[libdrm] libdrm: Allow dynamic drm majors on linux
diff mbox series

Message ID 20180831115419.4922-1-thellstrom@vmware.com
State New
Headers show
Series
  • [libdrm] libdrm: Allow dynamic drm majors on linux
Related show

Commit Message

Thomas Hellstrom Aug. 31, 2018, 11:54 a.m. UTC
To determine whether a device node is a drm device node or not, the code
currently compares the node's major number to the static drm major device
number.

This breaks the standalone vmwgfx driver on XWayland dri clients,
https://cgit.freedesktop.org/mesa/vmwgfx
and any future attempt to introduce dynamic device numbers for drm.

So instead of checking for the device major, instead check for the presence
of the /sys/dev/char/<major>:<minor>/device/drm directory.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
---
 xf86drm.c | 31 +++++++++++++++++++++++--------
 1 file changed, 23 insertions(+), 8 deletions(-)

Comments

Eric Engestrom Aug. 31, 2018, 12:15 p.m. UTC | #1
On Friday, 2018-08-31 13:54:19 +0200, Thomas Hellstrom wrote:
> To determine whether a device node is a drm device node or not, the code
> currently compares the node's major number to the static drm major device
> number.
> 
> This breaks the standalone vmwgfx driver on XWayland dri clients,
> https://cgit.freedesktop.org/mesa/vmwgfx
> and any future attempt to introduce dynamic device numbers for drm.
> 
> So instead of checking for the device major, instead check for the presence
> of the /sys/dev/char/<major>:<minor>/device/drm directory.

Just FYI, this means it now matches /dev/fb0 too.

I don't think this will be an issue, but just pointing it out so people notice.

> 
> Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
> ---
>  xf86drm.c | 31 +++++++++++++++++++++++--------
>  1 file changed, 23 insertions(+), 8 deletions(-)
> 
> diff --git a/xf86drm.c b/xf86drm.c
> index 7807dce9..4cfc5d3e 100644
> --- a/xf86drm.c
> +++ b/xf86drm.c
> @@ -2761,6 +2761,21 @@ char *drmGetDeviceNameFromFd(int fd)
>      return strdup(name);
>  }
>  
> +static bool
> +drmNodeIsDRM(int maj, int min)
> +{
> +#ifdef __linux__
> +  char path[64];
> +  struct stat sbuf;
> +
> +  snprintf(path, sizeof(path), "/sys/dev/char/%d:%d/device/drm",
> +	   maj, min);

Nit: mixing tabs and space on this line.

With that fixed:
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>

> +  return stat(path, &sbuf) == 0;
> +#else
> +  return maj == DRM_MAJOR;
> +#endif
> +}
> +
>  int drmGetNodeTypeFromFd(int fd)
>  {
>      struct stat sbuf;
> @@ -2772,7 +2787,7 @@ int drmGetNodeTypeFromFd(int fd)
>      maj = major(sbuf.st_rdev);
>      min = minor(sbuf.st_rdev);
>  
> -    if (maj != DRM_MAJOR || !S_ISCHR(sbuf.st_mode)) {
> +    if (!drmNodeIsDRM(maj, min) || !S_ISCHR(sbuf.st_mode)) {
>          errno = EINVAL;
>          return -1;
>      }
> @@ -2837,7 +2852,7 @@ static char *drmGetMinorNameForFD(int fd, int type)
>      maj = major(sbuf.st_rdev);
>      min = minor(sbuf.st_rdev);
>  
> -    if (maj != DRM_MAJOR || !S_ISCHR(sbuf.st_mode))
> +    if (!drmNodeIsDRM(maj, min) || !S_ISCHR(sbuf.st_mode))
>          return NULL;
>  
>      snprintf(buf, sizeof(buf), "/sys/dev/char/%d:%d/device/drm", maj, min);
> @@ -2871,7 +2886,7 @@ static char *drmGetMinorNameForFD(int fd, int type)
>      maj = major(sbuf.st_rdev);
>      min = minor(sbuf.st_rdev);
>  
> -    if (maj != DRM_MAJOR || !S_ISCHR(sbuf.st_mode))
> +    if (!drmNodeIsDRM(maj, min) || !S_ISCHR(sbuf.st_mode))
>          return NULL;
>  
>      switch (type) {
> @@ -3731,7 +3746,7 @@ process_device(drmDevicePtr *device, const char *d_name,
>      maj = major(sbuf.st_rdev);
>      min = minor(sbuf.st_rdev);
>  
> -    if (maj != DRM_MAJOR || !S_ISCHR(sbuf.st_mode))
> +    if (!drmNodeIsDRM(maj, min) || !S_ISCHR(sbuf.st_mode))
>          return -1;
>  
>      subsystem_type = drmParseSubsystemType(maj, min);
> @@ -3845,7 +3860,7 @@ int drmGetDevice2(int fd, uint32_t flags, drmDevicePtr *device)
>      maj = major(sbuf.st_rdev);
>      min = minor(sbuf.st_rdev);
>  
> -    if (maj != DRM_MAJOR || !S_ISCHR(sbuf.st_mode))
> +    if (!drmNodeIsDRM(maj, min) || !S_ISCHR(sbuf.st_mode))
>          return -EINVAL;
>  
>      node_type = drmGetMinorType(min);
> @@ -3911,7 +3926,7 @@ int drmGetDevice2(int fd, uint32_t flags, drmDevicePtr *device)
>      maj = major(sbuf.st_rdev);
>      min = minor(sbuf.st_rdev);
>  
> -    if (maj != DRM_MAJOR || !S_ISCHR(sbuf.st_mode))
> +    if (!drmNodeIsDRM(maj, min) || !S_ISCHR(sbuf.st_mode))
>          return -EINVAL;
>  
>      subsystem_type = drmParseSubsystemType(maj, min);
> @@ -4071,7 +4086,7 @@ char *drmGetDeviceNameFromFd2(int fd)
>      maj = major(sbuf.st_rdev);
>      min = minor(sbuf.st_rdev);
>  
> -    if (maj != DRM_MAJOR || !S_ISCHR(sbuf.st_mode))
> +    if (!drmNodeIsDRM(maj, min) || !S_ISCHR(sbuf.st_mode))
>          return NULL;
>  
>      snprintf(path, sizeof(path), "/sys/dev/char/%d:%d", maj, min);
> @@ -4097,7 +4112,7 @@ char *drmGetDeviceNameFromFd2(int fd)
>      maj = major(sbuf.st_rdev);
>      min = minor(sbuf.st_rdev);
>  
> -    if (maj != DRM_MAJOR || !S_ISCHR(sbuf.st_mode))
> +    if (!drmNodeIsDRM(maj, min) || !S_ISCHR(sbuf.st_mode))
>          return NULL;
>  
>      node_type = drmGetMinorType(min);
> -- 
> 2.18.0.rc1
> 
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel
Emil Velikov Aug. 31, 2018, 12:30 p.m. UTC | #2
Hi Thomas,

On 31 August 2018 at 12:54, Thomas Hellstrom <thellstrom@vmware.com> wrote:
> To determine whether a device node is a drm device node or not, the code
> currently compares the node's major number to the static drm major device
> number.
>
> This breaks the standalone vmwgfx driver on XWayland dri clients,
> https://cgit.freedesktop.org/mesa/vmwgfx

Any particular reason why the code doesn't use a fixed node there?
It will make the diff vs the in-kernel driver a bit smaller.

> and any future attempt to introduce dynamic device numbers for drm.
>
I'm not sure how well any such attempt will pan out, regardless of the
libdrm checks.

Namely: the static 226 has been used by a number of tools that
interpose the libc' ioctl function.
There could be others that also depend on it.

Personally, I'd go with the kernel developers decision.

Dave, Daniel, others
Should we keep or drop the major == 226 checks.

Thanks
Emil
Thomas Hellstrom Aug. 31, 2018, 1:05 p.m. UTC | #3
Hi, Emil
On 08/31/2018 02:30 PM, Emil Velikov wrote:
> Hi Thomas,
>
> On 31 August 2018 at 12:54, Thomas Hellstrom <thellstrom@vmware.com> wrote:
>> To determine whether a device node is a drm device node or not, the code
>> currently compares the node's major number to the static drm major device
>> number.
>>
>> This breaks the standalone vmwgfx driver on XWayland dri clients,
>>
> Any particular reason why the code doesn't use a fixed node there?
> It will make the diff vs the in-kernel driver a bit smaller.
Because then it won't be able to interoperate with other in-tree 
drivers, like virtual drm drivers or passthrough usb drm drivers.
There is no clean way to share the minor number allocation with in-tree 
drm, so standalone vmwgfx is using dynamic major allocation.

>
>> and any future attempt to introduce dynamic device numbers for drm.
>>
> I'm not sure how well any such attempt will pan out, regardless of the
> libdrm checks.
>
> Namely: the static 226 has been used by a number of tools that
> interpose the libc' ioctl function.
> There could be others that also depend on it.

True, in any case for existing drivers changing static 226 to something 
else is at least 10+ years away according to Linus' policy, so the main 
issue here is really to get rid of a big annoyance in the standalone 
vmwgfx case.

/Thomas


>
> Personally, I'd go with the kernel developers decision.
>
> Dave, Daniel, others
> Should we keep or drop the major == 226 checks.
>
> Thanks
> Emil
Michel Dänzer Aug. 31, 2018, 2:38 p.m. UTC | #4
[ Adding the amd-gfx list ]

On 2018-08-31 3:05 p.m., Thomas Hellstrom wrote:
> On 08/31/2018 02:30 PM, Emil Velikov wrote:
>> On 31 August 2018 at 12:54, Thomas Hellstrom <thellstrom@vmware.com>
>> wrote:
>>> To determine whether a device node is a drm device node or not, the code
>>> currently compares the node's major number to the static drm major
>>> device
>>> number.
>>>
>>> This breaks the standalone vmwgfx driver on XWayland dri clients,
>>>
>> Any particular reason why the code doesn't use a fixed node there?
>> It will make the diff vs the in-kernel driver a bit smaller.
> Because then it won't be able to interoperate with other in-tree
> drivers, like virtual drm drivers or passthrough usb drm drivers.
> There is no clean way to share the minor number allocation with in-tree
> drm, so standalone vmwgfx is using dynamic major allocation.

I wonder why I haven't heard of any of these issues with the standalone
version of amdgpu shipped in packaged AMD releases. Does that also use a
different major number? If yes, maybe it's just that nobody has tried
Xwayland clients with that driver. If no, how does it avoid the other
issues described above?
Thomas Hellstrom Aug. 31, 2018, 2:46 p.m. UTC | #5
On 08/31/2018 04:38 PM, Michel Dänzer wrote:
> [ Adding the amd-gfx list ]
>
> On 2018-08-31 3:05 p.m., Thomas Hellstrom wrote:
>> On 08/31/2018 02:30 PM, Emil Velikov wrote:
>>> On 31 August 2018 at 12:54, Thomas Hellstrom <thellstrom@vmware.com>
>>> wrote:
>>>> To determine whether a device node is a drm device node or not, the code
>>>> currently compares the node's major number to the static drm major
>>>> device
>>>> number.
>>>>
>>>> This breaks the standalone vmwgfx driver on XWayland dri clients,
>>>>
>>> Any particular reason why the code doesn't use a fixed node there?
>>> It will make the diff vs the in-kernel driver a bit smaller.
>> Because then it won't be able to interoperate with other in-tree
>> drivers, like virtual drm drivers or passthrough usb drm drivers.
>> There is no clean way to share the minor number allocation with in-tree
>> drm, so standalone vmwgfx is using dynamic major allocation.
> I wonder why I haven't heard of any of these issues with the standalone
> version of amdgpu shipped in packaged AMD releases. Does that also use a
> different major number? If yes, maybe it's just that nobody has tried
> Xwayland clients with that driver. If no, how does it avoid the other
> issues described above?
>
>
Is standalone AMD supposed to be able to coexist with in-tree drm drivers?

/Thomas
Michel Dänzer Aug. 31, 2018, 2:49 p.m. UTC | #6
On 2018-08-31 4:46 p.m., Thomas Hellstrom wrote:
> On 08/31/2018 04:38 PM, Michel Dänzer wrote:
>> [ Adding the amd-gfx list ]
>>
>> On 2018-08-31 3:05 p.m., Thomas Hellstrom wrote:
>>> On 08/31/2018 02:30 PM, Emil Velikov wrote:
>>>> On 31 August 2018 at 12:54, Thomas Hellstrom <thellstrom@vmware.com>
>>>> wrote:
>>>>> To determine whether a device node is a drm device node or not, the
>>>>> code
>>>>> currently compares the node's major number to the static drm major
>>>>> device
>>>>> number.
>>>>>
>>>>> This breaks the standalone vmwgfx driver on XWayland dri clients,
>>>>>
>>>> Any particular reason why the code doesn't use a fixed node there?
>>>> It will make the diff vs the in-kernel driver a bit smaller.
>>> Because then it won't be able to interoperate with other in-tree
>>> drivers, like virtual drm drivers or passthrough usb drm drivers.
>>> There is no clean way to share the minor number allocation with in-tree
>>> drm, so standalone vmwgfx is using dynamic major allocation.
>> I wonder why I haven't heard of any of these issues with the standalone
>> version of amdgpu shipped in packaged AMD releases. Does that also use a
>> different major number? If yes, maybe it's just that nobody has tried
>> Xwayland clients with that driver. If no, how does it avoid the other
>> issues described above?
>>
>>
> Is standalone AMD supposed to be able to coexist with in-tree drm drivers?

Yes, it does, it's working e.g. on laptops with an Intel integrated and
an AMD discrete GPU.
Thomas Hellstrom Aug. 31, 2018, 2:59 p.m. UTC | #7
On 08/31/2018 04:49 PM, Michel Dänzer wrote:
> On 2018-08-31 4:46 p.m., Thomas Hellstrom wrote:
>> On 08/31/2018 04:38 PM, Michel Dänzer wrote:
>>> [ Adding the amd-gfx list ]
>>>
>>> On 2018-08-31 3:05 p.m., Thomas Hellstrom wrote:
>>>> On 08/31/2018 02:30 PM, Emil Velikov wrote:
>>>>> On 31 August 2018 at 12:54, Thomas Hellstrom <thellstrom@vmware.com>
>>>>> wrote:
>>>>>> To determine whether a device node is a drm device node or not, the
>>>>>> code
>>>>>> currently compares the node's major number to the static drm major
>>>>>> device
>>>>>> number.
>>>>>>
>>>>>> This breaks the standalone vmwgfx driver on XWayland dri clients,
>>>>>>
>>>>> Any particular reason why the code doesn't use a fixed node there?
>>>>> It will make the diff vs the in-kernel driver a bit smaller.
>>>> Because then it won't be able to interoperate with other in-tree
>>>> drivers, like virtual drm drivers or passthrough usb drm drivers.
>>>> There is no clean way to share the minor number allocation with in-tree
>>>> drm, so standalone vmwgfx is using dynamic major allocation.
>>> I wonder why I haven't heard of any of these issues with the standalone
>>> version of amdgpu shipped in packaged AMD releases. Does that also use a
>>> different major number? If yes, maybe it's just that nobody has tried
>>> Xwayland clients with that driver. If no, how does it avoid the other
>>> issues described above?
>>>
>>>
>> Is standalone AMD supposed to be able to coexist with in-tree drm drivers?
> Yes, it does, it's working e.g. on laptops with an Intel integrated and
> an AMD discrete GPU.
>
>

Hmm. The symptoms with xf86-video-vmware are that when mesa initializes, 
we get:
MESA-LOADER: failed to retrieve device information
MESA-LOADER: failed to retrieve device information
MESA-LOADER: failed to retrieve device information

but then vmwgfx_dri.so loads anyway.

With XWayland, mesa just silently tries swrast instead of vmwgfx.

Not sure this has always been the case though. It might be due to a 
recent XWayland change. In any case, the change to libdrm silence the 
warnings on Xorg and makes mesa try vmwgfx on XWayland.

