Skip to contentSkip to conversation

xen/gntdev: Stop abusing DT of_dma_configure API

RFC PATCH Superseded Rob Herring 6 participants 9 messages Sep 25, 2019 – Oct 2, 2019

Subsystems (4)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 xen/gntdev: Stop abusing DT of_dma_configure API

As the comment says, this isn't a DT based device. of_dma_configure() is going to stop allowing a NULL DT node, so this needs to be fixed.

Not sure exactly what setup besides arch_setup_dma_ops is needed...

Cc Robin Murphy <[email protected]>
Cc Julien Grall <[email protected]>
Cc Nicolas Saenz Julienne <[email protected]>
Cc Oleksandr Andrushchenko <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc Juergen Gross <[email protected]>
Cc Stefano Stabellini <[email protected]>
---
 drivers/xen/gntdev.c | 11 -----------
 1 file changed, 11 deletions(-)
drivers/xen/gntdev.c +0 -11
index a446a7221e13..59906f9a40e4 100644
@@ -34,9 +34,6 @@
 #include <linux/slab.h>
 #include <linux/highmem.h>
 #include <linux/refcount.h>
-#ifdef CONFIG_XEN_GRANT_DMA_ALLOC
-#include <linux/of_device.h>
-#endif
 
 #include <xen/xen.h>
 #include <xen/grant_table.h>
@@ -625,14 +622,6 @@ static int gntdev_open(struct inode *inode, struct file *flip)
 	flip->private_data = priv;
 #ifdef CONFIG_XEN_GRANT_DMA_ALLOC
 	priv->dma_dev = gntdev_miscdev.this_device;
-
-	/*
-	 * The device is not spawn from a device tree, so arch_setup_dma_ops
-	 * is not called, thus leaving the device with dummy DMA ops.
-	 * Fix this by calling of_dma_configure() with a NULL node to set
-	 * default DMA ops.
-	 */
-	of_dma_configure(priv->dma_dev, NULL, true);
 #endif
 	pr_debug("priv %p\n", priv);
 
Signature · 1 line
2.20.1
#1 lore raw
Quoted 5 lines · Rob Herring

As the comment says, this isn't a DT based device. of_dma_configure() is going to stop allowing a NULL DT node, so this needs to be fixed.

Not sure exactly what setup besides arch_setup_dma_ops is needed...

Quoted 9 lines
Cc Robin Murphy <[email protected]>
Cc Julien Grall <[email protected]>
Cc Nicolas Saenz Julienne <[email protected]>
Cc Oleksandr Andrushchenko <[email protected]>
Cc Boris Ostrovsky <[email protected]>
Cc Juergen Gross <[email protected]>
Cc Stefano Stabellini <[email protected]>

Just so it isn't forgotten, the same applies here:

(quoted diff) +0 -0

b/drivers/gpu/drm/xen/xen_drm_front.c
index ba1828acd8c9..de316a891f39 100644

drivers/gpu/drm/xen/xen_drm_front.c +0 -13
@@ -11,7 +11,6 @@
 #include <linux/delay.h>
 #include <linux/dma-mapping.h>
 #include <linux/module.h>
-#include <linux/of_device.h>
 
 #include <drm/drm_atomic_helper.h>
 #include <drm/drm_drv.h>
@@ -718,19 +717,7 @@ static int xen_drv_probe(struct xenbus_device *xb_dev,
        struct device *dev = &xb_dev->dev;
        int ret;
 
-       /*
-        * The device is not spawn from a device tree, so arch_setup_dma_ops
-        * is not called, thus leaving the device with dummy DMA ops.
-        * This makes the device return error on PRIME buffer import, which
-        * is not correct: to fix this call of_dma_configure() with a NULL
-        * node to set default DMA ops.
-        */
        dev->coherent_dma_mask = DMA_BIT_MASK(32);
-       ret = of_dma_configure(dev, NULL, true);
-       if (ret < 0) {
-               DRM_ERROR("Cannot setup DMA ops, ret %d", ret);
-               return ret;
-       }
 
        front_info = devm_kzalloc(&xb_dev->dev,
                                  sizeof(*front_info), GFP_KERNEL);
#2 lore raw
In reply to Rob Herring

Hi Rob,

Quoted 2 lines · Rob Herring

As the comment says, this isn't a DT based device. of_dma_configure() is going to stop allowing a NULL DT node, so this needs to be fixed.

