diff options
Diffstat (limited to 'drivers/gpu/drm/i915/intel_lrc.c')
| -rw-r--r-- | drivers/gpu/drm/i915/intel_lrc.c | 979 |
1 files changed, 534 insertions, 445 deletions
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c index 7c4c8fb1dae4..174479232e94 100644 --- a/drivers/gpu/drm/i915/intel_lrc.c +++ b/drivers/gpu/drm/i915/intel_lrc.c @@ -137,6 +137,7 @@ #include <drm/i915_drm.h> #include "i915_drv.h" #include "i915_gem_render_state.h" +#include "i915_vgpu.h" #include "intel_lrc_reg.h" #include "intel_mocs.h" #include "intel_workarounds.h" @@ -164,7 +165,8 @@ #define WA_TAIL_BYTES (sizeof(u32) * WA_TAIL_DWORDS) static int execlists_context_deferred_alloc(struct i915_gem_context *ctx, - struct intel_engine_cs *engine); + struct intel_engine_cs *engine, + struct intel_context *ce); static void execlists_init_reg_state(u32 *reg_state, struct i915_gem_context *ctx, struct intel_engine_cs *engine, @@ -189,12 +191,7 @@ static inline bool need_preempt(const struct intel_engine_cs *engine, !i915_request_completed(last)); } -/** - * intel_lr_context_descriptor_update() - calculate & cache the descriptor - * descriptor for a pinned context - * @ctx: Context to work on - * @engine: Engine the descriptor will be used with - * +/* * The context descriptor encodes various attributes of a context, * including its GTT address and some flags. Because it's fairly * expensive to calculate, we'll just do it once and cache the result, @@ -204,7 +201,7 @@ static inline bool need_preempt(const struct intel_engine_cs *engine, * * bits 0-11: flags, GEN8_CTX_* (cached in ctx->desc_template) * bits 12-31: LRCA, GTT address of (the HWSP of) this context - * bits 32-52: ctx ID, a globally unique tag + * bits 32-52: ctx ID, a globally unique tag (highest bit used by GuC) * bits 53-54: mbz, reserved for use by hardware * bits 55-63: group ID, currently unused and set to 0 * @@ -222,9 +219,9 @@ static inline bool need_preempt(const struct intel_engine_cs *engine, */ static void intel_lr_context_descriptor_update(struct i915_gem_context *ctx, - struct intel_engine_cs *engine) + struct intel_engine_cs *engine, + struct intel_context *ce) { - struct intel_context *ce = to_intel_context(ctx, engine); u64 desc; BUILD_BUG_ON(MAX_CONTEXT_HW_ID > (BIT(GEN8_CTX_ID_WIDTH))); @@ -237,6 +234,11 @@ intel_lr_context_descriptor_update(struct i915_gem_context *ctx, /* bits 12-31 */ GEM_BUG_ON(desc & GENMASK_ULL(63, 32)); + /* + * The following 32bits are copied into the OA reports (dword 2). + * Consider updating oa_get_render_ctx_id in i915_perf.c when changing + * anything below. + */ if (INTEL_GEN(ctx->i915) >= 11) { GEM_BUG_ON(ctx->hw_id >= BIT(GEN11_SW_CTX_ID_WIDTH)); desc |= (u64)ctx->hw_id << GEN11_SW_CTX_ID_SHIFT; @@ -271,7 +273,7 @@ lookup_priolist(struct intel_engine_cs *engine, int prio) find_priolist: /* most positive priority is scheduled first, equal priorities fifo */ rb = NULL; - parent = &execlists->queue.rb_node; + parent = &execlists->queue.rb_root.rb_node; while (*parent) { rb = *parent; p = to_priolist(rb); @@ -309,10 +311,7 @@ find_priolist: p->priority = prio; INIT_LIST_HEAD(&p->requests); rb_link_node(&p->node, rb, parent); - rb_insert_color(&p->node, &execlists->queue); - - if (first) - execlists->first = &p->node; + rb_insert_color_cached(&p->node, &execlists->queue, first); return p; } @@ -418,9 +417,9 @@ execlists_update_context_pdps(struct i915_hw_ppgtt *ppgtt, u32 *reg_state) static u64 execlists_update_context(struct i915_request *rq) { - struct intel_context *ce = to_intel_context(rq->ctx, rq->engine); + struct intel_context *ce = rq->hw_context; struct i915_hw_ppgtt *ppgtt = - rq->ctx->ppgtt ?: rq->i915->mm.aliasing_ppgtt; + rq->gem_context->ppgtt ?: rq->i915->mm.aliasing_ppgtt; u32 *reg_state = ce->lrc_reg_state; reg_state[CTX_RING_TAIL+1] = intel_ring_set_tail(rq->ring, rq->tail); @@ -430,7 +429,7 @@ static u64 execlists_update_context(struct i915_request *rq) * PML4 is allocated during ppgtt init, so this is not needed * in 48-bit mode. */ - if (ppgtt && !i915_vm_is_48bit(&ppgtt->base)) + if (ppgtt && !i915_vm_is_48bit(&ppgtt->vm)) execlists_update_context_pdps(ppgtt, reg_state); return ce->lrc_desc; @@ -454,6 +453,16 @@ static void execlists_submit_ports(struct intel_engine_cs *engine) unsigned int n; /* + * We can skip acquiring intel_runtime_pm_get() here as it was taken + * on our behalf by the request (see i915_gem_mark_busy()) and it will + * not be relinquished until the device is idle (see + * i915_gem_idle_work_handler()). As a precaution, we make sure + * that all ELSP are drained i.e. we have processed the CSB, + * before allowing ourselves to idle and calling intel_runtime_pm_put(). + */ + GEM_BUG_ON(!engine->i915->gt.awake); + + /* * ELSQ note: the submit queue is not cleared after being submitted * to the HW so we need to make sure we always clean it up. This is * currently ensured by the fact that we always write the same number @@ -495,14 +504,14 @@ static void execlists_submit_ports(struct intel_engine_cs *engine) execlists_clear_active(execlists, EXECLISTS_ACTIVE_HWACK); } -static bool ctx_single_port_submission(const struct i915_gem_context *ctx) +static bool ctx_single_port_submission(const struct intel_context *ce) { return (IS_ENABLED(CONFIG_DRM_I915_GVT) && - i915_gem_context_force_single_submission(ctx)); + i915_gem_context_force_single_submission(ce->gem_context)); } -static bool can_merge_ctx(const struct i915_gem_context *prev, - const struct i915_gem_context *next) +static bool can_merge_ctx(const struct intel_context *prev, + const struct intel_context *next) { if (prev != next) return false; @@ -552,11 +561,24 @@ static void inject_preempt_context(struct intel_engine_cs *engine) if (execlists->ctrl_reg) writel(EL_CTRL_LOAD, execlists->ctrl_reg); - execlists_clear_active(&engine->execlists, EXECLISTS_ACTIVE_HWACK); - execlists_set_active(&engine->execlists, EXECLISTS_ACTIVE_PREEMPT); + execlists_clear_active(execlists, EXECLISTS_ACTIVE_HWACK); + execlists_set_active(execlists, EXECLISTS_ACTIVE_PREEMPT); +} + +static void complete_preempt_context(struct intel_engine_execlists *execlists) +{ + GEM_BUG_ON(!execlists_is_active(execlists, EXECLISTS_ACTIVE_PREEMPT)); + + if (inject_preempt_hang(execlists)) + return; + + execlists_cancel_port_requests(execlists); + __unwind_incomplete_requests(container_of(execlists, + struct intel_engine_cs, + execlists)); } -static bool __execlists_dequeue(struct intel_engine_cs *engine) +static void execlists_dequeue(struct intel_engine_cs *engine) { struct intel_engine_execlists * const execlists = &engine->execlists; struct execlist_port *port = execlists->port; @@ -566,9 +588,8 @@ static bool __execlists_dequeue(struct intel_engine_cs *engine) struct rb_node *rb; bool submit = false; - lockdep_assert_held(&engine->timeline.lock); - - /* Hardware submission is through 2 ports. Conceptually each port + /* + * Hardware submission is through 2 ports. Conceptually each port * has a (RING_START, RING_HEAD, RING_TAIL) tuple. RING_START is * static for a context, and unique to each, so we only execute * requests belonging to a single context from each ring. RING_HEAD @@ -589,9 +610,6 @@ static bool __execlists_dequeue(struct intel_engine_cs *engine) * and context switches) submission. */ - rb = execlists->first; - GEM_BUG_ON(rb_first(&execlists->queue) != rb); - if (last) { /* * Don't resubmit or switch until all outstanding @@ -602,8 +620,6 @@ static bool __execlists_dequeue(struct intel_engine_cs *engine) GEM_BUG_ON(!execlists_is_active(execlists, EXECLISTS_ACTIVE_USER)); GEM_BUG_ON(!port_count(&port[0])); - if (port_count(&port[0]) > 1) - return false; /* * If we write to ELSP a second time before the HW has had @@ -613,11 +629,11 @@ static bool __execlists_dequeue(struct intel_engine_cs *engine) * the HW to indicate that it has had a chance to respond. */ if (!execlists_is_active(execlists, EXECLISTS_ACTIVE_HWACK)) - return false; + return; if (need_preempt(engine, last, execlists->queue_priority)) { inject_preempt_context(engine); - return false; + return; } /* @@ -642,7 +658,7 @@ static bool __execlists_dequeue(struct intel_engine_cs *engine) * priorities of the ports haven't been switch. */ if (port_count(&port[1])) - return false; + return; /* * WaIdleLiteRestore:bdw,skl @@ -655,7 +671,7 @@ static bool __execlists_dequeue(struct intel_engine_cs *engine) last->tail = last->wa_tail; } - while (rb) { + while ((rb = rb_first_cached(&execlists->queue))) { struct i915_priolist *p = to_priolist(rb); struct i915_request *rq, *rn; @@ -671,7 +687,8 @@ static bool __execlists_dequeue(struct intel_engine_cs *engine) * second request, and so we never need to tell the * hardware about the first. */ - if (last && !can_merge_ctx(rq->ctx, last->ctx)) { + if (last && + !can_merge_ctx(rq->hw_context, last->hw_context)) { /* * If we are on the second port and cannot * combine this request with the last, then we @@ -690,14 +707,14 @@ static bool __execlists_dequeue(struct intel_engine_cs *engine) * the same context (even though a different * request) to the second port. */ - if (ctx_single_port_submission(last->ctx) || - ctx_single_port_submission(rq->ctx)) { + if (ctx_single_port_submission(last->hw_context) || + ctx_single_port_submission(rq->hw_context)) { __list_del_many(&p->requests, &rq->sched.link); goto done; } - GEM_BUG_ON(last->ctx == rq->ctx); + GEM_BUG_ON(last->hw_context == rq->hw_context); if (submit) port_assign(port, last); @@ -713,8 +730,7 @@ static bool __execlists_dequeue(struct intel_engine_cs *engine) submit = true; } - rb = rb_next(rb); - rb_erase(&p->node, &execlists->queue); + rb_erase_cached(&p->node, &execlists->queue); INIT_LIST_HEAD(&p->requests); if (p->priority != I915_PRIORITY_NORMAL) kmem_cache_free(engine->i915->priorities, p); @@ -740,35 +756,23 @@ done: execlists->queue_priority = port != execlists->port ? rq_prio(last) : INT_MIN; - execlists->first = rb; - if (submit) + if (submit) { port_assign(port, last); + execlists_submit_ports(engine); + } /* We must always keep the beast fed if we have work piled up */ - GEM_BUG_ON(execlists->first && !port_isset(execlists->port)); + GEM_BUG_ON(rb_first_cached(&execlists->queue) && + !port_isset(execlists->port)); /* Re-evaluate the executing context setup after each preemptive kick */ if (last) execlists_user_begin(execlists, execlists->port); - return submit; -} - -static void execlists_dequeue(struct intel_engine_cs *engine) -{ - struct intel_engine_execlists * const execlists = &engine->execlists; - unsigned long flags; - bool submit; - - spin_lock_irqsave(&engine->timeline.lock, flags); - submit = __execlists_dequeue(engine); - spin_unlock_irqrestore(&engine->timeline.lock, flags); - - if (submit) - execlists_submit_ports(engine); - - GEM_BUG_ON(port_isset(execlists->port) && - !execlists_is_active(execlists, EXECLISTS_ACTIVE_USER)); + /* If the engine is now idle, so should be the flag; and vice versa. */ + GEM_BUG_ON(execlists_is_active(&engine->execlists, + EXECLISTS_ACTIVE_USER) == + !port_isset(engine->execlists.port)); } void @@ -799,82 +803,27 @@ execlists_cancel_port_requests(struct intel_engine_execlists * const execlists) port++; } - execlists_user_end(execlists); + execlists_clear_all_active(execlists); } -static void clear_gtiir(struct intel_engine_cs *engine) +static void reset_csb_pointers(struct intel_engine_execlists *execlists) { - struct drm_i915_private *dev_priv = engine->i915; - int i; - /* - * Clear any pending interrupt state. - * - * We do it twice out of paranoia that some of the IIR are - * double buffered, and so if we only reset it once there may - * still be an interrupt pending. + * After a reset, the HW starts writing into CSB entry [0]. We + * therefore have to set our HEAD pointer back one entry so that + * the *first* entry we check is entry 0. To complicate this further, + * as we don't wait for the first interrupt after reset, we have to + * fake the HW write to point back to the last entry so that our + * inline comparison of our cached head position against the last HW + * write works even before the first interrupt. */ - if (INTEL_GEN(dev_priv) >= 11) { - static const struct { - u8 bank; - u8 bit; - } gen11_gtiir[] = { - [RCS] = {0, GEN11_RCS0}, - [BCS] = {0, GEN11_BCS}, - [_VCS(0)] = {1, GEN11_VCS(0)}, - [_VCS(1)] = {1, GEN11_VCS(1)}, - [_VCS(2)] = {1, GEN11_VCS(2)}, - [_VCS(3)] = {1, GEN11_VCS(3)}, - [_VECS(0)] = {1, GEN11_VECS(0)}, - [_VECS(1)] = {1, GEN11_VECS(1)}, - }; - unsigned long irqflags; - - GEM_BUG_ON(engine->id >= ARRAY_SIZE(gen11_gtiir)); - - spin_lock_irqsave(&dev_priv->irq_lock, irqflags); - for (i = 0; i < 2; i++) { - gen11_reset_one_iir(dev_priv, - gen11_gtiir[engine->id].bank, - gen11_gtiir[engine->id].bit); - } - spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags); - } else { - static const u8 gtiir[] = { - [RCS] = 0, - [BCS] = 0, - [VCS] = 1, - [VCS2] = 1, - [VECS] = 3, - }; - - GEM_BUG_ON(engine->id >= ARRAY_SIZE(gtiir)); - - for (i = 0; i < 2; i++) { - I915_WRITE(GEN8_GT_IIR(gtiir[engine->id]), - engine->irq_keep_mask); - POSTING_READ(GEN8_GT_IIR(gtiir[engine->id])); - } - GEM_BUG_ON(I915_READ(GEN8_GT_IIR(gtiir[engine->id])) & - engine->irq_keep_mask); - } + execlists->csb_head = execlists->csb_write_reset; + WRITE_ONCE(*execlists->csb_write, execlists->csb_write_reset); } -static void reset_irq(struct intel_engine_cs *engine) +static void nop_submission_tasklet(unsigned long data) { - /* Mark all CS interrupts as complete */ - smp_store_mb(engine->execlists.active, 0); - synchronize_hardirq(engine->i915->drm.irq); - - clear_gtiir(engine); - - /* - * The port is checked prior to scheduling a tasklet, but - * just in case we have suspended the tasklet to do the - * wedging make sure that when it wakes, it decides there - * is no work to do by clearing the irq_posted bit. - */ - clear_bit(ENGINE_IRQ_EXECLIST, &engine->irq_posted); + /* The driver is wedged; don't process any more events. */ } static void execlists_cancel_requests(struct intel_engine_cs *engine) @@ -901,13 +850,11 @@ static void execlists_cancel_requests(struct intel_engine_cs *engine) * submission's irq state, we also wish to remind ourselves that * it is irq state.) */ - local_irq_save(flags); + spin_lock_irqsave(&engine->timeline.lock, flags); /* Cancel the requests on the HW and clear the ELSP tracker. */ execlists_cancel_port_requests(execlists); - reset_irq(engine); - - spin_lock(&engine->timeline.lock); + execlists_user_end(execlists); /* Mark all executing requests as skipped. */ list_for_each_entry(rq, &engine->timeline.requests, link) { @@ -917,8 +864,7 @@ static void execlists_cancel_requests(struct intel_engine_cs *engine) } /* Flush the queued requests to the timeline list (for retiring). */ - rb = execlists->first; - while (rb) { + while ((rb = rb_first_cached(&execlists->queue))) { struct i915_priolist *p = to_priolist(rb); list_for_each_entry_safe(rq, rn, &p->requests, sched.link) { @@ -928,8 +874,7 @@ static void execlists_cancel_requests(struct intel_engine_cs *engine) __i915_request_submit(rq); } - rb = rb_next(rb); - rb_erase(&p->node, &execlists->queue); + rb_erase_cached(&p->node, &execlists->queue); INIT_LIST_HEAD(&p->requests); if (p->priority != I915_PRIORITY_NORMAL) kmem_cache_free(engine->i915->priorities, p); @@ -938,221 +883,198 @@ static void execlists_cancel_requests(struct intel_engine_cs *engine) /* Remaining _unready_ requests will be nop'ed when submitted */ execlists->queue_priority = INT_MIN; - execlists->queue = RB_ROOT; - execlists->first = NULL; + execlists->queue = RB_ROOT_CACHED; GEM_BUG_ON(port_isset(execlists->port)); - spin_unlock(&engine->timeline.lock); + GEM_BUG_ON(__tasklet_is_enabled(&execlists->tasklet)); + execlists->tasklet.func = nop_submission_tasklet; - local_irq_restore(flags); + spin_unlock_irqrestore(&engine->timeline.lock, flags); } -/* - * Check the unread Context Status Buffers and manage the submission of new - * contexts to the ELSP accordingly. - */ -static void execlists_submission_tasklet(unsigned long data) +static inline bool +reset_in_progress(const struct intel_engine_execlists *execlists) +{ + return unlikely(!__tasklet_is_enabled(&execlists->tasklet)); +} + +static void process_csb(struct intel_engine_cs *engine) { - struct intel_engine_cs * const engine = (struct intel_engine_cs *)data; struct intel_engine_execlists * const execlists = &engine->execlists; struct execlist_port *port = execlists->port; - struct drm_i915_private *dev_priv = engine->i915; - bool fw = false; + const u32 * const buf = execlists->csb_status; + u8 head, tail; /* - * We can skip acquiring intel_runtime_pm_get() here as it was taken - * on our behalf by the request (see i915_gem_mark_busy()) and it will - * not be relinquished until the device is idle (see - * i915_gem_idle_work_handler()). As a precaution, we make sure - * that all ELSP are drained i.e. we have processed the CSB, - * before allowing ourselves to idle and calling intel_runtime_pm_put(). + * Note that csb_write, csb_status may be either in HWSP or mmio. + * When reading from the csb_write mmio register, we have to be + * careful to only use the GEN8_CSB_WRITE_PTR portion, which is + * the low 4bits. As it happens we know the next 4bits are always + * zero and so we can simply masked off the low u8 of the register + * and treat it identically to reading from the HWSP (without having + * to use explicit shifting and masking, and probably bifurcating + * the code to handle the legacy mmio read). */ - GEM_BUG_ON(!dev_priv->gt.awake); + head = execlists->csb_head; + tail = READ_ONCE(*execlists->csb_write); + GEM_TRACE("%s cs-irq head=%d, tail=%d\n", engine->name, head, tail); + if (unlikely(head == tail)) + return; /* - * Prefer doing test_and_clear_bit() as a two stage operation to avoid - * imposing the cost of a locked atomic transaction when submitting a - * new request (outside of the context-switch interrupt). + * Hopefully paired with a wmb() in HW! + * + * We must complete the read of the write pointer before any reads + * from the CSB, so that we do not see stale values. Without an rmb + * (lfence) the HW may speculatively perform the CSB[] reads *before* + * we perform the READ_ONCE(*csb_write). */ - while (test_bit(ENGINE_IRQ_EXECLIST, &engine->irq_posted)) { - /* The HWSP contains a (cacheable) mirror of the CSB */ - const u32 *buf = - &engine->status_page.page_addr[I915_HWS_CSB_BUF0_INDEX]; - unsigned int head, tail; + rmb(); - if (unlikely(execlists->csb_use_mmio)) { - buf = (u32 * __force) - (dev_priv->regs + i915_mmio_reg_offset(RING_CONTEXT_STATUS_BUF_LO(engine, 0))); - execlists->csb_head = -1; /* force mmio read of CSB ptrs */ - } + do { + struct i915_request *rq; + unsigned int status; + unsigned int count; - /* Clear before reading to catch new interrupts */ - clear_bit(ENGINE_IRQ_EXECLIST, &engine->irq_posted); - smp_mb__after_atomic(); + if (++head == GEN8_CSB_ENTRIES) + head = 0; - if (unlikely(execlists->csb_head == -1)) { /* following a reset */ - if (!fw) { - intel_uncore_forcewake_get(dev_priv, - execlists->fw_domains); - fw = true; - } + /* + * We are flying near dragons again. + * + * We hold a reference to the request in execlist_port[] + * but no more than that. We are operating in softirq + * context and so cannot hold any mutex or sleep. That + * prevents us stopping the requests we are processing + * in port[] from being retired simultaneously (the + * breadcrumb will be complete before we see the + * context-switch). As we only hold the reference to the + * request, any pointer chasing underneath the request + * is subject to a potential use-after-free. Thus we + * store all of the bookkeeping within port[] as + * required, and avoid using unguarded pointers beneath + * request itself. The same applies to the atomic + * status notifier. + */ - head = readl(dev_priv->regs + i915_mmio_reg_offset(RING_CONTEXT_STATUS_PTR(engine))); - tail = GEN8_CSB_WRITE_PTR(head); - head = GEN8_CSB_READ_PTR(head); - execlists->csb_head = head; - } else { - const int write_idx = - intel_hws_csb_write_index(dev_priv) - - I915_HWS_CSB_BUF0_INDEX; + GEM_TRACE("%s csb[%d]: status=0x%08x:0x%08x, active=0x%x\n", + engine->name, head, + buf[2 * head + 0], buf[2 * head + 1], + execlists->active); + + status = buf[2 * head]; + if (status & (GEN8_CTX_STATUS_IDLE_ACTIVE | + GEN8_CTX_STATUS_PREEMPTED)) + execlists_set_active(execlists, + EXECLISTS_ACTIVE_HWACK); + if (status & GEN8_CTX_STATUS_ACTIVE_IDLE) + execlists_clear_active(execlists, + EXECLISTS_ACTIVE_HWACK); + + if (!(status & GEN8_CTX_STATUS_COMPLETED_MASK)) + continue; + + /* We should never get a COMPLETED | IDLE_ACTIVE! */ + GEM_BUG_ON(status & GEN8_CTX_STATUS_IDLE_ACTIVE); - head = execlists->csb_head; - tail = READ_ONCE(buf[write_idx]); - rmb(); /* Hopefully paired with a wmb() in HW */ + if (status & GEN8_CTX_STATUS_COMPLETE && + buf[2*head + 1] == execlists->preempt_complete_status) { + GEM_TRACE("%s preempt-idle\n", engine->name); + complete_preempt_context(execlists); + continue; } - GEM_TRACE("%s cs-irq head=%d [%d%s], tail=%d [%d%s]\n", - engine->name, - head, GEN8_CSB_READ_PTR(readl(dev_priv->regs + i915_mmio_reg_offset(RING_CONTEXT_STATUS_PTR(engine)))), fw ? "" : "?", - tail, GEN8_CSB_WRITE_PTR(readl(dev_priv->regs + i915_mmio_reg_offset(RING_CONTEXT_STATUS_PTR(engine)))), fw ? "" : "?"); - while (head != tail) { - struct i915_request *rq; - unsigned int status; - unsigned int count; + if (status & GEN8_CTX_STATUS_PREEMPTED && + execlists_is_active(execlists, + EXECLISTS_ACTIVE_PREEMPT)) + continue; - if (++head == GEN8_CSB_ENTRIES) - head = 0; + GEM_BUG_ON(!execlists_is_active(execlists, + EXECLISTS_ACTIVE_USER)); - /* We are flying near dragons again. - * - * We hold a reference to the request in execlist_port[] - * but no more than that. We are operating in softirq - * context and so cannot hold any mutex or sleep. That - * prevents us stopping the requests we are processing - * in port[] from being retired simultaneously (the - * breadcrumb will be complete before we see the - * context-switch). As we only hold the reference to the - * request, any pointer chasing underneath the request - * is subject to a potential use-after-free. Thus we - * store all of the bookkeeping within port[] as - * required, and avoid using unguarded pointers beneath - * request itself. The same applies to the atomic - * status notifier. + rq = port_unpack(port, &count); + GEM_TRACE("%s out[0]: ctx=%d.%d, global=%d (fence %llx:%d) (current %d), prio=%d\n", + engine->name, + port->context_id, count, + rq ? rq->global_seqno : 0, + rq ? rq->fence.context : 0, + rq ? rq->fence.seqno : 0, + intel_engine_get_seqno(engine), + rq ? rq_prio(rq) : 0); + + /* Check the context/desc id for this event matches */ + GEM_DEBUG_BUG_ON(buf[2 * head + 1] != port->context_id); + + GEM_BUG_ON(count == 0); + if (--count == 0) { + /* + * On the final event corresponding to the + * submission of this context, we expect either + * an element-switch event or a completion + * event (and on completion, the active-idle + * marker). No more preemptions, lite-restore + * or otherwise. */ + GEM_BUG_ON(status & GEN8_CTX_STATUS_PREEMPTED); + GEM_BUG_ON(port_isset(&port[1]) && + !(status & GEN8_CTX_STATUS_ELEMENT_SWITCH)); + GEM_BUG_ON(!port_isset(&port[1]) && + !(status & GEN8_CTX_STATUS_ACTIVE_IDLE)); - status = READ_ONCE(buf[2 * head]); /* maybe mmio! */ - GEM_TRACE("%s csb[%d]: status=0x%08x:0x%08x, active=0x%x\n", - engine->name, head, - status, buf[2*head + 1], - execlists->active); - - if (status & (GEN8_CTX_STATUS_IDLE_ACTIVE | - GEN8_CTX_STATUS_PREEMPTED)) - execlists_set_active(execlists, - EXECLISTS_ACTIVE_HWACK); - if (status & GEN8_CTX_STATUS_ACTIVE_IDLE) - execlists_clear_active(execlists, - EXECLISTS_ACTIVE_HWACK); - - if (!(status & GEN8_CTX_STATUS_COMPLETED_MASK)) - continue; - - /* We should never get a COMPLETED | IDLE_ACTIVE! */ - GEM_BUG_ON(status & GEN8_CTX_STATUS_IDLE_ACTIVE); - - if (status & GEN8_CTX_STATUS_COMPLETE && - buf[2*head + 1] == execlists->preempt_complete_status) { - GEM_TRACE("%s preempt-idle\n", engine->name); - - execlists_cancel_port_requests(execlists); - execlists_unwind_incomplete_requests(execlists); - - GEM_BUG_ON(!execlists_is_active(execlists, - EXECLISTS_ACTIVE_PREEMPT)); - execlists_clear_active(execlists, - EXECLISTS_ACTIVE_PREEMPT); - continue; - } - - if (status & GEN8_CTX_STATUS_PREEMPTED && - execlists_is_active(execlists, - EXECLISTS_ACTIVE_PREEMPT)) - continue; + /* + * We rely on the hardware being strongly + * ordered, that the breadcrumb write is + * coherent (visible from the CPU) before the + * user interrupt and CSB is processed. + */ + GEM_BUG_ON(!i915_request_completed(rq)); - GEM_BUG_ON(!execlists_is_active(execlists, - EXECLISTS_ACTIVE_USER)); + execlists_context_schedule_out(rq, + INTEL_CONTEXT_SCHEDULE_OUT); + i915_request_put(rq); - rq = port_unpack(port, &count); - GEM_TRACE("%s out[0]: ctx=%d.%d, global=%d (fence %llx:%d) (current %d), prio=%d\n", - engine->name, - port->context_id, count, - rq ? rq->global_seqno : 0, - rq ? rq->fence.context : 0, - rq ? rq->fence.seqno : 0, - intel_engine_get_seqno(engine), - rq ? rq_prio(rq) : 0); + GEM_TRACE("%s completed ctx=%d\n", + engine->name, port->context_id); - /* Check the context/desc id for this event matches */ - GEM_DEBUG_BUG_ON(buf[2 * head + 1] != port->context_id); - - GEM_BUG_ON(count == 0); - if (--count == 0) { - /* - * On the final event corresponding to the - * submission of this context, we expect either - * an element-switch event or a completion - * event (and on completion, the active-idle - * marker). No more preemptions, lite-restore - * or otherwise. - */ - GEM_BUG_ON(status & GEN8_CTX_STATUS_PREEMPTED); - GEM_BUG_ON(port_isset(&port[1]) && - !(status & GEN8_CTX_STATUS_ELEMENT_SWITCH)); - GEM_BUG_ON(!port_isset(&port[1]) && - !(status & GEN8_CTX_STATUS_ACTIVE_IDLE)); - - /* - * We rely on the hardware being strongly - * ordered, that the breadcrumb write is - * coherent (visible from the CPU) before the - * user interrupt and CSB is processed. - */ - GEM_BUG_ON(!i915_request_completed(rq)); - - execlists_context_schedule_out(rq, - INTEL_CONTEXT_SCHEDULE_OUT); - i915_request_put(rq); - - GEM_TRACE("%s completed ctx=%d\n", - engine->name, port->context_id); - - port = execlists_port_complete(execlists, port); - if (port_isset(port)) - execlists_user_begin(execlists, port); - else - execlists_user_end(execlists); - } else { - port_set(port, port_pack(rq, count)); - } + port = execlists_port_complete(execlists, port); + if (port_isset(port)) + execlists_user_begin(execlists, port); + else + execlists_user_end(execlists); + } else { + port_set(port, port_pack(rq, count)); } + } while (head != tail); - if (head != execlists->csb_head) { - execlists->csb_head = head; - writel(_MASKED_FIELD(GEN8_CSB_READ_PTR_MASK, head << 8), - dev_priv->regs + i915_mmio_reg_offset(RING_CONTEXT_STATUS_PTR(engine))); - } - } + execlists->csb_head = head; +} - if (!execlists_is_active(execlists, EXECLISTS_ACTIVE_PREEMPT)) +static void __execlists_submission_tasklet(struct intel_engine_cs *const engine) +{ + lockdep_assert_held(&engine->timeline.lock); + + process_csb(engine); + if (!execlists_is_active(&engine->execlists, EXECLISTS_ACTIVE_PREEMPT)) execlists_dequeue(engine); +} + +/* + * Check the unread Context Status Buffers and manage the submission of new + * contexts to the ELSP accordingly. + */ +static void execlists_submission_tasklet(unsigned long data) +{ + struct intel_engine_cs * const engine = (struct intel_engine_cs *)data; + unsigned long flags; - if (fw) - intel_uncore_forcewake_put(dev_priv, execlists->fw_domains); + GEM_TRACE("%s awake?=%d, active=%x\n", + engine->name, + engine->i915->gt.awake, + engine->execlists.active); - /* If the engine is now idle, so should be the flag; and vice versa. */ - GEM_BUG_ON(execlists_is_active(&engine->execlists, - EXECLISTS_ACTIVE_USER) == - !port_isset(engine->execlists.port)); + spin_lock_irqsave(&engine->timeline.lock, flags); + __execlists_submission_tasklet(engine); + spin_unlock_irqrestore(&engine->timeline.lock, flags); } static void queue_request(struct intel_engine_cs *engine, @@ -1163,16 +1085,30 @@ static void queue_request(struct intel_engine_cs *engine, &lookup_priolist(engine, prio)->requests); } -static void __submit_queue(struct intel_engine_cs *engine, int prio) +static void __update_queue(struct intel_engine_cs *engine, int prio) { engine->execlists.queue_priority = prio; - tasklet_hi_schedule(&engine->execlists.tasklet); +} + +static void __submit_queue_imm(struct intel_engine_cs *engine) +{ + struct intel_engine_execlists * const execlists = &engine->execlists; + + if (reset_in_progress(execlists)) + return; /* defer until we restart the engine following reset */ + + if (execlists->tasklet.func == execlists_submission_tasklet) + __execlists_submission_tasklet(engine); + else + tasklet_hi_schedule(&execlists->tasklet); } static void submit_queue(struct intel_engine_cs *engine, int prio) { - if (prio > engine->execlists.queue_priority) - __submit_queue(engine, prio); + if (prio > engine->execlists.queue_priority) { + __update_queue(engine, prio); + __submit_queue_imm(engine); + } } static void execlists_submit_request(struct i915_request *request) @@ -1184,11 +1120,12 @@ static void execlists_submit_request(struct i915_request *request) spin_lock_irqsave(&engine->timeline.lock, flags); queue_request(engine, &request->sched, rq_prio(request)); - submit_queue(engine, rq_prio(request)); - GEM_BUG_ON(!engine->execlists.first); + GEM_BUG_ON(RB_EMPTY_ROOT(&engine->execlists.queue.rb_root)); GEM_BUG_ON(list_empty(&request->sched.link)); + submit_queue(engine, rq_prio(request)); + spin_unlock_irqrestore(&engine->timeline.lock, flags); } @@ -1315,13 +1252,40 @@ static void execlists_schedule(struct i915_request *request, } if (prio > engine->execlists.queue_priority && - i915_sw_fence_done(&sched_to_request(node)->submit)) - __submit_queue(engine, prio); + i915_sw_fence_done(&sched_to_request(node)->submit)) { + /* defer submission until after all of our updates */ + __update_queue(engine, prio); + tasklet_hi_schedule(&engine->execlists.tasklet); + } } spin_unlock_irq(&engine->timeline.lock); } +static void execlists_context_destroy(struct intel_context *ce) +{ + GEM_BUG_ON(ce->pin_count); + + if (!ce->state) + return; + + intel_ring_free(ce->ring); + + GEM_BUG_ON(i915_gem_object_is_active(ce->state->obj)); + i915_gem_object_put(ce->state->obj); +} + +static void execlists_context_unpin(struct intel_context *ce) +{ + intel_ring_unpin(ce->ring); + + ce->state->obj->pin_global--; + i915_gem_object_unpin_map(ce->state->obj); + i915_vma_unpin(ce->state); + + i915_gem_context_put(ce->gem_context); +} + static int __context_pin(struct i915_gem_context *ctx, struct i915_vma *vma) { unsigned int flags; @@ -1345,21 +1309,15 @@ static int __context_pin(struct i915_gem_context *ctx, struct i915_vma *vma) return i915_vma_pin(vma, 0, GEN8_LR_CONTEXT_ALIGN, flags); } -static struct intel_ring * -execlists_context_pin(struct intel_engine_cs *engine, - struct i915_gem_context *ctx) +static struct intel_context * +__execlists_context_pin(struct intel_engine_cs *engine, + struct i915_gem_context *ctx, + struct intel_context *ce) { - struct intel_context *ce = to_intel_context(ctx, engine); void *vaddr; int ret; - lockdep_assert_held(&ctx->i915->drm.struct_mutex); - - if (likely(ce->pin_count++)) - goto out; - GEM_BUG_ON(!ce->pin_count); /* no overflow please! */ - - ret = execlists_context_deferred_alloc(ctx, engine); + ret = execlists_context_deferred_alloc(ctx, engine, ce); if (ret) goto err; GEM_BUG_ON(!ce->state); @@ -1378,17 +1336,17 @@ execlists_context_pin(struct intel_engine_cs *engine, if (ret) goto unpin_map; - intel_lr_context_descriptor_update(ctx, engine); + intel_lr_context_descriptor_update(ctx, engine, ce); ce->lrc_reg_state = vaddr + LRC_STATE_PN * PAGE_SIZE; ce->lrc_reg_state[CTX_RING_BUFFER_START+1] = i915_ggtt_offset(ce->ring->vma); + GEM_BUG_ON(!intel_ring_offset_valid(ce->ring, ce->ring->head)); ce->lrc_reg_state[CTX_RING_HEAD+1] = ce->ring->head; ce->state->obj->pin_global++; i915_gem_context_get(ctx); -out: - return ce->ring; + return ce; unpin_map: i915_gem_object_unpin_map(ce->state->obj); @@ -1399,33 +1357,33 @@ err: return ERR_PTR(ret); } -static void execlists_context_unpin(struct intel_engine_cs *engine, - struct i915_gem_context *ctx) +static const struct intel_context_ops execlists_context_ops = { + .unpin = execlists_context_unpin, + .destroy = execlists_context_destroy, +}; + +static struct intel_context * +execlists_context_pin(struct intel_engine_cs *engine, + struct i915_gem_context *ctx) { struct intel_context *ce = to_intel_context(ctx, engine); lockdep_assert_held(&ctx->i915->drm.struct_mutex); - GEM_BUG_ON(ce->pin_count == 0); - if (--ce->pin_count) - return; - - intel_ring_unpin(ce->ring); + if (likely(ce->pin_count++)) + return ce; + GEM_BUG_ON(!ce->pin_count); /* no overflow please! */ - ce->state->obj->pin_global--; - i915_gem_object_unpin_map(ce->state->obj); - i915_vma_unpin(ce->state); + ce->ops = &execlists_context_ops; - i915_gem_context_put(ctx); + return __execlists_context_pin(engine, ctx, ce); } static int execlists_request_alloc(struct i915_request *request) { - struct intel_context *ce = - to_intel_context(request->ctx, request->engine); int ret; - GEM_BUG_ON(!ce->pin_count); + GEM_BUG_ON(!request->hw_context->pin_count); /* Flush enough space to reduce the likelihood of waiting after * we start building the request - in which case we will just @@ -1538,29 +1496,56 @@ static u32 *gen8_init_indirectctx_bb(struct intel_engine_cs *engine, u32 *batch) return batch; } -static u32 *gen9_init_indirectctx_bb(struct intel_engine_cs *engine, u32 *batch) +struct lri { + i915_reg_t reg; + u32 value; +}; + +static u32 *emit_lri(u32 *batch, const struct lri *lri, unsigned int count) { - *batch++ = MI_ARB_ON_OFF | MI_ARB_DISABLE; + GEM_BUG_ON(!count || count > 63); - /* WaFlushCoherentL3CacheLinesAtContextSwitch:skl,bxt,glk */ - batch = gen8_emit_flush_coherentl3_wa(engine, batch); + *batch++ = MI_LOAD_REGISTER_IMM(count); + do { + *batch++ = i915_mmio_reg_offset(lri->reg); + *batch++ = lri->value; + } while (lri++, --count); + *batch++ = MI_NOOP; - *batch++ = MI_LOAD_REGISTER_IMM(3); + return batch; +} - /* WaDisableGatherAtSetShaderCommonSlice:skl,bxt,kbl,glk */ - *batch++ = i915_mmio_reg_offset(COMMON_SLICE_CHICKEN2); - *batch++ = _MASKED_BIT_DISABLE( - GEN9_DISABLE_GATHER_AT_SET_SHADER_COMMON_SLICE); +static u32 *gen9_init_indirectctx_bb(struct intel_engine_cs *engine, u32 *batch) +{ + static const struct lri lri[] = { + /* WaDisableGatherAtSetShaderCommonSlice:skl,bxt,kbl,glk */ + { + COMMON_SLICE_CHICKEN2, + __MASKED_FIELD(GEN9_DISABLE_GATHER_AT_SET_SHADER_COMMON_SLICE, + 0), + }, + + /* BSpec: 11391 */ + { + FF_SLICE_CHICKEN, + __MASKED_FIELD(FF_SLICE_CHICKEN_CL_PROVOKING_VERTEX_FIX, + FF_SLICE_CHICKEN_CL_PROVOKING_VERTEX_FIX), + }, + + /* BSpec: 11299 */ + { + _3D_CHICKEN3, + __MASKED_FIELD(_3D_CHICKEN_SF_PROVOKING_VERTEX_FIX, + _3D_CHICKEN_SF_PROVOKING_VERTEX_FIX), + } + }; - /* BSpec: 11391 */ - *batch++ = i915_mmio_reg_offset(FF_SLICE_CHICKEN); - *batch++ = _MASKED_BIT_ENABLE(FF_SLICE_CHICKEN_CL_PROVOKING_VERTEX_FIX); + *batch++ = MI_ARB_ON_OFF | MI_ARB_DISABLE; - /* BSpec: 11299 */ - *batch++ = i915_mmio_reg_offset(_3D_CHICKEN3); - *batch++ = _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_PROVOKING_VERTEX_FIX); + /* WaFlushCoherentL3CacheLinesAtContextSwitch:skl,bxt,glk */ + batch = gen8_emit_flush_coherentl3_wa(engine, batch); - *batch++ = MI_NOOP; + batch = emit_lri(batch, lri, ARRAY_SIZE(lri)); /* WaClearSlmSpaceAtContextSwitch:kbl */ /* Actual scratch location is at 128 bytes offset */ @@ -1652,7 +1637,7 @@ static int lrc_setup_wa_ctx(struct intel_engine_cs *engine) if (IS_ERR(obj)) return PTR_ERR(obj); - vma = i915_vma_instance(obj, &engine->i915->ggtt.base, NULL); + vma = i915_vma_instance(obj, &engine->i915->ggtt.vm, NULL); if (IS_ERR(vma)) { err = PTR_ERR(vma); goto err; @@ -1767,17 +1752,29 @@ static void enable_execlists(struct intel_engine_cs *engine) I915_WRITE(RING_MODE_GEN7(engine), _MASKED_BIT_ENABLE(GFX_RUN_LIST_ENABLE)); + I915_WRITE(RING_MI_MODE(engine->mmio_base), + _MASKED_BIT_DISABLE(STOP_RING)); + I915_WRITE(RING_HWS_PGA(engine->mmio_base), engine->status_page.ggtt_offset); POSTING_READ(RING_HWS_PGA(engine->mmio_base)); +} - /* Following the reset, we need to reload the CSB read/write pointers */ - engine->execlists.csb_head = -1; +static bool unexpected_starting_state(struct intel_engine_cs *engine) +{ + struct drm_i915_private *dev_priv = engine->i915; + bool unexpected = false; + + if (I915_READ(RING_MI_MODE(engine->mmio_base)) & STOP_RING) { + DRM_DEBUG_DRIVER("STOP_RING still set in RING_MI_MODE\n"); + unexpected = true; + } + + return unexpected; } static int gen8_init_common_ring(struct intel_engine_cs *engine) { - struct intel_engine_execlists * const execlists = &engine->execlists; int ret; ret = intel_mocs_init_engine(engine); @@ -1785,13 +1782,14 @@ static int gen8_init_common_ring(struct intel_engine_cs *engine) return ret; intel_engine_reset_breadcrumbs(engine); - intel_engine_init_hangcheck(engine); - enable_execlists(engine); + if (GEM_SHOW_DEBUG() && unexpected_starting_state(engine)) { + struct drm_printer p = drm_debug_printer(__func__); - /* After a GPU reset, we may have requests to replay */ - if (execlists->first) - tasklet_schedule(&execlists->tasklet); + intel_engine_dump(engine, &p, NULL); + } + + enable_execlists(engine); return 0; } @@ -1833,8 +1831,69 @@ static int gen9_init_render_ring(struct intel_engine_cs *engine) return 0; } -static void reset_common_ring(struct intel_engine_cs *engine, - struct i915_request *request) +static struct i915_request * +execlists_reset_prepare(struct intel_engine_cs *engine) +{ + struct intel_engine_execlists * const execlists = &engine->execlists; + struct i915_request *request, *active; + unsigned long flags; + + GEM_TRACE("%s\n", engine->name); + + /* + * Prevent request submission to the hardware until we have + * completed the reset in i915_gem_reset_finish(). If a request + * is completed by one engine, it may then queue a request + * to a second via its execlists->tasklet *just* as we are + * calling engine->init_hw() and also writing the ELSP. + * Turning off the execlists->tasklet until the reset is over + * prevents the race. + */ + __tasklet_disable_sync_once(&execlists->tasklet); + + spin_lock_irqsave(&engine->timeline.lock, flags); + + /* + * We want to flush the pending context switches, having disabled + * the tasklet above, we can assume exclusive access to the execlists. + * For this allows us to catch up with an inflight preemption event, + * and avoid blaming an innocent request if the stall was due to the + * preemption itself. + */ + process_csb(engine); + + /* + * The last active request can then be no later than the last request + * now in ELSP[0]. So search backwards from there, so that if the GPU + * has advanced beyond the last CSB update, it will be pardoned. + */ + active = NULL; + request = port_request(execlists->port); + if (request) { + /* + * Prevent the breadcrumb from advancing before we decide + * which request is currently active. + */ + intel_engine_stop_cs(engine); + + list_for_each_entry_from_reverse(request, + &engine->timeline.requests, + link) { + if (__i915_request_completed(request, + request->global_seqno)) + break; + + active = request; + } + } + + spin_unlock_irqrestore(&engine->timeline.lock, flags); + + return active; +} + +static void execlists_reset(struct intel_engine_cs *engine, + struct i915_request *request) { struct intel_engine_execlists * const execlists = &engine->execlists; unsigned long flags; @@ -1844,8 +1903,7 @@ static void reset_common_ring(struct intel_engine_cs *engine, engine->name, request ? request->global_seqno : 0, intel_engine_get_seqno(engine)); - /* See execlists_cancel_requests() for the irq/spinlock split. */ - local_irq_save(flags); + spin_lock_irqsave(&engine->timeline.lock, flags); /* * Catch up with any missed context-switch interrupts. @@ -1857,14 +1915,14 @@ static void reset_common_ring(struct intel_engine_cs *engine, * requests were completed. */ execlists_cancel_port_requests(execlists); - reset_irq(engine); /* Push back any incomplete requests for replay after the reset. */ - spin_lock(&engine->timeline.lock); __unwind_incomplete_requests(engine); - spin_unlock(&engine->timeline.lock); - local_irq_restore(flags); + /* Following the reset, we need to reload the CSB read/write pointers */ + reset_csb_pointers(&engine->execlists); + + spin_unlock_irqrestore(&engine->timeline.lock, flags); /* * If the request was innocent, we leave the request in the ELSP @@ -1888,35 +1946,52 @@ static void reset_common_ring(struct intel_engine_cs *engine, * future request will be after userspace has had the opportunity * to recreate its own state. */ - regs = to_intel_context(request->ctx, engine)->lrc_reg_state; - if (engine->default_state) { - void *defaults; - - defaults = i915_gem_object_pin_map(engine->default_state, - I915_MAP_WB); - if (!IS_ERR(defaults)) { - memcpy(regs, /* skip restoring the vanilla PPHWSP */ - defaults + LRC_STATE_PN * PAGE_SIZE, - engine->context_size - PAGE_SIZE); - i915_gem_object_unpin_map(engine->default_state); - } + regs = request->hw_context->lrc_reg_state; + if (engine->pinned_default_state) { + memcpy(regs, /* skip restoring the vanilla PPHWSP */ + engine->pinned_default_state + LRC_STATE_PN * PAGE_SIZE, + engine->context_size - PAGE_SIZE); } - execlists_init_reg_state(regs, request->ctx, engine, request->ring); + execlists_init_reg_state(regs, + request->gem_context, engine, request->ring); /* Move the RING_HEAD onto the breadcrumb, past the hanging batch */ regs[CTX_RING_BUFFER_START + 1] = i915_ggtt_offset(request->ring->vma); - regs[CTX_RING_HEAD + 1] = request->postfix; - request->ring->head = request->postfix; + request->ring->head = intel_ring_wrap(request->ring, request->postfix); + regs[CTX_RING_HEAD + 1] = request->ring->head; + intel_ring_update_space(request->ring); /* Reset WaIdleLiteRestore:bdw,skl as well */ unwind_wa_tail(request); } +static void execlists_reset_finish(struct intel_engine_cs *engine) +{ + struct intel_engine_execlists * const execlists = &engine->execlists; + + /* After a GPU reset, we may have requests to replay */ + if (!RB_EMPTY_ROOT(&execlists->queue.rb_root)) + tasklet_schedule(&execlists->tasklet); + + /* + * Flush the tasklet while we still have the forcewake to be sure + * that it is not allowed to sleep before we restart and reload a + * context. + * + * As before (with execlists_reset_prepare) we rely on the caller + * serialising multiple attempts to reset so that we know that we + * are the only one manipulating tasklet state. + */ + __tasklet_enable_sync_once(&execlists->tasklet); + + GEM_TRACE("%s\n", engine->name); +} + static int intel_logical_ring_emit_pdps(struct i915_request *rq) { - struct i915_hw_ppgtt *ppgtt = rq->ctx->ppgtt; + struct i915_hw_ppgtt *ppgtt = rq->gem_context->ppgtt; struct intel_engine_cs *engine = rq->engine; const int num_lri_cmds = GEN8_3LVL_PDPES * 2; u32 *cs; @@ -1955,15 +2030,15 @@ static int gen8_emit_bb_start(struct i915_request *rq, * it is unsafe in case of lite-restore (because the ctx is * not idle). PML4 is allocated during ppgtt init so this is * not needed in 48-bit.*/ - if (rq->ctx->ppgtt && - (intel_engine_flag(rq->engine) & rq->ctx->ppgtt->pd_dirty_rings) && - !i915_vm_is_48bit(&rq->ctx->ppgtt->base) && + if (rq->gem_context->ppgtt && + (intel_engine_flag(rq->engine) & rq->gem_context->ppgtt->pd_dirty_rings) && + !i915_vm_is_48bit(&rq->gem_context->ppgtt->vm) && !intel_vgpu_active(rq->i915)) { ret = intel_logical_ring_emit_pdps(rq); if (ret) return ret; - rq->ctx->ppgtt->pd_dirty_rings &= ~intel_engine_flag(rq->engine); + rq->gem_context->ppgtt->pd_dirty_rings &= ~intel_engine_flag(rq->engine); } cs = intel_ring_begin(rq, 6); @@ -2217,13 +2292,15 @@ void intel_logical_ring_cleanup(struct intel_engine_cs *engine) kfree(engine); } -static void execlists_set_default_submission(struct intel_engine_cs *engine) +void intel_execlists_set_default_submission(struct intel_engine_cs *engine) { engine->submit_request = execlists_submit_request; engine->cancel_requests = execlists_cancel_requests; engine->schedule = execlists_schedule; engine->execlists.tasklet.func = execlists_submission_tasklet; + engine->reset.prepare = execlists_reset_prepare; + engine->park = NULL; engine->unpark = NULL; @@ -2243,18 +2320,19 @@ logical_ring_default_vfuncs(struct intel_engine_cs *engine) { /* Default vfuncs which can be overriden by each engine. */ engine->init_hw = gen8_init_common_ring; - engine->reset_hw = reset_common_ring; - engine->context_pin = execlists_context_pin; - engine->context_unpin = execlists_context_unpin; + engine->reset.prepare = execlists_reset_prepare; + engine->reset.reset = execlists_reset; + engine->reset.finish = execlists_reset_finish; + engine->context_pin = execlists_context_pin; engine->request_alloc = execlists_request_alloc; engine->emit_flush = gen8_emit_flush; engine->emit_breadcrumb = gen8_emit_breadcrumb; engine->emit_breadcrumb_sz = gen8_emit_breadcrumb_sz; - engine->set_default_submission = execlists_set_default_submission; + engine->set_default_submission = intel_execlists_set_default_submission; if (INTEL_GEN(engine->i915) < 11) { engine->irq_enable = gen8_logical_ring_enable_irq; @@ -2294,28 +2372,11 @@ logical_ring_default_irqs(struct intel_engine_cs *engine) static void logical_ring_setup(struct intel_engine_cs *engine) { - struct drm_i915_private *dev_priv = engine->i915; - enum forcewake_domains fw_domains; - intel_engine_setup_common(engine); /* Intentionally left blank. */ engine->buffer = NULL; - fw_domains = intel_uncore_forcewake_for_reg(dev_priv, - RING_ELSP(engine), - FW_REG_WRITE); - - fw_domains |= intel_uncore_forcewake_for_reg(dev_priv, - RING_CONTEXT_STATUS_PTR(engine), - FW_REG_READ | FW_REG_WRITE); - - fw_domains |= intel_uncore_forcewake_for_reg(dev_priv, - RING_CONTEXT_STATUS_BUF_BASE(engine), - FW_REG_READ); - - engine->execlists.fw_domains = fw_domains; - tasklet_init(&engine->execlists.tasklet, execlists_submission_tasklet, (unsigned long)engine); @@ -2323,33 +2384,61 @@ logical_ring_setup(struct intel_engine_cs *engine) logical_ring_default_irqs(engine); } +static bool csb_force_mmio(struct drm_i915_private *i915) +{ + /* Older GVT emulation depends upon intercepting CSB mmio */ + return intel_vgpu_active(i915) && !intel_vgpu_has_hwsp_emulation(i915); +} + static int logical_ring_init(struct intel_engine_cs *engine) { + struct drm_i915_private *i915 = engine->i915; + struct intel_engine_execlists * const execlists = &engine->execlists; int ret; ret = intel_engine_init_common(engine); if (ret) goto error; - if (HAS_LOGICAL_RING_ELSQ(engine->i915)) { - engine->execlists.submit_reg = engine->i915->regs + + if (HAS_LOGICAL_RING_ELSQ(i915)) { + execlists->submit_reg = i915->regs + i915_mmio_reg_offset(RING_EXECLIST_SQ_CONTENTS(engine)); - engine->execlists.ctrl_reg = engine->i915->regs + + execlists->ctrl_reg = i915->regs + i915_mmio_reg_offset(RING_EXECLIST_CONTROL(engine)); } else { - engine->execlists.submit_reg = engine->i915->regs + + execlists->submit_reg = i915->regs + i915_mmio_reg_offset(RING_ELSP(engine)); } - engine->execlists.preempt_complete_status = ~0u; - if (engine->i915->preempt_context) { + execlists->preempt_complete_status = ~0u; + if (i915->preempt_context) { struct intel_context *ce = - to_intel_context(engine->i915->preempt_context, engine); + to_intel_context(i915->preempt_context, engine); - engine->execlists.preempt_complete_status = + execlists->preempt_complete_status = upper_32_bits(ce->lrc_desc); } + execlists->csb_read = + i915->regs + i915_mmio_reg_offset(RING_CONTEXT_STATUS_PTR(engine)); + if (csb_force_mmio(i915)) { + execlists->csb_status = (u32 __force *) + (i915->regs + i915_mmio_reg_offset(RING_CONTEXT_STATUS_BUF_LO(engine, 0))); + + execlists->csb_write = (u32 __force *)execlists->csb_read; + execlists->csb_write_reset = + _MASKED_FIELD(GEN8_CSB_WRITE_PTR_MASK, + GEN8_CSB_ENTRIES - 1); + } else { + execlists->csb_status = + &engine->status_page.page_addr[I915_HWS_CSB_BUF0_INDEX]; + + execlists->csb_write = + &engine->status_page.page_addr[intel_hws_csb_write_index(i915)]; + execlists->csb_write_reset = GEN8_CSB_ENTRIES - 1; + } + reset_csb_pointers(execlists); + return 0; error: @@ -2482,7 +2571,7 @@ static void execlists_init_reg_state(u32 *regs, struct drm_i915_private *dev_priv = engine->i915; struct i915_hw_ppgtt *ppgtt = ctx->ppgtt ?: dev_priv->mm.aliasing_ppgtt; u32 base = engine->mmio_base; - bool rcs = engine->id == RCS; + bool rcs = engine->class == RENDER_CLASS; /* A context is actually a big batch buffer with several * MI_LOAD_REGISTER_IMM commands followed by (reg, value) pairs. The @@ -2550,7 +2639,7 @@ static void execlists_init_reg_state(u32 *regs, CTX_REG(regs, CTX_PDP0_UDW, GEN8_RING_PDP_UDW(engine, 0), 0); CTX_REG(regs, CTX_PDP0_LDW, GEN8_RING_PDP_LDW(engine, 0), 0); - if (ppgtt && i915_vm_is_48bit(&ppgtt->base)) { + if (ppgtt && i915_vm_is_48bit(&ppgtt->vm)) { /* 64b PPGTT (48bit canonical) * PDP0_DESCRIPTOR contains the base address to PML4 and * other PDP Descriptors are ignored. @@ -2629,10 +2718,10 @@ err_unpin_ctx: } static int execlists_context_deferred_alloc(struct i915_gem_context *ctx, - struct intel_engine_cs *engine) + struct intel_engine_cs *engine, + struct intel_context *ce) { struct drm_i915_gem_object *ctx_obj; - struct intel_context *ce = to_intel_context(ctx, engine); struct i915_vma *vma; uint32_t context_size; struct intel_ring *ring; @@ -2654,7 +2743,7 @@ static int execlists_context_deferred_alloc(struct i915_gem_context *ctx, if (IS_ERR(ctx_obj)) return PTR_ERR(ctx_obj); - vma = i915_vma_instance(ctx_obj, &ctx->i915->ggtt.base, NULL); + vma = i915_vma_instance(ctx_obj, &ctx->i915->ggtt.vm, NULL); if (IS_ERR(vma)) { ret = PTR_ERR(vma); goto error_deref_obj; |