[Intel-gfx] [PATCH v3 34/37] drm/i915: support basic object migration
Matthew Auld
matthew.auld at intel.com
Fri Aug 9 22:26:40 UTC 2019
We are going want to able to move objects between different regions
like system memory and local memory. In the future everything should
be just another region.
Signed-off-by: Matthew Auld <matthew.auld at intel.com>
Signed-off-by: Abdiel Janulgue <abdiel.janulgue at linux.intel.com>
Signed-off-by: CQ Tang <cq.tang at intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
Cc: Abdiel Janulgue <abdiel.janulgue at linux.intel.com>
---
drivers/gpu/drm/i915/gem/i915_gem_object.c | 140 ++++++++++++++++++
drivers/gpu/drm/i915/gem/i915_gem_object.h | 8 +
drivers/gpu/drm/i915/gem/i915_gem_pages.c | 2 +-
.../drm/i915/selftests/intel_memory_region.c | 129 ++++++++++++++++
4 files changed, 278 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.c b/drivers/gpu/drm/i915/gem/i915_gem_object.c
index 24f737b00e84..5982aeaaa2e3 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_object.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_object.c
@@ -28,6 +28,8 @@
#include "i915_gem_clflush.h"
#include "i915_gem_context.h"
#include "i915_gem_object.h"
+#include "i915_gem_object_blt.h"
+#include "i915_gem_region.h"
#include "i915_globals.h"
#include "i915_trace.h"
@@ -170,6 +172,144 @@ static void __i915_gem_free_object_rcu(struct rcu_head *head)
atomic_dec(&i915->mm.free_count);
}
+
+int i915_gem_object_prepare_move(struct drm_i915_gem_object *obj)
+{
+ int err;
+
+ lockdep_assert_held(&obj->base.dev->struct_mutex);
+
+ if (obj->mm.madv != I915_MADV_WILLNEED)
+ return -EINVAL;
+
+ if (i915_gem_object_needs_bit17_swizzle(obj))
+ return -EINVAL;
+
+ if (atomic_read(&obj->mm.pages_pin_count) >
+ atomic_read(&obj->bind_count))
+ return -EBUSY;
+
+ if (obj->pin_global)
+ return -EBUSY;
+
+ i915_gem_object_release_mmap(obj);
+
+ GEM_BUG_ON(obj->mm.mapping);
+ GEM_BUG_ON(obj->base.filp && mapping_mapped(obj->base.filp->f_mapping));
+
+ err = i915_gem_object_wait(obj,
+ I915_WAIT_INTERRUPTIBLE |
+ I915_WAIT_LOCKED |
+ I915_WAIT_ALL,
+ MAX_SCHEDULE_TIMEOUT);
+ if (err)
+ return err;
+
+ return i915_gem_object_unbind(obj,
+ I915_GEM_OBJECT_UNBIND_ACTIVE);
+}
+
+int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
+ struct intel_context *ce,
+ enum intel_region_id id)
+{
+ struct drm_i915_private *i915 = to_i915(obj->base.dev);
+ struct drm_i915_gem_object *donor;
+ struct intel_memory_region *mem;
+ struct sg_table *pages = NULL;
+ unsigned int page_sizes;
+ int err = 0;
+
+ lockdep_assert_held(&i915->drm.struct_mutex);
+
+ GEM_BUG_ON(id >= INTEL_MEMORY_UKNOWN);
+ GEM_BUG_ON(obj->mm.region->id == id);
+ GEM_BUG_ON(obj->mm.madv != I915_MADV_WILLNEED);
+
+ mem = i915->regions[id];
+
+ donor = i915_gem_object_create_region(mem, obj->base.size, 0);
+ if (IS_ERR(donor))
+ return PTR_ERR(donor);
+
+ /* Copy backing-pages if we have to */
+ if (i915_gem_object_has_pages(obj)) {
+ err = i915_gem_object_pin_pages(obj);
+ if (err)
+ goto err_put_donor;
+
+ err = i915_gem_object_copy_blt(obj, donor, ce);
+ if (err)
+ goto err_put_donor;
+
+ i915_gem_object_lock(donor);
+ err = i915_gem_object_set_to_cpu_domain(donor, false);
+ i915_gem_object_unlock(donor);
+ if (err)
+ goto err_put_donor;
+
+ i915_retire_requests(i915);
+
+ i915_gem_object_unbind(donor, 0);
+ err = i915_gem_object_unbind(obj, 0);
+ if (err)
+ goto err_put_donor;
+
+ mutex_lock(&obj->mm.lock);
+
+ pages = __i915_gem_object_unset_pages(obj);
+ obj->ops->put_pages(obj, pages);
+
+ mutex_unlock(&obj->mm.lock);
+
+ page_sizes = donor->mm.page_sizes.phys;
+ pages = __i915_gem_object_unset_pages(donor);
+ }
+
+ if (obj->ops->release)
+ obj->ops->release(obj);
+
+ mutex_lock(&obj->mm.lock);
+
+ /* We need still need a little special casing for shmem */
+ if (obj->base.filp)
+ fput(fetch_and_zero(&obj->base.filp));
+ else if (donor->base.filp) {
+ atomic_long_inc(&donor->base.filp->f_count);
+ obj->base.filp = donor->base.filp;
+ }
+
+ obj->base.size = donor->base.size;
+ obj->mm.region = mem;
+ obj->flags = donor->flags;
+ obj->ops = donor->ops;
+ obj->cache_level = donor->cache_level;
+ obj->cache_coherent = donor->cache_coherent;
+ obj->cache_dirty = donor->cache_dirty;
+
+ list_replace_init(&donor->mm.blocks, &obj->mm.blocks);
+
+ mutex_lock(&mem->obj_lock);
+ list_add(&obj->mm.region_link, &mem->objects);
+ mutex_unlock(&mem->obj_lock);
+
+ /* set pages after migrated */
+ if (pages)
+ __i915_gem_object_set_pages(obj, pages, page_sizes);
+
+ mutex_unlock(&obj->mm.lock);
+
+ GEM_BUG_ON(i915_gem_object_has_pages(donor));
+ GEM_BUG_ON(i915_gem_object_has_pinned_pages(donor));
+
+err_put_donor:
+ i915_gem_object_put(donor);
+ if (i915_gem_object_has_pinned_pages(obj))
+ i915_gem_object_unpin_pages(obj);
+
+ return err;
+}
+
static void __i915_gem_free_objects(struct drm_i915_private *i915,
struct llist_node *freed)
{
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.h b/drivers/gpu/drm/i915/gem/i915_gem_object.h
index fd58b9aea180..660880a71554 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_object.h
+++ b/drivers/gpu/drm/i915/gem/i915_gem_object.h
@@ -40,8 +40,16 @@ int i915_gem_object_attach_phys(struct drm_i915_gem_object *obj, int align);
void i915_gem_close_object(struct drm_gem_object *gem, struct drm_file *file);
void i915_gem_free_object(struct drm_gem_object *obj);
+enum intel_region_id;
+int i915_gem_object_prepare_move(struct drm_i915_gem_object *obj);
+int i915_gem_object_migrate(struct drm_i915_gem_object *obj,
+ struct intel_context *ce,
+ enum intel_region_id id);
+
void i915_gem_flush_free_objects(struct drm_i915_private *i915);
+void __i915_gem_object_reset_page_iter(struct drm_i915_gem_object *obj);
+
struct sg_table *
__i915_gem_object_unset_pages(struct drm_i915_gem_object *obj);
void i915_gem_object_truncate(struct drm_i915_gem_object *obj);
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_pages.c b/drivers/gpu/drm/i915/gem/i915_gem_pages.c
index 0b73860deaf8..9e0a7af6fa1d 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_pages.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_pages.c
@@ -143,7 +143,7 @@ void i915_gem_object_writeback(struct drm_i915_gem_object *obj)
obj->ops->writeback(obj);
}
-static void __i915_gem_object_reset_page_iter(struct drm_i915_gem_object *obj)
+void __i915_gem_object_reset_page_iter(struct drm_i915_gem_object *obj)
{
struct radix_tree_iter iter;
void __rcu **slot;
diff --git a/drivers/gpu/drm/i915/selftests/intel_memory_region.c b/drivers/gpu/drm/i915/selftests/intel_memory_region.c
index 4123e81a2bda..2b62e3361a5f 100644
--- a/drivers/gpu/drm/i915/selftests/intel_memory_region.c
+++ b/drivers/gpu/drm/i915/selftests/intel_memory_region.c
@@ -500,6 +500,59 @@ static int igt_lmem_create(void *arg)
