Prevent simultaneous controller disabling/enabling tasks from interfering
with each other through a function to wait until the task successfully
transitioned the controller to the RESETTING state. This ensures disabling
the controller will not be interrupted by another reset path, otherwise
a concurrent reset may leave the controller in the wrong state.
Tested-by: Edmund Nadolski <edmund.nadolski@intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
A paused controller is doing critical internal activation work in the
background. Prevent subsequent controller resets from occurring during
this period by setting the controller state to RESETTING first. A helper
function, nvme_try_sched_reset_work(), is introduced for these paths so
they may continue with scheduling the reset_work after they've completed
their uninterruptible critical section.
Tested-by: Edmund Nadolski <edmund.nadolski@intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
A controller in the resetting state has not yet completed its recovery
actions. The pci and fc transports were already handling this, so update
the remaining transports to not attempt additional recovery in this
state. Instead, just restart the request timer.
Tested-by: Edmund Nadolski <edmund.nadolski@intel.com>
Reviewed-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
The admin only state was intended to fence off actions that don't
apply to a non-IO capable controller. The only actual user of this is
the scan_work, and pci was the only transport to ever set this state.
The consequence of having this state is placing an additional burden on
every other action that applies to both live and admin only controllers.
Remove the admin only state and place the admin only burden on the only
place that actually cares: scan_work.
This also prepares to make it easier to temporarily pause a LIVE state
so that we don't need to remember which state the controller had been in
prior to the pause.
Tested-by: Edmund Nadolski <edmund.nadolski@intel.com>
Reviewed-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
If a controller becomes degraded after a reset, we will not be able to
perform any IO. We currently teardown previously created request
queues and namespaces, but we had kept the unusable tagset. Free
it after all queues using it have been released.
Tested-by: Edmund Nadolski <edmund.nadolski@intel.com>
Reviewed-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
The return code from null_handle_zoned() sets the cmd->error value.
Returning OK status when an error occured overwrites the intended
cmd->error. Return the appropriate error code instead of setting the
error in the cmd.
Fixes: fceb5d1b19 ("null_blk: create a helper for zoned devices")
Cc: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The namespace disk names must be unique for the lifetime of the
subsystem. This was accomplished by using their parent subsystems'
instances which were allocated independently from the controllers
connected to that subsystem. This allowed name prefixes assigned to
namespaces to match a controller from an unrelated subsystem, and has
created confusion among users examining device nodes.
Ensure a namespace's subsystem instance never clashes with a controller
instance of another subsystem by transferring the instance ownership
to the parent subsystem from the first controller discovered in that
subsystem.
Reviewed-by: Logan Gunthorpe <logang@deltatee.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Minwoo Im <minwoo.im@samsung.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Reviewed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Some of the memory nodes described in HMAT may not be online at the
time the HMAT subsystem parses their nodes' attributes. Should the node be
set to online later, as can happen when using PMEM as RAM after boot, the
nodes will be missing their initiator links and performance attributes.
Regsiter a memory notifier callback and register the memory attributes
the first time its node is brought online if it wasn't registered.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Instead of registering the HMAT cache attributes in line with parsing
the table, save the attributes in the memory target and register them
after parsing completes. This will make it easier to register the
attributes later when hot add is supported.
Tested-by: Brice Goglin <Brice.Goglin@inria.fr>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Ensure the controller is not in the NEW state when nvme_probe() exits.
This will always allow a subsequent nvme_remove() to set the state to
DELETING, fixing a potential race between the initial asynchronous probe
and device removal.
Reported-by: Li Zhong <lizhongfs@gmail.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
The nvme pci driver prepares its devices for power loss during suspend
by shutting down the controllers. The power setting is deferred to
pci driver's power management before the platform removes power. The
suspend-to-idle mode, however, does not remove power.
NVMe devices that implement host managed power settings can achieve
lower power and better transition latencies than using generic PCI power
settings. Try to use this feature if the platform is not involved with
the suspend. If successful, restore the previous power state on resume.
Tested-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
Tested-by: Mario Limonciello <mario.limonciello@dell.com>
Reviewed-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
[hch: fixed the compilation for the !CONFIG_PM_SLEEP case]
Signed-off-by: Christoph Hellwig <hch@lst.de>
We're trying to append known effects to the ones reported in the
controller's log. The original patch accomplished this, but something
went wrong when patch was merged causing the effects log to override
the known effects.
Link: http://lists.infradead.org/pipermail/linux-nvme/2019-May/023710.html
Fixes: f4524cc456 ("nvme-pci: add known admin effects to augument admin effects log page")
Cc: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Keith Busch <keith.busch@intel.com>
A controller with multiple namespaces may have multiple request_queues with
their own timeout work. If a controller fails with IO outstanding to
diffent namespaces, each request queue may attempt to handle it, so
ensure there is no previously scheduled timeout work executing prior to
starting controller initialization by synchronizing with each queue.
Reviewed-by: Minwoo Im <minwoo.im.dev@gmail.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <keith.busch@intel.com>
The reset_work waits for queued IO to complete before setting the
controller to live. If any of these times out and requeues, we won't be
able to restart the controller because the reset_work is already running.
Flush all entered requests to a failed completion if a timeout occurs
in the connecting state, and ensure the controller can't transition to
the live state after we've unblocked it from waiting for completions.
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <keith.busch@intel.com>
The reset state doesn't dispatch commands that it needs to wait for
anymore. If a timeout occurs in this state, the reset work is already
disabling the controller, so just reset the request's timer.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <keith.busch@intel.com>
If a controller disabling didn't start a freeze, don't wait for the
operation to complete.
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <keith.busch@intel.com>
e8303bb7a7 ("PCI/LINK: Report degraded links via link bandwidth
notification") added dmesg logging whenever a link changes speed or width
to a state that is considered degraded. Unfortunately, it cannot
differentiate signal integrity-related link changes from those
intentionally initiated by an endpoint driver, including drivers that may
live in userspace or VMs when making use of vfio-pci. Some GPU drivers
actively manage the link state to save power, which generates a stream of
messages like this:
vfio-pci 0000:07:00.0: 32.000 Gb/s available PCIe bandwidth, limited by 2.5 GT/s x16 link at 0000:00:02.0 (capable of 64.000 Gb/s with 5 GT/s x16 link)
Since we can't distinguish the intentional changes from the signal
integrity issues, leave the reporting turned off by default. Add a Kconfig
option to turn it on if desired.
Fixes: e8303bb7a7 ("PCI/LINK: Report degraded links via link bandwidth notification")
Link: https://lore.kernel.org/linux-pci/20190501142942.26972-1-keith.busch@intel.com
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Just like IO queues, the admin queue also will not be restarted after a
controller shutdown. Unquiesce this queue so that we do not block
request dispatch on a permanently disabled controller.
Reported-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
We do not restart a controller in a deleting state for timeout errors.
When in this state, unblock potential request dispatchers with failed
completions by shutting down the controller on timeout detection.
Reported-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
The nvme target hadn't been taking the Get Log Page offset parameter
into consideration, and so has been returning corrupted log pages when
offsets are used. Since many tools, including nvme-cli, split the log
request to 4k, we've been breaking discovery log responses when more
than 3 subsystems exist.
Fix the returned data by internally generating the entire discovery
log page and copying only the requested bytes into the user buffer. The
command log page offset type has been modified to a native __le64 to
make it easier to extract the value from a command.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Tested-by: Minwoo Im <minwoo.im@samsung.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
We don't need to save the dma device as it's not used in the hot path
and hasn't in a long time. Shrink the struct nvme_queue removing this
unnecessary member.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
A negative value for the cq_vector used to mean the queue is either
disabled or a polled queue. However, we have a queue enabled flag,
so the cq_vector had been serving double duty.
Don't overload the meaning of cq_vector. Use a flag specific to the
polled queues instead.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Platforms may provide system memory where some physical address ranges
perform differently than others, or is cached by the system on the
memory side.
Add documentation describing a high level overview of such systems and the
perforamnce and caching attributes the kernel provides for applications
wishing to query this information.
Reviewed-by: Mike Rapoport <rppt@linux.ibm.com>
Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Tested-by: Brice Goglin <Brice.Goglin@inria.fr>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Save the best performance access attributes and register these with the
memory's node if HMAT provides the locality table. While HMAT does make
it possible to know performance for all possible initiator-target
pairings, we export only the local pairings at this time.
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Tested-by: Brice Goglin <Brice.Goglin@inria.fr>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If the HMAT Subsystem Address Range provides a valid processor proximity
domain for a memory domain, or a processor domain matches the performance
access of the valid processor proximity domain, register the memory
target with that initiator so this relationship will be visible under
the node's sysfs directory.
