drm/i915: Move context management under GEM

Keep track of the GEM contexts underneath i915->gem.contexts and assign
them their own lock for the purposes of list management.

v2: Focus on lock tracking; ctx->vm is protected by ctx->mutex
v3: Correct split with removal of logical HW ID
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191004134015.13204-15-chris@chris-wilson.co.uk

drivers/gpu/drm/i915/gem/i915_gem_context.h

@@ -11,7 +11,9 @@
 
 #include "gt/intel_context.h"
 
+#include "i915_drv.h"
 #include "i915_gem.h"
+#include "i915_gem_gtt.h"
 #include "i915_scheduler.h"
 #include "intel_device_info.h"
 
@@ -118,8 +120,8 @@ static inline bool i915_gem_context_is_kernel(struct i915_gem_context *ctx)
 }
 
 /* i915_gem_context.c */
-int __must_check i915_gem_contexts_init(struct drm_i915_private *dev_priv);
-void i915_gem_contexts_fini(struct drm_i915_private *dev_priv);
+int __must_check i915_gem_init_contexts(struct drm_i915_private *i915);
+void i915_gem_driver_release__contexts(struct drm_i915_private *i915);
 
 int i915_gem_context_open(struct drm_i915_private *i915,
 			  struct drm_file *file);
@@ -158,6 +160,27 @@ static inline void i915_gem_context_put(struct i915_gem_context *ctx)
 	kref_put(&ctx->ref, i915_gem_context_release);
 }
 
+static inline struct i915_address_space *
+i915_gem_context_vm(struct i915_gem_context *ctx)
+{
+	return rcu_dereference_protected(ctx->vm, lockdep_is_held(&ctx->mutex));
+}
+
+static inline struct i915_address_space *
+i915_gem_context_get_vm_rcu(struct i915_gem_context *ctx)
+{
+	struct i915_address_space *vm;
+
+	rcu_read_lock();
+	vm = rcu_dereference(ctx->vm);
+	if (!vm)
+		vm = &ctx->i915->ggtt.vm;
+	vm = i915_vm_get(vm);
+	rcu_read_unlock();
+
+	return vm;
+}
+
 static inline struct i915_gem_engines *
 i915_gem_context_engines(struct i915_gem_context *ctx)
 {

drivers/gpu/drm/i915/gem/i915_gem_context_types.h

@@ -88,7 +88,7 @@ struct i915_gem_context {
 	 * In other modes, this is a NULL pointer with the expectation that
 	 * the caller uses the shared global GTT.
 	 */
-	struct i915_address_space *vm;
+	struct i915_address_space __rcu *vm;
 
 	/**
 	 * @pid: process id of creator

drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c

@@ -728,7 +728,7 @@ static int eb_select_context(struct i915_execbuffer *eb)
 		return -ENOENT;
 
 	eb->gem_context = ctx;
-	if (ctx->vm)
+	if (rcu_access_pointer(ctx->vm))
 		eb->invalid_flags |= EXEC_OBJECT_NEEDS_GTT;
 
 	eb->context_flags = 0;

drivers/gpu/drm/i915/gem/i915_gem_userptr.c

@@ -766,7 +766,8 @@ i915_gem_userptr_ioctl(struct drm_device *dev,
 		 * On almost all of the older hw, we cannot tell the GPU that
 		 * a page is readonly.
 		 */
-		vm = dev_priv->kernel_context->vm;
+		vm = rcu_dereference_protected(dev_priv->kernel_context->vm,
+					       true); /* static vm */
 		if (!vm || !vm->has_read_only)
 			return -ENODEV;
 	}

drivers/gpu/drm/i915/gem/selftests/huge_pages.c

@@ -1322,15 +1322,15 @@ static int igt_ppgtt_pin_update(void *arg)
 	struct i915_gem_context *ctx = arg;
 	struct drm_i915_private *dev_priv = ctx->i915;
 	unsigned long supported = INTEL_INFO(dev_priv)->page_sizes;
-	struct i915_address_space *vm = ctx->vm;
 	struct drm_i915_gem_object *obj;
 	struct i915_gem_engines_iter it;
+	struct i915_address_space *vm;
 	struct intel_context *ce;
 	struct i915_vma *vma;
 	unsigned int flags = PIN_USER | PIN_OFFSET_FIXED;
 	unsigned int n;
 	int first, last;
-	int err;
+	int err = 0;
 
 	/*
 	 * Make sure there's no funny business when doing a PIN_UPDATE -- in the
@@ -1340,9 +1340,10 @@ static int igt_ppgtt_pin_update(void *arg)
 	 * huge-gtt-pages.
 	 */
 
-	if (!vm || !i915_vm_is_4lvl(vm)) {
+	vm = i915_gem_context_get_vm_rcu(ctx);
+	if (!i915_vm_is_4lvl(vm)) {
 		pr_info("48b PPGTT not supported, skipping\n");
-		return 0;
+		goto out_vm;
 	}
 
 	first = ilog2(I915_GTT_PAGE_SIZE_64K);
@@ -1451,6 +1452,8 @@ static int igt_ppgtt_pin_update(void *arg)
 	i915_vma_close(vma);
 out_put:
 	i915_gem_object_put(obj);
+out_vm:
+	i915_vm_put(vm);
 
 	return err;
 }
@@ -1460,7 +1463,7 @@ static int igt_tmpfs_fallback(void *arg)
 	struct i915_gem_context *ctx = arg;
 	struct drm_i915_private *i915 = ctx->i915;
 	struct vfsmount *gemfs = i915->mm.gemfs;
-	struct i915_address_space *vm = ctx->vm ?: &i915->ggtt.vm;
+	struct i915_address_space *vm = i915_gem_context_get_vm_rcu(ctx);
 	struct drm_i915_gem_object *obj;
 	struct i915_vma *vma;
 	u32 *vaddr;
@@ -1510,6 +1513,7 @@ static int igt_tmpfs_fallback(void *arg)
 out_restore:
 	i915->mm.gemfs = gemfs;
 
