drm/i915/gtt: Defer address space cleanup to an RCU worker

Enable RCU protection of i915_address_space and its ppgtt superclasses,
and defer its cleanup into a worker executed after an RCU grace period.

In the future we will be able to use the RCU protection to reduce the
locking around VM lookups, but the immediate benefit is being able to
defer the release into a kworker (process context). This is required as
we may need to sleep to reap the WC pages stashed away inside the ppgtt.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=110934Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190620183705.31006-1-chris@chris-wilson.co.uk

drivers/gpu/drm/i915/Makefile.header-test

@@ -6,6 +6,7 @@ header_test := \
 	i915_active_types.h \
 	i915_debugfs.h \
 	i915_drv.h \
+	i915_gem_gtt.h \
 	i915_irq.h \
 	i915_params.h \
 	i915_priolist_types.h \

drivers/gpu/drm/i915/i915_gem_gtt.c

@@ -482,9 +482,69 @@ static void vm_free_page(struct i915_address_space *vm, struct page *page)
 	spin_unlock(&vm->free_pages.lock);
 }
 
+static void i915_address_space_fini(struct i915_address_space *vm)
+{
+	spin_lock(&vm->free_pages.lock);
+	if (pagevec_count(&vm->free_pages.pvec))
+		vm_free_pages_release(vm, true);
+	GEM_BUG_ON(pagevec_count(&vm->free_pages.pvec));
+	spin_unlock(&vm->free_pages.lock);
+
+	drm_mm_takedown(&vm->mm);
+
+	mutex_destroy(&vm->mutex);
+}
+
+static void ppgtt_destroy_vma(struct i915_address_space *vm)
+{
+	struct list_head *phases[] = {
+		&vm->bound_list,
+		&vm->unbound_list,
+		NULL,
+	}, **phase;
+
+	mutex_lock(&vm->i915->drm.struct_mutex);
+	for (phase = phases; *phase; phase++) {
+		struct i915_vma *vma, *vn;
+
+		list_for_each_entry_safe(vma, vn, *phase, vm_link)
+			i915_vma_destroy(vma);
+	}
+	mutex_unlock(&vm->i915->drm.struct_mutex);
+}
+
+static void __i915_vm_release(struct work_struct *work)
+{
+	struct i915_address_space *vm =
+		container_of(work, struct i915_address_space, rcu.work);
+
+	ppgtt_destroy_vma(vm);
+
+	GEM_BUG_ON(!list_empty(&vm->bound_list));
+	GEM_BUG_ON(!list_empty(&vm->unbound_list));
+
+	vm->cleanup(vm);
+	i915_address_space_fini(vm);
+
+	kfree(vm);
+}
+
+void i915_vm_release(struct kref *kref)
+{
+	struct i915_address_space *vm =
+		container_of(kref, struct i915_address_space, ref);
+
+	GEM_BUG_ON(i915_is_ggtt(vm));
+	trace_i915_ppgtt_release(vm);
+
+	vm->closed = true;
+	queue_rcu_work(vm->i915->wq, &vm->rcu);
+}
+
 static void i915_address_space_init(struct i915_address_space *vm, int subclass)
 {
 	kref_init(&vm->ref);
+	INIT_RCU_WORK(&vm->rcu, __i915_vm_release);
 
 	/*
 	 * The vm->mutex must be reclaim safe (for use in the shrinker).
@@ -505,19 +565,6 @@ static void i915_address_space_init(struct i915_address_space *vm, int subclass)
 	INIT_LIST_HEAD(&vm->bound_list);
 }
 
-static void i915_address_space_fini(struct i915_address_space *vm)
-{
-	spin_lock(&vm->free_pages.lock);
-	if (pagevec_count(&vm->free_pages.pvec))
-		vm_free_pages_release(vm, true);
-	GEM_BUG_ON(pagevec_count(&vm->free_pages.pvec));
-	spin_unlock(&vm->free_pages.lock);
-
-	drm_mm_takedown(&vm->mm);
-
-	mutex_destroy(&vm->mutex);
-}
-
 static int __setup_page_dma(struct i915_address_space *vm,
 			    struct i915_page_dma *p,
 			    gfp_t gfp)
@@ -1909,62 +1956,15 @@ static void gen6_ppgtt_free_pd(struct gen6_ppgtt *ppgtt)
 			free_pt(&ppgtt->base.vm, pt);
 }
 
-struct gen6_ppgtt_cleanup_work {
-	struct work_struct base;
-	struct i915_vma *vma;
-};
-
-static void gen6_ppgtt_cleanup_work(struct work_struct *wrk)
-{
-	struct gen6_ppgtt_cleanup_work *work =
-		container_of(wrk, typeof(*work), base);
-	/* Side note, vma->vm is the GGTT not the ppgtt we just destroyed! */
-	struct drm_i915_private *i915 = work->vma->vm->i915;
-
-	mutex_lock(&i915->drm.struct_mutex);
-	i915_vma_destroy(work->vma);
-	mutex_unlock(&i915->drm.struct_mutex);
-
-	kfree(work);
-}
-
-static int nop_set_pages(struct i915_vma *vma)
-{
-	return -ENODEV;
-}
-
-static void nop_clear_pages(struct i915_vma *vma)
-{
-}
-
-static int nop_bind(struct i915_vma *vma,
-		    enum i915_cache_level cache_level,
-		    u32 unused)
-{
-	return -ENODEV;
-}
-
-static void nop_unbind(struct i915_vma *vma)
-{
-}
-
-static const struct i915_vma_ops nop_vma_ops = {
-	.set_pages = nop_set_pages,
-	.clear_pages = nop_clear_pages,
-	.bind_vma = nop_bind,
-	.unbind_vma = nop_unbind,
-};
-
 static void gen6_ppgtt_cleanup(struct i915_address_space *vm)
 {
 	struct gen6_ppgtt *ppgtt = to_gen6_ppgtt(i915_vm_to_ppgtt(vm));
-	struct gen6_ppgtt_cleanup_work *work = ppgtt->work;
+	struct drm_i915_private *i915 = vm->i915;
 
