Examine guest pagetable and bring the shadow back in sync. Caller is responsible
for local TLB flush before re-entering guest mode.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
There is not much point in write protecting large mappings. This
can only happen when a page is shadowed during the window between
is_largepage_backed and mmu_lock acquision. Zap the entry instead, so
the next pagefault will find a shadowed page via is_largepage_backed and
fallback to 4k translations.
Simplifies out of sync shadow.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Split the spte entry creation code into a new set_spte function.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Convert gfn_to_pfn to use get_user_pages_fast, which can do lockless
pagetable lookups on x86. Kernel compilation on 4-way guest is 3.7%
faster on VMX.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
EPT is 4 level by default in 32pae(48 bits), but the addr parameter
of kvm_shadow_walk->entry() only accept unsigned long as virtual
address, which is 32bit in 32pae. This result in SHADOW_PT_INDEX()
overflow when try to fetch level 4 index.
Fix it by extend kvm_shadow_walk->entry() to accept 64bit addr in
parameter.
Signed-off-by: Sheng Yang <sheng.yang@intel.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
The accessed bit was accidentally turned on in a random flag word, rather
than, the spte itself, which was lucky, since it used the non-EPT compatible
PT_ACCESSED_MASK.
Fix by turning the bit on in the spte and changing it to use the portable
accessed mask.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Otherwise, the cpu may allow writes to the tracked pages, and we lose
some display bits or fail to migrate correctly.
Signed-off-by: Avi Kivity <avi@qumranet.com>
We currently walk the shadow page tables in two places: direct map (for
real mode and two dimensional paging) and paging mode shadow. Since we
anticipate requiring a third walk (for invlpg), it makes sense to have
a generic facility for shadow walk.
This patch adds such a shadow walker, walks the page tables and calls a
method for every spte encountered. The method can examine the spte,
modify it, or even instantiate it. The walk can be aborted by returning
nonzero from the method.
Signed-off-by: Avi Kivity <avi@qumranet.com>
In all cases the shadow root level is available in mmu.shadow_root_level,
so there is no need to pass it as a parameter.
Signed-off-by: Avi Kivity <avi@qumranet.com>
The two paths are equivalent except for one argument, which is already
available. Merge the two codepaths.
Signed-off-by: Avi Kivity <avi@qumranet.com>
We're in a hot path. We can't use kmalloc() because
it might impact performance. So, we just stick the buffer that
we need into the kvm_vcpu_arch structure. This is used very
often, so it is not really a waste.
We also have to move the buffer structure's definition to the
arch-specific x86 kvm header.
Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
The twisty maze of conditionals can be reduced.
[joerg: fix tlb flushing]
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
As well as discard fake accessed bit and dirty bit of EPT.
Signed-off-by: Sheng Yang <sheng.yang@intel.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Synchronize changes to host virtual addresses which are part of
a KVM memory slot to the KVM shadow mmu. This allows pte operations
like swapping, page migration, and madvise() to transparently work
with KVM.
Signed-off-by: Andrea Arcangeli <andrea@qumranet.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
When an event (such as an interrupt) is injected, and the stack is
shadowed (and therefore write protected), the guest will exit. The
current code will see that the stack is shadowed and emulate a few
instructions, each time postponing the injection. Eventually the
injection may succeed, but at that time the guest may be unwilling
to accept the interrupt (for example, the TPR may have changed).
This occurs every once in a while during a Windows 2008 boot.
Fix by unshadowing the fault address if the fault was due to an event
injection.
Signed-off-by: Avi Kivity <avi@qumranet.com>
If NPT is enabled after loading both KVM modules on AMD and it should be
disabled, both KVM modules must be reloaded. If only the architecture module is
reloaded the behavior is undefined. With this patch it is possible to disable
NPT only by reloading the kvm_amd module.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
The direct mapped shadow code (used for real mode and two dimensional paging)
sets upper-level ptes using direct assignment rather than calling
set_shadow_pte(). A nonpae host will split this into two writes, which opens
up a race if another vcpu accesses the same memory area.
Fix by calling set_shadow_pte() instead of assigning directly.
Noticed by Izik Eidus.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Harden kvm_mmu_zap_page() against invalid root pages that
had been shadowed from memslots that are gone.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
kvm_mmu_zap_page() needs slots lock held (rmap_remove->gfn_to_memslot,
for example).
