| Message ID | 20250414081131.97374-2-ubizjak@gmail.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [1/2] KVM: x86: Use asm_inline() instead of asm() in kvm_hypercall[0-4]() | expand |
On Mon, Apr 14, 2025, Uros Bizjak wrote: > Micro-optimize vmx_do_interrupt_irqoff() by substituting > MOV %RBP,%RSP; POP %RBP instruction sequence with equivalent > LEAVE instruction. GCC compiler does this by default for > a generic tuning and for all modern processors: Out of curisoity, is LEAVE actually a performance win, or is the benefit essentially just the few code bytes saves? > DEF_TUNE (X86_TUNE_USE_LEAVE, "use_leave", > m_386 | m_CORE_ALL | m_K6_GEODE | m_AMD_MULTIPLE | m_ZHAOXIN > | m_TREMONT | m_CORE_HYBRID | m_CORE_ATOM | m_GENERIC) > > The new code also saves a couple of bytes, from: > > 27: 48 89 ec mov %rbp,%rsp > 2a: 5d pop %rbp > > to: > > 27: c9 leave > > No functional change intended. > > Signed-off-by: Uros Bizjak <ubizjak@gmail.com> > Cc: Sean Christopherson <seanjc@google.com> > Cc: Paolo Bonzini <pbonzini@redhat.com> > Cc: Vitaly Kuznetsov <vkuznets@redhat.com> > Cc: Thomas Gleixner <tglx@linutronix.de> > Cc: Ingo Molnar <mingo@kernel.org> > Cc: Borislav Petkov <bp@alien8.de> > Cc: Dave Hansen <dave.hansen@linux.intel.com> > Cc: "H. Peter Anvin" <hpa@zytor.com> > --- > arch/x86/kvm/vmx/vmenter.S | 3 +-- > 1 file changed, 1 insertion(+), 2 deletions(-) > > diff --git a/arch/x86/kvm/vmx/vmenter.S b/arch/x86/kvm/vmx/vmenter.S > index f6986dee6f8c..0a6cf5bff2aa 100644 > --- a/arch/x86/kvm/vmx/vmenter.S > +++ b/arch/x86/kvm/vmx/vmenter.S > @@ -59,8 +59,7 @@ > * without the explicit restore, thinks the stack is getting walloped. > * Using an unwind hint is problematic due to x86-64's dynamic alignment. > */ > - mov %_ASM_BP, %_ASM_SP > - pop %_ASM_BP > + leave > RET > .endm > > -- > 2.49.0 >
On Tue, Apr 15, 2025 at 3:05 AM Sean Christopherson <seanjc@google.com> wrote: > > On Mon, Apr 14, 2025, Uros Bizjak wrote: > > Micro-optimize vmx_do_interrupt_irqoff() by substituting > > MOV %RBP,%RSP; POP %RBP instruction sequence with equivalent > > LEAVE instruction. GCC compiler does this by default for > > a generic tuning and for all modern processors: > > Out of curisoity, is LEAVE actually a performance win, or is the benefit essentially > just the few code bytes saves? It is hard to say for out-of-order execution cores, especially when the stack engine is thrown to the mix (these two instructions, plus following RET, all update %rsp). The pragmatic solution was to do what the compiler does and use the compiler's choice, based on the tuning below. > > DEF_TUNE (X86_TUNE_USE_LEAVE, "use_leave", > > m_386 | m_CORE_ALL | m_K6_GEODE | m_AMD_MULTIPLE | m_ZHAOXIN > > | m_TREMONT | m_CORE_HYBRID | m_CORE_ATOM | m_GENERIC) The tuning is updated when a new target is introduced to the compiler and is based on various measurements by the processor manufacturer. The above covers the majority of recent processors (plus generic tuning), so I guess we won't fail by following the suit. OTOH, any performance difference will be negligible. > > The new code also saves a couple of bytes, from: > > > > 27: 48 89 ec mov %rbp,%rsp > > 2a: 5d pop %rbp > > > > to: > > > > 27: c9 leave Thanks, Uros.
diff --git a/arch/x86/kvm/vmx/vmenter.S b/arch/x86/kvm/vmx/vmenter.S index f6986dee6f8c..0a6cf5bff2aa 100644 --- a/arch/x86/kvm/vmx/vmenter.S +++ b/arch/x86/kvm/vmx/vmenter.S @@ -59,8 +59,7 @@ * without the explicit restore, thinks the stack is getting walloped. * Using an unwind hint is problematic due to x86-64's dynamic alignment. */ - mov %_ASM_BP, %_ASM_SP - pop %_ASM_BP + leave RET .endm
Micro-optimize vmx_do_interrupt_irqoff() by substituting MOV %RBP,%RSP; POP %RBP instruction sequence with equivalent LEAVE instruction. GCC compiler does this by default for a generic tuning and for all modern processors: DEF_TUNE (X86_TUNE_USE_LEAVE, "use_leave", m_386 | m_CORE_ALL | m_K6_GEODE | m_AMD_MULTIPLE | m_ZHAOXIN | m_TREMONT | m_CORE_HYBRID | m_CORE_ATOM | m_GENERIC) The new code also saves a couple of bytes, from: 27: 48 89 ec mov %rbp,%rsp 2a: 5d pop %rbp to: 27: c9 leave No functional change intended. Signed-off-by: Uros Bizjak <ubizjak@gmail.com> Cc: Sean Christopherson <seanjc@google.com> Cc: Paolo Bonzini <pbonzini@redhat.com> Cc: Vitaly Kuznetsov <vkuznets@redhat.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@kernel.org> Cc: Borislav Petkov <bp@alien8.de> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: "H. Peter Anvin" <hpa@zytor.com> --- arch/x86/kvm/vmx/vmenter.S | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)