cascardo/linux.git
9 years agokvm: vmx: Implement set_apic_access_page_addr
Tang Chen [Wed, 24 Sep 2014 07:57:54 +0000 (15:57 +0800)]
kvm: vmx: Implement set_apic_access_page_addr

Currently, the APIC access page is pinned by KVM for the entire life
of the guest.  We want to make it migratable in order to make memory
hot-unplug available for machines that run KVM.

This patch prepares to handle this for the case where there is no nested
virtualization, or where the nested guest does not have an APIC page of
its own.  All accesses to kvm->arch.apic_access_page are changed to go
through kvm_vcpu_reload_apic_access_page.

If the APIC access page is invalidated when the host is running, we update
the VMCS in the next guest entry.

If it is invalidated when the guest is running, the MMU notifier will force
an exit, after which we will handle everything as in the previous case.

If it is invalidated when a nested guest is running, the request will update
either the VMCS01 or the VMCS02.  Updating the VMCS01 is done at the
next L2->L1 exit, while updating the VMCS02 is done in prepare_vmcs02.

Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agokvm: x86: Add request bit to reload APIC access page address
Tang Chen [Wed, 24 Sep 2014 07:57:54 +0000 (15:57 +0800)]
kvm: x86: Add request bit to reload APIC access page address

Currently, the APIC access page is pinned by KVM for the entire life
of the guest.  We want to make it migratable in order to make memory
hot-unplug available for machines that run KVM.

This patch prepares to handle this in generic code, through a new
request bit (that will be set by the MMU notifier) and a new hook
that is called whenever the request bit is processed.

Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agokvm: Add arch specific mmu notifier for page invalidation
Tang Chen [Wed, 24 Sep 2014 07:57:57 +0000 (15:57 +0800)]
kvm: Add arch specific mmu notifier for page invalidation

This will be used to let the guest run while the APIC access page is
not pinned.  Because subsequent patches will fill in the function
for x86, place the (still empty) x86 implementation in the x86.c file
instead of adding an inline function in kvm_host.h.

Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agokvm: Rename make_all_cpus_request() to kvm_make_all_cpus_request() and make it non...
Tang Chen [Wed, 24 Sep 2014 07:57:55 +0000 (15:57 +0800)]
kvm: Rename make_all_cpus_request() to kvm_make_all_cpus_request() and make it non-static

Different architectures need different requests, and in fact we
will use this function in architecture-specific code later. This
will be outside kvm_main.c, so make it non-static and rename it to
kvm_make_all_cpus_request().

Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agokvm: Fix page ageing bugs
Andres Lagar-Cavilla [Mon, 22 Sep 2014 21:54:42 +0000 (14:54 -0700)]
kvm: Fix page ageing bugs

1. We were calling clear_flush_young_notify in unmap_one, but we are
within an mmu notifier invalidate range scope. The spte exists no more
(due to range_start) and the accessed bit info has already been
propagated (due to kvm_pfn_set_accessed). Simply call
clear_flush_young.

2. We clear_flush_young on a primary MMU PMD, but this may be mapped
as a collection of PTEs by the secondary MMU (e.g. during log-dirty).
This required expanding the interface of the clear_flush_young mmu
notifier, so a lot of code has been trivially touched.

3. In the absence of shadow_accessed_mask (e.g. EPT A bit), we emulate
the access bit by blowing the spte. This requires proper synchronizing
with MMU notifier consumers, like every other removal of spte's does.

Signed-off-by: Andres Lagar-Cavilla <andreslc@google.com>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agokvm/x86/mmu: Pass gfn and level to rmapp callback.
Andres Lagar-Cavilla [Tue, 23 Sep 2014 19:34:54 +0000 (12:34 -0700)]
kvm/x86/mmu: Pass gfn and level to rmapp callback.

Callbacks don't have to do extra computation to learn what the caller
(lvm_handle_hva_range()) knows very well. Useful for
debugging/tracing/printk/future.

Signed-off-by: Andres Lagar-Cavilla <andreslc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agox86: kvm: use alternatives for VMCALL vs. VMMCALL if kernel text is read-only
Paolo Bonzini [Mon, 22 Sep 2014 11:17:48 +0000 (13:17 +0200)]
x86: kvm: use alternatives for VMCALL vs. VMMCALL if kernel text is read-only

On x86_64, kernel text mappings are mapped read-only with CONFIG_DEBUG_RODATA.
In that case, KVM will fail to patch VMCALL instructions to VMMCALL
as required on AMD processors.

The failure mode is currently a divide-by-zero exception, which obviously
is a KVM bug that has to be fixed.  However, picking the right instruction
between VMCALL and VMMCALL will be faster and will help if you cannot upgrade
the hypervisor.

Reported-by: Chris Webb <chris@arachsys.com>
Tested-by: Chris Webb <chris@arachsys.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: x86@kernel.org
Acked-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agokvm: x86: use macros to compute bank MSRs
Chen Yucong [Tue, 23 Sep 2014 02:44:35 +0000 (10:44 +0800)]
kvm: x86: use macros to compute bank MSRs

Avoid open coded calculations for bank MSRs by using well-defined
macros that hide the index of higher bank MSRs.

No semantic changes.

Signed-off-by: Chen Yucong <slaoub@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86: Remove debug assertion of non-PAE reserved bits
Nadav Amit [Tue, 23 Sep 2014 07:01:57 +0000 (10:01 +0300)]
KVM: x86: Remove debug assertion of non-PAE reserved bits

Commit 346874c9507a ("KVM: x86: Fix CR3 reserved bits") removed non-PAE
reserved bits which were not according to Intel SDM.  However, residue was left
in a debug assertion (CR3_NONPAE_RESERVED_BITS).  Remove it.

Signed-off-by: Nadav Amit <namit@cs.technion.ac.il>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agokvm: don't take vcpu mutex for obviously invalid vcpu ioctls
David Matlack [Fri, 19 Sep 2014 23:03:25 +0000 (16:03 -0700)]
kvm: don't take vcpu mutex for obviously invalid vcpu ioctls

vcpu ioctls can hang the calling thread if issued while a vcpu is running.
However, invalid ioctls can happen when userspace tries to probe the kind
of file descriptors (e.g. isatty() calls ioctl(TCGETS)); in that case,
we know the ioctl is going to be rejected as invalid anyway and we can
fail before trying to take the vcpu mutex.

This patch does not change functionality, it just makes invalid ioctls
fail faster.

Cc: stable@vger.kernel.org
Signed-off-by: David Matlack <dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agokvm: Faults which trigger IO release the mmap_sem
Andres Lagar-Cavilla [Wed, 17 Sep 2014 17:51:48 +0000 (10:51 -0700)]
kvm: Faults which trigger IO release the mmap_sem

When KVM handles a tdp fault it uses FOLL_NOWAIT. If the guest memory
has been swapped out or is behind a filemap, this will trigger async
readahead and return immediately. The rationale is that KVM will kick
back the guest with an "async page fault" and allow for some other
guest process to take over.

If async PFs are enabled the fault is retried asap from an async
workqueue. If not, it's retried immediately in the same code path. In
either case the retry will not relinquish the mmap semaphore and will
block on the IO. This is a bad thing, as other mmap semaphore users
now stall as a function of swap or filemap latency.

This patch ensures both the regular and async PF path re-enter the
fault allowing for the mmap semaphore to be relinquished in the case
of IO wait.

