Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> writes:
> Migrate the request operations out of the main body of i915_gem.c and
> into their own C file for easier expansion.
>
> v2: Move __i915_add_request() across as well
>
> Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
Acked-by: Mika Kuoppala <mika.kuoppala@xxxxxxxxx>
> ---
> drivers/gpu/drm/i915/Makefile | 1 +
> drivers/gpu/drm/i915/i915_drv.h | 209 +---------
> drivers/gpu/drm/i915/i915_gem.c | 655 +------------------------------
> drivers/gpu/drm/i915/i915_gem_request.c | 658 ++++++++++++++++++++++++++++++++
> drivers/gpu/drm/i915/i915_gem_request.h | 238 ++++++++++++
> 5 files changed, 905 insertions(+), 856 deletions(-)
> create mode 100644 drivers/gpu/drm/i915/i915_gem_request.c
> create mode 100644 drivers/gpu/drm/i915/i915_gem_request.h
>
> diff --git a/drivers/gpu/drm/i915/Makefile b/drivers/gpu/drm/i915/Makefile
> index 75318ebb8d25..6092f0ea24df 100644
> --- a/drivers/gpu/drm/i915/Makefile
> +++ b/drivers/gpu/drm/i915/Makefile
> @@ -33,6 +33,7 @@ i915-y += i915_cmd_parser.o \
> i915_gem_gtt.o \
> i915_gem.o \
> i915_gem_render_state.o \
> + i915_gem_request.o \
> i915_gem_shrinker.o \
> i915_gem_stolen.o \
> i915_gem_tiling.o \
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index e73c0fc84c73..a4767c198413 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -61,6 +61,7 @@
> #include "i915_gem.h"
> #include "i915_gem_gtt.h"
> #include "i915_gem_render_state.h"
> +#include "i915_gem_request.h"
>
> #include "intel_gvt.h"
>
> @@ -2365,171 +2366,6 @@ static inline struct scatterlist *__sg_next(struct scatterlist *sg)
> (((__iter).curr += PAGE_SIZE) < (__iter).max) || \
> ((__iter) = __sgt_iter(__sg_next((__iter).sgp), false), 0))
>
> -/**
> - * Request queue structure.
> - *
> - * The request queue allows us to note sequence numbers that have been emitted
> - * and may be associated with active buffers to be retired.
> - *
> - * By keeping this list, we can avoid having to do questionable sequence
> - * number comparisons on buffer last_read|write_seqno. It also allows an
> - * emission time to be associated with the request for tracking how far ahead
> - * of the GPU the submission is.
> - *
> - * The requests are reference counted, so upon creation they should have an
> - * initial reference taken using kref_init
> - */
> -struct drm_i915_gem_request {
> - struct kref ref;
> -
> - /** On Which ring this request was generated */
> - struct drm_i915_private *i915;
> - struct intel_engine_cs *engine;
> - struct intel_signal_node signaling;
> -
> - /** GEM sequence number associated with the previous request,
> - * when the HWS breadcrumb is equal to this the GPU is processing
> - * this request.
> - */
> - u32 previous_seqno;
> -
> - /** GEM sequence number associated with this request,
> - * when the HWS breadcrumb is equal or greater than this the GPU
> - * has finished processing this request.
> - */
> - u32 seqno;
> -
> - /** Position in the ringbuffer of the start of the request */
> - u32 head;
> -
> - /**
> - * Position in the ringbuffer of the start of the postfix.
> - * This is required to calculate the maximum available ringbuffer
> - * space without overwriting the postfix.
> - */
> - u32 postfix;
> -
> - /** Position in the ringbuffer of the end of the whole request */
> - u32 tail;
> -
> - /** Preallocate space in the ringbuffer for the emitting the request */
> - u32 reserved_space;
> -
> - /**
> - * Context and ring buffer related to this request
> - * Contexts are refcounted, so when this request is associated with a
> - * context, we must increment the context's refcount, to guarantee that
> - * it persists while any request is linked to it. Requests themselves
> - * are also refcounted, so the request will only be freed when the last
> - * reference to it is dismissed, and the code in
> - * i915_gem_request_free() will then decrement the refcount on the
> - * context.
> - */
> - struct i915_gem_context *ctx;
> - struct intel_ringbuffer *ringbuf;
> -
> - /**
> - * Context related to the previous request.
> - * As the contexts are accessed by the hardware until the switch is
> - * completed to a new context, the hardware may still be writing
> - * to the context object after the breadcrumb is visible. We must
> - * not unpin/unbind/prune that object whilst still active and so
> - * we keep the previous context pinned until the following (this)
> - * request is retired.
> - */
> - struct i915_gem_context *previous_context;
> -
> - /** Batch buffer related to this request if any (used for
> - error state dump only) */
> - struct drm_i915_gem_object *batch_obj;
> -
> - /** Time at which this request was emitted, in jiffies. */
> - unsigned long emitted_jiffies;
> -
> - /** global list entry for this request */
> - struct list_head list;
> -
> - struct drm_i915_file_private *file_priv;
> - /** file_priv list entry for this request */
> - struct list_head client_list;
> -
> - /** process identifier submitting this request */
> - struct pid *pid;
> -
> - /**
> - * The ELSP only accepts two elements at a time, so we queue
> - * context/tail pairs on a given queue (ring->execlist_queue) until the
> - * hardware is available. The queue serves a double purpose: we also use
> - * it to keep track of the up to 2 contexts currently in the hardware
> - * (usually one in execution and the other queued up by the GPU): We
> - * only remove elements from the head of the queue when the hardware
> - * informs us that an element has been completed.
> - *
> - * All accesses to the queue are mediated by a spinlock
> - * (ring->execlist_lock).
> - */
> -
> - /** Execlist link in the submission queue.*/
> - struct list_head execlist_link;
> -
> - /** Execlists no. of times this request has been sent to the ELSP */
> - int elsp_submitted;
> -
> - /** Execlists context hardware id. */
> - unsigned ctx_hw_id;
> -};
> -
> -struct drm_i915_gem_request * __must_check
> -i915_gem_request_alloc(struct intel_engine_cs *engine,
> - struct i915_gem_context *ctx);
> -void i915_gem_request_free(struct kref *req_ref);
> -int i915_gem_request_add_to_client(struct drm_i915_gem_request *req,
> - struct drm_file *file);
> -
> -static inline uint32_t
> -i915_gem_request_get_seqno(struct drm_i915_gem_request *req)
> -{
> - return req ? req->seqno : 0;
> -}
> -
> -static inline struct intel_engine_cs *
> -i915_gem_request_get_engine(struct drm_i915_gem_request *req)
> -{
> - return req ? req->engine : NULL;
> -}
> -
> -static inline struct drm_i915_gem_request *
> -i915_gem_request_reference(struct drm_i915_gem_request *req)
> -{
> - if (req)
> - kref_get(&req->ref);
> - return req;
> -}
> -
> -static inline void
> -i915_gem_request_unreference(struct drm_i915_gem_request *req)
> -{
> - kref_put(&req->ref, i915_gem_request_free);
> -}
> -
> -static inline void i915_gem_request_assign(struct drm_i915_gem_request **pdst,
> - struct drm_i915_gem_request *src)
> -{
> - if (src)
> - i915_gem_request_reference(src);
> -
> - if (*pdst)
> - i915_gem_request_unreference(*pdst);
> -
> - *pdst = src;
> -}
> -
> -/*
> - * XXX: i915_gem_request_completed should be here but currently needs the
> - * definition of i915_seqno_passed() which is below. It will be moved in
> - * a later patch when the call to i915_seqno_passed() is obsoleted...
> - */
> -
> /*
> * A command that requires special handling by the command parser.
> */
> @@ -3297,37 +3133,6 @@ void i915_gem_track_fb(struct drm_i915_gem_object *old,
> struct drm_i915_gem_object *new,
> unsigned frontbuffer_bits);
>
> -/**
> - * Returns true if seq1 is later than seq2.
