diff options
Diffstat (limited to 'drivers/gpu/drm/i915/intel_lrc.c')
| -rw-r--r-- | drivers/gpu/drm/i915/intel_lrc.c | 726 |
1 files changed, 450 insertions, 276 deletions
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c index e0fca03..697af5a 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 "intel_lrc_reg.h" #include "intel_mocs.h" #define RING_EXECLIST_QFULL (1 << 0x2) @@ -156,60 +157,10 @@ #define GEN8_CTX_STATUS_COMPLETED_MASK \ (GEN8_CTX_STATUS_COMPLETE | GEN8_CTX_STATUS_PREEMPTED) -#define CTX_LRI_HEADER_0 0x01 -#define CTX_CONTEXT_CONTROL 0x02 -#define CTX_RING_HEAD 0x04 -#define CTX_RING_TAIL 0x06 -#define CTX_RING_BUFFER_START 0x08 -#define CTX_RING_BUFFER_CONTROL 0x0a -#define CTX_BB_HEAD_U 0x0c -#define CTX_BB_HEAD_L 0x0e -#define CTX_BB_STATE 0x10 -#define CTX_SECOND_BB_HEAD_U 0x12 -#define CTX_SECOND_BB_HEAD_L 0x14 -#define CTX_SECOND_BB_STATE 0x16 -#define CTX_BB_PER_CTX_PTR 0x18 -#define CTX_RCS_INDIRECT_CTX 0x1a -#define CTX_RCS_INDIRECT_CTX_OFFSET 0x1c -#define CTX_LRI_HEADER_1 0x21 -#define CTX_CTX_TIMESTAMP 0x22 -#define CTX_PDP3_UDW 0x24 -#define CTX_PDP3_LDW 0x26 -#define CTX_PDP2_UDW 0x28 -#define CTX_PDP2_LDW 0x2a -#define CTX_PDP1_UDW 0x2c -#define CTX_PDP1_LDW 0x2e -#define CTX_PDP0_UDW 0x30 -#define CTX_PDP0_LDW 0x32 -#define CTX_LRI_HEADER_2 0x41 -#define CTX_R_PWR_CLK_STATE 0x42 -#define CTX_GPGPU_CSR_BASE_ADDRESS 0x44 - -#define CTX_REG(reg_state, pos, reg, val) do { \ - (reg_state)[(pos)+0] = i915_mmio_reg_offset(reg); \ - (reg_state)[(pos)+1] = (val); \ -} while (0) - -#define ASSIGN_CTX_PDP(ppgtt, reg_state, n) do { \ - const u64 _addr = i915_page_dir_dma_addr((ppgtt), (n)); \ - reg_state[CTX_PDP ## n ## _UDW+1] = upper_32_bits(_addr); \ - reg_state[CTX_PDP ## n ## _LDW+1] = lower_32_bits(_addr); \ -} while (0) - -#define ASSIGN_CTX_PML4(ppgtt, reg_state) do { \ - reg_state[CTX_PDP0_UDW + 1] = upper_32_bits(px_dma(&ppgtt->pml4)); \ - reg_state[CTX_PDP0_LDW + 1] = lower_32_bits(px_dma(&ppgtt->pml4)); \ -} while (0) - -#define GEN8_CTX_RCS_INDIRECT_CTX_OFFSET_DEFAULT 0x17 -#define GEN9_CTX_RCS_INDIRECT_CTX_OFFSET_DEFAULT 0x26 -#define GEN10_CTX_RCS_INDIRECT_CTX_OFFSET_DEFAULT 0x19 - /* 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) -#define PREEMPT_ID 0x1 static int execlists_context_deferred_alloc(struct i915_gem_context *ctx, struct intel_engine_cs *engine); @@ -218,6 +169,23 @@ static void execlists_init_reg_state(u32 *reg_state, struct intel_engine_cs *engine, struct intel_ring *ring); +static inline struct i915_priolist *to_priolist(struct rb_node *rb) +{ + return rb_entry(rb, struct i915_priolist, node); +} + +static inline int rq_prio(const struct i915_request *rq) +{ + return rq->priotree.priority; +} + +static inline bool need_preempt(const struct intel_engine_cs *engine, + const struct i915_request *last, + int prio) +{ + return engine->i915->preempt_context && prio > max(rq_prio(last), 0); +} + /** * intel_lr_context_descriptor_update() - calculate & cache the descriptor * descriptor for a pinned context @@ -236,6 +204,18 @@ static void execlists_init_reg_state(u32 *reg_state, * bits 32-52: ctx ID, a globally unique tag * bits 53-54: mbz, reserved for use by hardware * bits 55-63: group ID, currently unused and set to 0 + * + * Starting from Gen11, the upper dword of the descriptor has a new format: + * + * bits 32-36: reserved + * bits 37-47: SW context ID + * bits 48:53: engine instance + * bit 54: mbz, reserved for use by hardware + * bits 55-60: SW counter + * bits 61-63: engine class + * + * engine info, SW context ID and SW counter need to form a unique number + * (Context ID) per lrc. */ static void intel_lr_context_descriptor_update(struct i915_gem_context *ctx, @@ -244,12 +224,32 @@ intel_lr_context_descriptor_update(struct i915_gem_context *ctx, struct intel_context *ce = &ctx->engine[engine->id]; u64 desc; - BUILD_BUG_ON(MAX_CONTEXT_HW_ID > (1<<GEN8_CTX_ID_WIDTH)); + BUILD_BUG_ON(MAX_CONTEXT_HW_ID > (BIT(GEN8_CTX_ID_WIDTH))); + BUILD_BUG_ON(GEN11_MAX_CONTEXT_HW_ID > (BIT(GEN11_SW_CTX_ID_WIDTH))); desc = ctx->desc_template; /* bits 0-11 */ + GEM_BUG_ON(desc & GENMASK_ULL(63, 12)); + desc |= i915_ggtt_offset(ce->state) + LRC_HEADER_PAGES * PAGE_SIZE; /* bits 12-31 */ - desc |= (u64)ctx->hw_id << GEN8_CTX_ID_SHIFT; /* bits 32-52 */ + GEM_BUG_ON(desc & GENMASK_ULL(63, 32)); + + 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; + /* bits 37-47 */ + + desc |= (u64)engine->instance << GEN11_ENGINE_INSTANCE_SHIFT; + /* bits 48-53 */ + + /* TODO: decide what to do with SW counter (bits 55-60) */ + + desc |= (u64)engine->class << GEN11_ENGINE_CLASS_SHIFT; + /* bits 61-63 */ + } else { + GEM_BUG_ON(ctx->hw_id >= BIT(GEN8_CTX_ID_WIDTH)); + desc |= (u64)ctx->hw_id << GEN8_CTX_ID_SHIFT; /* bits 32-52 */ + } ce->lrc_desc = desc; } @@ -273,7 +273,7 @@ find_priolist: parent = &execlists->queue.rb_node; while (*parent) { rb = *parent; - p = rb_entry(rb, typeof(*p), node); + p = to_priolist(rb); if (prio > p->priority) { parent = &rb->rb_left; } else if (prio < p->priority) { @@ -313,10 +313,10 @@ find_priolist: if (first) execlists->first = &p->node; - return ptr_pack_bits(p, first, 1); + return p; } -static void unwind_wa_tail(struct drm_i915_gem_request *rq) +static void unwind_wa_tail(struct i915_request *rq) { rq->tail = intel_ring_wrap(rq->ring, rq->wa_tail - WA_TAIL_BYTES); assert_ring_tail_valid(rq->ring, rq->tail); @@ -324,7 +324,7 @@ static void unwind_wa_tail(struct drm_i915_gem_request *rq) static void __unwind_incomplete_requests(struct intel_engine_cs *engine) { - struct drm_i915_gem_request *rq, *rn; + struct i915_request *rq, *rn; struct i915_priolist *uninitialized_var(p); int last_prio = I915_PRIORITY_INVALID; @@ -333,20 +333,16 @@ static void __unwind_incomplete_requests(struct intel_engine_cs *engine) list_for_each_entry_safe_reverse(rq, rn, &engine->timeline->requests, link) { - if (i915_gem_request_completed(rq)) + if (i915_request_completed(rq)) return; - __i915_gem_request_unsubmit(rq); + __i915_request_unsubmit(rq); unwind_wa_tail(rq); - GEM_BUG_ON(rq->priotree.priority == I915_PRIORITY_INVALID); - if (rq->priotree.priority != last_prio) { - p = lookup_priolist(engine, - &rq->priotree, - rq->priotree.priority); - p = ptr_mask_bits(p, 1); - - last_prio = rq->priotree.priority; + GEM_BUG_ON(rq_prio(rq) == I915_PRIORITY_INVALID); + if (rq_prio(rq) != last_prio) { + last_prio = rq_prio(rq); + p = lookup_priolist(engine, &rq->priotree, last_prio); } list_add(&rq->priotree.link, &p->requests); @@ -365,8 +361,7 @@ execlists_unwind_incomplete_requests(struct intel_engine_execlists *execlists) } static inline void -execlists_context_status_change(struct drm_i915_gem_request *rq, - unsigned long status) +execlists_context_status_change(struct i915_request *rq, unsigned long status) { /* * Only used when GVT-g is enabled now. When GVT-g is disabled, @@ -380,14 +375,14 @@ execlists_context_status_change(struct drm_i915_gem_request *rq, } static inline void -execlists_context_schedule_in(struct drm_i915_gem_request *rq) +execlists_context_schedule_in(struct i915_request *rq) { execlists_context_status_change(rq, INTEL_CONTEXT_SCHEDULE_IN); intel_engine_context_in(rq->engine); } static inline void -execlists_context_schedule_out(struct drm_i915_gem_request *rq) +execlists_context_schedule_out(struct i915_request *rq) { intel_engine_context_out(rq->engine); execlists_context_status_change(rq, INTEL_CONTEXT_SCHEDULE_OUT); @@ -402,7 +397,7 @@ execlists_update_context_pdps(struct i915_hw_ppgtt *ppgtt, u32 *reg_state) ASSIGN_CTX_PDP(ppgtt, reg_state, 0); } -static u64 execlists_update_context(struct drm_i915_gem_request *rq) +static u64 execlists_update_context(struct i915_request *rq) { struct intel_context *ce = &rq->ctx->engine[rq->engine->id]; struct i915_hw_ppgtt *ppgtt = @@ -422,19 +417,31 @@ static u64 execlists_update_context(struct drm_i915_gem_request *rq) return ce->lrc_desc; } -static inline void elsp_write(u64 desc, u32 __iomem *elsp) +static inline void write_desc(struct intel_engine_execlists *execlists, u64 desc, u32 port) { - writel(upper_32_bits(desc), elsp); - writel(lower_32_bits(desc), elsp); + if (execlists->ctrl_reg) { + writel(lower_32_bits(desc), execlists->submit_reg + port * 2); + writel(upper_32_bits(desc), execlists->submit_reg + port * 2 + 1); + } else { + writel(upper_32_bits(desc), execlists->submit_reg); + writel(lower_32_bits(desc), execlists->submit_reg); + } } static void execlists_submit_ports(struct intel_engine_cs *engine) { - struct execlist_port *port = engine->execlists.port; + struct intel_engine_execlists *execlists = &engine->execlists; + struct execlist_port *port = execlists->port; unsigned int n; - for (n = execlists_num_ports(&engine->execlists); n--; ) { - struct drm_i915_gem_request *rq; + /* + * 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 + * of elsq entries, keep this in mind before changing the loop below. + */ + for (n = execlists_num_ports(execlists); n--; ) { + struct i915_request *rq; unsigned int count; u64 desc; @@ -447,18 +454,24 @@ static void execlists_submit_ports(struct intel_engine_cs *engine) desc = execlists_update_context(rq); GEM_DEBUG_EXEC(port[n].context_id = upper_32_bits(desc)); - GEM_TRACE("%s in[%d]: ctx=%d.%d, seqno=%x\n", + GEM_TRACE("%s in[%d]: ctx=%d.%d, seqno=%x, prio=%d\n", engine->name, n, port[n].context_id, count, - rq->global_seqno); + rq->global_seqno, + rq_prio(rq)); } else { GEM_BUG_ON(!n); desc = 0; } - elsp_write(desc, engine->execlists.elsp); + write_desc(execlists, desc, n); } - execlists_clear_active(&engine->execlists, EXECLISTS_ACTIVE_HWACK); + + /* we need to manually load the submit queue */ + if (execlists->ctrl_reg) + writel(EL_CTRL_LOAD, execlists->ctrl_reg); + + execlists_clear_active(execlists, EXECLISTS_ACTIVE_HWACK); } static bool ctx_single_port_submission(const struct i915_gem_context *ctx) @@ -479,37 +492,47 @@ static bool can_merge_ctx(const struct i915_gem_context *prev, return true; } -static void port_assign(struct execlist_port *port, - struct drm_i915_gem_request *rq) +static void port_assign(struct execlist_port *port, struct i915_request *rq) { GEM_BUG_ON(rq == port_request(port)); if (port_isset(port)) - i915_gem_request_put(port_request(port)); + i915_request_put(port_request(port)); - port_set(port, port_pack(i915_gem_request_get(rq), port_count(port))); + port_set(port, port_pack(i915_request_get(rq), port_count(port))); } static void inject_preempt_context(struct intel_engine_cs *engine) { + struct intel_engine_execlists *execlists = &engine->execlists; struct intel_context *ce = &engine->i915->preempt_context->engine[engine->id]; unsigned int n; - GEM_BUG_ON(engine->i915->preempt_context->hw_id != PREEMPT_ID); - GEM_BUG_ON(!IS_ALIGNED(ce->ring->size, WA_TAIL_BYTES)); - - memset(ce->ring->vaddr + ce->ring->tail, 0, WA_TAIL_BYTES); - ce->ring->tail += WA_TAIL_BYTES; - ce->ring->tail &= (ce->ring->size - 1); - ce->lrc_reg_state[CTX_RING_TAIL+1] = ce->ring->tail; + GEM_BUG_ON(execlists->preempt_complete_status != + upper_32_bits(ce->lrc_desc)); + GEM_BUG_ON((ce->lrc_reg_state[CTX_CONTEXT_CONTROL + 1] & + _MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT | + CTX_CTRL_ENGINE_CTX_SAVE_INHIBIT)) != + _MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT | + CTX_CTRL_ENGINE_CTX_SAVE_INHIBIT)); + /* + * Switch to our empty preempt context so + * the state of the GPU is known (idle). + */ GEM_TRACE("%s\n", engine->name); - for (n = execlists_num_ports(&engine->execlists); --n; ) - elsp_write(0, engine->execlists.elsp); + for (n = execlists_num_ports(execlists); --n; ) + write_desc(execlists, 0, n); + + write_desc(execlists, ce->lrc_desc, n); + + /* we need to manually load the submit queue */ + if (execlists->ctrl_reg) + writel(EL_CTRL_LOAD, execlists->ctrl_reg); - elsp_write(ce->lrc_desc, engine->execlists.elsp); execlists_clear_active(&engine->execlists, EXECLISTS_ACTIVE_HWACK); + execlists_set_active(&engine->execlists, EXECLISTS_ACTIVE_PREEMPT); } static void execlists_dequeue(struct intel_engine_cs *engine) @@ -518,7 +541,7 @@ static void execlists_dequeue(struct intel_engine_cs *engine) struct execlist_port *port = execlists->port; const struct execlist_port * const last_port = &execlists->port[execlists->port_mask]; - struct drm_i915_gem_request *last = port_request(port); + struct i915_request *last = port_request(port); struct rb_node *rb; bool submit = false; @@ -546,8 +569,6 @@ static void execlists_dequeue(struct intel_engine_cs *engine) spin_lock_irq(&engine->timeline->lock); rb = execlists->first; GEM_BUG_ON(rb_first(&execlists->queue) != rb); - if (!rb) - goto unlock; if (last) { /* @@ -570,55 +591,49 @@ static void execlists_dequeue(struct intel_engine_cs *engine) if (!execlists_is_active(execlists, EXECLISTS_ACTIVE_HWACK)) goto unlock; - if (HAS_LOGICAL_RING_PREEMPTION(engine->i915) && - rb_entry(rb, struct i915_priolist, node)->priority > - max(last->priotree.priority, 0)) { - /* - * Switch to our empty preempt context so - * the state of the GPU is known (idle). - */ + if (need_preempt(engine, last, execlists->queue_priority)) { inject_preempt_context(engine); - execlists_set_active(execlists, - EXECLISTS_ACTIVE_PREEMPT); goto unlock; - } else { - /* - * In theory, we could coalesce more requests onto - * the second port (the first port is active, with - * no preemptions pending). However, that means we - * then have to deal with the possible lite-restore - * of the second port (as we submit the ELSP, there - * may be a context-switch) but also we may complete - * the resubmission before the context-switch. Ergo, - * coalescing onto the second port will cause a - * preemption event, but we cannot predict whether - * that will affect port[0] or port[1]. - * - * If the second port is already active, we can wait - * until the next context-switch before contemplating - * new requests. The GPU will be busy and we should be - * able to resubmit the new ELSP before it idles, - * avoiding pipeline bubbles (momentary pauses where - * the driver is unable to keep up the supply of new - * work). - */ - if (port_count(&port[1])) - goto unlock; - - /* WaIdleLiteRestore:bdw,skl - * Apply the wa NOOPs to prevent - * ring:HEAD == req:TAIL as we resubmit the - * request. See gen8_emit_breadcrumb() for - * where we prepare the padding after the - * end of the request. - */ - last->tail = last->wa_tail; } + + /* + * In theory, we could coalesce more requests onto + * the second port (the first port is active, with + * no preemptions pending). However, that means we + * then have to deal with the possible lite-restore + * of the second port (as we submit the ELSP, there + * may be a context-switch) but also we may complete + * the resubmission before the context-switch. Ergo, + * coalescing onto the second port will cause a + * preemption event, but we cannot predict whether + * that will affect port[0] or port[1]. + * + * If the second port is already active, we can wait + * until the next context-switch before contemplating + * new requests. The GPU will be busy and we should be + * able to resubmit the new ELSP before it idles, + * avoiding pipeline bubbles (momentary pauses where + * the driver is unable to keep up the supply of new + * work). However, we have to double check that the + * priorities of the ports haven't been switch. + */ + if (port_count(&port[1])) + goto unlock; + + /* + * WaIdleLiteRestore:bdw,skl + * Apply the wa NOOPs to prevent + * ring:HEAD == rq:TAIL as we resubmit the + * request. See gen8_emit_breadcrumb() for + * where we prepare the padding after the + * end of the request. + */ + last->tail = last->wa_tail; } - do { - struct i915_priolist *p = rb_entry(rb, typeof(*p), node); - struct drm_i915_gem_request *rq, *rn; + while (rb) { + struct i915_priolist *p = to_priolist(rb); + struct i915_request *rq, *rn; list_for_each_entry_safe(rq, rn, &p->requests, priotree.link) { /* @@ -668,8 +683,8 @@ static void execlists_dequeue(struct intel_engine_cs *engine) } INIT_LIST_HEAD(&rq->priotree.link); - __i915_gem_request_submit(rq); - trace_i915_gem_request_in(rq, port_index(port, execlists)); + __i915_request_submit(rq); + trace_i915_request_in(rq, port_index(port, execlists)); last = rq; submit = true; } @@ -679,11 +694,16 @@ static void execlists_dequeue(struct intel_engine_cs *engine) INIT_LIST_HEAD(&p->requests); if (p->priority != I915_PRIORITY_NORMAL) kmem_cache_free(engine->i915->priorities, p); - } while (rb); + } done: + execlists->queue_priority = rb ? to_priolist(rb)->priority : INT_MIN; execlists->first = rb; if (submit) port_assign(port, last); + + /* We must always keep the beast fed if we have work piled up */ + GEM_BUG_ON(execlists->first && !port_isset(execlists->port)); + unlock: spin_unlock_irq(&engine->timeline->lock); @@ -691,6 +711,9 @@ unlock: execlists_set_active(execlists, EXECLISTS_ACTIVE_USER); execlists_submit_ports(engine); } + + GEM_BUG_ON(port_isset(execlists->port) && + !execlists_is_active(execlists, EXECLISTS_ACTIVE_USER)); } void @@ -700,12 +723,17 @@ execlists_cancel_port_requests(struct intel_engine_execlists * const execlists) unsigned int num_ports = execlists_num_ports(execlists); while (num_ports-- && port_isset(port)) { - struct drm_i915_gem_request *rq = port_request(port); + struct i915_request *rq = port_request(port); GEM_BUG_ON(!execlists->active); intel_engine_context_out(rq->engine); - execlists_context_status_change(rq, INTEL_CONTEXT_SCHEDULE_PREEMPTED); - i915_gem_request_put(rq); + + execlists_context_status_change(rq, + i915_request_completed(rq) ? + INTEL_CONTEXT_SCHEDULE_OUT : + INTEL_CONTEXT_SCHEDULE_PREEMPTED); + + i915_request_put(rq); memset(port, 0, sizeof(*port)); port++; @@ -715,34 +743,50 @@ execlists_cancel_port_requests(struct intel_engine_execlists * const execlists) static void execlists_cancel_requests(struct intel_engine_cs *engine) { struct intel_engine_execlists * const execlists = &engine->execlists; - struct drm_i915_gem_request *rq, *rn; + struct i915_request *rq, *rn; struct rb_node *rb; unsigned long flags; GEM_TRACE("%s\n", engine->name); - spin_lock_irqsave(&engine->timeline->lock, flags); + /* + * Before we call engine->cancel_requests(), we should have exclusive + * access to the submission state. This is arranged for us by the + * caller disabling the interrupt generation, the tasklet and other + * threads that may then access the same state, giving us a free hand + * to reset state. However, we still need to let lockdep be aware that + * we know this state may be accessed in hardirq context, so we + * disable the irq around this manipulation and we want to keep + * the spinlock focused on its duties and not accidentally conflate + * coverage to the submission's irq state. (Similarly, although we + * shouldn't need to disable irq around the manipulation of the + * submission's irq state, we also wish to remind ourselves that + * it is irq state.) + */ + local_irq_save(flags); /* Cancel the requests on the HW and clear the ELSP tracker. */ execlists_cancel_port_requests(execlists); + spin_lock(&engine->timeline->lock); + /* Mark all executing requests as skipped. */ list_for_each_entry(rq, &engine->timeline->requests, link) { GEM_BUG_ON(!rq->global_seqno); - if (!i915_gem_request_completed(rq)) + if (!i915_request_completed(rq)) dma_fence_set_error(&rq->fence, -EIO); } /* Flush the queued requests to the timeline list (for retiring). */ rb = execlists->first; while (rb) { - struct i915_priolist *p = rb_entry(rb, typeof(*p), node); + struct i915_priolist *p = to_priolist(rb); list_for_each_entry_safe(rq, rn, &p->requests, priotree.link) { INIT_LIST_HEAD(&rq->priotree.link); dma_fence_set_error(&rq->fence, -EIO); - __i915_gem_request_submit(rq); + __i915_request_submit(rq); } rb = rb_next(rb); @@ -754,11 +798,13 @@ 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; GEM_BUG_ON(port_isset(execlists->port)); + spin_unlock(&engine->timeline->lock); + /* * The port is checked prior to scheduling a tasklet, but * just in case we have suspended the tasklet to do the @@ -770,7 +816,7 @@ static void execlists_cancel_requests(struct intel_engine_cs *engine) /* Mark all CS interrupts as complete */ execlists->active = 0; - spin_unlock_irqrestore(&engine->timeline->lock, flags); + local_irq_restore(flags); } /* @@ -783,8 +829,10 @@ static void execlists_submission_tasklet(unsigned long data) struct intel_engine_execlists * const execlists = &engine->execlists; struct execlist_port * const port = execlists->port; struct drm_i915_private *dev_priv = engine->i915; + bool fw = false; - /* We can skip acquiring intel_runtime_pm_get() here as it was taken + /* + * 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 @@ -793,9 +841,8 @@ static void execlists_submission_tasklet(unsigned long data) */ GEM_BUG_ON(!dev_priv->gt.awake); - intel_uncore_forcewake_get(dev_priv, execlists->fw_domains); - - /* Prefer doing test_and_clear_bit() as a two stage operation to avoid + /* + * 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). */ @@ -811,18 +858,17 @@ static void execlists_submission_tasklet(unsigned long data) execlists->csb_head = -1; /* force mmio read of CSB ptrs */ } - /* The write will be ordered by the uncached read (itself - * a memory barrier), so we do not need another in the form - * of a locked instruction. The race between the interrupt - * handler and the split test/clear is harmless as we order - * our clear before the CSB read. If the interrupt arrived - * first between the test and the clear, we read the updated - * CSB and clear the bit. If the interrupt arrives as we read - * the CSB or later (i.e. after we had cleared the bit) the bit - * is set and we do a new loop. - */ - __clear_bit(ENGINE_IRQ_EXECLIST, &engine->irq_posted); + /* Clear before reading to catch new interrupts */ + clear_bit(ENGINE_IRQ_EXECLIST, &engine->irq_posted); + smp_mb__after_atomic(); + if (unlikely(execlists->csb_head == -1)) { /* following a reset */ + if (!fw) { + intel_uncore_forcewake_get(dev_priv, + execlists->fw_domains); + fw = true; + } + 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); @@ -835,13 +881,13 @@ static void execlists_submission_tasklet(unsigned long data) head = execlists->csb_head; tail = READ_ONCE(buf[write_idx]); } - GEM_TRACE("%s cs-irq head=%d [%d], tail=%d [%d]\n", + 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)))), - tail, GEN8_CSB_WRITE_PTR(readl(dev_priv->regs + i915_mmio_reg_offset(RING_CONTEXT_STATUS_PTR(engine))))); + 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 drm_i915_gem_request *rq; + struct i915_request *rq; unsigned int status; unsigned int count; @@ -886,7 +932,7 @@ static void execlists_submission_tasklet(unsigned long data) GEM_BUG_ON(status & GEN8_CTX_STATUS_IDLE_ACTIVE); if (status & GEN8_CTX_STATUS_COMPLETE && - buf[2*head + 1] == PREEMPT_ID) { + buf[2*head + 1] == execlists->preempt_complete_status) { GEM_TRACE("%s preempt-idle\n", engine->name); execlists_cancel_port_requests(execlists); @@ -907,23 +953,28 @@ static void execlists_submission_tasklet(unsigned long data) GEM_BUG_ON(!execlists_is_active(execlists, EXECLISTS_ACTIVE_USER)); - /* Check the context/desc id for this event matches */ - GEM_DEBUG_BUG_ON(buf[2 * head + 1] != port->context_id); - rq = port_unpack(port, &count); - GEM_TRACE("%s out[0]: ctx=%d.%d, seqno=%x\n", + GEM_TRACE("%s out[0]: ctx=%d.%d, seqno=%x, prio=%d\n", engine->name, port->context_id, count, - rq ? rq->global_seqno : 0); + rq ? rq->global_seqno : 0, + 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) { GEM_BUG_ON(status & GEN8_CTX_STATUS_PREEMPTED); GEM_BUG_ON(port_isset(&port[1]) && !(status & GEN8_CTX_STATUS_ELEMENT_SWITCH)); - GEM_BUG_ON(!i915_gem_request_completed(rq)); + GEM_BUG_ON(!i915_request_completed(rq)); execlists_context_schedule_out(rq); - trace_i915_gem_request_out(rq); - i915_gem_request_put(rq); + trace_i915_request_out(rq); + i915_request_put(rq); + + GEM_TRACE("%s completed ctx=%d\n", + engine->name, port->context_id); execlists_port_complete(execlists, port); } else { @@ -948,21 +999,26 @@ static void execlists_submission_tasklet(unsigned long data) if (!execlists_is_active(execlists, EXECLISTS_ACTIVE_PREEMPT)) execlists_dequeue(engine); - intel_uncore_forcewake_put(dev_priv, execlists->fw_domains); + if (fw) + intel_uncore_forcewake_put(dev_priv, execlists->fw_domains); } -static void insert_request(struct intel_engine_cs *engine, - struct i915_priotree *pt, - int prio) +static void queue_request(struct intel_engine_cs *engine, + struct i915_priotree *pt, + int prio) { - struct i915_priolist *p = lookup_priolist(engine, pt, prio); + list_add_tail(&pt->link, &lookup_priolist(engine, pt, prio)->requests); +} - list_add_tail(&pt->link, &ptr_mask_bits(p, 1)->requests); - if (ptr_unmask_bits(p, 1)) +static void submit_queue(struct intel_engine_cs *engine, int prio) +{ + if (prio > engine->execlists.queue_priority) { + engine->execlists.queue_priority = prio; tasklet_hi_schedule(&engine->execlists.tasklet); + } } -static void execlists_submit_request(struct drm_i915_gem_request *request) +static void execlists_submit_request(struct i915_request *request) { struct intel_engine_cs *engine = request->engine; unsigned long flags; @@ -970,7 +1026,8 @@ static void execlists_submit_request(struct drm_i915_gem_request *request) /* Will be called from irq-context when using foreign fences. */ spin_lock_irqsave(&engine->timeline->lock, flags); - insert_request(engine, &request->priotree, request->priotree.priority); + queue_request(engine, &request->priotree, rq_prio(request)); + submit_queue(engine, rq_prio(request)); GEM_BUG_ON(!engine->execlists.first); GEM_BUG_ON(list_empty(&request->priotree.link)); @@ -978,9 +1035,9 @@ static void execlists_submit_request(struct drm_i915_gem_request *request) spin_unlock_irqrestore(&engine->timeline->lock, flags); } -static struct drm_i915_gem_request *pt_to_request(struct i915_priotree *pt) +static struct i915_request *pt_to_request(struct i915_priotree *pt) { - return container_of(pt, struct drm_i915_gem_request, priotree); + return container_of(pt, struct i915_request, priotree); } static struct intel_engine_cs * @@ -998,7 +1055,7 @@ pt_lock_engine(struct i915_priotree *pt, struct intel_engine_cs *locked) return engine; } -static void execlists_schedule(struct drm_i915_gem_request *request, int prio) +static void execlists_schedule(struct i915_request *request, int prio) { struct intel_engine_cs *engine; struct i915_dependency *dep, *p; @@ -1007,7 +1064,7 @@ static void execlists_schedule(struct drm_i915_gem_request *request, int prio) GEM_BUG_ON(prio == I915_PRIORITY_INVALID); - if (i915_gem_request_completed(request)) + if (i915_request_completed(request)) return; if (prio <= READ_ONCE(request->priotree.priority)) @@ -1019,13 +1076,14 @@ static void execlists_schedule(struct drm_i915_gem_request *request, int prio) stack.signaler = &request->priotree; list_add(&stack.dfs_link, &dfs); - /* Recursively bump all dependent priorities to match the new request. + /* + * Recursively bump all dependent priorities to match the new request. * * A naive approach would be to use recursion: * static void update_priorities(struct i915_priotree *pt, prio) { * list_for_each_entry(dep, &pt->signalers_list, signal_link) * update_priorities(dep->signal, prio) - * insert_request(pt); + * queue_request(pt); * } * but that may have unlimited recursion depth and so runs a very * real risk of overunning the kernel stack. Instead, we build @@ -1036,27 +1094,29 @@ static void execlists_schedule(struct drm_i915_gem_request *request, int prio) * end result is a topological list of requests in reverse order, the * last element in the list is the request we must execute first. */ - list_for_each_entry_safe(dep, p, &dfs, dfs_link) { + list_for_each_entry(dep, &dfs, dfs_link) { struct i915_priotree *pt = dep->signaler; - /* Within an engine, there can be no cycle, but we may + /* + * Within an engine, there can be no cycle, but we may * refer to the same dependency chain multiple times * (redundant dependencies are not eliminated) and across * engines. */ list_for_each_entry(p, &pt->signalers_list, signal_link) { - if (i915_gem_request_completed(pt_to_request(p->signaler))) + GEM_BUG_ON(p == dep); /* no cycles! */ + + if (i915_priotree_signaled(p->signaler)) continue; GEM_BUG_ON(p->signaler->priority < pt->priority); if (prio > READ_ONCE(p->signaler->priority)) list_move_tail(&p->dfs_link, &dfs); } - - list_safe_reset_next(dep, p, dfs_link); } - /* If we didn't need to bump any existing priorities, and we haven't + /* + * If we didn't need to bump any existing priorities, and we haven't * yet submitted this request (i.e. there is no