What type of offset is provided by nvidia_get_gem_map_offset?

[YusufKhan-gamedev]

When going through the uapi headers I saw the drm_nvidia_gem_map_offset_params definition:

struct drm_nvidia_gem_map_offset_params {
    uint32_t handle;             /* IN Handle to gem object */
    uint32_t __pad;

    uint64_t offset;             /* OUT Fake offset */
};

What does "fake" mean in this context?

[aaronp24]

For a number of reasons, it's not feasible to map memory pages from a drm 'gem' object into a process's virtual memory space while inside of an ioctl handler. So instead, applications have to use the mmap syscall to actually map pages. Applications choose which gem object to map by using an ioctl, and then pass the returned "fake offset" value as the off parameter to the mmap syscall. The drm driver then uses this offset value to look up the gem object to map.

The core drm code in the kernel does something similar. See drm_gem_create_mmap_offset_size in the kernel.

[YusufKhan-gamedev]

analogous to nouveau_bo_priv's map_handle member