On 06/09/2018 08:00, Chris Wilson wrote:
Quoting Tvrtko Ursulin (2018-09-05 15:25:44)
From: Lionel Landwerlin <lionel.g.landwer...@intel.com>
Verifies that the kernel programs slices correctly based by reading
the value of PWR_CLK_STATE register or MI_SET_PREDICATE on platforms
before Cannonlake.
v2: Add subslice tests (Lionel)
Use MI_SET_PREDICATE for further verification when available (Lionel)
v3: Rename to gem_ctx_rpcs (Lionel)
v4: Update kernel API (Lionel)
Add 0 value test (Lionel)
Exercise invalid values (Lionel)
v5: Add perf tests (Lionel)
v6: Add new sysfs entry tests (Lionel)
v7: Test rsvd fields
Update for kernel series changes
v8: Drop test_no_sseu_support() test (Kelvin)
Drop drm_intel_*() apis (Chris)
v9: by Chris:
Drop all do_ioctl/do_ioctl_err()
Use gem_context_[gs]et_param()
Use gem_read() instead of mapping memory
by Lionel:
Test dynamic sseu on/off more
Tvrtko Ursulin:
v10:
* Various style tweaks and refactorings.
* New test coverage.
I didn't notice any testing of:
- param->size
It exists in test_invalid_args.
- feeding garbage into param->value user pointer (always cleared before
use, perhaps just poison instead), along with abusive pointers.
Also in test_invalid_args - but only the null pointer. I can add an
unmapped or read-only one.
E.g., how does the code fare if we pass in an unfaulted GGTT mmap?
Would not fare well. :I It would be best to be able to reject them but
how? We'll hit the same problem in future other patches so to support
this, I think we need to refactor
Regards,
Tvrtko
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