Reviewed-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Andres Lagar-Cavilla <andreslc@google.com>
Acked-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agokvm: x86: fix two typos in comment
Tiejun Chen [Mon, 22 Sep 2014 02:31:38 +0000 (10:31 +0800)]
kvm: x86: fix two typos in comment

s/drity/dirty and s/vmsc01/vmcs01

Signed-off-by: Tiejun Chen <tiejun.chen@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: vmx: Inject #GP on invalid PAT CR
Nadav Amit [Thu, 18 Sep 2014 19:39:44 +0000 (22:39 +0300)]
KVM: vmx: Inject #GP on invalid PAT CR

Guest which sets the PAT CR to invalid value should get a #GP.  Currently, if
vmx supports loading PAT CR during entry, then the value is not checked.  This
patch makes the required check in that case.

Signed-off-by: Nadav Amit <namit@cs.technion.ac.il>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86: emulating descriptor load misses long-mode case
Nadav Amit [Thu, 18 Sep 2014 19:39:43 +0000 (22:39 +0300)]
KVM: x86: emulating descriptor load misses long-mode case

In 64-bit mode a #GP should be delivered to the guest "if the code segment
descriptor pointed to by the selector in the 64-bit gate doesn't have the L-bit
set and the D-bit clear." - Intel SDM "Interrupt 13—General Protection
Exception (#GP)".

This patch fixes the behavior of CS loading emulation code. Although the
comment says that segment loading is not supported in long mode, this function
is executed in long mode, so the fix is necassary.

Signed-off-by: Nadav Amit <namit@cs.technion.ac.il>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86: directly use kvm_make_request again
Liang Chen [Thu, 18 Sep 2014 16:38:37 +0000 (12:38 -0400)]
KVM: x86: directly use kvm_make_request again

A one-line wrapper around kvm_make_request is not particularly
useful. Replace kvm_mmu_flush_tlb() with kvm_make_request().

Signed-off-by: Liang Chen <liangchen.linux@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86: count actual tlb flushes
Radim Krčmář [Thu, 18 Sep 2014 16:38:36 +0000 (12:38 -0400)]
KVM: x86: count actual tlb flushes

- we count KVM_REQ_TLB_FLUSH requests, not actual flushes
  (KVM can have multiple requests for one flush)
- flushes from kvm_flush_remote_tlbs aren't counted
- it's easy to make a direct request by mistake

Solve these by postponing the counting to kvm_check_request().

Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Liang Chen <liangchen.linux@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: nested VMX: disable perf cpuid reporting
Marcelo Tosatti [Thu, 18 Sep 2014 21:24:57 +0000 (18:24 -0300)]
KVM: nested VMX: disable perf cpuid reporting

Initilization of L2 guest with -cpu host, on L1 guest with -cpu host
triggers:

(qemu) KVM: entry failed, hardware error 0x7
...
nested_vmx_run: VMCS MSR_{LOAD,STORE} unsupported

Nested VMX MSR load/store support is not sufficient to
allow perf for L2 guest.

Until properly fixed, trap CPUID and disable function 0xA.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86: Don't report guest userspace emulation error to userspace
Nadav Amit [Tue, 16 Sep 2014 23:50:50 +0000 (02:50 +0300)]
KVM: x86: Don't report guest userspace emulation error to userspace

Commit fc3a9157d314 ("KVM: X86: Don't report L2 emulation failures to
user-space") disabled the reporting of L2 (nested guest) emulation failures to
userspace due to race-condition between a vmexit and the instruction emulator.
The same rational applies also to userspace applications that are permitted by
the guest OS to access MMIO area or perform PIO.

This patch extends the current behavior - of injecting a #UD instead of
reporting it to userspace - also for guest userspace code.

Signed-off-by: Nadav Amit <namit@cs.technion.ac.il>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agokvm: Make init_rmode_tss() return 0 on success.
Paolo Bonzini [Tue, 16 Sep 2014 11:37:40 +0000 (13:37 +0200)]
kvm: Make init_rmode_tss() return 0 on success.

In init_rmode_tss(), there two variables indicating the return
value, r and ret, and it return 0 on error, 1 on success. The function
is only called by vmx_set_tss_addr(), and ret is redundant.

This patch removes the redundant variable, by making init_rmode_tss()
return 0 on success, -errno on failure.

Reviewed-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86: Warn if guest virtual address space is not 48-bits
Nadav Amit [Tue, 16 Sep 2014 12:10:03 +0000 (15:10 +0300)]
KVM: x86: Warn if guest virtual address space is not 48-bits

The KVM emulator code assumes that the guest virtual address space (in 64-bit)
is 48-bits wide.  Fail the KVM_SET_CPUID and KVM_SET_CPUID2 ioctl if
userspace tries to create a guest that does not obey this restriction.

Signed-off-by: Nadav Amit <namit@cs.technion.ac.il>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agokvm-vfio: do not use module_init
Paolo Bonzini [Wed, 24 Sep 2014 11:02:46 +0000 (13:02 +0200)]
kvm-vfio: do not use module_init

/me got confused between the kernel and QEMU.  In the kernel, you can
only have one module_init function, and it will prevent unloading the
module unless you also have the corresponding module_exit function.

So, commit 80ce1639727e (KVM: VFIO: register kvm_device_ops dynamically,
2014-09-02) broke unloading of the kvm module, by adding a module_init
function and no module_exit.

Repair it by making kvm_vfio_ops_init weak, and checking it in
kvm_init.

Cc: Will Deacon <will.deacon@arm.com>
Cc: Gleb Natapov <gleb@kernel.org>
Cc: Alex Williamson <Alex.Williamson@redhat.com>
Fixes: 80ce1639727e9d38729c34f162378508c307ca25
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: EVENTFD: Remove inclusion of irq.h
Christoffer Dall [Mon, 22 Sep 2014 21:33:08 +0000 (23:33 +0200)]
KVM: EVENTFD: Remove inclusion of irq.h

Commit c77dcac (KVM: Move more code under CONFIG_HAVE_KVM_IRQFD) added
functionality that depends on definitions in ioapic.h when
__KVM_HAVE_IOAPIC is defined.

At the same time, kvm-arm commit 0ba0951 (KVM: EVENTFD: remove inclusion
of irq.h) removed the inclusion of irq.h, an architecture-specific header
that is not present on ARM but which happened to include ioapic.h on x86.

Include ioapic.h directly in eventfd.c if __KVM_HAVE_IOAPIC is defined.
This fixes x86 and lets ARM use eventfd.c.

Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agokvm: Make init_rmode_identity_map() return 0 on success.
Tang Chen [Tue, 16 Sep 2014 10:41:59 +0000 (18:41 +0800)]
kvm: Make init_rmode_identity_map() return 0 on success.

In init_rmode_identity_map(), there two variables indicating the return
value, r and ret, and it return 0 on error, 1 on success. The function
is only called by vmx_create_vcpu(), and ret is redundant.

This patch removes the redundant variable, and makes init_rmode_identity_map()
return 0 on success, -errno on failure.

Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agokvm: Remove ept_identity_pagetable from struct kvm_arch.
Tang Chen [Tue, 16 Sep 2014 10:41:58 +0000 (18:41 +0800)]
kvm: Remove ept_identity_pagetable from struct kvm_arch.

kvm_arch->ept_identity_pagetable holds the ept identity pagetable page. But
it is never used to refer to the page at all.

