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

From: Paolo Bonzini
Date: Mon Jun 01 2015 - 05:17:10 EST




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@xxxxxxxxxxxxxxx>
> ---
> 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);
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/