Intel's MCA implementation broadcasts MCEs to all CPUs on the
node. This poses a problem for offlined CPUs which cannot
participate in the rendezvous process:
Kernel panic - not syncing: Timeout: Not all CPUs entered broadcast exception handler
Kernel Offset: disabled
Rebooting in 100 seconds..
More specifically, Linux does a soft offline of a CPU when
writing a 0 to /sys/devices/system/cpu/cpuX/online, which
doesn't prevent the #MC exception from being broadcasted to that
CPU.
Ensure that offline CPUs don't participate in the MCE rendezvous
and clear the RIP valid status bit so that a second MCE won't
cause a shutdown.
Without the patch, mce_start() will increment mce_callin and
wait for all CPUs. Offlined CPUs should avoid participating in
the rendezvous process altogether.
Signed-off-by: Ashok Raj <ashok.raj@intel.com>
[ Massage commit message. ]
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Tony Luck <tony.luck@intel.com>
Cc: <stable@vger.kernel.org>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-edac <linux-edac@vger.kernel.org>
Link: http://lkml.kernel.org/r/1449742346-21470-2-git-send-email-bp@alien8.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Pull perf fixes from Ingo Molnar:
"This tree includes four core perf fixes for misc bugs, three fixes to
x86 PMU drivers, and two updates to old email addresses"
* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
perf: Do not send exit event twice
perf/x86/intel: Fix INTEL_FLAGS_UEVENT_CONSTRAINT_DATALA_NA macro
perf/x86/intel: Make L1D_PEND_MISS.FB_FULL not constrained on Haswell
perf: Fix PERF_EVENT_IOC_PERIOD deadlock
treewide: Remove old email address
perf/x86: Fix LBR call stack save/restore
perf: Update email address in MAINTAINERS
perf/core: Robustify the perf_cgroup_from_task() RCU checks
perf/core: Fix RCU problem with cgroup context switching code
'range_new' doesn't seem to be used after init. It is only passed
to memset(), sum_ranges(), memcmp() and x86_get_mtrr_mem_range(), the
latter of which also only passes it on to various *range*
library functions.
So mark it __initdata to free up an extra page after init.
Its contents are wiped at every call to mtrr_calc_range_state(),
so it being static is not about preserving state between calls,
but simply to avoid a 4k+ stack frame. While there, add a
comment explaining this and why it's safe.
We could also mark nr_range_new as __initdata, but since it's
just a single int and also doesn't carry state between calls (it
is unconditionally assigned to before it is read), we might as
well make it an ordinary automatic variable.
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Toshi Kani <toshi.kani@hp.com>
Link: http://lkml.kernel.org/r/1449002691-20783-1-git-send-email-linux@rasmusvillemoes.dk
Signed-off-by: Ingo Molnar <mingo@kernel.org>
x86_init_rdrand() was added with 2 goals:
1. Sanity check that the built-in-self-test circuit on the Digital
Random Number Generator (DRNG) is not complaining. As RDRAND
HW self-checks on every invocation, this goal is achieved
by simply invoking RDRAND and checking its return code.
2. Force a full re-seed of the random number generator.
This was done out of paranoia to benefit the most un-sophisticated
DRNG implementation conceivable in the architecture,
an implementation that does not exist, and unlikely ever will.
This worst-case full-re-seed is achieved by invoking
a 64-bit RDRAND 8192 times.
Unfortunately, this worst-case re-seed costs O(1,000us).
Magnifying this cost, it is done from identify_cpu(), which is the
synchronous critical path to bring a processor on-line -- repeated
for every logical processor in the system at boot and resume from S3.
As it is very expensive, and of highly dubious value, we delete the
worst-case re-seed from the kernel.
We keep the 1st goal -- sanity check the hardware, and mark it absent
if it complains.
This change reduces the cost of x86_init_rdrand() by a factor of 1,000x,
to O(1us) from O(1,000us).
Signed-off-by: Len Brown <len.brown@intel.com>
Link: http://lkml.kernel.org/r/058618cc56ec6611171427ad7205e37e377aa8d4.1439738240.git.len.brown@intel.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
So sparse rightfully complains that the u64 MSR value we're
writing into the STAR MSR, i.e. 0xc0000081, is being truncated:
./arch/x86/include/asm/msr.h:193:36: warning: cast truncates
bits from constant value (23001000000000 becomes 0)
because the actual value doesn't fit into the unsigned 32-bit
quantity which are the @low and @high wrmsrl() parameters.
