]> git.baikalelectronics.ru Git - kernel.git/commitdiff
drm/i915: Pass intel_context to i915_request_create()
authorChris Wilson <chris@chris-wilson.co.uk>
Wed, 24 Apr 2019 20:07:16 +0000 (21:07 +0100)
committerChris Wilson <chris@chris-wilson.co.uk>
Wed, 24 Apr 2019 21:25:35 +0000 (22:25 +0100)
Start acquiring the logical intel_context and using that as our primary
means for request allocation. This is the initial step to allow us to
avoid requiring struct_mutex for request allocation along the
perma-pinned kernel context, but it also provides a foundation for
breaking up the complex request allocation to handle different scenarios
inside execbuf.

For the purpose of emitting a request from inside retirement (see the
next patch for engine power management), we also need to lift control
over the timeline mutex to the caller.

v2: Note that the request carries the active reference upon construction.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190424200717.1686-4-chris@chris-wilson.co.uk
14 files changed:
drivers/gpu/drm/i915/gt/intel_context.h
drivers/gpu/drm/i915/gt/intel_reset.c
drivers/gpu/drm/i915/gt/intel_ringbuffer.c
drivers/gpu/drm/i915/gt/intel_workarounds.c
drivers/gpu/drm/i915/i915_gem_context.c
drivers/gpu/drm/i915/i915_perf.c
drivers/gpu/drm/i915/i915_request.c
drivers/gpu/drm/i915/i915_request.h
drivers/gpu/drm/i915/intel_overlay.c
drivers/gpu/drm/i915/selftests/i915_active.c
drivers/gpu/drm/i915/selftests/i915_gem_coherency.c
drivers/gpu/drm/i915/selftests/i915_gem_object.c
drivers/gpu/drm/i915/selftests/i915_request.c
drivers/gpu/drm/i915/selftests/i915_timeline.c

index b732cf99efcb6e834cc1ae2c22f99102117c2cc8..60379eb37949fce3dd7b5903fe14389767e6c8d9 100644 (file)
@@ -105,4 +105,16 @@ static inline void intel_context_put(struct intel_context *ce)
        kref_put(&ce->ref, ce->ops->destroy);
 }
 
+static inline void intel_context_timeline_lock(struct intel_context *ce)
+       __acquires(&ce->ring->timeline->mutex)
+{
+       mutex_lock(&ce->ring->timeline->mutex);
+}
+
+static inline void intel_context_timeline_unlock(struct intel_context *ce)
+       __releases(&ce->ring->timeline->mutex)
+{
+       mutex_unlock(&ce->ring->timeline->mutex);
+}
+
 #endif /* __INTEL_CONTEXT_H__ */
index 7db498567843b3ef06c16c01dd2aadc81561b94f..cdf184403d463937277df0eabbd5e2b600305707 100644 (file)
@@ -783,7 +783,7 @@ static void restart_work(struct work_struct *work)
                if (!intel_engine_is_idle(engine))
                        continue;
 
-               rq = i915_request_alloc(engine, i915->kernel_context);
+               rq = i915_request_create(engine->kernel_context);
                if (!IS_ERR(rq))
                        i915_request_add(rq);
        }
index 5404fe382691a27a46e07eb899f563eee79320d8..f89541274d4454d96c80c6869bfea7ff716f9766 100644 (file)
@@ -1772,7 +1772,6 @@ static int switch_context(struct i915_request *rq)
        u32 hw_flags = 0;
        int ret, i;
 
-       lockdep_assert_held(&rq->i915->drm.struct_mutex);
        GEM_BUG_ON(HAS_EXECLISTS(rq->i915));
 
        if (ppgtt) {
@@ -1902,8 +1901,6 @@ static noinline int wait_for_space(struct intel_ring *ring, unsigned int bytes)
        struct i915_request *target;
        long timeout;
 
