If a CPU goes offline, interrupts affine to the CPU are moved away. If the
outgoing CPU is the last CPU in the affinity mask the migration code breaks
the affinity and sets it it all online cpus.
This is a problem for affinity managed interrupts as CPU hotplug is often
used for power management purposes. If the affinity is broken, the
interrupt is not longer affine to the CPUs to which it was allocated.
The affinity spreading allows to lay out multi queue devices in a way that
they are assigned to a single CPU or a group of CPUs. If the last CPU goes
offline, then the queue is not longer used, so the interrupt can be
shutdown gracefully and parked until one of the assigned CPUs comes online
again.
Add a graceful shutdown mechanism into the irq affinity breaking code path,
mark the irq as MANAGED_SHUTDOWN and leave the affinity mask unmodified.
In the online path, scan the active interrupts for managed interrupts and
if the interrupt is functional and the newly online CPU is part of the
affinity mask, restart the interrupt if it is marked MANAGED_SHUTDOWN or if
the interrupts is started up, try to add the CPU back to the effective
affinity mask.
Originally-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Keith Busch <keith.busch@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20170619235447.273417334@linutronix.de
Affinity managed interrupts should keep their assigned affinity accross CPU
hotplug. To avoid magic hackery in device drivers, the core code shall
manage them transparently and set these interrupts into a managed shutdown
state when the last CPU of the assigned affinity mask goes offline. The
interrupt will be restarted when one of the CPUs in the assigned affinity
mask comes back online.
Add the necessary logic to irq_startup(). If an interrupt is requested and
started up, the code checks whether it is affinity managed and if so, it
checks whether a CPU in the interrupts affinity mask is online. If not, it
puts the interrupt into managed shutdown state.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Keith Busch <keith.busch@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Christoph Hellwig <hch@lst.de>
Link: http://lkml.kernel.org/r/20170619235447.189851170@linutronix.de
In order to handle managed interrupts gracefully on irq_startup() so they
won't lose their assigned affinity, it's necessary to allow startups which
keep the interrupts in managed shutdown state, if none of the assigend CPUs
is online. This allows drivers to request interrupts w/o the CPUs being
online, which avoid online/offline churn in drivers.
Add a force argument which can override that decision and let only
request_irq() and enable_irq() allow the managed shutdown
handling. enable_irq() is required, because the interrupt might be
requested with IRQF_NOAUTOEN and enable_irq() invokes irq_startup() which
would then wreckage the assignment again. All other callers force startup
and potentially break the assigned affinity.
No functional change as this only adds the function argument.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Keith Busch <keith.busch@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Christoph Hellwig <hch@lst.de>
Link: http://lkml.kernel.org/r/20170619235447.112094565@linutronix.de
Affinity managed interrupts should keep their assigned affinity accross CPU
hotplug. To avoid magic hackery in device drivers, the core code shall
manage them transparently. This will set these interrupts into a managed
shutdown state when the last CPU of the assigned affinity mask goes
offline. The interrupt will be restarted when one of the CPUs in the
assigned affinity mask comes back online.
Introduce the necessary state flag and the accessor functions.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Keith Busch <keith.busch@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Christoph Hellwig <hch@lst.de>
Link: http://lkml.kernel.org/r/20170619235446.954523476@linutronix.de
There is currently no way to evaluate the effective affinity mask of a
given interrupt. Many irq chips allow only a single target CPU or a subset
of CPUs in the affinity mask.
Updating the mask at the time of setting the affinity to the subset would
be counterproductive because information for cpu hotplug about assigned
interrupt affinities gets lost. On CPU hotplug it's also pointless to force
migrate an interrupt, which is not targeted at the CPU effectively. But
currently the information is not available.
Provide a seperate mask to be updated by the irq_chip->irq_set_affinity()
implementations. Implement the read only proc files so the user can see the
effective mask as well w/o trying to deduce it from /proc/interrupts.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Keith Busch <keith.busch@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Christoph Hellwig <hch@lst.de>
Link: http://lkml.kernel.org/r/20170619235446.247834245@linutronix.de
The startup vs. setaffinity ordering of interrupts depends on the
IRQF_NOAUTOEN flag. Chained interrupts are not getting any affinity
assignment at all.