> - */
> -static inline bool
> -i915_seqno_passed(uint32_t seq1, uint32_t seq2)
> -{
> - return (int32_t)(seq1 - seq2) >= 0;
> -}
> -
> -static inline bool i915_gem_request_started(const struct drm_i915_gem_request *req)
> -{
> - return i915_seqno_passed(intel_engine_get_seqno(req->engine),
> - req->previous_seqno);
> -}
> -
> -static inline bool i915_gem_request_completed(const struct drm_i915_gem_request *req)
> -{
> - return i915_seqno_passed(intel_engine_get_seqno(req->engine),
> - req->seqno);
> -}
> -
> -bool __i915_spin_request(const struct drm_i915_gem_request *request,
> - int state, unsigned long timeout_us);
> -static inline bool i915_spin_request(const struct drm_i915_gem_request *request,
> - int state, unsigned long timeout_us)
> -{
> - return (i915_gem_request_started(request) &&
> - __i915_spin_request(request, state, timeout_us));
> -}
> -
> -int __must_check i915_gem_get_seqno(struct drm_i915_private *dev_priv, u32 *seqno);
> int __must_check i915_gem_set_seqno(struct drm_device *dev, u32 seqno);
>
> struct drm_i915_gem_request *
> @@ -3385,18 +3190,6 @@ void i915_gem_cleanup_engines(struct drm_device *dev);
> int __must_check i915_gem_wait_for_idle(struct drm_i915_private *dev_priv);
> int __must_check i915_gem_suspend(struct drm_device *dev);
> void i915_gem_resume(struct drm_device *dev);
> -void __i915_add_request(struct drm_i915_gem_request *req,
> - struct drm_i915_gem_object *batch_obj,
> - bool flush_caches);
> -#define i915_add_request(req) \
> - __i915_add_request(req, NULL, true)
> -#define i915_add_request_no_flush(req) \
> - __i915_add_request(req, NULL, false)
> -int __i915_wait_request(struct drm_i915_gem_request *req,
> - bool interruptible,
> - s64 *timeout,
> - struct intel_rps_client *rps);
> -int __must_check i915_wait_request(struct drm_i915_gem_request *req);
> int i915_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf);
> int __must_check
> i915_gem_object_wait_rendering(struct drm_i915_gem_object *obj,
> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index 8b42a5101f11..10e5db3f294e 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -1325,365 +1325,6 @@ put_rpm:
> return ret;
> }
>
> -static int
> -i915_gem_check_wedge(unsigned reset_counter, bool interruptible)
> -{
> - if (__i915_terminally_wedged(reset_counter))
> - return -EIO;
> -
> - if (__i915_reset_in_progress(reset_counter)) {
> - /* Non-interruptible callers can't handle -EAGAIN, hence return
> - * -EIO unconditionally for these. */
> - if (!interruptible)
> - return -EIO;
> -
> - return -EAGAIN;
> - }
> -
> - return 0;
> -}
> -
> -static unsigned long local_clock_us(unsigned *cpu)
> -{
> - unsigned long t;
> -
> - /* Cheaply and approximately convert from nanoseconds to microseconds.
> - * The result and subsequent calculations are also defined in the same
> - * approximate microseconds units. The principal source of timing
> - * error here is from the simple truncation.
> - *
> - * Note that local_clock() is only defined wrt to the current CPU;
> - * the comparisons are no longer valid if we switch CPUs. Instead of
> - * blocking preemption for the entire busywait, we can detect the CPU
> - * switch and use that as indicator of system load and a reason to
> - * stop busywaiting, see busywait_stop().
> - */
> - *cpu = get_cpu();
> - t = local_clock() >> 10;
> - put_cpu();
> -
> - return t;
> -}
> -
> -static bool busywait_stop(unsigned long timeout, unsigned cpu)
> -{
> - unsigned this_cpu;
> -
> - if (time_after(local_clock_us(&this_cpu), timeout))
> - return true;
> -
> - return this_cpu != cpu;
> -}
> -
> -bool __i915_spin_request(const struct drm_i915_gem_request *req,
> - int state, unsigned long timeout_us)
> -{
> - unsigned cpu;
> -
> - /* When waiting for high frequency requests, e.g. during synchronous
> - * rendering split between the CPU and GPU, the finite amount of time
> - * required to set up the irq and wait upon it limits the response
> - * rate. By busywaiting on the request completion for a short while we
> - * can service the high frequency waits as quick as possible. However,
> - * if it is a slow request, we want to sleep as quickly as possible.
> - * The tradeoff between waiting and sleeping is roughly the time it
> - * takes to sleep on a request, on the order of a microsecond.
> - */
> -
> - timeout_us += local_clock_us(&cpu);
> - do {
> - if (i915_gem_request_completed(req))
> - return true;
> -
> - if (signal_pending_state(state, current))
> - break;
> -
> - if (busywait_stop(timeout_us, cpu))
> - break;
> -
> - cpu_relax_lowlatency();
> - } while (!need_resched());
> -
> - return false;
> -}
> -
> -/**
> - * __i915_wait_request - wait until execution of request has finished
> - * @req: duh!
> - * @interruptible: do an interruptible wait (normally yes)
> - * @timeout: in - how long to wait (NULL forever); out - how much time remaining
> - * @rps: RPS client
> - *
> - * Note: It is of utmost importance that the passed in seqno and reset_counter
> - * values have been read by the caller in an smp safe manner. Where read-side
> - * locks are involved, it is sufficient to read the reset_counter before
> - * unlocking the lock that protects the seqno. For lockless tricks, the
> - * reset_counter _must_ be read before, and an appropriate smp_rmb must be
> - * inserted.
> - *
> - * Returns 0 if the request was found within the alloted time. Else returns the
> - * errno with remaining time filled in timeout argument.
> - */
> -int __i915_wait_request(struct drm_i915_gem_request *req,
> - bool interruptible,
> - s64 *timeout,
> - struct intel_rps_client *rps)
> -{
> - int state = interruptible ? TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE;
> - DEFINE_WAIT(reset);
> - struct intel_wait wait;
> - unsigned long timeout_remain;
> - s64 before = 0; /* Only to silence a compiler warning. */
> - int ret = 0;
> -
> - might_sleep();
> -
> - if (list_empty(&req->list))
> - return 0;
> -
> - if (i915_gem_request_completed(req))
> - return 0;
> -
> - timeout_remain = MAX_SCHEDULE_TIMEOUT;
> - if (timeout) {
> - if (WARN_ON(*timeout < 0))
> - return -EINVAL;
> -
> - if (*timeout == 0)
> - return -ETIME;
> -
> - timeout_remain = nsecs_to_jiffies_timeout(*timeout);
> -
> - /*
> - * Record current time in case interrupted by signal, or wedged.
> - */
> - before = ktime_get_raw_ns();
> - }
> -
> - trace_i915_gem_request_wait_begin(req);
> -
> - /* This client is about to stall waiting for the GPU. In many cases
> - * this is undesirable and limits the throughput of the system, as
> - * many clients cannot continue processing user input/output whilst
> - * blocked. RPS autotuning may take tens of milliseconds to respond
> - * to the GPU load and thus incurs additional latency for the client.
> - * We can circumvent that by promoting the GPU frequency to maximum
> - * before we wait. This makes the GPU throttle up much more quickly
> - * (good for benchmarks and user experience, e.g. window animations),
> - * but at a cost of spending more power processing the workload
> - * (bad for battery). Not all clients even want their results
> - * immediately and for them we should just let the GPU select its own
> - * frequency to maximise efficiency. To prevent a single client from
> - * forcing the clocks too high for the whole system, we only allow
> - * each client to waitboost once in a busy period.
> - */
> - if (INTEL_INFO(req->i915)->gen >= 6)
> - gen6_rps_boost(req->i915, rps, req->emitted_jiffies);
> -
> - /* Optimistic spin for the next ~jiffie before touching IRQs */
> - if (i915_spin_request(req, state, 5))
> - goto complete;
> -
> - set_current_state(state);
> - add_wait_queue(&req->i915->gpu_error.wait_queue, &reset);
> -
> - intel_wait_init(&wait, req->seqno);
> - if (intel_engine_add_wait(req->engine, &wait))
> - /* In order to check that we haven't missed the interrupt
> - * as we enabled it, we need to kick ourselves to do a
> - * coherent check on the seqno before we sleep.
