On Tue, 20 Jul 2021 at 23:04, Jason Ekstrand <jason@xxxxxxxxxxxxxx> wrote:
>
> On Tue, Jul 20, 2021 at 4:35 AM Matthew Auld
> <matthew.william.auld@xxxxxxxxx> wrote:
> >
> > On Thu, 15 Jul 2021 at 23:39, Jason Ekstrand <jason@xxxxxxxxxxxxxx> wrote:
> > >
> > > Instead of hand-rolling the same three calls in each function, pull them
> > > into an i915_gem_object_create_user helper. Apart from re-ordering of
> > > the placements array ENOMEM check, the only functional change here
> > > should be that i915_gem_dumb_create now calls i915_gem_flush_free_objects
> > > which it probably should have been calling all along.
> > >
> > > Signed-off-by: Jason Ekstrand <jason@xxxxxxxxxxxxxx>
> > > ---
> > > drivers/gpu/drm/i915/gem/i915_gem_create.c | 106 +++++++++------------
> > > 1 file changed, 43 insertions(+), 63 deletions(-)
> > >
> > > diff --git a/drivers/gpu/drm/i915/gem/i915_gem_create.c b/drivers/gpu/drm/i915/gem/i915_gem_create.c
> > > index 391c8c4a12172..69bf9ec777642 100644
> > > --- a/drivers/gpu/drm/i915/gem/i915_gem_create.c
> > > +++ b/drivers/gpu/drm/i915/gem/i915_gem_create.c
> > > @@ -11,13 +11,14 @@
> > > #include "i915_trace.h"
> > > #include "i915_user_extensions.h"
> > >
> > > -static u32 object_max_page_size(struct drm_i915_gem_object *obj)
> > > +static u32 object_max_page_size(struct intel_memory_region **placements,
> > > + unsigned int n_placements)
> > > {
> > > u32 max_page_size = 0;
> > > int i;
> > >
> > > - for (i = 0; i < obj->mm.n_placements; i++) {
> > > - struct intel_memory_region *mr = obj->mm.placements[i];
> > > + for (i = 0; i < n_placements; i++) {
> > > + struct intel_memory_region *mr = placements[i];
> > >
> > > GEM_BUG_ON(!is_power_of_2(mr->min_page_size));
> > > max_page_size = max_t(u32, max_page_size, mr->min_page_size);
> > > @@ -81,22 +82,35 @@ static int i915_gem_publish(struct drm_i915_gem_object *obj,
> > > return 0;
> > > }
> > >
> > > -static int
> > > -i915_gem_setup(struct drm_i915_gem_object *obj, u64 size)
> > > +static struct drm_i915_gem_object *
> > > +i915_gem_object_create_user(struct drm_i915_private *i915, u64 size,
> > > + struct intel_memory_region **placements,
> > > + unsigned int n_placements)
> > > {
> > > - struct intel_memory_region *mr = obj->mm.placements[0];
> > > + struct intel_memory_region *mr = placements[0];
> > > + struct drm_i915_gem_object *obj;
> > > unsigned int flags;
> > > int ret;
> > >
> > > - size = round_up(size, object_max_page_size(obj));
> > > + i915_gem_flush_free_objects(i915);
> > > +
> > > + obj = i915_gem_object_alloc();
> > > + if (!obj)
> > > + return ERR_PTR(-ENOMEM);
> > > +
> > > + size = round_up(size, object_max_page_size(placements, n_placements));
> > > if (size == 0)
> > > - return -EINVAL;
> > > + return ERR_PTR(-EINVAL);
> > >
> > > /* For most of the ABI (e.g. mmap) we think in system pages */
> > > GEM_BUG_ON(!IS_ALIGNED(size, PAGE_SIZE));
> > >
> > > if (i915_gem_object_size_2big(size))
> > > - return -E2BIG;
> > > + return ERR_PTR(-E2BIG);
> > > +
> > > + ret = object_set_placements(obj, placements, n_placements);
> > > + if (ret)
> > > + goto object_free;
> >
> > Thinking on this again, it might be way too thorny to expose
> > create_user as-is to other parts of i915, like we do in the last
> > patch. Since the caller will be expected to manually validate the
> > placements, otherwise we might crash and burn in weird ways as new
> > users pop up. i.e it needs the same validation that happens as part of
> > the extension. Also as new extensions arrive, like with PXP, that also
> > has to get bolted onto create_user, which might have its own hidden
> > constraints.
>
> Perhaps. Do you have a suggestion for how to make it available to
> selftests without exposing it to "the rest of i915"? If you want, I
> can make create_user duplicate the placements uniqueness check.
> That's really the only validation currently in the ioctl besides all
> the stuff for making sure that the class/instance provided by the user
> isn't bogus. But if we've got real i915_memory_region pointers, we
> don't need that.
Yeah, I guess the concern here was duplicated placements(that would
change the meaning of n_placements > 1), and then ofc regions not
supported by the device. Also maybe stolen which doesn't have a TTM
backend yet.
If this is just for the selftests, doing what the mman selftests do
with create_region + set_placements would be one option. Otherwise
maybe just add __two_underscores and a big comment, for why you
should be careful when using this?
>
> --Jason
