Re: [PATCH v6 05/38] KVM: x86: hyper-v: Handle HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST{,EX} calls gently

[Maxim Levitsky <mlevitsk@redhat.com>]

On Mon, 2022-06-06 at 10:36 +0200, Vitaly Kuznetsov wrote:
> Currently, HVCALL_FLUSH_VIRTUAL_ADDRESS_LIST{,EX} calls are handled
> the exact same way as HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE{,EX}: by
> flushing the whole VPID and this is sub-optimal. Switch to handling
> these requests with 'flush_tlb_gva()' hooks instead. Use the newly
> introduced TLB flush fifo to queue the requests.
> 
> Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
> ---
>  arch/x86/kvm/hyperv.c | 100 +++++++++++++++++++++++++++++++++++++-----
>  1 file changed, 88 insertions(+), 12 deletions(-)
> 
> diff --git a/arch/x86/kvm/hyperv.c b/arch/x86/kvm/hyperv.c
> index 762b0b699fdf..956072592e2f 100644
> --- a/arch/x86/kvm/hyperv.c
> +++ b/arch/x86/kvm/hyperv.c
> @@ -1806,32 +1806,82 @@ static u64 kvm_get_sparse_vp_set(struct kvm *kvm, struct kvm_hv_hcall *hc,
>                                   sparse_banks, consumed_xmm_halves, offset);
>  }
>  
> -static void hv_tlb_flush_enqueue(struct kvm_vcpu *vcpu)
> +static int kvm_hv_get_tlb_flush_entries(struct kvm *kvm, struct kvm_hv_hcall *hc, u64 entries[],
> +                                       int consumed_xmm_halves, gpa_t offset)
> +{
> +       return kvm_hv_get_hc_data(kvm, hc, hc->rep_cnt, hc->rep_cnt,
> +                                 entries, consumed_xmm_halves, offset);
> +}
> +
> +static void hv_tlb_flush_enqueue(struct kvm_vcpu *vcpu, u64 *entries, int count)
>  {
>         struct kvm_vcpu_hv_tlb_flush_fifo *tlb_flush_fifo;
>         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
>         u64 entry = KVM_HV_TLB_FLUSHALL_ENTRY;
> +       unsigned long flags;
>  
>         if (!hv_vcpu)
>                 return;
>  
>         tlb_flush_fifo = &hv_vcpu->tlb_flush_fifo;
>  
> -       kfifo_in_spinlocked(&tlb_flush_fifo->entries, &entry, 1, &tlb_flush_fifo->write_lock);
> +       spin_lock_irqsave(&tlb_flush_fifo->write_lock, flags);
> +
> +       /*
> +        * All entries should fit on the fifo leaving one free for 'flush all'
> +        * entry in case another request comes in. In case there's not enough
> +        * space, just put 'flush all' entry there.
> +        */
> +       if (count && entries && count < kfifo_avail(&tlb_flush_fifo->entries)) {
> +               WARN_ON(kfifo_in(&tlb_flush_fifo->entries, entries, count) != count);
> +               goto out_unlock;
> +       }
> +
> +       /*
> +        * Note: full fifo always contains 'flush all' entry, no need to check the
> +        * return value.
> +        */
> +       kfifo_in(&tlb_flush_fifo->entries, &entry, 1);
Very tiny nitpick: maybe call this flush_all_entry instead,
just so that it is a tiny bit easier to notice.


> +
> +out_unlock:
> +       spin_unlock_irqrestore(&tlb_flush_fifo->write_lock, flags);
>  }
>  
>  void kvm_hv_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
>  {
>         struct kvm_vcpu_hv_tlb_flush_fifo *tlb_flush_fifo;
>         struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
> +       u64 entries[KVM_HV_TLB_FLUSH_FIFO_SIZE];
> +       int i, j, count;
> +       gva_t gva;
>  
> -       kvm_vcpu_flush_tlb_guest(vcpu);
> -
> -       if (!hv_vcpu)
> +       if (!tdp_enabled || !hv_vcpu) {
I haven't noticed that in the review I did back then, but
any reason why !tdp_enabled? Just curious.

