start of the GPU reset, which is the last callsite to unlock cp.mutex, ie <radeon_fence_wait->radeon_gpu_reset->radeon_resume->rv770_resume->rv770= _startup->r600_cp_resume->radeon_ring_test->r600_ring_test->radeon_ring_unl= ock_commit> before the hang at <radeon_fence_wait->radeon_gpu_reset->->drm_helper_resume_force_mode->drm_c= rtc_helper_set_mode->mode_fixup->atombios_crtc_mode_fixup->radeon_pm_comput= e_clocks->radeon_pm_set_clocks>.
Looking through all the calls between these two spots, I just can't see anything that _could_ have locked cp.mutex, unless I'm misreading the disassembly of mutex_lock. Any thoughts? Dan --=20 Daniel J Blueman