-       lockdep_assert_held(&ring->vma->vm->i915->drm.struct_mutex);
-
        if (intel_ring_update_space(ring) >= bytes)
                return 0;
 
index f46ed0e2f07c5c17b07eb2e8a09b4628bd4fc460..364696221fd736bcdbdb67c3dc11f830b545c849 100644 (file)
@@ -1356,7 +1356,7 @@ static int engine_wa_list_verify(struct intel_engine_cs *engine,
        if (IS_ERR(vma))
                return PTR_ERR(vma);
 
-       rq = i915_request_alloc(engine, engine->kernel_context->gem_context);
+       rq = i915_request_create(engine->kernel_context);
        if (IS_ERR(rq)) {
                err = PTR_ERR(rq);
                goto err_vma;
index 37dff694456c2c82535ea354d60e099542a1558c..3eb1a664b5fab349c4e7f68a13867aa5506a3bd5 100644 (file)
@@ -942,7 +942,7 @@ int i915_gem_switch_to_kernel_context(struct drm_i915_private *i915,
                struct intel_ring *ring;
                struct i915_request *rq;
 
-               rq = i915_request_alloc(engine, i915->kernel_context);
+               rq = i915_request_create(engine->kernel_context);
                if (IS_ERR(rq))
                        return PTR_ERR(rq);
 
@@ -1188,7 +1188,7 @@ gen8_modify_rpcs(struct intel_context *ce, struct intel_sseu sseu)
        /* Submitting requests etc needs the hw awake. */
        wakeref = intel_runtime_pm_get(i915);
 
-       rq = i915_request_alloc(ce->engine, i915->kernel_context);
+       rq = i915_request_create(ce->engine->kernel_context);
        if (IS_ERR(rq)) {
                ret = PTR_ERR(rq);
                goto out_put;
index a87f790335c161ff4b98e7fa32d9b4711393f647..328a740e72cba715d84c3637334137c1f0294ab0 100644 (file)
@@ -1762,7 +1762,7 @@ static int gen8_configure_all_contexts(struct drm_i915_private *dev_priv,
         * Apply the configuration by doing one context restore of the edited
         * context image.
         */
-       rq = i915_request_alloc(engine, dev_priv->kernel_context);
+       rq = i915_request_create(engine->kernel_context);
        if (IS_ERR(rq))
                return PTR_ERR(rq);
 
index 9a2665ee012a998a611f698279bc3a278bb2bded..705c125bafc6ed551aaca71c992fed8b14b6e1ae 100644 (file)
@@ -581,7 +581,7 @@ static void ring_retire_requests(struct intel_ring *ring)
 }
 
 static noinline struct i915_request *
-i915_request_alloc_slow(struct intel_context *ce)
+request_alloc_slow(struct intel_context *ce, gfp_t gfp)
 {
        struct intel_ring *ring = ce->ring;
        struct i915_request *rq;
@@ -589,6 +589,9 @@ i915_request_alloc_slow(struct intel_context *ce)
        if (list_empty(&ring->request_list))
                goto out;
 
+       if (!gfpflags_allow_blocking(gfp))
+               goto out;
+
        /* Ratelimit ourselves to prevent oom from malicious clients */
        rq = list_last_entry(&ring->request_list, typeof(*rq), ring_link);
        cond_synchronize_rcu(rq->rcustate);
@@ -597,62 +600,21 @@ i915_request_alloc_slow(struct intel_context *ce)
        ring_retire_requests(ring);
 
 out:
-       return kmem_cache_alloc(global.slab_requests, GFP_KERNEL);
+       return kmem_cache_alloc(global.slab_requests, gfp);
 }
 
-/**
- * i915_request_alloc - allocate a request structure
- *
- * @engine: engine that we wish to issue the request on.
- * @ctx: context that the request will be associated with.
- *
- * Returns a pointer to the allocated request if successful,
- * or an error code if not.
- */
 struct i915_request *
-i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
+__i915_request_create(struct intel_context *ce, gfp_t gfp)
 {
-       struct drm_i915_private *i915 = engine->i915;
-       struct intel_context *ce;
-       struct i915_timeline *tl;
+       struct i915_timeline *tl = ce->ring->timeline;
        struct i915_request *rq;
        u32 seqno;
        int ret;
 
