NVMe hotplug must be monitored in the primary process -
DPDK doesn't support trying to handle it in the
secondary process.
This issue was somewhat masked previously in secondary
processes, since usually it would just probe(NULL) which
meant probe all attached NVMe controllers. So in the
secondary process, we would probe just once, and create
the hotplug fd - it would never actually try to monitor
it.
But when explicitly specifying multiple trids in a
secondary process, probe would get called multiple
times. First time would be fine since it only creates
the hotplug fd. But second time would segfault since
monitoring for hotplug requires checking the DPDK-allocated
context which doesn't exist in the secondary process.
Fixes issue #1063.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I2a9a91e222c206034293d90e30e3f598c8d7baa8
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/475015
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Alexey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Currently we have a mix of -1 and -EINVAL which
is confusing, especially since these types of failures
also result in the caller's callback routine getting
invoked.
While here, document this new -EFAULT return code for
all of the functions that could return it.
Fixes issue #797.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I8dfbba0ec0b83db0f2ec055b15830981af1965df
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/473054
Community-CI: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: yidong0635 <dongx.yi@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
This wasn't being done in the previous case which meant that I/O qpairs
were not being moved to the connecting state when connecting for the
first time. However, to prepare the way for a coherent state machine for
nvme qpairs, we need to ensure that all qpairs go through the same
states.
Change-Id: I3cfe799a003acd926b24c107ab1461a96239c1bb
Signed-off-by: Seth Howell <seth.howell@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/471753
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Alexey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
This reverts commit bc4e31d6b2.
This change was accidentally merged after it was decided to go with a
different architecture.
Change-Id: Ifc9d8b08bd1fcbc4ace8dd6fb4bd0014330916ed
Signed-off-by: Seth Howell <seth.howell@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/471144
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
The remove callback is a built in way of alerting the user application
that we have removed a controller. Once we fail a controller, we never
move it back out of that state so it is in essence removed.
Change-Id: Iaad6bef0994e9ddd5a424f6b83502f9191b2de49
Signed-off-by: Seth Howell <seth.howell@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/469637
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Making this structure available from the ctrlr allows us to call the
remove callback when the controller is failed/removed on transports
other than pcie.
Change-Id: I2c66dfef12b039c0d6daf7df83da745757818006
Signed-off-by: Seth Howell <seth.howell@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/469636
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
We were already passing up from each transport the number of completions
done during the transport specific call. So just use that return code
and batch all of the submissions together at one time in the generic
code.
This change and subsequent moves of code from the transport layer to the
genric layer are aimed at making reset handling at the generic NVMe
layer simpler.
Change-Id: I028aea86d76352363ffffe661deec2215bc9c450
Signed-off-by: Seth Howell <seth.howell@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/469757
Reviewed-by: Alexey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
The tailq and the requests all belong to the generic layer, might as
well put the queueing code there for better encapsulation.
Change-Id: Id5f08f798121b50a21044cfc61856999c50ca227
Signed-off-by: Seth Howell <seth.howell@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/469758
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
SPDK_ERRLOG lists the function name, so remove old references that
assume it doesn't and reprint the function name.
Change-Id: I69da6ca0a25bf0eda07d8dad52bcfadf964ac715
Signed-off-by: Seth Howell <seth.howell@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/469487
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
spdk_pci_device_claim() could create a file on the
filesystem that couldn't be deleted programatically.
It could only be overwritten - e.g. by another spdk
instance - but this didn't really work if that
another instance had less privileges and hence no
access to the previous file.
This is exactly the case we're seeing on our CI when
running SPDK as non-root. In general it's a good idea
not to leave any leftover files, so now we'll delete
the pci claim file when the spdk process exits.
spdk_pci_device_claim() used to return a file descriptor
that could be simply closed to "un-claim" the device.
It'll now return only a return code. The fd will be
stored inside spdk_pci_device and will be closed either
when user calls the newly introduced spdk_pci_device_unclaim(),
or when the device is detached.