> - */
> - goto wakeup;
> -
> - for (;;) {
> - if (signal_pending_state(state, current)) {
> - ret = -ERESTARTSYS;
> - break;
> - }
> -
> - timeout_remain = io_schedule_timeout(timeout_remain);
> - if (timeout_remain == 0) {
> - ret = -ETIME;
> - break;
> - }
> -
> - if (intel_wait_complete(&wait))
> - break;
> -
> - set_current_state(state);
> -
> -wakeup:
> - /* Carefully check if the request is complete, giving time
> - * for the seqno to be visible following the interrupt.
> - * We also have to check in case we are kicked by the GPU
> - * reset in order to drop the struct_mutex.
> - */
> - if (__i915_request_irq_complete(req))
> - break;
> -
> - /* Only spin if we know the GPU is processing this request */
> - if (i915_spin_request(req, state, 2))
> - break;
> - }
> - remove_wait_queue(&req->i915->gpu_error.wait_queue, &reset);
> -
> - intel_engine_remove_wait(req->engine, &wait);
> - __set_current_state(TASK_RUNNING);
> -complete:
> - trace_i915_gem_request_wait_end(req);
> -
> - if (timeout) {
> - s64 tres = *timeout - (ktime_get_raw_ns() - before);
> -
> - *timeout = tres < 0 ? 0 : tres;
> -
> - /*
> - * Apparently ktime isn't accurate enough and occasionally has a
> - * bit of mismatch in the jiffies<->nsecs<->ktime loop. So patch
> - * things up to make the test happy. We allow up to 1 jiffy.
> - *
> - * This is a regrssion from the timespec->ktime conversion.
> - */
> - if (ret == -ETIME && *timeout < jiffies_to_usecs(1)*1000)
> - *timeout = 0;
> - }
> -
> - if (rps && req->seqno == req->engine->last_submitted_seqno) {
> - /* The GPU is now idle and this client has stalled.
> - * Since no other client has submitted a request in the
> - * meantime, assume that this client is the only one
> - * supplying work to the GPU but is unable to keep that
> - * work supplied because it is waiting. Since the GPU is
> - * then never kept fully busy, RPS autoclocking will
> - * keep the clocks relatively low, causing further delays.
> - * Compensate by giving the synchronous client credit for
> - * a waitboost next time.
> - */
> - spin_lock(&req->i915->rps.client_lock);
> - list_del_init(&rps->link);
> - spin_unlock(&req->i915->rps.client_lock);
> - }
> -
> - return ret;
> -}
> -
> -int i915_gem_request_add_to_client(struct drm_i915_gem_request *req,
> - struct drm_file *file)
> -{
> - struct drm_i915_file_private *file_priv;
> -
> - WARN_ON(!req || !file || req->file_priv);
> -
> - if (!req || !file)
> - return -EINVAL;
> -
> - if (req->file_priv)
> - return -EINVAL;
> -
> - file_priv = file->driver_priv;
> -
> - spin_lock(&file_priv->mm.lock);
> - req->file_priv = file_priv;
> - list_add_tail(&req->client_list, &file_priv->mm.request_list);
> - spin_unlock(&file_priv->mm.lock);
> -
> - req->pid = get_pid(task_pid(current));
> -
> - return 0;
> -}
> -
> -static inline void
> -i915_gem_request_remove_from_client(struct drm_i915_gem_request *request)
> -{
> - struct drm_i915_file_private *file_priv = request->file_priv;
> -
> - if (!file_priv)
> - return;
> -
> - spin_lock(&file_priv->mm.lock);
> - list_del(&request->client_list);
> - request->file_priv = NULL;
> - spin_unlock(&file_priv->mm.lock);
> -
> - put_pid(request->pid);
> - request->pid = NULL;
> -}
> -
> -static void i915_gem_request_retire(struct drm_i915_gem_request *request)
> -{
> - trace_i915_gem_request_retire(request);
> -
> - /* We know the GPU must have read the request to have
> - * sent us the seqno + interrupt, so use the position
> - * of tail of the request to update the last known position
> - * of the GPU head.
> - *
> - * Note this requires that we are always called in request
> - * completion order.
> - */
> - request->ringbuf->last_retired_head = request->postfix;
> -
> - list_del_init(&request->list);
> - i915_gem_request_remove_from_client(request);
> -
> - if (request->previous_context) {
> - if (i915.enable_execlists)
> - intel_lr_context_unpin(request->previous_context,
> - request->engine);
> - }
> -
> - i915_gem_context_unreference(request->ctx);
> - i915_gem_request_unreference(request);
> -}
> -
> -static void
> -__i915_gem_request_retire__upto(struct drm_i915_gem_request *req)
> -{
> - struct intel_engine_cs *engine = req->engine;
> - struct drm_i915_gem_request *tmp;
> -
> - lockdep_assert_held(&engine->i915->drm.struct_mutex);
> -
> - if (list_empty(&req->list))
> - return;
> -
> - do {
> - tmp = list_first_entry(&engine->request_list,
> - typeof(*tmp), list);
> -
> - i915_gem_request_retire(tmp);
> - } while (tmp != req);
> -
> - WARN_ON(i915_verify_lists(engine->dev));
> -}
> -
> -/**
> - * Waits for a request to be signaled, and cleans up the
> - * request and object lists appropriately for that event.
> - * @req: request to wait on
> - */
> -int
> -i915_wait_request(struct drm_i915_gem_request *req)
> -{
> - struct drm_i915_private *dev_priv = req->i915;
> - bool interruptible;
> - int ret;
> -
> - interruptible = dev_priv->mm.interruptible;
> -
> - BUG_ON(!mutex_is_locked(&dev_priv->drm.struct_mutex));
> -
> - ret = __i915_wait_request(req, interruptible, NULL, NULL);
> - if (ret)
> - return ret;
> -
> - /* If the GPU hung, we want to keep the requests to find the guilty. */
> - if (!i915_reset_in_progress(&dev_priv->gpu_error))
> - __i915_gem_request_retire__upto(req);
> -
> - return 0;
> -}
> -
> /**
> * Ensures that all rendering to the object has completed and the object is
> * safe to unbind from the GTT or access from the CPU.
