diff --git a/drivers/gpu/drm/i915/gt/intel_ring_submission.c b/drivers/gpu/drm/i915/gt/intel_ring_submission.c
index ee241b7eaa3b..f0ce70861e93 100644
--- a/drivers/gpu/drm/i915/gt/intel_ring_submission.c
+++ b/drivers/gpu/drm/i915/gt/intel_ring_submission.c
@@ -1356,7 +1356,9 @@ static int load_pd_dir(struct i915_request *rq,
 	return rq->engine->emit_flush(rq, EMIT_FLUSH);
 }
 
-static inline int mi_set_context(struct i915_request *rq, u32 flags)
+static inline int mi_set_context(struct i915_request *rq,
+				 struct intel_context *ce,
+				 u32 flags)
 {
 	struct drm_i915_private *i915 = rq->i915;
 	struct intel_engine_cs *engine = rq->engine;
@@ -1431,7 +1433,7 @@ static inline int mi_set_context(struct i915_request *rq, u32 flags)
 
 	*cs++ = MI_NOOP;
 	*cs++ = MI_SET_CONTEXT;
-	*cs++ = i915_ggtt_offset(rq->context->state) | flags;
+	*cs++ = i915_ggtt_offset(ce->state) | flags;
 	/*
 	 * w/a: MI_SET_CONTEXT must always be followed by MI_NOOP
 	 * WaMiSetContext_Hang:snb,ivb,vlv
@@ -1546,13 +1548,56 @@ static int switch_mm(struct i915_request *rq, struct i915_address_space *vm)
 	return rq->engine->emit_flush(rq, EMIT_INVALIDATE);
 }
 
+static int clear_residuals(struct i915_request *rq)
+{
+	struct intel_engine_cs *engine = rq->engine;
+	int ret;
+
+	ret = switch_mm(rq, vm_alias(engine->kernel_context->vm));
+	if (ret)
+		return ret;
+
+	if (engine->kernel_context->state) {
+		ret = mi_set_context(rq,
+				     engine->kernel_context,
+				     MI_MM_SPACE_GTT | MI_RESTORE_INHIBIT);
+		if (ret)
+			return ret;
+	}
+
+	ret = engine->emit_bb_start(rq,
+				    engine->wa_ctx.vma->node.start, 0,
+				    0);
+	if (ret)
+		return ret;
+
+	ret = engine->emit_flush(rq, EMIT_FLUSH);
+	if (ret)
+		return ret;
+
+	/* Always invalidate before the next switch_mm() */
+	return engine->emit_flush(rq, EMIT_INVALIDATE);
+}
+
 static int switch_context(struct i915_request *rq)
 {
+	struct intel_engine_cs *engine = rq->engine;
 	struct intel_context *ce = rq->context;
+	void **residuals = NULL;
 	int ret;
 
 	GEM_BUG_ON(HAS_EXECLISTS(rq->i915));
 
+	if (engine->wa_ctx.vma && ce != engine->kernel_context) {
+		if (engine->wa_ctx.vma->private != ce) {
+			ret = clear_residuals(rq);
+			if (ret)
+				return ret;
+
+			residuals = &engine->wa_ctx.vma->private;
+		}
+	}
+
 	ret = switch_mm(rq, vm_alias(ce->vm));
 	if (ret)
 		return ret;
@@ -1560,7 +1605,7 @@ static int switch_context(struct i915_request *rq)
 	if (ce->state) {
 		u32 flags;
 
-		GEM_BUG_ON(rq->engine->id != RCS0);
+		GEM_BUG_ON(engine->id != RCS0);
 
 		/* For resource streamer on HSW+ and power context elsewhere */
 		BUILD_BUG_ON(HSW_MI_RS_SAVE_STATE_EN != MI_SAVE_EXT_STATE_EN);
@@ -1572,7 +1617,7 @@ static int switch_context(struct i915_request *rq)
 		else
 			flags |= MI_RESTORE_INHIBIT;
 
-		ret = mi_set_context(rq, flags);
+		ret = mi_set_context(rq, ce, flags);
 		if (ret)
 			return ret;
 	}
@@ -1581,6 +1626,20 @@ static int switch_context(struct i915_request *rq)
 	if (ret)
 		return ret;
 
