On Mon, Nov 25, 2024 at 03:18:57PM +0000, Mark Rutland wrote: > No; I meant that the test could use lm_alias() on the test vectors > before passing those to sg_set_buf(), when the test code knows by > construction that those vectors happen to be part of the kernel image. > That'll work on all architectures. Herbert doesn't want callers of the sign/verify API to do the mapping: https://lore.kernel.org/linux-crypto/Z0A2W1FTTPt9PeI5@xxxxxxxxxxxxxxxxxxx/ > That said, looking at the code it appears that testmgr.c can be built as > a module, so the test vectors could be module/vmalloc addresses rather > than virt/linear or image addresses. Given that, I don't think the > changes suggested here are sufficient, as module addresses should still > be rejected. Ah, I hadn't considered the modular case. Good point! Thanks for the explanation and for taking a look! Lukas
