diff options
Diffstat (limited to 'drivers/gpu/drm/i915/i915_request.c')
| -rw-r--r-- | drivers/gpu/drm/i915/i915_request.c | 78 |
1 files changed, 57 insertions, 21 deletions
diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c index 62fad16a55e8..4399941236cb 100644 --- a/drivers/gpu/drm/i915/i915_request.c +++ b/drivers/gpu/drm/i915/i915_request.c @@ -31,6 +31,8 @@ #include <linux/sched/signal.h> #include <linux/sched/mm.h> +#include <drm/drm_print.h> + #include "gem/i915_gem_context.h" #include "gt/intel_breadcrumbs.h" #include "gt/intel_context.h" @@ -43,16 +45,15 @@ #include "gt/intel_rps.h" #include "i915_active.h" +#include "i915_config.h" #include "i915_deps.h" #include "i915_driver.h" #include "i915_drv.h" #include "i915_trace.h" -#include "intel_pm.h" struct execute_cb { struct irq_work work; struct i915_sw_fence *fence; - struct i915_request *signal; }; static struct kmem_cache *slab_requests; @@ -134,9 +135,7 @@ static void i915_fence_release(struct dma_fence *fence) i915_sw_fence_fini(&rq->semaphore); /* - * Keep one request on each engine for reserved use under mempressure - * do not use with virtual engines as this really is only needed for - * kernel contexts. + * Keep one request on each engine for reserved use under mempressure. * * We do not hold a reference to the engine here and so have to be * very careful in what rq->engine we poke. The virtual engine is @@ -166,8 +165,7 @@ static void i915_fence_release(struct dma_fence *fence) * know that if the rq->execution_mask is a single bit, rq->engine * can be a physical engine with the exact corresponding mask. */ - if (!intel_engine_is_virtual(rq->engine) && - is_power_of_2(rq->execution_mask) && + if (is_power_of_2(rq->execution_mask) && !cmpxchg(&rq->engine->request_pool, NULL, rq)) return; @@ -277,11 +275,6 @@ i915_request_active_engine(struct i915_request *rq, return ret; } -static void __rq_init_watchdog(struct i915_request *rq) -{ - rq->watchdog.timer.function = NULL; -} - static enum hrtimer_restart __rq_watchdog_expired(struct hrtimer *hrtimer) { struct i915_request *rq = @@ -290,7 +283,7 @@ static enum hrtimer_restart __rq_watchdog_expired(struct hrtimer *hrtimer) if (!i915_request_completed(rq)) { if (llist_add(&rq->watchdog.link, >->watchdog.list)) - schedule_work(>->watchdog.work); + queue_work(gt->i915->unordered_wq, >->watchdog.work); } else { i915_request_put(rq); } @@ -298,6 +291,13 @@ static enum hrtimer_restart __rq_watchdog_expired(struct hrtimer *hrtimer) return HRTIMER_NORESTART; } +static void __rq_init_watchdog(struct i915_request *rq) +{ + struct i915_request_watchdog *wdg = &rq->watchdog; + + hrtimer_setup(&wdg->timer, __rq_watchdog_expired, CLOCK_MONOTONIC, HRTIMER_MODE_REL); +} + static void __rq_arm_watchdog(struct i915_request *rq) { struct i915_request_watchdog *wdg = &rq->watchdog; @@ -308,8 +308,6 @@ static void __rq_arm_watchdog(struct i915_request *rq) i915_request_get(rq); - hrtimer_init(&wdg->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); - wdg->timer.function = __rq_watchdog_expired; hrtimer_start_range_ns(&wdg->timer, ns_to_ktime(ce->watchdog.timeout_us * NSEC_PER_USEC), @@ -321,7 +319,7 @@ static void __rq_cancel_watchdog(struct i915_request *rq) { struct i915_request_watchdog *wdg = &rq->watchdog; - if (wdg->timer.function && hrtimer_try_to_cancel(&wdg->timer) > 0) + if (hrtimer_try_to_cancel(&wdg->timer) > 0) i915_request_put(rq); } @@ -476,7 +474,7 @@ static bool __request_in_flight(const struct i915_request *signal) * to avoid tearing.] * * Note that the read of *execlists->active may race with the promotion - * of execlists->pending[] to execlists->inflight[], overwritting + * of execlists->pending[] to execlists->inflight[], overwriting * the value at *execlists->active. This is fine. The promotion implies * that we received an ACK from the HW, and so the context is not * stuck -- if we do not see ourselves in *active, the inflight status @@ -1220,7 +1218,7 @@ emit_semaphore_wait(struct i915_request *to, /* * If this or its dependents are waiting on an external fence * that may fail catastrophically, then we want to avoid using - * sempahores as they bypass the fence signaling metadata, and we + * semaphores as they bypass the fence signaling metadata, and we * lose the fence->error propagation. */ if (from->sched.flags & I915_SCHED_HAS_EXTERNAL_CHAIN) @@ -1353,7 +1351,7 @@ __i915_request_await_external(struct i915_request *rq, struct dma_fence *fence) { mark_external(rq); return i915_sw_fence_await_dma_fence(&rq->submit, fence, - i915_fence_context_timeout(rq->engine->i915, + i915_fence_context_timeout(rq->i915, fence->context), I915_FENCE_GFP); } @@ -1621,6 +1619,20 @@ i915_request_await_object(struct i915_request *to, return ret; } +static void i915_request_await_huc(struct i915_request *rq) +{ + struct intel_huc *huc = &rq->context->engine->gt->uc.huc; + + /* don't stall kernel submissions! */ + if (!rcu_access_pointer(rq->context->gem_context)) + return; + + if (intel_huc_wait_required(huc)) + i915_sw_fence_await_sw_fence(&rq->submit, + &huc->delayed_load.fence, + &rq->hucq); +} + static struct i915_request * __i915_request_ensure_parallel_ordering(struct i915_request *rq, struct intel_timeline *timeline) @@ -1647,6 +1659,11 @@ __i915_request_ensure_parallel_ordering(struct i915_request *rq, request_to_parent(rq)->parallel.last_rq = i915_request_get(rq); + /* + * Users have to put a reference potentially got by + * __i915_active_fence_set() to the returned request + * when no longer needed + */ return to_request(__i915_active_fence_set(&timeline->last_request, &rq->fence)); } @@ -1693,6 +1710,10 @@ __i915_request_ensure_ordering(struct i915_request *rq, 0); } + /* + * Users have to put the reference to prev potentially got + * by __i915_active_fence_set() when no longer needed + */ return prev; } @@ -1703,6 +1724,16 @@ __i915_request_add_to_timeline(struct i915_request *rq) struct i915_request *prev; /* + * Media workloads may require HuC, so stall them until HuC loading is + * complete. Note that HuC not being loaded when a user submission + * arrives can only happen when HuC is loaded via GSC and in that case + * we still expect the window between us starting to accept submissions + * and HuC loading completion to be small (a few hundred ms). + */ + if (rq->engine->class == VIDEO_DECODE_CLASS) + i915_request_await_huc(rq); + + /* * Dependency tracking and request ordering along the timeline * is special cased so that we can eliminate redundant ordering * operations while building the request (we know that the timeline @@ -1736,6 +1767,8 @@ __i915_request_add_to_timeline(struct i915_request *rq) prev = __i915_request_ensure_ordering(rq, timeline); else prev = __i915_request_ensure_parallel_ordering(rq, timeline); + if (prev) + i915_request_put(prev); /* * Make sure that no request gazumped us - if it was allocated after @@ -2153,7 +2186,7 @@ void i915_request_show(struct drm_printer *m, const char *prefix, int indent) { - const char *name = rq->fence.ops->get_timeline_name((struct dma_fence *)&rq->fence); + const char __rcu *timeline; char buf[80] = ""; int x = 0; @@ -2189,6 +2222,8 @@ void i915_request_show(struct drm_printer *m, x = print_sched_attr(&rq->sched.attr, buf, x, sizeof(buf)); + rcu_read_lock(); + timeline = dma_fence_timeline_name((struct dma_fence *)&rq->fence); drm_printf(m, "%s%.*s%c %llx:%lld%s%s %s @ %dms: %s\n", prefix, indent, " ", queue_status(rq), @@ -2197,7 +2232,8 @@ void i915_request_show(struct drm_printer *m, fence_status(rq), buf, jiffies_to_msecs(jiffies - rq->emitted_jiffies), - name); + rcu_dereference(timeline)); + rcu_read_unlock(); } static bool engine_match_ring(struct intel_engine_cs *engine, struct i915_request *rq) |