We need to update the nvme queue's wait_queue_t entry during each
initialization since the nvme_thread may be ended and restarted when
the device is reset. If a device reset occurs during a large amount
of buffered IO, it would take a lot longer to complete the outstanding
requests due to the 1 second polling instead of waking up as completions
occur.

Fixes: b9afca3eSigned-off-by: NKeith Busch <keith.busch@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

This returns a more appropriate error for the "capacity exceeded"
status. In case other NVMe statuses have a better errno, this patch adds
a convience function to translate an NVMe status code to an errno for
IO commands, defaulting to the current -EIO.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

We can return -ENOIOCTLCMD and the ioctl will be handled by
fs/compat_ioctl.c instead. This removes a lot of duplicate code in the
nvme driver.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

If an nvme device is removed but user space has an open reference,
the nvme driver would have been holding an invalid reference to its pci
device. You may get a general protection fault on x86 h/w when the driver
uses that reference in dma_map_sg(), as is done in nvme_map_user_pages()
from the IOCTL interface.

This patch fixes the fault by taking a reference on the pci device and
holding it even after device removal until all opens on the nvme device
are closed.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Reported-by: NNilesh Choudhury <nilesh.choudhury@oracle.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

The use of "rcu_assign_pointer()" is NULLing out the pointer.
According to RCU_INIT_POINTER()'s block comment:
"1.   This use of RCU_INIT_POINTER() is NULLing out the pointer"
it is better to use it instead of rcu_assign_pointer() because it has a
smaller overhead.

The following Coccinelle semantic patch was used:
@@
@@

- rcu_assign_pointer
+ RCU_INIT_POINTER
  (..., NULL)
Signed-off-by: NAndreea-Cristina Bernat <bernat.ada@gmail.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

nvme_submit_io_cmd() uses smp_processor_id() to pick an IO queue index.
This patch fixes the case where there are more cpus from which the ioctl
call can originate than online queues, which can happen when a device
supports or was allocated fewer interrupt vectors than exist cpu cores.

Thanks to Keith Busch for the implementation suggestion.
Signed-off-by: NSam Bradshaw <sbradshaw@micron.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

This changes the order of deleting the gendisks so it happens after the
nvme IO queues are freed. If a device is removed while a filesystem has
associated dirty data, the removal will wait on these to complete before
proceeding from del_gendisk, which could have caused deadlock before.

The implication of this is that an orderly removal of a responsive
device won't necessarily wait for dirty data to be written, but we are
not guaranteed the device is even going to respond at this point either.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

Rather than relying on call_rcu, this patch directly frees the
nvme_queue's memory after ensuring no readers exist. Some arch specific
dma_free_coherent implementations may not be called from a call_rcu's
soft interrupt context, hence the change.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Reported-by: NMatthew Minter <matthew_minter@xyratex.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

The current implementation hard-codes the shutdown timeout to 2 seconds.
Some devices take longer than this to complete a normal shutdown.
Changing the shutdown timeout to a module parameter with a default
timeout of 5 seconds.
Signed-off-by: NDan McLeran <daniel.mcleran@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

Rather than skipping shutdown only for devices that have been removed,
skip the orderly shutdown on failed devices to avoid the long timeout
handling that inevitably happens when deleting queues on such a device.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

Fixing tabs inadvertently converted to spaces.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

Race conditions are theoretically possible between the NVMe PCI device
removal and the generic PCI bus rescan and device removal that can be
triggered via sysfs.

To avoid those race conditions make the NVMe code use
pci_stop_and_remove_bus_device_locked().
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

This is a minor refactor for handling devices that are incapable of IO.
The driver previously used special error codes to know that IO queues
are unavailable, but we have an online queue count now.

This also fixes an issue where the driver successfully sets the queue
count, but either is unable to allocate an IO queue or the device can't
create one for some reason.

If the driver can successfully enable the device and get responses to
admin commands, the driver will bring up a character device for managment
but not create block devices.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

Change the behavior of nvme_enable_ctrl to set EN.
Clear CC.SH for both nvme_enable_ctrl and nvme_disable_ctrl.
Remove reading of the CC register and manage the state in
dev->ctrl_config.
Signed-off-by: NDan McLeran <daniel.mcleran@intel.com>
[removed an unwanted write to CC]
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

Adds support for devices with max page size smaller than the host's.
In the case we encounter such a host/device combination, the driver will
split a page into as many PRP entries as necessary for the device's page
size capabilities. If the device's reported minimum page size is greater
than the host's, the driver will not attempt to enable the device and
return an error instead.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

Submits NVMe asynchronous event requests, one event up to the controller
maximum or number of possible different event types (8), whichever is
smaller. Events successfully returned by the controller are logged.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

Use the zeroing function instead of dma_alloc_coherent & memset(,0,)
Signed-off-by: NJoe Perches <joe@perches.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

Clear QUEUE_FLAG_ADD_RANDOM in all block drivers that set
QUEUE_FLAG_NONROT.

Historically, all block devices have automatically made entropy
contributions.  But as previously stated in commit e2e1a148 ("block: add
sysfs knob for turning off disk entropy contributions"):
    - On SSD disks, the completion times aren't as random as they
      are for rotational drives. So it's questionable whether they
      should contribute to the random pool in the first place.
    - Calling add_disk_randomness() has a lot of overhead.