potential race with * execlists_submit_request()), we can set our own priority and skip * acquiring the engine locks. @@ -1086,8 +1146,9 @@ static void execlists_schedule(struct drm_i915_gem_request *request, int prio) pt->priority = prio; if (!list_empty(&pt->link)) { __list_del_entry(&pt->link); - insert_request(engine, pt, prio); + queue_request(engine, pt, prio); } + submit_queue(engine, prio); } spin_unlock_irq(&engine->timeline->lock); @@ -1130,11 +1191,9 @@ execlists_context_pin(struct intel_engine_cs *engine, goto out; GEM_BUG_ON(!ce->pin_count); /* no overflow please! */ - if (!ce->state) { - ret = execlists_context_deferred_alloc(ctx, engine); - if (ret) - goto err; - } + ret = execlists_context_deferred_alloc(ctx, engine); + if (ret) + goto err; GEM_BUG_ON(!ce->state); ret = __context_pin(ctx, ce->state); @@ -1191,7 +1250,7 @@ static void execlists_context_unpin(struct intel_engine_cs *engine, i915_gem_context_put(ctx); } -static int execlists_request_alloc(struct drm_i915_gem_request *request) +static int execlists_request_alloc(struct i915_request *request) { struct intel_engine_cs *engine = request->engine; struct intel_context *ce = &request->ctx->engine[engine->id]; @@ -1368,6 +1427,40 @@ static u32 *gen9_init_indirectctx_bb(struct intel_engine_cs *engine, u32 *batch) return batch; } +static u32 * +gen10_init_indirectctx_bb(struct intel_engine_cs *engine, u32 *batch) +{ + int i; + + /* + * WaPipeControlBefore3DStateSamplePattern: cnl + * + * Ensure the engine is idle prior to programming a + * 3DSTATE_SAMPLE_PATTERN during a context restore. + */ + batch = gen8_emit_pipe_control(batch, + PIPE_CONTROL_CS_STALL, + 0); + /* + * WaPipeControlBefore3DStateSamplePattern says we need 4 dwords for + * the PIPE_CONTROL followed by 12 dwords of 0x0, so 16 dwords in + * total. However, a PIPE_CONTROL is 6 dwords long, not 4, which is + * confusing. Since gen8_emit_pipe_control() already advances the + * batch by 6 dwords, we advance the other 10 here, completing a + * cacheline. It's not clear if the workaround requires this padding + * before other commands, or if it's just the regular padding we would + * already have for the workaround bb, so leave it here for now. + */ + for (i = 0; i < 10; i++) + *batch++ = MI_NOOP; + + /* Pad to end of cacheline */ + while ((unsigned long)batch % CACHELINE_BYTES) + *batch++ = MI_NOOP; + + return batch; +} + #define CTX_WA_BB_OBJ_SIZE (PAGE_SIZE) static int lrc_setup_wa_ctx(struct intel_engine_cs *engine) @@ -1416,12 +1509,14 @@ static int intel_init_workaround_bb(struct intel_engine_cs *engine) unsigned int i; int ret; - if (WARN_ON(engine->id != RCS || !engine->scratch)) + if (GEM_WARN_ON(engine->id != RCS)) return -EINVAL; switch (INTEL_GEN(engine->i915)) { case 10: - return 0; + wa_bb_fn[0] = gen10_init_indirectctx_bb; + wa_bb_fn[1] = NULL; + break; case 9: wa_bb_fn[0] = gen9_init_indirectctx_bb; wa_bb_fn[1] = NULL; @@ -1451,7 +1546,8 @@ static int intel_init_workaround_bb(struct intel_engine_cs *engine) */ for (i = 0; i < ARRAY_SIZE(wa_bb_fn); i++) { wa_bb[i]->offset = batch_ptr - batch; - if (WARN_ON(!IS_ALIGNED(wa_bb[i]->offset, CACHELINE_BYTES))) { + if (GEM_WARN_ON(!IS_ALIGNED(wa_bb[i]->offset, + CACHELINE_BYTES))) { ret = -EINVAL; break; } @@ -1477,47 +1573,48 @@ static u8 gtiir[] = { [VECS] = 3, }; -static int gen8_init_common_ring(struct intel_engine_cs *engine) +static void enable_execlists(struct intel_engine_cs *engine) { struct drm_i915_private *dev_priv = engine->i915; - struct intel_engine_execlists * const execlists = &engine->execlists; - int ret; - ret = intel_mocs_init_engine(engine); - if (ret) - return ret; + I915_WRITE(RING_HWSTAM(engine->mmio_base), 0xffffffff); - intel_engine_reset_breadcrumbs(engine); - intel_engine_init_hangcheck(engine); + /* + * Make sure we're not enabling the new 12-deep CSB + * FIFO as that requires a slightly updated handling + * in the ctx switch irq. Since we're currently only + * using only 2 elements of the enhanced execlists the + * deeper FIFO it's not needed and it's not worth adding + * more statements to the irq handler to support it. + */ + if (INTEL_GEN(dev_priv) >= 11) + I915_WRITE(RING_MODE_GEN7(engine), + _MASKED_BIT_DISABLE(GEN11_GFX_DISABLE_LEGACY_MODE)); + else + I915_WRITE(RING_MODE_GEN7(engine), + _MASKED_BIT_ENABLE(GFX_RUN_LIST_ENABLE)); - I915_WRITE(RING_HWSTAM(engine->mmio_base), 0xffffffff); - I915_WRITE(RING_MODE_GEN7(engine), - _MASKED_BIT_ENABLE(GFX_RUN_LIST_ENABLE)); I915_WRITE(RING_HWS_PGA(engine->mmio_base), engine->status_page.ggtt_offset); POSTING_READ(RING_HWS_PGA(engine->mmio_base)); - DRM_DEBUG_DRIVER("Execlists enabled for %s\n", engine->name); + /* Following the reset, we need to reload the CSB read/write pointers */ + engine->execlists.csb_head = -1; +} - GEM_BUG_ON(engine->id >= ARRAY_SIZE(gtiir)); +static int gen8_init_common_ring(struct intel_engine_cs *engine) +{ + struct intel_engine_execlists * const execlists = &engine->execlists; + int ret; - /* - * Clear any pending interrupt state. - * - * We do it twice out of paranoia that some of the IIR are double - * buffered, and if we only reset it once there may still be - * an interrupt pending. - */ - I915_WRITE(GEN8_GT_IIR(gtiir[engine->id]), - GT_CONTEXT_SWITCH_INTERRUPT << engine->irq_shift); - I915_WRITE(GEN8_GT_IIR(gtiir[engine->id]), - GT_CONTEXT_SWITCH_INTERRUPT << engine->irq_shift); - clear_bit(ENGINE_IRQ_EXECLIST, &engine->irq_posted); - execlists->csb_head = -1; - execlists->active = 0; + ret = intel_mocs_init_engine(engine); + if (ret) + return ret; + + intel_engine_reset_breadcrumbs(engine); + intel_engine_init_hangcheck(engine); - execlists->elsp = - dev_priv->regs + i915_mmio_reg_offset(RING_ELSP(engine)); + enable_execlists(engine); /* After