Since HMAT requires valid address ranges have an equivalent SRAT entry,
verify each memory target satisfies this requirement.
Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Brice Goglin <Brice.Goglin@inria.fr>
Tested-by: Brice Goglin <Brice.Goglin@inria.fr>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
System memory may have caches to help improve access speed to frequently
requested address ranges. While the system provided cache is transparent
to the software accessing these memory ranges, applications can optimize
their own access based on cache attributes.
Provide a new API for the kernel to register these memory-side caches
under the memory node that provides it.
The new sysfs representation is modeled from the existing cpu cacheinfo
attributes, as seen from /sys/devices/system/cpu/<cpu>/cache/. Unlike CPU
cacheinfo though, the node cache level is reported from the view of the
memory. A higher level number is nearer to the CPU, while lower levels
are closer to the last level memory.
The exported attributes are the cache size, the line size, associativity
indexing, and write back policy, and add the attributes for the system
memory caches to sysfs stable documentation.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Brice Goglin <Brice.Goglin@inria.fr>
Tested-by: Brice Goglin <Brice.Goglin@inria.fr>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Heterogeneous memory systems provide memory nodes with different latency
and bandwidth performance attributes. Provide a new kernel interface
for subsystems to register the attributes under the memory target
node's initiator access class. If the system provides this information,
applications may query these attributes when deciding which node to
request memory.
The following example shows the new sysfs hierarchy for a node exporting
performance attributes:
# tree -P "read*|write*"/sys/devices/system/node/nodeY/accessZ/initiators/
/sys/devices/system/node/nodeY/accessZ/initiators/
|-- read_bandwidth
|-- read_latency
|-- write_bandwidth
`-- write_latency
The bandwidth is exported as MB/s and latency is reported in
nanoseconds. The values are taken from the platform as reported by the
manufacturer.
Memory accesses from an initiator node that is not one of the memory's
access "Z" initiator nodes linked in the same directory may observe
different performance than reported here. When a subsystem makes use
of this interface, initiators of a different access number may not have
the same performance relative to initiators in other access numbers, or
omitted from the any access class' initiators.
Descriptions for memory access initiator performance access attributes
are added to sysfs stable documentation.
Acked-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Tested-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Tested-by: Brice Goglin <Brice.Goglin@inria.fr>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Systems may be constructed with various specialized nodes. Some nodes
may provide memory, some provide compute devices that access and use
that memory, and others may provide both. Nodes that provide memory are
referred to as memory targets, and nodes that can initiate memory access
are referred to as memory initiators.
Memory targets will often have varying access characteristics from
different initiators, and platforms may have ways to express those
relationships. In preparation for these systems, provide interfaces for
the kernel to export the memory relationship among different nodes memory
targets and their initiators with symlinks to each other.
If a system provides access locality for each initiator-target pair, nodes
may be grouped into ranked access classes relative to other nodes. The
new interface allows a subsystem to register relationships of varying
classes if available and desired to be exported.
A memory initiator may have multiple memory targets in the same access
class. The target memory's initiators in a given class indicate the
nodes access characteristics share the same performance relative to other
linked initiator nodes. Each target within an initiator's access class,
though, do not necessarily perform the same as each other.
A memory target node may have multiple memory initiators. All linked
initiators in a target's class have the same access characteristics to
that target.
The following example show the nodes' new sysfs hierarchy for a memory
target node 'Y' with access class 0 from initiator node 'X':
# symlinks -v /sys/devices/system/node/nodeX/access0/
relative: /sys/devices/system/node/nodeX/access0/targets/nodeY -> ../../nodeY
# symlinks -v /sys/devices/system/node/nodeY/access0/
relative: /sys/devices/system/node/nodeY/access0/initiators/nodeX -> ../../nodeX
The new attributes are added to the sysfs stable documentation.
Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Tested-by: Brice Goglin <Brice.Goglin@inria.fr>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Parsing entries in an ACPI table had assumed a generic header
structure. There is no standard ACPI header, though, so less common
layouts with different field sizes required custom parsers to go through
their subtable entry list.
Create the infrastructure for adding different table types so parsing
the entries array may be more reused for all ACPI system tables and
the common code doesn't need to be duplicated.
