Re: [PATCH v2 04/16] KVM: arm64: Optimize host memory aborts

[Marc Zyngier]

On Mon, 26 Jul 2021 10:28:53 +0100,
Quentin Perret <qperret@xxxxxxxxxx> wrote:
> 
> The kvm_pgtable_stage2_find_range() function is used in the host memory
> abort path to try and look for the largest block mapping that can be
> used to map the faulting address. In order to do so, the function
> currently walks the stage-2 page-table and looks for existing
> incompatible mappings within the range of the largest possible block.
> If incompatible mappings are found, it tries the same procedure again,
> but using a smaller block range, and repeats until a matching range is
> found (potentially up to page granularity). While this approach has
> benefits (mostly in the fact that it proactively coalesces host stage-2
> mappings), it can be slow if the ranges are fragmented, and it isn't
> optimized to deal with CPUs faulting on the same IPA as all of them will
> do all the work every time.
> 
> To avoid these issues, remove kvm_pgtable_stage2_find_range(), and walk
> the page-table only once in the host_mem_abort() path to find the
> closest leaf to the input address. With this, use the corresponding
> range if it is invalid and not owned by another entity. If a valid leaf
> is found, return -EAGAIN similar to what is done in the
> kvm_pgtable_stage2_map() path to optimize concurrent faults.
> 
> Signed-off-by: Quentin Perret <qperret@xxxxxxxxxx>
> ---
>  arch/arm64/include/asm/kvm_pgtable.h  | 30 -----------
>  arch/arm64/kvm/hyp/nvhe/mem_protect.c | 39 +++++++++++++-
>  arch/arm64/kvm/hyp/pgtable.c          | 74 ---------------------------
>  3 files changed, 38 insertions(+), 105 deletions(-)
> 
> diff --git a/arch/arm64/include/asm/kvm_pgtable.h b/arch/arm64/include/asm/kvm_pgtable.h
> index 5a7a13bbd4a1..cec76a49f521 100644
> --- a/arch/arm64/include/asm/kvm_pgtable.h
> +++ b/arch/arm64/include/asm/kvm_pgtable.h
> @@ -141,16 +141,6 @@ enum kvm_pgtable_prot {
>  #define PAGE_HYP_RO		(KVM_PGTABLE_PROT_R)
>  #define PAGE_HYP_DEVICE		(PAGE_HYP | KVM_PGTABLE_PROT_DEVICE)
>  
> -/**
> - * struct kvm_mem_range - Range of Intermediate Physical Addresses
> - * @start:	Start of the range.
> - * @end:	End of the range.
> - */
> -struct kvm_mem_range {
> -	u64 start;
> -	u64 end;
> -};
> -
>  /**
>   * enum kvm_pgtable_walk_flags - Flags to control a depth-first page-table walk.
>   * @KVM_PGTABLE_WALK_LEAF:		Visit leaf entries, including invalid
> @@ -477,24 +467,4 @@ int kvm_pgtable_walk(struct kvm_pgtable *pgt, u64 addr, u64 size,
>   */
>  int kvm_pgtable_get_leaf(struct kvm_pgtable *pgt, u64 addr,
>  			 kvm_pte_t *ptep, u32 *level);
> -
> -/**
> - * kvm_pgtable_stage2_find_range() - Find a range of Intermediate Physical
> - *				     Addresses with compatible permission
> - *				     attributes.
> - * @pgt:	Page-table structure initialised by kvm_pgtable_stage2_init*().
> - * @addr:	Address that must be covered by the range.
> - * @prot:	Protection attributes that the range must be compatible with.
> - * @range:	Range structure used to limit the search space at call time and
> - *		that will hold the result.
> - *
> - * The offset of @addr within a page is ignored. An IPA is compatible with @prot
> - * iff its corresponding stage-2 page-table entry has default ownership and, if
> - * valid, is mapped with protection attributes identical to @prot.
> - *
> - * Return: 0 on success, negative error code on failure.
> - */
> -int kvm_pgtable_stage2_find_range(struct kvm_pgtable *pgt, u64 addr,
> -				  enum kvm_pgtable_prot prot,
> -				  struct kvm_mem_range *range);
>  #endif	/* __ARM64_KVM_PGTABLE_H__ */
> diff --git a/arch/arm64/kvm/hyp/nvhe/mem_protect.c b/arch/arm64/kvm/hyp/nvhe/mem_protect.c
> index 871149246f5f..01700a908bb7 100644
> --- a/arch/arm64/kvm/hyp/nvhe/mem_protect.c
> +++ b/arch/arm64/kvm/hyp/nvhe/mem_protect.c
> @@ -159,6 +159,11 @@ static int host_stage2_unmap_dev_all(void)
>  	return kvm_pgtable_stage2_unmap(pgt, addr, BIT(pgt->ia_bits) - addr);
>  }
>  
> +struct kvm_mem_range {
> +	u64 start;
> +	u64 end;
> +};
> +
>  static bool find_mem_range(phys_addr_t addr, struct kvm_mem_range *range)
>  {
>  	int cur, left = 0, right = hyp_memblock_nr;
> @@ -227,6 +232,38 @@ static inline int __host_stage2_idmap(u64 start, u64 end,
>  		__ret;							\
>  	 })
>
> +static int host_stage2_find_range(u64 addr, struct kvm_mem_range *range)

nit: I find 'find_range' a bit odd. We already have found a
range. We're just trying to narrow it down to something that fits in a
single block mapping. How about 'host_stage2_adjust_range'?

> +{
> +	u64 granule, start, end;
> +	kvm_pte_t pte;
> +	u32 level;
> +	int ret;
> +
> +	ret = kvm_pgtable_get_leaf(&host_kvm.pgt, addr, &pte, &level);
> +	if (ret)
> +		return ret;
> +
> +	if (kvm_pte_valid(pte))
> +		return -EAGAIN;
> +
> +	if (pte)
> +		return -EPERM;
> +
> +	do {
> +		granule = kvm_granule_size(level);
> +		start = ALIGN_DOWN(addr, granule);
> +		end = start + granule;
> +		level++;
> +	} while ((level < KVM_PGTABLE_MAX_LEVELS) &&
> +			(!kvm_level_supports_block_mapping(level) ||
> +			 start < range->start || range->end < end));
> +

This expression does my head in. You are trying to find the largest
block mapping that entirely fits in range, right? Can we just express
that directly (with a global negation for the purpose of the loop)?

	do {
		[...]
	} while (level < KVM_PGTABLE_MAX_LEVELS &&
		 !(kvm_level_supports_block_mapping(level) &&
		   start >= range->start &&
		   end <= range->end));

I personally find this much more readable, because it expresses the
condition we are looking for rather than a lot of conditions forcing
us to continue.

You could also use a kvm_mem_range for the iteration, and add a helper
that checks for the inclusion.

> +	range->start = start;
> +	range->end = end;
> +
> +	return 0;
> +}
> +
>  static int host_stage2_idmap(u64 addr)
>  {
>  	enum kvm_pgtable_prot prot = KVM_PGTABLE_PROT_R | KVM_PGTABLE_PROT_W;
> @@ -238,7 +275,7 @@ static int host_stage2_idmap(u64 addr)
>  		prot |= KVM_PGTABLE_PROT_X;
>  
>  	hyp_spin_lock(&host_kvm.lock);
> -	ret = kvm_pgtable_stage2_find_range(&host_kvm.pgt, addr, prot, &range);
> +	ret = host_stage2_find_range(addr, &range);
>  	if (ret)
>  		goto unlock;

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.