+	/*
+	 * Now past the point of no return, this request _will_ be emitted.
+	 *
+	 * Or at least this preamble will be emitted, the request may be
+	 * interrupted prior to submitting the user payload. If so, we
+	 * still submit the "empty" request in order to preserve global
+	 * state tracking such as this, our tracking of the current
+	 * dirty context.
+	 */
+	if (residuals) {
+		intel_context_put(*residuals);
+		*residuals = intel_context_get(ce);
+	}
+
 	return 0;
 }
 
@@ -1765,6 +1824,11 @@ static void ring_release(struct intel_engine_cs *engine)
 
 	intel_engine_cleanup_common(engine);
 
+	if (engine->wa_ctx.vma) {
+		intel_context_put(engine->wa_ctx.vma->private);
+		i915_vma_unpin_and_release(&engine->wa_ctx.vma, 0);
+	}
+
 	intel_ring_unpin(engine->legacy.ring);
 	intel_ring_put(engine->legacy.ring);
 
@@ -1912,6 +1976,60 @@ static void setup_vecs(struct intel_engine_cs *engine)
 	engine->emit_fini_breadcrumb = gen7_xcs_emit_breadcrumb;
 }
 
+static int gen7_ctx_switch_bb_setup(struct intel_engine_cs * const engine,
+				    struct i915_vma * const vma)
+{
+	return 0;
+}
+
+static int gen7_ctx_switch_bb_init(struct intel_engine_cs *engine)
+{
+	struct drm_i915_gem_object *obj;
+	struct i915_vma *vma;
+	int size;
+	int err;
+
+	size = gen7_ctx_switch_bb_setup(engine, NULL /* probe size */);
+	if (size <= 0)
+		return size;
+
+	size = ALIGN(size, PAGE_SIZE);
+	obj = i915_gem_object_create_internal(engine->i915, size);
+	if (IS_ERR(obj))
+		return PTR_ERR(obj);
+
+	vma = i915_vma_instance(obj, engine->gt->vm, NULL);
+	if (IS_ERR(vma)) {
+		err = PTR_ERR(vma);
+		goto err_obj;
+	}
+
+	vma->private = intel_context_create(engine); /* dummy residuals */
+	if (IS_ERR(vma->private)) {
+		err = PTR_ERR(vma->private);
+		goto err_obj;
+	}
+
+	err = i915_vma_pin(vma, 0, 0, PIN_USER | PIN_HIGH);
+	if (err)
+		goto err_private;
+
+	err = gen7_ctx_switch_bb_setup(engine, vma);
+	if (err)
+		goto err_unpin;
+
+	engine->wa_ctx.vma = vma;
+	return 0;
+
+err_unpin:
+	i915_vma_unpin(vma);
+err_private:
+	intel_context_put(vma->private);
+err_obj:
+	i915_gem_object_put(obj);
+	return err;
+}
+
 int intel_ring_submission_setup(struct intel_engine_cs *engine)
 {
 	struct intel_timeline *timeline;
@@ -1965,11 +2083,19 @@ int intel_ring_submission_setup(struct intel_engine_cs *engine)
 
 	GEM_BUG_ON(timeline->hwsp_ggtt != engine->status_page.vma);
 
+	if (IS_GEN(engine->i915, 7) && engine->class == RENDER_CLASS) {
+		err = gen7_ctx_switch_bb_init(engine);
+		if (err)
+			goto err_ring_unpin;
+	}
+
 	/* Finally, take ownership and responsibility for cleanup! */
 	engine->release = ring_release;
 
 	return 0;
 
