Skip to contentSkip to conversation

mfd: mfd-core: inherit only valid dma_masks/flags from parent

RFC PATCH In review Michael Walle 3 participants 5 messages Mar 10, 2020 – Mar 11, 2020 Multifunction devices (MFD)

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 mfd: mfd-core: inherit only valid dma_masks/flags from parent

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 <[email protected]>
Cc Rob Herring <[email protected]>
Cc Robin Murphy <[email protected]>
Cc Russell King <[email protected]>

---

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(-)
drivers/mfd/mfd-core.c +6 -3
index b9eb8f40c073..5d8ea5e8e93c 100644
@@ -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,
Signature · 1 line
2.20.1
#1 lore raw
Quoted 42 lines · 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 <[email protected]>
Cc Rob Herring <[email protected]>
Cc Robin Murphy <[email protected]>
Cc Russell King <[email protected]>

---

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(-)
drivers/mfd/mfd-core.c +0 -0
index b9eb8f40c073..5d8ea5e8e93c 100644

drivers/mfd/mfd-core.c static int mfd_add_device(struct device *parent, int id, · line 145

Quoted 10 lines
drivers/mfd/mfd-core.c +4 -3
@@ -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.

#2 lore raw

Am 2020-03-11 07:18, schrieb Christoph Hellwig:

Quoted 45 lines · Michael Walle
Quoted 44 lines

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 <[email protected]>
Cc Rob Herring <[email protected]>
Cc Robin Murphy <[email protected]>
Cc Russell King <[email protected]>

---

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(-)
drivers/mfd/mfd-core.c +0 -0
index b9eb8f40c073..5d8ea5e8e93c 100644
@@ -139,9 +139,12 @@ static int mfd_add_device(struct device *parent, 
 int id,
 
Quoted 12 lines

drivers/mfd/mfd-core.c static int mfd_add_device(struct device *parent, int id, · line 145

Quoted 9 lines
drivers/mfd/mfd-core.c +4 -3
 	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

#3 lore raw
In reply to Michael Walle
Quoted 20 lines · 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 <[email protected]>
Cc Rob Herring <[email protected]>
Cc Robin Murphy <[email protected]>
Cc Russell King <[email protected]>
Quoted 8 lines

---

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-----

drivers/mfd/mfd-core.c +1 -1
index f5a73af60dd4..1e4a6e8bd630 100644
@@ -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;
 
#4 lore raw

Am 2020-03-11 12:25, schrieb Robin Murphy:

Quoted 30 lines · Michael Walle
Quoted 20 lines

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 <[email protected]>
Cc Rob Herring <[email protected]>
Cc Robin Murphy <[email protected]>
Cc Russell King <[email protected]>
Quoted 8 lines

---

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.

Quoted 9 lines

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.

Quoted 13 lines

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-----

drivers/mfd/mfd-core.c +0 -0
index f5a73af60dd4..1e4a6e8bd630 100644

drivers/mfd/mfd-core.c static int mfd_add_device(struct device *parent, · line 142

Quoted 6 lines
drivers/mfd/mfd-core.c +1 -1
@@ -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

Quoted 2 lines
drivers/mfd/mfd-core.c +0 -0
 	pdev->dev.dma_parms = parent->dma_parms;
 	pdev->dev.coherent_dma_mask = parent->coherent_dma_mask;