[PATCH v8 01/29] drm/msm/dp: skip sink requests while the mainlink clocks are off

[Yongxing Mou <yongxing.mou@oss.qualcomm.com>]

msm_dp_ctrl_handle_sink_request() programs REG_DP_STATE_CTRL from all
three of its branches:

  DP_TEST_LINK_PHY_TEST_PATTERN -> msm_dp_ctrl_process_phy_test_request()
                                   -> msm_dp_ctrl_off_link()
  DP_LINK_STATUS_UPDATED        -> msm_dp_ctrl_link_maintenance()
                                   -> msm_dp_ctrl_push_idle()
  DP_TEST_LINK_TRAINING         -> msm_dp_ctrl_link_maintenance()
                                   -> msm_dp_ctrl_push_idle()

That register lives in the mainlink clock domain, and nothing on the
IRQ_HPD path guarantees those clocks are running. Disabling the
boot-time mode for the first modeset from userspace turns the link
clocks off. The sink answers the link going down with an IRQ_HPD
carrying DP_LINK_STATUS_UPDATED (DPCD 0x204 = 0x80), and handling it
walks straight into a write to an unclocked register block:

  [    4.572532] [drm:msm_dp_ctrl_off_link] disable
  [    4.573756] [drm:msm_dp_ctrl_link_clk_disable] disabled link clocks
  [    4.823194] [drm:msm_dp_link_process_request] sink request=0x80
  [    4.823199] [drm:msm_dp_bridge_hpd_notify] 128

The board hung on 3 of 3 boots with a 5120x2160 SST sink attached over
USB-C. On the hung boot captured with drm.debug=0x100 the printk ring
ends at that last line, the drm_dbg_dp() immediately before
msm_dp_ctrl_handle_sink_request(). In the RAM dumps CPU0 is still
running pmic_glink_altmode_worker, holding the bridge chain mutex taken
by drm_bridge_connector_hpd_notify(), and ctrl->link_clks_on is false.

Bail out when the mainlink clocks are off. All three branches need them
running to do anything useful, so gating the function entry costs
nothing: the first one resets the controller, which its own comment
notes "will need DP link related clocks to be running", and the other
two retrain the link.

Everything else reachable from the IRQ_HPD path - the MST ESI handling,
msm_dp_display_process_hpd_high() and the video pattern test response -
only talks to the sink over AUX and keeps working as before.

Reported-by: Joonhoe Kim <26rote@gmail.com>
Closes: https://lore.kernel.org/all/20261108011501.4466-1-26rote@gmail.com/
Fixes: c943b4948b58 ("drm/msm/dp: add displayPort driver support")
Cc: stable@vger.kernel.org
Signed-off-by: Yongxing Mou <yongxing.mou@oss.qualcomm.com>
---
 drivers/gpu/drm/msm/dp/dp_ctrl.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/drivers/gpu/drm/msm/dp/dp_ctrl.c b/drivers/gpu/drm/msm/dp/dp_ctrl.c
index abd3dcdad03c..32d4d5743eb3 100644
--- a/drivers/gpu/drm/msm/dp/dp_ctrl.c
+++ b/drivers/gpu/drm/msm/dp/dp_ctrl.c
@@ -2321,6 +2321,12 @@ void msm_dp_ctrl_handle_sink_request(struct msm_dp_ctrl *msm_dp_ctrl,
 	}
 
 	ctrl = container_of(msm_dp_ctrl, struct msm_dp_ctrl_private, msm_dp_ctrl);
+
+	if (!ctrl->link_clks_on) {
+		drm_dbg_dp(ctrl->drm_dev, "link clocks off, skipping sink request\n");
+		return;
+	}
+
 	sink_request = ctrl->link->sink_request;
 
 	if (sink_request & DP_TEST_LINK_PHY_TEST_PATTERN) {

-- 
2.43.0