We'll still need to clean up those files somewhere in
our test scripts (probably ./setup.sh cleanup) to
clean up after crashed processes or so - but we don't
necessarily want to run such scripts inside the autotest
whenever a non-root spdk is about to be started.
Change-Id: I797e079417bb56491013cc5b92f0f0d14f451d18
Signed-off-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/467107
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
If the CPU reorders the eventidx read before the shadow doorbell
write, it is indeterminate whether the controller will read the
updated shadow doorbell without an MMIO write. See
https://lkml.org/lkml/2018/8/14/1031 for details.
Signed-off-by: Benjamin Saunders <bsaunders@google.com>
Change-Id: I5aa08fdd5b32c7b81e8048ca6efe546318d80b5c
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/468188
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
This will be consistent with TCP and RDMA transport, and we will use
ctrlr->flags in nvme_ctrlr_init_cap() in next patch, the flags will
be cleared to 0 for now.
Change-Id: Ic360cd0c00d60c77452d19cdc1e7a32a5fc34df0
Signed-off-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/466678
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
We have defined NVMe controller initialization 'transport_retry_count' option, so
global 'spdk_nvme_retry_count' can be removed, we will remove the variable with
PCIe transport first, and make the retry count can be configured via RPC.
Change-Id: I4d54f78c8da2180d536635587e7291f44a57c4fb
Signed-off-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/464472
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
This change attempts to address the Trello request to decode I/O errors in
NVMe hello_world example.
See https://trello.com/c/MzJJw7hM/2-decode-io-errors-in-nvme-helloworld-example
As part of this change, spdk_nvme_cpl_get_status_string was declared
in nvme.h, and spdk_nvme_qpair_print_command and
spdk_nvme_qpair_print_completion were renamed and added to nvme.h,
allowing all three to used "externally."
To test the failing paths, two compile time defines were added to force a
write or read error (bad LBA) respectively.
As the example does a read after write, if the write fails, the example fails.
Signed-off-by: James Bergsten <jamesx.bergsten@intel.com>
Change-Id: Ib94b4a02495eb40966e3f49517a5bdf64485538a
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/457076
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
The __sync builtin based implementation generates full memory
barriers on some non-x86 platforms. Replace it with C11 atomic
builtins can make:
·arm and ppc from full barrier to half barrier
·x86 code same as before
Signed-off-by: Richael Zhuang <richael.zhuang@arm.com>
Change-Id: Ib6624ef8e45af497b9eced6ecfa7710bcc88a733
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/461590
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
The completion cb of outstanding_tr may submit new requeset to
the outstanding_tr list of the qpair, it's an endless loop.
We only abort the remaining outstanding trackers.
Fix#819
Change-Id: I342f52f4d1836f8ef620ef9e3add0b1986727282
Signed-off-by: JinYu <jin.yu@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/457755
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Adds fields to structure spdk_nvme_io_qpair_opts.
These fields allow specifying the locations of memory buffers used
for the submission and/or completion queues.
By default, vaddr is set to NULL meaning SPDK will allocate the memory to be used.
If vaddr is NULL then paddr must be set to 0.
If vaddr is non-NULL, and paddr is zero, SPDK derives the physical
address for the NVMe device, in this case the memory must be registered.
If a paddr value is non-zero, SPDK uses the vaddr and paddr as passed.
SPDK assumes that the memory passed is both virtually and physically
contiguous.
If these fields are used, SPDK will NOT impose any restriction
on the number of elements in the queues.
The buffer sizes are in number of bytes, and are used to confirm
that the buffers are large enough to contain the appropriate queue.
These fields are only used by PCIe attached NVMe devices. They
are presently ignored for other transports.
Signed-off-by: James Bergsten <jamesx.bergsten@intel.com>
Change-Id: Ibfab3939eefe48109335f43a1167082dd4865e7c
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/454074
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
We will need to put the recently completed nvme_request
object on the qpair's STAILQ. We don't reference any
real data from the nvme_request in the completion path
since we've already stashed the cb_fn and cb_arg in
the nvme_tracker. But we will need to reference the
STAILQ_ENTRY to put it back in the qpair's STAILQ, so
prefetch that cacheline.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Id76122afe4150c84a61fbe38bc874f10d606b3b3
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/456673
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Using AVX512 or AVX2 ends up being a small pessimization.
I think AVX works better for copies when there are
multiple cachelines to copy. I see a 2-3% improvement
in high IOPs benchmarks when reverting to SSE.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I3d70a1e359e98cec2a9da41ccf9af2de9baa5868
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/456247
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Profiling showed these weren't getting inlined - so add
the inline keyword to make sure it happens. This helps
improve performance a bit.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Ia86edccc9163258efdcddcce6989a71fb180caf6
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/456099
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
At 10M IO/s, we see a lot of CPU cycles wasted getting
the next tracker into cache. If we only get one
completion at a time, this is unavoidable, but when
there are multiple completions pending, we can prefetch
the second tracker while processing the completion for
the first.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I9de702bee3719e4494eec6f05b09be3672f1e0ac
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/456097
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
This avoids dereferencing the request to get the qpair
in cases where we already know the qpair. Adding a new
variant instead of just modifying nvme_free_request()
since there are 72 calls to this function and I don't
want to change all of them.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Ifd6fd964e546bcd71ff180fd71d5bf5cbab79d4f
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/455287
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
In some cases we have the qpair already when calling
this function. So pass the qpair to avoid having
to get it from the request. This shows about a 3%
performance improvement for high IOPs single core
tests.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I22fcca560492f4e7cf5ffedd252e41a027d0dd79
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/455286
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Currently the nvme driver will always log any
request completed with error status. Some
applications may not want this behavior. So provide
an option to disable it at the controller level.
When this option is enabled, any failed requests
from queues associated with that controller
(including the admin queue) will not log the
failed request.
Of course the application will still receive
the failed status code and can decide to do its
own logging there.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Ia093fcd23cf321a820fd53183ee7e2dac4f9d378
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/454081
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
The RDMA transport was the only one implementing this
function, and it only does a connect - not a disconnect
followed by a connect.
A later patch will add a matching disconnect function.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Ib68eb0ff2f8e59f437d6d8831bb37dfddf83e9a4
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/453929
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
This function returns a pointer to the PCIe I/O registers for a controller
or NULL if unsupported for this transport.
Used for PCIe only, other transports return NULL.
Use with caution.
Signed-off-by: James Bergsten <jamesx.bergsten@intel.com>
Change-Id: I849f9de9ad259a65b1eef9c1237345eb7195b9bf
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/452927
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
This transport function is a complete nop now, so
remove it.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I5cc6ac75795a3cf5311f24e2ac293fb53d4b9f8c
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/453487
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
This will allow us to move more of the reset-related
functionality to the common layer, as part of enabling
resets for fabrics controllers.
The transport qpair_enable and qpair_fail functions
acted similarly - so those are both removed now and
replaced with this new qpair_abort_reqs function.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I9486630ad5b807239b0b5bcde50e8cfd313695d3
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/453486
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Upcoming patches will move the actual is_enabled
logic to the common layer as well.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I9fd4d8712280295f57134ad66f8ccbfe9736d30a
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/453484
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
We submit AERs to all controllers - both pcie and
fabrics. But currently we only manually abort the
aers when disabling the qpair for pcie. Make this
common instead by creating a new transport function
for aborting aers.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I1e926b61b8035488cdc6e8cb4336b373732f985e
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/453482
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
This better explains what the function is doing,
and makes the name more general so we can use it
for the adminq as well.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I6b55761cb141a9a79cdef876be47995d8813b312
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/453480
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Send the given NVM I/O command, I/O buffers, lists and all to
the NVMe controller.
This is a low level interface for submitting I/O commands directly.
This can only be used on PCIe controllers and qpairs.
This function allows a caller to submit an I/O request that is
COMPLETELY pre-defined, right down to the "physical" memory buffers.
It is intended for testing hardware, specifying exact buffer location,
alignment, and offset. It also allows for specific choice of PRP
and SGLs.
