On Tue, Mar 12, 2024, Rick Edgecombe wrote: > Fix KVM_SET_MEMORY_ATTRIBUTES to not overflow lpage_info array and trigger > KASAN splat, as seen in the private_mem_conversions_test selftest. Ugh, that's embarrassing. > The issue can be observed simply by compiling the kernel with > CONFIG_KASAN_VMALLOC and running the selftest “private_mem_conversions_test”, Ah, less emabarrasing, as KASAN_VMALLOC isn't auto-selected by KASAN=y. > It is a little ambiguous whether the unaligned tail page should be Nit, it's the head page, not the tail page. Strictly speaking, it's probably both (or neither, if you're a half glass empty person), but the buggy code that is processing regions is specifically dealing with what it calls the head page. > expected to have KVM_LPAGE_MIXED_FLAG set. It is not functionally > required, as the unaligned tail pages will already have their > kvm_lpage_info count incremented. The comments imply not setting it on > unaligned head pages is intentional, so fix the callers to skip trying to > set KVM_LPAGE_MIXED_FLAG in this case, and in doing so not call > hugepage_has_attrs(). > Also rename hugepage_has_attrs() to __slot_hugepage_has_attrs() because it > is a delicate function that should not be widely used, and only is valid > for ranges covered by the passed slot. Eh, I vote to drop the rename. It's (a) a local static, (b) guarded by CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES=y, (c) pretty obvious from the @slot param that it works on a single slot, (d) the double underscores suggests there is an outer wrapper with the same name, which there is not, and (e) the rename adds noise to a diff that's destined for stable@. Other than the rename, code looks good. Thanks!
