On Wed, Dec 21, 2011 at 10:17:36AM -0500, Jerome Glisse wrote:
> On Mon, Dec 12, 2011 at 3:09 PM, Konrad Rzeszutek Wilk
> <konrad.wilk at oracle.com> wrote:
> > The code to figure out how many pages to shrink the pool
> > ends up capping the 'count' at _manager->options.max_size - which is OK.
> > Except that the 'count' is also used when accounting for how many pages
> > are recycled - which we end up with the invalid values. This fixes
> > it by using a different value for the amount of pages to shrink.
> >
> > Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk at oracle.com>
> > ---
> > ?drivers/gpu/drm/ttm/ttm_page_alloc_dma.c | ? 10 ++++++----
> > ?1 files changed, 6 insertions(+), 4 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> > b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> > index 6c06d0b..e57aa24 100644
> > --- a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> > +++ b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> > @@ -949,7 +949,7 @@ void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma,
> > struct device *dev)
> > ? ? ? ?struct dma_page *d_page, *next;
> > ? ? ? ?enum pool_type type;
> > ? ? ? ?bool is_cached = false;
> > - ? ? ? unsigned count = 0, i;
> > + ? ? ? unsigned count = 0, i, npages;
> > ? ? ? ?unsigned long irq_flags;
> >
> > ? ? ? ?type = ttm_to_type(ttm->page_flags, ttm->caching_state);
> > @@ -971,11 +971,13 @@ void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma,
> > struct device *dev)
> > ? ? ? ?pool->npages_in_use -= count;
> > ? ? ? ?if (is_cached) {
> > ? ? ? ? ? ? ? ?pool->nfrees += count;
> > + ? ? ? ? ? ? ? npages = count;
>
> This is wrong, i thought we didn't wanted to shrink cached page, your
> 3/3 patch remove this line. Thought i am kind of wondering why we
> don't shrink the cached pool, i can't remember the reason.
We delete the cached ones:
988 if (is_cached) {
989 list_for_each_entry_safe(d_page, next,
&ttm_dma->pages_list, page_list) {
990 ttm_mem_global_free_page(ttm->glob->mem_glob,
991 d_page->p);
992 ttm_dma_page_put(pool, d_page);
993 }
And then we end up calling in ttm_dma_page_pool_free for (count) pages
that we had already deleted. So it ends up being a bit redundant.
The 3/3 patch fixes that (by removing the call to ttm_dma_page_pool_free
for is_cached. This patch (2/3) tries to preserve the logic if it is either
is_cached or !is_cached.
>
> > ? ? ? ?} else {
> > ? ? ? ? ? ? ? ?pool->npages_free += count;
> > ? ? ? ? ? ? ? ?list_splice(&ttm_dma->pages_list, &pool->free_list);
> > + ? ? ? ? ? ? ? npages = count;
> > ? ? ? ? ? ? ? ?if (pool->npages_free > _manager->options.max_size) {
> > - ? ? ? ? ? ? ? ? ? ? ? count = pool->npages_free -
> > _manager->options.max_size;
> > + ? ? ? ? ? ? ? ? ? ? ? npages = pool->npages_free -
> > _manager->options.max_size;
> > ? ? ? ? ? ? ? ?}
> > ? ? ? ?}
> > ? ? ? ?spin_unlock_irqrestore(&pool->lock, irq_flags);
> > @@ -1000,8 +1002,8 @@ void ttm_dma_unpopulate(struct ttm_dma_tt *ttm_dma,
> > struct device *dev)
> > ? ? ? ?}
> >
> > ? ? ? ?/* shrink pool if necessary */
> > - ? ? ? if (count)
> > - ? ? ? ? ? ? ? ttm_dma_page_pool_free(pool, count);
> > + ? ? ? if (npages)
> > + ? ? ? ? ? ? ? ttm_dma_page_pool_free(pool, npages);
> > ? ? ? ?ttm->state = tt_unpopulated;
> > ?}
> > ?EXPORT_SYMBOL_GPL(ttm_dma_unpopulate);
> > --
> > 1.7.7.3
> >
>
> Otherwise Reviewed-by:Jerome Glisse <jglisse at redhat.com>
Thanks!