-       lockdep_assert_held(&i915->drm.struct_mutex);
-
-       /*
-        * Preempt contexts are reserved for exclusive use to inject a
-        * preemption context switch. They are never to be used for any trivial
-        * request!
-        */
-       GEM_BUG_ON(ctx == i915->preempt_context);
-
-       /*
-        * ABI: Before userspace accesses the GPU (e.g. execbuffer), report
-        * EIO if the GPU is already wedged.
-        */
-       ret = i915_terminally_wedged(i915);
-       if (ret)
-               return ERR_PTR(ret);
+       might_sleep_if(gfpflags_allow_blocking(gfp));
 
-       /*
-        * Pinning the contexts may generate requests in order to acquire
-        * GGTT space, so do this first before we reserve a seqno for
-        * ourselves.
-        */
-       ce = intel_context_pin(ctx, engine);
-       if (IS_ERR(ce))
-               return ERR_CAST(ce);
-
-       mutex_lock(&ce->ring->timeline->mutex);
-       intel_context_enter(ce);
-
-       /* Move our oldest request to the slab-cache (if not in use!) */
-       rq = list_first_entry(&ce->ring->request_list, typeof(*rq), ring_link);
-       if (!list_is_last(&rq->ring_link, &ce->ring->request_list) &&
-           i915_request_completed(rq))
-               i915_request_retire(rq);
+       /* Check that the caller provided an already pinned context */
+       __intel_context_pin(ce);
 
        /*
         * Beware: Dragons be flying overhead.
@@ -684,30 +646,26 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
         * Do not use kmem_cache_zalloc() here!
         */
        rq = kmem_cache_alloc(global.slab_requests,
-                             GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_NOWARN);
+                             gfp | __GFP_RETRY_MAYFAIL | __GFP_NOWARN);
        if (unlikely(!rq)) {
-               rq = i915_request_alloc_slow(ce);
+               rq = request_alloc_slow(ce, gfp);
                if (!rq) {
                        ret = -ENOMEM;
                        goto err_unreserve;
                }
        }
 
-       INIT_LIST_HEAD(&rq->active_list);
-       INIT_LIST_HEAD(&rq->execute_cb);
-
-       tl = ce->ring->timeline;
        ret = i915_timeline_get_seqno(tl, rq, &seqno);
        if (ret)
                goto err_free;
 
-       rq->i915 = i915;
-       rq->engine = engine;
-       rq->gem_context = ctx;
+       rq->i915 = ce->engine->i915;
        rq->hw_context = ce;
+       rq->gem_context = ce->gem_context;
+       rq->engine = ce->engine;
        rq->ring = ce->ring;
        rq->timeline = tl;
-       GEM_BUG_ON(rq->timeline == &engine->timeline);
+       GEM_BUG_ON(rq->timeline == &ce->engine->timeline);
        rq->hwsp_seqno = tl->hwsp_seqno;
        rq->hwsp_cacheline = tl->hwsp_cacheline;
        rq->rcustate = get_state_synchronize_rcu(); /* acts as smp_mb() */
@@ -728,6 +686,9 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
        rq->capture_list = NULL;
        rq->waitboost = false;
 
+       INIT_LIST_HEAD(&rq->active_list);
+       INIT_LIST_HEAD(&rq->execute_cb);
+
        /*
         * Reserve space in the ring buffer for all the commands required to
         * eventually emit this request. This is to guarantee that the
@@ -740,7 +701,8 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
         * around inside i915_request_add() there is sufficient space at
         * the beginning of the ring as well.
         */
-       rq->reserved_space = 2 * engine->emit_fini_breadcrumb_dw * sizeof(u32);
+       rq->reserved_space =
+               2 * rq->engine->emit_fini_breadcrumb_dw * sizeof(u32);
 
        /*
         * Record the position of the start of the request so that
@@ -750,20 +712,16 @@ i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
         */
        rq->head = rq->ring->emit;
 
-       ret = engine->request_alloc(rq);
+       ret = rq->engine->request_alloc(rq);
        if (ret)
                goto err_unwind;
 
-       /* Keep a second pin for the dual retirement along engine and ring */
-       __intel_context_pin(ce);
-
        rq->infix = rq->ring->emit; /* end of header; start of user payload */
 
-       /* Check that we didn't interrupt ourselves with a new request */
-       lockdep_assert_held(&rq->timeline->mutex);
-       GEM_BUG_ON(rq->timeline->seqno != rq->fence.seqno);
-       rq->cookie = lockdep_pin_lock(&rq->timeline->mutex);
+       /* Keep a second pin for the dual retirement along engine and ring */
+       __intel_context_pin(ce);
 
+       intel_context_mark_active(ce);
        return rq;
 
 err_unwind:
@@ -777,12 +735,86 @@ err_unwind:
 err_free:
        kmem_cache_free(global.slab_requests, rq);
 err_unreserve:
-       intel_context_exit(ce);
-       mutex_unlock(&ce->ring->timeline->mutex);
        intel_context_unpin(ce);
        return ERR_PTR(ret);
 }
 
+struct i915_request *
+i915_request_create(struct intel_context *ce)
+{
+       struct i915_request *rq;
+
+       intel_context_timeline_lock(ce);
+
+       /* Move our oldest request to the slab-cache (if not in use!) */
+       rq = list_first_entry(&ce->ring->request_list, typeof(*rq), ring_link);
+       if (!list_is_last(&rq->ring_link, &ce->ring->request_list) &&
+           i915_request_completed(rq))
+               i915_request_retire(rq);
+
+       intel_context_enter(ce);
+       rq = __i915_request_create(ce, GFP_KERNEL);
+       intel_context_exit(ce); /* active reference transferred to request */
+       if (IS_ERR(rq))
+               goto err_unlock;
+
+       /* Check that we do not interrupt ourselves with a new request */
+       rq->cookie = lockdep_pin_lock(&ce->ring->timeline->mutex);
+
+       return rq;
+
+err_unlock:
+       intel_context_timeline_unlock(ce);
+       return rq;
+}
+
+/**
+ * i915_request_alloc - allocate a request structure
+ *
+ * @engine: engine that we wish to issue the request on.
+ * @ctx: context that the request will be associated with.
+ *
+ * Returns a pointer to the allocated request if successful,
+ * or an error code if not.
+ */
+struct i915_request *
+i915_request_alloc(struct intel_engine_cs *engine, struct i915_gem_context *ctx)
+{
+       struct drm_i915_private *i915 = engine->i915;
+       struct intel_context *ce;
+       struct i915_request *rq;
+       int ret;
+
+       /*
+        * Preempt contexts are reserved for exclusive use to inject a
+        * preemption context switch. They are never to be used for any trivial
+        * request!
+        */
+       GEM_BUG_ON(ctx == i915->preempt_context);
+
+       /*
+        * ABI: Before userspace accesses the GPU (e.g. execbuffer), report
+        * EIO if the GPU is already wedged.
+        */
+       ret = i915_terminally_wedged(i915);
+       if (ret)
+               return ERR_PTR(ret);
+
+       /*
+        * Pinning the contexts may generate requests in order to acquire
+        * GGTT space, so do this first before we reserve a seqno for
+        * ourselves.
+        */
+       ce = intel_context_pin(ctx, engine);
+       if (IS_ERR(ce))
+               return ERR_CAST(ce);
+
+       rq = i915_request_create(ce);
+       intel_context_unpin(ce);
+
+       return rq;
+}
+
 static int
 emit_semaphore_wait(struct i915_request *to,
                    struct i915_request *from,
@@ -1043,8 +1075,7 @@ __i915_request_add_to_timeline(struct i915_request *rq)
         * precludes optimising to use semaphores serialisation of a single
         * timeline across engines.
         */
-       prev = i915_active_request_raw(&timeline->last_request,
-                                      &rq->i915->drm.struct_mutex);
+       prev = rcu_dereference_protected(timeline->last_request.request, 1);
        if (prev && !i915_request_completed(prev)) {
                if (is_power_of_2(prev->engine->mask | rq->engine->mask))
                        i915_sw_fence_await_sw_fence(&rq->submit,
@@ -1065,6 +1096,11 @@ __i915_request_add_to_timeline(struct i915_request *rq)
        list_add_tail(&rq->link, &timeline->requests);
        spin_unlock_irq(&timeline->lock);
 
+       /*
+        * Make sure that no request gazumped us - if it was allocated after
+        * our i915_request_alloc() and called __i915_request_add() before
+        * us, the timeline will hold its seqno which is later than ours.
+        */
        GEM_BUG_ON(timeline->seqno != rq->fence.seqno);
        __i915_active_request_set(&timeline->last_request, rq);
 
@@ -1076,36 +1112,23 @@ __i915_request_add_to_timeline(struct i915_request *rq)
  * request is not being tracked for completion but the work itself is
  * going to happen on the hardware. This would be a Bad Thing(tm).
  */
-void i915_request_add(struct i915_request *request)
+struct i915_request *__i915_request_commit(struct i915_request *rq)
 {
-       struct intel_engine_cs *engine = request->engine;
-       struct i915_timeline *timeline = request->timeline;
-       struct intel_ring *ring = request->ring;
+       struct intel_engine_cs *engine = rq->engine;
+       struct intel_ring *ring = rq->ring;
        struct i915_request *prev;
        u32 *cs;
 
        GEM_TRACE("%s fence %llx:%lld\n",
-                 engine->name, request->fence.context, request->fence.seqno);
-
-       lockdep_assert_held(&request->timeline->mutex);
-       lockdep_unpin_lock(&request->timeline->mutex, request->cookie);
-
-       trace_i915_request_add(request);
-
-       /*
-        * Make sure that no request gazumped us - if it was allocated after
-        * our i915_request_alloc() and called __i915_request_add() before
-        * us, the timeline will hold its seqno which is later than ours.
-        */
-       GEM_BUG_ON(timeline->seqno != request->fence.seqno);
+                 engine->name, rq->fence.context, rq->fence.seqno);
 
        /*
         * To ensure that this call will not fail, space for its emissions
         * should already have been reserved in the ring buffer. Let the ring
         * know that it is time to use that space up.
         */
-       GEM_BUG_ON(request->reserved_space > request->ring->space);
-       request->reserved_space = 0;
+       GEM_BUG_ON(rq->reserved_space > ring->space);
+       rq->reserved_space = 0;
 
        /*
         * Record the position of the start of the breadcrumb so that
@@ -1113,17 +1136,17 @@ void i915_request_add(struct i915_request *request)
         * GPU processing the request, we never over-estimate the
         * position of the ring's HEAD.
         */
-       cs = intel_ring_begin(request, engine->emit_fini_breadcrumb_dw);
+       cs = intel_ring_begin(rq, engine->emit_fini_breadcrumb_dw);
        GEM_BUG_ON(IS_ERR(cs));
-       request->postfix = intel_ring_offset(request, cs);
+       rq->postfix = intel_ring_offset(rq, cs);
 
-       prev = __i915_request_add_to_timeline(request);
+       prev = __i915_request_add_to_timeline(rq);
 
-       list_add_tail(&request->ring_link, &ring->request_list);
-       if (list_is_first(&request->ring_link, &ring->request_list))
-               list_add(&ring->active_link, &request->i915->gt.active_rings);
-       request->i915->gt.active_engines |= request->engine->mask;
-       request->emitted_jiffies = jiffies;
+       list_add_tail(&rq->ring_link, &ring->request_list);
+       if (list_is_first(&rq->ring_link, &ring->request_list))
+               list_add(&ring->active_link, &rq->i915->gt.active_rings);
+       rq->i915->gt.active_engines |= rq->engine->mask;
+       rq->emitted_jiffies = jiffies;
 
        /*
         * Let the backend know a new request has arrived that may need
@@ -1137,10 +1160,10 @@ void i915_request_add(struct i915_request *request)
         * run at the earliest possible convenience.
         */
        local_bh_disable();
-       i915_sw_fence_commit(&request->semaphore);
+       i915_sw_fence_commit(&rq->semaphore);
        rcu_read_lock(); /* RCU serialisation for set-wedged protection */
        if (engine->schedule) {
-               struct i915_sched_attr attr = request->gem_context->sched;
+               struct i915_sched_attr attr = rq->gem_context->sched;
 
                /*
                 * Boost actual workloads past semaphores!
@@ -1154,7 +1177,7 @@ void i915_request_add(struct i915_request *request)
                 * far in the distance past over useful work, we keep a history
                 * of any semaphore use along our dependency chain.
                 */
-               if (!(request->sched.flags & I915_SCHED_HAS_SEMAPHORE_CHAIN))
+               if (!(rq->sched.flags & I915_SCHED_HAS_SEMAPHORE_CHAIN))
                        attr.priority |= I915_PRIORITY_NOSEMAPHORE;
 
                /*
@@ -1163,15 +1186,29 @@ void i915_request_add(struct i915_request *request)
                 * Allow interactive/synchronous clients to jump ahead of
                 * the bulk clients. (FQ_CODEL)
                 */
-               if (list_empty(&request->sched.signalers_list))
+               if (list_empty(&rq->sched.signalers_list))
                        attr.priority |= I915_PRIORITY_NEWCLIENT;
 
-               engine->schedule(request, &attr);
+               engine->schedule(rq, &attr);
        }
        rcu_read_unlock();
-       i915_sw_fence_commit(&request->submit);
+       i915_sw_fence_commit(&rq->submit);
        local_bh_enable(); /* Kick the execlists tasklet if just scheduled */
 
+       return prev;
+}
+
+void i915_request_add(struct i915_request *rq)
+{
+       struct i915_request *prev;
+
+       lockdep_assert_held(&rq->timeline->mutex);
+       lockdep_unpin_lock(&rq->timeline->mutex, rq->cookie);
+
+       trace_i915_request_add(rq);
+
+       prev = __i915_request_commit(rq);
+
        /*
         * In typical scenarios, we do not expect the previous request on
         * the timeline to be still tracked by timeline->last_request if it
@@ -1192,7 +1229,7 @@ void i915_request_add(struct i915_request *request)
        if (prev && i915_request_completed(prev))
                i915_request_retire_upto(prev);
 
-       mutex_unlock(&request->timeline->mutex);
+       mutex_unlock(&rq->timeline->mutex);
 }
 
 static unsigned long local_clock_us(unsigned int *cpu)
index a982664618c2c977d89875dcc1dc4dd575e31106..36f13b74ec5821d3d87879e9b01626bc5e5de4f7 100644 (file)
@@ -239,6 +239,13 @@ static inline bool dma_fence_is_i915(const struct dma_fence *fence)
        return fence->ops == &i915_fence_ops;
 }
 
+struct i915_request * __must_check
+__i915_request_create(struct intel_context *ce, gfp_t gfp);
+struct i915_request * __must_check
+i915_request_create(struct intel_context *ce);
+
+struct i915_request *__i915_request_commit(struct i915_request *request);
+
 struct i915_request * __must_check
 i915_request_alloc(struct intel_engine_cs *engine,
                   struct i915_gem_context *ctx);
index eb317759b5d37f2656a7e5e1fa812b2ca41fb660..5c496b11ab5c7656d2e20f04c08b37bfa83ce459 100644 (file)
@@ -235,10 +235,9 @@ static int intel_overlay_do_wait_request(struct intel_overlay *overlay,
 
 static struct i915_request *alloc_request(struct intel_overlay *overlay)
 {
-       struct drm_i915_private *dev_priv = overlay->i915;
-       struct intel_engine_cs *engine = dev_priv->engine[RCS0];
+       struct intel_engine_cs *engine = overlay->i915->engine[RCS0];
 
-       return i915_request_alloc(engine, dev_priv->kernel_context);
+       return i915_request_create(engine->kernel_context);
 }
 
 /* overlay needs to be disable in OCMD reg */
index 27d8f853111be92fa340a2404224f5dba48cf57c..eee838dc0634e0c6a297ef22042bc2df1076d36a 100644 (file)
@@ -46,7 +46,7 @@ static int __live_active_setup(struct drm_i915_private *i915,
        for_each_engine(engine, i915, id) {
                struct i915_request *rq;
 
