PMIC interrupts each have an internal latched status bit which is
not visible from any register.  This status bit is set as soon as
the conditions specified in the interrupt type and polarity
registers are met even if the interrupt is not enabled.  When it
is set, nothing else changes within the PMIC and no interrupt
notification packets are sent.  If the internal latched status
bit is set when an interrupt is enabled, then the value is
immediately propagated into the interrupt latched status register
and an interrupt notification packet is sent out from the PMIC
over SPMI.

This PMIC hardware behavior can lead to a situation where the
handler for a level triggered interrupt is called immediately
after enable_irq() is called even though the interrupt physically
triggered while it was disabled within the genirq framework.
This situation takes place if the the interrupt fires twice after
calling disable_irq().  The first time it fires, the level flow
handler will mask and disregard it.  Unfortunately, the second
time it fires, the internal latched status bit is set within the
PMIC and no further notification is received.  When enable_irq()
is called later, the interrupt is unmasked (enabled in the PMIC)
which results in the PMIC immediately sending an interrupt
notification packet out over SPMI.  This breaks the semantics
of level triggered interrupts within the genirq framework since
they should be completely ignored while disabled.

The PMIC internal latched status behavior also affects how
interrupts are treated during suspend.  While entering suspend,
all interrupts not specified as wakeup mode are masked.  Upon
resume, these interrupts are unmasked.  Thus if any of the
non-wakeup PMIC interrupts fired while the system was suspended,
then the PMIC will send interrupt notification packets out via
SPMI as soon as they are unmasked during resume.  This behavior
violates genirq semantics as well since non-wakeup interrupts
should be completely ignored during suspend.

Modify the qpnpint_irq_unmask() function so that the interrupt
latched status clear register is written immediately before the
interrupt enable register.  This clears the internal latched
status bit of the interrupt so that it cannot trigger spuriously
immediately upon being enabled.

Also, while resuming an irq, an unmask could be called even if it
was not previously masked.  So, before writing these registers,
check if the interrupt is already enabled within the PMIC. If it
is, then no further register writes are required.  This
condition check ensures that a valid latched status register bit
is not cleared until it is properly handled.
Signed-off-by: Abhijeet Dharmapurikar <adharmap@codeaurora.org>
Signed-off-by: Kiran Gunda <kgunda@codeaurora.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

irq_enable is called when the device resumes. Note that the
irq_enable is called regardless of whether the interrupt was
marked enabled/disabled in the descriptor or whether it was
masked/unmasked at the controller while resuming.

The current driver unconditionally clears the interrupt in its
irq_enable callback. This is dangerous as any interrupts that
happen right before the resume could be missed.
Remove the irq_enable callback and use mask/unmask instead.

Also remove struct pmic_arb_irq_spec as it serves no real purpose.
It is used only in the translate function and the code is much
cleaner without it.
Signed-off-by: Abhijeet Dharmapurikar <adharmap@codeaurora.org>
Signed-off-by: Kiran Gunda <kgunda@codeaurora.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

We see a unmapped irqs trigger right around bootup. This could
likely be because the bootloader exited leaving the interrupts
in an unknown or unhandled state.  Ack and mask the interrupt
if one is found. A request_irq later will unmask it and also
setup proper mapping structures.

Also the current driver ensures that no read/write transaction
is in progress while it makes changes to the interrupt regions.
This is not necessary because read/writes over spmi and arbiter
interrupt control are independent operations. Hence, remove the
synchronized accesses to interrupt region.
Signed-off-by: Abhijeet Dharmapurikar <adharmap@codeaurora.org>
Signed-off-by: Kiran Gunda <kgunda@codeaurora.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

The current driver uses a mix of radix tree and a fwd lookup
table to translate between apid and ppid. It is buggy and confusing.

Instead simply use a radix tree for v1 hardware and use the
forward lookup table for v2.
Signed-off-by: Abhijeet Dharmapurikar <adharmap@codeaurora.org>
Signed-off-by: Kiran Gunda <kgunda@codeaurora.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

The driver currently uses "apid" and "chan" to mean apid. Remove
the use of chan and use only apid.

