[PATCH 3/5] DRM: Armada: Add support for ARGB 32x64 or 64x32 hardware cursors

[Alex Deucher]

On Thu, Oct 17, 2013 at 7:58 PM, Rob Clark <robdclark at gmail.com> wrote:
> On Mon, Oct 7, 2013 at 8:50 AM, Russell King - ARM Linux
> <linux at arm.linux.org.uk> wrote:
>> On Mon, Oct 07, 2013 at 03:29:15PM +0300, Siarhei Siamashka wrote:
>>> On Mon, 7 Oct 2013 11:32:50 +0100
>>> Russell King - ARM Linux <linux at arm.linux.org.uk> wrote:
>>> > However, what you're suggesting is dangerous.  What do you do when you're
>>> > presented with a cursor which is 64x64 ?  Do you:
>>> >
>>> > (a) not display it
>>> > (b) crash the X server
>>> >
>>> > There isn't a fallback to using software cursors available in the X server
>>> > because there's no error reporting for a failed hardware cursor update.
>>>
>>> Hmm, maybe I'm missing something, but doesn't returning FALSE from
>>> UseHWCursorARGB just make the X server fallback to using a software
>>> cursor?
>>>
>>> https://github.com/ssvb/xf86-video-fbturbo/blob/689c08256555/src/sunxi_disp_hwcursor.c#L72
>>>
>>> I'm just doing this. And also have hooks to notify the DRI2 code about
>>> the status of the cursor. In the case if the hardware cursor is not
>>> enabled, hardware overlays can't be safely used with the software
>>> cursor and we need to fallback to CPU/GPU copy instead of zero-copy
>>> buffers swapping.
>>>
>>> And I fully agree that it is the responsibility of the kernel to expose
>>> as much of the hardware capabilities as possible (without compromising
>>> reliability and/or security).
>>
>> If you wish to go in below the level presented by
>> hw/xfree86/modes/xf86Cursors.c (as you are doing), then yes you can do
>> this, and it's something which maybe should be looked into.  You have a
>> good reason to do this (due to your hardware cursor being incompatible
>> with overlays).
>>
>> We have no such restriction - the only issue here is the "odd" size of
>> cursors.  Practically, choosing one or other of 64x32 or 32x64 works
>> fine for the cases I've seen.  I haven't seen any cursors which are
>> 64x64 yet.  I'm not saying they don't exist though!  I can imagine that
>> some accessibility options might want to display a large cursor.
>>
>> Now, the way I handle this is that the kernel allows _either_ a
>> (1 to 32)x(1 to 64) or (1 to 64)x(1 to 32) cursor.  Both width and
>> height larger than 32 gets rejected with -EINVAL:
>>
>> +               /* maximum size is 64x32 or 32x64 */
>> +               if (w > 64 || h > 64 || (w > 32 && h > 32))
>> +                       return -ENOMEM;
>>
>> which is quite reasonable for this hardware.  However, there is no way
>> to export this from DRM - adding maximum cursor size properties wouldn't
>> really represent this.
>
> Until relatively recently, it had always been a device specific ddx
> which would somehow know what w/h values to pass to
> xf86_cursors_init().  I think for xf86-video-modesetting and wayland
> compositors, we probably need to come up with something better.
> Although off the top of my head I can't think of a good way to express
> 64x32 or 32x64.. that is a weird restriction.
>

We have a similar problem on our newer asics.  They require a 128x128
cursors so when you use xf86-video-modesetting, you end up with messed
up cursors because it assumes 64x64.  We could add cursor size to drm
caps.

Alex

> Anyways, right now I don't think there is anything different that I'd
> do in the kernel part.  I *suppose* you could try and figure out that
> all the alpha values are either 0x00 or 0xff and support 64x64 in that
> special case.  I'm not sure that is actually better.. it at least
> makes the failure vs success cases more confusing.  And unless you are
> rockin' the twm+xterm+xcalc setup, I doubt you will see many cursors
> with only 0x00 or 0xff alpha.
>
> So,
>
> Signed-off-by: Rob Clark <robdclark at gmail.com>
>
>
>> The only question is whether DRM should export some capabilities to
>> indicate what's possible with the cursor, so that X knows about this
>> quirk of this hardware.  Overall, I suspect that it's just rare and
>> not worth the effort.  I think it's just easier to pick one of these
>> and stick with it, and if we happen to encounter a cursor larger than
>> our chosen dimentions, XFree86 will already automatically fall back
>> to s/w cursor.
> _______________________________________________
> dri-devel mailing list
> dri-devel at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/dri-devel