[PATCH v15 10/17] RISC-V: KVM: Implement stage2 page table programming

[Anup Patel]

On Mon, Nov 16, 2020 at 2:59 PM Jiangyifei <jiangyifei at huawei.com> wrote:
>
>
> > -----Original Message-----
> > From: Anup Patel [mailto:anup.patel at wdc.com]
> > Sent: Monday, November 9, 2020 7:33 PM
> > To: Palmer Dabbelt <palmer at dabbelt.com>; Palmer Dabbelt
> > <palmerdabbelt at google.com>; Paul Walmsley <paul.walmsley at sifive.com>;
> > Albert Ou <aou at eecs.berkeley.edu>; Paolo Bonzini <pbonzini at redhat.com>
> > Cc: Alexander Graf <graf at amazon.com>; Atish Patra <atish.patra at wdc.com>;
> > Alistair Francis <Alistair.Francis at wdc.com>; Damien Le Moal
> > <damien.lemoal at wdc.com>; Anup Patel <anup at brainfault.org>;
> > kvm at vger.kernel.org; kvm-riscv at lists.infradead.org;
> > linux-riscv at lists.infradead.org; linux-kernel at vger.kernel.org; Anup Patel
> > <anup.patel at wdc.com>; Jiangyifei <jiangyifei at huawei.com>
> > Subject: [PATCH v15 10/17] RISC-V: KVM: Implement stage2 page table
> > programming
> >
> > This patch implements all required functions for programming the stage2 page
> > table for each Guest/VM.
> >
> > At high-level, the flow of stage2 related functions is similar from KVM
> > ARM/ARM64 implementation but the stage2 page table format is quite
> > different for KVM RISC-V.
> >
> > [jiangyifei: stage2 dirty log support]
> > Signed-off-by: Yifei Jiang <jiangyifei at huawei.com>
> > Signed-off-by: Anup Patel <anup.patel at wdc.com>
> > Acked-by: Paolo Bonzini <pbonzini at redhat.com>
> > Reviewed-by: Paolo Bonzini <pbonzini at redhat.com>
> > ---
> >  arch/riscv/include/asm/kvm_host.h     |  12 +
> >  arch/riscv/include/asm/pgtable-bits.h |   1 +
> >  arch/riscv/kvm/Kconfig                |   1 +
> >  arch/riscv/kvm/main.c                 |  19 +
> >  arch/riscv/kvm/mmu.c                  | 649
> > +++++++++++++++++++++++++-
> >  arch/riscv/kvm/vm.c                   |   6 -
> >  6 files changed, 672 insertions(+), 16 deletions(-)
> >
>
> ......
>
> >
> >  int kvm_riscv_stage2_map(struct kvm_vcpu *vcpu, @@ -69,27 +562,163 @@
> > int kvm_riscv_stage2_map(struct kvm_vcpu *vcpu,
> >                        gpa_t gpa, unsigned long hva,
> >                        bool writeable, bool is_write)
> >  {
> > -     /* TODO: */
> > -     return 0;
> > +     int ret;
> > +     kvm_pfn_t hfn;
> > +     short vma_pageshift;
> > +     gfn_t gfn = gpa >> PAGE_SHIFT;
> > +     struct vm_area_struct *vma;
> > +     struct kvm *kvm = vcpu->kvm;
> > +     struct kvm_mmu_page_cache *pcache = &vcpu->arch.mmu_page_cache;
> > +     bool logging = (memslot->dirty_bitmap &&
> > +                     !(memslot->flags & KVM_MEM_READONLY)) ? true : false;
> > +     unsigned long vma_pagesize;
> > +
> > +     mmap_read_lock(current->mm);
> > +
> > +     vma = find_vma_intersection(current->mm, hva, hva + 1);
> > +     if (unlikely(!vma)) {
> > +             kvm_err("Failed to find VMA for hva 0x%lx\n", hva);
> > +             mmap_read_unlock(current->mm);
> > +             return -EFAULT;
> > +     }
> > +
> > +     if (is_vm_hugetlb_page(vma))
> > +             vma_pageshift = huge_page_shift(hstate_vma(vma));
> > +     else
> > +             vma_pageshift = PAGE_SHIFT;
> > +     vma_pagesize = 1ULL << vma_pageshift;
> > +     if (logging || (vma->vm_flags & VM_PFNMAP))
> > +             vma_pagesize = PAGE_SIZE;
> > +
> > +     if (vma_pagesize == PMD_SIZE || vma_pagesize == PGDIR_SIZE)
> > +             gfn = (gpa & huge_page_mask(hstate_vma(vma))) >> PAGE_SHIFT;
> > +
> > +     mmap_read_unlock(current->mm);
> > +
> > +     if (vma_pagesize != PGDIR_SIZE &&
> > +         vma_pagesize != PMD_SIZE &&
> > +         vma_pagesize != PAGE_SIZE) {
> > +             kvm_err("Invalid VMA page size 0x%lx\n", vma_pagesize);
> > +             return -EFAULT;
> > +     }
> > +
> > +     /* We need minimum second+third level pages */
> > +     ret = stage2_cache_topup(pcache, stage2_pgd_levels,
> > +                              KVM_MMU_PAGE_CACHE_NR_OBJS);
> > +     if (ret) {
> > +             kvm_err("Failed to topup stage2 cache\n");
> > +             return ret;
> > +     }
> > +
> > +     hfn = gfn_to_pfn_prot(kvm, gfn, is_write, NULL);
> > +     if (hfn == KVM_PFN_ERR_HWPOISON) {
> > +             send_sig_mceerr(BUS_MCEERR_AR, (void __user *)hva,
> > +                             vma_pageshift, current);
> > +             return 0;
> > +     }
> > +     if (is_error_noslot_pfn(hfn))
> > +             return -EFAULT;
> > +
> > +     /*
> > +      * If logging is active then we allow writable pages only
> > +      * for write faults.
> > +      */
> > +     if (logging && !is_write)
> > +             writeable = false;
> > +
> > +     spin_lock(&kvm->mmu_lock);
> > +
> > +     if (writeable) {
>
> Hi Anup,
>
> What is the purpose of "writable = !memslot_is_readonly(slot)" in this series?

Where ? I don't see this line in any of the patches.

>
> When mapping the HVA to HPA above, it doesn't know that the PTE writeable of stage2 is "!memslot_is_readonly(slot)".
> This may causes the difference between the writability of HVA->HPA and GPA->HPA.
> For example, GPA->HPA is writeable, but HVA->HPA is not writeable.

Yes, this is possible particularly when Host kernel is updating writability
of HVA->HPA mappings for swapping in/out pages.

>
> Is it better that the writability of HVA->HPA is also determined by whether the memslot is readonly in this change?
> Like this:
> -    hfn = gfn_to_pfn_prot(kvm, gfn, is_write, NULL);
> +    hfn = gfn_to_pfn_prot(kvm, gfn, writeable, NULL);

The gfn_to_pfn_prot() needs to know what type of fault we
got (i.e read/write fault). Rest of the information (such as whether
slot is writable or not) is already available to gfn_to_pfn_prot().

The question here is should we pass "&writeable" or NULL as
last parameter to gfn_to_pfn_prot(). The recent JUMP label
support in Linux RISC-V causes problem on HW where PTE
'A' and 'D' bits are not updated by HW so I have to change
last parameter of gfn_to_pfn_prot() from "&writeable" to NULL.

I am still investigating this.

Regards,
Anup

>
> Regards,
> Yifei
>
> > +             kvm_set_pfn_dirty(hfn);
> > +             mark_page_dirty(kvm, gfn);
> > +             ret = stage2_map_page(kvm, pcache, gpa, hfn << PAGE_SHIFT,
> > +                                   vma_pagesize, false, true);
> > +     } else {
> > +             ret = stage2_map_page(kvm, pcache, gpa, hfn << PAGE_SHIFT,
> > +                                   vma_pagesize, true, true);
> > +     }
> > +
> > +     if (ret)
> > +             kvm_err("Failed to map in stage2\n");
> > +
> > +     spin_unlock(&kvm->mmu_lock);
> > +     kvm_set_pfn_accessed(hfn);
> > +     kvm_release_pfn_clean(hfn);
> > +     return ret;
> >  }
> >
>
> ......
>