In vcpu initialization, it indicates two things:
1. indicates if ept page is allocated
2. indicates if a memory slot for identity page is initialized

Actually, kvm_arch->ept_identity_pagetable_done is enough to tell if the ept
identity pagetable is initialized. So we can remove ept_identity_pagetable.

NOTE: In the original code, ept identity pagetable page is pinned in memroy.
      As a result, it cannot be migrated/hot-removed. After this patch, since
      kvm_arch->ept_identity_pagetable is removed, ept identity pagetable page
      is no longer pinned in memory. And it can be migrated/hot-removed.

Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Reviewed-by: Gleb Natapov <gleb@kernel.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: VFIO: register kvm_device_ops dynamically
Will Deacon [Tue, 2 Sep 2014 09:27:36 +0000 (10:27 +0100)]
KVM: VFIO: register kvm_device_ops dynamically

Now that we have a dynamic means to register kvm_device_ops, use that
for the VFIO kvm device, instead of relying on the static table.

This is achieved by a module_init call to register the ops with KVM.

Cc: Gleb Natapov <gleb@kernel.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Acked-by: Alex Williamson <Alex.Williamson@redhat.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: s390: register flic ops dynamically
Cornelia Huck [Tue, 2 Sep 2014 09:27:35 +0000 (10:27 +0100)]
KVM: s390: register flic ops dynamically

Using the new kvm_register_device_ops() interface makes us get rid of
an #ifdef in common code.

Cc: Gleb Natapov <gleb@kernel.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: ARM: vgic: register kvm_device_ops dynamically
Will Deacon [Tue, 2 Sep 2014 09:27:34 +0000 (10:27 +0100)]
KVM: ARM: vgic: register kvm_device_ops dynamically

Now that we have a dynamic means to register kvm_device_ops, use that
for the ARM VGIC, instead of relying on the static table.

Cc: Gleb Natapov <gleb@kernel.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: device: add simple registration mechanism for kvm_device_ops
Will Deacon [Tue, 2 Sep 2014 09:27:33 +0000 (10:27 +0100)]
KVM: device: add simple registration mechanism for kvm_device_ops

kvm_ioctl_create_device currently has knowledge of all the device types
and their associated ops. This is fairly inflexible when adding support
for new in-kernel device emulations, so move what we currently have out
into a table, which can support dynamic registration of ops by new
drivers for virtual hardware.

Cc: Alex Williamson <Alex.Williamson@redhat.com>
Cc: Alex Graf <agraf@suse.de>
Cc: Gleb Natapov <gleb@kernel.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agokvm: ioapic: conditionally delay irq delivery duringeoi broadcast
Zhang Haoyu [Thu, 11 Sep 2014 08:47:04 +0000 (16:47 +0800)]
kvm: ioapic: conditionally delay irq delivery duringeoi broadcast

Currently, we call ioapic_service() immediately when we find the irq is still
active during eoi broadcast. But for real hardware, there's some delay between
the EOI writing and irq delivery.  If we do not emulate this behavior, and
re-inject the interrupt immediately after the guest sends an EOI and re-enables
interrupts, a guest might spend all its time in the ISR if it has a broken
handler for a level-triggered interrupt.

Such livelock actually happens with Windows guests when resuming from
hibernation.

As there's no way to recognize the broken handle from new raised ones, this patch
delays an interrupt if 10.000 consecutive EOIs found that the interrupt was
still high.  The guest can then make a little forward progress, until a proper
IRQ handler is set or until some detection routine in the guest (such as
Linux's note_interrupt()) recognizes the situation.

Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Zhang Haoyu <zhanghy@sangfor.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86: Use kvm_make_request when applicable
Guo Hui Liu [Fri, 12 Sep 2014 05:43:19 +0000 (13:43 +0800)]
KVM: x86: Use kvm_make_request when applicable

This patch replace the set_bit method by kvm_make_request
to make code more readable and consistent.

Signed-off-by: Guo Hui Liu <liuguohui@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86: make apic_accept_irq tracepoint more generic
Paolo Bonzini [Thu, 11 Sep 2014 09:51:02 +0000 (11:51 +0200)]
KVM: x86: make apic_accept_irq tracepoint more generic

Initially the tracepoint was added only to the APIC_DM_FIXED case,
also because it reported coalesced interrupts that only made sense
for that case.  However, the coalesced argument is not used anymore
and tracing other delivery modes is useful, so hoist the call out
of the switch statement.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agokvm: Use APIC_DEFAULT_PHYS_BASE macro as the apic access page address.
Tang Chen [Thu, 11 Sep 2014 05:38:00 +0000 (13:38 +0800)]
kvm: Use APIC_DEFAULT_PHYS_BASE macro as the apic access page address.

We have APIC_DEFAULT_PHYS_BASE defined as 0xfee00000, which is also the address of
apic access page. So use this macro.

Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Reviewed-by: Gleb Natapov <gleb@kernel.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoMerge tag 'kvm-s390-next-20140910' of git://git.kernel.org/pub/scm/linux/kernel/git...
Paolo Bonzini [Thu, 11 Sep 2014 09:09:33 +0000 (11:09 +0200)]
Merge tag 'kvm-s390-next-20140910' of git://git./linux/kernel/git/kvms390/linux into kvm-next

KVM: s390: Fixes and features for next (3.18)

1. Crypto/CPACF support: To enable the MSA4 instructions we have to
   provide a common control structure for each SIE control block
2. Two cleanups found by a static code checker: one redundant assignment
   and one useless if
3. Fix the page handling of the diag10 ballooning interface. If the
   guest freed the pages at absolute 0 some checks and frees were
   incorrect
4. Limit guests to 16TB
5. Add __must_check to interrupt injection code

9 years agoKVM: s390/interrupt: remove double assignment
Christian Borntraeger [Wed, 3 Sep 2014 14:16:47 +0000 (16:16 +0200)]
KVM: s390/interrupt: remove double assignment

r is already initialized to 0.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
9 years agoKVM: s390/cmm: Fix prefix handling for diag 10 balloon
Christian Borntraeger [Wed, 3 Sep 2014 19:23:13 +0000 (21:23 +0200)]
KVM: s390/cmm: Fix prefix handling for diag 10 balloon

The old handling of prefix pages was broken in the diag10 ballooner.
We now rely on gmap_discard to check for start > end and do a
slow path if the prefix swap pages are affected:
1. discard the pages from start to prefix
2. discard the absolute 0 pages
3. discard the pages after prefix swap to end

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
9 years agoKVM: s390: get rid of constant condition in ipte_unlock_simple
Christian Borntraeger [Wed, 3 Sep 2014 19:17:03 +0000 (21:17 +0200)]
KVM: s390: get rid of constant condition in ipte_unlock_simple

Due to the earlier check we know that ipte_lock_count must be 0.
No need to add a useless if. Let's make clear that we are going
to always wakeup when we execute that code.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Acked-by: Heiko Carstens <heiko.carstens@de.ibm.com>
9 years agoKVM: s390: unintended fallthrough for external call
Christian Borntraeger [Wed, 3 Sep 2014 14:21:32 +0000 (16:21 +0200)]
KVM: s390: unintended fallthrough for external call

We must not fallthrough if the conditions for external call are not met.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Cc: stable@vger.kernel.org
9 years agoKVM: s390: Limit guest size to 16TB
Christian Borntraeger [Mon, 25 Aug 2014 10:38:57 +0000 (12:38 +0200)]
KVM: s390: Limit guest size to 16TB

Currently we fill up a full 5 level page table to hold the guest
mapping. Since commit "support gmap page tables with less than 5
levels" we can do better.
Having more than 4 TB might be useful for some testing scenarios,
so let's just limit ourselves to 16TB guest size.
Having more than that is totally untested as I do not have enough
swap space/memory.

