drm/i915: Flush request queue when waiting for ring space
authorChris Wilson <chris@chris-wilson.co.uk>
Mon, 5 May 2014 08:07:33 +0000 (09:07 +0100)
committerDaniel Vetter <daniel.vetter@ffwll.ch>
Wed, 7 May 2014 23:23:34 +0000 (01:23 +0200)
During the review of

commit 1f70999f9052f5a1b0ce1a55aff3808f2ec9fe42
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Mon Jan 27 22:43:07 2014 +0000

    drm/i915: Prevent recursion by retiring requests when the ring is full

Ville raised the point that our interaction with request->tail was
likely to foul up other uses elsewhere (such as hang check comparing
ACTHD against requests).

However, we also need to restore the implicit retire requests that certain
test cases depend upon (e.g. igt/gem_exec_lut_handle), this raises the
spectre that the ppgtt will randomly call i915_gpu_idle() and recurse
back into intel_ring_begin().

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=78023
Reviewed-by: Brad Volkin <bradley.d.volkin@intel.com>
[danvet: Remove now unused 'tail' variable as spotted by Brad.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drivers/gpu/drm/i915/i915_drv.h
drivers/gpu/drm/i915/i915_gem.c
drivers/gpu/drm/i915/intel_ringbuffer.c

index e3e499bfe6eecf3091ab94fa7045af7a9c9defc9..bd600db2349e30820380618627ef6437bb1dd47a 100644 (file)
@@ -2215,6 +2215,7 @@ struct drm_i915_gem_request *
 i915_gem_find_active_request(struct intel_ring_buffer *ring);
 
 bool i915_gem_retire_requests(struct drm_device *dev);
+void i915_gem_retire_requests_ring(struct intel_ring_buffer *ring);
 int __must_check i915_gem_check_wedge(struct i915_gpu_error *error,
                                      bool interruptible);
 static inline bool i915_reset_in_progress(struct i915_gpu_error *error)
index 8fd18246a1580ec23a99a91345741b0a38d8dd2d..a000a8bf955be357749272e45a64ea860ccbc17b 100644 (file)
@@ -64,7 +64,6 @@ static unsigned long i915_gem_inactive_scan(struct shrinker *shrinker,
 static unsigned long i915_gem_purge(struct drm_i915_private *dev_priv, long target);
 static unsigned long i915_gem_shrink_all(struct drm_i915_private *dev_priv);
 static void i915_gem_object_truncate(struct drm_i915_gem_object *obj);
-static void i915_gem_retire_requests_ring(struct intel_ring_buffer *ring);
 
 static bool cpu_cache_is_coherent(struct drm_device *dev,
                                  enum i915_cache_level level)
@@ -2448,7 +2447,7 @@ void i915_gem_reset(struct drm_device *dev)
 /**
  * This function clears the request list as sequence numbers are passed.
  */
-static void
+void
 i915_gem_retire_requests_ring(struct intel_ring_buffer *ring)
 {
        uint32_t seqno;
index e6c7a4dfea51f25c61aaa3a0aa4063da0bb57850..9907d66becddf41992b367ddaf7ebfbcab081d08 100644 (file)
  */
 #define CACHELINE_BYTES 64
 
-static inline int ring_space(struct intel_ring_buffer *ring)
+static inline int __ring_space(int head, int tail, int size)
 {
-       int space = (ring->head & HEAD_ADDR) - (ring->tail + I915_RING_FREE_SPACE);
+       int space = head - (tail + I915_RING_FREE_SPACE);
        if (space < 0)
-               space += ring->size;
+               space += size;
        return space;
 }
 
+static inline int ring_space(struct intel_ring_buffer *ring)
+{
+       return __ring_space(ring->head & HEAD_ADDR, ring->tail, ring->size);
+}
+
 static bool intel_ring_stopped(struct intel_ring_buffer *ring)
 {
        struct drm_i915_private *dev_priv = ring->dev->dev_private;
@@ -1482,7 +1487,7 @@ void intel_cleanup_ring_buffer(struct intel_ring_buffer *ring)
 static int intel_ring_wait_request(struct intel_ring_buffer *ring, int n)
 {
        struct drm_i915_gem_request *request;
-       u32 seqno = 0, tail;
+       u32 seqno = 0;
        int ret;
 
        if (ring->last_retired_head != -1) {
@@ -1495,26 +1500,10 @@ static int intel_ring_wait_request(struct intel_ring_buffer *ring, int n)
        }
 
        list_for_each_entry(request, &ring->request_list, list) {
-               int space;
-
-               if (request->tail == -1)
-                       continue;
-
-               space = request->tail - (ring->tail + I915_RING_FREE_SPACE);
-               if (space < 0)
-                       space += ring->size;
-               if (space >= n) {
+               if (__ring_space(request->tail, ring->tail, ring->size) >= n) {
                        seqno = request->seqno;
-                       tail = request->tail;
                        break;
                }
-
-               /* Consume this request in case we need more space than
-                * is available and so need to prevent a race between
-                * updating last_retired_head and direct reads of
-                * I915_RING_HEAD. It also provides a nice sanity check.
-                */
-               request->tail = -1;
        }
 
        if (seqno == 0)
@@ -1524,11 +1513,11 @@ static int intel_ring_wait_request(struct intel_ring_buffer *ring, int n)
        if (ret)
                return ret;
 
-       ring->head = tail;
-       ring->space = ring_space(ring);
-       if (WARN_ON(ring->space < n))
-               return -ENOSPC;
+       i915_gem_retire_requests_ring(ring);
+       ring->head = ring->last_retired_head;
+       ring->last_retired_head = -1;
 
+       ring->space = ring_space(ring);
        return 0;
 }