Since kvm_lock spinlock is held in mmu_shrink(), do a non-blocking
down_read_trylock().
Untested.
Signed-off-by: Avi Kivity <avi@qumranet.com>
SVM cannot benefit from page prefetching since guest page fault bypass
cannot by made to work there. Avoid accessing the guest page table in
this case.
Signed-off-by: Avi Kivity <avi@qumranet.com>
The KVM MMU tries to detect when a speculative pte update is not actually
used by demand fault, by checking the accessed bit of the shadow pte. If
the shadow pte has not been accessed, we deem that page table flooded and
remove the shadow page table, allowing further pte updates to proceed
without emulation.
However, if the pte itself points at a page table and only used for write
operations, the accessed bit will never be set since all access will happen
through the emulator.
This is exactly what happens with kscand on old (2.4.x) HIGHMEM kernels.
The kernel points a kmap_atomic() pte at a page table, and then
proceeds with read-modify-write operations to look at the dirty and accessed
bits. We get a false flood trigger on the kmap ptes, which results in the
mmu spending all its time setting up and tearing down shadows.
Fix by setting the shadow accessed bit on emulated accesses.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Noticed by sparse:
arch/x86/kvm/vmx.c:1583:6: warning: symbol 'vmx_disable_intercept_for_msr' was not declared. Should it be static?
arch/x86/kvm/x86.c:3406:5: warning: symbol 'kvm_task_switch_16' was not declared. Should it be static?
arch/x86/kvm/x86.c:3429:5: warning: symbol 'kvm_task_switch_32' was not declared. Should it be static?
arch/x86/kvm/mmu.c:1968:6: warning: symbol 'kvm_mmu_remove_one_alloc_mmu_page' was not declared. Should it be static?
arch/x86/kvm/mmu.c:2014:6: warning: symbol 'mmu_destroy_caches' was not declared. Should it be static?
arch/x86/kvm/lapic.c:862:5: warning: symbol 'kvm_lapic_get_base' was not declared. Should it be static?
arch/x86/kvm/i8254.c:94:5: warning: symbol 'pit_get_gate' was not declared. Should it be static?
arch/x86/kvm/i8254.c:196:5: warning: symbol '__pit_timer_fn' was not declared. Should it be static?
arch/x86/kvm/i8254.c:561:6: warning: symbol '__inject_pit_timer_intr' was not declared. Should it be static?
Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
KVM has a heuristic to unshadow guest pagetables when userspace accesses
them, on the assumption that most guests do not allow userspace to access
pagetables directly. Unfortunately, in addition to unshadowing the pagetables,
it also oopses.
This never triggers on ordinary guests since sane OSes will clear the
pagetables before assigning them to userspace, which will trigger the flood
heuristic, unshadowing the pagetables before the first userspace access. One
particular guest, though (Xenner) will run the kernel in userspace, triggering
the oops. Since the heuristic is incorrect in this case, we can simply
remove it.
Signed-off-by: Avi Kivity <avi@qumranet.com>
kvm_mmu_pte_write() does not handle 32-bit non-PAE large page backed
guests properly. It will instantiate two 2MB sptes pointing to the same
physical 2MB page when a guest large pte update is trapped.
Instead of duplicating code to handle this, disallow directory level
updates to happen through kvm_mmu_pte_write(), so the two 2MB sptes
emulating one guest 4MB pte can be correctly created by the page fault
handling path.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
rmap_next() does not work correctly after rmap_remove(), as it expects
the rmap chains not to change during iteration. Fix (for now) by restarting
iteration from the beginning.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Shadows for large guests can take a long time to tear down, so reschedule
occasionally to avoid softlockup warnings.
Signed-off-by: Avi Kivity <avi@qumranet.com>
nonpae guests can call rmap_write_protect twice per page (for page tables)
or four times per page (for page directories), triggering a bogus warning.
Remove the warning.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Currently EPT level is 4 for both pae and x86_64. The patch remove the #ifdef
for alloc root_hpa and free root_hpa to support EPT.
Signed-off-by: Sheng Yang <sheng.yang@intel.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
The function get_tdp_level() provided the number of tdp level for EPT and
NPT rather than the NPT specific macro.
