Quoting Tvrtko Ursulin (2018-01-19 10:29:12) > > On 19/01/2018 10:09, Chris Wilson wrote: > > Quoting Tvrtko Ursulin (2018-01-19 10:00:05) > >> From: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx> > >> > >> Setting up the workaround batch buffers can fail either due programming > >> errors which will be caught in development, or by the inability to > >> allocate a 4k object and pin it in GGTT at runtime. > >> > >> Since this is highly unlikely, and it is not deterministic to allow driver > >> operation to continue with unknown status of workarounds, it is better to > >> fail engine initialization explicitly under those circumstances. > > > > Not entirely. Failing the driver load leaves the system without a > > display / console. Disabling GPU execution is one response, but that > > is likely to happen if the w/a requirement was severe enough. > > > > We are not expecting to see an -EIO at this point in the init sequence > > and all other errors abort the driver load. > > Fair enough, I did not think about deeper consequences but only assumed > we would run without one engine. Assumptions assumptions! It's not a bad idea :) The next time we give the init a spring clean we may try that (minimising the impact of any specific failure). -Chris _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
