We have both KERN_TO_HYP and kern_hyp_va, which do the exact same
thing. Let's standardize on the latter.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NChristoffer Dall <christoffer.dall@linaro.org>

This is more of a safety measure than anything else: If we end-up
with an idmap page that intersect with the range picked for the
the HYP VA space, abort the KVM setup, as it is unsafe to go
further.

I cannot imagine it happening on 64bit (we have a mechanism to
work around it), but could potentially occur on a 32bit system with
the kernel loaded high enough in memory so that in conflicts with
the kernel VA.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NChristoffer Dall <christoffer.dall@linaro.org>

There is no way to free the boot PGD, because it doesn't exist
anymore as a standalone entity.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NChristoffer Dall <christoffer.dall@linaro.org>

Since we now only have one set of page tables, the concept of
boot_pgd is useless and can be removed. We still keep it as
an element of the "extended idmap" thing.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NChristoffer Dall <christoffer.dall@linaro.org>

We're in a position where we can now always have "merged" page
tables, where both the runtime mapping and the idmap coexist.

This results in some code being removed, but there is more to come.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NChristoffer Dall <christoffer.dall@linaro.org>

Declare the __hyp_text_start/end symbols in asm/virt.h so that
they can be reused without having to declare them locally.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NChristoffer Dall <christoffer.dall@linaro.org>

There should be no reason for mapping the HYP text read/write.

As such, let's have a new set of flags (PAGE_HYP_EXEC) that allows
execution, but makes the page as read-only, and update the two call
sites that deal with mapping code.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NChristoffer Dall <christoffer.dall@linaro.org>

Currently, create_hyp_mappings applies a "one size fits all" page
protection (PAGE_HYP). As we're heading towards separate protections
for different sections, let's make this protection a parameter, and
let the callers pass their prefered protection (PAGE_HYP for everyone
for the time being).
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NChristoffer Dall <christoffer.dall@linaro.org>

The ARMv8.1 architecture extensions introduce support for hardware
updates of the access and dirty information in page table entries. With
VTCR_EL2.HA enabled (bit 21), when the CPU accesses an IPA with the
PTE_AF bit cleared in the stage 2 page table, instead of raising an
Access Flag fault to EL2 the CPU sets the actual page table entry bit
(10). To ensure that kernel modifications to the page table do not
inadvertently revert a bit set by hardware updates, certain Stage 2
software pte/pmd operations must be performed atomically.

The main user of the AF bit is the kvm_age_hva() mechanism. The
kvm_age_hva_handler() function performs a "test and clear young" action
on the pte/pmd. This needs to be atomic in respect of automatic hardware
updates of the AF bit. Since the AF bit is in the same position for both
Stage 1 and Stage 2, the patch reuses the existing
ptep_test_and_clear_young() functionality if
__HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG is defined. Otherwise, the
existing pte_young/pte_mkold mechanism is preserved.

The kvm_set_s2pte_readonly() (and the corresponding pmd equivalent) have
to perform atomic modifications in order to avoid a race with updates of
the AF bit. The arm64 implementation has been re-written using
exclusives.

Currently, kvm_set_s2pte_writable() (and pmd equivalent) take a pointer
argument and modify the pte/pmd in place. However, these functions are
only used on local variables rather than actual page table entries, so
it makes more sense to follow the pte_mkwrite() approach for stage 1
attributes. The change to kvm_s2pte_mkwrite() makes it clear that these
functions do not modify the actual page table entries.

The (pte|pmd)_mkyoung() uses on Stage 2 entries (setting the AF bit
explicitly) do not need to be modified since hardware updates of the
dirty status are not supported by KVM, so there is no possibility of
losing such information.
Signed-off-by: NCatalin Marinas <catalin.marinas@arm.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Acked-by: NMarc Zyngier <marc.zyngier@arm.com>
Reviewed-by: NChristoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: NChristoffer Dall <christoffer.dall@linaro.org>

After the THP refcounting change, obtaining a compound pages from
get_user_pages() no longer allows us to assume the entire compound page
is immediately mappable from a secondary MMU.

A secondary MMU doesn't want to call get_user_pages() more than once for
each compound page, in order to know if it can map the whole compound
page.  So a secondary MMU needs to know from a single get_user_pages()
invocation when it can map immediately the entire compound page to avoid
a flood of unnecessary secondary MMU faults and spurious
atomic_inc()/atomic_dec() (pages don't have to be pinned by MMU notifier
users).

