Skip to contentSkip to conversation

drm/atomic: move waiting for hw_done to a helper

PATCH In review Maarten Lankhorst 3 participants 8 messages Jun 14, 2017 – Jun 26, 2017

Subsystems (3)
lore
Share & tools
lore

Maarten Lankhorst

author maintainer
#0 lore raw

PATCH v2 1/2 drm/atomic: move waiting for hw_done to a helper

Moving the wait to a helper allows callers outside of the core to wait for pending updates to complete.

This can be used to prevent races against nonblocking modesets. Only the hw_done call in swap_state is moved to a helper, doing it for the other callers requires too many changes and I think this is the only useful one to export.

---
 drivers/gpu/drm/drm_atomic_helper.c | 51 +++++++++++++++++++++++++++----------
 include/drm/drm_atomic_helper.h     |  1 +
 2 files changed, 38 insertions(+), 14 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +37 -14
index 86d3093c6c9b..1b068ce1aea9 100644
@@ -2067,6 +2067,41 @@ void drm_atomic_helper_cleanup_planes(struct drm_device *dev,
 EXPORT_SYMBOL(drm_atomic_helper_cleanup_planes);
 
 /**
+ * drm_atomic_helper_wait_for_hw_done - wait for all previous hw updates to complete.
+ * @crtc: crtc to wait on.
+ *
+ * This function waits for all previous hardware updates queued on @crtc
+ * to complete.
+ *
+ * Returns:
+ * 0 on success, negative error code on failure.
+ */
+int drm_atomic_helper_wait_for_hw_done(struct drm_crtc *crtc)
+{
+	struct drm_crtc_commit *commit;
+	long ret;
+
+	spin_lock(&crtc->commit_lock);
+	commit = list_first_entry_or_null(&crtc->commit_list,
+			struct drm_crtc_commit, commit_entry);
+	if (commit)
+		drm_crtc_commit_get(commit);
+	spin_unlock(&crtc->commit_lock);
+
+	if (!commit)
+		return 0;
+
+	ret = wait_for_completion_timeout(&commit->hw_done, 10 * HZ);
+	drm_crtc_commit_put(commit);
+
+	if (ret == 0)
+		return -EBUSY;
+
+	return ret > 0 ? 0 : ret;
+}
+EXPORT_SYMBOL(drm_atomic_helper_wait_for_hw_done);
+
+/**
  * drm_atomic_helper_swap_state - store atomic state into current sw state
  * @state: atomic state
  * @stall: stall for proceeding commits
@@ -2107,28 +2142,16 @@ void drm_atomic_helper_swap_state(struct drm_atomic_state *state,
 	struct drm_crtc_state *old_crtc_state, *new_crtc_state;
 	struct drm_plane *plane;
 	struct drm_plane_state *old_plane_state, *new_plane_state;
-	struct drm_crtc_commit *commit;
 	void *obj, *obj_state;
 	const struct drm_private_state_funcs *funcs;
 
 	if (stall) {
 		for_each_new_crtc_in_state(state, crtc, new_crtc_state, i) {
-			spin_lock(&crtc->commit_lock);
-			commit = list_first_entry_or_null(&crtc->commit_list,
-					struct drm_crtc_commit, commit_entry);
-			if (commit)
-				drm_crtc_commit_get(commit);
-			spin_unlock(&crtc->commit_lock);
+			ret = drm_atomic_helper_wait_for_hw_done(crtc);
 
-			if (!commit)
-				continue;
-
-			ret = wait_for_completion_timeout(&commit->hw_done,
-							  10*HZ);
-			if (ret == 0)
+			if (ret < 0)
 				DRM_ERROR("[CRTC:%d:%s] hw_done timed out\n",
 					  crtc->base.id, crtc->name);
-			drm_crtc_commit_put(commit);
 		}
 	}
 
include/drm/drm_atomic_helper.h +1 -0
index f0a8678ae98e..175a1a3fed36 100644
@@ -77,6 +77,7 @@ void
 drm_atomic_helper_disable_planes_on_crtc(struct drm_crtc_state *old_crtc_state,
 					 bool atomic);
 
+int drm_atomic_helper_wait_for_hw_done(struct drm_crtc *crtc);
 void drm_atomic_helper_swap_state(struct drm_atomic_state *state,
 				  bool stall);
 
Signature · 6 lines
2.11.0

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
author maintainer
#1 lore raw

PATCH v2 2/2 drm/i915: Wait for pending modesets to complete before trying link training, v2.

The nonblocking modeset tests were failing on systems with a DP output, because lijnk training could occur during the modeset.

With normal modesets we're protected by the connection_mutex lock, but nonblocking modesets drop locks before completing. To get around that, check if the most recent crtc_state has a modeset, and if this is the case, wait for all outstanding atomic updates on that crtc to complete.

