Am 19.10.18 um 09:44 schrieb Thomas Zimmermann: > Hi > > Am 19.10.18 um 09:24 schrieb Koenig, Christian: >> Hi Thomas, >> >> I honestly still find that way to complicated compared to what it >> actually does. >> >> Both ttm_mem_global and ttm_bo_global can just be some static variables. >> E.g. the whole handling with drm_global_reference is just superfluous. >> >> Can I just write a patch to clean up the mess we created? > 'Can I?' I don't know :) > > Sure, it's a mess. For example, if a driver inits drm_global_reference() > and is later unloaded, there are dangling function pointers to the > driver's code. > > I do a slow clean-up of the code because TTM-based drivers depend on the > current way of doing things. > > Maybe let me post my full patch set for RFC and discuss this instead. Yeah, the key point is that this is a lot of stuff to review while the end result is we just trow away most of it. I would rather just go ahead and do a few simple patches instead: 1. Use only a single instance of ttm_mem_global in ttm_memory.c and remove all "global" parameters from the functions. 2. Use only a single instance of ttm_bo_global in ttm_bo.c. 3. Remove bdev->global. The *_global_init()/*_global_release() should just increase/decrease a static reference count protected by a static lock. Regards, Christian. > > Best regards > Thomas > > >> Regards, >> Christian. >> >> Am 18.10.18 um 18:27 schrieb Thomas Zimmermann: >>> TTM provides global memory and a global BO that is shared by all >>> TTM-based drivers. The data structures are provided by struct drm_global >>> and its helpers. All TTM-based DRM drivers copy the initialization and >>> clean-up code for the global TTM state from each other; leading to code >>> duplication. >>> >>> The new structure struct ttm_global and its helpers provide a unified >>> implementation. Drivers only have to initialize it and forward the >>> contained BO global object to ttm_bo_device_init(). >>> >>> The amdgpu and radeon drivers are converted to struct ttm_global as >>> part of this patch set. All other TTM-based drivers share exactly the >>> same code patterns and can be converted in the same way. >>> >>> Future directions: I already have patches for converting the remaining >>> TTM drivers to struct ttm_global. These patches can be merged after >>> the structure has become available in upstream. struct ttm_global is >>> implemented on top of struct drm_global. The latter actually maintains >>> TTM state instead of DRM state. Once all drivers have been converted, >>> the code for struct drm_global can be merged into struct ttm_global. >>> >>> (Resending this patch set, as I forgot to CC the mailing lists as first.) >>> >>> Thomas Zimmermann (3): >>> drm/ttm: Provide struct ttm_global for referencing TTM global state >>> drm/amdgpu: Replace TTM initialization/release with ttm_global >>> drm/radeon: Replace TTM initialization/release with ttm_global >>> >>> drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 63 ++-------------- >>> drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 4 +- >>> drivers/gpu/drm/radeon/radeon.h | 4 +- >>> drivers/gpu/drm/radeon/radeon_ttm.c | 40 ++-------- >>> drivers/gpu/drm/ttm/Makefile | 2 +- >>> drivers/gpu/drm/ttm/ttm_global.c | 98 +++++++++++++++++++++++++ >>> include/drm/drm_global.h | 8 ++ >>> include/drm/ttm/ttm_global.h | 79 ++++++++++++++++++++ >>> 8 files changed, 200 insertions(+), 98 deletions(-) >>> create mode 100644 drivers/gpu/drm/ttm/ttm_global.c >>> create mode 100644 include/drm/ttm/ttm_global.h >>> >>> -- >>> 2.19.1 >>> _______________________________________________ dri-devel mailing list dri-devel@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/dri-devel