Ideally instead of the page->_mapcount < 1 check, get_user_pages()
should return the granularity of the "page" mapping in the "mm" passed
to get_user_pages().  However it's non trivial change to pass the "pmd"
status belonging to the "mm" walked by get_user_pages up the stack (up
to the caller of get_user_pages).  So the fix just checks if there is
not a single pte mapping on the page returned by get_user_pages, and in
turn if the caller can assume that the whole compound page is mapped in
the current "mm" (in a pmd_trans_huge()).  In such case the entire
compound page is safe to map into the secondary MMU without additional
get_user_pages() calls on the surrounding tail/head pages.  In addition
of being faster, not having to run other get_user_pages() calls also
reduces the memory footprint of the secondary MMU fault in case the pmd
split happened as result of memory pressure.

Without this fix after a MADV_DONTNEED (like invoked by QEMU during
postcopy live migration or balloning) or after generic swapping (with a
failure in split_huge_page() that would only result in pmd splitting and
not a physical page split), KVM would map the whole compound page into
the shadow pagetables, despite regular faults or userfaults (like
UFFDIO_COPY) may map regular pages into the primary MMU as result of the
pte faults, leading to the guest mode and userland mode going out of
sync and not working on the same memory at all times.

Any other secondary MMU notifier manager (KVM is just one of the many
MMU notifier users) will need the same information if it doesn't want to
run a flood of get_user_pages_fast and it can support multiple
granularity in the secondary MMU mappings, so I think it is justified to
be exposed not just to KVM.

The other option would be to move transparent_hugepage_adjust to
mm/huge_memory.c but that currently has all kind of KVM data structures
in it, so it's definitely not a cut-and-paste work, so I couldn't do a
fix as cleaner as this one for 4.6.
Signed-off-by: NAndrea Arcangeli <aarcange@redhat.com>
Cc: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: "Li, Liang Z" <liang.z.li@intel.com>
Cc: Amit Shah <amit.shah@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The ARM architecture mandates that when changing a page table entry
from a valid entry to another valid entry, an invalid entry is first
written, TLB invalidated, and only then the new entry being written.

The current code doesn't respect this, directly writing the new
entry and only then invalidating TLBs. Let's fix it up.

Cc: <stable@vger.kernel.org>
Reported-by: NChristoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NChristoffer Dall <christoffer.dall@linaro.org>

The current kvm implementation on arm64 does cpu-specific initialization
at system boot, and has no way to gracefully shutdown a core in terms of
kvm. This prevents kexec from rebooting the system at EL2.

This patch adds a cpu tear-down function and also puts an existing cpu-init
code into a separate function, kvm_arch_hardware_disable() and
kvm_arch_hardware_enable() respectively.
We don't need the arm64 specific cpu hotplug hook any more.

Since this patch modifies common code between arm and arm64, one stub
definition, __cpu_reset_hyp_mode(), is added on arm side to avoid
compilation errors.
Signed-off-by: NAKASHI Takahiro <takahiro.akashi@linaro.org>
[Rebase, added separate VHE init/exit path, changed resets use of
 kvm_call_hyp() to the __version, en/disabled hardware in init_subsystems(),
 added icache maintenance to __kvm_hyp_reset() and removed lr restore, removed
 guest-enter after teardown handling]
Signed-off-by: NJames Morse <james.morse@arm.com>
Acked-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

Now that we don't have any fake page table levels for arm64,
cleanup the common code to get rid of the dead code.

Cc: Marc Zyngier <marc.zyngier@arm.com>
Acked-by: NChristoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: NSuzuki K Poulose <suzuki.poulose@arm.com>

Now that we have switched to explicit page table routines,
get rid of the obsolete kvm_* wrappers.

Also, kvm_tlb_flush_vmid_by_ipa is now called only on stage2
page tables, hence get rid of the redundant check.

Cc: Marc Zyngier <marc.zyngier@arm.com>
Acked-by: NChristoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: NSuzuki K Poulose <suzuki.poulose@arm.com>

Now that the hyp page table is handled by different set of
routines, rename the original shared routines to stage2 handlers.
Also make explicit use of the stage2 page table helpers.

unmap_range has been merged to existing unmap_stage2_range.

Cc: Marc Zyngier <marc.zyngier@arm.com>
Acked-by: NChristoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: NSuzuki K Poulose <suzuki.poulose@arm.com>

We have common routines to modify hyp and stage2 page tables
based on the 'kvm' parameter. For a smoother transition to
using separate routines for each, duplicate the routines
and modify the copy to work on hyp.

