IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
free_moved_vector() accesses the per cpu vector array with this_cpu_write()
to clear the vector. The function has two call sites:
1) The vector cleanup IPI
2) The force_complete_move() code path
For #1 this_cpu_write() is correct as it runs on the CPU on which the
vector needs to be freed.
For #2 this_cpu_write() is wrong because the function is called from an
outgoing CPU which is not necessarily the CPU on which the previous vector
needs to be freed. As a result it sets the vector on the outgoing CPU to
NULL, which is pointless as that CPU does not handle interrupts
anymore. What's worse is that it leaves the vector on the previous target
CPU in place which later on triggers the BUG_ON(vector) in the vector
allocation code when the vector gets reused. That's possible because the
bitmap allocator entry of that CPU is freed correctly.
Always use the CPU to which the vector was associated and clear the vector
entry on that CPU. Fixup the tracepoint as well so it tracks on which CPU
the vector gets removed.
Fixes: 69cde0004a ("x86/vector: Use matrix allocator for vector assignment")
Reported-by: Petri Latvala <petri.latvala@intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Juergen Gross <jgross@suse.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Len Brown <lenb@kernel.org>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rui Zhang <rui.zhang@intel.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Alok Kataria <akataria@vmware.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Yu Chen <yu.c.chen@intel.com>
Link: https://lkml.kernel.org/r/alpine.DEB.2.20.1710161614430.1973@nanos
The core interrupt code can call the affinity setter for inactive
interrupts under certain circumstances.
For inactive intererupts which use managed or reservation mode this is a
pointless exercise as the activation will assign a vector which fits the
destination mask.
Check for this and return w/o going through the vector assignment.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Devices with many queues allocate a huge number of interrupts and get
assigned a vector for each of them, even if the queues are not active and
the interrupts never requested. This causes problems with the decision
whether the global vector space is sufficient for CPU hot unplug
operations.
Change it to a reservation scheme, which allows overcommitment.
When the interrupt is allocated and initialized the vector assignment
merily updates the reservation request counter in the matrix
allocator. This counter is used to emit warnings when the reservation
exceeds the available vector space, but does not affect CPU offline
operations. Like the managed interrupts the corresponding MSI/DMAR/IOAPIC
entries are directed to the special shutdown vector.
When the interrupt is requested, then the activation code tries to assign a
real vector. If that succeeds the interrupt is started up and functional.
If that fails, then subsequently request_irq() fails with -ENOSPC.
This allows a clear separation of inactive and active modes and simplifies
the final decisions whether the global vector space is sufficient for CPU
offline operations.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Juergen Gross <jgross@suse.com>
Tested-by: Yu Chen <yu.c.chen@intel.com>
Acked-by: Juergen Gross <jgross@suse.com>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: Alok Kataria <akataria@vmware.com>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Rui Zhang <rui.zhang@intel.com>
Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Len Brown <lenb@kernel.org>
Link: https://lkml.kernel.org/r/20170913213156.184211133@linutronix.de
Replace the magic vector allocation code by a simple bitmap matrix
allocator. This avoids loops and hoops over CPUs and vector arrays, so in
case of densly used vector spaces it's way faster.
This also gets rid of the magic 'spread the vectors accross priority
levels' heuristics in the current allocator:
The comment in __asign_irq_vector says:
* NOTE! The local APIC isn't very good at handling
* multiple interrupts at the same interrupt level.
* As the interrupt level is determined by taking the
* vector number and shifting that right by 4, we
* want to spread these out a bit so that they don't
* all fall in the same interrupt level.
After doing some palaeontological research the following was found the
following in the PPro Developer Manual Volume 3:
"7.4.2. Valid Interrupts
The local and I/O APICs support 240 distinct vectors in the range of 16
to 255. Interrupt priority is implied by its vector, according to the
following relationship: priority = vector / 16
One is the lowest priority and 15 is the highest. Vectors 16 through
31 are reserved for exclusive use by the processor. The remaining
vectors are for general use. The processor's local APIC includes an
in-service entry and a holding entry for each priority level. To avoid
losing inter- rupts, software should allocate no more than 2 interrupt
vectors per priority."
The current SDM tells nothing about that, instead it states:
"If more than one interrupt is generated with the same vector number,
the local APIC can set the bit for the vector both in the IRR and the
ISR. This means that for the Pentium 4 and Intel Xeon processors, the
IRR and ISR can queue two interrupts for each interrupt vector: one
in the IRR and one in the ISR. Any additional interrupts issued for
the same interrupt vector are collapsed into the single bit in the
IRR.
