[Intel-gfx] [PATCH 36/39] drm/i915: Use reservation_object to coordinate userptr get_pages()
Chris Wilson
chris at chris-wilson.co.uk
Fri Jun 14 07:10:20 UTC 2019
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c | 27 +--
.../gpu/drm/i915/gem/i915_gem_execbuffer.c | 3 -
drivers/gpu/drm/i915/gem/i915_gem_internal.c | 30 +--
.../gpu/drm/i915/gem/i915_gem_object_types.h | 11 +-
drivers/gpu/drm/i915/gem/i915_gem_pages.c | 137 ++++++++++-
drivers/gpu/drm/i915/gem/i915_gem_phys.c | 33 ++-
drivers/gpu/drm/i915/gem/i915_gem_shmem.c | 30 +--
drivers/gpu/drm/i915/gem/i915_gem_stolen.c | 25 +-
drivers/gpu/drm/i915/gem/i915_gem_userptr.c | 221 ++++--------------
.../drm/i915/gem/selftests/huge_gem_object.c | 17 +-
.../gpu/drm/i915/gem/selftests/huge_pages.c | 45 ++--
drivers/gpu/drm/i915/gvt/dmabuf.c | 17 +-
drivers/gpu/drm/i915/i915_drv.h | 9 +-
drivers/gpu/drm/i915/i915_gem.c | 5 +-
drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 14 +-
15 files changed, 300 insertions(+), 324 deletions(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c b/drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c
index 84992d590da5..a44d6d2ef7ed 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_dmabuf.c
@@ -225,33 +225,24 @@ struct dma_buf *i915_gem_prime_export(struct drm_device *dev,
return drm_gem_dmabuf_export(dev, &exp_info);
}
-static int i915_gem_object_get_pages_dmabuf(struct drm_i915_gem_object *obj)
+static struct sg_table *
+dmabuf_get_pages(struct i915_gem_object_get_pages_context *ctx,
+ unsigned int *sizes)
{
- struct sg_table *pages;
- unsigned int sg_page_sizes;
-
- pages = dma_buf_map_attachment(obj->base.import_attach,
- DMA_BIDIRECTIONAL);
- if (IS_ERR(pages))
- return PTR_ERR(pages);
-
- sg_page_sizes = i915_sg_page_sizes(pages->sgl);
-
- __i915_gem_object_set_pages(obj, pages, sg_page_sizes);
-
- return 0;
+ return dma_buf_map_attachment(ctx->object->base.import_attach,
+ DMA_BIDIRECTIONAL);
}
-static void i915_gem_object_put_pages_dmabuf(struct drm_i915_gem_object *obj,
- struct sg_table *pages)
+static void dmabuf_put_pages(struct drm_i915_gem_object *obj,
+ struct sg_table *pages)
{
dma_buf_unmap_attachment(obj->base.import_attach, pages,
DMA_BIDIRECTIONAL);
}
static const struct drm_i915_gem_object_ops i915_gem_object_dmabuf_ops = {
- .get_pages = i915_gem_object_get_pages_dmabuf,
- .put_pages = i915_gem_object_put_pages_dmabuf,
+ .get_pages = dmabuf_get_pages,
+ .put_pages = dmabuf_put_pages,
};
struct drm_gem_object *i915_gem_prime_import(struct drm_device *dev,
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
index 44bcb681c168..68faf1a71c97 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
@@ -1784,9 +1784,6 @@ static noinline int eb_relocate_slow(struct i915_execbuffer *eb)
goto out;
}
- /* A frequent cause for EAGAIN are currently unavailable client pages */
- flush_workqueue(eb->i915->mm.userptr_wq);
-
err = i915_mutex_lock_interruptible(dev);
if (err) {
mutex_lock(&dev->struct_mutex);
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_internal.c b/drivers/gpu/drm/i915/gem/i915_gem_internal.c
index 0c41e04ab8fa..aa0bd5de313b 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_internal.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_internal.c
@@ -32,11 +32,14 @@ static void internal_free_pages(struct sg_table *st)
kfree(st);
}
-static int i915_gem_object_get_pages_internal(struct drm_i915_gem_object *obj)
+static struct sg_table *
+internal_get_pages(struct i915_gem_object_get_pages_context *ctx,
+ unsigned int *sizes)
{
+ struct drm_i915_gem_object *obj = ctx->object;
struct drm_i915_private *i915 = to_i915(obj->base.dev);
- struct sg_table *st;
struct scatterlist *sg;
+ struct sg_table *st;
unsigned int sg_page_sizes;
unsigned int npages;
int max_order;
@@ -66,12 +69,12 @@ static int i915_gem_object_get_pages_internal(struct drm_i915_gem_object *obj)
create_st:
st = kmalloc(sizeof(*st), GFP_KERNEL);
if (!st)
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
npages = obj->base.size / PAGE_SIZE;
if (sg_alloc_table(st, npages, GFP_KERNEL)) {
kfree(st);
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
}
sg = st->sgl;
@@ -117,27 +120,26 @@ static int i915_gem_object_get_pages_internal(struct drm_i915_gem_object *obj)
goto err;
}
- /* Mark the pages as dontneed whilst they are still pinned. As soon
+ /*
+ * Mark the pages as dontneed whilst they are still pinned. As soon
* as they are unpinned they are allowed to be reaped by the shrinker,
* and the caller is expected to repopulate - the contents of this
* object are only valid whilst active and pinned.
*/
obj->mm.madv = I915_MADV_DONTNEED;
- __i915_gem_object_set_pages(obj, st, sg_page_sizes);
-
- return 0;
+ *sizes = sg_page_sizes;
+ return st;
err:
sg_set_page(sg, NULL, 0, 0);
sg_mark_end(sg);
internal_free_pages(st);
-
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
}
-static void i915_gem_object_put_pages_internal(struct drm_i915_gem_object *obj,
- struct sg_table *pages)
+static void internal_put_pages(struct drm_i915_gem_object *obj,
+ struct sg_table *pages)
{
i915_gem_gtt_finish_pages(obj, pages);
internal_free_pages(pages);
@@ -149,8 +151,8 @@ static void i915_gem_object_put_pages_internal(struct drm_i915_gem_object *obj,
static const struct drm_i915_gem_object_ops i915_gem_object_internal_ops = {
.flags = I915_GEM_OBJECT_HAS_STRUCT_PAGE |
I915_GEM_OBJECT_IS_SHRINKABLE,
- .get_pages = i915_gem_object_get_pages_internal,
- .put_pages = i915_gem_object_put_pages_internal,
+ .get_pages = internal_get_pages,
+ .put_pages = internal_put_pages,
};
/**
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 f792953b8a71..0ea404cfbc1c 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_object_types.h
+++ b/drivers/gpu/drm/i915/gem/i915_gem_object_types.h
@@ -19,6 +19,7 @@
struct drm_i915_gem_object;
struct intel_fronbuffer;
+struct task_struct;
/*
* struct i915_lut_handle tracks the fast lookups from handle to vma used
@@ -32,6 +33,11 @@ struct i915_lut_handle {
u32 handle;
};
+struct i915_gem_object_get_pages_context {
+ struct drm_i915_gem_object *object;
+ struct task_struct *task;
+};
+
struct drm_i915_gem_object_ops {
unsigned int flags;
#define I915_GEM_OBJECT_HAS_STRUCT_PAGE BIT(0)
@@ -52,9 +58,11 @@ struct drm_i915_gem_object_ops {
* being released or under memory pressure (where we attempt to
* reap pages for the shrinker).
*/
- int (*get_pages)(struct drm_i915_gem_object *obj);
+ struct sg_table *(*get_pages)(struct i915_gem_object_get_pages_context *ctx,
+ unsigned int *sizes);
void (*put_pages)(struct drm_i915_gem_object *obj,
struct sg_table *pages);
+
void (*truncate)(struct drm_i915_gem_object *obj);
void (*writeback)(struct drm_i915_gem_object *obj);
@@ -252,7 +260,6 @@ struct drm_i915_gem_object {
struct i915_mm_struct *mm;
struct i915_mmu_object *mmu_object;
- struct work_struct *work;
} userptr;
unsigned long scratch;
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_pages.c b/drivers/gpu/drm/i915/gem/i915_gem_pages.c
index 6bec301cee79..f65a983248c6 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_pages.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_pages.c
@@ -8,16 +8,49 @@
#include "i915_gem_object.h"
#include "i915_scatterlist.h"
-void __i915_gem_object_set_pages(struct drm_i915_gem_object *obj,
- struct sg_table *pages,
- unsigned int sg_page_sizes)
+static DEFINE_SPINLOCK(fence_lock);
+
+struct get_pages_work {
+ struct dma_fence dma; /* Must be first for dma_fence_free() */
+ struct i915_sw_fence wait;
+ struct work_struct work;
+ struct i915_gem_object_get_pages_context ctx;
+};
+
+static const char *get_pages_work_driver_name(struct dma_fence *fence)
+{
+ return DRIVER_NAME;
+}
+
+static const char *get_pages_work_timeline_name(struct dma_fence *fence)
+{
+ return "allocation";
+}
+
+static void get_pages_work_release(struct dma_fence *fence)
+{
+ struct get_pages_work *w = container_of(fence, typeof(*w), dma);
+
+ i915_sw_fence_fini(&w->wait);
+
+ BUILD_BUG_ON(offsetof(typeof(*w), dma));
+ dma_fence_free(&w->dma);
+}
+
+static const struct dma_fence_ops get_pages_work_ops = {
+ .get_driver_name = get_pages_work_driver_name,
+ .get_timeline_name = get_pages_work_timeline_name,
+ .release = get_pages_work_release,
+};
+
+static void __set_pages(struct drm_i915_gem_object *obj,
+ struct sg_table *pages,
+ unsigned int sg_page_sizes)
{
struct drm_i915_private *i915 = to_i915(obj->base.dev);
unsigned long supported = INTEL_INFO(i915)->page_sizes;
int i;
- lockdep_assert_held(&obj->mm.lock);
-
/* Make the pages coherent with the GPU (flushing any swapin). */
if (obj->cache_dirty) {
obj->write_domain = 0;
@@ -29,8 +62,6 @@ void __i915_gem_object_set_pages(struct drm_i915_gem_object *obj,
obj->mm.get_page.sg_pos = pages->sgl;
obj->mm.get_page.sg_idx = 0;
- obj->mm.pages = pages;
-
if (i915_gem_object_is_tiled(obj) &&
i915->quirks & QUIRK_PIN_SWIZZLED_PAGES) {
GEM_BUG_ON(obj->mm.quirked);
