On large systems with multiple devices of the same class (e.g. NVMe disks,
using managed interrupts), the kernel can affinitize these interrupts to a
small subset of CPUs instead of spreading them out evenly.
irq_matrix_alloc_managed() tries to select the CPU in the supplied cpumask
of possible target CPUs which has the lowest number of interrupt vectors
allocated.
This is done by searching the CPU with the highest number of available
vectors. While this is correct for non-managed CPUs it can select the wrong
CPU for managed interrupts. Under certain constellations this results in
affinitizing the managed interrupts of several devices to a single CPU in
a set.
The book keeping of available vectors works the following way:
1) Non-managed interrupts:
available is decremented when the interrupt is actually requested by
the device driver and a vector is assigned. It's incremented when the
interrupt and the vector are freed.
2) Managed interrupts:
Managed interrupts guarantee vector reservation when the MSI/MSI-X
functionality of a device is enabled, which is achieved by reserving
vectors in the bitmaps of the possible target CPUs. This reservation
decrements the available count on each possible target CPU.
When the interrupt is requested by the device driver then a vector is
allocated from the reserved region. The operation is reversed when the
interrupt is freed by the device driver. Neither of these operations
affect the available count.
The reservation persist up to the point where the MSI/MSI-X
functionality is disabled and only this operation increments the
available count again.
For non-managed interrupts the available count is the correct selection
criterion because the guaranteed reservations need to be taken into
account. Using the allocated counter could lead to a failing allocation in
the following situation (total vector space of 10 assumed):
CPU0 CPU1
available: 2 0
allocated: 5 3 <--- CPU1 is selected, but available space = 0
managed reserved: 3 7
while available yields the correct result.
For managed interrupts the available count is not the appropriate
selection criterion because as explained above the available count is not
affected by the actual vector allocation.
The following example illustrates that. Total vector space of 10
assumed. The starting point is:
CPU0 CPU1
available: 5 4
allocated: 2 3
managed reserved: 3 3
Allocating vectors for three non-managed interrupts will result in
affinitizing the first two to CPU0 and the third one to CPU1 because the
available count is adjusted with each allocation:
CPU0 CPU1
available: 5 4 <- Select CPU0 for 1st allocation
--> allocated: 3 3
available: 4 4 <- Select CPU0 for 2nd allocation
--> allocated: 4 3
available: 3 4 <- Select CPU1 for 3rd allocation
--> allocated: 4 4
But the allocation of three managed interrupts starting from the same
point will affinitize all of them to CPU0 because the available count is
not affected by the allocation (see above). So the end result is:
CPU0 CPU1
available: 5 4
allocated: 5 3
Introduce a "managed_allocated" field in struct cpumap to track the vector
allocation for managed interrupts separately. Use this information to
select the target CPU when a vector is allocated for a managed interrupt,
which results in more evenly distributed vector assignments. The above
example results in the following allocations:
CPU0 CPU1
managed_allocated: 0 0 <- Select CPU0 for 1st allocation
--> allocated: 3 3
managed_allocated: 1 0 <- Select CPU1 for 2nd allocation
--> allocated: 3 4
managed_allocated: 1 1 <- Select CPU0 for 3rd allocation
--> allocated: 4 4
The allocation of non-managed interrupts is not affected by this change and
is still evaluating the available count.
The overall distribution of interrupt vectors for both types of interrupts
might still not be perfectly even depending on the number of non-managed
and managed interrupts in a system, but due to the reservation guarantee
for managed interrupts this cannot be avoided.
Expose the new field in debugfs as well.
[ tglx: Clarified the background of the problem in the changelog and
described it independent of NVME ]
Signed-off-by: Long Li <longli@microsoft.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Michael Kelley <mikelley@microsoft.com>
Link: https://lkml.kernel.org/r/20181106040000.27316-1-longli@linuxonhyperv.com
IRQ_MATRIX_SIZE is the number of longs needed for a bitmap, multiplied by
the size of a long, yielding a byte count. But it is used to size an array
of longs, which is way more memory than is needed.
Change IRQ_MATRIX_SIZE so it is just the number of longs needed and the
arrays come out the correct size.
Fixes: 2f75d9e1c9 ("genirq: Implement bitmap matrix allocator")
Signed-off-by: Michael Kelley <mikelley@microsoft.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: KY Srinivasan <kys@microsoft.com>
Link: https://lkml.kernel.org/r/1541032428-10392-1-git-send-email-mikelley@microsoft.com
Linux spreads out the non managed interrupt across the possible target CPUs
to avoid vector space exhaustion.
Managed interrupts are treated differently, as for them the vectors are
reserved (with guarantee) when the interrupt descriptors are initialized.
When the interrupt is requested a real vector is assigned. The assignment
logic uses the first CPU in the affinity mask for assignment. If the
interrupt has more than one CPU in the affinity mask, which happens when a
multi queue device has less queues than CPUs, then doing the same search as
for non managed interrupts makes sense as it puts the interrupt on the
least interrupt plagued CPU. For single CPU affine vectors that's obviously
a NOOP.
Restructre the matrix allocation code so it does the 'best CPU' search, add
the sanity check for an empty affinity mask and adapt the call site in the
x86 vector management code.