A regular interrupt is started up and then the affinity is set. A
IRQF_NOAUTOEN marked interrupt is not started up, but the affinity is set
nevertheless.
Move the affinity setup to startup_irq() so the ordering is always the same
and chained interrupts get the proper default affinity assigned as well.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Keith Busch <keith.busch@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Christoph Hellwig <hch@lst.de>
Link: http://lkml.kernel.org/r/20170619235445.020534783@linutronix.de
If an CPU goes offline, the interrupts are migrated away, but a eventually
pending interrupt move, which has not yet been made effective is kept
pending even if the outgoing CPU is the sole target of the pending affinity
mask. What's worse is, that the pending affinity mask is discarded even if
it would contain a valid subset of the online CPUs.
Implement a helper function which allows to avoid these issues.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Keith Busch <keith.busch@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Christoph Hellwig <hch@lst.de>
Link: http://lkml.kernel.org/r/20170619235444.691345468@linutronix.de
Debugging (hierarchical) interupt domains is tedious as there is no
information about the hierarchy and no information about states of
interrupts in the various domain levels.
Add a debugfs directory 'irq' and subdirectories 'domains' and 'irqs'.
The domains directory contains the domain files. The content is information
about the domain. If the domain is part of a hierarchy then the parent
domains are printed as well.
# ls /sys/kernel/debug/irq/domains/
default INTEL-IR-2 INTEL-IR-MSI-2 IO-APIC-IR-2 PCI-MSI
DMAR-MSI INTEL-IR-3 INTEL-IR-MSI-3 IO-APIC-IR-3 unknown-1
INTEL-IR-0 INTEL-IR-MSI-0 IO-APIC-IR-0 IO-APIC-IR-4 VECTOR
INTEL-IR-1 INTEL-IR-MSI-1 IO-APIC-IR-1 PCI-HT
# cat /sys/kernel/debug/irq/domains/VECTOR
name: VECTOR
size: 0
mapped: 216
flags: 0x00000041
# cat /sys/kernel/debug/irq/domains/IO-APIC-IR-0
name: IO-APIC-IR-0
size: 24
mapped: 19
flags: 0x00000041
parent: INTEL-IR-3
name: INTEL-IR-3
size: 65536
mapped: 167
flags: 0x00000041
parent: VECTOR
name: VECTOR
size: 0
mapped: 216
flags: 0x00000041
Unfortunately there is no per cpu information about the VECTOR domain (yet).
The irqs directory contains detailed information about mapped interrupts.
# cat /sys/kernel/debug/irq/irqs/3
handler: handle_edge_irq
status: 0x00004000
istate: 0x00000000
ddepth: 1
wdepth: 0
dstate: 0x01018000
IRQD_IRQ_DISABLED
IRQD_SINGLE_TARGET
IRQD_MOVE_PCNTXT
node: 0
affinity: 0-143
effectiv: 0
pending:
domain: IO-APIC-IR-0
hwirq: 0x3
chip: IR-IO-APIC
flags: 0x10
IRQCHIP_SKIP_SET_WAKE
parent:
domain: INTEL-IR-3
hwirq: 0x20000
chip: INTEL-IR
flags: 0x0
parent:
domain: VECTOR
hwirq: 0x3
chip: APIC
flags: 0x0
This was developed to simplify the debugging of the managed affinity
changes.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Keith Busch <keith.busch@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Christoph Hellwig <hch@lst.de>
Link: http://lkml.kernel.org/r/20170619235444.537566163@linutronix.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
In order to provide proper debug interface it's required to have domain
names available when the domain is added. Non fwnode based architectures
like x86 have no way to do so.
It's not possible to use domain ops or host data for this as domain ops
might be the same for several instances, but the names have to be unique.
