Intel Xeons from Ivy Bridge onwards support a processor identification
number set in the factory. To the user this is a handy unique number to
identify a particular CPU. Intel can decode this to the fab/production
run to track errors. On systems that have it, include it in the machine
check record. I'm told that this would be helpful for users that run
large data centers with multi-socket servers to keep track of which CPUs
are seeing errors.
Boris:
* Add some clarifying comments and spacing.
* Mask out [63:2] in the disabled-but-not-locked case
* Call the MSR variable "val" for more readability.
Signed-off-by: Tony Luck <tony.luck@intel.com>
Cc: Ashok Raj <ashok.raj@intel.com>
Cc: linux-edac <linux-edac@vger.kernel.org>
Cc: x86-ml <x86@kernel.org>
Link: http://lkml.kernel.org/r/20161123114855.njguoaygp3qnbkia@pd.tnic
Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Install the callbacks via the state machine and let the core invoke
the callbacks on the already online CPUs.
Move the callbacks to online/offline as there is no point in having the
files around before the cpu is online and until its completely gone.
[ tglx: Move the callbacks to online/offline ]
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: rt@linuxtronix.de
Link: http://lkml.kernel.org/r/20161117183541.8588-4-bigeasy@linutronix.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
No point to have this file around before the cpu is online and no point to
have it around until the cpu is dead. Get rid of the explicit state.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Sebastian Siewior <bigeasy@linutronix.de>
No point to have the sysfs files around before the cpu is online and no
point to have them around until the cpu is dead. Get rid of the explicit
state.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Sebastian Siewior <bigeasy@linutronix.de>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Borislav Petkov <bp@alien8.de>
The Unified Memory Controllers (UMCs) on Fam17h log a normalized address
in their MCA_ADDR registers. We need to convert that normalized address
to a system physical address in order to support a few facilities:
1) To offline poisoned pages in DRAM proactively in the deferred error
handler.
2) To print sysaddr and page info for DRAM ECC errors in EDAC.
[ Boris: fixes/cleanups ontop:
* hi_addr_offset = 0 - no need for that branch. Stick it all under the
HiAddrOffsetEn case. It confines hi_addr_offset's declaration too.
* Move variables to the innermost scope they're used at so that we save
on stack and not blow it up immediately on function entry.
* Do not modify *sys_addr prematurely - we want to not exit early and
have modified *sys_addr some, which callers get to see. We either
convert to a sys_addr or we don't do anything. And we signal that with
the retval of the function.
* Rename label out -> out_err - because it is the error path.
* No need to pr_err of the conversion failed case: imagine a
sparsely-populated machine with UMCs which don't have DIMMs. Callers
should look at the retval instead and issue a printk only when really
necessary. No need for useless info in dmesg.
* s/temp_reg/tmp/ and other variable names shortening => shorter code.
* Use BIT() everywhere.
* Make error messages more informative.
* Small build fix for the !CONFIG_X86_MCE_AMD case.
* ... and more minor cleanups.
]
Signed-off-by: Yazen Ghannam <Yazen.Ghannam@amd.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Aravind Gopalakrishnan <aravindksg.lkml@gmail.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/20161122111133.mjzpvzhf7o7yl2oa@pd.tnic
[ Typo fixes. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
So adding thresholding_en et al was a good thing for removing the
per-CPU thresholding callback, i.e., threshold_cpu_callback.
But, in order for it to work and especially that test in
mce_threshold_create_device() so that all thresholding banks get
properly created and not the whole thing to fail with a NULL ptr
dereference at mce_cpu_pre_down() when we offline the CPUs, we need to
set the thresholding_en flag *before* we start creating the devices.
Yap, it failed because thresholding_en wasn't set at the time
we were creating the banks so we didn't create any and then at
mce_cpu_pre_down() -> mce_threshold_remove_device() time, we would blow
up.