The driver sets the CID. EVERYTHING else is assumed set by the caller.
Needless to say, this is potentially extremely dangerous for both the host
(accidental/malicionus storage usage/corruption), and the device.
Thus its intent is for very specific hardware testing and environment
reproduction.
Signed-off-by: James Bergsten <jamesx.bergsten@intel.com>
Change-Id: I595fe02fe0dfa9c3ceba1ac116b6900357b02d2c
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/451994
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
The mode of dmb oshld can guarantees cpu sequential execution,
which has less impact on performance.
Change-Id: If30b6a682a2216eecd1da039267ed4f5471afc38
Signed-off-by: h00448672 <heyang18@huawei.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/446827
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
This is an attempt to fix device hotremove with VFIO.
A soft device hotremove request through sysfs [1] would
currently just block until the SPDK process manually
releases that device - e.g. upon an RPC request.
VFIO won't get unbound from the device untill userspace
releases all its resources. VFIO can signal a pending
hotremove request by kicking any file descriptor provided
by the userspace - and DPDK does provide such descriptor -
but SPDK does not listen on it.
DPDK does offer handy API to listen and in this patch
we make use of it inside our env/pci layer. Within
a DPDK callback we set an internal per-device hotremove
flag, which upper-layer SPDK drivers can poll with a new
env API - spdk_pci_device_is_removed().
The VFIO hotremove event will be sent to primary
processes only, so that's where we listen.
We make use of this new API in the NVMe hotplug poller,
which will process it just like any other supported
hotremove event.
Fixes#595Fixes#690
[1] # echo 1 > /sys/bus/pci/devices/<bdf>/remove
Change-Id: I03d88271c2089c740e232056d9340e5a640d442c
Signed-off-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/448927
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
This avoids a data dependent load to find which
callback to call in the completion path.
Change-Id: Ifa20790a7af3332a74bc45037e589668744af797
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/450558
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
The same information can be obtained by checking the req
field for NULL.
Change-Id: I9689ee0be33537fd6d3f35d8c3710fabab5e1928
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/450557
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
When copying the NVMe command from the request to the actual
submission queue slot, use a non-temporal move instruction.
The submission queue slots are never read by software - only
written to. So don't pollute the CPU cache with their contents.
Change-Id: I112f721abfac03bd7b33ec9ddf783d4bf2952b42
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/450193
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Xiaodong Liu <xiaodong.liu@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
There are several places where we have the tracker
pointer, yet we go find the tracker again by getting
the tr->cid and using that index to find the tracker
again in the qpair's array. That's really silly.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I54acd642a2c9821f2b95e17563904b859495081a
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/450308
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Logically we should complete any I/O in the completion queue and
abort the rest of the I/O after we delete the submission queue
and completion queue, so that we would not lost any complete I/O.
We alse should complete I/O and abort I/O before destroy I/O qpair
even though the ctrlr/device has been removed.
Change-Id: Ieb28ad7b4a3a7be553f70178b29ca870b5413191
Signed-off-by: JinYu <jin.yu@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/449316
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Similar to recently added nvme_pcie_qpair_ring_sq_doorbell.
Prepares for using this code for coalescing cq doorbell
writes in the future.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I565e103acf73b3d305e72a4440e1cc678c95faa0
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/448871
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
The phys_addr param in spdk_*malloc() is about to be
deprecated, so use a separate spdk_vtophys() call to
retrieve physical addresses.
This patch also adds error checks against SPDK_VTOPHYS_ERROR.
The error handling paths are already there to account for
spdk_*malloc() failures themselves, so reuse them in case
of vtophys failures.
Change-Id: I377636e66b8c570d013c1bb2021f04bce4e6c0ce
Signed-off-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/416998
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Try to group data members that are used often into the
same cache lines. We still need to find more space in the second
cache line of spdk_nvme_pcie_qpair so that the important
parts of spdk_nvme_qpair fit.
Change-Id: Ib936cb2b1acc722de7ec313d6faa3812aacde394
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/447968
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