For the P6 family and Pentium processors, the IRR and ISR registers
can queue no more than two interrupts per interrupt vector and will
reject other interrupts that are received within the same vector."
Which means, that on P6/Pentium the APIC will reject a new message and
tell the sender to retry, which increases the load on the APIC bus and
nothing more.
There is no affirmative answer from Intel on that, but it's a sane approach
to remove that for the following reasons:
1) No other (relevant Open Source) operating systems bothers to
implement this or mentiones this at all.
2) The current allocator has no enforcement for this and especially the
legacy interrupts, which are the main source of interrupts on these
P6 and older systmes, are allocated linearly in the same priority
level and just work.
3) The current machines have no problem with that at all as verified
with some experiments.
4) AMD at least confirmed that such an issue is unknown.
5) P6 and older are dinosaurs almost 20 years EOL, so there is really
no reason to worry about that too much.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Juergen Gross <jgross@suse.com>
Tested-by: Yu Chen <yu.c.chen@intel.com>
Acked-by: Juergen Gross <jgross@suse.com>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: Alok Kataria <akataria@vmware.com>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Rui Zhang <rui.zhang@intel.com>
Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Len Brown <lenb@kernel.org>
Link: https://lkml.kernel.org/r/20170913213155.443678104@linutronix.de
Setting the interrupt affinity of a single interrupt to multiple CPUs has a
dubious value.
1) This only works on machines where the APIC uses logical destination
mode. If the APIC uses physical destination mode then it is already
restricted to a single CPU
2) Experiments have shown, that the benefit of multi CPU affinity is close
to zero and in some test even worse than setting the affinity to a
single CPU.
The reason for this is that the delivery targets the APIC with the
lowest ID first and only if that APIC is busy (servicing an interrupt,
i.e. ISR is not empty) it hands it over to the next APIC. In the
conducted tests the vast majority of interrupts ends up on the APIC
with the lowest ID anyway, so there is no natural spreading of the
interrupts possible.
Supporting multi CPU affinities adds a lot of complexity to the code, which
can turn the allocation search into a worst case of
nr_vectors * nr_online_cpus * nr_bits_in_target_mask
As a first step disable it by restricting the vector search to a single
CPU.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Juergen Gross <jgross@suse.com>
Tested-by: Yu Chen <yu.c.chen@intel.com>
Acked-by: Juergen Gross <jgross@suse.com>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: Alok Kataria <akataria@vmware.com>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Rui Zhang <rui.zhang@intel.com>
Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Len Brown <lenb@kernel.org>
Link: https://lkml.kernel.org/r/20170913213154.228824430@linutronix.de
Pull irq updates from Thomas Gleixner:
"The irq department delivers:
- Expand the generic infrastructure handling the irq migration on CPU
hotplug and convert X86 over to it. (Thomas Gleixner)
Aside of consolidating code this is a preparatory change for:
- Finalizing the affinity management for multi-queue devices. The
main change here is to shut down interrupts which are affine to a
outgoing CPU and reenabling them when the CPU comes online again.
That avoids moving interrupts pointlessly around and breaking and
reestablishing affinities for no value. (Christoph Hellwig)
Note: This contains also the BLOCK-MQ and NVME changes which depend
on the rework of the irq core infrastructure. Jens acked them and
agreed that they should go with the irq changes.
