We once had the following error in stderr when running piglit's spec@!opengl 1.1@copypixels-sync on sandybridge: i965: Failed to submit batchbuffer: Bad address Should I move this bug to mesa?
EFAULT means one of the bo was MADV_DONTNEED. We don't do a strict validation pass, so only fail if we happen to try and page in the object in question. (Less likely possibility is the user pointers passed to the ioctl are bogus.) Sounds like we could do with a validation layer.
(In reply to Chris Wilson from comment #1) > EFAULT means one of the bo was MADV_DONTNEED. We don't do a strict > validation pass, so only fail if we happen to try and page in the object in > question. (Less likely possibility is the user pointers passed to the ioctl > are bogus.) Sounds like we could do with a validation layer. Where do you need the validation, mesa or i915? I am wondering where this bug should reside.
I'd make the validation layer in userspace; part of the ioctl interception scrubbing tools. My gut feeling is that is a userspace bug; certainly easier to diagnose in userspace, and if that rules it out being a mesa bug then to delve back into the kernel.
(In reply to Chris Wilson from comment #3) > I'd make the validation layer in userspace; part of the ioctl interception > scrubbing tools. > > My gut feeling is that is a userspace bug; certainly easier to diagnose in > userspace, and if that rules it out being a mesa bug then to delve back into > the kernel. That makes sense, I am moving it to Mesa for their input.
We've seen this for a while, and can't pinpoint it. *** This bug has been marked as a duplicate of bug 104778 ***
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.