[ tglx: Added the empty mask check to the core and improved change log ]
Signed-off-by: Dou Liyang <douly.fnst@cn.fujitsu.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: hpa@zytor.com
Link: https://lkml.kernel.org/r/20180908175838.14450-2-dou_liyang@163.com
Linux finds the CPU which has the lowest vector allocation count to spread
out the non managed interrupts across the possible target CPUs, but does
not do so for managed interrupts.
Split out the CPU selection code into a helper function for reuse. No
functional change.
Signed-off-by: Dou Liyang <douly.fnst@cn.fujitsu.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: hpa@zytor.com
Link: https://lkml.kernel.org/r/20180908175838.14450-1-dou_liyang@163.com
Use the proper SPDX-Identifier format.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Cc: Kate Stewart <kstewart@linuxfoundation.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Link: https://lkml.kernel.org/r/20180314212030.492674761@linutronix.de
At CPU hotunplug the corresponding per cpu matrix allocator is shut down and
the allocated interrupt bits are discarded under the assumption that all
allocated bits have been either migrated away or shut down through the
managed interrupts mechanism.
This is not true because interrupts which are not started up might have a
vector allocated on the outgoing CPU. When the interrupt is started up
later or completely shutdown and freed then the allocated vector is handed
back, triggering warnings or causing accounting issues which result in
suspend failures and other issues.
Change the CPU hotplug mechanism of the matrix allocator so that the
remaining allocations at unplug time are preserved and global accounting at
hotplug is correctly readjusted to take the dormant vectors into account.
Fixes: 2f75d9e1c9 ("genirq: Implement bitmap matrix allocator")
Reported-by: Yuriy Vostrikov <delamonpansie@gmail.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Yuriy Vostrikov <delamonpansie@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/20180222112316.849980972@linutronix.de
Keith reported an issue with vector space exhaustion on a server machine
which is caused by the i40e driver allocating 168 MSI interrupts when the
driver is initialized, even when most of these interrupts are not used at
all.
The x86 vector allocation code tries to avoid the immediate allocation with
the reservation mode, but the card uses MSI and does not support MSI entry
masking, which prevents reservation mode and requires immediate vector
allocation.
The matrix allocator is a bit naive and prefers the first CPU in the
cpumask which describes the possible target CPUs for an allocation. That
results in allocating all 168 vectors on CPU0 which later causes vector
space exhaustion when the NVMe driver tries to allocate managed interrupts
on each CPU for the per CPU queues.
Avoid this by finding the CPU which has the lowest vector allocation count
to spread out the non managed interrupt accross the possible target CPUs.
Fixes: 2f75d9e1c9 ("genirq: Implement bitmap matrix allocator")
Reported-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Keith Busch <keith.busch@intel.com>
Link: https://lkml.kernel.org/r/alpine.DEB.2.20.1801171557330.1777@nanos
The previous commit which made the operator precedence in
irq_matrix_available() explicit made the implicit brokenness explicitely
wrong. It was wrong in the original commit already. The overworked
maintainer did not notice it either when merging the patch.
Replace the confusing '?' construct by a simple and obvious if ().
Fixes: 75f1133873 ("genirq/matrix: Make - vs ?: Precedence explicit")
Reported-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Kees Cook <keescook@chromium.org>
Noticed with a Clang build. This improves the readability of the ?:
expression, as it has lower precedence than the - expression. Show
explicitly that - is evaluated first.
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/20171122205645.GA27125@beast
Add tracepoints for the irq bitmap matrix allocator.
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/20170913213153.279468022@linutronix.de
Implement the infrastructure for a simple bitmap based allocator, which
will replace the x86 vector allocator. It's in the core code as other
architectures might be able to reuse/extend it. For now it only implements
allocations for single CPUs, but it's simple to add multi CPU allocation
support if required.
The concept is rather simple:
Global information:
system_vector bitmap
global accounting
PerCPU information:
allocation bitmap
managed allocation bitmap
local accounting
The system vector bitmap is used to exclude vectors system wide from the
allocation space.
The allocation bitmap is used to keep track of per cpu used vectors.
The managed allocation bitmap is used to reserve vectors for managed
interrupts.
When a regular (non managed) interrupt allocation happens then the
following rule applies:
tmpmap = system_map | alloc_map | managed_map
find_zero_bit(tmpmap)
Oring the bitmaps together gives the real available space. The same rule
applies for reserving a managed interrupt vector. But contrary to the
regular interrupts the reservation only marks the bit in the managed map
and therefor excludes it from the regular allocations. The managed map is
only cleaned out when the a managed interrupt is completely released and it
stays alive accross CPU offline/online operations.
For managed interrupt allocations the rule is:
tmpmap = managed_map & ~alloc_map
find_first_bit(tmpmap)
This returns the first bit which is in the managed map, but not yet
allocated in the allocation map. The allocation marks it in the allocation
map and hands it back to the caller for use.
The rest of the code are helper functions to handle the various
requirements and the accounting which are necessary to replace the x86
vector allocation code. The result is a single patch as the evolution of
this infrastructure cannot be represented in bits and pieces.
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: Chris Metcalf <cmetcalf@mellanox.com>
Cc: Len Brown <lenb@kernel.org>
Link: https://lkml.kernel.org/r/20170913213153.185437174@linutronix.de
Long Li