This fixes the following spew when running nonblocking tests:

[ 1061.653778] ------------[ cut here ]------------
[ 1061.653781] WARNING: CPU: 1 PID: 8165 at
drivers/gpu/drm/drm_irq.c:1217 drm_wait_one_vblank+0x165/0x1a0
[ 1061.653782] vblank not available on crtc 0, ret=-22
[ 1061.653783] Modules linked in: vgem snd_hda_intel i915

snd_hda_codec_hdmi snd_hda_codec_realtek snd_hda_codec_generic x86_pkg_temp_thermal intel_powerclamp coretemp crct10dif_pclmul crc32_pclmul ghash_clmulni_intel snd_hda_codec snd_hwdep snd_hda_core snd_pcm mei_me mei e1000e ptp pps_core i2c_hid [last unloaded: vgem]

[ 1061.653796] CPU: 1 PID: 8165 Comm: kworker/1:3 Tainted: G     U  W
4.10.0-rc1-CI-Custom_114+ #1
[ 1061.653797] Hardware name: GIGABYTE GB-BKi5A-7200/MFLP5AP-00, BIOS F1
07/27/2016
[ 1061.653812] Workqueue: events i915_hpd_poll_init_work [i915]
[ 1061.653813] Call Trace:
[ 1061.653816]  dump_stack+0x67/0x92
[ 1061.653818]  __warn+0xc6/0xe0
[ 1061.653819]  warn_slowpath_fmt+0x4a/0x50
[ 1061.653820]  ? drm_vblank_get+0x74/0xc0
[ 1061.653821]  drm_wait_one_vblank+0x165/0x1a0
[ 1061.653823]  ? drm_dp_dpcd_write+0x16/0x20
[ 1061.653837]  ? intel_dp_set_link_train+0x44/0xe0 [i915]
[ 1061.653851]  intel_dp_check_link_status+0x113/0x1a0 [i915]
[ 1061.653864]  intel_dp_detect+0x5a1/0xa40 [i915]
[ 1061.653866]  drm_helper_hpd_irq_event+0x8d/0x110
[ 1061.653880]  i915_hpd_poll_init_work+0xab/0xd0 [i915]
[ 1061.653882]  process_one_work+0x1f4/0x6d0
[ 1061.653883]  ? process_one_work+0x16e/0x6d0
[ 1061.653885]  worker_thread+0x49/0x4a0
[ 1061.653886]  kthread+0x107/0x140
[ 1061.653888]  ? process_one_work+0x6d0/0x6d0
[ 1061.653889]  ? kthread_create_on_node+0x40/0x40
[ 1061.653891]  ret_from_fork+0x27/0x40
[ 1061.653892] ---[ end trace f6037618668d4cbc ]---
[ 1061.656353] [drm:intel_dp_set_idle_link_train [i915]] *ERROR* Timed out waiting for DP idle patterns

Changes since v1:
- intel_dp_detect() now holds connection_mutex and has an acquire_ctx, rewrite patch to take advantage of it.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99272
Testcase: kms_atomic_transitions.1x-modeset-transitions-nonblocking

---
 drivers/gpu/drm/i915/intel_dp.c | 72 ++++++++++++++++++++++++++++++++++-------
 1 file changed, 60 insertions(+), 12 deletions(-)
drivers/gpu/drm/i915/intel_dp.c +60 -12
index 64fa774c855b..c7a66883e453 100644
@@ -4243,21 +4243,32 @@ static void
 intel_dp_check_link_status(struct intel_dp *intel_dp)
 {
 	struct intel_encoder *intel_encoder = &dp_to_dig_port(intel_dp)->base;
-	struct drm_device *dev = intel_dp_to_dev(intel_dp);
 	u8 link_status[DP_LINK_STATUS_SIZE];
+	struct intel_connector *intel_connector = intel_dp->attached_connector;
+	struct drm_crtc *crtc;
 
-	WARN_ON(!drm_modeset_is_locked(&dev->mode_config.connection_mutex));
-
-	if (!intel_dp_get_link_status(intel_dp, link_status)) {
-		DRM_ERROR("Failed to get link status\n");
+	crtc = intel_connector->base.state->crtc;
+	if (!crtc)
 		return;
-	}
 
-	if (!intel_encoder->base.crtc)
+	WARN_ON(!drm_modeset_is_locked(&crtc->mutex));
+
+	if (!crtc->state->active)
 		return;
 
-	if (!to_intel_crtc(intel_encoder->base.crtc)->active)
+	if (drm_atomic_crtc_needs_modeset(crtc->state)) {
+		int ret;
+
+		/* wait for atomic modeset to complete */
+		ret = drm_atomic_helper_wait_for_hw_done(crtc);
+		if (ret < 0)
+			DRM_ERROR("Waiting for hw_done timed out\n");
+	}
+
+	if (!intel_dp_get_link_status(intel_dp, link_status)) {
+		DRM_ERROR("Failed to get link status\n");
 		return;
+	}
 
 	/*
 	 * Validate the cached values of intel_dp->link_rate and
@@ -4291,7 +4302,6 @@ intel_dp_check_link_status(struct intel_dp *intel_dp)
 static bool
 intel_dp_short_pulse(struct intel_dp *intel_dp)
 {
-	struct drm_device *dev = intel_dp_to_dev(intel_dp);
 	struct intel_encoder *intel_encoder = &dp_to_dig_port(intel_dp)->base;
 	u8 sink_irq_vector = 0;
 	u8 old_sink_count = intel_dp->sink_count;
@@ -4331,9 +4341,8 @@ intel_dp_short_pulse(struct intel_dp *intel_dp)
 			DRM_DEBUG_DRIVER("CP or sink specific irq unhandled\n");
 	}
 
-	drm_modeset_lock(&dev->mode_config.connection_mutex, NULL);
 	intel_dp_check_link_status(intel_dp);
-	drm_modeset_unlock(&dev->mode_config.connection_mutex);
+
 	if (intel_dp->compliance.test_type == DP_TEST_LINK_TRAINING) {
 		DRM_DEBUG_KMS("Link Training Compliance Test requested\n");
 		/* Send a Hotplug Uevent to userspace to start modeset */
@@ -4735,8 +4744,19 @@ intel_dp_detect(struct drm_connector *connector,
 		      connector->base.id, connector->name);
 
