Merge tag 'drm-intel-gt-next-2021-10-21' of git://anongit.freedesktop.org/drm/drm-intel into drm-next

UAPI Changes:

- Expose multi-LRC submission interface

  Similar to the bonded submission interface but simplified.
  Comes with GuC only implementation for now. See kerneldoc
  for more details.

  Userspace changes: https://github.com/intel/media-driver/pull/1252

- Expose logical engine instance to user

  Needed by the multi-LRC submission interface for GuC

  Userspace changes: https://github.com/intel/media-driver/pull/1252

Driver Changes:

- Fix blank screen booting crashes when CONFIG_CC_OPTIMIZE_FOR_SIZE=y (Hugh)
- Add support for multi-LRC submission in the GuC backend (Matt B)
- Add extra cache flushing before making pages userspace visible (Matt A, Thomas)
- Mark internal GPU object pages dirty so they will be flushed properly (Matt A)

- Move remaining debugfs interfaces i915_wedged/i915_forcewake_user into gt (Andi)
- Replace the unconditional clflushes with drm_clflush_virt_range() (Ville)
- Remove IS_ACTIVE macro completely (Lucas)
- Improve kerneldocs for cache_dirty (Matt A)

- Add missing includes (Lucas)
- Selftest improvements (Matt R, Ran, Matt A)

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/YXFmLKoq8Fg9JxSd@jlahtine-mobl.ger.corp.intel.com

49 files changed, 3091 insertions, 639 deletions

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_busy.c b/drivers/gpu/drm/i915/gem/i915_gem_busy.c
index 6234e17259c1..7358bebef15c 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_busy.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_busy.c
@@ -4,6 +4,8 @@
  * Copyright © 2014-2016 Intel Corporation
  */
 
+#include <linux/dma-fence-array.h>
+
 #include "gt/intel_engine.h"
 
 #include "i915_gem_ioctls.h"
@@ -36,7 +38,7 @@ static __always_inline u32 __busy_write_id(u16 id)
 }
 
 static __always_inline unsigned int
-__busy_set_if_active(const struct dma_fence *fence, u32 (*flag)(u16 id))
+__busy_set_if_active(struct dma_fence *fence, u32 (*flag)(u16 id))
 {
 	const struct i915_request *rq;
 
@@ -46,29 +48,60 @@ __busy_set_if_active(const struct dma_fence *fence, u32 (*flag)(u16 id))
 	 * to eventually flush us, but to minimise latency just ask the
 	 * hardware.
 	 *
-	 * Note we only report on the status of native fences.
+	 * Note we only report on the status of native fences and we currently
+	 * have two native fences:
+	 *
+	 * 1. A composite fence (dma_fence_array) constructed of i915 requests
+	 * created during a parallel submission. In this case we deconstruct the
+	 * composite fence into individual i915 requests and check the status of
+	 * each request.
+	 *
+	 * 2. A single i915 request.
 	 */
-	if (!dma_fence_is_i915(fence))
+	if (dma_fence_is_array(fence)) {
+		struct dma_fence_array *array = to_dma_fence_array(fence);
+		struct dma_fence **child = array->fences;
+		unsigned int nchild = array->num_fences;
+
+		do {
+			struct dma_fence *current_fence = *child++;
+
+			/* Not an i915 fence, can't be busy per above */
+			if (!dma_fence_is_i915(current_fence) ||
+			    !test_bit(I915_FENCE_FLAG_COMPOSITE,
+				      &current_fence->flags)) {
+				return 0;
+			}
+
+			rq = to_request(current_fence);
+			if (!i915_request_completed(rq))
+				return flag(rq->engine->uabi_class);
+		} while (--nchild);
+
+		/* All requests in array complete, not busy */
 		return 0;
+	} else {
+		if (!dma_fence_is_i915(fence))
+			return 0;
 
-	/* opencode to_request() in order to avoid const warnings */
-	rq = container_of(fence, const struct i915_request, fence);
-	if (i915_request_completed(rq))
-		return 0;
+		rq = to_request(fence);
+		if (i915_request_completed(rq))
+			return 0;
 
-	/* Beware type-expansion follies! */
-	BUILD_BUG_ON(!typecheck(u16, rq->engine->uabi_class));
-	return flag(rq->engine->uabi_class);
+		/* Beware type-expansion follies! */
+		BUILD_BUG_ON(!typecheck(u16, rq->engine->uabi_class));
+		return flag(rq->engine->uabi_class);
+	}
 }
 
 static __always_inline unsigned int
-busy_check_reader(const struct dma_fence *fence)
+busy_check_reader(struct dma_fence *fence)
 {
 	return __busy_set_if_active(fence, __busy_read_flag);
 }
 
 static __always_inline unsigned int
-busy_check_writer(const struct dma_fence *fence)
+busy_check_writer(struct dma_fence *fence)
 {
 	if (!fence)
 		return 0;
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_context.c b/drivers/gpu/drm/i915/gem/i915_gem_context.c
index d225d3dd0b40..fb33d0322960 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_context.c
@@ -556,9 +556,147 @@ set_proto_ctx_engines_bond(struct i915_user_extension __user *base, void *data)
 	return 0;
 }
 
+static int
+set_proto_ctx_engines_parallel_submit(struct i915_user_extension __user *base,
+				      void *data)
+{
+	struct i915_context_engines_parallel_submit __user *ext =
+		container_of_user(base, typeof(*ext), base);
+	const struct set_proto_ctx_engines *set = data;
+	struct drm_i915_private *i915 = set->i915;
+	u64 flags;
+	int err = 0, n, i, j;
+	u16 slot, width, num_siblings;
+	struct intel_engine_cs **siblings = NULL;
+	intel_engine_mask_t prev_mask;
+
+	/* FIXME: This is NIY for execlists */
+	if (!(intel_uc_uses_guc_submission(&i915->gt.uc)))
+		return -ENODEV;
+
+	if (get_user(slot, &ext->engine_index))
+		return -EFAULT;
+
+	if (get_user(width, &ext->width))
+		return -EFAULT;
+
+	if (get_user(num_siblings, &ext->num_siblings))
+		return -EFAULT;
+
+	if (slot >= set->num_engines) {
+		drm_dbg(&i915->drm, "Invalid placement value, %d >= %d\n",
+			slot, set->num_engines);
+		return -EINVAL;
+	}
+
+	if (set->engines[slot].type != I915_GEM_ENGINE_TYPE_INVALID) {
+		drm_dbg(&i915->drm,
+			"Invalid placement[%d], already occupied\n", slot);
+		return -EINVAL;
+	}
+
+	if (get_user(flags, &ext->flags))
+		return -EFAULT;
+
+	if (flags) {
+		drm_dbg(&i915->drm, "Unknown flags 0x%02llx", flags);
+		return -EINVAL;
+	}
+
+	for (n = 0; n < ARRAY_SIZE(ext->mbz64); n++) {
+		err = check_user_mbz(&ext->mbz64[n]);
+		if (err)
+			return err;
+	}
+
+	if (width < 2) {
+		drm_dbg(&i915->drm, "Width (%d) < 2\n", width);
+		return -EINVAL;
+	}
+
+	if (num_siblings < 1) {
+		drm_dbg(&i915->drm, "Number siblings (%d) < 1\n",
+			num_siblings);
+		return -EINVAL;
+	}
+
+	siblings = kmalloc_array(num_siblings * width,
+				 sizeof(*siblings),
+				 GFP_KERNEL);
+	if (!siblings)
+		return -ENOMEM;
+
+	/* Create contexts / engines */
+	for (i = 0; i < width; ++i) {
+		intel_engine_mask_t current_mask = 0;
+		struct i915_engine_class_instance prev_engine;
+
+		for (j = 0; j < num_siblings; ++j) {
+			struct i915_engine_class_instance ci;
+
+			n = i * num_siblings + j;
+			if (copy_from_user(&ci, &ext->engines[n], sizeof(ci))) {
+				err = -EFAULT;
+				goto out_err;
+			}
+
+			siblings[n] =
+				intel_engine_lookup_user(i915, ci.engine_class,
+							 ci.engine_instance);
+			if (!siblings[n]) {
+				drm_dbg(&i915->drm,
+					"Invalid sibling[%d]: { class:%d, inst:%d }\n",
+					n, ci.engine_class, ci.engine_instance);
+				err = -EINVAL;
+				goto out_err;
+			}
+
+			if (n) {
+				if (prev_engine.engine_class !=
+				    ci.engine_class) {
+					drm_dbg(&i915->drm,
+						"Mismatched class %d, %d\n",
+						prev_engine.engine_class,
+						ci.engine_class);
+					err = -EINVAL;
+					goto out_err;
+				}
+			}
+
+			prev_engine = ci;
+			current_mask |= siblings[n]->logical_mask;
+		}
+
+		if (i > 0) {
+			if (current_mask != prev_mask << 1) {
+				drm_dbg(&i915->drm,
+					"Non contiguous logical mask 0x%x, 0x%x\n",
+					prev_mask, current_mask);
+				err = -EINVAL;
+				goto out_err;
+			}
+		}
+		prev_mask = current_mask;
+	}
+
+	set->engines[slot].type = I915_GEM_ENGINE_TYPE_PARALLEL;
+	set->engines[slot].num_siblings = num_siblings;
+	set->engines[slot].width = width;
+	set->engines[slot].siblings = siblings;
+
+	return 0;
+
+out_err:
+	kfree(siblings);
+
+	return err;
+}
+
 static const i915_user_extension_fn set_proto_ctx_engines_extensions[] = {
 	[I915_CONTEXT_ENGINES_EXT_LOAD_BALANCE] = set_proto_ctx_engines_balance,
 	[I915_CONTEXT_ENGINES_EXT_BOND] = set_proto_ctx_engines_bond,
+	[I915_CONTEXT_ENGINES_EXT_PARALLEL_SUBMIT] =
+		set_proto_ctx_engines_parallel_submit,
 };
 
 static int set_proto_ctx_engines(struct drm_i915_file_private *fpriv,
@@ -794,6 +932,7 @@ static int intel_context_set_gem(struct intel_context *ce,
 	GEM_BUG_ON(rcu_access_pointer(ce->gem_context));
 	RCU_INIT_POINTER(ce->gem_context, ctx);
 
+	GEM_BUG_ON(intel_context_is_pinned(ce));
 	ce->ring_size = SZ_16K;
 
 	i915_vm_put(ce->vm);
@@ -818,6 +957,25 @@ static int intel_context_set_gem(struct intel_context *ce,
 	return ret;
 }
 
+static void __unpin_engines(struct i915_gem_engines *e, unsigned int count)
+{
+	while (count--) {
+		struct intel_context *ce = e->engines[count], *child;
+
+		if (!ce || !test_bit(CONTEXT_PERMA_PIN, &ce->flags))
+			continue;
+
+		for_each_child(ce, child)
+			intel_context_unpin(child);
+		intel_context_unpin(ce);
+	}
+}
+
+static void unpin_engines(struct i915_gem_engines *e)
+{
+	__unpin_engines(e, e->num_engines);
+}
+
 static void __free_engines(struct i915_gem_engines *e, unsigned int count)
 {
 	while (count--) {
@@ -933,6 +1091,40 @@ free_engines:
 	return err;
 }
 
+static int perma_pin_contexts(struct intel_context *ce)
+{
+	struct intel_context *child;
+	int i = 0, j = 0, ret;
+
+	GEM_BUG_ON(!intel_context_is_parent(ce));
+
+	ret = intel_context_pin(ce);
+	if (unlikely(ret))
+		return ret;
+
+	for_each_child(ce, child) {
+		ret = intel_context_pin(child);
+		if (unlikely(ret))
+			goto unwind;
+		++i;
+	}
+
+	set_bit(CONTEXT_PERMA_PIN, &ce->flags);
+
+	return 0;
+
+unwind:
+	intel_context_unpin(ce);
+	for_each_child(ce, child) {
+		if (j++ < i)
+			intel_context_unpin(child);
+		else
+			break;
+	}
+
+	return ret;
+}
+
 static struct i915_gem_engines *user_engines(struct i915_gem_context *ctx,
 					     unsigned int num_engines,
 					     struct i915_gem_proto_engine *pe)
@@ -946,7 +1138,7 @@ static struct i915_gem_engines *user_engines(struct i915_gem_context *ctx,
 	e->num_engines = num_engines;
 
 	for (n = 0; n < num_engines; n++) {
-		struct intel_context *ce;
+		struct intel_context *ce, *child;
 		int ret;
 
 		switch (pe[n].type) {
@@ -956,7 +1148,13 @@ static struct i915_gem_engines *user_engines(struct i915_gem_context *ctx,
 
 		case I915_GEM_ENGINE_TYPE_BALANCED:
 			ce = intel_engine_create_virtual(pe[n].siblings,
-							 pe[n].num_siblings);
+							 pe[n].num_siblings, 0);
+			break;
+
+		case I915_GEM_ENGINE_TYPE_PARALLEL:
+			ce = intel_engine_create_parallel(pe[n].siblings,
+							  pe[n].num_siblings,
+							  pe[n].width);
 			break;
 
 		case I915_GEM_ENGINE_TYPE_INVALID:
@@ -977,6 +1175,30 @@ static struct i915_gem_engines *user_engines(struct i915_gem_context *ctx,
 			err = ERR_PTR(ret);
 			goto free_engines;
 		}
+		for_each_child(ce, child) {
+			ret = intel_context_set_gem(child, ctx, pe->sseu);
+			if (ret) {
+				err = ERR_PTR(ret);
+				goto free_engines;
+			}
+		}
+
+		/*
+		 * XXX: Must be done after calling intel_context_set_gem as that
+		 * function changes the ring size. The ring is allocated when
+		 * the context is pinned. If the ring size is changed after
+		 * allocation we have a mismatch of the ring size and will cause
+		 * the context to hang. Presumably with a bit of reordering we
+		 * could move the perma-pin step to the backend function
+		 * intel_engine_create_parallel.
+		 */
+		if (pe[n].type == I915_GEM_ENGINE_TYPE_PARALLEL) {
+			ret = perma_pin_contexts(ce);
+			if (ret) {
+				err = ERR_PTR(ret);
+				goto free_engines;
+			}
+		}
 	}
 
 	return e;
@@ -1219,6 +1441,7 @@ static void context_close(struct i915_gem_context *ctx)
 
 	/* Flush any concurrent set_engines() */
 	mutex_lock(&ctx->engines_mutex);
+	unpin_engines(__context_engines_static(ctx));
 	engines_idle_release(ctx, rcu_replace_pointer(ctx->engines, NULL, 1));
 	i915_gem_context_set_closed(ctx);
 	mutex_unlock(&ctx->engines_mutex);
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_context_types.h b/drivers/gpu/drm/i915/gem/i915_gem_context_types.h
index a627b09c4680..282cdb8a5c5a 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_context_types.h
+++ b/drivers/gpu/drm/i915/gem/i915_gem_context_types.h
@@ -78,13 +78,16 @@ enum i915_gem_engine_type {
 
 	/** @I915_GEM_ENGINE_TYPE_BALANCED: A load-balanced engine set */
 	I915_GEM_ENGINE_TYPE_BALANCED,
+
+	/** @I915_GEM_ENGINE_TYPE_PARALLEL: A parallel engine set */
+	I915_GEM_ENGINE_TYPE_PARALLEL,
 };
 
 /**
  * struct i915_gem_proto_engine - prototype engine
  *
  * This struct describes an engine that a context may contain.  Engines
- * have three types:
+ * have four types:
  *
  *  - I915_GEM_ENGINE_TYPE_INVALID: Invalid engines can be created but they
  *    show up as a NULL in i915_gem_engines::engines[i] and any attempt to
@@ -97,6 +100,10 @@ enum i915_gem_engine_type {
  *
  *  - I915_GEM_ENGINE_TYPE_BALANCED: A load-balanced engine set, described
  *    i915_gem_proto_engine::num_siblings and i915_gem_proto_engine::siblings.
+ *
+ *  - I915_GEM_ENGINE_TYPE_PARALLEL: A parallel submission engine set, described
+ *    i915_gem_proto_engine::width, i915_gem_proto_engine::num_siblings, and
+ *    i915_gem_proto_engine::siblings.
  */
 struct i915_gem_proto_engine {
 	/** @type: Type of this engine */
@@ -105,10 +112,13 @@ struct i915_gem_proto_engine {
 	/** @engine: Engine, for physical */
 	struct intel_engine_cs *engine;
 
-	/** @num_siblings: Number of balanced siblings */
+	/** @num_siblings: Number of balanced or parallel siblings */
 	unsigned int num_siblings;
 
-	/** @siblings: Balanced siblings */
+	/** @width: Width of each sibling */
+	unsigned int width;
+
+	/** @siblings: Balanced siblings or num_siblings * width for parallel */
 	struct intel_engine_cs **siblings;
 
 	/** @sseu: Client-set SSEU parameters */
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c b/drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c
index afa34111de02..1adcd8e02d29 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c
@@ -232,6 +232,7 @@ struct dma_buf *i915_gem_prime_export(struct drm_gem_object *gem_obj, int flags)
 
 static int i915_gem_object_get_pages_dmabuf(struct drm_i915_gem_object *obj)
 {
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
 	struct sg_table *pages;
 	unsigned int sg_page_sizes;
 
@@ -242,8 +243,11 @@ static int i915_gem_object_get_pages_dmabuf(struct drm_i915_gem_object *obj)
 	if (IS_ERR(pages))
 		return PTR_ERR(pages);
 
-	sg_page_sizes = i915_sg_dma_sizes(pages->sgl);
+	/* XXX: consider doing a vmap flush or something */
+	if (!HAS_LLC(i915) || i915_gem_object_can_bypass_llc(obj))
+		wbinvd_on_all_cpus();
 
+	sg_page_sizes = i915_sg_dma_sizes(pages->sgl);
 	__i915_gem_object_set_pages(obj, pages, sg_page_sizes);
 
 	return 0;
@@ -301,7 +305,8 @@ struct drm_gem_object *i915_gem_prime_import(struct drm_device *dev,
 	}
 
 	drm_gem_private_object_init(dev, &obj->base, dma_buf->size);
-	i915_gem_object_init(obj, &i915_gem_object_dmabuf_ops, &lock_class, 0);
+	i915_gem_object_init(obj, &i915_gem_object_dmabuf_ops, &lock_class,
+			     I915_BO_ALLOC_USER);
 	obj->base.import_attach = attach;
 	obj->base.resv = dma_buf->resv;
 
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
index 8b3a25bd93e6..4d7da07442f2 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
@@ -246,17 +246,25 @@ struct i915_execbuffer {
 	struct drm_i915_gem_exec_object2 *exec; /** ioctl execobj[] */
 	struct eb_vma *vma;
 
-	struct intel_engine_cs *engine; /** engine to queue the request to */
+	struct intel_gt *gt; /* gt for the execbuf */
 	struct intel_context *context; /* logical state for the request */
 	struct i915_gem_context *gem_context; /** caller's context */
 
-	struct i915_request *request; /** our request to build */
-	struct eb_vma *batch; /** identity of the batch obj/vma */
+	/** our requests to build */
+	struct i915_request *requests[MAX_ENGINE_INSTANCE + 1];
+	/** identity of the batch obj/vma */
+	struct eb_vma *batches[MAX_ENGINE_INSTANCE + 1];
 	struct i915_vma *trampoline; /** trampoline used for chaining */
 
+	/** used for excl fence in dma_resv objects when > 1 BB submitted */
+	struct dma_fence *composite_fence;
+
 	/** actual size of execobj[] as we may extend it for the cmdparser */
 	unsigned int buffer_count;
 
+	/* number of batches in execbuf IOCTL */
+	unsigned int num_batches;
+
 	/** list of vma not yet bound during reservation phase */
 	struct list_head unbound;
 
@@ -283,7 +291,8 @@ struct i915_execbuffer {
 
 	u64 invalid_flags; /** Set of execobj.flags that are invalid */
 
-	u64 batch_len; /** Length of batch within object */
+	/** Length of batch within object */
+	u64 batch_len[MAX_ENGINE_INSTANCE + 1];
 	u32 batch_start_offset; /** Location within object of batch */
 	u32 batch_flags; /** Flags composed for emit_bb_start() */
 	struct intel_gt_buffer_pool_node *batch_pool; /** pool node for batch buffer */
@@ -301,14 +310,13 @@ struct i915_execbuffer {
 };
 
 static int eb_parse(struct i915_execbuffer *eb);
-static struct i915_request *eb_pin_engine(struct i915_execbuffer *eb,
-					  bool throttle);
+static int eb_pin_engine(struct i915_execbuffer *eb, bool throttle);
 static void eb_unpin_engine(struct i915_execbuffer *eb);
 
 static inline bool eb_use_cmdparser(const struct i915_execbuffer *eb)
 {
-	return intel_engine_requires_cmd_parser(eb->engine) ||
-		(intel_engine_using_cmd_parser(eb->engine) &&
+	return intel_engine_requires_cmd_parser(eb->context->engine) ||
+		(intel_engine_using_cmd_parser(eb->context->engine) &&
 		 eb->args->batch_len);
 }
 
@@ -535,11 +543,21 @@ eb_validate_vma(struct i915_execbuffer *eb,
 	return 0;
 }
 
-static void
+static inline bool
+is_batch_buffer(struct i915_execbuffer *eb, unsigned int buffer_idx)
+{
+	return eb->args->flags & I915_EXEC_BATCH_FIRST ?
+		buffer_idx < eb->num_batches :
+		buffer_idx >= eb->args->buffer_count - eb->num_batches;
+}
+
+static int
 eb_add_vma(struct i915_execbuffer *eb,
-	   unsigned int i, unsigned batch_idx,
+	   unsigned int *current_batch,
+	   unsigned int i,
 	   struct i915_vma *vma)
 {
+	struct drm_i915_private *i915 = eb->i915;
 	struct drm_i915_gem_exec_object2 *entry = &eb->exec[i];
 	struct eb_vma *ev = &eb->vma[i];
 
@@ -566,15 +584,43 @@ eb_add_vma(struct i915_execbuffer *eb,
 	 * Note that actual hangs have only been observed on gen7, but for
 	 * paranoia do it everywhere.
 	 */
-	if (i == batch_idx) {
+	if (is_batch_buffer(eb, i)) {
 		if (entry->relocation_count &&
 		    !(ev->flags & EXEC_OBJECT_PINNED))
 			ev->flags |= __EXEC_OBJECT_NEEDS_BIAS;
 		if (eb->reloc_cache.has_fence)
 			ev->flags |= EXEC_OBJECT_NEEDS_FENCE;
 
-		eb->batch = ev;
+		eb->batches[*current_batch] = ev;
+
+		if (unlikely(ev->flags & EXEC_OBJECT_WRITE)) {
+			drm_dbg(&i915->drm,
+				"Attempting to use self-modifying batch buffer\n");
+			return -EINVAL;
+		}
+
+		if (range_overflows_t(u64,
+				      eb->batch_start_offset,
+				      eb->args->batch_len,
+				      ev->vma->size)) {
+			drm_dbg(&i915->drm, "Attempting to use out-of-bounds batch\n");
+			return -EINVAL;
+		}
+
+		if (eb->args->batch_len == 0)
+			eb->batch_len[*current_batch] = ev->vma->size -
+				eb->batch_start_offset;
+		else
+			eb->batch_len[*current_batch] = eb->args->batch_len;
+		if (unlikely(eb->batch_len[*current_batch] == 0)) { /* impossible! */
+			drm_dbg(&i915->drm, "Invalid batch length\n");
+			return -EINVAL;
+		}
+
+		++*current_batch;
 	}
+
+	return 0;
 }
 
 static inline int use_cpu_reloc(const struct reloc_cache *cache,
@@ -718,14 +764,6 @@ static int eb_reserve(struct i915_execbuffer *eb)
 	} while (1);
 }
 
-static unsigned int eb_batch_index(const struct i915_execbuffer *eb)
-{
-	if (eb->args->flags & I915_EXEC_BATCH_FIRST)
-		return 0;
-	else
-		return eb->buffer_count - 1;
-}
-
 static int eb_select_context(struct i915_execbuffer *eb)
 {
 	struct i915_gem_context *ctx;
@@ -846,9 +884,7 @@ static struct i915_vma *eb_lookup_vma(struct i915_execbuffer *eb, u32 handle)
 
 static int eb_lookup_vmas(struct i915_execbuffer *eb)
 {
-	struct drm_i915_private *i915 = eb->i915;
-	unsigned int batch = eb_batch_index(eb);
-	unsigned int i;
+	unsigned int i, current_batch = 0;
 	int err = 0;
 
 	INIT_LIST_HEAD(&eb->relocs);
@@ -868,7 +904,9 @@ static int eb_lookup_vmas(struct i915_execbuffer *eb)
 			goto err;
 		}
 
-		eb_add_vma(eb, i, batch, vma);
+		err = eb_add_vma(eb, &current_batch, i, vma);
+		if (err)
+			return err;
 
 		if (i915_gem_object_is_userptr(vma->obj)) {
 			err = i915_gem_object_userptr_submit_init(vma->obj);
@@ -891,26 +929,6 @@ static int eb_lookup_vmas(struct i915_execbuffer *eb)
 		}
 	}
 
-	if (unlikely(eb->batch->flags & EXEC_OBJECT_WRITE)) {
-		drm_dbg(&i915->drm,
-			"Attempting to use self-modifying batch buffer\n");
-		return -EINVAL;
-	}
-
-	if (range_overflows_t(u64,
-			      eb->batch_start_offset, eb->batch_len,
-			      eb->batch->vma->size)) {
-		drm_dbg(&i915->drm, "Attempting to use out-of-bounds batch\n");
-		return -EINVAL;
-	}
-
-	if (eb->batch_len == 0)
-		eb->batch_len = eb->batch->vma->size - eb->batch_start_offset;
-	if (unlikely(eb->batch_len == 0)) { /* impossible! */
-		drm_dbg(&i915->drm, "Invalid batch length\n");
-		return -EINVAL;
-	}
-
 	return 0;
 
 err:
@@ -1643,8 +1661,7 @@ static int eb_reinit_userptr(struct i915_execbuffer *eb)
 	return 0;
 }
 
-static noinline int eb_relocate_parse_slow(struct i915_execbuffer *eb,
-					   struct i915_request *rq)
+static noinline int eb_relocate_parse_slow(struct i915_execbuffer *eb)
 {
 	bool have_copy = false;
 	struct eb_vma *ev;
@@ -1660,21 +1677,6 @@ repeat:
 	eb_release_vmas(eb, false);
 	i915_gem_ww_ctx_fini(&eb->ww);
 
-	if (rq) {
-		/* nonblocking is always false */
-		if (i915_request_wait(rq, I915_WAIT_INTERRUPTIBLE,
-				      MAX_SCHEDULE_TIMEOUT) < 0) {
-			i915_request_put(rq);
-			rq = NULL;
-
-			err = -EINTR;
-			goto err_relock;
-		}
-
-		i915_request_put(rq);
-		rq = NULL;
-	}
-
 	/*
 	 * We take 3 passes through the slowpatch.
 	 *
@@ -1701,28 +1703,21 @@ repeat:
 	if (!err)
 		err = eb_reinit_userptr(eb);
 
-err_relock:
 	i915_gem_ww_ctx_init(&eb->ww, true);
 	if (err)
 		goto out;
 
 	/* reacquire the objects */
 repeat_validate:
-	rq = eb_pin_engine(eb, false);
-	if (IS_ERR(rq)) {
-		err = PTR_ERR(rq);
-		rq = NULL;
+	err = eb_pin_engine(eb, false);
+	if (err)
 		goto err;
-	}
-
-	/* We didn't throttle, should be NULL */
-	GEM_WARN_ON(rq);
 
 	err = eb_validate_vmas(eb);
 	if (err)
 		goto err;
 
-	GEM_BUG_ON(!eb->batch);
+	GEM_BUG_ON(!eb->batches[0]);
 
 	list_for_each_entry(ev, &eb->relocs, reloc_link) {
 		if (!have_copy) {
@@ -1786,46 +1781,23 @@ out:
 		}
 	}
 
-	if (rq)
-		i915_request_put(rq);
-
 	return err;
 }
 
 static int eb_relocate_parse(struct i915_execbuffer *eb)
 {
 	int err;
-	struct i915_request *rq = NULL;
 	bool throttle = true;
 
 retry:
-	rq = eb_pin_engine(eb, throttle);
-	if (IS_ERR(rq)) {
-		err = PTR_ERR(rq);
-		rq = NULL;
+	err = eb_pin_engine(eb, throttle);
+	if (err) {
 		if (err != -EDEADLK)
 			return err;
 
 		goto err;
 	}
 
-	if (rq) {
-		bool nonblock = eb->file->filp->f_flags & O_NONBLOCK;
-
-		/* Need to drop all locks now for throttling, take slowpath */
-		err = i915_request_wait(rq, I915_WAIT_INTERRUPTIBLE, 0);
-		if (err == -ETIME) {
-			if (nonblock) {
-				err = -EWOULDBLOCK;
-				i915_request_put(rq);
-				goto err;
-			}
-			goto slow;
-		}
-		i915_request_put(rq);
-		rq = NULL;
-	}
-
 	/* only throttle once, even if we didn't need to throttle */
 	throttle = false;
 
@@ -1865,7 +1837,7 @@ err:
 	return err;
 
 slow:
-	err = eb_relocate_parse_slow(eb, rq);
+	err = eb_relocate_parse_slow(eb);
 	if (err)
 		/*
 		 * If the user expects the execobject.offset and
@@ -1879,11 +1851,40 @@ slow:
 	return err;
 }
 
+/*
+ * Using two helper loops for the order of which requests / batches are created
+ * and added the to backend. Requests are created in order from the parent to
+ * the last child. Requests are added in the reverse order, from the last child
+ * to parent. This is done for locking reasons as the timeline lock is acquired
+ * during request creation and released when the request is added to the
+ * backend. To make lockdep happy (see intel_context_timeline_lock) this must be
+ * the ordering.
+ */
+#define for_each_batch_create_order(_eb, _i) \
+	for ((_i) = 0; (_i) < (_eb)->num_batches; ++(_i))
+#define for_each_batch_add_order(_eb, _i) \
+	BUILD_BUG_ON(!typecheck(int, _i)); \
+	for ((_i) = (_eb)->num_batches - 1; (_i) >= 0; --(_i))
+
+static struct i915_request *
+eb_find_first_request_added(struct i915_execbuffer *eb)
+{
+	int i;
+
+	for_each_batch_add_order(eb, i)
+		if (eb->requests[i])
+			return eb->requests[i];
+
+	GEM_BUG_ON("Request not found");
+
+	return NULL;
+}
+
 static int eb_move_to_gpu(struct i915_execbuffer *eb)
 {
 	const unsigned int count = eb->buffer_count;
 	unsigned int i = count;
-	int err = 0;
+	int err = 0, j;
 
 	while (i--) {
 		struct eb_vma *ev = &eb->vma[i];
@@ -1896,11 +1897,17 @@ static int eb_move_to_gpu(struct i915_execbuffer *eb)
 		if (flags & EXEC_OBJECT_CAPTURE) {
 			struct i915_capture_list *capture;
 
-			capture = kmalloc(sizeof(*capture), GFP_KERNEL);
-			if (capture) {
-				capture->next = eb->request->capture_list;
-				capture->vma = vma;
-				eb->request->capture_list = capture;
+			for_each_batch_create_order(eb, j) {
+				if (!eb->requests[j])
+					break;
+
+				capture = kmalloc(sizeof(*capture), GFP_KERNEL);
+				if (capture) {
+					capture->next =
+						eb->requests[j]->capture_list;
+					capture->vma = vma;
+					eb->requests[j]->capture_list = capture;
+				}
 			}
 		}
 
@@ -1915,20 +1922,43 @@ static int eb_move_to_gpu(struct i915_execbuffer *eb)
 		 *   !(obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_READ)
 		 * but gcc's optimiser doesn't handle that as well and emits
 		 * two jumps instead of one. Maybe one day...
+		 *
+		 * FIXME: There is also sync flushing in set_pages(), which
+		 * serves a different purpose(some of the time at least).
+		 *
+		 * We should consider:
+		 *
+		 *   1. Rip out the async flush code.
+		 *
+		 *   2. Or make the sync flushing use the async clflush path
+		 *   using mandatory fences underneath. Currently the below
+		 *   async flush happens after we bind the object.
 		 */
 		if (unlikely(obj->cache_dirty & ~obj->cache_coherent)) {
 			if (i915_gem_clflush_object(obj, 0))
 				flags &= ~EXEC_OBJECT_ASYNC;
 		}
 
+		/* We only need to await on the first request */
 		if (err == 0 && !(flags & EXEC_OBJECT_ASYNC)) {
 			err = i915_request_await_object
-				(eb->request, obj, flags & EXEC_OBJECT_WRITE);
+				(eb_find_first_request_added(eb), obj,
+				 flags & EXEC_OBJECT_WRITE);
 		}
 
-		if (err == 0)
-			err = i915_vma_move_to_active(vma, eb->request,
-						      flags | __EXEC_OBJECT_NO_RESERVE);
+		for_each_batch_add_order(eb, j) {
+			if (err)
+				break;
+			if (!eb->requests[j])
+				continue;
+
+			err = _i915_vma_move_to_active(vma, eb->requests[j],
+						       j ? NULL :
+						       eb->composite_fence ?
+						       eb->composite_fence :
+						       &eb->requests[j]->fence,
+						       flags | __EXEC_OBJECT_NO_RESERVE);
+		}
 	}
 
 #ifdef CONFIG_MMU_NOTIFIER
@@ -1959,11 +1989,16 @@ static int eb_move_to_gpu(struct i915_execbuffer *eb)
 		goto err_skip;
 
 	/* Unconditionally flush any chipset caches (for streaming writes). */
-	intel_gt_chipset_flush(eb->engine->gt);
+	intel_gt_chipset_flush(eb->gt);
 	return 0;
 
 err_skip:
-	i915_request_set_error_once(eb->request, err);
+	for_each_batch_create_order(eb, j) {
+		if (!eb->requests[j])
+			break;
+
+		i915_request_set_error_once(eb->requests[j], err);
+	}
 	return err;
 }
 
@@ -2058,14 +2093,17 @@ static int eb_parse(struct i915_execbuffer *eb)
 	int err;
 
 	if (!eb_use_cmdparser(eb)) {
-		batch = eb_dispatch_secure(eb, eb->batch->vma);
+		batch = eb_dispatch_secure(eb, eb->batches[0]->vma);
 		if (IS_ERR(batch))
 			return PTR_ERR(batch);
 
 		goto secure_batch;
 	}
 
-	len = eb->batch_len;
+	if (intel_context_is_parallel(eb->context))
+		return -EINVAL;
+
+	len = eb->batch_len[0];
 	if (!CMDPARSER_USES_GGTT(eb->i915)) {
 		/*
 		 * ppGTT backed shadow buffers must be mapped RO, to prevent
@@ -2079,11 +2117,11 @@ static int eb_parse(struct i915_execbuffer *eb)
 	} else {
 		len += I915_CMD_PARSER_TRAMPOLINE_SIZE;
 	}
-	if (unlikely(len < eb->batch_len)) /* last paranoid check of overflow */
+	if (unlikely(len < eb->batch_len[0])) /* last paranoid check of overflow */
 		return -EINVAL;
 
 	if (!pool) {
-		pool = intel_gt_get_buffer_pool(eb->engine->gt, len,
+		pool = intel_gt_get_buffer_pool(eb->gt, len,
 						I915_MAP_WB);
 		if (IS_ERR(pool))
 			return PTR_ERR(pool);
@@ -2108,7 +2146,7 @@ static int eb_parse(struct i915_execbuffer *eb)
 		trampoline = shadow;
 
 		shadow = shadow_batch_pin(eb, pool->obj,
-					  &eb->engine->gt->ggtt->vm,
+					  &eb->gt->ggtt->vm,
 					  PIN_GLOBAL);
 		if (IS_ERR(shadow)) {
 			err = PTR_ERR(shadow);
@@ -2130,26 +2168,29 @@ static int eb_parse(struct i915_execbuffer *eb)
 	if (err)
 		goto err_trampoline;
 
-	err = intel_engine_cmd_parser(eb->engine,
-				      eb->batch->vma,
+	err = intel_engine_cmd_parser(eb->context->engine,
+				      eb->batches[0]->vma,
 				      eb->batch_start_offset,
-				      eb->batch_len,
+				      eb->batch_len[0],
 				      shadow, trampoline);
 	if (err)
 		goto err_unpin_batch;
 
-	eb->batch = &eb->vma[eb->buffer_count++];
-	eb->batch->vma = i915_vma_get(shadow);
-	eb->batch->flags = __EXEC_OBJECT_HAS_PIN;
+	eb->batches[0] = &eb->vma[eb->buffer_count++];
+	eb->batches[0]->vma = i915_vma_get(shadow);
+	eb->batches[0]->flags = __EXEC_OBJECT_HAS_PIN;
 
 	eb->trampoline = trampoline;
 	eb->batch_start_offset = 0;
 
 secure_batch:
 	if (batch) {
-		eb->batch = &eb->vma[eb->buffer_count++];
-		eb->batch->flags = __EXEC_OBJECT_HAS_PIN;
-		eb->batch->vma = i915_vma_get(batch);
+		if (intel_context_is_parallel(eb->context))
+			return -EINVAL;
+
+		eb->batches[0] = &eb->vma[eb->buffer_count++];
+		eb->batches[0]->flags = __EXEC_OBJECT_HAS_PIN;
+		eb->batches[0]->vma = i915_vma_get(batch);
 	}
 	return 0;
 
@@ -2165,19 +2206,18 @@ err:
 	return err;
 }
 
-static int eb_submit(struct i915_execbuffer *eb, struct i915_vma *batch)
+static int eb_request_submit(struct i915_execbuffer *eb,
+			     struct i915_request *rq,
+			     struct i915_vma *batch,
+			     u64 batch_len)
 {
 	int err;
 
-	if (intel_context_nopreempt(eb->context))
-		__set_bit(I915_FENCE_FLAG_NOPREEMPT, &eb->request->fence.flags);
-
-	err = eb_move_to_gpu(eb);
-	if (err)
-		return err;
+	if (intel_context_nopreempt(rq->context))
+		__set_bit(I915_FENCE_FLAG_NOPREEMPT, &rq->fence.flags);
 
 	if (eb->args->flags & I915_EXEC_GEN7_SOL_RESET) {
-		err = i915_reset_gen7_sol_offsets(eb->request);
+		err = i915_reset_gen7_sol_offsets(rq);
 		if (err)
 			return err;
 	}
@@ -2188,26 +2228,26 @@ static int eb_submit(struct i915_execbuffer *eb, struct i915_vma *batch)
 	 * allows us to determine if the batch is still waiting on the GPU
 	 * or actually running by checking the breadcrumb.
 	 */
-	if (eb->engine->emit_init_breadcrumb) {
-		err = eb->engine->emit_init_breadcrumb(eb->request);
+	if (rq->context->engine->emit_init_breadcrumb) {
+		err = rq->context->engine->emit_init_breadcrumb(rq);
 		if (err)
 			return err;
 	}
 
-	err = eb->engine->emit_bb_start(eb->request,
-					batch->node.start +
-					eb->batch_start_offset,
-					eb->batch_len,
-					eb->batch_flags);
+	err = rq->context->engine->emit_bb_start(rq,
+						 batch->node.start +
+						 eb->batch_start_offset,
+						 batch_len,
+						 eb->batch_flags);
 	if (err)
 		return err;
 
 	if (eb->trampoline) {
+		GEM_BUG_ON(intel_context_is_parallel(rq->context));
 		GEM_BUG_ON(eb->batch_start_offset);
-		err = eb->engine->emit_bb_start(eb->request,
-						eb->trampoline->node.start +
-						eb->batch_len,
-						0, 0);
+		err = rq->context->engine->emit_bb_start(rq,
+							 eb->trampoline->node.start +
+							 batch_len, 0, 0);
 		if (err)
 			return err;
 	}
@@ -2215,6 +2255,27 @@ static int eb_submit(struct i915_execbuffer *eb, struct i915_vma *batch)
 	return 0;
 }
 
+static int eb_submit(struct i915_execbuffer *eb)
+{
+	unsigned int i;
+	int err;
+
+	err = eb_move_to_gpu(eb);
+
+	for_each_batch_create_order(eb, i) {
+		if (!eb->requests[i])
+			break;
+
+		trace_i915_request_queue(eb->requests[i], eb->batch_flags);
+		if (!err)
+			err = eb_request_submit(eb, eb->requests[i],
+						eb->batches[i]->vma,
+						eb->batch_len[i]);
+	}
+
+	return err;
+}
+
 static int num_vcs_engines(const struct drm_i915_private *i915)
 {
 	return hweight_long(VDBOX_MASK(&i915->gt));
@@ -2280,26 +2341,11 @@ static struct i915_request *eb_throttle(struct i915_execbuffer *eb, struct intel
 	return i915_request_get(rq);
 }
 
-static struct i915_request *eb_pin_engine(struct i915_execbuffer *eb, bool throttle)
+static int eb_pin_timeline(struct i915_execbuffer *eb, struct intel_context *ce,
+			   bool throttle)
 {
-	struct intel_context *ce = eb->context;
 	struct intel_timeline *tl;
 	struct i915_request *rq = NULL;
-	int err;
-
-	GEM_BUG_ON(eb->args->flags & __EXEC_ENGINE_PINNED);
-
-	if (unlikely(intel_context_is_banned(ce)))
-		return ERR_PTR(-EIO);
-
-	/*
-	 * Pinning the contexts may generate requests in order to acquire
-	 * GGTT space, so do this first before we reserve a seqno for
-	 * ourselves.
-	 */
-	err = intel_context_pin_ww(ce, &eb->ww);
-	if (err)
-		return ERR_PTR(err);
 
 	/*
 	 * Take a local wakeref for preparing to dispatch the execbuf as
@@ -2310,33 +2356,108 @@ static struct i915_request *eb_pin_engine(struct i915_execbuffer *eb, bool throt
 	 * taken on the engine, and the parent device.
 	 */
 	tl = intel_context_timeline_lock(ce);
-	if (IS_ERR(tl)) {
-		intel_context_unpin(ce);
-		return ERR_CAST(tl);
-	}
+	if (IS_ERR(tl))
+		return PTR_ERR(tl);
 
 	intel_context_enter(ce);
 	if (throttle)
 		rq = eb_throttle(eb, ce);
 	intel_context_timeline_unlock(tl);
 
+	if (rq) {
+		bool nonblock = eb->file->filp->f_flags & O_NONBLOCK;
+		long timeout = nonblock ? 0 : MAX_SCHEDULE_TIMEOUT;
+
+		if (i915_request_wait(rq, I915_WAIT_INTERRUPTIBLE,
+				      timeout) < 0) {
+			i915_request_put(rq);
+
+			tl = intel_context_timeline_lock(ce);
+			intel_context_exit(ce);
+			intel_context_timeline_unlock(tl);
+
+			if (nonblock)
+				return -EWOULDBLOCK;
+			else
+				return -EINTR;
+		}
+		i915_request_put(rq);
+	}
+
+	return 0;
+}
+
+static int eb_pin_engine(struct i915_execbuffer *eb, bool throttle)
+{
+	struct intel_context *ce = eb->context, *child;
+	int err;
+	int i = 0, j = 0;
+
+	GEM_BUG_ON(eb->args->flags & __EXEC_ENGINE_PINNED);
+
+	if (unlikely(intel_context_is_banned(ce)))
+		return -EIO;
+
+	/*
+	 * Pinning the contexts may generate requests in order to acquire
+	 * GGTT space, so do this first before we reserve a seqno for
+	 * ourselves.
+	 */
+	err = intel_context_pin_ww(ce, &eb->ww);
+	if (err)
+		return err;
+	for_each_child(ce, child) {
+		err = intel_context_pin_ww(child, &eb->ww);
+		GEM_BUG_ON(err);	/* perma-pinned should incr a counter */
+	}
+
+	for_each_child(ce, child) {
+		err = eb_pin_timeline(eb, child, throttle);
+		if (err)
+			goto unwind;
+		++i;
+	}
+	err = eb_pin_timeline(eb, ce, throttle);
+	if (err)
+		goto unwind;
+
 	eb->args->flags |= __EXEC_ENGINE_PINNED;
-	return rq;
+	return 0;
+
+unwind:
+	for_each_child(ce, child) {
+		if (j++ < i) {
+			mutex_lock(&child->timeline->mutex);
+			intel_context_exit(child);
+			mutex_unlock(&child->timeline->mutex);
+		}
+	}
+	for_each_child(ce, child)
+		intel_context_unpin(child);
+	intel_context_unpin(ce);
+	return err;
 }
 
 static void eb_unpin_engine(struct i915_execbuffer *eb)
 {
-	struct intel_context *ce = eb->context;
-	struct intel_timeline *tl = ce->timeline;
+	struct intel_context *ce = eb->context, *child;
 
 	if (!(eb->args->flags & __EXEC_ENGINE_PINNED))
 		return;
 
 	eb->args->flags &= ~__EXEC_ENGINE_PINNED;
 
-	mutex_lock(&tl->mutex);
+	for_each_child(ce, child) {
+		mutex_lock(&child->timeline->mutex);
+		intel_context_exit(child);
+		mutex_unlock(&child->timeline->mutex);
+
+		intel_context_unpin(child);
+	}
+
+	mutex_lock(&ce->timeline->mutex);
 	intel_context_exit(ce);
-	mutex_unlock(&tl->mutex);
+	mutex_unlock(&ce->timeline->mutex);
 
 	intel_context_unpin(ce);
 }
@@ -2387,7 +2508,7 @@ eb_select_legacy_ring(struct i915_execbuffer *eb)
 static int
 eb_select_engine(struct i915_execbuffer *eb)
 {
-	struct intel_context *ce;
+	struct intel_context *ce, *child;
 	unsigned int idx;
 	int err;
 
@@ -2400,6 +2521,20 @@ eb_select_engine(struct i915_execbuffer *eb)
 	if (IS_ERR(ce))
 		return PTR_ERR(ce);
 
+	if (intel_context_is_parallel(ce)) {
+		if (eb->buffer_count < ce->parallel.number_children + 1) {
+			intel_context_put(ce);
+			return -EINVAL;
+		}
+		if (eb->batch_start_offset || eb->args->batch_len) {
+			intel_context_put(ce);
+			return -EINVAL;
+		}
+	}
+	eb->num_batches = ce->parallel.number_children + 1;
+
+	for_each_child(ce, child)
+		intel_context_get(child);
 	intel_gt_pm_get(ce->engine->gt);
 
 	if (!test_bit(CONTEXT_ALLOC_BIT, &ce->flags)) {
@@ -2407,6 +2542,13 @@ eb_select_engine(struct i915_execbuffer *eb)
 		if (err)
 			goto err;
 	}
+	for_each_child(ce, child) {
+		if (!test_bit(CONTEXT_ALLOC_BIT, &child->flags)) {
+			err = intel_context_alloc_state(child);
+			if (err)
+				goto err;
+		}
+	}
 
 	/*
 	 * ABI: Before userspace accesses the GPU (e.g. execbuffer), report
@@ -2417,7 +2559,7 @@ eb_select_engine(struct i915_execbuffer *eb)
 		goto err;
 
 	eb->context = ce;
-	eb->engine = ce->engine;
+	eb->gt = ce->engine->gt;
 
 	/*
 	 * Make sure engine pool stays alive even if we call intel_context_put
@@ -2428,6 +2570,8 @@ eb_select_engine(struct i915_execbuffer *eb)
 
 err:
 	intel_gt_pm_put(ce->engine->gt);
+	for_each_child(ce, child)
+		intel_context_put(child);
 	intel_context_put(ce);
 	return err;
 }
@@ -2435,7 +2579,11 @@ err:
 static void
 eb_put_engine(struct i915_execbuffer *eb)
 {
-	intel_gt_pm_put(eb->engine->gt);
+	struct intel_context *child;
+
+	intel_gt_pm_put(eb->gt);
+	for_each_child(eb->context, child)
+		intel_context_put(child);
 	intel_context_put(eb->context);
 }
 
@@ -2658,7 +2806,8 @@ static void put_fence_array(struct eb_fence *fences, int num_fences)
 }
 
 static int
-await_fence_array(struct i915_execbuffer *eb)
+await_fence_array(struct i915_execbuffer *eb,
+		  struct i915_request *rq)
 {
 	unsigned int n;
 	int err;
@@ -2672,8 +2821,7 @@ await_fence_array(struct i915_execbuffer *eb)
 		if (!eb->fences[n].dma_fence)
 			continue;
 
-		err = i915_request_await_dma_fence(eb->request,
-						   eb->fences[n].dma_fence);
+		err = i915_request_await_dma_fence(rq, eb->fences[n].dma_fence);
 		if (err < 0)
 			return err;
 	}
@@ -2681,9 +2829,9 @@ await_fence_array(struct i915_execbuffer *eb)
 	return 0;
 }
 
-static void signal_fence_array(const struct i915_execbuffer *eb)
+static void signal_fence_array(const struct i915_execbuffer *eb,
+			       struct dma_fence * const fence)
 {
-	struct dma_fence * const fence = &eb->request->fence;
 	unsigned int n;
 
 	for (n = 0; n < eb->num_fences; n++) {
@@ -2731,9 +2879,9 @@ static void retire_requests(struct intel_timeline *tl, struct i915_request *end)
 			break;
 }
 
-static int eb_request_add(struct i915_execbuffer *eb, int err)
+static int eb_request_add(struct i915_execbuffer *eb, struct i915_request *rq,
+			  int err, bool last_parallel)
 {
-	struct i915_request *rq = eb->request;
 	struct intel_timeline * const tl = i915_request_timeline(rq);
 	struct i915_sched_attr attr = {};
 	struct i915_request *prev;
@@ -2755,6 +2903,17 @@ static int eb_request_add(struct i915_execbuffer *eb, int err)
 		err = -ENOENT; /* override any transient errors */
 	}
 
+	if (intel_context_is_parallel(eb->context)) {
+		if (err) {
+			__i915_request_skip(rq);
+			set_bit(I915_FENCE_FLAG_SKIP_PARALLEL,
+				&rq->fence.flags);
+		}
+		if (last_parallel)
+			set_bit(I915_FENCE_FLAG_SUBMIT_PARALLEL,
+				&rq->fence.flags);
+	}
+
 	__i915_request_queue(rq, &attr);
 
 	/* Try to clean up the client's timeline after submitting the request */
@@ -2766,6 +2925,25 @@ static int eb_request_add(struct i915_execbuffer *eb, int err)
 	return err;
 }
 
+static int eb_requests_add(struct i915_execbuffer *eb, int err)
+{
+	int i;
+
+	/*
+	 * We iterate in reverse order of creation to release timeline mutexes in
+	 * same order.
+	 */
+	for_each_batch_add_order(eb, i) {
+		struct i915_request *rq = eb->requests[i];
+
+		if (!rq)
+			continue;
+		err |= eb_request_add(eb, rq, err, i == 0);
+	}
+
+	return err;
+}
+
 static const i915_user_extension_fn execbuf_extensions[] = {
 	[DRM_I915_GEM_EXECBUFFER_EXT_TIMELINE_FENCES] = parse_timeline_fences,
 };
@@ -2792,6 +2970,185 @@ parse_execbuf2_extensions(struct drm_i915_gem_execbuffer2 *args,
 				    eb);
 }
 
+static void eb_requests_get(struct i915_execbuffer *eb)
+{
+	unsigned int i;
+
+	for_each_batch_create_order(eb, i) {
+		if (!eb->requests[i])
+			break;
+
+		i915_request_get(eb->requests[i]);
+	}
+}
+
+static void eb_requests_put(struct i915_execbuffer *eb)
+{
+	unsigned int i;
+
+	for_each_batch_create_order(eb, i) {
+		if (!eb->requests[i])
+			break;
+
+		i915_request_put(eb->requests[i]);
+	}
+}
+
+static struct sync_file *
+eb_composite_fence_create(struct i915_execbuffer *eb, int out_fence_fd)
+{
+	struct sync_file *out_fence = NULL;
+	struct dma_fence_array *fence_array;
+	struct dma_fence **fences;
+	unsigned int i;
+
+	GEM_BUG_ON(!intel_context_is_parent(eb->context));
+
+	fences = kmalloc_array(eb->num_batches, sizeof(*fences), GFP_KERNEL);
+	if (!fences)
+		return ERR_PTR(-ENOMEM);
+
+	for_each_batch_create_order(eb, i) {
+		fences[i] = &eb->requests[i]->fence;
+		__set_bit(I915_FENCE_FLAG_COMPOSITE,
+			  &eb->requests[i]->fence.flags);
+	}
+
+	fence_array = dma_fence_array_create(eb->num_batches,
+					     fences,
+					     eb->context->parallel.fence_context,
+					     eb->context->parallel.seqno,
+					     false);
+	if (!fence_array) {
+		kfree(fences);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	/* Move ownership to the dma_fence_array created above */
+	for_each_batch_create_order(eb, i)
+		dma_fence_get(fences[i]);
+
+	if (out_fence_fd != -1) {
+		out_fence = sync_file_create(&fence_array->base);
+		/* sync_file now owns fence_arry, drop creation ref */
+		dma_fence_put(&fence_array->base);
+		if (!out_fence)
+			return ERR_PTR(-ENOMEM);
+	}
+
+	eb->composite_fence = &fence_array->base;
+
+	return out_fence;
+}
+
+static struct sync_file *
+eb_fences_add(struct i915_execbuffer *eb, struct i915_request *rq,
+	      struct dma_fence *in_fence, int out_fence_fd)
+{
+	struct sync_file *out_fence = NULL;
+	int err;
+
+	if (unlikely(eb->gem_context->syncobj)) {
+		struct dma_fence *fence;
+
+		fence = drm_syncobj_fence_get(eb->gem_context->syncobj);
+		err = i915_request_await_dma_fence(rq, fence);
+		dma_fence_put(fence);
+		if (err)
+			return ERR_PTR(err);
+	}
+
+	if (in_fence) {
+		if (eb->args->flags & I915_EXEC_FENCE_SUBMIT)
+			err = i915_request_await_execution(rq, in_fence);
+		else
+			err = i915_request_await_dma_fence(rq, in_fence);
+		if (err < 0)
+			return ERR_PTR(err);
+	}
+
+	if (eb->fences) {
+		err = await_fence_array(eb, rq);
+		if (err)
+			return ERR_PTR(err);
+	}
+
+	if (intel_context_is_parallel(eb->context)) {
+		out_fence = eb_composite_fence_create(eb, out_fence_fd);
+		if (IS_ERR(out_fence))
+			return ERR_PTR(-ENOMEM);
+	} else if (out_fence_fd != -1) {
+		out_fence = sync_file_create(&rq->fence);
+		if (!out_fence)
+			return ERR_PTR(-ENOMEM);
+	}
+
+	return out_fence;
+}
+
+static struct intel_context *
+eb_find_context(struct i915_execbuffer *eb, unsigned int context_number)
+{
+	struct intel_context *child;
+
+	if (likely(context_number == 0))
+		return eb->context;
+
+	for_each_child(eb->context, child)
+		if (!--context_number)
+			return child;
+
+	GEM_BUG_ON("Context not found");
+
+	return NULL;
+}
+
+static struct sync_file *
+eb_requests_create(struct i915_execbuffer *eb, struct dma_fence *in_fence,
+		   int out_fence_fd)
+{
+	struct sync_file *out_fence = NULL;
+	unsigned int i;
+
+	for_each_batch_create_order(eb, i) {
+		/* Allocate a request for this batch buffer nice and early. */
+		eb->requests[i] = i915_request_create(eb_find_context(eb, i));
+		if (IS_ERR(eb->requests[i])) {
+			out_fence = ERR_PTR(PTR_ERR(eb->requests[i]));
+			eb->requests[i] = NULL;
+			return out_fence;
+		}
+
+		/*
+		 * Only the first request added (committed to backend) has to
+		 * take the in fences into account as all subsequent requests
+		 * will have fences inserted inbetween them.
+		 */
+		if (i + 1 == eb->num_batches) {
+			out_fence = eb_fences_add(eb, eb->requests[i],
+						  in_fence, out_fence_fd);
+			if (IS_ERR(out_fence))
+				return out_fence;
+		}
+
+		/*
+		 * Whilst this request exists, batch_obj will be on the
+		 * active_list, and so will hold the active reference. Only when
+		 * this request is retired will the batch_obj be moved onto
+		 * the inactive_list and lose its active reference. Hence we do
+		 * not need to explicitly hold another reference here.
+		 */
+		eb->requests[i]->batch = eb->batches[i]->vma;
+		if (eb->batch_pool) {
+			GEM_BUG_ON(intel_context_is_parallel(eb->context));
+			intel_gt_buffer_pool_mark_active(eb->batch_pool,
+							 eb->requests[i]);
+		}
+	}
+
+	return out_fence;
+}
+
 static int
 i915_gem_do_execbuffer(struct drm_device *dev,
 		       struct drm_file *file,
@@ -2802,7 +3159,6 @@ i915_gem_do_execbuffer(struct drm_device *dev,
 	struct i915_execbuffer eb;
 	struct dma_fence *in_fence = NULL;
 	struct sync_file *out_fence = NULL;
-	struct i915_vma *batch;
 	int out_fence_fd = -1;
 	int err;
 
