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

[Thomas Hellström]

On 3/8/23 10:15, Christian König wrote:
> 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.

Thanks, I need to fix this anyway for powerpc where it seems PMD_ORDER > 
MAX_ORDER :/

It might be we'd want to replace the ttm page arrays with scatter-gather 
tables at some point?
I think at least vmwgfx, i915 and xe would benefit from that...

/Thomas

>
> 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 at linux.intel.com>
>> ---
>>   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]);
>