[PATCH v7 00/26] drm/i915/display: All patches to make PREEMPT_RT work on i915 + xe.

[PATCH v7 00/26] drm/i915/display: All patches to make PREEMPT_RT work on i915 + xe.

[Maarten Lankhorst]

New version with patch from Sebastian to fix starvation of irq_work.

Still need feedback on the vblank approach, and check to see if there are some bugs left.
Hoping this passes BAT on i915, and still mostly the full run on xe.

Maarten Lankhorst (20):
  drm/vblank_work: Add methods to schedule vblank_work in 2 stages
  drm/vblank: Add a 2-stage version of drm_crtc_arm_vblank_event
  drm/intel/display: Make intel_crtc_arm_vblank_event static
  drm/intel/display: Convert vblank event handling to 2-stage arming
  drm/i915/display: Move vblank put until after critical section
  drm/i915/display: Remove locking from intel_vblank_evade critical
    section
  drm/i915/display: Handle vlv dsi workaround in scanline_in_safe_range
    too
  drm/i915/display: Make get_vblank_counter use intel_de_read_fw()
  drm/i915/display: Do not take uncore lock in i915_get_vblank_counter
  drm/i915/display: Make icl_dsi_frame_update use _fw too
  drm/i915/display: Use intel_de_read/write_fw in colorops
  drm/i915/display: Use intel_de_write_fw in intel_pipe_fastset
  drm/i915/display: Make set_pipeconf use the fw variants
  drm/i915/display: Fix intel_lpe_audio_irq_handler for PREEMPT-RT
  drm/i915/gt: Fix selftests on PREEMPT_RT
  drm/i915/gt: Set stop_timeout() correctly on PREEMPT-RT
  drm/i915/display: Remove uncore lock from vlv_atomic_update_fifo
  drm/i915: Use sleeping selftests for igt_atomic on PREEMPT_RT
  PREEMPT_RT injection
  FOR-CI: bump MAX_STACK_TRACE_ENTRIES

Mike Galbraith (1):
  drm/i915: Use preempt_disable/enable_rt() where recommended

Sebastian Andrzej Siewior (5):
  drm/i915/gt: Use spin_lock_irq() instead of local_irq_disable() +
    spin_lock()
  drm/i915: Drop the irqs_disabled() check
  drm/i915/guc: Consider also RCU depth in busy loop.
  Revert "drm/i915: Depend on !PREEMPT_RT."
  drm/i915/gt: Add a spinlock to prevent starvation of irq_work.

 drivers/gpu/drm/drm_vblank.c                  |  61 +++++++++-
 drivers/gpu/drm/drm_vblank_work.c             | 106 ++++++++++++----
 drivers/gpu/drm/i915/Kconfig                  |   1 -
 drivers/gpu/drm/i915/Kconfig.debug            |  15 ---
 drivers/gpu/drm/i915/display/i9xx_wm.c        |   4 -
 drivers/gpu/drm/i915/display/icl_dsi.c        |   4 +-
 drivers/gpu/drm/i915/display/intel_color.c    |   6 +-
 drivers/gpu/drm/i915/display/intel_crtc.c     |  86 ++++++-------
 drivers/gpu/drm/i915/display/intel_crtc.h     |   1 -
 drivers/gpu/drm/i915/display/intel_cursor.c   |   8 +-
 drivers/gpu/drm/i915/display/intel_de.h       |   8 ++
 drivers/gpu/drm/i915/display/intel_display.c  |  46 +++----
 .../gpu/drm/i915/display/intel_lpe_audio.c    |   2 +-
 drivers/gpu/drm/i915/display/intel_vblank.c   | 115 ++++++++++--------
 drivers/gpu/drm/i915/display/intel_vblank.h   |   1 +
 drivers/gpu/drm/i915/display/intel_vrr.c      |  18 +--
 drivers/gpu/drm/i915/gt/intel_breadcrumbs.c   |   8 +-
 .../gpu/drm/i915/gt/intel_breadcrumbs_types.h |   1 +
 drivers/gpu/drm/i915/gt/intel_engine_cs.c     |   2 +-
 .../drm/i915/gt/intel_execlists_submission.c  |  17 +--
 drivers/gpu/drm/i915/gt/selftest_engine_pm.c  |   8 +-
 drivers/gpu/drm/i915/gt/uc/intel_guc.h        |   2 +-
 drivers/gpu/drm/i915/i915_request.c           |   2 -
 drivers/gpu/drm/i915/intel_uncore.h           |  26 ++--
 drivers/gpu/drm/i915/selftests/igt_atomic.c   |   7 ++
 drivers/gpu/drm/xe/Kconfig.debug              |   5 +
 .../drm/xe/compat-i915-headers/intel_uncore.h |   7 ++
 include/drm/drm_vblank.h                      |  14 ++-
 include/drm/drm_vblank_work.h                 |  12 ++
 kernel/Kconfig.preempt                        |   4 +-
 lib/Kconfig.debug                             |   4 +-
 31 files changed, 391 insertions(+), 210 deletions(-)

-- 
2.51.0

[PATCH v7 01/26] drm/vblank_work: Add methods to schedule vblank_work in 2 stages

[Maarten Lankhorst]

In case of vblank evasion in intel/display, it's necessary to
perform some work in advance, so the critical section will always run in
constant time on PREEMPT_RT.

By preparing all the work in advance, the part that needs to finish in
constant time only has to write a single variable instead. This allows
PREEMPT_RT to keep the interrupts disabled at the most critical part,
without completely reworking all locks to be raw spinlocks.

Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
---
 drivers/gpu/drm/drm_vblank_work.c | 106 ++++++++++++++++++++++--------
 include/drm/drm_vblank_work.h     |  12 ++++
 2 files changed, 92 insertions(+), 26 deletions(-)

diff --git a/drivers/gpu/drm/drm_vblank_work.c b/drivers/gpu/drm/drm_vblank_work.c
index 70f0199251ea0..6acb240b9c112 100644
--- a/drivers/gpu/drm/drm_vblank_work.c
+++ b/drivers/gpu/drm/drm_vblank_work.c
@@ -54,7 +54,8 @@ void drm_handle_vblank_works(struct drm_vblank_crtc *vblank)
 	assert_spin_locked(&vblank->dev->event_lock);
 
 	list_for_each_entry_safe(work, next, &vblank->pending_work, node) {
-		if (!drm_vblank_passed(count, work->count))
+		/* READ_ONCE pairs with WRITE_ONCE in drm_vblank_work_enable() */
+		if (!READ_ONCE(work->armed) || !drm_vblank_passed(count, work->count))
 			continue;
 
 		list_del_init(&work->node);
@@ -86,30 +87,8 @@ void drm_vblank_cancel_pending_works(struct drm_vblank_crtc *vblank)
 	wake_up_all(&vblank->work_wait_queue);
 }
 
-/**
- * drm_vblank_work_schedule - schedule a vblank work
- * @work: vblank work to schedule
- * @count: target vblank count
- * @nextonmiss: defer until the next vblank if target vblank was missed
- *
- * Schedule @work for execution once the crtc vblank count reaches @count.
- *
- * If the crtc vblank count has already reached @count and @nextonmiss is
- * %false the work starts to execute immediately.
- *
- * If the crtc vblank count has already reached @count and @nextonmiss is
- * %true the work is deferred until the next vblank (as if @count has been
- * specified as crtc vblank count + 1).
- *
- * If @work is already scheduled, this function will reschedule said work
- * using the new @count. This can be used for self-rearming work items.
- *
- * Returns:
- * %1 if @work was successfully (re)scheduled, %0 if it was either already
- * scheduled or cancelled, or a negative error code on failure.
- */
-int drm_vblank_work_schedule(struct drm_vblank_work *work,
-			     u64 count, bool nextonmiss)
+static int __drm_vblank_work_schedule(struct drm_vblank_work *work,
+				      u64 count, bool nextonmiss, bool armed)
 {
 	struct drm_vblank_crtc *vblank = work->vblank;
 	struct drm_device *dev = vblank->dev;
@@ -139,6 +118,7 @@ int drm_vblank_work_schedule(struct drm_vblank_work *work,
 		rescheduling = true;
 	}
 
+	work->armed = armed;
 	work->count = count;
 	cur_vbl = drm_vblank_count(dev, vblank->pipe);
 	passed = drm_vblank_passed(cur_vbl, count);
@@ -147,7 +127,7 @@ int drm_vblank_work_schedule(struct drm_vblank_work *work,
 			     "crtc %d vblank %llu already passed (current %llu)\n",
 			     vblank->pipe, count, cur_vbl);
 
-	if (!nextonmiss && passed) {
+	if (!nextonmiss && passed && armed) {
 		drm_vblank_put(dev, vblank->pipe);
 		ret = kthread_queue_work(vblank->worker, &work->base);
 
@@ -167,8 +147,82 @@ int drm_vblank_work_schedule(struct drm_vblank_work *work,
 		wake_up_all(&vblank->work_wait_queue);
 	return ret;
 }
+
+/**
+ * drm_vblank_work_schedule - schedule a vblank work
+ * @work: vblank work to schedule
+ * @count: target vblank count
+ * @nextonmiss: defer until the next vblank if target vblank was missed
+ *
+ * Schedule @work for execution once the crtc vblank count reaches @count.
+ *
+ * If the crtc vblank count has already reached @count and @nextonmiss is
+ * %false the work starts to execute immediately.
+ *
+ * If the crtc vblank count has already reached @count and @nextonmiss is
+ * %true the work is deferred until the next vblank (as if @count has been
+ * specified as crtc vblank count + 1).
+ *
+ * If @work is already scheduled, this function will reschedule said work
+ * using the new @count. This can be used for self-rearming work items.
+ *
+ * Returns:
+ * %1 if @work was successfully (re)scheduled, %0 if it was either already
+ * scheduled or cancelled, or a negative error code on failure.
+ */
+int drm_vblank_work_schedule(struct drm_vblank_work *work,
+			     u64 count, bool nextonmiss)
+{
+	return __drm_vblank_work_schedule(work, count, nextonmiss, true);
+}
 EXPORT_SYMBOL(drm_vblank_work_schedule);
 
+
+/**
+ * drm_vblank_work_schedule_disabled - schedule a vblank work, withoug enabling
+ * @work: vblank work to schedule
+ * @count: target vblank count
+ *
+ * Schedule @work for execution once the crtc vblank count reaches @count.
+ *
+ * The vblank work will not be scheduled until drm_vblank_work_enable() is called.
+ * If the crtc vblank count has already reached @count, the work will still
+ * not be scheduled until the first following vblank.
+ *
+ * If @work is already scheduled, this function will reschedule said work
+ * using the new @count. This can be used for self-rearming work items.
+ *
+ * Returns:
+ * %1 if @work was successfully (re)scheduled, %0 if it was either already
+ * scheduled or cancelled, or a negative error code on failure.
+ */
+int drm_vblank_work_schedule_disabled(struct drm_vblank_work *work, u64 count)
+{
+	return __drm_vblank_work_schedule(work, count, true, false);
+}
+EXPORT_SYMBOL(drm_vblank_work_schedule_disabled);
+
+/**
+ * drm_vblank_work_enable - enable vblank work
+ * @work: vblank work to enable
+ *
+ * This function is specifically only for when drm_vblank_work_schedule_disabled() is
+ * called. It allows for the work to be armed in any context, without any locks.
+ *
+ * The work will be signalled earliest at the @count argument, if it has been passed,
+ * it will signalled at the next vblank.
+ *
+ * This is particularly useful for PREEMPT_RT, where the spin_lock is converted
+ * into a sleeping rtmutex, and vblank evasion requires some work to be
+ * scheduled on completion with interrupts disabled.
+ */
+void drm_vblank_work_enable(struct drm_vblank_work *work)
+{
+	WARN_ON(work->armed);
+	WRITE_ONCE(work->armed, true);
+}
+EXPORT_SYMBOL(drm_vblank_work_enable);
+
 /**
  * drm_vblank_work_cancel_sync - cancel a vblank work and wait for it to
  * finish executing
diff --git a/include/drm/drm_vblank_work.h b/include/drm/drm_vblank_work.h
index e04d436b72973..e19351200da24 100644
--- a/include/drm/drm_vblank_work.h
+++ b/include/drm/drm_vblank_work.h
@@ -47,6 +47,14 @@ struct drm_vblank_work {
 	 */
 	int cancelling;
 
+	/**
+	 * @armed: If false, the work item has been added to the
+	 * drm_vblank_crtc.pending_work list, but will not yet be signalled.
+	 *
+	 * Call drm_vblank_work_enable() to fire on next vblank.
+	 */
+	bool armed;
+
 	/**
 	 * @node: The position of this work item in
 	 * &drm_vblank_crtc.pending_work.
@@ -64,6 +72,10 @@ struct drm_vblank_work {
 
 int drm_vblank_work_schedule(struct drm_vblank_work *work,
 			     u64 count, bool nextonmiss);
+
+int drm_vblank_work_schedule_disabled(struct drm_vblank_work *work, u64 count);
+void drm_vblank_work_enable(struct drm_vblank_work *work);
+
 void drm_vblank_work_init(struct drm_vblank_work *work, struct drm_crtc *crtc,
 			  void (*func)(struct kthread_work *work));
 bool drm_vblank_work_cancel_sync(struct drm_vblank_work *work);
-- 
2.51.0

[PATCH v7 02/26] drm/vblank: Add a 2-stage version of drm_crtc_arm_vblank_event

[Maarten Lankhorst]

When trying to fix the hardware programming in intel/display, I had
to take all the vblank locks with local_irqs_disabled(). This
required converting the entire vblank code to raw spinlocks.

In the alternative approach, do all preparations in advance, and only
enable the vblank_event with interrupts disabled, this requires only
a simple write and prevents a complete re-architecture of the code.

Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
---
 drivers/gpu/drm/drm_vblank.c | 61 +++++++++++++++++++++++++++++++++++-
 include/drm/drm_vblank.h     | 14 ++++++++-
 2 files changed, 73 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/drm_vblank.c b/drivers/gpu/drm/drm_vblank.c
index f78bf37f1e0a7..388d94e2e961f 100644
--- a/drivers/gpu/drm/drm_vblank.c
+++ b/drivers/gpu/drm/drm_vblank.c
@@ -1118,12 +1118,68 @@ void drm_crtc_arm_vblank_event(struct drm_crtc *crtc,
 
 	assert_spin_locked(&dev->event_lock);
 
+	WARN_ON(e->postponed);
 	e->pipe = pipe;
 	e->sequence = drm_crtc_accurate_vblank_count(crtc) + 1;
 	list_add_tail(&e->base.link, &dev->vblank_event_list);
 }
 EXPORT_SYMBOL(drm_crtc_arm_vblank_event);
 
+/**
+ * drm_crtc_prepare_arm_vblank_event - arm vblank event *before* pageflip.
+ * @crtc: the source CRTC of the vblank event
+ * @e: the event to send
+ *
+ * See drm_crtc_arm_vblank_event(). This function is a 2-stage version of
+ * that call. This function is called *BEFORE* programming the hardware.
+ *
+ * After programming, call drm_crtc_arm_prepared_vblank_event() and the
+ * event will be scheduled on the next vblank.
+ *
+ * This is mainly useful for code that has to run on PREEMPT_RT kernels,
+ * with interrupts disabled, since all vblank spinlocks are converted to
+ * rtmutexes, and code running with irqs disabled cannot take any vblank lock.
+ *
+ * It also increases determinism for any hardware
+ * programming, since no vblank related locks are taking when arming.
+ */
+void drm_crtc_prepare_arm_vblank_event(struct drm_crtc *crtc,
+				       struct drm_pending_vblank_event *e)
+{
+	drm_crtc_arm_vblank_event(crtc, e);
+
+	/* Set the flag, so that the event is not fired yet */
+	e->postponed = true;
+}
+EXPORT_SYMBOL(drm_crtc_prepare_arm_vblank_event);
+
+/**
+ * drm_crtc_arm_prepared_vblank_event - arm prepared vblank event *after* pageflip.
+ * @crtc: the source CRTC of the vblank event
+ * @e: the event to send
+ *
+ * See drm_crtc_prepare_arm_vblank_event(). This function is a 2-stage version of
+ * that call. This function is called directly *AFTER* programming the hardware.
+ *
+ * Before this function is called, drm_crtc_prepare_arm_vblank_event() should be
+ * called instead.
+ *
+ * This is mainly useful for code that has to run on PREEMPT_RT kernels,
+ * with interrupts disabled, since all vblank spinlocks are converted to
+ * rtmutexes, and code running with irqs disabled cannot take any vblank lock.
+ *
+ * It also increases determinism for any hardware
+ * programming, since no vblank related locks are taking when arming.
+ */
+void drm_crtc_arm_prepared_vblank_event(struct drm_pending_vblank_event *e)
+{
+	WARN_ON(!e->postponed);
+
+	/* remove the flag to be processed as a normal event */
+	WRITE_ONCE(e->postponed, false);
+}
+EXPORT_SYMBOL(drm_crtc_arm_prepared_vblank_event);
+
 /**
  * drm_crtc_send_vblank_event - helper to send vblank event after pageflip
  * @crtc: the source CRTC of the vblank event
@@ -1381,6 +1437,8 @@ void drm_crtc_vblank_off(struct drm_crtc *crtc)
 	list_for_each_entry_safe(e, t, &dev->vblank_event_list, base.link) {
 		if (e->pipe != pipe)
 			continue;
+
+		WARN_ON(e->postponed);
 		drm_dbg_core(dev, "Sending premature vblank event on disable: "
 			     "wanted %llu, current %llu\n",
 			     e->sequence, seq);
@@ -1886,7 +1944,8 @@ static void drm_handle_vblank_events(struct drm_device *dev, unsigned int pipe)
 	seq = drm_vblank_count_and_time(dev, pipe, &now);
 
 	list_for_each_entry_safe(e, t, &dev->vblank_event_list, base.link) {
-		if (e->pipe != pipe)
+		/* Matches WRITE_ONCE in drm_crtc_arm_prepared_vblank_event() */
+		if (e->pipe != pipe || READ_ONCE(e->postponed))
 			continue;
 		if (!drm_vblank_passed(seq, e->sequence))
 			continue;
diff --git a/include/drm/drm_vblank.h b/include/drm/drm_vblank.h
index 2fcef9c0f5b1b..956d5621eb7f9 100644
--- a/include/drm/drm_vblank.h
+++ b/include/drm/drm_vblank.h
@@ -53,6 +53,13 @@ struct drm_pending_vblank_event {
 	 * @sequence: frame event should be triggered at
 	 */
 	u64 sequence;
+
+	/**
+	 * @postponed: whether drm_crtc_prepare_arm_vblank_event() is called,
+	 * and drm_crtc_arm_prepared_vblank_event has yet to be called to arm.
+	 */
+	bool postponed;
+
 	/**
 	 * @event: Actual event which will be sent to userspace.
 	 */
@@ -294,7 +301,12 @@ int drm_crtc_next_vblank_start(struct drm_crtc *crtc, ktime_t *vblanktime);
 void drm_crtc_send_vblank_event(struct drm_crtc *crtc,
 			       struct drm_pending_vblank_event *e);
 void drm_crtc_arm_vblank_event(struct drm_crtc *crtc,
-			      struct drm_pending_vblank_event *e);
+			       struct drm_pending_vblank_event *e);
+
+void drm_crtc_prepare_arm_vblank_event(struct drm_crtc *crtc,
+				       struct drm_pending_vblank_event *e);
+void drm_crtc_arm_prepared_vblank_event(struct drm_pending_vblank_event *e);
+
 void drm_vblank_set_event(struct drm_pending_vblank_event *e,
 			  u64 *seq,
 			  ktime_t *now);
-- 
2.51.0

[PATCH v7 03/26] drm/intel/display: Make intel_crtc_arm_vblank_event static

[Maarten Lankhorst]

Only used inside intel_crtc.c now, so no need to export it any more.

Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
---
 drivers/gpu/drm/i915/display/intel_crtc.c | 2 +-
 drivers/gpu/drm/i915/display/intel_crtc.h | 1 -
 2 files changed, 1 insertion(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/i915/display/intel_crtc.c b/drivers/gpu/drm/i915/display/intel_crtc.c
index b8189cd5d864a..296f7a7b962fa 100644
--- a/drivers/gpu/drm/i915/display/intel_crtc.c
+++ b/drivers/gpu/drm/i915/display/intel_crtc.c
@@ -638,7 +638,7 @@ static void dbg_vblank_evade(struct intel_crtc *crtc, ktime_t end)
 static void dbg_vblank_evade(struct intel_crtc *crtc, ktime_t end) {}
 #endif
 
-void intel_crtc_arm_vblank_event(struct intel_crtc_state *crtc_state)
+static void intel_crtc_arm_vblank_event(struct intel_crtc_state *crtc_state)
 {
 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
 	unsigned long irqflags;
diff --git a/drivers/gpu/drm/i915/display/intel_crtc.h b/drivers/gpu/drm/i915/display/intel_crtc.h
index 12507b51ee77e..f65cbafe2b42a 100644
--- a/drivers/gpu/drm/i915/display/intel_crtc.h
+++ b/drivers/gpu/drm/i915/display/intel_crtc.h
@@ -33,7 +33,6 @@ int intel_usecs_to_scanlines(const struct drm_display_mode *adjusted_mode,
 			     int usecs);
 int intel_scanlines_to_usecs(const struct drm_display_mode *adjusted_mode,
 			     int scanlines);
-void intel_crtc_arm_vblank_event(struct intel_crtc_state *crtc_state);
 void intel_crtc_prepare_vblank_event(struct intel_crtc_state *crtc_state,
 				     struct drm_pending_vblank_event **event);
 u32 intel_crtc_max_vblank_count(const struct intel_crtc_state *crtc_state);
-- 
2.51.0

[PATCH v7 04/26] drm/intel/display: Convert vblank event handling to 2-stage arming

[Maarten Lankhorst]

This is converts the vblank functions to be called with interrupts
disabled, even on PREEMPT_RT kernels.

Because the PREEMPT_RT kernel converts all spinlocks to rt-mutexes,
the normal vblank functions cannot be used inside the critical section.

Instead, prepare the vblank at the start, and then enable the vblank
work after the hardware programming is completed.

This allows us to keep programming the hardware with interrupts
disabled, and still schedule completion on PREEMPT_RT on next vblank.

Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
---
 drivers/gpu/drm/i915/display/intel_crtc.c | 84 ++++++++++++-----------
 1 file changed, 44 insertions(+), 40 deletions(-)

diff --git a/drivers/gpu/drm/i915/display/intel_crtc.c b/drivers/gpu/drm/i915/display/intel_crtc.c
index 296f7a7b962fa..ca85b6fe50c6f 100644
--- a/drivers/gpu/drm/i915/display/intel_crtc.c
+++ b/drivers/gpu/drm/i915/display/intel_crtc.c
@@ -480,6 +480,10 @@ static void intel_crtc_vblank_work_init(struct intel_crtc_state *crtc_state)
 
 	drm_vblank_work_init(&crtc_state->vblank_work, &crtc->base,
 			     intel_crtc_vblank_work);
+
+	drm_vblank_work_schedule_disabled(&crtc_state->vblank_work,
+					  drm_crtc_accurate_vblank_count(&crtc->base) + 1);
+
 	/*
 	 * Interrupt latency is critical for getting the vblank
 	 * work executed as early as possible during the vblank.
@@ -525,6 +529,21 @@ int intel_scanlines_to_usecs(const struct drm_display_mode *adjusted_mode,
 				adjusted_mode->crtc_clock);
 }
 
+static void intel_crtc_arm_vblank_event(struct intel_crtc_state *crtc_state)
+{
+	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
+	unsigned long irqflags;
+
+	if (!crtc_state->uapi.event)
+		return;
+
+	drm_WARN_ON(crtc->base.dev, drm_crtc_vblank_get(&crtc->base) != 0);
+
+	spin_lock_irqsave(&crtc->base.dev->event_lock, irqflags);
+	drm_crtc_prepare_arm_vblank_event(&crtc->base, crtc_state->uapi.event);
+	spin_unlock_irqrestore(&crtc->base.dev->event_lock, irqflags);
+}
+
 /**
  * intel_pipe_update_start() - start update of a set of display registers
  * @state: the atomic state
@@ -561,6 +580,8 @@ void intel_pipe_update_start(struct intel_atomic_state *state,
 
 	if (intel_crtc_needs_vblank_work(new_crtc_state))
 		intel_crtc_vblank_work_init(new_crtc_state);
+	else
+		intel_crtc_arm_vblank_event(new_crtc_state);
 
 	if (state->base.legacy_cursor_update) {
 		struct intel_plane *plane;
@@ -638,23 +659,6 @@ static void dbg_vblank_evade(struct intel_crtc *crtc, ktime_t end)
 static void dbg_vblank_evade(struct intel_crtc *crtc, ktime_t end) {}
 #endif
 
-static void intel_crtc_arm_vblank_event(struct intel_crtc_state *crtc_state)
-{
-	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
-	unsigned long irqflags;
-
-	if (!crtc_state->uapi.event)
-		return;
-
-	drm_WARN_ON(crtc->base.dev, drm_crtc_vblank_get(&crtc->base) != 0);
-
-	spin_lock_irqsave(&crtc->base.dev->event_lock, irqflags);
-	drm_crtc_arm_vblank_event(&crtc->base, crtc_state->uapi.event);
-	spin_unlock_irqrestore(&crtc->base.dev->event_lock, irqflags);
-
-	crtc_state->uapi.event = NULL;
-}
-
 void intel_crtc_prepare_vblank_event(struct intel_crtc_state *crtc_state,
 				     struct drm_pending_vblank_event **event)
 {
@@ -708,29 +712,10 @@ void intel_pipe_update_end(struct intel_atomic_state *state,
 	 * event outside of the critical section - the spinlock might spin for a
 	 * while ... */
 	if (intel_crtc_needs_vblank_work(new_crtc_state)) {
-		drm_vblank_work_schedule(&new_crtc_state->vblank_work,
-					 drm_crtc_accurate_vblank_count(&crtc->base) + 1,
-					 false);
-	} else {
-		intel_crtc_arm_vblank_event(new_crtc_state);
-	}
-
-	if (state->base.legacy_cursor_update) {
-		struct intel_plane *plane;
-		struct intel_plane_state *old_plane_state;
-		int i;
-
-		for_each_old_intel_plane_in_state(state, plane, old_plane_state, i) {
-			if (old_plane_state->hw.crtc == &crtc->base &&
-			    old_plane_state->unpin_work.vblank) {
-				drm_vblank_work_schedule(&old_plane_state->unpin_work,
-							 drm_crtc_accurate_vblank_count(&crtc->base) + 1,
-							 false);
-
-				/* Remove plane from atomic state, cleanup/free is done from vblank worker. */
-				memset(&state->base.planes[i], 0, sizeof(state->base.planes[i]));
-			}
-		}
+		drm_vblank_work_enable(&new_crtc_state->vblank_work);
+	} else if (new_crtc_state->uapi.event) {
+		drm_crtc_arm_prepared_vblank_event(new_crtc_state->uapi.event);
+		new_crtc_state->uapi.event = NULL;
 	}
 
 	/*
@@ -754,6 +739,25 @@ void intel_pipe_update_end(struct intel_atomic_state *state,
 
 	local_irq_enable();
 
+	/* Run after local_irq_enable(), not timing sensitive */
+	if (state->base.legacy_cursor_update) {
+		struct intel_plane *plane;
+		struct intel_plane_state *old_plane_state;
+		int i;
+
+		for_each_old_intel_plane_in_state(state, plane, old_plane_state, i) {
+			if (old_plane_state->hw.crtc == &crtc->base &&
+			    old_plane_state->unpin_work.vblank) {
+				drm_vblank_work_schedule(&old_plane_state->unpin_work,
+							 drm_crtc_accurate_vblank_count(&crtc->base) + 1,
+							 false);
+
+				/* Remove plane from atomic state, cleanup/free is done from vblank worker. */
+				memset(&state->base.planes[i], 0, sizeof(state->base.planes[i]));
+			}
+		}
+	}
+
 	if (intel_parent_vgpu_active(display))
 		goto out;
 
-- 
2.51.0

[PATCH v7 05/26] drm/i915/display: Move vblank put until after critical section

[Maarten Lankhorst]

drm_crtc_vblank_put may take some locks, this should probably
not be the first thing we do after entering the time sensitive
part.

A better place is after programming is completed. Add a flag
to put the vblank after completion.

In the case of drm_vblank_work_schedule, we may not even need
to disable the vblank interrupt any more if it takes its own
reference.

Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
Reviewed-by: Uma Shankar <uma.shankar@intel.com>
---
 drivers/gpu/drm/i915/display/intel_cursor.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/i915/display/intel_cursor.c b/drivers/gpu/drm/i915/display/intel_cursor.c
index 18d1014de3613..5248e478b64e8 100644
--- a/drivers/gpu/drm/i915/display/intel_cursor.c
+++ b/drivers/gpu/drm/i915/display/intel_cursor.c
@@ -817,6 +817,7 @@ intel_legacy_cursor_update(struct drm_plane *_plane,
 		to_intel_crtc_state(crtc->base.state);
 	struct intel_crtc_state *new_crtc_state;
 	struct intel_vblank_evade_ctx evade;
+	bool has_vblank = false;
 	int ret;
 
 	/*
@@ -914,6 +915,8 @@ intel_legacy_cursor_update(struct drm_plane *_plane,
 	intel_psr_lock(crtc_state);
 
 	if (!drm_WARN_ON(display->drm, drm_crtc_vblank_get(&crtc->base))) {
+		has_vblank = true;
+
 		/*
 		 * TODO: maybe check if we're still in PSR
 		 * and skip the vblank evasion entirely?
@@ -923,8 +926,6 @@ intel_legacy_cursor_update(struct drm_plane *_plane,
 		local_irq_disable();
 
 		intel_vblank_evade(&evade);
-
-		drm_crtc_vblank_put(&crtc->base);
 	} else {
 		local_irq_disable();
 	}
@@ -940,6 +941,9 @@ intel_legacy_cursor_update(struct drm_plane *_plane,
 
 	intel_psr_unlock(crtc_state);
 
+	if (has_vblank)
+		drm_crtc_vblank_put(&crtc->base);
+
 	if (old_plane_state->ggtt_vma != new_plane_state->ggtt_vma) {
 		drm_vblank_work_init(&old_plane_state->unpin_work, &crtc->base,
 				     intel_cursor_unpin_work);
-- 
2.51.0

[PATCH v7 06/26] drm/i915/display: Remove locking from intel_vblank_evade critical section

[Maarten Lankhorst]

finish_wait() may take a lock, which means that it can take any amount
of time. On PREEMPT-RT we should not be taking any lock after disabling
preemption, so ensure that the completion is done before disabling
interrupts.

This also has the benefit of making vblank evasion more deterministic,
by performing the final vblank check after all locking is done.

Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
---
 drivers/gpu/drm/i915/display/intel_crtc.c   |  2 +-
 drivers/gpu/drm/i915/display/intel_vblank.c | 30 +++++++++------------
 drivers/gpu/drm/i915/display/intel_vblank.h |  1 +
 3 files changed, 15 insertions(+), 18 deletions(-)

diff --git a/drivers/gpu/drm/i915/display/intel_crtc.c b/drivers/gpu/drm/i915/display/intel_crtc.c
index ca85b6fe50c6f..1022c2cf4eec2 100644
--- a/drivers/gpu/drm/i915/display/intel_crtc.c
+++ b/drivers/gpu/drm/i915/display/intel_crtc.c
@@ -688,7 +688,7 @@ void intel_pipe_update_end(struct intel_atomic_state *state,
 	struct intel_crtc_state *new_crtc_state =
 		intel_atomic_get_new_crtc_state(state, crtc);
 	enum pipe pipe = crtc->pipe;
-	int scanline_end = intel_get_crtc_scanline(crtc);
+	int scanline_end = __intel_get_crtc_scanline(crtc);
 	u32 end_vbl_count = intel_crtc_get_vblank_counter(crtc);
 	ktime_t end_vbl_time = ktime_get();
 
diff --git a/drivers/gpu/drm/i915/display/intel_vblank.c b/drivers/gpu/drm/i915/display/intel_vblank.c
index 1b7cfe226ff8f..73676d8ccec8b 100644
--- a/drivers/gpu/drm/i915/display/intel_vblank.c
+++ b/drivers/gpu/drm/i915/display/intel_vblank.c
@@ -241,7 +241,7 @@ int intel_crtc_scanline_offset(const struct intel_crtc_state *crtc_state)
  * intel_de_read_fw(), only for fast reads of display block, no need for
  * forcewake etc.
  */
-static int __intel_get_crtc_scanline(struct intel_crtc *crtc)
+int __intel_get_crtc_scanline(struct intel_crtc *crtc)
 {
 	struct intel_display *display = to_intel_display(crtc);
 	struct drm_vblank_crtc *vblank = drm_crtc_vblank_crtc(&crtc->base);
@@ -732,6 +732,16 @@ void intel_vblank_evade_init(const struct intel_crtc_state *old_crtc_state,
 		evade->min -= vblank_delay;
 }
 
+static bool scanline_in_safe_range(struct intel_vblank_evade_ctx *evade, int *scanline, bool unlocked)
+{
+	if (unlocked)
+		*scanline = intel_get_crtc_scanline(evade->crtc);
+	else
+		*scanline = __intel_get_crtc_scanline(evade->crtc);
+
+	return *scanline < evade->min || *scanline > evade->max;
+}
+
 /* must be called with vblank interrupt already enabled! */
 int intel_vblank_evade(struct intel_vblank_evade_ctx *evade)
 {
@@ -739,24 +749,12 @@ int intel_vblank_evade(struct intel_vblank_evade_ctx *evade)
 	struct intel_display *display = to_intel_display(crtc);
 	long timeout = msecs_to_jiffies_timeout(1);
 	wait_queue_head_t *wq = drm_crtc_vblank_waitqueue(&crtc->base);
-	DEFINE_WAIT(wait);
 	int scanline;
 
 	if (evade->min <= 0 || evade->max <= 0)
 		return 0;
 
-	for (;;) {
-		/*
-		 * prepare_to_wait() has a memory barrier, which guarantees
-		 * other CPUs can see the task state update by the time we
-		 * read the scanline.
-		 */
-		prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
-
-		scanline = intel_get_crtc_scanline(crtc);
-		if (scanline < evade->min || scanline > evade->max)
-			break;
-
+	while (!scanline_in_safe_range(evade, &scanline, false)) {
 		if (!timeout) {
 			drm_dbg_kms(display->drm,
 				    "Potential atomic update failure on pipe %c\n",
@@ -766,13 +764,11 @@ int intel_vblank_evade(struct intel_vblank_evade_ctx *evade)
 
 		local_irq_enable();
 
-		timeout = schedule_timeout(timeout);
+		timeout = wait_event_timeout(*wq, scanline_in_safe_range(evade, &scanline, true), timeout);
 
 		local_irq_disable();
 	}
 
-	finish_wait(wq, &wait);
-
 	/*
 	 * On VLV/CHV DSI the scanline counter would appear to
 	 * increment approx. 1/3 of a scanline before start of vblank.
diff --git a/drivers/gpu/drm/i915/display/intel_vblank.h b/drivers/gpu/drm/i915/display/intel_vblank.h
index 98d04cacd65f8..aa1974400e9fc 100644
--- a/drivers/gpu/drm/i915/display/intel_vblank.h
+++ b/drivers/gpu/drm/i915/display/intel_vblank.h
@@ -38,6 +38,7 @@ u32 g4x_get_vblank_counter(struct drm_crtc *crtc);
 bool intel_crtc_get_vblank_timestamp(struct drm_crtc *crtc, int *max_error,
 				     ktime_t *vblank_time, bool in_vblank_irq);
 int intel_get_crtc_scanline(struct intel_crtc *crtc);
+int __intel_get_crtc_scanline(struct intel_crtc *crtc);
 void intel_wait_for_pipe_scanline_stopped(struct intel_crtc *crtc);
 void intel_wait_for_pipe_scanline_moving(struct intel_crtc *crtc);
 void intel_crtc_update_active_timings(const struct intel_crtc_state *crtc_state,
-- 
2.51.0

[PATCH v7 07/26] drm/i915/display: Handle vlv dsi workaround in scanline_in_safe_range too

[Maarten Lankhorst]

Now that we have a macro, might as well handle the VLV dsi workaround
too.

This makes the vblank evasion code slightly more deterministic, by not
looping with interrupts disabled.

Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
Reviewed-by: Uma Shankar <uma.shankar@intel.com>
---
 drivers/gpu/drm/i915/display/intel_vblank.c | 36 ++++++++++-----------
 1 file changed, 18 insertions(+), 18 deletions(-)

diff --git a/drivers/gpu/drm/i915/display/intel_vblank.c b/drivers/gpu/drm/i915/display/intel_vblank.c
index 73676d8ccec8b..3eb6418723a84 100644
--- a/drivers/gpu/drm/i915/display/intel_vblank.c
+++ b/drivers/gpu/drm/i915/display/intel_vblank.c
@@ -739,6 +739,24 @@ static bool scanline_in_safe_range(struct intel_vblank_evade_ctx *evade, int *sc
 	else
 		*scanline = __intel_get_crtc_scanline(evade->crtc);
 
+	/*
+	 * On VLV/CHV DSI the scanline counter would appear to
+	 * increment approx. 1/3 of a scanline before start of vblank.
+	 * The registers still get latched at start of vblank however.
+	 * This means we must not write any registers on the first
+	 * line of vblank (since not the whole line is actually in
+	 * vblank). And unfortunately we can't use the interrupt to
+	 * wait here since it will fire too soon. We could use the
+	 * frame start interrupt instead since it will fire after the
+	 * critical scanline, but that would require more changes
+	 * in the interrupt code. So for now we'll just do the nasty
+	 * thing and poll for the bad scanline to pass us by.
+	 *
+	 * FIXME figure out if BXT+ DSI suffers from this as well
+	 */
+	if (evade->need_vlv_dsi_wa && *scanline == evade->vblank_start)
+		return false;
+
 	return *scanline < evade->min || *scanline > evade->max;
 }
 
@@ -769,24 +787,6 @@ int intel_vblank_evade(struct intel_vblank_evade_ctx *evade)
 		local_irq_disable();
 	}
 
-	/*
-	 * On VLV/CHV DSI the scanline counter would appear to
-	 * increment approx. 1/3 of a scanline before start of vblank.
-	 * The registers still get latched at start of vblank however.
-	 * This means we must not write any registers on the first
-	 * line of vblank (since not the whole line is actually in
-	 * vblank). And unfortunately we can't use the interrupt to
-	 * wait here since it will fire too soon. We could use the
-	 * frame start interrupt instead since it will fire after the
-	 * critical scanline, but that would require more changes
-	 * in the interrupt code. So for now we'll just do the nasty
-	 * thing and poll for the bad scanline to pass us by.
-	 *
-	 * FIXME figure out if BXT+ DSI suffers from this as well
-	 */
-	while (evade->need_vlv_dsi_wa && scanline == evade->vblank_start)
-		scanline = intel_get_crtc_scanline(crtc);
-
 	return scanline;
 }
 
-- 
2.51.0

[PATCH v7 08/26] drm/i915: Use preempt_disable/enable_rt() where recommended

[Maarten Lankhorst]

From: Mike Galbraith <umgwanakikbuti@gmail.com>

Mario Kleiner suggest in commit
  ad3543ede630f ("drm/intel: Push get_scanout_position() timestamping into kms driver.")

a spots where preemption should be disabled on PREEMPT_RT. The
difference is that on PREEMPT_RT the intel_uncore::lock disables neither
preemption nor interrupts and so region remains preemptible.

The area covers only register reads and writes. The part that worries me
is:
- __intel_get_crtc_scanline() the worst case is 100us if no match is
  found.

- intel_crtc_scanlines_since_frame_timestamp() not sure how long this
  may take in the worst case.

It was in the RT queue for a while and nobody complained.
Disable preemption on PREEPMPT_RT during timestamping.

[bigeasy: patch description.]

Cc: Mario Kleiner <mario.kleiner.de@gmail.com>
Signed-off-by: Mike Galbraith <umgwanakikbuti@gmail.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
Reviewed-by: Uma Shankar <uma.shankar@intel.com>
---
 drivers/gpu/drm/i915/display/intel_vblank.c | 43 ++++++++++++++++-----
 1 file changed, 33 insertions(+), 10 deletions(-)

diff --git a/drivers/gpu/drm/i915/display/intel_vblank.c b/drivers/gpu/drm/i915/display/intel_vblank.c
index 3eb6418723a84..8263cefaa0fb6 100644
--- a/drivers/gpu/drm/i915/display/intel_vblank.c
+++ b/drivers/gpu/drm/i915/display/intel_vblank.c
@@ -316,6 +316,20 @@ static void intel_vblank_section_exit(struct intel_display *display)
 	struct intel_uncore *uncore = to_intel_uncore(display->drm);
 	spin_unlock(&uncore->lock);
 }
+
+static void intel_vblank_section_enter_irqf(struct intel_display *display, unsigned long *flags)
+	__acquires(i915->uncore.lock)
+{
+	struct intel_uncore *uncore = to_intel_uncore(display->drm);
+	spin_lock_irqsave(&uncore->lock, *flags);
+}
+
+static void intel_vblank_section_exit_irqf(struct intel_display *display, unsigned long flags)
+	__releases(i915->uncore.lock)
+{
+	struct intel_uncore *uncore = to_intel_uncore(display->drm);
+	spin_unlock_irqrestore(&uncore->lock, flags);
+}
 #else
 static void intel_vblank_section_enter(struct intel_display *display)
 {
@@ -324,6 +338,17 @@ static void intel_vblank_section_enter(struct intel_display *display)
 static void intel_vblank_section_exit(struct intel_display *display)
 {
 }
+
+static void intel_vblank_section_enter_irqf(struct intel_display *display, unsigned long *flags)
+{
+	*flags = 0;
+}
+
+static void intel_vblank_section_exit_irqf(struct intel_display *display, unsigned long flags)
+{
+	if (flags)
+		return;
+}
 #endif
 
 static bool i915_get_crtc_scanoutpos(struct drm_crtc *_crtc,
@@ -360,10 +385,10 @@ static bool i915_get_crtc_scanoutpos(struct drm_crtc *_crtc,
 	 * timing critical raw register reads, potentially with
 	 * preemption disabled, so the following code must not block.
 	 */
-	local_irq_save(irqflags);
-	intel_vblank_section_enter(display);
+	intel_vblank_section_enter_irqf(display, &irqflags);
 
-	/* preempt_disable_rt() should go right here in PREEMPT_RT patchset. */
+	if (IS_ENABLED(CONFIG_PREEMPT_RT))
+		preempt_disable();
 
 	/* Get optional system timestamp before query. */
 	if (stime)
@@ -427,10 +452,10 @@ static bool i915_get_crtc_scanoutpos(struct drm_crtc *_crtc,
 	if (etime)
 		*etime = ktime_get();
 
-	/* preempt_enable_rt() should go right here in PREEMPT_RT patchset. */
+	if (IS_ENABLED(CONFIG_PREEMPT_RT))
+		preempt_enable();
 
-	intel_vblank_section_exit(display);
-	local_irq_restore(irqflags);
+	intel_vblank_section_exit_irqf(display, irqflags);
 
 	/*
 	 * While in vblank, position will be negative
@@ -468,13 +493,11 @@ int intel_get_crtc_scanline(struct intel_crtc *crtc)
 	unsigned long irqflags;
 	int position;
 
-	local_irq_save(irqflags);
-	intel_vblank_section_enter(display);
+	intel_vblank_section_enter_irqf(display, &irqflags);
 
 	position = __intel_get_crtc_scanline(crtc);
 
-	intel_vblank_section_exit(display);
-	local_irq_restore(irqflags);
+	intel_vblank_section_exit_irqf(display, irqflags);
 
 	return position;
 }
-- 
2.51.0

[PATCH v7 09/26] drm/i915/display: Make get_vblank_counter use intel_de_read_fw()

[Maarten Lankhorst]

Fixes the following lockdep splat on PREEMPT_RT:
<3> BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:48
<3> in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 1373, name: xe_module_load
<3> preempt_count: 1, expected: 0
<3> RCU nest depth: 0, expected: 0
<4> 11 locks held by xe_module_load/1373:
<4>  #0: ffff888107b691a0 (&dev->mutex){....}-{3:3}, at: __driver_attach+0x104/0x220
<4>  #1: ffff88813cd30280 (&dev->clientlist_mutex){+.+.}-{3:3}, at: drm_client_register+0x32/0xe0
<4>  #2: ffffffff837f88f8 (registration_lock){+.+.}-{3:3}, at: register_framebuffer+0x1b/0x50
<4>  #3: ffffffff835985e0 (console_lock){+.+.}-{0:0}, at: fbcon_fb_registered+0x6f/0x90
<4>  #4: ffff88812589e6a0 (&helper->lock){+.+.}-{3:3}, at: __drm_fb_helper_restore_fbdev_mode_unlocked+0x7b/0x110
<4>  #5: ffff88813cd30158 (&dev->master_mutex){+.+.}-{3:3}, at: drm_master_internal_acquire+0x20/0x50
<4>  #6: ffff88812589e488 (&client->modeset_mutex){+.+.}-{3:3}, at: drm_client_modeset_commit_locked+0x2a/0x1b0
<4>  #7: ffffc9000031eef0 (crtc_ww_class_acquire){+.+.}-{0:0}, at: drm_client_modeset_commit_atomic+0x4c/0x2b0
<4>  #8: ffffc9000031ef18 (crtc_ww_class_mutex){+.+.}-{3:3}, at: drm_client_modeset_commit_atomic+0x4c/0x2b0
<4>  #9: ffff888114f7b8b8 (&intel_dp->psr.lock){+.+.}-{3:3}, at: intel_psr_lock+0xc5/0xf0 [xe]
<4>  #10: ffff88812a0cbbc0 (&wl->lock){+.+.}-{2:2}, at: intel_dmc_wl_get+0x3c/0x140 [xe]

This splat will happen otherwise on all tracepoints too, for similar reasons.

Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
---
 drivers/gpu/drm/i915/display/intel_vblank.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/display/intel_vblank.c b/drivers/gpu/drm/i915/display/intel_vblank.c
index 8263cefaa0fb6..e204c260b9aef 100644
--- a/drivers/gpu/drm/i915/display/intel_vblank.c
+++ b/drivers/gpu/drm/i915/display/intel_vblank.c
@@ -132,7 +132,7 @@ u32 g4x_get_vblank_counter(struct drm_crtc *crtc)
 	if (!vblank->max_vblank_count)
 		return 0;
 
-	return intel_de_read(display, PIPE_FRMCOUNT_G4X(display, pipe));
+	return intel_de_read_fw(display, PIPE_FRMCOUNT_G4X(display, pipe));
 }
 
 static u32 intel_crtc_scanlines_since_frame_timestamp(struct intel_crtc *crtc)
-- 
2.51.0

[PATCH v7 10/26] drm/i915/display: Do not take uncore lock in i915_get_vblank_counter

[Maarten Lankhorst]

This fixes a lockdep splat that occurs in the code that should be run
with interrupts disabled. The uncore lock should not be taken and
released repeatedly in a timing sensitive path.

Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
---
 drivers/gpu/drm/i915/display/intel_de.h       |  8 ++++++
 drivers/gpu/drm/i915/display/intel_vblank.c   |  4 +--
 drivers/gpu/drm/i915/intel_uncore.h           | 26 +++++++++++++------
 .../drm/xe/compat-i915-headers/intel_uncore.h |  7 +++++
 4 files changed, 35 insertions(+), 10 deletions(-)

diff --git a/drivers/gpu/drm/i915/display/intel_de.h b/drivers/gpu/drm/i915/display/intel_de.h
index f30f3f8ebee1d..8990c2c778efe 100644
--- a/drivers/gpu/drm/i915/display/intel_de.h
+++ b/drivers/gpu/drm/i915/display/intel_de.h
@@ -68,6 +68,14 @@ intel_de_read64_2x32(struct intel_display *display,
 	return val;
 }
 
+static inline u64
+intel_de_read64_2x32_fw(struct intel_display *display,
+			i915_reg_t lower_reg, i915_reg_t upper_reg)
+{
+	return intel_uncore_read64_2x32_fw(__to_uncore(display),
+					   lower_reg, upper_reg);
+}
+
 static inline void
 intel_de_posting_read(struct intel_display *display, i915_reg_t reg)
 {
diff --git a/drivers/gpu/drm/i915/display/intel_vblank.c b/drivers/gpu/drm/i915/display/intel_vblank.c
index e204c260b9aef..749127ae7794c 100644
--- a/drivers/gpu/drm/i915/display/intel_vblank.c
+++ b/drivers/gpu/drm/i915/display/intel_vblank.c
@@ -109,8 +109,8 @@ u32 i915_get_vblank_counter(struct drm_crtc *crtc)
 	 * we get a low value that's stable across two reads of the high
 	 * register.
 	 */
-	frame = intel_de_read64_2x32(display, PIPEFRAMEPIXEL(display, pipe),
-				     PIPEFRAME(display, pipe));
+	frame = intel_de_read64_2x32_fw(display, PIPEFRAMEPIXEL(display, pipe),
+				        PIPEFRAME(display, pipe));
 
 	pixel = frame & PIPE_PIXEL_MASK;
 	frame = (frame >> PIPE_FRAME_LOW_SHIFT) & 0xffffff;
diff --git a/drivers/gpu/drm/i915/intel_uncore.h b/drivers/gpu/drm/i915/intel_uncore.h
index fafc2ca9a2376..507398a562649 100644
--- a/drivers/gpu/drm/i915/intel_uncore.h
+++ b/drivers/gpu/drm/i915/intel_uncore.h
@@ -449,13 +449,28 @@ static inline void intel_uncore_rmw_fw(struct intel_uncore *uncore,
 		intel_uncore_write_fw(uncore, reg, val);
 }
 
+static inline u64
+intel_uncore_read64_2x32_fw(struct intel_uncore *uncore,
+			    i915_reg_t lower_reg, i915_reg_t upper_reg)
+{
+	u32 upper, lower, old_upper, loop = 0;
+	upper = intel_uncore_read_fw(uncore, upper_reg);
+	do {
+		old_upper = upper;
+		lower = intel_uncore_read_fw(uncore, lower_reg);
+		upper = intel_uncore_read_fw(uncore, upper_reg);
+	} while (upper != old_upper && loop++ < 2);
+
+	return (u64)upper << 32 | lower;
+}
+
 static inline u64
 intel_uncore_read64_2x32(struct intel_uncore *uncore,
 			 i915_reg_t lower_reg, i915_reg_t upper_reg)
 {
-	u32 upper, lower, old_upper, loop = 0;
 	enum forcewake_domains fw_domains;
 	unsigned long flags;
+	u64 ret;
 
 	fw_domains = intel_uncore_forcewake_for_reg(uncore, lower_reg,
 						    FW_REG_READ);
@@ -466,17 +481,12 @@ intel_uncore_read64_2x32(struct intel_uncore *uncore,
 	spin_lock_irqsave(&uncore->lock, flags);
 	intel_uncore_forcewake_get__locked(uncore, fw_domains);
 
-	upper = intel_uncore_read_fw(uncore, upper_reg);
-	do {
-		old_upper = upper;
-		lower = intel_uncore_read_fw(uncore, lower_reg);
-		upper = intel_uncore_read_fw(uncore, upper_reg);
-	} while (upper != old_upper && loop++ < 2);
+	ret = intel_uncore_read64_2x32_fw(uncore, lower_reg, upper_reg);
 
 	intel_uncore_forcewake_put__locked(uncore, fw_domains);
 	spin_unlock_irqrestore(&uncore->lock, flags);
 
-	return (u64)upper << 32 | lower;
+	return ret;
 }
 
 static inline int intel_uncore_write_and_verify(struct intel_uncore *uncore,
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/intel_uncore.h b/drivers/gpu/drm/xe/compat-i915-headers/intel_uncore.h
index a8cfd65119e09..6959a761fa06a 100644
--- a/drivers/gpu/drm/xe/compat-i915-headers/intel_uncore.h
+++ b/drivers/gpu/drm/xe/compat-i915-headers/intel_uncore.h
@@ -74,6 +74,13 @@ intel_uncore_read64_2x32(struct intel_uncore *uncore,
 	return (u64)upper << 32 | lower;
 }
 
+static inline u64
+intel_uncore_read64_2x32_fw(struct intel_uncore *uncore,
+			    i915_reg_t i915_lower_reg, i915_reg_t i915_upper_reg)
+{
+	return intel_uncore_read64_2x32(uncore, i915_lower_reg, i915_upper_reg);
+}
+
 static inline void intel_uncore_posting_read(struct intel_uncore *uncore,
 					     i915_reg_t i915_reg)
 {
-- 
2.51.0

[PATCH v7 11/26] drm/i915/display: Make icl_dsi_frame_update use _fw too

[Maarten Lankhorst]

Don't use the dmc lock inside the vblank critical section,
not even as last call.

Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
Reviewed-by: Uma Shankar <uma.shankar@intel.com>
---
 drivers/gpu/drm/i915/display/icl_dsi.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/i915/display/icl_dsi.c b/drivers/gpu/drm/i915/display/icl_dsi.c
index c043279796781..ec4f5bb19450c 100644
--- a/drivers/gpu/drm/i915/display/icl_dsi.c
+++ b/drivers/gpu/drm/i915/display/icl_dsi.c
@@ -242,8 +242,8 @@ void icl_dsi_frame_update(struct intel_crtc_state *crtc_state)
 	else
 		return;
 
-	intel_de_rmw(display, DSI_CMD_FRMCTL(port), 0,
-		     DSI_FRAME_UPDATE_REQUEST);
+	intel_de_rmw_fw(display, DSI_CMD_FRMCTL(port), 0,
+			DSI_FRAME_UPDATE_REQUEST);
 }
 
 static void dsi_program_swing_and_deemphasis(struct intel_encoder *encoder)
-- 
2.51.0

[PATCH v7 12/26] drm/i915/display: Use intel_de_read/write_fw in colorops

[Maarten Lankhorst]

Using the _fw variants fixes a module load error on PTL:

