diff options
Diffstat (limited to 'drivers/gpu/drm/i915/gt/intel_lrc.c')
| -rw-r--r-- | drivers/gpu/drm/i915/gt/intel_lrc.c | 838 |
1 files changed, 670 insertions, 168 deletions
diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c b/drivers/gpu/drm/i915/gt/intel_lrc.c index 0cf0f6fae675..683014e7bc51 100644 --- a/drivers/gpu/drm/i915/gt/intel_lrc.c +++ b/drivers/gpu/drm/i915/gt/intel_lrc.c @@ -176,8 +176,6 @@ /* Typical size of the average request (2 pipecontrols and a MI_BB) */ #define EXECLISTS_REQUEST_SIZE 64 /* bytes */ -#define WA_TAIL_DWORDS 2 -#define WA_TAIL_BYTES (sizeof(u32) * WA_TAIL_DWORDS) struct virtual_engine { struct intel_engine_cs base; @@ -237,7 +235,8 @@ static void execlists_init_reg_state(u32 *reg_state, bool close); static void __execlists_update_reg_state(const struct intel_context *ce, - const struct intel_engine_cs *engine); + const struct intel_engine_cs *engine, + u32 head); static void mark_eio(struct i915_request *rq) { @@ -246,7 +245,7 @@ static void mark_eio(struct i915_request *rq) GEM_BUG_ON(i915_request_signaled(rq)); - dma_fence_set_error(&rq->fence, -EIO); + i915_request_set_error_once(rq, -EIO); i915_request_mark_complete(rq); } @@ -294,7 +293,7 @@ static inline struct i915_priolist *to_priolist(struct rb_node *rb) static inline int rq_prio(const struct i915_request *rq) { - return rq->sched.attr.priority; + return READ_ONCE(rq->sched.attr.priority); } static int effective_prio(const struct i915_request *rq) @@ -985,6 +984,8 @@ __unwind_incomplete_requests(struct intel_engine_cs *engine) GEM_BUG_ON(RB_EMPTY_ROOT(&engine->execlists.queue.rb_root)); list_move(&rq->sched.link, pl); + set_bit(I915_FENCE_FLAG_PQUEUE, &rq->fence.flags); + active = rq; } else { struct intel_engine_cs *owner = rq->context->engine; @@ -1003,7 +1004,7 @@ __unwind_incomplete_requests(struct intel_engine_cs *engine) i915_request_cancel_breadcrumb(rq); spin_unlock(&rq->lock); } - rq->engine = owner; + WRITE_ONCE(rq->engine, owner); owner->submit_request(rq); active = NULL; } @@ -1184,17 +1185,58 @@ static void reset_active(struct i915_request *rq, head = rq->tail; else head = active_request(ce->timeline, rq)->head; - ce->ring->head = intel_ring_wrap(ce->ring, head); - intel_ring_update_space(ce->ring); + head = intel_ring_wrap(ce->ring, head); /* Scrub the context image to prevent replaying the previous batch */ restore_default_state(ce, engine); - __execlists_update_reg_state(ce, engine); + __execlists_update_reg_state(ce, engine, head); /* We've switched away, so this should be a no-op, but intent matters */ ce->lrc_desc |= CTX_DESC_FORCE_RESTORE; } +static u32 intel_context_get_runtime(const struct intel_context *ce) +{ + /* + * We can use either ppHWSP[16] which is recorded before the context + * switch (and so excludes the cost of context switches) or use the + * value from the context image itself, which is saved/restored earlier + * and so includes the cost of the save. + */ + return READ_ONCE(ce->lrc_reg_state[CTX_TIMESTAMP]); +} + +static void st_update_runtime_underflow(struct intel_context *ce, s32 dt) +{ +#if IS_ENABLED(CONFIG_DRM_I915_SELFTEST) + ce->runtime.num_underflow += dt < 0; + ce->runtime.max_underflow = max_t(u32, ce->runtime.max_underflow, -dt); +#endif +} + +static void intel_context_update_runtime(struct intel_context *ce) +{ + u32 old; + s32 dt; + + if (intel_context_is_barrier(ce)) + return; + + old = ce->runtime.last; + ce->runtime.last = intel_context_get_runtime(ce); + dt = ce->runtime.last - old; + + if (unlikely(dt <= 0)) { + CE_TRACE(ce, "runtime underflow: last=%u, new=%u, delta=%d\n", + old, ce->runtime.last, dt); + st_update_runtime_underflow(ce, dt); + return; + } + + ewma_runtime_add(&ce->runtime.avg, dt); + ce->runtime.total += dt; +} + static inline struct intel_engine_cs * __execlists_schedule_in(struct i915_request *rq) { @@ -1209,12 +1251,12 @@ __execlists_schedule_in(struct i915_request *rq) if (IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM)) execlists_check_context(ce, engine); + ce->lrc_desc &= ~GENMASK_ULL(47, 37); if (ce->tag) { /* Use a fixed tag for OA and friends */ ce->lrc_desc |= (u64)ce->tag << 32; } else { /* We don't need a strict matching tag, just different values */ - ce->lrc_desc &= ~GENMASK_ULL(47, 37); ce->lrc_desc |= (u64)(++engine->context_tag % NUM_CONTEXT_TAG) << GEN11_SW_CTX_ID_SHIFT; @@ -1274,10 +1316,11 @@ __execlists_schedule_out(struct i915_request *rq, * If we have just completed this context, the engine may now be * idle and we want to re-enter powersaving. */ - if (list_is_last(&rq->link, &ce->timeline->requests) && + if (list_is_last_rcu(&rq->link, &ce->timeline->requests) && i915_request_completed(rq)) intel_engine_add_retire(engine, ce->timeline); + intel_context_update_runtime(ce); intel_engine_context_out(engine); execlists_context_status_change(rq, INTEL_CONTEXT_SCHEDULE_OUT); intel_gt_pm_put_async(engine->gt); @@ -1319,7 +1362,7 @@ static u64 execlists_update_context(struct i915_request *rq) { struct intel_context *ce = rq->context; u64 desc = ce->lrc_desc; - u32 tail; + u32 tail, prev; /* * WaIdleLiteRestore:bdw,skl @@ -1332,9 +1375,15 @@ static u64 execlists_update_context(struct i915_request *rq) * subsequent resubmissions (for lite restore). Should that fail us, * and we try and submit the same tail again, force the context * reload. + * + * If we need to return to a preempted context, we need to skip the + * lite-restore and force it to reload the RING_TAIL. Otherwise, the + * HW has a tendency to ignore us rewinding the TAIL to the end of + * an earlier request. */ tail = intel_ring_set_tail(rq->ring, rq->tail); - if (unlikely(ce->lrc_reg_state[CTX_RING_TAIL] == tail)) + prev = ce->lrc_reg_state[CTX_RING_TAIL]; + if (unlikely(intel_ring_direction(rq->ring, tail, prev) <= 0)) desc |= CTX_DESC_FORCE_RESTORE; ce->lrc_reg_state[CTX_RING_TAIL] = tail; rq->tail = rq->wa_tail; @@ -1387,15 +1436,26 @@ trace_ports(const struct intel_engine_execlists *execlists, ports[1] ? ports[1]->fence.seqno : 0); } +static inline bool +reset_in_progress(const struct intel_engine_execlists *execlists) +{ + return unlikely(!__tasklet_is_enabled(&execlists->tasklet)); +} + static __maybe_unused bool assert_pending_valid(const struct intel_engine_execlists *execlists, const char *msg) { struct i915_request * const *port, *rq; struct intel_context *ce = NULL; + bool sentinel = false; trace_ports(execlists, msg, execlists->pending); + /* We may be messing around with the lists during reset, lalala */ + if (reset_in_progress(execlists)) + return true; + if (!execlists->pending[0]) { GEM_TRACE_ERR("Nothing pending for promotion!\n"); return false; @@ -1422,6 +1482,26 @@ assert_pending_valid(const struct intel_engine_execlists *execlists, } ce = rq->context; + /* + * Sentinels are supposed to be lonely so they flush the + * current exection off the HW. Check that they are the + * only request in the pending submission. + */ + if (sentinel) { + GEM_TRACE_ERR("context:%llx after sentinel in pending[%zd]\n", + ce->timeline->fence_context, + port - execlists->pending); + return false; + } + + sentinel = i915_request_has_sentinel(rq); + if (sentinel && port != execlists->pending) { + GEM_TRACE_ERR("sentinel context:%llx not in prime position[%zd]\n", + ce->timeline->fence_context, + port - execlists->pending); + return false; + } + /* Hold tightly onto the lock to prevent concurrent retires! */ if (!spin_trylock_irqsave(&rq->lock, flags)) continue; @@ -1517,6 +1597,11 @@ static bool can_merge_ctx(const struct intel_context *prev, return true; } +static unsigned long i915_request_flags(const struct i915_request *rq) +{ + return READ_ONCE(rq->fence.flags); +} + static bool can_merge_rq(const struct i915_request *prev, const struct i915_request *next) { @@ -1534,13 +1619,15 @@ static bool can_merge_rq(const struct i915_request *prev, if (i915_request_completed(next)) return true; - if (unlikely((prev->fence.flags ^ next->fence.flags) & - (I915_FENCE_FLAG_NOPREEMPT | I915_FENCE_FLAG_SENTINEL))) + if (unlikely((i915_request_flags(prev) ^ i915_request_flags(next)) & + (BIT(I915_FENCE_FLAG_NOPREEMPT) | + BIT(I915_FENCE_FLAG_SENTINEL)))) return false; if (!can_merge_ctx(prev->context, next->context)) return false; + GEM_BUG_ON(i915_seqno_passed(prev->fence.seqno, next->fence.seqno)); return true; } @@ -1576,7 +1663,7 @@ static bool virtual_matches(const struct virtual_engine *ve, } static void virtual_xfer_breadcrumbs(struct virtual_engine *ve, - struct intel_engine_cs *engine) + struct i915_request *rq) { struct intel_engine_cs *old = ve->siblings[0]; @@ -1584,23 +1671,32 @@ static void virtual_xfer_breadcrumbs(struct virtual_engine *ve, spin_lock(&old->breadcrumbs.irq_lock); if (!list_empty(&ve->context.signal_link)) { - list_move_tail(&ve->context.signal_link, - &engine->breadcrumbs.signalers); - intel_engine_signal_breadcrumbs(engine); + list_del_init(&ve->context.signal_link); + + /* + * We cannot acquire the new engine->breadcrumbs.irq_lock + * (as we are holding a breadcrumbs.irq_lock already), + * so attach this request to the signaler on submission. + * The queued irq_work will occur when we finally drop + * the engine->active.lock after dequeue. + */ + set_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT, &rq->fence.flags); + + /* Also transfer the pending irq_work for the old breadcrumb. */ + intel_engine_signal_breadcrumbs(rq->engine); } spin_unlock(&old->breadcrumbs.irq_lock); } -static struct i915_request * -last_active(const struct intel_engine_execlists *execlists) -{ - struct i915_request * const *last = READ_ONCE(execlists->active); +#define for_each_waiter(p__, rq__) \ + list_for_each_entry_lockless(p__, \ + &(rq__)->sched.waiters_list, \ + wait_link) - while (*last && i915_request_completed(*last)) - last++; - - return *last; -} +#define for_each_signaler(p__, rq__) \ + list_for_each_entry_rcu(p__, \ + &(rq__)->sched.signalers_list, \ + signal_link) static void defer_request(struct i915_request *rq, struct list_head * const pl) { @@ -1619,7 +1715,7 @@ static void defer_request(struct i915_request *rq, struct list_head * const pl) GEM_BUG_ON(i915_request_is_active(rq)); list_move_tail(&rq->sched.link, pl); - list_for_each_entry(p, &rq->sched.waiters_list, wait_link) { + for_each_waiter(p, rq) { struct i915_request *w = container_of(p->waiter, typeof(*w), sched); @@ -1632,8 +1728,8 @@ static void defer_request(struct i915_request *rq, struct list_head * const pl) !i915_request_completed(rq)); GEM_BUG_ON(i915_request_is_active(w)); - if (list_empty(&w->sched.link)) - continue; /* Not yet submitted; unready */ + if (!i915_request_is_ready(w)) + continue; if (rq_prio(w) < rq_prio(rq)) continue; @@ -1665,11 +1761,9 @@ need_timeslice(struct intel_engine_cs *engine, const struct i915_request *rq) if (!intel_engine_has_timeslices(engine)) return false; - if (list_is_last(&rq->sched.link, &engine->active.requests)) - return false; - - hint = max(rq_prio(list_next_entry(rq, sched.link)), - engine->execlists.queue_priority_hint); + hint = engine->execlists.queue_priority_hint; + if (!list_is_last(&rq->sched.link, &engine->active.requests)) + hint = max(hint, rq_prio(list_next_entry(rq, sched.link))); return hint >= effective_prio(rq); } @@ -1692,12 +1786,13 @@ timeslice(const struct intel_engine_cs *engine) static unsigned long active_timeslice(const struct intel_engine_cs *engine) { - const struct i915_request *rq = *engine->execlists.active; + const struct intel_engine_execlists *execlists = &engine->execlists; + const struct i915_request *rq = *execlists->active; if (!rq || i915_request_completed(rq)) return 0; - if (engine->execlists.switch_priority_hint < effective_prio(rq)) + if (READ_ONCE(execlists->switch_priority_hint) < effective_prio(rq)) return 0; return timeslice(engine); @@ -1711,16 +1806,29 @@ static void set_timeslice(struct intel_engine_cs *engine) set_timer_ms(&engine->execlists.timer, active_timeslice(engine)); } +static void start_timeslice(struct intel_engine_cs *engine) +{ + struct intel_engine_execlists *execlists = &engine->execlists; + int prio = queue_prio(execlists); + + WRITE_ONCE(execlists->switch_priority_hint, prio); + if (prio == INT_MIN) + return; + + if (timer_pending(&execlists->timer)) + return; + + set_timer_ms(&execlists->timer, timeslice(engine)); +} + static void record_preemption(struct intel_engine_execlists *execlists) { (void)I915_SELFTEST_ONLY(execlists->preempt_hang.count++); } -static unsigned long active_preempt_timeout(struct intel_engine_cs *engine) +static unsigned long active_preempt_timeout(struct intel_engine_cs *engine, + const struct i915_request *rq) { - struct i915_request *rq; - - rq = last_active(&engine->execlists); if (!rq) return 0; @@ -1731,13 +1839,14 @@ static unsigned long active_preempt_timeout(struct intel_engine_cs *engine) return READ_ONCE(engine->props.preempt_timeout_ms); } -static void set_preempt_timeout(struct intel_engine_cs *engine) +static void set_preempt_timeout(struct intel_engine_cs *engine, + const struct i915_request *rq) { if (!intel_engine_has_preempt_reset(engine)) return; set_timer_ms(&engine->execlists.preempt, - active_preempt_timeout(engine)); + active_preempt_timeout(engine, rq)); } static inline void clear_ports(struct i915_request **ports, int count) @@ -1750,6 +1859,7 @@ static void execlists_dequeue(struct intel_engine_cs *engine) struct intel_engine_execlists * const execlists = &engine->execlists; struct i915_request **port = execlists->pending; struct i915_request ** const last_port = port + execlists->port_mask; + struct i915_request * const *active; struct i915_request *last; struct rb_node *rb; bool submit = false; @@ -1804,7 +1914,10 @@ static void execlists_dequeue(struct intel_engine_cs *engine) * i.e. we will retrigger preemption following the ack in case * of trouble. */ - last = last_active(execlists); + active = READ_ONCE(execlists->active); + while ((last = *active) && i915_request_completed(last)) + active++; + if (last) { if (need_preempt(engine, last, rb)) { ENGINE_TRACE(engine, @@ -1831,14 +1944,6 @@ static void execlists_dequeue(struct intel_engine_cs *engine) */ __unwind_incomplete_requests(engine); - /* - * If we need to return to the preempted context, we - * need to skip the lite-restore and force it to - * reload the RING_TAIL. Otherwise, the HW has a - * tendency to ignore us rewinding the TAIL to the - * end of an earlier request. - */ - last->context->lrc_desc |= CTX_DESC_FORCE_RESTORE; last = NULL; } else if (need_timeslice(engine, last) && timer_expired(&engine->execlists.timer)) { @@ -1882,11 +1987,7 @@ static void execlists_dequeue(struct intel_engine_cs *engine) * Even if ELSP[1] is occupied and not worthy * of timeslices, our queue might be. */ - if (!execlists->timer.expires && - need_timeslice(engine, last)) - set_timer_ms(&execlists->timer, - timeslice(engine)); - + start_timeslice(engine); return; } } @@ -1921,7 +2022,8 @@ static void execlists_dequeue(struct intel_engine_cs *engine) if (last && !can_merge_rq(last, rq)) { spin_unlock(&ve->base.active.lock); - return; /* leave this for another */ + start_timeslice(engine); + return; /* leave this for another sibling */ } ENGINE_TRACE(engine, @@ -1933,13 +2035,14 @@ static void execlists_dequeue(struct intel_engine_cs *engine) "", yesno(engine != ve->siblings[0])); - ve->request = NULL; - ve->base.execlists.queue_priority_hint = INT_MIN; + WRITE_ONCE(ve->request, NULL); + WRITE_ONCE(ve->base.execlists.queue_priority_hint, + INT_MIN); rb_erase_cached(rb, &execlists->virtual); RB_CLEAR_NODE(rb); GEM_BUG_ON(!(rq->execution_mask & engine->mask)); - rq->engine = engine; + WRITE_ONCE(rq->engine, engine); if (engine != ve->siblings[0]) { u32 *regs = ve->context.lrc_reg_state; @@ -1952,7 +2055,7 @@ static void execlists_dequeue(struct intel_engine_cs *engine) engine); if (!list_empty(&ve->context.signals)) - virtual_xfer_breadcrumbs(ve, engine); + virtual_xfer_breadcrumbs(ve, rq); /* * Move the bound engine to the top of the list @@ -2059,6 +2162,9 @@ static void execlists_dequeue(struct intel_engine_cs *engine) GEM_BUG_ON(last && !can_merge_ctx(last->context, rq->context)); + GEM_BUG_ON(last && + i915_seqno_passed(last->fence.seqno, + rq->fence.seqno)); submit = true; last = rq; @@ -2097,7 +2203,7 @@ done: * Skip if we ended up with exactly the same set of requests, * e.g. trying to timeslice a pair of ordered contexts */ - if (!memcmp(execlists->active, execlists->pending, + if (!memcmp(active, execlists->pending, (port - execlists->pending + 1) * sizeof(*port))) { do execlists_schedule_out(fetch_and_zero(port)); @@ -2108,7 +2214,7 @@ done: clear_ports(port + 1, last_port - port); execlists_submit_ports(engine); - set_preempt_timeout(engine); + set_preempt_timeout(engine, *active); } else { skip_submit: ring_set_paused(engine, 0); @@ -2129,6 +2235,7 @@ cancel_port_requests(struct intel_engine_execlists * const execlists) execlists_schedule_out(*port); clear_ports(execlists->inflight, ARRAY_SIZE(execlists->inflight)); + smp_wmb(); /* complete the seqlock for execlists_active() */ WRITE_ONCE(execlists->active, execlists->inflight); } @@ -2139,12 +2246,6 @@ invalidate_csb_entries(const u32 *first, const u32 *last) clflush((void *)last); } -static inline bool -reset_in_progress(const struct intel_engine_execlists *execlists) -{ - return unlikely(!__tasklet_is_enabled(&execlists->tasklet)); -} - /* * Starting with Gen12, the status has a new format: * @@ -2235,7 +2336,6 @@ static void process_csb(struct intel_engine_cs *engine) */ head = execlists->csb_head; tail = READ_ONCE(*execlists->csb_write); - ENGINE_TRACE(engine, "cs-irq head=%d, tail=%d\n", head, tail); if (unlikely(head == tail)) return; @@ -2249,6 +2349,7 @@ static void process_csb(struct intel_engine_cs *engine) */ rmb(); + ENGINE_TRACE(engine, "cs-irq head=%d, tail=%d\n", head, tail); do { bool promote; @@ -2283,11 +2384,13 @@ static void process_csb(struct intel_engine_cs *engine) if (promote) { struct i915_request * const *old = execlists->active; + GEM_BUG_ON(!assert_pending_valid(execlists, "promote")); + + ring_set_paused(engine, 0); + /* Point active to the new ELSP; prevent overwriting */ WRITE_ONCE(execlists->active, execlists->pending); - - if (!inject_preempt_hang(execlists)) - ring_set_paused(engine, 0); + smp_wmb(); /* notify execlists_active() */ /* cancel old inflight, prepare for switch */ trace_ports(execlists, "preempted", old); @@ -2295,12 +2398,12 @@ static void process_csb(struct intel_engine_cs *engine) execlists_schedule_out(*old++); /* switch pending to inflight */ - GEM_BUG_ON(!assert_pending_valid(execlists, "promote")); - WRITE_ONCE(execlists->active, - memcpy(execlists->inflight, - execlists->pending, - execlists_num_ports(execlists) * - sizeof(*execlists->pending))); + memcpy(execlists->inflight, + execlists->pending, + execlists_num_ports(execlists) * + sizeof(*execlists->pending)); + smp_wmb(); /* complete the seqlock */ + WRITE_ONCE(execlists->active, execlists->inflight); WRITE_ONCE(execlists->pending[0], NULL); } else { @@ -2315,8 +2418,37 @@ static void process_csb(struct intel_engine_cs *engine) * coherent (visible from the CPU) before the * user interrupt and CSB is processed. */ - GEM_BUG_ON(!i915_request_completed(*execlists->active) && - !reset_in_progress(execlists)); + if (GEM_SHOW_DEBUG() && + !i915_request_completed(*execlists->active) && + !reset_in_progress(execlists)) { + struct i915_request *rq __maybe_unused = + *execlists->active; + const u32 *regs __maybe_unused = + rq->context->lrc_reg_state; + + ENGINE_TRACE(engine, + "ring:{start:0x%08x, head:%04x, tail:%04x, ctl:%08x, mode:%08x}\n", + ENGINE_READ(engine, RING_START), + ENGINE_READ(engine, RING_HEAD) & HEAD_ADDR, + ENGINE_READ(engine, RING_TAIL) & TAIL_ADDR, + ENGINE_READ(engine, RING_CTL), + ENGINE_READ(engine, RING_MI_MODE)); + ENGINE_TRACE(engine, + "rq:{start:%08x, head:%04x, tail:%04x, seqno:%llx:%d, hwsp:%d}, ", + i915_ggtt_offset(rq->ring->vma), + rq->head, rq->tail, + rq->fence.context, + lower_32_bits(rq->fence.seqno), + hwsp_seqno(rq)); + ENGINE_TRACE(engine, + "ctx:{start:%08x, head:%04x, tail:%04x}, ", + regs[CTX_RING_START], + regs[CTX_RING_HEAD], + regs[CTX_RING_TAIL]); + + GEM_BUG_ON("context completed before request"); + } + execlists_schedule_out(*execlists->active++); GEM_BUG_ON(execlists->active - execlists->inflight > @@ -2344,31 +2476,348 @@ static void process_csb(struct intel_engine_cs *engine) static void __execlists_submission_tasklet(struct intel_engine_cs *const engine) { lockdep_assert_held(&engine->active.lock); - if (!engine->execlists.pending[0]) { + if (!READ_ONCE(engine->execlists.pending[0])) { rcu_read_lock(); /* protect peeking at execlists->active */ execlists_dequeue(engine); rcu_read_unlock(); } } -static noinline void preempt_reset(struct intel_engine_cs *engine) +static void __execlists_hold(struct i915_request *rq) +{ + LIST_HEAD(list); + + do { + struct i915_dependency *p; + + if (i915_request_is_active(rq)) + __i915_request_unsubmit(rq); + + clear_bit(I915_FENCE_FLAG_PQUEUE, &rq->fence.flags); + list_move_tail(&rq->sched.link, &rq->engine->active.hold); + i915_request_set_hold(rq); + RQ_TRACE(rq, "on hold\n"); + + for_each_waiter(p, rq) { + struct i915_request *w = + container_of(p->waiter, typeof(*w), sched); + + /* Leave semaphores spinning on the other engines */ + if (w->engine != rq->engine) + continue; + + if (!i915_request_is_ready(w)) + continue; + + if (i915_request_completed(w)) + continue; + + if (i915_request_on_hold(w)) + continue; + + list_move_tail(&w->sched.link, &list); + } + + rq = list_first_entry_or_null(&list, typeof(*rq), sched.link); + } while (rq); +} + +static bool execlists_hold(struct intel_engine_cs *engine, + struct i915_request *rq) +{ + spin_lock_irq(&engine->active.lock); + + if (i915_request_completed(rq)) { /* too late! */ + rq = NULL; + goto unlock; + } + + if (rq->engine != engine) { /* preempted virtual engine */ + struct virtual_engine *ve = to_virtual_engine(rq->engine); + + /* + * intel_context_inflight() is only protected by virtue + * of process_csb() being called only by the tasklet (or + * directly from inside reset while the tasklet is suspended). + * Assert that neither of those are allowed to run while we + * poke at the request queues. + */ + GEM_BUG_ON(!reset_in_progress(&engine->execlists)); + + /* + * An unsubmitted request along a virtual engine will + * remain on the active (this) engine until we are able + * to process the context switch away (and so mark the + * context as no longer in flight). That cannot have happened + * yet, otherwise we would not be hanging! + */ + spin_lock(&ve->base.active.lock); + GEM_BUG_ON(intel_context_inflight(rq->context) != engine); + GEM_BUG_ON(ve->request != rq); + ve->request = NULL; + spin_unlock(&ve->base.active.lock); + i915_request_put(rq); + + rq->engine = engine; + } + + /* + * Transfer this request onto the hold queue to prevent it + * being resumbitted to HW (and potentially completed) before we have + * released it. Since we may have already submitted following + * requests, we need to remove those as well. + */ + GEM_BUG_ON(i915_request_on_hold(rq)); + GEM_BUG_ON(rq->engine != engine); + __execlists_hold(rq); + GEM_BUG_ON(list_empty(&engine->active.hold)); + +unlock: + spin_unlock_irq(&engine->active.lock); + return rq; +} + +static bool hold_request(const struct i915_request *rq) +{ + struct i915_dependency *p; + bool result = false; + + /* + * If one of our ancestors is on hold, we must also be on hold, + * otherwise we will bypass it and execute before it. + */ + rcu_read_lock(); + for_each_signaler(p, rq) { + const struct i915_request *s = + container_of(p->signaler, typeof(*s), sched); + + if (s->engine != rq->engine) + continue; + + result = i915_request_on_hold(s); + if (result) + break; + } + rcu_read_unlock(); + + return result; +} + +static void __execlists_unhold(struct i915_request *rq) +{ + LIST_HEAD(list); + + do { + struct i915_dependency *p; + + RQ_TRACE(rq, "hold release\n"); + + GEM_BUG_ON(!i915_request_on_hold(rq)); + GEM_BUG_ON(!i915_sw_fence_signaled(&rq->submit)); + + i915_request_clear_hold(rq); + list_move_tail(&rq->sched.link, + i915_sched_lookup_priolist(rq->engine, + rq_prio(rq))); + set_bit(I915_FENCE_FLAG_PQUEUE, &rq->fence.flags); + + /* Also release any children on this engine that are ready */ + for_each_waiter(p, rq) { + struct i915_request *w = + container_of(p->waiter, typeof(*w), sched); + + /* Propagate any change in error status */ + if (rq->fence.error) + i915_request_set_error_once(w, rq->fence.error); + + if (w->engine != rq->engine) + continue; + + if (!i915_request_on_hold(w)) + continue; + + /* Check that no other parents are also on hold */ + if (hold_request(w)) + continue; + + list_move_tail(&w->sched.link, &list); + } + + rq = list_first_entry_or_null(&list, typeof(*rq), sched.link); + } while (rq); +} + +static void execlists_unhold(struct intel_engine_cs *engine, + struct i915_request *rq) +{ + spin_lock_irq(&engine->active.lock); + + /* + * Move this request back to the priority queue, and all of its + * children and grandchildren that were suspended along with it. + */ + __execlists_unhold(rq); + + if (rq_prio(rq) > engine->execlists.queue_priority_hint) { + engine->execlists.queue_priority_hint = rq_prio(rq); + tasklet_hi_schedule(&engine->execlists.tasklet); + } + + spin_unlock_irq(&engine->active.lock); +} + +struct execlists_capture { + struct work_struct work; + struct i915_request *rq; + struct i915_gpu_coredump *error; +}; + +static void execlists_capture_work(struct work_struct *work) +{ + struct execlists_capture *cap = container_of(work, typeof(*cap), work); + const gfp_t gfp = GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_NOWARN; + struct intel_engine_cs *engine = cap->rq->engine; + struct intel_gt_coredump *gt = cap->error->gt; + struct intel_engine_capture_vma *vma; + + /* Compress all the objects attached to the request, slow! */ + vma = intel_engine_coredump_add_request(gt->engine, cap->rq, gfp); + if (vma) { + struct i915_vma_compress *compress = + i915_vma_capture_prepare(gt); + + intel_engine_coredump_add_vma(gt->engine, vma, compress); + i915_vma_capture_finish(gt, compress); + } + + gt->simulated = gt->engine->simulated; + cap->error->simulated = gt->simulated; + + /* Publish the error state, and announce it to the world */ + i915_error_state_store(cap->error); + i915_gpu_coredump_put(cap->error); + + /* Return this request and all that depend upon it for signaling */ + execlists_unhold(engine, cap->rq); + i915_request_put(cap->rq); + + kfree(cap); +} + +static struct execlists_capture *capture_regs(struct intel_engine_cs *engine) +{ + const gfp_t gfp = GFP_ATOMIC | __GFP_NOWARN; + struct execlists_capture *cap; + + cap = kmalloc(sizeof(*cap), gfp); + if (!cap) + return NULL; + + cap->error = i915_gpu_coredump_alloc(engine->i915, gfp); + if (!cap->error) + goto err_cap; + + cap->error->gt = intel_gt_coredump_alloc(engine->gt, gfp); + if (!cap->error->gt) + goto err_gpu; + + cap->error->gt->engine = intel_engine_coredump_alloc(engine, gfp); + if (!cap->error->gt->engine) + goto err_gt; + + return cap; + +err_gt: + kfree(cap->error->gt); +err_gpu: + kfree(cap->error); +err_cap: + kfree(cap); + return NULL; +} + +static bool execlists_capture(struct intel_engine_cs *engine) +{ + struct execlists_capture *cap; + + if (!IS_ENABLED(CONFIG_DRM_I915_CAPTURE_ERROR)) + return true; + + /* + * We need to _quickly_ capture the engine state before we reset. + * We are inside an atomic section (softirq) here