diff --git a/drivers/gpu/drm/i915/gvt/scheduler.c b/drivers/gpu/drm/i915/gvt/scheduler.c index f6ded475bb2c..42cc61230ca7 100644 --- a/drivers/gpu/drm/i915/gvt/scheduler.c +++ b/drivers/gpu/drm/i915/gvt/scheduler.c @@ -723,8 +723,6 @@ int intel_vgpu_init_gvt_context(struct intel_vgpu *vgpu) if (IS_ERR(vgpu->shadow_ctx)) return PTR_ERR(vgpu->shadow_ctx); - vgpu->shadow_ctx->engine[RCS].initialised = true; - bitmap_zero(vgpu->shadow_ctx_desc_updated, I915_NUM_ENGINES); return 0; diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c index d89321f0468c..533ba096b9a6 100644 --- a/drivers/gpu/drm/i915/i915_debugfs.c +++ b/drivers/gpu/drm/i915/i915_debugfs.c @@ -1974,7 +1974,6 @@ static int i915_context_status(struct seq_file *m, void *unused) struct intel_context *ce = &ctx->engine[engine->id]; seq_printf(m, "%s: ", engine->name); - seq_putc(m, ce->initialised ? 'I' : 'i'); if (ce->state) describe_obj(m, ce->state->obj); if (ce->ring) diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c index 1b440f2b90a5..d97fe9c9439a 100644 --- a/drivers/gpu/drm/i915/i915_drv.c +++ b/drivers/gpu/drm/i915/i915_drv.c @@ -406,6 +406,9 @@ static int i915_getparam(struct drm_device *dev, void *data, */ value = 1; break; + case I915_PARAM_HAS_CONTEXT_ISOLATION: + value = intel_engines_has_context_isolation(dev_priv); + break; case I915_PARAM_SLICE_MASK: value = INTEL_INFO(dev_priv)->sseu.slice_mask; if (!value) diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index ed335612be25..4bf118304086 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -4972,6 +4972,120 @@ bool intel_sanitize_semaphores(struct drm_i915_private *dev_priv, int value) return true; } +static int __intel_engines_record_defaults(struct drm_i915_private *i915) +{ + struct i915_gem_context *ctx; + struct intel_engine_cs *engine; + enum intel_engine_id id; + int err; + + /* + * As we reset the gpu during very early sanitisation, the current + * register state on the GPU should reflect its defaults values. + * We load a context onto the hw (with restore-inhibit), then switch + * over to a second context to save that default register state. We + * can then prime every new context with that state so they all start + * from the same default HW values. + */ + + ctx = i915_gem_context_create_kernel(i915, 0); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + for_each_engine(engine, i915, id) { + struct drm_i915_gem_request *rq; + + rq = i915_gem_request_alloc(engine, ctx); + if (IS_ERR(rq)) { + err = PTR_ERR(rq); + goto out_ctx; + } + + err = i915_switch_context(rq); + if (engine->init_context) + err = engine->init_context(rq); + + __i915_add_request(rq, true); + if (err) + goto err_active; + } + + err = i915_gem_switch_to_kernel_context(i915); + if (err) + goto err_active; + + err = i915_gem_wait_for_idle(i915, I915_WAIT_LOCKED); + if (err) + goto err_active; + + assert_kernel_context_is_current(i915); + + for_each_engine(engine, i915, id) { + struct i915_vma *state; + + state = ctx->engine[id].state; + if (!state) + continue; + + /* + * As we will hold a reference to the logical state, it will + * not be torn down with the context, and importantly the + * object will hold onto its vma (making it possible for a + * stray GTT write to corrupt our defaults). Unmap the vma + * from the GTT to prevent such accidents and reclaim the + * space. + */ + err = i915_vma_unbind(state); + if (err) + goto err_active; + + err = i915_gem_object_set_to_cpu_domain(state->obj, false); + if (err) + goto err_active; + + engine->default_state = i915_gem_object_get(state->obj); + } + + if (IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM)) { + unsigned int found = intel_engines_has_context_isolation(i915); + + /* + * Make sure that classes with multiple engine instances all + * share the same basic configuration. + */ + for_each_engine(engine, i915, id) { + unsigned int bit = BIT(engine->uabi_class); + unsigned int expected = engine->default_state ? bit : 0; + + if ((found & bit) != expected) { + DRM_ERROR("mismatching default context state for class %d on engine %s\n", + engine->uabi_class, engine->name); + } + } + } + +out_ctx: + i915_gem_context_set_closed(ctx); + i915_gem_context_put(ctx); + return