| Message ID | 1309558978-4704-7-git-send-email-jbarnes@virtuousgeek.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Fri, 1 Jul 2011 15:22:56 -0700, Jesse Barnes <jbarnes@virtuousgeek.org> wrote: > If the receiver goes away, drop any associated CRTC. This will force a > full mode set on any subsequent setcrtc call, which is what we need if > the receiver is gone and the link is down. This doesn't look like a good solution to me -- we're smashing the connection between the DP output and the selected CRTC. If you plug the cable back in, won't this mess things up when we try to retrain again?
On Fri, 01 Jul 2011 16:48:27 -0700 Keith Packard <keithp@keithp.com> wrote: > On Fri, 1 Jul 2011 15:22:56 -0700, Jesse Barnes <jbarnes@virtuousgeek.org> wrote: > > > If the receiver goes away, drop any associated CRTC. This will force a > > full mode set on any subsequent setcrtc call, which is what we need if > > the receiver is gone and the link is down. > > This doesn't look like a good solution to me -- we're smashing the > connection between the DP output and the selected CRTC. If you plug the > cable back in, won't this mess things up when we try to retrain again? That depends on what behavior we want. With the previous fixes, if you unplug, we'll get a hotplug event, fail to detect a link, and tear down the receiver. With the old code you'd get bad behavior unless you had hit a DPMS path earlier. A subsequent hotplug will be ignored as far as link training goes, since we don't have a receiver configured. In both cases, we'll emit hotplug events to userspace, which is free to react (or not) however it wants. So I think we'd need to leave the receiver_configured bit set even if the hot plug re-train failed, and just try it again on the next hot plug (assuming we want to preserve user configs across hot plug events and not just let userspace handle the hotplug events).
On Fri, 1 Jul 2011 16:59:48 -0700, Jesse Barnes <jbarnes@virtuousgeek.org> wrote: > That depends on what behavior we want. With the previous fixes, if you > unplug, we'll get a hotplug event, fail to detect a link, and tear down > the receiver. With the old code you'd get bad behavior unless you had > hit a DPMS path earlier. Yeah, the old code was clearly wrong; this looks like something that needs fixing, I just don't know what the right fix is. > A subsequent hotplug will be ignored as far as link training goes, since > we don't have a receiver configured. Do we need separate variables for 'should be configured' and 'actually is configured' then? We need to know if there is a configuration we should be selecting, and then we need to know whether the display is actually using that configuration. > In both cases, we'll emit hotplug events to userspace, which is free to > react (or not) however it wants. Let's assume (for now) that user space doesn't do anything. That seems like a good first step to get right; we should assume that if user space sets some mode then it should work correctly. > So I think we'd need to leave the receiver_configured bit set even if > the hot plug re-train failed, and just try it again on the next hot > plug (assuming we want to preserve user configs across hot plug > events and not just let userspace handle the hotplug events). I don't think we want to lose the user config -- in the absence of any user-driven changes, we want to reset the mode exactly as it was when the monitor is plugged back in.
diff --git a/drivers/gpu/drm/i915/intel_dp.c b/drivers/gpu/drm/i915/intel_dp.c index f41aec3..d7a8d24 100644 --- a/drivers/gpu/drm/i915/intel_dp.c +++ b/drivers/gpu/drm/i915/intel_dp.c @@ -1533,7 +1533,7 @@ intel_dp_check_link_status(struct intel_dp *intel_dp) if (!intel_dp_get_link_status(intel_dp)) { intel_dp_link_down(intel_dp); - return; + goto out_gone; } /* Try to read receiver status if the link appears to be up */ @@ -1542,13 +1542,19 @@ intel_dp_check_link_status(struct intel_dp *intel_dp) sizeof (intel_dp->dpcd)); if (ret != sizeof(intel_dp->dpcd)) { intel_dp_link_down(intel_dp); - return; + goto out_gone; } if (!intel_channel_eq_ok(intel_dp)) { intel_dp_start_link_train(intel_dp); intel_dp_complete_link_train(intel_dp); } + + return; + +out_gone: + DRM_DEBUG_DRIVER("sink gone, clearing crtc for DP\n"); + intel_dp->base.base.crtc = NULL; } static enum drm_connector_status
If the receiver goes away, drop any associated CRTC. This will force a full mode set on any subsequent setcrtc call, which is what we need if the receiver is gone and the link is down. Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org> --- drivers/gpu/drm/i915/intel_dp.c | 10 ++++++++-- 1 files changed, 8 insertions(+), 2 deletions(-)