[PATCH v2 1/3] arm64: KVM: Fix stage-2 PGD allocation to have per-page refcounting

Andrew Jones drjones at redhat.com
Wed Mar 11 07:03:33 PDT 2015


On Wed, Mar 11, 2015 at 02:20:56PM +0100, Christoffer Dall wrote:
> On Wed, Mar 11, 2015 at 2:13 PM, Marc Zyngier <marc.zyngier at arm.com> wrote:
> > Hi Christoffer,
> >
> > On 11/03/15 12:16, Christoffer Dall wrote:
> >> Hi Marc,
> >>
> >> On Tue, Mar 10, 2015 at 07:06:59PM +0000, Marc Zyngier wrote:
> >>> We're using __get_free_pages with to allocate the guest's stage-2
> >>> PGD. The standard behaviour of this function is to return a set of
> >>> pages where only the head page has a valid refcount.
> >>>
> >>> This behaviour gets us into trouble when we're trying to increment
> >>> the refcount on a non-head page:
> >>>
> >>> page:ffff7c00cfb693c0 count:0 mapcount:0 mapping:          (null) index:0x0
> >>> flags: 0x4000000000000000()
> >>> page dumped because: VM_BUG_ON_PAGE((*({ __attribute__((unused)) typeof((&page->_count)->counter) __var = ( typeof((&page->_count)->counter)) 0; (volatile typeof((&page->_count)->counter) *)&((&page->_count)->counter); })) <= 0)
> >>> BUG: failure at include/linux/mm.h:548/get_page()!
> >>> Kernel panic - not syncing: BUG!
> >>> CPU: 1 PID: 1695 Comm: kvm-vcpu-0 Not tainted 4.0.0-rc1+ #3825
> >>> Hardware name: APM X-Gene Mustang board (DT)
> >>> Call trace:
> >>> [<ffff80000008a09c>] dump_backtrace+0x0/0x13c
> >>> [<ffff80000008a1e8>] show_stack+0x10/0x1c
> >>> [<ffff800000691da8>] dump_stack+0x74/0x94
> >>> [<ffff800000690d78>] panic+0x100/0x240
> >>> [<ffff8000000a0bc4>] stage2_get_pmd+0x17c/0x2bc
> >>> [<ffff8000000a1dc4>] kvm_handle_guest_abort+0x4b4/0x6b0
> >>> [<ffff8000000a420c>] handle_exit+0x58/0x180
> >>> [<ffff80000009e7a4>] kvm_arch_vcpu_ioctl_run+0x114/0x45c
> >>> [<ffff800000099df4>] kvm_vcpu_ioctl+0x2e0/0x754
> >>> [<ffff8000001c0a18>] do_vfs_ioctl+0x424/0x5c8
> >>> [<ffff8000001c0bfc>] SyS_ioctl+0x40/0x78
> >>> CPU0: stopping
> >>>
> >>> A possible approach for this is to split the compound page using
> >>> split_page() at allocation time, and change the teardown path to
> >>> free one page at a time.
> >>>
> >>> While we're at it, the PGD allocation code is reworked to reduce
> >>> duplication.
> >>>
> >>> This has been tested on an X-Gene platform with a 4kB/48bit-VA host
> >>> kernel, and kvmtool hacked to place memory in the second page of
> >>> the hardware PGD (PUD for the host kernel). Also regression-tested
> >>> on a Cubietruck (Cortex-A7).
> >>>
> >>> Reported-by: Mark Rutland <mark.rutland at arm.com>
> >>> Signed-off-by: Marc Zyngier <marc.zyngier at arm.com>
> >>> ---
> >>>  arch/arm/include/asm/kvm_mmu.h   |  9 ++---
> >>>  arch/arm/kvm/mmu.c               | 77 +++++++++++++++++++++++++++++++---------
> >>>  arch/arm64/include/asm/kvm_mmu.h | 46 +++---------------------
> >>>  3 files changed, 66 insertions(+), 66 deletions(-)
> >>>
> >>> diff --git a/arch/arm/include/asm/kvm_mmu.h b/arch/arm/include/asm/kvm_mmu.h
> >>> index 0187606..ff56f91 100644
> >>> --- a/arch/arm/include/asm/kvm_mmu.h
> >>> +++ b/arch/arm/include/asm/kvm_mmu.h
> >>> @@ -162,18 +162,13 @@ static inline bool kvm_page_empty(void *ptr)
> >>>
> >>>  #define KVM_PREALLOC_LEVEL  0
> >>>
> >>> -static inline int kvm_prealloc_hwpgd(struct kvm *kvm, pgd_t *pgd)
> >>> -{
> >>> -    return 0;
> >>> -}
> >>> -
> >>> -static inline void kvm_free_hwpgd(struct kvm *kvm) { }
> >>> -
> >>>  static inline void *kvm_get_hwpgd(struct kvm *kvm)
> >>>  {
> >>>      return kvm->arch.pgd;
> >>>  }
> >>>
> >>> +static inline unsigned int kvm_get_hwpgd_order(void) { return S2_PGD_ORDER; }
> >>> +
> >>>  struct kvm;
> >>>
> >>>  #define kvm_flush_dcache_to_poc(a,l)        __cpuc_flush_dcache_area((a), (l))
> >>> diff --git a/arch/arm/kvm/mmu.c b/arch/arm/kvm/mmu.c
> >>> index 69c2b4c..0a5457c 100644
> >>> --- a/arch/arm/kvm/mmu.c
> >>> +++ b/arch/arm/kvm/mmu.c
> >>> @@ -634,6 +634,31 @@ int create_hyp_io_mappings(void *from, void *to, phys_addr_t phys_addr)
> >>>                                   __phys_to_pfn(phys_addr), PAGE_HYP_DEVICE);
> >>>  }
> >>>
> >>> +/* Free the HW pgd, one page at a time */
> >>> +static void kvm_free_hwpgd(unsigned long hwpgd)
> >>> +{
> >>> +    int i;
> >>> +
> >>> +    for (i = 0; i < (1 << kvm_get_hwpgd_order()); i += PAGE_SIZE)
> >>> +            free_page(hwpgd + i);
> >>> +}
> >>> +
> >>> +/* Allocate the HW PGD, making sure that each page gets its own refcount */
> >>> +static int kvm_alloc_hwpgd(unsigned long *hwpgdp)
> >>
> >> I think this can be simplified somewhat by just returning an unsigned
> >> long that can be 0 in the error case which will make the caller look
> >> a little nicer too.
> >
> > Sure.
> >
> >>> +{
> >>> +    unsigned long hwpgd;
> >>> +    unsigned int order = kvm_get_hwpgd_order();
> >>> +
> >>> +    hwpgd = __get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
> >>> +    if (!hwpgd)
> >>> +            return -ENOMEM;
> >>> +
> >>> +    split_page(virt_to_page((void *)hwpgd), order);
> >>
> >> nit: alloc_pages_exact() and free_pages_exact() seem to do this for us,
> >> is it worth using those instead?
> >
> > Ah, I didn't know about these.
> >
> >> It would look something like this on top of your changes:
> >>
> >>
> >>  arch/arm/include/asm/kvm_mmu.h   |  5 ++++-
> >>  arch/arm/kvm/mmu.c               | 32 ++++++++++----------------------
> >>  arch/arm64/include/asm/kvm_mmu.h |  6 +++---
> >>  3 files changed, 17 insertions(+), 26 deletions(-)
> >>
> >> diff --git a/arch/arm/include/asm/kvm_mmu.h b/arch/arm/include/asm/kvm_mmu.h
> >> index fc05ba8..4cf48c3 100644
> >> --- a/arch/arm/include/asm/kvm_mmu.h
> >> +++ b/arch/arm/include/asm/kvm_mmu.h
> >> @@ -168,7 +168,10 @@ static inline void *kvm_get_hwpgd(struct kvm *kvm)
> >>       return kvm->arch.pgd;
> >>  }
> >>
> >> -static inline unsigned int kvm_get_hwpgd_order(void) { return S2_PGD_ORDER; }
> >> +static inline unsigned int kvm_get_hwpgd_size(void)
> >> +{
> >> +     return PTRS_PER_S2_PGD * sizeof(pgd_t);
> >> +}
> >>
> >>  struct kvm;
> >>
> >> diff --git a/arch/arm/kvm/mmu.c b/arch/arm/kvm/mmu.c
> >> index 8e91bea3..5656d79 100644
> >> --- a/arch/arm/kvm/mmu.c
> >> +++ b/arch/arm/kvm/mmu.c
> >> @@ -633,28 +633,17 @@ int create_hyp_io_mappings(void *from, void *to, phys_addr_t phys_addr)
> >>  }
> >>
> >>  /* Free the HW pgd, one page at a time */
> >> -static void kvm_free_hwpgd(unsigned long hwpgd)
> >> +static void kvm_free_hwpgd(void *hwpgd)
> >>  {
> >> -     int i;
> >> -
> >> -     for (i = 0; i < (1 << kvm_get_hwpgd_order()); i += PAGE_SIZE)
> >> -             free_page(hwpgd + i);
> >> +     free_pages_exact(hwpgd, kvm_get_hwpgd_size());
> >>  }
> >>
> >>  /* Allocate the HW PGD, making sure that each page gets its own refcount */
> >> -static int kvm_alloc_hwpgd(unsigned long *hwpgdp)
> >> +static void *kvm_alloc_hwpgd(void)
> >>  {
> >> -     unsigned long hwpgd;
> >> -     unsigned int order = kvm_get_hwpgd_order();
> >> -
> >> -     hwpgd = __get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
> >> -     if (!hwpgd)
> >> -             return -ENOMEM;
> >> -
> >> -     split_page(virt_to_page((void *)hwpgd), order);
> >> +     unsigned int size = kvm_get_hwpgd_size();
> >>
> >> -     *hwpgdp = hwpgd;
> >> -     return 0;
> >> +     return alloc_pages_exact(size, GFP_KERNEL | __GFP_ZERO);
> >>  }
> >>
> >>  /**
> >> @@ -670,18 +659,17 @@ static int kvm_alloc_hwpgd(unsigned long *hwpgdp)
> >>   */
> >>  int kvm_alloc_stage2_pgd(struct kvm *kvm)
> >>  {
> >> -     int ret;
> >>       pgd_t *pgd;
> >> -     unsigned long hwpgd;
> >> +     void *hwpgd;
> >>
> >>       if (kvm->arch.pgd != NULL) {
> >>               kvm_err("kvm_arch already initialized?\n");
> >>               return -EINVAL;
> >>       }
> >>
> >> -     ret = kvm_alloc_hwpgd(&hwpgd);
> >> -     if (ret)
> >> -             return ret;
> >> +     hwpgd = kvm_alloc_hwpgd();
> >> +     if (!hwpgd)
> >> +             return -ENOMEM;
> >>
> >>       /* When the kernel uses more levels of page tables than the
> >>        * guest, we allocate a fake PGD and pre-populate it to point
> >> @@ -829,7 +817,7 @@ void kvm_free_stage2_pgd(struct kvm *kvm)
> >>               return;
> >>
> >>       unmap_stage2_range(kvm, 0, KVM_PHYS_SIZE);
> >> -     kvm_free_hwpgd((unsigned long)kvm_get_hwpgd(kvm));
> >> +     kvm_free_hwpgd(kvm_get_hwpgd(kvm));
> >>       if (KVM_PREALLOC_LEVEL > 0)
> >>               kfree(kvm->arch.pgd);
> >>
> >> diff --git a/arch/arm64/include/asm/kvm_mmu.h b/arch/arm64/include/asm/kvm_mmu.h
> >> index 3668110..bbfb600 100644
> >> --- a/arch/arm64/include/asm/kvm_mmu.h
> >> +++ b/arch/arm64/include/asm/kvm_mmu.h
> >> @@ -189,11 +189,11 @@ static inline void *kvm_get_hwpgd(struct kvm *kvm)
> >>       return pmd_offset(pud, 0);
> >>  }
> >>
> >> -static inline unsigned int kvm_get_hwpgd_order(void)
> >> +static inline unsigned int kvm_get_hwpgd_size(void)
> >>  {
> >>       if (KVM_PREALLOC_LEVEL > 0)
> >> -             return PTRS_PER_S2_PGD_SHIFT;
> >> -     return S2_PGD_ORDER;
> >> +             return PTRS_PER_S2_PGD * PAGE_SIZE;
> >> +     return PTRS_PER_S2_PGD * sizeof(pgd_t);
> >>  }
> >>
> >>  static inline bool kvm_page_empty(void *ptr)
> >
> > Looks good to me. I'll fold that into the patch, amend the commit
> > message and resend.
> >
> 
> I've done that here:
> https://git.linaro.org/people/christoffer.dall/linux-kvm-arm.git/shortlog/refs/heads/pgd-fixes-alt
> 
> (except for the commit message, which I can also adjust).
> 
> I gave this a quick spin on Juno and TC2 (but couldn't test 64K pages
> because I don't have a platform for that handy this very second)...

Tested with 64K config. Works for me (test was just a guest boot)

drew

> 
> -Christoffer
> _______________________________________________
> kvmarm mailing list
> kvmarm at lists.cs.columbia.edu
> https://lists.cs.columbia.edu/mailman/listinfo/kvmarm



More information about the linux-arm-kernel mailing list