Am Donnerstag, den 28.04.2016, 14:55 +0100 schrieb Russell King - ARM Linux: > On Wed, Apr 27, 2016 at 02:38:18PM +0200, Lucas Stach wrote: > > If the MMU map fails caused by an unaligned SG entry, the unmap path > > is called to undo all already setup SG mappings. When encountering the > > unaligned SG the unmap path hangs the kernel with a BUG(), while the > > error is recoverable by just failing the submit that references the > > faulty object. > > I'm very tempted to NAK this, because I introduced it with good reason: > we should never see anything except a page-aligned number of bytes here. > > Are you seeing this trigger? > I've only seen this once and I don't know what caused it really. Maybe some unrelated corruption. The observation was that the common code in iommu_map() rightfully rejected to map things, as mapping something unaligned to the page size is totally bogus. iommu_unmap will detect this condition in the same way. So without this BUG_ON() the kernel would have been able to recover from the bogus input data (maybe allowing to debug things further), but with the BUG_ON() in place it just dies. Regards, Lucas _______________________________________________ dri-devel mailing list dri-devel@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/dri-devel
