On 2017å¹´11æ??22æ?¥ 13:36, Roger He wrote:
> to indicate page order for each element in the pool
>
> Change-Id: Ic609925ca5d2a5d4ad49d6becf505388ce3624cf
> Signed-off-by: Roger He <Hongbo.He at amd.com>
> ---
> drivers/gpu/drm/ttm/ttm_page_alloc.c | 42 ++++++++++++++++++++++++++----------
> 1 file changed, 31 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc.c b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> index 72ea037..0a0c653 100644
> --- a/drivers/gpu/drm/ttm/ttm_page_alloc.c
> +++ b/drivers/gpu/drm/ttm/ttm_page_alloc.c
> @@ -81,6 +81,7 @@ struct ttm_page_pool {
> char *name;
> unsigned long nfrees;
> unsigned long nrefills;
> + unsigned int order;
> };
>
> /**
> @@ -412,6 +413,7 @@ ttm_pool_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
> struct ttm_page_pool *pool;
> int shrink_pages = sc->nr_to_scan;
> unsigned long freed = 0;
> + unsigned int nr_free_pool;
>
> if (!mutex_trylock(&lock))
> return SHRINK_STOP;
> @@ -421,10 +423,15 @@ ttm_pool_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
> unsigned nr_free = shrink_pages;
> if (shrink_pages == 0)
> break;
> +
> pool = &_manager->pools[(i + pool_offset)%NUM_POOLS];
> /* OK to use static buffer since global mutex is held. */
> - shrink_pages = ttm_page_pool_free(pool, nr_free, true);
> - freed += nr_free - shrink_pages;
> + nr_free_pool = (nr_free >> pool->order);
How about nr_free_pool = roundup(nr_free >> pool->order, 1)?
 this way, your patch#4 is not need.
Regards,
David Zhou
> + if (nr_free_pool == 0)
> + continue;
> +
> + shrink_pages = ttm_page_pool_free(pool, nr_free_pool, true);
> + freed += ((nr_free_pool - shrink_pages) << pool->order);
> }
> mutex_unlock(&lock);
> return freed;
> @@ -436,9 +443,12 @@ ttm_pool_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
> {
> unsigned i;
> unsigned long count = 0;
> + struct ttm_page_pool *pool;
>
> - for (i = 0; i < NUM_POOLS; ++i)
> - count += _manager->pools[i].npages;
> + for (i = 0; i < NUM_POOLS; ++i) {
> + pool = &_manager->pools[i];
> + count += (pool->npages << pool->order);
> + }
>
> return count;
> }
> @@ -932,7 +942,7 @@ static int ttm_get_pages(struct page **pages, unsigned npages, int flags,
> }
>
> static void ttm_page_pool_init_locked(struct ttm_page_pool *pool, gfp_t flags,
> - char *name)
> + char *name, unsigned int order)
> {
> spin_lock_init(&pool->lock);
> pool->fill_lock = false;
> @@ -940,8 +950,18 @@ static void ttm_page_pool_init_locked(struct ttm_page_pool *pool, gfp_t flags,
> pool->npages = pool->nfrees = 0;
> pool->gfp_flags = flags;
> pool->name = name;
> + pool->order = order;
> }
>
> +/**
> + * Actually if TRANSPARENT_HUGEPAGE not enabled, we will not use
> + * wc_pool_huge and uc_pool_huge, so no matter whatever the page
> + * order are for those two pools
> + */
> +#ifndef CONFIG_TRANSPARENT_HUGEPAGE
> +#define HPAGE_PMD_ORDER 9
> +#endif
> +
> int ttm_page_alloc_init(struct ttm_mem_global *glob, unsigned max_pages)
> {
> int ret;
> @@ -952,23 +972,23 @@ int ttm_page_alloc_init(struct ttm_mem_global *glob, unsigned max_pages)
>
> _manager = kzalloc(sizeof(*_manager), GFP_KERNEL);
>
> - ttm_page_pool_init_locked(&_manager->wc_pool, GFP_HIGHUSER, "wc");
> + ttm_page_pool_init_locked(&_manager->wc_pool, GFP_HIGHUSER, "wc", 0);
>
> - ttm_page_pool_init_locked(&_manager->uc_pool, GFP_HIGHUSER, "uc");
> + ttm_page_pool_init_locked(&_manager->uc_pool, GFP_HIGHUSER, "uc", 0);
>
> ttm_page_pool_init_locked(&_manager->wc_pool_dma32,
> - GFP_USER | GFP_DMA32, "wc dma");
> + GFP_USER | GFP_DMA32, "wc dma", 0);
>
> ttm_page_pool_init_locked(&_manager->uc_pool_dma32,
> - GFP_USER | GFP_DMA32, "uc dma");
> + GFP_USER | GFP_DMA32, "uc dma", 0);
>
> ttm_page_pool_init_locked(&_manager->wc_pool_huge,
> GFP_TRANSHUGE & ~(__GFP_MOVABLE | __GFP_COMP),
> - "wc huge");
> + "wc huge", HPAGE_PMD_ORDER);
>
> ttm_page_pool_init_locked(&_manager->uc_pool_huge,
> GFP_TRANSHUGE & ~(__GFP_MOVABLE | __GFP_COMP)
> - , "uc huge");
> + , "uc huge", HPAGE_PMD_ORDER);
>
> _manager->options.max_size = max_pages;
> _manager->options.small = SMALL_ALLOCATION;
