Quite a few of the functions are used only in a single class. Moving
functions accordingly to improve the overall structure.
Furthermore, introduce a base class for the providers, which might also
come handy for future extensions.
Signed-off-by: Stefan Raspl <raspl@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Simplify a couple of initialization routines:
* TracepointProvider and DebugfsProvider: Pass pid into __init__() instead
  of switching to the requested value in an extra call after initializing
  to the default first.
* Pass a single options object into Stats.__init__(), delaying options
  evaluation accordingly, instead of evaluating options first and passing
  several parts of the options object to Stats.__init__() individually.
* Eliminate Stats.update_provider_pid(), since this 2-line function is now
  used in a single place only.
* Remove extra call to update_drilldown() in Tui.__init__() by getting the
  value of options.fields right initially when parsing options.
* Simplify get_providers() logic.
* Avoid duplicate fields initialization by handling it once in the
  providers' __init__() methods.
Signed-off-by: Stefan Raspl <raspl@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Signed-off-by: Stefan Raspl <raspl@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Function available_fields() is not used in any place.
Signed-off-by: Stefan Raspl <raspl@linux.vnet.ibm.com>
Reviewed-by: Janosch Frank <frankja@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Simplify line print logic for header and data lines in interactive mode
as previously suggested by Radim.
While at it, add a space between the first two columns to avoid the
total bleeding into the event name.
Furthermore, for column 'Current', differentiate between no events being
reported (empty 'Current' column) vs the case where events were reported
but the average was rounded down to zero ('0' in 'Current column), for
the folks who appreciate the difference.
Finally: Only skip events which were not reported at all yet, instead of
events that don't have a value in the current interval.
Considered using constants for the field widths in the format strings.
However, that would make things a bit more complicated, and considering
that there are only two places where output happens, I figured it isn't
worth the trouble.
Signed-off-by: Stefan Raspl <raspl@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Certain interactive commands will not modify any information displayed in
the header, hence we can skip them.
Signed-off-by: Stefan Raspl <raspl@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

We should not use the initial sleeptime for any key press that does not
switch to a different screen, as that introduces an unaesthetic flicker due
to two updates in quick succession.
Signed-off-by: Stefan Raspl <raspl@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

When an update interval is interrupted via key press (e.g. space), the
'Current' column value is calculated using the full interval length
instead of the elapsed time, which leads to lower than actual numbers.
Furthermore, the value should be rounded, not truncated.
This is fixed by using the actual elapsed time for the calculation.
Signed-off-by: Stefan Raspl <raspl@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Signed-off-by: Stefan Raspl <raspl@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

The guest-linear address field is set for VM exits due to attempts to
execute LMSW with a memory operand and VM exits due to attempts to
execute INS or OUTS for which the relevant segment is usable,
regardless of whether or not EPT is in use.

Fixes: 119a9c01 ("KVM: nVMX: pass valid guest linear-address to the L1")
Signed-off-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>

The XSS-exiting bitmap is a VMCS control field that does not change
while the CPU is in non-root mode. Transferring the unchanged value
from vmcs02 to vmcs12 is unnecessary.
Signed-off-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>

Bits 11:2 must be zero and the linear addess in bits 63:12 must be
canonical. Otherwise, WRMSR(BNDCFGS) should raise #GP.

Fixes: 0dd376e7 ("KVM: x86: add MSR_IA32_BNDCFGS to msrs_to_save")
Signed-off-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>

The BNDCFGS MSR should only be exposed to the guest if the guest
supports MPX. (cf. the TSC_AUX MSR and RDTSCP.)

Fixes: 0dd376e7 ("KVM: x86: add MSR_IA32_BNDCFGS to msrs_to_save")
Change-Id: I3ad7c01bda616715137ceac878f3fa7e66b6b387
Signed-off-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>

The MSR permission bitmaps are shared by all VMs. However, some VMs
may not be configured to support MPX, even when the host does. If the
host supports VMX and the guest does not, we should intercept accesses
to the BNDCFGS MSR, so that we can synthesize a #GP
fault. Furthermore, if the host does not support MPX and the
"ignore_msrs" kvm kernel parameter is set, then we should intercept
accesses to the BNDCFGS MSR, so that we can skip over the rdmsr/wrmsr
without raising a #GP fault.

