drm/msm: Add global_faults field in shmem

Instead of requiring synchronous a round trip to the host to check if the GPU has hung, stash the fault count in the shmem section shared with the guest. It is updated before notifying the guest of a fence that has been signaled so that the guest doesn't observe the fault count increasing after it has observed the fence signaling.

This will be useful for vulkan since the core vk runtime checks device status before every submit.

@digetx @pepp if you aren't already doing something like this, it would probably be useful for anv/radv as well