[igt-dev] [PATCH i-g-t 03/77] tests/i915/gem_exec_schedule: Convert to intel_ctx_t

[Dixit, Ashutosh]

On Wed, 16 Jun 2021 09:46:02 -0700, Jason Ekstrand wrote:
>
> On Tue, Jun 15, 2021 at 6:03 PM Dixit, Ashutosh wrote:
> > Also, i915_gem_vm_create_ioctl() in i915 has this:
> >
> >         if (!HAS_FULL_PPGTT(i915))
> >                 return -ENODEV;
> >
> > Considering all this, I'm wondering if we are better off just skipping the
> > gem_vm_create() above (and just use the default VM for the new context)?
>
> There isn't a single "default VM" which is the problem, at least not
> with HAS_FULL_PPGTT.  By default, if HAS_FULL_PPGTT, there is one VM
> per context.  We clone to ensure they're all using the same one.  If
> we don't HAS_FULL_PPGTT, then there is a single VM and we don't need
> to do anything in the test.  We do have a gem_uses_full_ppgtt() helper
> in IGT which we can predicate all the VM stuff on.  I'll do that.
>
> > Since in the new API there will no way to clone a VM, correct?
>
> You can't clone directly but you can assign the same one to both which
> is what this does.

Thanks, I follow now. I'll wait for the next version.

> > gem_vm_destroy() is also missing at the end of the function.
>
> Yup.  I'll fix that.