/Thomas
Emil Velikov Aug. 31, 2018, 3:27 p.m. UTC | #8
On 31 August 2018 at 15:38, Michel Dänzer <michel@daenzer.net> wrote:
>
> [ Adding the amd-gfx list ]
>
> On 2018-08-31 3:05 p.m., Thomas Hellstrom wrote:
>> On 08/31/2018 02:30 PM, Emil Velikov wrote:
>>> On 31 August 2018 at 12:54, Thomas Hellstrom <thellstrom@vmware.com>
>>> wrote:
>>>> To determine whether a device node is a drm device node or not, the code
>>>> currently compares the node's major number to the static drm major
>>>> device
>>>> number.
>>>>
>>>> This breaks the standalone vmwgfx driver on XWayland dri clients,
>>>>
>>> Any particular reason why the code doesn't use a fixed node there?
>>> It will make the diff vs the in-kernel driver a bit smaller.
>> Because then it won't be able to interoperate with other in-tree
>> drivers, like virtual drm drivers or passthrough usb drm drivers.
>> There is no clean way to share the minor number allocation with in-tree
>> drm, so standalone vmwgfx is using dynamic major allocation.
>
> I wonder why I haven't heard of any of these issues with the standalone
> version of amdgpu shipped in packaged AMD releases. Does that also use a
> different major number? If yes, maybe it's just that nobody has tried
> Xwayland clients with that driver. If no, how does it avoid the other
> issues described above?
>
AFAICT, the difference is that the standalone vmwgfx uses an internal
copy of drm core.
It doesn't reuse the in-kernel drm, hence it cannot know which minor it can use.

-Emil
Thomas Hellstrom Aug. 31, 2018, 3:30 p.m. UTC | #9
On 08/31/2018 05:27 PM, Emil Velikov wrote:
> On 31 August 2018 at 15:38, Michel Dänzer <michel@daenzer.net> wrote:
>> [ Adding the amd-gfx list ]
>>
>> On 2018-08-31 3:05 p.m., Thomas Hellstrom wrote:
>>> On 08/31/2018 02:30 PM, Emil Velikov wrote:
>>>> On 31 August 2018 at 12:54, Thomas Hellstrom <thellstrom@vmware.com>
>>>> wrote:
>>>>> To determine whether a device node is a drm device node or not, the code
>>>>> currently compares the node's major number to the static drm major
>>>>> device
>>>>> number.
>>>>>
>>>>> This breaks the standalone vmwgfx driver on XWayland dri clients,
>>>>>
>>>> Any particular reason why the code doesn't use a fixed node there?
>>>> It will make the diff vs the in-kernel driver a bit smaller.
>>> Because then it won't be able to interoperate with other in-tree
>>> drivers, like virtual drm drivers or passthrough usb drm drivers.
>>> There is no clean way to share the minor number allocation with in-tree
>>> drm, so standalone vmwgfx is using dynamic major allocation.
>> I wonder why I haven't heard of any of these issues with the standalone
>> version of amdgpu shipped in packaged AMD releases. Does that also use a
>> different major number? If yes, maybe it's just that nobody has tried
>> Xwayland clients with that driver. If no, how does it avoid the other
>> issues described above?
>>
> AFAICT, the difference is that the standalone vmwgfx uses an internal
> copy of drm core.
> It doesn't reuse the in-kernel drm, hence it cannot know which minor it can use.
>
> -Emil

Actually, standalone vmwgfx could perhaps also try to allocate minors 
from 63 and downwards. That might work, but needs some verification.

/Thomas
Christian König Aug. 31, 2018, 3:31 p.m. UTC | #10
Am 31.08.2018 um 17:27 schrieb Emil Velikov:
> On 31 August 2018 at 15:38, Michel Dänzer <michel@daenzer.net> wrote:
>> [ Adding the amd-gfx list ]
>>
>> On 2018-08-31 3:05 p.m., Thomas Hellstrom wrote:
>>> On 08/31/2018 02:30 PM, Emil Velikov wrote:
>>>> On 31 August 2018 at 12:54, Thomas Hellstrom <thellstrom@vmware.com>
>>>> wrote:
>>>>> To determine whether a device node is a drm device node or not, the code
>>>>> currently compares the node's major number to the static drm major
>>>>> device
>>>>> number.
>>>>>
>>>>> This breaks the standalone vmwgfx driver on XWayland dri clients,
>>>>>
>>>> Any particular reason why the code doesn't use a fixed node there?
>>>> It will make the diff vs the in-kernel driver a bit smaller.
>>> Because then it won't be able to interoperate with other in-tree
>>> drivers, like virtual drm drivers or passthrough usb drm drivers.
>>> There is no clean way to share the minor number allocation with in-tree
>>> drm, so standalone vmwgfx is using dynamic major allocation.
>> I wonder why I haven't heard of any of these issues with the standalone
>> version of amdgpu shipped in packaged AMD releases. Does that also use a
>> different major number? If yes, maybe it's just that nobody has tried
>> Xwayland clients with that driver. If no, how does it avoid the other
>> issues described above?
>>
> AFAICT, the difference is that the standalone vmwgfx uses an internal
> copy of drm core.
> It doesn't reuse the in-kernel drm, hence it cannot know which minor it can use.

The amdgpu pro package has it's own drm core copy as well and there it 
still works.

Not sure how our back-porting guys handle that.

Christian.

>
> -Emil
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel
Alex Deucher Sept. 2, 2018, 2:54 p.m. UTC | #11
On Fri, Aug 31, 2018 at 11:32 AM Christian König
<ckoenig.leichtzumerken@gmail.com> wrote:
>
> Am 31.08.2018 um 17:27 schrieb Emil Velikov:
> > On 31 August 2018 at 15:38, Michel Dänzer <michel@daenzer.net> wrote:
> >> [ Adding the amd-gfx list ]
> >>
> >> On 2018-08-31 3:05 p.m., Thomas Hellstrom wrote:
> >>> On 08/31/2018 02:30 PM, Emil Velikov wrote:
> >>>> On 31 August 2018 at 12:54, Thomas Hellstrom <thellstrom@vmware.com>
> >>>> wrote:
> >>>>> To determine whether a device node is a drm device node or not, the code
> >>>>> currently compares the node's major number to the static drm major
> >>>>> device
> >>>>> number.
> >>>>>
> >>>>> This breaks the standalone vmwgfx driver on XWayland dri clients,
> >>>>>
> >>>> Any particular reason why the code doesn't use a fixed node there?
> >>>> It will make the diff vs the in-kernel driver a bit smaller.
> >>> Because then it won't be able to interoperate with other in-tree
> >>> drivers, like virtual drm drivers or passthrough usb drm drivers.
> >>> There is no clean way to share the minor number allocation with in-tree
> >>> drm, so standalone vmwgfx is using dynamic major allocation.
> >> I wonder why I haven't heard of any of these issues with the standalone
> >> version of amdgpu shipped in packaged AMD releases. Does that also use a
> >> different major number? If yes, maybe it's just that nobody has tried
> >> Xwayland clients with that driver. If no, how does it avoid the other
> >> issues described above?
> >>
> > AFAICT, the difference is that the standalone vmwgfx uses an internal
> > copy of drm core.
> > It doesn't reuse the in-kernel drm, hence it cannot know which minor it can use.
>
> The amdgpu pro package has it's own drm core copy as well and there it
> still works.

We don't use our own copy of drm core in the kernel, we rely on the in
kernel one.  Just ttm and amdgpu are updated in the dkms packages.

Alex

>
> Not sure how our back-porting guys handle that.
>
> Christian.
>
> >
> > -Emil
> > _______________________________________________
> > dri-devel mailing list
> > dri-devel@lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/dri-devel
>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
Thomas Hellstrom Sept. 3, 2018, 9:16 a.m. UTC | #12
On 08/31/2018 05:30 PM, Thomas Hellstrom wrote:
> On 08/31/2018 05:27 PM, Emil Velikov wrote:
>> On 31 August 2018 at 15:38, Michel Dänzer <michel@daenzer.net> wrote:
>>> [ Adding the amd-gfx list ]
>>>
>>> On 2018-08-31 3:05 p.m., Thomas Hellstrom wrote:
>>>> On 08/31/2018 02:30 PM, Emil Velikov wrote:
>>>>> On 31 August 2018 at 12:54, Thomas Hellstrom <thellstrom@vmware.com>
>>>>> wrote:
>>>>>> To determine whether a device node is a drm device node or not, 
>>>>>> the code
>>>>>> currently compares the node's major number to the static drm major
>>>>>> device
>>>>>> number.
>>>>>>
>>>>>> This breaks the standalone vmwgfx driver on XWayland dri clients,
>>>>>>
>>>>> Any particular reason why the code doesn't use a fixed node there?
>>>>> It will make the diff vs the in-kernel driver a bit smaller.
>>>> Because then it won't be able to interoperate with other in-tree
>>>> drivers, like virtual drm drivers or passthrough usb drm drivers.
>>>> There is no clean way to share the minor number allocation with 
>>>> in-tree
>>>> drm, so standalone vmwgfx is using dynamic major allocation.
>>> I wonder why I haven't heard of any of these issues with the standalone
>>> version of amdgpu shipped in packaged AMD releases. Does that also 
>>> use a
>>> different major number? If yes, maybe it's just that nobody has tried
>>> Xwayland clients with that driver. If no, how does it avoid the other
>>> issues described above?
>>>
>> AFAICT, the difference is that the standalone vmwgfx uses an internal
>> copy of drm core.
>> It doesn't reuse the in-kernel drm, hence it cannot know which minor 
>> it can use.
>>
>> -Emil
>
> Actually, standalone vmwgfx could perhaps also try to allocate minors 
> from 63 and downwards. That might work, but needs some verification.
>

So unfortuntately this doesn't work since the in-tree drm's file 
operations are registered with the DRM_MAJOR.
So I still think the patch is the way to go. If people are concerned 
that also fbdev file descriptors are allowed, perhaps there are other 
sysfs traits we can look at?

