Home
last modified time | relevance | path

Searched hist:b2cc2a074de75671bbed5e2dda67a9252ef353ea (Results 1 – 1 of 1) sorted by relevance

/linux/arch/x86/kernel/cpu/
H A Dcommon.cdiff b2cc2a074de75671bbed5e2dda67a9252ef353ea Thu Sep 27 03:02:28 CEST 2012 H. Peter Anvin <hpa@linux.intel.com> x86, smep, smap: Make the switching functions one-way

There is no fundamental reason why we should switch SMEP and SMAP on
during early cpu initialization just to switch them off again. Now
with %eflags and %cr4 forced to be initialized to a clean state, we
only need the one-way enable. Also, make the functions inline to make
them (somewhat) harder to abuse.

This does mean that SMEP and SMAP do not get initialized anywhere near
as early. Even using early_param() instead of __setup() doesn't give
us control early enough to do this during the early cpu initialization
phase. This seems reasonable to me, because SMEP and SMAP should not
matter until we have userspace to protect ourselves from, but it does
potentially make it possible for a bug involving a "leak of
permissions to userspace" to get uncaught.

Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>