> @@ -1740,7 +1381,7 @@ i915_gem_object_retire_request(struct drm_i915_gem_object *obj,
> i915_gem_object_retire__write(obj);
>
> if (!i915_reset_in_progress(&req->i915->gpu_error))
> - __i915_gem_request_retire__upto(req);
> + i915_gem_request_retire_upto(req);
> }
>
> /* A nonblocking variant of the above wait. This is a highly dangerous routine
> @@ -2761,193 +2402,6 @@ i915_gem_object_retire__read(struct drm_i915_gem_object *obj, int ring)
> drm_gem_object_unreference(&obj->base);
> }
>
> -static int
> -i915_gem_init_seqno(struct drm_i915_private *dev_priv, u32 seqno)
> -{
> - struct intel_engine_cs *engine;
> - int ret;
> -
> - /* Carefully retire all requests without writing to the rings */
> - for_each_engine(engine, dev_priv) {
> - ret = intel_engine_idle(engine);
> - if (ret)
> - return ret;
> - }
> - i915_gem_retire_requests(dev_priv);
> -
> - /* If the seqno wraps around, we need to clear the breadcrumb rbtree */
> - if (!i915_seqno_passed(seqno, dev_priv->next_seqno)) {
> - while (intel_kick_waiters(dev_priv) ||
> - intel_kick_signalers(dev_priv))
> - yield();
> - }
> -
> - /* Finally reset hw state */
> - for_each_engine(engine, dev_priv)
> - intel_ring_init_seqno(engine, seqno);
> -
> - return 0;
> -}
> -
> -int i915_gem_set_seqno(struct drm_device *dev, u32 seqno)
> -{
> - struct drm_i915_private *dev_priv = to_i915(dev);
> - int ret;
> -
> - if (seqno == 0)
> - return -EINVAL;
> -
> - /* HWS page needs to be set less than what we
> - * will inject to ring
> - */
> - ret = i915_gem_init_seqno(dev_priv, seqno - 1);
> - if (ret)
> - return ret;
> -
> - /* Carefully set the last_seqno value so that wrap
> - * detection still works
> - */
> - dev_priv->next_seqno = seqno;
> - dev_priv->last_seqno = seqno - 1;
> - if (dev_priv->last_seqno == 0)
> - dev_priv->last_seqno--;
> -
> - return 0;
> -}
> -
> -int
> -i915_gem_get_seqno(struct drm_i915_private *dev_priv, u32 *seqno)
> -{
> - /* reserve 0 for non-seqno */
> - if (dev_priv->next_seqno == 0) {
> - int ret = i915_gem_init_seqno(dev_priv, 0);
> - if (ret)
> - return ret;
> -
> - dev_priv->next_seqno = 1;
> - }
> -
> - *seqno = dev_priv->last_seqno = dev_priv->next_seqno++;
> - return 0;
> -}
> -
> -static void i915_gem_mark_busy(const struct intel_engine_cs *engine)
> -{
> - struct drm_i915_private *dev_priv = engine->i915;
> -
> - dev_priv->gt.active_engines |= intel_engine_flag(engine);
> - if (dev_priv->gt.awake)
> - return;
> -
> - intel_runtime_pm_get_noresume(dev_priv);
> - dev_priv->gt.awake = true;
> -
> - intel_enable_gt_powersave(dev_priv);
> - i915_update_gfx_val(dev_priv);
> - if (INTEL_GEN(dev_priv) >= 6)
> - gen6_rps_busy(dev_priv);
> -
> - queue_delayed_work(dev_priv->wq,
> - &dev_priv->gt.retire_work,
> - round_jiffies_up_relative(HZ));
> -}
> -
> -/*
> - * NB: This function is not allowed to fail. Doing so would mean the the
> - * 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_add_request(struct drm_i915_gem_request *request,
> - struct drm_i915_gem_object *obj,
> - bool flush_caches)
> -{
> - struct intel_engine_cs *engine;
> - struct intel_ringbuffer *ringbuf;
> - u32 request_start;
> - u32 reserved_tail;
> - int ret;
> -
> - if (WARN_ON(request == NULL))
> - return;
> -
> - engine = request->engine;
> - ringbuf = request->ringbuf;
> -
> - /*
> - * 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.
> - */
> - request_start = intel_ring_get_tail(ringbuf);
> - reserved_tail = request->reserved_space;
> - request->reserved_space = 0;
> -
> - /*
> - * Emit any outstanding flushes - execbuf can fail to emit the flush
> - * after having emitted the batchbuffer command. Hence we need to fix
> - * things up similar to emitting the lazy request. The difference here
> - * is that the flush _must_ happen before the next request, no matter
> - * what.
> - */
> - if (flush_caches) {
> - if (i915.enable_execlists)
> - ret = logical_ring_flush_all_caches(request);
> - else
> - ret = intel_ring_flush_all_caches(request);
> - /* Not allowed to fail! */
> - WARN(ret, "*_ring_flush_all_caches failed: %d!\n", ret);
> - }
> -
> - trace_i915_gem_request_add(request);
> -
> - request->head = request_start;
> -
> - /* Whilst this request exists, batch_obj will be on the
> - * active_list, and so will hold the active reference. Only when this
> - * request is retired will the the batch_obj be moved onto the
> - * inactive_list and lose its active reference. Hence we do not need
> - * to explicitly hold another reference here.
> - */
> - request->batch_obj = obj;
> -
> - /* Seal the request and mark it as pending execution. Note that
> - * we may inspect this state, without holding any locks, during
> - * hangcheck. Hence we apply the barrier to ensure that we do not
> - * see a more recent value in the hws than we are tracking.
> - */
> - request->emitted_jiffies = jiffies;
> - request->previous_seqno = engine->last_submitted_seqno;
> - smp_store_mb(engine->last_submitted_seqno, request->seqno);
> - list_add_tail(&request->list, &engine->request_list);
> -
> - /* Record the position of the start of the request so that
> - * should we detect the updated seqno part-way through the
> - * GPU processing the request, we never over-estimate the
> - * position of the head.
> - */
> - request->postfix = intel_ring_get_tail(ringbuf);
> -
> - if (i915.enable_execlists)
> - ret = engine->emit_request(request);
> - else {
> - ret = engine->add_request(request);
> -
> - request->tail = intel_ring_get_tail(ringbuf);
> - }
> - /* Not allowed to fail! */
> - WARN(ret, "emit|add_request failed: %d!\n", ret);
> - /* Sanity check that the reserved size was large enough. */
> - ret = intel_ring_get_tail(ringbuf) - request_start;
> - if (ret < 0)
> - ret += ringbuf->size;
> - WARN_ONCE(ret > reserved_tail,
> - "Not enough space reserved (%d bytes) "
> - "for adding the request (%d bytes)\n",
> - reserved_tail, ret);
> -
> - i915_gem_mark_busy(engine);
> -}
> -
> static bool i915_context_is_banned(const struct i915_gem_context *ctx)
> {
> unsigned long elapsed;
> @@ -2979,101 +2433,6 @@ static void i915_set_reset_status(struct i915_gem_context *ctx,
> }
> }
>
> -void i915_gem_request_free(struct kref *req_ref)
> -{
> - struct drm_i915_gem_request *req = container_of(req_ref,
> - typeof(*req), ref);
> - kmem_cache_free(req->i915->requests, req);
> -}
> -
> -static inline int
> -__i915_gem_request_alloc(struct intel_engine_cs *engine,
> - struct i915_gem_context *ctx,
> - struct drm_i915_gem_request **req_out)
> -{
> - struct drm_i915_private *dev_priv = engine->i915;
> - unsigned reset_counter = i915_reset_counter(&dev_priv->gpu_error);
> - struct drm_i915_gem_request *req;
> - int ret;
> -
> - if (!req_out)
> - return -EINVAL;
> -
> - *req_out = NULL;
> -
> - /* ABI: Before userspace accesses the GPU (e.g. execbuffer), report
> - * EIO if the GPU is already wedged, or EAGAIN to drop the struct_mutex
> - * and restart.
> - */
> - ret = i915_gem_check_wedge(reset_counter, dev_priv->mm.interruptible);
> - if (ret)
> - return ret;
> -
> - req = kmem_cache_zalloc(dev_priv->requests, GFP_KERNEL);
> - if (req == NULL)
> - return -ENOMEM;
> -
> - ret = i915_gem_get_seqno(engine->i915, &req->seqno);
> - if (ret)
> - goto err;
> -
> - kref_init(&req->ref);
> - req->i915 = dev_priv;
> - req->engine = engine;
> - req->ctx = ctx;
> - i915_gem_context_reference(req->ctx);
> -
> - /*
> - * Reserve space in the ring buffer for all the commands required to
> - * eventually emit this request. This is to guarantee that the
> - * i915_add_request() call can't fail. Note that the reserve may need
> - * to be redone if the request is not actually submitted straight
> - * away, e.g. because a GPU scheduler has deferred it.
> - */
> - req->reserved_space = MIN_SPACE_FOR_ADD_REQUEST;
> -
> - if (i915.enable_execlists)
> - ret = intel_logical_ring_alloc_request_extras(req);
> - else
> - ret = intel_ring_alloc_request_extras(req);
> - if (ret)
> - goto err_ctx;
> -
> - *req_out = req;
> - return 0;
> -
> -err_ctx:
> - i915_gem_context_unreference(ctx);
> -err:
> - kmem_cache_free(dev_priv->requests, req);
> - return ret;
> -}
> -
> -/**
> - * i915_gem_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.
> - * This can be NULL if the request is not directly related to
> - * any specific user context, in which case this function will
> - * choose an appropriate context to use.
> - *
> - * Returns a pointer to the allocated request if successful,
> - * or an error code if not.
