Quoting Matthew Auld (2017-06-22 12:07:55)
> On 21 June 2017 at 23:51, Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> wrote:
> > Quoting Chris Wilson (2017-06-21 22:49:07)
> >> Quoting Matthew Auld (2017-06-21 21:33:36)
> >> > Support inserting 1G gtt pages into the 48b PPGTT.
> >> >
> >> > Signed-off-by: Matthew Auld <matthew.auld@xxxxxxxxx>
> >> > Cc: Joonas Lahtinen <joonas.lahtinen@xxxxxxxxxxxxxxx>
> >> > ---
> >> > drivers/gpu/drm/i915/i915_gem_gtt.c | 72 ++++++++++++++++++++++++++++++++++---
> >> > drivers/gpu/drm/i915/i915_gem_gtt.h | 2 ++
> >> > 2 files changed, 70 insertions(+), 4 deletions(-)
> >> >
> >> > diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
> >> > index de67084d5fcf..6fe10ee7dca8 100644
> >> > --- a/drivers/gpu/drm/i915/i915_gem_gtt.c
> >> > +++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
> >> > @@ -922,6 +922,65 @@ static void gen8_ppgtt_insert_3lvl(struct i915_address_space *vm,
> >> > cache_level);
> >> > }
> >> >
> >> > +static void gen8_ppgtt_insert_huge_entries(struct i915_vma *vma,
> >> > + struct i915_page_directory_pointer **pdps,
> >> > + struct sgt_dma *iter,
> >> > + enum i915_cache_level cache_level)
> >> > +{
> >> > + const gen8_pte_t pte_encode = gen8_pte_encode(0, cache_level);
> >> > + u64 start = vma->node.start;
> >> > +
> >> > + do {
> >> > + struct gen8_insert_pte idx = gen8_insert_pte(start);
> >> > + struct i915_page_directory_pointer *pdp = pdps[idx.pml4e];
> >> > + struct i915_page_directory *pd = pdp->page_directory[idx.pdpe];
> >> > + struct i915_page_table *pt = pd->page_table[idx.pde];
> >> > + dma_addr_t rem = iter->max - iter->dma;
> >> > + unsigned int page_size;
> >> > + gen8_pte_t encode = pte_encode;
> >> > + gen8_pte_t *vaddr;
> >> > + u16 index, max;
> >> > +
> >> > + if (unlikely(vma->page_sizes.sg & I915_GTT_PAGE_SIZE_1G) &&
> >> > + IS_ALIGNED(iter->dma, I915_GTT_PAGE_SIZE_1G) &&
> >> > + rem >= I915_GTT_PAGE_SIZE_1G && !(idx.pte | idx.pde)) {
> >> > + vaddr = kmap_atomic_px(pdp);
> >> > + index = idx.pdpe;
> >> > + max = GEN8_PML4ES_PER_PML4;
> >> > + page_size = I915_GTT_PAGE_SIZE_1G;
> >> > + encode |= GEN8_PDPE_PS_1G;
> >> > + } else {
> >> > + vaddr = kmap_atomic_px(pt);
> >> > + index = idx.pte;
> >> > + max = GEN8_PTES;
> >> > + page_size = I915_GTT_PAGE_SIZE;
> >> > + }
> >> > +
> >> > + do {
> >> > + vaddr[index++] = encode | iter->dma;
> >> > +
> >> > + start += page_size;
> >> > + iter->dma += page_size;
> >> > + if (iter->dma >= iter->max) {
> >> > + iter->sg = __sg_next(iter->sg);
> >> > + if (!iter->sg)
> >> > + break;
> >> > +
> >
> > GEM_BUG_ON(iter->sg->length < page_size);
>
> That should be expected behaviour, in that we need to downgrade to a
> smaller page size on the next iteration.
It still applies to just above where we set vaddr[index]. It fails here
because we have yet decided on our course of action. I still think there
is merit in having a confirmation that sg->length does meet our
criteria, considering that we set the page_sizes a long time ago.
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
