Re: [Xen-devel] [PATCH v2] xen, input: try to read screen resolution for xen-kbdfront

[Juergen Gross]

On 17/02/17 20:47, Dmitry Torokhov wrote:
> On Mon, Jan 30, 2017 at 01:27:29PM +0200, Oleksandr Andrushchenko wrote:
>> On 01/30/2017 01:23 PM, Juergen Gross wrote:
>>> On 27/01/17 17:10, Dmitry Torokhov wrote:
>>>> On January 27, 2017 12:31:19 AM PST, Juergen Gross <jgross@xxxxxxxx> wrote:
>>>>> On 27/01/17 09:26, Oleksandr Andrushchenko wrote:
>>>>>> On 01/27/2017 10:14 AM, Juergen Gross wrote:
>>>>>>> On 27/01/17 08:53, Oleksandr Andrushchenko wrote:
>>>>>>>> On 01/27/2017 09:46 AM, Juergen Gross wrote:
>>>>>>>>> On 27/01/17 08:21, Oleksandr Andrushchenko wrote:
>>>>>>>>>> On 01/27/2017 09:12 AM, Juergen Gross wrote:
>>>>>>>>>>> Instead of using the default resolution of 800*600 for the
>>>>> pointing
>>>>>>>>>>> device of xen-kbdfront try to read the resolution of the
>>>>> (virtual)
>>>>>>>>>>> framebuffer device. Use the default as fallback only.
>>>>>>>>>>>
>>>>>>>>>>> Signed-off-by: Juergen Gross <jgross@xxxxxxxx>
>>>>>>>>>>> ---
>>>>>>>>>>> V2: get framebuffer resolution only if CONFIG_FB (Dmitry
>>>>> Torokhov)
>>>>>>>>>>> ---
>>>>>>>>>>>     drivers/input/misc/xen-kbdfront.c | 15 ++++++++++++---
>>>>>>>>>>>     1 file changed, 12 insertions(+), 3 deletions(-)
>>>>>>>>>>>
>>>>>>>>>>> diff --git a/drivers/input/misc/xen-kbdfront.c
>>>>>>>>>>> b/drivers/input/misc/xen-kbdfront.c
>>>>>>>>>>> index 3900875..3aae9b4 100644
>>>>>>>>>>> --- a/drivers/input/misc/xen-kbdfront.c
>>>>>>>>>>> +++ b/drivers/input/misc/xen-kbdfront.c
>>>>>>>>>>> @@ -16,6 +16,7 @@
>>>>>>>>>>>     #include <linux/kernel.h>
>>>>>>>>>>>     #include <linux/errno.h>
>>>>>>>>>>>     #include <linux/module.h>
>>>>>>>>>>> +#include <linux/fb.h>
>>>>>>>>>>>     #include <linux/input.h>
>>>>>>>>>>>     #include <linux/slab.h>
>>>>>>>>>>>     @@ -108,7 +109,7 @@ static irqreturn_t input_handler(int rq,
>>>>> void
>>>>>>>>>>> *dev_id)
>>>>>>>>>>>     static int xenkbd_probe(struct xenbus_device *dev,
>>>>>>>>>>>                       const struct xenbus_device_id *id)
>>>>>>>>>>>     {
>>>>>>>>>>> -    int ret, i;
>>>>>>>>>>> +    int ret, i, width, height;
>>>>>>>>>>>         unsigned int abs;
>>>>>>>>>>>         struct xenkbd_info *info;
>>>>>>>>>>>         struct input_dev *kbd, *ptr;
>>>>>>>>>>> @@ -173,9 +174,17 @@ static int xenkbd_probe(struct
>>>>> xenbus_device
>>>>>>>>>>> *dev,
>>>>>>>>>>>         ptr->id.product = 0xfffe;
>>>>>>>>>>>           if (abs) {
>>>>>>>>>>> +        width = XENFB_WIDTH;
>>>>>>>>>>> +        height = XENFB_HEIGHT;
>>>>>>>>>>> +#ifdef CONFIG_FB
>>>>>>>>>>> +        if (registered_fb[0]) {
>>>>>>>>>> This still will not help if FB gets registered after kbd+ptr
>>>>>>>>> Hmm, so you think I should add a call to fb_register_client() to
>>>>> get
>>>>>>>>> events for new registered framebuffer devices?
>>>>>>>> yes, but also pay attention to CONFIG_FB_NOTIFY: you may still
>>>>>>>> end up w/o notification.
>>>>>>> Okay, that's not worse than today.
>>>>>> agree
>>>>>>>>> This would probably work. I'll have a try.
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Thanks,
>>>>>>>>>
>>>>>>>>> Juergen
>>>>>>>> My bigger concern here is that we try to tie keyboard and pointer
>>>>> device
>>>>>>>> to the framebuffer. IMO, these are independent parts of the system
>>>>> and
>>>>>>>> the relation
>>>>>>>> depends on the use-case. One can have graphics enabled w/o
>>>>> framebuffer
>>>>>>>> at all, e.g.
>>>>>>>> DRM/KMS + OpenGLES + Weston + kbd + ptr...
>>>>>>> Again: that's a use case which will work as today. The current
>>>>> defaults
>>>>>>> are being used.
>>>>>>>
>>>>>>> The question is whether we should add a module parameter switching
>>>>> off
>>>>>>> the automatic adaption of the resolution as there might be use cases
>>>>>>> where we don't want this feature.
>>>>>> I think for those who doesn't want this resolution there is
>>>>>> still a possibility to change it on backend's XenbusStateConnected
>>>>>> So, no need for module parameter, IMO
>>>>> Fine.
>>>>>
>>>>> I'll send V3 soon.
>>>> How about you do the axis adjustment from userspace (udev rule), and leave kernel as is?
>>> Hmm, is this a good idea?
>>>
>>> I'd need a udev rule to trigger when either the pointing device or a
>>> new frame buffer is showing up. In both cases I need to read the
>>> geometry of the frame buffer (in case it exists) and set the geometry
>>> of the pointing device (in case it exists) to the same values. This
>>> seems to be much more complicated than the required changes in the
>>> driver.
>>>
>>> I could be wrong, of course, especially as I'm no expert in writing
>>> udev rules. :-)
>> And you may also end up with thin Dom0 w/o udev at all...
> 
> So what piece of software is using resolution of this input device and
> why it has to match screen resolution? What happens when framebuffer is
> registered after input device is created? I see that in the later
> version of the patch you hook the notifier and change values, but how
> users would know about that?

The problem showed up in out internal QA: the installer is being tested
automatically in a virtual machine. The output of the (virtual)
framebuffer is captured automatically and verified to match certain
criteria. User input (via keyboard and pointing device) are entered
automatically, too.

As the installer has to work for different screen resolutions this is
a major requirement. And the pointing device has to match the resolution
of the screen, of course, to be able to reach all regions of the screen.

> I think tying input device to framebuffer is wrong.

Generally: might be.

For virtual machines: I think this makes sense in more than 99% of the
use cases.

Would you be okay if I add a module parameter for switching this
behavior off on demand?


Juergen