and we are delaying + * the forced preemption event. + */ + cap = capture_regs(engine); + if (!cap) + return true; + + spin_lock_irq(&engine->active.lock); + cap->rq = execlists_active(&engine->execlists); + if (cap->rq) { + cap->rq = active_request(cap->rq->context->timeline, cap->rq); + cap->rq = i915_request_get_rcu(cap->rq); + } + spin_unlock_irq(&engine->active.lock); + if (!cap->rq) + goto err_free; + + /* + * Remove the request from the execlists queue, and take ownership + * of the request. We pass it to our worker who will _slowly_ compress + * all the pages the _user_ requested for debugging their batch, after + * which we return it to the queue for signaling. + * + * By removing them from the execlists queue, we also remove the + * requests from being processed by __unwind_incomplete_requests() + * during the intel_engine_reset(), and so they will *not* be replayed + * afterwards. + * + * Note that because we have not yet reset the engine at this point, + * it is possible for the request that we have identified as being + * guilty, did in fact complete and we will then hit an arbitration + * point allowing the outstanding preemption to succeed. The likelihood + * of that is very low (as capturing of the engine registers should be + * fast enough to run inside an irq-off atomic section!), so we will + * simply hold that request accountable for being non-preemptible + * long enough to force the reset. + */ + if (!execlists_hold(engine, cap->rq)) + goto err_rq; + + INIT_WORK(&cap->work, execlists_capture_work); + schedule_work(&cap->work); + return true; + +err_rq: + i915_request_put(cap->rq); +err_free: + i915_gpu_coredump_put(cap->error); + kfree(cap); + return false; +} + +static void execlists_reset(struct intel_engine_cs *engine, const char *msg) { const unsigned int bit = I915_RESET_ENGINE + engine->id; unsigned long *lock = &engine->gt->reset.flags; - if (i915_modparams.reset < 3) + if (!intel_has_reset_engine(engine->gt)) return; if (test_and_set_bit(bit, lock)) return; + ENGINE_TRACE(engine, "reset for %s\n", msg); + /* Mark this tasklet as disabled to avoid waiting for it to complete */ tasklet_disable_nosync(&engine->execlists.tasklet); - ENGINE_TRACE(engine, "preempt timeout %lu+%ums\n", - READ_ONCE(engine->props.preempt_timeout_ms), - jiffies_to_msecs(jiffies - engine->execlists.preempt.expires)); - intel_engine_reset(engine, "preemption time out"); + ring_set_paused(engine, 1); /* Freeze the current request in place */ + if (execlists_capture(engine)) + intel_engine_reset(engine, msg); + else + ring_set_paused(engine, 0); tasklet_enable(&engine->execlists.tasklet); clear_and_wake_up_bit(bit, lock); @@ -2397,6 +2846,13 @@ static void execlists_submission_tasklet(unsigned long data) bool timeout = preempt_timeout(engine); process_csb(engine); + + if (unlikely(READ_ONCE(engine->execlists.error_interrupt))) { + engine->execlists.error_interrupt = 0; + if (ENGINE_READ(engine, RING_ESR)) /* confirm the error */ + execlists_reset(engine, "CS error"); + } + if (!READ_ONCE(engine->execlists.pending[0]) || timeout) { unsigned long flags; @@ -2405,8 +2861,8 @@ static void execlists_submission_tasklet(unsigned long data) spin_unlock_irqrestore(&engine->active.lock, flags); /* Recheck after serialising with direct-submission */ - if (timeout && preempt_timeout(engine)) - preempt_reset(engine); + if (unlikely(timeout && preempt_timeout(engine))) + execlists_reset(engine, "preemption time out"); } } @@ -2430,11 +2886,12 @@ static void execlists_preempt(struct timer_list *timer) } static void queue_request(struct intel_engine_cs *engine, - struct i915_sched_node *node, - int prio) + struct i915_request *rq) { - GEM_BUG_ON(!list_empty(&node->link)); - list_add_tail(&node->link, i915_sched_lookup_priolist(engine, prio)); + GEM_BUG_ON(!list_empty(&rq->sched.link)); + list_add_tail(&rq->sched.link, + i915_sched_lookup_priolist(engine, rq_prio(rq))); + set_bit(I915_FENCE_FLAG_PQUEUE, &rq->fence.flags); } static void __submit_queue_imm(struct intel_engine_cs *engine) @@ -2462,6 +2919,13 @@ static void submit_queue(struct intel_engine_cs *engine, __submit_queue_imm(engine); } +static bool ancestor_on_hold(const struct intel_engine_cs *engine, + const struct i915_request *rq) +{ + GEM_BUG_ON(i915_request_on_hold(rq)); + return !list_empty(&engine->active.hold) && hold_request(rq); +} + static void execlists_submit_request(struct i915_request *request) { struct intel_engine_cs *engine = request->engine; @@ -2470,12 +2934,18 @@ static void execlists_submit_request(struct i915_request *request) /* Will be called from irq-context when using foreign fences. */ spin_lock_irqsave(&engine->active.lock, flags); - queue_request(engine, &request->sched, rq_prio(request)); + if (unlikely(ancestor_on_hold(engine, request))) { + RQ_TRACE(request, "ancestor on hold\n"); + list_add_tail(&request->sched.link, &engine->active.hold); + i915_request_set_hold(request); + } else { + queue_request(engine, request); - GEM_BUG_ON(RB_EMPTY_ROOT(&engine->execlists.queue.rb_root)); - GEM_BUG_ON(list_empty(&request->sched.link)); + GEM_BUG_ON(RB_EMPTY_ROOT(&engine->execlists.queue.rb_root)); + GEM_BUG_ON(list_empty(&request->sched.link)); - submit_queue(engine, request); + submit_queue(engine, request); + } spin_unlock_irqrestore(&engine->active.lock, flags); } @@ -2531,22 +3001,23 @@ static void execlists_context_unpin(struct intel_context *ce) ce->engine); i915_gem_object_unpin_map(ce->state->obj); - intel_ring_reset(ce->ring, ce->ring->tail); } static