Message ID | 20200210114206.17115-1-andrew.cooper3@citrix.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | x86/svm: Reduce vmentry latency | expand |
On Mon, Feb 10, 2020 at 11:42:06AM +0000, Andrew Cooper wrote: > Writing to the stack pointer in the middle of a line of pop operations is > specifically recommended against by the optimisation guide, and is a technique > used by Speculative Load Hardening to combat SpectreRSB. > > In practice, it causes all further stack-relative accesses to block until the > write to the stack pointer retires, so the stack engine can get back in sync. > > Pop into any dead register to discard %rax's value without clobbering the > stack engine. Smaller compiled code, and runs faster. > > Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> Thanks.
On 10.02.2020 13:09, Roger Pau Monné wrote: > On Mon, Feb 10, 2020 at 11:42:06AM +0000, Andrew Cooper wrote: >> Writing to the stack pointer in the middle of a line of pop operations is >> specifically recommended against by the optimisation guide, and is a technique >> used by Speculative Load Hardening to combat SpectreRSB. >> >> In practice, it causes all further stack-relative accesses to block until the >> write to the stack pointer retires, so the stack engine can get back in sync. >> >> Pop into any dead register to discard %rax's value without clobbering the >> stack engine. Smaller compiled code, and runs faster. >> >> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> > > Reviewed-by: Roger Pau Monné <roger.pau@citrix.com> Acked-by: Jan Beulich <jbeulich@suse.com>
diff --git a/xen/arch/x86/hvm/svm/entry.S b/xen/arch/x86/hvm/svm/entry.S index e954d8e021..1d2df08e89 100644 --- a/xen/arch/x86/hvm/svm/entry.S +++ b/xen/arch/x86/hvm/svm/entry.S @@ -76,7 +76,7 @@ __UNLIKELY_END(nsvm_hap) pop %r10 pop %r9 pop %r8 - add $8,%rsp /* Skip %rax: restored by VMRUN. */ + pop %rcx /* Skip %rax: restored by VMRUN. */ pop %rcx pop %rdx pop %rsi
Writing to the stack pointer in the middle of a line of pop operations is specifically recommended against by the optimisation guide, and is a technique used by Speculative Load Hardening to combat SpectreRSB. In practice, it causes all further stack-relative accesses to block until the write to the stack pointer retires, so the stack engine can get back in sync. Pop into any dead register to discard %rax's value without clobbering the stack engine. Smaller compiled code, and runs faster. Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> --- CC: Jan Beulich <JBeulich@suse.com> CC: Wei Liu <wl@xen.org> CC: Roger Pau Monné <roger.pau@citrix.com> In a small test where I wired ICEBP to tighly re-enter the guest, this dropped the guests perviced time for ICEBP (as close to one vmexit and entry as I could realistically manage) by 20 ticks. Sadly, that also seems to be the granuarlity of measurement. The modal measurement (accounting for 80% of samples) was 1200 ticks, and reduced to 1180 with just this change in place. --- xen/arch/x86/hvm/svm/entry.S | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)