Marks the forked routines with _hyp_ and gets rid of the
kvm parameter which is no longer needed and is NULL for hyp.
Also, gets rid of calls to kvm_tlb_flush_by_vmid_ipa() calls
from the hyp versions. Uses explicit host page table accessors
instead of the kvm_* page table helpers.
Suggested-by: NChristoffer Dall <christoffer.dall@linaro.org>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: NChristoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: NSuzuki K Poulose <suzuki.poulose@arm.com>

We have stage2 page table helpers for both arm and arm64. Switch to
the stage2 helpers for routines that only deal with stage2 page table.

Cc: Marc Zyngier <marc.zyngier@arm.com>
Acked-by: NChristoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: NSuzuki K Poulose <suzuki.poulose@arm.com>

Get rid of kvm_pud_huge() which falls back to pud_huge. Use
pud_huge instead.
Acked-by: NChristoffer Dall <christoffer.dall@linaro.org>
Acked-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NSuzuki K Poulose <suzuki.poulose@arm.com>

Both arm and arm64 now provides a helper, pmd_thp_or_huge()
to check if the given pmd represents a huge page. Use that
instead of our own custom check.
Suggested-by: NMark Rutland <mark.rutland@arm.com>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Acked-by: NMarc Zyngier <marc.zyngier@arm.com>
Acked-by: NChristoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: NSuzuki K Poulose <suzuki.poulose@arm.com>

Rearrange the code for fake pgd handling, which is applicable
only for arm64. This will later be removed once we introduce
the stage2 page table walker macros.
Reviewed-by: NMarc Zyngier <marc.zyngier@arm.com>
Reviewed-by: NChristoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: NSuzuki K Poulose <suzuki.poulose@arm.com>

With the kernel running at EL2, there is no point trying to
configure page tables for HYP, as the kernel is already mapped.

Take this opportunity to refactor the whole init a bit, allowing
the various parts of the hypervisor bringup to be split across
multiple functions.
Reviewed-by: NChristoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>

So far, our handling of cache maintenance by VA has been pretty
simple: Either the access is in the guest RAM and generates a S2
fault, which results in the page being mapped RW, or we go down
the io_mem_abort() path, and nuke the guest.

The first one is fine, but the second one is extremely weird.
Treating the CM as an I/O is wrong, and nothing in the ARM ARM
indicates that we should generate a fault for something that
cannot end-up in the cache anyway (even if the guest maps it,
it will keep on faulting at stage-2 for emulation).

So let's just skip this instruction, and let the guest get away
with it.
Reviewed-by: NChristoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>

To date, we have implemented two I/O usage models for persistent memory,
PMEM (a persistent "ram disk") and DAX (mmap persistent memory into
userspace).  This series adds a third, DAX-GUP, that allows DAX mappings
to be the target of direct-i/o.  It allows userspace to coordinate
DMA/RDMA from/to persistent memory.

The implementation leverages the ZONE_DEVICE mm-zone that went into
4.3-rc1 (also discussed at kernel summit) to flag pages that are owned
and dynamically mapped by a device driver.  The pmem driver, after
mapping a persistent memory range into the system memmap via
devm_memremap_pages(), arranges for DAX to distinguish pfn-only versus
page-backed pmem-pfns via flags in the new pfn_t type.

The DAX code, upon seeing a PFN_DEV+PFN_MAP flagged pfn, flags the
resulting pte(s) inserted into the process page tables with a new
_PAGE_DEVMAP flag.  Later, when get_user_pages() is walking ptes it keys
off _PAGE_DEVMAP to pin the device hosting the page range active.
Finally, get_page() and put_page() are modified to take references
against the device driver established page mapping.

Finally, this need for "struct page" for persistent memory requires
memory capacity to store the memmap array.  Given the memmap array for a
large pool of persistent may exhaust available DRAM introduce a
mechanism to allocate the memmap from persistent memory.  The new
"struct vmem_altmap *" parameter to devm_memremap_pages() enables
arch_add_memory() to use reserved pmem capacity rather than the page
allocator.

This patch (of 18):

The core has developed a need for a "pfn_t" type [1].  Move the existing
pfn_t in KVM to kvm_pfn_t [2].

[1]: https://lists.01.org/pipermail/linux-nvdimm/2015-September/002199.html
[2]: https://lists.01.org/pipermail/linux-nvdimm/2015-September/002218.htmlSigned-off-by: NDan Williams <dan.j.williams@intel.com>
Acked-by: NChristoffer Dall <christoffer.dall@linaro.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Since commit a987370f ("arm64: KVM: Fix stage-2 PGD allocation to have
per-page refcounting") there is no reference to S2_PGD_ORDER, so kill it
for the good.
Acked-by: NChristoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: NVladimir Murzin <vladimir.murzin@arm.com>
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>

Commit e6fab544 ("ARM/arm64: KVM: test properly for a PTE's
uncachedness") modified the logic to test whether a HYP or stage-2
mapping needs flushing, from [incorrectly] interpreting the page table
attributes to [incorrectly] checking whether the PFN that backs the
mapping is covered by host system RAM. The PFN number is part of the
output of the translation, not the input, so we have to use pte_pfn()
on the contents of the PTE, not __phys_to_pfn() on the HYP virtual
address or stage-2 intermediate physical address.

