[v1] x86/hvm: Replace do_sched_op calls with their underlying logic

[PATCH] x86/hvm: Replace do_sched_op calls with their underlying logic

Posted by Teddy Astie 3 months, 1 week ago

do_sched_op(SCHEDOP_yield) just calls vcpu_yield(). Remove the indirection
through the hypercall handler and use the function directly.

Perform the same for SCHEDOP_block.

Not a functional change.

Signed-off-by: Teddy Astie <teddy.astie@vates.tech>
---
 xen/arch/x86/hvm/hvm.c               | 3 ++-
 xen/arch/x86/hvm/svm/svm.c           | 2 +-
 xen/arch/x86/hvm/viridian/viridian.c | 2 +-
 xen/arch/x86/hvm/vmx/vmx.c           | 2 +-
 4 files changed, 5 insertions(+), 4 deletions(-)

diff --git a/xen/arch/x86/hvm/hvm.c b/xen/arch/x86/hvm/hvm.c
index 4cb2e13046..5c3e9ad72e 100644
--- a/xen/arch/x86/hvm/hvm.c
+++ b/xen/arch/x86/hvm/hvm.c
@@ -1736,7 +1736,8 @@ void hvm_hlt(unsigned int eflags)
     if ( unlikely(!(eflags & X86_EFLAGS_IF)) )
         return hvm_vcpu_down(curr);
 
-    do_sched_op(SCHEDOP_block, guest_handle_from_ptr(NULL, void));
+    local_event_delivery_enable();
+    vcpu_block();
 
     TRACE(TRC_HVM_HLT, /* pending = */ vcpu_runnable(curr));
 }
diff --git a/xen/arch/x86/hvm/svm/svm.c b/xen/arch/x86/hvm/svm/svm.c
index e33a38c1e4..b0bcd4b1e7 100644
--- a/xen/arch/x86/hvm/svm/svm.c
+++ b/xen/arch/x86/hvm/svm/svm.c
@@ -2159,7 +2159,7 @@ static void svm_vmexit_do_pause(struct cpu_user_regs *regs)
      * Do something useful, like reschedule the guest
      */
     perfc_incr(pauseloop_exits);
-    do_sched_op(SCHEDOP_yield, guest_handle_from_ptr(NULL, void));
+    vcpu_yield();
 }
 
 static void
diff --git a/xen/arch/x86/hvm/viridian/viridian.c b/xen/arch/x86/hvm/viridian/viridian.c
index 33d54e587e..7ea6c90168 100644
--- a/xen/arch/x86/hvm/viridian/viridian.c
+++ b/xen/arch/x86/hvm/viridian/viridian.c
@@ -959,7 +959,7 @@ int viridian_hypercall(struct cpu_user_regs *regs)
         /*
          * See section 14.5.1 of the specification.
          */
-        do_sched_op(SCHEDOP_yield, guest_handle_from_ptr(NULL, void));
+        vcpu_yield();
         break;
 
     case HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE:
diff --git a/xen/arch/x86/hvm/vmx/vmx.c b/xen/arch/x86/hvm/vmx/vmx.c
index d8879c304e..1b9fbc4f4e 100644
--- a/xen/arch/x86/hvm/vmx/vmx.c
+++ b/xen/arch/x86/hvm/vmx/vmx.c
@@ -4768,7 +4768,7 @@ void asmlinkage vmx_vmexit_handler(struct cpu_user_regs *regs)
 
     case EXIT_REASON_PAUSE_INSTRUCTION:
         perfc_incr(pauseloop_exits);
-        do_sched_op(SCHEDOP_yield, guest_handle_from_ptr(NULL, void));
+        vcpu_yield();
         break;
 
     case EXIT_REASON_XSETBV:
-- 
2.50.1



Teddy Astie | Vates XCP-ng Developer

XCP-ng & Xen Orchestra - Vates solutions

web: https://vates.tech

Re: [PATCH] x86/hvm: Replace do_sched_op calls with their underlying logic

Posted by Jason Andryuk 3 months, 1 week ago

On 2025-07-22 14:07, Teddy Astie wrote:
> do_sched_op(SCHEDOP_yield) just calls vcpu_yield(). Remove the indirection
> through the hypercall handler and use the function directly.
> 
> Perform the same for SCHEDOP_block.
> 
> Not a functional change.
> 
> Signed-off-by: Teddy Astie <teddy.astie@vates.tech>
> ---
>   xen/arch/x86/hvm/hvm.c               | 3 ++-
>   xen/arch/x86/hvm/svm/svm.c           | 2 +-
>   xen/arch/x86/hvm/viridian/viridian.c | 2 +-
>   xen/arch/x86/hvm/vmx/vmx.c           | 2 +-
>   4 files changed, 5 insertions(+), 4 deletions(-)
> 
> diff --git a/xen/arch/x86/hvm/hvm.c b/xen/arch/x86/hvm/hvm.c
> index 4cb2e13046..5c3e9ad72e 100644
> --- a/xen/arch/x86/hvm/hvm.c
> +++ b/xen/arch/x86/hvm/hvm.c
> @@ -1736,7 +1736,8 @@ void hvm_hlt(unsigned int eflags)
>       if ( unlikely(!(eflags & X86_EFLAGS_IF)) )
>           return hvm_vcpu_down(curr);
>   
> -    do_sched_op(SCHEDOP_block, guest_handle_from_ptr(NULL, void));
> +    local_event_delivery_enable();
> +    vcpu_block();

I think it would be better to export and call 
vcpu_block_enable_events().  This ensures they stay in-sync.

The vcpu_vield() conversion looks good to me.

Regards,
Jason

>   
>       TRACE(TRC_HVM_HLT, /* pending = */ vcpu_runnable(curr));
>   }

Re: [PATCH] x86/hvm: Replace do_sched_op calls with their underlying logic

Posted by Andrew Cooper 3 months, 1 week ago

On 22/07/2025 1:05 pm, Jason Andryuk wrote:
> On 2025-07-22 14:07, Teddy Astie wrote:
>> do_sched_op(SCHEDOP_yield) just calls vcpu_yield(). Remove the
>> indirection
>> through the hypercall handler and use the function directly.
>>
>> Perform the same for SCHEDOP_block.
>>
>> Not a functional change.
>>
>> Signed-off-by: Teddy Astie <teddy.astie@vates.tech>
>> ---
>>   xen/arch/x86/hvm/hvm.c               | 3 ++-
>>   xen/arch/x86/hvm/svm/svm.c           | 2 +-
>>   xen/arch/x86/hvm/viridian/viridian.c | 2 +-
>>   xen/arch/x86/hvm/vmx/vmx.c           | 2 +-
>>   4 files changed, 5 insertions(+), 4 deletions(-)
>>
>> diff --git a/xen/arch/x86/hvm/hvm.c b/xen/arch/x86/hvm/hvm.c
>> index 4cb2e13046..5c3e9ad72e 100644
>> --- a/xen/arch/x86/hvm/hvm.c
>> +++ b/xen/arch/x86/hvm/hvm.c
>> @@ -1736,7 +1736,8 @@ void hvm_hlt(unsigned int eflags)
>>       if ( unlikely(!(eflags & X86_EFLAGS_IF)) )
>>           return hvm_vcpu_down(curr);
>>   -    do_sched_op(SCHEDOP_block, guest_handle_from_ptr(NULL, void));
>> +    local_event_delivery_enable();
>> +    vcpu_block();
>
> I think it would be better to export and call
> vcpu_block_enable_events().  This ensures they stay in-sync.
>
> The vcpu_vield() conversion looks good to me.

Agreed on both points.

~Andrew