drm/i915: Fix assert_plane() warning on bootup with external display
authorAzhar Shaikh <azhar.shaikh@intel.com>
Fri, 6 Jul 2018 18:37:30 +0000 (11:37 -0700)
committerVille Syrjälä <ville.syrjala@linux.intel.com>
Thu, 19 Jul 2018 20:20:44 +0000 (23:20 +0300)
On KBL, WHL RVPs, booting up with an external display connected, triggers
below warning, when the BiOS brings up the external display too.
This warning is not seen during hotplug.

[    3.615226] ------------[ cut here ]------------
[    3.619829] plane 1A assertion failure (expected on, current off)
[    3.632039] WARNING: CPU: 2 PID: 354 at drivers/gpu/drm/i915/intel_display.c:1294 assert_plane+0x71/0xbb
[    3.633920] iwlwifi 0000:00:14.3: loaded firmware version 38.c0e03d94.0 op_mode iwlmvm
[    3.647157] Modules linked in: iwlwifi cfg80211 btusb btrtl btbcm btintel bluetooth ecdh_generic
[    3.647163] CPU: 2 PID: 354 Comm: frecon Not tainted 4.17.0-rc7-50176-g655af12d39c2 #3
[    3.647165] Hardware name: Intel Corporation CoffeeLake Client Platform/WhiskeyLake U DDR4 ERB, BIOS CNLSFWR1.R00.X140.B00.1804040304 04/04/2018
[    3.684509] RIP: 0010:assert_plane+0x71/0xbb
[    3.764451] Call Trace:
[    3.766888]  intel_atomic_commit_tail+0xa97/0xb77
[    3.771569]  intel_atomic_commit+0x26a/0x279
[    3.771572]  drm_atomic_helper_set_config+0x5c/0x76
[    3.780670]  __drm_mode_set_config_internal+0x66/0x109
[    3.780672]  drm_mode_setcrtc+0x4c9/0x5cc
[    3.780674]  ? drm_mode_getcrtc+0x162/0x162
[    3.789774]  ? drm_mode_getcrtc+0x162/0x162
[    3.798108]  drm_ioctl_kernel+0x8d/0xe4
[    3.801926]  drm_ioctl+0x27d/0x368
[    3.805311]  ? drm_mode_getcrtc+0x162/0x162
[    3.805314]  ? selinux_file_ioctl+0x14e/0x199
[    3.805317]  vfs_ioctl+0x21/0x2f
[    3.813812]  do_vfs_ioctl+0x491/0x4b4
[    3.813813]  ? security_file_ioctl+0x37/0x4b
[    3.813816]  ksys_ioctl+0x55/0x75
[    3.820672]  __x64_sys_ioctl+0x1a/0x1e
[    3.820674]  do_syscall_64+0x51/0x5f
[    3.820678]  entry_SYSCALL_64_after_hwframe+0x44/0xa9
[    3.828221] RIP: 0033:0x7b5e04953967
[    3.835504] RSP: 002b:00007fff2eafb6f8 EFLAGS: 00000246 ORIG_RAX: 0000000000000010
[    3.835505] RAX: ffffffffffffffda RBX: 0000000000000002 RCX: 00007b5e04953967
[    3.835505] RDX: 00007fff2eafb730 RSI: 00000000c06864a2 RDI: 000000000000000f
[    3.835506] RBP: 00007fff2eafb720 R08: 0000000000000000 R09: 0000000000000000
[    3.835507] R10: 0000000000000070 R11: 0000000000000246 R12: 000000000000000f
[    3.879988] R13: 000056bc9dd7d210 R14: 00007fff2eafb730 R15: 00000000c06864a2
[    3.887081] Code: 48 c7 c7 06 71 a5 be 84 c0 48 c7 c2 06 fd a3 be 48 89 f9 48 0f 44 ca 84 db 48 0f 45 d7 48 c7 c7 df d3 a4 be 31 c0 e8 af a0 c0 ff <0f> 0b eb 2b 48 c7 c7 06 fd a3 be 84 c0 48 c7 c2 06 71 a5 be 48
[    3.905845] WARNING: CPU: 2 PID: 354 at drivers/gpu/drm/i915/intel_display.c:1294 assert_plane+0x71/0xbb
[    3.920964] ---[ end trace dac692f4ac46391a ]---

The warning is seen when mode_setcrtc() is called for pipeB
during bootup and before we get a mode_setcrtc() for pipeA,
while doing update_crtcs() in intel_atomic_commit_tail().
Now since, plane1A is still active after commit, update_crtcs()
is done for pipeA and eventually update_plane() for plane1A.

intel_plane_state->ctl for plane1A is not updated since set_modecrtc() is
called for pipeB. So intel_plane_state->ctl for plane 1A will be 0x0.
So doing an update_plane() for plane1A, will result in clearing
PLANE_CTL_ENABLE bit, and hence the warning.

To fix this warning, force all active planes to recompute their states
in probe.

Changes in v8:
- Actually add Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>

Changes in v7:
- Move call to intel_initial_commit() after sanitize_watermarks()
  Otherwise the plane update will still consult potentially bogus
  watermarks we read out from the hardware. (Ville)
- Carry Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
  from v6

Changes in v6:
- Handle EDEADLK for drm_atomic_get_crtc_state() and
  drm_atomic_add_affected_planes()
- Remove optimization of calling intel_initial_commit()
  only when there is more than one active pipe in probe.
- Avoid using intel_ types.

Changes in v5:
- Drop drm_modeset_lock_all_ctx() since locks will be taken later.

Changes in v4:
- Handle locking in intel_initial_commit()
- Move the for loop inside intel_initial_commit() so that
  drm_atomic_commit() is called only once
- Call intel_initial_commit() only for more than one active crtc on boot.
- Save the return value of intel_initial_commit() and print a message in
  case of an error

Changes in v3:
- Add comments

Changes in v2:
- Force all planes to recompute their states.(Ville Syrjälä)
- Update the commit message

Signed-off-by: Azhar Shaikh <azhar.shaikh@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/1530902250-44583-1-git-send-email-azhar.shaikh@intel.com
drivers/gpu/drm/i915/intel_display.c

index 8bd9080fce3463183af7ccf7ed8afacb59c1cef2..8719c1a9d1ceb6a24ba0137937bc11ce433b9eb7 100644 (file)
@@ -15065,12 +15065,61 @@ static void intel_update_fdi_pll_freq(struct drm_i915_private *dev_priv)
        DRM_DEBUG_DRIVER("FDI PLL freq=%d\n", dev_priv->fdi_pll_freq);
 }
 
+static int intel_initial_commit(struct drm_device *dev)
+{
+       struct drm_atomic_state *state = NULL;
+       struct drm_modeset_acquire_ctx ctx;
+       struct drm_crtc *crtc;
+       struct drm_crtc_state *crtc_state;
+       int ret = 0;
+
+       state = drm_atomic_state_alloc(dev);
+       if (!state)
+               return -ENOMEM;
+
+       drm_modeset_acquire_init(&ctx, 0);
+
+retry:
+       state->acquire_ctx = &ctx;
+
+       drm_for_each_crtc(crtc, dev) {
+               crtc_state = drm_atomic_get_crtc_state(state, crtc);
+               if (IS_ERR(crtc_state)) {
+                       ret = PTR_ERR(crtc_state);
+                       goto out;
+               }
+
+               if (crtc_state->active) {
+                       ret = drm_atomic_add_affected_planes(state, crtc);
+                       if (ret)
+                               goto out;
+               }
+       }
+
+       ret = drm_atomic_commit(state);
+
+out:
+       if (ret == -EDEADLK) {
+               drm_atomic_state_clear(state);
+               drm_modeset_backoff(&ctx);
+               goto retry;
+       }
+
+       drm_atomic_state_put(state);
+
+       drm_modeset_drop_locks(&ctx);
+       drm_modeset_acquire_fini(&ctx);
+
+       return ret;
+}
+
 int intel_modeset_init(struct drm_device *dev)
 {
        struct drm_i915_private *dev_priv = to_i915(dev);
        struct i915_ggtt *ggtt = &dev_priv->ggtt;
        enum pipe pipe;
        struct intel_crtc *crtc;
+       int ret;
 
        dev_priv->modeset_wq = alloc_ordered_workqueue("i915_modeset", 0);
 
@@ -15145,8 +15194,6 @@ int intel_modeset_init(struct drm_device *dev)
                      INTEL_INFO(dev_priv)->num_pipes > 1 ? "s" : "");
 
        for_each_pipe(dev_priv, pipe) {
-               int ret;
-
                ret = intel_crtc_init(dev_priv, pipe);
                if (ret) {
                        drm_mode_config_cleanup(dev);
@@ -15202,6 +15249,16 @@ int intel_modeset_init(struct drm_device *dev)
        if (!HAS_GMCH_DISPLAY(dev_priv))
                sanitize_watermarks(dev);
 
+       /*
+        * Force all active planes to recompute their states. So that on
+        * mode_setcrtc after probe, all the intel_plane_state variables
+        * are already calculated and there is no assert_plane warnings
+        * during bootup.
+        */
+       ret = intel_initial_commit(dev);
+       if (ret)
+               DRM_DEBUG_KMS("Initial commit in probe failed.\n");
+
        return 0;
 }