 	/* If full detect is not performed yet, do a full detect */
-	if (!intel_dp->detect_done)
+	if (!intel_dp->detect_done) {
+		struct drm_crtc *crtc;
+		int ret;
+
+		crtc = connector->state->crtc;
+		if (crtc) {
+			ret = drm_modeset_lock(&crtc->mutex, ctx);
+			if (ret)
+				return ret;
+		}
+
 		status = intel_dp_long_pulse(intel_dp->attached_connector);
+	}
 
 	intel_dp->detect_done = false;
 
@@ -5028,10 +5048,38 @@ intel_dp_hpd_pulse(struct intel_digital_port *intel_dig_port, bool long_hpd)
 	}
 
 	if (!intel_dp->is_mst) {
+		struct drm_modeset_acquire_ctx ctx;
+		struct drm_connector *connector = &intel_dp->attached_connector->base;
+		struct drm_crtc *crtc;
+		int iret;
+
+		drm_modeset_acquire_init(&ctx, 0);
+retry:
+		iret = drm_modeset_lock(&dev->mode_config.connection_mutex, &ctx);
+		if (iret)
+			goto err;
+
+		crtc = connector->state->crtc;
+		if (crtc) {
+			iret = drm_modeset_lock(&crtc->mutex, &ctx);
+			if (iret)
+				goto err;
+		}
+
 		if (!intel_dp_short_pulse(intel_dp)) {
 			intel_dp->detect_done = false;
 			goto put_power;
 		}
+
+err:
+		if (iret == -EDEADLK) {
+			drm_modeset_backoff(&ctx);
+			goto retry;
+		}
+
+		drm_modeset_drop_locks(&ctx);
+		drm_modeset_acquire_fini(&ctx);
+		WARN(iret, "Acquiring modeset locks failed with %i\n", iret);
 	}
 
 	ret = IRQ_HANDLED;
Signature · 6 lines
2.11.0

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#2 lore raw

Re: [PATCH v2 2/2] drm/i915: Wait for pending modesets to complete before trying link training, v2.

Quoted 150 lines · Maarten Lankhorst

The nonblocking modeset tests were failing on systems with a DP output, because lijnk training could occur during the modeset.

With normal modesets we're protected by the connection_mutex lock, but nonblocking modesets drop locks before completing. To get around that, check if the most recent crtc_state has a modeset, and if this is the case, wait for all outstanding atomic updates on that crtc to complete.

This fixes the following spew when running nonblocking tests:

[ 1061.653778] ------------[ cut here ]------------
[ 1061.653781] WARNING: CPU: 1 PID: 8165 at
drivers/gpu/drm/drm_irq.c:1217 drm_wait_one_vblank+0x165/0x1a0
[ 1061.653782] vblank not available on crtc 0, ret=-22
[ 1061.653783] Modules linked in: vgem snd_hda_intel i915

snd_hda_codec_hdmi snd_hda_codec_realtek snd_hda_codec_generic x86_pkg_temp_thermal intel_powerclamp coretemp crct10dif_pclmul crc32_pclmul ghash_clmulni_intel snd_hda_codec snd_hwdep snd_hda_core snd_pcm mei_me mei e1000e ptp pps_core i2c_hid [last unloaded: vgem]

[ 1061.653796] CPU: 1 PID: 8165 Comm: kworker/1:3 Tainted: G     U  W
4.10.0-rc1-CI-Custom_114+ #1
[ 1061.653797] Hardware name: GIGABYTE GB-BKi5A-7200/MFLP5AP-00, BIOS F1
07/27/2016
[ 1061.653812] Workqueue: events i915_hpd_poll_init_work [i915]
[ 1061.653813] Call Trace:
[ 1061.653816]  dump_stack+0x67/0x92
[ 1061.653818]  __warn+0xc6/0xe0
[ 1061.653819]  warn_slowpath_fmt+0x4a/0x50
[ 1061.653820]  ? drm_vblank_get+0x74/0xc0
[ 1061.653821]  drm_wait_one_vblank+0x165/0x1a0
[ 1061.653823]  ? drm_dp_dpcd_write+0x16/0x20
[ 1061.653837]  ? intel_dp_set_link_train+0x44/0xe0 [i915]
[ 1061.653851]  intel_dp_check_link_status+0x113/0x1a0 [i915]
[ 1061.653864]  intel_dp_detect+0x5a1/0xa40 [i915]
[ 1061.653866]  drm_helper_hpd_irq_event+0x8d/0x110
[ 1061.653880]  i915_hpd_poll_init_work+0xab/0xd0 [i915]
[ 1061.653882]  process_one_work+0x1f4/0x6d0
[ 1061.653883]  ? process_one_work+0x16e/0x6d0
[ 1061.653885]  worker_thread+0x49/0x4a0
[ 1061.653886]  kthread+0x107/0x140
[ 1061.653888]  ? process_one_work+0x6d0/0x6d0
[ 1061.653889]  ? kthread_create_on_node+0x40/0x40
[ 1061.653891]  ret_from_fork+0x27/0x40
[ 1061.653892] ---[ end trace f6037618668d4cbc ]---
[ 1061.656353] [drm:intel_dp_set_idle_link_train [i915]] *ERROR* Timed out waiting for DP idle patterns

Changes since v1:
- intel_dp_detect() now holds connection_mutex and has an acquire_ctx, rewrite patch to take advantage of it.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99272
Testcase: kms_atomic_transitions.1x-modeset-transitions-nonblocking

