Re: [PATCH v2] arm64: KVM: Save four instructions in __guest_enter/exit()

[Shanker Donthineni <shankerd@codeaurora.org>]

Hi Marc,


On 08/30/2016 05:54 AM, Marc Zyngier wrote:
> On 30/08/16 10:55, Christoffer Dall wrote:
>> On Mon, Aug 29, 2016 at 10:51:14PM -0500, Shanker Donthineni wrote:
>>> We are doing an unnecessary stack push/pop operation when restoring
>>> the guest registers x0-x18 in __guest_enter(). This patch saves the
>>> two instructions by using x18 as a base register. No need to store
>>> the vcpu context pointer in stack because it is redundant, the same
>>> information is available in tpidr_el2. The function __guest_exit()
>>> prototype is simplified and caller pushes the regs x0-x1 to stack
>>> instead of regs x0-x3.
>>>
>>> Signed-off-by: Shanker Donthineni <shankerd@codeaurora.org>
>> This looks reasonable to me:
>>
>> Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
>>
>> Unless Marc has any insight into this having a negative effect on ARM
>> CPUs, I'll go ahead an merge this.
> I've given it a go on Seattle, and couldn't observe any difference with
> the original code, which is pretty good news!
>
> I have some comments below, though:
>
>> -Christoffer
>>
>>> ---
>>> Changes since v1:
>>>    Incorporated Cristoffer suggestions.
>>>    __guest_exit prototype is changed to 'void __guest_exit(u64 reason,
> struct kvm_vcpu *vcpu)'.
>>>   arch/arm64/kvm/hyp/entry.S     | 101
> +++++++++++++++++++++--------------------
>>>   arch/arm64/kvm/hyp/hyp-entry.S |  11 +++--
>>>   2 files changed, 57 insertions(+), 55 deletions(-)
>>>
>>> diff --git a/arch/arm64/kvm/hyp/entry.S b/arch/arm64/kvm/hyp/entry.S
>>> index ce9e5e5..f70489a 100644
>>> --- a/arch/arm64/kvm/hyp/entry.S
>>> +++ b/arch/arm64/kvm/hyp/entry.S
>>> @@ -55,75 +55,76 @@
>>>    */
>>>   ENTRY(__guest_enter)
>>>   	// x0: vcpu
>>> -	// x1: host/guest context
>>> -	// x2-x18: clobbered by macros
>>> +	// x1: host context
>>> +	// x2-x17: clobbered by macros
>>> +	// x18: guest context
>>>   
>>>   	// Store the host regs
>>>   	save_callee_saved_regs x1
>>>   
>>> -	// Preserve vcpu & host_ctxt for use at exit time
>>> -	stp	x0, x1, [sp, #-16]!
>>> +	// Store the host_ctxt for use at exit time
>>> +	str	x1, [sp, #-16]!
>>>   
>>> -	add	x1, x0, #VCPU_CONTEXT
>>> +	add	x18, x0, #VCPU_CONTEXT
>>>   
>>> -	// Prepare x0-x1 for later restore by pushing them onto the stack
>>> -	ldp	x2, x3, [x1, #CPU_XREG_OFFSET(0)]
>>> -	stp	x2, x3, [sp, #-16]!
>>> +	// Restore guest regs x0-x17
>>> +	ldp	x0, x1,   [x18, #CPU_XREG_OFFSET(0)]
>>> +	ldp	x2, x3,   [x18, #CPU_XREG_OFFSET(2)]
>>> +	ldp	x4, x5,   [x18, #CPU_XREG_OFFSET(4)]
>>> +	ldp	x6, x7,   [x18, #CPU_XREG_OFFSET(6)]
>>> +	ldp	x8, x9,   [x18, #CPU_XREG_OFFSET(8)]
>>> +	ldp	x10, x11, [x18, #CPU_XREG_OFFSET(10)]
>>> +	ldp	x12, x13, [x18, #CPU_XREG_OFFSET(12)]
>>> +	ldp	x14, x15, [x18, #CPU_XREG_OFFSET(14)]
>>> +	ldp	x16, x17, [x18, #CPU_XREG_OFFSET(16)]
>>>   
>>> -	// x2-x18
>>> -	ldp	x2, x3,   [x1, #CPU_XREG_OFFSET(2)]
>>> -	ldp	x4, x5,   [x1, #CPU_XREG_OFFSET(4)]
>>> -	ldp	x6, x7,   [x1, #CPU_XREG_OFFSET(6)]
>>> -	ldp	x8, x9,   [x1, #CPU_XREG_OFFSET(8)]
>>> -	ldp	x10, x11, [x1, #CPU_XREG_OFFSET(10)]
>>> -	ldp	x12, x13, [x1, #CPU_XREG_OFFSET(12)]
>>> -	ldp	x14, x15, [x1, #CPU_XREG_OFFSET(14)]
>>> -	ldp	x16, x17, [x1, #CPU_XREG_OFFSET(16)]
>>> -	ldr	x18,      [x1, #CPU_XREG_OFFSET(18)]
>>> +	// Restore guest regs x19-x29, lr
>>> +	restore_callee_saved_regs x18
>>>   
>>> -	// x19-x29, lr
>>> -	restore_callee_saved_regs x1
>>> -
>>> -	// Last bits of the 64bit state
>>> -	ldp	x0, x1, [sp], #16
>>> +	// Restore guest reg x18
>>> +	ldr	x18,      [x18, #CPU_XREG_OFFSET(18)]
>>>   
>>>   	// Do not touch any register after this!
>>>   	eret
>>>   ENDPROC(__guest_enter)
>>>   
>>> +/*
>>> + * void __guest_exit(u64 exit_reason, struct kvm_vcpu *vcpu);
>>> + */
> I'm not sure this comment makes much sense as it stands. This is not a C
> function by any stretch of the imagination, but the continuation of
> __guest_enter. The calling convention is not the C one at all (see how
> the stack is involved), and caller-saved registers are going to be
> clobbered.

I'll remove this confusing comments.

>>>   ENTRY(__guest_exit)
>>> -	// x0: vcpu
>>> -	// x1: return code
>>> -	// x2-x3: free
>>> -	// x4-x29,lr: vcpu regs
>>> -	// vcpu x0-x3 on the stack
>>> -
>>> -	add	x2, x0, #VCPU_CONTEXT
>>> -
>>> -	stp	x4, x5,   [x2, #CPU_XREG_OFFSET(4)]
>>> -	stp	x6, x7,   [x2, #CPU_XREG_OFFSET(6)]
>>> -	stp	x8, x9,   [x2, #CPU_XREG_OFFSET(8)]
>>> -	stp	x10, x11, [x2, #CPU_XREG_OFFSET(10)]
>>> -	stp	x12, x13, [x2, #CPU_XREG_OFFSET(12)]
>>> -	stp	x14, x15, [x2, #CPU_XREG_OFFSET(14)]
>>> -	stp	x16, x17, [x2, #CPU_XREG_OFFSET(16)]
>>> -	str	x18,      [x2, #CPU_XREG_OFFSET(18)]
>>> -
>>> -	ldp	x6, x7, [sp], #16	// x2, x3
>>> -	ldp	x4, x5, [sp], #16	// x0, x1
>>> -
>>> -	stp	x4, x5, [x2, #CPU_XREG_OFFSET(0)]
>>> -	stp	x6, x7, [x2, #CPU_XREG_OFFSET(2)]
>>> +	// x0: return code
>>> +	// x1: vcpu
>>> +	// x2-x29,lr: vcpu regs
>>> +	// vcpu x0-x1 on the stack
>>> +
>>> +	add	x1, x1, #VCPU_CONTEXT
>>> +
>>> +	// Store the guest regs x2 and x3
>>> +	stp	x2, x3,   [x1, #CPU_XREG_OFFSET(2)]
>>> +
>>> +	// Retrieve the guest regs x0-x1 from the stack
>>> +	ldp	x2, x3, [sp], #16	// x0, x1
>>> +
>>> +	// Store the guest regs x0-x1 and x4-x18
>>> +	stp	x2, x3,   [x1, #CPU_XREG_OFFSET(0)]
>>> +	stp	x4, x5,   [x1, #CPU_XREG_OFFSET(4)]
>>> +	stp	x6, x7,   [x1, #CPU_XREG_OFFSET(6)]
>>> +	stp	x8, x9,   [x1, #CPU_XREG_OFFSET(8)]
>>> +	stp	x10, x11, [x1, #CPU_XREG_OFFSET(10)]
>>> +	stp	x12, x13, [x1, #CPU_XREG_OFFSET(12)]
>>> +	stp	x14, x15, [x1, #CPU_XREG_OFFSET(14)]
>>> +	stp	x16, x17, [x1, #CPU_XREG_OFFSET(16)]
>>> +	str	x18,      [x1, #CPU_XREG_OFFSET(18)]
>>> +
>>> +	// Store the guest regs x19-x29, lr
>>> +	save_callee_saved_regs x1
>>>   
>>> -	save_callee_saved_regs x2
>>> +	// Restore the host_ctxt from the stack
>>> +	ldr	x2, [sp], #16
>>>   
>>> -	// Restore vcpu & host_ctxt from the stack
>>> -	// (preserving return code in x1)
>>> -	ldp	x0, x2, [sp], #16
>>>   	// Now restore the host regs
>>>   	restore_callee_saved_regs x2
>>>   
>>> -	mov	x0, x1
>>>   	ret
>>>   ENDPROC(__guest_exit)
>>>   
>>> diff --git a/arch/arm64/kvm/hyp/hyp-entry.S
> b/arch/arm64/kvm/hyp/hyp-entry.S
>>> index f6d9694..06e8b3b 100644
>>> --- a/arch/arm64/kvm/hyp/hyp-entry.S
>>> +++ b/arch/arm64/kvm/hyp/hyp-entry.S
>>> @@ -121,14 +121,15 @@ el1_trap:
>>>   	cmp	x2, #ESR_ELx_EC_FP_ASIMD
>>>   	b.eq	__fpsimd_guest_restore
>>>   
>>> -	mrs	x0, tpidr_el2
>>> -	mov	x1, #ARM_EXCEPTION_TRAP
>>> +	ldp     x2, x3, [sp], #16
>>> +	mrs	x1, tpidr_el2
>>> +	mov	x0, #ARM_EXCEPTION_TRAP
>>>   	b	__guest_exit
>>>   
>>>   el1_irq:
>>> -	save_x0_to_x3
> So the save_x0_to_x3 macro now only has one single user (and so does
> restore_x0_to_x3).  Should we consider inline it?

Sure, I'll change to inline in v3 patch.

>>> -	mrs	x0, tpidr_el2
>>> -	mov	x1, #ARM_EXCEPTION_IRQ
>>> +	stp     x0, x1, [sp, #-16]!
>>> +	mrs	x1, tpidr_el2
>>> +	mov	x0, #ARM_EXCEPTION_IRQ
>>>   	b	__guest_exit
>>>   
>>>   ENTRY(__hyp_do_panic)
>>> -- 
>>> Qualcomm Datacenter Technologies, Inc. on behalf of the Qualcomm
> Technologies, Inc.
>>> Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a
> Linux Foundation Collaborative Project.
> Thanks,
>
> 	M.

-- 
Shanker Donthineni
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.