@@ -38,7 +69,8 @@ void __i915_gem_object_set_pages(struct drm_i915_gem_object *obj,
obj->mm.quirked = true;
}
- GEM_BUG_ON(!sg_page_sizes);
+ if (!sg_page_sizes)
+ sg_page_sizes = i915_sg_page_sizes(pages->sgl);
obj->mm.page_sizes.phys = sg_page_sizes;
/*
@@ -73,18 +105,105 @@ void __i915_gem_object_set_pages(struct drm_i915_gem_object *obj,
spin_unlock_irqrestore(&i915->mm.obj_lock, flags);
}
+}
+static void
+get_pages_worker(struct work_struct *_work)
+{
+ struct get_pages_work *work = container_of(_work, typeof(*work), work);
+ struct drm_i915_gem_object *obj = work->ctx.object;
+ struct sg_table *pages;
+ unsigned int sizes = 0;
+
+ if (!work->dma.error) {
+ pages = obj->ops->get_pages(&work->ctx, &sizes);
+ if (!IS_ERR(pages))
+ __set_pages(obj, pages, sizes);
+ else
+ dma_fence_set_error(&work->dma, PTR_ERR(pages));
+ } else {
+ pages = ERR_PTR(work->dma.error);
+ }
+
+ obj->mm.pages = pages;
complete_all(&obj->mm.completion);
+ atomic_dec(&obj->mm.pages_pin_count);
+
+ i915_gem_object_put(obj);
+ put_task_struct(work->ctx.task);
+
+ dma_fence_signal(&work->dma);
+ dma_fence_put(&work->dma);
+}
+
+static int __i915_sw_fence_call
+get_pages_notify(struct i915_sw_fence *fence, enum i915_sw_fence_notify state)
+{
+ struct get_pages_work *w = container_of(fence, typeof(*w), wait);
+
+ switch (state) {
+ case FENCE_COMPLETE:
+ if (fence->error)
+ dma_fence_set_error(&w->dma, fence->error);
+ queue_work(system_unbound_wq, &w->work);
+ break;
+
+ case FENCE_FREE:
+ dma_fence_put(&w->dma);
+ break;
+ }
+
+ return NOTIFY_DONE;
}
int ____i915_gem_object_get_pages(struct drm_i915_gem_object *obj)
{
+ struct get_pages_work *work;
+ int err;
+
if (unlikely(obj->mm.madv != I915_MADV_WILLNEED)) {
DRM_DEBUG("Attempting to obtain a purgeable object\n");
return -EFAULT;
}
- return obj->ops->get_pages(obj);
+ /* XXX inline? */
+
+ work = kmalloc(sizeof(*work), GFP_KERNEL);
+ if (!work)
+ return -ENOMEM;
+
+ dma_fence_init(&work->dma,
+ &get_pages_work_ops,
+ &fence_lock,
+ to_i915(obj->base.dev)->mm.unordered_timeline,
+ 0);
+ i915_sw_fence_init(&work->wait, get_pages_notify);
+
+ work->ctx.object = i915_gem_object_get(obj);
+
+ work->ctx.task = current;
+ get_task_struct(work->ctx.task);
+
+ INIT_WORK(&work->work, get_pages_worker);
+
+ i915_gem_object_lock(obj);
+ GEM_BUG_ON(!reservation_object_test_signaled_rcu(obj->resv, true));
+ err = i915_sw_fence_await_reservation(&work->wait,
+ obj->resv, NULL,
+ true, I915_FENCE_TIMEOUT,
+ I915_FENCE_GFP);
+ if (err == 0) {
+ reservation_object_add_excl_fence(obj->resv, &work->dma);
+ atomic_inc(&obj->mm.pages_pin_count);
+ } else {
+ dma_fence_set_error(&work->dma, err);
+ }
+ i915_gem_object_unlock(obj);
+
+ dma_fence_get(&work->dma);
+ i915_sw_fence_commit(&work->wait);
+
+ return err;
}
/* Ensure that the associated pages are gathered from the backing storage
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_phys.c b/drivers/gpu/drm/i915/gem/i915_gem_phys.c
index 2deac933cf59..6b4a5fb52055 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_phys.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_phys.c
@@ -17,18 +17,21 @@
#include "i915_gem_object.h"
#include "i915_scatterlist.h"
-static int i915_gem_object_get_pages_phys(struct drm_i915_gem_object *obj)
+static struct sg_table *
+phys_get_pages(struct i915_gem_object_get_pages_context *ctx,
+ unsigned int *sizes)
{
+ struct drm_i915_gem_object *obj = ctx->object;
struct address_space *mapping = obj->base.filp->f_mapping;
struct drm_dma_handle *phys;
- struct sg_table *st;
struct scatterlist *sg;
+ struct sg_table *st;
char *vaddr;
int i;
int err;
if (WARN_ON(i915_gem_object_needs_bit17_swizzle(obj)))
- return -EINVAL;
+ return ERR_PTR(-EINVAL);
/* Always aligning to the object size, allows a single allocation
* to handle all possible callers, and given typical object sizes,
@@ -38,7 +41,7 @@ static int i915_gem_object_get_pages_phys(struct drm_i915_gem_object *obj)
roundup_pow_of_two(obj->base.size),
roundup_pow_of_two(obj->base.size));
if (!phys)
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
vaddr = phys->vaddr;
for (i = 0; i < obj->base.size / PAGE_SIZE; i++) {
@@ -83,19 +86,16 @@ static int i915_gem_object_get_pages_phys(struct drm_i915_gem_object *obj)
obj->phys_handle = phys;
- __i915_gem_object_set_pages(obj, st, sg->length);
-
- return 0;
+ *sizes = sg->length;
+ return st;
err_phys:
drm_pci_free(obj->base.dev, phys);
