openwrt/staging/blogic.git
5 years agonvme: add support for the Write Zeroes command
Chaitanya Kulkarni [Tue, 18 Dec 2018 03:42:03 +0000 (22:42 -0500)]
nvme: add support for the Write Zeroes command

Allow write zeroes operations (REQ_OP_WRITE_ZEROES) on the block
device, if the device supports an optional command bit set for write
zeroes. Add support to setup write zeroes command. Set maximum possible
write zeroes sectors in one write zeroes command according to
nvme write zeroes command definition.

This patch was posted as a part of block-write-zeroes support
implementation (https://patchwork.kernel.org/patch/9454859/),
but did not make into mainline kernel as it got reverted due to
failure on the Linus's machine.

In this patch in order to be more cautious, we use NVMe controller's
maximum hardware sector size which is calculated based on the
controller's MDTS (Maximum Data Transfer Size) field to calculate
the maximum sectors for the write zeroes request.

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
[folded a fix from Keith Busch to properly respect
 NVME_QUIRK_DEALLOCATE_ZEROES]
Signed-off-by: Christoph Hellwig <hch@lst.de>
5 years agoblk-mq: save default hctx into ctx->hctxs for not-supported type
Jianchao Wang [Thu, 24 Jan 2019 10:25:33 +0000 (18:25 +0800)]
blk-mq: save default hctx into ctx->hctxs for not-supported type

Currently, we check whether the hctx type is supported every time
in hot path. Actually, this is not necessary, we could save the
default hctx into ctx->hctxs if the type is not supported when
map swqueues and use it directly with ctx->hctxs[type].

We also needn't check whether the poll is enabled or not, because
the caller would clear the REQ_HIPRI in that case.

Signed-off-by: Jianchao Wang <jianchao.w.wang@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
5 years agoblk-mq: save queue mapping result into ctx directly
Jianchao Wang [Thu, 24 Jan 2019 10:25:32 +0000 (18:25 +0800)]
blk-mq: save queue mapping result into ctx directly

Currently, the queue mapping result is saved in a two-dimensional
array. In the hot path, to get a hctx, we need do following:

  q->queue_hw_ctx[q->tag_set->map[type].mq_map[cpu]]

This isn't very efficient. We could save the queue mapping result into
ctx directly with different hctx type, like,

  ctx->hctxs[type]

Signed-off-by: Jianchao Wang <jianchao.w.wang@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
5 years agoblock, bfq: fix in-service-queue check for queue merging
Paolo Valente [Tue, 29 Jan 2019 11:06:38 +0000 (12:06 +0100)]
block, bfq: fix in-service-queue check for queue merging

When a new I/O request arrives for a bfq_queue, say Q, bfq checks
whether that request is close to
(a) the head request of some other queue waiting to be served, or
(b) the last request dispatched for the in-service queue (in case Q
itself is not the in-service queue)

If a queue, say Q2, is found for which the above condition holds, then
bfq merges Q and Q2, to hopefully get a more sequential I/O in the
resulting merged queue, and thus a possibly higher throughput.

Case (b) is checked by comparing the new request for Q with the last
request dispatched, assuming that the latter necessarily belonged to the
in-service queue. Unfortunately, this assumption is no longer always
correct, since commit d0edc2473be9 ("block, bfq: inject other-queue I/O
into seeky idle queues on NCQ flash").

When the assumption does not hold, queues that must not be merged may be
merged, causing unexpected loss of control on per-queue service
guarantees.

This commit solves this problem by adding an extra field, which stores
the actual last request dispatched for the in-service queue, and by
using this new field to correctly check case (b).

Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
5 years agoblock, bfq: do not overcharge writes in asymmetric scenarios
Paolo Valente [Tue, 29 Jan 2019 11:06:37 +0000 (12:06 +0100)]
block, bfq: do not overcharge writes in asymmetric scenarios

Writes tend to starve reads. bfq counters this problem by overcharging
writes with an inflated service w.r.t. the actual service (number of
sector written) they receive.

Yet his overcharging is useless, and actually causes unfairness in the
opposite direction, when bfq happens to be enforcing strong I/O control.
bfq does this enforcing when the scenario is asymmetric, i.e., when some
bfq_queue or group of bfq_queues is to be granted a different bandwidth
than some other bfq_queue or group of bfq_queues. So, in such a
scenario, this commit disables write overcharging.

Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
5 years agoblock, bfq: port commit "cfq-iosched: improve hw_tag detection"
Paolo Valente [Tue, 29 Jan 2019 11:06:36 +0000 (12:06 +0100)]
block, bfq: port commit "cfq-iosched: improve hw_tag detection"

The original commit is commit 1a1238a7dd48 ("cfq-iosched: improve hw_tag
detection") and has the following commit message:

If active queue hasn't enough requests and idle window opens, cfq will
not dispatch sufficient requests to hardware. In such situation, current
code will zero hw_tag. But this is because cfq doesn't dispatch enough
requests instead of hardware queue doesn't work. Don't zero hw_tag in
such case.

Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
5 years agoblock, bfq: reduce threshold for detecting command queueing
Paolo Valente [Tue, 29 Jan 2019 11:06:35 +0000 (12:06 +0100)]
block, bfq: reduce threshold for detecting command queueing

bfq simple heuristic from cfq for detecting whether the drive performs
command queueing: check whether the average number of in-flight requests
is above a given threshold. Unfortunately this heuristic does fail to
detect queueing (on drives with queueing) if processes doing I/O are few
and issue I/O with a low depth.

To reduce false negatives, this commit lowers the threshold.

Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
5 years agoblock, bfq: fix queue removal from weights tree
Paolo Valente [Tue, 29 Jan 2019 11:06:34 +0000 (12:06 +0100)]
block, bfq: fix queue removal from weights tree

bfq maintains an ordered list, through a red-black tree, of unique
weights of active bfq_queues. This list is used to detect whether there
are active queues with differentiated weights. The weight of a queue is
removed from the list when both the following two conditions become
true:

(1) the bfq_queue is flagged as inactive
(2) the has no in-flight request any longer;

Unfortunately, in the rare cases where condition (2) becomes true before
condition (1), the removal fails, because the function to remove the
weight of the queue (bfq_weights_tree_remove) is rightly invoked in the
path that deactivates the bfq_queue, but mistakenly invoked *before* the
function that actually performs the deactivation (bfq_deactivate_bfqq).

This commits moves the invocation of bfq_weights_tree_remove for
condition (1) to after bfq_deactivate_bfqq. As a consequence of this
move, it is necessary to add a further reference to the queue when the
weight of a queue is added, because the queue might otherwise be freed
before bfq_weights_tree_remove is invoked. This commit adds this
reference and makes all related modifications.

Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
5 years agoblock, bfq: fix sequential rq detection in rate estimation
Paolo Valente [Tue, 29 Jan 2019 11:06:33 +0000 (12:06 +0100)]
block, bfq: fix sequential rq detection in rate estimation

In bfq_update_peak_rate, to check whether an I/O request rq is
sequential, only the seek distance of rq w.r.t. the last request
dispatched is controlled. This is not sufficient for non-rotational
storage, where the size of rq is at least as relevant. This commit adds
the missing control.

Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
5 years agoblock, bfq: unconditionally plug I/O in asymmetric scenarios
Paolo Valente [Tue, 29 Jan 2019 11:06:32 +0000 (12:06 +0100)]
block, bfq: unconditionally plug I/O in asymmetric scenarios

bfq detects the creation of multiple bfq_queues shortly after each
other, namely a burst of queue creations in the terminology used in the
code. If the burst is large, then no queue in the burst is granted
- either I/O-dispatch plugging when the queue remains temporarily idle
  while in service;
- or weight raising, because it causes even longer plugging.

In fact, such a plugging tends to lower throughput, while these bursts
are typically due to applications or services that spawn multiple
processes, to reach a common goal as soon as possible. Examples are a
"git grep" or the booting of a system.

Unfortunately, disabling plugging may cause a loss of service guarantees
in asymmetric scenarios, i.e., if queue weights are differentiated or if
more than one group is active.

This commit addresses this issue by no longer disabling I/O-dispatch
plugging for queues in large bursts.

Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
5 years agoblock, bfq: do not plug I/O of in-service queue when harmful
Paolo Valente [Tue, 29 Jan 2019 11:06:31 +0000 (12:06 +0100)]
block, bfq: do not plug I/O of in-service queue when harmful

If the in-service bfq_queue is sync and remains temporarily idle, then
I/O dispatching (from other queues) may be plugged. It may be dome for
two reasons: either to boost throughput, or to preserve the bandwidth
share of the in-service queue. In the first case, if the I/O of the
in-service queue, when it finally arrives, consists only of one small
I/O request, then it makes sense to plug even the I/O of the in-service
queue. In fact, serving such a small request immediately is likely to
lower throughput instead of boosting it, whereas waiting a little bit is
likely to let that request grow, thanks to request merging, and become
more profitable in terms of throughput (this is likely to happen exactly
because the I/O of the queue has been detected to boost throughput).

On the opposite end, if I/O dispatching is being plugged only to
preserve the bandwidth of the in-service queue, then it would be better
not to plug also the I/O of the in-service queue, because such a
plugging is likely to cause only loss of bandwidth for the queue.

Unfortunately, no distinction is made between the two cases, and the I/O
of the in-service queue is always plugged in case just a small I/O
request arrives. This commit draws this missing distinction and does not
perform harmful plugging.

Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
5 years agoblock, bfq: split function bfq_better_to_idle
Paolo Valente [Tue, 29 Jan 2019 11:06:30 +0000 (12:06 +0100)]
block, bfq: split function bfq_better_to_idle

This is a preparatory commit for commits that need to check only one of
the two main reasons for idling. This change should also improve the
quality of the code a little bit, by splitting a function that contains
very long, non-trivial and little related comments.

Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
5 years agoblock, bfq: consider also ioprio classes in symmetry detection
Paolo Valente [Tue, 29 Jan 2019 11:06:29 +0000 (12:06 +0100)]
block, bfq: consider also ioprio classes in symmetry detection

In asymmetric scenarios, i.e., when some bfq_queue or bfq_group needs to
be guaranteed a different bandwidth than other bfq_queues or bfq_groups,
these service guaranteed can be provided only by plugging I/O dispatch,
completely or partially, when the queue in service remains temporarily
empty. A case where asymmetry is particularly strong is when some active
bfq_queues belong to a higher-priority class than some other active
bfq_queues. Unfortunately, this important case is not considered at all
in the code for detecting asymmetric scenarios. This commit adds the
missing logic.

Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
5 years agoblock, bfq: remove case of redirected bic from insert_request
Paolo Valente [Tue, 29 Jan 2019 11:06:28 +0000 (12:06 +0100)]
block, bfq: remove case of redirected bic from insert_request

Before commit 18e5a57d7987 ("block, bfq: postpone rq preparation to
insert or merge"), the destination queue for a request was chosen by a
different hook than the one that then inserted the request. So, between
the execution of the two hooks, the bic of the process generating the
request could happen to be redirected to a different bfq_queue. As a
consequence, the destination bfq_queue stored in the request could be
wrong. Such an event does not need to ba handled any longer.

Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
5 years agoblock, bfq: make sure queue budgets are not below service received
Paolo Valente [Tue, 29 Jan 2019 11:06:27 +0000 (12:06 +0100)]
block, bfq: make sure queue budgets are not below service received

With some unlucky sequences of events, the function bfq_updated_next_req
updates the current budget of a bfq_queue to a lower value than the
service received by the queue using such a budget. Unfortunately, if
this happens, then the return value of the function bfq_bfqq_budget_left
becomes inconsistent. This commit solves this problem by lower-bounding
the budget computed in bfq_updated_next_req to the service currently
charged to the queue.

Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
5 years agoblock, bfq: avoid selecting a queue w/o budget
Paolo Valente [Tue, 29 Jan 2019 11:06:26 +0000 (12:06 +0100)]
block, bfq: avoid selecting a queue w/o budget

To boost throughput on devices with internal queueing and in scenarios
where device idling is not strictly needed, bfq immediately starts
serving a new bfq_queue if the in-service bfq_queue remains without
pending I/O, even if new I/O may arrive soon for the latter queue. Then,
if such I/O actually arrives soon, bfq preempts the new in-service
bfq_queue so as to give the previous queue a chance to go on being
served (in case the previous queue should actually be the one to be
served, according to its timestamps).

However, the in-service bfq_queue, say Q, may also be without further
budget when it remains also pending I/O. Since bfq changes budgets
dynamically to fit the needs of bfq_queues, this happens more often than
one may expect. If this happens, then there is no point in trying to go
on serving Q when new I/O arrives for it soon: Q would be expired
immediately after being selected for service. This would only cause
useless overhead. This commit avoids such a useless selection.

Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
5 years agoblock, bfq: do not consider interactive queues in srt filtering
Paolo Valente [Tue, 29 Jan 2019 11:06:25 +0000 (12:06 +0100)]
block, bfq: do not consider interactive queues in srt filtering

The speed at which a bfq_queue receives I/O is one of the parameters by
which bfq decides whether the queue is soft real-time (i.e., whether the
queue contains the I/O of a soft real-time application). In particular,
when a bfq_queue remains without outstanding I/O requests, bfq computes
the minimum time instant, named soft_rt_next_start, at which the next
request of the queue may arrive for the queue to be deemed as soft real
time.

Unfortunately this filtering may cause problems with a queue in
interactive weight raising. In fact, such a queue may be conveying the
I/O needed to load a soft real-time application. The latter will
actually exhibit a soft real-time I/O pattern after it finally starts
doing its job. But, if soft_rt_next_start is updated for an interactive
bfq_queue, and the queue has received a lot of service before remaining
with no outstanding request (likely to happen on a fast device), then
soft_rt_next_start is assigned such a high value that, for a very long
time, the queue is prevented from being possibly considered as soft real
time.

This commit removes the updating of soft_rt_next_start for bfq_queues in
interactive weight raising.

Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
5 years agomtip32xx: ѕtop abusing the managed resource APIs
Christoph Hellwig [Mon, 14 Jan 2019 20:23:30 +0000 (21:23 +0100)]
mtip32xx: ѕtop abusing the managed resource APIs

The mtip32xx driver uses managed resources for DMA coherent memory
and irqs, but then always pairs them with free calls anyway, making
the resource tracking rather pointless.  Given some DMA allocations
are transient anyway, the irq freeing seems to require ordering vs
other hardware access the best solution seems to be to stop using
the managed resource API entirely.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
5 years agoLinux 5.0-rc4
Linus Torvalds [Sun, 27 Jan 2019 23:18:05 +0000 (15:18 -0800)]
Linux 5.0-rc4

5 years agoMerge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 27 Jan 2019 20:02:00 +0000 (12:02 -0800)]
Merge branch 'x86-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Thomas Gleixner:
 "A set of fixes for x86:

   - Fix the swapped outb() parameters in the KASLR code

   - Fix the PKEY handling at fork which missed to preserve the pkey
     state for the child. Comes with a test case to validate that.

   - Fix the entry stack handling for XEN PV to respect that XEN PV
     systems enter the function already on the current thread stack and
     not on the trampoline.

   - Fix kexec load failure caused by using a stale value when the
     kexec_buf structure is reused for subsequent allocations.

   - Fix a bogus sizeof() in the memory encryption code

   - Enforce PCI dependency for the Intel Low Power Subsystem

   - Enforce PCI_LOCKLESS_CONFIG when PCI is enabled"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/Kconfig: Select PCI_LOCKLESS_CONFIG if PCI is enabled
  x86/entry/64/compat: Fix stack switching for XEN PV
  x86/kexec: Fix a kexec_file_load() failure
  x86/mm/mem_encrypt: Fix erroneous sizeof()
  x86/selftests/pkeys: Fork() to check for state being preserved
  x86/pkeys: Properly copy pkey state at fork()
  x86/kaslr: Fix incorrect i8254 outb() parameters
  x86/intel/lpss: Make PCI dependency explicit

5 years agoMerge branch 'x86-timers-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 27 Jan 2019 19:57:46 +0000 (11:57 -0800)]
Merge branch 'x86-timers-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86 timer fixes from Thomas Gleixner:
 "Two commits which were missed to be sent during the merge window.

   - The TSC calibration fix turns out to be more urgent as recent
     Skylake-X systems seem to have massive trouble with calibration
     disturbance. This should go back into stable for that reason and it
     the risk of breakage is rather low.

   - Drop an unused define"

* 'x86-timers-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/hpet: Remove unused FSEC_PER_NSEC define
  x86/tsc: Make calibration refinement more robust

5 years agoMerge branch 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 27 Jan 2019 19:55:06 +0000 (11:55 -0800)]
Merge branch 'timers-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull timer fix from Thomas Glexiner:
 "A single regression fix to address the unintended breakage of posix
  cpu timers.

  This is caused by a new sanity check in the common code, which fails
  for posix cpu timers under certain conditions because the posix cpu
  timer code never updates the variable which is checked"

* 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  posix-cpu-timers: Unbreak timer rearming

5 years agoMerge branch 'locking-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 27 Jan 2019 19:52:50 +0000 (11:52 -0800)]
Merge branch 'locking-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull locking fixes from Thomas Gleixner:
 "A small series of fixes which all address possible missed wakeups:

   - Document and fix the wakeup ordering of wake_q

   - Add the missing barrier in rcuwait_wake_up(), which was documented
     in the comment but missing in the code

   - Fix the possible missed wakeups in the rwsem and futex code"

* 'locking-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  locking/rwsem: Fix (possible) missed wakeup
  futex: Fix (possible) missed wakeup
  sched/wake_q: Fix wakeup ordering for wake_q
  sched/wake_q: Document wake_q_add()
  sched/wait: Fix rcuwait_wake_up() ordering

5 years agoMerge branch 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 27 Jan 2019 19:25:38 +0000 (11:25 -0800)]
Merge branch 'irq-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

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

   - Fix a double increment in the irq descriptor allocator which
     resulted in a sanity check only being done for every second
     affinity mask

   - Add a missing device tree translation in the stm32-exti driver.
     Without that the interrupt association is completely wrong.

   - Initialize the mutex in the GIC-V3 MBI driver

   - Fix the alignment for aliasing devices in the GIC-V3-ITS driver so
     multi MSI allocations work correctly

   - Ensure that the initial affinity of a interrupt is not empty at
     startup time.

   - Drop bogus include in the madera irq chip driver

   - Fix KernelDoc regression"

* 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  irqchip/gic-v3-its: Align PCI Multi-MSI allocation on their size
  genirq/irqdesc: Fix double increment in alloc_descs()
  genirq: Fix the kerneldoc comment for struct irq_affinity_desc
  irqchip/madera: Drop GPIO includes
  irqchip/gic-v3-mbi: Fix uninitialized mbi_lock
  irqchip/stm32-exti: Add domain translate function
  genirq: Make sure the initial affinity is not empty

5 years agoMerge tag 'edac_fix_for_5.0' of git://git.kernel.org/pub/scm/linux/kernel/git/bp/bp
Linus Torvalds [Sun, 27 Jan 2019 19:00:37 +0000 (11:00 -0800)]
Merge tag 'edac_fix_for_5.0' of git://git./linux/kernel/git/bp/bp

Pull EDAC fix from Borislav Petkov:
 "Fix persistent register offsets of altera_edac, from Thor Thayer"

* tag 'edac_fix_for_5.0' of git://git.kernel.org/pub/scm/linux/kernel/git/bp/bp:
  EDAC, altera: Fix S10 persistent register offset

5 years agoMerge tag 'for-linus-20190127' of git://git.kernel.dk/linux-block
Linus Torvalds [Sun, 27 Jan 2019 18:58:20 +0000 (10:58 -0800)]
Merge tag 'for-linus-20190127' of git://git.kernel.dk/linux-block

Pull block revert from Jens Axboe:
 "Silly error snuck into a patch from the last series, let's do a revert
  to avoid a potential use-after-free"

* tag 'for-linus-20190127' of git://git.kernel.dk/linux-block:
  Revert "block: cover another queue enter recursion via BIO_QUEUE_ENTERED"

5 years agoMerge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Linus Torvalds [Sun, 27 Jan 2019 17:21:00 +0000 (09:21 -0800)]
Merge tag 'for-linus' of git://git./virt/kvm/kvm

Pull KVM fixes from Paolo Bonzini:
 "Quite a few fixes for x86: nested virtualization save/restore, AMD
  nested virtualization and virtual APIC, 32-bit fixes, an important fix
  to restore operation on older processors, and a bunch of hyper-v
  bugfixes. Several are marked stable.

  There are also fixes for GCC warnings and for a GCC/objtool interaction"

* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
  KVM: x86: Mark expected switch fall-throughs
  KVM: x86: fix TRACE_INCLUDE_PATH and remove -I. header search paths
  KVM: selftests: check returned evmcs version range
  x86/kvm/hyper-v: nested_enable_evmcs() sets vmcs_version incorrectly
  KVM: VMX: Move vmx_vcpu_run()'s VM-Enter asm blob to a helper function
  kvm: selftests: Fix region overlap check in kvm_util
  kvm: vmx: fix some -Wmissing-prototypes warnings
  KVM: nSVM: clear events pending from svm_complete_interrupts() when exiting to L1
  svm: Fix AVIC incomplete IPI emulation
  svm: Add warning message for AVIC IPI invalid target
  KVM: x86: WARN_ONCE if sending a PV IPI returns a fatal error
  KVM: x86: Fix PV IPIs for 32-bit KVM host
  x86/kvm/hyper-v: recommend using eVMCS only when it is enabled
  x86/kvm/hyper-v: don't recommend doing reset via synthetic MSR
  kvm: x86/vmx: Use kzalloc for cached_vmcs12
  KVM: VMX: Use the correct field var when clearing VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL
  KVM: x86: Fix single-step debugging
  x86/kvm/hyper-v: don't announce GUEST IDLE MSR support

5 years agoMerge tag 'dma-mapping-5.0-2' of git://git.infradead.org/users/hch/dma-mapping
Linus Torvalds [Sun, 27 Jan 2019 17:18:05 +0000 (09:18 -0800)]
Merge tag 'dma-mapping-5.0-2' of git://git.infradead.org/users/hch/dma-mapping

Pull dma-mapping fix from Christoph Hellwig:
 "Fix a xen-swiotlb regression on arm64"

* tag 'dma-mapping-5.0-2' of git://git.infradead.org/users/hch/dma-mapping:
  arm64/xen: fix xen-swiotlb cache flushing

5 years agoMerge tag 'libnvdimm-fixes-5.0-rc4' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 27 Jan 2019 17:11:51 +0000 (09:11 -0800)]
Merge tag 'libnvdimm-fixes-5.0-rc4' of git://git./linux/kernel/git/nvdimm/nvdimm

Pull libnvdimm fixes from Dan Williams:
 "A fix for namespace label support for non-Intel NVDIMMs that implement
  the ACPI standard label method.

  This has apparently never worked and could wait for v5.1. However it
  has enough visibility with hardware vendors [1] and distro bug
  trackers [2], and low enough risk that I decided it should go in for
  -rc4. The other fixups target the new, for v5.0, nvdimm security
  functionality. The larger init path fixup closes a memory leak and a
  potential userspace lockup due to missed notifications.

    [1] https://github.com/pmem/ndctl/issues/78
    [2] https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1811785

  These have all soaked in -next for a week with no reported issues.

  Summary:

   - Fix support for NVDIMMs that implement the ACPI standard label
     methods.

   - Fix error handling for security overwrite (memory leak / userspace
     hang condition), and another one-line security cleanup"

* tag 'libnvdimm-fixes-5.0-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm:
  acpi/nfit: Fix command-supported detection
  acpi/nfit: Block function zero DSMs
  libnvdimm/security: Require nvdimm_security_setup_events() to succeed
  nfit_test: fix security state pull for nvdimm security nfit_test

5 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
Linus Torvalds [Sun, 27 Jan 2019 17:07:03 +0000 (09:07 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/dtor/input

Pull input fixes from Dmitry Torokhov:
 "A fixup for the input_event fix for y2038 Sparc64, and couple other
  minor fixes"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input:
  Input: input_event - fix the CONFIG_SPARC64 mixup
  Input: olpc_apsp - assign priv->dev earlier
  Input: uinput - fix undefined behavior in uinput_validate_absinfo()
  Input: raspberrypi-ts - fix link error
  Input: xpad - add support for SteelSeries Stratus Duo
  Input: input_event - provide override for sparc64

5 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Sun, 27 Jan 2019 16:59:12 +0000 (08:59 -0800)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Count ttl-dropped frames properly in mac80211, from Bob Copeland.

 2) Integer overflow in ktime handling of bcm can code, from Oliver
    Hartkopp.

 3) Fix RX desc handling wrt. hw checksumming in ravb, from Simon
    Horman.

 4) Various hash key fixes in hv_netvsc, from Haiyang Zhang.

 5) Use after free in ax25, from Eric Dumazet.

 6) Several fixes to the SSN support in SCTP, from Xin Long.

 7) Do not process frames after a NAPI reschedule in ibmveth, from
    Thomas Falcon.

 8) Fix NLA_POLICY_NESTED arguments, from Johannes Berg.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (42 commits)
  qed: Revert error handling changes.
  cfg80211: extend range deviation for DMG
  cfg80211: reg: remove warn_on for a normal case
  mac80211: Add attribute aligned(2) to struct 'action'
  mac80211: don't initiate TDLS connection if station is not associated to AP
  nl80211: fix NLA_POLICY_NESTED() arguments
  ibmveth: Do not process frames after calling napi_reschedule
  net: dev_is_mac_header_xmit() true for ARPHRD_RAWIP
  net: usb: asix: ax88772_bind return error when hw_reset fail
  MAINTAINERS: Update cavium networking drivers
  net/mlx4_core: Fix error handling when initializing CQ bufs in the driver
  net/mlx4_core: Add masking for a few queries on HCA caps
  sctp: set flow sport from saddr only when it's 0
  sctp: set chunk transport correctly when it's a new asoc
  sctp: improve the events for sctp stream adding
  sctp: improve the events for sctp stream reset
  ip_tunnel: Make none-tunnel-dst tunnel port work with lwtunnel
  ax25: fix possible use-after-free
  sfc: suppress duplicate nvmem partition types in efx_ef10_mtd_probe
  hv_netvsc: fix typos in code comments
  ...

5 years agoRevert "block: cover another queue enter recursion via BIO_QUEUE_ENTERED"
Jens Axboe [Sun, 27 Jan 2019 13:35:28 +0000 (06:35 -0700)]
Revert "block: cover another queue enter recursion via BIO_QUEUE_ENTERED"

We can't touch a bio after ->make_request_fn(), for all we know it could
already have been completed by the time this function returns.

This reverts commit 698cef173983b086977e633e46476e0f925ca01e.

Reported-by: syzbot+4df6ca820108fd248943@syzkaller.appspotmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
5 years agoMerge tag '5.0-rc3-smb3-fixes' of git://git.samba.org/sfrench/cifs-2.6
Linus Torvalds [Sat, 26 Jan 2019 23:38:22 +0000 (15:38 -0800)]
Merge tag '5.0-rc3-smb3-fixes' of git://git.samba.org/sfrench/cifs-2.6

Pull smb3 fixes from Steve French:
 "A set of small smb3 fixes, some fixing various crediting issues
  discovered during xfstest runs, five for stable"

* tag '5.0-rc3-smb3-fixes' of git://git.samba.org/sfrench/cifs-2.6:
  cifs: print CIFSMaxBufSize as part of /proc/fs/cifs/DebugData
  smb3: add credits we receive from oplock/break PDUs
  CIFS: Fix mounts if the client is low on credits
  CIFS: Do not assume one credit for async responses
  CIFS: Fix credit calculations in compound mid callback
  CIFS: Fix credit calculation for encrypted reads with errors
  CIFS: Fix credits calculations for reads with errors
  CIFS: Do not reconnect TCP session in add_credits()
  smb3: Cleanup license mess
  CIFS: Fix possible hang during async MTU reads and writes
  cifs: fix memory leak of an allocated cifs_ntsd structure

5 years agoMerge tag 'vfio-v5.0-rc4' of git://github.com/awilliam/linux-vfio
Linus Torvalds [Sat, 26 Jan 2019 23:27:04 +0000 (15:27 -0800)]
Merge tag 'vfio-v5.0-rc4' of git://github.com/awilliam/linux-vfio

Pull VFIO fixes from Alex Williamson:

 - cleanup licenses in new files (Thomas Gleixner)

 - cleanup new compiler warnings (Alexey Kardashevskiy)

* tag 'vfio-v5.0-rc4' of git://github.com/awilliam/linux-vfio:
  vfio-pci/nvlink2: Fix ancient gcc warnings
  vfio/pci: Cleanup license mess

5 years agoMerge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Linus Torvalds [Sat, 26 Jan 2019 23:03:43 +0000 (15:03 -0800)]
Merge tag 'scsi-fixes' of git://git./linux/kernel/git/jejb/scsi

Pull SCSI fixes from James Bottomley:
 "Six fixes, all of which appear to have user visible consequences.

  The DMA one is a regression fix from the merge window and of the
  others, four are driver specific and one specific to the target code"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: ufs: Use explicit access size in ufshcd_dump_regs
  scsi: tcmu: fix use after free
  scsi: csiostor: fix NULL pointer dereference in csio_vport_set_state()
  scsi: lpfc: nvmet: avoid hang / use-after-free when destroying targetport
  scsi: lpfc: nvme: avoid hang / use-after-free when destroying localport
  scsi: communicate max segment size to the DMA mapping code

5 years agoMerge tag 'for-linus-20190125' of git://git.kernel.dk/linux-block
Linus Torvalds [Sat, 26 Jan 2019 20:42:41 +0000 (12:42 -0800)]
Merge tag 'for-linus-20190125' of git://git.kernel.dk/linux-block

Pull block fixes from Jens Axboe:
 "A collection of fixes for this release. This contains:

   - Silence sparse rightfully complaining about non-static wbt
     functions (Bart)

   - Fixes for the zoned comments/ioctl documentation (Damien)

   - direct-io fix that's been lingering for a while (Ernesto)

   - cgroup writeback fix (Tejun)

   - Set of NVMe patches for nvme-rdma/tcp (Sagi, Hannes, Raju)

   - Block recursion tracking fix (Ming)

   - Fix debugfs command flag naming for a few flags (Jianchao)"

* tag 'for-linus-20190125' of git://git.kernel.dk/linux-block:
  block: Fix comment typo
  uapi: fix ioctl documentation
  blk-wbt: Declare local functions static
  blk-mq: fix the cmd_flag_name array
  nvme-multipath: drop optimization for static ANA group IDs
  nvmet-rdma: fix null dereference under heavy load
  nvme-rdma: rework queue maps handling
  nvme-tcp: fix timeout handler
  nvme-rdma: fix timeout handler
  writeback: synchronize sync(2) against cgroup writeback membership switches
  block: cover another queue enter recursion via BIO_QUEUE_ENTERED
  direct-io: allow direct writes to empty inodes

5 years agoqed: Revert error handling changes.
David S. Miller [Fri, 25 Jan 2019 23:32:28 +0000 (15:32 -0800)]
qed: Revert error handling changes.

This is new code and not bug fixes.

This reverts all changes added by merge commit
8fb18be93efd7292d6ee403b9f61af1008239639

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge tag 'mmc-v5.0-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc
Linus Torvalds [Fri, 25 Jan 2019 23:07:03 +0000 (13:07 -1000)]
Merge tag 'mmc-v5.0-rc2' of git://git./linux/kernel/git/ulfh/mmc

Pull MMC fixes from Ulf Hansson:

 - sdhci-acpi: Fixup build dependency for PCI

 - sdhci-omap: Resolve Kconfig warnings on keystone

 - sdhci-iproc: Propagate errors from DT parsing

 - meson-gx: Fixup IRQ handling in release callback

 - meson-gx: Use signal re-sampling to fixup tuning

 - dw_mmc-bluefield: Fix the license information

* tag 'mmc-v5.0-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc:
  mmc: dw_mmc-bluefield: : Fix the license information
  mmc: meson-gx: enable signal re-sampling together with tuning
  mmc: sdhci-iproc: handle mmc_of_parse() errors during probe
  mmc: meson-gx: Free irq in release() callback
  mmc: host: Fix Kconfig warnings on keystone_defconfig
  mmc: sdhci-acpi: Make PCI dependency explicit

5 years agoMerge tag 'char-misc-5.0-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh...
Linus Torvalds [Fri, 25 Jan 2019 23:03:34 +0000 (13:03 -1000)]
Merge tag 'char-misc-5.0-rc4' of git://git./linux/kernel/git/gregkh/char-misc

Pull char/misc driver fixes from Greg KH:
 "Here are some small char and misc driver fixes to resolve some
  reported issues, as well as a number of binderfs fixups that were
  found after auditing the filesystem code by Al Viro. As binderfs
  hasn't been in a previous release yet, it's good to get these in now
  before the first users show up.

  All of these have been in linux-next for a bit with no reported
  issues"

* tag 'char-misc-5.0-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc: (26 commits)
  i3c: master: Fix an error checking typo in 'cdns_i3c_master_probe()'
  binderfs: switch from d_add() to d_instantiate()
  binderfs: drop lock in binderfs_binder_ctl_create
  binderfs: kill_litter_super() before cleanup
  binderfs: rework binderfs_binder_device_create()
  binderfs: rework binderfs_fill_super()
  binderfs: prevent renaming the control dentry
  binderfs: remove outdated comment
  binderfs: use __u32 for device numbers
  binderfs: use correct include guards in header
  misc: pvpanic: fix warning implicit declaration
  char/mwave: fix potential Spectre v1 vulnerability
  misc: ibmvsm: Fix potential NULL pointer dereference
  binderfs: fix error return code in binderfs_fill_super()
  mei: me: add denverton innovation engine device IDs
  mei: me: mark LBG devices as having dma support
  mei: dma: silent the reject message
  binderfs: handle !CONFIG_IPC_NS builds
  binderfs: reserve devices for initial mount
  binderfs: rename header to binderfs.h
  ...

5 years agoMerge tag 'staging-5.0-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh...
Linus Torvalds [Fri, 25 Jan 2019 23:02:12 +0000 (13:02 -1000)]
Merge tag 'staging-5.0-rc4' of git://git./linux/kernel/git/gregkh/staging

Pull staging driver fixes from Greg KH:
 "Here are some small staging driver fixes for 5.0-rc4.

  They resolve some reported bugs and add a new device id for one
  driver. Nothing major at all, but all good to have.

  All of these have been in linux-next for a while with no reported
  issues"

* tag 'staging-5.0-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging:
  staging: android: ion: Support cpu access during dma_buf_detach
  staging: rtl8723bs: Fix build error with Clang when inlining is disabled
  staging: rtl8188eu: Add device code for D-Link DWA-121 rev B1
  staging: vchiq: Fix local event signalling
  Staging: wilc1000: unlock on error in init_chip()
  staging: wilc1000: fix memory leak in wilc_add_rx_gtk
  staging: wilc1000: fix registration frame size

5 years agoMerge tag 'tty-5.0-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty
Linus Torvalds [Fri, 25 Jan 2019 22:58:40 +0000 (12:58 -1000)]
Merge tag 'tty-5.0-rc4' of git://git./linux/kernel/git/gregkh/tty

Pull tty/serial driver fixes from Greg KH:
 "Here are a number of small tty core and serial driver fixes for
  5.0-rc4 to resolve some reported issues.

  Nothing major, the small serial driver fixes, a tty core fixup for a
  crash that was reported, and some good vt fixes from Nicolas Pitre as
  he seems to be auditing that chunk of code a lot lately.

  All of these have been in linux-next for a while with no reported
  issues"

* tag 'tty-5.0-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty:
  serial: fsl_lpuart: fix maximum acceptable baud rate with over-sampling
  tty: serial: qcom_geni_serial: Allow mctrl when flow control is disabled
  tty: Handle problem if line discipline does not have receive_buf
  vgacon: unconfuse vc_origin when using soft scrollback
  vt: invoke notifier on screen size change
  vt: always call notifier with the console lock held
  vt: make vt_console_print() compatible with the unicode screen buffer
  tty/n_hdlc: fix __might_sleep warning
  serial: 8250: Fix serial8250 initialization crash
  uart: Fix crash in uart_write and uart_put_char

5 years agoMerge tag 'usb-5.0-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb
Linus Torvalds [Fri, 25 Jan 2019 22:57:09 +0000 (12:57 -1000)]
Merge tag 'usb-5.0-rc4' of git://git./linux/kernel/git/gregkh/usb

Pull USB/PHY fixes from Greg KH:
 "Here are a number of small USB and PHY driver fixes for 5.0-rc4.

  Nothing major at all, just the usual selection of USB gadget bugfixes,
  some new USB serial driver ids, some SPDX fixes, and some PHY driver
  fixes for reported issues.

  All of these have been in linux-next for a while with no reported
  issues"

* tag 'usb-5.0-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb:
  USB: serial: keyspan_usa: add proper SPDX lines for .h files
  USB: EHCI: ehci-mv: add MODULE_DEVICE_TABLE
  USB: leds: fix regression in usbport led trigger
  usb: chipidea: fix static checker warning for NULL pointer
  MAINTAINERS: email address update in MAINTAINERS entries
  USB: usbip: delete README file
  USB: serial: pl2303: add new PID to support PL2303TB
  usb: dwc2: gadget: Fix Remote Wakeup interrupt bit clearing
  phy: ath79-usb: Fix the main reset name to match the DT binding
  phy: ath79-usb: Fix the power on error path
  phy: fix build breakage: add PHY_MODE_SATA
  phy: ti: ensure priv is not null before dereferencing it
  USB: serial: ftdi_sio: fix GPIO not working in autosuspend
  usb: gadget: Potential NULL dereference on allocation error
  usb: dwc3: gadget: Fix the uninitialized link_state when udc starts
  usb: dwc3: gadget: Clear req->needs_extra_trb flag on cleanup
  usb: dwc3: gadget: synchronize_irq dwc irq in suspend
  USB: serial: simple: add Motorola Tetra TPG2200 device id

5 years agoMerge tag 'mac80211-for-davem-2019-01-25' of git://git.kernel.org/pub/scm/linux/kerne...
David S. Miller [Fri, 25 Jan 2019 18:59:36 +0000 (10:59 -0800)]
Merge tag 'mac80211-for-davem-2019-01-25' of git://git./linux/kernel/git/jberg/mac80211

Johannes Berg says:

====================
Just a few small fixes:
 * avoid trying to operate TDLS when not connection,
   this is not valid and led to issues
 * count TTL-dropped frames in mesh better
 * deal with new WiGig channels in regulatory code
 * remove a WARN_ON() that can trigger due to benign
   races during device/driver registration
 * fix nested netlink policy maxattrs (syzkaller)
 * fix hwsim n_limits (syzkaller)
 * propagate __aligned(2) to a surrounding struct
 * return proper error in virt_wifi error path
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoKVM: x86: Mark expected switch fall-throughs
Gustavo A. R. Silva [Fri, 25 Jan 2019 18:23:17 +0000 (12:23 -0600)]
KVM: x86: Mark expected switch fall-throughs

In preparation to enabling -Wimplicit-fallthrough, mark switch
cases where we are expecting to fall through.

This patch fixes the following warnings:

arch/x86/kvm/lapic.c:1037:27: warning: this statement may fall through [-Wimplicit-fallthrough=]
arch/x86/kvm/lapic.c:1876:3: warning: this statement may fall through [-Wimplicit-fallthrough=]
arch/x86/kvm/hyperv.c:1637:6: warning: this statement may fall through [-Wimplicit-fallthrough=]
arch/x86/kvm/svm.c:4396:6: warning: this statement may fall through [-Wimplicit-fallthrough=]
arch/x86/kvm/mmu.c:4372:36: warning: this statement may fall through [-Wimplicit-fallthrough=]
arch/x86/kvm/x86.c:3835:6: warning: this statement may fall through [-Wimplicit-fallthrough=]
arch/x86/kvm/x86.c:7938:23: warning: this statement may fall through [-Wimplicit-fallthrough=]
arch/x86/kvm/vmx/vmx.c:2015:6: warning: this statement may fall through [-Wimplicit-fallthrough=]
arch/x86/kvm/vmx/vmx.c:1773:6: warning: this statement may fall through [-Wimplicit-fallthrough=]

Warning level 3 was used: -Wimplicit-fallthrough=3

This patch is part of the ongoing efforts to enabling -Wimplicit-fallthrough.

Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: fix TRACE_INCLUDE_PATH and remove -I. header search paths
Masahiro Yamada [Fri, 25 Jan 2019 07:32:46 +0000 (16:32 +0900)]
KVM: x86: fix TRACE_INCLUDE_PATH and remove -I. header search paths

The header search path -I. in kernel Makefiles is very suspicious;
it allows the compiler to search for headers in the top of $(srctree),
where obviously no header file exists.

The reason of having -I. here is to make the incorrectly set
TRACE_INCLUDE_PATH working.

As the comment block in include/trace/define_trace.h says,
TRACE_INCLUDE_PATH should be a relative path to the define_trace.h

Fix the TRACE_INCLUDE_PATH, and remove the iffy include paths.

Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: selftests: check returned evmcs version range
Vitaly Kuznetsov [Thu, 17 Jan 2019 17:12:10 +0000 (18:12 +0100)]
KVM: selftests: check returned evmcs version range

Check that KVM_CAP_HYPERV_ENLIGHTENED_VMCS returns correct version range.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agox86/kvm/hyper-v: nested_enable_evmcs() sets vmcs_version incorrectly
Vitaly Kuznetsov [Thu, 17 Jan 2019 17:12:09 +0000 (18:12 +0100)]
x86/kvm/hyper-v: nested_enable_evmcs() sets vmcs_version incorrectly

Commit e2e871ab2f02 ("x86/kvm/hyper-v: Introduce nested_get_evmcs_version()
helper") broke EVMCS enablement: to set vmcs_version we now call
nested_get_evmcs_version() but this function checks
enlightened_vmcs_enabled flag which is not yet set so we end up returning
zero.

Fix the issue by re-arranging things in nested_enable_evmcs().

Fixes: e2e871ab2f02 ("x86/kvm/hyper-v: Introduce nested_get_evmcs_version() helper")
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: VMX: Move vmx_vcpu_run()'s VM-Enter asm blob to a helper function
Sean Christopherson [Wed, 16 Jan 2019 01:10:53 +0000 (17:10 -0800)]
KVM: VMX: Move vmx_vcpu_run()'s VM-Enter asm blob to a helper function

...along with the function's STACK_FRAME_NON_STANDARD tag.  Moving the
asm blob results in a significantly smaller amount of code that is
marked with STACK_FRAME_NON_STANDARD, which makes it far less likely
that gcc will split the function and trigger a spurious objtool warning.
As a bonus, removing STACK_FRAME_NON_STANDARD from vmx_vcpu_run() allows
the bulk of code to be properly checked by objtool.

Because %rbp is not loaded via VMCS fields, vmx_vcpu_run() must manually
save/restore the host's RBP and load the guest's RBP prior to calling
vmx_vmenter().  Modifying %rbp triggers objtool's stack validation code,
and so vmx_vcpu_run() is tagged with STACK_FRAME_NON_STANDARD since it's
impossible to avoid modifying %rbp.

Unfortunately, vmx_vcpu_run() is also a gigantic function that gcc will
split into separate functions, e.g. so that pieces of the function can
be inlined.  Splitting the function means that the compiled Elf file
will contain one or more vmx_vcpu_run.part.* functions in addition to
a vmx_vcpu_run function.  Depending on where the function is split,
objtool may warn about a "call without frame pointer save/setup" in
vmx_vcpu_run.part.* since objtool's stack validation looks for exact
names when whitelisting functions tagged with STACK_FRAME_NON_STANDARD.

Up until recently, the undesirable function splitting was effectively
blocked because vmx_vcpu_run() was tagged with __noclone.  At the time,
__noclone had an unintended side effect that put vmx_vcpu_run() into a
separate optimization unit, which in turn prevented gcc from inlining
the function (or any of its own function calls) and thus eliminated gcc's
motivation to split the function.  Removing the __noclone attribute
allowed gcc to optimize vmx_vcpu_run(), exposing the objtool warning.

Kudos to Qian Cai for root causing that the fnsplit optimization is what
caused objtool to complain.

Fixes: 453eafbe65f7 ("KVM: VMX: Move VM-Enter + VM-Exit handling to non-inline sub-routines")
Tested-by: Qian Cai <cai@lca.pw>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agokvm: selftests: Fix region overlap check in kvm_util
Ben Gardon [Wed, 16 Jan 2019 17:41:15 +0000 (09:41 -0800)]
kvm: selftests: Fix region overlap check in kvm_util

Fix a call to userspace_mem_region_find to conform to its spec of
taking an inclusive, inclusive range. It was previously being called
with an inclusive, exclusive range. Also remove a redundant region bounds
check in vm_userspace_mem_region_add. Region overlap checking is already
performed by the call to userspace_mem_region_find.

Tested: Compiled tools/testing/selftests/kvm with -static
Ran all resulting test binaries on an Intel Haswell test machine
All tests passed

Signed-off-by: Ben Gardon <bgardon@google.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agokvm: vmx: fix some -Wmissing-prototypes warnings
Yi Wang [Mon, 21 Jan 2019 07:27:05 +0000 (15:27 +0800)]
kvm: vmx: fix some -Wmissing-prototypes warnings

We get some warnings when building kernel with W=1:
arch/x86/kvm/vmx/vmx.c:426:5: warning: no previous prototype for ‘kvm_fill_hv_flush_list_func’ [-Wmissing-prototypes]
arch/x86/kvm/vmx/nested.c:58:6: warning: no previous prototype for ‘init_vmcs_shadow_fields’ [-Wmissing-prototypes]

Make them static to fix this.

Signed-off-by: Yi Wang <wang.yi59@zte.com.cn>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: nSVM: clear events pending from svm_complete_interrupts() when exiting to L1
Vitaly Kuznetsov [Mon, 7 Jan 2019 18:44:51 +0000 (19:44 +0100)]
KVM: nSVM: clear events pending from svm_complete_interrupts() when exiting to L1

kvm-unit-tests' eventinj "NMI failing on IDT" test results in NMI being
delivered to the host (L1) when it's running nested. The problem seems to
be: svm_complete_interrupts() raises 'nmi_injected' flag but later we
decide to reflect EXIT_NPF to L1. The flag remains pending and we do NMI
injection upon entry so it got delivered to L1 instead of L2.

It seems that VMX code solves the same issue in prepare_vmcs12(), this was
introduced with code refactoring in commit 5f3d5799974b ("KVM: nVMX: Rework
event injection and recovery").

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agosvm: Fix AVIC incomplete IPI emulation
Suravee Suthikulpanit [Tue, 22 Jan 2019 10:25:13 +0000 (10:25 +0000)]
svm: Fix AVIC incomplete IPI emulation

In case of incomplete IPI with invalid interrupt type, the current
SVM driver does not properly emulate the IPI, and fails to boot
FreeBSD guests with multiple vcpus when enabling AVIC.

Fix this by update APIC ICR high/low registers, which also
emulate sending the IPI.

Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agosvm: Add warning message for AVIC IPI invalid target
Suravee Suthikulpanit [Tue, 22 Jan 2019 10:24:19 +0000 (10:24 +0000)]
svm: Add warning message for AVIC IPI invalid target

Print warning message when IPI target ID is invalid due to one of
the following reasons:
  * In logical mode: cluster > max_cluster (64)
  * In physical mode: target > max_physical (512)
  * Address is not present in the physical or logical ID tables

Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: WARN_ONCE if sending a PV IPI returns a fatal error
Sean Christopherson [Wed, 23 Jan 2019 17:22:40 +0000 (09:22 -0800)]
KVM: x86: WARN_ONCE if sending a PV IPI returns a fatal error

KVM hypercalls return a negative value error code in case of a fatal
error, e.g. when the hypercall isn't supported or was made with invalid
parameters.  WARN_ONCE on fatal errors when sending PV IPIs as any such
error all but guarantees an SMP system will hang due to a missing IPI.

Fixes: aaffcfd1e82d ("KVM: X86: Implement PV IPIs in linux guest")
Cc: stable@vger.kernel.org
Cc: Wanpeng Li <wanpengli@tencent.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: Fix PV IPIs for 32-bit KVM host
Sean Christopherson [Wed, 23 Jan 2019 17:22:39 +0000 (09:22 -0800)]
KVM: x86: Fix PV IPIs for 32-bit KVM host

The recognition of the KVM_HC_SEND_IPI hypercall was unintentionally
wrapped in "#ifdef CONFIG_X86_64", causing 32-bit KVM hosts to reject
any and all PV IPI requests despite advertising the feature.  This
results in all KVM paravirtualized guests hanging during SMP boot due
to IPIs never being delivered.

Fixes: 4180bf1b655a ("KVM: X86: Implement "send IPI" hypercall")
Cc: stable@vger.kernel.org
Cc: Wanpeng Li <wanpengli@tencent.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agox86/kvm/hyper-v: recommend using eVMCS only when it is enabled
Vitaly Kuznetsov [Fri, 25 Jan 2019 11:19:34 +0000 (12:19 +0100)]
x86/kvm/hyper-v: recommend using eVMCS only when it is enabled

We shouldn't probably be suggesting using Enlightened VMCS when it's not
enabled (not supported from guest's point of view). Hyper-V on KVM seems
to be fine either way but let's be consistent.

Fixes: 2bc39970e932 ("x86/kvm/hyper-v: Introduce KVM_GET_SUPPORTED_HV_CPUID")
Reviewed-by: Liran Alon <liran.alon@oracle.com>
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agox86/kvm/hyper-v: don't recommend doing reset via synthetic MSR
Vitaly Kuznetsov [Fri, 25 Jan 2019 11:19:33 +0000 (12:19 +0100)]
x86/kvm/hyper-v: don't recommend doing reset via synthetic MSR

System reset through synthetic MSR is not recommended neither by genuine
Hyper-V nor my QEMU.

Fixes: 2bc39970e932 ("x86/kvm/hyper-v: Introduce KVM_GET_SUPPORTED_HV_CPUID")
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Liran Alon <liran.alon@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agokvm: x86/vmx: Use kzalloc for cached_vmcs12
Tom Roeder [Thu, 24 Jan 2019 21:48:20 +0000 (13:48 -0800)]
kvm: x86/vmx: Use kzalloc for cached_vmcs12

This changes the allocation of cached_vmcs12 to use kzalloc instead of
kmalloc. This removes the information leak found by Syzkaller (see
Reported-by) in this case and prevents similar leaks from happening
based on cached_vmcs12.

It also changes vmx_get_nested_state to copy out the full 4k VMCS12_SIZE
in copy_to_user rather than only the size of the struct.

Tested: rebuilt against head, booted, and ran the syszkaller repro
  https://syzkaller.appspot.com/text?tag=ReproC&x=174efca3400000 without
  observing any problems.

Reported-by: syzbot+ded1696f6b50b615b630@syzkaller.appspotmail.com
Fixes: 8fcc4b5923af5de58b80b53a069453b135693304
Cc: stable@vger.kernel.org
Signed-off-by: Tom Roeder <tmroeder@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: VMX: Use the correct field var when clearing VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL
Sean Christopherson [Mon, 14 Jan 2019 20:12:02 +0000 (12:12 -0800)]
KVM: VMX: Use the correct field var when clearing VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL

Fix a recently introduced bug that results in the wrong VMCS control
field being updated when applying a IA32_PERF_GLOBAL_CTRL errata.

Fixes: c73da3fcab43 ("KVM: VMX: Properly handle dynamic VM Entry/Exit controls")
Reported-by: Harald Arnesen <harald@skogtun.org>
Tested-by: Harald Arnesen <harald@skogtun.org>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: Fix single-step debugging
Alexander Popov [Mon, 21 Jan 2019 12:48:40 +0000 (15:48 +0300)]
KVM: x86: Fix single-step debugging

The single-step debugging of KVM guests on x86 is broken: if we run
gdb 'stepi' command at the breakpoint when the guest interrupts are
enabled, RIP always jumps to native_apic_mem_write(). Then other
nasty effects follow.

Long investigation showed that on Jun 7, 2017 the
commit c8401dda2f0a00cd25c0 ("KVM: x86: fix singlestepping over syscall")
introduced the kvm_run.debug corruption: kvm_vcpu_do_singlestep() can
be called without X86_EFLAGS_TF set.

Let's fix it. Please consider that for -stable.

Signed-off-by: Alexander Popov <alex.popov@linux.com>
Cc: stable@vger.kernel.org
Fixes: c8401dda2f0a00cd25c0 ("KVM: x86: fix singlestepping over syscall")
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agox86/kvm/hyper-v: don't announce GUEST IDLE MSR support
Vitaly Kuznetsov [Thu, 24 Jan 2019 14:27:09 +0000 (15:27 +0100)]
x86/kvm/hyper-v: don't announce GUEST IDLE MSR support

HV_X64_MSR_GUEST_IDLE_AVAILABLE appeared in kvm_vcpu_ioctl_get_hv_cpuid()
by mistake: it announces support for HV_X64_MSR_GUEST_IDLE (0x400000F0)
which we don't support in KVM (yet).

Fixes: 2bc39970e932 ("x86/kvm/hyper-v: Introduce KVM_GET_SUPPORTED_HV_CPUID")
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agocfg80211: extend range deviation for DMG
Chaitanya Tata [Fri, 18 Jan 2019 21:47:47 +0000 (03:17 +0530)]
cfg80211: extend range deviation for DMG

Recently, DMG frequency bands have been extended till 71GHz, so extend
the range check till 20GHz (45-71GHZ), else some channels will be marked
as disabled.

Signed-off-by: Chaitanya Tata <Chaitanya.Tata@bluwireless.co.uk>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
5 years agocfg80211: reg: remove warn_on for a normal case
Chaitanya Tata [Thu, 24 Jan 2019 10:43:02 +0000 (16:13 +0530)]
cfg80211: reg: remove warn_on for a normal case

If there are simulatenous queries of regdb, then there might be a case
where multiple queries can trigger request_firmware_no_wait and can have
parallel callbacks being executed asynchronously. In this scenario we
might hit the WARN_ON.

So remove the warn_on, as the code already handles multiple callbacks
gracefully.

Signed-off-by: Chaitanya Tata <chaitanya.tata@bluwireless.co.uk>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
5 years agomac80211: Add attribute aligned(2) to struct 'action'
Mathieu Malaterre [Thu, 24 Jan 2019 18:19:57 +0000 (19:19 +0100)]
mac80211: Add attribute aligned(2) to struct 'action'

During refactor in commit 9e478066eae4 ("mac80211: fix MU-MIMO
follow-MAC mode") a new struct 'action' was declared with packed
attribute as:

  struct {
          struct ieee80211_hdr_3addr hdr;
          u8 category;
          u8 action_code;
  } __packed action;

But since struct 'ieee80211_hdr_3addr' is declared with an aligned
keyword as:

  struct ieee80211_hdr {
   __le16 frame_control;
   __le16 duration_id;
   u8 addr1[ETH_ALEN];
   u8 addr2[ETH_ALEN];
   u8 addr3[ETH_ALEN];
   __le16 seq_ctrl;
   u8 addr4[ETH_ALEN];
  } __packed __aligned(2);

Solve the ambiguity of placing aligned structure in a packed one by
adding the aligned(2) attribute to struct 'action'.

This removes the following warning (W=1):

  net/mac80211/rx.c:234:2: warning: alignment 1 of 'struct <anonymous>' is less than 2 [-Wpacked-not-aligned]

Cc: Johannes Berg <johannes.berg@intel.com>
Suggested-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
5 years agomac80211: don't initiate TDLS connection if station is not associated to AP
Balaji Pothunoori [Mon, 21 Jan 2019 07:00:43 +0000 (12:30 +0530)]
mac80211: don't initiate TDLS connection if station is not associated to AP

Following call trace is observed while adding TDLS peer entry in driver
during TDLS setup.

Call Trace:
[<c1301476>] dump_stack+0x47/0x61
[<c10537d2>] __warn+0xe2/0x100
[<fa22415f>] ? sta_apply_parameters+0x49f/0x550 [mac80211]
[<c1053895>] warn_slowpath_null+0x25/0x30
[<fa22415f>] sta_apply_parameters+0x49f/0x550 [mac80211]
[<fa20ad42>] ? sta_info_alloc+0x1c2/0x450 [mac80211]
[<fa224623>] ieee80211_add_station+0xe3/0x160 [mac80211]
[<c1876fe3>] nl80211_new_station+0x273/0x420
[<c170f6d9>] genl_rcv_msg+0x219/0x3c0
[<c170f4c0>] ? genl_rcv+0x30/0x30
[<c170ee7e>] netlink_rcv_skb+0x8e/0xb0
[<c170f4ac>] genl_rcv+0x1c/0x30
[<c170e8aa>] netlink_unicast+0x13a/0x1d0
[<c170ec18>] netlink_sendmsg+0x2d8/0x390
[<c16c5acd>] sock_sendmsg+0x2d/0x40
[<c16c6369>] ___sys_sendmsg+0x1d9/0x1e0

Fixing this by allowing TDLS setup request only when we have completed
association.

Signed-off-by: Balaji Pothunoori <bpothuno@codeaurora.org>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
5 years agonl80211: fix NLA_POLICY_NESTED() arguments
Johannes Berg [Fri, 25 Jan 2019 08:26:32 +0000 (09:26 +0100)]
nl80211: fix NLA_POLICY_NESTED() arguments

syzbot reported an out-of-bounds read when passing certain
malformed messages into nl80211. The specific place where
this happened isn't interesting, the problem is that nested
policy parsing was referring to the wrong maximum attribute
and thus the policy wasn't long enough.

Fix this by referring to the correct attribute. Since this
is really not necessary, I'll come up with a separate patch
to just pass the policy instead of both, in the common case
we can infer the maxattr from the size of the policy array.

Reported-by: syzbot+4157b036c5f4713b1f2f@syzkaller.appspotmail.com
Cc: stable@vger.kernel.org
Fixes: 9bb7e0f24e7e ("cfg80211: add peer measurement with FTM initiator API")
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
5 years agoibmveth: Do not process frames after calling napi_reschedule
Thomas Falcon [Thu, 24 Jan 2019 17:17:01 +0000 (11:17 -0600)]
ibmveth: Do not process frames after calling napi_reschedule

The IBM virtual ethernet driver's polling function continues
to process frames after rescheduling NAPI, resulting in a warning
if it exhausted its budget. Do not restart polling after calling
napi_reschedule. Instead let frames be processed in the following
instance.

Signed-off-by: Thomas Falcon <tlfalcon@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: dev_is_mac_header_xmit() true for ARPHRD_RAWIP
Maciej Żenczykowski [Thu, 24 Jan 2019 11:07:02 +0000 (03:07 -0800)]
net: dev_is_mac_header_xmit() true for ARPHRD_RAWIP

__bpf_redirect() and act_mirred checks this boolean
to determine whether to prefix an ethernet header.

Signed-off-by: Maciej Żenczykowski <maze@google.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: usb: asix: ax88772_bind return error when hw_reset fail
Zhang Run [Thu, 24 Jan 2019 05:48:49 +0000 (13:48 +0800)]
net: usb: asix: ax88772_bind return error when hw_reset fail

The ax88772_bind() should return error code immediately when the PHY
was not reset properly through ax88772a_hw_reset().
Otherwise, The asix_get_phyid() will block when get the PHY
Identifier from the PHYSID1 MII registers through asix_mdio_read()
due to the PHY isn't ready. Furthermore, it will produce a lot of
error message cause system crash.As follows:
asix 1-1:1.0 (unnamed net_device) (uninitialized): Failed to write
 reg index 0x0000: -71
asix 1-1:1.0 (unnamed net_device) (uninitialized): Failed to send
 software reset: ffffffb9
asix 1-1:1.0 (unnamed net_device) (uninitialized): Failed to write
 reg index 0x0000: -71
asix 1-1:1.0 (unnamed net_device) (uninitialized): Failed to enable
 software MII access
asix 1-1:1.0 (unnamed net_device) (uninitialized): Failed to read
 reg index 0x0000: -71
asix 1-1:1.0 (unnamed net_device) (uninitialized): Failed to write
 reg index 0x0000: -71
asix 1-1:1.0 (unnamed net_device) (uninitialized): Failed to enable
 software MII access
asix 1-1:1.0 (unnamed net_device) (uninitialized): Failed to read
 reg index 0x0000: -71
...

Signed-off-by: Zhang Run <zhang.run@zte.com.cn>
Reviewed-by: Yang Wei <yang.wei9@zte.com.cn>
Tested-by: Marcel Ziswiler <marcel.ziswiler@toradex.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMAINTAINERS: Update cavium networking drivers
Sudarsana Reddy Kalluru [Thu, 24 Jan 2019 02:03:20 +0000 (18:03 -0800)]
MAINTAINERS: Update cavium networking drivers

Following Marvell's acquisition of Cavium, we need to update all the
Cavium drivers maintainer's entries to point to our new e-mail addresses.

Signed-off-by: Sudarsana Reddy Kalluru <Sudarsana.Kalluru@cavium.com>
Signed-off-by: Ameen Rahman <Ameen.Rahman@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge tag 'hyperv-fixes-signed' of git://git.kernel.org/pub/scm/linux/kernel/git...
David S. Miller [Fri, 25 Jan 2019 06:22:17 +0000 (22:22 -0800)]
Merge tag 'hyperv-fixes-signed' of git://git./linux/kernel/git/hyperv/linux

Sasha Levin says:

====================
Hyper-V hv_netvsc commits for 5.0

Three patches from Haiyang Zhang to fix settings hash key using ethtool,
and Adrian Vladu's first patch fixing a few spelling mistakes.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge tag 'linux-can-fixes-for-5.0-20190122' of git://git.kernel.org/pub/scm/linux...
David S. Miller [Fri, 25 Jan 2019 05:52:37 +0000 (21:52 -0800)]
Merge tag 'linux-can-fixes-for-5.0-20190122' of git://git./linux/kernel/git/mkl/linux-can

Marc Kleine-Budde says:

====================
pull-request: can 2019-01-22

this is a pull request of 4 patches for net/master.

The first patch by is by Manfred Schlaegl and reverts a patch that caused wrong
warning messages in certain use cases. The next patch is by Oliver Hartkopp for
the bcm that adds sanity checks for the timer value before using it to detect
potential interger overflows. The last two patches are for the flexcan driver,
YueHaibing's patch fixes the the return value in the error path of the
flexcan_setup_stop_mode() function. The second patch is by Uwe Kleine-König and
fixes a NULL pointer deref on older flexcan cores in flexcan_chip_start().
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'mlx4_core-fixes'
David S. Miller [Fri, 25 Jan 2019 05:48:26 +0000 (21:48 -0800)]
Merge branch 'mlx4_core-fixes'

Tariq Toukan says:

====================
mlx4_core fixes for 5.0-rc

This patchset includes two fixes for the mlx4_core driver.

First patch by Aya fixes inaccurate parsing of some FW fields, mistakenly
including additional (mostly reserved) bits.

Second patch by Jack fixes a wrong (yet harmless) error handling of
calls to copy_to_user() during the CQs init stage.

Series generated against net commit:
49a57857aeea Linux 5.0-rc3
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/mlx4_core: Fix error handling when initializing CQ bufs in the driver
Jack Morgenstein [Tue, 22 Jan 2019 13:19:45 +0000 (15:19 +0200)]
net/mlx4_core: Fix error handling when initializing CQ bufs in the driver

Procedure mlx4_init_user_cqes() handles returns by copy_to_user
incorrectly. copy_to_user() returns the number of bytes not copied.
Thus, a non-zero return should be treated as a -EFAULT error
(as is done elsewhere in the kernel). However, mlx4_init_user_cqes()
error handling simply returns the number of bytes not copied
(instead of -EFAULT).

Note, though, that this is a harmless bug: procedure mlx4_alloc_cq()
(which is the only caller of mlx4_init_user_cqes()) treats any
non-zero return as an error, but that returned error value is processed
internally, and not passed further up the call stack.

In addition, fixes the following sparse warning:
warning: incorrect type in argument 1 (different address spaces)
   expected void [noderef] <asn:1>*to
   got void *buf

Fixes: e45678973dcb ("{net, IB}/mlx4: Initialize CQ buffers in the driver when possible")
Reported by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/mlx4_core: Add masking for a few queries on HCA caps
Aya Levin [Tue, 22 Jan 2019 13:19:44 +0000 (15:19 +0200)]
net/mlx4_core: Add masking for a few queries on HCA caps

Driver reads the query HCA capabilities without the corresponding masks.
Without the correct masks, the base addresses of the queues are
unaligned.  In addition some reserved bits were wrongly read.  Using the
correct masks, ensures alignment of the base addresses and allows future
firmware versions safe use of the reserved bits.

Fixes: ab9c17a009ee ("mlx4_core: Modify driver initialization flow to accommodate SRIOV for Ethernet")
Fixes: 0ff1fb654bec ("{NET, IB}/mlx4: Add device managed flow steering firmware API")
Signed-off-by: Aya Levin <ayal@mellanox.com>
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosctp: set flow sport from saddr only when it's 0
Xin Long [Mon, 21 Jan 2019 18:42:41 +0000 (02:42 +0800)]
sctp: set flow sport from saddr only when it's 0

Now sctp_transport_pmtu() passes transport->saddr into .get_dst() to set
flow sport from 'saddr'. However, transport->saddr is set only when
transport->dst exists in sctp_transport_route().

If sctp_transport_pmtu() is called without transport->saddr set, like
when transport->dst doesn't exists, the flow sport will be set to 0
from transport->saddr, which will cause a wrong route to be got.

Commit 6e91b578bf3f ("sctp: re-use sctp_transport_pmtu in
sctp_transport_route") made the issue be triggered more easily
since sctp_transport_pmtu() would be called in sctp_transport_route()
after that.

In gerneral, fl4->fl4_sport should always be set to
htons(asoc->base.bind_addr.port), unless transport->asoc doesn't exist
in sctp_v4/6_get_dst(), which is the case:

  sctp_ootb_pkt_new() ->
    sctp_transport_route()

For that, we can simply handle it by setting flow sport from saddr only
when it's 0 in sctp_v4/6_get_dst().

Fixes: 6e91b578bf3f ("sctp: re-use sctp_transport_pmtu in sctp_transport_route")
Reported-by: Ying Xu <yinxu@redhat.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosctp: set chunk transport correctly when it's a new asoc
Xin Long [Mon, 21 Jan 2019 18:42:09 +0000 (02:42 +0800)]
sctp: set chunk transport correctly when it's a new asoc

In the paths:

  sctp_sf_do_unexpected_init() ->
    sctp_make_init_ack()
  sctp_sf_do_dupcook_a/b()() ->
    sctp_sf_do_5_1D_ce()

The new chunk 'retval' transport is set from the incoming chunk 'chunk'
transport. However, 'retval' transport belong to the new asoc, which
is a different one from 'chunk' transport's asoc.

It will cause that the 'retval' chunk gets set with a wrong transport.
Later when sending it and because of Commit b9fd683982c9 ("sctp: add
sctp_packet_singleton"), sctp_packet_singleton() will set some fields,
like vtag to 'retval' chunk from that wrong transport's asoc.

This patch is to fix it by setting 'retval' transport correctly which
belongs to the right asoc in sctp_make_init_ack() and
sctp_sf_do_5_1D_ce().

Fixes: b9fd683982c9 ("sctp: add sctp_packet_singleton")
Reported-by: Ying Xu <yinxu@redhat.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosctp: improve the events for sctp stream adding
Xin Long [Mon, 21 Jan 2019 18:40:12 +0000 (02:40 +0800)]
sctp: improve the events for sctp stream adding

This patch is to improve sctp stream adding events in 2 places:

  1. In sctp_process_strreset_addstrm_out(), move up SCTP_MAX_STREAM
     and in stream allocation failure checks, as the adding has to
     succeed after reconf_timer stops for the in stream adding
     request retransmission.

  3. In sctp_process_strreset_addstrm_in(), no event should be sent,
     as no in or out stream is added here.

Fixes: 50a41591f110 ("sctp: implement receiver-side procedures for the Add Outgoing Streams Request Parameter")
Fixes: c5c4ebb3ab87 ("sctp: implement receiver-side procedures for the Add Incoming Streams Request Parameter")
Reported-by: Ying Xu <yinxu@redhat.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosctp: improve the events for sctp stream reset
Xin Long [Mon, 21 Jan 2019 18:39:34 +0000 (02:39 +0800)]
sctp: improve the events for sctp stream reset

This patch is to improve sctp stream reset events in 4 places:

  1. In sctp_process_strreset_outreq(), the flag should always be set with
     SCTP_STREAM_RESET_INCOMING_SSN instead of OUTGOING, as receiver's in
     stream is reset here.
  2. In sctp_process_strreset_outreq(), move up SCTP_STRRESET_ERR_WRONG_SSN
     check, as the reset has to succeed after reconf_timer stops for the
     in stream reset request retransmission.
  3. In sctp_process_strreset_inreq(), no event should be sent, as no in
     or out stream is reset here.
  4. In sctp_process_strreset_resp(), SCTP_STREAM_RESET_INCOMING_SSN or
     OUTGOING event should always be sent for stream reset requests, no
     matter it fails or succeeds to process the request.

Fixes: 810544764536 ("sctp: implement receiver-side procedures for the Outgoing SSN Reset Request Parameter")
Fixes: 16e1a91965b0 ("sctp: implement receiver-side procedures for the Incoming SSN Reset Request Parameter")
Fixes: 11ae76e67a17 ("sctp: implement receiver-side procedures for the Reconf Response Parameter")
Reported-by: Ying Xu <yinxu@redhat.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoip_tunnel: Make none-tunnel-dst tunnel port work with lwtunnel
wenxu [Sat, 19 Jan 2019 05:11:25 +0000 (13:11 +0800)]
ip_tunnel: Make none-tunnel-dst tunnel port work with lwtunnel

ip l add dev tun type gretap key 1000
ip a a dev tun 10.0.0.1/24

Packets with tun-id 1000 can be recived by tun dev. But packet can't
be sent through dev tun for non-tunnel-dst

With this patch: tunnel-dst can be get through lwtunnel like beflow:
ip r a 10.0.0.7 encap ip dst 172.168.0.11 dev tun

Signed-off-by: wenxu <wenxu@ucloud.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge tag 'drm-fixes-2019-01-25-1' of git://anongit.freedesktop.org/drm/drm
Linus Torvalds [Thu, 24 Jan 2019 23:19:10 +0000 (12:19 +1300)]
Merge tag 'drm-fixes-2019-01-25-1' of git://anongit.freedesktop.org/drm/drm

Pull drm fixes from Dave Airlie:
 "Live from LCA pull, some fixes all over the place,

  i915:
   - GVT workload destruction fix

  msm:
   - A6XX opp-level fix
   - build fixes
   - hard-coded irq removal

  amdgpu:
   - overclocking fix
   - hybrid gfx fix

  sun4i:
   - fix TMDS clock usage"

* tag 'drm-fixes-2019-01-25-1' of git://anongit.freedesktop.org/drm/drm:
  drm/msm: avoid unused function warning
  drm/msm: Add __printf verification
  drm/msm: Fix A6XX support for opp-level
  drm/msm: honor GPU_READONLY flag
  drm/msm: drop interrupt-names
  drm/msm/gpu: Remove hardcoded interrupt name
  drm/msm/gpu: fix building without debugfs
  drm/i915/execlists: Mark up priority boost on preemption
  drm/i915/gvt: release shadow batch buffer and wa_ctx before destroy one workload
  drm/sun4i: hdmi: Fix usage of TMDS clock
  drm/amd/powerplay: OD setting fix on Vega10
  drm/amdgpu: Add APTX quirk for Lenovo laptop
  drm/msm: Unblock writer if reader closes file

5 years agoMerge tag 'drm-msm-fixes-2019-01-24' of git://people.freedesktop.org/~robclark/linux...
Dave Airlie [Thu, 24 Jan 2019 21:44:53 +0000 (07:44 +1000)]
Merge tag 'drm-msm-fixes-2019-01-24' of git://people.freedesktop.org/~robclark/linux into drm-fixes

A few fixes for v5.0.. the opp-level fix and removal of hard-coded irq
name is partially to make things smoother in v5.1 merge window to
avoid dependency on drm vs dt trees, but are otherwise sane changes.

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Rob Clark <robdclark@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/CAF6AEGsAEHd2tGRQxRTs+A-8y_tthPs2iUgCCCEwR5vDMXab4A@mail.gmail.com
5 years agoMerge tag 'drm-intel-fixes-2019-01-24' of git://anongit.freedesktop.org/drm/drm-intel...
Dave Airlie [Thu, 24 Jan 2019 20:57:45 +0000 (06:57 +1000)]
Merge tag 'drm-intel-fixes-2019-01-24' of git://anongit.freedesktop.org/drm/drm-intel into drm-fixes

drm/i915 fixes for v5.0-rc4:
- fix priority boost
- gvt: fix destroy of shadow batch and indirect ctx

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/87k1iu1a2e.fsf@intel.com
5 years agoMerge branch 'drm-fixes-5.0' of git://people.freedesktop.org/~agd5f/linux into drm...
Dave Airlie [Thu, 24 Jan 2019 20:53:16 +0000 (06:53 +1000)]
Merge branch 'drm-fixes-5.0' of git://people.freedesktop.org/~agd5f/linux into drm-fixes

- Overclock fix for vega10
- Hybrid gfx laptop fix

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexdeucher@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190123231004.3111-1-alexander.deucher@amd.com
5 years agocifs: print CIFSMaxBufSize as part of /proc/fs/cifs/DebugData
Ronnie Sahlberg [Thu, 24 Jan 2019 06:19:31 +0000 (16:19 +1000)]
cifs: print CIFSMaxBufSize as part of /proc/fs/cifs/DebugData

Was helpful in debug for some recent problems.

Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
5 years agosmb3: add credits we receive from oplock/break PDUs
Ronnie Sahlberg [Wed, 23 Jan 2019 06:20:38 +0000 (16:20 +1000)]
smb3: add credits we receive from oplock/break PDUs

Otherwise we gradually leak credits leading to potential
hung session.

Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
CC: Stable <stable@vger.kernel.org>
Reviewed-by: Pavel Shilovsky <pshilov@microsoft.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
5 years agoCIFS: Fix mounts if the client is low on credits
Pavel Shilovsky [Wed, 16 Jan 2019 19:48:42 +0000 (11:48 -0800)]
CIFS: Fix mounts if the client is low on credits

If the server doesn't grant us at least 3 credits during the mount
we won't be able to complete it because query path info operation
requires 3 credits. Use the cached file handle if possible to allow
the mount to succeed.

Signed-off-by: Pavel Shilovsky <pshilov@microsoft.com>
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
5 years agoCIFS: Do not assume one credit for async responses
Pavel Shilovsky [Tue, 15 Jan 2019 23:08:48 +0000 (15:08 -0800)]
CIFS: Do not assume one credit for async responses

If we don't receive a response we can't assume that the server
granted one credit. Assume zero credits in such cases.

Signed-off-by: Pavel Shilovsky <pshilov@microsoft.com>
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
5 years agoCIFS: Fix credit calculations in compound mid callback
Pavel Shilovsky [Wed, 23 Jan 2019 00:50:21 +0000 (16:50 -0800)]
CIFS: Fix credit calculations in compound mid callback

The current code doesn't do proper accounting for credits
in SMB1 case: it adds one credit per response only if we get
a complete response while it needs to return it unconditionally.
Fix this and also include malformed responses for SMB2+ into
accounting for credits because such responses have Credit
Granted field, thus nothing prevents to get a proper credit
value from them.

Signed-off-by: Pavel Shilovsky <pshilov@microsoft.com>
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
5 years agoCIFS: Fix credit calculation for encrypted reads with errors
Pavel Shilovsky [Fri, 18 Jan 2019 23:38:11 +0000 (15:38 -0800)]
CIFS: Fix credit calculation for encrypted reads with errors

We do need to account for credits received in error responses
to read requests on encrypted sessions.

Cc: <stable@vger.kernel.org>
Signed-off-by: Pavel Shilovsky <pshilov@microsoft.com>
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
5 years agoCIFS: Fix credits calculations for reads with errors
Pavel Shilovsky [Thu, 17 Jan 2019 23:29:26 +0000 (15:29 -0800)]
CIFS: Fix credits calculations for reads with errors

Currently we mark MID as malformed if we get an error from server
in a read response. This leads to not properly processing credits
in the readv callback. Fix this by marking such a response as
normal received response and process it appropriately.

Cc: <stable@vger.kernel.org>
Signed-off-by: Pavel Shilovsky <pshilov@microsoft.com>
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
5 years agoCIFS: Do not reconnect TCP session in add_credits()
Pavel Shilovsky [Sat, 19 Jan 2019 01:25:36 +0000 (17:25 -0800)]
CIFS: Do not reconnect TCP session in add_credits()

When executing add_credits() we currently call cifs_reconnect()
if the number of credits is zero and there are no requests in
flight. In this case we may call cifs_reconnect() recursively
twice and cause memory corruption given the following sequence
of functions:

mid1.callback() -> add_credits() -> cifs_reconnect() ->
-> mid2.callback() -> add_credits() -> cifs_reconnect().

Fix this by avoiding to call cifs_reconnect() in add_credits()
and checking for zero credits in the demultiplex thread.

Cc: <stable@vger.kernel.org>
Signed-off-by: Pavel Shilovsky <pshilov@microsoft.com>
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
5 years agoMerge tag 'drm-misc-fixes-2019-01-24' of git://anongit.freedesktop.org/drm/drm-misc...
Dave Airlie [Thu, 24 Jan 2019 20:46:51 +0000 (06:46 +1000)]
Merge tag 'drm-misc-fixes-2019-01-24' of git://anongit.freedesktop.org/drm/drm-misc into drm-fixes

drm-misc-fixes for v5.0-rc4:
- Small refcounting fix to sun4i's HDMI support.

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/588e9ecb-d80d-2cc6-254e-e5311f04224f@linux.intel.com
5 years agodrm/msm: avoid unused function warning
Arnd Bergmann [Thu, 10 Jan 2019 14:14:03 +0000 (15:14 +0100)]
drm/msm: avoid unused function warning

drivers/gpu/drm/msm/disp/dpu1/dpu_plane.c:368:13: error: 'dpu_plane_danger_signal_ctrl' defined but not used [-Werror=unused-function]

Fixes: 7b2e7adea732 ("drm/msm/dpu: Make dpu_plane_danger_signal_ctrl void")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Anders Roxell <anders.roxell@linaro.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
5 years agodrm/msm: Add __printf verification
Joe Perches [Thu, 17 Jan 2019 22:17:36 +0000 (14:17 -0800)]
drm/msm: Add __printf verification

Add a few __printf attribute specifiers to routines that
could use them.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Rob Clark <robdclark@gmail.com>
5 years agodrm/msm: Fix A6XX support for opp-level
Douglas Anderson [Wed, 16 Jan 2019 18:46:21 +0000 (10:46 -0800)]
drm/msm: Fix A6XX support for opp-level

The bindings for Qualcomm opp levels changed after being Acked but
before landing.  Thus the code in the GPU driver that was relying on
the old bindings is now broken.

Let's change the code to match the new bindings by adjusting the old
string 'qcom,level' to the new string 'opp-level'.  See the patch
("dt-bindings: opp: Introduce opp-level bindings").

NOTE: we will do additional cleanup to totally remove the string from
the code and use the new dev_pm_opp_get_level() but we'll do it in a
future patch.  This will facilitate getting the important code fix in
sooner without having to deal with cross-maintainer dependencies.

This patch needs to land before the patch ("arm64: dts: sdm845: Add
gpu and gmu device nodes") since if a tree contains the device tree
patch but not this one you'll get a crash at bootup.

Fixes: 4b565ca5a2cb ("drm/msm: Add A6XX device support")
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Reviewed-by: Jordan Crouse <jcrouse@codeaurora.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
5 years agodrm/msm: honor GPU_READONLY flag
Rob Clark [Wed, 9 Jan 2019 19:25:05 +0000 (14:25 -0500)]
drm/msm: honor GPU_READONLY flag

Signed-off-by: Rob Clark <robdclark@gmail.com>
5 years agodrm/msm: drop interrupt-names
Jordan Crouse [Tue, 18 Dec 2018 18:32:37 +0000 (11:32 -0700)]
drm/msm: drop interrupt-names

Each GPU core only uses one interrupt so we don't to look up
an interrupt by name and thereby we don't need interrupt-names.

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
5 years agodrm/msm/gpu: Remove hardcoded interrupt name
Jordan Crouse [Tue, 18 Dec 2018 18:32:36 +0000 (11:32 -0700)]
drm/msm/gpu: Remove hardcoded interrupt name

Every GPU core only has one interrupt so there isn't any
value in looking up the interrupt by name. Remove the name (which
is legacy anyway) and use platform_get_irq() instead.

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
Reviewed-by: Douglas Anderson <dianders@chromium.org>
Signed-off-by: Rob Clark <robdclark@gmail.com>
5 years agodrm/msm/gpu: fix building without debugfs
Arnd Bergmann [Mon, 13 Aug 2018 21:23:44 +0000 (23:23 +0200)]
drm/msm/gpu: fix building without debugfs

When debugfs is disabled, but coredump is turned on, the adreno driver fails to build:

drivers/gpu/drm/msm/adreno/a3xx_gpu.c:460:4: error: 'struct msm_gpu_funcs' has no member named 'show'
   .show = adreno_show,
    ^~~~
drivers/gpu/drm/msm/adreno/a3xx_gpu.c:460:11: note: (near initialization for 'funcs.base')
drivers/gpu/drm/msm/adreno/a3xx_gpu.c:460:11: error: initialization of 'void (*)(struct msm_gpu *, struct msm_gem_submit *, struct msm_file_private *)' from incompatible pointer type 'void (*)(struct msm_gpu *, struct msm_gpu_state *, struct drm_printer *)' [-Werror=incompatible-pointer-types]
drivers/gpu/drm/msm/adreno/a3xx_gpu.c:460:11: note: (near initialization for 'funcs.base.submit')
drivers/gpu/drm/msm/adreno/a4xx_gpu.c:546:4: error: 'struct msm_gpu_funcs' has no member named 'show'
drivers/gpu/drm/msm/adreno/a5xx_gpu.c:1460:4: error: 'struct msm_gpu_funcs' has no member named 'show'
drivers/gpu/drm/msm/adreno/a6xx_gpu.c:769:4: error: 'struct msm_gpu_funcs' has no member named 'show'
drivers/gpu/drm/msm/msm_gpu.c: In function 'msm_gpu_devcoredump_read':
drivers/gpu/drm/msm/msm_gpu.c:289:12: error: 'const struct msm_gpu_funcs' has no member named 'show'

Adjust the #ifdef to make it build again.

Fixes: c0fec7f562ec ("drm/msm/gpu: Capture the GPU state on a GPU hang")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Rob Clark <robdclark@gmail.com>