bdev channel is used in nbd fini process, so it should
be released in the latter part of nbd_stop
Change-Id: I87edea63d2d91954cc41cdb71261485ae24c0d9f
Signed-off-by: Liu Xiaodong <xiaodong.liu@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6280
Community-CI: Broadcom CI
Community-CI: Mellanox Build Bot
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: <dongx.yi@intel.com>
Similar issue was fixed in
813869d8231c502c7f738b7da3cc343b7f0bcf0f
nvmf: Fix possible race condition when adding IO qpair
This patch fixes the same issue which occurs a bit later,
when a message is delivered to another thread. This issue
occurred on CI, callstack is the following:
00:11:46.296 #6 0x00007f2705199f05 in __ubsan_handle_type_mismatch_v1 () from /lib64/libubsan.so.1
00:11:46.296 No symbol table info available.
00:11:46.296 #7 0x00007f27067ace6f in ctrlr_add_qpair_and_update_rsp (qpair=0x221edc0, ctrlr=0x1dc4ea0, rsp=0x2242918) at ctrlr.c:230
00:11:46.296 __PRETTY_FUNCTION__ = "ctrlr_add_qpair_and_update_rsp"
00:11:46.296 __func__ = "ctrlr_add_qpair_and_update_rsp"
00:11:46.296 #8 0x00007f27067b1d0b in nvmf_ctrlr_add_io_qpair (ctx=0x2242540) at ctrlr.c:534
00:11:46.296 req = 0x2242540
00:11:46.296 rsp = 0x2242918
00:11:46.296 qpair = 0x221edc0
00:11:46.296 ctrlr = 0x1dc4ea0
00:11:46.296 __func__ = "nvmf_ctrlr_add_io_qpair"
00:11:46.296 #9 0x00007f27062553ce in msg_queue_run_batch (thread=0x1cff540, max_msgs=8) at thread.c:553
where line 230 in ctrlr.c was
assert(ctrlr->admin_qpair->group->thread == spdk_get_thread());
That means that admin qpair was disconnected from the poll
group and controller is in the process of destruction
Change-Id: I818ba56adda5ed3488a8df78483c0b6839758192
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6364
Community-CI: Broadcom CI
Community-CI: Mellanox Build Bot
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 already have support for spdk_nvme_zns_zone_append(),
add support for spdk_nvme_zns_zone_appendv() (zone append with
NVME_PAYLOAD_TYPE_SGL).
_nvme_ns_cmd_rw() currently performs verification of the SGL,
if the parameter check_sgl is set. This parameter is set for all
calls with payload of type NVME_PAYLOAD_TYPE_SGL.
In order to be able to perform the same check_sgl verfication on
zone append vectors, we need to refactor _nvme_ns_cmd_rw() a bit.
Setting check_sgl ensures that _nvme_ns_cmd_split_request_sgl() or
_nvme_ns_cmd_split_request_prp() gets called.
These functions will split an oversized I/O into several different
requests. However, they also iterate the SGE entries, verifies that
the total payload size, total SGE entries is not too many, and that
buffers are properly aligned. A proper request will not get split.
For zone append, splitting a request into several is not allowed,
however, we still want the verification part to be done, such that
(e.g.) a non first/last SGE which is not page aligned, will cause
the whole request to be rejected.
(In the case of spdk_nvme_ns_cmd_write(), a non first/last SGE which
is not page aligned will instead cause the request to be split.)
An alternative would be to try to rip out the verification part from
_nvme_ns_cmd_split_request_sgl() and _nvme_ns_cmd_split_request_prp().
However, that is non-trivial, and would most likely end up with a lot
of duplicated code, which would easily get out of sync.
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Change-Id: I2728acdcadeb70b1f0ed628704df19e75d14dcca
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6248
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Use the new function spdk_nvme_zns_ns_get_zone_size_sectors() where
it is appropriate (in comparison to the existing
spdk_nvme_zns_ns_get_zone_size() variant).
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Change-Id: Ic929ffbc5a1f4a16ba6719a985c05ae625caed46
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6417
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
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>
Sometimes it is more optimal to get the zone size in number
of sectors, instead of in number of bytes.
Therefore, add a new spdk_nvme_zns_ns_get_zone_size_sectors()
function to get zone size in number of sectors.
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Change-Id: I0fe67e00a3d74dd27acfc895ae97448d995b89a3
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6416
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
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>
In order to make sure we do always response to the kernel module if
there are valid commands in the socket. If we do not see this,
we will see stuck request kernel info in nbd module. And the kernel
will print the timeout message of nbd module again and again.