And the fix is actually easy: we have thresholding on the system when we
have managed to set the thresholding vector to amd_threshold_interrupt()
earlier in mce_amd_feature_init() while we were picking apart the
thresholding banks and what is set and what not.
So let's do that.
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Yazen Ghannam <Yazen.Ghannam@amd.com>
Fixes: 4d7b02d58c ("x86/mcheck: Split threshold_cpu_callback into two callbacks")
Link: http://lkml.kernel.org/r/20161119103402.5227-1-bp@alien8.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The latest binutils are warning about a .fill directive with an explicit
value in a .bss section:
arch/x86/kernel/head_32.S: Assembler messages:
arch/x86/kernel/head_32.S:677: Warning: ignoring fill value in section `.bss..page_aligned'
arch/x86/kernel/head_32.S:679: Warning: ignoring fill value in section `.bss..page_aligned'
This comes from the 'ENTRY()' macro padding the space between the symbols
with 'nop' via:
.align 4,0x90
Open-coding the .globl directive without the padding avoids that warning,
as all the symbols are already page aligned.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
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: Josh Poimboeuf <jpoimboe@redhat.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/20161116141726.2013389-1-arnd@arndb.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Some devices on Fam17h can only be accessed through the System Management
Network (SMN). The SMN is accessed by a pair of index/data registers in PCI
config space. Add a pair of functions to read from and write to the SMN.
The Data Fabric on Fam17h allows multiple devices to use the same register
space. The registers of a specific device are accessed indirectly using the
device's DF InstanceId. Currently, we only need to read from these devices,
so only define a read function for now.
Signed-off-by: Yazen Ghannam <Yazen.Ghannam@amd.com>
Cc: linux-edac <linux-edac@vger.kernel.org>
Cc: x86-ml <x86@kernel.org>
Link: http://lkml.kernel.org/r/1478812257-5424-5-git-send-email-Yazen.Ghannam@amd.com
[ Boris: make __amd_smn_rw() even more compact. ]
Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Sparse populated CPUID leafs are collected in a software provided leaf to
avoid bloat of the x86_capability array, but there is no way to rebuild the
real leafs (e.g. for KVM CPUID enumeration) other than rereading the CPUID
leaf from the CPU. While this is possible it is problematic as it does not
take software disabled features into account. If a feature is disabled on
the host it should not be exposed to a guest either.
Add get_scattered_cpuid_leaf() which rebuilds the leaf from the scattered
cpuid table information and the active CPU features.
[ tglx: Rewrote changelog ]
Signed-off-by: He Chen <he.chen@linux.intel.com>
Reviewed-by: Borislav Petkov <bp@suse.de>
Cc: Luwei Kang <luwei.kang@intel.com>
Cc: kvm@vger.kernel.org
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Piotr Luc <Piotr.Luc@intel.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Link: http://lkml.kernel.org/r/1478856336-9388-3-git-send-email-he.chen@linux.intel.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
The error count field in MCA_MISC does not get reset by hardware when the
threshold has been reached. Software is expected to reset it. Currently,
the threshold limit only gets reset during init or when a user writes to
sysfs.
If the user is not monitoring threshold interrupts and resetting
the limit then the user will only see 1 interrupt when the limit is first
hit. So if, for example, the limit is set to 10 then only 1 interrupt will
be recorded after 10 errors even if 100 errors have occurred. The user may
then assume that only 10 errors have occurred.
Signed-off-by: Yazen Ghannam <Yazen.Ghannam@amd.com>
Cc: Aravind Gopalakrishnan <aravindksg.lkml@gmail.com>
Cc: linux-edac <linux-edac@vger.kernel.org>
Cc: x86-ml <x86@kernel.org>
Link: http://lkml.kernel.org/r/1479244433-69267-1-git-send-email-Yazen.Ghannam@amd.com
Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
The screen_info object was extended to support 64-bit lfb_base addresses
in:
ae2ee627dc ("efifb: Add support for 64-bit frame buffer addresses")
However, the x86 simple-framebuffer setup code never made use of it. Fix
it to properly assemble and verify the lfb_base before advertising
simple-framebuffer devices.