Extend the irqchip fwnode to allow transporting the domain name. If no node
is supplied, create a 'unknown-N' placeholder.
Warn if an invalid node is supplied and treat it like no node. This happens
e.g. with i2 devices on x86 which hand in an ACPI type node which has no
interface for retrieving the name.
[ Folded a fix from Marc to make DT name parsing work ]
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Keith Busch <keith.busch@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Christoph Hellwig <hch@lst.de>
Link: http://lkml.kernel.org/r/20170619235443.588784933@linutronix.de
Pull timer fixes from Thomas Gleixner:
"Three fixlets for timers:
- Two hot-fixes for the alarmtimer based posix timers, which prevent
a nasty DOS by self rescheduling timers. The proper cleanup of that
mess is queued for 4.13
- Make a function static"
* 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
tick/broadcast: Make tick_broadcast_setup_oneshot() static
alarmtimer: Rate limit periodic intervals
alarmtimer: Prevent overflow of relative timers
Pull scheduler fixes from Thomas Gleixner:
"Two small fixes for the schedulre core:
- Use the proper switch_mm() variant in idle_task_exit() because that
code is not called with interrupts disabled.
- Fix a confusing typo in a printk"
* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
sched/core: Idle_task_exit() shouldn't use switch_mm_irqs_off()
sched/fair: Fix typo in printk message
Pull irq fix from Thomas Gleixner:
"Add a missing resource release to an error path"
* 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
genirq: Release resources in __setup_irq() error path
Revert commit 39b64aa1c0 (cpufreq: schedutil: Reduce frequencies
slower) that introduced unintentional changes in behavior leading
to adverse effects on some systems.
Reported-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Pull CPU hotplug fix from Ingo Molnar:
"An error handling corner case fix"
* 'smp-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
cpu/hotplug: Drop the device lock on error
Pull RCU fixes from Ingo Molnar:
"Fix an SRCU bug affecting KVM IRQ injection"
* 'rcu-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
srcu: Allow use of Classic SRCU from both process and interrupt context
srcu: Allow use of Tiny/Tree SRCU from both process and interrupt context
Pull perf fixes from Ingo Molnar:
"This is mostly tooling fixes, plus an instruction pointer filtering
fix.
It's more fixes than usual - Arnaldo got back from a longer vacation
and there was a backlog"
* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (27 commits)
perf symbols: Kill dso__build_id_is_kmod()
perf symbols: Keep DSO->symtab_type after decompress
perf tests: Decompress kernel module before objdump
perf tools: Consolidate error path in __open_dso()
perf tools: Decompress kernel module when reading DSO data
perf annotate: Use dso__decompress_kmodule_path()
perf tools: Introduce dso__decompress_kmodule_{fd,path}
perf tools: Fix a memory leak in __open_dso()
perf annotate: Fix symbolic link of build-id cache
perf/core: Drop kernel samples even though :u is specified
perf script python: Remove dups in documentation examples
perf script python: Updated trace_unhandled() signature
perf script python: Fix wrong code snippets in documentation
perf script: Fix documentation errors
perf script: Fix outdated comment for perf-trace-python
perf probe: Fix examples section of documentation
perf report: Ensure the perf DSO mapping matches what libdw sees
perf report: Include partial stacks unwound with libdw
perf annotate: Add missing powerpc triplet
perf test: Disable breakpoint signal tests for powerpc
...
Pull RCU fix from Paul E. McKenney:
" This series enables srcu_read_lock() and srcu_read_unlock() to be used from
interrupt handlers, which fixes a bug in KVM's use of SRCU in delivery
of interrupts to guest OSes. "
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull power management fixes from Rafael Wysocki:
"These revert one problematic commit related to system sleep and fix
one recent intel_pstate regression.
Specifics:
- Revert a recent commit that attempted to avoid spurious wakeups
from suspend-to-idle via ACPI SCI, but introduced regressions on
some systems (Rafael Wysocki).