@@ -2826,12 +3182,15 @@ i915_gem_do_execbuffer(struct drm_device *dev,
 
 	eb.buffer_count = args->buffer_count;
 	eb.batch_start_offset = args->batch_start_offset;
-	eb.batch_len = args->batch_len;
 	eb.trampoline = NULL;
 
 	eb.fences = NULL;
 	eb.num_fences = 0;
 
+	memset(eb.requests, 0, sizeof(struct i915_request *) *
+	       ARRAY_SIZE(eb.requests));
+	eb.composite_fence = NULL;
+
 	eb.batch_flags = 0;
 	if (args->flags & I915_EXEC_SECURE) {
 		if (GRAPHICS_VER(i915) >= 11)
@@ -2915,70 +3274,25 @@ i915_gem_do_execbuffer(struct drm_device *dev,
 
 	ww_acquire_done(&eb.ww.ctx);
 
-	batch = eb.batch->vma;
-
-	/* Allocate a request for this batch buffer nice and early. */
-	eb.request = i915_request_create(eb.context);
-	if (IS_ERR(eb.request)) {
-		err = PTR_ERR(eb.request);
-		goto err_vma;
-	}
-
-	if (unlikely(eb.gem_context->syncobj)) {
-		struct dma_fence *fence;
-
-		fence = drm_syncobj_fence_get(eb.gem_context->syncobj);
-		err = i915_request_await_dma_fence(eb.request, fence);
-		dma_fence_put(fence);
-		if (err)
-			goto err_ext;
-	}
-
-	if (in_fence) {
-		if (args->flags & I915_EXEC_FENCE_SUBMIT)
-			err = i915_request_await_execution(eb.request,
-							   in_fence);
-		else
-			err = i915_request_await_dma_fence(eb.request,
-							   in_fence);
-		if (err < 0)
-			goto err_request;
-	}
-
-	if (eb.fences) {
-		err = await_fence_array(&eb);
-		if (err)
+	out_fence = eb_requests_create(&eb, in_fence, out_fence_fd);
+	if (IS_ERR(out_fence)) {
+		err = PTR_ERR(out_fence);
+		if (eb.requests[0])
 			goto err_request;
+		else
+			goto err_vma;
 	}
 
-	if (out_fence_fd != -1) {
-		out_fence = sync_file_create(&eb.request->fence);
-		if (!out_fence) {
-			err = -ENOMEM;
-			goto err_request;
-		}
-	}
-
-	/*
-	 * Whilst this request exists, batch_obj will be on the
-	 * active_list, and so will hold the active reference. Only when this
-	 * request is retired will the the batch_obj be moved onto the
-	 * inactive_list and lose its active reference. Hence we do not need
-	 * to explicitly hold another reference here.
-	 */
-	eb.request->batch = batch;
-	if (eb.batch_pool)
-		intel_gt_buffer_pool_mark_active(eb.batch_pool, eb.request);
-
-	trace_i915_request_queue(eb.request, eb.batch_flags);
-	err = eb_submit(&eb, batch);
+	err = eb_submit(&eb);
 
 err_request:
-	i915_request_get(eb.request);
-	err = eb_request_add(&eb, err);
+	eb_requests_get(&eb);
+	err = eb_requests_add(&eb, err);
 
 	if (eb.fences)
-		signal_fence_array(&eb);
+		signal_fence_array(&eb, eb.composite_fence ?
+				   eb.composite_fence :
+				   &eb.requests[0]->fence);
 
 	if (out_fence) {
 		if (err == 0) {
@@ -2993,10 +3307,15 @@ err_request:
 
 	if (unlikely(eb.gem_context->syncobj)) {
 		drm_syncobj_replace_fence(eb.gem_context->syncobj,
-					  &eb.request->fence);
+					  eb.composite_fence ?
+					  eb.composite_fence :
+					  &eb.requests[0]->fence);
 	}
 
-	i915_request_put(eb.request);
+	if (!out_fence && eb.composite_fence)
+		dma_fence_put(eb.composite_fence);
+
+	eb_requests_put(&eb);
 
 err_vma:
 	eb_release_vmas(&eb, true);
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_internal.c b/drivers/gpu/drm/i915/gem/i915_gem_internal.c
index e5ae9c06510c..a57a6b7013c2 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_internal.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_internal.c
@@ -134,6 +134,8 @@ static void i915_gem_object_put_pages_internal(struct drm_i915_gem_object *obj,
 	internal_free_pages(pages);
 
 	obj->mm.dirty = false;
+
+	__start_cpu_write(obj);
 }
 
 static const struct drm_i915_gem_object_ops i915_gem_object_internal_ops = {
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.c b/drivers/gpu/drm/i915/gem/i915_gem_object.c
index 76ce6a1500bc..1e426a42a36c 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_object.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_object.c
@@ -128,6 +128,32 @@ void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj,
 		!(obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_WRITE);
 }
 
+bool i915_gem_object_can_bypass_llc(struct drm_i915_gem_object *obj)
+{
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+
+	/*
+	 * This is purely from a security perspective, so we simply don't care
+	 * about non-userspace objects being able to bypass the LLC.
+	 */
+	if (!(obj->flags & I915_BO_ALLOC_USER))
+		return false;
+
+	/*
+	 * EHL and JSL add the 'Bypass LLC' MOCS entry, which should make it
+	 * possible for userspace to bypass the GTT caching bits set by the
+	 * kernel, as per the given object cache_level. This is troublesome
+	 * since the heavy flush we apply when first gathering the pages is
+	 * skipped if the kernel thinks the object is coherent with the GPU. As
+	 * a result it might be possible to bypass the cache and read the
+	 * contents of the page directly, which could be stale data. If it's
+	 * just a case of userspace shooting themselves in the foot then so be
+	 * it, but since i915 takes the stance of always zeroing memory before
+	 * handing it to userspace, we need to prevent this.
+	 */
+	return IS_JSL_EHL(i915);
+}
+
 static void i915_gem_close_object(struct drm_gem_object *gem, struct drm_file *file)
 {
 	struct drm_i915_gem_object *obj = to_intel_bo(gem);
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.h b/drivers/gpu/drm/i915/gem/i915_gem_object.h
index 9df3ee60604e..59201801cec5 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_object.h
+++ b/drivers/gpu/drm/i915/gem/i915_gem_object.h
@@ -514,6 +514,7 @@ i915_gem_object_finish_access(struct drm_i915_gem_object *obj)
 
 void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj,
 					 unsigned int cache_level);
+bool i915_gem_object_can_bypass_llc(struct drm_i915_gem_object *obj);
 void i915_gem_object_flush_if_display(struct drm_i915_gem_object *obj);
 void i915_gem_object_flush_if_display_locked(struct drm_i915_gem_object *obj);
 
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object_types.h b/drivers/gpu/drm/i915/gem/i915_gem_object_types.h
index 7c3da4e3e737..da85169006d4 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_object_types.h
+++ b/drivers/gpu/drm/i915/gem/i915_gem_object_types.h
@@ -427,6 +427,33 @@ struct drm_i915_gem_object {
 	 * can freely bypass the CPU cache when touching the pages with the GPU,
 	 * where the kernel is completely unaware. On such platform we need
 	 * apply the sledgehammer-on-acquire regardless of the @cache_coherent.
+	 *
+	 * Special care is taken on non-LLC platforms, to prevent potential
+	 * information leak. The driver currently ensures:
+	 *
+	 *   1. All userspace objects, by default, have @cache_level set as
+	 *   I915_CACHE_NONE. The only exception is userptr objects, where we
+	 *   instead force I915_CACHE_LLC, but we also don't allow userspace to
+	 *   ever change the @cache_level for such objects. Another special case
+	 *   is dma-buf, which doesn't rely on @cache_dirty,  but there we
+	 *   always do a forced flush when acquiring the pages, if there is a
+	 *   chance that the pages can be read directly from main memory with
+	 *   the GPU.
+	 *
+	 *   2. All I915_CACHE_NONE objects have @cache_dirty initially true.
+	 *
+	 *   3. All swapped-out objects(i.e shmem) have @cache_dirty set to
+	 *   true.
+	 *
+	 *   4. The @cache_dirty is never freely reset before the initial
+	 *   flush, even if userspace adjusts the @cache_level through the
+	 *   i915_gem_set_caching_ioctl.
+	 *
+	 *   5. All @cache_dirty objects(including swapped-in) are initially
+	 *   flushed with a synchronous call to drm_clflush_sg in
+	 *   __i915_gem_object_set_pages. The @cache_dirty can be freely reset
+	 *   at this point. All further asynchronous clfushes are never security
+	 *   critical, i.e userspace is free to race against itself.
 	 */
 	unsigned int cache_dirty:1;
 
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_shmem.c b/drivers/gpu/drm/i915/gem/i915_gem_shmem.c
index 11f072193f3b..d77da59fae04 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_shmem.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_shmem.c
@@ -182,22 +182,7 @@ rebuild_st:
 	if (i915_gem_object_needs_bit17_swizzle(obj))
 		i915_gem_object_do_bit_17_swizzle(obj, st);
 
-	/*
-	 * EHL and JSL add the 'Bypass LLC' MOCS entry, which should make it
-	 * possible for userspace to bypass the GTT caching bits set by the
-	 * kernel, as per the given object cache_level. This is troublesome
-	 * since the heavy flush we apply when first gathering the pages is
-	 * skipped if the kernel thinks the object is coherent with the GPU. As
-	 * a result it might be possible to bypass the cache and read the
-	 * contents of the page directly, which could be stale data. If it's
-	 * just a case of userspace shooting themselves in the foot then so be
-	 * it, but since i915 takes the stance of always zeroing memory before
-	 * handing it to userspace, we need to prevent this.
-	 *
-	 * By setting cache_dirty here we make the clflush in set_pages
-	 * unconditional on such platforms.
-	 */
-	if (IS_JSL_EHL(i915) && obj->flags & I915_BO_ALLOC_USER)
+	if (i915_gem_object_can_bypass_llc(obj))
 		obj->cache_dirty = true;
 
 	__i915_gem_object_set_pages(obj, st, sg_page_sizes);
@@ -301,6 +286,8 @@ __i915_gem_object_release_shmem(struct drm_i915_gem_object *obj,
 				struct sg_table *pages,
 				bool needs_clflush)
 {
+	struct drm_i915_private *i915 = to_i915(obj->base.dev);
+
 	GEM_BUG_ON(obj->mm.madv == __I915_MADV_PURGED);
 
 	if (obj->mm.madv == I915_MADV_DONTNEED)
@@ -312,6 +299,16 @@ __i915_gem_object_release_shmem(struct drm_i915_gem_object *obj,
 		drm_clflush_sg(pages);
 
 	__start_cpu_write(obj);
+	/*
+	 * On non-LLC platforms, force the flush-on-acquire if this is ever
+	 * swapped-in. Our async flush path is not trust worthy enough yet(and
+	 * happens in the wrong order), and with some tricks it's conceivable
+	 * for userspace to change the cache-level to I915_CACHE_NONE after the
+	 * pages are swapped-in, and since execbuf binds the object before doing
+	 * the async flush, we have a race window.
+	 */
+	if (!HAS_LLC(i915))
+		obj->cache_dirty = true;
 }
 
 void i915_gem_object_put_pages_shmem(struct drm_i915_gem_object *obj, struct sg_table *pages)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_userptr.c b/drivers/gpu/drm/i915/gem/i915_gem_userptr.c
index 8ea0fa665e53..3173c9f9a040 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_userptr.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_userptr.c
@@ -165,8 +165,11 @@ alloc_table:
 		goto err;
 	}
 
-	sg_page_sizes = i915_sg_dma_sizes(st->sgl);
+	WARN_ON_ONCE(!(obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_WRITE));
+	if (i915_gem_object_can_bypass_llc(obj))
+		obj->cache_dirty = true;
 
+	sg_page_sizes = i915_sg_dma_sizes(st->sgl);
 	__i915_gem_object_set_pages(obj, st, sg_page_sizes);
 
 	return 0;
@@ -546,7 +549,8 @@ i915_gem_userptr_ioctl(struct drm_device *dev,
 		return -ENOMEM;
 
 	drm_gem_private_object_init(dev, &obj->base, args->user_size);
-	i915_gem_object_init(obj, &i915_gem_userptr_ops, &lock_class, 0);
+	i915_gem_object_init(obj, &i915_gem_userptr_ops, &lock_class,
+			     I915_BO_ALLOC_USER);
 	obj->mem_flags = I915_BO_FLAG_STRUCT_PAGE;
 	obj->read_domains = I915_GEM_DOMAIN_CPU;
 	obj->write_domain = I915_GEM_DOMAIN_CPU;
diff --git a/drivers/gpu/drm/i915/gem/selftests/huge_pages.c b/drivers/gpu/drm/i915/gem/selftests/huge_pages.c
index 41d0680f3bd7..b2003133deaf 100644
--- a/drivers/gpu/drm/i915/gem/selftests/huge_pages.c
+++ b/drivers/gpu/drm/i915/gem/selftests/huge_pages.c
@@ -136,6 +136,8 @@ static void put_huge_pages(struct drm_i915_gem_object *obj,
 	huge_pages_free_pages(pages);
 
 	obj->mm.dirty = false;
+
+	__start_cpu_write(obj);
 }
 
 static const struct drm_i915_gem_object_ops huge_page_ops = {
@@ -152,6 +154,7 @@ huge_pages_object(struct drm_i915_private *i915,
 {
 	static struct lock_class_key lock_class;
 	struct drm_i915_gem_object *obj;
+	unsigned int cache_level;
 
 	GEM_BUG_ON(!size);
 	GEM_BUG_ON(!IS_ALIGNED(size, BIT(__ffs(page_mask))));
@@ -173,7 +176,9 @@ huge_pages_object(struct drm_i915_private *i915,
 
 	obj->write_domain = I915_GEM_DOMAIN_CPU;
 	obj->read_domains = I915_GEM_DOMAIN_CPU;
-	obj->cache_level = I915_CACHE_NONE;
+
+	cache_level = HAS_LLC(i915) ? I915_CACHE_LLC : I915_CACHE_NONE;
+	i915_gem_object_set_cache_coherency(obj, cache_level);
 
 	obj->mm.page_mask = page_mask;
 
diff --git a/drivers/gpu/drm/i915/gem/selftests/i915_gem_client_blt.c b/drivers/gpu/drm/i915/gem/selftests/i915_gem_client_blt.c
index ecbcbb86ae1e..8402ed925a69 100644
--- a/drivers/gpu/drm/i915/gem/selftests/i915_gem_client_blt.c
+++ b/drivers/gpu/drm/i915/gem/selftests/i915_gem_client_blt.c
@@ -17,13 +17,20 @@
 #include "huge_gem_object.h"
 #include "mock_context.h"
 
+enum client_tiling {
+	CLIENT_TILING_LINEAR,
+	CLIENT_TILING_X,
+	CLIENT_TILING_Y,
+	CLIENT_NUM_TILING_TYPES
+};
+
 #define WIDTH 512
 #define HEIGHT 32
 
 struct blit_buffer {
 	struct i915_vma *vma;
 	u32 start_val;
-	u32 tiling;
+	enum client_tiling tiling;
 };
 
 struct tiled_blits {
@@ -53,9 +60,9 @@ static int prepare_blit(const struct tiled_blits *t,
 	*cs++ = MI_LOAD_REGISTER_IMM(1);
 	*cs++ = i915_mmio_reg_offset(BCS_SWCTRL);
 	cmd = (BCS_SRC_Y | BCS_DST_Y) << 16;
-	if (src->tiling == I915_TILING_Y)
+	if (src->tiling == CLIENT_TILING_Y)
 		cmd |= BCS_SRC_Y;
-	if (dst->tiling == I915_TILING_Y)
+	if (dst->tiling == CLIENT_TILING_Y)
 		cmd |= BCS_DST_Y;
 	*cs++ = cmd;
 
@@ -172,7 +179,7 @@ static int tiled_blits_create_buffers(struct tiled_blits *t,
 
 		t->buffers[i].vma = vma;
 		t->buffers[i].tiling =
-			i915_prandom_u32_max_state(I915_TILING_Y + 1, prng);
+			i915_prandom_u32_max_state(CLIENT_TILING_Y + 1, prng);
 	}
 
 	return 0;
@@ -197,17 +204,17 @@ static u64 swizzle_bit(unsigned int bit, u64 offset)
 static u64 tiled_offset(const struct intel_gt *gt,
 			u64 v,
 			unsigned int stride,
-			unsigned int tiling)
+			enum client_tiling tiling)
 {
 	unsigned int swizzle;
 	u64 x, y;
 
-	if (tiling == I915_TILING_NONE)
+	if (tiling == CLIENT_TILING_LINEAR)
 		return v;
 
 	y = div64_u64_rem(v, stride, &x);
 
-	if (tiling == I915_TILING_X) {
+	if (tiling == CLIENT_TILING_X) {
 		v = div64_u64_rem(y, 8, &y) * stride * 8;
 		v += y * 512;
 		v += div64_u64_rem(x, 512, &x) << 12;
@@ -244,12 +251,12 @@ static u64 tiled_offset(const struct intel_gt *gt,
 	return v;
 }
 
-static const char *repr_tiling(int tiling)
+static const char *repr_tiling(enum client_tiling tiling)
 {
 	switch (tiling) {
-	case I915_TILING_NONE: return "linear";
-	case I915_TILING_X: return "X";
-	case I915_TILING_Y: return "Y";
+	case CLIENT_TILING_LINEAR: return "linear";
+	case CLIENT_TILING_X: return "X";
+	case CLIENT_TILING_Y: return "Y";
 	default: return "unknown";
 	}
 }
diff --git a/drivers/gpu/drm/i915/gt/intel_context.c b/drivers/gpu/drm/i915/gt/intel_context.c
index e9a0cad5c34d..488acd39ff67 100644
--- a/drivers/gpu/drm/i915/gt/intel_context.c
+++ b/drivers/gpu/drm/i915/gt/intel_context.c
@@ -240,6 +240,8 @@ int __intel_context_do_pin_ww(struct intel_context *ce,
 	if (err)
 		goto err_post_unpin;
 
+	intel_engine_pm_might_get(ce->engine);
+
 	if (unlikely(intel_context_is_closed(ce))) {
 		err = -ENOENT;
 		goto err_unlock;
@@ -362,8 +364,8 @@ static int __intel_context_active(struct i915_active *active)
 	return 0;
 }
 
-static int sw_fence_dummy_notify(struct i915_sw_fence *sf,
-				 enum i915_sw_fence_notify state)
+static int __i915_sw_fence_call
+sw_fence_dummy_notify(struct i915_sw_fence *sf, enum i915_sw_fence_notify state)
 {
 	return NOTIFY_DONE;
 }
@@ -399,6 +401,10 @@ intel_context_init(struct intel_context *ce, struct intel_engine_cs *engine)
 	ce->guc_id.id = GUC_INVALID_LRC_ID;
 	INIT_LIST_HEAD(&ce->guc_id.link);
 
+	INIT_LIST_HEAD(&ce->destroyed_link);
+
+	INIT_LIST_HEAD(&ce->parallel.child_list);
+
 	/*
 	 * Initialize fence to be complete as this is expected to be complete
 	 * unless there is a pending schedule disable outstanding.
@@ -413,10 +419,17 @@ intel_context_init(struct intel_context *ce, struct intel_engine_cs *engine)
 
 void intel_context_fini(struct intel_context *ce)
 {
+	struct intel_context *child, *next;
+
 	if (ce->timeline)
 		intel_timeline_put(ce->timeline);
 	i915_vm_put(ce->vm);
 
+	/* Need to put the creation ref for the children */
+	if (intel_context_is_parent(ce))
+		for_each_child_safe(ce, child, next)
+			intel_context_put(child);
+
 	mutex_destroy(&ce->pin_mutex);
 	i915_active_fini(&ce->active);
 	i915_sw_fence_fini(&ce->guc_state.blocked);
@@ -515,24 +528,53 @@ retry:
 
 struct i915_request *intel_context_find_active_request(struct intel_context *ce)
 {
+	struct intel_context *parent = intel_context_to_parent(ce);
 	struct i915_request *rq, *active = NULL;
 	unsigned long flags;
 
 	GEM_BUG_ON(!intel_engine_uses_guc(ce->engine));
 
-	spin_lock_irqsave(&ce->guc_state.lock, flags);
-	list_for_each_entry_reverse(rq, &ce->guc_state.requests,
+	/*
+	 * We search the parent list to find an active request on the submitted
+	 * context. The parent list contains the requests for all the contexts
+	 * in the relationship so we have to do a compare of each request's
+	 * context.
+	 */
+	spin_lock_irqsave(&parent->guc_state.lock, flags);
+	list_for_each_entry_reverse(rq, &parent->guc_state.requests,
 				    sched.link) {
+		if (rq->context != ce)
+			continue;
 		if (i915_request_completed(rq))
 			break;
 
 		active = rq;
 	}
-	spin_unlock_irqrestore(&ce->guc_state.lock, flags);
+	spin_unlock_irqrestore(&parent->guc_state.lock, flags);
 
 	return active;
 }
 
+void intel_context_bind_parent_child(struct intel_context *parent,
+				     struct intel_context *child)
+{
+	/*
+	 * Callers responsibility to validate that this function is used
+	 * correctly but we use GEM_BUG_ON here ensure that they do.
+	 */
+	GEM_BUG_ON(!intel_engine_uses_guc(parent->engine));
+	GEM_BUG_ON(intel_context_is_pinned(parent));
+	GEM_BUG_ON(intel_context_is_child(parent));
+	GEM_BUG_ON(intel_context_is_pinned(child));
+	GEM_BUG_ON(intel_context_is_child(child));
+	GEM_BUG_ON(intel_context_is_parent(child));
+
+	parent->parallel.child_index = parent->parallel.number_children++;
+	list_add_tail(&child->parallel.child_link,
+		      &parent->parallel.child_list);
+	child->parallel.parent = parent;
+}
+
 #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
 #include "selftest_context.c"
 #endif
diff --git a/drivers/gpu/drm/i915/gt/intel_context.h b/drivers/gpu/drm/i915/gt/intel_context.h
index c41098950746..246c37d72cd7 100644
--- a/drivers/gpu/drm/i915/gt/intel_context.h
+++ b/drivers/gpu/drm/i915/gt/intel_context.h
@@ -44,6 +44,54 @@ void intel_context_free(struct intel_context *ce);
 int intel_context_reconfigure_sseu(struct intel_context *ce,
 				   const struct intel_sseu sseu);
 
+#define PARENT_SCRATCH_SIZE	PAGE_SIZE
+
+static inline bool intel_context_is_child(struct intel_context *ce)
+{
+	return !!ce->parallel.parent;
+}
+
+static inline bool intel_context_is_parent(struct intel_context *ce)
+{
+	return !!ce->parallel.number_children;
+}
+
+static inline bool intel_context_is_pinned(struct intel_context *ce);
+
+static inline struct intel_context *
+intel_context_to_parent(struct intel_context *ce)
+{
+	if (intel_context_is_child(ce)) {
+		/*
+		 * The parent holds ref count to the child so it is always safe
+		 * for the parent to access the child, but the child has a
+		 * pointer to the parent without a ref. To ensure this is safe
+		 * the child should only access the parent pointer while the
+		 * parent is pinned.
+		 */
+		GEM_BUG_ON(!intel_context_is_pinned(ce->parallel.parent));
+
+		return ce->parallel.parent;
+	} else {
+		return ce;
+	}
+}
+
+static inline bool intel_context_is_parallel(struct intel_context *ce)
+{
+	return intel_context_is_child(ce) || intel_context_is_parent(ce);
+}
+
+void intel_context_bind_parent_child(struct intel_context *parent,
+				     struct intel_context *child);
+
+#define for_each_child(parent, ce)\
+	list_for_each_entry(ce, &(parent)->parallel.child_list,\
+			    parallel.child_link)
+#define for_each_child_safe(parent, ce, cn)\
+	list_for_each_entry_safe(ce, cn, &(parent)->parallel.child_list,\
+				 parallel.child_link)
+
 /**
  * intel_context_lock_pinned - Stablises the 'pinned' status of the HW context
  * @ce - the context
@@ -193,7 +241,13 @@ intel_context_timeline_lock(struct intel_context *ce)
 	struct intel_timeline *tl = ce->timeline;
 	int err;
 
-	err = mutex_lock_interruptible(&tl->mutex);
+	if (intel_context_is_parent(ce))
+		err = mutex_lock_interruptible_nested(&tl->mutex, 0);
+	else if (intel_context_is_child(ce))
+		err = mutex_lock_interruptible_nested(&tl->mutex,
+						      ce->parallel.child_index + 1);
+	else
+		err = mutex_lock_interruptible(&tl->mutex);
 	if (err)
 		return ERR_PTR(err);
 
diff --git a/drivers/gpu/drm/i915/gt/intel_context_types.h b/drivers/gpu/drm/i915/gt/intel_context_types.h
index 12252c411159..9e0177dc5484 100644
--- a/drivers/gpu/drm/i915/gt/intel_context_types.h
+++ b/drivers/gpu/drm/i915/gt/intel_context_types.h
@@ -55,9 +55,13 @@ struct intel_context_ops {
 	void (*reset)(struct intel_context *ce);
 	void (*destroy)(struct kref *kref);
 
-	/* virtual engine/context interface */
+	/* virtual/parallel engine/context interface */
 	struct intel_context *(*create_virtual)(struct intel_engine_cs **engine,
-						unsigned int count);
+						unsigned int count,
+						unsigned long flags);
+	struct intel_context *(*create_parallel)(struct intel_engine_cs **engines,
+						 unsigned int num_siblings,
+						 unsigned int width);
 	struct intel_engine_cs *(*get_sibling)(struct intel_engine_cs *engine,
 					       unsigned int sibling);
 };
@@ -113,6 +117,7 @@ struct intel_context {
 #define CONTEXT_NOPREEMPT		8
 #define CONTEXT_LRCA_DIRTY		9
 #define CONTEXT_GUC_INIT		10
+#define CONTEXT_PERMA_PIN		11
 
 	struct {
 		u64 timeout_us;
@@ -197,22 +202,80 @@ struct intel_context {
 	struct {
 		/**
 		 * @id: handle which is used to uniquely identify this context
-		 * with the GuC, protected by guc->contexts_lock
+		 * with the GuC, protected by guc->submission_state.lock
 		 */
 		u16 id;
 		/**
 		 * @ref: the number of references to the guc_id, when
 		 * transitioning in and out of zero protected by
-		 * guc->contexts_lock
+		 * guc->submission_state.lock
 		 */
 		atomic_t ref;
 		/**
 		 * @link: in guc->guc_id_list when the guc_id has no refs but is
-		 * still valid, protected by guc->contexts_lock
+		 * still valid, protected by guc->submission_state.lock
 		 */
 		struct list_head link;
 	} guc_id;
 
