On Thu, Jul 26, 2012 at 12:44:43PM -0600, Peter Grehan wrote: > >Yes, CPUID or LFENCE is enough to fix the failure. > > CPUID causes an unconditional exit in VT-x/SVM so it would be best to > avoid that if possible.
The only place where the use of CPUID in tsc patch seems to be reasonable is at the SMP test, since it is too much burden to check SSE2/Intel/AMD nuances at the boot time, to be redone later anyway. Would it be a serious issue for monitors to get several thousands of CPUID traps at boot only ?
pgp2Id6FuYF3k.pgp
Description: PGP signature
