On Tue, Feb 23, 2021 at 3:36 AM Jian Cai <[email protected]> wrote: > This patch adds CONFIG_HARDEN_SLS_ALL that can be used to turn on > -mharden-sls=all, which mitigates the straight-line speculation > vulnerability, speculative execution of the instruction following some > unconditional jumps. Notice -mharden-sls= has other options as below, > and this config turns on the strongest option. > > all: enable all mitigations against Straight Line Speculation that are > implemented. > none: disable all mitigations against Straight Line Speculation. > retbr: enable the mitigation against Straight Line Speculation for RET and BR > instructions. > blr: enable the mitigation against Straight Line Speculation for BLR > instructions.
I heard about compiler protection for this, so nice to see it happening! Would you happen to know if there is any plan to do the same for GCC? I know you folks at Google like LLVM, but if you know let us know. > +config HARDEN_SLS_ALL > + bool "enable SLS vulnerability hardening" I would go in and also edit arch/arm/mm/Kconfig under: config HARDEN_BRANCH_PREDICTOR add select HARDEN_SLS_ALL Because if the user wants hardening for branch prediction in general then the user certainly wants this as well, if available. The help text for that option literally says: "This config option will take CPU-specific actions to harden the branch predictor against aliasing attacks and may rely on specific instruction sequences or control bits being set by the system firmware." Notice this only turns on for CPUs with CPU_SPECTRE defined which makes sense. Also it is default y which fulfils Will's request that it be turned on by default where applicable. Notably it will not be turned on for pre-v7 silicon which would be unhelpful as they don't suffer from these bugs. Reading Kristofs compiler patch here: https://reviews.llvm.org/rG195f44278c4361a4a32377a98a1e3a15203d3647 I take it that for affected CPUs we should also patch all assembly in the kernel containing a RET, BR or BLR with DSB SYS followed by ISB? I suppose we would also need to look for any mov PC, <> code... I guess we can invent a "SB" macro to mimic what Aarch64 is doing so the code is easy to read. (Thinking aloud.) Yours, Linus Walleij
