drm/vblank: Add vblank works

Add some kind of vblank workers. The interface is similar to regular
delayed works, and is mostly based off kthread_work. It allows for
scheduling delayed works that execute once a particular vblank sequence
has passed. It also allows for accurate flushing of scheduled vblank
works - in that flushing waits for both the vblank sequence and job
execution to complete, or for the work to get cancelled - whichever
comes first.

Whatever hardware programming we do in the work must be fast (must at
least complete during the vblank or scanout period, sometimes during the
first few scanlines of the vblank). As such we use a high-priority
per-CRTC thread to accomplish this.

Changes since v7:
* Stuff drm_vblank_internal.h and drm_vblank_work_internal.h contents
  into drm_internal.h
* Get rid of unnecessary spinlock in drm_crtc_vblank_on()
* Remove !vblank->worker check
* Grab vbl_lock in drm_vblank_work_schedule()
* Mention self-rearming work items in drm_vblank_work_schedule() kdocs
* Return 1 from drm_vblank_work_schedule() if the work was scheduled
  successfully, 0 or error code otherwise
* Use drm_dbg_core() instead of DRM_DEV_ERROR() in
  drm_vblank_work_schedule()
* Remove vblank->worker checks in drm_vblank_destroy_worker() and
  drm_vblank_flush_worker()
Changes since v6:
* Get rid of ->pending and seqcounts, and implement flushing through
  simpler means - danvet
* Get rid of work_lock, just use drm_device->event_lock
* Move drm_vblank_work item cleanup into drm_crtc_vblank_off() so that
  we ensure that all vblank work has finished before disabling vblanks
* Add checks into drm_crtc_vblank_reset() so we yell if it gets called
  while there's vblank workers active
* Grab event_lock in both drm_crtc_vblank_on()/drm_crtc_vblank_off(),
  the main reason for this is so that other threads calling
  drm_vblank_work_schedule() are blocked from attempting to schedule
  while we're in the middle of enabling/disabling vblanks.
* Move drm_handle_vblank_works() call below drm_handle_vblank_events()
* Simplify drm_vblank_work_cancel_sync()
* Fix drm_vblank_work_cancel_sync() documentation
* Move wake_up_all() calls out of spinlock where we can. The only one I
  left was the call to wake_up_all() in drm_vblank_handle_works() as
  this seemed like it made more sense just living in that function
  (which is all technically under lock)
* Move drm_vblank_work related functions into their own source files
* Add drm_vblank_internal.h so we can export some functions we don't
  want drivers using, but that we do need to use in drm_vblank_work.c
* Add a bunch of documentation
Changes since v4:
* Get rid of kthread interfaces we tried adding and move all of the
  locking into drm_vblank.c. For implementing drm_vblank_work_flush(),
  we now use a wait_queue and sequence counters in order to
  differentiate between multiple work item executions.
* Get rid of drm_vblank_work_cancel() - this would have been pretty
  difficult to actually reimplement and it occurred to me that neither
  nouveau or i915 are even planning to use this function. Since there's
  also no async cancel function for most of the work interfaces in the
  kernel, it seems a bit unnecessary anyway.
* Get rid of to_drm_vblank_work() since we now are also able to just
  pass the struct drm_vblank_work to work item callbacks anyway
Changes since v3:
* Use our own spinlocks, don't integrate so tightly with kthread_works
Changes since v2:
* Use kthread_workers instead of reinventing the wheel.

Cc: Tejun Heo <tj@kernel.org>
Cc: dri-devel@lists.freedesktop.org
Cc: nouveau@lists.freedesktop.org
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Co-developed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Lyude Paul <lyude@redhat.com>
Acked-by: Dave Airlie <airlied@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200627194657.156514-4-lyude@redhat.com

1 files changed, 267 insertions, 0 deletions

diff --git a/drivers/gpu/drm/drm_vblank_work.c b/drivers/gpu/drm/drm_vblank_work.c
new file mode 100644
index 000000000000..7ac0fc0a9415
--- /dev/null
+++ b/drivers/gpu/drm/drm_vblank_work.c
@@ -0,0 +1,267 @@
+// SPDX-License-Identifier: MIT
+
+#include <uapi/linux/sched/types.h>
+
+#include <drm/drm_print.h>
+#include <drm/drm_vblank.h>
+#include <drm/drm_vblank_work.h>
+#include <drm/drm_crtc.h>
+
+#include "drm_internal.h"
+
+/**
+ * DOC: vblank works
+ *
+ * Many DRM drivers need to program hardware in a time-sensitive manner, many
+ * times with a deadline of starting and finishing within a certain region of
+ * the scanout. Most of the time the safest way to accomplish this is to
+ * simply do said time-sensitive programming in the driver's IRQ handler,
+ * which allows drivers to avoid being preempted during these critical
+ * regions. Or even better, the hardware may even handle applying such
+ * time-critical programming independently of the CPU.
+ *
+ * While there's a decent amount of hardware that's designed so that the CPU
+ * doesn't need to be concerned with extremely time-sensitive programming,
+ * there's a few situations where it can't be helped. Some unforgiving
+ * hardware may require that certain time-sensitive programming be handled
+ * completely by the CPU, and said programming may even take too long to
+ * handle in an IRQ handler. Another such situation would be where the driver
+ * needs to perform a task that needs to complete within a specific scanout
+ * period, but might possibly block and thus cannot be handled in an IRQ
+ * context. Both of these situations can't be solved perfectly in Linux since
+ * we're not a realtime kernel, and thus the scheduler may cause us to miss
+ * our deadline if it decides to preempt us. But for some drivers, it's good
+ * enough if we can lower our chance of being preempted to an absolute
+ * minimum.
+ *
+ * This is where &drm_vblank_work comes in. &drm_vblank_work provides a simple
+ * generic delayed work implementation which delays work execution until a
+ * particular vblank has passed, and then executes the work at realtime
+ * priority. This provides the best possible chance at performing
+ * time-sensitive hardware programming on time, even when the system is under
+ * heavy load. &drm_vblank_work also supports rescheduling, so that self
+ * re-arming work items can be easily implemented.
+ */
+
+void drm_handle_vblank_works(struct drm_vblank_crtc *vblank)
+{
+	struct drm_vblank_work *work, *next;
+	u64 count = atomic64_read(&vblank->count);
+	bool wake = false;
+
+	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))
+			continue;
+
+		list_del_init(&work->node);
+		drm_vblank_put(vblank->dev, vblank->pipe);
+		kthread_queue_work(vblank->worker, &work->base);
+		wake = true;
+	}
+	if (wake)
+		wake_up_all(&vblank->work_wait_queue);
+}
+
+/* Handle cancelling any pending vblank work items and drop respective vblank
+ * references in response to vblank interrupts being disabled.
+ */
+void drm_vblank_cancel_pending_works(struct drm_vblank_crtc *vblank)
+{
+	struct drm_vblank_work *work, *next;
+
+	assert_spin_locked(&vblank->dev->event_lock);
+
+	list_for_each_entry_safe(work, next, &vblank->pending_work, node) {
+		list_del_init(&work->node);
+		drm_vblank_put(vblank->dev, vblank->pipe);
+	}
+
+	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)
+{
+	struct drm_vblank_crtc *vblank = work->vblank;
+	struct drm_device *dev = vblank->dev;
+	u64 cur_vbl;
+	unsigned long irqflags;
+	bool passed, inmodeset, rescheduling = false, wake = false;
+	int ret = 0;
+
+	spin_lock_irqsave(&dev->event_lock, irqflags);
+	if (work->cancelling)
+		goto out;
+
+	spin_lock(&dev->vbl_lock);
+	inmodeset = vblank->inmodeset;
+	spin_unlock(&dev->vbl_lock);
+	if (inmodeset)
+		goto out;
+
+	if (list_empty(&work->node)) {
+		ret = drm_vblank_get(dev, vblank->pipe);
+		if (ret < 0)
+			goto out;
+	} else if (work->count == count) {
+		/* Already scheduled w/ same vbl count */
+		goto out;
+	} else {
+		rescheduling = true;
+	}
+
+	work->count = count;
+	cur_vbl = drm_vblank_count(dev, vblank->pipe);
+	passed = drm_vblank_passed(cur_vbl, count);
+	if (passed)
+		drm_dbg_core(dev,
+			     "crtc %d vblank %llu already passed (current %llu)\n",
+			     vblank->pipe, count, cur_vbl);
+
+	if (!nextonmiss && passed) {
+		drm_vblank_put(dev, vblank->pipe);
+		ret = kthread_queue_work(vblank->worker, &work->base);
+
+		if (rescheduling) {
+			list_del_init(&work->node);
+			wake = true;
+		}
+	} else {
+		if (!rescheduling)
+			list_add_tail(&work->node, &vblank->pending_work);
+		ret = true;
+	}
+
+out:
+	spin_unlock_irqrestore(&dev->event_lock, irqflags);
+	if (wake)
+		wake_up_all(&vblank->work_wait_queue);
+	return ret;
+}
+EXPORT_SYMBOL(drm_vblank_work_schedule);
+
+/**
+ * drm_vblank_work_cancel_sync - cancel a vblank work and wait for it to
+ * finish executing
+ * @work: vblank work to cancel
+ *
+ * Cancel an already scheduled vblank work and wait for its
+ * execution to finish.
+ *
+ * On return, @work is guaranteed to no longer be scheduled or running, even
+ * if it's self-arming.
+ *
+ * Returns:
+ * %True if the work was cancelled before it started to execute, %false
+ * otherwise.
+ */
+bool drm_vblank_work_cancel_sync(struct drm_vblank_work *work)
+{
+	struct drm_vblank_crtc *vblank = work->vblank;
+	struct drm_device *dev = vblank->dev;
+	bool ret = false;
+
+	spin_lock_irq(&dev->event_lock);
+	if (!list_empty(&work->node)) {
+		list_del_init(&work->node);
+		drm_vblank_put(vblank->dev, vblank->pipe);
+		ret = true;
+	}
+
+	work->cancelling++;
+	spin_unlock_irq(&dev->event_lock);
+
+	wake_up_all(&vblank->work_wait_queue);
+
+	if (kthread_cancel_work_sync(&work->base))
+		ret = true;
+
+	spin_lock_irq(&dev->event_lock);
+	work->cancelling--;
+	spin_unlock_irq(&dev->event_lock);
+
+	return ret;
+}
+EXPORT_SYMBOL(drm_vblank_work_cancel_sync);
+
+/**
+ * drm_vblank_work_flush - wait for a scheduled vblank work to finish
+ * executing
+ * @work: vblank work to flush
+ *
+ * Wait until @work has finished executing once.
+ */
+void drm_vblank_work_flush(struct drm_vblank_work *work)
+{
+	struct drm_vblank_crtc *vblank = work->vblank;
+	struct drm_device *dev = vblank->dev;
+
+	spin_lock_irq(&dev->event_lock);
+	wait_event_lock_irq(vblank->work_wait_queue, list_empty(&work->node),
+			    dev->event_lock);
+	spin_unlock_irq(&dev->event_lock);
+
+	kthread_flush_work(&work->base);
+}
+EXPORT_SYMBOL(drm_vblank_work_flush);
+
+/**
+ * drm_vblank_work_init - initialize a vblank work item
+ * @work: vblank work item
+ * @crtc: CRTC whose vblank will trigger the work execution
+ * @func: work function to be executed
+ *
+ * Initialize a vblank work item for a specific crtc.
+ */
+void drm_vblank_work_init(struct drm_vblank_work *work, struct drm_crtc *crtc,
+			  void (*func)(struct kthread_work *work))
+{
+	kthread_init_work(&work->base, func);
+	INIT_LIST_HEAD(&work->node);
+	work->vblank = &crtc->dev->vblank[drm_crtc_index(crtc)];
+}
+EXPORT_SYMBOL(drm_vblank_work_init);
+
+int drm_vblank_worker_init(struct drm_vblank_crtc *vblank)
+{
+	struct sched_param param = {
+		.sched_priority = MAX_RT_PRIO - 1,
+	};
+	struct kthread_worker *worker;
+
+	INIT_LIST_HEAD(&vblank->pending_work);
+	init_waitqueue_head(&vblank->work_wait_queue);
+	worker = kthread_create_worker(0, "card%d-crtc%d",
+				       vblank->dev->primary->index,
+				       vblank->pipe);
+	if (IS_ERR(worker))
+		return PTR_ERR(worker);
+
+	vblank->worker = worker;
+
+	return sched_setscheduler(vblank->worker->task, SCHED_FIFO, &param);
+}