[Intel-xe] [PATCH v2 1/6] drm/xe/bo: Don't limit the TT manager to half of system memory

[Thomas Hellström]

Hi,

On 6/19/23 18:46, Matthew Auld wrote:
> On Mon, 19 Jun 2023 at 16:22, Thomas Hellström
> <thomas.hellstrom at linux.intel.com> wrote:
>> TTM enforces this limit a page-allocation time, swapping out TT memory
>> if needed.
> It seems that limit can be exceeded by just creating one crazy big
> object, or having a few as part of the same exec() call. If unable to
> lock an object it seems to just skip it (-EBUSY), and if there are no
> objects left to iterate over it just carries on with the allocation.
> The hard limit in sys_mgr seemed to prevent that. Do we know if this
> new behaviour is now expected?

Hm. You're right. Once there was a quite elaborate accounting mechanism 
in TTM, and IIRC bypassing it in that way wasn't possible. Now it looks 
more like an opportunistic swapout.

Sigh, Yes I guess then we have to leave this where it is and adjust the 
IGT accordingly. I figure I need to resurrect that TTM shrinker series.

/Thomas


>> Signed-off-by: Thomas Hellström <thomas.hellstrom at linux.intel.com>
>> Reviewed-by: Matthew Brost <matthew.brost at intel.com>
>> ---
>>   drivers/gpu/drm/xe/xe_ttm_sys_mgr.c | 2 --
>>   1 file changed, 2 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/xe/xe_ttm_sys_mgr.c b/drivers/gpu/drm/xe/xe_ttm_sys_mgr.c
>> index 3e1fa0c832ca..961bbc29dcdf 100644
>> --- a/drivers/gpu/drm/xe/xe_ttm_sys_mgr.c
>> +++ b/drivers/gpu/drm/xe/xe_ttm_sys_mgr.c
>> @@ -106,8 +106,6 @@ int xe_ttm_sys_mgr_init(struct xe_device *xe)
>>
>>          si_meminfo(&si);
>>          gtt_size = (u64)si.totalram * si.mem_unit;
>> -       /* TTM limits allocation of all TTM devices by 50% of system memory */
>> -       gtt_size /= 2;
>>
>>          man->use_tt = true;
>>          man->func = &xe_ttm_sys_mgr_func;
>> --
>> 2.40.1
>>