Using RCU for lockless shadow walking can increase the amount of memory
in use by the system, since RCU grace periods are unpredictable. We also
have an unconditional write to a shared variable (reader_counter), which
isn't good for scaling.
Replace that with a scheme similar to x86's get_user_pages_fast(): disable
interrupts during lockless shadow walk to force the freer
(kvm_mmu_commit_zap_page()) to wait for the TLB flush IPI to find the
processor with interrupts enabled.
We also add a new vcpu->mode, READING_SHADOW_PAGE_TABLES, to prevent
kvm_flush_remote_tlbs() from avoiding the IPI.
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
On x86_64, we can defer %ds and %es reload to the heavyweight context switch,
since nothing in the lightweight paths uses the host %ds or %es (they are
ignored by the processor). Furthermore we can avoid the load if the segments
are null, by letting the hardware load the null segments for us. This is the
expected case.
On i386, we could avoid the reload entirely, since the entry.S paths take care
of reload, except for the SYSEXIT path which leaves %ds and %es set to __USER_DS.
So we set them to the same values as well.
Saves about 70 cycles out of 1600 (around 4%; noisy measurements).
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
The vmx exit code unconditionally restores %ds and %es to __USER_DS. This
can override the user's values, since %ds and %es are not saved and restored
in x86_64 syscalls. In practice, this isn't dangerous since nobody uses
segment registers in long mode, least of all programs that use KVM.
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
The instruction emulation for bsrw is broken in KVM because
the code always uses bsr with 32 or 64 bit operand size for
emulation. Fix that by using emulate_2op_SrcV_nobyte() macro
to use guest operand size for emulation.
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Although ModRM byte is fetched for group decoding, it is soon pushed
back to make decode_modrm() fetch it later again.
Now that ModRM flag can be found in the top level opcode tables, fetch
ModRM byte before group decoding to make the code simpler.
Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Signed-off-by: Avi Kivity <avi@redhat.com>
Needed for the following patch which simplifies ModRM fetching code.
Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Signed-off-by: Avi Kivity <avi@redhat.com>
This cpuid range does not exist on real HW and Intel spec says that
"Information returned for highest basic information leaf" will be
returned. Not very well defined.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
cpuid eax should return the max leaf so that
guests can find out the valid range.
This matches Xen et al.
Update documentation to match.
Tested with -cpu host.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
We can't run PIT IRQ injection work in the interrupt context of the host
timer. This would allow the user to influence the handler complexity by
asking for a broadcast to a large number of VCPUs. Therefore, this work
was pushed into workqueue context in 9d244caf2e. However, this prevents
prioritizing the PIT injection over other task as workqueues share
kernel threads.
This replaces the workqueue with a kthread worker and gives that thread
a name in the format "kvm-pit/<owner-process-pid>". That allows to
identify and adjust the kthread priority according to the VM process
parameters.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
The patch introduces a bitmap that will hold reasons apic should be
checked during vmexit. This is in a preparation for vp eoi patch
that will add one more check on vmexit. With the bitmap we can do
if(apic_attention) to check everything simultaneously which will
add zero overhead on the fast path.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Currently, MSI messages can only be injected to in-kernel irqchips by
defining a corresponding IRQ route for each message. This is not only
unhandy if the MSI messages are generated "on the fly" by user space,
IRQ routes are a limited resource that user space has to manage
carefully.
By providing a direct injection path, we can both avoid using up limited
resources and simplify the necessary steps for user land.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
MMIO that are split across a page boundary are currently broken - the
code does not expect to be aborted by the exit to userspace for the
first MMIO fragment.
This patch fixes the problem by generalizing the current code for handling
16-byte MMIOs to handle a number of "fragments", and changes the MMIO
code to create those fragments.
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Merge reason: development work has dependency on kvm patches merged
upstream.
Conflicts:
Documentation/feature-removal-schedule.txt
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Pull KVM updates from Marcelo Tosatti.
* git://git.kernel.org/pub/scm/virt/kvm/kvm:
KVM: lock slots_lock around device assignment
KVM: VMX: Fix kvm_set_shared_msr() called in preemptible context
KVM: unmap pages from the iommu when slots are removed
KVM: PMU emulation: GLOBAL_CTRL MSR should be enabled on reset
kvm_set_shared_msr() may not be called in preemptible context,
but vmx_set_msr() does so:
BUG: using smp_processor_id() in preemptible [00000000] code: qemu-kvm/22713
caller is kvm_set_shared_msr+0x32/0xa0 [kvm]
Pid: 22713, comm: qemu-kvm Not tainted 3.4.0-rc3+ #39
Call Trace:
[<ffffffff8131fa82>] debug_smp_processor_id+0xe2/0x100
[<ffffffffa0328ae2>] kvm_set_shared_msr+0x32/0xa0 [kvm]
[<ffffffffa03a103b>] vmx_set_msr+0x28b/0x2d0 [kvm_intel]
...
Making kvm_set_shared_msr() work in preemptible is cleaner, but
it's used in the fast path. Making two variants is overkill, so
this patch just disables preemption around the call.
Reported-by: Dave Jones <davej@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Its much cleaner to use PT_PAGE_TABLE_LEVEL than its numeric value.
Signed-off-by: Davidlohr Bueso <dave@gnu.org>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Intel spec says that TMR needs to be set/cleared
when IRR is set, but kvm also clears it on EOI.
I did some tests on a real (AMD based) system,
and I see same TMR values both before
and after EOI, so I think it's a minor bug in kvm.
This patch fixes TMR to be set/cleared on IRR set
only as per spec.
And now that we don't clear TMR, we can save
an atomic read of TMR on EOI that's not propagated
to ioapic, by checking whether ioapic needs
a specific vector first and calculating
the mode afterwards.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
General support for the MMX instruction set. Special care is taken
to trap pending x87 exceptions so that they are properly reflected
to the guest.
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
An Ubuntu 9.10 Karmic Koala guest is unable to boot or install due to
missing movdqa emulation:
kvm_exit: reason EXCEPTION_NMI rip 0x7fef3e025a7b info 7fef3e799000 80000b0e
kvm_page_fault: address 7fef3e799000 error_code f
kvm_emulate_insn: 0:7fef3e025a7b: 66 0f 7f 07 (prot64)
movdqa %xmm0,(%rdi)
[avi: mark it explicitly aligned]
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
x86 defines three classes of vector instructions: explicitly
aligned (#GP(0) if unaligned, explicitly unaligned, and default
(which depends on the encoding: AVX is unaligned, SSE is aligned).
Add support for marking an instruction as explicitly aligned or
unaligned, and mark MOVDQU as unaligned.
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Enable x86 feature-based autoloading for the kvm-amd module on CPUs
with X86_FEATURE_SVM.
Signed-off-by: Josh Triplett <josh@joshtriplett.org>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
This can happen if the instruction is much longer than the maximum length,
or if insn->opnd_bytes is manually changed.
This patch also fixes warnings from -Wswitch-default flag.
Reported-by: Prashanth Nageshappa <prashanth@linux.vnet.ibm.com>
Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Jim Keniston <jkenisto@linux.vnet.ibm.com>
Cc: Linux-mm <linux-mm@kvack.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
Cc: Anton Arapov <anton@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: yrl.pp-manager.tt@hitachi.com
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20120413032427.32577.42602.stgit@localhost.localdomain
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The 'max' range needs to be unsigned, since the size of the user address
space is bigger than 2GB.
We know that 'count' is positive in 'long' (that is checked in the
caller), so we will truncate 'max' down to something that fits in a
signed long, but before we actually do that, that comparison needs to be
done in unsigned.
Bug introduced in commit 92ae03f2ef ("x86: merge 32/64-bit versions of
'strncpy_from_user()' and speed it up"). On x86-64 you can't trigger
this, since the user address space is much smaller than 63 bits, and on
x86-32 it works in practice, since you would seldom hit the strncpy
limits anyway.
I had actually tested the corner-cases, I had only tested them on
x86-64. Besides, I had only worried about the case of a pointer *close*
to the end of the address space, rather than really far away from it ;)
This also changes the "we hit the user-specified maximum" to return
'res', for the trivial reason that gcc seems to generate better code
that way. 'res' and 'count' are the same in that case, so it really
doesn't matter which one we return.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull x86 fixes from Thomas Gleixner.
* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86: Use correct byte-sized register constraint in __add()
x86: Use correct byte-sized register constraint in __xchg_op()
x86: vsyscall: Use NULL instead 0 for a pointer argument
This merges the 32- and 64-bit versions of the x86 strncpy_from_user()
by just rewriting it in C rather than the ancient inline asm versions
that used lodsb/stosb and had been duplicated for (trivial) differences
between the 32-bit and 64-bit versions.
While doing that, it also speeds them up by doing the accesses a word at
a time. Finally, the new routines also properly handle the case of
hitting the end of the address space, which we have never done correctly
before (fs/namei.c has a hack around it for that reason).
Despite all these improvements, it actually removes more lines than it
adds, due to the de-duplication. Also, we no longer export (or define)
the legacy __strncpy_from_user() function (that was defined to not do
the user permission checks), since it's not actually used anywhere, and
the user address space checks are built in to the new code.
Other architecture maintainers have been notified that the old hack in
fs/namei.c will be going away in the 3.5 merge window, in case they
copied the x86 approach of being a bit cavalier about the end of the
address space.
Cc: linux-arch@vger.kernel.org
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Peter Anvin" <hpa@zytor.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
On reset all MPU counters should be enabled in GLOBAL_CTRL MSR.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
while we can't just use -U$(SUBARCH), we still need to kill idiotic define
(implicit -Di386=1), both for SUBARCH=i386 and SUBARCH=x86/CONFIG_64BIT=n
builds.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Iteration using rmap_next(), the actual body is pte_list_next(), is
inefficient: every time we call it we start from checking whether rmap
holds a single spte or points to a descriptor which links more sptes.
In the case of shadow paging, this quadratic total iteration cost is a
problem. Even for two dimensional paging, with EPT/NPT on, in which we
almost always have a single mapping, the extra checks at the end of the
iteration should be eliminated.
This patch fixes this by introducing rmap_iterator which keeps the
iteration context for the next search. Furthermore the implementation
of rmap_next() is splitted into two functions, rmap_get_first() and
rmap_get_next(), to avoid repeatedly checking whether the rmap being
iterated on has only one spte.
Although there seemed to be only a slight change for EPT/NPT, the actual
improvement was significant: we observed that GET_DIRTY_LOG for 1GB
dirty memory became 15% faster than before. This is probably because
the new code is easy to make branch predictions.
Note: we just remove pte_list_next() because we can think of parent_ptes
as a reverse mapping.
Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Signed-off-by: Avi Kivity <avi@redhat.com>
We have PTE_LIST_EXT + 1 pointers in this structure and these 40/20
bytes do not fit cache lines well. Furthermore, some allocators may
use 64/32-byte objects for the pte_list_desc cache.
This patch solves this problem by changing PTE_LIST_EXT from 4 to 3.
For shadow paging, the new size is still large enough to hold both the
kernel and process mappings for usual anonymous pages. For file
mappings, there may be a slight change in the cache usage.
Note: with EPT/NPT we almost always have a single spte in each reverse
mapping and we will not see any change by this.
Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Signed-off-by: Avi Kivity <avi@redhat.com>
Since most guests will have paging enabled for memory management, add likely() optimization
around CR0.PG checks.
Signed-off-by: Davidlohr Bueso <dave@gnu.org>
Signed-off-by: Avi Kivity <avi@redhat.com>
Enable x86 feature-based autoloading for the kvm-intel module on CPUs
with X86_FEATURE_VMX.
Signed-off-by: Josh Triplett <josh@joshtriplett.org>
Acked-By: Kay Sievers <kay@vrfy.org>
Signed-off-by: Avi Kivity <avi@redhat.com>
We have seen some problems of the current implementation of
get_dirty_log() which uses synchronize_srcu_expedited() for updating
dirty bitmaps; e.g. it is noticeable that this sometimes gives us ms
order of latency when we use VGA displays.
Furthermore the recent discussion on the following thread
"srcu: Implement call_srcu()"
http://lkml.org/lkml/2012/1/31/211
also motivated us to implement get_dirty_log() without SRCU.
This patch achieves this goal without sacrificing the performance of
both VGA and live migration: in practice the new code is much faster
than the old one unless we have too many dirty pages.
Implementation:
The key part of the implementation is the use of xchg() operation for
clearing dirty bits atomically. Since this allows us to update only
BITS_PER_LONG pages at once, we need to iterate over the dirty bitmap
until every dirty bit is cleared again for the next call.
Although some people may worry about the problem of using the atomic
memory instruction many times to the concurrently accessible bitmap,
it is usually accessed with mmu_lock held and we rarely see concurrent
accesses: so what we need to care about is the pure xchg() overheads.
Another point to note is that we do not use for_each_set_bit() to check
which ones in each BITS_PER_LONG pages are actually dirty. Instead we
simply use __ffs() in a loop. This is much faster than repeatedly call
find_next_bit().
Performance:
The dirty-log-perf unit test showed nice improvements, some times faster
than before, except for some extreme cases; for such cases the speed of
getting dirty page information is much faster than we process it in the
userspace.
For real workloads, both VGA and live migration, we have observed pure
improvements: when the guest was reading a file during live migration,
we originally saw a few ms of latency, but with the new method the
latency was less than 200us.
Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Signed-off-by: Avi Kivity <avi@redhat.com>
Dropped such mappings when we enabled dirty logging and we will never
create new ones until we stop the logging.
For this we introduce a new function which can be used to write protect
a range of PT level pages: although we do not need to care about a range
of pages at this point, the following patch will need this feature to
optimize the write protection of many pages.
Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Signed-off-by: Avi Kivity <avi@redhat.com>
We will use this in the following patch to implement another function
which needs to write protect pages using the rmap information.
