4 - For discrete memory manager, merge enough dg1 to be able to refactor it to
5 TTM. Then land pci ids (just in case that turns up an uapi problem). TTM has
6 improved a lot the past 2 years, there's no reason anymore not to use it.
8 - Come up with a plan what to do with drm/scheduler and how to get there.
10 - Roll out dma_fence critical section annotations.
12 - There's a lot of complexity added past few years to make relocations faster.
13 That doesn't make sense given hw and gpu apis moved away from this model years
15 1. Land a modern pre-bound uapi like VM_BIND
16 2. Any complexity added in this area past few years which can't be justified
17 with VM_BIND using userspace should be removed. Looking at amdgpu dma_resv on
18 the bo and vm, plus some lru locks is all that needed. No complex rcu,
19 refcounts, caching, ... on everything.
20 This is the matching task on the vm side compared to ttm/dma_resv on the
23 - i915_sw_fence seems to be the main structure for the i915-gem dma_fence model.
24 How-to-dma_fence is core and drivers really shouldn't build their own world
25 here, treating everything else as a fixed platform. i915_sw_fence concepts
26 should be moved to dma_fence, drm/scheduler or atomic commit helpers. Or
27 removed if dri-devel consensus is that it's not a good idea. Once that's done
28 maybe even remove it if there's nothing left.
31 - i915_utils.h needs to be moved to the right places.
33 - dma_fence_work should be in drivers/dma-buf
35 - i915_mm.c should be moved to the right places. Some of the helpers also look a
38 https://lore.kernel.org/linux-mm/20210301083320.943079-1-hch@lst.de/
40 - tasklet helpers in i915_gem.h also look a bit misplaced and should
41 probably be moved to tasklet headers.