And this can't work on arch not selecting CONFIG_OF and can select CONFIG_XEN_GRANT_DMA_ALLOC.

We are lucky enough on x86 because, AFAICT, arch_setup_dma_ops is just a nop.

Quoted 2 lines

Not sure exactly what setup besides arch_setup_dma_ops is needed...

We probably want to update dma_mask, coherent_dma_mask and dma_pfn_offset.

Also, while look at of_configure_dma, I noticed that we consider the DMA will not be coherent for the grant-table. Oleksandr, do you know why they can't be coherent?

Cheers,

Signature · 1 line
Julien Grall
#3 lore raw
Quoted 4 lines · Julien Grall

Hi Rob,

On 9/25/19 10:50 PM, Rob Herring wrote:

Quoted 9 lines
Quoted 2 lines

As the comment says, this isn't a DT based device. of_dma_configure() is going to stop allowing a NULL DT node, so this needs to be fixed.

And this can't work on arch not selecting CONFIG_OF and can select CONFIG_XEN_GRANT_DMA_ALLOC.

We are lucky enough on x86 because, AFAICT, arch_setup_dma_ops is just a nop.

No luck is needed as [1] does nothing for those platforms not using CONFIG_OF

Quoted 9 lines
Quoted 2 lines

Not sure exactly what setup besides arch_setup_dma_ops is needed...

We probably want to update dma_mask, coherent_dma_mask and dma_pfn_offset.

Also, while look at of_configure_dma, I noticed that we consider the DMA will not be coherent for the grant-table. Oleksandr, do you know why they can't be coherent?

The main and the only reason to use of_configure_dma is that if we don't then we
are about to stay with dma_dummy_ops [2]. It effectively means that operations on dma-bufs
will end up returning errors, like [3], [4], thus not making it possible for Xen PV DRM and DMA
part of gntdev driver to do what we need (dma-bufs in our use-cases allow zero-copying
while using graphics buffers and many more).

I didn't find any better way of achieving that, but of_configure_dma... If there is any better solution which will not break the existing functionality then
I will definitely change the drivers so we do not abuse DT ) Before that, please keep in mind that merging this RFC will break Xen PV DRM +
DMA buf support in gntdev...
Hope we can work out some acceptable solution, so everyone is happy

Quoted 3 lines

Cheers,

Thank you,
Oleksandr

[1]
https://elixir.bootlin.com/linux/v5.3.1/source/include/linux/of_device.h#L109
[2] https://elixir.bootlin.com/linux/v5.3.1/source/kernel/dma/dummy.c#L33
[3] https://elixir.bootlin.com/linux/v5.3.1/source/kernel/dma/dummy.c#L11
[4] https://elixir.bootlin.com/linux/v5.3.1/source/kernel/dma/dummy.c#L18

#4 lore raw
Quoted 36 lines · Oleksandr Andrushchenko
Quoted 4 lines · Julien Grall

Hi Rob,

On 9/25/19 10:50 PM, Rob Herring wrote:

Quoted 9 lines
Quoted 2 lines

As the comment says, this isn't a DT based device. of_dma_configure() is going to stop allowing a NULL DT node, so this needs to be fixed.

And this can't work on arch not selecting CONFIG_OF and can select CONFIG_XEN_GRANT_DMA_ALLOC.

We are lucky enough on x86 because, AFAICT, arch_setup_dma_ops is just a nop.

No luck is needed as [1] does nothing for those platforms not using CONFIG_OF

Quoted 9 lines
Quoted 2 lines

Not sure exactly what setup besides arch_setup_dma_ops is needed...

We probably want to update dma_mask, coherent_dma_mask and dma_pfn_offset.

Also, while look at of_configure_dma, I noticed that we consider the DMA will not be coherent for the grant-table. Oleksandr, do you know why they can't be coherent?

The main and the only reason to use of_configure_dma is that if we don't then we
are about to stay with dma_dummy_ops [2]. It effectively means that operations on dma-bufs
will end up returning errors, like [3], [4], thus not making it possible for Xen PV DRM and DMA
part of gntdev driver to do what we need (dma-bufs in our use-cases allow zero-copying
while using graphics buffers and many more).