There are more reliable sources for randomness than non-rotational block
devices.  From a security perspective it is better to err on the side of
caution than to allow entropy contributions from unreliable "random"
sources.
Signed-off-by: NMike Snitzer <snitzer@redhat.com>
Signed-off-by: NJens Axboe <axboe@fb.com>

There is a potential dead lock if a cpu event occurs during nvme probe
since it registered with hot cpu notification. This fixes the race by
having the module register with notification outside of probe rather
than have each device register.

The actual work is done in a scheduled work queue instead of in the
notifier since assigning IO queues has the potential to block if the
driver creates additional queues.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>

It's positively immoral to have a global variable called 'io_timeout'.
Keep the module parameter called io_timeout, though.
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>

Recently, a new sysfs control "iostats" was added to selectively
enable or disable io statistics collection for request queues.  This
patch hooks that control.

IO statistics collection is rather expensive on large, multi-node
machines with drives pushing millions of iops.  Having the ability to
disable collection if not needed can improve throughput significantly.

As a data point, on a quad E5-4640, I see more than 50% throughput
improvement when io statistics accounting is disabled during heavily
multi-threaded small block random read benchmarks where device
performance is in the million iops+ range.
Signed-off-by: NSam Bradshaw <sbradshaw@micron.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>

The routines to get and lock nvme queues required the caller to "put"
or "unlock" them even if getting one returned NULL. This patch fixes that.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>

Signed-off-by: NKeith Busch <keith.busch@intel.com>
[made admin_timeout static]
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>

This was originally set to 4 times the IO timeout, but that was when
the IO timeout was 5 seconds instead of 30. 20 seconds for total time
to failure seemed more reasonable than 2 minutes for most, but other
users have requested to make this a module parameter instead.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
[renamed the module parameter to retry_time]
[made retry_time static]
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>

kmalloc() used by the nvme_alloc_iod() to allocate memory for 'iod'
can fail. So check the return value.
Signed-off-by: NSantosh Y <santosh.sy@samsung.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>

Quiesce and shutdown the device prior to reset, then restart the device and
resume IO after.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>

Since _nvme_check_size() wasn't being called from anywhere, the compiler
was optimising it away ... along with all the link-time build failures
that would result if any of the structures were the wrong size.  Call it
from nvme_exit() for no particular reason.
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>

It is possible a filesystem may send a flush flagged bio with write
data. There is no such composite NVMe command, so the driver sends flush
and write separately.

The device is allowed to execute these commands in any order, so it was
possible the driver ends the bio after the write completes, but while the
flush is still active. We don't want to let a filesystem believe flush
succeeded before it really has; this could cause data corruption on a
power loss between these events. To fix, this patch splits the flush
and write into chained bios.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>

This configures an nvme request_queue as flush capable if the device
has a volatile write cache present.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>

Adding tracepoints for bio_complete and block_split into nvme to help
with gathering IO info using blktrace and blkparse.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>

If a misbehaving device posts a CQE with a command id < depth but for
one that was never allocated, the command info will have a callback
function set to NULL and we don't want to try invoking that.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>

Help people diagnose what is going wrong at initialisation time by
printing out which command has gone wrong and what the device returned.
Also fix the error message printed while waiting for reset.
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>
Reviewed-by: NKeith Busch <keith.busch@intel.com>

Make the copyright dates accurate and remove the final paragraph that
includes the address of the FSF.
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>

For commands returned with failed status, queue these for resubmission
and continue retrying them until success or for a limited amount of
time. The final timeout was arbitrarily chosen so requests can't be
retried indefinitely.

Since these are requeued on the nvmeq that submitted the command, the
callbacks have to take an nvmeq instead of an nvme_dev as a parameter
so that we can use the locked queue to append the iod to retry later.

The nvme_iod conviently can be used to track how long we've been trying
to successfully complete an iod request. The nvme_iod also provides the
nvme prp dma mappings, so I had to move a few things around so we can
keep those mappings.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
[fixed checkpatch issue with long line]
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>

Some programs require HDIO_GETGEO work, which requires we implement
getgeo.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>

Done to ensure nvme_thread is not running when there
are no devices to poll.
Signed-off-by: NDan McLeran <daniel.mcleran@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>

Increase the default timeout to 30 seconds to match SCSI.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
[use byte instead of ushort]
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>

Registers with hot cpu notification to rebalance, and potentially allocate
additional, io queues.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>

The device's IO queues are associated with CPUs, so we can use a per-cpu
variable to map the a qid to a cpu. This provides a convienient way
to optimally assign queues to multiple cpus when the device supports
fewer queues than the host has cpus. The previous implementation may
have assigned these poorly in these situations. This patch addresses
this by sharing queues among cpus that are "close" together and should
have a lower lock contention penalty.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>

Checkpatch has started warning against using DEFINE_PCI_DEVICE_TABLE,
so replace it.  Also update the copyright date and bump the module
version number to 0.9.
Signed-off-by: NMatthew Wilcox <matthew.r.wilcox@intel.com>