This makes the function much simpler since the complex retry logic is now handled else where.

Signed-off-by: Christian König christian.koenig@amd.com Acked-by: Daniel Vetter daniel@ffwll.ch --- drivers/gpu/drm/i915/gem/i915_gem_busy.c | 35 ++++++++++-------------- 1 file changed, 14 insertions(+), 21 deletions(-)

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_busy.c b/drivers/gpu/drm/i915/gem/i915_gem_busy.c index 7358bebef15c..470fdfd61a0f 100644 --- a/drivers/gpu/drm/i915/gem/i915_gem_busy.c +++ b/drivers/gpu/drm/i915/gem/i915_gem_busy.c @@ -115,8 +115,8 @@ i915_gem_busy_ioctl(struct drm_device *dev, void *data, { struct drm_i915_gem_busy *args = data; struct drm_i915_gem_object *obj; - struct dma_resv_list *list; - unsigned int seq; + struct dma_resv_iter cursor; + struct dma_fence *fence; int err;

err = -ENOENT; @@ -142,27 +142,20 @@ i915_gem_busy_ioctl(struct drm_device *dev, void *data, * to report the overall busyness. This is what the wait-ioctl does. * */ -retry: - seq = raw_read_seqcount(&obj->base.resv->seq); - - /* Translate the exclusive fence to the READ *and* WRITE engine */ - args->busy = busy_check_writer(dma_resv_excl_fence(obj->base.resv)); - - /* Translate shared fences to READ set of engines */ - list = dma_resv_shared_list(obj->base.resv); - if (list) { - unsigned int shared_count = list->shared_count, i; - - for (i = 0; i < shared_count; ++i) { - struct dma_fence *fence = - rcu_dereference(list->shared[i]); - + args->busy = 0; + dma_resv_iter_begin(&cursor, obj->base.resv, true); + dma_resv_for_each_fence_unlocked(&cursor, fence) { + if (dma_resv_iter_is_restarted(&cursor)) + args->busy = 0; + + if (dma_resv_iter_is_exclusive(&cursor)) + /* Translate the exclusive fence to the READ *and* WRITE engine */ + args->busy |= busy_check_writer(fence); + else + /* Translate shared fences to READ set of engines */ args->busy |= busy_check_reader(fence); - } } - - if (args->busy && read_seqcount_retry(&obj->base.resv->seq, seq)) - goto retry; + dma_resv_iter_end(&cursor);

err = 0; out:

Simplifying the code a bit.

v2: use dma_resv_for_each_fence instead, according to Tvrtko the lock is held here anyway. v3: back to using dma_resv_for_each_fence_unlocked.

Signed-off-by: Christian König christian.koenig@amd.com Acked-by: Daniel Vetter daniel@ffwll.ch --- drivers/gpu/drm/i915/i915_sw_fence.c | 53 ++++++---------------------- 1 file changed, 11 insertions(+), 42 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_sw_fence.c b/drivers/gpu/drm/i915/i915_sw_fence.c index c589a681da77..7ea0dbf81530 100644 --- a/drivers/gpu/drm/i915/i915_sw_fence.c +++ b/drivers/gpu/drm/i915/i915_sw_fence.c @@ -572,56 +572,25 @@ int i915_sw_fence_await_reservation(struct i915_sw_fence *fence, unsigned long timeout, gfp_t gfp) { - struct dma_fence *excl; + struct dma_resv_iter cursor; + struct dma_fence *f; int ret = 0, pending;

debug_fence_assert(fence); might_sleep_if(gfpflags_allow_blocking(gfp));

- if (write) { - struct dma_fence **shared; - unsigned int count, i; - - ret = dma_resv_get_fences(resv, &excl, &count, &shared); - if (ret) - return ret; - - for (i = 0; i < count; i++) { - if (shared[i]->ops == exclude) - continue; - - pending = i915_sw_fence_await_dma_fence(fence, - shared[i], - timeout, - gfp); - if (pending < 0) { - ret = pending; - break; - } - - ret |= pending; - } - - for (i = 0; i < count; i++) - dma_fence_put(shared[i]); - kfree(shared); - } else { - excl = dma_resv_get_excl_unlocked(resv); - } - - if (ret >= 0 && excl && excl->ops != exclude) { - pending = i915_sw_fence_await_dma_fence(fence, - excl, - timeout, + dma_resv_iter_begin(&cursor, resv, write); + dma_resv_for_each_fence_unlocked(&cursor, f) { + pending = i915_sw_fence_await_dma_fence(fence, f, timeout, gfp); - if (pending < 0) + if (pending < 0) { ret = pending; - else - ret |= pending; - } - - dma_fence_put(excl); + break; + }

+ ret |= pending; + } + dma_resv_iter_end(&cursor); return ret; }

From: Maarten Lankhorst maarten.lankhorst@linux.intel.com

The i915_request fence wait behaves differently for timeout = 0 compared to expected dma-fence behavior.

i915 behavior: - Unsignaled: -ETIME - Signaled: 0 (= timeout)

Expected: - Unsignaled: 0 - Signaled: 1

Signed-off-by: Maarten Lankhorst maarten.lankhorst@linux.intel.com Acked-by: Daniel Vetter daniel@ffwll.ch Acked-by: Christian König christian.koenig@amd.com --- drivers/gpu/drm/i915/i915_request.c | 57 ++++++++++++++++++++++++----- drivers/gpu/drm/i915/i915_request.h | 5 +++ 2 files changed, 52 insertions(+), 10 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c index 820a1f38b271..42cd17357771 100644 --- a/drivers/gpu/drm/i915/i915_request.c +++ b/drivers/gpu/drm/i915/i915_request.c @@ -96,9 +96,9 @@ static signed long i915_fence_wait(struct dma_fence *fence, bool interruptible, signed long timeout) { - return i915_request_wait(to_request(fence), - interruptible | I915_WAIT_PRIORITY, - timeout); + return i915_request_wait_timeout(to_request(fence), + interruptible | I915_WAIT_PRIORITY, + timeout); }

struct kmem_cache *i915_request_slab_cache(void) @@ -1857,23 +1857,27 @@ static void request_wait_wake(struct dma_fence *fence, struct dma_fence_cb *cb) }

/** - * i915_request_wait - wait until execution of request has finished + * i915_request_wait_timeout - wait until execution of request has finished * @rq: the request to wait upon * @flags: how to wait * @timeout: how long to wait in jiffies * - * i915_request_wait() waits for the request to be completed, for a + * i915_request_wait_timeout() waits for the request to be completed, for a * maximum of @timeout jiffies (with MAX_SCHEDULE_TIMEOUT implying an * unbounded wait). * * Returns the remaining time (in jiffies) if the request completed, which may - * be zero or -ETIME if the request is unfinished after the timeout expires. + * be zero if the request is unfinished after the timeout expires. + * If the timeout is 0, it will return 1 if the fence is signaled. + * * May return -EINTR is called with I915_WAIT_INTERRUPTIBLE and a signal is * pending before the request completes. + * + * NOTE: This function has the same wait semantics as dma-fence. */ -long i915_request_wait(struct i915_request *rq, - unsigned int flags, - long timeout) +long i915_request_wait_timeout(struct i915_request *rq, + unsigned int flags, + long timeout) { const int state = flags & I915_WAIT_INTERRUPTIBLE ? TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE; @@ -1883,7 +1887,7 @@ long i915_request_wait(struct i915_request *rq, GEM_BUG_ON(timeout < 0);

if (dma_fence_is_signaled(&rq->fence)) - return timeout; + return timeout ?: 1;

if (!timeout) return -ETIME; @@ -1992,6 +1996,39 @@ long i915_request_wait(struct i915_request *rq, return timeout; }

+/** + * i915_request_wait - wait until execution of request has finished + * @rq: the request to wait upon + * @flags: how to wait + * @timeout: how long to wait in jiffies + * + * i915_request_wait() waits for the request to be completed, for a + * maximum of @timeout jiffies (with MAX_SCHEDULE_TIMEOUT implying an + * unbounded wait). + * + * Returns the remaining time (in jiffies) if the request completed, which may + * be zero or -ETIME if the request is unfinished after the timeout expires. + * May return -EINTR is called with I915_WAIT_INTERRUPTIBLE and a signal is + * pending before the request completes. + * + * NOTE: This function behaves differently from dma-fence wait semantics for + * timeout = 0. It returns 0 on success, and -ETIME if not signaled. + */ +long i915_request_wait(struct i915_request *rq, + unsigned int flags, + long timeout) +{ + long ret = i915_request_wait_timeout(rq, flags, timeout); + + if (!ret) + return -ETIME; + + if (ret > 0 && !timeout) + return 0; + + return ret; +} + static int print_sched_attr(const struct i915_sched_attr *attr, char *buf, int x, int len) { diff --git a/drivers/gpu/drm/i915/i915_request.h b/drivers/gpu/drm/i915/i915_request.h index dc359242d1ae..3c6e8acd1457 100644 --- a/drivers/gpu/drm/i915/i915_request.h +++ b/drivers/gpu/drm/i915/i915_request.h @@ -414,6 +414,11 @@ void i915_request_unsubmit(struct i915_request *request);

void i915_request_cancel(struct i915_request *rq, int error);

+long i915_request_wait_timeout(struct i915_request *rq, + unsigned int flags, + long timeout) + __attribute__((nonnull(1))); + long i915_request_wait(struct i915_request *rq, unsigned int flags, long timeout)