We continue to allow ucontrol the full size.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
9 years agoKVM: s390: add __must_check to interrupt deliver functions
Christian Borntraeger [Mon, 25 Aug 2014 10:27:29 +0000 (12:27 +0200)]
KVM: s390: add __must_check to interrupt deliver functions

We now propagate interrupt injection errors back to the ioctl. We
should mark functions that might fail with __must_check.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Acked-by: Jens Freimann <jfrei@linux.vnet.ibm.com>
9 years agoKVM: CPACF: Enable MSA4 instructions for kvm guest
Tony Krowiak [Fri, 27 Jun 2014 18:46:01 +0000 (14:46 -0400)]
KVM: CPACF: Enable MSA4 instructions for kvm guest

We have to provide a per guest crypto block for the CPUs to
enable MSA4 instructions. According to icainfo on z196 or
later this enables CCM-AES-128, CMAC-AES-128, CMAC-AES-192
and CMAC-AES-256.

Signed-off-by: Tony Krowiak <akrowiak@linux.vnet.ibm.com>
Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: Michael Mueller <mimu@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
[split MSA4/protected key into two patches]

9 years agoKVM: fix api documentation of KVM_GET_EMULATED_CPUID
Alex Bennée [Tue, 9 Sep 2014 16:27:19 +0000 (17:27 +0100)]
KVM: fix api documentation of KVM_GET_EMULATED_CPUID

It looks like when this was initially merged it got accidentally included
in the following section. I've just moved it back in the correct section
and re-numbered it as other ioctls have been added since.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Acked-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: document KVM_SET_GUEST_DEBUG api
Alex Bennée [Tue, 9 Sep 2014 16:27:18 +0000 (17:27 +0100)]
KVM: document KVM_SET_GUEST_DEBUG api

In preparation for working on the ARM implementation I noticed the debug
interface was missing from the API document. I've pieced together the
expected behaviour from the code and commit messages written it up as
best I can.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: remove redundant assignments in __kvm_set_memory_region
Christian Borntraeger [Thu, 4 Sep 2014 19:13:33 +0000 (21:13 +0200)]
KVM: remove redundant assignments in __kvm_set_memory_region

__kvm_set_memory_region sets r to EINVAL very early.
Doing it again is not necessary. The same is true later on, where
r is assigned -ENOMEM twice.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: remove redundant assigment of return value in kvm_dev_ioctl
Christian Borntraeger [Thu, 4 Sep 2014 19:13:32 +0000 (21:13 +0200)]
KVM: remove redundant assigment of return value in kvm_dev_ioctl

The first statement of kvm_dev_ioctl is
        long r = -EINVAL;

No need to reassign the same value.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: remove redundant check of in_spin_loop
Christian Borntraeger [Thu, 4 Sep 2014 19:13:31 +0000 (21:13 +0200)]
KVM: remove redundant check of in_spin_loop

The expression `vcpu->spin_loop.in_spin_loop' is always true,
because it is evaluated only when the condition
`!vcpu->spin_loop.in_spin_loop' is false.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86: propagate exception from permission checks on the nested page fault
Paolo Bonzini [Tue, 2 Sep 2014 11:23:06 +0000 (13:23 +0200)]
KVM: x86: propagate exception from permission checks on the nested page fault

Currently, if a permission error happens during the translation of
the final GPA to HPA, walk_addr_generic returns 0 but does not fill
in walker->fault.  To avoid this, add an x86_exception* argument
to the translate_gpa function, and let it fill in walker->fault.
The nested_page_fault field will be true, since the walk_mmu is the
nested_mmu and translate_gpu instead operates on the "outer" (NPT)
instance.

Reported-by: Valentine Sinitsyn <valentine.sinitsyn@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86: skip writeback on injection of nested exception
Paolo Bonzini [Thu, 4 Sep 2014 17:46:15 +0000 (19:46 +0200)]
KVM: x86: skip writeback on injection of nested exception

If a nested page fault happens during emulation, we will inject a vmexit,
not a page fault.  However because writeback happens after the injection,
we will write ctxt->eip from L2 into the L1 EIP.  We do not write back
if an instruction caused an interception vmexit---do the same for page
faults.

Suggested-by: Gleb Natapov <gleb@kernel.org>
Reviewed-by: Gleb Natapov <gleb@kernel.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: nSVM: propagate the NPF EXITINFO to the guest
Paolo Bonzini [Tue, 2 Sep 2014 11:18:37 +0000 (13:18 +0200)]
KVM: nSVM: propagate the NPF EXITINFO to the guest

This is similar to what the EPT code does with the exit qualification.
This allows the guest to see a valid value for bits 33:32.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86: reserve bit 8 of non-leaf PDPEs and PML4Es in 64-bit mode on AMD
Paolo Bonzini [Tue, 2 Sep 2014 11:24:12 +0000 (13:24 +0200)]
KVM: x86: reserve bit 8 of non-leaf PDPEs and PML4Es in 64-bit mode on AMD

Bit 8 would be the "global" bit, which does not quite make sense for non-leaf
page table entries.  Intel ignores it; AMD ignores it in PDEs, but reserves it
in PDPEs and PML4Es.  The SVM test is relying on this behavior, so enforce it.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: mmio: cleanup kvm_set_mmio_spte_mask
Tiejun Chen [Mon, 1 Sep 2014 10:44:04 +0000 (18:44 +0800)]
KVM: mmio: cleanup kvm_set_mmio_spte_mask

Just reuse rsvd_bits() inside kvm_set_mmio_spte_mask()
for slightly better code.

Signed-off-by: Tiejun Chen <tiejun.chen@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agokvm: x86: fix stale mmio cache bug
David Matlack [Mon, 18 Aug 2014 22:46:07 +0000 (15:46 -0700)]
kvm: x86: fix stale mmio cache bug

The following events can lead to an incorrect KVM_EXIT_MMIO bubbling
up to userspace:

(1) Guest accesses gpa X without a memory slot. The gfn is cached in
struct kvm_vcpu_arch (mmio_gfn). On Intel EPT-enabled hosts, KVM sets
the SPTE write-execute-noread so that future accesses cause
EPT_MISCONFIGs.

(2) Host userspace creates a memory slot via KVM_SET_USER_MEMORY_REGION
covering the page just accessed.

(3) Guest attempts to read or write to gpa X again. On Intel, this
generates an EPT_MISCONFIG. The memory slot generation number that
was incremented in (2) would normally take care of this but we fast
path mmio faults through quickly_check_mmio_pf(), which only checks
the per-vcpu mmio cache. Since we hit the cache, KVM passes a
KVM_EXIT_MMIO up to userspace.

This patch fixes the issue by using the memslot generation number
to validate the mmio cache.

