On Tue, Dec 09, 2014 at 06:04:33PM +0200, Mika Kuoppala wrote: > This is a similar to 'i915_gpu_state' but more dangerous > as it doesn't try to lock nor idle the gpu before grabbing > the state. But the obvious advantages are that one can > inspect the current running gpu state and also get a state > dump even if mutex is hold. > > As we don't want a unsuspecting user to trip into this by > accident, there needs to be write into the debugfs node > before any unsafe capturing can take place. It shouldn't be called unsafe. It is precisely the mode in which the regular error capture runs and so the kernel is required to be robust. So just document that i915_gpu_state runs asynchronously to any client and the gpu, and caveat emptor. -Chris -- Chris Wilson, Intel Open Source Technology Centre _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/intel-gfx