a GPU reset, we may have requests to replay */ if (execlists->first) @@ -1559,8 +1656,33 @@ static int gen9_init_render_ring(struct intel_engine_cs *engine) return init_workarounds_ring(engine); } +static void reset_irq(struct intel_engine_cs *engine) +{ + struct drm_i915_private *dev_priv = engine->i915; + int i; + + GEM_BUG_ON(engine->id >= ARRAY_SIZE(gtiir)); + + /* + * Clear any pending interrupt state. + * + * We do it twice out of paranoia that some of the IIR are double + * buffered, and if we only reset it once there may still be + * an interrupt pending. + */ + for (i = 0; i < 2; i++) { + I915_WRITE(GEN8_GT_IIR(gtiir[engine->id]), + GT_CONTEXT_SWITCH_INTERRUPT << engine->irq_shift); + POSTING_READ(GEN8_GT_IIR(gtiir[engine->id])); + } + GEM_BUG_ON(I915_READ(GEN8_GT_IIR(gtiir[engine->id])) & + (GT_CONTEXT_SWITCH_INTERRUPT << engine->irq_shift)); + + clear_bit(ENGINE_IRQ_EXECLIST, &engine->irq_posted); +} + static void reset_common_ring(struct intel_engine_cs *engine, - struct drm_i915_gem_request *request) + struct i915_request *request) { struct intel_engine_execlists * const execlists = &engine->execlists; struct intel_context *ce; @@ -1568,7 +1690,11 @@ static void reset_common_ring(struct intel_engine_cs *engine, GEM_TRACE("%s seqno=%x\n", engine->name, request ? request->global_seqno : 0); - spin_lock_irqsave(&engine->timeline->lock, flags); + + /* See execlists_cancel_requests() for the irq/spinlock split. */ + local_irq_save(flags); + + reset_irq(engine); /* * Catch up with any missed context-switch interrupts. @@ -1582,11 +1708,17 @@ static void reset_common_ring(struct intel_engine_cs *engine, execlists_cancel_port_requests(execlists); /* Push back any incomplete requests for replay after the reset. */ + spin_lock(&engine->timeline->lock); __unwind_incomplete_requests(engine); + spin_unlock(&engine->timeline->lock); - spin_unlock_irqrestore(&engine->timeline->lock, flags); + /* Mark all CS interrupts as complete */ + execlists->active = 0; - /* If the request was innocent, we leave the request in the ELSP + local_irq_restore(flags); + + /* + * If the request was innocent, we leave the request in the ELSP * and will try to replay it on restarting. The context image may * have been corrupted by the reset, in which case we may have * to service a new GPU hang, but more likely we can continue on @@ -1599,7 +1731,8 @@ static void reset_common_ring(struct intel_engine_cs *engine, if (!request || request->fence.error != -EIO) return; - /* We want a simple context + ring to execute the breadcrumb update. + /* + * We want a simple context + ring to execute the breadcrumb update. * We cannot rely on the context being intact across the GPU hang, * so clear it and rebuild just what we need for the breadcrumb. * All pending requests for this context will be zapped, and any @@ -1622,15 +1755,15 @@ static void reset_common_ring(struct intel_engine_cs *engine, unwind_wa_tail(request); } -static int intel_logical_ring_emit_pdps(struct drm_i915_gem_request *req) +static int intel_logical_ring_emit_pdps(struct i915_request *rq) { - struct i915_hw_ppgtt *ppgtt = req->ctx->ppgtt; - struct intel_engine_cs *engine = req->engine; + struct i915_hw_ppgtt *ppgtt = rq->ctx->ppgtt; + struct intel_engine_cs *engine = rq->engine; const int num_lri_cmds = GEN8_3LVL_PDPES * 2; u32 *cs; int i; - cs = intel_ring_begin(req, num_lri_cmds * 2 + 2); + cs = intel_ring_begin(rq, num_lri_cmds * 2 + 2); if (IS_ERR(cs)) return PTR_ERR(cs); @@ -1645,12 +1778,12 @@ static int intel_logical_ring_emit_pdps(struct drm_i915_gem_request *req) } *cs++ = MI_NOOP; - intel_ring_advance(req, cs); + intel_ring_advance(rq, cs); return 0; } -static int gen8_emit_bb_start(struct drm_i915_gem_request *req, +static int gen8_emit_bb_start(struct i915_request *rq, u64 offset, u32 len, const unsigned int flags) { @@ -1663,18 +1796,18 @@ static int gen8_emit_bb_start(struct drm_i915_gem_request *req, * 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 (req->ctx->ppgtt && - (intel_engine_flag(req->engine) & req->ctx->ppgtt->pd_dirty_rings) && - !i915_vm_is_48bit(&req->ctx->ppgtt->base) && - !intel_vgpu_active(req->i915)) { - ret = intel_logical_ring_emit_pdps(req); + if (rq->ctx->ppgtt && + (intel_engine_flag(rq->engine) & rq->ctx->ppgtt->pd_dirty_rings) && + !i915_vm_is_48bit(&rq->ctx->ppgtt->base) && + !intel_vgpu_active(rq->i915)) { + ret = intel_logical_ring_emit_pdps(rq); if (ret) return ret; - req->ctx->ppgtt->pd_dirty_rings &= ~intel_engine_flag(req->engine); + rq->ctx->ppgtt->pd_dirty_rings &= ~intel_engine_flag(rq->engine); } - cs = intel_ring_begin(req, 4); + cs = intel_ring_begin(rq, 4); if (IS_ERR(cs)) return PTR_ERR(cs); @@ -1703,7 +1836,7 @@ static int gen8_emit_bb_start(struct drm_i915_gem_request *req, (flags & I915_DISPATCH_RS ? MI_BATCH_RESOURCE_STREAMER : 0); *cs++ = lower_32_bits(offset); *cs++ = upper_32_bits(offset); - intel_ring_advance(req, cs); + intel_ring_advance(rq, cs); return 0; } @@ -1722,7 +1855,7 @@ static void gen8_logical_ring_disable_irq(struct intel_engine_cs *engine) I915_WRITE_IMR(engine, ~engine->irq_keep_mask); } -static int gen8_emit_flush(struct drm_i915_gem_request *request, u32 mode) +static int gen8_emit_flush(struct i915_request *request, u32 mode) { u32 cmd, *cs; @@ -1754,7 +1887,7 @@ static int gen8_emit_flush(struct drm_i915_gem_request *request, u32 mode) return 0; } -static int gen8_emit_flush_render(struct drm_i915_gem_request *request, +static int gen8_emit_flush_render(struct i915_request *request, u32 mode) { struct intel_engine_cs *engine = request->engine; @@ -1829,7 +1962,7 @@ static int gen8_emit_flush_render(struct drm_i915_gem_request *request, * used as a workaround for not being allowed to do lite * restore