In particular, this means if VIDEO_CAPABILITY_64BIT_BASE is set, the
screen_info->ext_lfb_base field will contain the upper 32bit of the
actual lfb_base. Make sure the address is not 0 (i.e., unset), as well as
does not overflow the physical address type.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Matt Fleming <matt.fleming@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Tom Gundersen <teg@jklm.no>
Link: http://lkml.kernel.org/r/20161115120158.15388-2-dh.herrmann@gmail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Initially I wanted to remove mcheck_cpu_init() from identify_cpu() and let it
become an independent early hotplug callback. The main problem here was that
the init on the boot CPU may happen too late
(device_initcall_sync(mcheck_init_device)) and nobody wanted to risk receiving
and MCE event at boot time leading to a shutdown (if the MCE feature is not yet
enabled).
Here is attempt two: the timming stays as-is but the ordering of the functions
is changed:
- mcheck_cpu_init() (which is run from identify_cpu()) will setup the timer
struct but won't fire the timer. This is moved to CPU_ONLINE since its
cleanup part is in CPU_DOWN_PREPARE. So if it is okay to stop the timer early
in the shutdown phase, it should be okay to start it late in the bring up phase.
- CPU_DOWN_PREPARE disables the MCE feature flags for !INTEL CPUs in
mce_disable_cpu(). If a failure occures it would be re-enabled on all vendor
CPUs (including Intel where it was not disabled during shutdown). To keep this
working I am moving it to CPU_ONLINE. smp_call_function_single() is dropped
beause the notifier runs nowdays on the target CPU.
- CPU_ONLINE is invoking mce_device_create() + mce_threshold_create_device()
but its cleanup part is in CPU_DEAD (mce_threshold_remove_device() and
mce_device_remove()). In order to keep this symmetrical I am moving the clean
up from CPU_DEAD to CPU_DOWN_PREPARE.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Acked-by: Borislav Petkov <bp@alien8.de>
Cc: Tony Luck <tony.luck@intel.com>
Cc: rt@linutronix.de
Cc: linux-edac@vger.kernel.org
Link: http://lkml.kernel.org/r/20161110174447.11848-6-bigeasy@linutronix.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
The threshold_cpu_callback callbacks looks like one of the notifier and
its arguments are almost the same. Split this out and have one ONLINE
and one DEAD callback. This will come handy later once the main code
gets changed to use the callback mechanism.
Also, handle threshold_cpu_callback_online() return value so we don't
continue if the function fails.
Boris Petkov removed the callback pointer and replaced it with proper
functions.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Acked-by: Borislav Petkov <bp@alien8.de>
Cc: Tony Luck <tony.luck@intel.com>
Cc: rt@linutronix.de
Cc: linux-edac@vger.kernel.org
Link: http://lkml.kernel.org/r/20161110174447.11848-5-bigeasy@linutronix.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
This was originally a part of commit 186f43608a:
("x86/kernel: Audit and remove any unnecessary uses of module.h")
...but without the asm/desc.h addition. As such, Ingo reported a
build failure on i386 allnoconfig with SMP=y during his pre-merge
testing. For expediency the chunk was just dropped at that time.
The failure was as follows:
arch/x86/kernel/setup_percpu.c: In function ‘setup_percpu_segment’:
arch/x86/kernel/setup_percpu.c:159:2: error: implicit declaration of function ‘pack_descriptor’ [-Werror=implicit-function-declaration]
arch/x86/kernel/setup_percpu.c:162:2: error: implicit declaration of function ‘write_gdt_entry’ [-Werror=implicit-function-declaration]
arch/x86/kernel/setup_percpu.c:162:18: error: implicit declaration of function ‘get_cpu_gdt_table’ [-Werror=implicit-function-declaration]
As pack_descriptor(), write_gdt_entry() and get_cpu_gdt_table() all
live in the file arch/x86/include/asm/desc.h -- calling that header
out explicitly should fix things.
Reported-by: Ingo Molnar <mingo@redhat.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.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/20161114190443.10873-1-paul.gortmaker@windriver.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull x86 fixes from Ingo Molnar:
"Misc fixes:
- fix an Intel/MID boot crash/hang bug
- fix a cache topology mis-parsing bug on certain AMD CPUs
- fix a virtualization firmware bug by adding a check+quirk
workaround on the kernel side"
* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/cpu: Deal with broken firmware (VMWare/XEN)
x86/cpu/AMD: Fix cpu_llc_id for AMD Fam17h systems
x86/platform/intel-mid: Retrofit pci_platform_pm_ops ->get_state hook
apm_bios_call() can fail, and return a status in its argument structure.
If that status however is zero during a call from
apm_get_power_status(), we end up using data that may have never been
set, as reported by "gcc -Wmaybe-uninitialized":
arch/x86/kernel/apm_32.c: In function ‘apm’:
arch/x86/kernel/apm_32.c:1729:17: error: ‘bx’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
arch/x86/kernel/apm_32.c:1835:5: error: ‘cx’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
arch/x86/kernel/apm_32.c:1730:17: note: ‘cx’ was declared here
arch/x86/kernel/apm_32.c:1842:27: error: ‘dx’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
arch/x86/kernel/apm_32.c:1731:17: note: ‘dx’ was declared here
This changes the function to return "APM_NO_ERROR" here, which makes the
code more robust to broken BIOS versions, and avoids the warning.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Jiri Kosina <jkosina@suse.cz>
Reviewed-by: Luis R. Rodriguez <mcgrof@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We did have logic in the MCE code which would TSC-timestamp an error
record only when it is exact - i.e., when it wasn't detected by polling.
This isn't the case anymore. So let's fix that:
We have a valid TSC timestamp in the error record only when it has been
a precise detection, i.e., either in the #MC handler or in one of the
interrupt handlers (thresholding, deferred, ...).
All other error records still have mce.time which contains the wall
time in order to be able to place the error record in time at least
approximately.
Also, this fixes another bug where machine_check_poll() would clear
mce.tsc unconditionally even if we requested precise MCP_TIMESTAMP
logging.
The proper fix would be to generate timestamp only when it has been
requested and not always. But that would require a more thorough code
audit of all mce_gather_info/mce_setup() users. Add a FIXME for now.
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Tony <tony.luck@intel.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: kernel test robot <xiaolong.ye@intel.com>
Cc: linux-edac <linux-edac@vger.kernel.org>
Cc: lkp@01.org
Link: http://lkml.kernel.org/r/20161110131053.kybsijfs5venpjnf@pd.tnic
Signed-off-by: Ingo Molnar <mingo@kernel.org>
With CONFIG_OF enabled on x86, we get the following error on boot:
"
Failed to find cpu0 device node
Unable to detect cache hierarchy from DT for CPU 0
"
and the cacheinfo fails to get populated in the corresponding sysfs
entries. This is because cache_setup_of_node looks for of_node for
setting up the shared cpu_map without checking that it's already
populated in the architecture specific callback.
In order to indicate that the shared cpu_map is already populated, this
patch introduces a boolean `cpu_map_populated` in struct cpu_cacheinfo
that can be used by the generic code to skip cache_shared_cpu_map_setup.
This patch also sets that boolean for x86.
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The following RCU lockdep warning led to adding irq_enter()/irq_exit() into
smp_reschedule_interrupt():
RCU used illegally from idle CPU!
rcu_scheduler_active = 1, debug_locks = 0
RCU used illegally from extended quiescent state!
no locks held by swapper/1/0.
do_trace_write_msr
native_write_msr
native_apic_msr_eoi_write
smp_reschedule_interrupt
reschedule_interrupt
As Peterz pointed out:
| So now we're making a very frequent interrupt slower because of debug
| code.
|
| The thing is, many many smp_reschedule_interrupt() invocations don't
| actually execute anything much at all and are only sent to tickle the
| return to user path (which does the actual preemption).
|
| Having to do the whole irq_enter/irq_exit dance just for this unlikely
| debug case totally blows.