Cc: stable@vger.kernel.org
Signed-off-by: David Matlack <dmatlack@google.com>
[xiaoguangrong: adjust the code to make it simpler for stable-tree fix.]
Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Reviewed-by: David Matlack <dmatlack@google.com>
Reviewed-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Tested-by: David Matlack <dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agokvm: fix potentially corrupt mmio cache
David Matlack [Mon, 18 Aug 2014 22:46:06 +0000 (15:46 -0700)]
kvm: fix potentially corrupt mmio cache

vcpu exits and memslot mutations can run concurrently as long as the
vcpu does not aquire the slots mutex. Thus it is theoretically possible
for memslots to change underneath a vcpu that is handling an exit.

If we increment the memslot generation number again after
synchronize_srcu_expedited(), vcpus can safely cache memslot generation
without maintaining a single rcu_dereference through an entire vm exit.
And much of the x86/kvm code does not maintain a single rcu_dereference
of the current memslots during each exit.

We can prevent the following case:

   vcpu (CPU 0)                             | thread (CPU 1)
--------------------------------------------+--------------------------
1  vm exit                                  |
2  srcu_read_unlock(&kvm->srcu)             |
3  decide to cache something based on       |
     old memslots                           |
4                                           | change memslots
                                            | (increments generation)
5                                           | synchronize_srcu(&kvm->srcu);
6  retrieve generation # from new memslots  |
7  tag cache with new memslot generation    |
8  srcu_read_unlock(&kvm->srcu)             |
...                                         |
   <action based on cache occurs even       |
    though the caching decision was based   |
    on the old memslots>                    |
...                                         |
   <action *continues* to occur until next  |
    memslot generation change, which may    |
    be never>                               |
                                            |

By incrementing the generation after synchronizing with kvm->srcu readers,
we ensure that the generation retrieved in (6) will become invalid soon
after (8).

Keeping the existing increment is not strictly necessary, but we
do keep it and just move it for consistency from update_memslots to
install_new_memslots.  It invalidates old cached MMIOs immediately,
instead of having to wait for the end of synchronize_srcu_expedited,
which makes the code more clearly correct in case CPU 1 is preempted
right after synchronize_srcu() returns.

To avoid halving the generation space in SPTEs, always presume that the
low bit of the generation is zero when reconstructing a generation number
out of an SPTE.  This effectively disables MMIO caching in SPTEs during
the call to synchronize_srcu_expedited.  Using the low bit this way is
somewhat like a seqcount---where the protected thing is a cache, and
instead of retrying we can simply punt if we observe the low bit to be 1.

Cc: stable@vger.kernel.org
Signed-off-by: David Matlack <dmatlack@google.com>
Reviewed-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Reviewed-by: David Matlack <dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: do not bias the generation number in kvm_current_mmio_generation
Paolo Bonzini [Wed, 20 Aug 2014 12:29:21 +0000 (14:29 +0200)]
KVM: do not bias the generation number in kvm_current_mmio_generation

The next patch will give a meaning (a la seqcount) to the low bit of the
generation number.  Ensure that it matches between kvm->memslots->generation
and kvm_current_mmio_generation().

Cc: stable@vger.kernel.org
Reviewed-by: David Matlack <dmatlack@google.com>
Reviewed-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86: use guest maxphyaddr to check MTRR values
Paolo Bonzini [Fri, 29 Aug 2014 16:56:01 +0000 (18:56 +0200)]
KVM: x86: use guest maxphyaddr to check MTRR values

The check introduced in commit d7a2a246a1b5 (KVM: x86: #GP when attempts to write reserved bits of Variable Range MTRRs, 2014-08-19)
will break if the guest maxphyaddr is higher than the host's (which
sometimes happens depending on your hardware and how QEMU is
configured).

To fix this, use cpuid_maxphyaddr similar to how the APIC_BASE MSR
does already.

Reported-by: Jan Kiszka <jan.kiszka@siemens.com>
Tested-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: remove garbage arg to *hardware_{en,dis}able
Radim Krčmář [Thu, 28 Aug 2014 13:13:03 +0000 (15:13 +0200)]
KVM: remove garbage arg to *hardware_{en,dis}able

In the beggining was on_each_cpu(), which required an unused argument to
kvm_arch_ops.hardware_{en,dis}able, but this was soon forgotten.

Remove unnecessary arguments that stem from this.

Signed-off-by: Radim KrÄ\8dmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: static inline empty kvm_arch functions
Radim Krčmář [Thu, 28 Aug 2014 13:13:02 +0000 (15:13 +0200)]
KVM: static inline empty kvm_arch functions

Using static inline is going to save few bytes and cycles.
For example on powerpc, the difference is 700 B after stripping.
(5 kB before)

This patch also deals with two overlooked empty functions:
kvm_arch_flush_shadow was not removed from arch/mips/kvm/mips.c
  2df72e9bc KVM: split kvm_arch_flush_shadow
and kvm_arch_sched_in never made it into arch/ia64/kvm/kvm-ia64.c.
  e790d9ef6 KVM: add kvm_arch_sched_in

Signed-off-by: Radim KrÄ\8dmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: forward declare structs in kvm_types.h
Paolo Bonzini [Fri, 29 Aug 2014 12:01:17 +0000 (14:01 +0200)]
KVM: forward declare structs in kvm_types.h

Opaque KVM structs are useful for prototypes in asm/kvm_host.h, to avoid
"'struct foo' declared inside parameter list" warnings (and consequent
breakage due to conflicting types).

Move them from individual files to a generic place in linux/kvm_types.h.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86: remove Aligned bit from movntps/movntpd
Paolo Bonzini [Mon, 14 Jul 2014 10:54:48 +0000 (12:54 +0200)]
KVM: x86: remove Aligned bit from movntps/movntpd

These are not explicitly aligned, and do not require alignment on AVX.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86 emulator: emulate MOVNTDQ
Alex Williamson [Fri, 11 Jul 2014 17:56:31 +0000 (11:56 -0600)]
KVM: x86 emulator: emulate MOVNTDQ

Windows 8.1 guest with NVIDIA driver and GPU fails to boot with an
emulation failure.  The KVM spew suggests the fault is with lack of
movntdq emulation (courtesy of Paolo):

Code=02 00 00 b8 08 00 00 00 f3 0f 6f 44 0a f0 f3 0f 6f 4c 0a e0 <66> 0f e7 41 f0 66 0f e7 49 e0 48 83 e9 40 f3 0f 6f 44 0a 10 f3 0f 6f 0c 0a 66 0f e7 41 10

$ as -o a.out
        .section .text
        .byte 0x66, 0x0f, 0xe7, 0x41, 0xf0
        .byte 0x66, 0x0f, 0xe7, 0x49, 0xe0
$ objdump -d a.out
    0:  66 0f e7 41 f0          movntdq %xmm0,-0x10(%rcx)
    5:  66 0f e7 49 e0          movntdq %xmm1,-0x20(%rcx)

Add the necessary emulation.

Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: vmx: VMXOFF emulation in vm86 should cause #UD
Nadav Amit [Fri, 29 Aug 2014 08:26:55 +0000 (11:26 +0300)]
KVM: vmx: VMXOFF emulation in vm86 should cause #UD

Unlike VMCALL, the instructions VMXOFF, VMLAUNCH and VMRESUME should cause a UD
exception in real-mode or vm86.  However, the emulator considers all these
instructions the same for the matter of mode checks, and emulation upon exit
due to #UD exception.