Reviewed-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Tested-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Tested-by: Brice Goglin <Brice.Goglin@inria.fr>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The field is defined to be a 24 byte array, we don't need to multiply
the sizeof() that field by the number of dwords it covers.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
A write or flush IO passthrough command is expected to change the
logical block content, so don't warn on these as no additional handling
is necessary.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The reset work holds a mutex to prevent races with removal modifying the
same resources, but was unlocking only on success. Unlock on failure
too.
Fixes: 5c959d73db ("nvme-pci: fix rapid add remove sequence")
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
A surprise removal may fail to tear down request queues if it is racing
with the initial asynchronous probe. If that happens, the remove path
won't see the queue resources to tear down, and the controller reset
path may create a new request queue on a removed device, but will not
be able to make forward progress, deadlocking the pci removal.
Protect setting up non-blocking resources from a shutdown by holding the
same mutex, and transition to the CONNECTING state after these resources
are initialized so the probe path may see the dead controller state
before dispatching new IO.
Link: https://bugzilla.kernel.org/show_bug.cgi?id=202081
Reported-by: Alex Gagniuc <Alex_Gagniuc@Dellteam.com>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Tested-by: Alex Gagniuc <mr.nuke.me@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
If a controller supports the NS Change Notification, the namespace
scan_work is automatically triggered after attaching a new namespace.
Occasionally the namespace scan_work may append the new namespace to the
list before the admin command effects handling is completed. The effects
handling unfreezes namespaces, but if it unfreezes the newly attached
namespace, its request_queue freeze depth will be off and we'll hit the
warning in blk_mq_unfreeze_queue().
On the next namespace add, we will fail to freeze that queue due to the
previous bad accounting and deadlock waiting for frozen.
Fix that by preventing scan work from altering the namespace list while
command effects handling needs to pair freeze with unfreeze.
Reported-by: Wen Xiong <wenxiong@us.ibm.com>
Tested-by: Wen Xiong <wenxiong@us.ibm.com>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
We need to preserve the leading zeros in the vid and ssvid when generating
a unique NQN. Truncating these may lead to naming collisions.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
A controller may have an internal state that is not able to successfully
process commands for a short duration. In such states, an immediate
command requeue is expected to fail. The driver may exceed its max
retry count, which permanently ends the command in failure when the same
command would succeed after waiting for the controller to be ready.
NVMe ratified TP 4033 provides a delay hint in the completion status
code for failed commands. Implement the retry delay based on the command
completion status and the controller's requested delay.
Note that requeued commands are handled per request_queue, not per
individual request. If multiple commands fail, the controller should
consistently report the desired delay time for retryable commands in
all CQEs, otherwise the requeue list may be kicked too soon.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Free the kobject name that was allocated for the controller device on
failure rather than its parent.
Fixes: d22524a478 ("nvme: switch controller refcounting to use struct device")
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
There are no more users relying on blk-mq request states to prevent
double completions, so replace the relatively expensive cmpxchg operation
with WRITE_ONCE.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The scsi timeout error handling had been directly updating the block
layer's request state to prevent a error handling and a natural completion
from completing the same request twice. Fix this layering violation
by having scsi control the fate of its commands with scsi owned flags
rather than use blk-mq's.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
A driver may have internal state to cleanup if we're pretending a request
didn't complete. Return 'false' if the command wasn't actually completed
due to the timeout error injection, and true otherwise.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If the kernel allocates a bounce buffer for user read data, this memory
needs to be cleared before copying it to the user, otherwise it may leak
kernel memory to user space.
Laurence Oberman <loberman@redhat.com>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The nvme pci driver had been adding its CMB resource to the P2P DMA
subsystem everytime on on a controller reset. This results in the
following warning:
------------[ cut here ]------------
nvme 0000:00:03.0: Conflicting mapping in same section
WARNING: CPU: 7 PID: 81 at kernel/memremap.c:155 devm_memremap_pages+0xa6/0x380
...
Call Trace:
pci_p2pdma_add_resource+0x153/0x370
nvme_reset_work+0x28c/0x17b1 [nvme]
? add_timer+0x107/0x1e0
? dequeue_entity+0x81/0x660
? dequeue_entity+0x3b0/0x660
? pick_next_task_fair+0xaf/0x610
? __switch_to+0xbc/0x410
process_one_work+0x1cf/0x350
worker_thread+0x215/0x3d0
? process_one_work+0x350/0x350
kthread+0x107/0x120
? kthread_park+0x80/0x80
ret_from_fork+0x1f/0x30
---[ end trace f7ea76ac6ee72727 ]---
nvme nvme0: failed to register the CMB
This patch fixes this by registering the CMB with P2P only once.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Logan Gunthorpe <logang@deltatee.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Getting pages from ZONE_DEVICE memory needs to check the backing device's
live-ness, which is tracked in the device's dev_pagemap metadata. This
metadata is stored in a radix tree and looking it up adds measurable
software overhead.
This patch avoids repeating this relatively costly operation when
dev_pagemap is used by caching the last dev_pagemap while getting user
pages. The gup_benchmark kernel self test reports this reduces time to
get user pages to as low as 1/3 of the previous time.
Link: http://lkml.kernel.org/r/20181012173040.15669-1-keith.busch@intel.com
Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Move the bus ops fallback into separate functions. No functional change
here.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
The aer_inject module was directly calling aer_irq(). This required the
AER driver export its private IRQ handler for no other reason than to
support error injection. A driver should not have to expose its private
interfaces, so use the IRQ subsystem to route injection to the AER driver,
and make aer_irq() a private interface.
This provides additional benefits:
First, directly calling the IRQ handler bypassed the IRQ subsytem so the
injection wasn't really synthesizing what happens if a shared AER interrupt
occurs.
The error injection had to provide the callback data directly, which may be
racing with a removal that is freeing that structure. The IRQ subsystem
can handle that race.
Finally, using the IRQ subsystem automatically reacts to threaded IRQs,
keeping the error injection abstracted from that implementation detail.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
The port services driver already provides a method to find the pcie_device
for a service. Export that function, use it from the aer_inject module,
and remove the duplicate functionality.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
A removal waits for the reset_work to complete. If a surprise removal
occurs around the same time as an error triggered controller reset, and
reset work happened to dispatch a command to the removed controller, the
command won't be recovered since the timeout work doesn't do anything
during error recovery. We wouldn't want to wait for timeout handling
anyway, so this patch fixes this by disabling the controller and killing
admin queues prior to syncing with the reset_work.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
The nvme namespace paths were being updated only when the current path
was not set or nonoptimized. If a new path comes online that is a better
path for its NUMA node, the multipath selector may continue using the
previously set path on a potentially further node.
This patch re-runs the path assignment after successfully adding a new
optimized path.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
The threaded IRQ is naturally single threaded as desired, so use that to
simplify the AER bottom half handler. Since the root port structure has
much less to do now, remove the rpc construction helper routine.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Use the recommended kernel API for writing to a concurrently-accessed
kfifo. No functional change here.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
The kernel provides a generic FIFO implementation, so no need to reinvent
that capability in a driver. Replace the AER-specific implementation with
the kernel-provided kfifo. Since the interrupt handler producer and work
queue consumer run single threaded, there is no need for additional
locking, so remove that lock, too.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
The AER struct aer_rpc was carrying a copy of the error source simply as a
temperary variable. Remove that from the structure and use a stack
variable for the purpose.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
The error recovery callbacks are only run on child devices. A Root Port is
never a child device, so this error resume callback was never invoked.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
The code had been clearing a namespace being deleted as the current
path while that namespace was still in the path siblings list. It is
possible a new IO could set that namespace back to the current path
since it appeared to be an eligable path to select, which may result in
a use-after-free error.
This patch ensures a namespace being removed is not eligable to be reset
as a current path prior to clearing it as the current path.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
The PCI bus config accessors could be inlined into other accessor
functions, which makes it so they can't be traced. Force them to never be
inlined so that ftrace can hook into these functions.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
The spec has timing requirements when waiting for a link to become active
after a conventional reset. Implement those hard delays when waiting for
an active link so pciehp and dpc drivers don't need to duplicate this.
For devices that don't support data link layer active reporting, wait the
fixed time recommended by the PCIe spec.
Signed-off-by: Keith Busch <keith.busch@intel.com>
[bhelgaas: changelog]
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Sinan Kaya <okaya@kernel.org>
Bring surprise removals and permanent failures together so we no longer
need separate flags. The implementation enforces that error handling will
not be able to override a surprise removal's permanent channel failure.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Sinan Kaya <okaya@kernel.org>
A device still participates in error recovery even if it doesn't have
the error callbacks.
Always provide the status for user event watchers.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Sinan Kaya <okaya@kernel.org>
There is no point in having a generic broadcast function if it needs to
have special cases for each callback it broadcasts.
Abstract the error broadcast to only the necessary information and removes
the now unnecessary helper to walk the bus.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Sinan Kaya <okaya@kernel.org>
If an Endpoint reported an error with ERR_FATAL, we previously ran driver
error recovery callbacks only for the Endpoint's driver. But if we reset a
Link to recover from the error, all downstream components are affected,
including the Endpoint, any multi-function peers, and children of those
peers.
Initiate the Link reset from the deepest Downstream Port that is
reliable, and call the error recovery callbacks for all its children.
If a Downstream Port (including a Root Port) reports an error, we assume
the Port itself is reliable and we need to reset its downstream Link. In
all other cases (Switch Upstream Ports, Endpoints, Bridges, etc), we assume
the Link leading to the component needs to be reset, so we initiate the
reset at the parent Downstream Port.
This allows two other clean-ups. First, we currently only use a Link
reset, which can only be initiated using a Downstream Port, so we can
remove checks for Endpoints. Second, the Downstream Port where we initiate
the Link reset is reliable (unlike components downstream from it), so the
special cases for error detect and resume are no longer necessary.
Signed-off-by: Keith Busch <keith.busch@intel.com>
[bhelgaas: changelog]
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Sinan Kaya <okaya@kernel.org>
We don't need to be paranoid about the topology changing while handling an
error. If the device has changed in a hotplug capable slot, we can rely on
the presence detection handling to react to a changing topology.
Restore the fatal error handling behavior that existed before merging DPC
with AER with 7e9084b367 ("PCI/AER: Handle ERR_FATAL with removal and
re-enumeration of devices").
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Sinan Kaya <okaya@kernel.org>
A recent commit runs tag iterator callbacks under the rcu read lock,
but existing callbacks do not satisfy the non-blocking requirement.
The commit intended to prevent an iterator from accessing a queue that's
being modified. This patch fixes the original issue by taking a queue
reference instead of reading it, which allows callbacks to make blocking
calls.
Fixes: f5bbbbe4d6 ("blk-mq: sync the update nr_hw_queues with blk_mq_queue_tag_busy_iter")
Acked-by: Jianchao Wang <jianchao.w.wang@oracle.com>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The AER driver has never read the config space of an endpoint that reported
a fatal error because the link to that device is considered unreliable.
An ERR_FATAL from an upstream port almost certainly indicates an error on
its upstream link, so we can't expect to reliably read its config space for
the same reason.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Sinan Kaya <okaya@kernel.org>
Error handling may be running in parallel with a hot removal. Reference
count the device during AER handling so the device can not be freed while
AER wants to reference it.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Sinan Kaya <okaya@kernel.org>
This patch provides DPC save and restore capabilities. This is necessary
for the driver to observe DPC events in the event the configuration space
needs to be restored after a reset.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Sinan Kaya <okaya@kernel.org>
The port's config space may be cleared after a link reset, which wipes out
the bridge's bus and memory windows. Restore the config space that was
saved during probe so we can access downstream devices.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Sinan Kaya <okaya@kernel.org>
The PCI port driver saves the PCI state after initializing the device with
the applicable service devices. This was, however, before the service
drivers were even registered because PCI probe happens before the
device_initcall initialized those service drivers. The config space state
that the services set up were not being saved. The end result would cause
PCI devices to not react to events that the drivers think they did if the
PCI state ever needed to be restored.
Fix this by changing the service drivers from using the init calls to
having the portdrv driver calling the services directly. This will get the
state saved as desired, while making the relationship between the port
driver and the services under it more explicit in the code.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Sinan Kaya <okaya@kernel.org>
If both hot-add and power fault were observed in a single interrupt, we
handled the hot-add first, then the power fault, in this path:
pciehp_ist
if (events & (PDC | DLLSC))
pciehp_handle_presence_or_link_change
case OFF_STATE:
pciehp_enable_slot
__pciehp_enable_slot
board_added
pciehp_power_on_slot
ctrl->power_fault_detected = 0
pcie_write_cmd(ctrl, PCI_EXP_SLTCTL_PWR_ON, PCI_EXP_SLTCTL_PCC)
pciehp_green_led_on(p_slot) # power LED on
pciehp_set_attention_status(p_slot, 0) # attention LED off
if ((events & PFD) && !ctrl->power_fault_detected)
ctrl->power_fault_detected = 1
pciehp_set_attention_status(1) # attention LED on
pciehp_green_led_off(slot) # power LED off
This left the attention indicator on (even though the hot-add succeeded)
and the power indicator off (even though the slot power was on).