Use the wrmsr_notrace() variant in native_apic_msr_write_eoi, annotate the
kvm variant with notrace and add a native_apic_eoi callback to the apic
structure so KVM guests are covered as well.
This allows to revert the irq_enter/irq_exit dance in
smp_reschedule_interrupt().
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Cc: kvm@vger.kernel.org
Cc: Mike Galbraith <efault@gmx.de>
Cc: Borislav Petkov <bp@alien8.de>
Link: http://lkml.kernel.org/r/1478488420-5982-3-git-send-email-wanpeng.li@hotmail.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Both ACPI and MP specifications require that the APIC id in the respective
tables must be the same as the APIC id in CPUID.
The kernel retrieves the physical package id from the APIC id during the
ACPI/MP table scan and builds the physical to logical package map. The
physical package id which is used after a CPU comes up is retrieved from
CPUID. So we rely on ACPI/MP tables and CPUID agreeing in that respect.
There exist VMware and XEN implementations which violate the spec. As a
result the physical to logical package map, which relies on the ACPI/MP
tables does not work on those systems, because the CPUID initialized
physical package id does not match the firmware id. This causes system
crashes and malfunction due to invalid package mappings.
The only way to cure this is to sanitize the physical package id after the
CPUID enumeration and yell when the APIC ids are different. Fix up the
initial APIC id, which is fine as it is only used printout purposes.
If the physical package IDs differ yell and use the package information
from the ACPI/MP tables so the existing logical package map just works.
Chas provided the resulting dmesg output for his affected 4 virtual
sockets, 1 core per socket VM:
[Firmware Bug]: CPU1: APIC id mismatch. Firmware: 1 CPUID: 2
[Firmware Bug]: CPU1: Using firmware package id 1 instead of 2
....
Reported-and-tested-by: "Charles (Chas) Williams" <ciwillia@brocade.com>,
Reported-by: M. Vefa Bicakci <m.v.b@runbox.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Alok Kataria <akataria@vmware.com>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: #4.6+ <stable@vger,kernel.org>
Link: http://lkml.kernel.org/r/alpine.DEB.2.20.1611091613540.3501@nanos
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
These changes do not affect current hw - just a cleanup:
Currently, we assume that a system has a single Last Level Cache (LLC)
per node, and that the cpu_llc_id is thus equal to the node_id. This no
longer applies since Fam17h can have multiple last level caches within a
node.
So group the cpu_llc_id assignment by topology feature and family in
order to make the computation of cpu_llc_id on the different families
more clear.
Here is how the LLC ID is being computed on the different families:
The NODEID_MSR feature only applies to Fam10h in which case the LLC is
at the node level.
The TOPOEXT feature is used on families 15h, 16h and 17h. So far we only
see multiple last level caches if L3 caches are available. Otherwise,
the cpu_llc_id will default to be the phys_proc_id.
We have L3 caches only on families 15h and 17h:
- on Fam15h, the LLC is at the node level.
- on Fam17h, the LLC is at the core complex level and can be found by
right shifting the APIC ID. Also, keep the family checks explicit so that
new families will fall back to the default, which will be node_id for
TOPOEXT systems.
Single node systems in families 10h and 15h will have a Node ID of 0
which will be the same as the phys_proc_id, so we don't need to check
for multiple nodes before using the node_id.
Tested-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Yazen Ghannam <Yazen.Ghannam@amd.com>
[ Rewrote the commit message. ]
Signed-off-by: Borislav Petkov <bp@suse.de>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Aravind Gopalakrishnan <aravindksg.lkml@gmail.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20161108153054.bs3sajbyevq6a6uu@pd.tnic
Signed-off-by: Ingo Molnar <mingo@kernel.org>