+	i915_vm_put(vm);
 	return err;
 }
 
@@ -1517,14 +1521,14 @@ static int igt_shrink_thp(void *arg)
 {
 	struct i915_gem_context *ctx = arg;
 	struct drm_i915_private *i915 = ctx->i915;
-	struct i915_address_space *vm = ctx->vm ?: &i915->ggtt.vm;
+	struct i915_address_space *vm = i915_gem_context_get_vm_rcu(ctx);
 	struct drm_i915_gem_object *obj;
 	struct i915_gem_engines_iter it;
 	struct intel_context *ce;
 	struct i915_vma *vma;
 	unsigned int flags = PIN_USER;
 	unsigned int n;
-	int err;
+	int err = 0;
 
 	/*
 	 * Sanity check shrinking huge-paged object -- make sure nothing blows
@@ -1533,12 +1537,14 @@ static int igt_shrink_thp(void *arg)
 
 	if (!igt_can_allocate_thp(i915)) {
 		pr_info("missing THP support, skipping\n");
-		return 0;
+		goto out_vm;
 	}
 
 	obj = i915_gem_object_create_shmem(i915, SZ_2M);
-	if (IS_ERR(obj))
-		return PTR_ERR(obj);
+	if (IS_ERR(obj)) {
+		err = PTR_ERR(obj);
+		goto out_vm;
+	}
 
 	vma = i915_vma_instance(obj, vm, NULL);
 	if (IS_ERR(vma)) {
@@ -1607,6 +1613,8 @@ static int igt_shrink_thp(void *arg)
 	i915_vma_close(vma);
 out_put:
 	i915_gem_object_put(obj);
+out_vm:
+	i915_vm_put(vm);
 
 	return err;
 }
@@ -1675,6 +1683,7 @@ int i915_gem_huge_page_live_selftests(struct drm_i915_private *i915)
 	};
 	struct drm_file *file;
 	struct i915_gem_context *ctx;
+	struct i915_address_space *vm;
 	intel_wakeref_t wakeref;
 	int err;
 
@@ -1699,8 +1708,11 @@ int i915_gem_huge_page_live_selftests(struct drm_i915_private *i915)
 		goto out_unlock;
 	}
 
-	if (ctx->vm)
-		ctx->vm->scrub_64K = true;
+	mutex_lock(&ctx->mutex);
+	vm = i915_gem_context_vm(ctx);
+	if (vm)
+		WRITE_ONCE(vm->scrub_64K, true);
+	mutex_unlock(&ctx->mutex);
 
 	err = i915_subtests(tests, ctx);
 

drivers/gpu/drm/i915/gem/selftests/mock_context.c

@@ -42,7 +42,10 @@ mock_context(struct drm_i915_private *i915,
 		if (!ppgtt)
 			goto err_put;
 
+		mutex_lock(&ctx->mutex);
 		__set_ppgtt(ctx, &ppgtt->vm);
+		mutex_unlock(&ctx->mutex);
+
 		i915_vm_put(&ppgtt->vm);
 	}
 
@@ -65,7 +68,7 @@ void mock_context_close(struct i915_gem_context *ctx)
 
 void mock_init_contexts(struct drm_i915_private *i915)
 {
-	init_contexts(i915);
+	init_contexts(&i915->gem.contexts);
 }
 
 struct i915_gem_context *
@@ -74,8 +77,6 @@ live_context(struct drm_i915_private *i915, struct drm_file *file)
 	struct i915_gem_context *ctx;
 	int err;
 
-	lockdep_assert_held(&i915->drm.struct_mutex);
-
 	ctx = i915_gem_create_context(i915, 0);
 	if (IS_ERR(ctx))
 		return ctx;

drivers/gpu/drm/i915/gt/intel_context.c

@@ -221,12 +221,20 @@ intel_context_init(struct intel_context *ce,
 		   struct i915_gem_context *ctx,
 		   struct intel_engine_cs *engine)
 {
+	struct i915_address_space *vm;
+
 	GEM_BUG_ON(!engine->cops);
 
 	kref_init(&ce->ref);
 
 	ce->gem_context = ctx;
-	ce->vm = i915_vm_get(ctx->vm ?: &engine->gt->ggtt->vm);
+	rcu_read_lock();
+	vm = rcu_dereference(ctx->vm);
+	if (vm)
+		ce->vm = i915_vm_get(vm);
+	else
+		ce->vm = i915_vm_get(&engine->gt->ggtt->vm);
+	rcu_read_unlock();
 	if (ctx->timeline)
 		ce->timeline = intel_timeline_get(ctx->timeline);
 

drivers/gpu/drm/i915/gt/selftest_context.c

@@ -155,13 +155,9 @@ static int live_context_size(void *arg)
 	 * HW tries to write past the end of one.
 	 */
 
-	mutex_lock(&gt->i915->drm.struct_mutex);
-
 	fixme = kernel_context(gt->i915);
-	if (IS_ERR(fixme)) {
-		err = PTR_ERR(fixme);
-		goto unlock;
-	}
+	if (IS_ERR(fixme))
+		return PTR_ERR(fixme);
 
 	for_each_engine(engine, gt->i915, id) {
 		struct {
@@ -201,8 +197,6 @@ static int live_context_size(void *arg)
 	}
 
 	kernel_context_close(fixme);
-unlock:
-	mutex_unlock(&gt->i915->drm.struct_mutex);
 	return err;
 }
 
@@ -305,12 +299,10 @@ static int live_active_context(void *arg)
 	if (IS_ERR(file))
 		return PTR_ERR(file);
 
-	mutex_lock(&gt->i915->drm.struct_mutex);
-
 	fixme = live_context(gt->i915, file);
 	if (IS_ERR(fixme)) {
 		err = PTR_ERR(fixme);
-		goto unlock;
+		goto out_file;
 	}
 
 	for_each_engine(engine, gt->i915, id) {
@@ -323,8 +315,7 @@ static int live_active_context(void *arg)
 			break;
 	}
 
-unlock:
-	mutex_unlock(&gt->i915->drm.struct_mutex);
+out_file:
 	mock_file_free(gt->i915, file);
 	return err;
 }
@@ -418,12 +409,10 @@ static int live_remote_context(void *arg)
 	if (IS_ERR(file))
 		return PTR_ERR(file);
 
-	mutex_lock(&gt->i915->drm.struct_mutex);
-
 	fixme = live_context(gt->i915, file);
 	if (IS_ERR(fixme)) {
 		err = PTR_ERR(fixme);
-		goto unlock;
+		goto out_file;
 	}
 
 	for_each_engine(engine, gt->i915, id) {
@@ -436,8 +425,7 @@ static int live_remote_context(void *arg)
 			break;
 	}
 
-unlock:
-	mutex_unlock(&gt->i915->drm.struct_mutex);
+out_file:
 	mock_file_free(gt->i915, file);
 	return err;
 }

drivers/gpu/drm/i915/gt/selftest_hangcheck.c

@@ -58,9 +58,7 @@ static int hang_init(struct hang *h, struct intel_gt *gt)
 	memset(h, 0, sizeof(*h));
 	h->gt = gt;
 
-	mutex_lock(&gt->i915->drm.struct_mutex);
 	h->ctx = kernel_context(gt->i915);
-	mutex_unlock(&gt->i915->drm.struct_mutex);
 	if (IS_ERR(h->ctx))
 		return PTR_ERR(h->ctx);
 
@@ -133,7 +131,7 @@ static struct i915_request *
 hang_create_request(struct hang *h, struct intel_engine_cs *engine)
 {
 	struct intel_gt *gt = h->gt;
-	struct i915_address_space *vm = h->ctx->vm ?: &engine->gt->ggtt->vm;
+	struct i915_address_space *vm = i915_gem_context_get_vm_rcu(h->ctx);
 	struct drm_i915_gem_object *obj;
 	struct i915_request *rq = NULL;
 	struct i915_vma *hws, *vma;
@@ -143,12 +141,15 @@ hang_create_request(struct hang *h, struct intel_engine_cs *engine)
 	int err;
 
 	obj = i915_gem_object_create_internal(gt->i915, PAGE_SIZE);
-	if (IS_ERR(obj))
+	if (IS_ERR(obj)) {
+		i915_vm_put(vm);
 		return ERR_CAST(obj);
+	}
 
 	vaddr = i915_gem_object_pin_map(obj, i915_coherent_map_type(gt->i915));
 	if (IS_ERR(vaddr)) {
 		i915_gem_object_put(obj);
+		i915_vm_put(vm);
 		return ERR_CAST(vaddr);
 	}
 
@@ -159,16 +160,22 @@ hang_create_request(struct hang *h, struct intel_engine_cs *engine)
 	h->batch = vaddr;
 
 	vma = i915_vma_instance(h->obj, vm, NULL);
-	if (IS_ERR(vma))
+	if (IS_ERR(vma)) {
+		i915_vm_put(vm);
 		return ERR_CAST(vma);
+	}
 
 	hws = i915_vma_instance(h->hws, vm, NULL);
-	if (IS_ERR(hws))
+	if (IS_ERR(hws)) {
+		i915_vm_put(vm);
 		return ERR_CAST(hws);
+	}
 
 	err = i915_vma_pin(vma, 0, 0, PIN_USER);
-	if (err)
+	if (err) {
+		i915_vm_put(vm);
 		return ERR_PTR(err);
+	}
 
 	err = i915_vma_pin(hws, 0, 0, PIN_USER);
 	if (err)
@@ -266,6 +273,7 @@ hang_create_request(struct hang *h, struct intel_engine_cs *engine)
 	i915_vma_unpin(hws);
 unpin_vma:
 	i915_vma_unpin(vma);
+	i915_vm_put(vm);
 	return err ? ERR_PTR(err) : rq;
 }
 
@@ -382,9 +390,7 @@ static int igt_reset_nop(void *arg)
 	if (IS_ERR(file))
 		return PTR_ERR(file);
 
-	mutex_lock(&gt->i915->drm.struct_mutex);
 	ctx = live_context(gt->i915, file);
-	mutex_unlock(&gt->i915->drm.struct_mutex);
 	if (IS_ERR(ctx)) {
 		err = PTR_ERR(ctx);
 		goto out;
@@ -458,9 +464,7 @@ static int igt_reset_nop_engine(void *arg)
 	if (IS_ERR(file))
 		return PTR_ERR(file);
 
-	mutex_lock(&gt->i915->drm.struct_mutex);
 	ctx = live_context(gt->i915, file);
-	mutex_unlock(&gt->i915->drm.struct_mutex);
 	if (IS_ERR(ctx)) {
 		err = PTR_ERR(ctx);
 		goto out;
@@ -705,9 +709,7 @@ static int active_engine(void *data)
 		return PTR_ERR(file);
 
 	for (count = 0; count < ARRAY_SIZE(ctx); count++) {
-		mutex_lock(&engine->i915->drm.struct_mutex);
 		ctx[count] = live_context(engine->i915, file);
-		mutex_unlock(&engine->i915->drm.struct_mutex);
 		if (IS_ERR(ctx[count])) {
 			err = PTR_ERR(ctx[count]);
 			while (--count)
@@ -1291,6 +1293,7 @@ static int igt_reset_evict_ppgtt(void *arg)
 {
 	struct intel_gt *gt = arg;
 	struct i915_gem_context *ctx;
+	struct i915_address_space *vm;
 	struct drm_file *file;
 	int err;
 
@@ -1298,18 +1301,20 @@ static int igt_reset_evict_ppgtt(void *arg)
 	if (IS_ERR(file))
 		return PTR_ERR(file);
 
-	mutex_lock(&gt->i915->drm.struct_mutex);
 	ctx = live_context(gt->i915, file);
-	mutex_unlock(&gt->i915->drm.struct_mutex);
 	if (IS_ERR(ctx)) {
 		err = PTR_ERR(ctx);
 		goto out;
 	}
 
 	err = 0;
-	if (ctx->vm) /* aliasing == global gtt locking, covered above */
-		err = __igt_reset_evict_vma(gt, ctx->vm,
+	vm = i915_gem_context_get_vm_rcu(ctx);
+	if (!i915_is_ggtt(vm)) {
+		/* aliasing == global gtt locking, covered above */
+		err = __igt_reset_evict_vma(gt, vm,
 					    evict_vma, EXEC_OBJECT_WRITE);
+	}
+	i915_vm_put(vm);
 
 out:
 	mock_file_free(gt->i915, file);

drivers/gpu/drm/i915/gt/selftest_lrc.c

@@ -1631,7 +1631,11 @@ static int smoke_submit(struct preempt_smoke *smoke,
 	int err = 0;
 
 	if (batch) {
-		vma = i915_vma_instance(batch, ctx->vm, NULL);
+		struct i915_address_space *vm;
+
+		vm = i915_gem_context_get_vm_rcu(ctx);
+		vma = i915_vma_instance(batch, vm, NULL);
+		i915_vm_put(vm);
 		if (IS_ERR(vma))
 			return PTR_ERR(vma);
 

drivers/gpu/drm/i915/gt/selftest_workarounds.c

@@ -260,7 +260,6 @@ switch_to_scratch_context(struct intel_engine_cs *engine,
 		rq = igt_spinner_create_request(spin, ce, MI_NOOP);
 
 	intel_context_put(ce);
-	kernel_context_close(ctx);
 
 	if (IS_ERR(rq)) {
 		spin = NULL;
@@ -279,6 +278,7 @@ switch_to_scratch_context(struct intel_engine_cs *engine,
 	if (err && spin)
 		igt_spinner_end(spin);
 
+	kernel_context_close(ctx);
 	return err;
 }
 
@@ -355,6 +355,7 @@ static int check_whitelist_across_reset(struct intel_engine_cs *engine,
 static struct i915_vma *create_batch(struct i915_gem_context *ctx)
 {
 	struct drm_i915_gem_object *obj;
+	struct i915_address_space *vm;
 	struct i915_vma *vma;
 	int err;
 
@@ -362,7 +363,9 @@ static struct i915_vma *create_batch(struct i915_gem_context *ctx)
 	if (IS_ERR(obj))
 		return ERR_CAST(obj);
 
-	vma = i915_vma_instance(obj, ctx->vm, NULL);
+	vm = i915_gem_context_get_vm_rcu(ctx);
+	vma = i915_vma_instance(obj, vm, NULL);
+	i915_vm_put(vm);
 	if (IS_ERR(vma)) {
 		err = PTR_ERR(vma);
 		goto err_obj;
@@ -463,12 +466,15 @@ static int check_dirty_whitelist(struct i915_gem_context *ctx,
 		0xffff00ff,
 		0xffffffff,
 	};
+	struct i915_address_space *vm;
 	struct i915_vma *scratch;
 	struct i915_vma *batch;
 	int err = 0, i, v;
 	u32 *cs, *results;
 
-	scratch = create_scratch(ctx->vm, 2 * ARRAY_SIZE(values) + 1);
+	vm = i915_gem_context_get_vm_rcu(ctx);
+	scratch = create_scratch(vm, 2 * ARRAY_SIZE(values) + 1);
+	i915_vm_put(vm);
 	if (IS_ERR(scratch))
 		return PTR_ERR(scratch);
 
@@ -1010,6 +1016,7 @@ static int live_isolated_whitelist(void *arg)
 		return 0;
 
 	for (i = 0; i < ARRAY_SIZE(client); i++) {
+		struct i915_address_space *vm;
 		struct i915_gem_context *c;
 
 		c = kernel_context(i915);
@@ -1018,22 +1025,27 @@ static int live_isolated_whitelist(void *arg)
 			goto err;
 		}
 
-		client[i].scratch[0] = create_scratch(c->vm, 1024);
+		vm = i915_gem_context_get_vm_rcu(c);
+
+		client[i].scratch[0] = create_scratch(vm, 1024);
 		if (IS_ERR(client[i].scratch[0])) {
 			err = PTR_ERR(client[i].scratch[0]);
+			i915_vm_put(vm);
 			kernel_context_close(c);
 			goto err;
 		}
 
-		client[i].scratch[1] = create_scratch(c->vm, 1024);
+		client[i].scratch[1] = create_scratch(vm, 1024);
 		if (IS_ERR(client[i].scratch[1])) {
 			err = PTR_ERR(client[i].scratch[1]);
 			i915_vma_unpin_and_release(&client[i].scratch[0], 0);
+			i915_vm_put(vm);
 			kernel_context_close(c);
 			goto err;
 		}
 
 		client[i].ctx = c;
+		i915_vm_put(vm);
 	}
 
 	for_each_engine(engine, i915, id) {

drivers/gpu/drm/i915/gvt/scheduler.c

@@ -365,7 +365,8 @@ static void set_context_ppgtt_from_shadow(struct intel_vgpu_workload *workload,
 					  struct i915_gem_context *ctx)
 {
 	struct intel_vgpu_mm *mm = workload->shadow_mm;
-	struct i915_ppgtt *ppgtt = i915_vm_to_ppgtt(ctx->vm);
+	struct i915_ppgtt *ppgtt =
+		i915_vm_to_ppgtt(i915_gem_context_get_vm_rcu(ctx));
 	int i = 0;
 
 	if (mm->ppgtt_mm.root_entry_type == GTT_TYPE_PPGTT_ROOT_L4_ENTRY) {
@@ -378,6 +379,8 @@ static void set_context_ppgtt_from_shadow(struct intel_vgpu_workload *workload,
 			px_dma(pd) = mm->ppgtt_mm.shadow_pdps[i];
 		}
 	}
+
+	i915_vm_put(&ppgtt->vm);
 }
 
 static int
@@ -1213,20 +1216,18 @@ int intel_vgpu_setup_submission(struct intel_vgpu *vgpu)
 	struct intel_vgpu_submission *s = &vgpu->submission;
 	struct intel_engine_cs *engine;
 	struct i915_gem_context *ctx;
+	struct i915_ppgtt *ppgtt;
 	enum intel_engine_id i;
 	int ret;
 
-	mutex_lock(&i915->drm.struct_mutex);
-
 	ctx = i915_gem_context_create_kernel(i915, I915_PRIORITY_MAX);
-	if (IS_ERR(ctx)) {
-		ret = PTR_ERR(ctx);
-		goto out_unlock;
-	}
+	if (IS_ERR(ctx))
+		return PTR_ERR(ctx);
 
 	i915_gem_context_set_force_single_submission(ctx);
 