As a result, the hypervisor behaves incorrectly on vm86 mode. VMXOFF, VMLAUNCH
or VMRESUME cause on vm86 exit due to #UD. The hypervisor then emulates these
instruction and inject #GP to the guest instead of #UD.

This patch creates a new group for these instructions and mark only VMCALL as
an instruction which can be emulated.

Signed-off-by: Nadav Amit <namit@cs.technion.ac.il>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86: fix some sparse warnings
Paolo Bonzini [Tue, 26 Aug 2014 11:27:46 +0000 (13:27 +0200)]
KVM: x86: fix some sparse warnings

Sparse reports the following easily fixed warnings:

   arch/x86/kvm/vmx.c:8795:48: sparse: Using plain integer as NULL pointer
   arch/x86/kvm/vmx.c:2138:5: sparse: symbol vmx_read_l1_tsc was not declared. Should it be static?
   arch/x86/kvm/vmx.c:6151:48: sparse: Using plain integer as NULL pointer
   arch/x86/kvm/vmx.c:8851:6: sparse: symbol vmx_sched_in was not declared. Should it be static?

   arch/x86/kvm/svm.c:2162:5: sparse: symbol svm_read_l1_tsc was not declared. Should it be static?

Cc: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: nVMX: nested TPR shadow/threshold emulation
Wanpeng Li [Thu, 21 Aug 2014 11:46:50 +0000 (19:46 +0800)]
KVM: nVMX: nested TPR shadow/threshold emulation

This patch fix bug https://bugzilla.kernel.org/show_bug.cgi?id=61411

TPR shadow/threshold feature is important to speed up the Windows guest.
Besides, it is a must feature for certain VMM.

We map virtual APIC page address and TPR threshold from L1 VMCS. If
TPR_BELOW_THRESHOLD VM exit is triggered by L2 guest and L1 interested
in, we inject it into L1 VMM for handling.

Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Wanpeng Li <wanpeng.li@linux.intel.com>
[Add PAGE_ALIGNED check, do not write useless virtual APIC page address
 if TPR shadowing is disabled. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: nVMX: introduce nested_get_vmcs12_pages
Wanpeng Li [Thu, 21 Aug 2014 11:46:49 +0000 (19:46 +0800)]
KVM: nVMX: introduce nested_get_vmcs12_pages

Introduce function nested_get_vmcs12_pages() to check the valid
of nested apic access page and virtual apic page earlier.

Signed-off-by: Wanpeng Li <wanpeng.li@linux.intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: Unconditionally export KVM_CAP_USER_NMI
Christoffer Dall [Tue, 26 Aug 2014 12:00:38 +0000 (14:00 +0200)]
KVM: Unconditionally export KVM_CAP_USER_NMI

The idea between capabilities and the KVM_CHECK_EXTENSION ioctl is that
userspace can, at run-time, determine if a feature is supported or not.
This allows KVM to being supporting a new feature with a new kernel
version without any need to update user space.  Unfortunately, since the
definition of KVM_CAP_USER_NMI was guarded by #ifdef
__KVM_HAVE_USER_NMI, such discovery still required a user space update.

Therefore, unconditionally export KVM_CAP_USER_NMI and change the
the typo in the comment for the IOCTL number definition as well.

Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: Unconditionally export KVM_CAP_READONLY_MEM
Christoffer Dall [Tue, 26 Aug 2014 12:00:37 +0000 (14:00 +0200)]
KVM: Unconditionally export KVM_CAP_READONLY_MEM

The idea between capabilities and the KVM_CHECK_EXTENSION ioctl is that
userspace can, at run-time, determine if a feature is supported or not.
This allows KVM to being supporting a new feature with a new kernel
version without any need to update user space.  Unfortunately, since the
definition of KVM_CAP_READONLY_MEM was guarded by #ifdef
__KVM_HAVE_READONLY_MEM, such discovery still required a user space
update.

Therefore, unconditionally export KVM_CAP_READONLY_MEM and change the
in-kernel conditional to rely on __KVM_HAVE_READONLY_MEM.

Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: s390/mm: fix up indentation of set_guest_storage_key
Christian Borntraeger [Wed, 27 Aug 2014 10:20:02 +0000 (12:20 +0200)]
KVM: s390/mm: fix up indentation of set_guest_storage_key

commit ab3f285f227f ("KVM: s390/mm: try a cow on read only pages for
key ops")' misaligned a code block. Let's fixup the indentation.

Reported-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoMerge tag 'kvm-s390-next-20140825' of git://git.kernel.org/pub/scm/linux/kernel/git...
Paolo Bonzini [Tue, 26 Aug 2014 12:31:44 +0000 (14:31 +0200)]
Merge tag 'kvm-s390-next-20140825' of git://git./linux/kernel/git/kvms390/linux into HEAD

KVM: s390: Fixes and features for 3.18 part 1

1. The usual cleanups: get rid of duplicate code, use defines, factor
   out the sync_reg handling, additional docs for sync_regs, better
   error handling on interrupt injection
2. We use KVM_REQ_TLB_FLUSH instead of open coding tlb flushes
3. Additional registers for kvm_run sync regs. This is usually not
   needed in the fast path due to eventfd/irqfd, but kvm stat claims
   that we reduced the overhead of console output by ~50% on my system
4. A rework of the gmap infrastructure. This is the 2nd step towards
   host large page support (after getting rid of the storage key
   dependency). We introduces two radix trees to store the guest-to-host
   and host-to-guest translations. This gets us rid of most of
   the page-table walks in the gmap code. Only one in __gmap_link is left,
   this one is required to link the shadow page table to the process page
   table. Finally this contains the plumbing to support gmap page tables
   with less than 5 levels.

9 years agoKVM: s390/mm: remove outdated gmap data structures
Martin Schwidefsky [Fri, 1 Aug 2014 13:03:33 +0000 (15:03 +0200)]
KVM: s390/mm: remove outdated gmap data structures

The radix tree rework removed all code that uses the gmap_rmap
and gmap_pgtable data structures. Remove these outdated definitions.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
9 years agoKVM: s390/mm: support gmap page tables with less than 5 levels
Martin Schwidefsky [Tue, 1 Jul 2014 12:36:04 +0000 (14:36 +0200)]
KVM: s390/mm: support gmap page tables with less than 5 levels

Add an addressing limit to the gmap address spaces and only allocate
the page table levels that are needed for the given limit. The limit
is fixed and can not be changed after a gmap has been created.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
9 years agoKVM: s390/mm: use radix trees for guest to host mappings
Martin Schwidefsky [Wed, 30 Apr 2014 14:04:25 +0000 (16:04 +0200)]
KVM: s390/mm: use radix trees for guest to host mappings

Store the target address for the gmap segments in a radix tree
instead of using invalid segment table entries. gmap_translate
becomes a simple radix_tree_lookup, gmap_fault is split into the
address translation with gmap_translate and the part that does
the linking of the gmap shadow page table with the process page
table.
A second radix tree is used to keep the pointers to the segment
table entries for segments that are mapped in the guest address
space. On unmap of a segment the pointer is retrieved from the
radix tree and is used to carry out the segment invalidation in
the gmap shadow page table. As the radix tree can only store one
pointer, each host segment may only be mapped to exactly one
guest location.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
9 years agokvm: x86: fix tracing for 32-bit
Paolo Bonzini [Mon, 25 Aug 2014 14:08:21 +0000 (16:08 +0200)]
kvm: x86: fix tracing for 32-bit

Fix commit 7b46268d29543e313e731606d845e65c17f232e4, which mistakenly
included the new tracepoint under #ifdef CONFIG_X86_64.