Fix this by checking for power faults before checking for new devices.
Prior to 0e94916e60, this was successful because everything was chained
through work queues and the order was:
INT_PRESENCE_ON -> INT_POWER_FAULT -> ENABLE_REQ
The ENABLE_REQ cleared the power fault at the end, but now everything is
handled inline with the interrupt thread, such that the work ENABLE_REQ was
doing happens before power fault handling now.
Fixes: 0e94916e60 ("PCI: pciehp: Handle events synchronously")
Signed-off-by: Keith Busch <keith.busch@intel.com>
[bhelgaas: changelog]
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Lukas Wunner <lukas@wunner.de>
The 'available_size' attribute showing the combined total of all
unallocated space isn't always useful to know how large of a namespace
a user may be able to allocate if the region is fragmented. This patch
will export the largest extent of unallocated space that may be allocated
to create a new namespace.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Vishal Verma <vishal.l.verma@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
This patch will find the max contiguous area to determine the largest
pmem namespace size that can be created. If the requested size exceeds
the largest available, ENOSPC error will be returned.
This fixes the allocation underrun error and wrong error return code
that have otherwise been observed as the following kernel warning:
WARNING: CPU: <CPU> PID: <PID> at drivers/nvdimm/namespace_devs.c:913 size_store
Fixes: a1f3e4d6a0 ("libnvdimm, region: update nd_region_available_dpa() for multi-pmem support")
Cc: <stable@vger.kernel.org>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Vishal Verma <vishal.l.verma@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
The scsi block layer requires requests claimed by the error handling be
completed by the error handler. A previous commit allowed completions
to proceed for blk-mq, breaking that assumption.
This patch prevents completions that may race with the timeout handler
by marking the state to complete, restoring the previous behavior.
Fixes: 12f5b931 ("blk-mq: Remove generation seqeunce")
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This is preparing for drivers that want to directly alter the state of
their requests. No functional change here.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This will print the disk name to the nvme event trace for io requests so
a user can better distinguish traffic to different disks. This can be used
to create disk based filters. For example, to see only nvme0n2 traffic:
echo "disk == \"nvme0n2\"" > /sys/kernel/debug/tracing/events/nvme/filter
Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
[hch: turned __assign_disk_name into an inline function]
Signed-off-by: Christoph Hellwig <hch@lst.de>
This appends the controller instance to the nvme trace buffer to
distinguish which controller is dispatching and completing a command.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
We can not match a command to its completion based on the command
id alone. We need the submitting queue identifier to pair with the
completion, so this patch adds that to the trace buffer.
This patch is also collapsing the admin and IO submission traces into a
single one so we don't need to duplicate this and creating unnecessary
code branches: we know if the command is an admin vs IO based on the qid.
And since we're here, the patch fixes code formatting in the area.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
[hch: move the qid helper to nvme.h and made it an inline function]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Simplify waiting for the contained link to become inactive, removing the
indirection to a unnecessary DPC-specific handler.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Sinan Kaya <okaya@kernel.org>
Reviewed-by: Oza Pawandeep <poza@codeaurora.org>
A DPC enabled device suppresses ERR_(NON)FATAL messages, preventing the AER
handler from reporting error details. If the DPC trigger reason says the
downstream port detected the error, collect the AER uncorrectable status
for logging, then clear the status.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Sinan Kaya <okaya@kernel.org>
Reviewed-by: Oza Pawandeep <poza@codeaurora.org>
We don't need to save the rp pio status across multiple contexts as all
DPC event handling occurs in a single work queue context.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Sinan Kaya <okaya@kernel.org>
Reviewed-by: Oza Pawandeep <poza@codeaurora.org>
Move all event handling to the existing work queue, which will
make it simpler to pass event information to the handler.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Sinan Kaya <okaya@kernel.org>
Reviewed-by: Oza Pawandeep <poza@codeaurora.org>
Now that the DPC driver clears the interrupt status before exiting the
IRQ handler, we don't need to abuse the DPC control register to know if
a shared interrupt is for a new DPC event: a DPC port can not trigger
a second interrupt until the host clears the trigger status later in the
work queue handler.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Sinan Kaya <okaya@kernel.org>
Reviewed-by: Oza Pawandeep <poza@codeaurora.org>
Keith Busch