err; + +err_active: + /* + * If we have to abandon now, we expect the engines to be idle + * and ready to be torn-down. First try to flush any remaining + * request, ensure we are pointing at the kernel context and + * then remove it. + */ + if (WARN_ON(i915_gem_switch_to_kernel_context(i915))) + goto out_ctx; + + if (WARN_ON(i915_gem_wait_for_idle(i915, I915_WAIT_LOCKED))) + goto out_ctx; + + i915_gem_contexts_lost(i915); + goto out_ctx; +} + int i915_gem_init(struct drm_i915_private *dev_priv) { int ret; @@ -5038,6 +5152,7 @@ int i915_gem_init(struct drm_i915_private *dev_priv) */ intel_init_clock_gating(dev_priv); + ret = __intel_engines_record_defaults(dev_priv); out_unlock: if (ret == -EIO) { /* Allow engine initialisation to fail by marking the GPU as diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c index c05c3d7d21a5..2db040695035 100644 --- a/drivers/gpu/drm/i915/i915_gem_context.c +++ b/drivers/gpu/drm/i915/i915_gem_context.c @@ -418,8 +418,8 @@ out: return ctx; } -static struct i915_gem_context * -create_kernel_context(struct drm_i915_private *i915, int prio) +struct i915_gem_context * +i915_gem_context_create_kernel(struct drm_i915_private *i915, int prio) { struct i915_gem_context *ctx; @@ -473,7 +473,7 @@ int i915_gem_contexts_init(struct drm_i915_private *dev_priv) ida_init(&dev_priv->contexts.hw_ida); /* lowest priority; idle task */ - ctx = create_kernel_context(dev_priv, I915_PRIORITY_MIN); + ctx = i915_gem_context_create_kernel(dev_priv, I915_PRIORITY_MIN); if (IS_ERR(ctx)) { DRM_ERROR("Failed to create default global context\n"); err = PTR_ERR(ctx); @@ -487,7 +487,7 @@ int i915_gem_contexts_init(struct drm_i915_private *dev_priv) dev_priv->kernel_context = ctx; /* highest priority; preempting task */ - ctx = create_kernel_context(dev_priv, INT_MAX); + ctx = i915_gem_context_create_kernel(dev_priv, INT_MAX); if (IS_ERR(ctx)) { DRM_ERROR("Failed to create default preempt context\n"); err = PTR_ERR(ctx); @@ -522,28 +522,6 @@ void i915_gem_contexts_lost(struct drm_i915_private *dev_priv) engine->context_unpin(engine, engine->last_retired_context); engine->last_retired_context = NULL; } - - /* Force the GPU state to be restored on enabling */ - if (!i915_modparams.enable_execlists) { - struct i915_gem_context *ctx; - - list_for_each_entry(ctx, &dev_priv->contexts.list, link) { - if (!i915_gem_context_is_default(ctx)) - continue; - - for_each_engine(engine, dev_priv, id) - ctx->engine[engine->id].initialised = false; - - ctx->remap_slice = ALL_L3_SLICES(dev_priv); - } - - for_each_engine(engine, dev_priv, id) { - struct intel_context *kce = - &dev_priv->kernel_context->engine[engine->id]; - - kce->initialised = true; - } - } } void i915_gem_contexts_fini(struct drm_i915_private *i915) @@ -718,9 +696,6 @@ static inline bool skip_rcs_switch(struct i915_hw_ppgtt *ppgtt, if (to->remap_slice) return false; - if (!to->engine[RCS].initialised) - return false; - if (ppgtt && (intel_engine_flag(engine) & ppgtt->pd_dirty_rings)) return false; @@ -795,11 +770,14 @@ static int do_rcs_switch(struct drm_i915_gem_request *req) return ret; } - if (!to->engine[RCS].initialised || i915_gem_context_is_default(to)) - /* NB: If we inhibit the restore, the context is not allowed to - * die because future work may end up depending on valid address - * space. This means we must enforce that a page table load - * occur when this occurs. */ + if (i915_gem_context_is_kernel(to)) + /* + * The kernel context(s) is treated as pure scratch and is not + * expected to retain any state (as we sacrifice it during + * suspend and on resume it may be corrupted). This is ok, + * as nothing actually executes using the kernel context; it + * is purely used for flushing user contexts. + */ hw_flags = MI_RESTORE_INHIBIT; else if (ppgtt && intel_engine_flag(engine) & ppgtt->pd_dirty_rings) hw_flags = MI_FORCE_RESTORE; @@ -843,15 +821,6 @@ static int do_rcs_switch(struct drm_i915_gem_request *req) to->remap_slice &= ~(1<engine[RCS].initialised) { - if (engine->init_context) { - ret = engine->init_context(req); - if (ret) - return ret; - } - to->engine[RCS].initialised = true; - } - return 0; } diff --git a/drivers/gpu/drm/i915/i915_gem_context.h b/drivers/gpu/drm/i915/i915_gem_context.h index 44688e22a5c2..4bfb72f8e1cb 100644 --- a/drivers/gpu/drm/i915/i915_gem_context.h +++ b/drivers/gpu/drm/i915/i915_gem_context.h @@ -157,7 +157,6 @@ struct i915_gem_context { u32 *lrc_reg_state; u64 lrc_desc; int pin_count; - bool initialised; } engine[I915_NUM_ENGINES]; /** ring_size: size for allocating the per-engine ring buffer */ @@ -292,6 +291,9 @@ int i915_gem_context_setparam_ioctl(struct drm_device *dev, void *data, int i915_gem_context_reset_stats_ioctl(struct drm_device *dev, void *data, struct drm_file *file); +struct i915_gem_context * +i915_gem_context_create_kernel(struct drm_i915_private *i915, int prio); + static inline struct i915_gem_context * i915_gem_context_get(struct i915_gem_context *ctx) { diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c index 6d0b38189a7d..868c07a693b5 100644 --- a/drivers/gpu/drm/i915/intel_engine_cs.c +++ b/drivers/gpu/drm/i915/intel_engine_cs.c @@ -687,6 +687,9 @@ void intel_engine_cleanup_common(struct intel_engine_cs *engine) intel_engine_cleanup_cmd_parser(engine); i915_gem_batch_pool_fini(&engine->batch_pool); + if (engine->default_state) + i915_gem_object_put(engine->default_state); + if (HAS_LOGICAL_RING_PREEMPTION(engine->i915)) engine->context_unpin(engine, engine->i915->preempt_context); engine->context_unpin(engine, engine->i915->kernel_context); @@ -1705,6 +1708,20 @@ bool intel_engine_can_store_dword(struct intel_engine_cs *engine) } } +unsigned int intel_engines_has_context_isolation(struct drm_i915_private *i915) +{ + struct intel_engine_cs *engine; + enum intel_engine_id id; + unsigned int which; + + which = 0; + for_each_engine(engine, i915, id) + if (engine->default_state) + which |= BIT(engine->uabi_class); + + return which; +} + static void print_request(struct drm_printer *m, struct drm_i915_gem_request *rq, const char *prefix) diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c index 1baaab35905c..0c93f27f36ee 100644 --- a/drivers/gpu/drm/i915/intel_lrc.c +++ b/drivers/gpu/drm/i915/intel_lrc.c @@ -1180,7 +1180,6 @@ static int execlists_request_alloc(struct drm_i915_gem_request *request) struct intel_engine_cs *engine = request->engine; struct intel_context *ce = &request->ctx->engine[engine->id]; u32 *cs; - int ret; GEM_BUG_ON(!ce->pin_count); @@ -1194,14 +1193,6 @@ static int execlists_request_alloc(struct drm_i915_gem_request *request) if (IS_ERR(cs)) return PTR_ERR(cs); - if (!ce->initialised) { - ret = engine->init_context(request); - if (ret) - return ret; - - ce->initialised = true; - } - /* Note that after this point, we have committed to using * this request as it is being used to both track the * state of engine initialisation and liveness of the @@ -2133,7 +2124,6 @@ static void execlists_init_reg_state(u32 *regs, CTX_REG(regs, CTX_CONTEXT_CONTROL, RING_CONTEXT_CONTROL(engine), _MASKED_BIT_ENABLE(CTX_CTRL_INHIBIT_SYN_CTX_SWITCH | - CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT | (HAS_RESOURCE_STREAMER(dev_priv) ? CTX_CTRL_RS_CTX_ENABLE : 0))); CTX_REG(regs, CTX_RING_HEAD, RING_HEAD(base), 0); @@ -2210,6 +2200,7 @@ populate_lr_context(struct i915_gem_context *ctx, struct intel_ring *ring) { void *vaddr; + u32 *regs; int ret; ret = i915_gem_object_set_to_cpu_domain(ctx_obj, true); @@ -2226,11 +2217,31 @@ populate_lr_context(struct i915_gem_context *ctx, } ctx_obj->mm.dirty = true; + if (engine->default_state) { + /* + * We only want to copy over the template context state; + * skipping over the headers reserved for GuC communication, + * leaving those as zero. + */ + const unsigned long start = LRC_HEADER_PAGES * PAGE_SIZE; + void *defaults; + + defaults = i915_gem_object_pin_map(engine->default_state, + I915_MAP_WB); + if (IS_ERR(defaults)) + return PTR_ERR(defaults); + + memcpy(vaddr + start, defaults + start, engine->context_size); + i915_gem_object_unpin_map(engine->default_state); + } + /* The second page of the context object contains some fields which must * be set up prior to the first execution. */ - - execlists_init_reg_state(vaddr + LRC_STATE_PN * PAGE_SIZE, - ctx, engine, ring); + regs = vaddr + LRC_STATE_PN * PAGE_SIZE; + execlists_init_reg_state(regs, ctx, engine, ring); + if (!engine->default_state) + regs[CTX_CONTEXT_CONTROL + 1] |= + _MASKED_BIT_ENABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT); i915_gem_object_unpin_map(ctx_obj); @@ -2283,7 +2294,6 @@ static int execlists_context_deferred_alloc(struct i915_gem_context *ctx, ce->ring = ring; ce->state = vma; - ce->initialised |= engine->init_context == NULL; return 0; diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c index 7e2a671882fb..464dc58af27b 100644 --- a/drivers/gpu/drm/i915/intel_ringbuffer.c +++ b/drivers/gpu/drm/i915/intel_ringbuffer.c @@ -1384,11 +1384,34 @@ alloc_context_vma(struct intel_engine_cs *engine) struct drm_i915_private *i915 = engine->i915; struct drm_i915_gem_object *obj; struct i915_vma *vma; + int err; obj = i915_gem_object_create(i915, engine->context_size); if (IS_ERR(obj)) return ERR_CAST(obj); + if (engine->default_state) { + void *defaults, *vaddr; + + vaddr = i915_gem_object_pin_map(obj, I915_MAP_WB); + if (IS_ERR(vaddr)) { + err = PTR_ERR(vaddr); + goto err_obj; + } + + defaults = i915_gem_object_pin_map(engine->default_state, + I915_MAP_WB); + if (IS_ERR(defaults)) { + err = PTR_ERR(defaults); + goto err_map; + } + + memcpy(vaddr, defaults, engine->context_size); + + i915_gem_object_unpin_map(engine->default_state); + i915_gem_object_unpin_map(obj); + } + /* * Try to make the context utilize L3 as well as LLC. * @@ -1410,10 +1433,18 @@ alloc_context_vma(struct intel_engine_cs *engine) } vma = i915_vma_instance(obj, &i915->ggtt.base, NULL); - if (IS_ERR(vma)) - i915_gem_object_put(obj); + if (IS_ERR(vma)) { + err = PTR_ERR(vma); + goto err_obj; + } return vma; + +err_map: + i915_gem_object_unpin_map(obj); +err_obj: + i915_gem_object_put(obj); + return ERR_PTR(err); } static struct intel_ring * @@ -1449,16 +1480,6 @@ intel_ring_context_pin(struct intel_engine_cs *engine, ce->state->obj->pin_global++; } - /* The kernel context is only used as a placeholder for flushing the - * active context. It is never used for submitting user rendering and - * as such never requires the golden render context, and so we can skip - * emitting it when we switch to the kernel context. This is required - * as during eviction we cannot allocate and pin the renderstate in - * order to initialise the context. - */ - if (i915_gem_context_is_kernel(ctx)) - ce->initialised = true; - i915_gem_context_get(ctx); out: diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h index f3dbfe7ae6e4..337222859166 100644 --- a/drivers/gpu/drm/i915/intel_ringbuffer.h +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h @@ -306,6 +306,7 @@ struct intel_engine_cs { struct intel_ring *buffer; struct intel_timeline *timeline; + struct drm_i915_gem_object *default_state; struct intel_render_state *render_state; atomic_t irq_count; @@ -932,6 +933,7 @@ void intel_engines_park(struct drm_i915_private *i915); void intel_engines_unpark(struct drm_i915_private *i915); void intel_engines_reset_default_submission(struct drm_i915_private *i915); +unsigned int intel_engines_has_context_isolation(struct drm_i915_private *i915); bool intel_engine_can_store_dword(struct intel_engine_cs *engine); diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h index 1f7dfb22a7c2..6c02ced663f8 100644 --- a/include/uapi/drm/i915_drm.h +++ b/include/uapi/drm/i915_drm.h @@ -466,6 +466,21 @@ typedef struct drm_i915_irq_wait { */ #define I915_PARAM_HAS_EXEC_FENCE_ARRAY 49 +/* + * Query whether every context (both per-file default and user created) is + * isolated (insofar as HW supports). If this parameter is not true, then + * freshly created contexts may inherit values from an existing context, + * rather than default HW values. If true, it also ensures (insofar as HW + * supports) that all state set by this context will not leak to any other + * context. + * + * As not every engine across every gen support contexts, the returned + * value reports the support of context isolation for individual engines by + * returning a bitmask of each engine class set to true if that class supports + * isolation. + */ +#define I915_PARAM_HAS_CONTEXT_ISOLATION 50 + typedef struct drm_i915_getparam { __s32 param; /*