Note that there is a small change in debug printing for large pages:
we do not differentiate them from others to avoid duplicating code.
Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Signed-off-by: Avi Kivity <avi@redhat.com>
check_and_clear_guest_paused does not need to be exported as it isn't used
by any modules, remove the export.
Signed-off-by: Eric B Munson <emunson@mgebm.net>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Now that we have a flag that will tell the guest it was suspended, create an
interface for that communication using a KVM ioctl.
Signed-off-by: Eric B Munson <emunson@mgebm.net>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
When a host stops or suspends a VM it will set a flag to show this. The
watchdog will use these functions to determine if a softlockup is real, or the
result of a suspended VM.
Signed-off-by: Eric B Munson <emunson@mgebm.net>
asm-generic changes Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This flag will be used to check if the vm was stopped by the host when a soft
lockup was detected. The host will set the flag when it stops the guest. On
resume, the guest will check this flag if a soft lockup is detected and skip
issuing the warning.
Signed-off-by: Eric B Munson <emunson@mgebm.net>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
The kvm_vcpu_kick function performs roughly the same funcitonality on
most all architectures, so we shouldn't have separate copies.
PowerPC keeps a pointer to interchanging waitqueues on the vcpu_arch
structure and to accomodate this special need a
__KVM_HAVE_ARCH_VCPU_GET_WQ define and accompanying function
kvm_arch_vcpu_wq have been defined. For all other architectures this
is a generic inline that just returns &vcpu->wq;
Acked-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Christoffer Dall <c.dall@virtualopensystems.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Also count the exits of fast-path.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Acked-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Intel recently release 2 new features, HLE and RTM.
Refer to http://software.intel.com/file/41417.
This patch expose them to guest.
Signed-off-by: Liu, Jinsong <jinsong.liu@intel.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Pull a few KVM fixes from Avi Kivity:
"A bunch of powerpc KVM fixes, a guest and a host RCU fix (unrelated),
and a small build fix."
* 'kvm-updates/3.4' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
KVM: Resolve RCU vs. async page fault problem
KVM: VMX: vmx_set_cr0 expects kvm->srcu locked
KVM: PMU: Fix integer constant is too large warning in kvm_pmu_set_msr()
KVM: PPC: Book3S: PR: Fix preemption
KVM: PPC: Save/Restore CR over vcpu_run
KVM: PPC: Book3S HV: Save and restore CR in __kvmppc_vcore_entry
KVM: PPC: Book3S HV: Fix kvm_alloc_linear in case where no linears exist
KVM: PPC: Book3S: Compile fix for ppc32 in HIOR access code
* one is a workaround that will be removed in v3.5 with proper fix in the tip/x86 tree,
* the other is to fix drivers to load on PV (a previous patch made them only
load in PVonHVM mode).
The rest are just minor fixes in the various drivers and some cleanup in the
core code.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
iQEcBAABAgAGBQJPfyVUAAoJEFjIrFwIi8fJUjUH/jbY5JavRqSlNELZW2A4Ta76
8p00LqLHw/C56iHZcWKke8mqtWNb+ZfcQt7ZYcxDIYa4QWBL28x0OLAO2tOBIt37
ZjYESWSdFJaJvmpADluWtFyGyZ9TYJllDTBm/jWj1ZtKSZvR1YkhuMXCS0f4AmGQ
xFzSWJZUDdiOAqpN+VQD8wP00gfR8knQLg16XE2fvFdQo4XwpCtqLfHV/5pMMGdy
Cs/ep6rq/7cdv/nshKOcBnw7RW8l3Xoi/28ht8k3DvAQ2VtFq1Tugv2G9pcCHwQG
DIBkB3SOU6/v6P5at5+egKS5xR1fJetCWlkMd8kkbcdz2NPI4UDMkvOW6Q8yQls=
=6Ve+
-----END PGP SIGNATURE-----
Merge tag 'stable/for-linus-3.4-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen
Pull xen fixes from Konrad Rzeszutek Wilk:
"Two fixes for regressions:
* one is a workaround that will be removed in v3.5 with proper fix in
the tip/x86 tree,
* the other is to fix drivers to load on PV (a previous patch made
them only load in PVonHVM mode).
The rest are just minor fixes in the various drivers and some cleanup
in the core code."
* tag 'stable/for-linus-3.4-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen:
xen/pcifront: avoid pci_frontend_enable_msix() falsely returning success
xen/pciback: fix XEN_PCI_OP_enable_msix result
xen/smp: Remove unnecessary call to smp_processor_id()
xen/x86: Workaround 'x86/ioapic: Add register level checks to detect bogus io-apic entries'
xen: only check xen_platform_pci_unplug if hvm
