If simultaneous NMIs happen, we're supposed to queue the second
and next (collapsing them), but currently we sometimes collapse
the second into the first.
Fix by using a counter for pending NMIs instead of a bool; since
the counter limit depends on whether the processor is currently
in an NMI handler, which can only be checked in vcpu context
(via the NMI mask), we add a new KVM_REQ_NMI to request recalculation
of the counter.
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
KVM assumed in several places that reading the TSC MSR returns the value for
L1. This is incorrect, because when L2 is running, the correct TSC read exit
emulation is to return L2's value.
We therefore add a new x86_ops function, read_l1_tsc, to use in places that
specifically need to read the L1 TSC, NOT the TSC of the current level of
guest.
Note that one change, of one line in kvm_arch_vcpu_load, is made redundant
by a different patch sent by Zachary Amsden (and not yet applied):
kvm_arch_vcpu_load() should not read the guest TSC, and if it didn't, of
course we didn't have to change the call of kvm_get_msr() to read_l1_tsc().
[avi: moved callback to kvm_x86_ops tsc block]
Signed-off-by: Nadav Har'El <nyh@il.ibm.com>
Acked-by: Zachary Amsdem <zamsden@gmail.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Architecturally, PDPTEs are cached in the PDPTRs when CR3 is reloaded.
On SVM, it is not possible to implement this, but on VMX this is possible
and was indeed implemented until nested SVM changed this to unconditionally
read PDPTEs dynamically. This has noticable impact when running PAE guests.
Fix by changing the MMU to read PDPTRs from the cache, falling back to
reading from memory for the nested MMU.
Signed-off-by: Avi Kivity <avi@redhat.com>
Tested-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
The vmexit tracepoints format the exit_reason to make it human-readable.
Since the exit_reason depends on the instruction set (vmx or svm),
formatting is handled with ftrace_print_symbols_seq() by referring to
the appropriate exit reason table.
However, the ftrace_print_symbols_seq() function is not meant to be used
directly in tracepoints since it does not export the formatting table
which userspace tools like trace-cmd and perf use to format traces.
In practice perf dies when formatting vmexit-related events and
trace-cmd falls back to printing the numeric value (with extra
formatting code in the kvm plugin to paper over this limitation). Other
userspace consumers of vmexit-related tracepoints would be in similar
trouble.
To avoid significant changes to the kvm_exit tracepoint, this patch
moves the vmx and svm exit reason tables into arch/x86/kvm/trace.h and
selects the right table with __print_symbolic() depending on the
instruction set. Note that __print_symbolic() is designed for exporting
the formatting table to userspace and allows trace-cmd and perf to work.
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
The patch raises the hard limit of VCPU count to 254.
This will allow developers to easily work on scalability
and will allow users to test high VCPU setups easily without
patching the kernel.
To prevent possible issues with current setups, KVM_CAP_NR_VCPUS
now returns the recommended VCPU limit (which is still 64) - this
should be a safe value for everybody, while a new KVM_CAP_MAX_VCPUS
returns the hard limit which is now 254.
Cc: Avi Kivity <avi@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Pekka Enberg <penberg@kernel.org>
Suggested-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Use rcu to protect shadow pages table to be freed, so we can safely walk it,
it should run fastly and is needed by mmio page fault
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
The idea is from Avi:
| Maybe it's time to kill off bypass_guest_pf=1. It's not as effective as
| it used to be, since unsync pages always use shadow_trap_nonpresent_pte,
| and since we convert between the two nonpresent_ptes during sync and unsync.
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
If the page fault is caused by mmio, we can cache the mmio info, later, we do
not need to walk guest page table and quickly know it is a mmio fault while we
emulate the mmio instruction
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
To implement steal time, we need the hypervisor to pass the guest
information about how much time was spent running other processes
outside the VM, while the vcpu had meaningful work to do - halt
time does not count.
This information is acquired through the run_delay field of
delayacct/schedstats infrastructure, that counts time spent in a
runqueue but not running.
