[Qemu-devel] [RfC PATCH] hid: handle full ptr queues in post_load

[Qemu-devel] [RfC PATCH] hid: handle full ptr queues in post_load

[Gerd Hoffmann]

Cc: Gonglei <arei.gonglei@huawei.com>
Cc: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
---
 hw/input/hid.c | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)

diff --git a/hw/input/hid.c b/hw/input/hid.c
index 8f6fbb3..6fb963f 100644
--- a/hw/input/hid.c
+++ b/hw/input/hid.c
@@ -519,6 +519,27 @@ static int hid_post_load(void *opaque, int version_id)
     HIDState *s = opaque;
 
     hid_set_next_idle(s);
+
+    if (s->n == QUEUE_LENGTH && (s->kind == HID_TABLET ||
+                                 s->kind == HID_MOUSE)) {
+        /*
+         * Handle ptr device migration from old qemu with full queue.
+         *
+         * Throw away everything but the last event, so we propagate
+         * at least the current button state to the guest.  Also keep
+         * current position for the tablet, signal "no motion" for the
+         * mouse.
+         */
+        HIDPointerEvent evt;
+        evt = s->ptr.queue[(s->head+s->n) & QUEUE_MASK];
+        if (s->kind == HID_MOUSE) {
+            evt.xdx = 0;
+            evt.ydy = 0;
+        }
+        s->ptr.queue[0] = evt;
+        s->head = 0;
+        s->n = 1;
+    }
     return 0;
 }
 
-- 
1.8.3.1

Re: [Qemu-devel] [RfC PATCH] hid: handle full ptr queues in post_load

[Gonglei]

On 2014/11/27 17:04, Gerd Hoffmann wrote:

> Cc: Gonglei <arei.gonglei@huawei.com>
> Cc: Dr. David Alan Gilbert <dgilbert@redhat.com>
> Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
> ---
>  hw/input/hid.c | 21 +++++++++++++++++++++
>  1 file changed, 21 insertions(+)
> 
> diff --git a/hw/input/hid.c b/hw/input/hid.c
> index 8f6fbb3..6fb963f 100644
> --- a/hw/input/hid.c
> +++ b/hw/input/hid.c
> @@ -519,6 +519,27 @@ static int hid_post_load(void *opaque, int version_id)
>      HIDState *s = opaque;
>  
>      hid_set_next_idle(s);
> +
> +    if (s->n == QUEUE_LENGTH && (s->kind == HID_TABLET ||
> +                                 s->kind == HID_MOUSE)) {
> +        /*
> +         * Handle ptr device migration from old qemu with full queue.
> +         *
> +         * Throw away everything but the last event, so we propagate
> +         * at least the current button state to the guest.  Also keep
> +         * current position for the tablet, signal "no motion" for the
> +         * mouse.
> +         */
> +        HIDPointerEvent evt;
> +        evt = s->ptr.queue[(s->head+s->n) & QUEUE_MASK];

s->n is QUEUE_LENGTH, can we directly delete it?
  evt = s->ptr.queue[s->head & QUEUE_MASK]

Best regards,
-Gonglei

> +        if (s->kind == HID_MOUSE) {
> +            evt.xdx = 0;
> +            evt.ydy = 0;
> +        }
> +        s->ptr.queue[0] = evt;
> +        s->head = 0;
> +        s->n = 1;
> +    }
>      return 0;
>  }
>  

Re: [Qemu-devel] [RfC PATCH] hid: handle full ptr queues in post_load

[Gerd Hoffmann]

  Hi,

> > +        evt = s->ptr.queue[(s->head+s->n) & QUEUE_MASK];
> 
> s->n is QUEUE_LENGTH, can we directly delete it?
>   evt = s->ptr.queue[s->head & QUEUE_MASK]

I prefer to make clear in the code that we want the last ring element
not the first and leave in the "+n", even if we could take it out.

cheers,
  Gerd

Re: [Qemu-devel] [RfC PATCH] hid: handle full ptr queues in post_load

[Gonglei]

On 2014/11/27 19:04, Gerd Hoffmann wrote:

>   Hi,
> 
>>> +        evt = s->ptr.queue[(s->head+s->n) & QUEUE_MASK];
>>
>> s->n is QUEUE_LENGTH, can we directly delete it?
>>   evt = s->ptr.queue[s->head & QUEUE_MASK]
> 
> I prefer to make clear in the code that we want the last ring element
> not the first and leave in the "+n", even if we could take it out.
> 
OK.

Another question, whether or not we handle this scenario
before calling hid_set_next_idle(s) ? Maybe it is safer, because
hid_idle_timer will call  hid_pointer_event(), callback function
of hs->event(hs).

Thanks,
-Gonglei

Re: [Qemu-devel] [RfC PATCH] hid: handle full ptr queues in post_load

[Gerd Hoffmann]

On Do, 2014-11-27 at 19:16 +0800, Gonglei wrote:
> On 2014/11/27 19:04, Gerd Hoffmann wrote:
> 
> >   Hi,
> > 
> >>> +        evt = s->ptr.queue[(s->head+s->n) & QUEUE_MASK];
> >>
> >> s->n is QUEUE_LENGTH, can we directly delete it?
> >>   evt = s->ptr.queue[s->head & QUEUE_MASK]
> > 
> > I prefer to make clear in the code that we want the last ring element
> > not the first and leave in the "+n", even if we could take it out.
> > 
> OK.
> 
> Another question, whether or not we handle this scenario
> before calling hid_set_next_idle(s) ? Maybe it is safer, because
> hid_idle_timer will call  hid_pointer_event(), callback function
> of hs->event(hs).

Not needed, the timer will not called before vmload is completely
finished.

cheers,
  Gerd

Re: [Qemu-devel] [RfC PATCH] hid: handle full ptr queues in post_load

[Gonglei]

On 2014/11/27 19:19, Gerd Hoffmann wrote:

> On Do, 2014-11-27 at 19:16 +0800, Gonglei wrote:
>> On 2014/11/27 19:04, Gerd Hoffmann wrote:
>>
>>>   Hi,
>>>
>>>>> +        evt = s->ptr.queue[(s->head+s->n) & QUEUE_MASK];
>>>>
>>>> s->n is QUEUE_LENGTH, can we directly delete it?
>>>>   evt = s->ptr.queue[s->head & QUEUE_MASK]
>>>
>>> I prefer to make clear in the code that we want the last ring element
>>> not the first and leave in the "+n", even if we could take it out.
>>>
>> OK.
>>
>> Another question, whether or not we handle this scenario
>> before calling hid_set_next_idle(s) ? Maybe it is safer, because
>> hid_idle_timer will call  hid_pointer_event(), callback function
>> of hs->event(hs).
> 
> Not needed, the timer will not called before vmload is completely
> finished.
> 

Yep, as the Qemu big lock.
I will test this patch later, thanks for your work. :)

Regards,
-Gonglei

Re: [Qemu-devel] [RfC PATCH] hid: handle full ptr queues in post_load

[Gonglei]

On 2014/11/27 19:22, Gonglei wrote:

> On 2014/11/27 19:19, Gerd Hoffmann wrote:
> 
>> On Do, 2014-11-27 at 19:16 +0800, Gonglei wrote:
>>> On 2014/11/27 19:04, Gerd Hoffmann wrote:
>>>
>>>>   Hi,
>>>>
>>>>>> +        evt = s->ptr.queue[(s->head+s->n) & QUEUE_MASK];
>>>>>
>>>>> s->n is QUEUE_LENGTH, can we directly delete it?
>>>>>   evt = s->ptr.queue[s->head & QUEUE_MASK]
>>>>
>>>> I prefer to make clear in the code that we want the last ring element
>>>> not the first and leave in the "+n", even if we could take it out.
>>>>
>>> OK.
>>>
>>> Another question, whether or not we handle this scenario
>>> before calling hid_set_next_idle(s) ? Maybe it is safer, because
>>> hid_idle_timer will call  hid_pointer_event(), callback function
>>> of hs->event(hs).
>>
>> Not needed, the timer will not called before vmload is completely
>> finished.
>>
> 
> Yep, as the Qemu big lock.
> I will test this patch later, thanks for your work. :)
> 
Patch works for me.  Thanks.

Tested-by: Gonglei <arei.gonglei@huawei.com>
Reviewed-by: Gonglei <arei.gonglei@huawei.com>

Regards,
-Gonglei