---
 drivers/gpu/drm/i915/intel_dp.c | 72 ++++++++++++++++++++++++++++++++++-------
 1 file changed, 60 insertions(+), 12 deletions(-)
drivers/gpu/drm/i915/intel_dp.c +36 -12
index 64fa774c855b..c7a66883e453 100644
@@ -4243,21 +4243,32 @@ static void
 intel_dp_check_link_status(struct intel_dp *intel_dp)
 {
 	struct intel_encoder *intel_encoder = &dp_to_dig_port(intel_dp)->base;
-	struct drm_device *dev = intel_dp_to_dev(intel_dp);
 	u8 link_status[DP_LINK_STATUS_SIZE];
+	struct intel_connector *intel_connector = intel_dp->attached_connector;
+	struct drm_crtc *crtc;
 
-	WARN_ON(!drm_modeset_is_locked(&dev->mode_config.connection_mutex));
-
-	if (!intel_dp_get_link_status(intel_dp, link_status)) {
-		DRM_ERROR("Failed to get link status\n");
+	crtc = intel_connector->base.state->crtc;
+	if (!crtc)
 		return;
-	}
 
-	if (!intel_encoder->base.crtc)
+	WARN_ON(!drm_modeset_is_locked(&crtc->mutex));
+
+	if (!crtc->state->active)
 		return;
 
-	if (!to_intel_crtc(intel_encoder->base.crtc)->active)
+	if (drm_atomic_crtc_needs_modeset(crtc->state)) {
+		int ret;
+
+		/* wait for atomic modeset to complete */
+		ret = drm_atomic_helper_wait_for_hw_done(crtc);
+		if (ret < 0)
+			DRM_ERROR("Waiting for hw_done timed out\n");
+	}
+
+	if (!intel_dp_get_link_status(intel_dp, link_status)) {
+		DRM_ERROR("Failed to get link status\n");
 		return;
+	}
 
 	/*
 	 * Validate the cached values of intel_dp->link_rate and
@@ -4291,7 +4302,6 @@ intel_dp_check_link_status(struct intel_dp *intel_dp)
 static bool
 intel_dp_short_pulse(struct intel_dp *intel_dp)
 {
-	struct drm_device *dev = intel_dp_to_dev(intel_dp);
 	struct intel_encoder *intel_encoder = &dp_to_dig_port(intel_dp)->base;
 	u8 sink_irq_vector = 0;
 	u8 old_sink_count = intel_dp->sink_count;
@@ -4331,9 +4341,8 @@ intel_dp_short_pulse(struct intel_dp *intel_dp)
 			DRM_DEBUG_DRIVER("CP or sink specific irq unhandled\n");
 	}
 
-	drm_modeset_lock(&dev->mode_config.connection_mutex, NULL);
 	intel_dp_check_link_status(intel_dp);
-	drm_modeset_unlock(&dev->mode_config.connection_mutex);
+
 	if (intel_dp->compliance.test_type == DP_TEST_LINK_TRAINING) {
 		DRM_DEBUG_KMS("Link Training Compliance Test requested\n");
 		/* Send a Hotplug Uevent to userspace to start modeset */
@@ -4735,8 +4744,19 @@ intel_dp_detect(struct drm_connector *connector,
 		      connector->base.id, connector->name);
 
 	/* If full detect is not performed yet, do a full detect */
-	if (!intel_dp->detect_done)
+	if (!intel_dp->detect_done) {
+		struct drm_crtc *crtc;
+		int ret;
+
+		crtc = connector->state->crtc;
+		if (crtc) {
+			ret = drm_modeset_lock(&crtc->mutex, ctx);
+			if (ret)
+				return ret;
+		}
+
 		status = intel_dp_long_pulse(intel_dp->attached_connector);
+	}
 
 	intel_dp->detect_done = false;
 
@@ -5028,10 +5048,38 @@ intel_dp_hpd_pulse(struct intel_digital_port *intel_dig_port, bool long_hpd)
 	}
 
 	if (!intel_dp->is_mst) {
+		struct drm_modeset_acquire_ctx ctx;
+		struct drm_connector *connector = &intel_dp->attached_connector->base;
+		struct drm_crtc *crtc;
+		int iret;

drivers/gpu/drm/i915/intel_dp.c intel_dp_hpd_pulse(struct intel_digital_port *intel_dig_port, bool long_hpd) · line 5064

Quoted 10 lines
drivers/gpu/drm/i915/intel_dp.c +10 -0
+
+		drm_modeset_acquire_init(&ctx, 0);
+retry:
+		iret = drm_modeset_lock(&dev->mode_config.connection_mutex, &ctx);
+		if (iret)
+			goto err;
+
+		crtc = connector->state->crtc;
+		if (crtc) {
+			iret = drm_modeset_lock(&crtc->mutex, &ctx);

Please no modeset locks in the dig_work. Hmm. Looks like we already have it there though :( I thought we had moved the link retraining fully into the hoplug_work, but I guess not. Can you try to do that instead?

Quoted 28 lines
drivers/gpu/drm/i915/intel_dp.c +14 -0
+			if (iret)
+				goto err;
+		}
+
 		if (!intel_dp_short_pulse(intel_dp)) {
 			intel_dp->detect_done = false;
 			goto put_power;
 		}
+
+err:
+		if (iret == -EDEADLK) {
+			drm_modeset_backoff(&ctx);
+			goto retry;
+		}
+
+		drm_modeset_drop_locks(&ctx);
+		drm_modeset_acquire_fini(&ctx);
+		WARN(iret, "Acquiring modeset locks failed with %i\n", iret);
 	}
 
 	ret = IRQ_HANDLED;
Signature · 6 lines
2.11.0

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 6 lines
Ville Syrjälä
Intel OTC
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
author maintainer
#3 lore raw

Re: [PATCH v2 2/2] drm/i915: Wait for pending modesets to complete before trying link training, v2.

Op 14-06-17 om 15:44 schreef Ville Syrjälä:

Quoted 156 lines · Maarten Lankhorst
Quoted 155 lines

The nonblocking modeset tests were failing on systems with a DP output, because lijnk training could occur during the modeset.

With normal modesets we're protected by the connection_mutex lock, but nonblocking modesets drop locks before completing. To get around that, check if the most recent crtc_state has a modeset, and if this is the case, wait for all outstanding atomic updates on that crtc to complete.

This fixes the following spew when running nonblocking tests:

[ 1061.653778] ------------[ cut here ]------------
[ 1061.653781] WARNING: CPU: 1 PID: 8165 at
drivers/gpu/drm/drm_irq.c:1217 drm_wait_one_vblank+0x165/0x1a0
[ 1061.653782] vblank not available on crtc 0, ret=-22
[ 1061.653783] Modules linked in: vgem snd_hda_intel i915

snd_hda_codec_hdmi snd_hda_codec_realtek snd_hda_codec_generic x86_pkg_temp_thermal intel_powerclamp coretemp crct10dif_pclmul crc32_pclmul ghash_clmulni_intel snd_hda_codec snd_hwdep snd_hda_core snd_pcm mei_me mei e1000e ptp pps_core i2c_hid [last unloaded: vgem]

[ 1061.653796] CPU: 1 PID: 8165 Comm: kworker/1:3 Tainted: G     U  W
4.10.0-rc1-CI-Custom_114+ #1
[ 1061.653797] Hardware name: GIGABYTE GB-BKi5A-7200/MFLP5AP-00, BIOS F1
07/27/2016
[ 1061.653812] Workqueue: events i915_hpd_poll_init_work [i915]
[ 1061.653813] Call Trace:
[ 1061.653816]  dump_stack+0x67/0x92
[ 1061.653818]  __warn+0xc6/0xe0
[ 1061.653819]  warn_slowpath_fmt+0x4a/0x50
[ 1061.653820]  ? drm_vblank_get+0x74/0xc0
[ 1061.653821]  drm_wait_one_vblank+0x165/0x1a0
[ 1061.653823]  ? drm_dp_dpcd_write+0x16/0x20
[ 1061.653837]  ? intel_dp_set_link_train+0x44/0xe0 [i915]
[ 1061.653851]  intel_dp_check_link_status+0x113/0x1a0 [i915]
[ 1061.653864]  intel_dp_detect+0x5a1/0xa40 [i915]
[ 1061.653866]  drm_helper_hpd_irq_event+0x8d/0x110
[ 1061.653880]  i915_hpd_poll_init_work+0xab/0xd0 [i915]
[ 1061.653882]  process_one_work+0x1f4/0x6d0
[ 1061.653883]  ? process_one_work+0x16e/0x6d0
[ 1061.653885]  worker_thread+0x49/0x4a0
[ 1061.653886]  kthread+0x107/0x140
[ 1061.653888]  ? process_one_work+0x6d0/0x6d0
[ 1061.653889]  ? kthread_create_on_node+0x40/0x40
[ 1061.653891]  ret_from_fork+0x27/0x40
[ 1061.653892] ---[ end trace f6037618668d4cbc ]---
[ 1061.656353] [drm:intel_dp_set_idle_link_train [i915]] *ERROR* Timed out waiting for DP idle patterns

Changes since v1:
- intel_dp_detect() now holds connection_mutex and has an acquire_ctx, rewrite patch to take advantage of it.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99272
Testcase: kms_atomic_transitions.1x-modeset-transitions-nonblocking

---
 drivers/gpu/drm/i915/intel_dp.c | 72 ++++++++++++++++++++++++++++++++++-------
 1 file changed, 60 insertions(+), 12 deletions(-)
drivers/gpu/drm/i915/intel_dp.c +41 -12
index 64fa774c855b..c7a66883e453 100644
@@ -4243,21 +4243,32 @@ static void
 intel_dp_check_link_status(struct intel_dp *intel_dp)
 {
 	struct intel_encoder *intel_encoder = &dp_to_dig_port(intel_dp)->base;
-	struct drm_device *dev = intel_dp_to_dev(intel_dp);
 	u8 link_status[DP_LINK_STATUS_SIZE];
+	struct intel_connector *intel_connector = intel_dp->attached_connector;
+	struct drm_crtc *crtc;
 
-	WARN_ON(!drm_modeset_is_locked(&dev->mode_config.connection_mutex));
-
-	if (!intel_dp_get_link_status(intel_dp, link_status)) {
-		DRM_ERROR("Failed to get link status\n");
+	crtc = intel_connector->base.state->crtc;
+	if (!crtc)
 		return;
-	}
 
-	if (!intel_encoder->base.crtc)
+	WARN_ON(!drm_modeset_is_locked(&crtc->mutex));
+
+	if (!crtc->state->active)
 		return;
 
-	if (!to_intel_crtc(intel_encoder->base.crtc)->active)
+	if (drm_atomic_crtc_needs_modeset(crtc->state)) {
+		int ret;
+
+		/* wait for atomic modeset to complete */
+		ret = drm_atomic_helper_wait_for_hw_done(crtc);
+		if (ret < 0)
+			DRM_ERROR("Waiting for hw_done timed out\n");
+	}
+
+	if (!intel_dp_get_link_status(intel_dp, link_status)) {
+		DRM_ERROR("Failed to get link status\n");
 		return;
+	}
 
 	/*
 	 * Validate the cached values of intel_dp->link_rate and
@@ -4291,7 +4302,6 @@ intel_dp_check_link_status(struct intel_dp *intel_dp)
 static bool
 intel_dp_short_pulse(struct intel_dp *intel_dp)
 {
-	struct drm_device *dev = intel_dp_to_dev(intel_dp);
 	struct intel_encoder *intel_encoder = &dp_to_dig_port(intel_dp)->base;
 	u8 sink_irq_vector = 0;
 	u8 old_sink_count = intel_dp->sink_count;
@@ -4331,9 +4341,8 @@ intel_dp_short_pulse(struct intel_dp *intel_dp)
 			DRM_DEBUG_DRIVER("CP or sink specific irq unhandled\n");
 	}
 
-	drm_modeset_lock(&dev->mode_config.connection_mutex, NULL);
 	intel_dp_check_link_status(intel_dp);
-	drm_modeset_unlock(&dev->mode_config.connection_mutex);
+
 	if (intel_dp->compliance.test_type == DP_TEST_LINK_TRAINING) {
 		DRM_DEBUG_KMS("Link Training Compliance Test requested\n");
 		/* Send a Hotplug Uevent to userspace to start modeset */
@@ -4735,8 +4744,19 @@ intel_dp_detect(struct drm_connector *connector,
 		      connector->base.id, connector->name);
 
 	/* If full detect is not performed yet, do a full detect */
-	if (!intel_dp->detect_done)
+	if (!intel_dp->detect_done) {
+		struct drm_crtc *crtc;
+		int ret;
+
+		crtc = connector->state->crtc;
+		if (crtc) {
+			ret = drm_modeset_lock(&crtc->mutex, ctx);
+			if (ret)
+				return ret;
+		}
+
 		status = intel_dp_long_pulse(intel_dp->attached_connector);
+	}
 
 	intel_dp->detect_done = false;
 
@@ -5028,10 +5048,38 @@ intel_dp_hpd_pulse(struct intel_digital_port *intel_dig_port, bool long_hpd)
 	}
 
 	if (!intel_dp->is_mst) {
+		struct drm_modeset_acquire_ctx ctx;
+		struct drm_connector *connector = &intel_dp->attached_connector->base;
+		struct drm_crtc *crtc;
+		int iret;
+
+		drm_modeset_acquire_init(&ctx, 0);
+retry:
+		iret = drm_modeset_lock(&dev->mode_config.connection_mutex, &ctx);
+		if (iret)
Quoted 8 lines

drivers/gpu/drm/i915/intel_dp.c intel_dp_hpd_pulse(struct intel_digital_port *intel_dig_port, bool long_hpd) · line 5064

Quoted 5 lines
drivers/gpu/drm/i915/intel_dp.c +5 -0
+			goto err;
+
+		crtc = connector->state->crtc;
+		if (crtc) {
+			iret = drm_modeset_lock(&crtc->mutex, &ctx);

Please no modeset locks in the dig_work. Hmm. Looks like we already have it there though :( I thought we had moved the link retraining fully into the hoplug_work, but I guess not. Can you try to do that instead?

I see you've had a series to fix this, what happened to it?

Quoted 20 lines
Quoted 20 lines
drivers/gpu/drm/i915/intel_dp.c +14 -0
+			if (iret)
+				goto err;
+		}
+
 		if (!intel_dp_short_pulse(intel_dp)) {
 			intel_dp->detect_done = false;
 			goto put_power;
 		}
+
+err:
+		if (iret == -EDEADLK) {
+			drm_modeset_backoff(&ctx);
+			goto retry;
+		}
+
+		drm_modeset_drop_locks(&ctx);
+		drm_modeset_acquire_fini(&ctx);
+		WARN(iret, "Acquiring modeset locks failed with %i\n", iret);
 	}
 
Quoted 8 lines
Quoted 8 lines
drivers/gpu/drm/i915/intel_dp.c +0 -0
 	ret = IRQ_HANDLED;
Signature · 6 lines
2.11.0

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#4 lore raw

✓ Fi.CI.BAT: success for series starting with [v2,1/2] drm/atomic: move waiting for hw_done to a helper

In reply to Maarten Lankhorst
== Series Details ==

== Series Details ==

Series: series starting with [v2,1/2] drm/atomic: move waiting for hw_done to a helper
URL : https://patchwork.freedesktop.org/series/25761/
State : success

== Summary ==

Series 25761v1 Series without cover letter
https://patchwork.freedesktop.org/api/1.0/series/25761/revisions/1/mbox/

fi-bdw-5557u     total:278  pass:267  dwarn:0   dfail:0   fail:0   skip:11  time:447s
fi-bdw-gvtdvm    total:278  pass:256  dwarn:8   dfail:0   fail:0   skip:14  time:432s
fi-bsw-n3050     total:278  pass:242  dwarn:0   dfail:0   fail:0   skip:36  time:580s
fi-bxt-j4205     total:278  pass:259  dwarn:0   dfail:0   fail:0   skip:19  time:506s
fi-byt-j1900     total:278  pass:254  dwarn:0   dfail:0   fail:0   skip:24  time:484s
fi-byt-n2820     total:278  pass:250  dwarn:0   dfail:0   fail:0   skip:28  time:482s
fi-glk-2a        total:278  pass:259  dwarn:0   dfail:0   fail:0   skip:19  time:586s
fi-hsw-4770      total:278  pass:262  dwarn:0   dfail:0   fail:0   skip:16  time:434s
fi-hsw-4770r     total:278  pass:262  dwarn:0   dfail:0   fail:0   skip:16  time:418s
fi-ilk-650       total:278  pass:228  dwarn:0   dfail:0   fail:0   skip:50  time:417s
fi-ivb-3520m     total:278  pass:260  dwarn:0   dfail:0   fail:0   skip:18  time:495s
fi-ivb-3770      total:278  pass:260  dwarn:0   dfail:0   fail:0   skip:18  time:471s
fi-kbl-7500u     total:278  pass:260  dwarn:0   dfail:0   fail:0   skip:18  time:466s
fi-skl-6260u     total:278  pass:268  dwarn:0   dfail:0   fail:0   skip:10  time:454s
fi-skl-6700hq    total:278  pass:228  dwarn:1   dfail:0   fail:27  skip:22  time:401s
fi-skl-6700k     total:278  pass:256  dwarn:4   dfail:0   fail:0   skip:18  time:462s
fi-skl-6770hq    total:278  pass:268  dwarn:0   dfail:0   fail:0   skip:10  time:475s
fi-skl-gvtdvm    total:278  pass:265  dwarn:0   dfail:0   fail:0   skip:13  time:434s
fi-snb-2520m     total:278  pass:250  dwarn:0   dfail:0   fail:0   skip:28  time:532s
fi-snb-2600      total:278  pass:248  dwarn:0   dfail:0   fail:1   skip:29  time:401s

d4bedb8b0f9ba91df2e8cb136a489145a83e96a7 drm-tip: 2017y-06m-13d-14h-22m-46s UTC integration manifest b79675f drm/i915: Wait for pending modesets to complete before trying link training, v2. 48dd8f5 drm/atomic: move waiting for hw_done to a helper

== Logs ==

For more details see: https://intel-gfx-ci.01.org/CI/Patchwork_4949/
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#5 lore raw
In reply to Maarten Lankhorst
Quoted 41 lines · Maarten Lankhorst

Moving the wait to a helper allows callers outside of the core to wait for pending updates to complete.

This can be used to prevent races against nonblocking modesets. Only the hw_done call in swap_state is moved to a helper, doing it for the other callers requires too many changes and I think this is the only useful one to export.

---
 drivers/gpu/drm/drm_atomic_helper.c | 51 +++++++++++++++++++++++++++----------
 include/drm/drm_atomic_helper.h     |  1 +
 2 files changed, 38 insertions(+), 14 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +19 -0
index 86d3093c6c9b..1b068ce1aea9 100644
@@ -2067,6 +2067,41 @@ void drm_atomic_helper_cleanup_planes(struct drm_device *dev,
 EXPORT_SYMBOL(drm_atomic_helper_cleanup_planes);
 
 /**
+ * drm_atomic_helper_wait_for_hw_done - wait for all previous hw updates to complete.
+ * @crtc: crtc to wait on.
+ *
+ * This function waits for all previous hardware updates queued on @crtc
+ * to complete.
+ *
+ * Returns:
+ * 0 on success, negative error code on failure.
+ */
+int drm_atomic_helper_wait_for_hw_done(struct drm_crtc *crtc)
+{
+	struct drm_crtc_commit *commit;
+	long ret;
+
+	spin_lock(&crtc->commit_lock);
+	commit = list_first_entry_or_null(&crtc->commit_list,
+			struct drm_crtc_commit, commit_entry);
+	if (commit)
+		drm_crtc_commit_get(commit);

drivers/gpu/drm/drm_atomic_helper.c void drm_atomic_helper_cleanup_planes(struct drm_device *dev, · line 2098

Quoted 10 lines
drivers/gpu/drm/drm_atomic_helper.c +10 -0
+	spin_unlock(&crtc->commit_lock);
+
+	if (!commit)
+		return 0;
+
+	ret = wait_for_completion_timeout(&commit->hw_done, 10 * HZ);
+	drm_crtc_commit_put(commit);
+
+	if (ret == 0)
+		return -EBUSY;

-ETIMEDOUT ?

Quoted 59 lines
drivers/gpu/drm/drm_atomic_helper.c +8 -14
+
+	return ret > 0 ? 0 : ret;
+}
+EXPORT_SYMBOL(drm_atomic_helper_wait_for_hw_done);
+
+/**
  * drm_atomic_helper_swap_state - store atomic state into current sw state
  * @state: atomic state
  * @stall: stall for proceeding commits
@@ -2107,28 +2142,16 @@ void drm_atomic_helper_swap_state(struct drm_atomic_state *state,
 	struct drm_crtc_state *old_crtc_state, *new_crtc_state;
 	struct drm_plane *plane;
 	struct drm_plane_state *old_plane_state, *new_plane_state;
-	struct drm_crtc_commit *commit;
 	void *obj, *obj_state;
 	const struct drm_private_state_funcs *funcs;
 
 	if (stall) {
 		for_each_new_crtc_in_state(state, crtc, new_crtc_state, i) {
-			spin_lock(&crtc->commit_lock);
-			commit = list_first_entry_or_null(&crtc->commit_list,
-					struct drm_crtc_commit, commit_entry);
-			if (commit)
-				drm_crtc_commit_get(commit);
-			spin_unlock(&crtc->commit_lock);
+			ret = drm_atomic_helper_wait_for_hw_done(crtc);
 