Steal time is a per-cpu information, so the traditional MSR-based
infrastructure is used. A new msr, KVM_MSR_STEAL_TIME, holds the
memory area address containing information about steal time
This patch contains the hypervisor part of the steal time infrasructure,
and can be backported independently of the guest portion.
[avi, yongjie: export delayacct_on, to avoid build failures in some configs]
Signed-off-by: Glauber Costa <glommer@redhat.com>
Tested-by: Eric B Munson <emunson@mgebm.net>
CC: Rik van Riel <riel@redhat.com>
CC: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
CC: Peter Zijlstra <peterz@infradead.org>
CC: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Yongjie Ren <yongjie.ren@intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
When CR0.WP=0, we sometimes map user pages as kernel pages (to allow
the kernel to write to them). Unfortunately this also allows the kernel
to fetch from these pages, even if CR4.SMEP is set.
Adjust for this by also setting NX on the spte in these circumstances.
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch allows the guest to enable the VMXE bit in CR4, which is a
prerequisite to running VMXON.
Whether to allow setting the VMXE bit now depends on the architecture (svm
or vmx), so its checking has moved to kvm_x86_ops->set_cr4(). This function
now returns an int: If kvm_x86_ops->set_cr4() returns 1, __kvm_set_cr4()
will also return 1, and this will cause kvm_set_cr4() will throw a #GP.
Turning on the VMXE bit is allowed only when the nested VMX feature is
enabled, and turning it off is forbidden after a vmxon.
Signed-off-by: Nadav Har'El <nyh@il.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Parent pte rmap and page rmap are very similar, so use the same arithmetic
for them
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Abstract the operation of rmap to spte_list, then we can use it for the
reverse mapping of parent pte in the later patch
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Simply return from kvm_mmu_pte_write path if no shadow page is
write-protected, then we can avoid to walk all shadow pages and hold
mmu-lock
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
We clean up a failed VMREAD by clearing the output register. Do
it in the exception handler instead of unconditionally. This is
worthwhile since there are more than a hundred call sites.
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Since the emulator now checks segment limits and access rights, it
generates a lot more accesses to the vmcs segment fields. Undo some
of the performance hit by cacheing those fields in a read-only cache
(the entire cache is invalidated on any write, or on guest exit).
Signed-off-by: Avi Kivity <avi@redhat.com>
Avoid using ctxt->vcpu; we can do everything with ->get_cr() and ->set_cr().
A side effect is that we no longer activate the fpu on emulated CLTS; but that
should be very rare.
Signed-off-by: Avi Kivity <avi@redhat.com>
The last_guest_tsc is used in vcpu_load to adjust the
tsc_offset since tsc-scaling is merged. So the
last_guest_tsc needs to be updated in vcpu_put instead of
the the last_host_tsc. This is fixed with this patch.
Reported-by: Jan Kiszka <jan.kiszka@web.de>
Tested-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Currently we sync registers back and forth before/after exiting
to userspace for IO, but during IO device model shouldn't need to
read/write the registers, so we can as well skip those sync points. The
only exaception is broken vmware backdor interface. The new code sync
registers content during IO only if registers are read from/written to
by userspace in the middle of the IO operation and this almost never
happens in practise.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
This patch implements two new vm-ioctls to get and set the
virtual_tsc_khz if the machine supports tsc-scaling. Setting
the tsc-frequency is only possible before userspace creates
any vcpu.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
With TSC scaling in SVM the tsc-offset needs to be
calculated differently. This patch propagates this
calculation into the architecture specific modules so that
this complexity can be handled there.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch implements a call-back into the architecture code
to allow the propagation of changes to the virtual tsc_khz
of the vcpu.
On SVM it updates the tsc_ratio variable, on VMX it does
nothing.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch changes the kvm_guest_time_update function to use
TSC frequency the guest actually has for updating its clock.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
The mmu_seq verification can be removed since we get the pfn in the
protection of mmu_lock.