We will get back to the problem it tried to address in the next
cycle.
- Fix a possible division by 0 during intel_pstate initialization
due to a missing check (Rafael Wysocki)"
* tag 'pm-4.12-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
Revert "ACPI / sleep: Ignore spurious SCI wakeups from suspend-to-idle"
cpufreq: intel_pstate: Avoid division by 0 in min_perf_pct_min()
Pull printk fix from Petr Mladek:
"This reverts a fix added into 4.12-rc1. It caused the kernel log to be
printed on another console when two consoles of the same type were
defined, e.g. console=ttyS0 console=ttyS1.
This configuration was never supported by kernel itself, but it
started to make sense with systemd. In other words, the commit broke
userspace"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/pmladek/printk:
Revert "printk: fix double printing with earlycon"
Linu Cherian reported a WARN in cleanup_srcu_struct() when shutting
down a guest running iperf on a VFIO assigned device. This happens
because irqfd_wakeup() calls srcu_read_lock(&kvm->irq_srcu) in interrupt
context, while a worker thread does the same inside kvm_set_irq(). If the
interrupt happens while the worker thread is executing __srcu_read_lock(),
updates to the Classic SRCU ->lock_count[] field or the Tree SRCU
->srcu_lock_count[] field can be lost.
The docs say you are not supposed to call srcu_read_lock() and
srcu_read_unlock() from irq context, but KVM interrupt injection happens
from (host) interrupt context and it would be nice if SRCU supported the
use case. KVM is using SRCU here not really for the "sleepable" part,
but rather due to its IPI-free fast detection of grace periods. It is
therefore not desirable to switch back to RCU, which would effectively
revert commit 719d93cd5f ("kvm/irqchip: Speed up KVM_SET_GSI_ROUTING",
2014-01-16).
However, the docs are overly conservative. You can have an SRCU instance
only has users in irq context, and you can mix process and irq context
as long as process context users disable interrupts. In addition,
__srcu_read_unlock() actually uses this_cpu_dec() on both Tree SRCU and
Classic SRCU. For those two implementations, only srcu_read_lock()
is unsafe.
When Classic SRCU's __srcu_read_unlock() was changed to use this_cpu_dec(),
in commit 5a41344a3d ("srcu: Simplify __srcu_read_unlock() via
this_cpu_dec()", 2012-11-29), __srcu_read_lock() did two increments.
Therefore it kept __this_cpu_inc(), with preempt_disable/enable in
the caller. Tree SRCU however only does one increment, so on most
architectures it is more efficient for __srcu_read_lock() to use
this_cpu_inc(), and any performance differences appear to be down in
the noise.
Cc: stable@vger.kernel.org
Fixes: 719d93cd5f ("kvm/irqchip: Speed up KVM_SET_GSI_ROUTING")
Reported-by: Linu Cherian <linuc.decode@gmail.com>
Suggested-by: Linu Cherian <linuc.decode@gmail.com>
Cc: kvm@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Linu Cherian reported a WARN in cleanup_srcu_struct() when shutting
down a guest running iperf on a VFIO assigned device. This happens
because irqfd_wakeup() calls srcu_read_lock(&kvm->irq_srcu) in interrupt
context, while a worker thread does the same inside kvm_set_irq(). If the
interrupt happens while the worker thread is executing __srcu_read_lock(),
updates to the Classic SRCU ->lock_count[] field or the Tree SRCU
->srcu_lock_count[] field can be lost.
The docs say you are not supposed to call srcu_read_lock() and
srcu_read_unlock() from irq context, but KVM interrupt injection happens
from (host) interrupt context and it would be nice if SRCU supported the
use case. KVM is using SRCU here not really for the "sleepable" part,
but rather due to its IPI-free fast detection of grace periods. It is
therefore not desirable to switch back to RCU, which would effectively
revert commit 719d93cd5f ("kvm/irqchip: Speed up KVM_SET_GSI_ROUTING",
2014-01-16).