> - */
> -struct drm_i915_gem_request *
> -i915_gem_request_alloc(struct intel_engine_cs *engine,
> - struct i915_gem_context *ctx)
> -{
> - struct drm_i915_gem_request *req;
> - int err;
> -
> - if (ctx == NULL)
> - ctx = engine->i915->kernel_context;
> - err = __i915_gem_request_alloc(engine, ctx, &req);
> - return err ? ERR_PTR(err) : req;
> -}
> -
> struct drm_i915_gem_request *
> i915_gem_find_active_request(struct intel_engine_cs *engine)
> {
> @@ -3147,14 +2506,14 @@ static void i915_gem_reset_engine_cleanup(struct intel_engine_cs *engine)
> * implicit references on things like e.g. ppgtt address spaces through
> * the request.
> */
> - while (!list_empty(&engine->request_list)) {
> + if (!list_empty(&engine->request_list)) {
> struct drm_i915_gem_request *request;
>
> - request = list_first_entry(&engine->request_list,
> - struct drm_i915_gem_request,
> - list);
> + request = list_last_entry(&engine->request_list,
> + struct drm_i915_gem_request,
> + list);
>
> - i915_gem_request_retire(request);
> + i915_gem_request_retire_upto(request);
> }
>
> /* Having flushed all requests from all queues, we know that all
> @@ -3222,7 +2581,7 @@ i915_gem_retire_requests_ring(struct intel_engine_cs *engine)
> if (!i915_gem_request_completed(request))
> break;
>
> - i915_gem_request_retire(request);
> + i915_gem_request_retire_upto(request);
> }
>
> /* Move any buffers on the active list that are no longer referenced
> diff --git a/drivers/gpu/drm/i915/i915_gem_request.c b/drivers/gpu/drm/i915/i915_gem_request.c
> new file mode 100644
> index 000000000000..9e9aa6b725f7
> --- /dev/null
> +++ b/drivers/gpu/drm/i915/i915_gem_request.c
> @@ -0,0 +1,658 @@
> +/*
> + * Copyright © 2008-2015 Intel Corporation
> + *
> + * Permission is hereby granted, free of charge, to any person obtaining a
> + * copy of this software and associated documentation files (the "Software"),
> + * to deal in the Software without restriction, including without limitation
> + * the rights to use, copy, modify, merge, publish, distribute, sublicense,
> + * and/or sell copies of the Software, and to permit persons to whom the
> + * Software is furnished to do so, subject to the following conditions:
> + *
> + * The above copyright notice and this permission notice (including the next
> + * paragraph) shall be included in all copies or substantial portions of the
> + * Software.
> + *
> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
> + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
> + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
> + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
> + * IN THE SOFTWARE.
> + *
> + */
> +
> +#include "i915_drv.h"
> +
> +int i915_gem_request_add_to_client(struct drm_i915_gem_request *req,
> + struct drm_file *file)
> +{
> + struct drm_i915_private *dev_private;
> + struct drm_i915_file_private *file_priv;
> +
> + WARN_ON(!req || !file || req->file_priv);
> +
> + if (!req || !file)
> + return -EINVAL;
> +
> + if (req->file_priv)
> + return -EINVAL;
> +
> + dev_private = req->i915;
> + file_priv = file->driver_priv;
> +
> + spin_lock(&file_priv->mm.lock);
> + req->file_priv = file_priv;
> + list_add_tail(&req->client_list, &file_priv->mm.request_list);
> + spin_unlock(&file_priv->mm.lock);
> +
> + req->pid = get_pid(task_pid(current));
> +
> + return 0;
> +}
> +
> +static inline void
> +i915_gem_request_remove_from_client(struct drm_i915_gem_request *request)
> +{
> + struct drm_i915_file_private *file_priv = request->file_priv;
> +
> + if (!file_priv)
> + return;
> +
> + spin_lock(&file_priv->mm.lock);
> + list_del(&request->client_list);
> + request->file_priv = NULL;
> + spin_unlock(&file_priv->mm.lock);
> +
> + put_pid(request->pid);
> + request->pid = NULL;
> +}
> +
> +static void i915_gem_request_retire(struct drm_i915_gem_request *request)
> +{
> + trace_i915_gem_request_retire(request);
> + list_del_init(&request->list);
> +
> + /* We know the GPU must have read the request to have
> + * sent us the seqno + interrupt, so use the position
> + * of tail of the request to update the last known position
> + * of the GPU head.
> + *
> + * Note this requires that we are always called in request
> + * completion order.
> + */
> + request->ringbuf->last_retired_head = request->postfix;
> +
> + i915_gem_request_remove_from_client(request);
> +
> + if (request->previous_context) {
> + if (i915.enable_execlists)
> + intel_lr_context_unpin(request->previous_context,
> + request->engine);
> + }
> +
> + i915_gem_context_unreference(request->ctx);
> + i915_gem_request_unreference(request);
> +}
> +
> +void i915_gem_request_retire_upto(struct drm_i915_gem_request *req)
> +{
> + struct intel_engine_cs *engine = req->engine;
> + struct drm_i915_gem_request *tmp;
> +
> + lockdep_assert_held(&req->i915->drm.struct_mutex);
> +
> + if (list_empty(&req->list))
> + return;
> +
> + do {
> + tmp = list_first_entry(&engine->request_list,
> + typeof(*tmp), list);
> +
> + i915_gem_request_retire(tmp);
> + } while (tmp != req);
> +
> + WARN_ON(i915_verify_lists(engine->dev));
> +}
> +
> +static int i915_gem_check_wedge(unsigned int reset_counter, bool interruptible)
> +{
> + if (__i915_terminally_wedged(reset_counter))
> + return -EIO;
> +
> + if (__i915_reset_in_progress(reset_counter)) {
> + /* Non-interruptible callers can't handle -EAGAIN, hence return
> + * -EIO unconditionally for these.
> + */
> + if (!interruptible)
> + return -EIO;
> +
> + return -EAGAIN;
> + }
> +
> + return 0;
> +}
> +
> +static int i915_gem_init_seqno(struct drm_i915_private *dev_priv, u32 seqno)
> +{
> + struct intel_engine_cs *engine;
> + int ret;
> +
> + /* Carefully retire all requests without writing to the rings */
> + for_each_engine(engine, dev_priv) {
> + ret = intel_engine_idle(engine);
> + if (ret)
> + return ret;
> + }
> + i915_gem_retire_requests(dev_priv);
> +
> + /* If the seqno wraps around, we need to clear the breadcrumb rbtree */
> + if (!i915_seqno_passed(seqno, dev_priv->next_seqno)) {
> + while (intel_kick_waiters(dev_priv) ||
> + intel_kick_signalers(dev_priv))
> + yield();
> + }
> +
> + /* Finally reset hw state */
> + for_each_engine(engine, dev_priv)
> + intel_ring_init_seqno(engine, seqno);
> +
> + return 0;
> +}
> +
> +int i915_gem_set_seqno(struct drm_device *dev, u32 seqno)
> +{
> + struct drm_i915_private *dev_priv = to_i915(dev);
> + int ret;
> +
> + if (seqno == 0)
> + return -EINVAL;
> +
> + /* HWS page needs to be set less than what we
> + * will inject to ring
> + */
> + ret = i915_gem_init_seqno(dev_priv, seqno - 1);
> + if (ret)
> + return ret;
> +
> + /* Carefully set the last_seqno value so that wrap
> + * detection still works
> + */
> + dev_priv->next_seqno = seqno;
> + dev_priv->last_seqno = seqno - 1;
> + if (dev_priv->last_seqno == 0)
> + dev_priv->last_seqno--;
> +
> + return 0;
> +}
> +
> +static int i915_gem_get_seqno(struct drm_i915_private *dev_priv, u32 *seqno)
> +{
> + /* reserve 0 for non-seqno */
> + if (unlikely(dev_priv->next_seqno == 0)) {
> + int ret;
> +
> + ret = i915_gem_init_seqno(dev_priv, 0);
> + if (ret)
> + return ret;
> +
> + dev_priv->next_seqno = 1;
> + }
> +
> + *seqno = dev_priv->last_seqno = dev_priv->next_seqno++;
> + return 0;
> +}
> +
> +static inline int
> +__i915_gem_request_alloc(struct intel_engine_cs *engine,
> + struct i915_gem_context *ctx,
> + struct drm_i915_gem_request **req_out)
> +{
> + struct drm_i915_private *dev_priv = engine->i915;
> + unsigned int reset_counter = i915_reset_counter(&dev_priv->gpu_error);
> + struct drm_i915_gem_request *req;
> + int ret;
> +
> + if (!req_out)
> + return -EINVAL;
> +
> + *req_out = NULL;
> +
> + /* ABI: Before userspace accesses the GPU (e.g. execbuffer), report
> + * EIO if the GPU is already wedged, or EAGAIN to drop the struct_mutex
> + * and restart.