Signed-off-by: Sheng Yang <sheng.yang@intel.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Move some definitions to mmu.h in order to allow building common table
entries between EPT and non-EPT.
Signed-off-by: Sheng Yang <sheng.yang@intel.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
kvm_pv_mmu_op should not take mmap_sem. All gfn_to_page() callers down
in the MMU processing will take it if necessary, so as it is it can
deadlock.
Apparently a leftover from the days before slots_lock.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
This patch introduces a gfn_to_pfn() function and corresponding functions like
kvm_release_pfn_dirty(). Using these new functions, we can modify the x86
MMU to no longer assume that it can always get a struct page for any given gfn.
We don't want to eliminate gfn_to_page() entirely because a number of places
assume they can do gfn_to_page() and then kmap() the results. When we support
IO memory, gfn_to_page() will fail for IO pages although gfn_to_pfn() will
succeed.
This does not implement support for avoiding reference counting for reserved
RAM or for IO memory. However, it should make those things pretty straight
forward.
Since we're only introducing new common symbols, I don't think it will break
the non-x86 architectures but I haven't tested those. I've tested Intel,
AMD, NPT, and hugetlbfs with Windows and Linux guests.
[avi: fix overflow when shifting left pfns by adding casts]
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Zdenek reported a bug where a looping "dmsetup status" eventually hangs
on SMP guests.
The problem is that kvm_mmu_get_page() prepopulates the shadow MMU
before write protecting the guest page tables. By doing so, it leaves a
window open where the guest can mark a pte as present while the host has
shadow cached such pte as "notrap". Accesses to such address will fault
in the guest without the host having a chance to fix the situation.
Fix by moving the write protection before the pte prefetch.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
If the accessed bit is not set, the guest has never accessed this page
(at least through this spte), so there's no need to mark the page
accessed. This provides more accurate data for the eviction algortithm.
Noted by Andrea Arcangeli.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Allow the Linux memory manager to reclaim memory in the kvm shadow cache.
Signed-off-by: Izik Eidus <izike@qumranet.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Unify slots_lock acquision around vcpu_run(). This is simpler and less
error-prone.
Also fix some callsites that were not grabbing the lock properly.
[avi: drop slots_lock while in guest mode to avoid holding the lock
for indefinite periods]
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
When mmu_set_spte() checks if a page related to spte should be release as
dirty or clean, it check if the shadow pte was writeble, but in case
rmap_write_protect() is called called it is possible for shadow ptes that were
writeble to become readonly and therefor mmu_set_spte will release the pages
as clean.
This patch fix this issue by marking the page as dirty inside
rmap_write_protect().
Signed-off-by: Izik Eidus <izike@qumranet.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
If we populate a shadow pte due to a fault (and not speculatively due to a
pte write) then we can set the accessed bit on it, as we know it will be
set immediately on the next guest instruction. This saves a read-modify-write
operation.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Hypercall based pte updates are faster than faults, and also allow use
of the lazy MMU mode to batch operations.
Don't report the feature if two dimensional paging is enabled.
[avi:
- one mmu_op hypercall instead of one per op
- allow 64-bit gpa on hypercall
- don't pass host errors (-ENOMEM) to guest]
[akpm: warning fix on i386]
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Create large pages mappings if the guest PTE's are marked as such and
the underlying memory is hugetlbfs backed. If the largepage contains
write-protected pages, a large pte is not used.
Gives a consistent 2% improvement for data copies on ram mounted
filesystem, without NPT/EPT.
Anthony measures a 4% improvement on 4-way kernbench, with NPT.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Mark zapped root pagetables as invalid and ignore such pages during lookup.
This is a problem with the cr3-target feature, where a zapped root table fools
the faulting code into creating a read-only mapping. The result is a lockup
if the instruction can't be emulated.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
This patch contains the changes to the KVM MMU necessary for support of the
Nested Paging feature in AMD Barcelona and Phenom Processors.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
The mapping function for the nonpaging case in the softmmu does basically the
same as required for Nested Paging. Make this function generic so it can be
used for both.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
The generic x86 code has to know if the specific implementation uses Nested
Paging. In the generic code Nested Paging is called Two Dimensional Paging
(TDP) to avoid confusion with (future) TDP implementations of other vendors.