Quoted 8 lines

I didn't find any better way of achieving that, but of_configure_dma... If there is any better solution which will not break the existing functionality then
I will definitely change the drivers so we do not abuse DT ) Before that, please keep in mind that merging this RFC will break Xen PV DRM +
DMA buf support in gntdev...
Hope we can work out some acceptable solution, so everyone is happy

As I mentioned elsewhere, the recent dma-direct rework means that dma_dummy_ops are now only explicitly installed for the ACPI error case, so - much as I may dislike it - you should get regular (direct/SWIOTLB) ops by default again.

Coherency is trickier - if the guest is allocating buffers for the PV device, which may be shared directly with hardware by the host driver, then the coherency of the PV device should really reflect that of the underlying hardware to avoid potential problems. There are some cases where the stage 2 attributes alone wouldn't be enough to correct a mismatch.

Robin.

#5 lore raw
Quoted 46 lines · Robin Murphy
Quoted 36 lines · Oleksandr Andrushchenko
Quoted 4 lines · Julien Grall

Hi Rob,

On 9/25/19 10:50 PM, Rob Herring wrote:

Quoted 9 lines
Quoted 2 lines

As the comment says, this isn't a DT based device. of_dma_configure() is going to stop allowing a NULL DT node, so this needs to be fixed.

And this can't work on arch not selecting CONFIG_OF and can select CONFIG_XEN_GRANT_DMA_ALLOC.

We are lucky enough on x86 because, AFAICT, arch_setup_dma_ops is just a nop.

No luck is needed as [1] does nothing for those platforms not using CONFIG_OF

Quoted 9 lines
Quoted 2 lines

Not sure exactly what setup besides arch_setup_dma_ops is needed...

We probably want to update dma_mask, coherent_dma_mask and dma_pfn_offset.

Also, while look at of_configure_dma, I noticed that we consider the DMA will not be coherent for the grant-table. Oleksandr, do you know why they can't be coherent?

The main and the only reason to use of_configure_dma is that if we don't then we
are about to stay with dma_dummy_ops [2]. It effectively means that operations on dma-bufs
will end up returning errors, like [3], [4], thus not making it possible for Xen PV DRM and DMA
part of gntdev driver to do what we need (dma-bufs in our use-cases allow zero-copying
while using graphics buffers and many more).

Quoted 8 lines

I didn't find any better way of achieving that, but of_configure_dma... If there is any better solution which will not break the existing functionality then
I will definitely change the drivers so we do not abuse DT ) Before that, please keep in mind that merging this RFC will break Xen PV DRM +
DMA buf support in gntdev...
Hope we can work out some acceptable solution, so everyone is happy

Quoted 4 lines

As I mentioned elsewhere, the recent dma-direct rework means that dma_dummy_ops are now only explicitly installed for the ACPI error case, so - much as I may dislike it - you should get regular (direct/SWIOTLB) ops by default again.

Ah, my bad, I missed that change. So, if no dummy dma ops are to be used then
I believe we can apply both changes, e.g. remove of_dma_configure from both of the drivers.

Quoted 9 lines

Coherency is trickier - if the guest is allocating buffers for the PV device, which may be shared directly with hardware by the host driver, then the coherency of the PV device should really reflect that of the underlying hardware to avoid potential problems. There are some cases where the stage 2 attributes alone wouldn't be enough to correct a mismatch.

Robin.

Thank you,
Oleksandr

#6 lore raw
Quoted 48 lines · Oleksandr Andrushchenko
Quoted 46 lines · Robin Murphy
Quoted 44 lines · Oleksandr Andrushchenko
Quoted 4 lines · Julien Grall

Hi Rob,

On 9/25/19 10:50 PM, Rob Herring wrote:

Quoted 9 lines
Quoted 2 lines

As the comment says, this isn't a DT based device. of_dma_configure() is going to stop allowing a NULL DT node, so this needs to be fixed.

And this can't work on arch not selecting CONFIG_OF and can select CONFIG_XEN_GRANT_DMA_ALLOC.

We are lucky enough on x86 because, AFAICT, arch_setup_dma_ops is just a nop.

No luck is needed as [1] does nothing for those platforms not using CONFIG_OF

Quoted 9 lines
Quoted 2 lines

Not sure exactly what setup besides arch_setup_dma_ops is needed...