/Thomas




> /Thomas
>
Daniel Vetter Sept. 3, 2018, 4:33 p.m. UTC | #13
On Mon, Sep 03, 2018 at 11:16:29AM +0200, Thomas Hellstrom wrote:
> On 08/31/2018 05:30 PM, Thomas Hellstrom wrote:
> > On 08/31/2018 05:27 PM, Emil Velikov wrote:
> > > On 31 August 2018 at 15:38, Michel Dänzer <michel@daenzer.net> wrote:
> > > > [ Adding the amd-gfx list ]
> > > > 
> > > > On 2018-08-31 3:05 p.m., Thomas Hellstrom wrote:
> > > > > On 08/31/2018 02:30 PM, Emil Velikov wrote:
> > > > > > On 31 August 2018 at 12:54, Thomas Hellstrom <thellstrom@vmware.com>
> > > > > > wrote:
> > > > > > > To determine whether a device node is a drm device
> > > > > > > node or not, the code
> > > > > > > currently compares the node's major number to the static drm major
> > > > > > > device
> > > > > > > number.
> > > > > > > 
> > > > > > > This breaks the standalone vmwgfx driver on XWayland dri clients,
> > > > > > > 
> > > > > > Any particular reason why the code doesn't use a fixed node there?
> > > > > > It will make the diff vs the in-kernel driver a bit smaller.
> > > > > Because then it won't be able to interoperate with other in-tree
> > > > > drivers, like virtual drm drivers or passthrough usb drm drivers.
> > > > > There is no clean way to share the minor number allocation
> > > > > with in-tree
> > > > > drm, so standalone vmwgfx is using dynamic major allocation.
> > > > I wonder why I haven't heard of any of these issues with the standalone
> > > > version of amdgpu shipped in packaged AMD releases. Does that
> > > > also use a
> > > > different major number? If yes, maybe it's just that nobody has tried
> > > > Xwayland clients with that driver. If no, how does it avoid the other
> > > > issues described above?
> > > > 
> > > AFAICT, the difference is that the standalone vmwgfx uses an internal
> > > copy of drm core.
> > > It doesn't reuse the in-kernel drm, hence it cannot know which minor
> > > it can use.
> > > 
> > > -Emil
> > 
> > Actually, standalone vmwgfx could perhaps also try to allocate minors
> > from 63 and downwards. That might work, but needs some verification.
> > 
> 
> So unfortuntately this doesn't work since the in-tree drm's file operations
> are registered with the DRM_MAJOR.
> So I still think the patch is the way to go. If people are concerned that
> also fbdev file descriptors are allowed, perhaps there are other sysfs
> traits we can look at?

Somewhat out of curiosity, but why do you have to overwrite all of drm?
amdgpu seems to be able to pull their stunt off without ...
-Daniel
Thomas Hellstrom Sept. 3, 2018, 5 p.m. UTC | #14
On 09/03/2018 06:33 PM, Daniel Vetter wrote:
> On Mon, Sep 03, 2018 at 11:16:29AM +0200, Thomas Hellstrom wrote:
>> On 08/31/2018 05:30 PM, Thomas Hellstrom wrote:
>>> On 08/31/2018 05:27 PM, Emil Velikov wrote:
>>>> On 31 August 2018 at 15:38, Michel Dänzer <michel@daenzer.net> wrote:
>>>>> [ Adding the amd-gfx list ]
>>>>>
>>>>> On 2018-08-31 3:05 p.m., Thomas Hellstrom wrote:
>>>>>> On 08/31/2018 02:30 PM, Emil Velikov wrote:
>>>>>>> On 31 August 2018 at 12:54, Thomas Hellstrom <thellstrom@vmware.com>
>>>>>>> wrote:
>>>>>>>> To determine whether a device node is a drm device
>>>>>>>> node or not, the code
>>>>>>>> currently compares the node's major number to the static drm major
>>>>>>>> device
>>>>>>>> number.
>>>>>>>>
>>>>>>>> This breaks the standalone vmwgfx driver on XWayland dri clients,
>>>>>>>>
>>>>>>> Any particular reason why the code doesn't use a fixed node there?
>>>>>>> It will make the diff vs the in-kernel driver a bit smaller.
>>>>>> Because then it won't be able to interoperate with other in-tree
>>>>>> drivers, like virtual drm drivers or passthrough usb drm drivers.
>>>>>> There is no clean way to share the minor number allocation
>>>>>> with in-tree
>>>>>> drm, so standalone vmwgfx is using dynamic major allocation.
>>>>> I wonder why I haven't heard of any of these issues with the standalone
>>>>> version of amdgpu shipped in packaged AMD releases. Does that
>>>>> also use a
>>>>> different major number? If yes, maybe it's just that nobody has tried
>>>>> Xwayland clients with that driver. If no, how does it avoid the other
>>>>> issues described above?
>>>>>
>>>> AFAICT, the difference is that the standalone vmwgfx uses an internal
>>>> copy of drm core.
>>>> It doesn't reuse the in-kernel drm, hence it cannot know which minor
>>>> it can use.
>>>>
>>>> -Emil
>>> Actually, standalone vmwgfx could perhaps also try to allocate minors
>>> from 63 and downwards. That might work, but needs some verification.
>>>
>> So unfortuntately this doesn't work since the in-tree drm's file operations
>> are registered with the DRM_MAJOR.
>> So I still think the patch is the way to go. If people are concerned that
>> also fbdev file descriptors are allowed, perhaps there are other sysfs
>> traits we can look at?
> Somewhat out of curiosity, but why do you have to overwrite all of drm?
> amdgpu seems to be able to pull their stunt off without ...
> -Daniel

At the time we launched the standalone vmwgfx, the DRM <-> driver 
interface was moving considerably more rapidly than the DRM <-> kernel 
interface. I think that's still the case. Hence less work for us. Also 
meant we can install the full driver stack with latest features on 
fairly old VMs without backported DRM functionality.