This patch exports the availability of TDP to the generic x86 code.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Currently an mmio guest pte is encoded in the shadow pagetable as a
not-present trapping pte, with the SHADOW_IO_MARK bit set. However
nothing is ever done with this information, so maintaining it is a
useless complication.
This patch moves the check for mmio to before shadow ptes are instantiated,
so the shadow code is never invoked for ptes that reference mmio. The code
is simpler, and with future work, can be made to handle mmio concurrently.
Signed-off-by: Avi Kivity <avi@qumranet.com>
A guest partial guest pte write will leave shadow_trap_nonpresent_pte
in spte, which generates a vmexit at the next guest access through that pte.
This patch improves this by reading the full guest pte in advance and thus
being able to update the spte and eliminate the vmexit.
This helps pae guests which use two 32-bit writes to set a single 64-bit pte.
[truncation fix by Eric]
Signed-off-by: Yaozu (Eddie) Dong <eddie.dong@intel.com>
Signed-off-by: Feng (Eric) Liu <eric.e.liu@intel.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
While backporting 72dc67a696, a gfn_to_page()
call was duplicated instead of moved (due to an unrelated patch not being
present in mainline). This caused a page reference leak, resulting in a
fairly massive memory leak.
Fix by removing the extraneous gfn_to_page() call.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Do not assume that a shadow mapping will always point to the same host
frame number. Fixes crash with madvise(MADV_DONTNEED).
[avi: move after first printk(), add another printk()]
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
For improved concurrency, the guest walk is performed concurrently with other
vcpus. This means that we need to revalidate the guest ptes once we have
write-protected the guest page tables, at which point they can no longer be
modified.
The current code attempts to avoid this check if the shadow page table is not
new, on the assumption that if it has existed before, the guest could not have
modified the pte without the shadow lock. However the assumption is incorrect,
as the racing vcpu could have modified the pte, then instantiated the shadow
page, before our vcpu regains control:
vcpu0 vcpu1
fault
walk pte
modify pte
fault in same pagetable
instantiate shadow page
lookup shadow page
conclude it is old
instantiate spte based on stale guest pte
We could do something clever with generation counters, but a test run by
Marcelo suggests this is unnecessary and we can just do the revalidation
unconditionally. The pte will be in the processor cache and the check can
be quite fast.
Signed-off-by: Avi Kivity <avi@qumranet.com>
the cr3 variable is now inside the vcpu->arch structure.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
This patch replaces the mmap_sem lock for the memory slots with a new
kvm private lock, it is needed beacuse untill now there were cases where
kvm accesses user memory while holding the mmap semaphore.
Signed-off-by: Izik Eidus <izike@qumranet.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Right now rmap_remove won't set the page as dirty if the shadow pte
pointed to this page had write access and then it became readonly.
This patches fixes that, by setting the page as dirty for spte changes from
write to readonly access.
Signed-off-by: Izik Eidus <izike@qumranet.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
If some other cpu steals mmu pages between our check and an attempt to
allocate, we can run out of mmu pages. Fix by moving the check into the
same critical section as the allocation.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Convert the synchronization of the shadow handling to a separate mmu_lock
spinlock.
Also guard fetch() by mmap_sem in read-mode to protect against alias
and memslot changes.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Since gfn_to_page() is a sleeping function, and we want to make the core mmu
spinlocked, we need to pass the page from the walker context (which can sleep)
to the shadow context (which cannot).
[marcelo: avoid recursive locking of mmap_sem]
Signed-off-by: Avi Kivity <avi@qumranet.com>
Do not hold kvm->lock mutex across the entire pagefault code,
only acquire it in places where it is necessary, such as mmu
hash list, active list, rmap and parent pte handling.
Allow concurrent guest walkers by switching walk_addr() to use
mmap_sem in read-mode.
And get rid of the lockless __gfn_to_page.
[avi: move kvm_mmu_pte_write() locking inside the function]
[avi: add locking for real mode]
[avi: fix cmpxchg locking]
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Host side TLB flush can be merged together if multiple
spte need to be write-protected.
Signed-off-by: Yaozu (Eddie) Dong <eddie.dong@intel.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>
This paves the way for multiple architecture support. Note that while
ioapic.c could potentially be shared with ia64, it is also moved.
Signed-off-by: Avi Kivity <avi@qumranet.com>
Marcelo Tosatti