[Intel-gfx] [PATCH] [drm][i915] Increase LSPCON timeout

Thu Aug 16 21:35:26 UTC 2018

tor 2018-08-16 klockan 11:23 -0700 skrev Rodrigo Vivi:
> On Thu, Aug 16, 2018 at 10:36:43AM +0200, Fredrik Schön wrote:
> > Shashank,
> > 
> > Den tors 16 aug. 2018 kl 10:15 skrev Sharma, Shashank
> > <shashank.sharma at intel.com>:
> > > 
> > > Hey Chris,
> > > 
> > > 
> > > On 8/16/2018 1:13 PM, Chris Wilson wrote:
> > > > Quoting Sharma, Shashank (2018-08-16 08:33:36)
> > > > > Regards
> > > > > 
> > > > > Shashank
> > > > > 
> > > > > 
> > > > > On 8/16/2018 12:47 PM, Jani Nikula wrote:
> > > > > > On Wed, 15 Aug 2018, Rodrigo Vivi <rodrigo.vivi at intel.com>
> > > > > > wrote:
> > > > > > > On Wed, Aug 15, 2018 at 03:39:40PM -0700, Rodrigo Vivi
> > > > > > > wrote:
> > > > > > > > On Mon, Aug 13, 2018 at 07:51:33PM +0200, Fredrik Schön
> > > > > > > > wrote:
> > > > > > > > 
> > > > > > > > First of all we need to fix the commit subject:
> > > > > > > > 
> > > > > > > > drm/i915: Increase LSPCON timeout
> > > > > > > > 
> > > > > > > > (this can be done when merging, no need to resend)
> > > > > > > > 
> > > > > > > > > 100 ms is not enough time for the LSPCON adapter on
> > > > > > > > > Intel NUC devices to
> > > > > > > > > settle. This causes dropped display modes at driver
> > > > > > > > > initialisation.
> > > > > > > > > 
> > > > > > > > > Increase timeout to 1000 ms.
> > > > > > > > > 
> > > > > > > > > Fixes: 
> > > > > > > > > https://bugzilla.redhat.com/show_bug.cgi?id=1570392
> > > > > > > > 
> > > > > > > > Missusage of "Fixes:" tag, please read
> > > > > > > > 
> > > > > > > > 
https://01.org/linuxgraphics/gfx-docs/maintainer-tools/drm-intel.html#fixes
> > > > > > > > 
> > > > > > > > But also no need for resending... could be fixed when
> > > > > > > > merging
> > > > > > > > 
> > > > > > > > The right one would be:
> > > > > > > > 
> > > > > > > > Bugzilla: 
> > > > > > > > https://bugzilla.redhat.com/show_bug.cgi?id=1570392
> > > > > > > > Fixes: 357c0ae9198a ("drm/i915/lspcon: Wait for
> > > > > > > > expected LSPCON mode to settle")
> > > > > > > > Cc: Shashank Sharma <shashank.sharma at intel.com>
> > > > > > > > Cc: Imre Deak <imre.deak at intel.com>
> > > > > > > > Cc: Jani Nikula <jani.nikula at intel.com>
> > > > > > > > Cc: <stable at vger.kernel.org> # v4.11+
> > > > > > > > 
> > > > > > > > Since initial 100 seemed to be empirical and this
> > > > > > > > increase seems to
> > > > > > > > help other cases I'm in favor of this move so
> > > > > > > > 
> > > > > > > > Reviewed-by: Rodrigo Vivi <rodrigo.vivi at intel.com>
> > > > > > > > 
> > > > > > > > However I will wait a bit before merging it
> > > > > > > > so Imre, Shashank, and/or Jani can take a look here...
> > > > > > > 
> > > > > > > now, really cc'ing them...
> > > > > > 
> > > > > > Shashank? Does this slow down non-LSPCON paths?
> > > > > 
> > > > > This will slow down the lspcon probing and resume part, but
> > > > > both of them
> > > > > happen only when LSPCON device is found. So to answer your
> > > > > question,
> > > > > this will not slow down the non-lspcon path, but will slow
> > > > > down the
> > > > > LSPCON connector resume and probe time. but I would
> > > > > recommend, instead
> > > > > of increasing it to 1000 ms in a single shot, we might want
> > > > > to gradually
> > > > > pick this up, on a wake-and-check way.
> > > > 
> > > > wait_for() checks every [10us, 1ms] until the condition is met,
> > > > or it
> > > > times out. So, so long as we don't enter this path for !LPSCON
> > > > where we
> > > > know that it will timeout, the wait_for() will only take as
> > > > long as is
> > > > required for the connector to settle.
> > > 
> > > We wont hit !LSPCON timeout case here, as we have already read
> > > the
> > > dongle signature successfully by now.  But I was thinking that,
> > > if the
> > > spec recommends max wait time as 100ms (which is of course
> > > doesn't seem
> > > enough), if we can't detect i2c-over-aux after first 500ms, I
> > > guess we
> > > wont be able to do that in next 500ms too. So is it really ok to
> > > wait
> > > this long in the resume sequence ?
> > > 
> > > I guess Fredrik can provide some inputs here on if there are some
> > > experiments behind this number of 1000ms, or this is just a safe
> > > bet ?
> > > > 
> > 
> > My first patch attempt - which is attached to the Redhat and FDO
> > Bugzilla
> > bugs - added a retry loop around the original 100 ms timeout. The
> > retry loop
> > did trigger, but never more than once in a row in my testing.
> > 
> > So possibly 200 ms would be a sufficient timeout, but as the
> > wait_for() loop
> > terminates early on success I suggested a conservative value of
> > 1000 ms.
> 
> Since Shashank mentioned 100us came from some spec, maybe the double
> is already
> a conservative value.
> 
> Since there is the concerns of delaying something when LSPCON fails
> and we are possibly looping on connectors somewhere/somehow I believe
> we need
> to have a balanced approach here.
> 
> could you please try the 200 ms approach on your case there for a
> while and
> see how it goes?
> 

I ran a few stress tests using Nicholas test case from [1]. I can
quickly reproduce the failure with timeouts 100 ms, 110 ms, 130 ms, 150
ms and 170 ms. I am unable to reproduce any failures with timeouts 190
ms (n=18) and 200 ms (n=20+16).

So while 200 ms appears to work on my hardware with reasonable
confidence, I wouldn't call 200 conservative. But then again, I do not
know the specifications. I'm just being empirical.

[1] https://bugs.freedesktop.org/show_bug.cgi?id=107503#c15

> > 
> > > > Can we do other connectors at the same time, or does probing
> > > > LSPCON
> > > > block the system?
> > > 
> > > We can do other connectors at the same time in DRM layer at-
> > > least,
> > > LSPCON blocks only this connector. I was curious if are we doing
> > > this
> > > during the resume scenario or is this in the sequential
> > > get_connector()
> > > type of call  ?
> > > - Shashank
> > > > -Chris
> > 
> > /F
> > _______________________________________________
> > Intel-gfx mailing list
> > Intel-gfx at lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/intel-gfx