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch adds all necessary intercept checks for
instructions that access the crX registers.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch adds a callback into kvm_x86_ops so that svm and
vmx code can do intercept checks on emulated instructions.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Since sse instructions can issue 16-byte mmios, we need to support them. We
can't increase the kvm_run mmio buffer size to 16 bytes without breaking
compatibility, so instead we break the large mmios into two smaller 8-byte
ones. Since the bus is 64-bit we aren't breaking any atomicity guarantees.
Signed-off-by: Avi Kivity <avi@redhat.com>
We may read the cpl quite often in the same vmexit (instruction privilege
check, memory access checks for instruction and operands), so we gain
a bit if we cache the value.
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch does:
- call vcpu->arch.mmu.update_pte directly
- use gfn_to_pfn_atomic in update_pte path
The suggestion is from Avi.
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
No need to record the gfn to verifier the pte has the same mode as
current vcpu, it's because we only speculatively update the pte only
if the pte and vcpu have the same mode
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Code under this lock requires non-preemptibility. Ensure this also over
-rt by converting it to raw spinlock.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Code under this lock requires non-preemptibility. Ensure this also over
-rt by converting it to raw spinlock.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
For GRU and EPT, we need gup-fast to set referenced bit too (this is why
it's correct to return 0 when shadow_access_mask is zero, it requires
gup-fast to set the referenced bit). qemu-kvm access already sets the
young bit in the pte if it isn't zero-copy, if it's zero copy or a shadow
paging EPT minor fault we relay on gup-fast to signal the page is in
use...
We also need to check the young bits on the secondary pagetables for NPT
and not nested shadow mmu as the data may never get accessed again by the
primary pte.
Without this closer accuracy, we'd have to remove the heuristic that
avoids collapsing hugepages in hugepage virtual regions that have not even
a single subpage in use.
->test_young is full backwards compatible with GRU and other usages that
don't have young bits in pagetables set by the hardware and that should
nuke the secondary mmu mappings when ->clear_flush_young runs just like
EPT does.
Removing the heuristic that checks the young bit in
khugepaged/collapse_huge_page completely isn't so bad either probably but
I thought it was worth it and this makes it reliable.
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
It only allows to audit one guest in the system since:
- 'audit_point' is a glob variable
- mmu_audit_disable() is called in kvm_mmu_destroy(), so audit is disabled
after a guest exited
this patch fix those issues then allow to audit more guests at the same time
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Instead of syncing the guest cr3 every exit, which is expensince on vmx
with ept enabled, sync it only on demand.
[sheng: fix incorrect cr3 seen by Windows XP]
Signed-off-by: Sheng Yang <sheng@linux.intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
In case of a nested page fault or an intercepted #PF newer SVM
implementations provide a copy of the faulting instruction bytes
in the VMCB.
Use these bytes to feed the instruction emulator and avoid the costly
guest instruction fetch in this case.
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
emulate_instruction had many callers, but only one used all
parameters. One parameter was unused, another one is now
hidden by a wrapper function (required for a future addition
anyway), so most callers use now a shorter parameter list.
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
move the complete_insn_gp() helper function out of the VMX part
into the generic x86 part to make it usable by SVM.
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
The handling of CR8 writes in KVM is currently somewhat cumbersome.
This patch makes it look like the other CR register handlers
and fixes a possible issue in VMX, where the RIP would be incremented
despite an injected #GP.
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Retry #PF for softmmu only when the current vcpu has the same cr3 as the time
when #PF occurs
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
It's the speculative path if 'no_apf = 1' and we will specially handle this
speculative path in the later patch, so 'prefault' is better to fit the sense.
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Since vmx blocks INIT signals, we disable virtualization extensions during
reboot. This leads to virtualization instructions faulting; we trap these
faults and spin while the reboot continues.
Unfortunately spinning on a non-preemptible kernel may block a task that
reboot depends on; this causes the reboot to hang.
Fix by skipping over the instruction and hoping for the best.
Signed-off-by: Avi Kivity <avi@redhat.com>