with HEAD==TAIL (WaIdleLiteRestore). */ -static void gen8_emit_wa_tail(struct drm_i915_gem_request *request, u32 *cs) +static void gen8_emit_wa_tail(struct i915_request *request, u32 *cs) { /* Ensure there's always at least one preemption point per-request. */ *cs++ = MI_ARB_CHECK; @@ -1837,7 +1970,7 @@ static void gen8_emit_wa_tail(struct drm_i915_gem_request *request, u32 *cs) request->wa_tail = intel_ring_offset(request, cs); } -static void gen8_emit_breadcrumb(struct drm_i915_gem_request *request, u32 *cs) +static void gen8_emit_breadcrumb(struct i915_request *request, u32 *cs) { /* w/a: bit 5 needs to be zero for MI_FLUSH_DW address. */ BUILD_BUG_ON(I915_GEM_HWS_INDEX_ADDR & (1 << 5)); @@ -1853,8 +1986,7 @@ static void gen8_emit_breadcrumb(struct drm_i915_gem_request *request, u32 *cs) } static const int gen8_emit_breadcrumb_sz = 6 + WA_TAIL_DWORDS; -static void gen8_emit_breadcrumb_rcs(struct drm_i915_gem_request *request, - u32 *cs) +static void gen8_emit_breadcrumb_rcs(struct i915_request *request, u32 *cs) { /* We're using qword write, seqno should be aligned to 8 bytes. */ BUILD_BUG_ON(I915_GEM_HWS_INDEX & 1); @@ -1870,15 +2002,15 @@ static void gen8_emit_breadcrumb_rcs(struct drm_i915_gem_request *request, } static const int gen8_emit_breadcrumb_rcs_sz = 8 + WA_TAIL_DWORDS; -static int gen8_init_rcs_context(struct drm_i915_gem_request *req) +static int gen8_init_rcs_context(struct i915_request *rq) { int ret; - ret = intel_ring_workarounds_emit(req); + ret = intel_ring_workarounds_emit(rq); if (ret) return ret; - ret = intel_rcs_context_init_mocs(req); + ret = intel_rcs_context_init_mocs(rq); /* * Failing to program the MOCS is non-fatal.The system will not * run at peak performance. So generate an error and carry on. @@ -1886,7 +2018,7 @@ static int gen8_init_rcs_context(struct drm_i915_gem_request *req) if (ret) DRM_ERROR("MOCS failed to program: expect performance issues.\n"); - return i915_gem_render_state_emit(req); + return i915_gem_render_state_emit(rq); } /** @@ -1917,6 +2049,7 @@ void intel_logical_ring_cleanup(struct intel_engine_cs *engine) intel_engine_cleanup_common(engine); lrc_destroy_wa_ctx(engine); + engine->i915 = NULL; dev_priv->engine[engine->id] = NULL; kfree(engine); @@ -1933,6 +2066,12 @@ static void execlists_set_default_submission(struct intel_engine_cs *engine) engine->unpark = NULL; engine->flags |= I915_ENGINE_SUPPORTS_STATS; + + engine->i915->caps.scheduler = + I915_SCHEDULER_CAP_ENABLED | + I915_SCHEDULER_CAP_PRIORITY; + if (engine->i915->preempt_context) + engine->i915->caps.scheduler |= I915_SCHEDULER_CAP_PREEMPTION; } static void @@ -1953,8 +2092,17 @@ logical_ring_default_vfuncs(struct intel_engine_cs *engine) engine->set_default_submission = execlists_set_default_submission; - engine->irq_enable = gen8_logical_ring_enable_irq; - engine->irq_disable = gen8_logical_ring_disable_irq; + if (INTEL_GEN(engine->i915) < 11) { + engine->irq_enable = gen8_logical_ring_enable_irq; + engine->irq_disable = gen8_logical_ring_disable_irq; + } else { + /* + * TODO: On Gen11 interrupt masks need to be clear + * to allow C6 entry. Keep interrupts enabled at + * and take the hit of generating extra interrupts + * until a more refined solution exists. + */ + } engine->emit_bb_start = gen8_emit_bb_start; } @@ -2006,6 +2154,21 @@ static int logical_ring_init(struct intel_engine_cs *engine) if (ret) goto error; + if (HAS_LOGICAL_RING_ELSQ(engine->i915)) { + engine->execlists.submit_reg = engine->i915->regs + + i915_mmio_reg_offset(RING_EXECLIST_SQ_CONTENTS(engine)); + engine->execlists.ctrl_reg = engine->i915->regs + + i915_mmio_reg_offset(RING_EXECLIST_CONTROL(engine)); + } else { + engine->execlists.submit_reg = engine->i915->regs + + i915_mmio_reg_offset(RING_ELSP(engine)); + } + + engine->execlists.preempt_complete_status = ~0u; + if (engine->i915->preempt_context) + engine->execlists.preempt_complete_status = + upper_32_bits(engine->i915->preempt_context->engine[engine->id].lrc_desc); + return 0; error: @@ -2085,7 +2248,7 @@ make_rpcs(struct drm_i915_private *dev_priv) if (INTEL_INFO(dev_priv)->sseu.has_subslice_pg) { rpcs |= GEN8_RPCS_SS_CNT_ENABLE; - rpcs |= hweight8(INTEL_INFO(dev_priv)->sseu.subslice_mask) << + rpcs |= hweight8(INTEL_INFO(dev_priv)->sseu.subslice_mask[0]) << GEN8_RPCS_SS_CNT_SHIFT; rpcs |= GEN8_RPCS_ENABLE; } @@ -2109,6 +2272,10 @@ static u32 intel_lr_indirect_ctx_offset(struct intel_engine_cs *engine) default: MISSING_CASE(INTEL_GEN(engine->i915)); /* fall through */ + case 11: + indirect_ctx_offset = + GEN11_CTX_RCS_INDIRECT_CTX_OFFSET_DEFAULT; + break; case 10: indirect_ctx_offset = GEN10_CTX_RCS_INDIRECT_CTX_OFFSET_DEFAULT; @@ -2147,6 +2314,8 @@ static void execlists_init_reg_state(u32 *regs, MI_LRI_FORCE_POSTED; CTX_REG(regs, CTX_CONTEXT_CONTROL, RING_CONTEXT_CONTROL(engine), + _MASKED_BIT_DISABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT | + CTX_CTRL_ENGINE_CTX_SAVE_INHIBIT) | _MASKED_BIT_ENABLE(CTX_CTRL_INHIBIT_SYN_CTX_SWITCH | (HAS_RESOURCE_STREAMER(dev_priv) ? CTX_CTRL_RS_CTX_ENABLE : 0))); @@ -2266,6 +2435,10 @@ populate_lr_context(struct i915_gem_context *ctx, if (!engine->default_state) regs[CTX_CONTEXT_CONTROL + 1] |= _MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT); + if (ctx == ctx->i915->preempt_context && INTEL_GEN(engine->i915) < 11) + regs[CTX_CONTEXT_CONTROL + 1] |= + _MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT | + CTX_CTRL_ENGINE_CTX_SAVE_INHIBIT); i915_gem_object_unpin_map(ctx_obj); @@ -2282,7 +2455,8 @@ static int execlists_context_deferred_alloc(struct i915_gem_context *ctx, struct intel_ring *ring; int ret; - WARN_ON(ce->state); + if (ce->state) + return 0; context_size = round_up(engine->context_size, I915_GTT_PAGE_SIZE); |