-			if (!commit)
-				continue;
-
-			ret = wait_for_completion_timeout(&commit->hw_done,
-							  10*HZ);
-			if (ret == 0)
+			if (ret < 0)
 				DRM_ERROR("[CRTC:%d:%s] hw_done timed out\n",
 					  crtc->base.id, crtc->name);
-			drm_crtc_commit_put(commit);
 		}
 	}
 
include/drm/drm_atomic_helper.h +1 -0
index f0a8678ae98e..175a1a3fed36 100644
@@ -77,6 +77,7 @@ void
 drm_atomic_helper_disable_planes_on_crtc(struct drm_crtc_state *old_crtc_state,
 					 bool atomic);
 
+int drm_atomic_helper_wait_for_hw_done(struct drm_crtc *crtc);
 void drm_atomic_helper_swap_state(struct drm_atomic_state *state,
 				  bool stall);
 
Signature · 6 lines
2.11.0

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 6 lines
Ville Syrjälä
Intel OTC
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
author maintainer
#6 lore raw

Op 14-06-17 om 16:52 schreef Ville Syrjälä:

Quoted 45 lines · Maarten Lankhorst
Quoted 44 lines

Moving the wait to a helper allows callers outside of the core to wait for pending updates to complete.

This can be used to prevent races against nonblocking modesets. Only the hw_done call in swap_state is moved to a helper, doing it for the other callers requires too many changes and I think this is the only useful one to export.

---
 drivers/gpu/drm/drm_atomic_helper.c | 51 +++++++++++++++++++++++++++----------
 include/drm/drm_atomic_helper.h     |  1 +
 2 files changed, 38 insertions(+), 14 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +22 -0
index 86d3093c6c9b..1b068ce1aea9 100644
@@ -2067,6 +2067,41 @@ void drm_atomic_helper_cleanup_planes(struct drm_device *dev,
 EXPORT_SYMBOL(drm_atomic_helper_cleanup_planes);
 
 /**
+ * drm_atomic_helper_wait_for_hw_done - wait for all previous hw updates to complete.
+ * @crtc: crtc to wait on.
+ *
+ * This function waits for all previous hardware updates queued on @crtc
+ * to complete.
+ *
+ * Returns:
+ * 0 on success, negative error code on failure.
+ */
+int drm_atomic_helper_wait_for_hw_done(struct drm_crtc *crtc)
+{
+	struct drm_crtc_commit *commit;
+	long ret;
+
+	spin_lock(&crtc->commit_lock);
+	commit = list_first_entry_or_null(&crtc->commit_list,
+			struct drm_crtc_commit, commit_entry);
+	if (commit)
+		drm_crtc_commit_get(commit);
+	spin_unlock(&crtc->commit_lock);
+
+	if (!commit)
Quoted 8 lines

drivers/gpu/drm/drm_atomic_helper.c void drm_atomic_helper_cleanup_planes(struct drm_device *dev, · line 2098

Quoted 7 lines
drivers/gpu/drm/drm_atomic_helper.c +7 -0
+		return 0;
+
+	ret = wait_for_completion_timeout(&commit->hw_done, 10 * HZ);
+	drm_crtc_commit_put(commit);
+
+	if (ret == 0)
+		return -EBUSY;

-ETIMEDOUT ?

ETIMEDOUT -> /* Connection timed out */

No I don't think that would match.
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#7 lore raw
Quoted 54 lines · Maarten Lankhorst

Op 14-06-17 om 16:52 schreef Ville Syrjälä:

Quoted 45 lines · Maarten Lankhorst
Quoted 44 lines

Moving the wait to a helper allows callers outside of the core to wait for pending updates to complete.

This can be used to prevent races against nonblocking modesets. Only the hw_done call in swap_state is moved to a helper, doing it for the other callers requires too many changes and I think this is the only useful one to export.

---
 drivers/gpu/drm/drm_atomic_helper.c | 51 +++++++++++++++++++++++++++----------
 include/drm/drm_atomic_helper.h     |  1 +
 2 files changed, 38 insertions(+), 14 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +22 -0
index 86d3093c6c9b..1b068ce1aea9 100644
@@ -2067,6 +2067,41 @@ void drm_atomic_helper_cleanup_planes(struct drm_device *dev,
 EXPORT_SYMBOL(drm_atomic_helper_cleanup_planes);
 
 /**
+ * drm_atomic_helper_wait_for_hw_done - wait for all previous hw updates to complete.
+ * @crtc: crtc to wait on.
+ *
+ * This function waits for all previous hardware updates queued on @crtc
+ * to complete.
+ *
+ * Returns:
+ * 0 on success, negative error code on failure.
+ */
+int drm_atomic_helper_wait_for_hw_done(struct drm_crtc *crtc)
+{
+	struct drm_crtc_commit *commit;
+	long ret;
+
+	spin_lock(&crtc->commit_lock);
+	commit = list_first_entry_or_null(&crtc->commit_list,
+			struct drm_crtc_commit, commit_entry);
+	if (commit)
+		drm_crtc_commit_get(commit);
+	spin_unlock(&crtc->commit_lock);
+
+	if (!commit)
Quoted 8 lines

drivers/gpu/drm/drm_atomic_helper.c void drm_atomic_helper_cleanup_planes(struct drm_device *dev, · line 2098

Quoted 7 lines
drivers/gpu/drm/drm_atomic_helper.c +7 -0
+		return 0;
+
+	ret = wait_for_completion_timeout(&commit->hw_done, 10 * HZ);
+	drm_crtc_commit_put(commit);
+
+	if (ret == 0)
+		return -EBUSY;

-ETIMEDOUT ?

Quoted 1 line

ETIMEDOUT -> /* Connection timed out */

It's the standard "something timed out" errno. Used all over drm.

Quoted 2 lines

No I don't think that would match.

Signature · 6 lines
Ville Syrjälä
Intel OTC
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel