[RFC] xen/gntdev: Stop abusing DT of_dma_configure API

Message ID 20190925215006.12056-1-robh@kernel.org
State New
Headers show
Series
  • [RFC] xen/gntdev: Stop abusing DT of_dma_configure API
Related show

Commit Message

Rob Herring Sept. 25, 2019, 9:50 p.m.
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 <robin.murphy@arm.com>
Cc: Julien Grall <julien.grall@arm.com>
Cc: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>
Cc: Oleksandr Andrushchenko <oleksandr_andrushchenko@epam.com>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Juergen Gross <jgross@suse.com>
Cc: Stefano Stabellini <sstabellini@kernel.org>
Cc: xen-devel@lists.xenproject.org
Signed-off-by: Rob Herring <robh@kernel.org>

---
 drivers/xen/gntdev.c | 11 -----------
 1 file changed, 11 deletions(-)

-- 
2.20.1

Comments

Nicolas Saenz Julienne Sept. 26, 2019, 9:06 a.m. | #1
On Wed, 2019-09-25 at 16:50 -0500, Rob Herring wrote:
> 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 <robin.murphy@arm.com>

> Cc: Julien Grall <julien.grall@arm.com>

> Cc: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>

> Cc: Oleksandr Andrushchenko <oleksandr_andrushchenko@epam.com>

> Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>

> Cc: Juergen Gross <jgross@suse.com>

> Cc: Stefano Stabellini <sstabellini@kernel.org>

> Cc: xen-devel@lists.xenproject.org

> Signed-off-by: Rob Herring <robh@kernel.org>


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

diff --git a/drivers/gpu/drm/xen/xen_drm_front.c
b/drivers/gpu/drm/xen/xen_drm_front.c
index ba1828acd8c9..de316a891f39 100644
--- a/drivers/gpu/drm/xen/xen_drm_front.c
+++ b/drivers/gpu/drm/xen/xen_drm_front.c
@@ -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);
Julien Grall Sept. 26, 2019, 9:49 a.m. | #2
Hi Rob,


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

> 

> 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,

-- 
Julien Grall
Oleksandr Andrushchenko Sept. 26, 2019, 10:17 a.m. | #3
On 9/26/19 12:49 PM, Julien Grall wrote:
> Hi Rob,

>

>

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

>> 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
>>

>> 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
>

> 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
Robin Murphy Sept. 26, 2019, 10:46 a.m. | #4
On 2019-09-26 11:17 am, Oleksandr Andrushchenko wrote:
> 

> On 9/26/19 12:49 PM, Julien Grall wrote:

>> Hi Rob,

>>

>>

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

>>> 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

>>>

>>> 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


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.
Oleksandr Andrushchenko Sept. 26, 2019, 11:16 a.m. | #5
On 9/26/19 1:46 PM, Robin Murphy wrote:
> On 2019-09-26 11:17 am, Oleksandr Andrushchenko wrote:

>>

>> On 9/26/19 12:49 PM, Julien Grall wrote:

>>> Hi Rob,

>>>

>>>

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

>>>> 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

>>>>

>>>> 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

>

> 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.
>

> 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
Rob Herring Sept. 26, 2019, 7:27 p.m. | #6
On Thu, Sep 26, 2019 at 6:16 AM Oleksandr Andrushchenko
<Oleksandr_Andrushchenko@epam.com> wrote:
>

> On 9/26/19 1:46 PM, Robin Murphy wrote:

> > On 2019-09-26 11:17 am, Oleksandr Andrushchenko wrote:

> >>

> >> On 9/26/19 12:49 PM, Julien Grall wrote:

> >>> Hi Rob,

> >>>

> >>>

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

> >>>> 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

> >>>>

> >>>> 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

> >

> > 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
Stefano Stabellini Oct. 1, 2019, 6:23 p.m. | #7
On Thu, 26 Sep 2019, Rob Herring wrote:
> On Thu, Sep 26, 2019 at 6:16 AM Oleksandr Andrushchenko

> <Oleksandr_Andrushchenko@epam.com> wrote:

> >

> > On 9/26/19 1:46 PM, Robin Murphy wrote:

> > > On 2019-09-26 11:17 am, Oleksandr Andrushchenko wrote:

> > >>

> > >> On 9/26/19 12:49 PM, Julien Grall wrote:

> > >>> Hi Rob,

> > >>>

> > >>>

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

> > >>>> 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

> > >>>>

> > >>>> 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

> > >

> > > 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.


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?
Oleksandr Andrushchenko Oct. 2, 2019, 4:42 a.m. | #8
On 10/1/19 9:23 PM, Stefano Stabellini wrote:
> On Thu, 26 Sep 2019, Rob Herring wrote:

>> On Thu, Sep 26, 2019 at 6:16 AM Oleksandr Andrushchenko

>> <Oleksandr_Andrushchenko@epam.com> wrote:

>>> On 9/26/19 1:46 PM, Robin Murphy wrote:

>>>> On 2019-09-26 11:17 am, Oleksandr Andrushchenko wrote:

>>>>> On 9/26/19 12:49 PM, Julien Grall wrote:

>>>>>> Hi Rob,

>>>>>>

>>>>>>

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

>>>>>>> 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

>>>>>>> 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

>>>> 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.

> 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?

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

Patch

diff --git a/drivers/xen/gntdev.c b/drivers/xen/gntdev.c
index a446a7221e13..59906f9a40e4 100644
--- a/drivers/xen/gntdev.c
+++ b/drivers/xen/gntdev.c
@@ -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);