-	i915_context_ppgtt_root_save(s, i915_vm_to_ppgtt(ctx->vm));
+	ppgtt = i915_vm_to_ppgtt(i915_gem_context_get_vm_rcu(ctx));
+	i915_context_ppgtt_root_save(s, ppgtt);
 
 	for_each_engine(engine, i915, i) {
 		struct intel_context *ce;
@@ -1271,12 +1272,12 @@ int intel_vgpu_setup_submission(struct intel_vgpu *vgpu)
 	atomic_set(&s->running_workload_num, 0);
 	bitmap_zero(s->tlb_handle_pending, I915_NUM_ENGINES);
 
+	i915_vm_put(&ppgtt->vm);
 	i915_gem_context_put(ctx);
-	mutex_unlock(&i915->drm.struct_mutex);
 	return 0;
 
 out_shadow_ctx:
-	i915_context_ppgtt_root_restore(s, i915_vm_to_ppgtt(ctx->vm));
+	i915_context_ppgtt_root_restore(s, ppgtt);
 	for_each_engine(engine, i915, i) {
 		if (IS_ERR(s->shadow[i]))
 			break;
@@ -1284,9 +1285,8 @@ int intel_vgpu_setup_submission(struct intel_vgpu *vgpu)
 		intel_context_unpin(s->shadow[i]);
 		intel_context_put(s->shadow[i]);
 	}
+	i915_vm_put(&ppgtt->vm);
 	i915_gem_context_put(ctx);
-out_unlock:
-	mutex_unlock(&i915->drm.struct_mutex);
 	return ret;
 }
 

drivers/gpu/drm/i915/i915_debugfs.c

@@ -316,12 +316,18 @@ static void print_context_stats(struct seq_file *m,
 				struct drm_i915_private *i915)
 {
 	struct file_stats kstats = {};
-	struct i915_gem_context *ctx;
+	struct i915_gem_context *ctx, *cn;
 
-	list_for_each_entry(ctx, &i915->contexts.list, link) {
+	spin_lock(&i915->gem.contexts.lock);
+	list_for_each_entry_safe(ctx, cn, &i915->gem.contexts.list, link) {
 		struct i915_gem_engines_iter it;
 		struct intel_context *ce;
 
+		if (!kref_get_unless_zero(&ctx->ref))
+			continue;
+
+		spin_unlock(&i915->gem.contexts.lock);
+
 		for_each_gem_engine(ce,
 				    i915_gem_context_lock_engines(ctx), it) {
 			intel_context_lock_pinned(ce);
@@ -338,7 +344,9 @@ static void print_context_stats(struct seq_file *m,
 		i915_gem_context_unlock_engines(ctx);
 
 		if (!IS_ERR_OR_NULL(ctx->file_priv)) {
-			struct file_stats stats = { .vm = ctx->vm, };
+			struct file_stats stats = {
+				.vm = rcu_access_pointer(ctx->vm),
+			};
 			struct drm_file *file = ctx->file_priv->file;
 			struct task_struct *task;
 			char name[80];
@@ -355,7 +363,12 @@ static void print_context_stats(struct seq_file *m,
 
 			print_file_stats(m, name, stats);
 		}
+
+		spin_lock(&i915->gem.contexts.lock);
+		list_safe_reset_next(ctx, cn, link);
+		i915_gem_context_put(ctx);
 	}
+	spin_unlock(&i915->gem.contexts.lock);
 
 	print_file_stats(m, "[k]contexts", kstats);
 }
@@ -363,7 +376,6 @@ static void print_context_stats(struct seq_file *m,
 static int i915_gem_object_info(struct seq_file *m, void *data)
 {
 	struct drm_i915_private *i915 = node_to_i915(m->private);
-	int ret;
 
 	seq_printf(m, "%u shrinkable [%u free] objects, %llu bytes\n",
 		   i915->mm.shrink_count,
@@ -372,12 +384,7 @@ static int i915_gem_object_info(struct seq_file *m, void *data)
 
 	seq_putc(m, '\n');
 
-	ret = mutex_lock_interruptible(&i915->drm.struct_mutex);
-	if (ret)
-		return ret;
-
 	print_context_stats(m, i915);
-	mutex_unlock(&i915->drm.struct_mutex);
 
 	return 0;
 }
@@ -1579,19 +1586,19 @@ static void describe_ctx_ring(struct seq_file *m, struct intel_ring *ring)
 
 static int i915_context_status(struct seq_file *m, void *unused)
 {
-	struct drm_i915_private *dev_priv = node_to_i915(m->private);
-	struct drm_device *dev = &dev_priv->drm;
-	struct i915_gem_context *ctx;
-	int ret;
-
-	ret = mutex_lock_interruptible(&dev->struct_mutex);
-	if (ret)
-		return ret;
+	struct drm_i915_private *i915 = node_to_i915(m->private);
+	struct i915_gem_context *ctx, *cn;
 
-	list_for_each_entry(ctx, &dev_priv->contexts.list, link) {
+	spin_lock(&i915->gem.contexts.lock);
+	list_for_each_entry_safe(ctx, cn, &i915->gem.contexts.list, link) {
 		struct i915_gem_engines_iter it;
 		struct intel_context *ce;
 
+		if (!kref_get_unless_zero(&ctx->ref))
+			continue;
+
+		spin_unlock(&i915->gem.contexts.lock);
+
 		seq_puts(m, "HW context ");
 		if (ctx->pid) {
 			struct task_struct *task;
@@ -1626,9 +1633,12 @@ static int i915_context_status(struct seq_file *m, void *unused)
 		i915_gem_context_unlock_engines(ctx);
 
 		seq_putc(m, '\n');
-	}
 
-	mutex_unlock(&dev->struct_mutex);
+		spin_lock(&i915->gem.contexts.lock);
+		list_safe_reset_next(ctx, cn, link);
+		i915_gem_context_put(ctx);
+	}
+	spin_unlock(&i915->gem.contexts.lock);
 
 	return 0;
 }

drivers/gpu/drm/i915/i915_drv.c

@@ -1702,10 +1702,8 @@ static void i915_driver_postclose(struct drm_device *dev, struct drm_file *file)
 {
 	struct drm_i915_file_private *file_priv = file->driver_priv;
 
-	mutex_lock(&dev->struct_mutex);
 	i915_gem_context_close(file);
 	i915_gem_release(dev, file);
-	mutex_unlock(&dev->struct_mutex);
 
 	kfree_rcu(file_priv, rcu);
 

drivers/gpu/drm/i915/i915_drv.h

@@ -1536,13 +1536,6 @@ struct drm_i915_private {
 	int audio_power_refcount;
 	u32 audio_freq_cntrl;
 
-	struct {
-		struct mutex mutex;
-		struct list_head list;
-		struct llist_head free_list;
-		struct work_struct free_work;
-	} contexts;
-
 	u32 fdi_rx_config;
 
 	/* Shadow for DISPLAY_PHY_CONTROL which can't be safely read */
@@ -1698,6 +1691,14 @@ struct drm_i915_private {
 
 	struct {
 		struct notifier_block pm_notifier;
+
+		struct i915_gem_contexts {
+			spinlock_t lock; /* locks list */
+			struct list_head list;
+
+			struct llist_head free_list;
+			struct work_struct free_work;
+		} contexts;
 	} gem;
 
 	/* For i945gm vblank irq vs. C3 workaround */

drivers/gpu/drm/i915/i915_gem.c

@@ -1266,7 +1266,7 @@ int i915_gem_init(struct drm_i915_private *dev_priv)
 		goto err_unlock;
 	}
 
-	ret = i915_gem_contexts_init(dev_priv);
+	ret = i915_gem_init_contexts(dev_priv);
 	if (ret) {
 		GEM_BUG_ON(ret == -EIO);
 		goto err_scratch;
@@ -1348,7 +1348,7 @@ int i915_gem_init(struct drm_i915_private *dev_priv)
 	}
 err_context:
 	if (ret != -EIO)
-		i915_gem_contexts_fini(dev_priv);
+		i915_gem_driver_release__contexts(dev_priv);
 err_scratch:
 	intel_gt_driver_release(&dev_priv->gt);
 err_unlock:
@@ -1416,11 +1416,9 @@ void i915_gem_driver_remove(struct drm_i915_private *dev_priv)
 
 void i915_gem_driver_release(struct drm_i915_private *dev_priv)
 {
-	mutex_lock(&dev_priv->drm.struct_mutex);
 	intel_engines_cleanup(dev_priv);
-	i915_gem_contexts_fini(dev_priv);
+	i915_gem_driver_release__contexts(dev_priv);
 	intel_gt_driver_release(&dev_priv->gt);
-	mutex_unlock(&dev_priv->drm.struct_mutex);
 
 	intel_wa_list_free(&dev_priv->gt_wa_list);
 
@@ -1430,7 +1428,7 @@ void i915_gem_driver_release(struct drm_i915_private *dev_priv)
 
 	i915_gem_drain_freed_objects(dev_priv);
 
-	WARN_ON(!list_empty(&dev_priv->contexts.list));
+	WARN_ON(!list_empty(&dev_priv->gem.contexts.list));
 }
 
 void i915_gem_init_mmio(struct drm_i915_private *i915)

drivers/gpu/drm/i915/i915_gem_gtt.c

@@ -1366,7 +1366,9 @@ static int gen8_init_scratch(struct i915_address_space *vm)
 	if (vm->has_read_only &&
 	    vm->i915->kernel_context &&
 	    vm->i915->kernel_context->vm) {
-		struct i915_address_space *clone = vm->i915->kernel_context->vm;
+		struct i915_address_space *clone =
+			rcu_dereference_protected(vm->i915->kernel_context->vm,
+						  true); /* static */
 
 		GEM_BUG_ON(!clone->has_read_only);
 

drivers/gpu/drm/i915/i915_perf.c

@@ -1853,8 +1853,8 @@ static int gen8_configure_all_contexts(struct i915_perf_stream *stream,
 	};
 #undef ctx_flexeuN
 	struct intel_engine_cs *engine;
-	struct i915_gem_context *ctx;
-	int i;
+	struct i915_gem_context *ctx, *cn;
+	int i, err;
 
 	for (i = 2; i < ARRAY_SIZE(regs); i++)
 		regs[i].value = oa_config_flex_reg(oa_config, regs[i].reg);
@@ -1877,16 +1877,27 @@ static int gen8_configure_all_contexts(struct i915_perf_stream *stream,
 	 * context. Contexts idle at the time of reconfiguration are not
 	 * trapped behind the barrier.
 	 */
-	list_for_each_entry(ctx, &i915->contexts.list, link) {
-		int err;
-
+	spin_lock(&i915->gem.contexts.lock);
+	list_for_each_entry_safe(ctx, cn, &i915->gem.contexts.list, link) {
 		if (ctx == i915->kernel_context)
 			continue;
 
+		if (!kref_get_unless_zero(&ctx->ref))
+			continue;
+
+		spin_unlock(&i915->gem.contexts.lock);
+
 		err = gen8_configure_context(ctx, regs, ARRAY_SIZE(regs));
-		if (err)
+		if (err) {
+			i915_gem_context_put(ctx);
 			return err;
+		}
+
+		spin_lock(&i915->gem.contexts.lock);
+		list_safe_reset_next(ctx, cn, link);
+		i915_gem_context_put(ctx);
 	}
+	spin_unlock(&i915->gem.contexts.lock);
 
 	/*
 	 * After updating all other contexts, we need to modify ourselves.
@@ -1895,7 +1906,6 @@ static int gen8_configure_all_contexts(struct i915_perf_stream *stream,
 	 */
 	for_each_uabi_engine(engine, i915) {
 		struct intel_context *ce = engine->kernel_context;
-		int err;
 
 		if (engine->class != RENDER_CLASS)
 			continue;

drivers/gpu/drm/i915/i915_sysfs.c

@@ -176,16 +176,12 @@ i915_l3_read(struct file *filp, struct kobject *kobj,
 	count = min_t(size_t, GEN7_L3LOG_SIZE - offset, count);
 	memset(buf, 0, count);
 
-	ret = i915_mutex_lock_interruptible(&i915->drm);
-	if (ret)
-		return ret;
-
+	spin_lock(&i915->gem.contexts.lock);
 	if (i915->l3_parity.remap_info[slice])
 		memcpy(buf,
 		       i915->l3_parity.remap_info[slice] + offset / sizeof(u32),
 		       count);
-
-	mutex_unlock(&i915->drm.struct_mutex);
+	spin_unlock(&i915->gem.contexts.lock);
 
 	return count;
 }
@@ -198,8 +194,8 @@ i915_l3_write(struct file *filp, struct kobject *kobj,
 	struct device *kdev = kobj_to_dev(kobj);
 	struct drm_i915_private *i915 = kdev_minor_to_i915(kdev);
 	int slice = (int)(uintptr_t)attr->private;
+	u32 *remap_info, *freeme = NULL;
 	struct i915_gem_context *ctx;
-	u32 **remap_info;
 	int ret;
 
 	ret = l3_access_valid(i915, offset);
@@ -209,37 +205,36 @@ i915_l3_write(struct file *filp, struct kobject *kobj,
 	if (count < sizeof(u32))
 		return -EINVAL;
 
-	ret = i915_mutex_lock_interruptible(&i915->drm);
-	if (ret)
-		return ret;
+	remap_info = kzalloc(GEN7_L3LOG_SIZE, GFP_KERNEL);
+	if (!remap_info)
+		return -ENOMEM;
 
-	remap_info = &i915->l3_parity.remap_info[slice];
-	if (!*remap_info) {
-		*remap_info = kzalloc(GEN7_L3LOG_SIZE, GFP_KERNEL);
-		if (!*remap_info) {
-			ret = -ENOMEM;
-			goto out;
-		}
+	spin_lock(&i915->gem.contexts.lock);
+
+	if (i915->l3_parity.remap_info[slice]) {
+		freeme = remap_info;
+		remap_info = i915->l3_parity.remap_info[slice];
+	} else {
+		i915->l3_parity.remap_info[slice] = remap_info;
 	}
 
 	count = round_down(count, sizeof(u32));
-	memcpy(*remap_info + offset / sizeof(u32), buf, count);
+	memcpy(remap_info + offset / sizeof(u32), buf, count);
 
 	/* NB: We defer the remapping until we switch to the context */
-	list_for_each_entry(ctx, &i915->contexts.list, link)
+	list_for_each_entry(ctx, &i915->gem.contexts.list, link)
 		ctx->remap_slice |= BIT(slice);
 
+	spin_unlock(&i915->gem.contexts.lock);
+	kfree(freeme);
+
 	/*
 	 * TODO: Ideally we really want a GPU reset here to make sure errors
 	 * aren't propagated. Since I cannot find a stable way to reset the GPU
 	 * at this point it is left as a TODO.
 	*/
 
-	ret = count;
-out:
-	mutex_unlock(&i915->drm.struct_mutex);
-
-	return ret;
+	return count;
 }
 
 static const struct bin_attribute dpf_attrs = {

drivers/gpu/drm/i915/i915_trace.h

@@ -952,7 +952,7 @@ DECLARE_EVENT_CLASS(i915_context,
 	TP_fast_assign(
 			__entry->dev = ctx->i915->drm.primary->index;
 			__entry->ctx = ctx;
-			__entry->vm = ctx->vm;
+			__entry->vm = rcu_access_pointer(ctx->vm);
 	),
 
 	TP_printk("dev=%u, ctx=%p, ctx_vm=%p",

drivers/gpu/drm/i915/selftests/i915_gem.c

@@ -138,11 +138,9 @@ static int igt_gem_suspend(void *arg)
 		return PTR_ERR(file);
 
 	err = -ENOMEM;
-	mutex_lock(&i915->drm.struct_mutex);
 	ctx = live_context(i915, file);
 	if (!IS_ERR(ctx))
 		err = switch_to_context(i915, ctx);
-	mutex_unlock(&i915->drm.struct_mutex);
 	if (err)
 		goto out;
 
@@ -157,9 +155,7 @@ static int igt_gem_suspend(void *arg)
 
 	pm_resume(i915);
 
-	mutex_lock(&i915->drm.struct_mutex);
 	err = switch_to_context(i915, ctx);
-	mutex_unlock(&i915->drm.struct_mutex);
 out:
 	mock_file_free(i915, file);
 	return err;
@@ -177,11 +173,9 @@ static int igt_gem_hibernate(void *arg)
 		return PTR_ERR(file);
 
 	err = -ENOMEM;
-	mutex_lock(&i915->drm.struct_mutex);
 	ctx = live_context(i915, file);
 	if (!IS_ERR(ctx))
 		err = switch_to_context(i915, ctx);
-	mutex_unlock(&i915->drm.struct_mutex);
 	if (err)
 		goto out;
 
@@ -196,9 +190,7 @@ static int igt_gem_hibernate(void *arg)
 
 	pm_resume(i915);
 
-	mutex_lock(&i915->drm.struct_mutex);
 	err = switch_to_context(i915, ctx);
-	mutex_unlock(&i915->drm.struct_mutex);
 out:
 	mock_file_free(i915, file);
 	return err;

drivers/gpu/drm/i915/selftests/i915_gem_evict.c

@@ -473,7 +473,6 @@ static int igt_evict_contexts(void *arg)
 		}
 
 		count = 0;
-		mutex_lock(&i915->drm.struct_mutex);
 		onstack_fence_init(&fence);
 		do {
 			struct i915_request *rq;
@@ -510,8 +509,6 @@ static int igt_evict_contexts(void *arg)
 			count++;
 			err = 0;
 		} while(1);
-		mutex_unlock(&i915->drm.struct_mutex);
-
 		onstack_fence_fini(&fence);
 		pr_info("Submitted %lu contexts/requests on %s\n",
 			count, engine->name);

drivers/gpu/drm/i915/selftests/i915_gem_gtt.c

@@ -1246,6 +1246,7 @@ static int exercise_mock(struct drm_i915_private *i915,
 				     unsigned long end_time))
 {
 	const u64 limit = totalram_pages() << PAGE_SHIFT;
+	struct i915_address_space *vm;
 	struct i915_gem_context *ctx;
 	IGT_TIMEOUT(end_time);
 	int err;
@@ -1254,7 +1255,9 @@ static int exercise_mock(struct drm_i915_private *i915,
 	if (!ctx)
 		return -ENOMEM;
 
-	err = func(i915, ctx->vm, 0, min(ctx->vm->total, limit), end_time);
+	vm = i915_gem_context_get_vm_rcu(ctx);
+	err = func(i915, vm, 0, min(vm->total, limit), end_time);
+	i915_vm_put(vm);
 
 	mock_context_close(ctx);
 	return err;
@@ -1801,15 +1804,15 @@ static int igt_cs_tlb(void *arg)
 		goto out_unlock;
 	}
 
