On Tue, Dec 12, 2023 at 10:55:13AM +0000, Will Deacon wrote: > As discussed off-list, the vague concern was if kernel_neon_begin() is > nested off the back of a user fault. The BUG_ON() should fire in that case, > so we're all good. I think we should really document all these rules, and Samuel's series to add the portable FPU API would be the right place for it. I'd also really love to see objtool support for enforcing the various rules where possible.
