On 07/10/2025 16:04, Christian König wrote:
On 07.10.25 16:27, Tvrtko Ursulin wrote:
On 07/10/2025 15:03, Christian König wrote:
On 07.10.25 16:00, Tvrtko Ursulin wrote:
Please not in the header. Neither drivers nor other TTM modules should mess
with such properties.
That is all internal to the pool.
Hmm IMHO it is not that bad. Especially that ttm_pool.c and ttm_tt.c need to
have access to them. Alternatiev is a new header for internal helpers which
sounds a bit too much. But if you insist I can create it.
Wait a second why is ttm_tt.c still needing this? For the DMA32 eviction?
Apparently so, goes back to:
680dcede2762 ("drm/ttm: switch back to static allocation limits for now"
Then there is the newer usage for ttm->use_dma_alloc from:
71ce046327cf ("drm/ttm: Make sure the mapped tt pages are decrypted when
needed")
Ah yes that was the DMA layer hack for encryption Zack came up with.
In this case please put the functions into ttm_bo_internal.h. It should already
be in the ttm directory for TTM internal stuff.
Then ttm_pool.c will have to start including ttm_bo_internal.h which
feels it is making the component separation worse. Add
ttm_pool_internal.h? Keep in ttm_pool.h?
Regards,
Tvrtko
Going to comment on the other patch as well.
Thanks,
Christian.
Regards,
Tvrtko