We probably want to update dma_mask, coherent_dma_mask and dma_pfn_offset.

Also, while look at of_configure_dma, I noticed that we consider the DMA will not be coherent for the grant-table. Oleksandr, do you know why they can't be coherent?

The main and the only reason to use of_configure_dma is that if we don't then we
are about to stay with dma_dummy_ops [2]. It effectively means that operations on dma-bufs
will end up returning errors, like [3], [4], thus not making it possible for Xen PV DRM and DMA
part of gntdev driver to do what we need (dma-bufs in our use-cases allow zero-copying
while using graphics buffers and many more).

I didn't find any better way of achieving that, but of_configure_dma... If there is any better solution which will not break the existing functionality then
I will definitely change the drivers so we do not abuse DT ) Before that, please keep in mind that merging this RFC will break Xen PV DRM +
DMA buf support in gntdev...
Hope we can work out some acceptable solution, so everyone is happy

Quoted 8 lines
Quoted 4 lines

As I mentioned elsewhere, the recent dma-direct rework means that dma_dummy_ops are now only explicitly installed for the ACPI error case, so - much as I may dislike it - you should get regular (direct/SWIOTLB) ops by default again.

Ah, my bad, I missed that change. So, if no dummy dma ops are to be used then
I believe we can apply both changes, e.g. remove of_dma_configure from both of the drivers.

What about the dma masks? I think there's a default setup, but it is considered a driver bug to not set its mask. xen_drm_front sets the coherent_dma_mask (why only 32-bits though?), but not the dma_mask. gntdev is doing neither. I could copy out what of_dma_configure does but better for the Xen folks to decide what is needed or not and test the change. I'm not setup to test any of this.

Rob

#7 lore raw
Quoted 59 lines · Rob Herring
Quoted 48 lines · Oleksandr Andrushchenko
Quoted 46 lines · Robin Murphy
Quoted 44 lines · Oleksandr Andrushchenko
Quoted 4 lines · Julien Grall

Hi Rob,

On 9/25/19 10:50 PM, Rob Herring wrote:

Quoted 9 lines
Quoted 2 lines

As the comment says, this isn't a DT based device. of_dma_configure() is going to stop allowing a NULL DT node, so this needs to be fixed.

And this can't work on arch not selecting CONFIG_OF and can select CONFIG_XEN_GRANT_DMA_ALLOC.

We are lucky enough on x86 because, AFAICT, arch_setup_dma_ops is just a nop.

No luck is needed as [1] does nothing for those platforms not using CONFIG_OF

Quoted 9 lines
Quoted 2 lines

Not sure exactly what setup besides arch_setup_dma_ops is needed...

We probably want to update dma_mask, coherent_dma_mask and dma_pfn_offset.

Also, while look at of_configure_dma, I noticed that we consider the DMA will not be coherent for the grant-table. Oleksandr, do you know why they can't be coherent?

The main and the only reason to use of_configure_dma is that if we don't then we
are about to stay with dma_dummy_ops [2]. It effectively means that operations on dma-bufs
will end up returning errors, like [3], [4], thus not making it possible for Xen PV DRM and DMA
part of gntdev driver to do what we need (dma-bufs in our use-cases allow zero-copying
while using graphics buffers and many more).

I didn't find any better way of achieving that, but of_configure_dma... If there is any better solution which will not break the existing functionality then
I will definitely change the drivers so we do not abuse DT ) Before that, please keep in mind that merging this RFC will break Xen PV DRM +
DMA buf support in gntdev...
Hope we can work out some acceptable solution, so everyone is happy

Quoted 8 lines
Quoted 4 lines

As I mentioned elsewhere, the recent dma-direct rework means that dma_dummy_ops are now only explicitly installed for the ACPI error case, so - much as I may dislike it - you should get regular (direct/SWIOTLB) ops by default again.

Ah, my bad, I missed that change. So, if no dummy dma ops are to be used then
I believe we can apply both changes, e.g. remove of_dma_configure from both of the drivers.

Quoted 6 lines

What about the dma masks? I think there's a default setup, but it is considered a driver bug to not set its mask. xen_drm_front sets the coherent_dma_mask (why only 32-bits though?), but not the dma_mask. gntdev is doing neither. I could copy out what of_dma_configure does but better for the Xen folks to decide what is needed or not and test the change. I'm not setup to test any of this.

FYI I have seen the issue Oleksandr is talking about too. I confirm that the only reason for the of_configure_dma call is to get away from the dummy_dma_ops and use the default dma_ops instead. I think this should be mentioned in the commit message so that if one day the behavior regarding dummy_dma_ops changes one more time, hopefully we'll be able to figure out the issue more easily with bisection.

In regards to the coherent_dma_mask and dma_mask, I can't see why gntdev would have any dma addressing limitations, so we should be able to set both to 64 bits. I also can't see why xen_drm_front would limit it to 32 bits, after all this is just the frontend, if anything it would be the backend that has a limitation. So, we should be able to set both dma_mask and coherent_dma_mask in xen_drm_front to 64 bits. Oleksandr, can you confirm?

#8 lore raw
Quoted 66 lines · Stefano Stabellini
Quoted 52 lines · Rob Herring
Quoted 42 lines · Oleksandr Andrushchenko
Quoted 41 lines · Robin Murphy
Quoted 40 lines · Oleksandr Andrushchenko
Quoted 12 lines · Julien Grall

Hi Rob,

Quoted 2 lines · Rob Herring

As the comment says, this isn't a DT based device. of_dma_configure() is going to stop allowing a NULL DT node, so this needs to be fixed.

And this can't work on arch not selecting CONFIG_OF and can select CONFIG_XEN_GRANT_DMA_ALLOC.

We are lucky enough on x86 because, AFAICT, arch_setup_dma_ops is just a nop.

No luck is needed as [1] does nothing for those platforms not using CONFIG_OF

Quoted 7 lines
Quoted 1 line

Not sure exactly what setup besides arch_setup_dma_ops is needed...

We probably want to update dma_mask, coherent_dma_mask and dma_pfn_offset.

Also, while look at of_configure_dma, I noticed that we consider the DMA will not be coherent for the grant-table. Oleksandr, do you know why they can't be coherent?

The main and the only reason to use of_configure_dma is that if we don't then we
are about to stay with dma_dummy_ops [2]. It effectively means that operations on dma-bufs
will end up returning errors, like [3], [4], thus not making it possible for Xen PV DRM and DMA
part of gntdev driver to do what we need (dma-bufs in our use-cases allow zero-copying
while using graphics buffers and many more).

I didn't find any better way of achieving that, but of_configure_dma... If there is any better solution which will not break the existing functionality then
I will definitely change the drivers so we do not abuse DT ) Before that, please keep in mind that merging this RFC will break Xen PV DRM +
DMA buf support in gntdev...
Hope we can work out some acceptable solution, so everyone is happy

Quoted 8 lines
Quoted 4 lines

As I mentioned elsewhere, the recent dma-direct rework means that dma_dummy_ops are now only explicitly installed for the ACPI error case, so - much as I may dislike it - you should get regular (direct/SWIOTLB) ops by default again.

Ah, my bad, I missed that change. So, if no dummy dma ops are to be used then
I believe we can apply both changes, e.g. remove of_dma_configure from both of the drivers.

Quoted 6 lines

What about the dma masks? I think there's a default setup, but it is considered a driver bug to not set its mask. xen_drm_front sets the coherent_dma_mask (why only 32-bits though?), but not the dma_mask. gntdev is doing neither. I could copy out what of_dma_configure does but better for the Xen folks to decide what is needed or not and test the change. I'm not setup to test any of this.

FYI I have seen the issue Oleksandr is talking about too. I confirm that the only reason for the of_configure_dma call is to get away from the dummy_dma_ops and use the default dma_ops instead. I think this should be mentioned in the commit message so that if one day the behavior regarding dummy_dma_ops changes one more time, hopefully we'll be able to figure out the issue more easily with bisection.

Quoted 7 lines

In regards to the coherent_dma_mask and dma_mask, I can't see why gntdev would have any dma addressing limitations, so we should be able to set both to 64 bits. I also can't see why xen_drm_front would limit it to 32 bits, after all this is just the frontend, if anything it would be the backend that has a limitation. So, we should be able to set both dma_mask and coherent_dma_mask in xen_drm_front to 64 bits. Oleksandr, can you confirm?

I am totally fine with 64-bits in both cases and
agree with what Stefano says.