On a SPMI bus there is allocation to manage up to 4K peripherals.
However, in practice only few peripherals are instantiated
and only few among the instantiated ones actually interrupt.

APID is CPU's way of keeping track of peripherals that could interrupt.
There is a table that maps the 256 interrupting peripherals to
a number between 0 and 255. This number is called APID. Information about
that interrupting peripheral is stored in registers offset by its
corresponding apid.
Signed-off-by: Abhijeet Dharmapurikar <adharmap@codeaurora.org>
Signed-off-by: Kiran Gunda <kgunda@codeaurora.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

Usually *_dev best used for structures that embed a struct device in
them. spmi_pmic_arb_dev doesn't embed one. It is simply a driver data
structure. Use an appropriate name for it.

Also there are many places in the driver that left shift the bit to
generate a bit mask. Replace it with the BIT() macro.
Signed-off-by: Abhijeet Dharmapurikar <adharmap@codeaurora.org>
Signed-off-by: Kiran Gunda <kgunda@codeaurora.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

The system crashes due to bad access when reading from an non configured
peripheral and when writing to peripheral which is not owned by current
ee. This patch verifies ownership to avoid crashing on
write.
For reads, since the forward mapping table, data_channel->ppid, is
towards the end of the block, we use the core size to figure the
max number of ppids supported. The table starts at an offset of 0x800
within the block, so size - 0x800 will give us the area used by the
table. Since each table is 4 bytes long (core_size - 0x800) / 4 will
gives us the number of data_channel supported.
This new protection is functional on hw v2.
Signed-off-by: Abhijeet Dharmapurikar <adharmap@codeaurora.org>
Signed-off-by: Kiran Gunda <kgunda@codeaurora.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

This reverts commit 7975bd4c, because
VPD relies on driver core to handle deferrals returned by
coreboot_table_find().
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Reviewed-by: Guenter Roeck <groeck@chromium.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

include/linux/i2c is not for client devices. Move the header file to a
more appropriate location.
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

include/linux/i2c is not for client devices. Move the header file to a
more appropriate location.
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

convert list_for_each() to list_for_each_entry() where
applicable.
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

Make this static as it's only referenced in this source and
it does not need global scope.

Cleans up a sparse warning:

drivers/platform/goldfish/goldfish_pipe.c: warning: symbol
  'pipe_dev' was not declared. Should it be static?
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

notifcation -> notification
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

mei_cl_bus_rescan is used only in bus.c,
so make it local to the file and mark static.
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

Structures and functions should be ordered such that forward declaration
use is minimized.

MODULE_* macros should immediately follow the structures and functions
upon which they act.

Remaining MODULE_* macros should be at the end of the file in
alphabetical order.
Signed-off-by: Andrew F. Davis <afd@ti.com>
Acked-by: Evgeniy Polyakov <zbr@ioremap.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

Extend the disabling of preemption to include the hypercall so that
another thread can't get the CPU and corrupt the per-cpu page used
for hypercall arguments.

Cc: <Stable@vger.kernel.org> #4.11
Signed-off-by: Michael Kelley <mikelley@microsoft.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

Instead of open coded variant use generic helper to convert UUID strings
to binary format.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

Commit c0bb0392 ("Drivers: hv: vmbus: Raise retry/wait limits in
vmbus_post_msg()") increased the retry/wait limits of vmbus_post_msg()
to address the new DoS protection policies in WS2016.
Increase the time between retries to make the
code more robust.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

It was found that ICTIMESYNCFLAG_SYNC packets are handled incorrectly
on WS2012R2, e.g. after the guest is paused and resumed its time is set
to something different from host's time. The problem is that we call
adj_guesttime() with reftime=0 for these old hosts and we don't account
for that in 'if (adj_flags & ICTIMESYNCFLAG_SYNC)' branch and
hv_set_host_time().

While we could've solved this by adding a check like
'if (ts_srv_version > TS_VERSION_3)' to hv_set_host_time() I prefer
to do some refactoring. We don't need to have two separate containers
for host samples, struct host_ts which we use for PTP is enough.

Throw away 'struct adj_time_work' and create hv_get_adj_host_time()
accessor to host_ts to avoid code duplication.

Fixes: 3716a49a ("hv_utils: implement Hyper-V PTP source")
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

Turns out that our implementation of .getcrosststamp() never actually
worked. Hyper-V is sending time samples every 5 seconds and this is
too much for get_device_system_crosststamp() as it's interpolation
algorithm (which nobody is currently using in kernel, btw) accounts
for a 'slow' device but we're not slow in Hyper-V, our time reference
is too far away.

.getcrosststamp() is not currently used, get_device_system_crosststamp()
almost always returns -EINVAL and client falls back to using PTP_SYS_OFFSET
so this patch doesn't change much. I also tried doing interpolation
manually (e.g. the same way hv_ptp_gettime() works and it turns out that
we're getting even lower quality:

PTP_SYS_OFFSET_PRECISE with manual interpolation:
* PHC0                     0   3    37     4  -3974ns[-5338ns] +/-  977ns
* PHC0                     0   3    77     7  +2227ns[+3184ns] +/-  576ns
* PHC0                     0   3   177    10  +3060ns[+5220ns] +/-  548ns
* PHC0                     0   3   377    12  +3937ns[+4371ns] +/- 1414ns
* PHC0                     0   3   377     6   +764ns[+1240ns] +/- 1047ns
* PHC0                     0   3   377     7  -1210ns[-3731ns] +/-  479ns
* PHC0                     0   3   377     9   +153ns[-1019ns] +/-  406ns
* PHC0                     0   3   377    12   -872ns[-1793ns] +/-  443ns
* PHC0                     0   3   377     5   +701ns[+3599ns] +/-  426ns
* PHC0                     0   3   377     5   -923ns[ -375ns] +/- 1062ns

PTP_SYS_OFFSET:
* PHC0                     0   3     7     5    +72ns[+8020ns] +/-  251ns
* PHC0                     0   3    17     5   -885ns[-3661ns] +/-  254ns
* PHC0                     0   3    37     6   -454ns[-5732ns] +/-  258ns
* PHC0                     0   3    77    10  +1183ns[+3754ns] +/-  164ns
* PHC0                     0   3   377     5   +579ns[+1137ns] +/-  110ns
* PHC0                     0   3   377     7   +501ns[+1064ns] +/-   96ns
* PHC0                     0   3   377     9  +1641ns[+3342ns] +/-  106ns
* PHC0                     0   3   377     8    -47ns[  +77ns] +/-  160ns
* PHC0                     0   3   377     5    +54ns[ +107ns] +/-  102ns
* PHC0                     0   3   377     8   -354ns[ -617ns] +/-   89ns

This fact wasn't noticed during the initial testing of the PTP device
somehow but got revealed now. Let's just drop .getcrosststamp()
implementation for now as it doesn't seem to be suitable for us.
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

The current code uses the MSR based mechanism to get the current tick.
Use the current clock source as that might be more optimal.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

There is no reason why VPD should register platform device and driver,
given that we do not use their respective kobjects to attach attributes,
nor do we need suspend/resume hooks, or any other features of device
core.
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Reviewed-by: Guenter Roeck <groeck@chromium.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

ro_vpd and rw_vpd are static module-scope variables that are guaranteed
to be initialized with zeroes, there is no need for explicit memset().
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Reviewed-by: Guenter Roeck <groeck@chromium.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

When creating name for the "raw" attribute, let's switch to using
kaspeintf() instead of doing it by hand. Also make sure we handle
errors.
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Reviewed-by: Guenter Roeck <groeck@chromium.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

Instead of open-coding kstrndup with kzalloc + memcpy, let's use
the helper.
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Reviewed-by: Guenter Roeck <groeck@chromium.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

The recent coreboot memory console update (firmware: google: memconsole:
Adapt to new coreboot ring buffer format) introduced a small security
issue in the driver: The new driver implementation parses the memory
console structure again on every access. This is intentional so that
additional lines added concurrently by runtime firmware can be read out.