+	/**
+	 * @destroyed_link: link in guc->submission_state.destroyed_contexts, in
+	 * list when context is pending to be destroyed (deregistered with the
+	 * GuC), protected by guc->submission_state.lock
+	 */
+	struct list_head destroyed_link;
+
+	/** @parallel: sub-structure for parallel submission members */
+	struct {
+		union {
+			/**
+			 * @child_list: parent's list of children
+			 * contexts, no protection as immutable after context
+			 * creation
+			 */
+			struct list_head child_list;
+			/**
+			 * @child_link: child's link into parent's list of
+			 * children
+			 */
+			struct list_head child_link;
+		};
+		/** @parent: pointer to parent if child */
+		struct intel_context *parent;
+		/**
+		 * @last_rq: last request submitted on a parallel context, used
+		 * to insert submit fences between requests in the parallel
+		 * context
+		 */
+		struct i915_request *last_rq;
+		/**
+		 * @fence_context: fence context composite fence when doing
+		 * parallel submission
+		 */
+		u64 fence_context;
+		/**
+		 * @seqno: seqno for composite fence when doing parallel
+		 * submission
+		 */
+		u32 seqno;
+		/** @number_children: number of children if parent */
+		u8 number_children;
+		/** @child_index: index into child_list if child */
+		u8 child_index;
+		/** @guc: GuC specific members for parallel submission */
+		struct {
+			/** @wqi_head: head pointer in work queue */
+			u16 wqi_head;
+			/** @wqi_tail: tail pointer in work queue */
+			u16 wqi_tail;
+			/**
+			 * @parent_page: page in context state (ce->state) used
+			 * by parent for work queue, process descriptor
+			 */
+			u8 parent_page;
+		} guc;
+	} parallel;
+
 #ifdef CONFIG_DRM_I915_SELFTEST
 	/**
 	 * @drop_schedule_enable: Force drop of schedule enable G2H for selftest
diff --git a/drivers/gpu/drm/i915/gt/intel_engine.h b/drivers/gpu/drm/i915/gt/intel_engine.h
index 452248884ef1..08559ace0ada 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine.h
+++ b/drivers/gpu/drm/i915/gt/intel_engine.h
@@ -2,6 +2,7 @@
 #ifndef _INTEL_RINGBUFFER_H_
 #define _INTEL_RINGBUFFER_H_
 
+#include <asm/cacheflush.h>
 #include <drm/drm_util.h>
 
 #include <linux/hashtable.h>
@@ -281,9 +282,19 @@ intel_engine_has_preempt_reset(const struct intel_engine_cs *engine)
 	return intel_engine_has_preemption(engine);
 }
 
+#define FORCE_VIRTUAL	BIT(0)
 struct intel_context *
 intel_engine_create_virtual(struct intel_engine_cs **siblings,
-			    unsigned int count);
+			    unsigned int count, unsigned long flags);
+
+static inline struct intel_context *
+intel_engine_create_parallel(struct intel_engine_cs **engines,
+			     unsigned int num_engines,
+			     unsigned int width)
+{
+	GEM_BUG_ON(!engines[0]->cops->create_parallel);
+	return engines[0]->cops->create_parallel(engines, num_engines, width);
+}
 
 static inline bool
 intel_virtual_engine_has_heartbeat(const struct intel_engine_cs *engine)
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
index 2ae57e4656a3..ff6753ccb129 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
@@ -290,7 +290,8 @@ static void nop_irq_handler(struct intel_engine_cs *engine, u16 iir)
 	GEM_DEBUG_WARN_ON(iir);
 }
 
-static int intel_engine_setup(struct intel_gt *gt, enum intel_engine_id id)
+static int intel_engine_setup(struct intel_gt *gt, enum intel_engine_id id,
+			      u8 logical_instance)
 {
 	const struct engine_info *info = &intel_engines[id];
 	struct drm_i915_private *i915 = gt->i915;
@@ -335,6 +336,7 @@ static int intel_engine_setup(struct intel_gt *gt, enum intel_engine_id id)
 
 	engine->class = info->class;
 	engine->instance = info->instance;
+	engine->logical_mask = BIT(logical_instance);
 	__sprint_engine_name(engine);
 
 	engine->props.heartbeat_interval_ms =
@@ -588,6 +590,37 @@ static intel_engine_mask_t init_engine_mask(struct intel_gt *gt)
 	return info->engine_mask;
 }
 
+static void populate_logical_ids(struct intel_gt *gt, u8 *logical_ids,
+				 u8 class, const u8 *map, u8 num_instances)
+{
+	int i, j;
+	u8 current_logical_id = 0;
+
+	for (j = 0; j < num_instances; ++j) {
+		for (i = 0; i < ARRAY_SIZE(intel_engines); ++i) {
+			if (!HAS_ENGINE(gt, i) ||
+			    intel_engines[i].class != class)
+				continue;
+
+			if (intel_engines[i].instance == map[j]) {
+				logical_ids[intel_engines[i].instance] =
+					current_logical_id++;
+				break;
+			}
+		}
+	}
+}
+
+static void setup_logical_ids(struct intel_gt *gt, u8 *logical_ids, u8 class)
+{
+	int i;
+	u8 map[MAX_ENGINE_INSTANCE + 1];
+
+	for (i = 0; i < MAX_ENGINE_INSTANCE + 1; ++i)
+		map[i] = i;
+	populate_logical_ids(gt, logical_ids, class, map, ARRAY_SIZE(map));
+}
+
 /**
  * intel_engines_init_mmio() - allocate and prepare the Engine Command Streamers
  * @gt: pointer to struct intel_gt
@@ -599,7 +632,8 @@ int intel_engines_init_mmio(struct intel_gt *gt)
 	struct drm_i915_private *i915 = gt->i915;
 	const unsigned int engine_mask = init_engine_mask(gt);
 	unsigned int mask = 0;
-	unsigned int i;
+	unsigned int i, class;
+	u8 logical_ids[MAX_ENGINE_INSTANCE + 1];
 	int err;
 
 	drm_WARN_ON(&i915->drm, engine_mask == 0);
@@ -609,15 +643,23 @@ int intel_engines_init_mmio(struct intel_gt *gt)
 	if (i915_inject_probe_failure(i915))
 		return -ENODEV;
 
-	for (i = 0; i < ARRAY_SIZE(intel_engines); i++) {
-		if (!HAS_ENGINE(gt, i))
-			continue;
+	for (class = 0; class < MAX_ENGINE_CLASS + 1; ++class) {
+		setup_logical_ids(gt, logical_ids, class);
 
-		err = intel_engine_setup(gt, i);
-		if (err)
-			goto cleanup;
+		for (i = 0; i < ARRAY_SIZE(intel_engines); ++i) {
+			u8 instance = intel_engines[i].instance;
+
+			if (intel_engines[i].class != class ||
+			    !HAS_ENGINE(gt, i))
+				continue;
 
-		mask |= BIT(i);
+			err = intel_engine_setup(gt, i,
+						 logical_ids[instance]);
+			if (err)
+				goto cleanup;
+
+			mask |= BIT(i);
+		}
 	}
 
 	/*
@@ -1911,16 +1953,16 @@ ktime_t intel_engine_get_busy_time(struct intel_engine_cs *engine, ktime_t *now)
 
 struct intel_context *
 intel_engine_create_virtual(struct intel_engine_cs **siblings,
-			    unsigned int count)
+			    unsigned int count, unsigned long flags)
 {
 	if (count == 0)
 		return ERR_PTR(-EINVAL);
 
-	if (count == 1)
+	if (count == 1 && !(flags & FORCE_VIRTUAL))
 		return intel_context_create(siblings[0]);
 
 	GEM_BUG_ON(!siblings[0]->cops->create_virtual);
-	return siblings[0]->cops->create_virtual(siblings, count);
+	return siblings[0]->cops->create_virtual(siblings, count, flags);
 }
 
 struct i915_request *
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_pm.c b/drivers/gpu/drm/i915/gt/intel_engine_pm.c
index dacd62773735..a1334b48dde7 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_pm.c
+++ b/drivers/gpu/drm/i915/gt/intel_engine_pm.c
@@ -162,6 +162,19 @@ static bool switch_to_kernel_context(struct intel_engine_cs *engine)
 	unsigned long flags;
 	bool result = true;
 
+	/*
+	 * This is execlist specific behaviour intended to ensure the GPU is
+	 * idle by switching to a known 'safe' context. With GuC submission, the
+	 * same idle guarantee is achieved by other means (disabling
+	 * scheduling). Further, switching to a 'safe' context has no effect
+	 * with GuC submission as the scheduler can just switch back again.
+	 *
+	 * FIXME: Move this backend scheduler specific behaviour into the
+	 * scheduler backend.
+	 */
+	if (intel_engine_uses_guc(engine))
+		return true;
+
 	/* GPU is pointing to the void, as good as in the kernel context. */
 	if (intel_gt_is_wedged(engine->gt))
 		return true;
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_pm.h b/drivers/gpu/drm/i915/gt/intel_engine_pm.h
index 8520c595f5e1..d68675925b79 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_pm.h
+++ b/drivers/gpu/drm/i915/gt/intel_engine_pm.h
@@ -6,9 +6,11 @@
 #ifndef INTEL_ENGINE_PM_H
 #define INTEL_ENGINE_PM_H
 
+#include "i915_drv.h"
 #include "i915_request.h"
 #include "intel_engine_types.h"
 #include "intel_wakeref.h"
+#include "intel_gt_pm.h"
 
 static inline bool
 intel_engine_pm_is_awake(const struct intel_engine_cs *engine)
@@ -16,6 +18,11 @@ intel_engine_pm_is_awake(const struct intel_engine_cs *engine)
 	return intel_wakeref_is_active(&engine->wakeref);
 }
 
+static inline void __intel_engine_pm_get(struct intel_engine_cs *engine)
+{
+	__intel_wakeref_get(&engine->wakeref);
+}
+
 static inline void intel_engine_pm_get(struct intel_engine_cs *engine)
 {
 	intel_wakeref_get(&engine->wakeref);
@@ -26,6 +33,21 @@ static inline bool intel_engine_pm_get_if_awake(struct intel_engine_cs *engine)
 	return intel_wakeref_get_if_active(&engine->wakeref);
 }
 
+static inline void intel_engine_pm_might_get(struct intel_engine_cs *engine)
+{
+	if (!intel_engine_is_virtual(engine)) {
+		intel_wakeref_might_get(&engine->wakeref);
+	} else {
+		struct intel_gt *gt = engine->gt;
+		struct intel_engine_cs *tengine;
+		intel_engine_mask_t tmp, mask = engine->mask;
+
+		for_each_engine_masked(tengine, gt, mask, tmp)
+			intel_wakeref_might_get(&tengine->wakeref);
+	}
+	intel_gt_pm_might_get(engine->gt);
+}
+
 static inline void intel_engine_pm_put(struct intel_engine_cs *engine)
 {
 	intel_wakeref_put(&engine->wakeref);
@@ -47,6 +69,21 @@ static inline void intel_engine_pm_flush(struct intel_engine_cs *engine)
 	intel_wakeref_unlock_wait(&engine->wakeref);
 }
 
+static inline void intel_engine_pm_might_put(struct intel_engine_cs *engine)
+{
+	if (!intel_engine_is_virtual(engine)) {
+		intel_wakeref_might_put(&engine->wakeref);
+	} else {
+		struct intel_gt *gt = engine->gt;
+		struct intel_engine_cs *tengine;
+		intel_engine_mask_t tmp, mask = engine->mask;
+
+		for_each_engine_masked(tengine, gt, mask, tmp)
+			intel_wakeref_might_put(&tengine->wakeref);
+	}
+	intel_gt_pm_might_put(engine->gt);
+}
+
 static inline struct i915_request *
 intel_engine_create_kernel_request(struct intel_engine_cs *engine)
 {
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h b/drivers/gpu/drm/i915/gt/intel_engine_types.h
index 9167ce52487c..e0f773585c29 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_types.h
+++ b/drivers/gpu/drm/i915/gt/intel_engine_types.h
@@ -269,6 +269,13 @@ struct intel_engine_cs {
 	unsigned int guc_id;
 
 	intel_engine_mask_t mask;
+	/**
+	 * @logical_mask: logical mask of engine, reported to user space via
+	 * query IOCTL and used to communicate with the GuC in logical space.
+	 * The logical instance of a physical engine can change based on product
+	 * and fusing.
+	 */
+	intel_engine_mask_t logical_mask;
 
 	u8 class;
 	u8 instance;
diff --git a/drivers/gpu/drm/i915/gt/intel_execlists_submission.c b/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
index 73a79c2acd3a..bedb80057046 100644
--- a/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
+++ b/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
@@ -201,7 +201,8 @@ static struct virtual_engine *to_virtual_engine(struct intel_engine_cs *engine)
 }
 
 static struct intel_context *
-execlists_create_virtual(struct intel_engine_cs **siblings, unsigned int count);
+execlists_create_virtual(struct intel_engine_cs **siblings, unsigned int count,
+			 unsigned long flags);
 
 static struct i915_request *
 __active_request(const struct intel_timeline * const tl,
@@ -3784,7 +3785,8 @@ unlock:
 }
 
 static struct intel_context *
-execlists_create_virtual(struct intel_engine_cs **siblings, unsigned int count)
+execlists_create_virtual(struct intel_engine_cs **siblings, unsigned int count,
+			 unsigned long flags)
 {
 	struct virtual_engine *ve;
 	unsigned int n;
@@ -3877,6 +3879,7 @@ execlists_create_virtual(struct intel_engine_cs **siblings, unsigned int count)
 
 		ve->siblings[ve->num_siblings++] = sibling;
 		ve->base.mask |= sibling->mask;
+		ve->base.logical_mask |= sibling->logical_mask;
 
 		/*
 		 * All physical engines must be compatible for their emission
diff --git a/drivers/gpu/drm/i915/gt/intel_gt_debugfs.c b/drivers/gpu/drm/i915/gt/intel_gt_debugfs.c
index 1fe19ccd2794..f103664b71d4 100644
--- a/drivers/gpu/drm/i915/gt/intel_gt_debugfs.c
+++ b/drivers/gpu/drm/i915/gt/intel_gt_debugfs.c
@@ -13,6 +13,59 @@
 #include "pxp/intel_pxp_debugfs.h"
 #include "uc/intel_uc_debugfs.h"
 
+int intel_gt_debugfs_reset_show(struct intel_gt *gt, u64 *val)
+{
+	int ret = intel_gt_terminally_wedged(gt);
+
+	switch (ret) {
+	case -EIO:
+		*val = 1;
+		return 0;
+	case 0:
+		*val = 0;
+		return 0;
+	default:
+		return ret;
+	}
+}
+
+int intel_gt_debugfs_reset_store(struct intel_gt *gt, u64 val)
+{
+	/* Flush any previous reset before applying for a new one */
+	wait_event(gt->reset.queue,
+		   !test_bit(I915_RESET_BACKOFF, &gt->reset.flags));
+
+	intel_gt_handle_error(gt, val, I915_ERROR_CAPTURE,
+			      "Manually reset engine mask to %llx", val);
+	return 0;
+}
+
+/*
+ * keep the interface clean where the first parameter
+ * is a 'struct intel_gt *' instead of 'void *'
+ */
+static int __intel_gt_debugfs_reset_show(void *data, u64 *val)
+{
+	return intel_gt_debugfs_reset_show(data, val);
+}
+
+static int __intel_gt_debugfs_reset_store(void *data, u64 val)
+{
+	return intel_gt_debugfs_reset_store(data, val);
+}
+
+DEFINE_SIMPLE_ATTRIBUTE(reset_fops, __intel_gt_debugfs_reset_show,
+			__intel_gt_debugfs_reset_store, "%llu\n");
+
+static void gt_debugfs_register(struct intel_gt *gt, struct dentry *root)
+{
+	static const struct intel_gt_debugfs_file files[] = {
+		{ "reset", &reset_fops, NULL },
+	};
+
+	intel_gt_debugfs_register_files(root, files, ARRAY_SIZE(files), gt);
+}
+
 void intel_gt_debugfs_register(struct intel_gt *gt)
 {
 	struct dentry *root;
@@ -24,6 +77,8 @@ void intel_gt_debugfs_register(struct intel_gt *gt)
 	if (IS_ERR(root))
 		return;
 
+	gt_debugfs_register(gt, root);
+
 	intel_gt_engines_debugfs_register(gt, root);
 	intel_gt_pm_debugfs_register(gt, root);
 	intel_sseu_debugfs_register(gt, root);
diff --git a/drivers/gpu/drm/i915/gt/intel_gt_debugfs.h b/drivers/gpu/drm/i915/gt/intel_gt_debugfs.h
index 8b6fca09897c..e307ceb99031 100644
--- a/drivers/gpu/drm/i915/gt/intel_gt_debugfs.h
+++ b/drivers/gpu/drm/i915/gt/intel_gt_debugfs.h
@@ -35,4 +35,8 @@ void intel_gt_debugfs_register_files(struct dentry *root,
 				     const struct intel_gt_debugfs_file *files,
 				     unsigned long count, void *data);
 
+/* functions that need to be accessed by the upper level non-gt interfaces */
+int intel_gt_debugfs_reset_show(struct intel_gt *gt, u64 *val);
+int intel_gt_debugfs_reset_store(struct intel_gt *gt, u64 val);
+
 #endif /* INTEL_GT_DEBUGFS_H */
diff --git a/drivers/gpu/drm/i915/gt/intel_gt_pm.h b/drivers/gpu/drm/i915/gt/intel_gt_pm.h
index d0588d8aaa44..bc898df7a48c 100644
--- a/drivers/gpu/drm/i915/gt/intel_gt_pm.h
+++ b/drivers/gpu/drm/i915/gt/intel_gt_pm.h
@@ -31,6 +31,11 @@ static inline bool intel_gt_pm_get_if_awake(struct intel_gt *gt)
 	return intel_wakeref_get_if_active(&gt->wakeref);
 }
 
+static inline void intel_gt_pm_might_get(struct intel_gt *gt)
+{
+	intel_wakeref_might_get(&gt->wakeref);
+}
+
 static inline void intel_gt_pm_put(struct intel_gt *gt)
 {
 	intel_wakeref_put(&gt->wakeref);
@@ -41,6 +46,15 @@ static inline void intel_gt_pm_put_async(struct intel_gt *gt)
 	intel_wakeref_put_async(&gt->wakeref);
 }
 
+static inline void intel_gt_pm_might_put(struct intel_gt *gt)
+{
+	intel_wakeref_might_put(&gt->wakeref);
+}
+
+#define with_intel_gt_pm(gt, tmp) \
+	for (tmp = 1, intel_gt_pm_get(gt); tmp; \
+	     intel_gt_pm_put(gt), tmp = 0)
+
 static inline int intel_gt_pm_wait_for_idle(struct intel_gt *gt)
 {
 	return intel_wakeref_wait_for_idle(&gt->wakeref);
diff --git a/drivers/gpu/drm/i915/gt/intel_gt_pm_debugfs.c b/drivers/gpu/drm/i915/gt/intel_gt_pm_debugfs.c
index b3ddb9106125..404dfa7673c6 100644
--- a/drivers/gpu/drm/i915/gt/intel_gt_pm_debugfs.c
+++ b/drivers/gpu/drm/i915/gt/intel_gt_pm_debugfs.c
@@ -20,6 +20,46 @@
 #include "intel_uncore.h"
 #include "vlv_sideband.h"
 
+int intel_gt_pm_debugfs_forcewake_user_open(struct intel_gt *gt)
+{
+	atomic_inc(&gt->user_wakeref);
+	intel_gt_pm_get(gt);
+	if (GRAPHICS_VER(gt->i915) >= 6)
+		intel_uncore_forcewake_user_get(gt->uncore);
+
+	return 0;
+}
+
+int intel_gt_pm_debugfs_forcewake_user_release(struct intel_gt *gt)
+{
+	if (GRAPHICS_VER(gt->i915) >= 6)
+		intel_uncore_forcewake_user_put(gt->uncore);
+	intel_gt_pm_put(gt);
+	atomic_dec(&gt->user_wakeref);
+
+	return 0;
+}
+
+static int forcewake_user_open(struct inode *inode, struct file *file)
+{
+	struct intel_gt *gt = inode->i_private;
+
+	return intel_gt_pm_debugfs_forcewake_user_open(gt);
+}
+
+static int forcewake_user_release(struct inode *inode, struct file *file)
+{
+	struct intel_gt *gt = inode->i_private;
+
+	return intel_gt_pm_debugfs_forcewake_user_release(gt);
+}
+
+static const struct file_operations forcewake_user_fops = {
+	.owner = THIS_MODULE,
+	.open = forcewake_user_open,
+	.release = forcewake_user_release,
+};
+
 static int fw_domains_show(struct seq_file *m, void *data)
 {
 	struct intel_gt *gt = m->private;
@@ -628,6 +668,7 @@ void intel_gt_pm_debugfs_register(struct intel_gt *gt, struct dentry *root)
 		{ "drpc", &drpc_fops, NULL },
 		{ "frequency", &frequency_fops, NULL },
 		{ "forcewake", &fw_domains_fops, NULL },
+		{ "forcewake_user", &forcewake_user_fops, NULL},
 		{ "llc", &llc_fops, llc_eval },
 		{ "rps_boost", &rps_boost_fops, rps_eval },
 	};
diff --git a/drivers/gpu/drm/i915/gt/intel_gt_pm_debugfs.h b/drivers/gpu/drm/i915/gt/intel_gt_pm_debugfs.h
index 2b824289582b..a8457887ec65 100644
--- a/drivers/gpu/drm/i915/gt/intel_gt_pm_debugfs.h
+++ b/drivers/gpu/drm/i915/gt/intel_gt_pm_debugfs.h
@@ -13,4 +13,8 @@ struct drm_printer;
 void intel_gt_pm_debugfs_register(struct intel_gt *gt, struct dentry *root);
 void intel_gt_pm_frequency_dump(struct intel_gt *gt, struct drm_printer *m);
 
+/* functions that need to be accessed by the upper level non-gt interfaces */
+int intel_gt_pm_debugfs_forcewake_user_open(struct intel_gt *gt);
+int intel_gt_pm_debugfs_forcewake_user_release(struct intel_gt *gt);
+
 #endif /* INTEL_GT_PM_DEBUGFS_H */
diff --git a/drivers/gpu/drm/i915/gt/intel_llc.c b/drivers/gpu/drm/i915/gt/intel_llc.c
index 9ca88f8ac5dd..08d7d5ae263a 100644
--- a/drivers/gpu/drm/i915/gt/intel_llc.c
+++ b/drivers/gpu/drm/i915/gt/intel_llc.c
@@ -3,6 +3,7 @@
  * Copyright © 2019 Intel Corporation
  */
 
+#include <asm/tsc.h>
 #include <linux/cpufreq.h>
 
 #include "i915_drv.h"
diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c b/drivers/gpu/drm/i915/gt/intel_lrc.c
index 3ef9eaf8c50e..56156cf18c41 100644
--- a/drivers/gpu/drm/i915/gt/intel_lrc.c
+++ b/drivers/gpu/drm/i915/gt/intel_lrc.c
@@ -942,6 +942,11 @@ __lrc_alloc_state(struct intel_context *ce, struct intel_engine_cs *engine)
 		context_size += PAGE_SIZE;
 	}
 
+	if (intel_context_is_parent(ce) && intel_engine_uses_guc(engine)) {
+		ce->parallel.guc.parent_page = context_size / PAGE_SIZE;
+		context_size += PARENT_SCRATCH_SIZE;
+	}
+
 	obj = i915_gem_object_create_lmem(engine->i915, context_size,
 					  I915_BO_ALLOC_PM_VOLATILE);
 	if (IS_ERR(obj))
diff --git a/drivers/gpu/drm/i915/gt/intel_ring_submission.c b/drivers/gpu/drm/i915/gt/intel_ring_submission.c
index 593524195707..586dca1731ce 100644
--- a/drivers/gpu/drm/i915/gt/intel_ring_submission.c
+++ b/drivers/gpu/drm/i915/gt/intel_ring_submission.c
@@ -292,7 +292,7 @@ static void xcs_sanitize(struct intel_engine_cs *engine)
 	sanitize_hwsp(engine);
 
 	/* And scrub the dirty cachelines for the HWSP */
-	clflush_cache_range(engine->status_page.addr, PAGE_SIZE);
+	drm_clflush_virt_range(engine->status_page.addr, PAGE_SIZE);
 
 	intel_engine_reset_pinned_contexts(engine);
 }
diff --git a/drivers/gpu/drm/i915/gt/intel_timeline.c b/drivers/gpu/drm/i915/gt/intel_timeline.c
index 1257f4f11e66..438bbc7b8147 100644
--- a/drivers/gpu/drm/i915/gt/intel_timeline.c
+++ b/drivers/gpu/drm/i915/gt/intel_timeline.c
@@ -64,7 +64,7 @@ intel_timeline_pin_map(struct intel_timeline *timeline)
 
 	timeline->hwsp_map = vaddr;
 	timeline->hwsp_seqno = memset(vaddr + ofs, 0, TIMELINE_SEQNO_BYTES);
-	clflush(vaddr + ofs);
+	drm_clflush_virt_range(vaddr + ofs, TIMELINE_SEQNO_BYTES);
 
 	return 0;
 }
@@ -225,7 +225,7 @@ void intel_timeline_reset_seqno(const struct intel_timeline *tl)
 
 	memset(hwsp_seqno + 1, 0, TIMELINE_SEQNO_BYTES - sizeof(*hwsp_seqno));
 	WRITE_ONCE(*hwsp_seqno, tl->seqno);
-	clflush(hwsp_seqno);
+	drm_clflush_virt_range(hwsp_seqno, TIMELINE_SEQNO_BYTES);
 }
 
 void intel_timeline_enter(struct intel_timeline *tl)
diff --git a/drivers/gpu/drm/i915/gt/selftest_execlists.c b/drivers/gpu/drm/i915/gt/selftest_execlists.c
index 25a8c4f62b0d..b367ecfa42de 100644
--- a/drivers/gpu/drm/i915/gt/selftest_execlists.c
+++ b/drivers/gpu/drm/i915/gt/selftest_execlists.c
@@ -3733,7 +3733,7 @@ static int nop_virtual_engine(struct intel_gt *gt,
 	GEM_BUG_ON(!nctx || nctx > ARRAY_SIZE(ve));
 
 	for (n = 0; n < nctx; n++) {
-		ve[n] = intel_engine_create_virtual(siblings, nsibling);
+		ve[n] = intel_engine_create_virtual(siblings, nsibling, 0);
 		if (IS_ERR(ve[n])) {
 			err = PTR_ERR(ve[n]);
 			nctx = n;
@@ -3929,7 +3929,7 @@ static int mask_virtual_engine(struct intel_gt *gt,
 	 * restrict it to our desired engine within the virtual engine.
 	 */
 
-	ve = intel_engine_create_virtual(siblings, nsibling);
+	ve = intel_engine_create_virtual(siblings, nsibling, 0);
 	if (IS_ERR(ve)) {
 		err = PTR_ERR(ve);
 		goto out_close;
@@ -4060,7 +4060,7 @@ static int slicein_virtual_engine(struct intel_gt *gt,
 		i915_request_add(rq);
 	}
 
-	ce = intel_engine_create_virtual(siblings, nsibling);
+	ce = intel_engine_create_virtual(siblings, nsibling, 0);
 	if (IS_ERR(ce)) {
 		err = PTR_ERR(ce);
 		goto out;
@@ -4112,7 +4112,7 @@ static int sliceout_virtual_engine(struct intel_gt *gt,
 
 	/* XXX We do not handle oversubscription and fairness with normal rq */
 	for (n = 0; n < nsibling; n++) {
-		ce = intel_engine_create_virtual(siblings, nsibling);
+		ce = intel_engine_create_virtual(siblings, nsibling, 0);
 		if (IS_ERR(ce)) {
 			err = PTR_ERR(ce);
 			goto out;
@@ -4214,7 +4214,7 @@ static int preserved_virtual_engine(struct intel_gt *gt,
 	if (err)
 		goto out_scratch;
 
-	ve = intel_engine_create_virtual(siblings, nsibling);
+	ve = intel_engine_create_virtual(siblings, nsibling, 0);
 	if (IS_ERR(ve)) {
 		err = PTR_ERR(ve);
 		goto out_scratch;
@@ -4354,7 +4354,7 @@ static int reset_virtual_engine(struct intel_gt *gt,
 	if (igt_spinner_init(&spin, gt))
 		return -ENOMEM;
 
-	ve = intel_engine_create_virtual(siblings, nsibling);
+	ve = intel_engine_create_virtual(siblings, nsibling, 0);
 	if (IS_ERR(ve)) {
 		err = PTR_ERR(ve);
 		goto out_spin;
diff --git a/drivers/gpu/drm/i915/gt/uc/abi/guc_actions_abi.h b/drivers/gpu/drm/i915/gt/uc/abi/guc_actions_abi.h
index 8ff582222aff..ba10bd374cee 100644
--- a/drivers/gpu/drm/i915/gt/uc/abi/guc_actions_abi.h
+++ b/drivers/gpu/drm/i915/gt/uc/abi/guc_actions_abi.h
@@ -142,6 +142,7 @@ enum intel_guc_action {
 	INTEL_GUC_ACTION_REGISTER_COMMAND_TRANSPORT_BUFFER = 0x4505,
 	INTEL_GUC_ACTION_DEREGISTER_COMMAND_TRANSPORT_BUFFER = 0x4506,
 	INTEL_GUC_ACTION_DEREGISTER_CONTEXT_DONE = 0x4600,
+	INTEL_GUC_ACTION_REGISTER_CONTEXT_MULTI_LRC = 0x4601,
 	INTEL_GUC_ACTION_RESET_CLIENT = 0x5507,
 	INTEL_GUC_ACTION_LIMIT
 };
diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc.c b/drivers/gpu/drm/i915/gt/uc/intel_guc.c
index 8f8182bf7c11..6e228343e8cb 100644
--- a/drivers/gpu/drm/i915/gt/uc/intel_guc.c
+++ b/drivers/gpu/drm/i915/gt/uc/intel_guc.c
@@ -756,3 +756,32 @@ void intel_guc_load_status(struct intel_guc *guc, struct drm_printer *p)
 		}
 	}
 }
+
+void intel_guc_write_barrier(struct intel_guc *guc)
+{
+	struct intel_gt *gt = guc_to_gt(guc);
+
+	if (i915_gem_object_is_lmem(guc->ct.vma->obj)) {
+		/*
+		 * Ensure intel_uncore_write_fw can be used rather than
+		 * intel_uncore_write.
+		 */
+		GEM_BUG_ON(guc->send_regs.fw_domains);
+
+		/*
+		 * This register is used by the i915 and GuC for MMIO based
+		 * communication. Once we are in this code CTBs are the only
+		 * method the i915 uses to communicate with the GuC so it is
+		 * safe to write to this register (a value of 0 is NOP for MMIO
+		 * communication). If we ever start mixing CTBs and MMIOs a new
+		 * register will have to be chosen. This function is also used
+		 * to enforce ordering of a work queue item write and an update
+		 * to the process descriptor. When a work queue is being used,
+		 * CTBs are also the only mechanism of communication.
+		 */
+		intel_uncore_write_fw(gt->uncore, GEN11_SOFT_SCRATCH(0), 0);
+	} else {
+		/* wmb() sufficient for a barrier if in smem */
+		wmb();
+	}
+}
diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc.h b/drivers/gpu/drm/i915/gt/uc/intel_guc.h
index 5dd174babf7a..31cf9fb48c7e 100644
--- a/drivers/gpu/drm/i915/gt/uc/intel_guc.h
+++ b/drivers/gpu/drm/i915/gt/uc/intel_guc.h
@@ -46,6 +46,15 @@ struct intel_guc {
 	 * submitted until the stalled request is processed.
 	 */
 	struct i915_request *stalled_request;
+	/**
+	 * @submission_stall_reason: reason why submission is stalled
+	 */
+	enum {
+		STALL_NONE,
+		STALL_REGISTER_CONTEXT,
+		STALL_MOVE_LRC_TAIL,
+		STALL_ADD_REQUEST,
+	} submission_stall_reason;
 
 	/* intel_guc_recv interrupt related state */
 	/** @irq_lock: protects GuC irq state */
@@ -71,16 +80,41 @@ struct intel_guc {
 	} interrupts;
 
 	/**
-	 * @contexts_lock: protects guc_ids, guc_id_list, ce->guc_id.id, and
-	 * ce->guc_id.ref when transitioning in and out of zero
-	 */
-	spinlock_t contexts_lock;
-	/** @guc_ids: used to allocate unique ce->guc_id.id values */
-	struct ida guc_ids;
-	/**
-	 * @guc_id_list: list of intel_context with valid guc_ids but no refs
+	 * @submission_state: sub-structure for submission state protected by
+	 * single lock
 	 */
-	struct list_head guc_id_list;
+	struct {
+		/**
+		 * @lock: protects everything in submission_state,
+		 * ce->guc_id.id, and ce->guc_id.ref when transitioning in and
+		 * out of zero
+		 */
+		spinlock_t lock;
+		/**
+		 * @guc_ids: used to allocate new guc_ids, single-lrc
+		 */
+		struct ida guc_ids;
+		/**
+		 * @guc_ids_bitmap: used to allocate new guc_ids, multi-lrc
+		 */
+		unsigned long *guc_ids_bitmap;
+		/**
+		 * @guc_id_list: list of intel_context with valid guc_ids but no
+		 * refs
+		 */
+		struct list_head guc_id_list;
+		/**
+		 * @destroyed_contexts: list of contexts waiting to be destroyed
+		 * (deregistered with the GuC)
+		 */
+		struct list_head destroyed_contexts;
+		/**
+		 * @destroyed_worker: worker to deregister contexts, need as we
+		 * need to take a GT PM reference and can't from destroy
+		 * function as it might be in an atomic context (no sleeping)
+		 */
+		struct work_struct destroyed_worker;
+	} submission_state;
 
 	/**
 	 * @submission_supported: tracks whether we support GuC submission on
@@ -342,4 +376,6 @@ void intel_guc_submission_cancel_requests(struct intel_guc *guc);
 
 void intel_guc_load_status(struct intel_guc *guc, struct drm_printer *p);
 
+void intel_guc_write_barrier(struct intel_guc *guc);
+
 #endif
diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_ads.c b/drivers/gpu/drm/i915/gt/uc/intel_guc_ads.c
index 2c6ea64af7ec..621c893a009f 100644
--- a/drivers/gpu/drm/i915/gt/uc/intel_guc_ads.c
+++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_ads.c
@@ -176,7 +176,7 @@ static void guc_mapping_table_init(struct intel_gt *gt,
 	for_each_engine(engine, gt, id) {
 		u8 guc_class = engine_class_to_guc_class(engine->class);
 
-		system_info->mapping_table[guc_class][engine->instance] =
+		system_info->mapping_table[guc_class][ilog2(engine->logical_mask)] =
 			engine->instance;
 	}
 }
diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c b/drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c
index 0a3504bc0b61..a0cc34be7b56 100644
--- a/drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c
+++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c
@@ -383,28 +383,6 @@ static u32 ct_get_next_fence(struct intel_guc_ct *ct)
 	return ++ct->requests.last_fence;
 }
 
-static void write_barrier(struct intel_guc_ct *ct)
-{
-	struct intel_guc *guc = ct_to_guc(ct);
-	struct intel_gt *gt = guc_to_gt(guc);
-
-	if (i915_gem_object_is_lmem(guc->ct.vma->obj)) {
-		GEM_BUG_ON(guc->send_regs.fw_domains);
-		/*
-		 * This register is used by the i915 and GuC for MMIO based
-		 * communication. Once we are in this code CTBs are the only
-		 * method the i915 uses to communicate with the GuC so it is
-		 * safe to write to this register (a value of 0 is NOP for MMIO
-		 * communication). If we ever start mixing CTBs and MMIOs a new
-		 * register will have to be chosen.
-		 */
-		intel_uncore_write_fw(gt->uncore, GEN11_SOFT_SCRATCH(0), 0);
-	} else {
-		/* wmb() sufficient for a barrier if in smem */
-		wmb();
-	}
-}
-
 static int ct_write(struct intel_guc_ct *ct,
 		    const u32 *action,
 		    u32 len /* in dwords */,
@@ -474,7 +452,7 @@ static int ct_write(struct intel_guc_ct *ct,
 	 * make sure H2G buffer update and LRC tail update (if this triggering a
 	 * submission) are visible before updating the descriptor tail
 	 */
-	write_barrier(ct);
+	intel_guc_write_barrier(ct_to_guc(ct));
 
 	/* update local copies */
 	ctb->tail = tail;
diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_fwif.h b/drivers/gpu/drm/i915/gt/uc/intel_guc_fwif.h
index fa4be13c8854..722933e26347 100644
--- a/drivers/gpu/drm/i915/gt/uc/intel_guc_fwif.h
+++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_fwif.h
@@ -52,27 +52,27 @@
 
 #define GUC_DOORBELL_INVALID		256
 
-#define GUC_WQ_SIZE			(PAGE_SIZE * 2)
-
-/* Work queue item header definitions */
+/*
+ * Work queue item header definitions
+ *
+ * Work queue is circular buffer used to submit complex (multi-lrc) submissions
+ * to the GuC. A work queue item is an entry in the circular buffer.
+ */
 #define WQ_STATUS_ACTIVE		1
 #define WQ_STATUS_SUSPENDED		2
 #define WQ_STATUS_CMD_ERROR		3
 #define WQ_STATUS_ENGINE_ID_NOT_USED	4
 #define WQ_STATUS_SUSPENDED_FROM_RESET	5
-#define WQ_TYPE_SHIFT			0
-#define   WQ_TYPE_BATCH_BUF		(0x1 << WQ_TYPE_SHIFT)
-#define   WQ_TYPE_PSEUDO		(0x2 << WQ_TYPE_SHIFT)
-#define   WQ_TYPE_INORDER		(0x3 << WQ_TYPE_SHIFT)
-#define   WQ_TYPE_NOOP			(0x4 << WQ_TYPE_SHIFT)
-#define WQ_TARGET_SHIFT			10
-#define WQ_LEN_SHIFT			16
-#define WQ_NO_WCFLUSH_WAIT		(1 << 27)
-#define WQ_PRESENT_WORKLOAD		(1 << 28)
-
-#define WQ_RING_TAIL_SHIFT		20
-#define WQ_RING_TAIL_MAX		0x7FF	/* 2^11 QWords */
-#define WQ_RING_TAIL_MASK		(WQ_RING_TAIL_MAX << WQ_RING_TAIL_SHIFT)
+#define WQ_TYPE_BATCH_BUF		0x1
+#define WQ_TYPE_PSEUDO			0x2
+#define WQ_TYPE_INORDER			0x3
+#define WQ_TYPE_NOOP			0x4
+#define WQ_TYPE_MULTI_LRC		0x5
+#define WQ_TYPE_MASK			GENMASK(7, 0)
+#define WQ_LEN_MASK			GENMASK(26, 16)
+
+#define WQ_GUC_ID_MASK			GENMASK(15, 0)
+#define WQ_RING_TAIL_MASK		GENMASK(28, 18)
 
 #define GUC_STAGE_DESC_ATTR_ACTIVE	BIT(0)
 #define GUC_STAGE_DESC_ATTR_PENDING_DB	BIT(1)
@@ -186,7 +186,7 @@ struct guc_process_desc {
 	u32 wq_status;
 	u32 engine_presence;
 	u32 priority;
-	u32 reserved[30];
+	u32 reserved[36];
 } __packed;
 
 #define CONTEXT_REGISTRATION_FLAG_KMD	BIT(0)
diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
index ba0de35f6323..d7710debcd47 100644
--- a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
+++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
@@ -11,6 +11,7 @@
 #include "gt/intel_context.h"
 #include "gt/intel_engine_pm.h"
 #include "gt/intel_engine_heartbeat.h"
+#include "gt/intel_gpu_commands.h"
 #include "gt/intel_gt.h"
 #include "gt/intel_gt_irq.h"
 #include "gt/intel_gt_pm.h"
@@ -68,7 +69,7 @@
  * fence is used to stall all requests associated with this guc_id until the
  * corresponding G2H returns indicating the guc_id has been deregistered.
  *
- * guc_ids:
+ * submission_state.guc_ids:
  * Unique number associated with private GuC context data passed in during
  * context registration / submission / deregistration. 64k available. Simple ida
  * is used for allocation.
@@ -89,9 +90,9 @@
  * sched_engine can be submitting at a time. Currently only one sched_engine is
  * used for all of GuC submission but that could change in the future.
  *
- * guc->contexts_lock
- * Protects guc_id allocation for the given GuC, i.e. only one context can be
- * doing guc_id allocation operations at a time for each GuC in the system.
+ * guc->submission_state.lock
+ * Global lock for GuC submission state. Protects guc_ids and destroyed contexts
+ * list.
  *
  * ce->guc_state.lock
  * Protects everything under ce->guc_state. Ensures that a context is in the
@@ -103,7 +104,7 @@
  *
  * Lock ordering rules:
  * sched_engine->lock -> ce->guc_state.lock
- * guc->contexts_lock -> ce->guc_state.lock
+ * guc->submission_state.lock -> ce->guc_state.lock
  *
  * Reset races:
  * When a full GT reset is triggered it is assumed that some G2H responses to
@@ -124,11 +125,27 @@ struct guc_virtual_engine {
 };
 
 static struct intel_context *
-guc_create_virtual(struct intel_engine_cs **siblings, unsigned int count);
+guc_create_virtual(struct intel_engine_cs **siblings, unsigned int count,
+		   unsigned long flags);
+
+static struct intel_context *
+guc_create_parallel(struct intel_engine_cs **engines,
+		    unsigned int num_siblings,
+		    unsigned int width);
 
 #define GUC_REQUEST_SIZE 64 /* bytes */
 
 /*
+ * We reserve 1/16 of the guc_ids for multi-lrc as these need to be contiguous
+ * per the GuC submission interface. A different allocation algorithm is used
+ * (bitmap vs. ida) between multi-lrc and single-lrc hence the reason to
+ * partition the guc_id space. We believe the number of multi-lrc contexts in
+ * use should be low and 1/16 should be sufficient. Minimum of 32 guc_ids for
+ * multi-lrc.
+ */
+#define NUMBER_MULTI_LRC_GUC_ID		(GUC_MAX_LRC_DESCRIPTORS / 16)
+
+/*
  * Below is a set of functions which control the GuC scheduling state which
  * require a lock.
  */
@@ -324,6 +341,12 @@ static inline void decr_context_committed_requests(struct intel_context *ce)
 	GEM_BUG_ON(ce->guc_state.number_committed_requests < 0);
 }
 
+static struct intel_context *
+request_to_scheduling_context(struct i915_request *rq)
+{
+	return intel_context_to_parent(rq->context);
+}
+
 static inline bool context_guc_id_invalid(struct intel_context *ce)
 {
 	return ce->guc_id.id == GUC_INVALID_LRC_ID;
@@ -344,6 +367,104 @@ static inline struct i915_priolist *to_priolist(struct rb_node *rb)
 	return rb_entry(rb, struct i915_priolist, node);
 }
 
+/*
+ * When using multi-lrc submission a scratch memory area is reserved in the
+ * parent's context state for the process descriptor, work queue, and handshake
+ * between the parent + children contexts to insert safe preemption points
+ * between each of the BBs. Currently the scratch area is sized to a page.
+ *
+ * The layout of this scratch area is below:
+ * 0						guc_process_desc
+ * + sizeof(struct guc_process_desc)		child go
+ * + CACHELINE_BYTES				child join[0]
+ * ...
+ * + CACHELINE_BYTES				child join[n - 1]
+ * ...						unused
+ * PARENT_SCRATCH_SIZE / 2			work queue start
+ * ...						work queue
+ * PARENT_SCRATCH_SIZE - 1			work queue end
+ */
+#define WQ_SIZE			(PARENT_SCRATCH_SIZE / 2)
+#define WQ_OFFSET		(PARENT_SCRATCH_SIZE - WQ_SIZE)
+
+struct sync_semaphore {
+	u32 semaphore;
+	u8 unused[CACHELINE_BYTES - sizeof(u32)];
+};
+
+struct parent_scratch {
+	struct guc_process_desc pdesc;
+
+	struct sync_semaphore go;
+	struct sync_semaphore join[MAX_ENGINE_INSTANCE + 1];
+
+	u8 unused[WQ_OFFSET - sizeof(struct guc_process_desc) -
+		sizeof(struct sync_semaphore) * (MAX_ENGINE_INSTANCE + 2)];
+
+	u32 wq[WQ_SIZE / sizeof(u32)];
+};
+
+static u32 __get_parent_scratch_offset(struct intel_context *ce)
+{
+	GEM_BUG_ON(!ce->parallel.guc.parent_page);
+
+	return ce->parallel.guc.parent_page * PAGE_SIZE;
+}
+
+static u32 __get_wq_offset(struct intel_context *ce)
+{
+	BUILD_BUG_ON(offsetof(struct parent_scratch, wq) != WQ_OFFSET);
+
+	return __get_parent_scratch_offset(ce) + WQ_OFFSET;
+}
+
+static struct parent_scratch *
+__get_parent_scratch(struct intel_context *ce)
+{
+	BUILD_BUG_ON(sizeof(struct parent_scratch) != PARENT_SCRATCH_SIZE);
+	BUILD_BUG_ON(sizeof(struct sync_semaphore) != CACHELINE_BYTES);
+
+	/*
+	 * Need to subtract LRC_STATE_OFFSET here as the
+	 * parallel.guc.parent_page is the offset into ce->state while
+	 * ce->lrc_reg_reg is ce->state + LRC_STATE_OFFSET.
+	 */
+	return (struct parent_scratch *)
+		(ce->lrc_reg_state +
+		 ((__get_parent_scratch_offset(ce) -
+		   LRC_STATE_OFFSET) / sizeof(u32)));
+}
+
+static struct guc_process_desc *
+__get_process_desc(struct intel_context *ce)
+{
+	struct parent_scratch *ps = __get_parent_scratch(ce);
+
+	return &ps->pdesc;
+}
+
+static u32 *get_wq_pointer(struct guc_process_desc *desc,
+			   struct intel_context *ce,
+			   u32 wqi_size)
+{
+	/*
+	 * Check for space in work queue. Caching a value of head pointer in
+	 * intel_context structure in order reduce the number accesses to shared
+	 * GPU memory which may be across a PCIe bus.
+	 */
+#define AVAILABLE_SPACE	\
+	CIRC_SPACE(ce->parallel.guc.wqi_tail, ce->parallel.guc.wqi_head, WQ_SIZE)
+	if (wqi_size > AVAILABLE_SPACE) {
+		ce->parallel.guc.wqi_head = READ_ONCE(desc->head);
+
+		if (wqi_size > AVAILABLE_SPACE)
+			return NULL;
+	}
+#undef AVAILABLE_SPACE
+
+	return &__get_parent_scratch(ce)->wq[ce->parallel.guc.wqi_tail / sizeof(u32)];
+}
+
 static struct guc_lrc_desc *__get_lrc_desc(struct intel_guc *guc, u32 index)
 {
 	struct guc_lrc_desc *base = guc->lrc_desc_pool_vaddr;
@@ -503,10 +624,10 @@ int intel_guc_wait_for_idle(struct intel_guc *guc, long timeout)
 
 static int guc_lrc_desc_pin(struct intel_context *ce, bool loop);
 
-static int guc_add_request(struct intel_guc *guc, struct i915_request *rq)
+static int __guc_add_request(struct intel_guc *guc, struct i915_request *rq)
 {
 	int err = 0;
-	struct intel_context *ce = rq->context;
+	struct intel_context *ce = request_to_scheduling_context(rq);
 	u32 action[3];
 	int len = 0;
 	u32 g2h_len_dw = 0;
@@ -527,26 +648,17 @@ static int guc_add_request(struct intel_guc *guc, struct i915_request *rq)
 	GEM_BUG_ON(!atomic_read(&ce->guc_id.ref));
 	GEM_BUG_ON(context_guc_id_invalid(ce));
 
-	/*
-	 * Corner case where the GuC firmware was blown away and reloaded while
-	 * this context was pinned.
-	 */
-	if (unlikely(!lrc_desc_registered(guc, ce->guc_id.id))) {
-		err = guc_lrc_desc_pin(ce, false);
-		if (unlikely(err))
-			return err;
-	}
-
 	spin_lock(&ce->guc_state.lock);
 
 	/*
 	 * The request / context will be run on the hardware when scheduling
-	 * gets enabled in the unblock.
+	 * gets enabled in the unblock. For multi-lrc we still submit the
+	 * context to move the LRC tails.
 	 */
-	if (unlikely(context_blocked(ce)))
+	if (unlikely(context_blocked(ce) && !intel_context_is_parent(ce)))
 		goto out;
 
-	enabled = context_enabled(ce);
+	enabled = context_enabled(ce) || context_blocked(ce);
 
 	if (!enabled) {
 		action[len++] = INTEL_GUC_ACTION_SCHED_CONTEXT_MODE_SET;
@@ -565,6 +677,18 @@ static int guc_add_request(struct intel_guc *guc, struct i915_request *rq)
 		trace_intel_context_sched_enable(ce);
 		atomic_inc(&guc->outstanding_submission_g2h);
 		set_context_enabled(ce);
+
+		/*
+		 * Without multi-lrc KMD does the submission step (moving the
+		 * lrc tail) so enabling scheduling is sufficient to submit the
+		 * context. This isn't the case in multi-lrc submission as the
+		 * GuC needs to move the tails, hence the need for another H2G
+		 * to submit a multi-lrc context after enabling scheduling.
+		 */
+		if (intel_context_is_parent(ce)) {
+			action[0] = INTEL_GUC_ACTION_SCHED_CONTEXT;
+			err = intel_guc_send_nb(guc, action, len - 1, 0);
+		}
 	} else if (!enabled) {
 		clr_context_pending_enable(ce);
 		intel_context_put(ce);
@@ -577,6 +701,18 @@ out:
 	return err;
 }
 
+static int guc_add_request(struct intel_guc *guc, struct i915_request *rq)
+{
+	int ret = __guc_add_request(guc, rq);
+
+	if (unlikely(ret == -EBUSY)) {
+		guc->stalled_request = rq;
+		guc->submission_stall_reason = STALL_ADD_REQUEST;
+	}
+
+	return ret;
+}
+
 static inline void guc_set_lrc_tail(struct i915_request *rq)
 {
 	rq->context->lrc_reg_state[CTX_RING_TAIL] =
@@ -588,6 +724,135 @@ static inline int rq_prio(const struct i915_request *rq)
 	return rq->sched.attr.priority;
 }
 
+static bool is_multi_lrc_rq(struct i915_request *rq)
+{
+	return intel_context_is_parallel(rq->context);
+}
+
+static bool can_merge_rq(struct i915_request *rq,
+			 struct i915_request *last)
+{
+	return request_to_scheduling_context(rq) ==
+		request_to_scheduling_context(last);
+}
+
+static u32 wq_space_until_wrap(struct intel_context *ce)
+{
+	return (WQ_SIZE - ce->parallel.guc.wqi_tail);
+}
+
+static void write_wqi(struct guc_process_desc *desc,
+		      struct intel_context *ce,
+		      u32 wqi_size)
+{
+	BUILD_BUG_ON(!is_power_of_2(WQ_SIZE));
+
+	/*
+	 * Ensure WQI are visible before updating tail
+	 */
+	intel_guc_write_barrier(ce_to_guc(ce));
+
+	ce->parallel.guc.wqi_tail = (ce->parallel.guc.wqi_tail + wqi_size) &
+		(WQ_SIZE - 1);
+	WRITE_ONCE(desc->tail, ce->parallel.guc.wqi_tail);
+}
+
+static int guc_wq_noop_append(struct intel_context *ce)
+{
+	struct guc_process_desc *desc = __get_process_desc(ce);
+	u32 *wqi = get_wq_pointer(desc, ce, wq_space_until_wrap(ce));
+	u32 len_dw = wq_space_until_wrap(ce) / sizeof(u32) - 1;
+
+	if (!wqi)
+		return -EBUSY;
+
+	GEM_BUG_ON(!FIELD_FIT(WQ_LEN_MASK, len_dw));
+
+	*wqi = FIELD_PREP(WQ_TYPE_MASK, WQ_TYPE_NOOP) |
+		FIELD_PREP(WQ_LEN_MASK, len_dw);
+	ce->parallel.guc.wqi_tail = 0;
+
+	return 0;
+}
+
+static int __guc_wq_item_append(struct i915_request *rq)
+{
+	struct intel_context *ce = request_to_scheduling_context(rq);
+	struct intel_context *child;
+	struct guc_process_desc *desc = __get_process_desc(ce);
+	unsigned int wqi_size = (ce->parallel.number_children + 4) *
+		sizeof(u32);
+	u32 *wqi;
+	u32 len_dw = (wqi_size / sizeof(u32)) - 1;
+	int ret;
+
+	/* Ensure context is in correct state updating work queue */
+	GEM_BUG_ON(!atomic_read(&ce->guc_id.ref));
+	GEM_BUG_ON(context_guc_id_invalid(ce));
+	GEM_BUG_ON(context_wait_for_deregister_to_register(ce));
+	GEM_BUG_ON(!lrc_desc_registered(ce_to_guc(ce), ce->guc_id.id));
+
+	/* Insert NOOP if this work queue item will wrap the tail pointer. */
+	if (wqi_size > wq_space_until_wrap(ce)) {
+		ret = guc_wq_noop_append(ce);
+		if (ret)
+			return ret;
+	}
+
+	wqi = get_wq_pointer(desc, ce, wqi_size);
+	if (!wqi)
+		return -EBUSY;
+
+	GEM_BUG_ON(!FIELD_FIT(WQ_LEN_MASK, len_dw));
+
+	*wqi++ = FIELD_PREP(WQ_TYPE_MASK, WQ_TYPE_MULTI_LRC) |
+		FIELD_PREP(WQ_LEN_MASK, len_dw);
+	*wqi++ = ce->lrc.lrca;
+	*wqi++ = FIELD_PREP(WQ_GUC_ID_MASK, ce->guc_id.id) |
+	       FIELD_PREP(WQ_RING_TAIL_MASK, ce->ring->tail / sizeof(u64));
+	*wqi++ = 0;	/* fence_id */
+	for_each_child(ce, child)
+		*wqi++ = child->ring->tail / sizeof(u64);
+
+	write_wqi(desc, ce, wqi_size);
+
+	return 0;
+}
+
+static int guc_wq_item_append(struct intel_guc *guc,
+			      struct i915_request *rq)
+{
+	struct intel_context *ce = request_to_scheduling_context(rq);
+	int ret = 0;
+
+	if (likely(!intel_context_is_banned(ce))) {
+		ret = __guc_wq_item_append(rq);
+
+		if (unlikely(ret == -EBUSY)) {
+			guc->stalled_request = rq;
+			guc->submission_stall_reason = STALL_MOVE_LRC_TAIL;
+		}
+	}
+
+	return ret;
+}
+
+static bool multi_lrc_submit(struct i915_request *rq)
+{
+	struct intel_context *ce = request_to_scheduling_context(rq);
+
+	intel_ring_set_tail(rq->ring, rq->tail);
+
+	/*
+	 * We expect the front end (execbuf IOCTL) to set this flag on the last
+	 * request generated from a multi-BB submission. This indicates to the
+	 * backend (GuC interface) that we should submit this context thus
+	 * submitting all the requests generated in parallel.
+	 */
+	return test_bit(I915_FENCE_FLAG_SUBMIT_PARALLEL, &rq->fence.flags) ||
+		intel_context_is_banned(ce);
+}
+
 static int guc_dequeue_one_context(struct intel_guc *guc)
 {
 	struct i915_sched_engine * const sched_engine = guc->sched_engine;
@@ -601,7 +866,17 @@ static int guc_dequeue_one_context(struct intel_guc *guc)
 	if (guc->stalled_request) {
 		submit = true;
 		last = guc->stalled_request;
-		goto resubmit;
+
+		switch (guc->submission_stall_reason) {
+		case STALL_REGISTER_CONTEXT:
+			goto register_context;
+		case STALL_MOVE_LRC_TAIL:
+			goto move_lrc_tail;
+		case STALL_ADD_REQUEST:
+			goto add_request;
+		default:
+			MISSING_CASE(guc->submission_stall_reason);
+		}
 	}
 
 	while ((rb = rb_first_cached(&sched_engine->queue))) {
@@ -609,8 +884,8 @@ static int guc_dequeue_one_context(struct intel_guc *guc)
 		struct i915_request *rq, *rn;
 
 		priolist_for_each_request_consume(rq, rn, p) {
-			if (last && rq->context != last->context)
-				goto done;
+			if (last && !can_merge_rq(rq, last))
+				goto register_context;
 
 			list_del_init(&rq->sched.link);
 
@@ -618,33 +893,84 @@ static int guc_dequeue_one_context(struct intel_guc *guc)
 
 			trace_i915_request_in(rq, 0);
 			last = rq;
-			submit = true;
+
+			if (is_multi_lrc_rq(rq)) {
+				/*
+				 * We need to coalesce all multi-lrc requests in
+				 * a relationship into a single H2G. We are
+				 * guaranteed that all of these requests will be
+				 * submitted sequentially.
+				 */
+				if (multi_lrc_submit(rq)) {
+					submit = true;
+					goto register_context;
+				}
+			} else {
+				submit = true;
+			}
 		}
 
 		rb_erase_cached(&p->node, &sched_engine->queue);
 		i915_priolist_free(p);
 	}
-done:
+
+register_context:
 	if (submit) {
-		guc_set_lrc_tail(last);
-resubmit:
+		struct intel_context *ce = request_to_scheduling_context(last);
+
+		if (unlikely(!lrc_desc_registered(guc, ce->guc_id.id) &&
+			     !intel_context_is_banned(ce))) {
+			ret = guc_lrc_desc_pin(ce, false);
+			if (unlikely(ret == -EPIPE)) {
+				goto deadlk;
+			} else if (ret == -EBUSY) {
+				guc->stalled_request = last;
+				guc->submission_stall_reason =
+					STALL_REGISTER_CONTEXT;
+				goto schedule_tasklet;
+			} else if (ret != 0) {
+				GEM_WARN_ON(ret);	/* Unexpected */
+				goto deadlk;
+			}
+		}
+
+move_lrc_tail:
+		if (is_multi_lrc_rq(last)) {
+			ret = guc_wq_item_append(guc, last);
+			if (ret == -EBUSY) {
+				goto schedule_tasklet;
+			} else if (ret != 0) {
+				GEM_WARN_ON(ret);	/* Unexpected */
+				goto deadlk;
+			}
+		} else {
+			guc_set_lrc_tail(last);
+		}
+
+add_request:
 		ret = guc_add_request(guc, last);
-		if (unlikely(ret == -EPIPE))
+		if (unlikely(ret == -EPIPE)) {
+			goto deadlk;
+		} else if (ret == -EBUSY) {
+			goto schedule_tasklet;
+		} else if (ret != 0) {
+			GEM_WARN_ON(ret);	/* Unexpected */
 			goto deadlk;
-		else if (ret == -EBUSY) {
-			tasklet_schedule(&sched_engine->tasklet);
-			guc->stalled_request = last;
-			return false;
 		}
 	}
 
 	guc->stalled_request = NULL;
+	guc->submission_stall_reason = STALL_NONE;
 	return submit;
 
 deadlk:
 	sched_engine->tasklet.callback = NULL;
 	tasklet_disable_nosync(&sched_engine->tasklet);
 	return false;
+
+schedule_tasklet:
+	tasklet_schedule(&sched_engine->tasklet);
+	return false;
 }
 
 static void guc_submission_tasklet(struct tasklet_struct *t)
@@ -719,6 +1045,7 @@ static void scrub_guc_desc_for_outstanding_g2h(struct intel_guc *guc)
 			if (deregister)
 				guc_signal_context_fence(ce);
 			if (destroyed) {
+				intel_gt_pm_put_async(guc_to_gt(guc));
 				release_guc_id(guc, ce);
 				__guc_context_destroy(ce);
 			}
@@ -797,6 +1124,8 @@ static void guc_flush_submissions(struct intel_guc *guc)
 	spin_unlock_irqrestore(&sched_engine->lock, flags);
 }
 
+static void guc_flush_destroyed_contexts(struct intel_guc *guc);
+
 void intel_guc_submission_reset_prepare(struct intel_guc *guc)
 {
 	int i;
@@ -815,6 +1144,7 @@ void intel_guc_submission_reset_prepare(struct intel_guc *guc)
 	spin_unlock_irq(&guc_to_gt(guc)->irq_lock);
 
 	guc_flush_submissions(guc);
+	guc_flush_destroyed_contexts(guc);
 
 	/*
 	 * Handle any outstanding G2Hs before reset. Call IRQ handler directly
@@ -929,10 +1259,15 @@ __unwind_incomplete_requests(struct intel_context *ce)
 
 static void __guc_reset_context(struct intel_context *ce, bool stalled)
 {
+	bool local_stalled;
 	struct i915_request *rq;
 	unsigned long flags;
 	u32 head;
+	int i, number_children = ce->parallel.number_children;
 	bool skip = false;
+	struct intel_context *parent = ce;
+
+	GEM_BUG_ON(intel_context_is_child(ce));
 
 	intel_context_get(ce);
 
@@ -958,25 +1293,38 @@ static void __guc_reset_context(struct intel_context *ce, bool stalled)
 	if (unlikely(skip))
 		goto out_put;
 
-	rq = intel_context_find_active_request(ce);
-	if (!rq) {
-		head = ce->ring->tail;
-		stalled = false;
-		goto out_replay;
-	}
+	/*
+	 * For each context in the relationship find the hanging request
+	 * resetting each context / request as needed
+	 */
+	for (i = 0; i < number_children + 1; ++i) {
+		if (!intel_context_is_pinned(ce))
+			goto next_context;
 
-	if (!i915_request_started(rq))
-		stalled = false;
+		local_stalled = false;
+		rq = intel_context_find_active_request(ce);
+		if (!rq) {
+			head = ce->ring->tail;
+			goto out_replay;
+		}
 
-	GEM_BUG_ON(i915_active_is_idle(&ce->active));
-	head = intel_ring_wrap(ce->ring, rq->head);
-	__i915_request_reset(rq, stalled);
+		if (i915_request_started(rq))
+			local_stalled = true;
 
+		GEM_BUG_ON(i915_active_is_idle(&ce->active));
+		head = intel_ring_wrap(ce->ring, rq->head);
+
+		__i915_request_reset(rq, local_stalled && stalled);
 out_replay:
-	guc_reset_state(ce, head, stalled);
-	__unwind_incomplete_requests(ce);
+		guc_reset_state(ce, head, local_stalled && stalled);
+next_context:
+		if (i != number_children)
+			ce = list_next_entry(ce, parallel.child_link);
+	}
+
+	__unwind_incomplete_requests(parent);
 out_put:
-	intel_context_put(ce);
+	intel_context_put(parent);
 }
 
 void intel_guc_submission_reset(struct intel_guc *guc, bool stalled)
@@ -997,7 +1345,8 @@ void intel_guc_submission_reset(struct intel_guc *guc, bool stalled)
 
 		xa_unlock(&guc->context_lookup);
 
-		if (intel_context_is_pinned(ce))
+		if (intel_context_is_pinned(ce) &&
+		    !intel_context_is_child(ce))
 			__guc_reset_context(ce, stalled);
 
 		intel_context_put(ce);
@@ -1089,7 +1438,8 @@ void intel_guc_submission_cancel_requests(struct intel_guc *guc)
 
 		xa_unlock(&guc->context_lookup);
 
-		if (intel_context_is_pinned(ce))
+		if (intel_context_is_pinned(ce) &&
+		    !intel_context_is_child(ce))
 			guc_cancel_context_requests(ce);
 
 		intel_context_put(ce);
@@ -1126,6 +1476,8 @@ void intel_guc_submission_reset_finish(struct intel_guc *guc)
 	intel_gt_unpark_heartbeats(guc_to_gt(guc));
 }
 
+static void destroyed_worker_func(struct work_struct *w);
+
 /*
  * Set up the memory resources to be shared with the GuC (via the GGTT)
  * at firmware loading time.
@@ -1148,9 +1500,17 @@ int intel_guc_submission_init(struct intel_guc *guc)
 
 	xa_init_flags(&guc->context_lookup, XA_FLAGS_LOCK_IRQ);
 
-	spin_lock_init(&guc->contexts_lock);
-	INIT_LIST_HEAD(&guc->guc_id_list);
-	ida_init(&guc->guc_ids);
+	spin_lock_init(&guc->submission_state.lock);
+	INIT_LIST_HEAD(&guc->submission_state.guc_id_list);
+	ida_init(&guc->submission_state.guc_ids);
+	INIT_LIST_HEAD(&guc->submission_state.destroyed_contexts);
+	INIT_WORK(&guc->submission_state.destroyed_worker,
+		  destroyed_worker_func);
+
+	guc->submission_state.guc_ids_bitmap =
+		bitmap_zalloc(NUMBER_MULTI_LRC_GUC_ID, GFP_KERNEL);
+	if (!guc->submission_state.guc_ids_bitmap)
+		return -ENOMEM;
 
 	return 0;
 }
@@ -1160,8 +1520,10 @@ void intel_guc_submission_fini(struct intel_guc *guc)
 	if (!guc->lrc_desc_pool)
 		return;
 
+	guc_flush_destroyed_contexts(guc);
 	guc_lrc_desc_pool_destroy(guc);
 	i915_sched_engine_put(guc->sched_engine);
+	bitmap_free(guc->submission_state.guc_ids_bitmap);
 }
 
 static inline void queue_request(struct i915_sched_engine *sched_engine,
@@ -1178,16 +1540,22 @@ static inline void queue_request(struct i915_sched_engine *sched_engine,
 static int guc_bypass_tasklet_submit(struct intel_guc *guc,
 				     struct i915_request *rq)
 {
-	int ret;
+	int ret = 0;
 
 	__i915_request_submit(rq);
 
 	trace_i915_request_in(rq, 0);
 
-	guc_set_lrc_tail(rq);
-	ret = guc_add_request(guc, rq);
-	if (ret == -EBUSY)
-		guc->stalled_request = rq;
+	if (is_multi_lrc_rq(rq)) {
+		if (multi_lrc_submit(rq)) {
+			ret = guc_wq_item_append(guc, rq);
+			if (!ret)
+				ret = guc_add_request(guc, rq);
+		}
+	} else {
+		guc_set_lrc_tail(rq);
+		ret = guc_add_request(guc, rq);
+	}
 
 	if (unlikely(ret == -EPIPE))
 		disable_submission(guc);
@@ -1195,6 +1563,16 @@ static int guc_bypass_tasklet_submit(struct intel_guc *guc,
 	return ret;
 }
 
+static bool need_tasklet(struct intel_guc *guc, struct i915_request *rq)
+{
+	struct i915_sched_engine *sched_engine = rq->engine->sched_engine;
+	struct intel_context *ce = request_to_scheduling_context(rq);
+
+	return submission_disabled(guc) || guc->stalled_request ||
+		!i915_sched_engine_is_empty(sched_engine) ||
+		!lrc_desc_registered(guc, ce->guc_id.id);
+}
+
 static void guc_submit_request(struct i915_request *rq)
 {
 	struct i915_sched_engine *sched_engine = rq->engine->sched_engine;
@@ -1204,8 +1582,7 @@ static void guc_submit_request(struct i915_request *rq)
 	/* Will be called from irq-context when using foreign fences. */
 	spin_lock_irqsave(&sched_engine->lock, flags);
 
-	if (submission_disabled(guc) || guc->stalled_request ||
-	    !i915_sched_engine_is_empty(sched_engine))
+	if (need_tasklet(guc, rq))
 		queue_request(sched_engine, rq, rq_prio(rq));
 	else if (guc_bypass_tasklet_submit(guc, rq) == -EBUSY)
 		tasklet_hi_schedule(&sched_engine->tasklet);
@@ -1213,17 +1590,43 @@ static void guc_submit_request(struct i915_request *rq)
 	spin_unlock_irqrestore(&sched_engine->lock, flags);
 }
 
-static int new_guc_id(struct intel_guc *guc)
+static int new_guc_id(struct intel_guc *guc, struct intel_context *ce)
 {
-	return ida_simple_get(&guc->guc_ids, 0,
-			      GUC_MAX_LRC_DESCRIPTORS, GFP_KERNEL |
-			      __GFP_RETRY_MAYFAIL | __GFP_NOWARN);
+	int ret;
+
+	GEM_BUG_ON(intel_context_is_child(ce));
+
+	if (intel_context_is_parent(ce))
+		ret = bitmap_find_free_region(guc->submission_state.guc_ids_bitmap,
+					      NUMBER_MULTI_LRC_GUC_ID,
+					      order_base_2(ce->parallel.number_children
+							   + 1));
+	else
+		ret = ida_simple_get(&guc->submission_state.guc_ids,
+				     NUMBER_MULTI_LRC_GUC_ID,
+				     GUC_MAX_LRC_DESCRIPTORS,
+				     GFP_KERNEL | __GFP_RETRY_MAYFAIL |
+				     __GFP_NOWARN);
+	if (unlikely(ret < 0))
+		return ret;
+
+	ce->guc_id.id = ret;
+	return 0;
 }
 
 static void __release_guc_id(struct intel_guc *guc, struct intel_context *ce)
 {
+	GEM_BUG_ON(intel_context_is_child(ce));
+
 	if (!context_guc_id_invalid(ce)) {
-		ida_simple_remove(&guc->guc_ids, ce->guc_id.id);
+		if (intel_context_is_parent(ce))
+			bitmap_release_region(guc->submission_state.guc_ids_bitmap,
+					      ce->guc_id.id,
+					      order_base_2(ce->parallel.number_children
+							   + 1));
+		else
+			ida_simple_remove(&guc->submission_state.guc_ids,
+					  ce->guc_id.id);
 		reset_lrc_desc(guc, ce->guc_id.id);
 		set_context_guc_id_invalid(ce);
 	}
@@ -1235,54 +1638,69 @@ static void release_guc_id(struct intel_guc *guc, struct intel_context *ce)
 {
 	unsigned long flags;
 
-	spin_lock_irqsave(&guc->contexts_lock, flags);
+	spin_lock_irqsave(&guc->submission_state.lock, flags);
 	__release_guc_id(guc, ce);
-	spin_unlock_irqrestore(&guc->contexts_lock, flags);
+	spin_unlock_irqrestore(&guc->submission_state.lock, flags);
 }
 
-static int steal_guc_id(struct intel_guc *guc)
+static int steal_guc_id(struct intel_guc *guc, struct intel_context *ce)
 {
-	struct intel_context *ce;
-	int guc_id;
+	struct intel_context *cn;
 
-	lockdep_assert_held(&guc->contexts_lock);
+	lockdep_assert_held(&guc->submission_state.lock);
+	GEM_BUG_ON(intel_context_is_child(ce));
+	GEM_BUG_ON(intel_context_is_parent(ce));
 
-	if (!list_empty(&guc->guc_id_list)) {
-		ce = list_first_entry(&guc->guc_id_list,
+	if (!list_empty(&guc->submission_state.guc_id_list)) {
+		cn = list_first_entry(&guc->submission_state.guc_id_list,
 				      struct intel_context,
 				      guc_id.link);
 
-		GEM_BUG_ON(atomic_read(&ce->guc_id.ref));
-		GEM_BUG_ON(context_guc_id_invalid(ce));
+		GEM_BUG_ON(atomic_read(&cn->guc_id.ref));
+		GEM_BUG_ON(context_guc_id_invalid(cn));
+		GEM_BUG_ON(intel_context_is_child(cn));
+		GEM_BUG_ON(intel_context_is_parent(cn));
 
-		list_del_init(&ce->guc_id.link);
-		guc_id = ce->guc_id.id;
+		list_del_init(&cn->guc_id.link);
+		ce->guc_id = cn->guc_id;
 
 		spin_lock(&ce->guc_state.lock);
-		clr_context_registered(ce);
+		clr_context_registered(cn);
 		spin_unlock(&ce->guc_state.lock);
 
-		set_context_guc_id_invalid(ce);
-		return guc_id;
+		set_context_guc_id_invalid(cn);
+
+		return 0;
 	} else {
 		return -EAGAIN;
 	}
 }
 
-static int assign_guc_id(struct intel_guc *guc, u16 *out)
+static int assign_guc_id(struct intel_guc *guc, struct intel_context *ce)
 {
 	int ret;
 
-	lockdep_assert_held(&guc->contexts_lock);
+	lockdep_assert_held(&guc->submission_state.lock);
+	GEM_BUG_ON(intel_context_is_child(ce));
 
-	ret = new_guc_id(guc);
+	ret = new_guc_id(guc, ce);
 	if (unlikely(ret < 0)) {
-		ret = steal_guc_id(guc);
+		if (intel_context_is_parent(ce))
+			return -ENOSPC;
+
+		ret = steal_guc_id(guc, ce);
 		if (ret < 0)
 			return ret;
 	}
 
-	*out = ret;
+	if (intel_context_is_parent(ce)) {
+		struct intel_context *child;
+		int i = 1;
+
+		for_each_child(ce, child)
+			child->guc_id.id = ce->guc_id.id + i++;
+	}
+
 	return 0;
 }
 
@@ -1295,12 +1713,12 @@ static int pin_guc_id(struct intel_guc *guc, struct intel_context *ce)
 	GEM_BUG_ON(atomic_read(&ce->guc_id.ref));
 
 try_again:
-	spin_lock_irqsave(&guc->contexts_lock, flags);
+	spin_lock_irqsave(&guc->submission_state.lock, flags);
 
 	might_lock(&ce->guc_state.lock);
 
 	if (context_guc_id_invalid(ce)) {
-		ret = assign_guc_id(guc, &ce->guc_id.id);
+		ret = assign_guc_id(guc, ce);
 		if (ret)
 			goto out_unlock;
 		ret = 1;	/* Indidcates newly assigned guc_id */
@@ -1310,7 +1728,7 @@ try_again:
 	atomic_inc(&ce->guc_id.ref);
 
 out_unlock:
-	spin_unlock_irqrestore(&guc->contexts_lock, flags);
+	spin_unlock_irqrestore(&guc->submission_state.lock, flags);
 
 	/*
 	 * -EAGAIN indicates no guc_id are available, let's retire any
@@ -1342,15 +1760,42 @@ static void unpin_guc_id(struct intel_guc *guc, struct intel_context *ce)
 	unsigned long flags;
 
 	GEM_BUG_ON(atomic_read(&ce->guc_id.ref) < 0);
+	GEM_BUG_ON(intel_context_is_child(ce));
 
-	if (unlikely(context_guc_id_invalid(ce)))
+	if (unlikely(context_guc_id_invalid(ce) ||
+		     intel_context_is_parent(ce)))
 		return;
 
-	spin_lock_irqsave(&guc->contexts_lock, flags);
+	spin_lock_irqsave(&guc->submission_state.lock, flags);
 	if (!context_guc_id_invalid(ce) && list_empty(&ce->guc_id.link) &&
 	    !atomic_read(&ce->guc_id.ref))
-		list_add_tail(&ce->guc_id.link, &guc->guc_id_list);
-	spin_unlock_irqrestore(&guc->contexts_lock, flags);
+		list_add_tail(&ce->guc_id.link,
+			      &guc->submission_state.guc_id_list);
+	spin_unlock_irqrestore(&guc->submission_state.lock, flags);
+}
+
+static int __guc_action_register_multi_lrc(struct intel_guc *guc,
+					   struct intel_context *ce,
+					   u32 guc_id,
+					   u32 offset,
+					   bool loop)
+{
+	struct intel_context *child;
+	u32 action[4 + MAX_ENGINE_INSTANCE];
+	int len = 0;
+
+	GEM_BUG_ON(ce->parallel.number_children > MAX_ENGINE_INSTANCE);
+
+	action[len++] = INTEL_GUC_ACTION_REGISTER_CONTEXT_MULTI_LRC;
+	action[len++] = guc_id;
+	action[len++] = ce->parallel.number_children + 1;
+	action[len++] = offset;
+	for_each_child(ce, child) {
+		offset += sizeof(struct guc_lrc_desc);
+		action[len++] = offset;
+	}
+
+	return guc_submission_send_busy_loop(guc, action, len, 0, loop);
 }
 
 static int __guc_action_register_context(struct intel_guc *guc,
@@ -1375,9 +1820,15 @@ static int register_context(struct intel_context *ce, bool loop)
 		ce->guc_id.id * sizeof(struct guc_lrc_desc);
 	int ret;
 
+	GEM_BUG_ON(intel_context_is_child(ce));
 	trace_intel_context_register(ce);
 
-	ret = __guc_action_register_context(guc, ce->guc_id.id, offset, loop);
+	if (intel_context_is_parent(ce))
+		ret = __guc_action_register_multi_lrc(guc, ce, ce->guc_id.id,
+						      offset, loop);
+	else
+		ret = __guc_action_register_context(guc, ce->guc_id.id, offset,
+						    loop);
 	if (likely(!ret)) {
 		unsigned long flags;
 
@@ -1406,26 +1857,31 @@ static int deregister_context(struct intel_context *ce, u32 guc_id)
 {
 	struct intel_guc *guc = ce_to_guc(ce);
 
+	GEM_BUG_ON(intel_context_is_child(ce));
 	trace_intel_context_deregister(ce);
 
 	return __guc_action_deregister_context(guc, guc_id);
 }
 
-static intel_engine_mask_t adjust_engine_mask(u8 class, intel_engine_mask_t mask)
+static inline void clear_children_join_go_memory(struct intel_context *ce)
 {
-	switch (class) {
-	case RENDER_CLASS:
-		return mask >> RCS0;
-	case VIDEO_ENHANCEMENT_CLASS:
-		return mask >> VECS0;
-	case VIDEO_DECODE_CLASS:
-		return mask >> VCS0;
-	case COPY_ENGINE_CLASS:
-		return mask >> BCS0;
-	default:
-		MISSING_CASE(class);
-		return 0;
-	}
+	struct parent_scratch *ps = __get_parent_scratch(ce);
+	int i;
+
+	ps->go.semaphore = 0;
+	for (i = 0; i < ce->parallel.number_children + 1; ++i)
+		ps->join[i].semaphore = 0;
+}
+
+static inline u32 get_children_go_value(struct intel_context *ce)
+{
+	return __get_parent_scratch(ce)->go.semaphore;
+}
+
+static inline u32 get_children_join_value(struct intel_context *ce,
+					  u8 child_index)
+{
+	return __get_parent_scratch(ce)->join[child_index].semaphore;
 }
 
 static void guc_context_policy_init(struct intel_engine_cs *engine,
@@ -1450,6 +1906,7 @@ static int guc_lrc_desc_pin(struct intel_context *ce, bool loop)
 	struct guc_lrc_desc *desc;
 	bool context_registered;
 	intel_wakeref_t wakeref;
+	struct intel_context *child;
 	int ret = 0;
 
 	GEM_BUG_ON(!engine->mask);
@@ -1469,14 +1926,50 @@ static int guc_lrc_desc_pin(struct intel_context *ce, bool loop)
 
 	desc = __get_lrc_desc(guc, desc_idx);
 	desc->engine_class = engine_class_to_guc_class(engine->class);
-	desc->engine_submit_mask = adjust_engine_mask(engine->class,
-						      engine->mask);
+	desc->engine_submit_mask = engine->logical_mask;
 	desc->hw_context_desc = ce->lrc.lrca;
 	desc->priority = ce->guc_state.prio;
 	desc->context_flags = CONTEXT_REGISTRATION_FLAG_KMD;
 	guc_context_policy_init(engine, desc);
 
 	/*
+	 * If context is a parent, we need to register a process descriptor
+	 * describing a work queue and register all child contexts.
+	 */
+	if (intel_context_is_parent(ce)) {
+		struct guc_process_desc *pdesc;
+
+		ce->parallel.guc.wqi_tail = 0;
+		ce->parallel.guc.wqi_head = 0;
+
+		desc->process_desc = i915_ggtt_offset(ce->state) +
+			__get_parent_scratch_offset(ce);
+		desc->wq_addr = i915_ggtt_offset(ce->state) +
+			__get_wq_offset(ce);
+		desc->wq_size = WQ_SIZE;
+
+		pdesc = __get_process_desc(ce);
+		memset(pdesc, 0, sizeof(*(pdesc)));
+		pdesc->stage_id = ce->guc_id.id;
+		pdesc->wq_base_addr = desc->wq_addr;
+		pdesc->wq_size_bytes = desc->wq_size;
+		pdesc->wq_status = WQ_STATUS_ACTIVE;
+
+		for_each_child(ce, child) {
+			desc = __get_lrc_desc(guc, child->guc_id.id);
+
+			desc->engine_class =
+				engine_class_to_guc_class(engine->class);
+			desc->hw_context_desc = child->lrc.lrca;
+			desc->priority = ce->guc_state.prio;
+			desc->context_flags = CONTEXT_REGISTRATION_FLAG_KMD;
+			guc_context_policy_init(engine, desc);
+		}
+
+		clear_children_join_go_memory(ce);
+	}
+
+	/*
 	 * The context_lookup xarray is used to determine if the hardware
 	 * context is currently registered. There are two cases in which it
 	 * could be registered either the guc_id has been stolen from another
@@ -1559,7 +2052,12 @@ static int guc_context_pre_pin(struct intel_context *ce,
 
 static int guc_context_pin(struct intel_context *ce, void *vaddr)
 {
-	return __guc_context_pin(ce, ce->engine, vaddr);
+	int ret = __guc_context_pin(ce, ce->engine, vaddr);
+
+	if (likely(!ret && !intel_context_is_barrier(ce)))
+		intel_engine_pm_get(ce->engine);
+
+	return ret;
 }
 
 static void guc_context_unpin(struct intel_context *ce)
@@ -1568,6 +2066,9 @@ static void guc_context_unpin(struct intel_context *ce)
 
 	unpin_guc_id(guc, ce);
 	lrc_unpin(ce);
+
+	if (likely(!intel_context_is_barrier(ce)))
+		intel_engine_pm_put_async(ce->engine);
 }
 
 static void guc_context_post_unpin(struct intel_context *ce)
@@ -1602,6 +2103,7 @@ static void __guc_context_sched_disable(struct intel_guc *guc,
 
 	GEM_BUG_ON(guc_id == GUC_INVALID_LRC_ID);
 
+	GEM_BUG_ON(intel_context_is_child(ce));
 	trace_intel_context_sched_disable(ce);
 
 	guc_submission_send_busy_loop(guc, action, ARRAY_SIZE(action),
@@ -1653,6 +2155,8 @@ static struct i915_sw_fence *guc_context_block(struct intel_context *ce)
 	u16 guc_id;
 	bool enabled;
 
+	GEM_BUG_ON(intel_context_is_child(ce));
+
 	spin_lock_irqsave(&ce->guc_state.lock, flags);
 
 	incr_context_blocked(ce);
@@ -1707,6 +2211,7 @@ static void guc_context_unblock(struct intel_context *ce)
 	bool enable;
 
 	GEM_BUG_ON(context_enabled(ce));
+	GEM_BUG_ON(intel_context_is_child(ce));
 
 	spin_lock_irqsave(&ce->guc_state.lock, flags);
 
@@ -1733,11 +2238,14 @@ static void guc_context_unblock(struct intel_context *ce)
 static void guc_context_cancel_request(struct intel_context *ce,
 				       struct i915_request *rq)
 {
+	struct intel_context *block_context =
+		request_to_scheduling_context(rq);
+
 	if (i915_sw_fence_signaled(&rq->submit)) {
 		struct i915_sw_fence *fence;
 
 		intel_context_get(ce);
-		fence = guc_context_block(ce);
+		fence = guc_context_block(block_context);
 		i915_sw_fence_wait(fence);
 		if (!i915_request_completed(rq)) {
 			__i915_request_skip(rq);
@@ -1751,7 +2259,7 @@ static void guc_context_cancel_request(struct intel_context *ce,
 		 */
 		flush_work(&ce_to_guc(ce)->ct.requests.worker);
 
-		guc_context_unblock(ce);
+		guc_context_unblock(block_context);
 		intel_context_put(ce);
 	}
 }
@@ -1777,6 +2285,8 @@ static void guc_context_ban(struct intel_context *ce, struct i915_request *rq)
 	intel_wakeref_t wakeref;
 	unsigned long flags;
 
+	GEM_BUG_ON(intel_context_is_child(ce));
+
 	guc_flush_submissions(guc);
 
 	spin_lock_irqsave(&ce->guc_state.lock, flags);
@@ -1827,6 +2337,8 @@ static void guc_context_sched_disable(struct intel_context *ce)
 	intel_wakeref_t wakeref;
 	u16 guc_id;
 
+	GEM_BUG_ON(intel_context_is_child(ce));
+
 	spin_lock_irqsave(&ce->guc_state.lock, flags);
 
 	/*
@@ -1857,11 +2369,30 @@ unpin:
 static inline void guc_lrc_desc_unpin(struct intel_context *ce)
 {
 	struct intel_guc *guc = ce_to_guc(ce);
+	struct intel_gt *gt = guc_to_gt(guc);
+	unsigned long flags;
+	bool disabled;
 
+	GEM_BUG_ON(!intel_gt_pm_is_awake(gt));
 	GEM_BUG_ON(!lrc_desc_registered(guc, ce->guc_id.id));
 	GEM_BUG_ON(ce != __get_context(guc, ce->guc_id.id));
 	GEM_BUG_ON(context_enabled(ce));
 
+	/* Seal race with Reset */
+	spin_lock_irqsave(&ce->guc_state.lock, flags);
+	disabled = submission_disabled(guc);
+	if (likely(!disabled)) {
+		__intel_gt_pm_get(gt);
+		set_context_destroyed(ce);
+		clr_context_registered(ce);
+	}
+	spin_unlock_irqrestore(&ce->guc_state.lock, flags);
+	if (unlikely(disabled)) {
+		release_guc_id(guc, ce);
+		__guc_context_destroy(ce);
+		return;
+	}
+
 	deregister_context(ce, ce->guc_id.id);
 }
 
@@ -1889,78 +2420,86 @@ static void __guc_context_destroy(struct intel_context *ce)
 	}
 }
 
+static void guc_flush_destroyed_contexts(struct intel_guc *guc)
+{
+	struct intel_context *ce, *cn;
+	unsigned long flags;
+
+	GEM_BUG_ON(!submission_disabled(guc) &&
+		   guc_submission_initialized(guc));
+
+	spin_lock_irqsave(&guc->submission_state.lock, flags);
+	list_for_each_entry_safe(ce, cn,
+				 &guc->submission_state.destroyed_contexts,
+				 destroyed_link) {
+		list_del_init(&ce->destroyed_link);
+		__release_guc_id(guc, ce);
+		__guc_context_destroy(ce);
+	}
+	spin_unlock_irqrestore(&guc->submission_state.lock, flags);
+}
+
+static void deregister_destroyed_contexts(struct intel_guc *guc)
+{
+	struct intel_context *ce, *cn;
+	unsigned long flags;
+
+	spin_lock_irqsave(&guc->submission_state.lock, flags);
+	list_for_each_entry_safe(ce, cn,
+				 &guc->submission_state.destroyed_contexts,
+				 destroyed_link) {
+		list_del_init(&ce->destroyed_link);
+		guc_lrc_desc_unpin(ce);
+	}
+	spin_unlock_irqrestore(&guc->submission_state.lock, flags);
+}
+
+static void destroyed_worker_func(struct work_struct *w)
+{
+	struct intel_guc *guc = container_of(w, struct intel_guc,
+					     submission_state.destroyed_worker);
+	struct intel_gt *gt = guc_to_gt(guc);
+	int tmp;
+
+	with_intel_gt_pm(gt, tmp)
+		deregister_destroyed_contexts(guc);
+}
+
 static void guc_context_destroy(struct kref *kref)
 {
 	struct intel_context *ce = container_of(kref, typeof(*ce), ref);
-	struct intel_runtime_pm *runtime_pm = ce->engine->uncore->rpm;
 	struct intel_guc *guc = ce_to_guc(ce);
-	intel_wakeref_t wakeref;
 	unsigned long flags;
-	bool disabled;
+	bool destroy;
 
 	/*
 	 * If the guc_id is invalid this context has been stolen and we can free
 	 * it immediately. Also can be freed immediately if the context is not
 	 * registered with the GuC or the GuC is in the middle of a reset.
 	 */
-	if (context_guc_id_invalid(ce)) {
-		__guc_context_destroy(ce);
-		return;
-	} else if (submission_disabled(guc) ||
-		   !lrc_desc_registered(guc, ce->guc_id.id)) {
-		release_guc_id(guc, ce);
-		__guc_context_destroy(ce);
-		return;
-	}
-
-	/*
-	 * We have to acquire the context spinlock and check guc_id again, if it
-	 * is valid it hasn't been stolen and needs to be deregistered. We
-	 * delete this context from the list of unpinned guc_id available to
-	 * steal to seal a race with guc_lrc_desc_pin(). When the G2H CTB
-	 * returns indicating this context has been deregistered the guc_id is
-	 * returned to the pool of available guc_id.
-	 */
-	spin_lock_irqsave(&guc->contexts_lock, flags);
-	if (context_guc_id_invalid(ce)) {
-		spin_unlock_irqrestore(&guc->contexts_lock, flags);
-		__guc_context_destroy(ce);
-		return;
-	}
-
-	if (!list_empty(&ce->guc_id.link))
-		list_del_init(&ce->guc_id.link);
-	spin_unlock_irqrestore(&guc->contexts_lock, flags);
-
-	/* Seal race with Reset */
-	spin_lock_irqsave(&ce->guc_state.lock, flags);
-	disabled = submission_disabled(guc);
-	if (likely(!disabled)) {
-		set_context_destroyed(ce);
-		clr_context_registered(ce);
+	spin_lock_irqsave(&guc->submission_state.lock, flags);
+	destroy = submission_disabled(guc) || context_guc_id_invalid(ce) ||
+		!lrc_desc_registered(guc, ce->guc_id.id);
+	if (likely(!destroy)) {
+		if (!list_empty(&ce->guc_id.link))
+			list_del_init(&ce->guc_id.link);
+		list_add_tail(&ce->destroyed_link,
+			      &guc->submission_state.destroyed_contexts);
+	} else {
+		__release_guc_id(guc, ce);
 	}
-	spin_unlock_irqrestore(&ce->guc_state.lock, flags);
-	if (unlikely(disabled)) {
-		release_guc_id(guc, ce);
+	spin_unlock_irqrestore(&guc->submission_state.lock, flags);
+	if (unlikely(destroy)) {
 		__guc_context_destroy(ce);
 		return;
 	}
 
 	/*
-	 * We defer GuC context deregistration until the context is destroyed
-	 * in order to save on CTBs. With this optimization ideally we only need
-	 * 1 CTB to register the context during the first pin and 1 CTB to
-	 * deregister the context when the context is destroyed. Without this
-	 * optimization, a CTB would be needed every pin & unpin.
-	 *
-	 * XXX: Need to acqiure the runtime wakeref as this can be triggered
-	 * from context_free_worker when runtime wakeref is not held.
-	 * guc_lrc_desc_unpin requires the runtime as a GuC register is written
-	 * in H2G CTB to deregister the context. A future patch may defer this
-	 * H2G CTB if the runtime wakeref is zero.
+	 * We use a worker to issue the H2G to deregister the context as we can
+	 * take the GT PM for the first time which isn't allowed from an atomic
+	 * context.
 	 */
-	with_intel_runtime_pm(runtime_pm, wakeref)
-		guc_lrc_desc_unpin(ce);
+	queue_work(system_unbound_wq, &guc->submission_state.destroyed_worker);
 }
 
 static int guc_context_alloc(struct intel_context *ce)
@@ -2056,9 +2595,10 @@ static inline bool new_guc_prio_higher(u8 old_guc_prio, u8 new_guc_prio)
 
 static void add_to_context(struct i915_request *rq)
 {
-	struct intel_context *ce = rq->context;
+	struct intel_context *ce = request_to_scheduling_context(rq);
 	u8 new_guc_prio = map_i915_prio_to_guc_prio(rq_prio(rq));
 
+	GEM_BUG_ON(intel_context_is_child(ce));
 	GEM_BUG_ON(rq->guc_prio == GUC_PRIO_FINI);
 
 	spin_lock(&ce->guc_state.lock);
@@ -2091,7 +2631,9 @@ static void guc_prio_fini(struct i915_request *rq, struct intel_context *ce)
 
 static void remove_from_context(struct i915_request *rq)
 {
-	struct intel_context *ce = rq->context;
+	struct intel_context *ce = request_to_scheduling_context(rq);
+
+	GEM_BUG_ON(intel_context_is_child(ce));
 
 	spin_lock_irq(&ce->guc_state.lock);
 
@@ -2132,6 +2674,7 @@ static const struct intel_context_ops guc_context_ops = {
 	.destroy = guc_context_destroy,
 
 	.create_virtual = guc_create_virtual,
+	.create_parallel = guc_create_parallel,
 };
 
 static void submit_work_cb(struct irq_work *wrk)
@@ -2168,6 +2711,8 @@ static void guc_signal_context_fence(struct intel_context *ce)
 {
 	unsigned long flags;
 
+	GEM_BUG_ON(intel_context_is_child(ce));
+
 	spin_lock_irqsave(&ce->guc_state.lock, flags);
 	clr_context_wait_for_deregister_to_register(ce);
 	__guc_signal_context_fence(ce);
@@ -2198,7 +2743,7 @@ static void guc_context_init(struct intel_context *ce)
 
 static int guc_request_alloc(struct i915_request *rq)
 {
-	struct intel_context *ce = rq->context;
+	struct intel_context *ce = request_to_scheduling_context(rq);
 	struct intel_guc *guc = ce_to_guc(ce);
 	unsigned long flags;
 	int ret;
@@ -2302,8 +2847,30 @@ static int guc_virtual_context_pre_pin(struct intel_context *ce,
 static int guc_virtual_context_pin(struct intel_context *ce, void *vaddr)
 {
 	struct intel_engine_cs *engine = guc_virtual_get_sibling(ce->engine, 0);
+	int ret = __guc_context_pin(ce, engine, vaddr);
+	intel_engine_mask_t tmp, mask = ce->engine->mask;
 
-	return __guc_context_pin(ce, engine, vaddr);
+	if (likely(!ret))
+		for_each_engine_masked(engine, ce->engine->gt, mask, tmp)
+			intel_engine_pm_get(engine);
+
+	return ret;
+}
+
+static void guc_virtual_context_unpin(struct intel_context *ce)
+{
+	intel_engine_mask_t tmp, mask = ce->engine->mask;
+	struct intel_engine_cs *engine;
+	struct intel_guc *guc = ce_to_guc(ce);
+
+	GEM_BUG_ON(context_enabled(ce));
+	GEM_BUG_ON(intel_context_is_barrier(ce));
+
+	unpin_guc_id(guc, ce);
+	lrc_unpin(ce);
+
+	for_each_engine_masked(engine, ce->engine->gt, mask, tmp)
+		intel_engine_pm_put_async(engine);
 }
 
 static void guc_virtual_context_enter(struct intel_context *ce)
@@ -2340,7 +2907,98 @@ static const struct intel_context_ops virtual_guc_context_ops = {
 
 	.pre_pin = guc_virtual_context_pre_pin,
 	.pin = guc_virtual_context_pin,
-	.unpin = guc_context_unpin,
+	.unpin = guc_virtual_context_unpin,
+	.post_unpin = guc_context_post_unpin,
+
+	.ban = guc_context_ban,
+
+	.cancel_request = guc_context_cancel_request,
+
+	.enter = guc_virtual_context_enter,
+	.exit = guc_virtual_context_exit,
+
+	.sched_disable = guc_context_sched_disable,
+
+	.destroy = guc_context_destroy,
+
+	.get_sibling = guc_virtual_get_sibling,
+};
+
+static int guc_parent_context_pin(struct intel_context *ce, void *vaddr)
+{
+	struct intel_engine_cs *engine = guc_virtual_get_sibling(ce->engine, 0);
+	struct intel_guc *guc = ce_to_guc(ce);
+	int ret;
+
+	GEM_BUG_ON(!intel_context_is_parent(ce));
+	GEM_BUG_ON(!intel_engine_is_virtual(ce->engine));
+
+	ret = pin_guc_id(guc, ce);
+	if (unlikely(ret < 0))
+		return ret;
+
+	return __guc_context_pin(ce, engine, vaddr);
+}
+
+static int guc_child_context_pin(struct intel_context *ce, void *vaddr)
+{
+	struct intel_engine_cs *engine = guc_virtual_get_sibling(ce->engine, 0);
+
+	GEM_BUG_ON(!intel_context_is_child(ce));
+	GEM_BUG_ON(!intel_engine_is_virtual(ce->engine));
+
+	__intel_context_pin(ce->parallel.parent);
+	return __guc_context_pin(ce, engine, vaddr);
+}
+
+static void guc_parent_context_unpin(struct intel_context *ce)
+{
+	struct intel_guc *guc = ce_to_guc(ce);
+
+	GEM_BUG_ON(context_enabled(ce));
+	GEM_BUG_ON(intel_context_is_barrier(ce));
+	GEM_BUG_ON(!intel_context_is_parent(ce));
+	GEM_BUG_ON(!intel_engine_is_virtual(ce->engine));
+
+	if (ce->parallel.last_rq)
+		i915_request_put(ce->parallel.last_rq);
+	unpin_guc_id(guc, ce);
+	lrc_unpin(ce);
+}
+
+static void guc_child_context_unpin(struct intel_context *ce)
+{
+	GEM_BUG_ON(context_enabled(ce));
+	GEM_BUG_ON(intel_context_is_barrier(ce));
+	GEM_BUG_ON(!intel_context_is_child(ce));
+	GEM_BUG_ON(!intel_engine_is_virtual(ce->engine));
+
+	lrc_unpin(ce);
+}
+
+static void guc_child_context_post_unpin(struct intel_context *ce)
+{
+	GEM_BUG_ON(!intel_context_is_child(ce));
+	GEM_BUG_ON(!intel_context_is_pinned(ce->parallel.parent));
+	GEM_BUG_ON(!intel_engine_is_virtual(ce->engine));
+
+	lrc_post_unpin(ce);
+	intel_context_unpin(ce->parallel.parent);
+}
+
+static void guc_child_context_destroy(struct kref *kref)
+{
+	struct intel_context *ce = container_of(kref, typeof(*ce), ref);
+
+	__guc_context_destroy(ce);
+}
+
+static const struct intel_context_ops virtual_parent_context_ops = {
+	.alloc = guc_virtual_context_alloc,
+
+	.pre_pin = guc_context_pre_pin,
+	.pin = guc_parent_context_pin,
+	.unpin = guc_parent_context_unpin,
 	.post_unpin = guc_context_post_unpin,
 
 	.ban = guc_context_ban,
@@ -2357,6 +3015,110 @@ static const struct intel_context_ops virtual_guc_context_ops = {
 	.get_sibling = guc_virtual_get_sibling,
 };
 
+static const struct intel_context_ops virtual_child_context_ops = {
+	.alloc = guc_virtual_context_alloc,
+
+	.pre_pin = guc_context_pre_pin,
+	.pin = guc_child_context_pin,
+	.unpin = guc_child_context_unpin,
+	.post_unpin = guc_child_context_post_unpin,
+
+	.cancel_request = guc_context_cancel_request,
+
+	.enter = guc_virtual_context_enter,
+	.exit = guc_virtual_context_exit,
+
+	.destroy = guc_child_context_destroy,
+
+	.get_sibling = guc_virtual_get_sibling,
+};
+
+/*
+ * The below override of the breadcrumbs is enabled when the user configures a
+ * context for parallel submission (multi-lrc, parent-child).
+ *
+ * The overridden breadcrumbs implements an algorithm which allows the GuC to
+ * safely preempt all the hw contexts configured for parallel submission
+ * between each BB. The contract between the i915 and GuC is if the parent
+ * context can be preempted, all the children can be preempted, and the GuC will
+ * always try to preempt the parent before the children. A handshake between the
+ * parent / children breadcrumbs ensures the i915 holds up its end of the deal
+ * creating a window to preempt between each set of BBs.
+ */
+static int emit_bb_start_parent_no_preempt_mid_batch(struct i915_request *rq,
+						     u64 offset, u32 len,
+						     const unsigned int flags);
+static int emit_bb_start_child_no_preempt_mid_batch(struct i915_request *rq,
+						    u64 offset, u32 len,
+						    const unsigned int flags);
+static u32 *
+emit_fini_breadcrumb_parent_no_preempt_mid_batch(struct i915_request *rq,
+						 u32 *cs);
+static u32 *
+emit_fini_breadcrumb_child_no_preempt_mid_batch(struct i915_request *rq,
+						u32 *cs);
+
+static struct intel_context *
+guc_create_parallel(struct intel_engine_cs **engines,
+		    unsigned int num_siblings,
+		    unsigned int width)
+{
+	struct intel_engine_cs **siblings = NULL;
+	struct intel_context *parent = NULL, *ce, *err;
+	int i, j;
+
+	siblings = kmalloc_array(num_siblings,
+				 sizeof(*siblings),
+				 GFP_KERNEL);
+	if (!siblings)
+		return ERR_PTR(-ENOMEM);
+
+	for (i = 0; i < width; ++i) {
+		for (j = 0; j < num_siblings; ++j)
+			siblings[j] = engines[i * num_siblings + j];
+
+		ce = intel_engine_create_virtual(siblings, num_siblings,
+						 FORCE_VIRTUAL);
+		if (!ce) {
+			err = ERR_PTR(-ENOMEM);
+			goto unwind;
+		}
+
+		if (i == 0) {
+			parent = ce;
+			parent->ops = &virtual_parent_context_ops;
+		} else {
+			ce->ops = &virtual_child_context_ops;
+			intel_context_bind_parent_child(parent, ce);
+		}
+	}
+
+	parent->parallel.fence_context = dma_fence_context_alloc(1);
+
+	parent->engine->emit_bb_start =
+		emit_bb_start_parent_no_preempt_mid_batch;
+	parent->engine->emit_fini_breadcrumb =
+		emit_fini_breadcrumb_parent_no_preempt_mid_batch;
+	parent->engine->emit_fini_breadcrumb_dw =
+		12 + 4 * parent->parallel.number_children;
+	for_each_child(parent, ce) {
+		ce->engine->emit_bb_start =
+			emit_bb_start_child_no_preempt_mid_batch;
+		ce->engine->emit_fini_breadcrumb =
+			emit_fini_breadcrumb_child_no_preempt_mid_batch;
+		ce->engine->emit_fini_breadcrumb_dw = 16;
+	}
+
+	kfree(siblings);
+	return parent;
+
+unwind:
+	if (parent)
+		intel_context_put(parent);
+	kfree(siblings);
+	return err;
+}
+
 static bool
 guc_irq_enable_breadcrumbs(struct intel_breadcrumbs *b)
 {
@@ -2416,7 +3178,7 @@ static void guc_init_breadcrumbs(struct intel_engine_cs *engine)
 static void guc_bump_inflight_request_prio(struct i915_request *rq,
 					   int prio)
 {
-	struct intel_context *ce = rq->context;
+	struct intel_context *ce = request_to_scheduling_context(rq);
 	u8 new_guc_prio = map_i915_prio_to_guc_prio(prio);
 
 	/* Short circuit function */
@@ -2439,7 +3201,7 @@ static void guc_bump_inflight_request_prio(struct i915_request *rq,
 
 static void guc_retire_inflight_request_prio(struct i915_request *rq)
 {
-	struct intel_context *ce = rq->context;
+	struct intel_context *ce = request_to_scheduling_context(rq);
 
 	spin_lock(&ce->guc_state.lock);
 	guc_prio_fini(rq, ce);
@@ -2753,6 +3515,12 @@ g2h_context_lookup(struct intel_guc *guc, u32 desc_idx)
 		return NULL;
 	}
 
+	if (unlikely(intel_context_is_child(ce))) {
+		drm_err(&guc_to_gt(guc)->i915->drm,
+			"Context is child, desc_idx %u", desc_idx);
+		return NULL;
+	}
+
 	return ce;
 }
 
@@ -2796,6 +3564,7 @@ int intel_guc_deregister_done_process_msg(struct intel_guc *guc,
 		intel_context_put(ce);
 	} else if (context_destroyed(ce)) {
 		/* Context has been destroyed */
+		intel_gt_pm_put_async(guc_to_gt(guc));
 		release_guc_id(guc, ce);
 		__guc_context_destroy(ce);
 	}
@@ -3122,6 +3891,25 @@ static inline void guc_log_context_priority(struct drm_printer *p,
 	drm_printf(p, "\n");
 }
 
+static inline void guc_log_context(struct drm_printer *p,
+				   struct intel_context *ce)
+{
+	drm_printf(p, "GuC lrc descriptor %u:\n", ce->guc_id.id);
+	drm_printf(p, "\tHW Context Desc: 0x%08x\n", ce->lrc.lrca);
+	drm_printf(p, "\t\tLRC Head: Internal %u, Memory %u\n",
+		   ce->ring->head,
+		   ce->lrc_reg_state[CTX_RING_HEAD]);
+	drm_printf(p, "\t\tLRC Tail: Internal %u, Memory %u\n",
+		   ce->ring->tail,
+		   ce->lrc_reg_state[CTX_RING_TAIL]);
+	drm_printf(p, "\t\tContext Pin Count: %u\n",
+		   atomic_read(&ce->pin_count));
+	drm_printf(p, "\t\tGuC ID Ref Count: %u\n",
+		   atomic_read(&ce->guc_id.ref));
+	drm_printf(p, "\t\tSchedule State: 0x%x\n\n",
+		   ce->guc_state.sched_state);
+}
+
 void intel_guc_submission_print_context_info(struct intel_guc *guc,
 					     struct drm_printer *p)
 {
@@ -3131,28 +3919,310 @@ void intel_guc_submission_print_context_info(struct intel_guc *guc,
 
 	xa_lock_irqsave(&guc->context_lookup, flags);
 	xa_for_each(&guc->context_lookup, index, ce) {
-		drm_printf(p, "GuC lrc descriptor %u:\n", ce->guc_id.id);
-		drm_printf(p, "\tHW Context Desc: 0x%08x\n", ce->lrc.lrca);
-		drm_printf(p, "\t\tLRC Head: Internal %u, Memory %u\n",
-			   ce->ring->head,
-			   ce->lrc_reg_state[CTX_RING_HEAD]);
-		drm_printf(p, "\t\tLRC Tail: Internal %u, Memory %u\n",
-			   ce->ring->tail,
-			   ce->lrc_reg_state[CTX_RING_TAIL]);
-		drm_printf(p, "\t\tContext Pin Count: %u\n",
-			   atomic_read(&ce->pin_count));
-		drm_printf(p, "\t\tGuC ID Ref Count: %u\n",
-			   atomic_read(&ce->guc_id.ref));
-		drm_printf(p, "\t\tSchedule State: 0x%x\n\n",
-			   ce->guc_state.sched_state);
+		GEM_BUG_ON(intel_context_is_child(ce));
 
+		guc_log_context(p, ce);
 		guc_log_context_priority(p, ce);
+
+		if (intel_context_is_parent(ce)) {
+			struct guc_process_desc *desc = __get_process_desc(ce);
+			struct intel_context *child;
+
+			drm_printf(p, "\t\tNumber children: %u\n",
+				   ce->parallel.number_children);
+			drm_printf(p, "\t\tWQI Head: %u\n",
+				   READ_ONCE(desc->head));
+			drm_printf(p, "\t\tWQI Tail: %u\n",
+				   READ_ONCE(desc->tail));
+			drm_printf(p, "\t\tWQI Status: %u\n\n",
+				   READ_ONCE(desc->wq_status));
+
+			if (ce->engine->emit_bb_start ==
+			    emit_bb_start_parent_no_preempt_mid_batch) {
+				u8 i;
+
+				drm_printf(p, "\t\tChildren Go: %u\n\n",
+					   get_children_go_value(ce));
+				for (i = 0; i < ce->parallel.number_children; ++i)
+					drm_printf(p, "\t\tChildren Join: %u\n",
+						   get_children_join_value(ce, i));
+			}
+
+			for_each_child(ce, child)
+				guc_log_context(p, child);
+		}
 	}
 	xa_unlock_irqrestore(&guc->context_lookup, flags);
 }
 
+static inline u32 get_children_go_addr(struct intel_context *ce)
+{
+	GEM_BUG_ON(!intel_context_is_parent(ce));
+
+	return i915_ggtt_offset(ce->state) +
+		__get_parent_scratch_offset(ce) +
+		offsetof(struct parent_scratch, go.semaphore);
+}
+
+static inline u32 get_children_join_addr(struct intel_context *ce,
+					 u8 child_index)
+{
+	GEM_BUG_ON(!intel_context_is_parent(ce));
+
+	return i915_ggtt_offset(ce->state) +
+		__get_parent_scratch_offset(ce) +
+		offsetof(struct parent_scratch, join[child_index].semaphore);
+}
+
+#define PARENT_GO_BB			1
+#define PARENT_GO_FINI_BREADCRUMB	0
+#define CHILD_GO_BB			1
+#define CHILD_GO_FINI_BREADCRUMB	0
+static int emit_bb_start_parent_no_preempt_mid_batch(struct i915_request *rq,
+						     u64 offset, u32 len,
+						     const unsigned int flags)
+{
+	struct intel_context *ce = rq->context;
+	u32 *cs;
+	u8 i;
+
+	GEM_BUG_ON(!intel_context_is_parent(ce));
+
+	cs = intel_ring_begin(rq, 10 + 4 * ce->parallel.number_children);
+	if (IS_ERR(cs))
+		return PTR_ERR(cs);
+
+	/* Wait on children */
+	for (i = 0; i < ce->parallel.number_children; ++i) {
+		*cs++ = (MI_SEMAPHORE_WAIT |
+			 MI_SEMAPHORE_GLOBAL_GTT |
+			 MI_SEMAPHORE_POLL |
+			 MI_SEMAPHORE_SAD_EQ_SDD);
+		*cs++ = PARENT_GO_BB;
+		*cs++ = get_children_join_addr(ce, i);
+		*cs++ = 0;
+	}
+
+	/* Turn off preemption */
+	*cs++ = MI_ARB_ON_OFF | MI_ARB_DISABLE;
+	*cs++ = MI_NOOP;
+
+	/* Tell children go */
+	cs = gen8_emit_ggtt_write(cs,
+				  CHILD_GO_BB,
+				  get_children_go_addr(ce),
+				  0);
+
+	/* Jump to batch */
+	*cs++ = MI_BATCH_BUFFER_START_GEN8 |
+		(flags & I915_DISPATCH_SECURE ? 0 : BIT(8));
+	*cs++ = lower_32_bits(offset);
+	*cs++ = upper_32_bits(offset);
+	*cs++ = MI_NOOP;
+
+	intel_ring_advance(rq, cs);
+
+	return 0;
+}
+
+static int emit_bb_start_child_no_preempt_mid_batch(struct i915_request *rq,
+						    u64 offset, u32 len,
+						    const unsigned int flags)
+{
+	struct intel_context *ce = rq->context;
+	struct intel_context *parent = intel_context_to_parent(ce);
+	u32 *cs;
+
+	GEM_BUG_ON(!intel_context_is_child(ce));
+
+	cs = intel_ring_begin(rq, 12);
+	if (IS_ERR(cs))
+		return PTR_ERR(cs);
+
+	/* Signal parent */
+	cs = gen8_emit_ggtt_write(cs,
+				  PARENT_GO_BB,
+				  get_children_join_addr(parent,
+							 ce->parallel.child_index),
+				  0);
+
+	/* Wait on parent for go */
+	*cs++ = (MI_SEMAPHORE_WAIT |
+		 MI_SEMAPHORE_GLOBAL_GTT |
+		 MI_SEMAPHORE_POLL |
+		 MI_SEMAPHORE_SAD_EQ_SDD);
+	*cs++ = CHILD_GO_BB;
+	*cs++ = get_children_go_addr(parent);
+	*cs++ = 0;
+
+	/* Turn off preemption */
+	*cs++ = MI_ARB_ON_OFF | MI_ARB_DISABLE;
+
+	/* Jump to batch */
+	*cs++ = MI_BATCH_BUFFER_START_GEN8 |
+		(flags & I915_DISPATCH_SECURE ? 0 : BIT(8));
+	*cs++ = lower_32_bits(offset);
+	*cs++ = upper_32_bits(offset);
+
+	intel_ring_advance(rq, cs);
+
+	return 0;
+}
+
+static u32 *
+__emit_fini_breadcrumb_parent_no_preempt_mid_batch(struct i915_request *rq,
+						   u32 *cs)
+{
+	struct intel_context *ce = rq->context;
+	u8 i;
+
+	GEM_BUG_ON(!intel_context_is_parent(ce));
+
+	/* Wait on children */
+	for (i = 0; i < ce->parallel.number_children; ++i) {
+		*cs++ = (MI_SEMAPHORE_WAIT |
+			 MI_SEMAPHORE_GLOBAL_GTT |
+			 MI_SEMAPHORE_POLL |
+			 MI_SEMAPHORE_SAD_EQ_SDD);
+		*cs++ = PARENT_GO_FINI_BREADCRUMB;
+		*cs++ = get_children_join_addr(ce, i);
+		*cs++ = 0;
+	}
+
+	/* Turn on preemption */
+	*cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE;
+	*cs++ = MI_NOOP;
+
+	/* Tell children go */
+	cs = gen8_emit_ggtt_write(cs,
+				  CHILD_GO_FINI_BREADCRUMB,
+				  get_children_go_addr(ce),
+				  0);
+
+	return cs;
+}
+
+/*
+ * If this true, a submission of multi-lrc requests had an error and the
+ * requests need to be skipped. The front end (execuf IOCTL) should've called
+ * i915_request_skip which squashes the BB but we still need to emit the fini
+ * breadrcrumbs seqno write. At this point we don't know how many of the
+ * requests in the multi-lrc submission were generated so we can't do the
+ * handshake between the parent and children (e.g. if 4 requests should be
+ * generated but 2nd hit an error only 1 would be seen by the GuC backend).
+ * Simply skip the handshake, but still emit the breadcrumbd seqno, if an error
+ * has occurred on any of the requests in submission / relationship.
+ */
+static inline bool skip_handshake(struct i915_request *rq)
+{
+	return test_bit(I915_FENCE_FLAG_SKIP_PARALLEL, &rq->fence.flags);
+}
+
+static u32 *
+emit_fini_breadcrumb_parent_no_preempt_mid_batch(struct i915_request *rq,
+						 u32 *cs)
+{
+	struct intel_context *ce = rq->context;
+
+	GEM_BUG_ON(!intel_context_is_parent(ce));
+
+	if (unlikely(skip_handshake(rq))) {
+		/*
+		 * NOP everything in __emit_fini_breadcrumb_parent_no_preempt_mid_batch,
+		 * the -6 comes from the length of the emits below.
+		 */
+		memset(cs, 0, sizeof(u32) *
+		       (ce->engine->emit_fini_breadcrumb_dw - 6));
+		cs += ce->engine->emit_fini_breadcrumb_dw - 6;
+	} else {
+		cs = __emit_fini_breadcrumb_parent_no_preempt_mid_batch(rq, cs);
+	}
+
+	/* Emit fini breadcrumb */
+	cs = gen8_emit_ggtt_write(cs,
+				  rq->fence.seqno,
+				  i915_request_active_timeline(rq)->hwsp_offset,
+				  0);
+
+	/* User interrupt */
+	*cs++ = MI_USER_INTERRUPT;
+	*cs++ = MI_NOOP;
+
+	rq->tail = intel_ring_offset(rq, cs);
+
+	return cs;
+}
+
+static u32 *
+__emit_fini_breadcrumb_child_no_preempt_mid_batch(struct i915_request *rq,
+						  u32 *cs)
+{
+	struct intel_context *ce = rq->context;
+	struct intel_context *parent = intel_context_to_parent(ce);
+
+	GEM_BUG_ON(!intel_context_is_child(ce));
+
+	/* Turn on preemption */
+	*cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE;
+	*cs++ = MI_NOOP;
+
+	/* Signal parent */
+	cs = gen8_emit_ggtt_write(cs,
+				  PARENT_GO_FINI_BREADCRUMB,
+				  get_children_join_addr(parent,
+							 ce->parallel.child_index),
+				  0);
+
+	/* Wait parent on for go */
+	*cs++ = (MI_SEMAPHORE_WAIT |
+		 MI_SEMAPHORE_GLOBAL_GTT |
+		 MI_SEMAPHORE_POLL |
+		 MI_SEMAPHORE_SAD_EQ_SDD);
+	*cs++ = CHILD_GO_FINI_BREADCRUMB;
+	*cs++ = get_children_go_addr(parent);
+	*cs++ = 0;
+
+	return cs;
+}
+
+static u32 *
+emit_fini_breadcrumb_child_no_preempt_mid_batch(struct i915_request *rq,
+						u32 *cs)
+{
+	struct intel_context *ce = rq->context;
+
+	GEM_BUG_ON(!intel_context_is_child(ce));
+
+	if (unlikely(skip_handshake(rq))) {
+		/*
+		 * NOP everything in __emit_fini_breadcrumb_child_no_preempt_mid_batch,
+		 * the -6 comes from the length of the emits below.
+		 */
+		memset(cs, 0, sizeof(u32) *
+		       (ce->engine->emit_fini_breadcrumb_dw - 6));
+		cs += ce->engine->emit_fini_breadcrumb_dw - 6;
+	} else {
+		cs = __emit_fini_breadcrumb_child_no_preempt_mid_batch(rq, cs);
+	}
+
+	/* Emit fini breadcrumb */
+	cs = gen8_emit_ggtt_write(cs,
+				  rq->fence.seqno,
+				  i915_request_active_timeline(rq)->hwsp_offset,
+				  0);
+
+	/* User interrupt */
+	*cs++ = MI_USER_INTERRUPT;
+	*cs++ = MI_NOOP;
+
+	rq->tail = intel_ring_offset(rq, cs);
+
+	return cs;
+}
+
 static struct intel_context *
-guc_create_virtual(struct intel_engine_cs **siblings, unsigned int count)
+guc_create_virtual(struct intel_engine_cs **siblings, unsigned int count,
+		   unsigned long flags)
 {
 	struct guc_virtual_engine *ve;
 	struct intel_guc *guc;
@@ -3201,6 +4271,7 @@ guc_create_virtual(struct intel_engine_cs **siblings, unsigned int count)
 		}
 