Signed-off-by: Ziye Yang <ziye.yang@intel.com>
Change-Id: I2ecc3e9c948231a712778f0126e2ecc6220e1d3c
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6276
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: <dongx.yi@intel.com>
Reviewed-by: Xiaodong Liu <xiaodong.liu@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Community-CI: Mellanox Build Bot
The current implementation treats HPDA/CPDA as the absolute offset
to the beginning of the PDU where the payload data starts. This is
incorrect. The HPDA/CPDA actually specify where the payload data
should start such that the starting location is a multiple of HPDA
(for C2H PDU) or CPDA (for H2C PDU or CapsuleCmd PDU).
The other issue fixed is that the current implementation calculates
padding only when header digest is enabled. This is also incorrect.
Signed-off-by: Wenhua Liu <liuw@vmware.com>
Change-Id: If7a3896a4c1d73f6d062bd3dbe6a912d31771180
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6256
Community-CI: Broadcom CI
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
For correct behaviour, pthread_mutex should not be locked after it has
been destroyed.
g_bdev_mgr.mutex is statically initialized. It is destroyed in
bdev_mgr_unregister_cb, but not re-initialized in spdk_bdev_initialize.
Repeated calls to initialize/unregister occur during unit tests.
Remove the destroy from bdev_mgr_unregister_cb, which seems
the simplest way of resolving the issue.
The sequence: spdk_put_io_channel(), spdk_bdev_close(),
spdk_bdev_unregister() occurs during unit tests.
spdk_bdev_unregister() destroys internal.mutex which is then
locked by a call to bdev_channel_destroy() resulting from the
earlier spdk_put_io_channel(). Move the destroy and the free of
internal.qos into bdev_destroy_cb so that they don't occur until
all of the channels have been released. Remove the no longer
required bdev_fini.
Repeat calls to spdk_bdev_unregister that occur after an unregister has
completed will lock internal.mutex which has been destroyed by the
previous unregister. This occurs during unit tests. Defer locking
internal.mutex until after the internal.status has been checked for
SPDK_BDEV_STATUS_REMOVING. This is the only place where
internal.status is set to removing and g_bdev_mgr.mutex alone is
sufficient to ensure atomicity here.
Tested with a pthreads library that contains debugging code to
check the mutex state and a modified version of bdev_io_types_test
to call get_io_channel on a different thread.
Suggested-by: Jim Harris <james.r.harris@intel.com>
Signed-off-by: Nick Connolly <nick.connolly@mayadata.io>
Change-Id: I81cc46a1b8a766700253829b19cc86c7f0eb79f2
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6217
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Community-CI: Broadcom CI
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Vhost is enabled by default, so rte_net was always included.
When disabled, rte_power failed as it depends on rte_ethdev and rte_net.
rte_vhost was only possible to enable on Linux, so there
is no conflict with adding it next to rte_power under this condition.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I2e183004d6457e404471740a0540dcb08aa738d8
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6398
Reviewed-by: Ziye Yang <ziye.yang@intel.com>
Reviewed-by: Jacek Kalwas <jacek.kalwas@intel.com>
Reviewed-by: Nick Connolly <nick.connolly@mayadata.io>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Identify application prints the PMR details if it is supported
Signed-off-by: Krishna Kanth Reddy <krish.reddy@samsung.com>
Change-Id: Iaba4c15e18e1402035b11a34b2defe8078855751
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6209
Community-CI: Broadcom CI
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
In dpdk 19.11 version, RTE_VHOST_USER_ASYNC_COPY is not define.
After dpdk 20.08, we can use RTE_VHOST_USER_ASYNC_COPY.
Use version check to avoid this problem.
Signed-off-by: sunshihao <sunshihao@huawei.com>
Change-Id: Iaf9914e8380f3d54cded1e2f16af6a7dc3504f95
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6274
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Community-CI: Mellanox Build Bot
There is a special case when using 8-byte metadata + PI + PRACT
where no metadata is transferred to/from controller.
Since _nvme_ns_cmd_rw() already calculates the proper sector size
using _nvme_get_host_buffer_sector_size(), which takes PRACT into
account, change the sectors_per_max_io calculation to also take
PRACT into account.
This will avoid certain requests that don't need splitting getting
split.
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Change-Id: I8d450d37c2458453701189f0e0eca4b8fe71173b
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6247
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
After setting io timeout, host can avoid nbd io
stuck or kernel hang occasionally caused by nbd
stop or underlying bdev removal.
Change-Id: I4ba2a0af7ff7bed369cdaf86121f082136dc1a0b
Signed-off-by: Liu Xiaodong <xiaodong.liu@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6191
Community-CI: Broadcom CI
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: <dongx.yi@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Ziye Yang <ziye.yang@intel.com>
Aspects of bit fields are 'implementation defined'. On some platforms
alignment will occur if two adjacent fields are of different types. This
occurs in spdk_nvme_feat_async_event_configutation after the crit_warn
member which is effectively an int8_t, followed by an int16_t. There
isn't a generic way of changing the compiler's behaviour, so the best
options are:
- Change crit_warn to a uint32_t bit field and copy the value to/from
a spdk_nvme_critical_warning_state variable to use it. This requires
changes to code using the field.
- Adjust the structure definition to use smaller types to avoid the
problem. This preserves existing semantics, but the field order will
need to be reviewed if big-endian support is ever added (other places
in nvme_spec.h will need similar attention). A second reserved field
is required.
Use smaller types which seems the most straightforward option. Adjust
the use of the spdk_nvme_feat_async_event_configuration reserved fields
in lib/nvmf/ctrlr.c.
The new structure is binary compatible and the fields behave in the same
way, with the exception of an additional reserved field, so updating
CHANGELOG.md probably isn't necessary.
Signed-off-by: Nick Connolly <nick.connolly@mayadata.io>
Change-Id: I7d8163c84b4f410fc95a5b7064506ad7b4b62c6c
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6340
Community-CI: Broadcom CI
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
__builtin_clzl takes an unsigned long argument which may be smaller
than uint64_t on some platforms. GCC silently ignores the mismatch,
returning the wrong answer at runtime. Use __builtin_clzll instead and
add static assertions to detect the issue.
Attribute 'target_clones' requires 'ifunc' support which only applies to
ELF targets. Add check for defined(__ELF__).
Signed-off-by: Nick Connolly <nick.connolly@mayadata.io>
Change-Id: Iff76640b34223649de531250ad40471d829512c7
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6317
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
insert_queue() will copy it to internal data structure, so that
before successful map we don't need to consider the error path.
Change-Id: Id7ea2ef73da7914ea430ea568e7981657016d3f7
Signed-off-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6310
Community-CI: Broadcom CI
Community-CI: Mellanox Build Bot
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>
The CQ is created first, so it's more reasonable to connect
the IO queue pair after creating the SQ.
Change-Id: I196c19a54a015310a3777d9bfca7db8735a4d5b2
Signed-off-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6309
Community-CI: Broadcom CI
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>
add_qp() function is only called when creating NVMe SQ/CQ, so unpack
it into the caller to make the code more clear.
Change-Id: Id5cc1152b1684df980909b2f7d73ed2788c0efb2
Signed-off-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6308
Community-CI: Broadcom CI
Community-CI: Mellanox Build Bot
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>
The clients(QEMU and NVMe vfio-user driver) use shared memory with
NVMf vfio-user target for zero-copied IO processing, when memory
region hotplug happens from clients, the backend target can get
notification via callbacks. Here we rename them to reflect the
action. For now NVMe vfio-user example applications use static
memory model and single memory segment.
Change-Id: Icecbe13883668dd8267019b5fe57d0fef1c68b81
Signed-off-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6307
Community-CI: Broadcom CI
Community-CI: Mellanox Build Bot
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 can use the NVMf library ABORT implementation directly, so remove
it in vfio-user.
Change-Id: I0f204a869c53c6a6ce67ad900a64d5bb59ac2aab
Signed-off-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6306
Community-CI: Broadcom CI
Community-CI: Mellanox Build Bot
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>
The group poll context is for queue pair state, so we don't need to
check controller state here, and for the disconnect case below, the ADMIN
queue pair will be removed from group poll.
Also add spdk_unlikely in the poll context.
Change-Id: I5ef32ef3cf41ad757a7cb167e1e1fa32c52a84d6
Signed-off-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6227
Community-CI: Broadcom CI
Community-CI: Mellanox Build Bot
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>
Zone append commands cannot be split.
_nvme_ns_cmd_rw() should never cause a NVME_PAYLOAD_TYPE_CONTIG
zone append request to be split.
This is currently true, but add an assert to make sure that
any refactoring to _nvme_ns_cmd_rw() does not break this promise.
Also add error handling, since release builds are built with
asserts disabled.
Follow-up patches will refactor _nvme_ns_cmd_rw().
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Change-Id: I5fd2440c4c9d6bd8d56f30354b208a9047b64729
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6246
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Community-CI: Mellanox Build Bot
This function allows applications to specify whether
they wish to allow probing a newly attached NVMe
PCIe SSD.
The env layer will only even probe devices that have
been allowed. By default, this is all devices, but
if the user has specified some list of
allowed PCI addresses (via spdk_env_opts pci_allowed)
then newly attached PCIe devices are implicitly not
allowed. This API allows applications to add
device addresses to the allowed list after the
application has started.