Reported-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoMerge tag 'kvm-s390-20140825' of git://git.kernel.org/pub/scm/linux/kernel/git/kvms39...
Paolo Bonzini [Mon, 25 Aug 2014 13:37:00 +0000 (15:37 +0200)]
Merge tag 'kvm-s390-20140825' of git://git./linux/kernel/git/kvms390/linux into HEAD

Here are two fixes for s390 KVM code that prevent:
1. a malicious user to trigger a kernel BUG
2. a malicious user to change the storage key of read-only pages

9 years agoKVM: s390/mm: cleanup gmap function arguments, variable names
Martin Schwidefsky [Tue, 29 Apr 2014 07:34:41 +0000 (09:34 +0200)]
KVM: s390/mm: cleanup gmap function arguments, variable names

Make the order of arguments for the gmap calls more consistent,
if the gmap pointer is passed it is always the first argument.
In addition distinguish between guest address and user address
by naming the variables gaddr for a guest address and vmaddr for
a user address.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
9 years agoKVM: s390/mm: readd address parameter to gmap_do_ipte_notify
Martin Schwidefsky [Wed, 30 Apr 2014 12:46:26 +0000 (14:46 +0200)]
KVM: s390/mm: readd address parameter to gmap_do_ipte_notify

Revert git commit c3a23b9874c1 ("remove unnecessary parameter from
gmap_do_ipte_notify").

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
9 years agoKVM: s390/mm: readd address parameter to pgste_ipte_notify
Martin Schwidefsky [Wed, 30 Apr 2014 12:44:44 +0000 (14:44 +0200)]
KVM: s390/mm: readd address parameter to pgste_ipte_notify

Revert git commit 1b7fd6952063 ("remove unecessary parameter from
pgste_ipte_notify")

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
9 years agoKVM: s390: don't use kvm lock in interrupt injection code
Jens Freimann [Mon, 11 Aug 2014 13:39:43 +0000 (15:39 +0200)]
KVM: s390: don't use kvm lock in interrupt injection code

The kvm lock protects us against vcpus going away, but they only go
away when the virtual machine is shut down. We don't need this
mutex here, so let's get rid of it.

Signed-off-by: Jens Freimann <jfrei@linux.vnet.ibm.com>
Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
9 years agoKVM: s390: return -EFAULT if lowcore is not mapped during irq delivery
Jens Freimann [Thu, 17 Apr 2014 08:10:30 +0000 (10:10 +0200)]
KVM: s390: return -EFAULT if lowcore is not mapped during irq delivery

Currently we just kill the userspace process and exit the thread
immediatly without making sure that we don't hold any locks etc.

Improve this by making KVM_RUN return -EFAULT if the lowcore is not
mapped during interrupt delivery. To achieve this we need to pass
the return code of guest memory access routines used in interrupt
delivery all the way back to the KVM_RUN ioctl.

Signed-off-by: Jens Freimann <jfrei@linux.vnet.ibm.com>
Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
9 years agoKVM: s390: implement KVM_REQ_TLB_FLUSH and make use of it
David Hildenbrand [Tue, 29 Jul 2014 06:53:36 +0000 (08:53 +0200)]
KVM: s390: implement KVM_REQ_TLB_FLUSH and make use of it

Use the KVM_REQ_TLB_FLUSH request in order to trigger tlb flushes instead
of manipulating the SIE control block whenever we need it. Also trigger it for
a control register sync directly instead of (ab)using kvm_s390_set_prefix().

Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
9 years agoKVM: s390: synchronize more registers with kvm_run
David Hildenbrand [Thu, 17 Jul 2014 08:47:43 +0000 (10:47 +0200)]
KVM: s390: synchronize more registers with kvm_run

In order to reduce the number of syscalls when dropping to user space, this
patch enables the synchronization of the following "registers" with kvm_run:
- ARCH0: CPU timer, clock comparator, TOD programmable register,
         guest breaking-event register, program parameter
- PFAULT: pfault parameters (token, select, compare)

The registers are grouped to reduce the overhead when syncing.

As this grows the number of sync registers quite a bit, let's move the code
synchronizing registers with kvm_run from kvm_arch_vcpu_ioctl_run() into
separate helper routines.

Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
9 years agoKVM: s390: no special machine check delivery
Christian Borntraeger [Mon, 4 Aug 2014 14:54:22 +0000 (16:54 +0200)]
KVM: s390: no special machine check delivery

The load PSW handler does not have to inject pending machine checks.
This can wait until the CPU runs the generic interrupt injection code.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
9 years agoKVM: s390: clear kvm_dirty_regs when dropping to user space
David Hildenbrand [Tue, 29 Jul 2014 06:22:33 +0000 (08:22 +0200)]
KVM: s390: clear kvm_dirty_regs when dropping to user space

We should make sure that all kvm_dirty_regs bits are cleared before dropping
to user space. Until now, some would remain pending.

Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
9 years agoKVM: clarify the idea of kvm_dirty_regs
David Hildenbrand [Tue, 29 Jul 2014 06:19:26 +0000 (08:19 +0200)]
KVM: clarify the idea of kvm_dirty_regs

This patch clarifies that kvm_dirty_regs are just a hint to the kernel and
that the kernel might just ignore some flags and sync the values (like done for
acrs and gprs now).

Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
9 years agoKVM: s390: factor out get_ilc() function
Jens Freimann [Wed, 23 Jul 2014 14:36:06 +0000 (16:36 +0200)]
KVM: s390: factor out get_ilc() function

Let's make this a reusable function.

Signed-off-by: Jens Freimann <jfrei@linux.vnet.ibm.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
9 years agoKVM: s390/mm: try a cow on read only pages for key ops
Christian Borntraeger [Tue, 19 Aug 2014 14:19:35 +0000 (16:19 +0200)]
KVM: s390/mm: try a cow on read only pages for key ops

The PFMF instruction handler  blindly wrote the storage key even if
the page was mapped R/O in the host. Lets try a COW before continuing
and bail out in case of errors.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: Dominik Dingel <dingel@linux.vnet.ibm.com>
Cc: stable@vger.kernel.org
9 years agoKVM: s390: add defines for pfault init delivery code
Jens Freimann [Wed, 16 Apr 2014 11:57:18 +0000 (13:57 +0200)]
KVM: s390: add defines for pfault init delivery code

Get rid of open coded values for pfault init.

Signed-off-by: Jens Freimann <jfrei@linux.vnet.ibm.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
9 years agoKVM: s390: Fix user triggerable bug in dead code
Christian Borntraeger [Wed, 6 Aug 2014 14:17:58 +0000 (16:17 +0200)]
KVM: s390: Fix user triggerable bug in dead code

In the early days, we had some special handling for the
KVM_EXIT_S390_SIEIC exit, but this was gone in 2009 with commit
d7b0b5eb3000 (KVM: s390: Make psw available on all exits, not
just a subset).

