openwrt/staging/blogic.git
10 years agokvm/x86: implement hv EOI assist
Michael S. Tsirkin [Wed, 7 May 2014 13:29:48 +0000 (16:29 +0300)]
kvm/x86: implement hv EOI assist

It seems that it's easy to implement the EOI assist
on top of the PV EOI feature: simply convert the
page address to the format expected by PV EOI.

Notes:
-"No EOI required" is set only if interrupt injected
 is edge triggered; this is true because level interrupts are going
 through IOAPIC which disables PV EOI.
 In any case, if guest triggers EOI the bit will get cleared on exit.
-For migration, set of HV_X64_MSR_APIC_ASSIST_PAGE sets
 KVM_PV_EOI_EN internally, so restoring HV_X64_MSR_APIC_ASSIST_PAGE
 seems sufficient
 In any case, bit is cleared on exit so worst case it's never re-enabled
-no handling of PV EOI data is performed at HV_X64_MSR_EOI write;
 HV_X64_MSR_EOI is a separate optimization - it's an X2APIC
 replacement that lets you do EOI with an MSR and not IO.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
10 years agoKVM: x86: Mark bit 7 in long-mode PDPTE according to 1GB pages support
Nadav Amit [Wed, 7 May 2014 12:32:50 +0000 (15:32 +0300)]
KVM: x86: Mark bit 7 in long-mode PDPTE according to 1GB pages support

In long-mode, bit 7 in the PDPTE is not reserved only if 1GB pages are
supported by the CPU. Currently the bit is considered by KVM as always
reserved.

Signed-off-by: Nadav Amit <namit@cs.technion.ac.il>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
10 years agoKVM: vmx: handle_dr does not handle RSP correctly
Nadav Amit [Wed, 7 May 2014 12:32:49 +0000 (15:32 +0300)]
KVM: vmx: handle_dr does not handle RSP correctly

The RSP register is not automatically cached, causing mov DR instruction with
RSP to fail.  Instead the regular register accessing interface should be used.

Signed-off-by: Nadav Amit <namit@cs.technion.ac.il>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
10 years agoKVM: nVMX: move vmclear and vmptrld pre-checks to nested_vmx_check_vmptr
Bandan Das [Tue, 6 May 2014 06:19:18 +0000 (02:19 -0400)]
KVM: nVMX: move vmclear and vmptrld pre-checks to nested_vmx_check_vmptr

Some checks are common to all, and moreover,
according to the spec, the check for whether any bits
beyond the physical address width are set are also
applicable to all of them

Signed-off-by: Bandan Das <bsd@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
10 years agoKVM: nVMX: fail on invalid vmclear/vmptrld pointer
Bandan Das [Tue, 6 May 2014 06:19:17 +0000 (02:19 -0400)]
KVM: nVMX: fail on invalid vmclear/vmptrld pointer

The spec mandates that if the vmptrld or vmclear
address is equal to the vmxon region pointer, the
instruction should fail with error "VMPTRLD with
VMXON pointer" or "VMCLEAR with VMXON pointer"

Signed-off-by: Bandan Das <bsd@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
10 years agoKVM: nVMX: additional checks on vmxon region
Bandan Das [Tue, 6 May 2014 06:19:16 +0000 (02:19 -0400)]
KVM: nVMX: additional checks on vmxon region

Currently, the vmxon region isn't used in the nested case.
However, according to the spec, the vmxon instruction performs
additional sanity checks on this region and the associated
pointer. Modify emulated vmxon to better adhere to the spec
requirements

Signed-off-by: Bandan Das <bsd@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
10 years agoKVM: nVMX: rearrange get_vmx_mem_address
Bandan Das [Tue, 6 May 2014 06:19:15 +0000 (02:19 -0400)]
KVM: nVMX: rearrange get_vmx_mem_address

Our common function for vmptr checks (in 2/4) needs to fetch
the memory address

Signed-off-by: Bandan Das <bsd@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
10 years agoMerge tag 'kvm-s390-20140506' of git://git.kernel.org/pub/scm/linux/kernel/git/kvms39...
Paolo Bonzini [Tue, 6 May 2014 15:20:37 +0000 (17:20 +0200)]
Merge tag 'kvm-s390-20140506' of git://git./linux/kernel/git/kvms390/linux into kvm-next

1. Fixes an error return code for the breakpoint setup

2. External interrupt fixes
2.1. Some interrupt conditions like cpu timer or clock comparator
stay pending even after the interrupt is injected. If the external
new PSW is enabled for interrupts this will result in an endless
loop. Usually this indicates a programming error in the guest OS.
Lets detect such situations and go to userspace. We will provide
a QEMU patch that sets the guest in panicked/crashed state to avoid
wasting CPU cycles.
2.2 Resend external interrupts back to the guest if the HW could
not do it.
-

10 years agoKVM: s390: Fix external interrupt interception
Thomas Huth [Wed, 15 Jan 2014 15:46:07 +0000 (16:46 +0100)]
KVM: s390: Fix external interrupt interception

The external interrupt interception can only occur in rare cases, e.g.
when the PSW of the interrupt handler has a bad value. The old handler
for this interception simply ignored these events (except for increasing
the exit_external_interrupt counter), but for proper operation we either
have to inject the interrupts manually or we should drop to userspace in
case of errors.

Signed-off-by: Thomas Huth <thuth@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>
10 years agoKVM: s390: Add clock comparator and CPU timer IRQ injection
Thomas Huth [Wed, 26 Mar 2014 15:11:54 +0000 (16:11 +0100)]
KVM: s390: Add clock comparator and CPU timer IRQ injection

Add an interface to inject clock comparator and CPU timer interrupts
into the guest. This is needed for handling the external interrupt
interception.

Signed-off-by: Thomas Huth <thuth@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>
10 years agoKVM: s390: return -EFAULT if copy_from_user() fails
Dan Carpenter [Sat, 3 May 2014 20:18:11 +0000 (23:18 +0300)]
KVM: s390: return -EFAULT if copy_from_user() fails

When copy_from_user() fails, this code returns the number of bytes
remaining instead of a negative error code.  The positive number is
returned to the user but otherwise it is harmless.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: x86: improve the usability of the 'kvm_pio' tracepoint
Ulrich Obergfell [Fri, 2 May 2014 15:57:47 +0000 (17:57 +0200)]
KVM: x86: improve the usability of the 'kvm_pio' tracepoint

This patch moves the 'kvm_pio' tracepoint to emulator_pio_in_emulated()
and emulator_pio_out_emulated(), and it adds an argument (a pointer to
the 'pio_data'). A single 8-bit or 16-bit or 32-bit data item is fetched
from 'pio_data' (depending on 'size'), and the value is included in the
trace record ('val'). If 'count' is greater than one, this is indicated
by the string "(...)" in the trace output.

Signed-off-by: Ulrich Obergfell <uobergfe@redhat.com>
Reviewed-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
10 years agokvm/irqchip: Speed up KVM_SET_GSI_ROUTING
Christian Borntraeger [Thu, 16 Jan 2014 12:44:20 +0000 (13:44 +0100)]
kvm/irqchip: Speed up KVM_SET_GSI_ROUTING

When starting lots of dataplane devices the bootup takes very long on
Christian's s390 with irqfd patches. With larger setups he is even
able to trigger some timeouts in some components.  Turns out that the
KVM_SET_GSI_ROUTING ioctl takes very long (strace claims up to 0.1 sec)
when having multiple CPUs.  This is caused by the  synchronize_rcu and
the HZ=100 of s390.  By changing the code to use a private srcu we can
speed things up.  This patch reduces the boot time till mounting root
from 8 to 2 seconds on my s390 guest with 100 disks.

Uses of hlist_for_each_entry_rcu, hlist_add_head_rcu, hlist_del_init_rcu
are fine because they do not have lockdep checks (hlist_for_each_entry_rcu
uses rcu_dereference_raw rather than rcu_dereference, and write-sides
do not do rcu lockdep at all).

Note that we're hardly relying on the "sleepable" part of srcu.  We just
want SRCU's faster detection of grace periods.

Testing was done by Andrew Theurer using netperf tests STREAM, MAERTS
and RR.  The difference between results "before" and "after" the patch
has mean -0.2% and standard deviation 0.6%.  Using a paired t-test on the
data points says that there is a 2.5% probability that the patch is the
cause of the performance difference (rather than a random fluctuation).

(Restricting the t-test to RR, which is the most likely to be affected,
changes the numbers to respectively -0.3% mean, 0.7% stdev, and 8%
probability that the numbers actually say something about the patch.
The probability increases mostly because there are fewer data points).

Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Tested-by: Christian Borntraeger <borntraeger@de.ibm.com> # s390
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
10 years agoMerge tag 'kvm-s390-20140429' of git://git.kernel.org/pub/scm/linux/kernel/git/kvms39...
Paolo Bonzini [Wed, 30 Apr 2014 10:29:41 +0000 (12:29 +0200)]
Merge tag 'kvm-s390-20140429' of git://git./linux/kernel/git/kvms390/linux into kvm-next

1. Guest handling fixes
The handling of MVPG, PFMF and Test Block is fixed to better follow
the architecture. None of these fixes is critical for any current
Linux guests, but let's play safe.

2. Optimization for single CPU guests
We can enable the IBS facility if only one VCPU is running (!STOPPED
state). We also enable this optimization for guest > 1 VCPU as soon
as all but one VCPU is in stopped state. Thus will help guests that
have tools like cpuplugd (from s390-utils) that do dynamic offline/
online of CPUs.

3. NOTES
There is one non-s390 change in include/linux/kvm_host.h that
introduces 2 defines for VCPU requests:
define KVM_REQ_ENABLE_IBS        23
define KVM_REQ_DISABLE_IBS       24

10 years agoKVM: x86: expose invariant tsc cpuid bit (v2)
Marcelo Tosatti [Sun, 27 Apr 2014 01:30:23 +0000 (22:30 -0300)]
KVM: x86: expose invariant tsc cpuid bit (v2)

Invariant TSC is a property of TSC, no additional
support code necessary.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
10 years agoKVM: s390: enable IBS for single running VCPUs
David Hildenbrand [Fri, 14 Mar 2014 10:00:21 +0000 (11:00 +0100)]
KVM: s390: enable IBS for single running VCPUs

This patch enables the IBS facility when a single VCPU is running.
The facility is dynamically turned on/off as soon as other VCPUs
enter/leave the stopped state.

When this facility is operating, some instructions can be executed
faster for single-cpu guests.

Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Reviewed-by: Dominik Dingel <dingel@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: introduce kvm_s390_vcpu_{start,stop}
David Hildenbrand [Fri, 14 Mar 2014 09:59:29 +0000 (10:59 +0100)]
KVM: s390: introduce kvm_s390_vcpu_{start,stop}

This patch introduces two new functions to set/clear the CPUSTAT_STOPPED bit and
makes use of it at all applicable places. These functions prepare the additional
execution of code when starting/stopping a vcpu.

The CPUSTAT_STOPPED bit should not be touched outside of these functions.

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>
10 years agoKVM: s390: Add low-address protection to TEST BLOCK
Thomas Huth [Fri, 7 Mar 2014 11:14:23 +0000 (12:14 +0100)]
KVM: s390: Add low-address protection to TEST BLOCK

TEST BLOCK is also subject to the low-address protection, so we need
to check the destination address in our handler.

Signed-off-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: Fixes for PFMF
Thomas Huth [Mon, 9 Sep 2013 15:58:38 +0000 (17:58 +0200)]
KVM: s390: Fixes for PFMF

Add a check for low-address protection to the PFMF handler and
convert real-addresses to absolute if necessary, as it is defined
in the Principles of Operations specification.

Signed-off-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: Add a function for checking the low-address protection
Thomas Huth [Mon, 3 Mar 2014 22:34:42 +0000 (23:34 +0100)]
KVM: s390: Add a function for checking the low-address protection

The s390 architecture has a special protection mechanism that can
be used to prevent write access to the vital data in the low-core
memory area. This patch adds a new helper function that can be used
to check for such write accesses and in case of protection, it also
sets up the exception data accordingly.

Signed-off-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: Handle MVPG partial execution interception
Thomas Huth [Mon, 3 Feb 2014 09:42:30 +0000 (10:42 +0100)]
KVM: s390: Handle MVPG partial execution interception

When the guest executes the MVPG instruction with DAT disabled,
and the source or destination page is not mapped in the host,
the so-called partial execution interception occurs. We need to
handle this event by setting up a mapping for the corresponding
user pages.

Signed-off-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: async_pf: change async_pf_execute() to use get_user_pages(tsk => NULL)
Oleg Nesterov [Mon, 28 Apr 2014 15:03:00 +0000 (17:03 +0200)]
KVM: async_pf: change async_pf_execute() to use get_user_pages(tsk => NULL)

async_pf_execute() passes tsk == current to gup(), this is doesn't
hurt but unnecessary and misleading. "tsk" is only used to account
the number of faults and current is the random workqueue thread.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Suggested-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
10 years agoKVM: async_pf: kill the unnecessary use_mm/unuse_mm async_pf_execute()
Oleg Nesterov [Mon, 21 Apr 2014 13:25:58 +0000 (15:25 +0200)]
KVM: async_pf: kill the unnecessary use_mm/unuse_mm async_pf_execute()

async_pf_execute() has no reasons to adopt apf->mm, gup(current, mm)
should work just fine even if current has another or NULL ->mm.

Recently kvm_async_page_present_sync() was added insedie the "use_mm"
section, but it seems that it doesn't need current->mm too.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
10 years agoKVM: MMU: flush tlb out of mmu lock when write-protect the sptes
Xiao Guangrong [Thu, 17 Apr 2014 09:06:16 +0000 (17:06 +0800)]
KVM: MMU: flush tlb out of mmu lock when write-protect the sptes

Now we can flush all the TLBs out of the mmu lock without TLB corruption when
write-proect the sptes, it is because:
- we have marked large sptes readonly instead of dropping them that means we
  just change the spte from writable to readonly so that we only need to care
  the case of changing spte from present to present (changing the spte from
  present to nonpresent will flush all the TLBs immediately), in other words,
  the only case we need to care is mmu_spte_update()

- in mmu_spte_update(), we haved checked
  SPTE_HOST_WRITEABLE | PTE_MMU_WRITEABLE instead of PT_WRITABLE_MASK, that
  means it does not depend on PT_WRITABLE_MASK anymore

Acked-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
10 years agoKVM: MMU: flush tlb if the spte can be locklessly modified
Xiao Guangrong [Thu, 17 Apr 2014 09:06:15 +0000 (17:06 +0800)]
KVM: MMU: flush tlb if the spte can be locklessly modified

Relax the tlb flush condition since we will write-protect the spte out of mmu
lock. Note lockless write-protection only marks the writable spte to readonly
and the spte can be writable only if both SPTE_HOST_WRITEABLE and
SPTE_MMU_WRITEABLE are set (that are tested by spte_is_locklessly_modifiable)

This patch is used to avoid this kind of race:

      VCPU 0                         VCPU 1
lockless wirte protection:
      set spte.w = 0
                                 lock mmu-lock

                                 write protection the spte to sync shadow page,
                                 see spte.w = 0, then without flush tlb

 unlock mmu-lock

                                 !!! At this point, the shadow page can still be
                                     writable due to the corrupt tlb entry
     Flush all TLB

Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
10 years agoKVM: MMU: lazily drop large spte
Xiao Guangrong [Thu, 17 Apr 2014 09:06:14 +0000 (17:06 +0800)]
KVM: MMU: lazily drop large spte

Currently, kvm zaps the large spte if write-protected is needed, the later
read can fault on that spte. Actually, we can make the large spte readonly
instead of making them un-present, the page fault caused by read access can
be avoided

The idea is from Avi:
| As I mentioned before, write-protecting a large spte is a good idea,
| since it moves some work from protect-time to fault-time, so it reduces
| jitter.  This removes the need for the return value.

This version has fixed the issue reported in 6b73a9606, the reason of that
issue is that fast_page_fault() directly sets the readonly large spte to
writable but only dirty the first page into the dirty-bitmap that means
other pages are missed. Fixed it by only the normal sptes (on the
PT_PAGE_TABLE_LEVEL level) can be fast fixed

Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
10 years agoKVM: MMU: properly check last spte in fast_page_fault()
Xiao Guangrong [Thu, 17 Apr 2014 09:06:13 +0000 (17:06 +0800)]
KVM: MMU: properly check last spte in fast_page_fault()

Using sp->role.level instead of @level since @level is not got from the
page table hierarchy

There is no issue in current code since the fast page fault currently only
fixes the fault caused by dirty-log that is always on the last level
(level = 1)

This patch makes the code more readable and avoids potential issue in the
further development

Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
10 years agoRevert "KVM: Simplify kvm->tlbs_dirty handling"
Xiao Guangrong [Thu, 17 Apr 2014 09:06:12 +0000 (17:06 +0800)]
Revert "KVM: Simplify kvm->tlbs_dirty handling"

This reverts commit 5befdc385ddb2d5ae8995ad89004529a3acf58fc.

Since we will allow flush tlb out of mmu-lock in the later
patch

Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
10 years agoKVM: x86: Processor mode may be determined incorrectly
Nadav Amit [Fri, 18 Apr 2014 04:11:34 +0000 (07:11 +0300)]
KVM: x86: Processor mode may be determined incorrectly

If EFER.LMA is off, cs.l does not determine execution mode.
Currently, the emulation engine assumes differently.

Signed-off-by: Nadav Amit <namit@cs.technion.ac.il>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
10 years agoKVM: x86: IN instruction emulation should ignore REP-prefix
Nadav Amit [Fri, 18 Apr 2014 00:35:10 +0000 (03:35 +0300)]
KVM: x86: IN instruction emulation should ignore REP-prefix

The IN instruction is not be affected by REP-prefix as INS is.  Therefore, the
emulation should ignore the REP prefix as well.  The current emulator
implementation tries to perform writeback when IN instruction with REP-prefix
is emulated. This causes it to perform wrong memory write or spurious #GP
exception to be injected to the guest.

Signed-off-by: Nadav Amit <namit@cs.technion.ac.il>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
10 years agoKVM: x86: Fix CR3 reserved bits
Nadav Amit [Fri, 18 Apr 2014 00:35:09 +0000 (03:35 +0300)]
KVM: x86: Fix CR3 reserved bits

According to Intel specifications, PAE and non-PAE does not have any reserved
bits.  In long-mode, regardless to PCIDE, only the high bits (above the
physical address) are reserved.

Signed-off-by: Nadav Amit <namit@cs.technion.ac.il>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
10 years agoKVM: x86: Fix wrong/stuck PMU when guest does not use PMI
Nadav Amit [Fri, 18 Apr 2014 00:35:08 +0000 (03:35 +0300)]
KVM: x86: Fix wrong/stuck PMU when guest does not use PMI

If a guest enables a performance counter but does not enable PMI, the
hypervisor currently does not reprogram the performance counter once it
overflows.  As a result the host performance counter is kept with the original
sampling period which was configured according to the value of the guest's
counter when the counter was enabled.

Such behaviour can cause very bad consequences. The most distrubing one can
cause the guest not to make any progress at all, and keep exiting due to host
PMI before any guest instructions is exeucted. This situation occurs when the
performance counter holds a very high value when the guest enables the
performance counter. As a result the host's sampling period is configured to be
very short. The host then never reconfigures the sampling period and get stuck
at entry->PMI->exit loop. We encountered such a scenario in our experiments.

The solution is to reprogram the counter even if the guest does not use PMI.

Signed-off-by: Nadav Amit <namit@cs.technion.ac.il>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
10 years agoKVM: nVMX: Advertise support for interrupt acknowledgement
Bandan Das [Sat, 19 Apr 2014 22:17:46 +0000 (18:17 -0400)]
KVM: nVMX: Advertise support for interrupt acknowledgement

Some Type 1 hypervisors such as XEN won't enable VMX without it present

Signed-off-by: Bandan Das <bsd@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
10 years agoKVM: nVMX: Ack and write vector info to intr_info if L1 asks us to
Bandan Das [Sat, 19 Apr 2014 22:17:45 +0000 (18:17 -0400)]
KVM: nVMX: Ack and write vector info to intr_info if L1 asks us to

This feature emulates the "Acknowledge interrupt on exit" behavior.
We can safely emulate it for L1 to run L2 even if L0 itself has it
disabled (to run L1).

Signed-off-by: Bandan Das <bsd@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
10 years agoKVM: nVMX: Don't advertise single context invalidation for invept
Bandan Das [Sat, 19 Apr 2014 22:17:44 +0000 (18:17 -0400)]
KVM: nVMX: Don't advertise single context invalidation for invept

For single context invalidation, we fall through to global
invalidation in handle_invept() except for one case - when
the operand supplied by L1 is different from what we have in
vmcs12. However, typically hypervisors will only call invept
for the currently loaded eptp, so the condition will
never be true.

Signed-off-by: Bandan Das <bsd@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
10 years agoKVM: VMX: Advance rip to after an ICEBP instruction
Huw Davies [Wed, 16 Apr 2014 09:02:51 +0000 (10:02 +0100)]
KVM: VMX: Advance rip to after an ICEBP instruction

When entering an exception after an ICEBP, the saved instruction
pointer should point to after the instruction.

This fixes the bug here: https://bugs.launchpad.net/qemu/+bug/1119686

Signed-off-by: Huw Davies <huw@codeweavers.com>
Reviewed-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
10 years agoMerge tag 'kvm-s390-20140422' of git://git.kernel.org/pub/scm/linux/kernel/git/kvms39...
Marcelo Tosatti [Tue, 22 Apr 2014 13:51:06 +0000 (10:51 -0300)]
Merge tag 'kvm-s390-20140422' of git://git./linux/kernel/git/kvms390/linux into queue

Lazy storage key handling
-------------------------
Linux does not use the ACC and F bits of the storage key. Newer Linux
versions also do not use the storage keys for dirty and reference
tracking. We can optimize the guest handling for those guests for faults
as well as page-in and page-out by simply not caring about the guest
visible storage key. We trap guest storage key instruction to enable
those keys only on demand.

Migration bitmap

Until now s390 never provided a proper dirty bitmap.  Let's provide a
proper migration bitmap for s390. We also change the user dirty tracking
to a fault based mechanism. This makes the host completely independent
from the storage keys. Long term this will allow us to back guest memory
with large pages.

per-VM device attributes
------------------------
To avoid the introduction of new ioctls, let's provide the
attribute semanantic also on the VM-"device".

Userspace controlled CMMA
-------------------------
The CMMA assist is changed from "always on" to "on if requested" via
per-VM device attributes. In addition a callback to reset all usage
states is provided.

Proper guest DAT handling for intercepts
----------------------------------------
While instructions handled by SIE take care of all addressing aspects,
KVM/s390 currently does not care about guest address translation of
intercepts. This worked out fine, because
- the s390 Linux kernel has a 1:1 mapping between kernel virtual<->real
 for all pages up to memory size
- intercepts happen only for a small amount of cases
- all of these intercepts happen to be in the kernel text for current
  distros

Of course we need to be better for other intercepts, kernel modules etc.
We provide the infrastructure and rework all in-kernel intercepts to work
on logical addresses (paging etc) instead of real ones. The code has
been running internally for several months now, so it is time for going
public.

GDB support
-----------
We provide breakpoints, single stepping and watchpoints.

Fixes/Cleanups
--------------
- Improve program check delivery
- Factor out the handling of transactional memory  on program checks
- Use the existing define __LC_PGM_TDB
- Several cleanups in the lowcore structure
- Documentation

NOTES
-----
- All patches touching base s390 are either ACKed or written by the s390
  maintainers
- One base KVM patch "KVM: add kvm_is_error_gpa() helper"
- One patch introduces the notion of VM device attributes

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Conflicts:
include/uapi/linux/kvm.h

10 years agoKVM: s390: Factor out handle_itdb to handle TX aborts
Michael Mueller [Thu, 13 Mar 2014 11:16:45 +0000 (12:16 +0100)]
KVM: s390: Factor out handle_itdb to handle TX aborts

Factor out the new function handle_itdb(), which copies the ITDB into
guest lowcore to fully handle a TX abort.

Signed-off-by: Michael Mueller <mimu@linux.vnet.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: replace TDB_ADDR by __LC_PGM_TDB
Michael Mueller [Thu, 13 Mar 2014 18:29:09 +0000 (19:29 +0100)]
KVM: s390: replace TDB_ADDR by __LC_PGM_TDB

The generically assembled low core labels already contain the
address for the TDB.

Signed-off-by: Michael Mueller <mimu@linux.vnet.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: Drop pending interrupts on guest exit
Christian Borntraeger [Tue, 25 Mar 2014 16:09:08 +0000 (17:09 +0100)]
KVM: s390: Drop pending interrupts on guest exit

On hard exits (abort, sigkill) we have have some kvm_s390_interrupt_info
structures hanging around. Delete those on exit to avoid memory leaks.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
CC: stable@vger.kernel.org
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
10 years agoKVM: s390: no timer interrupts when single-stepping a guest
David Hildenbrand [Tue, 18 Mar 2014 09:06:14 +0000 (10:06 +0100)]
KVM: s390: no timer interrupts when single-stepping a guest

When a guest is single-stepped, we want to disable timer interrupts. Otherwise,
the guest will continuously execute the external interrupt handler and make
debugging of code where timer interrupts are enabled almost impossible.

The delivery of timer interrupts can be enforced in such sections by setting a
breakpoint and continuing execution.

In order to disable timer interrupts, they are disabled in the control register
of the guest just before SIE entry and are suppressed in the interrupt
check/delivery methods.

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>
10 years agoKVM: s390: move timer interrupt checks into own functions
David Hildenbrand [Tue, 18 Mar 2014 09:03:26 +0000 (10:03 +0100)]
KVM: s390: move timer interrupt checks into own functions

This patch moves the checks for enabled timer (clock-comparator) interrupts and pending
timer interrupts into own functions, making the code better readable and easier to
maintain.

The method kvm_cpu_has_pending_timer is filled with life.

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>
10 years agoKVM: s390: add documentation for diag 501
David Hildenbrand [Mon, 3 Feb 2014 16:18:02 +0000 (17:18 +0100)]
KVM: s390: add documentation for diag 501

Added documentation for diag 501, stating that no subfunctions are provided and
no parameters are used.

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>
10 years agoKVM: s390: hardware support for guest debugging
David Hildenbrand [Thu, 23 Jan 2014 11:26:52 +0000 (12:26 +0100)]
KVM: s390: hardware support for guest debugging

This patch adds support to debug the guest using the PER facility on s390.
Single-stepping, hardware breakpoints and hardware watchpoints are supported. In
order to use the PER facility of the guest without it noticing it, the control
registers of the guest have to be patched and access to them has to be
intercepted(stctl, stctg, lctl, lctlg).

All PER program interrupts have to be intercepted and only the relevant PER
interrupts for the guest have to be given back. Special care has to be taken
about repeated exits on the same hardware breakpoint. The intervention of the
host in the guests PER configuration is not fully transparent. PER instruction
nullification can not be used by the guest and too many storage alteration
events may be reported to the guest (if it is activated for special address
ranges only) when the host concurrently debugging it.

Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: kernel header addition for guest debugging
David Hildenbrand [Thu, 23 Jan 2014 10:03:51 +0000 (11:03 +0100)]
KVM: s390: kernel header addition for guest debugging

This patch adds the structs to the kernel headers needed to pass information
from/to userspace in order to debug a guest on s390 with hardware support.

Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: emulate stctl and stctg
David Hildenbrand [Thu, 23 Jan 2014 09:47:13 +0000 (10:47 +0100)]
KVM: s390: emulate stctl and stctg

Introduce the methods to emulate the stctl and stctg instruction. Added tracing
code.

Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: deliver program irq parameters and use correct ilc
David Hildenbrand [Mon, 3 Mar 2014 09:55:13 +0000 (10:55 +0100)]
KVM: s390: deliver program irq parameters and use correct ilc

When a program interrupt was to be delivered until now, no program interrupt
parameters were stored in the low-core of the target vcpu.

This patch enables the delivery of those program interrupt parameters, takes
care of concurrent PER events which can be injected in addition to any program
interrupt and uses the correct instruction length code (depending on the
interception code) for the injection of program interrupts.

Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: extract irq parameters of intercepted program irqs
David Hildenbrand [Mon, 3 Mar 2014 09:54:33 +0000 (10:54 +0100)]
KVM: s390: extract irq parameters of intercepted program irqs

Whenever a program interrupt is intercepted, some parameters are stored in the
sie control block. These parameters have to be extracted in order to be
reinjected correctly. This patch also takes care of intercepted PER events which
can occurr in addition to any program interrupt.

Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agos390: add fields to lowcore definition
Jens Freimann [Wed, 26 Feb 2014 16:03:29 +0000 (17:03 +0100)]
s390: add fields to lowcore definition

This patch adds fields which are currently missing but needed for the correct
injection of interrupts.

This is based on a patch by David Hildenbrand

Signed-off-by: Jens Freimann <jfrei@linux.vnet.ibm.com>
Acked-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agos390: rename and split lowcore field per_perc_atmid
Jens Freimann [Wed, 26 Feb 2014 15:32:46 +0000 (16:32 +0100)]
s390: rename and split lowcore field per_perc_atmid

per_perc_atmid is currently a two-byte field that combines two
fields, the PER code and the PER Addressing-and-Translation-Mode
Identification (ATMID)

Let's make them accessible indepently and also rename per_cause to
per_code.

Signed-off-by: Jens Freimann <jfrei@linux.vnet.ibm.com>
Acked-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agos390: fix name of lowcore field at offset 0xa3
Jens Freimann [Mon, 10 Feb 2014 09:55:37 +0000 (10:55 +0100)]
s390: fix name of lowcore field at offset 0xa3

According to the Principles of Operation, at offset 0xA3
in the lowcore we have the "Architectural-Mode identification",
not an "access identification".

Signed-off-by: Jens Freimann <jfrei@linux.vnet.ibm.com>
Acked-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: cleanup kvm_s390_real_to_abs()
Heiko Carstens [Wed, 1 Jan 2014 15:19:55 +0000 (16:19 +0100)]
KVM: s390: cleanup kvm_s390_real_to_abs()

Add kerneldoc comment to kvm_s390_real_to_abs() and change the code
so it matches the coding style of the rest of gaccess.h.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: remove old guest access functions
Heiko Carstens [Thu, 2 Jan 2014 09:46:01 +0000 (10:46 +0100)]
KVM: s390: remove old guest access functions

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: convert handle_stsi()
Heiko Carstens [Wed, 1 Jan 2014 15:58:59 +0000 (16:58 +0100)]
KVM: s390: convert handle_stsi()

Convert handle_stsi() to new guest access functions.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: convert handle lctl[g]()
Heiko Carstens [Wed, 1 Jan 2014 15:59:21 +0000 (16:59 +0100)]
KVM: s390: convert handle lctl[g]()

Convert handle lctl[g]() to new guest access functions.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: convert handle_stidp()
Heiko Carstens [Wed, 1 Jan 2014 15:58:16 +0000 (16:58 +0100)]
KVM: s390: convert handle_stidp()

Convert handle_stidp() to new guest access functions.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: convert handle_lpsw[e]()
Heiko Carstens [Wed, 1 Jan 2014 15:57:42 +0000 (16:57 +0100)]
KVM: s390: convert handle_lpsw[e]()

Convert handle_lpsw[e]() to new guest access functions.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: reinject io interrupt on tpi failure
Cornelia Huck [Wed, 8 Jan 2014 17:07:54 +0000 (18:07 +0100)]
KVM: s390: reinject io interrupt on tpi failure

The tpi instruction should be suppressed on addressing and protection
exceptions, so we need to re-inject the dequeued io interrupt in that
case.

Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: convert handle_tpi()
Heiko Carstens [Wed, 1 Jan 2014 15:55:48 +0000 (16:55 +0100)]
KVM: s390: convert handle_tpi()

Convert handle_tpi() to new guest access functions.

The code now sets up a structure which is copied with a single call to
guest space instead of issuing several separate guest access calls.
This is necessary since the to be copied data may cross a page boundary.
If a protection exception happens while accessing any of the pages, the
instruction is suppressed and may not have modified any memory contents.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: convert handle_test_block()
Heiko Carstens [Wed, 1 Jan 2014 15:53:49 +0000 (16:53 +0100)]
KVM: s390: convert handle_test_block()

Convert handle_test_block() to new guest access functions.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: convert handle_store_cpu_address()
Heiko Carstens [Wed, 1 Jan 2014 15:53:27 +0000 (16:53 +0100)]
KVM: s390: convert handle_store_cpu_address()

Convert handle_store_cpu_address() to new guest access functions.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: convert handle_store_prefix()
Heiko Carstens [Wed, 1 Jan 2014 15:52:47 +0000 (16:52 +0100)]
KVM: s390: convert handle_store_prefix()

Convert handle_store_prefix() to new guest access functions.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: convert handle_set_clock()
Heiko Carstens [Wed, 1 Jan 2014 15:50:11 +0000 (16:50 +0100)]
KVM: s390: convert handle_set_clock()

Convert handle_set_clock() to new guest access functions.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: convert __sigp_set_prefix()/handle_set_prefix()
Heiko Carstens [Wed, 1 Jan 2014 15:47:12 +0000 (16:47 +0100)]
KVM: s390: convert __sigp_set_prefix()/handle_set_prefix()

Convert __sigp_set_prefix() and handle_set_prefix() to new guest
access functions.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: convert kvm_s390_store_status_unloaded()
Heiko Carstens [Wed, 1 Jan 2014 15:45:58 +0000 (16:45 +0100)]
KVM: s390: convert kvm_s390_store_status_unloaded()

Convert kvm_s390_store_status_unloaded() to new guest access functions.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: convert handle_prog()
Heiko Carstens [Wed, 1 Jan 2014 15:37:47 +0000 (16:37 +0100)]
KVM: s390: convert handle_prog()

Convert handle_prog() to new guest access functions.
Also make the code a bit more readable and look at the return code
of write_guest_lc() which was missing before.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: convert pfault code
Heiko Carstens [Wed, 1 Jan 2014 15:36:07 +0000 (16:36 +0100)]
KVM: s390: convert pfault code

Convert pfault code to new guest access functions.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Dominik Dingel <dingel@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: convert handle_stfl()
Heiko Carstens [Wed, 1 Jan 2014 15:56:41 +0000 (16:56 +0100)]
KVM: s390: convert handle_stfl()

Convert handle_stfl() to new guest access functions.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: convert local irqs in __do_deliver_interrupt()
Jens Freimann [Wed, 12 Feb 2014 13:05:38 +0000 (14:05 +0100)]
KVM: s390: convert local irqs in  __do_deliver_interrupt()

Convert local irqs in __do_deliver_interrupt() to new guest
access functions.

Signed-off-by: Jens Freimann <jfrei@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: convert __do_deliver_interrupt()
Heiko Carstens [Thu, 2 Jan 2014 09:59:41 +0000 (10:59 +0100)]
KVM: s390: convert __do_deliver_interrupt()

Convert __do_deliver_interrupt() to new guest access functions.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: make use of ipte lock
Heiko Carstens [Fri, 10 Jan 2014 13:33:28 +0000 (14:33 +0100)]
KVM: s390: make use of ipte lock

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390/sclp: correctly set eca siif bit
Heiko Carstens [Mon, 30 Dec 2013 11:54:14 +0000 (12:54 +0100)]
KVM: s390/sclp: correctly set eca siif bit

Check if siif is available before setting.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: add architecture compliant guest access functions
Heiko Carstens [Wed, 1 Jan 2014 15:26:52 +0000 (16:26 +0100)]
KVM: s390: add architecture compliant guest access functions

The new guest memory access function write_guest() and read_guest() can be
used to access guest memory in an architecture compliant way.
These functions will look at the vcpu's PSW and select the correct address
space for memory access and also perform correct address wrap around.
In case DAT is turned on, page tables will be walked otherwise access will
happen to real or absolute memory.

Any access exception will be recognized and exception data will be stored
in the vcpu's kvm_vcpu_arch.pgm member. Subsequently an exception can be
injected if necessary.

Missing are:
- key protection checks
- access register mode support
- program event recording support

This patch also adds write_guest_real(), read_guest_real(),
write_guest_absolute() and read_guest_absolute() guest functions which can
be used to access real and absolute storage. These functions currently do
not perform any access checks, since there is no use case (yet?).

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: add lowcore access functions
Heiko Carstens [Wed, 1 Jan 2014 15:23:29 +0000 (16:23 +0100)]
KVM: s390: add lowcore access functions

put_guest_lc, read_guest_lc and write_guest_lc are guest access
functions which shall only be used to access the lowcore of a vcpu.
These functions should be used for e.g. interrupt handlers where no
guest memory access protection facilities, like key or low address
protection, are applicable.

At a later point guest vcpu lowcore access should happen via pinned
prefix pages, so that these pages can be accessed directly via the
kernel mapping. All of these *_lc functions can be removed then.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: add 'pgm' member to kvm_vcpu_arch and helper function
Heiko Carstens [Wed, 1 Jan 2014 15:31:49 +0000 (16:31 +0100)]
KVM: s390: add 'pgm' member to kvm_vcpu_arch and helper function

Add a 'struct kvm_s390_pgm_info pgm' member to kvm_vcpu_arch. This
structure will be used if during instruction emulation in the context
of a vcpu exception data needs to be stored somewhere.

Also add a helper function kvm_s390_inject_prog_cond() which can inject
vcpu's last exception if needed.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: add kvm_s390_logical_to_effective() helper
Heiko Carstens [Wed, 1 Jan 2014 15:21:47 +0000 (16:21 +0100)]
KVM: s390: add kvm_s390_logical_to_effective() helper

Add kvm_s390_logical_to_effective() helper which converts a guest vcpu's
logical storage address to a guest vcpu effective address by applying the
rules of the vcpu's addressing mode defined by PSW bits 31 and 32
(extendended and basic addressing mode).
Depending on the vcpu's addressing mode the upper 40 bits (24 bit addressing
mode), 33 bits (31 bit addressing mode) or no bits (64 bit addressing mode)
will be zeroed and the remaining bits will be returned.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agos390/ctl_reg: add union type for control register 0
Heiko Carstens [Wed, 1 Jan 2014 15:08:37 +0000 (16:08 +0100)]
s390/ctl_reg: add union type for control register 0

Add 'union ctlreg0_bits' to easily allow setting and testing bits of
control register 0 bits.
This patch only adds the bits needed for the new guest access functions.
Other bits and control registers can be added when needed.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agos390/ptrace: add struct psw and accessor function
Heiko Carstens [Wed, 1 Jan 2014 15:08:06 +0000 (16:08 +0100)]
s390/ptrace: add struct psw and accessor function

Introduce a 'struct psw' which makes it easier to decode and test if
certain bits in a psw are set or are not set.
In addition also add a 'psw_bits()' helper define which allows to
directly modify and test a psw_t structure. E.g.

psw_t psw;
psw_bits(psw).t = 1; /* set dat bit */

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: export test_vfacility()
Heiko Carstens [Tue, 17 Dec 2013 08:08:28 +0000 (09:08 +0100)]
KVM: s390: export test_vfacility()

Make test_vfacility() available for other files. This is needed for the
new guest access functions, which test if certain facilities are available
for a guest.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: add kvm_is_error_gpa() helper
Heiko Carstens [Wed, 1 Jan 2014 15:09:21 +0000 (16:09 +0100)]
KVM: add kvm_is_error_gpa() helper

It's quite common (in the s390 guest access code) to test if a guest
physical address points to a valid guest memory area or not.
So add a simple helper function in common code, since this might be
of interest for other architectures as well.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: allow injecting every kind of interrupt
Jens Freimann [Tue, 11 Feb 2014 10:07:05 +0000 (11:07 +0100)]
KVM: s390: allow injecting every kind of interrupt

Add a new data structure and function that allows to inject
all kinds of interrupt as defined in the PoP

Signed-off-by: Jens Freimann <jfrei@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: Exploiting generic userspace interface for cmma
Dominik Dingel [Wed, 9 Apr 2014 11:13:00 +0000 (13:13 +0200)]
KVM: s390: Exploiting generic userspace interface for cmma

To enable CMMA and to reset its state we use the vm kvm_device ioctls,
encapsulating attributes within the KVM_S390_VM_MEM_CTRL group.

Signed-off-by: Dominik Dingel <dingel@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: make cmma usage conditionally
Dominik Dingel [Tue, 25 Mar 2014 12:47:11 +0000 (13:47 +0100)]
KVM: s390: make cmma usage conditionally

When userspace reset the guest without notifying kvm, the CMMA state
of the pages might be unused, resulting in guest data corruption.
To avoid this, CMMA must be enabled only if userspace understands
the implications.

CMMA must be enabled before vCPU creation. It can't be switched off
once enabled.  All subsequently created vCPUs will be enabled for
CMMA according to the CMMA state of the VM.

Signed-off-by: Dominik Dingel <dingel@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
[remove now unnecessary calls to page_table_reset_pgste]

10 years agoKVM: s390: Per-vm kvm device controls
Dominik Dingel [Wed, 9 Apr 2014 11:13:00 +0000 (13:13 +0200)]
KVM: s390: Per-vm kvm device controls

We sometimes need to get/set attributes specific to a virtual machine
and so need something else than ONE_REG.

Let's copy the KVM_DEVICE approach, and define the respective ioctls
for the vm file descriptor.

Signed-off-by: Dominik Dingel <dingel@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Acked-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: Add proper dirty bitmap support to S390 kvm.
Jason J. Herne [Thu, 2 Aug 2012 14:10:17 +0000 (10:10 -0400)]
KVM: s390: Add proper dirty bitmap support to S390 kvm.

Replace the kvm_s390_sync_dirty_log() stub with code to construct the KVM
dirty_bitmap from S390 memory change bits.  Also add code to properly clear
the dirty_bitmap size when clearing the bitmap.

Signed-off-by: Jason J. Herne <jjherne@us.ibm.com>
CC: Dominik Dingel <dingel@linux.vnet.ibm.com>
[Dominik Dingel: use gmap_test_and_clear_dirty, locking fixes]
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390/mm: new gmap_test_and_clear_dirty function
Dominik Dingel [Mon, 24 Mar 2014 13:27:58 +0000 (14:27 +0100)]
KVM: s390/mm: new gmap_test_and_clear_dirty function

For live migration kvm needs to test and clear the dirty bit of guest pages.

That for is ptep_test_and_clear_user_dirty, to be sure we are not racing with
other code, we protect the pte. This needs to be done within
the architecture memory management code.

Signed-off-by: Dominik Dingel <dingel@linux.vnet.ibm.com>
Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390/mm: use software dirty bit detection for user dirty tracking
Martin Schwidefsky [Fri, 18 Oct 2013 10:03:41 +0000 (12:03 +0200)]
KVM: s390/mm: use software dirty bit detection for user dirty tracking

Switch the user dirty bit detection used for migration from the hardware
provided host change-bit in the pgste to a fault based detection method.
This reduced the dependency of the host from the storage key to a point
where it becomes possible to enable the RCP bypass for KVM guests.

The fault based dirty detection will only indicate changes caused
by accesses via the guest address space. The hardware based method
can detect all changes, even those caused by I/O or accesses via the
kernel page table. The KVM/qemu code needs to take this into account.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Dominik Dingel <dingel@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: Don't enable skeys by default
Dominik Dingel [Tue, 14 Jan 2014 17:11:14 +0000 (18:11 +0100)]
KVM: s390: Don't enable skeys by default

The first invocation of storage key operations on a given cpu will be intercepted.

On these intercepts we will enable storage keys for the guest and remove the
previously added intercepts.

Signed-off-by: Dominik Dingel <dingel@linux.vnet.ibm.com>
Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: Allow skeys to be enabled for the current process
Dominik Dingel [Tue, 14 Jan 2014 17:10:17 +0000 (18:10 +0100)]
KVM: s390: Allow skeys to be enabled for the current process

Introduce a new function s390_enable_skey(), which enables storage key
handling via setting the use_skey flag in the mmu context.

This function is only useful within the context of kvm.

Note that enabling storage keys will cause a one-time hickup when
walking the page table; however, it saves us special effort for cases
like clear reset while making it possible for us to be architecture
conform.

s390_enable_skey() takes the page table lock to prevent reseting
storage keys triggered from multiple vcpus.

Signed-off-by: Dominik Dingel <dingel@linux.vnet.ibm.com>
Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: Clear storage keys
Dominik Dingel [Wed, 29 Jan 2014 15:02:32 +0000 (16:02 +0100)]
KVM: s390: Clear storage keys

page_table_reset_pgste() already does a complete page table walk to
reset the pgste. Enhance it to initialize the storage keys to
PAGE_DEFAULT_KEY if requested by the caller. This will be used
for lazy storage key handling. Also provide an empty stub for
!CONFIG_PGSTE

Lets adopt the current code (diag 308) to not clear the keys.

Signed-off-by: Dominik Dingel <dingel@linux.vnet.ibm.com>
Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: s390: Adding skey bit to mmu context
Dominik Dingel [Tue, 14 Jan 2014 14:02:11 +0000 (15:02 +0100)]
KVM: s390: Adding skey bit to mmu context

For lazy storage key handling, we need a mechanism to track if the
process ever issued a storage key operation.

This patch adds the basic infrastructure for making the storage
key handling optional, but still leaves it enabled for now by default.

Signed-off-by: Dominik Dingel <dingel@linux.vnet.ibm.com>
Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
10 years agoKVM: x86: Fix CR3 and LDT sel should not be saved in TSS
Nadav Amit [Mon, 7 Apr 2014 15:37:47 +0000 (18:37 +0300)]
KVM: x86: Fix CR3 and LDT sel should not be saved in TSS

According to Intel specifications, only general purpose registers and segment
selectors should be saved in the old TSS during 32-bit task-switch.

Signed-off-by: Nadav Amit <namit@cs.technion.ac.il>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
10 years agoKVM: VMX: speed up wildcard MMIO EVENTFD
Michael S. Tsirkin [Mon, 31 Mar 2014 18:50:44 +0000 (21:50 +0300)]
KVM: VMX: speed up wildcard MMIO EVENTFD

With KVM, MMIO is much slower than PIO, due to the need to
do page walk and emulation. But with EPT, it does not have to be: we
know the address from the VMCS so if the address is unique, we can look
up the eventfd directly, bypassing emulation.

Unfortunately, this only works if userspace does not need to match on
access length and data.  The implementation adds a separate FAST_MMIO
bus internally. This serves two purposes:
    - minimize overhead for old userspace that does not use eventfd with lengtth = 0
    - minimize disruption in other code (since we don't know the length,
      devices on the MMIO bus only get a valid address in write, this
      way we don't need to touch all devices to teach them to handle
      an invalid length)

At the moment, this optimization only has effect for EPT on x86.

It will be possible to speed up MMIO for NPT and MMU using the same
idea in the future.

With this patch applied, on VMX MMIO EVENTFD is essentially as fast as PIO.
I was unable to detect any measureable slowdown to non-eventfd MMIO.

Making MMIO faster is important for the upcoming virtio 1.0 which
includes an MMIO signalling capability.

The idea was suggested by Peter Anvin.  Lots of thanks to Gleb for
pre-review and suggestions.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
10 years agoKVM: support any-length wildcard ioeventfd
Michael S. Tsirkin [Mon, 31 Mar 2014 18:50:38 +0000 (21:50 +0300)]
KVM: support any-length wildcard ioeventfd

It is sometimes benefitial to ignore IO size, and only match on address.
In hindsight this would have been a better default than matching length
when KVM_IOEVENTFD_FLAG_DATAMATCH is not set, In particular, this kind
of access can be optimized on VMX: there no need to do page lookups.
This can currently be done with many ioeventfds but in a suboptimal way.

However we can't change kernel/userspace ABI without risk of breaking
some applications.
Use len = 0 to mean "ignore length for matching" in a more optimal way.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
10 years agoKVM: x86: Fix page-tables reserved bits
Nadav Amit [Fri, 4 Apr 2014 03:31:04 +0000 (06:31 +0300)]
KVM: x86: Fix page-tables reserved bits

KVM does not handle the reserved bits of x86 page tables correctly:
In PAE, bits 5:8 are reserved in the PDPTE.
In IA-32e, bit 8 is not reserved.

Signed-off-by: Nadav Amit <namit@cs.technion.ac.il>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
10 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux
Linus Torvalds [Wed, 16 Apr 2014 18:28:25 +0000 (11:28 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/s390/linux

Pull s390 patches from Martin Schwidefsky:
 "An update to the oops output with additional information about the
  crash.  The renameat2 system call is enabled.  Two patches in regard
  to the PTR_ERR_OR_ZERO cleanup.  And a bunch of bug fixes"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux:
  s390/sclp_cmd: replace PTR_RET with PTR_ERR_OR_ZERO
  s390/sclp: replace PTR_RET with PTR_ERR_OR_ZERO
  s390/sclp_vt220: Fix kernel panic due to early terminal input
  s390/compat: fix typo
  s390/uaccess: fix possible register corruption in strnlen_user_srst()
  s390: add 31 bit warning message
  s390: wire up sys_renameat2
  s390: show_registers() should not map user space addresses to kernel symbols
  s390/mm: print control registers and page table walk on crash
  s390/smp: fix smp_stop_cpu() for !CONFIG_SMP
  s390: fix control register update

10 years agoMerge tag 'please-pull-ia64-erratum' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Wed, 16 Apr 2014 18:22:45 +0000 (11:22 -0700)]
Merge tag 'please-pull-ia64-erratum' of git://git./linux/kernel/git/aegl/linux

Pull itanium erratum fix from Tony Luck:
 "Small workaround for a rare, but annoying, erratum #237"

* tag 'please-pull-ia64-erratum' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux:
  [IA64] Change default PSR.ac from '1' to '0' (Fix erratum #237)

10 years ago[IA64] Change default PSR.ac from '1' to '0' (Fix erratum #237)
Tony Luck [Fri, 28 Mar 2014 21:42:07 +0000 (14:42 -0700)]
[IA64] Change default PSR.ac from '1' to '0' (Fix erratum #237)

April 2014 Itanium processor specification update:

http://www.intel.com/content/www/us/en/processors/itanium/itanium-specification-update.html

describes this erratum:

=========================================================================
237. Under a complex set of conditions, store to load forwarding for a
sub 8-byte load may complete incorrectly

Problem: A load instruction may complete incorrectly when a code sequence
using 4-byte or smaller load and store operations to the same address
is executed in combination with specific timing of all the following
concurrent conditions: store to load forwarding, alignment checking
enabled, a mis-predicted branch, and complex cache utilization activity.

Implication: The affected sub 8-byte instruction may complete
incorrectly resulting in unpredictable system behavior. There is an
extremely low probability of exposure due to the significant number of
complex microarchitectural concurrent conditions required to encounter
the erratum.

Workaround: Set PSR.ac = 0 to completely avoid the erratum. Disabling
Hyper-Threading will significantly reduce exposure to the conditions
that contribute to encountering the erratum.

Status: See the Summary Table of Changes for the affected steppings.
=========================================================================

[Table of changes essentially lists all models from McKinley to Tukwila]

The PSR.ac bit controls whether the processor will always generate
an unaligned reference trap (0x5a00) for a misaligned data access
(when PSR.ac=1) or if it will let the access succeed when running
on a cpu that implements logic to handle some unaligned accesses.

Way back in 2008 in commit b704882e70d87d7f56db5ff17e2253f3fa90e4f3
  [IA64] Rationalize kernel mode alignment checking
we made the decision to always enable strict checking. We were
already doing so in trap/interrupt context because the common
preamble code set this bit - but the rest of supervisor code
(and by inheritance user code) ran with PSR.ac=0.

We now reverse that decision and set PSR.ac=0 everywhere in the
kernel (also inherited by user processes). This will avoid the
erratum using the method described in the Itanium specification
update.  Net effect for users is that the processor will handle
unaligned access when it can (typically with a tiny performance
bubble in the pipeline ... but much less invasive than taking a
trap and having the OS perform the access).

Signed-off-by: Tony Luck <tony.luck@intel.com>
10 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Wed, 16 Apr 2014 03:30:30 +0000 (20:30 -0700)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Fix BPF filter validation of netlink attribute accesses, from
    Mathias Kruase.

 2) Netfilter conntrack generation seqcount not initialized properly,
    from Andrey Vagin.

 3) Fix comparison mask computation on big-endian in nft_cmp_fast(),
    from Patrick McHardy.

 4) Properly limit MTU over ipv6, from Eric Dumazet.

 5) Fix seccomp system call argument population on 32-bit, from Daniel
    Borkmann.

 6) skb_network_protocol() should not use hard-coded ETH_HLEN, instead
    skb->mac_len needs to be used.  From Vlad Yasevich.

 7) We have several cases of using socket based communications to
    implement a tunnel.  For example, some tunnels are encapsulations
    over UDP so we use an internal kernel UDP socket to do the
    transmits.

    These tunnels should behave just like other software devices and
    pass the packets on down to the next layer.

    Most importantly we want the top-level socket (eg TCP) that created
    the traffic to be charged for the SKB memory.

    However, once you get into the IP output path, we have code that
    assumed that whatever was attached to skb->sk is an IP socket.

    To keep the top-level socket being charged for the SKB memory,
    whilst satisfying the needs of the IP output path, we now pass in an
    explicit 'sk' argument.

    From Eric Dumazet.

 8) ping_init_sock() leaks group info, from Xiaoming Wang.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (33 commits)
  cxgb4: use the correct max size for firmware flash
  qlcnic: Fix MSI-X initialization code
  ip6_gre: don't allow to remove the fb_tunnel_dev
  ipv4: add a sock pointer to dst->output() path.
  ipv4: add a sock pointer to ip_queue_xmit()
  driver/net: cosa driver uses udelay incorrectly
  at86rf230: fix __at86rf230_read_subreg function
  at86rf230: remove check if AVDD settled
  net: cadence: Add architecture dependencies
  net: Start with correct mac_len in skb_network_protocol
  Revert "net: sctp: Fix a_rwnd/rwnd management to reflect real state of the receiver's buffer"
  cxgb4: Save the correct mac addr for hw-loopback connections in the L2T
  net: filter: seccomp: fix wrong decoding of BPF_S_ANC_SECCOMP_LD_W
  seccomp: fix populating a0-a5 syscall args in 32-bit x86 BPF
  qlcnic: Do not disable SR-IOV when VFs are assigned to VMs
  qlcnic: Fix QLogic application/driver interface for virtual NIC configuration
  qlcnic: Fix PVID configuration on eSwitch port.
  qlcnic: Fix max ring count calculation
  qlcnic: Fix to send INIT_NIC_FUNC as first mailbox.
  qlcnic: Fix panic due to uninitialzed delayed_work struct in use.
  ...

10 years agocxgb4: use the correct max size for firmware flash
Steve Wise [Tue, 15 Apr 2014 19:22:34 +0000 (14:22 -0500)]
cxgb4: use the correct max size for firmware flash

The wrong max fw size was being used and causing false
"too big" errors running ethtool -f.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: David S. Miller <davem@davemloft.net>