> + */
> + ret = i915_gem_check_wedge(reset_counter, dev_priv->mm.interruptible);
> + if (ret)
> + return ret;
> +
> + req = kmem_cache_zalloc(dev_priv->requests, GFP_KERNEL);
> + if (!req)
> + return -ENOMEM;
> +
> + ret = i915_gem_get_seqno(dev_priv, &req->seqno);
> + if (ret)
> + goto err;
> +
> + kref_init(&req->ref);
> + req->i915 = dev_priv;
> + req->engine = engine;
> + req->ctx = ctx;
> + i915_gem_context_reference(ctx);
> +
> + /*
> + * Reserve space in the ring buffer for all the commands required to
> + * eventually emit this request. This is to guarantee that the
> + * i915_add_request() call can't fail. Note that the reserve may need
> + * to be redone if the request is not actually submitted straight
> + * away, e.g. because a GPU scheduler has deferred it.
> + */
> + req->reserved_space = MIN_SPACE_FOR_ADD_REQUEST;
> +
> + if (i915.enable_execlists)
> + ret = intel_logical_ring_alloc_request_extras(req);
> + else
> + ret = intel_ring_alloc_request_extras(req);
> + if (ret)
> + goto err_ctx;
> +
> + *req_out = req;
> + return 0;
> +
> +err_ctx:
> + i915_gem_context_unreference(ctx);
> +err:
> + kmem_cache_free(dev_priv->requests, req);
> + return ret;
> +}
> +
> +/**
> + * i915_gem_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.
> + * This can be NULL if the request is not directly related to
> + * any specific user context, in which case this function will
> + * choose an appropriate context to use.
> + *
> + * Returns a pointer to the allocated request if successful,
> + * or an error code if not.
> + */
> +struct drm_i915_gem_request *
> +i915_gem_request_alloc(struct intel_engine_cs *engine,
> + struct i915_gem_context *ctx)
> +{
> + struct drm_i915_gem_request *req;
> + int err;
> +
> + if (!ctx)
> + ctx = engine->i915->kernel_context;
> + err = __i915_gem_request_alloc(engine, ctx, &req);
> + return err ? ERR_PTR(err) : req;
> +}
> +
> +static void i915_gem_mark_busy(const struct intel_engine_cs *engine)
> +{
> + struct drm_i915_private *dev_priv = engine->i915;
> +
> + dev_priv->gt.active_engines |= intel_engine_flag(engine);
> + if (dev_priv->gt.awake)
> + return;
> +
> + intel_runtime_pm_get_noresume(dev_priv);
> + dev_priv->gt.awake = true;
> +
> + intel_enable_gt_powersave(dev_priv);
> + i915_update_gfx_val(dev_priv);
> + if (INTEL_GEN(dev_priv) >= 6)
> + gen6_rps_busy(dev_priv);
> +
> + queue_delayed_work(dev_priv->wq,
> + &dev_priv->gt.retire_work,
> + round_jiffies_up_relative(HZ));
> +}
> +
> +/*
> + * NB: This function is not allowed to fail. Doing so would mean the the
> + * 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_add_request(struct drm_i915_gem_request *request,
> + struct drm_i915_gem_object *obj,
> + bool flush_caches)
> +{
> + struct intel_engine_cs *engine;
> + struct intel_ringbuffer *ringbuf;
> + u32 request_start;
> + u32 reserved_tail;
> + int ret;
> +
> + if (WARN_ON(!request))
> + return;
> +
> + engine = request->engine;
> + ringbuf = request->ringbuf;
> +
> + /*
> + * 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.
> + */
> + request_start = intel_ring_get_tail(ringbuf);
> + reserved_tail = request->reserved_space;
> + request->reserved_space = 0;
> +
> + /*
> + * Emit any outstanding flushes - execbuf can fail to emit the flush
> + * after having emitted the batchbuffer command. Hence we need to fix
> + * things up similar to emitting the lazy request. The difference here
> + * is that the flush _must_ happen before the next request, no matter
> + * what.
> + */
> + if (flush_caches) {
> + if (i915.enable_execlists)
> + ret = logical_ring_flush_all_caches(request);
> + else
> + ret = intel_ring_flush_all_caches(request);
> + /* Not allowed to fail! */
> + WARN(ret, "*_ring_flush_all_caches failed: %d!\n", ret);
> + }
> +
> + trace_i915_gem_request_add(request);
> +
> + request->head = request_start;
> +
> + /* Whilst this request exists, batch_obj will be on the
> + * active_list, and so will hold the active reference. Only when this
> + * request is retired will the the batch_obj be moved onto the
> + * inactive_list and lose its active reference. Hence we do not need
> + * to explicitly hold another reference here.
> + */
> + request->batch_obj = obj;
> +
> + /* Seal the request and mark it as pending execution. Note that
> + * we may inspect this state, without holding any locks, during
> + * hangcheck. Hence we apply the barrier to ensure that we do not
> + * see a more recent value in the hws than we are tracking.
> + */
> + request->emitted_jiffies = jiffies;
> + request->previous_seqno = engine->last_submitted_seqno;
> + smp_store_mb(engine->last_submitted_seqno, request->seqno);
> + list_add_tail(&request->list, &engine->request_list);
> +
> + /* Record the position of the start of the request so that
> + * should we detect the updated seqno part-way through the
> + * GPU processing the request, we never over-estimate the
> + * position of the head.
> + */
> + request->postfix = intel_ring_get_tail(ringbuf);
> +
> + if (i915.enable_execlists) {
> + ret = engine->emit_request(request);
> + } else {
> + ret = engine->add_request(request);
> +
> + request->tail = intel_ring_get_tail(ringbuf);
> + }
> + /* Not allowed to fail! */
> + WARN(ret, "emit|add_request failed: %d!\n", ret);
> + /* Sanity check that the reserved size was large enough. */
> + ret = intel_ring_get_tail(ringbuf) - request_start;
> + if (ret < 0)
> + ret += ringbuf->size;
> + WARN_ONCE(ret > reserved_tail,
> + "Not enough space reserved (%d bytes) "
> + "for adding the request (%d bytes)\n",
> + reserved_tail, ret);
> +
> + i915_gem_mark_busy(engine);
> +}
> +
> +static unsigned long local_clock_us(unsigned int *cpu)
> +{
> + unsigned long t;
> +
> + /* Cheaply and approximately convert from nanoseconds to microseconds.
> + * The result and subsequent calculations are also defined in the same
> + * approximate microseconds units. The principal source of timing
> + * error here is from the simple truncation.
> + *
> + * Note that local_clock() is only defined wrt to the current CPU;
> + * the comparisons are no longer valid if we switch CPUs. Instead of
> + * blocking preemption for the entire busywait, we can detect the CPU
> + * switch and use that as indicator of system load and a reason to
> + * stop busywaiting, see busywait_stop().
> + */
> + *cpu = get_cpu();
> + t = local_clock() >> 10;
> + put_cpu();
> +
> + return t;
> +}
> +
> +static bool busywait_stop(unsigned long timeout, unsigned int cpu)
> +{
> + unsigned int this_cpu;
> +
> + if (time_after(local_clock_us(&this_cpu), timeout))
> + return true;
> +
> + return this_cpu != cpu;
> +}
> +
> +bool __i915_spin_request(const struct drm_i915_gem_request *req,
> + int state, unsigned long timeout_us)
> +{
> + unsigned int cpu;
> +
> + /* When waiting for high frequency requests, e.g. during synchronous
> + * rendering split between the CPU and GPU, the finite amount of time
> + * required to set up the irq and wait upon it limits the response
> + * rate. By busywaiting on the request completion for a short while we
> + * can service the high frequency waits as quick as possible. However,
> + * if it is a slow request, we want to sleep as quickly as possible.