 	/* FIXME remove the struct_mutex to bring the locking under control */
-	INIT_WORK(&work->base, gen6_ppgtt_cleanup_work);
-	work->vma = ppgtt->vma;
-	work->vma->ops = &nop_vma_ops;
-	schedule_work(&work->base);
+	mutex_lock(&i915->drm.struct_mutex);
+	i915_vma_destroy(ppgtt->vma);
+	mutex_unlock(&i915->drm.struct_mutex);
 
 	gen6_ppgtt_free_pd(ppgtt);
 	gen6_ppgtt_free_scratch(vm);
@@ -2146,16 +2146,10 @@ static struct i915_ppgtt *gen6_ppgtt_create(struct drm_i915_private *i915)
 
 	ppgtt->base.vm.pte_encode = ggtt->vm.pte_encode;
 
-	ppgtt->work = kmalloc(sizeof(*ppgtt->work), GFP_KERNEL);
-	if (!ppgtt->work) {
-		err = -ENOMEM;
-		goto err_free;
-	}
-
 	ppgtt->base.pd = __alloc_pd();
 	if (!ppgtt->base.pd) {
 		err = -ENOMEM;
-		goto err_work;
+		goto err_free;
 	}
 
 	err = gen6_ppgtt_init_scratch(ppgtt);
@@ -2174,8 +2168,6 @@ static struct i915_ppgtt *gen6_ppgtt_create(struct drm_i915_private *i915)
 	gen6_ppgtt_free_scratch(&ppgtt->base.vm);
 err_pd:
 	kfree(ppgtt->base.pd);
-err_work:
-	kfree(ppgtt->work);
 err_free:
 	kfree(ppgtt);
 	return ERR_PTR(err);
@@ -2250,42 +2242,6 @@ i915_ppgtt_create(struct drm_i915_private *i915)
 	return ppgtt;
 }
 
-static void ppgtt_destroy_vma(struct i915_address_space *vm)
-{
-	struct list_head *phases[] = {
-		&vm->bound_list,
-		&vm->unbound_list,
-		NULL,
-	}, **phase;
-
-	vm->closed = true;
-	for (phase = phases; *phase; phase++) {
-		struct i915_vma *vma, *vn;
-
-		list_for_each_entry_safe(vma, vn, *phase, vm_link)
-			i915_vma_destroy(vma);
-	}
-}
-
-void i915_vm_release(struct kref *kref)
-{
-	struct i915_address_space *vm =
-		container_of(kref, struct i915_address_space, ref);
-
-	GEM_BUG_ON(i915_is_ggtt(vm));
-	trace_i915_ppgtt_release(vm);
-
-	ppgtt_destroy_vma(vm);
-
-	GEM_BUG_ON(!list_empty(&vm->bound_list));
-	GEM_BUG_ON(!list_empty(&vm->unbound_list));
-
-	vm->cleanup(vm);
-	i915_address_space_fini(vm);
-
-	kfree(vm);
-}
-
 /* Certain Gen5 chipsets require require idling the GPU before
  * unmapping anything from the GTT when VT-d is enabled.
  */

drivers/gpu/drm/i915/i915_gem_gtt.h

@@ -35,8 +35,12 @@
 #define __I915_GEM_GTT_H__
 
 #include <linux/io-mapping.h>
+#include <linux/kref.h>
 #include <linux/mm.h>
 #include <linux/pagevec.h>
+#include <linux/workqueue.h>
+
+#include <drm/drm_mm.h>
 
 #include "gt/intel_reset.h"
 #include "i915_gem_fence_reg.h"
@@ -280,6 +284,7 @@ struct pagestash {
 
 struct i915_address_space {
 	struct kref ref;
+	struct rcu_work rcu;
 
 	struct drm_mm mm;
 	struct drm_i915_private *i915;
@@ -425,8 +430,6 @@ struct gen6_ppgtt {
 
 	unsigned int pin_count;
 	bool scan_for_unused_pt;
-
-	struct gen6_ppgtt_cleanup_work *work;
 };
 
 #define __to_gen6_ppgtt(base) container_of(base, struct gen6_ppgtt, base)

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

@@ -208,9 +208,7 @@ static int igt_ppgtt_alloc(void *arg)
 	}
 
 err_ppgtt_cleanup:
-	mutex_lock(&dev_priv->drm.struct_mutex);
 	i915_vm_put(&ppgtt->vm);
-	mutex_unlock(&dev_priv->drm.struct_mutex);
 	return err;
 }