However, the docs are overly conservative. You can have an SRCU instance
only has users in irq context, and you can mix process and irq context
as long as process context users disable interrupts. In addition,
__srcu_read_unlock() actually uses this_cpu_dec() on both Tree SRCU and
Classic SRCU. For those two implementations, only srcu_read_lock()
is unsafe.
When Classic SRCU's __srcu_read_unlock() was changed to use this_cpu_dec(),
in commit 5a41344a3d ("srcu: Simplify __srcu_read_unlock() via
this_cpu_dec()", 2012-11-29), __srcu_read_lock() did two increments.
Therefore it kept __this_cpu_inc(), with preempt_disable/enable in
the caller. Tree SRCU however only does one increment, so on most
architectures it is more efficient for __srcu_read_lock() to use
this_cpu_inc(), and any performance differences appear to be down in
the noise.
Unlike Classic and Tree SRCU, Tiny SRCU does increments and decrements on
a single variable. Therefore, as Peter Zijlstra pointed out, Tiny SRCU's
implementation already supports mixed-context use of srcu_read_lock()
and srcu_read_unlock(), at least as long as uses of srcu_read_lock()
and srcu_read_unlock() in each handler are nested and paired properly.
In other words, it is still illegal to (say) invoke srcu_read_lock()
in an interrupt handler and to invoke the matching srcu_read_unlock()
in a softirq handler. Therefore, the only change required for Tiny SRCU
is to its comments.
Fixes: 719d93cd5f ("kvm/irqchip: Speed up KVM_SET_GSI_ROUTING")
Reported-by: Linu Cherian <linuc.decode@gmail.com>
Suggested-by: Linu Cherian <linuc.decode@gmail.com>
Cc: kvm@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Tested-by: Paolo Bonzini <pbonzini@redhat.com>
This reverts commit cf39bf58af.
The commit regression to users that define both console=ttyS1
and console=ttyS0 on the command line, see
https://lkml.kernel.org/r/20170509082915.GA13236@bistromath.localdomain
The kernel log messages always appeared only on one serial port. It is
even documented in Documentation/admin-guide/serial-console.rst:
"Note that you can only define one console per device type (serial,
video)."
The above mentioned commit changed the order in which the command line
parameters are searched. As a result, the kernel log messages go to
the last mentioned ttyS* instead of the first one.
We long thought that using two console=ttyS* on the command line
did not make sense. But then we realized that console= parameters
were handled also by systemd, see
http://0pointer.de/blog/projects/serial-console.html
"By default systemd will instantiate one serial-getty@.service on
the main kernel console, if it is not a virtual terminal."
where
"[4] If multiple kernel consoles are used simultaneously, the main
console is the one listed first in /sys/class/tty/console/active,
which is the last one listed on the kernel command line."
This puts the original report into another light. The system is running
in qemu. The first serial port is used to store the messages into a file.
The second one is used to login to the system via a socket. It depends
on systemd and the historic kernel behavior.
By other words, systemd causes that it makes sense to define both
console=ttyS1 console=ttyS0 on the command line. The kernel fix
caused regression related to userspace (systemd) and need to be
reverted.
In addition, it went out that the fix helped only partially.
The messages still were duplicated when the boot console was
removed early by late_initcall(printk_late_init). Then the entire
log was replayed when the same console was registered as a normal one.
Link: 20170606160339.GC7604@pathway.suse.cz
Cc: Aleksey Makarov <aleksey.makarov@linaro.org>
Cc: Sabrina Dubroca <sd@queasysnail.net>
Cc: Sudeep Holla <sudeep.holla@arm.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Peter Hurley <peter@hurleysoftware.com>
Cc: Jiri Slaby <jslaby@suse.com>
Cc: Robin Murphy <robin.murphy@arm.com>,
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: "Nair, Jayachandran" <Jayachandran.Nair@cavium.com>
Cc: linux-serial@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Reported-by: Sabrina Dubroca <sd@queasysnail.net>
Acked-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Petr Mladek <pmladek@suse.com>