-	vm = ctx->vm;
-	if (!vm)
-		goto out_unlock;
+	vm = i915_gem_context_get_vm_rcu(ctx);
+	if (i915_is_ggtt(vm))
+		goto out_vm;
 
 	/* Create two pages; dummy we prefill the TLB, and intended */
 	bbe = i915_gem_object_create_internal(i915, PAGE_SIZE);
 	if (IS_ERR(bbe)) {
 		err = PTR_ERR(bbe);
-		goto out_unlock;
+		goto out_vm;
 	}
 
 	batch = i915_gem_object_pin_map(bbe, I915_MAP_WC);
@@ -2014,6 +2017,8 @@ static int igt_cs_tlb(void *arg)
 	i915_gem_object_put(act);
 out_put_bbe:
 	i915_gem_object_put(bbe);
+out_vm:
+	i915_vm_put(vm);
 out_unlock:
 	mutex_unlock(&i915->drm.struct_mutex);
 	mock_file_free(i915, file);

drivers/gpu/drm/i915/selftests/i915_request.c

@@ -181,9 +181,7 @@ static int igt_request_rewind(void *arg)
 	struct intel_context *ce;
 	int err = -EINVAL;
 
-	mutex_lock(&i915->drm.struct_mutex);
 	ctx[0] = mock_context(i915, "A");
-	mutex_unlock(&i915->drm.struct_mutex);
 
 	ce = i915_gem_context_get_engine(ctx[0], RCS0);
 	GEM_BUG_ON(IS_ERR(ce));
@@ -197,9 +195,7 @@ static int igt_request_rewind(void *arg)
 	i915_request_get(request);
 	i915_request_add(request);
 
-	mutex_lock(&i915->drm.struct_mutex);
 	ctx[1] = mock_context(i915, "B");
-	mutex_unlock(&i915->drm.struct_mutex);
 
 	ce = i915_gem_context_get_engine(ctx[1], RCS0);
 	GEM_BUG_ON(IS_ERR(ce));
@@ -438,9 +434,7 @@ static int mock_breadcrumbs_smoketest(void *arg)
 	}
 
 	for (n = 0; n < t.ncontexts; n++) {
-		mutex_lock(&t.engine->i915->drm.struct_mutex);
 		t.contexts[n] = mock_context(t.engine->i915, "mock");
-		mutex_unlock(&t.engine->i915->drm.struct_mutex);
 		if (!t.contexts[n]) {
 			ret = -ENOMEM;
 			goto out_contexts;
@@ -734,9 +728,9 @@ static int live_empty_request(void *arg)
 static struct i915_vma *recursive_batch(struct drm_i915_private *i915)
 {
 	struct i915_gem_context *ctx = i915->kernel_context;
-	struct i915_address_space *vm = ctx->vm ?: &i915->ggtt.vm;
 	struct drm_i915_gem_object *obj;
 	const int gen = INTEL_GEN(i915);
+	struct i915_address_space *vm;
 	struct i915_vma *vma;
 	u32 *cmd;
 	int err;
@@ -745,7 +739,9 @@ static struct i915_vma *recursive_batch(struct drm_i915_private *i915)
 	if (IS_ERR(obj))
 		return ERR_CAST(obj);
 
+	vm = i915_gem_context_get_vm_rcu(ctx);
 	vma = i915_vma_instance(obj, vm, NULL);
+	i915_vm_put(vm);
 	if (IS_ERR(vma)) {
 		err = PTR_ERR(vma);
 		goto err;
@@ -1220,9 +1216,7 @@ static int live_breadcrumbs_smoketest(void *arg)
 	}
 
 	for (n = 0; n < t[0].ncontexts; n++) {
-		mutex_lock(&i915->drm.struct_mutex);
 		t[0].contexts[n] = live_context(i915, file);
-		mutex_unlock(&i915->drm.struct_mutex);
 		if (!t[0].contexts[n]) {
 			ret = -ENOMEM;
 			goto out_contexts;

drivers/gpu/drm/i915/selftests/i915_vma.c

@@ -24,6 +24,7 @@
 
 #include <linux/prime_numbers.h>
 
+#include "gem/i915_gem_context.h"
 #include "gem/selftests/mock_context.h"
 
 #include "i915_scatterlist.h"
@@ -38,7 +39,7 @@ static bool assert_vma(struct i915_vma *vma,
 {
 	bool ok = true;
 
-	if (vma->vm != ctx->vm) {
+	if (vma->vm != rcu_access_pointer(ctx->vm)) {
 		pr_err("VMA created with wrong VM\n");
 		ok = false;
 	}
@@ -113,11 +114,13 @@ static int create_vmas(struct drm_i915_private *i915,
 	list_for_each_entry(obj, objects, st_link) {
 		for (pinned = 0; pinned <= 1; pinned++) {
 			list_for_each_entry(ctx, contexts, link) {
-				struct i915_address_space *vm = ctx->vm;
+				struct i915_address_space *vm;
 				struct i915_vma *vma;
 				int err;
 
+				vm = i915_gem_context_get_vm_rcu(ctx);
 				vma = checked_vma_instance(obj, vm, NULL);
+				i915_vm_put(vm);
 				if (IS_ERR(vma))
 					return PTR_ERR(vma);
 

drivers/gpu/drm/i915/selftests/mock_gem_device.c

@@ -59,11 +59,9 @@ static void mock_device_release(struct drm_device *dev)
 
 	i915_gem_drain_workqueue(i915);
 
-	mutex_lock(&i915->drm.struct_mutex);
 	for_each_engine(engine, i915, id)
 		mock_engine_free(engine);
-	i915_gem_contexts_fini(i915);
-	mutex_unlock(&i915->drm.struct_mutex);
+	i915_gem_driver_release__contexts(i915);
 
 	intel_timelines_fini(i915);
 
@@ -206,7 +204,7 @@ struct drm_i915_private *mock_gem_device(void)
 	return i915;
 
 err_context:
-	i915_gem_contexts_fini(i915);
+	i915_gem_driver_release__contexts(i915);
 err_engine:
 	mock_engine_free(i915->engine[RCS0]);
 err_unlock: