Skip to content
  1. May 27, 2014
    • Paolo Bonzini's avatar
      KVM: lapic: sync highest ISR to hardware apic on EOI · fc57ac2c
      Paolo Bonzini authored 
      

When Hyper-V enlightenments are in effect, Windows prefers to issue an
Hyper-V MSR write to issue an EOI rather than an x2apic MSR write.
The Hyper-V MSR write is not handled by the processor, and besides
being slower, this also causes bugs with APIC virtualization.  The
reason is that on EOI the processor will modify the highest in-service
interrupt (SVI) field of the VMCS, as explained in section 29.1.4 of
the SDM; every other step in EOI virtualization is already done by
apic_send_eoi or on VM entry, but this one is missing.

We need to do the same, and be careful not to muck with the isr_count
and highest_isr_cache fields that are unused when virtual interrupt
delivery is enabled.

Cc: stable@vger.kernel.org
Reviewed-by: default avatarYang Zhang <yang.z.zhang@intel.com>
Signed-off-by: default avatarPaolo Bonzini <pbonzini@redhat.com>
      fc57ac2c
  3. May 22, 2014
    • Nadav Amit's avatar
      KVM: vmx: DR7 masking on task switch emulation is wrong · 1f854112
      Nadav Amit authored 
      

The DR7 masking which is done on task switch emulation should be in hex format
(clearing the local breakpoints enable bits 0,2,4 and 6).

Signed-off-by: default avatarNadav Amit <namit@cs.technion.ac.il>
Signed-off-by: default avatarPaolo Bonzini <pbonzini@redhat.com>
      1f854112
    • Dave Hansen's avatar
      x86: fix page fault tracing when KVM guest support enabled · 65a7f03f
      Dave Hansen authored 
      I noticed on some of my systems that page fault tracing doesn't
work:

	cd /sys/kernel/debug/tracing
	echo 1 > events/exceptions/enable
	cat trace;
	# nothing shows up

I eventually traced it down to CONFIG_KVM_GUEST.  At least in a
KVM VM, enabling that option breaks page fault tracing, and
disabling fixes it.  I tried on some old kernels and this does
not appear to be a regression: it never worked.

There are two page-fault entry functions today.  One when tracing
is on and another when it is off.  The KVM code calls do_page_fault()
directly instead of calling the traced version:

> dotraplinkage void __kprobes
> do_async_page_fault(struct pt_regs *regs, unsigned long
> error_code)
> {
>         enum ctx_state prev_state;
>
>         switch (kvm_read_and_reset_pf_reason()) {
>         default:
>                 do_page_fault(regs, error_code);
>                 break;
>         case KVM_PV_REASON_PAGE_NOT_PRESENT:

I'm also having problems with the page fault tracing on bare
metal (same symptom of no trace output).  I'm unsure if it's
related.

Steven had an alternative to this which has zero overhead when
tracing is off where this includes the standard noops even when
tracing is disabled.  I'm unconvinced that the extra complexity
of his apporach:

	http://lkml.kernel.org/r/20140508194508.561ed220@gandalf.local.home



is worth it, expecially considering that the KVM code is already
making page fault entry slower here.  This solution is
dirt-simple.

Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: x86@kernel.org
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Gleb Natapov <gleb@redhat.com>
Cc: kvm@vger.kernel.org
Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: default avatarDave Hansen <dave.hansen@linux.intel.com>
Acked-by: default avatar"H. Peter Anvin" <hpa@zytor.com>
Acked-by: default avatarSteven Rostedt <rostedt@goodmis.org>
Signed-off-by: default avatarPaolo Bonzini <pbonzini@redhat.com>
      65a7f03f
    • Paolo Bonzini's avatar
      KVM: x86: get CPL from SS.DPL · ae9fedc7
      Paolo Bonzini authored 
      

CS.RPL is not equal to the CPL in the few instructions between
setting CR0.PE and reloading CS.  And CS.DPL is also not equal
to the CPL for conforming code segments.

However, SS.DPL *is* always equal to the CPL except for the weird
case of SYSRET on AMD processors, which sets SS.DPL=SS.RPL from the
value in the STAR MSR, but force CPL=3 (Intel instead forces
SS.DPL=SS.RPL=CPL=3).

So this patch:

- modifies SVM to update the CPL from SS.DPL rather than CS.RPL;
the above case with SYSRET is not broken further, and the way
to fix it would be to pass the CPL to userspace and back

- modifies VMX to always return the CPL from SS.DPL (except
forcing it to 0 if we are emulating real mode via vm86 mode;
in vm86 mode all DPLs have to be 3, but real mode does allow
privileged instructions).  It also removes the CPL cache,
which becomes a duplicate of the SS access rights cache.

This fixes doing KVM_IOCTL_SET_SREGS exactly after setting
CR0.PE=1 but before CS has been reloaded.

Signed-off-by: default avatarPaolo Bonzini <pbonzini@redhat.com>
      ae9fedc7
    • Paolo Bonzini's avatar
      KVM: x86: check CS.DPL against RPL during task switch · 5045b468
      Paolo Bonzini authored 
      

Table 7-1 of the SDM mentions a check that the code segment's
DPL must match the selector's RPL.  This was not done by KVM,
fix it.

Signed-off-by: default avatarPaolo Bonzini <pbonzini@redhat.com>
      5045b468
    • Paolo Bonzini's avatar
      KVM: x86: drop set_rflags callback · fb5e336b
      Paolo Bonzini authored 
      

Not needed anymore now that the CPL is computed directly
during task switch.

Signed-off-by: default avatarPaolo Bonzini <pbonzini@redhat.com>
      fb5e336b
    • Paolo Bonzini's avatar
      KVM: x86: use new CS.RPL as CPL during task switch · 2356aaeb
      Paolo Bonzini authored 
      

During task switch, all of CS.DPL, CS.RPL, SS.DPL must match (in addition
to all the other requirements) and will be the new CPL.  So far this
worked by carefully setting the CS selector and flag before doing the
task switch; setting CS.selector will already change the CPL.

However, this will not work once we get the CPL from SS.DPL, because
then you will have to set the full segment descriptor cache to change
the CPL.  ctxt->ops->cpl(ctxt) will then return the old CPL during the
task switch, and the check that SS.DPL == CPL will fail.

Temporarily assume that the CPL comes from CS.RPL during task switch
to a protected-mode task.  This is the same approach used in QEMU's
emulation code, which (until version 2.0) manually tracks the CPL.

Signed-off-by: default avatarPaolo Bonzini <pbonzini@redhat.com>
      2356aaeb
  5. May 17, 2014
    • Paolo Bonzini's avatar
      Merge tag 'kvm-s390-20140516' of... · afa538f0
      Paolo Bonzini authored 
      Merge tag 'kvm-s390-20140516' of git://git.kernel.org/pub/scm/linux/kernel/git/kvms390/linux into kvm-next

1. Correct locking for lazy storage key handling
   A test loop with multiple CPUs triggered a race in the lazy storage
   key handling as introduced by commit 934bc131
   (KVM: s390: Allow skeys to be enabled for the current process). This
   race should not happen with Linux guests, but let's fix it anyway.
   Patch touches !/kvm/ code, but is from the s390 maintainer.

2. Better handling of broken guests
   If we detect a program check loop we stop the guest instead of
   wasting CPU cycles.

3. Better handling on MVPG emulation
   The move page handling is improved to be architecturally correct.

3. Trace point rework
   Let's rework the kvm trace points to have a common header file (for
   later perf usage) and provided a table based instruction decoder.

4. Interpretive execution of SIGP external call
   Let the hardware handle most cases of SIGP external call (IPI) and
   wire up the fixup code for the corner cases.

5. Initial preparations for the IBC facility
   Prepare the code to handle instruction blocking
      afa538f0
  7. May 16, 2014
  9. May 13, 2014
  11. May 08, 2014
    • Gabriel L. Somlo's avatar
      kvm: x86: emulate monitor and mwait instructions as nop · 87c00572
      Gabriel L. Somlo authored 
      

Treat monitor and mwait instructions as nop, which is architecturally
correct (but inefficient) behavior. We do this to prevent misbehaving
guests (e.g. OS X <= 10.7) from crashing after they fail to check for
monitor/mwait availability via cpuid.

Since mwait-based idle loops relying on these nop-emulated instructions
would keep the host CPU pegged at 100%, do NOT advertise their presence
via cpuid, to prevent compliant guests from using them inadvertently.

Signed-off-by: default avatarGabriel L. Somlo <somlo@cmu.edu>
Signed-off-by: default avatarPaolo Bonzini <pbonzini@redhat.com>
      87c00572
    • Michael S. Tsirkin's avatar
      kvm/x86: implement hv EOI assist · b63cf42f
      Michael S. Tsirkin authored 
      

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: default avatarMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: default avatarPaolo Bonzini <pbonzini@redhat.com>
      b63cf42f
  13. May 07, 2014
  15. May 06, 2014
  17. May 05, 2014
    • Christian Borntraeger's avatar
      kvm/irqchip: Speed up KVM_SET_GSI_ROUTING · 719d93cd
      Christian Borntraeger authored 
      

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: default avatarChristian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: default avatarChristian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: default avatarPaolo Bonzini <pbonzini@redhat.com>
      719d93cd
  19. Apr 30, 2014
    • Paolo Bonzini's avatar
      Merge tag 'kvm-s390-20140429' of... · 57b5981c
      Paolo Bonzini authored 
      Merge tag 'kvm-s390-20140429' of git://git.kernel.org/pub/scm/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
      57b5981c
  21. Apr 29, 2014