[Intel-gfx] [igt-dev] [RFC PATCH 2/4] lib/i915/intel_memory_region: Add lib to manage memory regions

[Joonas Lahtinen]

+ Abdiel/intel-gfx

Quoting Joonas Lahtinen (2019-08-14 15:46:01)
> Quoting Chris Wilson (2019-08-14 13:31:10)
> > Quoting Lukasz Kalamarz (2019-08-14 11:21:38)
> > > +/**
> > > + *  gem_get_page_size:
> > > + *  @fd: open i915 drm file descriptor
> > > + *  @mem_region_type: used memory_region type
> > > + *
> > > + *  With introduction of LMEM we observe different page sizes for those two
> > > + *  memory regions. Without this helper function we may be prone to forget
> > > + *  about setting proper page size.
> > > + */
> > > +uint32_t gem_get_batch_size(int fd, uint8_t mem_region_type)
> > > +{
> > > +       return (mem_region_type == LOCAL_I915_DEVICE_MEMORY) ? 65536 : 4096;
> > 
> > You have to be kidding me. This, this constitutes a forward thinking uAPI?
> 
> We should be just fine requesting the minimum BO size we need, letting the KMD
> round the sizes up and reading back the created BO size.
> 
> (Now that the regression has been fixed, too :) )
> 
> So the code logic needs to be updated to follow that.

On a second thought we may be better off rounding the backing storage
size up transparently.

I guess the prime question is why would the userspace/IGT care for
actual backing storage size?

Abdiel/Matt, how painful would it be to round the backing storage size
up, irrespective of the BO size?

Regards, Joonas