Re: [PATCH 01/13] drm/fb-helper: Push down modeset lock into FB helpers

From: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
To: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Intel Graphics Development <intel-gfx@lists.freedesktop.org>,
	Thierry Reding <treding@nvidia.com>,
	DRI Development <dri-devel@lists.freedesktop.org>
Subject: Re: [PATCH 01/13] drm/fb-helper: Push down modeset lock into FB helpers
Date: Thu, 29 Jun 2017 13:13:13 +0200	[thread overview]
Message-ID: <104c45a9-bb18-a542-5e9f-0aeb47afe865@linux.intel.com> (raw)
In-Reply-To: <CAKMK7uGrRK9tpWw1F-JxHeNnyHNLvUmnSAfYqJZwxxR92v40GQ@mail.gmail.com>

Op 29-06-17 om 11:44 schreef Daniel Vetter:
> On Thu, Jun 29, 2017 at 11:33 AM, Maarten Lankhorst
> <maarten.lankhorst@linux.intel.com> wrote:
>> Op 29-06-17 om 11:23 schreef Daniel Vetter:
>>> On Thu, Jun 29, 2017 at 11:10 AM, Maarten Lankhorst
>>> <maarten.lankhorst@linux.intel.com> wrote:
>>>> Op 27-06-17 om 16:59 schreef Daniel Vetter:
>>>>> From: Thierry Reding <treding@nvidia.com>
>>>>>
>>>>> Move the modeset locking from drivers into FB helpers.
>>>>>
>>>>> v2: Also handle intel_connector_add_to_fbdev.
>>>>>
>>>>> Tested-by: John Stultz <john.stultz@linaro.org>
>>>>> Signed-off-by: Thierry Reding <treding@nvidia.com> (v1)
>>>>> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
>>>>> ---
>>>>>  drivers/gpu/drm/drm_fb_helper.c        | 40 +++++++++++++++++++++++++++++-----
>>>>>  drivers/gpu/drm/i915/intel_dp_mst.c    |  6 -----
>>>>>  drivers/gpu/drm/radeon/radeon_dp_mst.c |  7 ------
>>>>>  3 files changed, 34 insertions(+), 19 deletions(-)
>>>> I know we suck at DP-MST. But I fear the unregister leaves open a race with mst_port being unset without connection_mutex..
>>>>
>>>> best_encoder() and mode_valid() appear to use it. It's probably harmless and I have no good solution, so maybe just annotate it in the patch? Might also affect i915_hpd_poll_init_work, though I think the race in itself is harmless.
>>> Hm, but this did race even before that already ... Should I just wrap
>>> the mst_port = NULL in the connection_mutex? With a big warning that
>>> we should have proper refcounting for this instead (both connectors
>>> and all the mst things are refcounted already).
>>> -Daniel
>> I think leave open the race, the DP-MST modeset code itself doesn't use the connector->mst_port, just the detect callbacks do. In case of nonblocking modeset mst_port might already fall away so it's not like the race is any worse now.
>>
>> I don't even know how to solve it, except by not unplugging. :)
> Why does grabbing the lock not fix the race? ->detect eventually calls
> drm_dp_get_validated_port_ref, which will bail on a zombie mst port..
Oh ok. But there will always be a race. If DP-MST fails immediately after drm_dp_get_validated_port_ref,
you'd still not be able to do anything about it. :)
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel