Re: [PATCH 07/15] KVM: MTRR: improve kvm_mtrr_get_guest_memory_type

[Paolo Bonzini <pbonzini@redhat.com>]

On 30/05/2015 12:59, Xiao Guangrong wrote:
>  - kvm_mtrr_get_guest_memory_type() only checks one page in MTRRs so that
>    it's unnecessary to check to see if the range is partially covered in
>    MTRR
> 
>  - optimize the check of overlap memory type and add some comments to explain
>    the precedence
> 
> Signed-off-by: Xiao Guangrong <guangrong.xiao@linux.intel.com>
> ---
>  arch/x86/kvm/mtrr.c | 89 ++++++++++++++++++++++++++---------------------------
>  1 file changed, 44 insertions(+), 45 deletions(-)
> 
> diff --git a/arch/x86/kvm/mtrr.c b/arch/x86/kvm/mtrr.c
> index bc9c6da..d3c06d2 100644
> --- a/arch/x86/kvm/mtrr.c
> +++ b/arch/x86/kvm/mtrr.c
> @@ -231,24 +231,16 @@ int kvm_mtrr_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
>  	return 0;
>  }
>  
> -/*
> - * The function is based on mtrr_type_lookup() in
> - * arch/x86/kernel/cpu/mtrr/generic.c
> - */
> -static int get_mtrr_type(struct kvm_mtrr *mtrr_state,
> -			 u64 start, u64 end)
> +u8 kvm_mtrr_get_guest_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn)
>  {
> -	u64 base, mask;
> -	u8 prev_match, curr_match;
> -	int i, num_var_ranges = KVM_NR_VAR_MTRR;
> +	struct kvm_mtrr *mtrr_state = &vcpu->arch.mtrr_state;
> +	u64 base, mask, start = gfn_to_gpa(gfn);
> +	int i, num_var_ranges = KVM_NR_VAR_MTRR, type_mask, type = -1;

Do not mix initialized and uninitialized variables on the same line
(preexisting, I know, but let's fix it instead of making it worse :)).
Please put each initialized variable on a separate line.

Also please initialize type_mask here (more on this below).

>  
>  	/* MTRR is completely disabled, use UC for all of physical memory. */
>  	if (!mtrr_state->mtrr_enabled)
>  		return MTRR_TYPE_UNCACHABLE;
>  
> -	/* Make end inclusive end, instead of exclusive */
> -	end--;
> -
>  	/* Look in fixed ranges. Just return the type as per start */
>  	if (mtrr_state->fixed_mtrr_enabled && (start < 0x100000)) {
>  		int idx;
> @@ -273,9 +265,9 @@ static int get_mtrr_type(struct kvm_mtrr *mtrr_state,
>  	 * Look of multiple ranges matching this address and pick type
>  	 * as per MTRR precedence
>  	 */
> -	prev_match = 0xFF;
> +	type_mask = (1 << MTRR_TYPE_WRBACK) | (1 << MTRR_TYPE_WRTHROUGH);
>  	for (i = 0; i < num_var_ranges; ++i) {
> -		unsigned short start_state, end_state;
> +		int curr_type;
>  
>  		if (!(mtrr_state->var_ranges[i].mask & (1 << 11)))
>  			continue;
> @@ -283,50 +275,57 @@ static int get_mtrr_type(struct kvm_mtrr *mtrr_state,
>  		base = mtrr_state->var_ranges[i].base & PAGE_MASK;
>  		mask = mtrr_state->var_ranges[i].mask & PAGE_MASK;
>  
> -		start_state = ((start & mask) == (base & mask));
> -		end_state = ((end & mask) == (base & mask));
> -		if (start_state != end_state)
> -			return 0xFE;
> -
>  		if ((start & mask) != (base & mask))
>  			continue;
>  
> -		curr_match = mtrr_state->var_ranges[i].base & 0xff;
> -		if (prev_match == 0xFF) {
> -			prev_match = curr_match;
> +		/*
> +		 * Please refer to Intel SDM Volume 3: 11.11.4.1 MTRR
> +		 * Precedences.
> +		 */
> +
> +		curr_type = mtrr_state->var_ranges[i].base & 0xff;
> +		if (type == -1) {
> +			type = curr_type;
>  			continue;
>  		}
>  
> -		if (prev_match == MTRR_TYPE_UNCACHABLE ||
> -		    curr_match == MTRR_TYPE_UNCACHABLE)
> +		/*
> +		 * If two or more variable memory ranges match and the
> +		 * memory types are identical, then that memory type is
> +		 * used.
> +		 */
> +		if (type == curr_type)
> +			continue;
> +
> +		/*
> +		 * If two or more variable memory ranges match and one of
> +		 * the memory types is UC, the UC memory type used.
> +		 */
> +		if (curr_type == MTRR_TYPE_UNCACHABLE)
>  			return MTRR_TYPE_UNCACHABLE;
>  
> -		if ((prev_match == MTRR_TYPE_WRBACK &&
> -		     curr_match == MTRR_TYPE_WRTHROUGH) ||
> -		    (prev_match == MTRR_TYPE_WRTHROUGH &&
> -		     curr_match == MTRR_TYPE_WRBACK)) {
> -			prev_match = MTRR_TYPE_WRTHROUGH;
> -			curr_match = MTRR_TYPE_WRTHROUGH;
> +		/*
> +		 * If two or more variable memory ranges match and the
> +		 * memory types are WT and WB, the WT memory type is used.
> +		 */
> +		if (((1 << type) & type_mask) &&
> +		      ((1 << curr_type) & type_mask)) {

Please inline definition of type_mask in the "if", or rename it to
"wt_wb_mask" and make it const.  Or another suggestion below...

> +			type = MTRR_TYPE_WRTHROUGH;
> +			continue;
>  		}
>  
> -		if (prev_match != curr_match)
> -			return MTRR_TYPE_UNCACHABLE;
> +		/*
> +		 * For overlaps not defined by the above rules, processor
> +		 * behavior is undefined.
> +		 */

Perhaps just use type = MIN(type, curr_type), which also happens to get
WT vs. WB right?  You can also add a

	BUILD_BUG_ON(MTRR_TYPE_WRTHROUGH > MTRR_TYPE_WRBACK);

to ensure that the WT vs. WB precedence is correct.

Paolo

> +
> +		 /* We use WB for this undefined behavior. :( */
> +		 return MTRR_TYPE_WRBACK;
>  	}
>  
> -	if (prev_match != 0xFF)
> -		return prev_match;
> +	if (type != -1)
> +		return type;
>  
>  	return mtrr_state->def_type;
>  }
> -
> -u8 kvm_mtrr_get_guest_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn)
> -{
> -	u8 mtrr;
> -
> -	mtrr = get_mtrr_type(&vcpu->arch.mtrr_state, gfn << PAGE_SHIFT,
> -			     (gfn << PAGE_SHIFT) + PAGE_SIZE);
> -	if (mtrr == 0xfe || mtrr == 0xff)
> -		mtrr = MTRR_TYPE_WRBACK;
> -	return mtrr;
> -}
>  EXPORT_SYMBOL_GPL(kvm_mtrr_get_guest_memory_type);
>