Fixes: e6fab544 ("ARM/arm64: KVM: test properly for a PTE's uncachedness")
Cc: stable@vger.kernel.org
Tested-by: NPavel Fedin <p.fedin@samsung.com>
Signed-off-by: NArd Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>

The open coded tests for checking whether a PTE maps a page as
uncached use a flawed '(pte_val(xxx) & CONST) != CONST' pattern,
which is not guaranteed to work since the type of a mapping is
not a set of mutually exclusive bits

For HYP mappings, the type is an index into the MAIR table (i.e, the
index itself does not contain any information whatsoever about the
type of the mapping), and for stage-2 mappings it is a bit field where
normal memory and device types are defined as follows:

    #define MT_S2_NORMAL            0xf
    #define MT_S2_DEVICE_nGnRE      0x1

I.e., masking *and* comparing with the latter matches on the former,
and we have been getting lucky merely because the S2 device mappings
also have the PTE_UXN bit set, or we would misidentify memory mappings
as device mappings.

Since the unmap_range() code path (which contains one instance of the
flawed test) is used both for HYP mappings and stage-2 mappings, and
considering the difference between the two, it is non-trivial to fix
this by rewriting the tests in place, as it would involve passing
down the type of mapping through all the functions.

However, since HYP mappings and stage-2 mappings both deal with host
physical addresses, we can simply check whether the mapping is backed
by memory that is managed by the host kernel, and only perform the
D-cache maintenance if this is the case.

Cc: stable@vger.kernel.org
Signed-off-by: NArd Biesheuvel <ard.biesheuvel@linaro.org>
Tested-by: NPavel Fedin <p.fedin@samsung.com>
Reviewed-by: NChristoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: NChristoffer Dall <christoffer.dall@linaro.org>

A critical bug has been found in device memory stage1 translation for
VMs with more then 4GB of address space. Once vm_pgoff size is smaller
then pa (which is true for LPAE case, u32 and u64 respectively) some
more significant bits of pa may be lost as a shift operation is performed
on u32 and later cast onto u64.

Example: vm_pgoff(u32)=0x00210030, PAGE_SHIFT=12
        expected pa(u64):   0x0000002010030000
        produced pa(u64):   0x0000000010030000

The fix is to change the order of operations (casting first onto phys_addr_t
and then shifting).
Reviewed-by: NMarc Zyngier <marc.zyngier@arm.com>
[maz: fixed changelog and patch formatting]
Cc: stable@vger.kernel.org
Signed-off-by: NMarek Majtyka <marek.majtyka@tieto.com>
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>

No need to cast the void pointer returned by kmalloc() in
arch/arm/kvm/mmu.c::kvm_alloc_stage2_pgd().
Signed-off-by: NFiro Yang <firogm@gmail.com>
Acked-by: NPaolo Bonzini <pbonzini@redhat.com>
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>

This lets the function access the new memory slot without going through
kvm_memslots and id_to_memslot.  It will simplify the code when more
than one address space will be supported.

Unfortunately, the "const"ness of the new argument must be casted
away in two places.  Fixing KVM to accept const struct kvm_memory_slot
pointers would require modifications in pretty much all architectures,
and is left for later.
Reviewed-by: NRadim Krcmar <rkrcmar@redhat.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Prepare for the case of multiple address spaces.
Reviewed-by: NRadim Krcmar <rkrcmar@redhat.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Architecture-specific helpers are not supposed to muck with
struct kvm_userspace_memory_region contents.  Add const to
enforce this.

In order to eliminate the only write in __kvm_set_memory_region,
the cleaning of deleted slots is pulled up from update_memslots
to __kvm_set_memory_region.
Reviewed-by: NTakuya Yoshikawa <yoshikawa_takuya_b1@lab.ntt.co.jp>
Reviewed-by: NRadim Krcmar <rkrcmar@redhat.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

kvm_memslots provides lockdep checking.  Use it consistently instead of
explicit dereferencing of kvm->memslots.
Reviewed-by: NRadim Krcmar <rkrcmar@redhat.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

This patch modifies the HYP init code so it can deal with system
RAM residing at an offset which exceeds the reach of VA_BITS.

Like for EL1, this involves configuring an additional level of
translation for the ID map. However, in case of EL2, this implies
that all translations use the extra level, as we cannot seamlessly
switch between translation tables with different numbers of
translation levels.

So add an extra translation table at the root level. Since the
ID map and the runtime HYP map are guaranteed not to overlap, they
can share this root level, and we can essentially merge these two
tables into one.
Tested-by: NMarc Zyngier <marc.zyngier@arm.com>
Reviewed-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NArd Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

The HYP init bounce page is a runtime construct that ensures that the
HYP init code does not cross a page boundary. However, this is something
we can do perfectly well at build time, by aligning the code appropriately.

For arm64, we just align to 4 KB, and enforce that the code size is less
than 4 KB, regardless of the chosen page size.

For ARM, the whole code is less than 256 bytes, so we tweak the linker
script to align at a power of 2 upper bound of the code size

Note that this also fixes a benign off-by-one error in the original bounce
page code, where a bounce page would be allocated unnecessarily if the code
was exactly 1 page in size.

On ARM, it also fixes an issue with very large kernels reported by Arnd
Bergmann, where stub sections with linker emitted veneers could erroneously
trigger the size/alignment ASSERT() in the linker script.
Tested-by: NMarc Zyngier <marc.zyngier@arm.com>
Reviewed-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NArd Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: NWill Deacon <will.deacon@arm.com>

Now that we have page aging in Stage-2, it becomes obvious that
we're doing way too much work handling the fault.

The page is not going anywhere (it is still mapped), the page
tables are already allocated, and all we want is to flip a bit
in the PMD or PTE. Also, we can avoid any form of TLB invalidation,
since a page with the AF bit off is not allowed to be cached.

An obvious solution is to have a separate handler for FSC_ACCESS,
where we pride ourselves to only do the very minimum amount of
work.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Acked-by: NChristoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: NChristoffer Dall <christoffer.dall@linaro.org>

Until now, KVM/arm didn't care much for page aging (who was swapping
anyway?), and simply provided empty hooks to the core KVM code. With
server-type systems now being available, things are quite different.

This patch implements very simple support for page aging, by clearing
the Access flag in the Stage-2 page tables. On access fault, the current
fault handling will write the PTE or PMD again, putting the Access flag
back on.

It should be possible to implement a much faster handling for Access
faults, but that's left for a later patch.

With this in place, performance in VMs is degraded much more gracefully.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Acked-by: NChristoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: NChristoffer Dall <christoffer.dall@linaro.org>

So far, handle_hva_to_gpa was never required to return a value.
As we prepare to age pages at Stage-2, we need to be able to
return a value from the iterator (kvm_test_age_hva).

Adapt the code to handle this situation. No semantic change.
Acked-by: NChristoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NChristoffer Dall <christoffer.dall@linaro.org>

The kernel's pgd_index macro is designed to index a normal, page
sized array. KVM is a bit diffferent, as we can use concatenated
pages to have a bigger address space (for example 40bit IPA with
4kB pages gives us an 8kB PGD.

In the above case, the use of pgd_index will always return an index
inside the first 4kB, which makes a guest that has memory above
0x8000000000 rather unhappy, as it spins forever in a page fault,
whist the host happilly corrupts the lower pgd.

The obvious fix is to get our own kvm_pgd_index that does the right
thing(tm).

Tested on X-Gene with a hacked kvmtool that put memory at a stupidly
high address.
Reviewed-by: NChristoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NChristoffer Dall <christoffer.dall@linaro.org>

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.  It turns out that alloc_pages_exact() and
free_pages_exact() does exactly that.

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).

 [ Reworked to use alloc_pages_exact() and free_pages_exact() and to
   return pointers directly instead of by reference as arguments
    - Christoffer ]
Reported-by: NMark Rutland <mark.rutland@arm.com>
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NChristoffer Dall <christoffer.dall@linaro.org>

When handling a fault in stage-2, we need to resync I$ and D$, just
to be sure we don't leave any old cache line behind.

That's very good, except that we do so using the *user* address.
Under heavy load (swapping like crazy), we may end up in a situation
where the page gets mapped in stage-2 while being unmapped from
userspace by another CPU.

At that point, the DC/IC instructions can generate a fault, which
we handle with kvm->mmu_lock held. The box quickly deadlocks, user
is unhappy.

Instead, perform this invalidation through the kernel mapping,
which is guaranteed to be present. The box is much happier, and so
am I.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NChristoffer Dall <christoffer.dall@linaro.org>