However, if an attacker can write to the structure, they could increase
the size value to a point where the driver would read potentially
sensitive memory areas from outside the original console buffer during
the next access. This can be done through /dev/mem, since the console
buffer usually resides in firmware-reserved memory that is not covered
by STRICT_DEVMEM.

This patch resolves that problem by reading the buffer's size value only
once during boot (where we can still trust the structure). Other parts
of the structure can still be modified at runtime, but the driver's
bounds checks make sure that it will never read outside the buffer.

Fixes: a5061d02 ("firmware: google: memconsole: Adapt to new coreboot ring buffer format")
Signed-off-by: Julius Werner <jwerner@chromium.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

We want the fixes in here as well to handle merge issues.
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

The code to fetch a 64-bit value from user space was entirely buggered,
and has been since the code was merged in early 2016 in commit
b2f68038 ("x86/mm/32: Add support for 64-bit __get_user() on 32-bit
kernels").

Happily the buggered routine is almost certainly entirely unused, since
the normal way to access user space memory is just with the non-inlined
"get_user()", and the inlined version didn't even historically exist.

The normal "get_user()" case is handled by external hand-written asm in
arch/x86/lib/getuser.S that doesn't have either of these issues.

There were two independent bugs in __get_user_asm_u64():

 - it still did the STAC/CLAC user space access marking, even though
   that is now done by the wrapper macros, see commit 11f1a4b9
   ("x86: reorganize SMAP handling in user space accesses").

   This didn't result in a semantic error, it just means that the
   inlined optimized version was hugely less efficient than the
   allegedly slower standard version, since the CLAC/STAC overhead is
   quite high on modern Intel CPU's.

 - the double register %eax/%edx was marked as an output, but the %eax
   part of it was touched early in the asm, and could thus clobber other
   inputs to the asm that gcc didn't expect it to touch.

   In particular, that meant that the generated code could look like
   this:

        mov    (%eax),%eax
        mov    0x4(%eax),%edx

   where the load of %edx obviously was _supposed_ to be from the 32-bit
   word that followed the source of %eax, but because %eax was
   overwritten by the first instruction, the source of %edx was
   basically random garbage.

The fixes are trivial: remove the extraneous STAC/CLAC entries, and mark
the 64-bit output as early-clobber to let gcc know that no inputs should
alias with the output register.

Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: stable@kernel.org   # v4.8+
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Al noticed that unsafe_put_user() had type problems, and fixed them in
commit a7cc722f ("fix unsafe_put_user()"), which made me look more
at those functions.

It turns out that unsafe_get_user() had a type issue too: it limited the
largest size of the type it could handle to "unsigned long".  Which is
fine with the current users, but doesn't match our existing normal
get_user() semantics, which can also handle "u64" even when that does
not fit in a long.

While at it, also clean up the type cast in unsafe_put_user().  We
actually want to just make it an assignment to the expected type of the
pointer, because we actually do want warnings from types that don't
convert silently.  And it makes the code more readable by not having
that one very long and complex line.

[ This patch might become stable material if we ever end up back-porting
  any new users of the unsafe uaccess code, but as things stand now this
  doesn't matter for any current existing uses. ]

Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Pull misc uaccess fixes from Al Viro:
 "Fix for unsafe_put_user() (no callers currently in mainline, but
  anyone starting to use it will step into that) + alpha osf_wait4()
  infoleak fix"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  osf_wait4(): fix infoleak
  fix unsafe_put_user()

Pull scheduler fix from Thomas Gleixner:
 "A single scheduler fix:

  Prevent idle task from ever being preempted. That makes sure that
  synchronize_rcu_tasks() which is ignoring idle task does not pretend
  that no task is stuck in preempted state. If that happens and idle was
  preempted on a ftrace trampoline the machine crashes due to
  inconsistent state"

* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched/core: Call __schedule() from do_idle() without enabling preemption

Pull irq fixes from Thomas Gleixner:
 "A set of small fixes for the irq subsystem:

   - Cure a data ordering problem with chained interrupts

   - Three small fixlets for the mbigen irq chip"

* 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  genirq: Fix chained interrupt data ordering
  irqchip/mbigen: Fix the clear register offset calculation
  irqchip/mbigen: Fix potential NULL dereferencing
  irqchip/mbigen: Fix memory mapping code

failing sys_wait4() won't fill struct rusage...

Cc: stable@vger.kernel.org
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

__put_user_size() relies upon its first argument having the same type as what
the second one points to; the only other user makes sure of that and
unsafe_put_user() should do the same.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

Pull tracing fixes from Steven Rostedt:

 - Fix a bug caused by not cleaning up the new instance unique triggers
   when deleting an instance. It also creates a selftest that triggers
   that bug.

 - Fix the delayed optimization happening after kprobes boot up self
   tests being removed by freeing of init memory.

 - Comment kprobes on why the delay optimization is not a problem for
   removal of modules, to keep other developers from searching that
   riddle.

 - Fix another case of rcu not watching in stack trace tracing.

* tag 'trace-v4.12-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace:
  tracing: Make sure RCU is watching before calling a stack trace
  kprobes: Document how optimized kprobes are removed from module unload
  selftests/ftrace: Add test to remove instance with active event triggers
  selftests/ftrace: Fix bashisms
  ftrace: Remove #ifdef from code and add clear_ftrace_function_probes() stub
  ftrace/instances: Clear function triggers when removing instances
  ftrace: Simplify glob handling in unregister_ftrace_function_probe_func()
  tracing/kprobes: Enforce kprobes teardown after testing
  tracing: Move postpone selftests to core from early_initcall

Pull block fixes from Jens Axboe:
 "A small collection of fixes that should go into this cycle.

   - a pull request from Christoph for NVMe, which ended up being
     manually applied to avoid pulling in newer bits in master. Mostly
     fibre channel fixes from James, but also a few fixes from Jon and
     Vijay

   - a pull request from Konrad, with just a single fix for xen-blkback
     from Gustavo.

   - a fuseblk bdi fix from Jan, fixing a regression in this series with
     the dynamic backing devices.

   - a blktrace fix from Shaohua, replacing sscanf() with kstrtoull().

   - a request leak fix for drbd from Lars, fixing a regression in the
     last series with the kref changes. This will go to stable as well"

* 'for-linus' of git://git.kernel.dk/linux-block:
  nvmet: release the sq ref on rdma read errors
  nvmet-fc: remove target cpu scheduling flag
  nvme-fc: stop queues on error detection
  nvme-fc: require target or discovery role for fc-nvme targets
  nvme-fc: correct port role bits
  nvme: unmap CMB and remove sysfs file in reset path
  blktrace: fix integer parse
  fuseblk: Fix warning in super_setup_bdi_name()
  block: xen-blkback: add null check to avoid null pointer dereference
  drbd: fix request leak introduced by locking/atomic, kref: Kill kref_sub()

On rdma read errors, release the sq ref that was taken
when the req was initialized. This avoids a hang in
nvmet_sq_destroy() when the queue is being freed.
Signed-off-by: Vijay Immanuel <vijayi@attalasystems.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>

Remove NVMET_FCTGTFEAT_NEEDS_CMD_CPUSCHED. It's unnecessary.
Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>

Per the recommendation by Sagi on:
http://lists.infradead.org/pipermail/linux-nvme/2017-April/009261.html

Rather than waiting for reset work thread to stop queues and abort the ios,
immediately stop the queues on error detection. Reset thread will restop
the queues (as it's called on other paths), but it does not appear to have
a side effect.
Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>