
> drivers but couldn't find a clue why they shouldn't have the same
> problem.

The underlying issue is that about 99% of MFD children should not be 
going through dma_configure() at all because their parent 'real' device 
is not on a DMA-capable bus, but as they are platform devices we are 
forced to give them the benefit of the doubt. For DT systems the only 
vaguely-reasonable heuristic to distinguish between "platform" meaning 
"SoC memory-mapped device" and "platform" meaning "random crap made up 
by Linux" is whether the device has a populated OF node, but MFD's trick 
of hanging the parent device's OF node onto its synthesised children 
kicks a hole right through even that.

Modulo any other concerns with the existing code, does the change below 
make things work the way you want? It's still a bit of a bodge, but 
short of invasive large-scale changes with bus types I don't see a way 
to do the 'right' thing :/

Robin.

----->8-----
diff --git a/drivers/mfd/mfd-core.c b/drivers/mfd/mfd-core.c
index f5a73af60dd4..1e4a6e8bd630 100644
--- a/drivers/mfd/mfd-core.c
+++ b/drivers/mfd/mfd-core.c
@@ -138,7 +138,7 @@ static int mfd_add_device(struct device *parent, int id,

  	pdev->dev.parent = parent;
  	pdev->dev.type = &mfd_dev_type;
-	pdev->dev.dma_mask = parent->dma_mask;
+	pdev->dma_mask = parent->dma_mask ? *parent->dma_mask : 0;
  	pdev->dev.dma_parms = parent->dma_parms;
  	pdev->dev.coherent_dma_mask = parent->coherent_dma_mask;

Re: [RFC PATCH] mfd: mfd-core: inherit only valid dma_masks/flags from parent

[Michael Walle]

Am 2020-03-11 12:25, schrieb Robin Murphy:
> On 2020-03-10 11:09 pm, Michael Walle wrote:
>> Only copy the dma_masks and flags from the parent device, if the 
>> parent
>> has a valid dma_mask/flags. Commit cdfee5623290 ("driver core:
>> initialize a default DMA mask for platform device") initialize the DMA
>> masks of a platform device. But if the parent doesn't have a dma_mask
>> set, for example if it's an I2C device, the dma_mask of the child
>> platform device will be set to zero again. Which leads to many "DMA 
>> mask
>> not set" warnings, if the MFD cell has the of_compatible property set.
>> 
>> [    1.877937] sl28cpld-pwm sl28cpld-pwm: DMA mask not set
>> [    1.883282] sl28cpld-pwm sl28cpld-pwm.0: DMA mask not set
>> [    1.888795] sl28cpld-gpio sl28cpld-gpio: DMA mask not set
>> 
>> Thus a MFD child should just inherit valid dma_masks and keep the
>> platform default otherwise.
>> 
>> Cc: Christoph Hellwig <hch@lst.de>
>> Cc: Rob Herring <robh@kernel.org>
>> Cc: Robin Murphy <robin.murphy@arm.com>
>> Cc: Russell King <linux@armlinux.org.uk>
>> Signed-off-by: Michael Walle <michael@walle.cc>
>> ---
>> 
>> Hi,
>> 
>> I don't know if that is the correct way of handling things. Maybe I'm
>> also doing something wrong in my driver, I had a look at other I2C MFD
>> drivers but couldn't find a clue why they shouldn't have the same
>> problem.
> 
> The underlying issue is that about 99% of MFD children should not be
> going through dma_configure() at all because their parent 'real'
> device is not on a DMA-capable bus, but as they are platform devices
> we are forced to give them the benefit of the doubt. For DT systems
> the only vaguely-reasonable heuristic to distinguish between
> "platform" meaning "SoC memory-mapped device" and "platform" meaning
> "random crap made up by Linux" is whether the device has a populated
> OF node, but MFD's trick of hanging the parent device's OF node onto
> its synthesised children kicks a hole right through even that.

Thanks for the explanation.

> Modulo any other concerns with the existing code, does the change
> below make things work the way you want? It's still a bit of a bodge,
> but short of invasive large-scale changes with bus types I don't see a
> way to do the 'right' thing :/
> 
> Robin.
> 
> ----->8-----
> diff --git a/drivers/mfd/mfd-core.c b/drivers/mfd/mfd-core.c
> index f5a73af60dd4..1e4a6e8bd630 100644
> --- a/drivers/mfd/mfd-core.c
> +++ b/drivers/mfd/mfd-core.c
> @@ -138,7 +138,7 @@ static int mfd_add_device(struct device *parent, 
> int id,
> 
>  	pdev->dev.parent = parent;
>  	pdev->dev.type = &mfd_dev_type;
> -	pdev->dev.dma_mask = parent->dma_mask;
> +	pdev->dma_mask = parent->dma_mask ? *parent->dma_mask : 0;

That works.

-michael

>  	pdev->dev.dma_parms = parent->dma_parms;
>  	pdev->dev.coherent_dma_mask = parent->coherent_dma_mask;