[Intel-gfx] [PATCH 4/4] drm/i915: Use __sg_alloc_table_from_pages for userptr allocations
Tvrtko Ursulin
tursulin at ursulin.net
Mon Jan 16 14:12:10 UTC 2017
From: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
With the addition of __sg_alloc_table_from_pages we can control
the maximum coallescing size and eliminate a separate path for
allocating backing store here.
Similar to 871dfbd67d4e ("drm/i915: Allow compaction upto
SWIOTLB max segment size") this enables more compact sg lists to
be created and so has a beneficial effect on workloads with many
and/or large objects of this class.
v2:
* Rename helper to i915_sg_segment_size and fix swiotlb override.
* Commit message update.
v3:
* Actually include the swiotlb override fix.
v4:
* Regroup parameters a bit. (Chris Wilson)
v5:
* Rebase for swiotlb_max_segment.
* Add DMA map failure handling as in abb0deacb5a6
("drm/i915: Fallback to single PAGE_SIZE segments for DMA remapping").
v6: Handle swiotlb_max_segment() returning 1. (Joonas Lahtinen)
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
Cc: Chris Wilson <chris at chris-wilson.co.uk>
Cc: linux-kernel at vger.kernel.org
Reviewed-by: Chris Wilson <chris at chris-wilson.co.uk> (v4)
Cc: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
---
drivers/gpu/drm/i915/i915_drv.h | 15 +++++++
drivers/gpu/drm/i915/i915_gem.c | 6 +--
drivers/gpu/drm/i915/i915_gem_userptr.c | 79 ++++++++++++---------------------
3 files changed, 45 insertions(+), 55 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index f861418122ef..531f47fae143 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -2599,6 +2599,21 @@ static inline struct scatterlist *__sg_next(struct scatterlist *sg)
(((__iter).curr += PAGE_SIZE) < (__iter).max) || \
((__iter) = __sgt_iter(__sg_next((__iter).sgp), false), 0))
+static inline unsigned int i915_sg_segment_size(void)
+{
+ unsigned int size = swiotlb_max_segment();
+
+ if (size == 0)
+ return SCATTERLIST_MAX_SEGMENT;
+
+ size = rounddown(size, PAGE_SIZE);
+ /* swiotlb_max_segment_size can return 1 byte when it means one page. */
+ if (size < PAGE_SIZE)
+ size = PAGE_SIZE;
+
+ return size;
+}
+
static inline const struct intel_device_info *
intel_info(const 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 d4c59b53532e..4dca3b52f495 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -2254,7 +2254,7 @@ i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj)
struct sgt_iter sgt_iter;
struct page *page;
unsigned long last_pfn = 0; /* suppress gcc warning */
- unsigned int max_segment;
+ unsigned int max_segment = i915_sg_segment_size();
int ret;
gfp_t gfp;
@@ -2265,10 +2265,6 @@ i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj)
GEM_BUG_ON(obj->base.read_domains & I915_GEM_GPU_DOMAINS);
GEM_BUG_ON(obj->base.write_domain & I915_GEM_GPU_DOMAINS);
- max_segment = swiotlb_max_segment();
- if (!max_segment)
- max_segment = rounddown(UINT_MAX, PAGE_SIZE);
-
st = kmalloc(sizeof(*st), GFP_KERNEL);
if (st == NULL)
return ERR_PTR(-ENOMEM);
diff --git a/drivers/gpu/drm/i915/i915_gem_userptr.c b/drivers/gpu/drm/i915/i915_gem_userptr.c
index 6a8fa085b74e..95b62b9c5cd6 100644
--- a/drivers/gpu/drm/i915/i915_gem_userptr.c
+++ b/drivers/gpu/drm/i915/i915_gem_userptr.c
@@ -390,64 +390,42 @@ struct get_pages_work {
struct task_struct *task;
};
-#if IS_ENABLED(CONFIG_SWIOTLB)
-#define swiotlb_active() swiotlb_nr_tbl()
-#else
-#define swiotlb_active() 0
-#endif
-
-static int
-st_set_pages(struct sg_table **st, struct page **pvec, int num_pages)
-{
- struct scatterlist *sg;
- int ret, n;
-
- *st = kmalloc(sizeof(**st), GFP_KERNEL);
- if (*st == NULL)
- return -ENOMEM;
-
- if (swiotlb_active()) {
- ret = sg_alloc_table(*st, num_pages, GFP_KERNEL);
- if (ret)
- goto err;
-
- for_each_sg((*st)->sgl, sg, num_pages, n)
- sg_set_page(sg, pvec[n], PAGE_SIZE, 0);
- } else {
- ret = sg_alloc_table_from_pages(*st, pvec, num_pages,
- 0, num_pages << PAGE_SHIFT,
- GFP_KERNEL);
- if (ret)
- goto err;
- }
-
- return 0;
-
-err:
- kfree(*st);
- *st = NULL;
- return ret;
-}
-
static struct sg_table *
-__i915_gem_userptr_set_pages(struct drm_i915_gem_object *obj,
- struct page **pvec, int num_pages)
+__i915_gem_userptr_alloc_pages(struct drm_i915_gem_object *obj,
+ struct page **pvec, int num_pages)
{
- struct sg_table *pages;
+ unsigned int max_segment = i915_sg_segment_size();
+ struct sg_table *st;
int ret;
- ret = st_set_pages(&pages, pvec, num_pages);
- if (ret)
+ st = kmalloc(sizeof(*st), GFP_KERNEL);
+ if (!st)
+ return ERR_PTR(-ENOMEM);
+
+alloc_table:
+ ret = __sg_alloc_table_from_pages(st, pvec, num_pages,
+ 0, num_pages << PAGE_SHIFT,
+ max_segment,
+ GFP_KERNEL);
+ if (ret) {
+ kfree(st);
return ERR_PTR(ret);
+ }
- ret = i915_gem_gtt_prepare_pages(obj, pages);
+ ret = i915_gem_gtt_prepare_pages(obj, st);
if (ret) {
- sg_free_table(pages);
- kfree(pages);
+ sg_free_table(st);
+
+ if (max_segment > PAGE_SIZE) {
+ max_segment = PAGE_SIZE;
+ goto alloc_table;
+ }
+
+ kfree(st);
return ERR_PTR(ret);
}
- return pages;
+ return st;
}
static int
@@ -531,7 +509,8 @@ __i915_gem_userptr_get_pages_worker(struct work_struct *_work)
struct sg_table *pages = ERR_PTR(ret);
if (pinned == npages) {
- pages = __i915_gem_userptr_set_pages(obj, pvec, npages);
+ pages = __i915_gem_userptr_alloc_pages(obj, pvec,
+ npages);
if (!IS_ERR(pages)) {
__i915_gem_object_set_pages(obj, pages);
pinned = 0;
@@ -653,7 +632,7 @@ i915_gem_userptr_get_pages(struct drm_i915_gem_object *obj)
else if (pinned < num_pages)
pages = __i915_gem_userptr_get_pages_schedule(obj, &active);
else
- pages = __i915_gem_userptr_set_pages(obj, pvec, num_pages);
+ pages = __i915_gem_userptr_alloc_pages(obj, pvec, num_pages);
if (IS_ERR(pages)) {
__i915_gem_userptr_set_active(obj, active);
release_pages(pvec, pinned, 0);
--
2.7.4
More information about the Intel-gfx
mailing list