| Message ID | 20190805085350.23309-1-jnwang@suse.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v1] x86/spec-ctrl: Remove EAGER_FPU when PV/HVM is disabled. | expand |
On 05/08/2019 09:54, Jin Nan Wang wrote: > In commit ac3f9a72141a48d40fabfff561d5a7dc0e1b810d: > For one, "no-xen" should not imply "no-eager-fpu", as "eager FPU" mode > is to guard guests, not Xen itself, which is also expressed so by > print_details(). > > So when spec-ctrl=pv=off,hvm=off; EAGER_FPU shouldn't be display in lines: > Current[Wrong]: > (XEN) Support for HVM VMs: EAGER_FPU MD_CLEAR > (XEN) Support for PV VMs: EAGER_FPU MD_CLEAR > > expectation: > (XEN) Support for HVM VMs: MD_CLEAR > (XEN) Support for PV VMs: MD_CLEAR Eager FPU is a performance win across the board. It will become an even bigger performance win when we follow in Linux's footsteps and remove the eager option, and associated logic. I don't see any value in turning it off for speculation related reasons. ~Andrew
On 8/5/19 5:04 PM, Andrew Cooper wrote: > On 05/08/2019 09:54, Jin Nan Wang wrote: >> In commit ac3f9a72141a48d40fabfff561d5a7dc0e1b810d: >> For one, "no-xen" should not imply "no-eager-fpu", as "eager FPU" mode >> is to guard guests, not Xen itself, which is also expressed so by >> print_details(). >> >> So when spec-ctrl=pv=off,hvm=off; EAGER_FPU shouldn't be display in lines: >> Current[Wrong]: >> (XEN) Support for HVM VMs: EAGER_FPU MD_CLEAR >> (XEN) Support for PV VMs: EAGER_FPU MD_CLEAR >> >> expectation: >> (XEN) Support for HVM VMs: MD_CLEAR >> (XEN) Support for PV VMs: MD_CLEAR > Eager FPU is a performance win across the board. It will become an even > bigger performance win when we follow in Linux's footsteps and remove > the eager option, and associated logic. OK. > I don't see any value in turning it off for speculation related reasons. Got it. > ~Andrew >
diff --git a/xen/arch/x86/spec_ctrl.c b/xen/arch/x86/spec_ctrl.c index 468a847598..5f519406fb 100644 --- a/xen/arch/x86/spec_ctrl.c +++ b/xen/arch/x86/spec_ctrl.c @@ -128,12 +128,14 @@ static int __init parse_spec_ctrl(const char *s) opt_msr_sc_pv = val; opt_rsb_pv = val; opt_md_clear_pv = val; + opt_eager_fpu = val; } else if ( (val = parse_boolean("hvm", s, ss)) >= 0 ) { opt_msr_sc_hvm = val; opt_rsb_hvm = val; opt_md_clear_hvm = val; + opt_eager_fpu = val; } else if ( (val = parse_boolean("msr-sc", s, ss)) >= 0 ) {
In commit ac3f9a72141a48d40fabfff561d5a7dc0e1b810d: For one, "no-xen" should not imply "no-eager-fpu", as "eager FPU" mode is to guard guests, not Xen itself, which is also expressed so by print_details(). So when spec-ctrl=pv=off,hvm=off; EAGER_FPU shouldn't be display in lines: Current[Wrong]: (XEN) Support for HVM VMs: EAGER_FPU MD_CLEAR (XEN) Support for PV VMs: EAGER_FPU MD_CLEAR expectation: (XEN) Support for HVM VMs: MD_CLEAR (XEN) Support for PV VMs: MD_CLEAR --- xen/arch/x86/spec_ctrl.c | 2 ++ 1 file changed, 2 insertions(+)