> +               kvm_vcpu_flush_tlb_guest(vcpu);
>                 return;
> +       }
>  
>         tlb_flush_fifo = &hv_vcpu->tlb_flush_fifo;
>  
> +       count = kfifo_out(&tlb_flush_fifo->entries, entries, KVM_HV_TLB_FLUSH_FIFO_SIZE);
> +
> +       for (i = 0; i < count; i++) {
> +               if (entries[i] == KVM_HV_TLB_FLUSHALL_ENTRY)
> +                       goto out_flush_all;
> +
> +               /*
> +                * Lower 12 bits of 'address' encode the number of additional
> +                * pages to flush.
> +                */
> +               gva = entries[i] & PAGE_MASK;
> +               for (j = 0; j < (entries[i] & ~PAGE_MASK) + 1; j++)
> +                       static_call(kvm_x86_flush_tlb_gva)(vcpu, gva + j * PAGE_SIZE);
> +
> +               ++vcpu->stat.tlb_flush;
> +       }
> +       return;
> +
> +out_flush_all:
> +       kvm_vcpu_flush_tlb_guest(vcpu);
>         kfifo_reset_out(&tlb_flush_fifo->entries);
>  }
>  
> @@ -1841,11 +1891,21 @@ static u64 kvm_hv_flush_tlb(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc)
>         struct hv_tlb_flush_ex flush_ex;
>         struct hv_tlb_flush flush;
>         DECLARE_BITMAP(vcpu_mask, KVM_MAX_VCPUS);
> +       /*
> +        * Normally, there can be no more than 'KVM_HV_TLB_FLUSH_FIFO_SIZE'
> +        * entries on the TLB flush fifo. The last entry, however, needs to be
> +        * always left free for 'flush all' entry which gets placed when
> +        * there is not enough space to put all the requested entries.
> +        */
> +       u64 __tlb_flush_entries[KVM_HV_TLB_FLUSH_FIFO_SIZE - 1];
> +       u64 *tlb_flush_entries;
>         u64 valid_bank_mask;
>         u64 sparse_banks[KVM_HV_MAX_SPARSE_VCPU_SET_BITS];
>         struct kvm_vcpu *v;
>         unsigned long i;
>         bool all_cpus;
> +       int consumed_xmm_halves = 0;
> +       gpa_t data_offset;
>  
>         /*
>          * The Hyper-V TLFS doesn't allow more than 64 sparse banks, e.g. the
> @@ -1861,10 +1921,12 @@ static u64 kvm_hv_flush_tlb(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc)
>                         flush.address_space = hc->ingpa;
>                         flush.flags = hc->outgpa;
>                         flush.processor_mask = sse128_lo(hc->xmm[0]);
> +                       consumed_xmm_halves = 1;
>                 } else {
>                         if (unlikely(kvm_read_guest(kvm, hc->ingpa,
>                                                     &flush, sizeof(flush))))
>                                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
> +                       data_offset = sizeof(flush);
>                 }
>  
>                 trace_kvm_hv_flush_tlb(flush.processor_mask,
> @@ -1888,10 +1950,12 @@ static u64 kvm_hv_flush_tlb(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc)
>                         flush_ex.flags = hc->outgpa;
>                         memcpy(&flush_ex.hv_vp_set,
>                                &hc->xmm[0], sizeof(hc->xmm[0]));
> +                       consumed_xmm_halves = 2;
>                 } else {
>                         if (unlikely(kvm_read_guest(kvm, hc->ingpa, &flush_ex,
>                                                     sizeof(flush_ex))))
>                                 return HV_STATUS_INVALID_HYPERCALL_INPUT;
> +                       data_offset = sizeof(flush_ex);
>                 }
>  
>                 trace_kvm_hv_flush_tlb_ex(flush_ex.hv_vp_set.valid_bank_mask,
> @@ -1907,25 +1971,37 @@ static u64 kvm_hv_flush_tlb(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc)
>                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
>  
>                 if (all_cpus)
> -                       goto do_flush;
> +                       goto read_flush_entries;
>  
>                 if (!hc->var_cnt)
>                         goto ret_success;
>  
> -               if (kvm_get_sparse_vp_set(kvm, hc, sparse_banks, 2,
> -                                         offsetof(struct hv_tlb_flush_ex,
> -                                                  hv_vp_set.bank_contents)))
> +               if (kvm_get_sparse_vp_set(kvm, hc, sparse_banks, consumed_xmm_halves,
> +                                         data_offset))
> +                       return HV_STATUS_INVALID_HYPERCALL_INPUT;
> +               data_offset += hc->var_cnt * sizeof(sparse_banks[0]);
> +               consumed_xmm_halves += hc->var_cnt;
> +       }
> +
> +read_flush_entries:
> +       if (hc->code == HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE ||
> +           hc->code == HVCALL_FLUSH_VIRTUAL_ADDRESS_SPACE_EX ||
> +           hc->rep_cnt > ARRAY_SIZE(__tlb_flush_entries)) {
> +               tlb_flush_entries = NULL;
> +       } else {
> +               if (kvm_hv_get_tlb_flush_entries(kvm, hc, __tlb_flush_entries,
> +                                               consumed_xmm_halves, data_offset))
>                         return HV_STATUS_INVALID_HYPERCALL_INPUT;
> +               tlb_flush_entries = __tlb_flush_entries;
>         }
>  
> -do_flush:
>         /*
>          * vcpu->arch.cr3 may not be up-to-date for running vCPUs so we can't
>          * analyze it here, flush TLB regardless of the specified address space.
>          */
>         if (all_cpus) {
>                 kvm_for_each_vcpu(i, v, kvm)
> -                       hv_tlb_flush_enqueue(v);
> +                       hv_tlb_flush_enqueue(v, tlb_flush_entries, hc->rep_cnt);
>  
>                 kvm_make_all_cpus_request(kvm, KVM_REQ_HV_TLB_FLUSH);
>         } else {
> @@ -1935,7 +2011,7 @@ static u64 kvm_hv_flush_tlb(struct kvm_vcpu *vcpu, struct kvm_hv_hcall *hc)
>                         v = kvm_get_vcpu(kvm, i);
>                         if (!v)
>                                 continue;
> -                       hv_tlb_flush_enqueue(v);
> +                       hv_tlb_flush_enqueue(v, tlb_flush_entries, hc->rep_cnt);
>                 }
>  
>                 kvm_make_vcpus_request_mask(kvm, KVM_REQ_HV_TLB_FLUSH, vcpu_mask);


Besides the nitpick, dont see anything wrong, but I might have missed something.

Best regards,
	Maxim Levitsky