This API will be useful for use cases where multiple
SPDK processes are running on one server, and assignment
of PCIe SSDs to those processes are based on some function
of the SSD's PCIe address.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I919bc267f2ad9130ab5c875ff760a301028b047e
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6184
Community-CI: Broadcom CI
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: <dongx.yi@intel.com>
The env layer has a pci_allowed list, which specifies
that only a subset of PCI devices may be attached
by the associated process.
But that doesn't cover PCI devices that are hot-inserted
after the application starts, which is common for
storage/NVMe.
So add a new spdk_pci_device_allow() API which allows
an application to add new devices to the allowed list.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I7bd5ff428d84480d46bc236698daadd019b20b8e
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6183
Community-CI: Broadcom CI
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: <dongx.yi@intel.com>
Fixes#1777.
When a qpair cannot be allocated because the transport connection fails,
the qpair was freed without unlinking it from the other structures.
This was leading to a segfault when attempting to create and free other
qpairs.
Also added a unit test to cover this case.
Change-Id: I74b78d1847f90117248b07203b43a11ff5cfa5d6
Signed-off-by: Vasuki Manikarnike <vasuki.manikarnike@hpe.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6272
Community-CI: Broadcom CI
Community-CI: Mellanox Build Bot
Reviewed-by: Michael Haeuptle <michaelhaeuptle@gmail.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: <dongx.yi@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
To allow SO_MINOR updates on LTS for the whole year it is supported,
the major version for all components needs to be increased.
This is to prevent scenario where two versions exists with matching
versions, but conflicting ABI.
Ex. Next SPDK release adds an API call increasing the minor version,
then LTS needs just a subset of those additions.
Increasing major so version after LTS, allows the quarterly releases
to update versions as needed. Yet allowing LTS to increase minor
version separately.
Disabled test for increasing SO version without ABI change, as
that is goal of this patch. This check shall be removed with SPDK 21.04
release.
This patch:
- increases SO_VER by 1 for all components
- resets SO_MINOR to 0 for all components
- removes suppressions for ABI tests
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I44d01154430a074103bd21c7084f44932e81fe72
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6167
Community-CI: Broadcom CI
Community-CI: Mellanox Build Bot
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>
In dpdk 19.11, rte_kernel_driver is the old version, add version check before use the members.
Signed-off-by: sunshihao <sunshihao@huawei.com>
Change-Id: Ic1db37cc0760c7d03692fd2cdcbb6ff1e41f872d
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6252
Community-CI: Mellanox Build Bot
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>
Developer convenience - make this based on a specific version.
Signed-off-by: paul luse <paul.e.luse@intel.com>
Change-Id: I228b8aff6e8957cad5e8c1fae5615b113e16cfb5
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/5950
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>
Community-CI: Mellanox Build Bot
In the completion handler there's no need to do an MMIO read
unless the completion record indicates there's an error.
Signed-off-by: paul luse <paul.e.luse@intel.com>
Change-Id: Ic4850d803a23413f9813da30ac6f1b611804f1b5
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/5847
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Ziye Yang <ziye.yang@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
For flow control reasons we have to resize the bit arrays we
use to manage flow as channels come and go. However since
channels are assigned to devices, until the channel count
reaches the device count there's no sharing so no resize of
the array is needed. So, when we use a device for the first
time there's no need to run through the rest of the channels
and re-balance.
Same thing is done on destruction. The code to free idxd
specific resources was moved from the rebalance function to
the idxd put channel function which is a much more logical
place for it as well.
Signed-off-by: paul luse <paul.e.luse@intel.com>
Change-Id: Ib4df163286906f413dd6429dc6833af7b68e208c
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/5846
Community-CI: Broadcom CI
Community-CI: Mellanox Build Bot
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>
Config #1 remains what is shown as an example in the spec. Change
config #0 to just have 1 work group and 1 work queue all backed
by 4 engines. As the majority of initial use cases will not be
implementing separate priorities and/or different back end
targets (mem, pmem, etc) having just 1 group and work queue makes
the most sense as it allows the silicon to decide which engine to
use.
Also, having multiple work queues spreads out the available
entires such that if we're not using all of the work queues then
we're not using all of the resources. As channels are created
they are assigned the next available device. As a channel is
assigned a device that is already in use it will round robin
work queues. If then, for example, we have 16 devices then only
the first work queue will ever be used for the first 16 threads
which seems and if there are even just 2 work queues per device
it would take 32 threads to use all of the resources at the
device.
By haing just one work queue per device we always have the max
number of work queue entries available regardless of how many threads
are being used.