+err_ring_unpin:
+	intel_ring_unpin(ring);
 err_ring:
 	intel_ring_put(ring);
 err_timeline_unpin:
@@ -1980,3 +2106,7 @@ err:
 	intel_engine_cleanup_common(engine);
 	return err;
 }
+
+#if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
+#include "selftest_ring_submission.c"
+#endif
diff --git a/drivers/gpu/drm/i915/gt/selftest_ring_submission.c b/drivers/gpu/drm/i915/gt/selftest_ring_submission.c
new file mode 100644
index 000000000000..fcfddcd2a63d
--- /dev/null
+++ b/drivers/gpu/drm/i915/gt/selftest_ring_submission.c
@@ -0,0 +1,290 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2020 Intel Corporation
+ */
+
+#include "intel_engine_pm.h"
+#include "selftests/igt_flush_test.h"
+
+static struct i915_vma *create_wally(struct intel_engine_cs *engine)
+{
+	struct drm_i915_gem_object *obj;
+	struct i915_vma *vma;
+	u32 *cs;
+	int err;
+
+	obj = i915_gem_object_create_internal(engine->i915, 4096);
+	if (IS_ERR(obj))
+		return ERR_CAST(obj);
+
+	vma = i915_vma_instance(obj, engine->gt->vm, NULL);
+	if (IS_ERR(vma)) {
+		i915_gem_object_put(obj);
+		return vma;
+	}
+
+	err = i915_vma_pin(vma, 0, 0, PIN_USER | PIN_HIGH);
+	if (err) {
+		i915_gem_object_put(obj);
+		return ERR_PTR(err);
+	}
+
+	cs = i915_gem_object_pin_map(obj, I915_MAP_WC);
+	if (IS_ERR(cs)) {
+		i915_gem_object_put(obj);
+		return ERR_CAST(cs);
+	}
+
+	if (INTEL_GEN(engine->i915) >= 6) {
+		*cs++ = MI_STORE_DWORD_IMM_GEN4;
+		*cs++ = 0;
+	} else if (INTEL_GEN(engine->i915) >= 4) {
+		*cs++ = MI_STORE_DWORD_IMM_GEN4 | MI_USE_GGTT;
+		*cs++ = 0;
+	} else {
+		*cs++ = MI_STORE_DWORD_IMM | MI_MEM_VIRTUAL;
+	}
+	*cs++ = vma->node.start + 4000;
+	*cs++ = STACK_MAGIC;
+
+	*cs++ = MI_BATCH_BUFFER_END;
+	i915_gem_object_unpin_map(obj);
+
+	vma->private = intel_context_create(engine); /* dummy residuals */
+	if (IS_ERR(vma->private)) {
+		vma = ERR_CAST(vma->private);
+		i915_gem_object_put(obj);
+	}
+
+	return vma;
+}
+
+static int context_sync(struct intel_context *ce)
+{
+	struct i915_request *rq;
+	int err = 0;
+
+	rq = intel_context_create_request(ce);
+	if (IS_ERR(rq))
+		return PTR_ERR(rq);
+
+	i915_request_get(rq);
+	i915_request_add(rq);
+
+	if (i915_request_wait(rq, 0, HZ / 5) < 0)
+		err = -ETIME;
+	i915_request_put(rq);
+
+	return err;
+}
+
+static int new_context_sync(struct intel_engine_cs *engine)
+{
+	struct intel_context *ce;
+	int err;
+
+	ce = intel_context_create(engine);
+	if (IS_ERR(ce))
+		return PTR_ERR(ce);
+
+	err = context_sync(ce);
+	intel_context_put(ce);
+
+	return err;
+}
+
+static int mixed_contexts_sync(struct intel_engine_cs *engine, u32 *result)
+{
+	int pass;
+	int err;
+
+	for (pass = 0; pass < 2; pass++) {
+		WRITE_ONCE(*result, 0);
+		err = context_sync(engine->kernel_context);
+		if (err || READ_ONCE(*result)) {
+			if (!err) {
+				pr_err("pass[%d] wa_bb emitted for the kernel context\n",
+				       pass);
+				err = -EINVAL;
+			}
+			return err;
+		}
+
+		WRITE_ONCE(*result, 0);
+		err = new_context_sync(engine);
+		if (READ_ONCE(*result) != STACK_MAGIC) {
+			if (!err) {
+				pr_err("pass[%d] wa_bb *NOT* emitted after the kernel context\n",
+				       pass);
+				err = -EINVAL;
+			}
+			return err;
+		}
+
+		WRITE_ONCE(*result, 0);
+		err = new_context_sync(engine);
+		if (READ_ONCE(*result) != STACK_MAGIC) {
+			if (!err) {
+				pr_err("pass[%d] wa_bb *NOT* emitted for the user context switch\n",
+				       pass);
+				err = -EINVAL;
+			}
+			return err;
+		}
+	}
+
+	return 0;
+}
+
+static int double_context_sync_00(struct intel_engine_cs *engine, u32 *result)
+{
+	struct intel_context *ce;
+	int err, i;
+
+	ce = intel_context_create(engine);
+	if (IS_ERR(ce))
+		return PTR_ERR(ce);
+
+	for (i = 0; i < 2; i++) {
+		WRITE_ONCE(*result, 0);
+		err = context_sync(ce);
+		if (err)
+			break;
+	}
+	intel_context_put(ce);
+	if (err)
+		return err;
+
+	if (READ_ONCE(*result)) {
+		pr_err("wa_bb emitted between the same user context\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int kernel_context_sync_00(struct intel_engine_cs *engine, u32 *result)
+{
+	struct intel_context *ce;
+	int err, i;
+
+	ce = intel_context_create(engine);
+	if (IS_ERR(ce))
+		return PTR_ERR(ce);
+
+	for (i = 0; i < 2; i++) {
+		WRITE_ONCE(*result, 0);
+		err = context_sync(ce);
+		if (err)
+			break;
+
+		err = context_sync(engine->kernel_context);
+		if (err)
+			break;
+	}
+	intel_context_put(ce);
+	if (err)
+		return err;
+
+	if (READ_ONCE(*result)) {
+		pr_err("wa_bb emitted between the same user context [with intervening kernel]\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int __live_ctx_switch_wa(struct intel_engine_cs *engine)
+{
+	struct i915_vma *bb;
+	u32 *result;
+	int err;
+
+	bb = create_wally(engine);
+	if (IS_ERR(bb))
+		return PTR_ERR(bb);
+
+	result = i915_gem_object_pin_map(bb->obj, I915_MAP_WC);
+	if (IS_ERR(result)) {
+		intel_context_put(bb->private);
+		i915_vma_unpin_and_release(&bb, 0);
+		return PTR_ERR(result);
+	}
+	result += 1000;
+
+	engine->wa_ctx.vma = bb;
+
+	err = mixed_contexts_sync(engine, result);
+	if (err)
+		goto out;
+
+	err = double_context_sync_00(engine, result);
+	if (err)
+		goto out;
+
+	err = kernel_context_sync_00(engine, result);
+	if (err)
+		goto out;
+
+out:
+	intel_context_put(engine->wa_ctx.vma->private);
+	i915_vma_unpin_and_release(&engine->wa_ctx.vma, I915_VMA_RELEASE_MAP);
+	return err;
+}
+
+static int live_ctx_switch_wa(void *arg)
+{
+	struct intel_gt *gt = arg;
+	struct intel_engine_cs *engine;
+	enum intel_engine_id id;
+
+	/*
+	 * Exercise the inter-context wa batch.
+	 *
+	 * Between each user context we run a wa batch, and since it may
+	 * have implications for user visible state, we have to check that
+	 * we do actually execute it.
+	 *
+	 * The trick we use is to replace the normal wa batch with a custom
+	 * one that writes to a marker within it, and we can then look for
+	 * that marker to confirm if the batch was run when we expect it,
+	 * and equally important it was wasn't run when we don't!
+	 */
+
+	for_each_engine(engine, gt, id) {
+		struct i915_vma *saved_wa;
+		int err;
+
+		if (!intel_engine_can_store_dword(engine))
+			continue;
+
+		if (IS_GEN_RANGE(gt->i915, 4, 5))
+			continue; /* MI_STORE_DWORD is privileged! */
+
+		saved_wa = fetch_and_zero(&engine->wa_ctx.vma);
+
+		intel_engine_pm_get(engine);
+		err = __live_ctx_switch_wa(engine);
+		intel_engine_pm_put(engine);
+		if (igt_flush_test(gt->i915))
+			err = -EIO;
+
+		engine->wa_ctx.vma = saved_wa;
+		if (err)
+			return err;
+	}
+
+	return 0;
+}
+
+int intel_ring_submission_live_selftests(struct drm_i915_private *i915)
+{
+	static const struct i915_subtest tests[] = {
+		SUBTEST(live_ctx_switch_wa),
+	};
+
+	if (HAS_EXECLISTS(i915))
+		return 0;
+
+	return intel_gt_live_subtests(tests, &i915->gt);
+}
diff --git a/drivers/gpu/drm/i915/selftests/i915_live_selftests.h b/drivers/gpu/drm/i915/selftests/i915_live_selftests.h
index 34138c7bdd15..0a953bfc0585 100644
--- a/drivers/gpu/drm/i915/selftests/i915_live_selftests.h
+++ b/drivers/gpu/drm/i915/selftests/i915_live_selftests.h
@@ -43,6 +43,7 @@ selftest(reset, intel_reset_live_selftests)
 selftest(memory_region, intel_memory_region_live_selftests)
 selftest(hangcheck, intel_hangcheck_live_selftests)
 selftest(execlists, intel_execlists_live_selftests)
+selftest(ring_submission, intel_ring_submission_live_selftests)
 selftest(perf, i915_perf_live_selftests)
 /* Here be dragons: keep last to run last! */
 selftest(late_gt_pm, intel_gt_pm_late_selftests)