 		ve->base.mask |= sibling->mask;
+		ve->base.logical_mask |= sibling->logical_mask;
 
 		if (n != 0 && ve->base.class != sibling->class) {
 			DRM_DEBUG("invalid mixing of engine class, sibling %d, already %d\n",
@@ -3259,4 +4330,5 @@ bool intel_guc_virtual_engine_has_heartbeat(const struct intel_engine_cs *ve)
 
 #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
 #include "selftest_guc.c"
+#include "selftest_guc_multi_lrc.c"
 #endif
diff --git a/drivers/gpu/drm/i915/gt/uc/selftest_guc_multi_lrc.c b/drivers/gpu/drm/i915/gt/uc/selftest_guc_multi_lrc.c
new file mode 100644
index 000000000000..50953c8e8b53
--- /dev/null
+++ b/drivers/gpu/drm/i915/gt/uc/selftest_guc_multi_lrc.c
@@ -0,0 +1,179 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright �� 2019 Intel Corporation
+ */
+
+#include "selftests/igt_spinner.h"
+#include "selftests/igt_reset.h"
+#include "selftests/intel_scheduler_helpers.h"
+#include "gt/intel_engine_heartbeat.h"
+#include "gem/selftests/mock_context.h"
+
+static void logical_sort(struct intel_engine_cs **engines, int num_engines)
+{
+	struct intel_engine_cs *sorted[MAX_ENGINE_INSTANCE + 1];
+	int i, j;
+
+	for (i = 0; i < num_engines; ++i)
+		for (j = 0; j < MAX_ENGINE_INSTANCE + 1; ++j) {
+			if (engines[j]->logical_mask & BIT(i)) {
+				sorted[i] = engines[j];
+				break;
+			}
+		}
+
+	memcpy(*engines, *sorted,
+	       sizeof(struct intel_engine_cs *) * num_engines);
+}
+
+static struct intel_context *
+multi_lrc_create_parent(struct intel_gt *gt, u8 class,
+			unsigned long flags)
+{
+	struct intel_engine_cs *siblings[MAX_ENGINE_INSTANCE + 1];
+	struct intel_engine_cs *engine;
+	enum intel_engine_id id;
+	int i = 0;
+
+	for_each_engine(engine, gt, id) {
+		if (engine->class != class)
+			continue;
+
+		siblings[i++] = engine;
+	}
+
+	if (i <= 1)
+		return ERR_PTR(0);
+
+	logical_sort(siblings, i);
+
+	return intel_engine_create_parallel(siblings, 1, i);
+}
+
+static void multi_lrc_context_unpin(struct intel_context *ce)
+{
+	struct intel_context *child;
+
+	GEM_BUG_ON(!intel_context_is_parent(ce));
+
+	for_each_child(ce, child)
+		intel_context_unpin(child);
+	intel_context_unpin(ce);
+}
+
+static void multi_lrc_context_put(struct intel_context *ce)
+{
+	GEM_BUG_ON(!intel_context_is_parent(ce));
+
+	/*
+	 * Only the parent gets the creation ref put in the uAPI, the parent
+	 * itself is responsible for creation ref put on the children.
+	 */
+	intel_context_put(ce);
+}
+
+static struct i915_request *
+multi_lrc_nop_request(struct intel_context *ce)
+{
+	struct intel_context *child;
+	struct i915_request *rq, *child_rq;
+	int i = 0;
+
+	GEM_BUG_ON(!intel_context_is_parent(ce));
+
+	rq = intel_context_create_request(ce);
+	if (IS_ERR(rq))
+		return rq;
+
+	i915_request_get(rq);
+	i915_request_add(rq);
+
+	for_each_child(ce, child) {
+		child_rq = intel_context_create_request(child);
+		if (IS_ERR(child_rq))
+			goto child_error;
+
+		if (++i == ce->parallel.number_children)
+			set_bit(I915_FENCE_FLAG_SUBMIT_PARALLEL,
+				&child_rq->fence.flags);
+		i915_request_add(child_rq);
+	}
+
+	return rq;
+
+child_error:
+	i915_request_put(rq);
+
+	return ERR_PTR(-ENOMEM);
+}
+
+static int __intel_guc_multi_lrc_basic(struct intel_gt *gt, unsigned int class)
+{
+	struct intel_context *parent;
+	struct i915_request *rq;
+	int ret;
+
+	parent = multi_lrc_create_parent(gt, class, 0);
+	if (IS_ERR(parent)) {
+		pr_err("Failed creating contexts: %ld", PTR_ERR(parent));
+		return PTR_ERR(parent);
+	} else if (!parent) {
+		pr_debug("Not enough engines in class: %d", class);
+		return 0;
+	}
+
+	rq = multi_lrc_nop_request(parent);
+	if (IS_ERR(rq)) {
+		ret = PTR_ERR(rq);
+		pr_err("Failed creating requests: %d", ret);
+		goto out;
+	}
+
+	ret = intel_selftest_wait_for_rq(rq);
+	if (ret)
+		pr_err("Failed waiting on request: %d", ret);
+
+	i915_request_put(rq);
+
+	if (ret >= 0) {
+		ret = intel_gt_wait_for_idle(gt, HZ * 5);
+		if (ret < 0)
+			pr_err("GT failed to idle: %d\n", ret);
+	}
+
+out:
+	multi_lrc_context_unpin(parent);
+	multi_lrc_context_put(parent);
+	return ret;
+}
+
+static int intel_guc_multi_lrc_basic(void *arg)
+{
+	struct intel_gt *gt = arg;
+	unsigned int class;
+	int ret;
+
+	for (class = 0; class < MAX_ENGINE_CLASS + 1; ++class) {
+		ret = __intel_guc_multi_lrc_basic(gt, class);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+int intel_guc_multi_lrc_live_selftests(struct drm_i915_private *i915)
+{
+	static const struct i915_subtest tests[] = {
+		SUBTEST(intel_guc_multi_lrc_basic),
+	};
+	struct intel_gt *gt = &i915->gt;
+
+	if (intel_gt_is_wedged(gt))
+		return 0;
+
+	if (!intel_uc_uses_guc_submission(&gt->uc))
+		return 0;
+
+	return intel_gt_live_subtests(tests, gt);
+}
diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index 54ae75db121f..fe638b5da7c0 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -35,6 +35,7 @@
 #include "gt/intel_gt.h"
 #include "gt/intel_gt_buffer_pool.h"
 #include "gt/intel_gt_clock_utils.h"
+#include "gt/intel_gt_debugfs.h"
 #include "gt/intel_gt_pm.h"
 #include "gt/intel_gt_pm_debugfs.h"
 #include "gt/intel_gt_requests.h"
@@ -553,36 +554,18 @@ static int i915_wa_registers(struct seq_file *m, void *unused)
 	return 0;
 }
 
-static int
-i915_wedged_get(void *data, u64 *val)
+static int i915_wedged_get(void *data, u64 *val)
 {
 	struct drm_i915_private *i915 = data;
-	int ret = intel_gt_terminally_wedged(&i915->gt);
 
-	switch (ret) {
-	case -EIO:
-		*val = 1;
-		return 0;
-	case 0:
-		*val = 0;
-		return 0;
-	default:
-		return ret;
-	}
+	return intel_gt_debugfs_reset_show(&i915->gt, val);
 }
 
-static int
-i915_wedged_set(void *data, u64 val)
+static int i915_wedged_set(void *data, u64 val)
 {
 	struct drm_i915_private *i915 = data;
 
-	/* Flush any previous reset before applying for a new one */
-	wait_event(i915->gt.reset.queue,
-		   !test_bit(I915_RESET_BACKOFF, &i915->gt.reset.flags));
-
-	intel_gt_handle_error(&i915->gt, val, I915_ERROR_CAPTURE,
-			      "Manually set wedged engine mask = %llx", val);
-	return 0;
+	return intel_gt_debugfs_reset_store(&i915->gt, val);
 }
 
 DEFINE_SIMPLE_ATTRIBUTE(i915_wedged_fops,
@@ -727,27 +710,15 @@ static int i915_sseu_status(struct seq_file *m, void *unused)
 static int i915_forcewake_open(struct inode *inode, struct file *file)
 {
 	struct drm_i915_private *i915 = inode->i_private;
-	struct intel_gt *gt = &i915->gt;
-
-	atomic_inc(&gt->user_wakeref);
-	intel_gt_pm_get(gt);
-	if (GRAPHICS_VER(i915) >= 6)
-		intel_uncore_forcewake_user_get(gt->uncore);
 
-	return 0;
+	return intel_gt_pm_debugfs_forcewake_user_open(&i915->gt);
 }
 
 static int i915_forcewake_release(struct inode *inode, struct file *file)
 {
 	struct drm_i915_private *i915 = inode->i_private;
-	struct intel_gt *gt = &i915->gt;
 
-	if (GRAPHICS_VER(i915) >= 6)
-		intel_uncore_forcewake_user_put(&i915->uncore);
-	intel_gt_pm_put(gt);
-	atomic_dec(&gt->user_wakeref);
-
-	return 0;
+	return intel_gt_pm_debugfs_forcewake_user_release(&i915->gt);
 }
 
 static const struct file_operations i915_forcewake_fops = {
diff --git a/drivers/gpu/drm/i915/i915_query.c b/drivers/gpu/drm/i915/i915_query.c
index 5e2b909827f4..51b368be0fc4 100644
--- a/drivers/gpu/drm/i915/i915_query.c
+++ b/drivers/gpu/drm/i915/i915_query.c
@@ -124,7 +124,9 @@ query_engine_info(struct drm_i915_private *i915,
 	for_each_uabi_engine(engine, i915) {
 		info.engine.engine_class = engine->uabi_class;
 		info.engine.engine_instance = engine->uabi_instance;
+		info.flags = I915_ENGINE_INFO_HAS_LOGICAL_INSTANCE;
 		info.capabilities = engine->uabi_capabilities;
+		info.logical_instance = ilog2(engine->logical_mask);
 
 		if (copy_to_user(info_ptr, &info, sizeof(info)))
 			return -EFAULT;
diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
index ed64fa9defdf..2c3cd6e635b5 100644
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -1335,6 +1335,25 @@ i915_request_await_external(struct i915_request *rq, struct dma_fence *fence)
 	return err;
 }
 
+static inline bool is_parallel_rq(struct i915_request *rq)
+{
+	return intel_context_is_parallel(rq->context);
+}
+
+static inline struct intel_context *request_to_parent(struct i915_request *rq)
+{
+	return intel_context_to_parent(rq->context);
+}
+
+static bool is_same_parallel_context(struct i915_request *to,
+				     struct i915_request *from)
+{
+	if (is_parallel_rq(to))
+		return request_to_parent(to) == request_to_parent(from);
+
+	return false;
+}
+
 int
 i915_request_await_execution(struct i915_request *rq,
 			     struct dma_fence *fence)
@@ -1366,11 +1385,14 @@ i915_request_await_execution(struct i915_request *rq,
 		 * want to run our callback in all cases.
 		 */
 
-		if (dma_fence_is_i915(fence))
+		if (dma_fence_is_i915(fence)) {
+			if (is_same_parallel_context(rq, to_request(fence)))
+				continue;
 			ret = __i915_request_await_execution(rq,
 							     to_request(fence));
-		else
+		} else {
 			ret = i915_request_await_external(rq, fence);
+		}
 		if (ret < 0)
 			return ret;
 	} while (--nchild);
@@ -1471,10 +1493,13 @@ i915_request_await_dma_fence(struct i915_request *rq, struct dma_fence *fence)
 						 fence))
 			continue;
 
-		if (dma_fence_is_i915(fence))
+		if (dma_fence_is_i915(fence)) {
+			if (is_same_parallel_context(rq, to_request(fence)))
+				continue;
 			ret = i915_request_await_request(rq, to_request(fence));
-		else
+		} else {
 			ret = i915_request_await_external(rq, fence);
+		}
 		if (ret < 0)
 			return ret;
 
@@ -1550,35 +1575,51 @@ i915_request_await_object(struct i915_request *to,
 }
 
 static struct i915_request *
-__i915_request_add_to_timeline(struct i915_request *rq)
+__i915_request_ensure_parallel_ordering(struct i915_request *rq,
+					struct intel_timeline *timeline)
 {
-	struct intel_timeline *timeline = i915_request_timeline(rq);
 	struct i915_request *prev;
 
-	/*
-	 * Dependency tracking and request ordering along the timeline
-	 * is special cased so that we can eliminate redundant ordering
-	 * operations while building the request (we know that the timeline
-	 * itself is ordered, and here we guarantee it).
-	 *
-	 * As we know we will need to emit tracking along the timeline,
-	 * we embed the hooks into our request struct -- at the cost of
-	 * having to have specialised no-allocation interfaces (which will
-	 * be beneficial elsewhere).
-	 *
-	 * A second benefit to open-coding i915_request_await_request is
-	 * that we can apply a slight variant of the rules specialised
-	 * for timelines that jump between engines (such as virtual engines).
-	 * If we consider the case of virtual engine, we must emit a dma-fence
-	 * to prevent scheduling of the second request until the first is
-	 * complete (to maximise our greedy late load balancing) and this
-	 * precludes optimising to use semaphores serialisation of a single
-	 * timeline across engines.
-	 */
+	GEM_BUG_ON(!is_parallel_rq(rq));
+
+	prev = request_to_parent(rq)->parallel.last_rq;
+	if (prev) {
+		if (!__i915_request_is_complete(prev)) {
+			i915_sw_fence_await_sw_fence(&rq->submit,
+						     &prev->submit,
+						     &rq->submitq);
+
+			if (rq->engine->sched_engine->schedule)
+				__i915_sched_node_add_dependency(&rq->sched,
+								 &prev->sched,
+								 &rq->dep,
+								 0);
+		}
+		i915_request_put(prev);
+	}
+
+	request_to_parent(rq)->parallel.last_rq = i915_request_get(rq);
+
+	return to_request(__i915_active_fence_set(&timeline->last_request,
+						  &rq->fence));
+}
+
+static struct i915_request *
+__i915_request_ensure_ordering(struct i915_request *rq,
+			       struct intel_timeline *timeline)
+{
+	struct i915_request *prev;
+
+	GEM_BUG_ON(is_parallel_rq(rq));
+
 	prev = to_request(__i915_active_fence_set(&timeline->last_request,
 						  &rq->fence));
+
 	if (prev && !__i915_request_is_complete(prev)) {
 		bool uses_guc = intel_engine_uses_guc(rq->engine);
+		bool pow2 = is_power_of_2(READ_ONCE(prev->engine)->mask |
+					  rq->engine->mask);
+		bool same_context = prev->context == rq->context;
 
 		/*
 		 * The requests are supposed to be kept in order. However,
@@ -1586,13 +1627,11 @@ __i915_request_add_to_timeline(struct i915_request *rq)
 		 * is used as a barrier for external modification to this
 		 * context.
 		 */
-		GEM_BUG_ON(prev->context == rq->context &&
+		GEM_BUG_ON(same_context &&
 			   i915_seqno_passed(prev->fence.seqno,
 					     rq->fence.seqno));
 
-		if ((!uses_guc &&
-		     is_power_of_2(READ_ONCE(prev->engine)->mask | rq->engine->mask)) ||
-		    (uses_guc && prev->context == rq->context))
+		if ((same_context && uses_guc) || (!uses_guc && pow2))
 			i915_sw_fence_await_sw_fence(&rq->submit,
 						     &prev->submit,
 						     &rq->submitq);
@@ -1607,6 +1646,50 @@ __i915_request_add_to_timeline(struct i915_request *rq)
 							 0);
 	}
 
+	return prev;
+}
+
+static struct i915_request *
+__i915_request_add_to_timeline(struct i915_request *rq)
+{
+	struct intel_timeline *timeline = i915_request_timeline(rq);
+	struct i915_request *prev;
+
+	/*
+	 * Dependency tracking and request ordering along the timeline
+	 * is special cased so that we can eliminate redundant ordering
+	 * operations while building the request (we know that the timeline
+	 * itself is ordered, and here we guarantee it).
+	 *
+	 * As we know we will need to emit tracking along the timeline,
+	 * we embed the hooks into our request struct -- at the cost of
+	 * having to have specialised no-allocation interfaces (which will
+	 * be beneficial elsewhere).
+	 *
+	 * A second benefit to open-coding i915_request_await_request is
+	 * that we can apply a slight variant of the rules specialised
+	 * for timelines that jump between engines (such as virtual engines).
+	 * If we consider the case of virtual engine, we must emit a dma-fence
+	 * to prevent scheduling of the second request until the first is
+	 * complete (to maximise our greedy late load balancing) and this
+	 * precludes optimising to use semaphores serialisation of a single
+	 * timeline across engines.
+	 *
+	 * We do not order parallel submission requests on the timeline as each
+	 * parallel submission context has its own timeline and the ordering
+	 * rules for parallel requests are that they must be submitted in the
+	 * order received from the execbuf IOCTL. So rather than using the
+	 * timeline we store a pointer to last request submitted in the
+	 * relationship in the gem context and insert a submission fence
+	 * between that request and request passed into this function or
+	 * alternatively we use completion fence if gem context has a single
+	 * timeline and this is the first submission of an execbuf IOCTL.
+	 */
+	if (likely(!is_parallel_rq(rq)))
+		prev = __i915_request_ensure_ordering(rq, timeline);
+	else
+		prev = __i915_request_ensure_parallel_ordering(rq, timeline);
+
 	/*
 	 * Make sure that no request gazumped us - if it was allocated after
 	 * our i915_request_alloc() and called __i915_request_add() before
diff --git a/drivers/gpu/drm/i915/i915_request.h b/drivers/gpu/drm/i915/i915_request.h
index 7bd9ed20623e..dc359242d1ae 100644
--- a/drivers/gpu/drm/i915/i915_request.h
+++ b/drivers/gpu/drm/i915/i915_request.h
@@ -139,6 +139,29 @@ enum {
 	 * the GPU. Here we track such boost requests on a per-request basis.
 	 */
 	I915_FENCE_FLAG_BOOST,
+
+	/*
+	 * I915_FENCE_FLAG_SUBMIT_PARALLEL - request with a context in a
+	 * parent-child relationship (parallel submission, multi-lrc) should
+	 * trigger a submission to the GuC rather than just moving the context
+	 * tail.
+	 */
+	I915_FENCE_FLAG_SUBMIT_PARALLEL,
+
+	/*
+	 * I915_FENCE_FLAG_SKIP_PARALLEL - request with a context in a
+	 * parent-child relationship (parallel submission, multi-lrc) that
+	 * hit an error while generating requests in the execbuf IOCTL.
+	 * Indicates this request should be skipped as another request in
+	 * submission / relationship encoutered an error.
+	 */
+	I915_FENCE_FLAG_SKIP_PARALLEL,
+
+	/*
+	 * I915_FENCE_FLAG_COMPOSITE - Indicates fence is part of a composite
+	 * fence (dma_fence_array) and i915 generated for parallel submission.
+	 */
+	I915_FENCE_FLAG_COMPOSITE,
 };
 
 /**
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 4b7fc4647e46..90546fa58fc1 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -1234,9 +1234,10 @@ int __i915_vma_move_to_active(struct i915_vma *vma, struct i915_request *rq)
 	return i915_active_add_request(&vma->active, rq);
 }
 
-int i915_vma_move_to_active(struct i915_vma *vma,
-			    struct i915_request *rq,
-			    unsigned int flags)
+int _i915_vma_move_to_active(struct i915_vma *vma,
+			     struct i915_request *rq,
+			     struct dma_fence *fence,
+			     unsigned int flags)
 {
 	struct drm_i915_gem_object *obj = vma->obj;
 	int err;
@@ -1257,9 +1258,11 @@ int i915_vma_move_to_active(struct i915_vma *vma,
 			intel_frontbuffer_put(front);
 		}
 
-		dma_resv_add_excl_fence(vma->resv, &rq->fence);
-		obj->write_domain = I915_GEM_DOMAIN_RENDER;
-		obj->read_domains = 0;
+		if (fence) {
+			dma_resv_add_excl_fence(vma->resv, fence);
+			obj->write_domain = I915_GEM_DOMAIN_RENDER;
+			obj->read_domains = 0;
+		}
 	} else {
 		if (!(flags & __EXEC_OBJECT_NO_RESERVE)) {
 			err = dma_resv_reserve_shared(vma->resv, 1);
@@ -1267,8 +1270,10 @@ int i915_vma_move_to_active(struct i915_vma *vma,
 				return err;
 		}
 
-		dma_resv_add_shared_fence(vma->resv, &rq->fence);
-		obj->write_domain = 0;
+		if (fence) {
+			dma_resv_add_shared_fence(vma->resv, fence);
+			obj->write_domain = 0;
+		}
 	}
 
 	if (flags & EXEC_OBJECT_NEEDS_FENCE && vma->fence)
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index ed69f66c7ab0..648dbe744c96 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -57,9 +57,16 @@ static inline bool i915_vma_is_active(const struct i915_vma *vma)
 
 int __must_check __i915_vma_move_to_active(struct i915_vma *vma,
 					   struct i915_request *rq);
-int __must_check i915_vma_move_to_active(struct i915_vma *vma,
-					 struct i915_request *rq,
-					 unsigned int flags);
+int __must_check _i915_vma_move_to_active(struct i915_vma *vma,
+					  struct i915_request *rq,
+					  struct dma_fence *fence,
+					  unsigned int flags);
+static inline int __must_check
+i915_vma_move_to_active(struct i915_vma *vma, struct i915_request *rq,
+			unsigned int flags)
+{
+	return _i915_vma_move_to_active(vma, rq, &rq->fence, flags);
+}
 
 #define __i915_vma_flags(v) ((unsigned long *)&(v)->flags.counter)
 
diff --git a/drivers/gpu/drm/i915/intel_wakeref.h b/drivers/gpu/drm/i915/intel_wakeref.h
index 545c8f277c46..4f4c2e15e736 100644
--- a/drivers/gpu/drm/i915/intel_wakeref.h
+++ b/drivers/gpu/drm/i915/intel_wakeref.h
@@ -123,6 +123,12 @@ enum {
 	__INTEL_WAKEREF_PUT_LAST_BIT__
 };
 
+static inline void
+intel_wakeref_might_get(struct intel_wakeref *wf)
+{
+	might_lock(&wf->mutex);
+}
+
 /**
  * intel_wakeref_put_flags: Release the wakeref
  * @wf: the wakeref
@@ -170,6 +176,12 @@ intel_wakeref_put_delay(struct intel_wakeref *wf, unsigned long delay)
 			    FIELD_PREP(INTEL_WAKEREF_PUT_DELAY, delay));
 }
 
+static inline void
+intel_wakeref_might_put(struct intel_wakeref *wf)
+{
+	might_lock(&wf->mutex);
+}
+
 /**
  * intel_wakeref_lock: Lock the wakeref (mutex)
  * @wf: the wakeref
diff --git a/drivers/gpu/drm/i915/selftests/i915_live_selftests.h b/drivers/gpu/drm/i915/selftests/i915_live_selftests.h
index 3cf6758931f9..bdd290f2bf3c 100644
--- a/drivers/gpu/drm/i915/selftests/i915_live_selftests.h
+++ b/drivers/gpu/drm/i915/selftests/i915_live_selftests.h
@@ -48,5 +48,6 @@ selftest(ring_submission, intel_ring_submission_live_selftests)
 selftest(perf, i915_perf_live_selftests)
 selftest(slpc, intel_slpc_live_selftests)
 selftest(guc, intel_guc_live_selftests)
+selftest(guc_multi_lrc, intel_guc_multi_lrc_live_selftests)
 /* Here be dragons: keep last to run last! */
 selftest(late_gt_pm, intel_gt_pm_late_selftests)
diff --git a/drivers/gpu/drm/i915/selftests/mock_region.c b/drivers/gpu/drm/i915/selftests/mock_region.c
index efa86dffe3c6..75793008c4ef 100644
--- a/drivers/gpu/drm/i915/selftests/mock_region.c
+++ b/drivers/gpu/drm/i915/selftests/mock_region.c
@@ -6,8 +6,6 @@
 #include <drm/ttm/ttm_placement.h>
 #include <linux/scatterlist.h>
 
-#include <drm/ttm/ttm_placement.h>
-
 #include "gem/i915_gem_region.h"
 #include "intel_memory_region.h"
 #include "intel_region_ttm.h"