We currently require that our per-engine reset can be called from any
context, even hardirq, and in the future wish to perform the device
reset without holding struct_mutex (which requires some lockless
shenanigans that demand the lowlevel intel_reset_gpu() be able to be
used in atomic context). Te
On 12/12/2018 13:41, Chris Wilson wrote:
We currently require that our per-engine reset can be called from any
context, even hardirq, and in the future wish to perform the device
reset without holding struct_mutex (which requires some lockless
shenanigans that demand the lowlevel intel_reset_gpu
Quoting Tvrtko Ursulin (2018-12-13 08:28:00)
>
> On 12/12/2018 13:41, Chris Wilson wrote:
> > +static int igt_atomic_reset(void *arg)
> > +{
> > + static const struct atomic_section phases[] = {
> > + { "preempt", __preempt_begin, __preempt_end },
> > + { "softirq", __s