<4> [363.690050] Call Trace:
<4> [363.690052]  <TASK>
<4> [363.690055]  dump_stack_lvl+0x91/0xf0
<4> [363.690067]  dump_stack+0x10/0x20
<4> [363.690074]  __might_resched+0x174/0x260
<4> [363.690087]  rt_spin_lock+0x63/0x200
<4> [363.690092]  ? intel_dmc_wl_get+0x3c/0x140 [xe]
<4> [363.690470]  ? __lock_acquire+0x1195/0x2860
<4> [363.690487]  intel_dmc_wl_get+0x3c/0x140 [xe]
<4> [363.690842]  intel_color_plane_commit_arm+0xbc/0x140 [xe]
<4> [363.691246]  icl_plane_update_arm+0x23f/0x280 [xe]
<4> [363.691605]  intel_plane_update_arm+0x74/0x170 [xe]
<4> [363.691970]  intel_crtc_planes_update_arm+0x3cb/0x4c0 [xe]
<4> [363.692338]  intel_update_crtc+0x1c3/0x840 [xe]
<4> [363.692742]  ? intel_pre_update_crtc+0x2ce/0x470 [xe]
<4> [363.693125]  ? intel_enable_crtc+0x123/0x150 [xe]
<4> [363.693508]  skl_commit_modeset_enables+0x4c4/0x720 [xe]
<4> [363.693888]  intel_atomic_commit_tail+0xd9d/0x1b30 [xe]
<4> [363.694274]  intel_atomic_commit+0x2e8/0x330 [xe]
<4> [363.694621]  ? intel_atomic_commit+0x2e8/0x330 [xe]
<4> [363.694956]  drm_atomic_commit+0xaf/0xf0
<4> [363.694962]  ? __pfx___drm_printfn_info+0x10/0x10
<4> [363.694978]  drm_client_modeset_commit_atomic+0x25c/0x2b0
<4> [363.695018]  drm_client_modeset_commit_locked+0x63/0x1b0
<4> [363.695029]  drm_client_modeset_commit+0x26/0x50
<4> [363.695035]  __drm_fb_helper_restore_fbdev_mode_unlocked+0xdc/0x110
<4> [363.695045]  drm_fb_helper_set_par+0x2f/0x50
<4> [363.695052]  intel_fbdev_set_par+0x39/0x90 [xe]
<4> [363.695365]  fbcon_init+0x283/0x680
<4> [363.695382]  visual_init+0xf2/0x190
<4> [363.695396]  do_bind_con_driver.isra.0+0x1f1/0x4c0
<4> [363.695416]  do_take_over_console+0x181/0x220
<4> [363.695422]  ? vprintk_default+0x1d/0x30
<4> [363.695436]  do_fbcon_takeover+0x85/0x160
<4> [363.695447]  do_fb_registered+0x24c/0x2b0
<4> [363.695460]  fbcon_fb_registered+0x3a/0x90
<4> [363.695469]  do_register_framebuffer+0x216/0x320
<4> [363.695488]  register_framebuffer+0x23/0x50
<4> [363.695494]  __drm_fb_helper_initial_config_and_unlock+0x3ea/0x670
<4> [363.695502]  ? trace_hardirqs_on+0x1e/0xd0
<4> [363.695526]  drm_fb_helper_initial_config+0x3f/0x50
<4> [363.695534]  drm_fbdev_client_hotplug+0x80/0xd0
<4> [363.695543]  drm_client_register+0x8a/0xe0
<4> [363.695556]  drm_fbdev_client_setup+0x127/0x1f0
<4> [363.695563]  drm_client_setup+0xa7/0xe0
<4> [363.695569]  drm_client_setup_with_color_mode+0x24/0x40
<4> [363.695575]  intel_fbdev_setup+0x1c6/0x510 [xe]
<4> [363.695857]  intel_display_driver_register+0xb5/0x100 [xe]
<4> [363.696188]  ? __pfx___drm_printfn_dbg+0x10/0x10
<4> [363.696194]  ? intel_display_driver_register+0x2e/0x100 [xe]
<4> [363.696515]  xe_display_register+0x29/0x40 [xe]
<4> [363.696858]  xe_device_probe+0x51a/0x9e0 [xe]
<4> [363.697102]  ? __drmm_add_action+0x98/0x110
<4> [363.697108]  ? __pfx___drmm_mutex_release+0x10/0x10
<4> [363.697116]  ? __drmm_add_action_or_reset+0x1e/0x50
<4> [363.697130]  xe_pci_probe+0x396/0x620 [xe]
<4> [363.697423]  local_pci_probe+0x47/0xb0
<4> [363.697431]  pci_device_probe+0xf3/0x260
<4> [363.697444]  really_probe+0xf1/0x3c0
<4> [363.697451]  __driver_probe_device+0x8c/0x180
<4> [363.697458]  driver_probe_device+0x24/0xd0
<4> [363.697464]  __driver_attach+0x10f/0x220
<4> [363.697468]  ? __pfx___driver_attach+0x10/0x10
<4> [363.697472]  bus_for_each_dev+0x7f/0xe0
<4> [363.697484]  driver_attach+0x1e/0x30
<4> [363.697487]  bus_add_driver+0x154/0x290
<4> [363.697498]  driver_register+0x5e/0x130
<4> [363.697504]  __pci_register_driver+0x84/0xa0
<4> [363.697509]  xe_register_pci_driver+0x23/0x30 [xe]
<4> [363.697762]  xe_init+0x2c/0x110 [xe]
<4> [363.698007]  ? __pfx_xe_init+0x10/0x10 [xe]
<4> [363.698239]  do_one_initcall+0x60/0x3f0
<4> [363.698250]  ? __kmalloc_cache_noprof+0x470/0x690
<4> [363.698267]  do_init_module+0x97/0x2b0
<4> [363.698275]  load_module+0x2d08/0x2e30
<4> [363.698280]  ? __kernel_read+0x164/0x310
<4> [363.698312]  ? kernel_read_file+0x2ca/0x340
<4> [363.698328]  init_module_from_file+0x96/0xe0
<4> [363.698330]  ? init_module_from_file+0x96/0xe0
<4> [363.698357]  idempotent_init_module+0x11d/0x340
<4> [363.698384]  __x64_sys_finit_module+0x73/0xe0
<4> [363.698393]  x64_sys_call+0x1e3d/0x26a0
<4> [363.698399]  do_syscall_64+0x93/0xab0
<4> [363.698413]  ? ext4_llseek+0xc3/0x130
<4> [363.698425]  ? ksys_lseek+0x55/0xd0
<4> [363.698438]  ? do_syscall_64+0x1b7/0xab0
<4> [363.698444]  ? switch_fpu_return+0xea/0x150
<4> [363.698454]  ? do_syscall_64+0x1b7/0xab0
<4> [363.698465]  ? kernfs_fop_read_iter+0x197/0x210
<4> [363.698470]  ? rw_verify_area+0x16f/0x200
<4> [363.698482]  ? vfs_read+0x22a/0x360
<4> [363.698498]  ? do_syscall_64+0x1b7/0xab0
<4> [363.698506]  ? ksys_read+0x72/0xf0
<4> [363.698520]  ? do_syscall_64+0x1b7/0xab0
<4> [363.698526]  ? do_syscall_64+0x1b7/0xab0
<4> [363.698532]  ? do_syscall_64+0x1b7/0xab0
<4> [363.698535]  ? exc_page_fault+0xbd/0x2c0
<4> [363.698545]  entry_SYSCALL_64_after_hwframe+0x76/0x7e
<4> [363.698549] RIP: 0033:0x715af255025d

And a similar backtrace in hsw_color_commit_arm().

Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
Reviewed-by: Uma Shankar <uma.shankar@intel.com>
---
 drivers/gpu/drm/i915/display/intel_color.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/i915/display/intel_color.c b/drivers/gpu/drm/i915/display/intel_color.c
index e7950655434b8..9cbc719370456 100644
--- a/drivers/gpu/drm/i915/display/intel_color.c
+++ b/drivers/gpu/drm/i915/display/intel_color.c
@@ -1050,7 +1050,7 @@ static void hsw_color_commit_arm(struct intel_dsb *dsb,
 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
 	struct intel_display *display = to_intel_display(crtc);
 
-	intel_de_write(display, GAMMA_MODE(crtc->pipe),
+	intel_de_write_fw(display, GAMMA_MODE(crtc->pipe),
 		       crtc_state->gamma_mode);
 
 	intel_de_write_fw(display, PIPE_CSC_MODE(crtc->pipe),
@@ -4078,7 +4078,7 @@ static void glk_load_lut_3d(struct intel_dsb *dsb,
 	int i, lut_size = drm_color_lut32_size(blob);
 	enum pipe pipe = crtc->pipe;
 
-	if (!dsb && intel_de_read(display, LUT_3D_CTL(pipe)) & LUT_3D_READY) {
+	if (!dsb && intel_de_read_fw(display, LUT_3D_CTL(pipe)) & LUT_3D_READY) {
 		drm_err(display->drm, "[CRTC:%d:%s] 3D LUT not ready, not loading LUTs\n",
 			crtc->base.base.id, crtc->base.name);
 		return;
@@ -4096,7 +4096,7 @@ static void glk_lut_3d_commit(struct intel_dsb *dsb, struct intel_crtc *crtc, bo
 	enum pipe pipe = crtc->pipe;
 	u32 val = 0;
 
-	if (!dsb && intel_de_read(display, LUT_3D_CTL(pipe)) & LUT_3D_READY) {
+	if (!dsb && intel_de_read_fw(display, LUT_3D_CTL(pipe)) & LUT_3D_READY) {
 		drm_err(display->drm, "[CRTC:%d:%s] 3D LUT not ready, not committing change\n",
 			crtc->base.base.id, crtc->base.name);
 		return;
-- 
2.51.0

[PATCH v7 13/26] drm/i915/display: Use intel_de_write_fw in intel_pipe_fastset

[Maarten Lankhorst]

intel_set_pipe_src_size(), hsw_set_linetime_wm(),
intel_cpu_transcoder_set_m1_n1() and intel_set_transcoder_timings_lrr()
are called from an atomic context on PREEMPT_RT, and should be using the
_fw functions.

This likely prevents a deadlock on i915.

Again noticed when trying to disable preemption in vblank evasion:
<3> BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:48
<3> in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 1505, name: kms_cursor_lega
<3> preempt_count: 1, expected: 0
<3> RCU nest depth: 0, expected: 0
<4> 4 locks held by kms_cursor_lega/1505:
<4>  #0: ffffc90003c6f988 (crtc_ww_class_acquire){+.+.}-{0:0}, at: drm_mode_atomic_ioctl+0x13b/0xe90
<4>  #1: ffffc90003c6f9b0 (crtc_ww_class_mutex){+.+.}-{3:3}, at: drm_mode_atomic_ioctl+0x13b/0xe90
<4>  #2: ffff888135b838b8 (&intel_dp->psr.lock){+.+.}-{3:3}, at: intel_psr_lock+0xc5/0xf0 [xe]
<4>  #3: ffff88812607bbc0 (&wl->lock){+.+.}-{2:2}, at: intel_dmc_wl_get+0x3c/0x140 [xe]
<4> CPU: 6 UID: 0 PID: 1505 Comm: kms_cursor_lega Tainted: G     U              6.18.0-rc3-lgci-xe-xe-pw-156729v1+ #1 PREEMPT_{RT,(lazy)}
<4> Tainted: [U]=USER
<4> Hardware name: Intel Corporation Panther Lake Client Platform/PTL-UH LP5 T3 RVP1, BIOS PTLPFWI1.R00.3383.D02.2509240621 09/24/2025
<4> Call Trace:
<4>  <TASK>
<4>  dump_stack_lvl+0xc1/0xf0
<4>  dump_stack+0x10/0x20
<4>  __might_resched+0x174/0x260
<4>  rt_spin_lock+0x63/0x200
<4>  ? intel_dmc_wl_get+0x3c/0x140 [xe]
<4>  intel_dmc_wl_get+0x3c/0x140 [xe]
<4>  intel_set_pipe_src_size+0x89/0xe0 [xe]
<4>  intel_update_crtc+0x3c1/0x950 [xe]
<4>  ? intel_pre_update_crtc+0x258/0x400 [xe]
<4>  skl_commit_modeset_enables+0x217/0x720 [xe]
<4>  intel_atomic_commit_tail+0xd4e/0x1af0 [xe]
<4>  ? lock_release+0xce/0x2a0
<4>  intel_atomic_commit+0x2e5/0x330 [xe]
<4>  ? intel_atomic_commit+0x2e5/0x330 [xe]
<4>  drm_atomic_commit+0xaf/0xf0
<4>  ? __pfx___drm_printfn_info+0x10/0x10
<4>  drm_mode_atomic_ioctl+0xbd5/0xe90
<4>  ? lock_acquire+0xc4/0x2e0
<4>  ? __pfx_drm_mode_atomic_ioctl+0x10/0x10
<4>  drm_ioctl_kernel+0xb6/0x120
<4>  drm_ioctl+0x2d7/0x5a0
<4>  ? __pfx_drm_mode_atomic_ioctl+0x10/0x10
<4>  ? rt_spin_unlock+0xa0/0x140
<4>  ? __pm_runtime_resume+0x53/0x90
<4>  xe_drm_ioctl+0x56/0x90 [xe]
<4>  __x64_sys_ioctl+0xa8/0x110
<4>  ? lock_acquire+0xc4/0x2e0
<4>  x64_sys_call+0x1144/0x26a0
<4>  do_syscall_64+0x93/0xae0
<4>  ? lock_release+0xce/0x2a0
<4>  ? __task_pid_nr_ns+0xd9/0x270
<4>  ? do_syscall_64+0x1b7/0xae0
<4>  ? find_held_lock+0x31/0x90
<4>  ? __task_pid_nr_ns+0xcf/0x270
<4>  ? __lock_acquire+0x43e/0x2860
<4>  ? __task_pid_nr_ns+0xd9/0x270
<4>  ? lock_acquire+0xc4/0x2e0
<4>  ? find_held_lock+0x31/0x90
<4>  ? __task_pid_nr_ns+0xcf/0x270
<4>  ? lock_release+0xce/0x2a0
<4>  ? __task_pid_nr_ns+0xd9/0x270
<4>  ? do_syscall_64+0x1b7/0xae0
<4>  ? do_syscall_64+0x1b7/0xae0
<4>  entry_SYSCALL_64_after_hwframe+0x76/0x7e

Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
---
 drivers/gpu/drm/i915/display/intel_display.c | 36 ++++++++++----------
 drivers/gpu/drm/i915/display/intel_vrr.c     | 18 +++++-----
 2 files changed, 27 insertions(+), 27 deletions(-)

diff --git a/drivers/gpu/drm/i915/display/intel_display.c b/drivers/gpu/drm/i915/display/intel_display.c
index b18ce0c36a649..9cfa3fa915cf5 100644
--- a/drivers/gpu/drm/i915/display/intel_display.c
+++ b/drivers/gpu/drm/i915/display/intel_display.c
@@ -1606,9 +1606,9 @@ static void hsw_set_linetime_wm(const struct intel_crtc_state *crtc_state)
 	struct intel_display *display = to_intel_display(crtc_state);
 	struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc);
 
-	intel_de_write(display, WM_LINETIME(crtc->pipe),
-		       HSW_LINETIME(crtc_state->linetime) |
-		       HSW_IPS_LINETIME(crtc_state->ips_linetime));
+	intel_de_write_fw(display, WM_LINETIME(crtc->pipe),
+			  HSW_LINETIME(crtc_state->linetime) |
+			  HSW_IPS_LINETIME(crtc_state->ips_linetime));
 }
 
 static void hsw_set_frame_start_delay(const struct intel_crtc_state *crtc_state)
@@ -2604,14 +2604,14 @@ void intel_set_m_n(struct intel_display *display,
 		   i915_reg_t data_m_reg, i915_reg_t data_n_reg,
 		   i915_reg_t link_m_reg, i915_reg_t link_n_reg)
 {
-	intel_de_write(display, data_m_reg, TU_SIZE(m_n->tu) | m_n->data_m);
-	intel_de_write(display, data_n_reg, m_n->data_n);
-	intel_de_write(display, link_m_reg, m_n->link_m);
+	intel_de_write_fw(display, data_m_reg, TU_SIZE(m_n->tu) | m_n->data_m);
+	intel_de_write_fw(display, data_n_reg, m_n->data_n);
+	intel_de_write_fw(display, link_m_reg, m_n->link_m);
 	/*
 	 * On BDW+ writing LINK_N arms the double buffered update
 	 * of all the M/N registers, so it must be written last.
 	 */
-	intel_de_write(display, link_n_reg, m_n->link_n);
+	intel_de_write_fw(display, link_n_reg, m_n->link_n);
 }
 
 bool intel_cpu_transcoder_has_m2_n2(struct intel_display *display,
@@ -2798,9 +2798,9 @@ static void intel_set_transcoder_timings_lrr(const struct intel_crtc_state *crtc
 	}
 
 	if (DISPLAY_VER(display) >= 13) {
-		intel_de_write(display,
-			       TRANS_SET_CONTEXT_LATENCY(display, cpu_transcoder),
-			       crtc_state->set_context_latency);
+		intel_de_write_fw(display,
+				  TRANS_SET_CONTEXT_LATENCY(display, cpu_transcoder),
+				  crtc_state->set_context_latency);
 
 		/*
 		 * VBLANK_START not used by hw, just clear it
@@ -2816,9 +2816,9 @@ static void intel_set_transcoder_timings_lrr(const struct intel_crtc_state *crtc
 	 * The hardware actually ignores TRANS_VBLANK.VBLANK_END in DP mode.
 	 * But let's write it anyway to keep the state checker happy.
 	 */
-	intel_de_write(display, TRANS_VBLANK(display, cpu_transcoder),
-		       VBLANK_START(crtc_vblank_start - 1) |
-		       VBLANK_END(crtc_vblank_end - 1));
+	intel_de_write_fw(display, TRANS_VBLANK(display, cpu_transcoder),
+			  VBLANK_START(crtc_vblank_start - 1) |
+			  VBLANK_END(crtc_vblank_end - 1));
 	/*
 	 * For platforms that always use VRR Timing Generator, the VTOTAL.Vtotal
 	 * bits are not required. Since the support for these bits is going to
@@ -2832,9 +2832,9 @@ static void intel_set_transcoder_timings_lrr(const struct intel_crtc_state *crtc
 	 * The double buffer latch point for TRANS_VTOTAL
 	 * is the transcoder's undelayed vblank.
 	 */
-	intel_de_write(display, TRANS_VTOTAL(display, cpu_transcoder),
-		       VACTIVE(crtc_vdisplay - 1) |
-		       VTOTAL(crtc_vtotal - 1));
+	intel_de_write_fw(display, TRANS_VTOTAL(display, cpu_transcoder),
+			  VACTIVE(crtc_vdisplay - 1) |
+			  VTOTAL(crtc_vtotal - 1));
 
 	intel_vrr_set_fixed_rr_timings(crtc_state);
 	intel_vrr_transcoder_enable(crtc_state);
@@ -2851,8 +2851,8 @@ static void intel_set_pipe_src_size(const struct intel_crtc_state *crtc_state)
 	/* pipesrc controls the size that is scaled from, which should
 	 * always be the user's requested size.
 	 */
-	intel_de_write(display, PIPESRC(display, pipe),
-		       PIPESRC_WIDTH(width - 1) | PIPESRC_HEIGHT(height - 1));
+	intel_de_write_fw(display, PIPESRC(display, pipe),
+			  PIPESRC_WIDTH(width - 1) | PIPESRC_HEIGHT(height - 1));
 }
 
 static bool intel_pipe_is_interlaced(const struct intel_crtc_state *crtc_state)
diff --git a/drivers/gpu/drm/i915/display/intel_vrr.c b/drivers/gpu/drm/i915/display/intel_vrr.c
index 8a957804cb970..a2d507b238322 100644
--- a/drivers/gpu/drm/i915/display/intel_vrr.c
+++ b/drivers/gpu/drm/i915/display/intel_vrr.c
@@ -318,12 +318,12 @@ void intel_vrr_set_fixed_rr_timings(const struct intel_crtc_state *crtc_state)
 	if (!intel_vrr_possible(crtc_state))
 		return;
 
-	intel_de_write(display, TRANS_VRR_VMIN(display, cpu_transcoder),
-		       intel_vrr_fixed_rr_hw_vmin(crtc_state) - 1);
-	intel_de_write(display, TRANS_VRR_VMAX(display, cpu_transcoder),
-		       intel_vrr_fixed_rr_hw_vmax(crtc_state) - 1);
-	intel_de_write(display, TRANS_VRR_FLIPLINE(display, cpu_transcoder),
-		       intel_vrr_fixed_rr_hw_flipline(crtc_state) - 1);
+	intel_de_write_fw(display, TRANS_VRR_VMIN(display, cpu_transcoder),
+			  intel_vrr_fixed_rr_hw_vmin(crtc_state) - 1);
+	intel_de_write_fw(display, TRANS_VRR_VMAX(display, cpu_transcoder),
+			  intel_vrr_fixed_rr_hw_vmax(crtc_state) - 1);
+	intel_de_write_fw(display, TRANS_VRR_FLIPLINE(display, cpu_transcoder),
+			  intel_vrr_fixed_rr_hw_flipline(crtc_state) - 1);
 }
 
 static
@@ -908,8 +908,8 @@ static void intel_vrr_tg_enable(const struct intel_crtc_state *crtc_state,
 	enum transcoder cpu_transcoder = crtc_state->cpu_transcoder;
 	u32 vrr_ctl;
 
-	intel_de_write(display, TRANS_PUSH(display, cpu_transcoder),
-		       trans_vrr_push(crtc_state, false));
+	intel_de_write_fw(display, TRANS_PUSH(display, cpu_transcoder),
+			  trans_vrr_push(crtc_state, false));
 
 	vrr_ctl = VRR_CTL_VRR_ENABLE | trans_vrr_ctl(crtc_state);
 
@@ -921,7 +921,7 @@ static void intel_vrr_tg_enable(const struct intel_crtc_state *crtc_state,
 	if (cmrr_enable)
 		vrr_ctl |= VRR_CTL_CMRR_ENABLE;
 
-	intel_de_write(display, TRANS_VRR_CTL(display, cpu_transcoder), vrr_ctl);
+	intel_de_write_fw(display, TRANS_VRR_CTL(display, cpu_transcoder), vrr_ctl);
 }
 
 static void intel_vrr_tg_disable(const struct intel_crtc_state *old_crtc_state)
-- 
2.51.0

[PATCH v7 14/26] drm/i915/display: Make set_pipeconf use the fw variants

[Maarten Lankhorst]

The calls are used inside the critical section when updating
the gamma mode, and thus should use the _fw variants to prevent
locks.

Fixes following splat:
| BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:48
| in_atomic(): 0, irqs_disabled(): 1, non_block: 0, pid: 2115, name: modprobe
| preempt_count: 0, expected: 0
| RCU nest depth: 0, expected: 0
| 4 locks held by modprobe/2115:
|  #0: ffff99b9425161a0 (&dev->mutex){....}-{4:4}, at: __driver_attach+0xaf/0x1c0
|  #1: ffffaa224810f6c0 (crtc_ww_class_acquire){+.+.}-{0:0}, at: intel_initial_commit+0x4c/0x200 [i915]
|  #2: ffffaa224810f6e8 (crtc_ww_class_mutex){+.+.}-{4:4}, at: intel_initial_commit+0x4c/0x200 [i915]
|  #3: ffff99b94a6c9030 (&uncore->lock){+.+.}-{3:3}, at: gen6_write32+0x50/0x290 [i915]
| irq event stamp: 513344
| hardirqs last  enabled at (513343): [<ffffffff8ba8d84c>] _raw_spin_unlock_irqrestore+0x4c/0x60
| hardirqs last disabled at (513344): [<ffffffffc1543646>] intel_pipe_update_start+0x216/0x2c0 [i915]
| softirqs last  enabled at (512766): [<ffffffff8af045cf>] __local_bh_enable_ip+0x10f/0x170
| softirqs last disabled at (512712): [<ffffffffc14dfb6a>] __i915_request_queue+0x3a/0x70 [i915]
| CPU: 3 UID: 0 PID: 2115 Comm: modprobe Tainted: G        W           6.18.0-rc1+ #17 PREEMPT_{RT,(lazy)}
| Tainted: [W]=WARN
| Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./Z68 Pro3-M, BIOS P2.30 06/29/2012
| Call Trace:
|  <TASK>
|  dump_stack_lvl+0x68/0x90
|  __might_resched.cold+0xf0/0x12b
|  rt_spin_lock+0x5f/0x200
|  gen6_write32+0x50/0x290 [i915]
|  ilk_set_pipeconf+0x12d/0x230 [i915]
|  ilk_color_commit_arm+0x2d/0x70 [i915]
|  intel_update_crtc+0x15b/0x690 [i915]
|  intel_commit_modeset_enables+0xa6/0xd0 [i915]
|  intel_atomic_commit_tail+0xd55/0x19a0 [i915]
|  intel_atomic_commit+0x25d/0x2a0 [i915]
|  drm_atomic_commit+0xad/0xe0 [drm]
|  intel_initial_commit+0x16c/0x200 [i915]
|  intel_display_driver_probe+0x2e/0x80 [i915]
|  i915_driver_probe+0x791/0xc10 [i915]
|  i915_pci_probe+0xd7/0x190 [i915]

Reported-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
---
 drivers/gpu/drm/i915/display/intel_display.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/i915/display/intel_display.c b/drivers/gpu/drm/i915/display/intel_display.c
index 9cfa3fa915cf5..3a0cab5db2490 100644
--- a/drivers/gpu/drm/i915/display/intel_display.c
+++ b/drivers/gpu/drm/i915/display/intel_display.c
@@ -3035,8 +3035,9 @@ void i9xx_set_pipeconf(const struct intel_crtc_state *crtc_state)
 
 	val |= TRANSCONF_FRAME_START_DELAY(crtc_state->framestart_delay - 1);
 
-	intel_de_write(display, TRANSCONF(display, cpu_transcoder), val);
-	intel_de_posting_read(display, TRANSCONF(display, cpu_transcoder));
+	intel_de_write_fw(display, TRANSCONF(display, cpu_transcoder), val);
+	/* posting read */
+	intel_de_read_fw(display, TRANSCONF(display, cpu_transcoder));
 }
 
 static enum intel_output_format
@@ -3231,8 +3232,9 @@ void ilk_set_pipeconf(const struct intel_crtc_state *crtc_state)
 	val |= TRANSCONF_FRAME_START_DELAY(crtc_state->framestart_delay - 1);
 	val |= TRANSCONF_MSA_TIMING_DELAY(crtc_state->msa_timing_delay);
 
-	intel_de_write(display, TRANSCONF(display, cpu_transcoder), val);
-	intel_de_posting_read(display, TRANSCONF(display, cpu_transcoder));
+	intel_de_write_fw(display, TRANSCONF(display, cpu_transcoder), val);
+	/* posting read */
+	intel_de_read_fw(display, TRANSCONF(display, cpu_transcoder));
 }
 
 static void hsw_set_transconf(const struct intel_crtc_state *crtc_state)
-- 
2.51.0

[PATCH v7 15/26] drm/i915/display: Fix intel_lpe_audio_irq_handler for PREEMPT-RT

[Maarten Lankhorst]

The LPE audio interrupt comes from the i915 interrupt handler. It
should be in irq disabled context.

With PREEMPT_RT enabled, the IRQ handler is threaded.
Because intel_lpe_audio_irq_handler() may be called in threaded IRQ context,
generic_handle_irq_safe API disables the interrupts before calling LPE's
interrupt top half handler.

This fixes braswell audio issues with RT enabled.

Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Uma Shankar <uma.shankar@intel.com>
---
 drivers/gpu/drm/i915/display/intel_lpe_audio.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/display/intel_lpe_audio.c b/drivers/gpu/drm/i915/display/intel_lpe_audio.c
index 117b60656ca19..306b16889bc92 100644
--- a/drivers/gpu/drm/i915/display/intel_lpe_audio.c
+++ b/drivers/gpu/drm/i915/display/intel_lpe_audio.c
@@ -262,7 +262,7 @@ void intel_lpe_audio_irq_handler(struct intel_display *display)
 	if (!HAS_LPE_AUDIO(display))
 		return;
 
-	ret = generic_handle_irq(display->audio.lpe.irq);
+	ret = generic_handle_irq_safe(display->audio.lpe.irq);
 	if (ret)
 		drm_err_ratelimited(display->drm,
 				    "error handling LPE audio irq: %d\n", ret);
-- 
2.51.0

[PATCH v7 16/26] drm/i915/gt: Use spin_lock_irq() instead of local_irq_disable() + spin_lock()

[Maarten Lankhorst]

From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

execlists_dequeue() is invoked from a function which uses
local_irq_disable() to disable interrupts so the spin_lock() behaves
like spin_lock_irq().
This breaks PREEMPT_RT because local_irq_disable() + spin_lock() is not
the same as spin_lock_irq().

execlists_dequeue_irq() and execlists_dequeue() has each one caller
only. If intel_engine_cs::active::lock is acquired and released with the
_irq suffix then it behaves almost as if execlists_dequeue() would be
invoked with disabled interrupts. The difference is the last part of the
function which is then invoked with enabled interrupts.
I can't tell if this makes a difference. From looking at it, it might
work to move the last unlock at the end of the function as I didn't find
anything that would acquire the lock again.

Reported-by: Clark Williams <williams@redhat.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
---
 .../drm/i915/gt/intel_execlists_submission.c    | 17 +++++------------
 1 file changed, 5 insertions(+), 12 deletions(-)

diff --git a/drivers/gpu/drm/i915/gt/intel_execlists_submission.c b/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
index 1359fc9cb88ef..e11db81dca9c5 100644
--- a/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
+++ b/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
@@ -1300,7 +1300,7 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
 	 * and context switches) submission.
 	 */
 
-	spin_lock(&sched_engine->lock);
+	spin_lock_irq(&sched_engine->lock);
 
 	/*
 	 * If the queue is higher priority than the last
@@ -1400,7 +1400,7 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
 				 * Even if ELSP[1] is occupied and not worthy
 				 * of timeslices, our queue might be.
 				 */
-				spin_unlock(&sched_engine->lock);
+				spin_unlock_irq(&sched_engine->lock);
 				return;
 			}
 		}
@@ -1426,7 +1426,7 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
 
 		if (last && !can_merge_rq(last, rq)) {
 			spin_unlock(&ve->base.sched_engine->lock);
-			spin_unlock(&engine->sched_engine->lock);
+			spin_unlock_irq(&engine->sched_engine->lock);
 			return; /* leave this for another sibling */
 		}
 
@@ -1588,7 +1588,7 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
 	 */
 	sched_engine->queue_priority_hint = queue_prio(sched_engine);
 	i915_sched_engine_reset_on_empty(sched_engine);
-	spin_unlock(&sched_engine->lock);
+	spin_unlock_irq(&sched_engine->lock);
 
 	/*
 	 * We can skip poking the HW if we ended up with exactly the same set
@@ -1614,13 +1614,6 @@ static void execlists_dequeue(struct intel_engine_cs *engine)
 	}
 }
 
-static void execlists_dequeue_irq(struct intel_engine_cs *engine)
-{
-	local_irq_disable(); /* Suspend interrupts across request submission */
-	execlists_dequeue(engine);
-	local_irq_enable(); /* flush irq_work (e.g. breadcrumb enabling) */
-}
-
 static void clear_ports(struct i915_request **ports, int count)
 {
 	memset_p((void **)ports, NULL, count);
@@ -2475,7 +2468,7 @@ static void execlists_submission_tasklet(struct tasklet_struct *t)
 	}
 
 	if (!engine->execlists.pending[0]) {
-		execlists_dequeue_irq(engine);
+		execlists_dequeue(engine);
 		start_timeslice(engine);
 	}
 
-- 
2.51.0

[PATCH v7 17/26] drm/i915: Drop the irqs_disabled() check

[Maarten Lankhorst]

From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

The !irqs_disabled() check triggers on PREEMPT_RT even with
i915_sched_engine::lock acquired. The reason is the lock is transformed
into a sleeping lock on PREEMPT_RT and does not disable interrupts.

There is no need to check for disabled interrupts. The lockdep
annotation below already check if the lock has been acquired by the
caller and will yell if the interrupts are not disabled.

Remove the !irqs_disabled() check.

Reported-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
---
 drivers/gpu/drm/i915/i915_request.c | 2 --
 1 file changed, 2 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
index d2c7b1090df08..f66f8efc70629 100644
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -610,7 +610,6 @@ bool __i915_request_submit(struct i915_request *request)
 
 	RQ_TRACE(request, "\n");
 
-	GEM_BUG_ON(!irqs_disabled());
 	lockdep_assert_held(&engine->sched_engine->lock);
 
 	/*
@@ -719,7 +718,6 @@ void __i915_request_unsubmit(struct i915_request *request)
 	 */
 	RQ_TRACE(request, "\n");
 
-	GEM_BUG_ON(!irqs_disabled());
 	lockdep_assert_held(&engine->sched_engine->lock);
 
 	/*
-- 
2.51.0

[PATCH v7 18/26] drm/i915/guc: Consider also RCU depth in busy loop.

[Maarten Lankhorst]

From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

intel_guc_send_busy_loop() looks at in_atomic() and irqs_disabled() to
decide if it should busy-spin while waiting or if it may sleep.
Both checks will report false on PREEMPT_RT if sleeping spinlocks are
acquired leading to RCU splats while the function sleeps.

Check also if RCU has been disabled.

Reported-by: "John B. Wyatt IV" <jwyatt@redhat.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
---
 drivers/gpu/drm/i915/gt/uc/intel_guc.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc.h b/drivers/gpu/drm/i915/gt/uc/intel_guc.h
index 053780f562c1a..b25fa8f4dc4bd 100644
--- a/drivers/gpu/drm/i915/gt/uc/intel_guc.h
+++ b/drivers/gpu/drm/i915/gt/uc/intel_guc.h
@@ -362,7 +362,7 @@ static inline int intel_guc_send_busy_loop(struct intel_guc *guc,
 {
 	int err;
 	unsigned int sleep_period_ms = 1;
-	bool not_atomic = !in_atomic() && !irqs_disabled();
+	bool not_atomic = !in_atomic() && !irqs_disabled() && !rcu_preempt_depth();
 
 	/*
 	 * FIXME: Have caller pass in if we are in an atomic context to avoid
-- 
2.51.0

[PATCH v7 19/26] drm/i915/gt: Fix selftests on PREEMPT_RT

[Maarten Lankhorst]

Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
---
 drivers/gpu/drm/i915/gt/selftest_engine_pm.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/i915/gt/selftest_engine_pm.c b/drivers/gpu/drm/i915/gt/selftest_engine_pm.c
index 10e556a7eac45..c1eff9edd8a5e 100644
--- a/drivers/gpu/drm/i915/gt/selftest_engine_pm.c
+++ b/drivers/gpu/drm/i915/gt/selftest_engine_pm.c
@@ -277,11 +277,11 @@ static int live_engine_busy_stats(void *arg)
 		st_engine_heartbeat_disable(engine);
 
 		ENGINE_TRACE(engine, "measuring idle time\n");
-		preempt_disable();
+		migrate_disable();
 		de = intel_engine_get_busy_time(engine, &t[0]);
 		udelay(100);
 		de = ktime_sub(intel_engine_get_busy_time(engine, &t[1]), de);
-		preempt_enable();
+		migrate_enable();
 		dt = ktime_sub(t[1], t[0]);
 		if (de < 0 || de > 10) {
 			pr_err("%s: reported %lldns [%d%%] busyness while sleeping [for %lldns]\n",
@@ -316,11 +316,11 @@ static int live_engine_busy_stats(void *arg)
 		}
 
 		ENGINE_TRACE(engine, "measuring busy time\n");
-		preempt_disable();
+		migrate_disable();
 		de = intel_engine_get_busy_time(engine, &t[0]);
 		mdelay(100);
 		de = ktime_sub(intel_engine_get_busy_time(engine, &t[1]), de);
-		preempt_enable();
+		migrate_enable();
 		dt = ktime_sub(t[1], t[0]);
 		if (100 * de < 95 * dt || 95 * de > 100 * dt) {
 			pr_err("%s: reported %lldns [%d%%] busyness while spinning [for %lldns]\n",
-- 
2.51.0

[PATCH v7 20/26] drm/i915/gt: Set stop_timeout() correctly on PREEMPT-RT

[Maarten Lankhorst]

Also check if RCU is disabled for PREEMPT-RT, which is the case when
local_bh_disable() is called.

Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
---
 drivers/gpu/drm/i915/gt/intel_engine_cs.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
index c0fd349a4600c..9dd9665128caa 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
@@ -1607,7 +1607,7 @@ u64 intel_engine_get_last_batch_head(const struct intel_engine_cs *engine)
 
 static unsigned long stop_timeout(const struct intel_engine_cs *engine)
 {
-	if (in_atomic() || irqs_disabled()) /* inside atomic preempt-reset? */
+	if (in_atomic() || irqs_disabled() || rcu_preempt_depth()) /* inside atomic preempt-reset? */
 		return 0;
 
 	/*
-- 
2.51.0

[PATCH v7 21/26] drm/i915/display: Remove uncore lock from vlv_atomic_update_fifo

[Maarten Lankhorst]

TODO: Grab uncore lock during entire vblank evasion before disabling
interrupts, and check what breaks?

Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
---
 drivers/gpu/drm/i915/display/i9xx_wm.c | 4 ----
 1 file changed, 4 deletions(-)

diff --git a/drivers/gpu/drm/i915/display/i9xx_wm.c b/drivers/gpu/drm/i915/display/i9xx_wm.c
index 9e170e79dcf67..8e1b3aced9a98 100644
--- a/drivers/gpu/drm/i915/display/i9xx_wm.c
+++ b/drivers/gpu/drm/i915/display/i9xx_wm.c
@@ -1863,7 +1863,6 @@ static void vlv_atomic_update_fifo(struct intel_atomic_state *state,
 				   struct intel_crtc *crtc)
 {
 	struct intel_display *display = to_intel_display(crtc);
-	struct intel_uncore *uncore = to_intel_uncore(display->drm);
 	const struct intel_crtc_state *crtc_state =
 		intel_atomic_get_new_crtc_state(state, crtc);
 	const struct vlv_fifo_state *fifo_state =
@@ -1892,7 +1891,6 @@ static void vlv_atomic_update_fifo(struct intel_atomic_state *state,
 	 * intel_pipe_update_start() has already disabled interrupts
 	 * for us, so a plain spin_lock() is sufficient here.
 	 */
-	spin_lock(&uncore->lock);
 
 	switch (crtc->pipe) {
 	case PIPE_A:
@@ -1951,8 +1949,6 @@ static void vlv_atomic_update_fifo(struct intel_atomic_state *state,
 	}
 
 	intel_de_read_fw(display, DSPARB(display));
-
-	spin_unlock(&uncore->lock);
 }
 
 #undef VLV_FIFO
-- 
2.51.0

[PATCH v7 22/26] drm/i915: Use sleeping selftests for igt_atomic on PREEMPT_RT

[Maarten Lankhorst]

This makes the i915 selftests slightly happier, especially
related to GPU reset.

I believe this may be a better approach than trying to convert
uncore->lock to raw_spinlock

Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
---
 drivers/gpu/drm/i915/selftests/igt_atomic.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/drivers/gpu/drm/i915/selftests/igt_atomic.c b/drivers/gpu/drm/i915/selftests/igt_atomic.c
index fb506b6990956..8ae39cf570b76 100644
--- a/drivers/gpu/drm/i915/selftests/igt_atomic.c
+++ b/drivers/gpu/drm/i915/selftests/igt_atomic.c
@@ -39,7 +39,14 @@ static void __hardirq_end(void)
 	local_irq_enable();
 }
 
+static void __maybe_unused __nop(void)
+{}
+
 const struct igt_atomic_section igt_atomic_phases[] = {
+#if IS_ENABLED(CONFIG_PREEMPT_RT)
+	{ "sleeping", __nop, __nop },
+	{ },
+#endif
 	{ "preempt", __preempt_begin, __preempt_end },
 	{ "softirq", __softirq_begin, __softirq_end },
 	{ "hardirq", __hardirq_begin, __hardirq_end },
-- 
2.51.0

[PATCH v7 23/26] Revert "drm/i915: Depend on !PREEMPT_RT."

[Maarten Lankhorst]

From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

Once the known issues are addressed, it should be safe to enable the
driver.

Acked-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
---
 drivers/gpu/drm/i915/Kconfig | 1 -
 1 file changed, 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/Kconfig b/drivers/gpu/drm/i915/Kconfig
index 5e939004b6463..40a9234e6e5dc 100644
--- a/drivers/gpu/drm/i915/Kconfig
+++ b/drivers/gpu/drm/i915/Kconfig
@@ -3,7 +3,6 @@ config DRM_I915
 	tristate "Intel 8xx/9xx/G3x/G4x/HD Graphics"
 	depends on DRM
 	depends on X86 && PCI
-	depends on !PREEMPT_RT
 	select INTEL_GTT if X86
 	select INTERVAL_TREE
 	# we need shmfs for the swappable backing store, and in particular
-- 
2.51.0

[PATCH v7 24/26] PREEMPT_RT injection

[Maarten Lankhorst]

Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
---
 drivers/gpu/drm/i915/Kconfig.debug | 15 ---------------
 drivers/gpu/drm/xe/Kconfig.debug   |  5 +++++
 kernel/Kconfig.preempt             |  4 ++--
 3 files changed, 7 insertions(+), 17 deletions(-)

diff --git a/drivers/gpu/drm/i915/Kconfig.debug b/drivers/gpu/drm/i915/Kconfig.debug
index 3562a02ef7adc..0ab10ff41e38d 100644
--- a/drivers/gpu/drm/i915/Kconfig.debug
+++ b/drivers/gpu/drm/i915/Kconfig.debug
@@ -233,21 +233,6 @@ config DRM_I915_LOW_LEVEL_TRACEPOINTS
 
 	  If in doubt, say "N".
 
-config DRM_I915_DEBUG_VBLANK_EVADE
-	bool "Enable extra debug warnings for vblank evasion"
-	depends on DRM_I915
-	default n
-	help
-	  Choose this option to turn on extra debug warnings for the
-	  vblank evade mechanism. This gives a warning every time the
-	  the deadline allotted for the vblank evade critical section
-	  is exceeded, even if there isn't an actual risk of missing
-	  the vblank.
-
-	  Recommended for driver developers only.
-
-	  If in doubt, say "N".
-
 config DRM_I915_DEBUG_RUNTIME_PM
 	bool "Enable extra state checking for runtime PM"
 	depends on DRM_I915
diff --git a/drivers/gpu/drm/xe/Kconfig.debug b/drivers/gpu/drm/xe/Kconfig.debug
index 01227c77f6d70..1d5f11c6e88f3 100644
--- a/drivers/gpu/drm/xe/Kconfig.debug
+++ b/drivers/gpu/drm/xe/Kconfig.debug
@@ -30,6 +30,11 @@ config DRM_XE_DEBUG
 
 	  If in doubt, say "N".
 
+config DRM_I915_DEBUG_VBLANK_EVADE
+        def_bool y
+        depends on DRM_XE
+
+
 config DRM_XE_DEBUG_VM
 	bool "Enable extra VM debugging info"
 	default n
diff --git a/kernel/Kconfig.preempt b/kernel/Kconfig.preempt
index 88c594c6d7fcd..11bec9be95405 100644
--- a/kernel/Kconfig.preempt
+++ b/kernel/Kconfig.preempt
@@ -90,9 +90,9 @@ config PREEMPT_LAZY
 endchoice
 
 config PREEMPT_RT
-	bool "Fully Preemptible Kernel (Real-Time)"
-	depends on EXPERT && ARCH_SUPPORTS_RT && !COMPILE_TEST
+	def_bool y
 	select PREEMPTION
+	depends on ARCH_SUPPORTS_RT
 	help
 	  This option turns the kernel into a real-time kernel by replacing
 	  various locking primitives (spinlocks, rwlocks, etc.) with
-- 
2.51.0

[PATCH v7 25/26] FOR-CI: bump MAX_STACK_TRACE_ENTRIES

[Maarten Lankhorst]

We're hitting a bug in CI where MAX_STACK_TRACE_ENTRIES is set too low.
My guess is the repeated loading/unloading is creating multiples of the
same entries. As a hack just reset lockdep. This might only be necessary
for CI + PREEMPT_RT.

Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
---
 lib/Kconfig.debug | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug
index 56ec7173cb7b6..b8d7e6dc345e9 100644
--- a/lib/Kconfig.debug
+++ b/lib/Kconfig.debug
@@ -1625,8 +1625,8 @@ config LOCKDEP_CHAINS_BITS
 config LOCKDEP_STACK_TRACE_BITS
 	int "Size for MAX_STACK_TRACE_ENTRIES (as Nth power of 2)"
 	depends on LOCKDEP && !LOCKDEP_SMALL
-	range 10 26
-	default 19
+	range 22 26
+	default 22
 	help
 	  Try increasing this value if you hit "BUG: MAX_STACK_TRACE_ENTRIES too low!" message.
 
-- 
2.51.0

[PATCH v7 26/26] drm/i915/gt: Add a spinlock to prevent starvation of irq_work.

[Maarten Lankhorst]

From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

IRQ-Work (FIFO-1) will be preempted by the threaded-interrupt (FIFO-50)
and the interrupt will poll on signaler_active while the irq-work can't
make progress.

Solve this by adding a spinlock to prevent starvation and force
completion.

Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
---
 drivers/gpu/drm/i915/gt/intel_breadcrumbs.c       | 8 +++++++-
 drivers/gpu/drm/i915/gt/intel_breadcrumbs_types.h | 1 +
 2 files changed, 8 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c b/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
index c10ac0ab3bfa8..c2b174bfa1418 100644
--- a/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
+++ b/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
@@ -209,6 +209,7 @@ static void signal_irq_work(struct irq_work *work)
 		intel_breadcrumbs_disarm_irq(b);
 
 	rcu_read_lock();
+	spin_lock(&b->signaler_active_sync);
 	atomic_inc(&b->signaler_active);
 	list_for_each_entry_rcu(ce, &b->signalers, signal_link) {
 		struct i915_request *rq;
@@ -246,6 +247,7 @@ static void signal_irq_work(struct irq_work *work)
 		}
 	}
 	atomic_dec(&b->signaler_active);
+	spin_unlock(&b->signaler_active_sync);
 	rcu_read_unlock();
 
 	llist_for_each_safe(signal, sn, signal) {
@@ -290,6 +292,7 @@ intel_breadcrumbs_create(struct intel_engine_cs *irq_engine)
 	init_llist_head(&b->signaled_requests);
 
 	spin_lock_init(&b->irq_lock);
+	spin_lock_init(&b->signaler_active_sync);
 	init_irq_work(&b->irq_work, signal_irq_work);
 
 	b->irq_engine = irq_engine;
@@ -487,8 +490,11 @@ void intel_context_remove_breadcrumbs(struct intel_context *ce,
 	if (release)
 		intel_context_put(ce);
 
-	while (atomic_read(&b->signaler_active))
+	while (atomic_read(&b->signaler_active)) {
+		spin_lock(&b->signaler_active_sync);
+		spin_unlock(&b->signaler_active_sync);
 		cpu_relax();
+	}
 }
 
 static void print_signals(struct intel_breadcrumbs *b, struct drm_printer *p)
diff --git a/drivers/gpu/drm/i915/gt/intel_breadcrumbs_types.h b/drivers/gpu/drm/i915/gt/intel_breadcrumbs_types.h
index bdf09fd67b6e7..28dae32628aab 100644
--- a/drivers/gpu/drm/i915/gt/intel_breadcrumbs_types.h
+++ b/drivers/gpu/drm/i915/gt/intel_breadcrumbs_types.h
@@ -40,6 +40,7 @@ struct intel_breadcrumbs {
 	struct list_head signalers;
 	struct llist_head signaled_requests;
 	atomic_t signaler_active;
+	spinlock_t signaler_active_sync;
 
 	spinlock_t irq_lock; /* protects the interrupt from hardirq context */
 	struct irq_work irq_work; /* for use from inside irq_lock */
-- 
2.51.0

Re: [PATCH v7 26/26] drm/i915/gt: Add a spinlock to prevent starvation of irq_work.

[Sebastian Andrzej Siewior]

On 2026-03-10 12:57:08 [+0100], Maarten Lankhorst wrote:
> From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
> 
> IRQ-Work (FIFO-1) will be preempted by the threaded-interrupt (FIFO-50)
> and the interrupt will poll on signaler_active while the irq-work can't
> make progress.

The threaded-interrupt is the interrupt.

| On PREEMPT_RT the irq_work can be preempted by threaded-interrupt which
| will be poll for completion but the irq_work routine can't make
| progress.

> Solve this by adding a spinlock to prevent starvation and force
> completion.

| Solve this by adding a spinlock_t to prevent starvation by forcing a
| context switch if lock is held based on `signaler_active'. On
| !PREEMPT_RT `signaler_active' can only be non-zero if multiple CPUs are
| involved and spinning on the lock leds to the same result.

> Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>

If I am the From: then I should have the Signed-off-by, too. Let me do
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

so it can be picked up.

You did suggest the following:

--- a/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
+++ b/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
@@ -209,7 +209,7 @@ static void signal_irq_work(struct irq_work *work)
 		intel_breadcrumbs_disarm_irq(b);
 
 	rcu_read_lock();
-	atomic_inc(&b->signaler_active);
+	spin_lock(&b->signaler_active_sync);
 	list_for_each_entry_rcu(ce, &b->signalers, signal_link) {
 		struct i915_request *rq;
 
@@ -245,7 +245,7 @@ static void signal_irq_work(struct irq_work *work)
 				i915_request_put(rq);
 		}
 	}
-	atomic_dec(&b->signaler_active);
+	spin_unlock(&b->signaler_active_sync);
 	rcu_read_unlock();
 
 	llist_for_each_safe(signal, sn, signal) {
@@ -290,6 +290,7 @@ intel_breadcrumbs_create(struct intel_engine_cs *irq_engine)
 	init_llist_head(&b->signaled_requests);
 
 	spin_lock_init(&b->irq_lock);
+	spin_lock_init(&b->signaler_active_sync);
 	init_irq_work(&b->irq_work, signal_irq_work);
 
 	b->irq_engine = irq_engine;
@@ -487,8 +488,10 @@ void intel_context_remove_breadcrumbs(struct intel_context *ce,
 	if (release)
 		intel_context_put(ce);
 
-	while (atomic_read(&b->signaler_active))
-		cpu_relax();
+	while (spin_is_locked(&b->signaler_active_sync)) {
+		spin_lock_irqsave(&b->signaler_active_sync, flags);
+		spin_unlock_irqrestore(&b->signaler_active_sync, flags);

And this does not work because spin_is_locked() returns true and spins
forever. This fails because there is a "corner case" where
spin_is_locked() returns but the lock has no lock owner as in locked.
This happens if there is a waiter which did not yet acquire the lock. 

So if you happy with this, we could keep it ;)

+	}
 }
 
 static void print_signals(struct intel_breadcrumbs *b, struct drm_printer *p)
diff --git a/drivers/gpu/drm/i915/gt/intel_breadcrumbs_types.h b/drivers/gpu/drm/i915/gt/intel_breadcrumbs_types.h
index bdf09fd67b6e7..d071f4f1b1910 100644
--- a/drivers/gpu/drm/i915/gt/intel_breadcrumbs_types.h
+++ b/drivers/gpu/drm/i915/gt/intel_breadcrumbs_types.h
@@ -39,7 +39,7 @@ struct intel_breadcrumbs {
 	spinlock_t signalers_lock; /* protects the list of signalers */
 	struct list_head signalers;
 	struct llist_head signaled_requests;
-	atomic_t signaler_active;
+	spinlock_t signaler_active_sync;
 
 	spinlock_t irq_lock; /* protects the interrupt from hardirq context */
 	struct irq_work irq_work; /* for use from inside irq_lock */


Sebastian

Re: [PATCH v7 26/26] drm/i915/gt: Add a spinlock to prevent starvation of irq_work.

[Sebastian Andrzej Siewior]

On 2026-03-10 12:57:08 [+0100], Maarten Lankhorst wrote:
> --- a/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
> +++ b/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
> @@ -487,8 +490,11 @@ void intel_context_remove_breadcrumbs(struct intel_context *ce,
>  	if (release)
>  		intel_context_put(ce);
>  
> -	while (atomic_read(&b->signaler_active))
> +	while (atomic_read(&b->signaler_active)) {
> +		spin_lock(&b->signaler_active_sync);
> +		spin_unlock(&b->signaler_active_sync);
>  		cpu_relax();

Forgot to mention: This is fine on PREEMPT_RT but on !RT you might need

               spin_lock_irqsave(&b->signaler_active_sync, flags);
               spin_unlock_irqrestore(&b->signaler_active_sync, flags);

if the function can be called from !IRQ context which might be the case
due to the irqsave() earlier in this function.

> +	}
>  }

Sebastian

Re: [PATCH v7 26/26] drm/i915/gt: Add a spinlock to prevent starvation of irq_work.

[Maarten Lankhorst]

Hey,

Den 2026-03-10 kl. 18:04, skrev Sebastian Andrzej Siewior:
> On 2026-03-10 12:57:08 [+0100], Maarten Lankhorst wrote:
>> From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
>>
>> IRQ-Work (FIFO-1) will be preempted by the threaded-interrupt (FIFO-50)
>> and the interrupt will poll on signaler_active while the irq-work can't
>> make progress.
> 
> The threaded-interrupt is the interrupt.
> 
> | On PREEMPT_RT the irq_work can be preempted by threaded-interrupt which
> | will be poll for completion but the irq_work routine can't make
> | progress.
> 
>> Solve this by adding a spinlock to prevent starvation and force
>> completion.
> 
> | Solve this by adding a spinlock_t to prevent starvation by forcing a
> | context switch if lock is held based on `signaler_active'. On
> | !PREEMPT_RT `signaler_active' can only be non-zero if multiple CPUs are
> | involved and spinning on the lock leds to the same result.
> 
>> Signed-off-by: Maarten Lankhorst <dev@lankhorst.se>
> 
> If I am the From: then I should have the Signed-off-by, too. Let me do
> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
> 
> so it can be picked up.
> 
> You did suggest the following:
> 
> --- a/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
> +++ b/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
> @@ -209,7 +209,7 @@ static void signal_irq_work(struct irq_work *work)
>  		intel_breadcrumbs_disarm_irq(b);
>  
>  	rcu_read_lock();
> -	atomic_inc(&b->signaler_active);
> +	spin_lock(&b->signaler_active_sync);
>  	list_for_each_entry_rcu(ce, &b->signalers, signal_link) {
>  		struct i915_request *rq;
>  
> @@ -245,7 +245,7 @@ static void signal_irq_work(struct irq_work *work)
>  				i915_request_put(rq);
>  		}
>  	}
> -	atomic_dec(&b->signaler_active);
> +	spin_unlock(&b->signaler_active_sync);
>  	rcu_read_unlock();
>  
>  	llist_for_each_safe(signal, sn, signal) {
> @@ -290,6 +290,7 @@ intel_breadcrumbs_create(struct intel_engine_cs *irq_engine)
>  	init_llist_head(&b->signaled_requests);
>  
>  	spin_lock_init(&b->irq_lock);
> +	spin_lock_init(&b->signaler_active_sync);
>  	init_irq_work(&b->irq_work, signal_irq_work);
>  
>  	b->irq_engine = irq_engine;
> @@ -487,8 +488,10 @@ void intel_context_remove_breadcrumbs(struct intel_context *ce,
>  	if (release)
>  		intel_context_put(ce);
>  
> -	while (atomic_read(&b->signaler_active))
> -		cpu_relax();
> +	while (spin_is_locked(&b->signaler_active_sync)) {
> +		spin_lock_irqsave(&b->signaler_active_sync, flags);
> +		spin_unlock_irqrestore(&b->signaler_active_sync, flags);
> 
> And this does not work because spin_is_locked() returns true and spins
> forever. This fails because there is a "corner case" where
> spin_is_locked() returns but the lock has no lock owner as in locked.
> This happens if there is a waiter which did not yet acquire the lock. 
> 
> So if you happy with this, we could keep it ;)

It seems CI is a lot happier too now.

Xe:
https://patchwork.freedesktop.org/series/159034/#rev14

BAT passes, the full run has some minor issues but only kms_vblank systematic.
Likely due to the fundamental changes of the PREEMPT_RT kernel itself, nothing driver specific.

i915:
https://patchwork.freedesktop.org/series/159035/#rev14

A few new warnings, and some noise. The most worrying part is the i915 execlists selftest
failing on nearly all platforms:

<3>[  504.279024] i915/intel_execlists_live_selftests: live_preempt_user failed with error -62
...
<7>[  506.799685] [IGT] i915_selftest: finished subtest execlists, FAIL

I don't know what's going on there yet, likely needs more debugging. Could be the test itself
being written incorrectly or something else entirely.

Otherwise things are looking good! Have you uncovered anything else?

Kind regards,
~Maarten Lankhorst

Re: [PATCH v7 26/26] drm/i915/gt: Add a spinlock to prevent starvation of irq_work.

[Maarten Lankhorst]

Hey,

Den 2026-03-10 kl. 18:22, skrev Sebastian Andrzej Siewior:
> On 2026-03-10 12:57:08 [+0100], Maarten Lankhorst wrote:
>> --- a/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
>> +++ b/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
>> @@ -487,8 +490,11 @@ void intel_context_remove_breadcrumbs(struct intel_context *ce,
>>  	if (release)
>>  		intel_context_put(ce);
>>  
>> -	while (atomic_read(&b->signaler_active))
>> +	while (atomic_read(&b->signaler_active)) {
>> +		spin_lock(&b->signaler_active_sync);
>> +		spin_unlock(&b->signaler_active_sync);
>>  		cpu_relax();
> 
> Forgot to mention: This is fine on PREEMPT_RT but on !RT you might need
> 
>                spin_lock_irqsave(&b->signaler_active_sync, flags);
>                spin_unlock_irqrestore(&b->signaler_active_sync, flags);
> 
> if the function can be called from !IRQ context which might be the case
> due to the irqsave() earlier in this function.
Yeah seems a bit of an oversight. I'll change it to use flags.

Is it correct to assume that irq_work disables interrupts on !RT?

Kind regards,
~Maarten Lankhorst

Claude review: drm/i915/display: All patches to make PREEMPT_RT work on i915 + xe.

[Claude Code Review Bot]

Overall Series Review

Subject: drm/i915/display: All patches to make PREEMPT_RT work on i915 + xe.
Author: Maarten Lankhorst <dev@lankhorst.se>
Patches: 31
Reviewed: 2026-03-11T13:21:17.008561

---

This is v7 of a 26-patch series to make the i915 (and xe) display driver work with PREEMPT_RT kernels. The core strategy is sound: instead of converting all vblank-related spinlocks to raw spinlocks, the series introduces a 2-stage arming mechanism for vblank events and vblank work, so that the critical section with interrupts disabled only needs to perform a simple `WRITE_ONCE()` rather than acquire sleeping locks.

The series covers:
1. **DRM core changes** (patches 1-2): 2-stage vblank work/event arming
2. **Display critical section rework** (patches 3-7): Converting i915 display to use the new 2-stage mechanism, removing locks from the vblank evade critical section
3. **RT-specific fixes** (patches 8-18): `_fw` register access variants, lock conversions, RT-aware busy loops
4. **Selftests/config** (patches 19-26): Selftest fixes, removing `!PREEMPT_RT` dependency, CI hacks

**Strengths:** The overall approach of 2-stage arming is elegant and avoids a massive raw spinlock conversion. Many individual fixes (LPE audio, execlists locking, GUC busy loop) are well-motivated with lockdep splats.

**Concerns:**
- Patches 24-25 are clearly FOR-CI/debug patches that should not be merged. Patch 24 force-enables PREEMPT_RT and removes the vblank evade debug config. Patch 25 bumps lockdep stack trace limits as a CI workaround.
- Patch 21 has a TODO in its commit message, suggesting unfinished work.
- Several patches lack Signed-off-by chains or commit message body (patch 19).
- The mbox ordering doesn't match the patch numbering, which may cause confusion.

---
Generated by Claude Code Patch Reviewer