Re: [PATCH v2 6/7] drm/ttm: Reduce the number of used allocation orders for TTM pages

[Christian König <christian.koenig@xxxxxxx>]

Am 07.03.23 um 15:46 schrieb Thomas Hellström:
> When swapping out, we will split multi-order pages both in order to
> move them to the swap-cache and to be able to return memory to the
> swap cache as soon as possible on a page-by-page basis.
> Reduce the page max order to the system PMD size, as we can then be nicer
> to the system and avoid splitting gigantic pages.

Mhm, we actually have a todo to start supporting giant pages at some time.

Using the folio directly just saves tons of overhead when you don't need 
to allocate 2MiG page array any more for each 1GiB you allocate.

But that probably needs tons of work anyway, so feel free to add my rb 
for now.

Regards,
Christian.

> Looking forward to when we might be able to swap out PMD size folios
> without splitting, this will also be a benefit.
>
> v2:
> - Include all orders up to the PMD size (Christian König)
>
> Signed-off-by: Thomas Hellström <thomas.hellstrom@xxxxxxxxxxxxxxx>
> ---
>   drivers/gpu/drm/ttm/ttm_pool.c | 27 ++++++++++++++++-----------
>   1 file changed, 16 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c
> index 0b6e20613d19..939845d853af 100644
> --- a/drivers/gpu/drm/ttm/ttm_pool.c
> +++ b/drivers/gpu/drm/ttm/ttm_pool.c
> @@ -47,6 +47,9 @@
>   
>   #include "ttm_module.h"
>   
> +#define TTM_MAX_ORDER (PMD_SHIFT - PAGE_SHIFT)
> +#define TTM_DIM_ORDER (TTM_MAX_ORDER + 1)
> +
>   /**
>    * struct ttm_pool_dma - Helper object for coherent DMA mappings
>    *
> @@ -65,11 +68,11 @@ module_param(page_pool_size, ulong, 0644);
>   
>   static atomic_long_t allocated_pages;
>   
> -static struct ttm_pool_type global_write_combined[MAX_ORDER];
> -static struct ttm_pool_type global_uncached[MAX_ORDER];
> +static struct ttm_pool_type global_write_combined[TTM_DIM_ORDER];
> +static struct ttm_pool_type global_uncached[TTM_DIM_ORDER];
>   
> -static struct ttm_pool_type global_dma32_write_combined[MAX_ORDER];
> -static struct ttm_pool_type global_dma32_uncached[MAX_ORDER];
> +static struct ttm_pool_type global_dma32_write_combined[TTM_DIM_ORDER];
> +static struct ttm_pool_type global_dma32_uncached[TTM_DIM_ORDER];
>   
>   static spinlock_t shrinker_lock;
>   static struct list_head shrinker_list;
> @@ -431,7 +434,7 @@ int ttm_pool_alloc(struct ttm_pool *pool, struct ttm_tt *tt,
>   	else
>   		gfp_flags |= GFP_HIGHUSER;
>   
> -	for (order = min_t(unsigned int, MAX_ORDER - 1, __fls(num_pages));
> +	for (order = min_t(unsigned int, TTM_MAX_ORDER, __fls(num_pages));
>   	     num_pages;
>   	     order = min_t(unsigned int, order, __fls(num_pages))) {
>   		struct ttm_pool_type *pt;
> @@ -550,7 +553,7 @@ void ttm_pool_init(struct ttm_pool *pool, struct device *dev,
>   
>   	if (use_dma_alloc) {
>   		for (i = 0; i < TTM_NUM_CACHING_TYPES; ++i)
> -			for (j = 0; j < MAX_ORDER; ++j)
> +			for (j = 0; j < TTM_DIM_ORDER; ++j)
>   				ttm_pool_type_init(&pool->caching[i].orders[j],
>   						   pool, i, j);
>   	}
> @@ -570,7 +573,7 @@ void ttm_pool_fini(struct ttm_pool *pool)
>   
>   	if (pool->use_dma_alloc) {
>   		for (i = 0; i < TTM_NUM_CACHING_TYPES; ++i)
> -			for (j = 0; j < MAX_ORDER; ++j)
> +			for (j = 0; j < TTM_DIM_ORDER; ++j)
>   				ttm_pool_type_fini(&pool->caching[i].orders[j]);
>   	}
>   
> @@ -624,7 +627,7 @@ static void ttm_pool_debugfs_header(struct seq_file *m)
>   	unsigned int i;
>   
>   	seq_puts(m, "\t ");
> -	for (i = 0; i < MAX_ORDER; ++i)
> +	for (i = 0; i < TTM_DIM_ORDER; ++i)
>   		seq_printf(m, " ---%2u---", i);
>   	seq_puts(m, "\n");
>   }
> @@ -635,7 +638,7 @@ static void ttm_pool_debugfs_orders(struct ttm_pool_type *pt,
>   {
>   	unsigned int i;
>   
> -	for (i = 0; i < MAX_ORDER; ++i)
> +	for (i = 0; i < TTM_DIM_ORDER; ++i)
>   		seq_printf(m, " %8u", ttm_pool_type_count(&pt[i]));
>   	seq_puts(m, "\n");
>   }
> @@ -738,13 +741,15 @@ int ttm_pool_mgr_init(unsigned long num_pages)
>   {
>   	unsigned int i;
>   
> +	BUILD_BUG_ON(TTM_DIM_ORDER > MAX_ORDER);
> +
>   	if (!page_pool_size)
>   		page_pool_size = num_pages;
>   
>   	spin_lock_init(&shrinker_lock);
>   	INIT_LIST_HEAD(&shrinker_list);
>   
> -	for (i = 0; i < MAX_ORDER; ++i) {
> +	for (i = 0; i < TTM_DIM_ORDER; ++i) {
>   		ttm_pool_type_init(&global_write_combined[i], NULL,
>   				   ttm_write_combined, i);
>   		ttm_pool_type_init(&global_uncached[i], NULL, ttm_uncached, i);
> @@ -777,7 +782,7 @@ void ttm_pool_mgr_fini(void)
>   {
>   	unsigned int i;
>   
> -	for (i = 0; i < MAX_ORDER; ++i) {
> +	for (i = 0; i < TTM_DIM_ORDER; ++i) {
>   		ttm_pool_type_fini(&global_write_combined[i]);
>   		ttm_pool_type_fini(&global_uncached[i]);
>