This is not a problem, practically, because gcc is actually
being smart enough here and does the right thing:
.loc 3 87 0
xorl %esi, %esi # we needz a 32-bit zero
movl $2293776, %edx # 0x00230010 == (__USER32_CS << 16) | __KERNEL_CS go into the high bits
movl $-1073741695, %ecx # MSR_STAR, i.e., 0xc0000081
movl %esi, %eax # low order 32 bits in the MSR which are 0
#APP
# 87 "./arch/x86/include/asm/msr.h" 1
wrmsr
More specifically, MSR_STAR[31:0] is being set to 0. That field
is reserved on Intel and on AMD it is 32-bit SYSCALL Target EIP.
I'd strongly guess because Intel doesn't have SYSCALL in
compat/legacy mode and we're using SYSENTER and INT80 there. And
for compat syscalls in long mode we use CSTAR.
So let's fix the sparse warning by writing SYSRET and SYSCALL CS
and SS into the high 32-bit half of STAR and 0 in the low half
explicitly.
[ Actually, if we had to be precise, we would have to read what's in
STAR[31:0] and write it back unchanged on Intel and write 0 on AMD. I
guess the current writing to 0 is still ok since Intel can apparently
stomach it. ]
The resulting code is identical to what we have above:
.loc 3 87 0
xorl %esi, %esi # tmp104
movl $2293776, %eax #, tmp103
movl $-1073741695, %ecx #, tmp102
movl %esi, %edx # tmp104, tmp104
...
wrmsr
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1448273546-2567-6-git-send-email-bp@alien8.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
We used to have a special ring buffer for deferred errors that
was used to mark problem pages. We replaced that with a generic
pool. Then later converted mce_log() to also use the same pool.
As a result, we end up adding all deferred errors to the pool
twice.
Rearrange this code. Make sure to set the m.severity and
m.usable_addr fields for deferred errors. Then if flags and
mca_cfg.dont_log_ce mean we call mce_log() we are done, because
that will add this entry to the generic pool.
If we skipped mce_log(), then we still want to take action for
the deferred error, so add to the pool.
Change the name of the boolean "error_logged" to "error_seen",
we should set it whether of not we logged an error because the
return value from machine_check_poll() is used to decide whether
storms have subsided or not.
Reported-by: Gong Chen <gong.chen@linux.intel.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-edac <linux-edac@vger.kernel.org>
Link: http://lkml.kernel.org/r/1448350880-5573-2-git-send-email-bp@alien8.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The earlier constraint fix for Broadwell CYCLE_ACTIVITY.*
forced umask 8 to counter 2. For this it used UEVENT,
to match the complete umask.
The event list for Broadwell has an additional
STALLS_L1D_PENDIND event that uses umask 8, but also
sets other bits in the umask. The earlier strict umask match
didn't handle this case.
Add a new UBIT_EVENT constraint macro that only matches
the specified bits in the umask. Then use that macro
to handle CYCLE_ACTIVITY.* on Broadwell.
The documented event also uses cmask, but there's no
need to let the event scheduler know about the cmask,
as the scheduling restriction is only tied to the umask.
Reported-by: Grant Ayers <ayers@cs.stanford.edu>
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Link: http://lkml.kernel.org/r/1447719667-9998-1-git-send-email-andi@firstfloor.org
[ Filled in the missing email address of Grant Ayers - hopefully I got the right one. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
With LBRv5 reading the extra LBR flags like mispredict, TSX, cycles is
not free anymore, as it has moved to a separate MSR.
For callstack mode we don't need any of this information; so we can
avoid the unnecessary MSR read. Add flags to the perf interface where
perf record can request not collecting this information.
Add branch_sample_type flags for CYCLES and FLAGS. It's a bit unusual
for branch_sample_types to be negative (disable), not positive (enable),
but since the legacy ABI reported the flags we need some form of
explicit disabling to avoid breaking the ABI.
After we have the flags the x86 perf code can keep track if any users
need the flags. If noone needs it the information is not collected.
This cuts down the cost of LBR callstack on Skylake significantly.
Profiling a kernel build with LBR call stack the average run time of
the PMI handler drops by 43%.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: acme@kernel.org
Cc: jolsa@kernel.org
Link: http://lkml.kernel.org/r/1445366797-30894-2-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Change the perf user stack walking to use the new
__copy_from_user_nmi(), and split each access into word sized transfer
sizes. This allows to inline the complete access and optimize it all
into a single load.
The main advantage is that this avoids the overhead of double page
faults. When normal copy_from_user() fails it reexecutes the copy to
compute an accurate number of non copied bytes. This leads to
executing the expensive page fault twice.
While walking stacks having a fault at some point is relatively common
(typically when some part of the program isn't compiled with frame
pointers), so this is a large overhead.
With the optimized copies we avoid this problem because they only do
all accesses once. And of course they're much faster too when the
access does not fault because they're just single instructions instead
of complex function calls.
While profiling a kernel build with -g, the patch brings down the
average time of the PMI handler from 966ns to 552ns (-43%).
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Link: http://lkml.kernel.org/r/1445551641-13379-2-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull x86 fixes from Thomas Gleixner:
"This update contains:
- MPX updates for handling 32bit processes
- A fix for a long standing bug in 32bit signal frame handling
related to FPU/XSAVE state
- Handle get_xsave_addr() correctly in KVM
- Fix SMAP check under paravirtualization
- Add a comment to the static function trace entry to avoid further
confusion about the difference to dynamic tracing"
* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/cpu: Fix SMAP check in PVOPS environments
x86/ftrace: Add comment on static function tracing
x86/fpu: Fix get_xsave_addr() behavior under virtualization
x86/fpu: Fix 32-bit signal frame handling
x86/mpx: Fix 32-bit address space calculation
x86/mpx: Do proper get_user() when running 32-bit binaries on 64-bit kernels
There appears to be no formal statement of what pv_irq_ops.save_fl() is
supposed to return precisely. Native returns the full flags, while lguest and
Xen only return the Interrupt Flag, and both have comments by the
implementations stating that only the Interrupt Flag is looked at. This may
have been true when initially implemented, but no longer is.
To make matters worse, the Xen PVOP leaves the upper bits undefined, making
the BUG_ON() undefined behaviour. Experimentally, this now trips for 32bit PV
guests on Broadwell hardware. The BUG_ON() is consistent for an individual
build, but not consistent for all builds. It has also been a sitting timebomb
since SMAP support was introduced.
Use native_save_fl() instead, which will obtain an accurate view of the AC
flag.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: David Vrabel <david.vrabel@citrix.com>
Tested-by: Rusty Russell <rusty@rustcorp.com.au>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: <lguest@lists.ozlabs.org>
Cc: Xen-devel <xen-devel@lists.xen.org>
CC: stable@vger.kernel.org
Link: http://lkml.kernel.org/r/1433323874-6927-1-git-send-email-andrew.cooper3@citrix.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Pull perf updates from Thomas Gleixner:
"Mostly updates to the perf tool plus two fixes to the kernel core code:
- Handle tracepoint filters correctly for inherited events (Peter
Zijlstra)
- Prevent a deadlock in perf_lock_task_context (Paul McKenney)
- Add missing newlines to some pr_err() calls (Arnaldo Carvalho de
Melo)
- Print full source file paths when using 'perf annotate --print-line
--full-paths' (Michael Petlan)
- Fix 'perf probe -d' when just one out of uprobes and kprobes is
enabled (Wang Nan)
- Add compiler.h to list.h to fix 'make perf-tar-src-pkg' generated
tarballs, i.e. out of tree building (Arnaldo Carvalho de Melo)
- Add the llvm-src-base.c and llvm-src-kbuild.c files, generated by
the 'perf test' LLVM entries, when running it in-tree, to
.gitignore (Yunlong Song)
- libbpf error reporting improvements, using a strerror interface to
more precisely tell the user about problems with the provided
scriptlet, be it in C or as a ready made object file (Wang Nan)
- Do not be case sensitive when searching for matching 'perf test'
entries (Arnaldo Carvalho de Melo)
- Inform the user about objdump failures in 'perf annotate' (Andi
Kleen)
- Improve the LLVM 'perf test' entry, introduce a new ones for BPF
and kbuild tests to check the environment used by clang to compile
.c scriptlets (Wang Nan)"
* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (32 commits)
perf/x86/intel/rapl: Remove the unused RAPL_EVENT_DESC() macro
tools include: Add compiler.h to list.h
perf probe: Verify parameters in two functions
perf session: Add missing newlines to some pr_err() calls
perf annotate: Support full source file paths for srcline fix
perf test: Add llvm-src-base.c and llvm-src-kbuild.c to .gitignore
perf: Fix inherited events vs. tracepoint filters
perf: Disable IRQs across RCU RS CS that acquires scheduler lock
perf test: Do not be case sensitive when searching for matching tests
perf test: Add 'perf test BPF'
perf test: Enhance the LLVM tests: add kbuild test
perf test: Enhance the LLVM test: update basic BPF test program
perf bpf: Improve BPF related error messages
perf tools: Make fetch_kernel_version() publicly available
bpf tools: Add new API bpf_object__get_kversion()
bpf tools: Improve libbpf error reporting
perf probe: Cleanup find_perf_probe_point_from_map to reduce redundancy
perf annotate: Inform the user about objdump failures in --stdio
perf stat: Make stat options global
perf sched latency: Fix thread pid reuse issue
...
Pull x86 fixes from Thomas Gleixner:
"A couple of fixes and updates related to x86:
- Fix the W+X check regression on XEN
- The real fix for the low identity map trainwreck
- Probe legacy PIC early instead of unconditionally allocating legacy
irqs
- Add cpu verification to long mode entry
- Adjust the cache topology to AMD Fam17H systems
- Let Merrifield use the TSC across S3"
* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/cpu: Call verify_cpu() after having entered long mode too
x86/setup: Fix low identity map for >= 2GB kernel range
x86/mm: Skip the hypervisor range when walking PGD
x86/AMD: Fix last level cache topology for AMD Fam17h systems
x86/irq: Probe for PIC presence before allocating descs for legacy IRQs
x86/cpu/intel: Enable X86_FEATURE_NONSTOP_TSC_S3 for Merrifield
Pull x86 cpu changes from Ingo Molnar:
"Two changes in this cycle: a Kconfig help text enhancement, and an AMD
CLZERO instruction capability detection and enumeration"
* 'x86-cpu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/cpu: Add CLZERO detection
x86/Kconfig/cpus: Fix/complete CPU type help texts
Pull RAS changes from Ingo Molnar:
"The main system reliability related changes were from x86, but also
some generic RAS changes:
- AMD MCE error injection subsystem enhancements. (Aravind
Gopalakrishnan)
- Fix MCE and CPU hotplug interaction bug. (Ashok Raj)
- kcrash bootup robustness fix. (Baoquan He)
- kcrash cleanups. (Borislav Petkov)
- x86 microcode driver rework: simplify it by unmodularizing it and
other cleanups. (Borislav Petkov)"
* 'ras-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (23 commits)
x86/mce: Add a default case to the switch in __mcheck_cpu_ancient_init()
x86/mce: Add a Scalable MCA vendor flags bit
MAINTAINERS: Unify the microcode driver section
x86/microcode/intel: Move #ifdef DEBUG inside the function
x86/microcode/amd: Remove maintainers from comments
x86/microcode: Remove modularization leftovers
x86/microcode: Merge the early microcode loader
x86/microcode: Unmodularize the microcode driver
x86/mce: Fix thermal throttling reporting after kexec
kexec/crash: Say which char is the unrecognized
x86/setup/crash: Check memblock_reserve() retval
x86/setup/crash: Cleanup some more
x86/setup/crash: Remove alignment variable
x86/setup: Cleanup crashkernel reservation functions
x86/amd_nb, EDAC: Rename amd_get_node_id()
x86/setup: Do not reserve crashkernel high memory if low reservation failed
x86/microcode/amd: Do not overwrite final patch levels
x86/microcode/amd: Extract current patch level read to a function
x86/ras/mce_amd_inj: Inject bank 4 errors on the NBC
x86/ras/mce_amd_inj: Trigger deferred and thresholding errors interrupts
...
The per CPU thermal vector init code checks if the thermal
vector is already installed and complains and bails out if it
is.
This happens after kexec, as kernel shut down does not clear the
thermal vector APIC register.
This causes two problems:
1. So we always do not fully initialize thermal reports after
kexec. The CPU is still likely initialized, as the previous
kernel should have done it. But we don't set up the software
pointer to the thermal vector, so reporting may end up with a
unknown thermal interrupt message.
2. Also it complains for every logical CPU, even though the
value is actually derived from BP only.
The problem is that we end up with one message per CPU, so on
larger systems it becomes very noisy and messes up the otherwise
nicely formatted CPU bootup numbers in the kernel log.
Just remove the check. I checked the code and there's no valid
code paths where the thermal init code for a CPU could be called
multiple times.
Why the kernel does not clean up this value on shutdown:
The thermal monitoring is controlled per logical CPU thread.
Normal shutdown code is just running on one CPU. To disable it
we would need a broadcast NMI to all CPUs on shut down. That's
overkill for this. So we just ignore it after kexec.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Tony Luck <tony.luck@intel.com>
Cc: linux-edac <linux-edac@vger.kernel.org>
Link: http://lkml.kernel.org/r/1445246268-26285-9-git-send-email-bp@alien8.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pave the way for checking the current patch level of the
microcode in a core. We want to be able to do stuff depending on
the patch level - in this case decide whether to update or not.
But that will be added in a later patch.
Drop unused local var uci assignment, while at it.
Integrate a fix for 32-bit and CONFIG_PARAVIRT from Takashi Iwai:
Use native_rdmsr() in check_current_patch_level() because with
CONFIG_PARAVIRT enabled and on 32-bit, where we run before
paging has been enabled, we cannot deref pv_info yet. Or we
could, but we'd need to access its physical address. This way of
fixing it is simpler. See:
https://bugzilla.suse.com/show_bug.cgi?id=943179 for the background.
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: Borislav Petkov <bp@alien8.de>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Takashi Iwai <tiwai@suse.com>:
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Tony Luck <tony.luck@intel.com>
Link: http://lkml.kernel.org/r/1444641762-9437-6-git-send-email-bp@alien8.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