- Consolidation of irq domain code (Marc Zyngier)
- State tracking consolidation in the core code (Jeffy Chen)
- Add debug infrastructure for hierarchical irq domains (Thomas
Gleixner)
- Infrastructure enhancement for managing generic interrupt chips via
devmem (Bartosz Golaszewski)
- Constification work all over the place (Tobias Klauser)
- Two new interrupt controller drivers for MVEBU (Thomas Petazzoni)
- The usual set of fixes, updates and enhancements all over the
place"
* 'irq-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (112 commits)
irqchip/or1k-pic: Fix interrupt acknowledgement
irqchip/irq-mvebu-gicp: Allocate enough memory for spi_bitmap
irqchip/gic-v3: Fix out-of-bound access in gic_set_affinity
nvme: Allocate queues for all possible CPUs
blk-mq: Create hctx for each present CPU
blk-mq: Include all present CPUs in the default queue mapping
genirq: Avoid unnecessary low level irq function calls
genirq: Set irq masked state when initializing irq_desc
genirq/timings: Add infrastructure for estimating the next interrupt arrival time
genirq/timings: Add infrastructure to track the interrupt timings
genirq/debugfs: Remove pointless NULL pointer check
irqchip/gic-v3-its: Don't assume GICv3 hardware supports 16bit INTID
irqchip/gic-v3-its: Add ACPI NUMA node mapping
irqchip/gic-v3-its-platform-msi: Make of_device_ids const
irqchip/gic-v3-its: Make of_device_ids const
irqchip/irq-mvebu-icu: Add new driver for Marvell ICU
irqchip/irq-mvebu-gicp: Add new driver for Marvell GICP
dt-bindings/interrupt-controller: Add DT binding for the Marvell ICU
genirq/irqdomain: Remove auto-recursive hierarchy support
irqchip/MSI: Use irq_domain_update_bus_token instead of an open coded access
...
When a CPU is about to be offlined we call fixup_irqs() that resets IRQ
affinities related to the CPU in question. The same thing is also done when
the system is suspended to S-states like S3 (mem).
For each IRQ we try to complete any on-going move regardless whether the
IRQ is actually part of x86_vector_domain. For each IRQ descriptor we fetch
its chip_data, assume it is of type struct apic_chip_data and manipulate it
by clearing old_domain mask etc. For irq_chips that are not part of the
x86_vector_domain, like those created by various GPIO drivers, will find
their chip_data being changed unexpectly.
Below is an example where GPIO chip owned by pinctrl-sunrisepoint.c gets
corrupted after resume:
# cat /sys/kernel/debug/gpio
gpiochip0: GPIOs 360-511, parent: platform/INT344B:00, INT344B:00:
gpio-511 ( |sysfs ) in hi
# rtcwake -s10 -mmem
<10 seconds passes>
# cat /sys/kernel/debug/gpio
gpiochip0: GPIOs 360-511, parent: platform/INT344B:00, INT344B:00:
gpio-511 ( |sysfs ) in ?
Note '?' in the output. It means the struct gpio_chip ->get function is
NULL whereas before suspend it was there.
Fix this by first checking that the IRQ belongs to x86_vector_domain before
we try to use the chip_data as struct apic_chip_data.
Reported-and-tested-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Cc: stable@vger.kernel.org # 4.4+
Link: http://lkml.kernel.org/r/20161003101708.34795-1-mika.westerberg@linux.intel.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
If x86_vector_alloc_irq() fails x86_vector_free_irqs() is invoked to cleanup
the already allocated vectors. This subsequently calls clear_vector_irq().
The failed irq has no vector assigned, which triggers the BUG_ON(!vector) in
clear_vector_irq().
We cannot suppress the call to x86_vector_free_irqs() for the failed
interrupt, because the other data related to this irq must be cleaned up as
well. So calling clear_vector_irq() with vector == 0 is legitimate.
Remove the BUG_ON and return if vector is zero,
[ tglx: Massaged changelog ]
Fixes: b5dc8e6c21 "x86/irq: Use hierarchical irqdomain to manage CPU interrupt vectors"
Signed-off-by: Keith Busch <keith.busch@intel.com>
Cc: stable@vger.kernel.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Harry reported, that he's able to trigger a system freeze with cpu hot
unplug. The freeze turned out to be a live lock caused by recent changes in
irq_force_complete_move().
When fixup_irqs() and from there irq_force_complete_move() is called on the
dying cpu, then all other cpus are in stop machine an wait for the dying cpu
to complete the teardown. If there is a move of an interrupt pending then
irq_force_complete_move() sends the cleanup IPI to the cpus in the old_domain
mask and waits for them to clear the mask. That's obviously impossible as
those cpus are firmly stuck in stop machine with interrupts disabled.
I should have known that, but I completely overlooked it being concentrated on
the locking issues around the vectors. And the existance of the call to
__irq_complete_move() in the code, which actually sends the cleanup IPI made
it reasonable to wait for that cleanup to complete. That call was bogus even
before the recent changes as it was just a pointless distraction.
We have to look at two cases:
1) The move_in_progress flag of the interrupt is set
This means the ioapic has been updated with the new vector, but it has not
fired yet. In theory there is a race:
set_ioapic(new_vector) <-- Interrupt is raised before update is effective,
i.e. it's raised on the old vector.
So if the target cpu cannot handle that interrupt before the old vector is
cleaned up, we get a spurious interrupt and in the worst case the ioapic
irq line becomes stale, but my experiments so far have only resulted in
spurious interrupts.
But in case of cpu hotplug this should be a non issue because if the
affinity update happens right before all cpus rendevouz in stop machine,
there is no way that the interrupt can be blocked on the target cpu because
all cpus loops first with interrupts enabled in stop machine, so the old
vector is not yet cleaned up when the interrupt fires.
So the only way to run into this issue is if the delivery of the interrupt
on the apic/system bus would be delayed beyond the point where the target
cpu disables interrupts in stop machine. I doubt that it can happen, but at
least there is a theroretical chance. Virtualization might be able to
expose this, but AFAICT the IOAPIC emulation is not as stupid as the real
hardware.
I've spent quite some time over the weekend to enforce that situation,
though I was not able to trigger the delayed case.
2) The move_in_progress flag is not set and the old_domain cpu mask is not
empty.
That means, that an interrupt was delivered after the change and the
cleanup IPI has been sent to the cpus in old_domain, but not all CPUs have
responded to it yet.
In both cases we can assume that the next interrupt will arrive on the new
vector, so we can cleanup the old vectors on the cpus in the old_domain cpu
mask.
Fixes: 98229aa36c "x86/irq: Plug vector cleanup race"
Reported-by: Harry Junior <harryjr@outlook.fr>
Tested-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Joe Lawrence <joe.lawrence@stratus.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Ben Hutchings <ben@decadent.org.uk>
Cc: stable@vger.kernel.org
Link: http://lkml.kernel.org/r/alpine.DEB.2.11.1603140931430.3657@nanos
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
We still can end up with a stale vector due to the following:
CPU0 CPU1 CPU2
lock_vector()
data->move_in_progress=0
sendIPI()
unlock_vector()
set_affinity()
assign_irq_vector()
lock_vector() handle_IPI
move_in_progress = 1 lock_vector()
unlock_vector()
move_in_progress == 1
So we need to serialize the vector assignment against a pending cleanup. The
solution is rather simple now. We not only check for the move_in_progress flag
in assign_irq_vector(), we also check whether there is still a cleanup pending
in the old_domain cpumask. If so, we return -EBUSY to the caller and let him
deal with it. Though we have to be careful in the cpu unplug case. If the
cleanout has not yet completed then the following setaffinity() call would
return -EBUSY. Add code which prevents this.
Full context is here: http://lkml.kernel.org/r/5653B688.4050809@stratus.com
Reported-and-tested-by: Joe Lawrence <joe.lawrence@stratus.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Borislav Petkov <bp@alien8.de>
Cc: Jiang Liu <jiang.liu@linux.intel.com>
Cc: Jeremiah Mahler <jmmahler@gmail.com>
Cc: andy.shevchenko@gmail.com
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: stable@vger.kernel.org #4.3+
Link: http://lkml.kernel.org/r/20151231160107.207265407@linutronix.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
send_cleanup_vector() fiddles with the old_domain mask unprotected because it
relies on the protection by the move_in_progress flag. But this is fatal, as
the flag is reset after the IPI has been sent. So a cpu which receives the IPI
can still see the flag set and therefor ignores the cleanup request. If no
other cleanup request happens then the vector stays stale on that cpu and in
case of an irq removal the vector still persists. That can lead to use after
free when the next cleanup IPI happens.
Protect the code with vector_lock and clear move_in_progress before sending
the IPI.
This does not plug the race which Joe reported because:
CPU0 CPU1 CPU2
lock_vector()
data->move_in_progress=0
sendIPI()
unlock_vector()
set_affinity()
assign_irq_vector()
lock_vector() handle_IPI
move_in_progress = 1 lock_vector()
unlock_vector()
move_in_progress == 1
The full fix comes with a later patch.
Reported-and-tested-by: Joe Lawrence <joe.lawrence@stratus.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Borislav Petkov <bp@alien8.de>
Cc: Jiang Liu <jiang.liu@linux.intel.com>
Cc: Jeremiah Mahler <jmmahler@gmail.com>
Cc: andy.shevchenko@gmail.com
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: stable@vger.kernel.org #4.3+
Link: http://lkml.kernel.org/r/20151231160106.892412198@linutronix.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