-               rq = i915_request_alloc(engine, i915->kernel_context);
+               rq = i915_request_create(engine->kernel_context);
                if (IS_ERR(rq)) {
                        err = PTR_ERR(rq);
                        break;
index e43630b40fce4c463714c4df75aaa3ceaa8e2080..046a38743152c4b66de08339c2957dcf1920dab9 100644 (file)
@@ -202,7 +202,7 @@ static int gpu_set(struct drm_i915_gem_object *obj,
        if (IS_ERR(vma))
                return PTR_ERR(vma);
 
-       rq = i915_request_alloc(i915->engine[RCS0], i915->kernel_context);
+       rq = i915_request_create(i915->engine[RCS0]->kernel_context);
        if (IS_ERR(rq)) {
                i915_vma_unpin(vma);
                return PTR_ERR(rq);
index 12fc53c694a67021e1411a47bb28f354418c9f1a..12203d665a4edb3e5f19e5913ad82e9b3202bb85 100644 (file)
@@ -468,7 +468,7 @@ static int make_obj_busy(struct drm_i915_gem_object *obj)
        if (err)
                return err;
 
-       rq = i915_request_alloc(i915->engine[RCS0], i915->kernel_context);
+       rq = i915_request_create(i915->engine[RCS0]->kernel_context);
        if (IS_ERR(rq)) {
                i915_vma_unpin(vma);
                return PTR_ERR(rq);
index e6ffe2240126af596f4d1af997fb211a169a3cd3..098d7b3aa131794ec76dc7e8df63dcf0798fdf2a 100644 (file)
@@ -551,8 +551,7 @@ static int live_nop_request(void *arg)
                        times[1] = ktime_get_raw();
 
                        for (n = 0; n < prime; n++) {
-                               request = i915_request_alloc(engine,
-                                                            i915->kernel_context);
+                               request = i915_request_create(engine->kernel_context);
                                if (IS_ERR(request)) {
                                        err = PTR_ERR(request);
                                        goto out_unlock;
@@ -649,7 +648,7 @@ empty_request(struct intel_engine_cs *engine,
        struct i915_request *request;
        int err;
 
-       request = i915_request_alloc(engine, engine->i915->kernel_context);
+       request = i915_request_create(engine->kernel_context);
        if (IS_ERR(request))
                return request;
 
@@ -853,7 +852,7 @@ static int live_all_engines(void *arg)
        }
 
        for_each_engine(engine, i915, id) {
-               request[id] = i915_request_alloc(engine, i915->kernel_context);
+               request[id] = i915_request_create(engine->kernel_context);
                if (IS_ERR(request[id])) {
                        err = PTR_ERR(request[id]);
                        pr_err("%s: Request allocation failed with err=%d\n",
@@ -962,7 +961,7 @@ static int live_sequential_engines(void *arg)
                        goto out_unlock;
                }
 
-               request[id] = i915_request_alloc(engine, i915->kernel_context);
+               request[id] = i915_request_create(engine->kernel_context);
                if (IS_ERR(request[id])) {
                        err = PTR_ERR(request[id]);
                        pr_err("%s: Request allocation failed for %s with err=%d\n",
index bd96afcadfe7c21ce170d146ca8d623ee43756fe..ff9ebe50fae8073ac361072fec65cf9178f7cc0c 100644 (file)
@@ -454,7 +454,7 @@ tl_write(struct i915_timeline *tl, struct intel_engine_cs *engine, u32 value)
                goto out;
        }
 
-       rq = i915_request_alloc(engine, engine->i915->kernel_context);
+       rq = i915_request_create(engine->kernel_context);
        if (IS_ERR(rq))
                goto out_unpin;
 
@@ -678,7 +678,7 @@ static int live_hwsp_wrap(void *arg)
                if (!intel_engine_can_store_dword(engine))
                        continue;
 
-               rq = i915_request_alloc(engine, i915->kernel_context);
+               rq = i915_request_create(engine->kernel_context);
                if (IS_ERR(rq)) {
                        err = PTR_ERR(rq);
                        goto out;