[PATCH xserver v2 22/22] xwayland: Guard against very late vblanks

Michel Dänzer michel at daenzer.net
Wed Mar 7 10:37:41 UTC 2018


On 2018-02-28 07:00 PM, Roman Gilg wrote:
> On Wed, Feb 28, 2018 at 6:43 PM, Michel Dänzer <michel at daenzer.net> wrote:
>> I'm unable to reproduce the issue you described with Steam (without this
>> patch series applied). I'd really like to know where the bogus target
>> MSC values you're seeing are coming from. What are the values of the
>> window_msc parameter and window_priv->msc_offset in
>> present_window_to_crtc_msc?
> 
> You don't have to go that far. I did read out on current master high
> target msc values directly for stuff->target_msc in
> proc_present_pixmap. But this happened only on like two or three
> occasions directly at the startup of the Steam client while hundreds
> of other present_pixmap requests before and after had normal
> stuff->target_msc values.
> 
> So just put a line
> 
> ErrorF("XXX %lu\n", stuff->target_msc);
> 
> in proc_present_pixmap and if you look carefully through the debug
> output there should be one or two very high numbers in comparison to
> all the other values.

As discussed on IRC, I've been unable to reproduce the issue you're
seeing yet, despite trying various things. At this point, the most
likely trigger of your symptoms is that in Mesa's
dri3_handle_present_event, the

         if (draw->recv_sbc > draw->send_sbc)
            draw->recv_sbc -= 0x100000000;

is incorrectly hit, e.g. due to processing events from presentations of
another client/context. We'll need more information from you to validate
that hypothesis.


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


More information about the xorg-devel mailing list