diff options
Diffstat (limited to 'drivers/gpu/drm/qxl/TODO')
-rw-r--r-- | drivers/gpu/drm/qxl/TODO | 42 |
1 files changed, 42 insertions, 0 deletions
diff --git a/drivers/gpu/drm/qxl/TODO b/drivers/gpu/drm/qxl/TODO new file mode 100644 index 00000000000..f29c1e9ab5b --- /dev/null +++ b/drivers/gpu/drm/qxl/TODO @@ -0,0 +1,42 @@ +airlied: + drop unmap ioctl, userspace can use munmap + reorder qxl_drm_command, uint64_t together. + implement our own gamma, used for fades. + no need to print a kernel git version, we have that in dmesg anyawys + +Allocation of on device memory in smaller chunks then PAGE_SIZE. +For two users: + internal: can be done with preallocated lists, or on the fly allocating of chunks of them + - specialize for each message. + ioctl (xf86-video-qxl): can again do our own memory allocator on top of pages, but now we don't + know the sizes in advance. + can solve both with mspace, adopted to be on top of TTM. + +qxl_image: allocate chunks. + we can do this without memory allocator + +remove the garbage qxl_io_log +use QXL_INFO / QXL_DEBUG macros that check a debugfs settable variable to toggle qxl_io_log. + +cleanup + +running checkpatch.pl + +Dynamic output number (QXL_NUM_OUTPUTS define must go) + +module parameter log in [0,2] => qxl_log_level + +coalesce updates (for fb mainly - let userspace do it for itself) + +qxl_image hashing: either implement (good for modesetting) or ditch (xf86-video-qxl +does it's own hashing, we don't see/don't check for the created images) + +==== +Misc +==== + +// TODO: why have a seperate mmap and alloc ioctl? also, why not just unmmap on the last +// decref, i.e. when destroyed? is there a huge performance gain to unmapping stuff before +// destroying? and can anyone use an allocated but not mapped memory from userspace?? + +qxlfb_create_pinned_object - 1024x768 limit is actually used? |