Now this switch statement is just a sanity check for userspace
not messing with the kvm_run structure. Unfortunately, this
allows userspace to trigger a kernel BUG. Let's just remove
this switch statement.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Cc: stable@vger.kernel.org
9 years agoKVM: trace kvm_ple_window grow/shrink
Radim Krčmář [Thu, 21 Aug 2014 16:08:09 +0000 (18:08 +0200)]
KVM: trace kvm_ple_window grow/shrink

Tracepoint for dynamic PLE window, fired on every potential change.

Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: VMX: dynamise PLE window
Radim Krčmář [Thu, 21 Aug 2014 16:08:08 +0000 (18:08 +0200)]
KVM: VMX: dynamise PLE window

Window is increased on every PLE exit and decreased on every sched_in.
The idea is that we don't want to PLE exit if there is no preemption
going on.
We do this with sched_in() because it does not hold rq lock.

There are two new kernel parameters for changing the window:
 ple_window_grow and ple_window_shrink
ple_window_grow affects the window on PLE exit and ple_window_shrink
does it on sched_in;  depending on their value, the window is modifier
like this: (ple_window is kvm_intel's global)

  ple_window_shrink/ |
  ple_window_grow    | PLE exit           | sched_in
  -------------------+--------------------+---------------------
  < 1                |  = ple_window      |  = ple_window
  < ple_window       | *= ple_window_grow | /= ple_window_shrink
  otherwise          | += ple_window_grow | -= ple_window_shrink

A third new parameter, ple_window_max, controls the maximal ple_window;
it is internally rounded down to a closest multiple of ple_window_grow.

VCPU's PLE window is never allowed below ple_window.

Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: VMX: make PLE window per-VCPU
Radim Krčmář [Thu, 21 Aug 2014 16:08:07 +0000 (18:08 +0200)]
KVM: VMX: make PLE window per-VCPU

Change PLE window into per-VCPU variable, seeded from module parameter,
to allow greater flexibility.

Brings in a small overhead on every vmentry.

Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86: introduce sched_in to kvm_x86_ops
Radim Krčmář [Thu, 21 Aug 2014 16:08:06 +0000 (18:08 +0200)]
KVM: x86: introduce sched_in to kvm_x86_ops

sched_in preempt notifier is available for x86, allow its use in
specific virtualization technlogies as well.

Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: add kvm_arch_sched_in
Radim Krčmář [Thu, 21 Aug 2014 16:08:05 +0000 (18:08 +0200)]
KVM: add kvm_arch_sched_in

Introduce preempt notifiers for architecture specific code.
Advantage over creating a new notifier in every arch is slightly simpler
code and guaranteed call order with respect to kvm_sched_in.

Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86: Replace X86_FEATURE_NX offset with the definition
Nadav Amit [Wed, 20 Aug 2014 13:38:19 +0000 (16:38 +0300)]
KVM: x86: Replace X86_FEATURE_NX offset with the definition

Replace reference to X86_FEATURE_NX using bit shift with the defined
X86_FEATURE_NX.

Signed-off-by: Nadav Amit <namit@cs.technion.ac.il>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: avoid unnecessary synchronize_rcu
Christian Borntraeger [Tue, 19 Aug 2014 14:45:56 +0000 (16:45 +0200)]
KVM: avoid unnecessary synchronize_rcu

We dont have to wait for a grace period if there is no oldpid that
we are going to free. putpid also checks for NULL, so this patch
only fences synchronize_rcu.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: emulate: warn on invalid or uninitialized exception numbers
Paolo Bonzini [Wed, 20 Aug 2014 08:08:23 +0000 (10:08 +0200)]
KVM: emulate: warn on invalid or uninitialized exception numbers

These were reported when running Jailhouse on AMD processors.

Initialize ctxt->exception.vector with an invalid exception number,
and warn if it remained invalid even though the emulator got
an X86EMUL_PROPAGATE_FAULT return code.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: emulate: do not return X86EMUL_PROPAGATE_FAULT explicitly
Paolo Bonzini [Wed, 20 Aug 2014 08:05:08 +0000 (10:05 +0200)]
KVM: emulate: do not return X86EMUL_PROPAGATE_FAULT explicitly

Always get it through emulate_exception or emulate_ts.  This
ensures that the ctxt->exception fields have been populated.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86: Clarify PMU related features bit manipulation
Nadav Amit [Wed, 20 Aug 2014 10:25:52 +0000 (13:25 +0300)]
KVM: x86: Clarify PMU related features bit manipulation

kvm_pmu_cpuid_update makes a lot of bit manuiplation operations, when in fact
there are already unions that can be used instead. Changing the bit
manipulation to the union for clarity. This patch does not change the
functionality.

Signed-off-by: Nadav Amit <namit@cs.technion.ac.il>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: vmx: fix ept reserved bits for 1-GByte page
Wanpeng Li [Wed, 20 Aug 2014 07:31:53 +0000 (15:31 +0800)]
KVM: vmx: fix ept reserved bits for 1-GByte page

EPT misconfig handler in kvm will check which reason lead to EPT
misconfiguration after vmexit. One of the reasons is that an EPT
paging-structure entry is configured with settings reserved for
future functionality. However, the handler can't identify if
paging-structure entry of reserved bits for 1-GByte page are
configured, since PDPTE which point to 1-GByte page will reserve
bits 29:12 instead of bits 7:3 which are reserved for PDPTE that
references an EPT Page Directory. This patch fix it by reserve
bits 29:12 for 1-GByte page.

Signed-off-by: Wanpeng Li <wanpeng.li@linux.intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86: recalculate_apic_map after enabling apic
Nadav Amit [Mon, 18 Aug 2014 21:03:00 +0000 (00:03 +0300)]
KVM: x86: recalculate_apic_map after enabling apic

Currently, recalculate_apic_map ignores vcpus whose lapic is software disabled
through the spurious interrupt vector. However, once it is re-enabled, the map
is not recalculated. Therefore, if the guest OS configured DFR while lapic is
software-disabled, the map may be incorrect. This patch recalculates apic map
after software enabling the lapic.

Signed-off-by: Nadav Amit <namit@cs.technion.ac.il>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86: Clear apic tsc-deadline after deadline
Nadav Amit [Mon, 18 Aug 2014 19:42:13 +0000 (22:42 +0300)]
KVM: x86: Clear apic tsc-deadline after deadline

Intel SDM 10.5.4.1 says "When the timer generates an interrupt, it disarms
itself and clears the IA32_TSC_DEADLINE MSR".

This patch clears the MSR upon timer interrupt delivery which delivered on
deadline mode.  Since the MSR may be reconfigured while an interrupt is
pending, causing the new value to be overriden, pending timer interrupts are
checked before setting a new deadline.

Signed-off-by: Nadav Amit <namit@cs.technion.ac.il>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoKVM: x86: #GP when attempts to write reserved bits of Variable Range MTRRs
Wanpeng Li [Tue, 19 Aug 2014 09:04:40 +0000 (17:04 +0800)]
KVM: x86: #GP when attempts to write reserved bits of Variable Range MTRRs

Section 11.11.2.3 of the SDM mentions "All other bits in the IA32_MTRR_PHYSBASEn
and IA32_MTRR_PHYSMASKn registers are reserved; the processor generates a
general-protection exception(#GP) if software attempts to write to them". This
patch do it in kvm.

Signed-off-by: Wanpeng Li <wanpeng.li@linux.intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>