In kvm_arch_prepare_memory_region(), if no error occurs, a spin_lock()/
spin_unlock() call can be avoided.

Switch to kvm_riscv_gstage_iounmap() that is the same as the current code,
but with a better semantic.
It also embeds the locking logic. So it is avoided if ret == 0.

Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
---
I don't use cross-compiler, so this patch is NOT even compile tested.
---
 arch/riscv/kvm/mmu.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/arch/riscv/kvm/mmu.c b/arch/riscv/kvm/mmu.c
index 3620ecac2fa1..c8834e463763 100644
--- a/arch/riscv/kvm/mmu.c
+++ b/arch/riscv/kvm/mmu.c
@@ -537,10 +537,8 @@ int kvm_arch_prepare_memory_region(struct kvm *kvm,
 	if (change == KVM_MR_FLAGS_ONLY)
 		goto out;
 
-	spin_lock(&kvm->mmu_lock);
 	if (ret)
-		gstage_unmap_range(kvm, base_gpa, size, false);
-	spin_unlock(&kvm->mmu_lock);
+		kvm_riscv_gstage_iounmap(kvm, base_gpa, size);
 
 out:
 	mmap_read_unlock(current->mm);
-- 
2.34.1

On Sun, Nov 13, 2022 at 7:22 PM Christophe JAILLET
<christophe.jaillet@wanadoo.fr> wrote:
>
> In kvm_arch_prepare_memory_region(), if no error occurs, a spin_lock()/
> spin_unlock() call can be avoided.
>
> Switch to kvm_riscv_gstage_iounmap() that is the same as the current code,
> but with a better semantic.
> It also embeds the locking logic. So it is avoided if ret == 0.
>
> Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>

Looks good to me.

Reviewed-by: Anup Patel <anup@brainfault.org>

I have tested this on QEMU virt machine for RV64.

Queued this patch for Linux-6.2

Thanks,
Anup

> ---
> I don't use cross-compiler, so this patch is NOT even compile tested.
> ---
>  arch/riscv/kvm/mmu.c | 4 +---
>  1 file changed, 1 insertion(+), 3 deletions(-)
>
> diff --git a/arch/riscv/kvm/mmu.c b/arch/riscv/kvm/mmu.c
> index 3620ecac2fa1..c8834e463763 100644
> --- a/arch/riscv/kvm/mmu.c
> +++ b/arch/riscv/kvm/mmu.c
> @@ -537,10 +537,8 @@ int kvm_arch_prepare_memory_region(struct kvm *kvm,
>         if (change == KVM_MR_FLAGS_ONLY)
>                 goto out;
>
> -       spin_lock(&kvm->mmu_lock);
>         if (ret)
> -               gstage_unmap_range(kvm, base_gpa, size, false);
> -       spin_unlock(&kvm->mmu_lock);
> +               kvm_riscv_gstage_iounmap(kvm, base_gpa, size);
>
>  out:
>         mmap_read_unlock(current->mm);
> --
> 2.34.1
>