-
- return err;
+ return ERR_PTR(err);
}
static void
-i915_gem_object_put_pages_phys(struct drm_i915_gem_object *obj,
- struct sg_table *pages)
+phys_put_pages(struct drm_i915_gem_object *obj, struct sg_table *pages)
{
__i915_gem_object_release_shmem(obj, pages, false);
@@ -139,8 +139,8 @@ i915_gem_object_release_phys(struct drm_i915_gem_object *obj)
}
static const struct drm_i915_gem_object_ops i915_gem_phys_ops = {
- .get_pages = i915_gem_object_get_pages_phys,
- .put_pages = i915_gem_object_put_pages_phys,
+ .get_pages = phys_get_pages,
+ .put_pages = phys_put_pages,
.release = i915_gem_object_release_phys,
};
@@ -193,15 +193,12 @@ int i915_gem_object_attach_phys(struct drm_i915_gem_object *obj, int align)
if (!IS_ERR_OR_NULL(pages))
i915_gem_shmem_ops.put_pages(obj, pages);
mutex_unlock(&obj->mm.lock);
+
+ wait_for_completion(&obj->mm.completion);
return 0;
err_xfer:
obj->ops = &i915_gem_shmem_ops;
- if (!IS_ERR_OR_NULL(pages)) {
- unsigned int sg_page_sizes = i915_sg_page_sizes(pages->sgl);
-
- __i915_gem_object_set_pages(obj, pages, sg_page_sizes);
- }
err_unlock:
mutex_unlock(&obj->mm.lock);
return err;
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_shmem.c b/drivers/gpu/drm/i915/gem/i915_gem_shmem.c
index 19d9ecdb2894..c43304e3bada 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_shmem.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_shmem.c
@@ -22,11 +22,13 @@ static void check_release_pagevec(struct pagevec *pvec)
cond_resched();
}
-static int shmem_get_pages(struct drm_i915_gem_object *obj)
+static struct sg_table *
+shmem_get_pages(struct i915_gem_object_get_pages_context *ctx,
+ unsigned int *sizes)
{
+ struct drm_i915_gem_object *obj = ctx->object;
struct drm_i915_private *i915 = to_i915(obj->base.dev);
const unsigned long page_count = obj->base.size / PAGE_SIZE;
- unsigned long i;
struct address_space *mapping;
struct sg_table *st;
struct scatterlist *sg;
@@ -37,31 +39,24 @@ static int shmem_get_pages(struct drm_i915_gem_object *obj)
unsigned int sg_page_sizes;
struct pagevec pvec;
gfp_t noreclaim;
+ unsigned long i;
int ret;
- /*
- * Assert that the object is not currently in any GPU domain. As it
- * wasn't in the GTT, there shouldn't be any way it could have been in
- * a GPU cache
- */
- GEM_BUG_ON(obj->read_domains & I915_GEM_GPU_DOMAINS);
- GEM_BUG_ON(obj->write_domain & I915_GEM_GPU_DOMAINS);
-
/*
* If there's no chance of allocating enough pages for the whole
* object, bail early.
*/
if (page_count > totalram_pages())
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
st = kmalloc(sizeof(*st), GFP_KERNEL);
if (!st)
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
rebuild_st:
if (sg_alloc_table(st, page_count, GFP_KERNEL)) {
kfree(st);
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
}
/*
@@ -179,9 +174,8 @@ static int shmem_get_pages(struct drm_i915_gem_object *obj)
if (i915_gem_object_needs_bit17_swizzle(obj))
i915_gem_object_do_bit_17_swizzle(obj, st);
- __i915_gem_object_set_pages(obj, st, sg_page_sizes);
-
- return 0;
+ *sizes = sg_page_sizes;
+ return st;
err_sg:
sg_mark_end(sg);
@@ -209,7 +203,7 @@ static int shmem_get_pages(struct drm_i915_gem_object *obj)
if (ret == -ENOSPC)
ret = -ENOMEM;
- return ret;
+ return ERR_PTR(ret);
}
static void
@@ -276,8 +270,6 @@ __i915_gem_object_release_shmem(struct drm_i915_gem_object *obj,
struct sg_table *pages,
bool needs_clflush)
{
- GEM_BUG_ON(obj->mm.madv == __I915_MADV_PURGED);
-
if (obj->mm.madv == I915_MADV_DONTNEED)
obj->mm.dirty = false;
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_stolen.c b/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
index 7066044d63cf..28ea06e667cd 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
@@ -499,22 +499,19 @@ i915_pages_create_for_stolen(struct drm_device *dev,
return st;
}
-static int i915_gem_object_get_pages_stolen(struct drm_i915_gem_object *obj)
+static struct sg_table *
+stolen_get_pages(struct i915_gem_object_get_pages_context *ctx,
+ unsigned int *sizes)
{
- struct sg_table *pages =
- i915_pages_create_for_stolen(obj->base.dev,
- obj->stolen->start,
- obj->stolen->size);
- if (IS_ERR(pages))
- return PTR_ERR(pages);
-
- __i915_gem_object_set_pages(obj, pages, obj->stolen->size);
+ struct drm_i915_gem_object *obj = ctx->object;
- return 0;
+ return i915_pages_create_for_stolen(obj->base.dev,
+ obj->stolen->start,
+ obj->stolen->size);
}
-static void i915_gem_object_put_pages_stolen(struct drm_i915_gem_object *obj,
- struct sg_table *pages)
+static void
+stolen_put_pages(struct drm_i915_gem_object *obj, struct sg_table *pages)
{
/* Should only be called from i915_gem_object_release_stolen() */
sg_free_table(pages);
@@ -536,8 +533,8 @@ i915_gem_object_release_stolen(struct drm_i915_gem_object *obj)
}
static const struct drm_i915_gem_object_ops i915_gem_object_stolen_ops = {
- .get_pages = i915_gem_object_get_pages_stolen,
- .put_pages = i915_gem_object_put_pages_stolen,
+ .get_pages = stolen_get_pages,
+ .put_pages = stolen_put_pages,
.release = i915_gem_object_release_stolen,
};
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_userptr.c b/drivers/gpu/drm/i915/gem/i915_gem_userptr.c
index f093deaeb5c0..6748e15bf89a 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_userptr.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_userptr.c
@@ -378,7 +378,7 @@ __i915_mm_struct_free(struct kref *kref)
mutex_unlock(&mm->i915->mm_lock);
INIT_WORK(&mm->work, __i915_mm_struct_free__worker);
- queue_work(mm->i915->mm.userptr_wq, &mm->work);
+ schedule_work(&mm->work);
}
static void
@@ -393,19 +393,12 @@ i915_gem_userptr_release__mm_struct(struct drm_i915_gem_object *obj)
obj->userptr.mm = NULL;
}
-struct get_pages_work {
- struct work_struct work;
- struct drm_i915_gem_object *obj;
- struct task_struct *task;
-};
-
static struct sg_table *
__i915_gem_userptr_alloc_pages(struct drm_i915_gem_object *obj,
struct page **pvec, int num_pages)
{
unsigned int max_segment = i915_sg_segment_size();
struct sg_table *st;
- unsigned int sg_page_sizes;
int ret;
st = kmalloc(sizeof(*st), GFP_KERNEL);
@@ -435,131 +428,23 @@ __i915_gem_userptr_alloc_pages(struct drm_i915_gem_object *obj,
return ERR_PTR(ret);
}
- sg_page_sizes = i915_sg_page_sizes(st->sgl);
-
- __i915_gem_object_set_pages(obj, st, sg_page_sizes);
-
return st;
}
-static void
-__i915_gem_userptr_get_pages_worker(struct work_struct *_work)
-{
- struct get_pages_work *work = container_of(_work, typeof(*work), work);
- struct drm_i915_gem_object *obj = work->obj;
- const int npages = obj->base.size >> PAGE_SHIFT;
- struct page **pvec;
- int pinned, ret;
-
- ret = -ENOMEM;
- pinned = 0;
-
- pvec = kvmalloc_array(npages, sizeof(struct page *), GFP_KERNEL);
- if (pvec != NULL) {
- struct mm_struct *mm = obj->userptr.mm->mm;
- unsigned int flags = 0;
-
- if (!i915_gem_object_is_readonly(obj))
- flags |= FOLL_WRITE;
-
- ret = -EFAULT;
- if (mmget_not_zero(mm)) {
- down_read(&mm->mmap_sem);
- while (pinned < npages) {
- ret = get_user_pages_remote
- (work->task, mm,
- obj->userptr.ptr + pinned * PAGE_SIZE,
- npages - pinned,
- flags,
- pvec + pinned, NULL, NULL);
- if (ret < 0)
- break;
-
- pinned += ret;
- }
- up_read(&mm->mmap_sem);
- mmput(mm);
- }
- }
-
- mutex_lock(&obj->mm.lock);
- if (obj->userptr.work == &work->work) {
- struct sg_table *pages = ERR_PTR(ret);
-
- if (pinned == npages) {
- pages = __i915_gem_userptr_alloc_pages(obj, pvec,
- npages);
- if (!IS_ERR(pages)) {
- pinned = 0;
- pages = NULL;
- }
- }
-
- obj->userptr.work = ERR_CAST(pages);
- if (IS_ERR(pages))
- __i915_gem_userptr_set_active(obj, false);
- }
- mutex_unlock(&obj->mm.lock);
-
- release_pages(pvec, pinned);
- kvfree(pvec);
-
- i915_gem_object_put(obj);
- put_task_struct(work->task);
- kfree(work);
-}
-
static struct sg_table *
-__i915_gem_userptr_get_pages_schedule(struct drm_i915_gem_object *obj)
-{
- struct get_pages_work *work;
-
- /* Spawn a worker so that we can acquire the
- * user pages without holding our mutex. Access
- * to the user pages requires mmap_sem, and we have
- * a strict lock ordering of mmap_sem, struct_mutex -
- * we already hold struct_mutex here and so cannot
- * call gup without encountering a lock inversion.
- *
- * Userspace will keep on repeating the operation
- * (thanks to EAGAIN) until either we hit the fast
- * path or the worker completes. If the worker is
- * cancelled or superseded, the task is still run
- * but the results ignored. (This leads to
- * complications that we may have a stray object
- * refcount that we need to be wary of when
- * checking for existing objects during creation.)
- * If the worker encounters an error, it reports
- * that error back to this function through
- * obj->userptr.work = ERR_PTR.
- */
- work = kmalloc(sizeof(*work), GFP_KERNEL);
- if (work == NULL)
- return ERR_PTR(-ENOMEM);
-
- obj->userptr.work = &work->work;
-
- work->obj = i915_gem_object_get(obj);
-
- work->task = current;
- get_task_struct(work->task);
-
- INIT_WORK(&work->work, __i915_gem_userptr_get_pages_worker);
- queue_work(to_i915(obj->base.dev)->mm.userptr_wq, &work->work);
-
- return ERR_PTR(-EAGAIN);
-}
-
-static int i915_gem_userptr_get_pages(struct drm_i915_gem_object *obj)
+userptr_get_pages(struct i915_gem_object_get_pages_context *ctx,
+ unsigned int *sizes)
{
+ struct drm_i915_gem_object *obj = ctx->object;
const int num_pages = obj->base.size >> PAGE_SHIFT;
struct mm_struct *mm = obj->userptr.mm->mm;
- struct page **pvec;
struct sg_table *pages;
- bool active;
+ struct page **pvec;
int pinned;
+ int ret;
- /* If userspace should engineer that these pages are replaced in
+ /*
+ * If userspace should engineer that these pages are replaced in
* the vma between us binding this page into the GTT and completion
* of rendering... Their loss. If they change the mapping of their
* pages they need to create a new bo to point to the new vma.
@@ -576,59 +461,58 @@ static int i915_gem_userptr_get_pages(struct drm_i915_gem_object *obj)
* egregious cases from causing harm.
*/
- if (obj->userptr.work) {
- /* active flag should still be held for the pending work */
- if (IS_ERR(obj->userptr.work))
- return PTR_ERR(obj->userptr.work);
- else
- return -EAGAIN;
- }
+ pvec = kvmalloc_array(num_pages, sizeof(struct page *),
+ GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN);
+ if (!pvec)
+ return ERR_PTR(-ENOMEM);
+
+ __i915_gem_userptr_set_active(obj, true);
- pvec = NULL;
pinned = 0;
+ ret = -EFAULT;
+ if (mmget_not_zero(mm)) {
+ unsigned int flags;
- if (mm == current->mm) {
- pvec = kvmalloc_array(num_pages, sizeof(struct page *),
- GFP_KERNEL |
- __GFP_NORETRY |
- __GFP_NOWARN);
- if (pvec) /* defer to worker if malloc fails */
- pinned = __get_user_pages_fast(obj->userptr.ptr,
- num_pages,
- !i915_gem_object_is_readonly(obj),
- pvec);
- }
+ flags = 0;
+ if (!i915_gem_object_is_readonly(obj))
+ flags |= FOLL_WRITE;
- active = false;
- if (pinned < 0) {
- pages = ERR_PTR(pinned);
- pinned = 0;
- } else if (pinned < num_pages) {
- pages = __i915_gem_userptr_get_pages_schedule(obj);
- active = pages == ERR_PTR(-EAGAIN);
- } else {
- pages = __i915_gem_userptr_alloc_pages(obj, pvec, num_pages);
- active = !IS_ERR(pages);
+ down_read(&mm->mmap_sem);
+ while (pinned < num_pages) {
+ ret = get_user_pages_remote
+ (ctx->task, mm,
+ obj->userptr.ptr + pinned * PAGE_SIZE,
+ num_pages - pinned,
+ flags,
+ pvec + pinned, NULL, NULL);
+ if (ret < 0)
+ break;
+
+ pinned += ret;
+ }
+ up_read(&mm->mmap_sem);
+ mmput(mm);
}
- if (active)
- __i915_gem_userptr_set_active(obj, true);
- if (IS_ERR(pages))
+ if (ret)
+ pages = ERR_PTR(ret);
+ else
+ pages = __i915_gem_userptr_alloc_pages(obj, pvec, num_pages);
+ if (IS_ERR(pages)) {
release_pages(pvec, pinned);
+ __i915_gem_userptr_set_active(obj, false);
+ }
kvfree(pvec);
- return PTR_ERR_OR_ZERO(pages);
+ return pages;
}
static void
-i915_gem_userptr_put_pages(struct drm_i915_gem_object *obj,
- struct sg_table *pages)
+userptr_put_pages(struct drm_i915_gem_object *obj, struct sg_table *pages)
{
struct sgt_iter sgt_iter;
struct page *page;
- /* Cancel any inflight work and force them to restart their gup */
- obj->userptr.work = NULL;
__i915_gem_userptr_set_active(obj, false);
if (!pages)
return;
@@ -669,8 +553,8 @@ static const struct drm_i915_gem_object_ops i915_gem_userptr_ops = {
.flags = I915_GEM_OBJECT_HAS_STRUCT_PAGE |
I915_GEM_OBJECT_IS_SHRINKABLE |
I915_GEM_OBJECT_ASYNC_CANCEL,
- .get_pages = i915_gem_userptr_get_pages,
- .put_pages = i915_gem_userptr_put_pages,
+ .get_pages = userptr_get_pages,
+ .put_pages = userptr_put_pages,
.dmabuf_export = i915_gem_userptr_dmabuf_export,
.release = i915_gem_userptr_release,
};
@@ -786,22 +670,13 @@ i915_gem_userptr_ioctl(struct drm_device *dev,
return 0;
}
-int i915_gem_init_userptr(struct drm_i915_private *dev_priv)
+void i915_gem_init_userptr(struct drm_i915_private *dev_priv)
{
mutex_init(&dev_priv->mm_lock);
hash_init(dev_priv->mm_structs);
-
- dev_priv->mm.userptr_wq =
- alloc_workqueue("i915-userptr-acquire",
- WQ_HIGHPRI | WQ_UNBOUND,
- 0);
- if (!dev_priv->mm.userptr_wq)
- return -ENOMEM;
-
- return 0;
}
void i915_gem_cleanup_userptr(struct drm_i915_private *dev_priv)
{
- destroy_workqueue(dev_priv->mm.userptr_wq);
+ mutex_destroy(&dev_priv->mm_lock);
}
diff --git a/drivers/gpu/drm/i915/gem/selftests/huge_gem_object.c b/drivers/gpu/drm/i915/gem/selftests/huge_gem_object.c
index 3c5d17b2b670..02e6edce715e 100644
--- a/drivers/gpu/drm/i915/gem/selftests/huge_gem_object.c
+++ b/drivers/gpu/drm/i915/gem/selftests/huge_gem_object.c
@@ -21,9 +21,12 @@ static void huge_free_pages(struct drm_i915_gem_object *obj,
kfree(pages);
}
-static int huge_get_pages(struct drm_i915_gem_object *obj)
+static struct sg_table *
+huge_get_pages(struct i915_gem_object_get_pages_context *ctx,
+ unsigned int *sizes)
{
#define GFP (GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY)
+ struct drm_i915_gem_object *obj = ctx->object;
const unsigned long nreal = obj->scratch / PAGE_SIZE;
const unsigned long npages = obj->base.size / PAGE_SIZE;
struct scatterlist *sg, *src, *end;
@@ -32,11 +35,11 @@ static int huge_get_pages(struct drm_i915_gem_object *obj)
pages = kmalloc(sizeof(*pages), GFP);
if (!pages)
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
if (sg_alloc_table(pages, npages, GFP)) {
kfree(pages);
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
}
sg = pages->sgl;
@@ -64,14 +67,12 @@ static int huge_get_pages(struct drm_i915_gem_object *obj)
if (i915_gem_gtt_prepare_pages(obj, pages))
goto err;
- __i915_gem_object_set_pages(obj, pages, PAGE_SIZE);
-
- return 0;
+ *sizes = PAGE_SIZE;
+ return pages;
err:
huge_free_pages(obj, pages);
-
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
#undef GFP
}
diff --git a/drivers/gpu/drm/i915/gem/selftests/huge_pages.c b/drivers/gpu/drm/i915/gem/selftests/huge_pages.c
index b03a29366bd1..3b93a7337f27 100644
--- a/drivers/gpu/drm/i915/gem/selftests/huge_pages.c
+++ b/drivers/gpu/drm/i915/gem/selftests/huge_pages.c
@@ -51,9 +51,12 @@ static void huge_pages_free_pages(struct sg_table *st)
kfree(st);
}
-static int get_huge_pages(struct drm_i915_gem_object *obj)
+static struct sg_table *
+get_huge_pages(struct i915_gem_object_get_pages_context *ctx,
+ unsigned int *sizes)
{
#define GFP (GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY)
+ struct drm_i915_gem_object *obj = ctx->object;
unsigned int page_mask = obj->mm.page_mask;
struct sg_table *st;
struct scatterlist *sg;
@@ -62,11 +65,11 @@ static int get_huge_pages(struct drm_i915_gem_object *obj)
st = kmalloc(sizeof(*st), GFP);
if (!st)
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
if (sg_alloc_table(st, obj->base.size >> PAGE_SHIFT, GFP)) {
kfree(st);
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
}
rem = obj->base.size;
@@ -114,16 +117,14 @@ static int get_huge_pages(struct drm_i915_gem_object *obj)
obj->mm.madv = I915_MADV_DONTNEED;
GEM_BUG_ON(sg_page_sizes != obj->mm.page_mask);
- __i915_gem_object_set_pages(obj, st, sg_page_sizes);
-
- return 0;
+ *sizes = sg_page_sizes;
+ return st;
err:
sg_set_page(sg, NULL, 0, 0);
sg_mark_end(sg);
huge_pages_free_pages(st);
-
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
}
static void put_huge_pages(struct drm_i915_gem_object *obj,
@@ -175,8 +176,11 @@ huge_pages_object(struct drm_i915_private *i915,
return obj;
}
-static int fake_get_huge_pages(struct drm_i915_gem_object *obj)
+static struct sg_table *
+fake_get_huge_pages(struct i915_gem_object_get_pages_context *ctx,
+ unsigned int *sizes)
{
+ struct drm_i915_gem_object *obj = ctx->object;
struct drm_i915_private *i915 = to_i915(obj->base.dev);
const u64 max_len = rounddown_pow_of_two(UINT_MAX);
struct sg_table *st;
@@ -186,11 +190,11 @@ static int fake_get_huge_pages(struct drm_i915_gem_object *obj)
st = kmalloc(sizeof(*st), GFP);
if (!st)
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
if (sg_alloc_table(st, obj->base.size >> PAGE_SHIFT, GFP)) {
kfree(st);
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
}
/* Use optimal page sized chunks to fill in the sg table */
@@ -227,13 +231,15 @@ static int fake_get_huge_pages(struct drm_i915_gem_object *obj)
obj->mm.madv = I915_MADV_DONTNEED;
- __i915_gem_object_set_pages(obj, st, sg_page_sizes);
-
- return 0;
+ *sizes = sg_page_sizes;
+ return st;
}
-static int fake_get_huge_pages_single(struct drm_i915_gem_object *obj)
+static struct sg_table *
+fake_get_huge_pages_single(struct i915_gem_object_get_pages_context *ctx,
+ unsigned int *sizes)
{
+ struct drm_i915_gem_object *obj = ctx->object;
struct drm_i915_private *i915 = to_i915(obj->base.dev);
struct sg_table *st;
struct scatterlist *sg;
@@ -241,11 +247,11 @@ static int fake_get_huge_pages_single(struct drm_i915_gem_object *obj)