void __execlists_update_reg_state(const struct intel_context *ce, - const struct intel_engine_cs *engine) + const struct intel_engine_cs *engine, + u32 head) { struct intel_ring *ring = ce->ring; u32 *regs = ce->lrc_reg_state; - GEM_BUG_ON(!intel_ring_offset_valid(ring, ring->head)); + GEM_BUG_ON(!intel_ring_offset_valid(ring, head)); GEM_BUG_ON(!intel_ring_offset_valid(ring, ring->tail)); regs[CTX_RING_START] = i915_ggtt_offset(ring->vma); - regs[CTX_RING_HEAD] = ring->head; + regs[CTX_RING_HEAD] = head; regs[CTX_RING_TAIL] = ring->tail; + regs[CTX_RING_CTL] = RING_CTL_SIZE(ring->size) | RING_VALID; /* RPCS */ if (engine->class == RENDER_CLASS) { @@ -2574,7 +3045,7 @@ __execlists_context_pin(struct intel_context *ce, ce->lrc_desc = lrc_descriptor(ce, engine) | CTX_DESC_FORCE_RESTORE; ce->lrc_reg_state = vaddr + LRC_STATE_PN * PAGE_SIZE; - __execlists_update_reg_state(ce, engine); + __execlists_update_reg_state(ce, engine, ce->ring->tail); return 0; } @@ -2594,28 +3065,12 @@ static void execlists_context_reset(struct intel_context *ce) CE_TRACE(ce, "reset\n"); GEM_BUG_ON(!intel_context_is_pinned(ce)); - /* - * Because we emit WA_TAIL_DWORDS there may be a disparity - * between our bookkeeping in ce->ring->head and ce->ring->tail and - * that stored in context. As we only write new commands from - * ce->ring->tail onwards, everything before that is junk. If the GPU - * starts reading from its RING_HEAD from the context, it may try to - * execute that junk and die. - * - * The contexts that are stilled pinned on resume belong to the - * kernel, and are local to each engine. All other contexts will - * have their head/tail sanitized upon pinning before use, so they - * will never see garbage, - * - * So to avoid that we reset the context images upon resume. For - * simplicity, we just zero everything out. - */ intel_ring_reset(ce->ring, ce->ring->emit); /* Scrub away the garbage */ execlists_init_reg_state(ce->lrc_reg_state, ce, ce->engine, ce->ring, true); - __execlists_update_reg_state(ce, ce->engine); + __execlists_update_reg_state(ce, ce->engine, ce->ring->tail); ce->lrc_desc |= CTX_DESC_FORCE_RESTORE; } @@ -2637,7 +3092,8 @@ static int gen8_emit_init_breadcrumb(struct i915_request *rq) { u32 *cs; - GEM_BUG_ON(!i915_request_timeline(rq)->has_initial_breadcrumb); + if (!i915_request_timeline(rq)->has_initial_breadcrumb) + return 0; cs = intel_ring_begin(rq, 6); if (IS_ERR(cs)) @@ -2930,7 +3386,7 @@ static int lrc_setup_wa_ctx(struct intel_engine_cs *engine) goto err; } - err = i915_vma_pin(vma, 0, 0, PIN_GLOBAL | PIN_HIGH); + err = i915_ggtt_pin(vma, 0, PIN_HIGH); if (err) goto err; @@ -3019,6 +3475,49 @@ static int intel_init_workaround_bb(struct intel_engine_cs *engine) return ret; } +static void enable_error_interrupt(struct intel_engine_cs *engine) +{ + u32 status; + + engine->execlists.error_interrupt = 0; + ENGINE_WRITE(engine, RING_EMR, ~0u); + ENGINE_WRITE(engine, RING_EIR, ~0u); /* clear all existing errors */ + + status = ENGINE_READ(engine, RING_ESR); + if (unlikely(status)) { + dev_err(engine->i915->drm.dev, + "engine '%s' resumed still in error: %08x\n", + engine->name, status); + __intel_gt_reset(engine->gt, engine->mask); + } + + /* + * On current gen8+, we have 2 signals to play with + * + * - I915_ERROR_INSTUCTION (bit 0) + * + * Generate an error if the command parser encounters an invalid + * instruction + * + * This is a fatal error. + * + * - CP_PRIV (bit 2) + * + * Generate an error on privilege violation (where the CP replaces + * the instruction with a no-op). This also fires for writes into + * read-only scratch pages. + * + * This is a non-fatal error, parsing continues. + * + * * there are a few others defined for odd HW that we do not use + * + * Since CP_PRIV fires for cases where we have chosen to ignore the + * error (as the HW is validating and suppressing the mistakes), we + * only unmask the instruction error bit. + */ + ENGINE_WRITE(engine, RING_EMR, ~I915_ERROR_INSTRUCTION); +} + static void enable_execlists(struct intel_engine_cs *engine) { u32 mode; @@ -3040,6 +3539,8 @@ static void enable_execlists(struct intel_engine_cs *engine) i915_ggtt_offset(engine->status_page.vma)); ENGINE_POSTING_READ(engine, RING_HWS_PGA); + enable_error_interrupt(engine); + engine->context_tag = 0; } @@ -3057,9 +3558,6 @@ static bool unexpected_starting_state(struct intel_engine_cs *engine) static int execlists_resume(struct intel_engine_cs *engine) { - intel_engine_apply_workarounds(engine); - intel_engine_apply_whitelist(engine); - intel_mocs_init_engine(engine); intel_engine_reset_breadcrumbs(engine); @@ -3170,6 +3668,7 @@ static void __execlists_reset(struct intel_engine_cs *engine, bool stalled) struct intel_engine_execlists * const execlists = &engine->execlists; struct intel_context *ce; struct i915_request *rq; + u32 head; mb(); /* paranoia: read the CSB pointers from after the reset */ clflush(execlists->csb_write); @@ -3189,23 +3688,24 @@ static void __execlists_reset(struct intel_engine_cs *engine, bool stalled) if (!rq) goto unwind; - /* We still have requests in-flight; the engine should be active */ - GEM_BUG_ON(!intel_engine_pm_is_awake(engine)); - ce = rq->context; GEM_BUG_ON(!i915_vma_is_pinned(ce->state)); if (i915_request_completed(rq)) { /* Idle context; tidy up the ring so we can restart afresh */ - ce->ring->head = intel_ring_wrap(ce->ring, rq->tail); + head = intel_ring_wrap(ce->ring, rq->tail); goto out_replay; } + /* We still have requests in-flight; the engine should be active */ + GEM_BUG_ON(!intel_engine_pm_is_awake(engine)); + /* Context has requests still in-flight; it should not be idle! */ GEM_BUG_ON(i915_active_is_idle(&ce->active)); + rq = active_request(ce->timeline, rq); - ce->ring->head = intel_ring_wrap(ce->ring, rq->head); - GEM_BUG_ON(ce->ring->head == ce->ring->tail); + head = intel_ring_wrap(ce->ring, rq->head); + GEM_BUG_ON(head == ce->ring->tail); /* * If this request hasn't started yet, e.g. it is waiting on a @@ -3250,10 +3750,9 @@ static void __execlists_reset(struct intel_engine_cs *engine, bool stalled) out_replay: ENGINE_TRACE(engine, "replay {head:%04x, tail:%04x}\n", - ce->ring->head, ce->ring->tail); - intel_ring_update_space(ce->ring); + head, ce->ring->tail); __execlists_reset_reg_state(ce, engine); - __execlists_update_reg_state(ce, engine); + __execlists_update_reg_state(ce, engine, head); ce->lrc_desc |= CTX_DESC_FORCE_RESTORE; /* paranoid: GPU was reset! */ unwind: @@ -3277,7 +3776,10 @@ static void execlists_reset_rewind(struct intel_engine_cs *engine, bool stalled) static void nop_submission_tasklet(unsigned long data) { + struct intel_engine_cs * const engine = (struct intel_engine_cs *)data; + /* The driver is wedged; don't process any more events. */ + WRITE_ONCE(engine->execlists.queue_priority_hint, INT_MIN); } static void execlists_reset_cancel(struct intel_engine_cs *engine) @@ -3325,6 +3827,10 @@ static void execlists_reset_cancel(struct intel_engine_cs *engine) i915_priolist_free(p); } + /* On-hold requests will be flushed to timeline upon their release */ + list_for_each_entry(rq, &engine->active.hold, sched.link) + mark_eio(rq); + /* Cancel all attached virtual engines */ while ((rb = rb_first_cached(&execlists->virtual))) { struct virtual_engine *ve = @@ -3669,26 +4175,6 @@ static int gen12_emit_flush_render(struct i915_request *request, *cs++ = preparser_disable(false); intel_ring_advance(request, cs); - - /* - * Wa_1604544889:tgl - */ - if (IS_TGL_REVID(request->i915, TGL_REVID_A0, TGL_REVID_A0)) { - flags = 0; - flags |= PIPE_CONTROL_CS_STALL; - flags |= PIPE_CONTROL_HDC_PIPELINE_FLUSH; - - flags |= PIPE_CONTROL_STORE_DATA_INDEX; - flags |= PIPE_CONTROL_QW_WRITE; - - cs = intel_ring_begin(request, 6); - if (IS_ERR(cs)) - return PTR_ERR(cs); - - cs = gen8_emit_pipe_control(cs, flags, - LRC_PPHWSP_SCRATCH_ADDR); - intel_ring_advance(request, cs); - } } return 0; @@ -3962,6 +4448,7 @@ logical_ring_default_irqs(struct intel_engine_cs *engine) engine->irq_enable_mask = GT_RENDER_USER_INTERRUPT << shift; engine->irq_keep_mask = GT_CONTEXT_SWITCH_INTERRUPT << shift; + engine->irq_keep_mask |= GT_CS_MASTER_ERROR_INTERRUPT << shift; } static void rcs_submission_override(struct intel_engine_cs *engine) @@ -4203,8 +4690,13 @@ populate_lr_context(struct intel_context *ce, inhibit = false; } - /* The second page of the context object contains some fields which must - * be set up prior to the first execution. */ + /* Clear the ppHWSP (inc. per-context counters) */ + memset(vaddr, 0, PAGE_SIZE); + + /* + * The second page of the context object contains some registers which + * must be set up prior to the first execution. + */ execlists_init_reg_state(vaddr + LRC_STATE_PN * PAGE_SIZE, ce, engine, ring, inhibit); @@ -4242,8 +4734,17 @@ static int __execlists_context_alloc(struct intel_context *ce, if (!ce->timeline) { struct intel_timeline *tl; + struct i915_vma *hwsp; - tl = intel_timeline_create(engine->gt, NULL); + /* + * Use the static global HWSP for the kernel context, and + * a dynamically allocated cacheline for everyone else. + */ + hwsp = NULL; + if (unlikely(intel_context_is_barrier(ce))) + hwsp = engine->status_page.vma; + + tl = intel_timeline_create(engine->gt, hwsp); if (IS_ERR(tl)) { ret = PTR_ERR(tl); goto error_deref_obj; @@ -4412,7 +4913,7 @@ static intel_engine_mask_t virtual_submission_mask(struct virtual_engine *ve) mask = rq->execution_mask; if (unlikely(!mask)) { /* Invalid selection, submit to a random engine in error */ - i915_request_skip(rq, -ENODEV); + i915_request_set_error_once(rq, -ENODEV); mask = ve->siblings[0]->mask; } @@ -4426,7 +4927,7 @@ static intel_engine_mask_t virtual_submission_mask(struct virtual_engine *ve) static void virtual_submission_tasklet(unsigned long data) { struct virtual_engine * const ve = (struct virtual_engine *)data; - const int prio = ve->base.execlists.queue_priority_hint; + const int prio = READ_ONCE(ve->base.execlists.queue_priority_hint); intel_engine_mask_t mask; unsigned int n; @@ -4822,11 +5323,15 @@ void intel_execlists_show_requests(struct intel_engine_cs *engine, show_request(m, last, "\t\tE "); } - last = NULL; - count = 0; + if (execlists->switch_priority_hint != INT_MIN) + drm_printf(m, "\t\tSwitch priority hint: %d\n", + READ_ONCE(execlists->switch_priority_hint)); if (execlists->queue_priority_hint != INT_MIN) drm_printf(m, "\t\tQueue priority hint: %d\n", - execlists->queue_priority_hint); + READ_ONCE(execlists->queue_priority_hint)); + + last = NULL; + count = 0; for (rb = rb_first_cached(&execlists->queue); rb; rb = rb_next(rb)) { struct i915_priolist *p = rb_entry(rb, typeof(*p), node); int i; @@ -4892,10 +5397,7 @@ void intel_lr_context_reset(struct intel_engine_cs *engine, restore_default_state(ce, engine); /* Rerun the request; its payload has been neutered (if guilty). */ - ce->ring->head = head; - intel_ring_update_space(ce->ring); - - __execlists_update_reg_state(ce, engine); + __execlists_update_reg_state(ce, engine, head); } bool |