[PATCH 2/2] drm: Return -ENOTSUPP when called for KMS cap with a non-KMS driver

[Michel Dänzer]

On 30/11/16 06:07 PM, Daniel Vetter wrote:
> On Wed, Nov 30, 2016 at 05:30:02PM +0900, Michel Dänzer wrote:
>> From: Michel Dänzer <michel.daenzer at amd.com>
>>
>> This is an attempt to make the previous fix a bit more robust going
>> forward.
>>
>> Signed-off-by: Michel Dänzer <michel.daenzer at amd.com>
>> ---
>>  drivers/gpu/drm/drm_ioctl.c | 23 +++++++++++++++++------
>>  1 file changed, 17 insertions(+), 6 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/drm_ioctl.c b/drivers/gpu/drm/drm_ioctl.c
>> index 71c3473..32f484b 100644
>> --- a/drivers/gpu/drm/drm_ioctl.c
>> +++ b/drivers/gpu/drm/drm_ioctl.c
>> @@ -229,6 +229,19 @@ static int drm_getcap(struct drm_device *dev, void *data, struct drm_file *file_
>>  	struct drm_crtc *crtc;
>>  
>>  	req->value = 0;
>> +
>> +	/* Only allow non-KMS caps with non-KMS drivers */
>> +	switch (req->capability) {
>> +	case DRM_CAP_DUMB_BUFFER:
> 
> Dumb buffers are only meant to be used for kms drivers, should be
> disallowed too.
> 
>> +	case DRM_CAP_VBLANK_HIGH_CRTC:
> 
> Might be good to have a comment here that we need to allow this for old
> ums?

This is effectively KMS-only as well, per Alex (thanks!).


>> +	case DRM_CAP_PRIME:
>> +	case DRM_CAP_TIMESTAMP_MONOTONIC:
> 
> This is pretty new, I don't think any of the old ums drivers was ever
> updated to use it. Should probably disallow it too.

DRM_CAP_TIMESTAMP_MONOTONIC is driver independent, I don't see why it
wouldn't work fine with UMS drivers. OTOH, I don't think DRM_CAP_PRIME
can work with UMS.


>> +		break;
>> +	default:
>> +		if (!drm_core_check_feature(dev, DRIVER_MODESET))
>> +			return -ENOTSUPP;
>> +	}
> 
> And one code org bikeshed: I don't like the duplicated switch, could we
> instead split it it into two disjoint sets like this?
> 
> 	switch (req->capability) {
> 	case DRM_CAP_PRIME:
> 		req->value |= dev->driver->prime_fd_to_handle ? DRM_PRIME_CAP_IMPORT : 0;
> 		req->value |= dev->driver->prime_handle_to_fd ? DRM_PRIME_CAP_EXPORT : 0;
> 		break;
> 	... all other non-modeset caps ...
> 	}
> 
> 	if (!drm_core_check_feature(dev, DRIVER_MODESET))
> 		return -ENOTSUPP;
> 
> 	switch (req->capability) {
> 	... handle remaining caps needed for DRIVER_MODSET ...
> 	default:
> 		return -EINVAL;
> 	}
> 
> That way it would be a bit more obvious that people who add a new cap need
> to make a decision where to put it (and by default put it in the bottom
> pile).

Your pseudo-code wouldn't work correctly, but I get your idea. :)


v2 addressing review feedback coming up soon.


-- 
Earthling Michel Dänzer               |               http://www.amd.com
Libre software enthusiast             |             Mesa and X developer