On Tue Nov 8, 2022 at 12:28 AM AEST, Christophe Leroy wrote: > > > Le 07/11/2022 à 04:31, Rohan McLure a écrit : > > Add Kconfig option for enabling clearing of registers on arrival in an > > interrupt handler. This reduces the speculation influence of registers > > on kernel internals. The option will be consumed by 64-bit systems that > > feature speculation and wish to implement this mitigation. > > > > This patch only introduces the Kconfig option, no actual mitigations. > > If that has to do with speculation, do we need a new Kconfig option ? > Can't we use CONFIG_PPC_BARRIER_NOSPEC for that ?
NOSPEC barrier adds runtime-patchable hardware barrier and that config is a build implementation detail. Also that spec barrier is for bounds checks speculation that is easy to get the kernel to do something like speculatively branch to arbitrary address. Interrupt/syscall register sanitization is more handwavy. It could be a bandaid for cases where the above speculation barrier was missed for exampel. But at some point, at least for syscalls, registers have to contain some values influenced by userspace so if we were paranoid we would have to put barriers before every branch while any registers contained a value from userspace. A security option menu might be a good idea though. There's some other build time options like rop protection that we might want to add. Thanks, Nick
