[Intel-gfx] i915 timeouts delaying boot under GVT

[Christoph Hellwig]

Hi all,

when botting a current 4.14-rc tree in a VM using GVT-g (with the host
also running a current 4.14-rc tree), I see bunch of long timeouts
followed by i915 errors:

[    4.252066] i915 0000:00:03.0: [drm] VGT balloon successfully
[    5.095190] i915 0000:00:03.0: [drm] *ERROR* Failed to disable SAGV (-110)
[   15.334559] [drm:drm_atomic_helper_wait_for_flip_done] *ERROR* [CRTC:51:pipe
A] flip_done timed out
[   15.346934] [drm] Initialized i915 1.6.0 20201103 for 0000:00:03.0 on minor
0

I did a hackjob to track them down and just if out the offending code,
which speeds up the boot by ~11 seconds but is probably dangerous as hell:

diff --git a/drivers/gpu/drm/i915/display/intel_display.c b/drivers/gpu/drm/i915/display/intel_display.c
index 2d5d21740c25..ee82fd67f386 100644
--- a/drivers/gpu/drm/i915/display/intel_display.c
+++ b/drivers/gpu/drm/i915/display/intel_display.c
@@ -10696,7 +10696,7 @@ static void intel_atomic_commit_tail(struct intel_atomic_state *state)
 	 * - switch over to the vblank wait helper in the core after that since
 	 *   we don't need out special handling any more.
 	 */
-	drm_atomic_helper_wait_for_flip_done(dev, &state->base);
+//	drm_atomic_helper_wait_for_flip_done(dev, &state->base);
 
 	for_each_new_intel_crtc_in_state(state, crtc, new_crtc_state, i) {
 		if (new_crtc_state->uapi.async_flip)
diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c
index 45fefa0ed160..f03ce729cc4b 100644
--- a/drivers/gpu/drm/i915/intel_pm.c
+++ b/drivers/gpu/drm/i915/intel_pm.c
@@ -3753,7 +3753,7 @@ intel_disable_sagv(struct drm_i915_private *dev_priv)
 	if (!intel_has_sagv(dev_priv))
 		return 0;
 
-	if (dev_priv->sagv_status == I915_SAGV_DISABLED)
+	if (1 || dev_priv->sagv_status == I915_SAGV_DISABLED)
 		return 0;
 
 	drm_dbg_kms(&dev_priv->drm, "Disabling SAGV\n");