return err;
}
+static int igt_smem_create_migrate(void *arg)
+{
+ struct drm_i915_private *i915 = arg;
+ struct intel_context *ce = i915->engine[BCS0]->kernel_context;
+ struct drm_i915_gem_object *obj;
+ int err;
+
+ /* Switch object backing-store on create */
+ obj = i915_gem_object_create_lmem(i915, PAGE_SIZE, 0);
+ if (IS_ERR(obj))
+ return PTR_ERR(obj);
+
+ err = i915_gem_object_migrate(obj, ce, INTEL_MEMORY_SMEM);
+ if (err)
+ goto out_put;
+
+ err = i915_gem_object_pin_pages(obj);
+ if (err)
+ goto out_put;
+
+ i915_gem_object_unpin_pages(obj);
+out_put:
+ i915_gem_object_put(obj);
+
+ return err;
+}
+
+static int igt_lmem_create_migrate(void *arg)
+{
+ struct drm_i915_private *i915 = arg;
+ struct intel_context *ce = i915->engine[BCS0]->kernel_context;
+ struct drm_i915_gem_object *obj;
+ int err;
+
+ /* Switch object backing-store on create */
+ obj = i915_gem_object_create_shmem(i915, PAGE_SIZE);
+ if (IS_ERR(obj))
+ return PTR_ERR(obj);
+
+ err = i915_gem_object_migrate(obj, ce, INTEL_MEMORY_LMEM);
+ if (err)
+ goto out_put;
+
+ err = i915_gem_object_pin_pages(obj);
+ if (err)
+ goto out_put;
+
+ i915_gem_object_unpin_pages(obj);
+out_put:
+ i915_gem_object_put(obj);
+
+ return err;
+}
static int igt_lmem_write_gpu(void *arg)
{
struct drm_i915_private *i915 = arg;
@@ -626,6 +679,79 @@ static int igt_lmem_write_cpu(void *arg)
return err;
}
+static int igt_lmem_pages_migrate(void *arg)
+{
+ struct drm_i915_private *i915 = arg;
+ struct intel_context *ce = i915->engine[BCS0]->kernel_context;
+ struct drm_i915_gem_object *obj;
+ IGT_TIMEOUT(end_time);
+ I915_RND_STATE(prng);
+ u32 sz;
+ int err;
+
+ sz = round_up(prandom_u32_state(&prng) % SZ_32M, PAGE_SIZE);
+
+ obj = i915_gem_object_create_lmem(i915, sz, 0);
+ if (IS_ERR(obj))
+ return PTR_ERR(obj);
+
+ err = i915_gem_object_fill_blt(obj, ce, 0);
+ if (err)
+ goto out_put;
+
+ do {
+ err = i915_gem_object_prepare_move(obj);
+ if (err)
+ goto out_put;
+
+ if (i915_gem_object_is_lmem(obj)) {
+ err = i915_gem_object_migrate(obj, ce, INTEL_MEMORY_SMEM);
+ if (err)
+ goto out_put;
+
+ if (i915_gem_object_is_lmem(obj)) {
+ pr_err("object still backed by lmem\n");
+ err = -EINVAL;
+ }
+
+ if (!list_empty(&obj->mm.blocks)) {
+ pr_err("object leaking memory region\n");
+ err = -EINVAL;
+ }
+
+ if (!i915_gem_object_has_struct_page(obj)) {
+ pr_err("object not backed by struct page\n");
+ err = -EINVAL;
+ }
+
+ } else {
+ err = i915_gem_object_migrate(obj, ce, INTEL_MEMORY_LMEM);
+ if (err)
+ goto out_put;
+
+ if (i915_gem_object_has_struct_page(obj)) {
+ pr_err("object still backed by struct page\n");
+ err = -EINVAL;
+ }
+
+ if (!i915_gem_object_is_lmem(obj)) {
+ pr_err("object not backed by lmem\n");
+ err = -EINVAL;
+ }
+ }
+
+ if (!err)
+ err = i915_gem_object_fill_blt(obj, ce, 0xdeadbeaf);
+ if (err)
+ break;
+ } while (!__igt_timeout(end_time, NULL));
+
+out_put:
+ i915_gem_object_put(obj);
+
+ return err;
+}
+
int intel_memory_region_mock_selftests(void)
{
static const struct i915_subtest tests[] = {
@@ -669,6 +795,9 @@ int intel_memory_region_live_selftests(struct drm_i915_private *i915)
SUBTEST(igt_lmem_create),
SUBTEST(igt_lmem_write_cpu),
SUBTEST(igt_lmem_write_gpu),
+ SUBTEST(igt_smem_create_migrate),
+ SUBTEST(igt_lmem_create_migrate),
+ SUBTEST(igt_lmem_pages_migrate),
};
int err;
--
2.20.1
More information about the Intel-gfx
mailing list