Re: [PATCH] KVM: PPC: paravirt: Avoid patching paravirt template

[Scott Wood <scottwood@freescale.com>]

On 11/23/2011 04:05 AM, Liu Yu wrote:
> Currently we patch the whole code include paravirt template code.
> This doesn't lead into issue for now,
> but it makes some previlege instructions branch to paravirt code twice.

Ouch.  This will do bad things to the scratch area.

> Signed-off-by: Liu Yu <yu.liu@freescale.com>
> ---
>  arch/powerpc/kernel/kvm.c |   17 +++++++++++++++++
>  1 files changed, 17 insertions(+), 0 deletions(-)
> 
> diff --git a/arch/powerpc/kernel/kvm.c b/arch/powerpc/kernel/kvm.c
> index 3953fbd..485748c 100644
> --- a/arch/powerpc/kernel/kvm.c
> +++ b/arch/powerpc/kernel/kvm.c
> @@ -171,6 +171,10 @@ static void kvm_patch_ins_mtmsrd(u32 *inst, u32 rt)
>  	if (!p)
>  		return;
>  
> +	if ((inst > kvm_emulate_mtmsrd) &&
> +	    (inst < kvm_emulate_mtmsrd + kvm_emulate_mtmsrd_len))
> +		return;
> +
>  	/* Find out where we are and put everything there */
>  	distance_start = (ulong)p - (ulong)inst;
>  	next_inst = ((ulong)inst + 4);
> @@ -220,6 +224,11 @@ static void kvm_patch_ins_mtmsr(u32 *inst, u32 rt)
>  	int distance_end;
>  	ulong next_inst;
>  
> +	if ((inst > kvm_emulate_mtmsr) &&
> +	    (inst < kvm_emulate_mtmsr + kvm_emulate_mtmsr_len))
> +		return;
> +
> +
>  	p = kvm_alloc(kvm_emulate_mtmsr_len * 4);
>  	if (!p)
>  		return;
> @@ -281,6 +290,10 @@ static void kvm_patch_ins_wrtee(u32 *inst, u32 rt, int imm_one)
>  	int distance_end;
>  	ulong next_inst;
>  
> +	if ((inst > kvm_emulate_wrtee) &&
> +	    (inst < kvm_emulate_wrtee + kvm_emulate_wrtee_len))
> +		return;
> +
>  	p = kvm_alloc(kvm_emulate_wrtee_len * 4);
>  	if (!p)
>  		return;
> @@ -338,6 +351,10 @@ static void kvm_patch_ins_wrteei_0(u32 *inst)
>  	int distance_end;
>  	ulong next_inst;
>  
> +	if ((inst > kvm_emulate_wrteei_0) &&
> +	    (inst < kvm_emulate_wrteei_0 + kvm_emulate_wrteei_0_len))
> +		return;
> +
>  	p = kvm_alloc(kvm_emulate_wrteei_0_len * 4);
>  	if (!p)
>  		return;

Instead of excluding only the template code for the instruction type
being patched, we should exclude the entire template region in the main
loop.

-Scott