On 29 June 2017 at 07:36, Zhenyu Wang <zhen...@linux.intel.com> wrote:
> On 2017.06.27 15:54:29 +0100, Matthew Auld wrote:
>> In preparation for supporting huge gtt pages for the ppgtt, we introduce
>> page size members for gem objects. We fill in the page sizes by
>> scanning the sg table.
>>
>> v2: pass the sg_mask to set_pages
>>
>> v3: calculate the sg_mask inline with populating the sg_table where
>> possible, and pass to set_pages along with the pages.
>>
>> Signed-off-by: Matthew Auld <matthew.a...@intel.com>
>> Cc: Joonas Lahtinen <joonas.lahti...@linux.intel.com>
>> Cc: Chris Wilson <ch...@chris-wilson.co.uk>
>> Cc: Daniel Vetter <dan...@ffwll.ch>
>> ---
>> drivers/gpu/drm/i915/i915_drv.h | 5 ++-
>> drivers/gpu/drm/i915/i915_gem.c | 43
>> ++++++++++++++++++++----
>> drivers/gpu/drm/i915/i915_gem_dmabuf.c | 17 ++++++++--
>> drivers/gpu/drm/i915/i915_gem_internal.c | 5 ++-
>> drivers/gpu/drm/i915/i915_gem_object.h | 20 ++++++++++-
>> drivers/gpu/drm/i915/i915_gem_stolen.c | 13 ++++---
>> drivers/gpu/drm/i915/i915_gem_userptr.c | 26 ++++++++++----
>> drivers/gpu/drm/i915/selftests/huge_gem_object.c | 4 ++-
>> drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 3 +-
>> 9 files changed, 110 insertions(+), 26 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/i915/i915_drv.h
>> b/drivers/gpu/drm/i915/i915_drv.h
>> index fe225cb9b622..0539f210622f 100644
>> --- a/drivers/gpu/drm/i915/i915_drv.h
>> +++ b/drivers/gpu/drm/i915/i915_drv.h
>> @@ -2951,6 +2951,8 @@ intel_info(const struct drm_i915_private *dev_priv)
>> #define USES_PPGTT(dev_priv) (i915.enable_ppgtt)
>> #define USES_FULL_PPGTT(dev_priv) (i915.enable_ppgtt >= 2)
>> #define USES_FULL_48BIT_PPGTT(dev_priv) (i915.enable_ppgtt == 3)
>> +#define HAS_PAGE_SIZE(dev_priv, page_size) \
>> + ((dev_priv)->info.page_size_mask & (page_size))
>>
>> #define HAS_OVERLAY(dev_priv) ((dev_priv)->info.has_overlay)
>> #define OVERLAY_NEEDS_PHYSICAL(dev_priv) \
>> @@ -3335,7 +3337,8 @@ i915_gem_object_get_dma_address(struct
>> drm_i915_gem_object *obj,
>> unsigned long n);
>>
>> void __i915_gem_object_set_pages(struct drm_i915_gem_object *obj,
>> - struct sg_table *pages);
>> + struct sg_table *pages,
>> + unsigned int sg_mask);
>> int __i915_gem_object_get_pages(struct drm_i915_gem_object *obj);
>>
>> static inline int __must_check
>> diff --git a/drivers/gpu/drm/i915/i915_gem.c
>> b/drivers/gpu/drm/i915/i915_gem.c
>> index 80539a821004..4d29a5cfa0c4 100644
>> --- a/drivers/gpu/drm/i915/i915_gem.c
>> +++ b/drivers/gpu/drm/i915/i915_gem.c
>> @@ -163,7 +163,8 @@ i915_gem_get_aperture_ioctl(struct drm_device *dev, void
>> *data,
>> }
>>
>> static struct sg_table *
>> -i915_gem_object_get_pages_phys(struct drm_i915_gem_object *obj)
>> +i915_gem_object_get_pages_phys(struct drm_i915_gem_object *obj,
>> + unsigned int *sg_mask)
>> {
>> struct address_space *mapping = obj->base.filp->f_mapping;
>> drm_dma_handle_t *phys;
>> @@ -223,6 +224,8 @@ i915_gem_object_get_pages_phys(struct
>> drm_i915_gem_object *obj)
>> sg->offset = 0;
>> sg->length = obj->base.size;
>>
>> + *sg_mask = sg->length;
>> +
>> sg_dma_address(sg) = phys->busaddr;
>> sg_dma_len(sg) = obj->base.size;
>>
>> @@ -2314,6 +2317,8 @@ void __i915_gem_object_put_pages(struct
>> drm_i915_gem_object *obj,
>> if (!IS_ERR(pages))
>> obj->ops->put_pages(obj, pages);
>>
>> + obj->mm.page_sizes.phys = obj->mm.page_sizes.sg = 0;
>> +
>> unlock:
>> mutex_unlock(&obj->mm.lock);
>> }
>> @@ -2345,7 +2350,8 @@ static bool i915_sg_trim(struct sg_table *orig_st)
>> }
>>
>> static struct sg_table *
>> -i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj)
>> +i915_gem_object_get_pages_gtt(struct drm_i915_gem_object *obj,
>> + unsigned int *sg_mask)
>> {
>> struct drm_i915_private *dev_priv = to_i915(obj->base.dev);
>> const unsigned long page_count = obj->base.size / PAGE_SIZE;
>> @@ -2392,6 +2398,7 @@ i915_gem_object_get_pages_gtt(struct
>> drm_i915_gem_object *obj)
>>
>> sg = st->sgl;
>> st->nents = 0;
>> + *sg_mask = 0;
>> for (i = 0; i < page_count; i++) {
>> const unsigned int shrink[] = {
>> I915_SHRINK_BOUND | I915_SHRINK_UNBOUND |
>> I915_SHRINK_PURGEABLE,
>> @@ -2443,8 +2450,10 @@ i915_gem_object_get_pages_gtt(struct
>> drm_i915_gem_object *obj)
>> if (!i ||
>> sg->length >= max_segment ||
>> page_to_pfn(page) != last_pfn + 1) {
>> - if (i)
>> + if (i) {
>> + *sg_mask |= sg->length;
>> sg = sg_next(sg);
>> + }
>> st->nents++;
>> sg_set_page(sg, page, PAGE_SIZE, 0);
>> } else {
>> @@ -2455,8 +2464,10 @@ i915_gem_object_get_pages_gtt(struct
>> drm_i915_gem_object *obj)
>> /* Check that the i965g/gm workaround works. */
>> WARN_ON((gfp & __GFP_DMA32) && (last_pfn >= 0x00100000UL));
>> }
>> - if (sg) /* loop terminated early; short sg table */
>> + if (sg) { /* loop terminated early; short sg table */
>> + *sg_mask |= sg->length;
>> sg_mark_end(sg);
>> + }
>>
>> /* Trim unused sg entries to avoid wasting memory. */
>> i915_sg_trim(st);
>> @@ -2510,8 +2521,13 @@ i915_gem_object_get_pages_gtt(struct
>> drm_i915_gem_object *obj)
>> }
>>
>> void __i915_gem_object_set_pages(struct drm_i915_gem_object *obj,
>> - struct sg_table *pages)
>> + struct sg_table *pages,
>> + unsigned int sg_mask)
>> {
>> + struct drm_i915_private *i915 = to_i915(obj->base.dev);
>> + unsigned long supported_page_sizes = INTEL_INFO(i915)->page_size_mask;
>> + unsigned int bit;
>> +
>> lockdep_assert_held(&obj->mm.lock);
>>
>> obj->mm.get_page.sg_pos = pages->sgl;
>> @@ -2525,11 +2541,24 @@ void __i915_gem_object_set_pages(struct
>> drm_i915_gem_object *obj,
>> __i915_gem_object_pin_pages(obj);
>> obj->mm.quirked = true;
>> }
>> +
>> + GEM_BUG_ON(!sg_mask);
>> +
>> + obj->mm.page_sizes.phys = sg_mask;
>> +
>> + obj->mm.page_sizes.sg = 0;
>> + for_each_set_bit(bit, &supported_page_sizes, BITS_PER_LONG) {
>> + if (obj->mm.page_sizes.phys & ~0u << bit)
>> + obj->mm.page_sizes.sg |= BIT(bit);
>> + }
>> +
>> + GEM_BUG_ON(!HAS_PAGE_SIZE(i915, obj->mm.page_sizes.sg));
>
> We need to fallback to default supported page size when vGPU is active
> (intel_vgpu_active() is true).
> Currently gvt gtt handling can't support huge page entry yet, we need to
> check either hypervisor
> mm can support huge guest page or just do emulation in gvt.
That should already be the case, since the guest doesn't support the
48b PPGTT. But it looks you are working to change that, so I guess we
should just do this anyway?
>
> --
> Open Source Technology Center, Intel ltd.
>
> $gpg --keyserver wwwkeys.pgp.net --recv-keys 4D781827
>
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
>
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
