drm/i915/breadcrumbs: Tweak commentary
authorChris Wilson <chris@chris-wilson.co.uk>
Wed, 15 Mar 2017 22:22:59 +0000 (22:22 +0000)
committerChris Wilson <chris@chris-wilson.co.uk>
Thu, 16 Mar 2017 08:49:28 +0000 (08:49 +0000)
Tvrtko spotted a stale reference to b->lock (now b->rb_lock) so review
the comments and try to improve them in passing.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170315222259.1469-1-chris@chris-wilson.co.uk
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
drivers/gpu/drm/i915/intel_breadcrumbs.c

index cb6985acc542cbe4c1d664d15c61bfb38e1983d3..ba986edee3127ea24023026692ff8b7082066662 100644 (file)
@@ -85,7 +85,7 @@ static void intel_breadcrumbs_hangcheck(unsigned long data)
                return;
        }
 
-       /* We keep the hangcheck time alive until we disarm the irq, even
+       /* We keep the hangcheck timer alive until we disarm the irq, even
         * if there are no waiters at present.
         *
         * If the waiter was currently running, assume it hasn't had a chance
@@ -110,12 +110,11 @@ static void intel_breadcrumbs_fake_irq(unsigned long data)
        struct intel_engine_cs *engine = (struct intel_engine_cs *)data;
        struct intel_breadcrumbs *b = &engine->breadcrumbs;
 
-       /*
-        * The timer persists in case we cannot enable interrupts,
+       /* The timer persists in case we cannot enable interrupts,
         * or if we have previously seen seqno/interrupt incoherency
-        * ("missed interrupt" syndrome). Here the worker will wake up
-        * every jiffie in order to kick the oldest waiter to do the
-        * coherent seqno check.
+        * ("missed interrupt" syndrome, better known as a "missed breadcrumb").
+        * Here the worker will wake up every jiffie in order to kick the
+        * oldest waiter to do the coherent seqno check.
         */
 
        spin_lock_irq(&b->irq_lock);
@@ -290,7 +289,12 @@ static inline void __intel_breadcrumbs_finish(struct intel_breadcrumbs *b,
        GEM_BUG_ON(b->irq_wait == wait);
 
        /* This request is completed, so remove it from the tree, mark it as
-        * complete, and *then* wake up the associated task.
+        * complete, and *then* wake up the associated task. N.B. when the
+        * task wakes up, it will find the empty rb_node, discern that it
+        * has already been removed from the tree and skip the serialisation
+        * of the b->rb_lock and b->irq_lock. This means that the destruction
+        * of the intel_wait is not serialised with the interrupt handler
+        * by the waiter - it must instead be serialised by the caller.
         */
        rb_erase(&wait->node, &b->waiters);
        RB_CLEAR_NODE(&wait->node);
@@ -397,6 +401,11 @@ static bool __intel_engine_add_wait(struct intel_engine_cs *engine,
        }
 
        if (completed) {
+               /* Advance the bottom-half (b->irq_wait) before we wake up
+                * the waiters who may scribble over their intel_wait
+                * just as the interrupt handler is dereferencing it via
+                * b->irq_wait.
+                */
                if (!first) {
                        struct rb_node *next = rb_next(completed);
                        GEM_BUG_ON(next == &wait->node);
@@ -653,7 +662,7 @@ void intel_engine_enable_signaling(struct drm_i915_gem_request *request)
        /* Note that we may be called from an interrupt handler on another
         * device (e.g. nouveau signaling a fence completion causing us
         * to submit a request, and so enable signaling). As such,
-        * we need to make sure that all other users of b->lock protect
+        * we need to make sure that all other users of b->rb_lock protect
         * against interrupts, i.e. use spin_lock_irqsave.
         */