linux-stable.git/arch/mips/kernel/smp.c, branch linux-4.8.y

MIPS: SMP: Fix possibility of deadlock when bringing CPUs online

2016-09-24T23:43:52+00:00

This patch fixes the possibility of a deadlock when bringing up
secondary CPUs.
The deadlock occurs because the set_cpu_online() is called before
synchronise_count_slave(). This can cause a deadlock if the boot CPU,
having scheduled another thread, attempts to send an IPI to the
secondary CPU, which it sees has been marked online. The secondary is
blocked in synchronise_count_slave() waiting for the boot CPU to enter
synchronise_count_master(), but the boot cpu is blocked in
smp_call_function_many() waiting for the secondary to respond to it's
IPI request.

Fix this by marking the CPU online in cpu_callin_map and synchronising
counters before declaring the CPU online and calculating the maps for
IPIs.

Signed-off-by: Matt Redfearn 
Reported-by: Justin Chen 
Tested-by: Justin Chen 
Cc: Florian Fainelli 
Cc: stable@vger.kernel.org # v4.1+
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/14302/
Signed-off-by: Ralf Baechle

MIPS: c-r4k: Exclude sibling CPUs in SMP calls

2016-07-29T08:19:29+00:00

When performing SMP calls to foreign cores, exclude sibling CPUs from
the provided map, as we already handle the local core on the current
CPU. This prevents an SMP call from for example core 0, VPE 1 to VPE 0
on the same core.

In the process the cpu_foreign_map cpumask is turned into an array of
cpumasks, so that each CPU has its own version of it which excludes
sibling CPUs. r4k_op_needs_ipi() is also updated to reflect that cache
management SMP calls are not needed when all CPUs are siblings (i.e.
there are no foreign CPUs according to the new cpu_foreign_map[]
semantics which exclude siblings).

Signed-off-by: James Hogan 
Cc: Paul Burton 
Cc: Leonid Yegoshin 
Cc: Felix Fietkau 
Cc: Jayachandran C. 
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/13801/
Signed-off-by: Ralf Baechle

MIPS: SMP: Drop stop_this_cpu() cpu_foreign_map hack

2016-07-29T08:19:28+00:00

Commit cccf34e9411c ("MIPS: c-r4k: Fix cache flushing for MT cores")
added the cpu_foreign_map cpumask containing a single VPE from each
online core, and recalculated it when secondary CPUs are brought up.

stop_this_cpu() was also updated to recalculate cpu_foreign_map, but
with an additional hack before marking the CPU as offline to copy
cpu_online_mask into cpu_foreign_map and perform an SMP memory barrier.

This appears to have been intended to prevent cache management IPIs
being missed when the VPE representing the core in cpu_foreign_map is
taken offline while other VPEs remain online. Unfortunately there is
nothing in this hack to prevent r4k_on_each_cpu() from reading the old
cpu_foreign_map, and smp_call_function_many() from reading that new
cpu_online_mask with the core's representative VPE marked offline. It
then wouldn't send an IPI to any online VPEs of that core.

stop_this_cpu() is only actually called in panic and system shutdown /
halt / reboot situations, in which case all CPUs are going down and we
don't really need to care about cache management, so drop this hack.

Note that the __cpu_disable() case for CPU hotplug is handled in the
previous commit, and no synchronisation is needed there due to the use
of stop_machine() which prevents hotplug from taking place while any CPU
has disabled preemption (as r4k_on_each_cpu() does).

Signed-off-by: James Hogan 
Cc: Paul Burton 
Cc: Leonid Yegoshin 
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/13796/
Signed-off-by: Ralf Baechle

MIPS: SMP: Update cpu_foreign_map on CPU disable

2016-07-29T08:19:28+00:00

When a CPU is disabled via CPU hotplug, cpu_foreign_map is not updated.
This could result in cache management SMP calls being sent to offline
CPUs instead of online siblings in the same core.

Add a call to calculate_cpu_foreign_map() in the various MIPS cpu
disable callbacks after set_cpu_online(). All cases are updated for
consistency and to keep cpu_foreign_map strictly up to date, not just
those which may support hardware multithreading.

Fixes: cccf34e9411c ("MIPS: c-r4k: Fix cache flushing for MT cores")
Signed-off-by: James Hogan 
Cc: Paul Burton 
Cc: David Daney 
Cc: Kevin Cernekee 
Cc: Florian Fainelli 
Cc: Huacai Chen 
Cc: Hongliang Tao 
Cc: Hua Yan 
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/13799/
Signed-off-by: Ralf Baechle

MIPS: SMP: Clear ASID without confusing has_valid_asid()

2016-07-29T08:19:28+00:00

The SMP flush_tlb_*() functions may clear the memory map's ASIDs for
other CPUs if the mm has only a single user (the current CPU) in order
to avoid SMP calls. However this makes it appear to has_valid_asid(),
which is used by various cache flush functions, as if the CPUs have
never run in the mm, and therefore can't have cached any of its memory.

For flush_tlb_mm() this doesn't sound unreasonable.

flush_tlb_range() corresponds to flush_cache_range() which does do full
indexed cache flushes, but only on the icache if the specified mapping
is executable, otherwise it doesn't guarantee that there are no cache
contents left for the mm.

flush_tlb_page() corresponds to flush_cache_page(), which will perform
address based cache ops on the specified page only, and also only
touches the icache if the page is executable. It does not guarantee that
there are no cache contents left for the mm.

For example, this affects flush_cache_range() which uses the
has_valid_asid() optimisation. It is required to flush the icache when
mappings are made executable (e.g. using mprotect) so they are
immediately usable. If some code is changed to non executable in order
to be modified then it will not be flushed from the icache during that
time, but the ASID on other CPUs may still be cleared for TLB flushing.
When the code is changed back to executable, flush_cache_range() will
assume the code hasn't run on those other CPUs due to the zero ASID, and
won't invalidate the icache on them.

This is fixed by clearing the other CPUs ASIDs to 1 instead of 0 for the
above two flush_tlb_*() functions when the corresponding cache flushes
are likely to be incomplete (non executable range flush, or any page
flush). This ASID appears valid to has_valid_asid(), but still triggers
ASID regeneration due to the upper ASID version bits being 0, which is
less than the minimum ASID version of 1 and so always treated as stale.

Signed-off-by: James Hogan 
Cc: Paul Burton 
Cc: Leonid Yegoshin 
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/13795/
Signed-off-by: Ralf Baechle

MIPS: Don't BUG_ON when no IPI domain is found

2016-05-09T10:00:00+00:00

Commit fbde2d7d8290 ("MIPS: Add generic SMP IPI support") introduced
code that BUG_ON's in the case of a kernel that supports IPI domains but
does not have one at runtime. This case is possible on Malta where for
IPIs we may use either the GIC (which has an IPI IRQ domain
implementation) or core-local software interrupts between VPEs (which do
not currently have an IPI IRQ domain implementation). We can not know
which will be used until runtime when we know whether a GIC is actually
present, and if we run on a system with multiple VPEs and no GIC then
the BUG_ON is hit.

Commit 19fb5818ed60 ("IPS: Fix broken malta qemu") worked around this
for the single-core single-VPE case typically seen using QEMU, but does
not catch the multi-VPE case. This patch removes the insufficient CPU
presence check that was added and works around the bug differently,
effectively reverting that commit.

A simple way to reproduce this bug is by using QEMU, which partially
implements the MT ASE but does not implement the GIC as of version 2.5.
Using "-cpu 34Kf -smp 2" will present a system with 2 VPEs in one core &
no GIC, hitting the BUG_ON.

Given that we're post-merge-window on the way to v4.6, avoid this by
just returning from mips_smp_ipi_init when no IPI IRQ domain is found.
Ideally at some point all IPI implementations would be converted to the
same IPI IRQ domain interface & we'd be able to restore the check.

Signed-off-by: Paul Burton 
Cc: Qais Yousef 
Fixes: fbde2d7d8290 ("MIPS: Add generic SMP IPI support")
Fixes: 19fb5818ed60 ("IPS: Fix broken malta qemu")
Reverts: 19fb5818ed60 ("IPS: Fix broken malta qemu")
Cc: Qais Yousef 
Cc: James Hogan 
Cc: Alex Smith 
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/13007/
Signed-off-by: Ralf Baechle

MIPS: Fix broken malta qemu

2016-03-29T12:18:18+00:00

Malta defconfig compiles with GIC on. Hence when compiling for SMP it causes
the new IPI code to be activated. But on qemu malta there's no GIC causing a
BUG_ON(!ipidomain) to be hit in mips_smp_ipi_init().

Since in that configuration one can only run a single core SMP (!), skip IPI
initialisation if we detect that this is the case. It is a sensible
behaviour to introduce and should keep such possible configuration to run
rather than die hard unnecessarily.

Signed-off-by: Qais Yousef 
Reported-by: Guenter Roeck 
Tested-by: Guenter Roeck 
Cc: Thomas Gleixner 
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/12892/
Signed-off-by: Ralf Baechle

Merge branch 'smp-hotplug-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

2016-03-15T20:50:29+00:00

Pull cpu hotplug updates from Thomas Gleixner:
"This is the first part of the ongoing cpu hotplug rework:

- Initial implementation of the state machine

- Runs all online and prepare down callbacks on the plugged cpu and
not on some random processor

- Replaces busy loop waiting with completions

- Adds tracepoints so the states can be followed"

More detailed commentary on this work from an earlier email:
"What's wrong with the current cpu hotplug infrastructure?

- Asymmetry

The hotplug notifier mechanism is asymmetric versus the bringup and
teardown. This is mostly caused by the notifier mechanism.

- Largely undocumented dependencies

While some notifiers use explicitely defined notifier priorities,
we have quite some notifiers which use numerical priorities to
express dependencies without any documentation why.

- Control processor driven

Most of the bringup/teardown of a cpu is driven by a control
processor. While it is understandable, that preperatory steps,
like idle thread creation, memory allocation for and initialization
of essential facilities needs to be done before a cpu can boot,
there is no reason why everything else must run on a control
processor. Before this patch series, bringup looks like this:

Control CPU Booting CPU

do preparatory steps
kick cpu into life

do low level init

sync with booting cpu sync with control cpu

bring the rest up

- All or nothing approach

There is no way to do partial bringups. That's something which is
really desired because we waste e.g. at boot substantial amount of
time just busy waiting that the cpu comes to life. That's stupid
as we could very well do preparatory steps and the initial IPI for
other cpus and then go back and do the necessary low level
synchronization with the freshly booted cpu.

- Minimal debuggability

Due to the notifier based design, it's impossible to switch between
two stages of the bringup/teardown back and forth in order to test
the correctness. So in many hotplug notifiers the cancel
mechanisms are either not existant or completely untested.

- Notifier [un]registering is tedious

To [un]register notifiers we need to protect against hotplug at
every callsite. There is no mechanism that bringup/teardown
callbacks are issued on the online cpus, so every caller needs to
do it itself. That also includes error rollback.

What's the new design?

The base of the new design is a symmetric state machine, where both
the control processor and the booting/dying cpu execute a well
defined set of states. Each state is symmetric in the end, except
for some well defined exceptions, and the bringup/teardown can be
stopped and reversed at almost all states.

So the bringup of a cpu will look like this in the future:

Control CPU Booting CPU

do preparatory steps
kick cpu into life

do low level init

sync with booting cpu sync with control cpu

bring itself up

The synchronization step does not require the control cpu to wait.
That mechanism can be done asynchronously via a worker or some
other mechanism.

The teardown can be made very similar, so that the dying cpu cleans
up and brings itself down. Cleanups which need to be done after
the cpu is gone, can be scheduled asynchronously as well.

There is a long way to this, as we need to refactor the notion when a
cpu is available. Today we set the cpu online right after it comes
out of the low level bringup, which is not really correct.

The proper mechanism is to set it to available, i.e. cpu local
threads, like softirqd, hotplug thread etc. can be scheduled on that
cpu, and once it finished all booting steps, it's set to online, so
general workloads can be scheduled on it. The reverse happens on
teardown. First thing to do is to forbid scheduling of general
workloads, then teardown all the per cpu resources and finally shut it
off completely.

This patch series implements the basic infrastructure for this at the
core level. This includes the following:

- Basic state machine implementation with well defined states, so
ordering and prioritization can be expressed.

- Interfaces to [un]register state callbacks

This invokes the bringup/teardown callback on all online cpus with
the proper protection in place and [un]installs the callbacks in
the state machine array.

For callbacks which have no particular ordering requirement we have
a dynamic state space, so that drivers don't have to register an
explicit hotplug state.

If a callback fails, the code automatically does a rollback to the
previous state.

- Sysfs interface to drive the state machine to a particular step.

This is only partially functional today. Full functionality and
therefor testability will be achieved once we converted all
existing hotplug notifiers over to the new scheme.

- Run all CPU_ONLINE/DOWN_PREPARE notifiers on the booting/dying
processor:

Control CPU Booting CPU

do preparatory steps
kick cpu into life

do low level init

sync with booting cpu sync with control cpu
wait for boot
bring itself up

Signal completion to control cpu

In a previous step of this work we've done a full tree mechanical
conversion of all hotplug notifiers to the new scheme. The balance
is a net removal of about 4000 lines of code.

This is not included in this series, as we decided to take a
different approach. Instead of mechanically converting everything
over, we will do a proper overhaul of the usage sites one by one so
they nicely fit into the symmetric callback scheme.

I decided to do that after I looked at the ugliness of some of the
converted sites and figured out that their hotplug mechanism is
completely buggered anyway. So there is no point to do a
mechanical conversion first as we need to go through the usage
sites one by one again in order to achieve a full symmetric and
testable behaviour"

* 'smp-hotplug-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (23 commits)
cpu/hotplug: Document states better
cpu/hotplug: Fix smpboot thread ordering
cpu/hotplug: Remove redundant state check
cpu/hotplug: Plug death reporting race
rcu: Make CPU_DYING_IDLE an explicit call
cpu/hotplug: Make wait for dead cpu completion based
cpu/hotplug: Let upcoming cpu bring itself fully up
arch/hotplug: Call into idle with a proper state
cpu/hotplug: Move online calls to hotplugged cpu
cpu/hotplug: Create hotplug threads
cpu/hotplug: Split out the state walk into functions
cpu/hotplug: Unpark smpboot threads from the state machine
cpu/hotplug: Move scheduler cpu_online notifier to hotplug core
cpu/hotplug: Implement setup/removal interface
cpu/hotplug: Make target state writeable
cpu/hotplug: Add sysfs state interface
cpu/hotplug: Hand in target state to _cpu_up/down
cpu/hotplug: Convert the hotplugged cpu work to a state machine
cpu/hotplug: Convert to a state machine for the control processor
cpu/hotplug: Add tracepoints
...

Merge branch 'irq-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

2016-03-15T19:48:48+00:00

Pull irq updates from Thomas Gleixner:
 "The 4.6 pile of irq updates contains:

   - Support for IPI irqdomains to support proper integration of IPIs to
     and from coprocessors.  The first user of this new facility is
     MIPS.  The relevant MIPS patches come with the core to avoid merge
     ordering issues and have been acked by Ralf.

   - A new command line option to set the default interrupt affinity
     mask at boot time.

   - Support for some more new ARM and MIPS interrupt controllers:
     tango, alpine-msix and bcm6345-l1

   - Two small cleanups for x86/apic which we merged into irq/core to
     avoid yet another branch in x86 with two tiny commits.

   - The usual set of updates, cleanups in drivers/irqchip.  Mostly in
     the area of ARM-GIC, arada-37-xp and atmel chips.  Nothing
     outstanding here"

* 'irq-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (56 commits)
  irqchip/irq-alpine-msi: Release the correct domain on error
  irqchip/mxs: Fix error check of of_io_request_and_map()
  irqchip/sunxi-nmi: Fix error check of of_io_request_and_map()
  genirq: Export IRQ functions for module use
  irqchip/gic/realview: Support more RealView DCC variants
  Documentation/bindings: Document the Alpine MSIX driver
  irqchip: Add the Alpine MSIX interrupt controller
  irqchip/gic-v3: Always return IRQ_SET_MASK_OK_DONE in gic_set_affinity
  irqchip/gic-v3-its: Mark its_init() and its children as __init
  irqchip/gic-v3: Remove gic_root_node variable from the ITS code
  irqchip/gic-v3: ACPI: Add redistributor support via GICC structures
  irqchip/gic-v3: Add ACPI support for GICv3/4 initialization
  irqchip/gic-v3: Refactor gic_of_init() for GICv3 driver
  x86/apic: Deinline _flat_send_IPI_mask, save ~150 bytes
  x86/apic: Deinline __default_send_IPI_*, save ~200 bytes
  dt-bindings: interrupt-controller: Add SoC-specific compatible string to Marvell ODMI
  irqchip/mips-gic: Add new DT property to reserve IPIs
  MIPS: Delete smp-gic.c
  MIPS: Make smp CMP, CPS and MT use the new generic IPI functions
  MIPS: Add generic SMP IPI support
  ...

MIPS: smp.c: Fix uninitialised temp_foreign_map

2016-03-13T09:59:19+00:00

When calculate_cpu_foreign_map() recalculates the cpu_foreign_map
cpumask it uses the local variable temp_foreign_map without initialising
it to zero. Since the calculation only ever sets bits in this cpumask
any existing bits at that memory location will remain set and find their
way into cpu_foreign_map too. This could potentially lead to cache
operations suboptimally doing smp calls to multiple VPEs in the same
core, even though the VPEs share primary caches.

Therefore initialise temp_foreign_map using cpumask_clear() before use.

Fixes: cccf34e9411c ("MIPS: c-r4k: Fix cache flushing for MT cores")
Signed-off-by: James Hogan 
Cc: Paul Burton 
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/12759/
Signed-off-by: Ralf Baechle