Fixes: da8999d3 ("KVM: x86: Intel MPX vmx and msr handle")
Signed-off-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>

em_fxstor previously called fxstor_fixup.  Both created instances of
struct fxregs_state on the stack, which triggered the warning:

arch/x86/kvm/emulate.c:4018:12: warning: stack frame size of 1080 bytes
in function
      'em_fxrstor' [-Wframe-larger-than=]
static int em_fxrstor(struct x86_emulate_ctxt *ctxt)
           ^
with CONFIG_FRAME_WARN set to 1024.

This patch does the fixup in em_fxstor now, avoiding one additional
struct fxregs_state, and now fxstor_fixup can be removed as it has no
other call sites.

Further, the calculation for offsets into xmm_space can be shared
between em_fxstor and em_fxsave.
Signed-off-by: Nick Desaulniers <nick.desaulniers@gmail.com>
[Clean up calculation of offsets and fix it for 64-bit mode. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

This should have been indented one more character over and it should use
tabs.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>

I moved the || to the line before.  Also I replaced some spaces with a
tab on the "return 0;" line.  It looks OK in the diff but originally
that line was only indented 7 spaces.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Joerg Roedel <jroedel@suse.de>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>

When spin_lock_irqsave() deadlock occurs inside the guest, vcpu threads,
other than the lock-holding one, would enter into S state because of
pvspinlock. Then inject NMI via libvirt API "inject-nmi", the NMI could
not be injected into vm.

The reason is:
1 It sets nmi_queued to 1 when calling ioctl KVM_NMI in qemu, and sets
cpu->kvm_vcpu_dirty to true in do_inject_external_nmi() meanwhile.
2 It sets nmi_queued to 0 in process_nmi(), before entering guest, because
cpu->kvm_vcpu_dirty is true.

It's not enough just to check nmi_queued to decide whether to stay in
vcpu_block() or not. NMI should be injected immediately at any situation.
Add checking nmi_pending, and testing KVM_REQ_NMI replaces nmi_queued
in vm_vcpu_has_events().

Do the same change for SMIs.
Signed-off-by: Zhuang Yanying <ann.zhuangyanying@huawei.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

This is a fix for the problem [1], where VMCB.CPL was set to 0 and interrupt
was taken on userspace stack.  The root cause lies in the specific AMD CPU
behaviour which manifests itself as unusable segment attributes on SYSRET.
The corresponding work around for the kernel is the following:

61f01dd9 ("x86_64, asm: Work around AMD SYSRET SS descriptor attribute issue")

In other turn virtualization side treated unusable segment incorrectly and
restored CPL from SS attributes, which were zeroed out few lines above.

In current patch it is assured only that P bit is cleared in VMCB.save state
and segment attributes are not zeroed out if segment is not presented or is
unusable, therefore CPL can be safely restored from DPL field.

This is only one part of the fix, since QEMU side should be fixed accordingly
not to zero out attributes on its side.  Corresponding patch will follow.

[1] Message id: CAJrWOzD6Xq==b-zYCDdFLgSRMPM-NkNuTSDFEtX=7MreT45i7Q@mail.gmail.com
Signed-off-by: Roman Pen <roman.penyaev@profitbricks.com>
Signed-off-by: Mikhail Sennikovskii <mikhail.sennikovskii@profitbricks.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim KrÄmář <rkrcmar@redhat.com>
Cc: kvm@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Commit 19bca6ab ("KVM: SVM: Fix cross vendor migration issue with
unusable bit") added checking type when setting unusable.
So unusable can be set if present is 0 OR type is 0.
According to the AMD processor manual, long mode ignores the type value
in segment descriptor. And type can be 0 if it is read-only data segment.
Therefore type value is not related to unusable flag.

This patch is based on linux-next v4.12.0-rc3.
Signed-off-by: Gioh Kim <gi-oh.kim@profitbricks.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

kvm_skip_emulated_instruction() will return 0 if userspace is
single-stepping the guest.

kvm_skip_emulated_instruction() uses return status convention of exit
handler: 0 means "exit to userspace" and 1 means "continue vm entries".
The problem is that nested_vmx_check_vmptr() return status means
something else: 0 is ok, 1 is error.

This means we would continue executing after a failure.  Static checker
noticed it because vmptr was not initialized.
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Fixes: 6affcbed ("KVM: x86: Add kvm_skip_emulated_instruction and use it.")
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Intel SDM says, that at most one LAPIC should be configured with ExtINT
delivery. KVM configures all LAPICs this way. This causes pic_unlock()
to kick the first available vCPU from the internal KVM data structures.
If this vCPU is not the BSP, but some not-yet-booted AP, the BSP may
never realize that there is an interrupt.

Fix that by enabling ExtINT delivery only for the BSP.

This allows booting a Linux guest without a TSC in the above situation.
Otherwise the BSP gets stuck in calibrate_delay_converge().
Signed-off-by: Jan H. Schönherr <jschoenh@amazon.de>
Reviewed-by: Wanpeng Li <wanpeng.li@hotmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

The decision whether or not to exit from L2 to L1 on an lmsw instruction is
based on bogus values: instead of using the information encoded within the
exit qualification, it uses the data also used for the mov-to-cr
instruction, which boils down to using whatever is in %eax at that point.

Use the correct values instead.

Without this fix, an L1 may not get notified when a 32-bit Linux L2
switches its secondary CPUs to protected mode; the L1 is only notified on
the next modification of CR0. This short time window poses a problem, when
there is some other reason to exit to L1 in between. Then, L2 will be
resumed in real mode and chaos ensues.
Signed-off-by: Jan H. Schönherr <jschoenh@amazon.de>
Reviewed-by: Wanpeng Li <wanpeng.li@hotmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Preemption can occur during cancel preemption timer, and there will be
inconsistent status in lapic, vmx and vmcs field.

          CPU0                    CPU1

  preemption timer vmexit
  handle_preemption_timer(vCPU0)
    kvm_lapic_expired_hv_timer
      vmx_cancel_hv_timer
        vmx->hv_deadline_tsc = -1
        vmcs_clear_bits
        /* hv_timer_in_use still true */
  sched_out
                           sched_in
                           kvm_arch_vcpu_load
                             vmx_set_hv_timer
                               write vmx->hv_deadline_tsc
                               vmcs_set_bits
                           /* back in kvm_lapic_expired_hv_timer */
                           hv_timer_in_use = false
                           ...
                           vmx_vcpu_run
                             vmx_arm_hv_run
                               write preemption timer deadline
                             spurious preemption timer vmexit
                               handle_preemption_timer(vCPU0)
                                 kvm_lapic_expired_hv_timer
                                   WARN_ON(!apic->lapic_timer.hv_timer_in_use);

This can be reproduced sporadically during boot of L2 on a
preemptible L1, causing a splat on L1.

 WARNING: CPU: 3 PID: 1952 at arch/x86/kvm/lapic.c:1529 kvm_lapic_expired_hv_timer+0xb5/0xd0 [kvm]
 CPU: 3 PID: 1952 Comm: qemu-system-x86 Not tainted 4.12.0-rc1+ #24 RIP: 0010:kvm_lapic_expired_hv_timer+0xb5/0xd0 [kvm]
  Call Trace:
  handle_preemption_timer+0xe/0x20 [kvm_intel]
  vmx_handle_exit+0xc9/0x15f0 [kvm_intel]
  ? lock_acquire+0xdb/0x250
  ? lock_acquire+0xdb/0x250
  ? kvm_arch_vcpu_ioctl_run+0xdf3/0x1ce0 [kvm]
  kvm_arch_vcpu_ioctl_run+0xe55/0x1ce0 [kvm]
  kvm_vcpu_ioctl+0x384/0x7b0 [kvm]
  ? kvm_vcpu_ioctl+0x384/0x7b0 [kvm]
  ? __fget+0xf3/0x210
  do_vfs_ioctl+0xa4/0x700
  ? __fget+0x114/0x210
  SyS_ioctl+0x79/0x90
  do_syscall_64+0x8f/0x750
  ? trace_hardirqs_on_thunk+0x1a/0x1c
  entry_SYSCALL64_slow_path+0x25/0x25

This patch fixes it by disabling preemption while cancelling
preemption timer.  This way cancel_hv_timer is atomic with
respect to kvm_arch_vcpu_load.

Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

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>

In order to create an association, the remoteport must be
serving either a target role or a discovery role.
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>

FC Port roles is a bit mask, not individual values.
Correct nvme definitions to unique bits.
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>

CMB doesn't get unmapped until removal while getting remapped on every
reset. Add the unmapping and sysfs file removal to the reset path in
nvme_pci_disable to match the mapping path in nvme_pci_enable.

Fixes: 202021c1 ("nvme : Add sysfs entry for NVMe CMBs when appropriate")
Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>
Acked-by: Keith Busch <keith.busch@intel.com>
Reviewed-By: Stephen Bates <sbates@raithlin.com>
Cc: <stable@vger.kernel.org> # 4.9+
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>