st = kmalloc(sizeof(*st), GFP);
if (!st)
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
if (sg_alloc_table(st, 1, GFP)) {
kfree(st);
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
}
sg = st->sgl;
@@ -261,9 +267,8 @@ static int fake_get_huge_pages_single(struct drm_i915_gem_object *obj)
obj->mm.madv = I915_MADV_DONTNEED;
- __i915_gem_object_set_pages(obj, st, sg->length);
-
- return 0;
+ *sizes = sg->length;
+ return st;
#undef GFP
}
diff --git a/drivers/gpu/drm/i915/gvt/dmabuf.c b/drivers/gpu/drm/i915/gvt/dmabuf.c
index 41c8ebc60c63..c2a81939f698 100644
--- a/drivers/gpu/drm/i915/gvt/dmabuf.c
+++ b/drivers/gpu/drm/i915/gvt/dmabuf.c
@@ -36,9 +36,11 @@
#define GEN8_DECODE_PTE(pte) (pte & GENMASK_ULL(63, 12))
-static int vgpu_gem_get_pages(
- struct drm_i915_gem_object *obj)
+static struct sg_table *
+vgpu_gem_get_pages(struct i915_gem_object_get_pages_context *ctx,
+ unsigned int *sizes)
{
+ struct drm_i915_gem_object *obj = ctx->object;
struct drm_i915_private *dev_priv = to_i915(obj->base.dev);
struct sg_table *st;
struct scatterlist *sg;
@@ -49,17 +51,17 @@ static int vgpu_gem_get_pages(
fb_info = (struct intel_vgpu_fb_info *)obj->gvt_info;
if (WARN_ON(!fb_info))
- return -ENODEV;
+ return ERR_PTR(-ENODEV);
st = kmalloc(sizeof(*st), GFP_KERNEL);
if (unlikely(!st))
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
page_num = obj->base.size >> PAGE_SHIFT;
ret = sg_alloc_table(st, page_num, GFP_KERNEL);
if (ret) {
kfree(st);
- return ret;
+ return ERR_PTR(ret);
}
gtt_entries = (gen8_pte_t __iomem *)dev_priv->ggtt.gsm +
(fb_info->start >> PAGE_SHIFT);
@@ -71,9 +73,8 @@ static int vgpu_gem_get_pages(
sg_dma_len(sg) = PAGE_SIZE;
}
- __i915_gem_object_set_pages(obj, st, PAGE_SIZE);
-
- return 0;
+ *sizes = PAGE_SIZE;
+ return st;
}
static void vgpu_gem_put_pages(struct drm_i915_gem_object *obj,
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 9a1ec487e8b1..ef9e4cb49b4a 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -786,13 +786,6 @@ struct i915_gem_mm {
struct notifier_block vmap_notifier;
struct shrinker shrinker;
- /**
- * Workqueue to fault in userptr pages, flushed by the execbuf
- * when required but otherwise left to userspace to try again
- * on EAGAIN.
- */
- struct workqueue_struct *userptr_wq;
-
u64 unordered_timeline;
/* the indicator for dispatch video commands on two BSD rings */
@@ -2514,7 +2507,7 @@ static inline bool intel_vgpu_active(struct drm_i915_private *dev_priv)
}
/* i915_gem.c */
-int i915_gem_init_userptr(struct drm_i915_private *dev_priv);
+void i915_gem_init_userptr(struct drm_i915_private *dev_priv);
void i915_gem_cleanup_userptr(struct drm_i915_private *dev_priv);
void i915_gem_sanitize(struct drm_i915_private *i915);
int i915_gem_init_early(struct drm_i915_private *dev_priv);
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 21416b87905e..1571c707ad15 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -1544,10 +1544,7 @@ int i915_gem_init(struct drm_i915_private *dev_priv)
dev_priv->mm.unordered_timeline = dma_fence_context_alloc(1);
i915_timelines_init(dev_priv);
-
- ret = i915_gem_init_userptr(dev_priv);
- if (ret)
- return ret;
+ i915_gem_init_userptr(dev_priv);
ret = intel_uc_init_misc(dev_priv);
if (ret)
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
index 664b7e3e1d7c..dae1f8634a38 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
@@ -54,10 +54,13 @@ static void fake_free_pages(struct drm_i915_gem_object *obj,
kfree(pages);
}
-static int fake_get_pages(struct drm_i915_gem_object *obj)
+static struct sg_table *
+fake_get_pages(struct i915_gem_object_get_pages_context *ctx,
+ unsigned int *sizes)
{
#define GFP (GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY)
#define PFN_BIAS 0x1000
+ struct drm_i915_gem_object *obj = ctx->object;
struct sg_table *pages;
struct scatterlist *sg;
unsigned int sg_page_sizes;
@@ -65,12 +68,12 @@ static int fake_get_pages(struct drm_i915_gem_object *obj)
pages = kmalloc(sizeof(*pages), GFP);
if (!pages)
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
rem = round_up(obj->base.size, BIT(31)) >> 31;
if (sg_alloc_table(pages, rem, GFP)) {
kfree(pages);
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
}
sg_page_sizes = 0;
@@ -90,9 +93,8 @@ static int fake_get_pages(struct drm_i915_gem_object *obj)
obj->mm.madv = I915_MADV_DONTNEED;
- __i915_gem_object_set_pages(obj, pages, sg_page_sizes);
-
- return 0;
+ *sizes = sg_page_sizes;
+ return pages;
#undef GFP
}
--
2.20.1
More information about the Intel-gfx
mailing list