The midgard/bifrost GPUs need to allocate GPU heap memory which is allocated on GPU page faults and not pinned in memory. The vendor driver calls this functionality GROW_ON_GPF. This implementation assumes that BOs allocated with the PANFROST_BO_NOEXEC flag are never mmapped or exported. Both of those may actually work, but I'm unsure if there's some interaction there. It would cause the whole object to be pinned in memory which would defeat the point of this. On faults, we map in 2MB at a time in order to utilize huge pages (if enabled). Currently, once we've mapped pages in, they are only unmapped if the BO is freed. Once we add shrinker support, we can unmap pages with the shrinker.
Showing the changes to include/uapi (1 of 6 files). Show all
Unified · Split
| @@ -85,6 +85,7 @@ struct drm_panfrost_wait_bo { · 84 unmodified lines | |||
| 85 | }; | 85 | }; |
| 86 | 86 | ||
| 87 | #define PANFROST_BO_NOEXEC 1 | 87 | #define PANFROST_BO_NOEXEC 1 |
| 88 | #define PANFROST_BO_HEAP 2 | ||
| 88 | 89 | ||
| 89 | /** | 90 | /** |
| 90 | * struct drm_panfrost_create_bo - ioctl argument for creating Panfrost BOs. | 91 | * struct drm_panfrost_create_bo - ioctl argument for creating Panfrost BOs. |