/Thomas
Dave Airlie Sept. 4, 2018, 10:33 p.m. UTC | #15
On Tue, 4 Sep 2018 at 03:00, Thomas Hellstrom <thellstrom@vmware.com> wrote:
>
> On 09/03/2018 06:33 PM, Daniel Vetter wrote:
> > On Mon, Sep 03, 2018 at 11:16:29AM +0200, Thomas Hellstrom wrote:
> >> On 08/31/2018 05:30 PM, Thomas Hellstrom wrote:
> >>> On 08/31/2018 05:27 PM, Emil Velikov wrote:
> >>>> On 31 August 2018 at 15:38, Michel Dänzer <michel@daenzer.net> wrote:
> >>>>> [ Adding the amd-gfx list ]
> >>>>>
> >>>>> On 2018-08-31 3:05 p.m., Thomas Hellstrom wrote:
> >>>>>> On 08/31/2018 02:30 PM, Emil Velikov wrote:
> >>>>>>> On 31 August 2018 at 12:54, Thomas Hellstrom <thellstrom@vmware.com>
> >>>>>>> wrote:
> >>>>>>>> To determine whether a device node is a drm device
> >>>>>>>> node or not, the code
> >>>>>>>> currently compares the node's major number to the static drm major
> >>>>>>>> device
> >>>>>>>> number.
> >>>>>>>>
> >>>>>>>> This breaks the standalone vmwgfx driver on XWayland dri clients,
> >>>>>>>>
> >>>>>>> Any particular reason why the code doesn't use a fixed node there?
> >>>>>>> It will make the diff vs the in-kernel driver a bit smaller.
> >>>>>> Because then it won't be able to interoperate with other in-tree
> >>>>>> drivers, like virtual drm drivers or passthrough usb drm drivers.
> >>>>>> There is no clean way to share the minor number allocation
> >>>>>> with in-tree
> >>>>>> drm, so standalone vmwgfx is using dynamic major allocation.
> >>>>> I wonder why I haven't heard of any of these issues with the standalone
> >>>>> version of amdgpu shipped in packaged AMD releases. Does that
> >>>>> also use a
> >>>>> different major number? If yes, maybe it's just that nobody has tried
> >>>>> Xwayland clients with that driver. If no, how does it avoid the other
> >>>>> issues described above?
> >>>>>
> >>>> AFAICT, the difference is that the standalone vmwgfx uses an internal
> >>>> copy of drm core.
> >>>> It doesn't reuse the in-kernel drm, hence it cannot know which minor
> >>>> it can use.
> >>>>
> >>>> -Emil
> >>> Actually, standalone vmwgfx could perhaps also try to allocate minors
> >>> from 63 and downwards. That might work, but needs some verification.
> >>>
> >> So unfortuntately this doesn't work since the in-tree drm's file operations
> >> are registered with the DRM_MAJOR.
> >> So I still think the patch is the way to go. If people are concerned that
> >> also fbdev file descriptors are allowed, perhaps there are other sysfs
> >> traits we can look at?
> > Somewhat out of curiosity, but why do you have to overwrite all of drm?
> > amdgpu seems to be able to pull their stunt off without ...
> > -Daniel
>
> At the time we launched the standalone vmwgfx, the DRM <-> driver
> interface was moving considerably more rapidly than the DRM <-> kernel
> interface. I think that's still the case. Hence less work for us. Also
> meant we can install the full driver stack with latest features on
> fairly old VMs without backported DRM functionality.
>

I think this should be fine for 99% of drm usage, there may be corner
cases in wierd places, but I can't point to any that really matter
(maybe strace?)

Acked-by: Dave Airlie <airlied@redhat.com>

Dave.
Emil Velikov Sept. 5, 2018, 9:33 a.m. UTC | #16
On 4 September 2018 at 23:33, Dave Airlie <airlied@gmail.com> wrote:
> On Tue, 4 Sep 2018 at 03:00, Thomas Hellstrom <thellstrom@vmware.com> wrote:
>>
>> On 09/03/2018 06:33 PM, Daniel Vetter wrote:
>> > On Mon, Sep 03, 2018 at 11:16:29AM +0200, Thomas Hellstrom wrote:
>> >> On 08/31/2018 05:30 PM, Thomas Hellstrom wrote:
>> >>> On 08/31/2018 05:27 PM, Emil Velikov wrote:
>> >>>> On 31 August 2018 at 15:38, Michel Dänzer <michel@daenzer.net> wrote:
>> >>>>> [ Adding the amd-gfx list ]
>> >>>>>
>> >>>>> On 2018-08-31 3:05 p.m., Thomas Hellstrom wrote:
>> >>>>>> On 08/31/2018 02:30 PM, Emil Velikov wrote:
>> >>>>>>> On 31 August 2018 at 12:54, Thomas Hellstrom <thellstrom@vmware.com>
>> >>>>>>> wrote:
>> >>>>>>>> To determine whether a device node is a drm device
>> >>>>>>>> node or not, the code
>> >>>>>>>> currently compares the node's major number to the static drm major
>> >>>>>>>> device
>> >>>>>>>> number.
>> >>>>>>>>
>> >>>>>>>> This breaks the standalone vmwgfx driver on XWayland dri clients,
>> >>>>>>>>
>> >>>>>>> Any particular reason why the code doesn't use a fixed node there?
>> >>>>>>> It will make the diff vs the in-kernel driver a bit smaller.
>> >>>>>> Because then it won't be able to interoperate with other in-tree
>> >>>>>> drivers, like virtual drm drivers or passthrough usb drm drivers.
>> >>>>>> There is no clean way to share the minor number allocation
>> >>>>>> with in-tree
>> >>>>>> drm, so standalone vmwgfx is using dynamic major allocation.
>> >>>>> I wonder why I haven't heard of any of these issues with the standalone
>> >>>>> version of amdgpu shipped in packaged AMD releases. Does that
>> >>>>> also use a
>> >>>>> different major number? If yes, maybe it's just that nobody has tried
>> >>>>> Xwayland clients with that driver. If no, how does it avoid the other
>> >>>>> issues described above?
>> >>>>>
>> >>>> AFAICT, the difference is that the standalone vmwgfx uses an internal
>> >>>> copy of drm core.
>> >>>> It doesn't reuse the in-kernel drm, hence it cannot know which minor
>> >>>> it can use.
>> >>>>
>> >>>> -Emil
>> >>> Actually, standalone vmwgfx could perhaps also try to allocate minors
>> >>> from 63 and downwards. That might work, but needs some verification.
>> >>>
>> >> So unfortuntately this doesn't work since the in-tree drm's file operations
>> >> are registered with the DRM_MAJOR.
>> >> So I still think the patch is the way to go. If people are concerned that
>> >> also fbdev file descriptors are allowed, perhaps there are other sysfs
>> >> traits we can look at?
>> > Somewhat out of curiosity, but why do you have to overwrite all of drm?
>> > amdgpu seems to be able to pull their stunt off without ...
>> > -Daniel
>>
>> At the time we launched the standalone vmwgfx, the DRM <-> driver
>> interface was moving considerably more rapidly than the DRM <-> kernel
>> interface. I think that's still the case. Hence less work for us. Also
>> meant we can install the full driver stack with latest features on
>> fairly old VMs without backported DRM functionality.
>>
>
> I think this should be fine for 99% of drm usage, there may be corner
> cases in wierd places, but I can't point to any that really matter
> (maybe strace?)
>
Having a closer look, I think this will break the Firefox/Chrome sandboxing :-\
I cannot see the path /sys/dev/char/%d:%d/device/drm in the allowed
list [1] [2].

Thomas, can you please send a patch to the respective teams or give
them a heads up?

Thanks
Emil

[1] https://hg.mozilla.org/releases/mozilla-beta/file/264fcd3206a6/security/sandbox/linux/broker/SandboxBrokerPolicyFactory.cpp
[2] https://chromium.googlesource.com/chromium/src/+/8655d49f657d3878c937f1387b3d31fa66c8e76a/content/gpu/gpu_sandbox_hook_linux.cc
Thomas Hellstrom Sept. 5, 2018, 10:10 a.m. UTC | #17
Hi, Emil,

On 09/05/2018 11:33 AM, Emil Velikov wrote:
> On 4 September 2018 at 23:33, Dave Airlie <airlied@gmail.com> wrote:
>> On Tue, 4 Sep 2018 at 03:00, Thomas Hellstrom <thellstrom@vmware.com> wrote:
>>> On 09/03/2018 06:33 PM, Daniel Vetter wrote:
>>>> On Mon, Sep 03, 2018 at 11:16:29AM +0200, Thomas Hellstrom wrote:
>>>>> On 08/31/2018 05:30 PM, Thomas Hellstrom wrote:
>>>>>> On 08/31/2018 05:27 PM, Emil Velikov wrote:
>>>>>>> On 31 August 2018 at 15:38, Michel Dänzer <michel@daenzer.net> wrote:
>>>>>>>> [ Adding the amd-gfx list ]
>>>>>>>>
>>>>>>>> On 2018-08-31 3:05 p.m., Thomas Hellstrom wrote:
>>>>>>>>> On 08/31/2018 02:30 PM, Emil Velikov wrote:
>>>>>>>>>> On 31 August 2018 at 12:54, Thomas Hellstrom <thellstrom@vmware.com>
>>>>>>>>>> wrote:
>>>>>>>>>>> To determine whether a device node is a drm device
>>>>>>>>>>> node or not, the code
>>>>>>>>>>> currently compares the node's major number to the static drm major
>>>>>>>>>>> device
>>>>>>>>>>> number.
>>>>>>>>>>>
>>>>>>>>>>> This breaks the standalone vmwgfx driver on XWayland dri clients,
>>>>>>>>>>>
>>>>>>>>>> Any particular reason why the code doesn't use a fixed node there?
>>>>>>>>>> It will make the diff vs the in-kernel driver a bit smaller.
>>>>>>>>> Because then it won't be able to interoperate with other in-tree
>>>>>>>>> drivers, like virtual drm drivers or passthrough usb drm drivers.
>>>>>>>>> There is no clean way to share the minor number allocation
>>>>>>>>> with in-tree
>>>>>>>>> drm, so standalone vmwgfx is using dynamic major allocation.
>>>>>>>> I wonder why I haven't heard of any of these issues with the standalone
>>>>>>>> version of amdgpu shipped in packaged AMD releases. Does that
>>>>>>>> also use a
>>>>>>>> different major number? If yes, maybe it's just that nobody has tried
>>>>>>>> Xwayland clients with that driver. If no, how does it avoid the other
>>>>>>>> issues described above?
>>>>>>>>
>>>>>>> AFAICT, the difference is that the standalone vmwgfx uses an internal
>>>>>>> copy of drm core.
>>>>>>> It doesn't reuse the in-kernel drm, hence it cannot know which minor
>>>>>>> it can use.
>>>>>>>
>>>>>>> -Emil
>>>>>> Actually, standalone vmwgfx could perhaps also try to allocate minors
>>>>>> from 63 and downwards. That might work, but needs some verification.
>>>>>>
>>>>> So unfortuntately this doesn't work since the in-tree drm's file operations
>>>>> are registered with the DRM_MAJOR.
>>>>> So I still think the patch is the way to go. If people are concerned that
>>>>> also fbdev file descriptors are allowed, perhaps there are other sysfs
>>>>> traits we can look at?
>>>> Somewhat out of curiosity, but why do you have to overwrite all of drm?
>>>> amdgpu seems to be able to pull their stunt off without ...
>>>> -Daniel
>>> At the time we launched the standalone vmwgfx, the DRM <-> driver
>>> interface was moving considerably more rapidly than the DRM <-> kernel
>>> interface. I think that's still the case. Hence less work for us. Also
>>> meant we can install the full driver stack with latest features on
>>> fairly old VMs without backported DRM functionality.
>>>
>> I think this should be fine for 99% of drm usage, there may be corner
>> cases in wierd places, but I can't point to any that really matter
>> (maybe strace?)
>>
> Having a closer look, I think this will break the Firefox/Chrome sandboxing :-\
> I cannot see the path /sys/dev/char/%d:%d/device/drm in the allowed
> list [1] [2].
Thanks for pointing this out.

The function drmGetMinorNameForFD() already opens this path, so any 
user-space using that function would not work before either.

Also mozilla/firefox adds /sys/dev/char/226:* Which means that while it 
still won't work on standalone vmwgfx, there should at least be no 
regression.

For Chromium it seems they allow /sys/dev/char/ for AmdGpu, but only 
under ChromOS, so I'll ping those to be safe.

I also won't be doing an immediate release after pushing.

Thanks,
Thomas


> Thomas, can you please send a patch to the respective teams or give
> them a heads up?
>
> Thanks
> Emil
>
Emil Velikov Sept. 5, 2018, 1:07 p.m. UTC | #18
On 5 September 2018 at 11:10, Thomas Hellstrom <thellstrom@vmware.com> wrote:
> Hi, Emil,
>
>
> On 09/05/2018 11:33 AM, Emil Velikov wrote:
>>
>> On 4 September 2018 at 23:33, Dave Airlie <airlied@gmail.com> wrote:
>>>
>>> On Tue, 4 Sep 2018 at 03:00, Thomas Hellstrom <thellstrom@vmware.com>
>>> wrote:
>>>>
>>>> On 09/03/2018 06:33 PM, Daniel Vetter wrote:
>>>>>
>>>>> On Mon, Sep 03, 2018 at 11:16:29AM +0200, Thomas Hellstrom wrote:
>>>>>>
>>>>>> On 08/31/2018 05:30 PM, Thomas Hellstrom wrote:
>>>>>>>
>>>>>>> On 08/31/2018 05:27 PM, Emil Velikov wrote:
>>>>>>>>
>>>>>>>> On 31 August 2018 at 15:38, Michel Dänzer <michel@daenzer.net>
>>>>>>>> wrote:
>>>>>>>>>
>>>>>>>>> [ Adding the amd-gfx list ]
>>>>>>>>>
>>>>>>>>> On 2018-08-31 3:05 p.m., Thomas Hellstrom wrote:
>>>>>>>>>>
>>>>>>>>>> On 08/31/2018 02:30 PM, Emil Velikov wrote:
>>>>>>>>>>>
>>>>>>>>>>> On 31 August 2018 at 12:54, Thomas Hellstrom
>>>>>>>>>>> <thellstrom@vmware.com>
>>>>>>>>>>> wrote:
>>>>>>>>>>>>
>>>>>>>>>>>> To determine whether a device node is a drm device
>>>>>>>>>>>> node or not, the code
>>>>>>>>>>>> currently compares the node's major number to the static drm
>>>>>>>>>>>> major
>>>>>>>>>>>> device
>>>>>>>>>>>> number.
>>>>>>>>>>>>
>>>>>>>>>>>> This breaks the standalone vmwgfx driver on XWayland dri
>>>>>>>>>>>> clients,
>>>>>>>>>>>>
>>>>>>>>>>> Any particular reason why the code doesn't use a fixed node
>>>>>>>>>>> there?
>>>>>>>>>>> It will make the diff vs the in-kernel driver a bit smaller.
>>>>>>>>>>
>>>>>>>>>> Because then it won't be able to interoperate with other in-tree
>>>>>>>>>> drivers, like virtual drm drivers or passthrough usb drm drivers.
>>>>>>>>>> There is no clean way to share the minor number allocation
>>>>>>>>>> with in-tree
>>>>>>>>>> drm, so standalone vmwgfx is using dynamic major allocation.
>>>>>>>>>
>>>>>>>>> I wonder why I haven't heard of any of these issues with the
>>>>>>>>> standalone
>>>>>>>>> version of amdgpu shipped in packaged AMD releases. Does that
>>>>>>>>> also use a
>>>>>>>>> different major number? If yes, maybe it's just that nobody has
>>>>>>>>> tried
>>>>>>>>> Xwayland clients with that driver. If no, how does it avoid the
>>>>>>>>> other
>>>>>>>>> issues described above?
>>>>>>>>>
>>>>>>>> AFAICT, the difference is that the standalone vmwgfx uses an
>>>>>>>> internal
>>>>>>>> copy of drm core.
>>>>>>>> It doesn't reuse the in-kernel drm, hence it cannot know which minor
>>>>>>>> it can use.
>>>>>>>>
>>>>>>>> -Emil
>>>>>>>
>>>>>>> Actually, standalone vmwgfx could perhaps also try to allocate minors
>>>>>>> from 63 and downwards. That might work, but needs some verification.
>>>>>>>
>>>>>> So unfortuntately this doesn't work since the in-tree drm's file
>>>>>> operations
>>>>>> are registered with the DRM_MAJOR.
>>>>>> So I still think the patch is the way to go. If people are concerned
>>>>>> that
>>>>>> also fbdev file descriptors are allowed, perhaps there are other sysfs
>>>>>> traits we can look at?
>>>>>
>>>>> Somewhat out of curiosity, but why do you have to overwrite all of drm?
>>>>> amdgpu seems to be able to pull their stunt off without ...
>>>>> -Daniel
>>>>
>>>> At the time we launched the standalone vmwgfx, the DRM <-> driver
>>>> interface was moving considerably more rapidly than the DRM <-> kernel
>>>> interface. I think that's still the case. Hence less work for us. Also
>>>> meant we can install the full driver stack with latest features on
>>>> fairly old VMs without backported DRM functionality.
>>>>
>>> I think this should be fine for 99% of drm usage, there may be corner
>>> cases in wierd places, but I can't point to any that really matter
>>> (maybe strace?)
>>>
>> Having a closer look, I think this will break the Firefox/Chrome
>> sandboxing :-\
>> I cannot see the path /sys/dev/char/%d:%d/device/drm in the allowed
>> list [1] [2].
>
> Thanks for pointing this out.
>
> The function drmGetMinorNameForFD() already opens this path, so any
> user-space using that function would not work before either.
>
> Also mozilla/firefox adds /sys/dev/char/226:* Which means that while it
> still won't work on standalone vmwgfx, there should at least be no
> regression.
>
> For Chromium it seems they allow /sys/dev/char/ for AmdGpu, but only under
> ChromOS, so I'll ping those to be safe.
>
> I also won't be doing an immediate release after pushing.
>
In that case, please give me 24h to do a libdrm release before pushing.
I had to push some workarounds for the sandboxing mentioned earlier :-\