Signed-off-by: paul luse <paul.e.luse@intel.com>
Change-Id: Ie15ff6bdea12525fe3bfc769613084ddd2de50bf
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/5845
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Ziye Yang <ziye.yang@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
If rdma_qp_disconnect is not correctly sent out, we will not wait
for the event.
Change-Id: I99701e421dc93909d481ccf35e9bfd8004e60da8
Signed-off-by: Ziye Yang <ziye.yang@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6163
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: <dongx.yi@intel.com>
Those spdk threads were neither used nor tracked in event framework.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I1becee6390e172a10e08754703e018f7c44e30df
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6213
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
spdk_vhost_init() relies on having a thread on each reactor.
Every vhost controller could be created on the same core and
even passing --cpumask when creating would not affect it.
This has happened before, see patch (7cc83b62).
This patch modifies the g_vhost_core_mask to match the actual
cores in use.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I42a07c5f99690bfa4ecd2a5b9b7b04d1aa7d2800
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6188
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Documentation for vhost target states that CPU mask must be a subset
of application CPU mask. This wasn't enforced right now and allowed
the cpumask on controller creation so long as at least single
CPU core overlapped with application's CPU mask.
This might have been misleading and covered up user configuration errors.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I03f959ec37efd0be9b98cff9c93c5f996b04af35
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6212
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
As of DPDK main commit 682a6454 ("power: add ethdev
power management"), rte_power depends on rte_ethdev.
So add rte_ethdev library, but only on Linux since
that is the only place where we use rte_power.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Ibc3473d6ec75c14a8c0c5517616950017cb2ea96
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6210
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: <dongx.yi@intel.com>
Add the real support in nvme tcp transport.
Change-Id: I2aa9b0284d6fe009925e67f602a055e787f77987
Signed-off-by: Ziye Yang <ziye.yang@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/5734
Community-CI: Broadcom CI
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>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
This patch is used to add spdk_nvme_poll_group_get_optimal
public API.
Signed-off-by: Ziye Yang <ziye.yang@intel.com>
Change-Id: Iee34c89e0e1ff1f81167b18e198c144ca28f71de
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/3311
Community-CI: Broadcom CI
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
This reverts commit 72eed604b407e8cb53ce41e97d573b801c6ea50e.
spdk_vhost_init() relies on having a thread on each reactor.
Without the revert, every vhost controller is created on the same
core and even passing --cpumask when creating does not affect it.
Proper fix would be to change spdk_vhost_init(),
but would require additional testing. For now revert the patch
that spawned always idle threads.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I0dbd6c9fe1d9d23ada260da8fc7b48086223c632
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/6186
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>
Community-CI: Mellanox Build Bot
Recently, we discovered that under the SPDK + UIO mode, it is possible that vm may be wrong when doing io.
such as:
[root@VM_6_151_centos ~]# mkfs.xfs /dev/vdc
meta-data=/dev/vdc isize=512 agcount=4, agsize=234421142 blks
= sectsz=4096 attr=2, projid32bit=1
= crc=1 finobt=0, sparse=0
data = bsize=4096 blocks=937684566, imaxpct=5
= sunit=0 swidth=0 blks
naming =version 2 bsize=4096 ascii-ci=0 ftype=1
log =internal log bsize=4096 blocks=457853, version=2
= sectsz=4096 sunit=1 blks, lazy-count=1
realtime =none extsz=4096 blocks=0, rtextents=0
bad magic number
bad magic number
Metadata corruption detected at xfs_sb block 0x0/0x1000
libxfs_writebufr: write verifer failed on xfs_sb bno 0x0/0x1000
the github issue link: https://github.com/spdk/spdk/issues/1738
After investigation, we found as below:
The qemu set to VHOST_USER_SET_MEM_TABLE then call vhost_user_set_mem_table(dpdk version spdk-20.11)->mmap(without MAP_POPULATE)
the mmap will only return the virtual address not map the virtual address to the physical address actually.
after call vtophys_notify->vtophys_get_paddr_pagemap, we will get a wrong physical(the virtual address we not access)
So this patch is to set MAP_POPULATE during we use SPDK + UIO.
Signed-off-by: andypma <andypma@tencent.com>
Reviewed-by: Zhigang Lu <tonnylu@tencent.com>
Reviewed-by: Xun Ni <richardni@tencent.com>
Reviewed-by: YuZhoujian <windyu@tencent.com>
Reviewed-by: LuWeiKang <luweikang@tencent.com>
Change-Id: Ib921b6d7381f46bb5894ed787cef52e02524197a
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/5970
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Ziye Yang <ziye.yang@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
