lkml.org 
[lkml]   [2022]   [Feb]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
SubjectRe: [PATCH v2 09/18] KVM: x86/mmu: look for a cached PGD when going from 32-bit to 64-bit
From
Date
On Thu, 2022-02-17 at 16:03 -0500, Paolo Bonzini wrote:
> Right now, PGD caching avoids placing a PAE root in the cache by using the
> old value of mmu->root_level and mmu->shadow_root_level; it does not look
> for a cached PGD if the old root is a PAE one, and then frees it using
> kvm_mmu_free_roots.
>
> Change the logic instead to free the uncacheable root early.
> This way, __kvm_new_mmu_pgd is able to look up the cache when going from
> 32-bit to 64-bit (if there is a hit, the invalid root becomes the least
> recently used). An example of this is nested virtualization with shadow
> paging, when a 64-bit L1 runs a 32-bit L2.
>
> As a side effect (which is actually the reason why this patch was
> written), PGD caching does not use the old value of mmu->root_level
> and mmu->shadow_root_level anymore.

Which is great - I hated this code!!!

>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
> arch/x86/kvm/mmu/mmu.c | 82 ++++++++++++++++++++++++++++++------------
> 1 file changed, 59 insertions(+), 23 deletions(-)
>
> diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
> index 0f2de811e871..da324a317000 100644
> --- a/arch/x86/kvm/mmu/mmu.c
> +++ b/arch/x86/kvm/mmu/mmu.c
> @@ -4107,52 +4107,88 @@ static inline bool is_root_usable(struct kvm_mmu_root_info *root, gpa_t pgd,
> union kvm_mmu_page_role role)
> {
> return (role.direct || pgd == root->pgd) &&
> - VALID_PAGE(root->hpa) && to_shadow_page(root->hpa) &&
> + VALID_PAGE(root->hpa) &&
> role.word == to_shadow_page(root->hpa)->role.word;
> }
>
> /*
> - * Find out if a previously cached root matching the new pgd/role is available.
> - * The current root is also inserted into the cache.
> - * If a matching root was found, it is assigned to kvm_mmu->root.hpa and true is
> - * returned.
> - * Otherwise, the LRU root from the cache is assigned to kvm_mmu->root.hpa and
> - * false is returned. This root should now be freed by the caller.
> + * Find out if a previously cached root matching the new pgd/role is available,
> + * and insert the current root as the MRU in the cache.
> + * If a matching root is found, it is assigned to kvm_mmu->root and
> + * true is returned.
> + * If no match is found, kvm_mmu->root is left invalid, the LRU root is
> + * evicted to make room for the current root, and false is returned.
> */
> -static bool cached_root_available(struct kvm_vcpu *vcpu, gpa_t new_pgd,
> - union kvm_mmu_page_role new_role)
> +static bool cached_root_find_and_keep_current(struct kvm *kvm, struct kvm_mmu *mmu,
> + gpa_t new_pgd,
> + union kvm_mmu_page_role new_role)
> {
> uint i;
> - struct kvm_mmu *mmu = vcpu->arch.mmu;
>
> if (is_root_usable(&mmu->root, new_pgd, new_role))
> return true;
>
> for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
> + /*
> + * The swaps end up rotating the cache like this:
> + * C 0 1 2 3 (on entry to the function)
> + * 0 C 1 2 3
> + * 1 C 0 2 3
> + * 2 C 0 1 3
> + * 3 C 0 1 2 (on exit from the loop)
> + */

Thanks a million for documenting this! I remember it took
me too much time to figure out what all of these swaps do.

I do want to mention that it would be nice to mention that
the above trace is when none of the root in the cache match.


> swap(mmu->root, mmu->prev_roots[i]);
> -
> if (is_root_usable(&mmu->root, new_pgd, new_role))
> - break;
Maybe even add comment that if that break happens, the cache would look like
'2 C 0 1 3' with 2 beeing the matching root.

> + return true;
> }
>
> - return i < KVM_MMU_NUM_PREV_ROOTS;
> + kvm_mmu_free_roots(kvm, mmu, KVM_MMU_ROOT_CURRENT);
> + return false;
> }
>
> -static bool fast_pgd_switch(struct kvm_vcpu *vcpu, gpa_t new_pgd,
> - union kvm_mmu_page_role new_role)
> +/*
> + * Find out if a previously cached root matching the new pgd/role is available.
> + * On entry, mmu->root is invalid.
> + * If a matching root is found, it is assigned to kvm_mmu->root, the LRU entry
> + * of the cache becomes invalid, and true is returned.
> + * If no match is found, kvm_mmu->root is left invalid and false is returned.
> + */
> +static bool cached_root_find_without_current(struct kvm *kvm, struct kvm_mmu *mmu,
> + gpa_t new_pgd,
> + union kvm_mmu_page_role new_role)
> {
> - struct kvm_mmu *mmu = vcpu->arch.mmu;
> + uint i;
> +
> + for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
> + if (is_root_usable(&mmu->prev_roots[i], new_pgd, new_role))
> + goto hit;
>
> + return false;
> +
> +hit:
> + swap(mmu->root, mmu->prev_roots[i]);
> + /* Bubble up the remaining roots. */
> + for (; i < KVM_MMU_NUM_PREV_ROOTS - 1; i++)
> + mmu->prev_roots[i] = mmu->prev_roots[i + 1];
> + mmu->prev_roots[i].hpa = INVALID_PAGE;

I'll would have invalidated the 'pgd' value as well, just in case.

> + return true;
> +}

Since we just have 4 pointers in the LRU cache + root pointer, I wonder if something
dumber/slower could work for both cases.

Like something like that (not tested at all):

for (i = 0 ; i < KVM_MMU_NUM_PREV_ROOTS ; i++)
tmp_prev_roots[i] = mmu->prev_roots[i];

j = 0;

/* current mmu root becomes MRU*/
if (VALID_PAGE(mmu->root.hpa)
mmu->prev_roots[j++] = mmu->root;


for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS && j < KVM_MMU_NUM_PREV_ROOTS; i++)
if (is_root_usable(&tmp_prev_roots[i], new_pgd, new_role))
/* TODO: can also add a warning here if current root is already usable */
mmu->root = mmu->tmp_prev_roots[i];
else
mmu->prev_roots[j++] = mmu->tmp_prev_roots[i];

for (; j < KVM_MMU_NUM_PREV_ROOTS ; j++)
mmu->prev_roots[j++].hpa = INVALID_PAGE;


> +
> +static bool fast_pgd_switch(struct kvm *kvm, struct kvm_mmu *mmu,
> + gpa_t new_pgd, union kvm_mmu_page_role new_role)
> +{
> /*
> - * For now, limit the fast switch to 64-bit hosts+VMs in order to avoid
> + * For now, limit the caching to 64-bit hosts+VMs in order to avoid
> * having to deal with PDPTEs. We may add support for 32-bit hosts/VMs
> * later if necessary.
> */
> - if (mmu->shadow_root_level >= PT64_ROOT_4LEVEL &&
> - mmu->root_level >= PT64_ROOT_4LEVEL)
> - return cached_root_available(vcpu, new_pgd, new_role);
> + if (VALID_PAGE(mmu->root.hpa) && !to_shadow_page(mmu->root.hpa))
> + kvm_mmu_free_roots(kvm, mmu, KVM_MMU_ROOT_CURRENT);
>
> - return false;
> + if (VALID_PAGE(mmu->root.hpa))
> + return cached_root_find_and_keep_current(kvm, mmu, new_pgd, new_role);
> + else
> + return cached_root_find_without_current(kvm, mmu, new_pgd, new_role);
> }
>
> static void __kvm_mmu_new_pgd(struct kvm_vcpu *vcpu, gpa_t new_pgd,
> @@ -4160,8 +4196,8 @@ static void __kvm_mmu_new_pgd(struct kvm_vcpu *vcpu, gpa_t new_pgd,
> {
> struct kvm_mmu *mmu = vcpu->arch.mmu;
>
> - if (!fast_pgd_switch(vcpu, new_pgd, new_role)) {
> - kvm_mmu_free_roots(vcpu->kvm, mmu, KVM_MMU_ROOT_CURRENT);
> + if (!fast_pgd_switch(vcpu->kvm, mmu, new_pgd, new_role)) {
> + /* kvm_mmu_ensure_valid_pgd will set up a new root. */
I also agree with Sean's comment on this.
> return;
> }
>


I don't see any bugs in the code however, thus:

Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>

Best regards,
Maxim Levitsky


\
 
 \ /
  Last update: 2022-02-23 17:02    [W:0.386 / U:0.368 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site