Thanks
Emil
Thomas Hellstrom Sept. 5, 2018, 1:20 p.m. UTC | #19
On 09/05/2018 03:07 PM, Emil Velikov wrote:
> On 5 September 2018 at 11:10, Thomas Hellstrom <thellstrom@vmware.com> wrote:
>> Hi, Emil,
>>
>>
>> On 09/05/2018 11:33 AM, Emil Velikov wrote:
>>> On 4 September 2018 at 23:33, Dave Airlie <airlied@gmail.com> wrote:
>>>> On Tue, 4 Sep 2018 at 03:00, Thomas Hellstrom <thellstrom@vmware.com>
>>>> wrote:
>>>>> On 09/03/2018 06:33 PM, Daniel Vetter wrote:
>>>>>> On Mon, Sep 03, 2018 at 11:16:29AM +0200, Thomas Hellstrom wrote:
>>>>>>> On 08/31/2018 05:30 PM, Thomas Hellstrom wrote:
>>>>>>>> On 08/31/2018 05:27 PM, Emil Velikov wrote:
>>>>>>>>> On 31 August 2018 at 15:38, Michel Dänzer <michel@daenzer.net>
>>>>>>>>> wrote:
>>>>>>>>>> [ Adding the amd-gfx list ]
>>>>>>>>>>
>>>>>>>>>> On 2018-08-31 3:05 p.m., Thomas Hellstrom wrote:
>>>>>>>>>>> On 08/31/2018 02:30 PM, Emil Velikov wrote:
>>>>>>>>>>>> On 31 August 2018 at 12:54, Thomas Hellstrom
>>>>>>>>>>>> <thellstrom@vmware.com>
>>>>>>>>>>>> wrote:
>>>>>>>>>>>>> To determine whether a device node is a drm device
>>>>>>>>>>>>> node or not, the code
>>>>>>>>>>>>> currently compares the node's major number to the static drm
>>>>>>>>>>>>> major
>>>>>>>>>>>>> device
>>>>>>>>>>>>> number.
>>>>>>>>>>>>>
>>>>>>>>>>>>> This breaks the standalone vmwgfx driver on XWayland dri
>>>>>>>>>>>>> clients,
>>>>>>>>>>>>>
>>>>>>>>>>>> Any particular reason why the code doesn't use a fixed node
>>>>>>>>>>>> there?
>>>>>>>>>>>> It will make the diff vs the in-kernel driver a bit smaller.
>>>>>>>>>>> Because then it won't be able to interoperate with other in-tree
>>>>>>>>>>> drivers, like virtual drm drivers or passthrough usb drm drivers.
>>>>>>>>>>> There is no clean way to share the minor number allocation
>>>>>>>>>>> with in-tree
>>>>>>>>>>> drm, so standalone vmwgfx is using dynamic major allocation.
>>>>>>>>>> I wonder why I haven't heard of any of these issues with the
>>>>>>>>>> standalone
>>>>>>>>>> version of amdgpu shipped in packaged AMD releases. Does that
>>>>>>>>>> also use a
>>>>>>>>>> different major number? If yes, maybe it's just that nobody has
>>>>>>>>>> tried
>>>>>>>>>> Xwayland clients with that driver. If no, how does it avoid the
>>>>>>>>>> other
>>>>>>>>>> issues described above?
>>>>>>>>>>
>>>>>>>>> AFAICT, the difference is that the standalone vmwgfx uses an
>>>>>>>>> internal
>>>>>>>>> copy of drm core.
>>>>>>>>> It doesn't reuse the in-kernel drm, hence it cannot know which minor
>>>>>>>>> it can use.
>>>>>>>>>
>>>>>>>>> -Emil
>>>>>>>> Actually, standalone vmwgfx could perhaps also try to allocate minors
>>>>>>>> from 63 and downwards. That might work, but needs some verification.
>>>>>>>>
>>>>>>> So unfortuntately this doesn't work since the in-tree drm's file
>>>>>>> operations
>>>>>>> are registered with the DRM_MAJOR.
>>>>>>> So I still think the patch is the way to go. If people are concerned
>>>>>>> that
>>>>>>> also fbdev file descriptors are allowed, perhaps there are other sysfs
>>>>>>> traits we can look at?
>>>>>> Somewhat out of curiosity, but why do you have to overwrite all of drm?
>>>>>> amdgpu seems to be able to pull their stunt off without ...
>>>>>> -Daniel
>>>>> At the time we launched the standalone vmwgfx, the DRM <-> driver
>>>>> interface was moving considerably more rapidly than the DRM <-> kernel
>>>>> interface. I think that's still the case. Hence less work for us. Also
>>>>> meant we can install the full driver stack with latest features on
>>>>> fairly old VMs without backported DRM functionality.
>>>>>
>>>> I think this should be fine for 99% of drm usage, there may be corner
>>>> cases in wierd places, but I can't point to any that really matter
>>>> (maybe strace?)
>>>>
>>> Having a closer look, I think this will break the Firefox/Chrome
>>> sandboxing :-\
>>> I cannot see the path /sys/dev/char/%d:%d/device/drm in the allowed
>>> list [1] [2].
>> Thanks for pointing this out.
>>
>> The function drmGetMinorNameForFD() already opens this path, so any
>> user-space using that function would not work before either.
>>
>> Also mozilla/firefox adds /sys/dev/char/226:* Which means that while it
>> still won't work on standalone vmwgfx, there should at least be no
>> regression.
>>
>> For Chromium it seems they allow /sys/dev/char/ for AmdGpu, but only under
>> ChromOS, so I'll ping those to be safe.
>>
>> I also won't be doing an immediate release after pushing.
>>
> In that case, please give me 24h to do a libdrm release before pushing.
> I had to push some workarounds for the sandboxing mentioned earlier :-\
>
> Thanks
> Emil

Ouch, I just pushed the patch, but feel free to cut the release just 
before that commit.

/Thomas
Emil Velikov Sept. 5, 2018, 1:53 p.m. UTC | #20
On 5 September 2018 at 14:20, Thomas Hellstrom <thellstrom@vmware.com> wrote:

>> In that case, please give me 24h to do a libdrm release before pushing.
>> I had to push some workarounds for the sandboxing mentioned earlier :-\
>>
>> Thanks
>> Emil
>
>
> Ouch, I just pushed the patch, but feel free to cut the release just before
> that commit.
>
That doesn't quite work. Barring any objections I'll: revert, release, reapply.

-Emil
Thomas Hellström (VMware) Sept. 30, 2018, 5:31 p.m. UTC | #21
Hi, Emil,

On 09/05/2018 03:53 PM, Emil Velikov wrote:
> On 5 September 2018 at 14:20, Thomas Hellstrom <thellstrom@vmware.com> wrote:
>
>>> In that case, please give me 24h to do a libdrm release before pushing.
>>> I had to push some workarounds for the sandboxing mentioned earlier :-\
>>>
>>> Thanks
>>> Emil
>>
>> Ouch, I just pushed the patch, but feel free to cut the release just before
>> that commit.
>>
> That doesn't quite work. Barring any objections I'll: revert, release, reapply.
>
> -Emil
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

What happened here? I can't really see my commit nor a revert nor a 
release in libdrm.

/Thomas
Thomas Hellstrom Oct. 2, 2018, 7:55 p.m. UTC | #22
Ping?

On 09/30/2018 07:31 PM, Thomas Hellstrom wrote:
> Hi, Emil,
>
> On 09/05/2018 03:53 PM, Emil Velikov wrote:
>> On 5 September 2018 at 14:20, Thomas Hellstrom 
>> <thellstrom@vmware.com> wrote:
>>
>>>> In that case, please give me 24h to do a libdrm release before 
>>>> pushing.
>>>> I had to push some workarounds for the sandboxing mentioned earlier 
>>>> :-\
>>>>
>>>> Thanks
>>>> Emil
>>>
>>> Ouch, I just pushed the patch, but feel free to cut the release just 
>>> before
>>> that commit.
>>>
>> That doesn't quite work. Barring any objections I'll: revert, 
>> release, reapply.
>>
>> -Emil
>> _______________________________________________
>> dri-devel mailing list
>> dri-devel@lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
>
> What happened here? I can't really see my commit nor a revert nor a 
> release in libdrm.
>
> /Thomas
>
>
Emil Velikov Oct. 4, 2018, 2:12 p.m. UTC | #23
On Sun, 30 Sep 2018 at 18:31, Thomas Hellstrom <thomas@shipmail.org> wrote:
>
> Hi, Emil,
>
> On 09/05/2018 03:53 PM, Emil Velikov wrote:
> > On 5 September 2018 at 14:20, Thomas Hellstrom <thellstrom@vmware.com> wrote:
> >
> >>> In that case, please give me 24h to do a libdrm release before pushing.
> >>> I had to push some workarounds for the sandboxing mentioned earlier :-\
> >>>
> >>> Thanks
> >>> Emil
> >>
> >> Ouch, I just pushed the patch, but feel free to cut the release just before
> >> that commit.
> >>
> > That doesn't quite work. Barring any objections I'll: revert, release, reapply.
> >
> > -Emil
> > _______________________________________________
> > dri-devel mailing list
> > dri-devel@lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/dri-devel
>
> What happened here? I can't really see my commit nor a revert nor a
> release in libdrm.
>
Coming back from holidays+XDC. I' m doing a release in a moment and
will pick your patch just after that.

Hmm you said you pushed the patch, yet it's not in master ... Not sure
what happened there.
Either way - it'll be there shortly.

Thanks
Emil
Thomas Hellstrom Oct. 4, 2018, 4:43 p.m. UTC | #24
Hi, Emil,

On 10/04/2018 04:12 PM, Emil Velikov wrote:
> On Sun, 30 Sep 2018 at 18:31, Thomas Hellstrom <thomas@shipmail.org> wrote:
>> Hi, Emil,
>>
>> On 09/05/2018 03:53 PM, Emil Velikov wrote:
>>> On 5 September 2018 at 14:20, Thomas Hellstrom <thellstrom@vmware.com> wrote:
>>>
>>>>> In that case, please give me 24h to do a libdrm release before pushing.
>>>>> I had to push some workarounds for the sandboxing mentioned earlier :-\
>>>>>
>>>>> Thanks
>>>>> Emil
>>>> Ouch, I just pushed the patch, but feel free to cut the release just before
>>>> that commit.
>>>>
>>> That doesn't quite work. Barring any objections I'll: revert, release, reapply.
>>>
>>> -Emil
>>> _______________________________________________
>>> dri-devel mailing list
>>> dri-devel@lists.freedesktop.org
>>> https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.freedesktop.org%2Fmailman%2Flistinfo%2Fdri-devel&amp;data=02%7C01%7Cthellstrom%40vmware.com%7C01a5434f4ae94d41e02108d62a042d8b%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C1%7C0%7C636742594763251508&amp;sdata=f4pVmMl%2B2GfI8HIR4GriTD1Ed2eHyEdAttMbcFoavp0%3D&amp;reserved=0
>> What happened here? I can't really see my commit nor a revert nor a
>> release in libdrm.
>>
> Coming back from holidays+XDC. I' m doing a release in a moment and
> will pick your patch just after that.
>
> Hmm you said you pushed the patch, yet it's not in master ... Not sure
> what happened there.
> Either way - it'll be there shortly.

Yes, that's strange. I'm offsite too so I can't check the system from 
which I pushed it. But anyway, I'll push it later today if you haven't 
already.

Thanks,
Thomas


>
> Thanks
> Emil

Patch
diff mbox series

diff --git a/xf86drm.c b/xf86drm.c
index 7807dce9..4cfc5d3e 100644
--- a/xf86drm.c
+++ b/xf86drm.c
@@ -2761,6 +2761,21 @@  char *drmGetDeviceNameFromFd(int fd)
     return strdup(name);
 }
 
+static bool
+drmNodeIsDRM(int maj, int min)
+{
+#ifdef __linux__
+  char path[64];
+  struct stat sbuf;
+
+  snprintf(path, sizeof(path), "/sys/dev/char/%d:%d/device/drm",
+	   maj, min);
+  return stat(path, &sbuf) == 0;
+#else
+  return maj == DRM_MAJOR;
+#endif
+}
+
 int drmGetNodeTypeFromFd(int fd)
 {
     struct stat sbuf;
@@ -2772,7 +2787,7 @@  int drmGetNodeTypeFromFd(int fd)
     maj = major(sbuf.st_rdev);
     min = minor(sbuf.st_rdev);
 
-    if (maj != DRM_MAJOR || !S_ISCHR(sbuf.st_mode)) {
+    if (!drmNodeIsDRM(maj, min) || !S_ISCHR(sbuf.st_mode)) {
         errno = EINVAL;
         return -1;
     }
@@ -2837,7 +2852,7 @@  static char *drmGetMinorNameForFD(int fd, int type)
     maj = major(sbuf.st_rdev);
     min = minor(sbuf.st_rdev);
 
-    if (maj != DRM_MAJOR || !S_ISCHR(sbuf.st_mode))
+    if (!drmNodeIsDRM(maj, min) || !S_ISCHR(sbuf.st_mode))
         return NULL;
 
     snprintf(buf, sizeof(buf), "/sys/dev/char/%d:%d/device/drm", maj, min);
@@ -2871,7 +2886,7 @@  static char *drmGetMinorNameForFD(int fd, int type)
     maj = major(sbuf.st_rdev);
     min = minor(sbuf.st_rdev);
 
-    if (maj != DRM_MAJOR || !S_ISCHR(sbuf.st_mode))
+    if (!drmNodeIsDRM(maj, min) || !S_ISCHR(sbuf.st_mode))
         return NULL;
 
     switch (type) {
@@ -3731,7 +3746,7 @@  process_device(drmDevicePtr *device, const char *d_name,
     maj = major(sbuf.st_rdev);
     min = minor(sbuf.st_rdev);
 
-    if (maj != DRM_MAJOR || !S_ISCHR(sbuf.st_mode))
+    if (!drmNodeIsDRM(maj, min) || !S_ISCHR(sbuf.st_mode))
         return -1;
 
     subsystem_type = drmParseSubsystemType(maj, min);
@@ -3845,7 +3860,7 @@  int drmGetDevice2(int fd, uint32_t flags, drmDevicePtr *device)
     maj = major(sbuf.st_rdev);
     min = minor(sbuf.st_rdev);
 
-    if (maj != DRM_MAJOR || !S_ISCHR(sbuf.st_mode))
+    if (!drmNodeIsDRM(maj, min) || !S_ISCHR(sbuf.st_mode))
         return -EINVAL;
 
     node_type = drmGetMinorType(min);
@@ -3911,7 +3926,7 @@  int drmGetDevice2(int fd, uint32_t flags, drmDevicePtr *device)
     maj = major(sbuf.st_rdev);
     min = minor(sbuf.st_rdev);
 
-    if (maj != DRM_MAJOR || !S_ISCHR(sbuf.st_mode))
+    if (!drmNodeIsDRM(maj, min) || !S_ISCHR(sbuf.st_mode))
         return -EINVAL;
 
     subsystem_type = drmParseSubsystemType(maj, min);
@@ -4071,7 +4086,7 @@  char *drmGetDeviceNameFromFd2(int fd)
     maj = major(sbuf.st_rdev);
     min = minor(sbuf.st_rdev);
 
-    if (maj != DRM_MAJOR || !S_ISCHR(sbuf.st_mode))
+    if (!drmNodeIsDRM(maj, min) || !S_ISCHR(sbuf.st_mode))
         return NULL;
 
     snprintf(path, sizeof(path), "/sys/dev/char/%d:%d", maj, min);
@@ -4097,7 +4112,7 @@  char *drmGetDeviceNameFromFd2(int fd)
     maj = major(sbuf.st_rdev);
     min = minor(sbuf.st_rdev);
 
-    if (maj != DRM_MAJOR || !S_ISCHR(sbuf.st_mode))
+    if (!drmNodeIsDRM(maj, min) || !S_ISCHR(sbuf.st_mode))
         return NULL;
 
     node_type = drmGetMinorType(min);