> + * The tradeoff between waiting and sleeping is roughly the time it
> + * takes to sleep on a request, on the order of a microsecond.
> + */
> +
> + timeout_us += local_clock_us(&cpu);
> + do {
> + if (i915_gem_request_completed(req))
> + return true;
> +
> + if (signal_pending_state(state, current))
> + break;
> +
> + if (busywait_stop(timeout_us, cpu))
> + break;
> +
> + cpu_relax_lowlatency();
> + } while (!need_resched());
> +
> + return false;
> +}
> +
> +/**
> + * __i915_wait_request - wait until execution of request has finished
> + * @req: duh!
> + * @interruptible: do an interruptible wait (normally yes)
> + * @timeout: in - how long to wait (NULL forever); out - how much time remaining
> + * @rps: client to charge for RPS boosting
> + *
> + * Note: It is of utmost importance that the passed in seqno and reset_counter
> + * values have been read by the caller in an smp safe manner. Where read-side
> + * locks are involved, it is sufficient to read the reset_counter before
> + * unlocking the lock that protects the seqno. For lockless tricks, the
> + * reset_counter _must_ be read before, and an appropriate smp_rmb must be
> + * inserted.
> + *
> + * Returns 0 if the request was found within the alloted time. Else returns the
> + * errno with remaining time filled in timeout argument.
> + */
> +int __i915_wait_request(struct drm_i915_gem_request *req,
> + bool interruptible,
> + s64 *timeout,
> + struct intel_rps_client *rps)
> +{
> + int state = interruptible ? TASK_INTERRUPTIBLE : TASK_UNINTERRUPTIBLE;
> + DEFINE_WAIT(reset);
> + struct intel_wait wait;
> + unsigned long timeout_remain;
> + int ret = 0;
> +
> + might_sleep();
> +
> + if (list_empty(&req->list))
> + return 0;
> +
> + if (i915_gem_request_completed(req))
> + return 0;
> +
> + timeout_remain = MAX_SCHEDULE_TIMEOUT;
> + if (timeout) {
> + if (WARN_ON(*timeout < 0))
> + return -EINVAL;
> +
> + if (*timeout == 0)
> + return -ETIME;
> +
> + /* Record current time in case interrupted, or wedged */
> + timeout_remain = nsecs_to_jiffies_timeout(*timeout);
> + *timeout += ktime_get_raw_ns();
> + }
> +
> + trace_i915_gem_request_wait_begin(req);
> +
> + /* This client is about to stall waiting for the GPU. In many cases
> + * this is undesirable and limits the throughput of the system, as
> + * many clients cannot continue processing user input/output whilst
> + * blocked. RPS autotuning may take tens of milliseconds to respond
> + * to the GPU load and thus incurs additional latency for the client.
> + * We can circumvent that by promoting the GPU frequency to maximum
> + * before we wait. This makes the GPU throttle up much more quickly
> + * (good for benchmarks and user experience, e.g. window animations),
> + * but at a cost of spending more power processing the workload
> + * (bad for battery). Not all clients even want their results
> + * immediately and for them we should just let the GPU select its own
> + * frequency to maximise efficiency. To prevent a single client from
> + * forcing the clocks too high for the whole system, we only allow
> + * each client to waitboost once in a busy period.
> + */
> + if (INTEL_GEN(req->i915) >= 6)
> + gen6_rps_boost(req->i915, rps, req->emitted_jiffies);
> +
> + /* Optimistic spin for the next ~jiffie before touching IRQs */
> + if (i915_spin_request(req, state, 5))
> + goto complete;
> +
> + set_current_state(state);
> + add_wait_queue(&req->i915->gpu_error.wait_queue, &reset);
> +
> + intel_wait_init(&wait, req->seqno);
> + if (intel_engine_add_wait(req->engine, &wait))
> + /* In order to check that we haven't missed the interrupt
> + * as we enabled it, we need to kick ourselves to do a
> + * coherent check on the seqno before we sleep.
> + */
> + goto wakeup;
> +
> + for (;;) {
> + if (signal_pending_state(state, current)) {
> + ret = -ERESTARTSYS;
> + break;
> + }
> +
> + timeout_remain = io_schedule_timeout(timeout_remain);
> + if (timeout_remain == 0) {
> + ret = -ETIME;
> + break;
> + }
> +
> + if (intel_wait_complete(&wait))
> + break;
> +
> + set_current_state(state);
> +
> +wakeup:
> + /* Carefully check if the request is complete, giving time
> + * for the seqno to be visible following the interrupt.
> + * We also have to check in case we are kicked by the GPU
> + * reset in order to drop the struct_mutex.
> + */
> + if (__i915_request_irq_complete(req))
> + break;
> +
> + /* Only spin if we know the GPU is processing this request */
> + if (i915_spin_request(req, state, 2))
> + break;
> + }
> + remove_wait_queue(&req->i915->gpu_error.wait_queue, &reset);
> +
> + intel_engine_remove_wait(req->engine, &wait);
> + __set_current_state(TASK_RUNNING);
> +complete:
> + trace_i915_gem_request_wait_end(req);
> +
> + if (timeout) {
> + *timeout -= ktime_get_raw_ns();
> + if (*timeout < 0)
> + *timeout = 0;
> +
> + /*
> + * Apparently ktime isn't accurate enough and occasionally has a
> + * bit of mismatch in the jiffies<->nsecs<->ktime loop. So patch
> + * things up to make the test happy. We allow up to 1 jiffy.
> + *
> + * This is a regrssion from the timespec->ktime conversion.
> + */
> + if (ret == -ETIME && *timeout < jiffies_to_usecs(1)*1000)
> + *timeout = 0;
> + }
> +
> + if (rps && req->seqno == req->engine->last_submitted_seqno) {
> + /* The GPU is now idle and this client has stalled.
> + * Since no other client has submitted a request in the
> + * meantime, assume that this client is the only one
> + * supplying work to the GPU but is unable to keep that
> + * work supplied because it is waiting. Since the GPU is
> + * then never kept fully busy, RPS autoclocking will
> + * keep the clocks relatively low, causing further delays.
> + * Compensate by giving the synchronous client credit for
> + * a waitboost next time.
> + */
> + spin_lock(&req->i915->rps.client_lock);
> + list_del_init(&rps->link);
> + spin_unlock(&req->i915->rps.client_lock);
> + }
> +
> + return ret;
> +}
> +
> +/**
> + * Waits for a request to be signaled, and cleans up the
> + * request and object lists appropriately for that event.
> + */
> +int i915_wait_request(struct drm_i915_gem_request *req)
> +{
> + int ret;
> +
> + GEM_BUG_ON(!req);
> + lockdep_assert_held(&req->i915->drm.struct_mutex);
> +
> + ret = __i915_wait_request(req, req->i915->mm.interruptible, NULL, NULL);
> + if (ret)
> + return ret;
> +
> + /* If the GPU hung, we want to keep the requests to find the guilty. */
> + if (!i915_reset_in_progress(&req->i915->gpu_error))
> + i915_gem_request_retire_upto(req);
> +
> + return 0;
> +}
> +
> +void i915_gem_request_free(struct kref *req_ref)
> +{
> + struct drm_i915_gem_request *req =
> + container_of(req_ref, typeof(*req), ref);
> + kmem_cache_free(req->i915->requests, req);
> +}
> diff --git a/drivers/gpu/drm/i915/i915_gem_request.h b/drivers/gpu/drm/i915/i915_gem_request.h
> new file mode 100644
> index 000000000000..ea700befcc28
> --- /dev/null
> +++ b/drivers/gpu/drm/i915/i915_gem_request.h
> @@ -0,0 +1,238 @@
> +/*
> + * Copyright © 2008-2015 Intel Corporation
> + *
> + * Permission is hereby granted, free of charge, to any person obtaining a
> + * copy of this software and associated documentation files (the "Software"),
> + * to deal in the Software without restriction, including without limitation
> + * the rights to use, copy, modify, merge, publish, distribute, sublicense,
> + * and/or sell copies of the Software, and to permit persons to whom the
> + * Software is furnished to do so, subject to the following conditions:
> + *
> + * The above copyright notice and this permission notice (including the next
> + * paragraph) shall be included in all copies or substantial portions of the
> + * Software.
> + *
> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
> + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
> + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
> + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
> + * IN THE SOFTWARE.
> + *
> + */
> +
> +#ifndef I915_GEM_REQUEST_H
> +#define I915_GEM_REQUEST_H
> +
> +/**
> + * Request queue structure.
> + *
> + * The request queue allows us to note sequence numbers that have been emitted
> + * and may be associated with active buffers to be retired.
> + *
> + * By keeping this list, we can avoid having to do questionable sequence
> + * number comparisons on buffer last_read|write_seqno. It also allows an
> + * emission time to be associated with the request for tracking how far ahead
> + * of the GPU the submission is.
> + *
> + * The requests are reference counted, so upon creation they should have an
> + * initial reference taken using kref_init
> + */
> +struct drm_i915_gem_request {
> + struct kref ref;
> +
> + /** On Which ring this request was generated */
> + struct drm_i915_private *i915;
> +
> + /**
> + * Context and ring buffer related to this request
> + * Contexts are refcounted, so when this request is associated with a
> + * context, we must increment the context's refcount, to guarantee that
> + * it persists while any request is linked to it. Requests themselves
> + * are also refcounted, so the request will only be freed when the last
> + * reference to it is dismissed, and the code in
> + * i915_gem_request_free() will then decrement the refcount on the
> + * context.
> + */
> + struct i915_gem_context *ctx;
> + struct intel_engine_cs *engine;
> + struct intel_ringbuffer *ringbuf;
> + struct intel_signal_node signaling;
> +
> + /** GEM sequence number associated with the previous request,
> + * when the HWS breadcrumb is equal to this the GPU is processing
> + * this request.
> + */
> + u32 previous_seqno;
> +
> + /** GEM sequence number associated with this request,
> + * when the HWS breadcrumb is equal or greater than this the GPU
> + * has finished processing this request.
> + */
> + u32 seqno;
> +
> + /** Position in the ringbuffer of the start of the request */
> + u32 head;
> +
> + /**
> + * Position in the ringbuffer of the start of the postfix.
> + * This is required to calculate the maximum available ringbuffer
> + * space without overwriting the postfix.
> + */
> + u32 postfix;
> +
> + /** Position in the ringbuffer of the end of the whole request */
> + u32 tail;
> +
> + /** Preallocate space in the ringbuffer for the emitting the request */
> + u32 reserved_space;
> +
> + /**
> + * Context related to the previous request.
> + * As the contexts are accessed by the hardware until the switch is
> + * completed to a new context, the hardware may still be writing
> + * to the context object after the breadcrumb is visible. We must
> + * not unpin/unbind/prune that object whilst still active and so
> + * we keep the previous context pinned until the following (this)
> + * request is retired.
> + */
> + struct i915_gem_context *previous_context;
> +
> + /** Batch buffer related to this request if any (used for
> + * error state dump only).
> + */
> + struct drm_i915_gem_object *batch_obj;
> +
> + /** Time at which this request was emitted, in jiffies. */
> + unsigned long emitted_jiffies;
> +
> + /** global list entry for this request */
> + struct list_head list;
> +
> + struct drm_i915_file_private *file_priv;
> + /** file_priv list entry for this request */
> + struct list_head client_list;
> +
> + /** process identifier submitting this request */
> + struct pid *pid;
> +
> + /**
> + * The ELSP only accepts two elements at a time, so we queue
> + * context/tail pairs on a given queue (ring->execlist_queue) until the
> + * hardware is available. The queue serves a double purpose: we also use
> + * it to keep track of the up to 2 contexts currently in the hardware
> + * (usually one in execution and the other queued up by the GPU): We
> + * only remove elements from the head of the queue when the hardware
> + * informs us that an element has been completed.
> + *
> + * All accesses to the queue are mediated by a spinlock
> + * (ring->execlist_lock).
> + */
> +
> + /** Execlist link in the submission queue.*/
> + struct list_head execlist_link;
> +
> + /** Execlists no. of times this request has been sent to the ELSP */
> + int elsp_submitted;
> +
> + /** Execlists context hardware id. */
> + unsigned int ctx_hw_id;
> +};
> +
> +struct drm_i915_gem_request * __must_check
> +i915_gem_request_alloc(struct intel_engine_cs *engine,
> + struct i915_gem_context *ctx);
> +void i915_gem_request_free(struct kref *req_ref);
> +int i915_gem_request_add_to_client(struct drm_i915_gem_request *req,
> + struct drm_file *file);
> +void i915_gem_request_retire_upto(struct drm_i915_gem_request *req);
> +
> +static inline u32
> +i915_gem_request_get_seqno(struct drm_i915_gem_request *req)
> +{
> + return req ? req->seqno : 0;
> +}
> +
> +static inline struct intel_engine_cs *
> +i915_gem_request_get_engine(struct drm_i915_gem_request *req)
> +{
> + return req ? req->engine : NULL;
> +}
> +
> +static inline struct drm_i915_gem_request *
> +i915_gem_request_reference(struct drm_i915_gem_request *req)
> +{
> + if (req)
> + kref_get(&req->ref);
> + return req;
> +}
> +
> +static inline void
> +i915_gem_request_unreference(struct drm_i915_gem_request *req)
> +{
> + kref_put(&req->ref, i915_gem_request_free);
> +}
> +
> +static inline void i915_gem_request_assign(struct drm_i915_gem_request **pdst,
> + struct drm_i915_gem_request *src)
> +{
> + if (src)
> + i915_gem_request_reference(src);
> +
> + if (*pdst)
> + i915_gem_request_unreference(*pdst);
> +
> + *pdst = src;
> +}
> +
> +void __i915_add_request(struct drm_i915_gem_request *req,
> + struct drm_i915_gem_object *batch_obj,
> + bool flush_caches);
> +#define i915_add_request(req) \
> + __i915_add_request(req, NULL, true)
> +#define i915_add_request_no_flush(req) \
> + __i915_add_request(req, NULL, false)
> +
> +struct intel_rps_client;
> +
> +int __i915_wait_request(struct drm_i915_gem_request *req,
> + bool interruptible,
> + s64 *timeout,
> + struct intel_rps_client *rps);
> +int __must_check i915_wait_request(struct drm_i915_gem_request *req);
> +
> +static inline u32 intel_engine_get_seqno(struct intel_engine_cs *engine);
> +
> +/**
> + * Returns true if seq1 is later than seq2.
> + */
> +static inline bool i915_seqno_passed(u32 seq1, u32 seq2)
> +{
> + return (s32)(seq1 - seq2) >= 0;
> +}
> +
> +static inline bool
> +i915_gem_request_started(const struct drm_i915_gem_request *req)
> +{
> + return i915_seqno_passed(intel_engine_get_seqno(req->engine),
> + req->previous_seqno);
> +}
> +
> +static inline bool
> +i915_gem_request_completed(const struct drm_i915_gem_request *req)
> +{
> + return i915_seqno_passed(intel_engine_get_seqno(req->engine),
> + req->seqno);
> +}
> +
> +bool __i915_spin_request(const struct drm_i915_gem_request *request,
> + int state, unsigned long timeout_us);
> +static inline bool i915_spin_request(const struct drm_i915_gem_request *request,
> + int state, unsigned long timeout_us)
> +{
> + return (i915_gem_request_started(request) &&
> + __i915_spin_request(request, state, timeout_us));
> +}
> +
> +#endif /* I915_GEM_REQUEST_H */
> --
> 2.8.1
>
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
