[PATCH RESEND v3 2/3] drm/ttm: Reduce the number of used allocation orders for TTM pages

[Daniel Vetter]

On Tue, Apr 04, 2023 at 10:06:49PM +0200, Thomas Hellström wrote:
> 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.
> 
> 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)
> v3:
> - Avoid compilation errors for architectures with special PFN_SHIFTs
> 
> Signed-off-by: Thomas Hellström <thomas.hellstrom at linux.intel.com>
> Reviewed-by: Christian König <christian.koenig at amd.com>

Apparently this fails on ppc build testing. Please supply build fix asap
(or I guess we need to revert). I'm kinda not clear why this only showed
up when I merged the drm-misc-next pr into drm-next ...
-Daniel

> ---
>  drivers/gpu/drm/ttm/ttm_pool.c | 30 +++++++++++++++++++-----------
>  1 file changed, 19 insertions(+), 11 deletions(-)
> 
> diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c
> index dfce896c4bae..18c342a919a2 100644
> --- a/drivers/gpu/drm/ttm/ttm_pool.c
> +++ b/drivers/gpu/drm/ttm/ttm_pool.c
> @@ -47,6 +47,11 @@
>  
>  #include "ttm_module.h"
>  
> +#define TTM_MAX_ORDER (PMD_SHIFT - PAGE_SHIFT)
> +#define __TTM_DIM_ORDER (TTM_MAX_ORDER + 1)
> +/* Some architectures have a weird PMD_SHIFT */
> +#define TTM_DIM_ORDER (__TTM_DIM_ORDER <= MAX_ORDER ? __TTM_DIM_ORDER : MAX_ORDER)
> +
>  /**
>   * struct ttm_pool_dma - Helper object for coherent DMA mappings
>   *
> @@ -65,11 +70,11 @@ 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;
> @@ -444,7 +449,7 @@ 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;
> @@ -563,7 +568,7 @@ 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);
>  	}
> @@ -583,7 +588,7 @@ 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]);
>  	}
>  
> @@ -637,7 +642,7 @@ 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");
>  }
> @@ -648,7 +653,7 @@ 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");
>  }
> @@ -751,13 +756,16 @@ int ttm_pool_mgr_init(unsigned long num_pages)
>  {
>  	unsigned int i;
>  
> +	BUILD_BUG_ON(TTM_DIM_ORDER > MAX_ORDER);
> +	BUILD_BUG_ON(TTM_DIM_ORDER < 1);
> +
>  	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);
> @@ -790,7 +798,7 @@ 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]);
>  
> -- 
> 2.39.2
> 

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch