Currently each controller is associated with one of the poll groups.
For each controller, all sessions are associated with the poll
groups with which the corresponding controller is associated.
Vhost poll group does not have any polling loop but its usage was
very complex.
Association of controller with poll group is done based on the
specified cpumask, and poll group is created per CPU core.
This is as same as association of thread with CPU core.
So in this patch, replace poll group per session by thread per
controller.
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: Ifa1e136caae11959f7b097b06a22910bc2169b30
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1146
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: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
The next patch will create a SPDK thread in vhost_dev_register()
and exit the SPDK thread in vhost_dev_unregister().
As a preparation, simplify error paths in vhost_dev_register()
by changing some gotos to return, moving free after the out label,
and moving insertion after succeeding vhost_register_unix_socket().
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: Id1a2a50b4ba5732f91598a326a08de5c652fd136
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1145
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
This way if nothing prints it's just using software. If CDMA
is being used it will print that. Will do the same for DSA
once added.
Signed-off-by: paul luse <paul.e.luse@intel.com>
Change-Id: I44c49cf3a495d5e8440e80fa42fbb6f1338c5aac
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1183
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
This is a preparation to support thread_get_io_channels RPC
in the next patch. The next patch will refer only name of struct
io_device, and so adding a helper function is better than making
struct io_device public.
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: I9bc9dd78f71aa1e32847dc4c67eb79dae9cfbf9d
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/886
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>
Add an new JSON RPC thread_get_pollers to retrieve pollers of all
the threads. By adding a helper function spdk_poller_state_str(),
output poller state as string to improve readability. Most of the
code of thread_get_stats and thread_get_pollers are common and so
unify these two RPCs as possible as we can.
Sample output of thread_get_stats RPC in doc/jsonrpc.md was wrong
because thread_get_stats doesn't output ticks. Fix this together.
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: I003ffe569d3c0651ae65c5858eff8287f7e9031d
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/604
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>
This was recently made asynchronous to support virtualized transports.
However, we're moving to add a new call to associated a listener with a
subsystem to transports and the operation that needed to be asynchronous
will actually be performed there. For simplicity, make this synchronous
again.
Change-Id: Ie98136a19c58f0f9bba0d140476de3bbb38e12d7
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/881
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: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
This gets rid of some duplicate lines of code.
Change-Id: I24d4864921f6030672f3640b33f88f37a9e8175a
Signed-off-by: Seth Howell <seth.howell@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1136
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: Aleksey Marchuk <alexeymar@mellanox.com>
Further part of the series will hold array of md pages
in the ctx. Callers of _spdk_bs_load_replay_md_parse_page()
will make select which page to parse.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I3fb70660672ba74bdb338eb1233409103903b215
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/983
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>
This is refactoring change for future patches.
struct spdk_bs_load_ctx will contain array of pages
instead of single one. Having to change just single
line for selection of page will make it easier to
read next patches in series.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: If3dc1e7da7e61c7b4866307d859e55131a32d38b
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/982
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>
New spdk_ftl_conf parameter l2p_path, l2p_path defines
location of existing pmem file or device to use as l2p table.
ftl_dev_l2p_alloc now has two flows:
1. If l2p_path is set perform pmem_map_file (PMDK) of l2p_size on it
2. Else malloc l2p table as usual
l2p_get/set will use atomic_load/store on both of those storage
locations as PMDK pmem_memcpy family of functions is not thread safe.
Signed-off-by: Maciej Szczepaniak <maciej.szczepaniak@intel.com>
Change-Id: I91806feb7aa0ef8057792bc120b09a39c63c8640
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/649
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Wojciech Malikowski <wojciech.malikowski@intel.com>
Reviewed-by: Konrad Sztyber <konrad.sztyber@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
When using vfio with enable_unsafe_noiommu_mode=Y, force iova-mode
as "pa" here for DPDK guesses it's "va", which cause the following
error: "EAL: Expecting 'PA' IOVA mode but current mode is 'VA',
not initializing".
Signed-off-by: Richael Zhuang <richael.zhuang@arm.com>
Change-Id: I7c343498c5d6976a7c75d75438d6f9c35f1b6160
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1071
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 some cases, conn->sock will be lost, while conn pointer will still be.
A runtime error will apperars: null pointer of type 'struct spdk_sock'.
This only happens when ./configure with asan.
So I add a sssert to display this error explicitly when it occurs.
Change-Id: I7d012ac1a29a7fb0bce4815e0622582f23222a25
Signed-off-by: WANGHAILIANG <hailiangx.e.wang@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1102
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Also rename __initialize_cache()/__free_cache() to the new name
to reflect the logic.
Change-Id: I69bec4a10b2f21a7c40475fc2a99919bc526c556
Signed-off-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/976
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>
Currently we will free the cache buffers in the file R/W
data path, however we can start a thread to do the cache
buffer reclaim regulary, so that we don't need to free
the cache buffer in the Rocksdb thread context. By
doing this, we can also remove the global cache buffer
lock in following patches.
Change-Id: Icb0fbc49baecd1e9d86a5fcfe400758dfc3c53a2
Signed-off-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/941
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>
Replace g_tmp_cpuset by tmp_cpuset local in vhost_get_poll_group()
without any side effect.
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: Ibb01d2c77d6d82d7cc81df722e7848b357d4e9e7
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/840
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
We use `spdk_map_bar_rte()` to read mapped addresses
from PCI BARs.
This function is currently checking for NULL in each pair.
But in PCI memory, some registers can be left unused,
in which case they are set to 0.
As a result, we may read some NULL pointers from BARs,
which is OK.
To check if given address is indeed invalid, we should first
check if it is used.
So it is best to delegate such checks to the
user of this function.
In fact, users already do the NULL check where it is needed
(ex: virtio_pci.c:390, nvme_pcie.c:589)
so this patch just removes them from `spdk_map_bar_rte()`.
This solves github issue #1206
Change-Id: I88021ceca1b9e9d503b224f790819999cd16da01
Signed-off-by: Vitaliy Mysak <vitaliy.mysak@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1129
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>
Fix Segmentation fault on the target side.
Issue:
rdma.c:2752:spdk_nvmf_rdma_listen: *NOTICE*: *** NVMe/RDMA Target Listening on 192.168.35.11 port 4420 ***
rdma.c: 789:nvmf_rdma_resources_create: *ERROR*: Unable to allocate sufficient memory for RDMA queue.
rdma.c:3385:spdk_nvmf_rdma_poll_group_create: *ERROR*: Unable to allocate resources for shared receive queue.
Segmentation fault (core dumped)
GDB:
Program terminated with signal 11, Segmentation fault.
736 if (resources->cmds_mr) {
(gdb) bt
736 if (resources->cmds_mr) {
(gdb) bt
0 nvmf_rdma_resources_destroy (resources=0x0) at rdma.c:736
1 0x0000000000497516 in spdk_nvmf_rdma_poll_group_destroy (group=group@entry=0x2fe1300) at rdma.c:3489
2 0x00000000004978bb in spdk_nvmf_rdma_poll_group_create (transport=0x2fe11d0) at rdma.c:3371
3 0x000000000048df70 in spdk_nvmf_transport_poll_group_create (transport=0x2fe11d0) at transport.c:267
4 0x000000000048a450 in spdk_nvmf_poll_group_add_transport (group=0x2f49af0, transport=<optimized out>) at nvmf.c:941
5 0x000000000048a6cb in spdk_nvmf_tgt_create_poll_group (io_device=0x2fce600, ctx_buf=0x2f49af0) at nvmf.c:122
6 0x00000000004a0492 in spdk_get_io_channel (io_device=0x2fce600) at thread.c:1324
7 0x000000000048a0e9 in spdk_nvmf_poll_group_create (tgt=<optimized out>) at nvmf.c:723
8 0x000000000047f230 in nvmf_tgt_create_poll_group (ctx=<optimized out>) at nvmf_tgt.c:356
9 0x000000000049f92b in spdk_on_thread (ctx=0x2f81b20) at thread.c:1065
10 0x000000000049f17d in _spdk_msg_queue_run_batch (max_msgs=<optimized out>, thread=0x1e67e90) at thread.c:554
11 spdk_thread_poll (thread=thread@entry=0x1e67e90, max_msgs=max_msgs@entry=0, now=now@entry=947267017376702) at thread.c:623
12 0x000000000049af86 in _spdk_reactor_run (arg=0x1e678c0) at reactor.c:342
13 0x000000000049b3a9 in spdk_reactors_start () at reactor.c:448
14 0x0000000000499a00 in spdk_app_start (opts=opts@entry=0x7ffc2a5e0ce0, start_fn=start_fn@entry=0x40aa80 <nvmf_tgt_started>,
arg1=arg1@entry=0x0) at app.c:690
15 0x0000000000408237 in main (argc=5, argv=0x7ffc2a5e0e98) at nvmf_main.c:75
Signed-off-by: yidong0635 <dongx.yi@intel.com>
Change-Id: Id9bf081964d0cf3575757e80fc7582b80776d554
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1073
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
To avoid the strange formatting, typedef has been used. But this
comment is hard to get the meaning. So stop breaking after return
type for this case.
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: Ia03d6ec50610c395007fe172018b890733dce599
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1052
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
To avoid the strange formatting, typedef has been used. But this
comment is hard to get the meaning. So stop breaking after return
type for this case.
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: I317604de67aab9f201d691e0f886bd673f451f3f
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1051
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
To avoid the starnge formatting, typedef has been used. But this comment
is hard to get the meaning. So stop breaking after return type for this case.
The strange formating is
struct spdk_scsi_dev *
spdk_scsi_dev_construct(const char *name, const char *bdev_name_list[],
int *lun_id_list, int num_luns, uint8_t protocol_id,
void (*hotremove_cb)(const struct spdk_scsi_lun *, void *),
void *hotremove_ctx)
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: I20c30b225b3b96fa9207b4a89e3210c8a97fb1d8
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1050
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Add an new API spdk_poller_register_named() to set arbitrary name
to the created poller. If NULL, the name is set to the pointer of
the poller function.
To set the name to the string of the poller function name conveniently,
add an new macro SPDK_POLLER_REGISTER() together in this patch.
All debug or error logs are changed to output poller name from pointer.
The added name will be used in the new RPC thread_get_pollers.
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: I3be558dd795252f797e3e81fa2db2e8b128cf004
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/506
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Maciej Szwed <maciej.szwed@intel.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
thread_get_pollers RPC which will be added in the upcoming patches
will need to access internal of all pollers.
Following the last patch, expose struct spdk_poller internally among
SPDK libraries.
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: I6844fc70165b4f127c49680ce592ac7b8c326cac
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/594
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
thread_get_pollers RPC which will be added in the upcoming patches
will need to access all pollers each thread has.
To avoid adding JSON related code into lib/thread/thread.c, expose
struct spdk_thread internally among SPDK libraries and RPC code will
access it.
The next patch will expose struct spdk_poller internally among
SPDK library for the same purpose.
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: I8da039db3021966ca1e28f6f086bb4c2a8eeb84a
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/973
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Use the payload to construct iovs so that split desc
or packed desc can both work.
Change-Id: Ib2c60b6eb53d0d615594e97b3d3ae1a365908b6c
Signed-off-by: Jin Yu <jin.yu@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1000
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>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Replaced single global write buffer with the per-io_channel write
buffers. This means that the "rwb" module and all of its references
were removed and replaced with the recently added interfaces.
Change-Id: Idc899d3a4d63a8a2bede1ac26549ed06e9a2e784
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/909
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: Wojciech Malikowski <wojciech.malikowski@intel.com>
Added queue responsible for keeping track of full, ready to be written
batches. A batch might be put on this queue in case it'd already been
filled, but could not be written out due to lack of resources or it was
written out, but failed and needs to be resent.
Change-Id: Iba49cd359425300d21b8100f13f17189e99d4c7c
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/908
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: Maciej Szczepaniak <maciej.szczepaniak@intel.com>
Reviewed-by: Wojciech Malikowski <wojciech.malikowski@intel.com>
Metadata buffer was added to ftl_batch structure. It'll make it
possible to send writes with a separate metadata buffer.
Change-Id: I475c0982809cb0cc9aa6bcbc2a4935a6f15f19af
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/907
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>
Reviewed-by: Wojciech Malikowski <wojciech.malikowski@intel.com>
Reviewed-by: Maciej Szczepaniak <maciej.szczepaniak@intel.com>
Two functions were added:
- ftl_get_addr_from_entry() returning an address pointing at a given
entry
- and it's inverse, ftl_get_entry_from_addr(), returning a write buffer
entry for a given address
Change-Id: Iea1e2e59f240f50183fdb549280a0f0b722efbef
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/906
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Wojciech Malikowski <wojciech.malikowski@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
This patch adds the ftl_batch strucutre, which describes a single batch
of data to be written to the disk. It's comprised of multiple write
buffer entries (the actual number depends on the write unit size of the
underyling device).
Additionally, a function responsible for filling out a batch was added.
It iterates over available IO channels and dequeues submitted write
buffer entries until a batch is completed. The IO channel queue is
shifted, so that each time the function is called a subsequent IO
channel is used first, which guarantees that all channels are treated
fairly.
Change-Id: Ie61d8c6cb51d5c5175540c975447bc27590c5cb4
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/905
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Wojciech Malikowski <wojciech.malikowski@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Added functions responsible for acquiring/releasing write buffer entries
from the per-io_channel caches. They're also in charge of limiting user
writes by reducing the number of available entries to the user.
Change-Id: I23adfe36065139036f9b72a016e34fbf6eb29230
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/904
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Wojciech Malikowski <wojciech.malikowski@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
This patch introduces the per-io_channel write buffer entry, the
structures necessary for its management, and performs their
initialization. They will be utilized by the upcomming patches.
Change-Id: I772caa445d1887166b9bcd7679d32d2ed2461ef3
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/903
Reviewed-by: Wojciech Malikowski <wojciech.malikowski@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
The IO channel pointers are now stored inside an array of the device
they belong to. Once write buffer entries are tied to IO channels,
it'll provide a method for dereferencing an entry from its address.
Change-Id: Iaf401525eb0f5af8dc6047a1dc8bae11b56761d7
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/901
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Wojciech Malikowski <wojciech.malikowski@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
This is the first patch of a series that replaces single global write
buffer with per-io_channel buffers. This change is intended to improve
performance for multithreaded workloads.
This patch changes the way the ftl_io_channels are allocated by only
keeping an ftl_io_channel pointer inside spdk_io_channel's context. It
allows for delaying IO channel destruction, which in turn allows the FTL
to iterate over all exisiting IO channels without locking.
Change-Id: I5e0cab8043a2b5f747e971dd3d65ed2546c8cf26
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/900
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Wojciech Malikowski <wojciech.malikowski@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Add a flag resched to check if reschedule operation is requested
to struct spdk_lw_thread. Add _reactor_resquest_thread_reschedule()
to set the resched flag, and add it to the case SPDK_THREAD_OP_RESCHED
in spdk_reactor_thread_op(), and return true in the case
SPDK_THREAD_OP_RESCHED in spdk_reactor_thread_op_supported().
Then _spdk_reactor_run() checks if the resched flag is true for each
thread. If true, set the resched flag to false, and remove the
thread and call _reactor_schedule_thread(). Add continue to avoid
use-after-free issue for both reschedule and terminate cases.
This idea follows voluntary thread termination and will remove our
worries for all complicated rare cases.
Add unit test case to verify this update.
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: I656872d32dbb469ae70f771cd0419a77236bfe18
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/500
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>
Remove checks for open bands inside because
we should only support closed bands relocation.
Change-Id: Ice5f13d579a046a9cb46684488801a2fa0c15a9d
Signed-off-by: Wojciech Malikowski <wojciech.malikowski@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/876
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>
Since ftl supports only one core thread ftl_thread
structure is not needed anymore.
Change-Id: I66eda9249d74b3eff0c4473dcd0aba4a6135e296
Signed-off-by: Wojciech Malikowski <wojciech.malikowski@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/548
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>
Reviewed-by: Maciej Szczepaniak <maciej.szczepaniak@intel.com>
Reviewed-by: Konrad Sztyber <konrad.sztyber@intel.com>
Since relocation can be done only on core thread there is
no need for spdk_ring. Move queue size is limited by
reloc->max_qdepth so we are trying to iterate over all
move elements in move queue when processing band relocation.
Change-Id: Ic4cfaeeab4b5f3dd3e1dbd967908a3e6c9b3ed46
Signed-off-by: Wojciech Malikowski <wojciech.malikowski@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/547
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>
Add transport_ack_timeout parameter to nvme controller opts.
This parameter allows to configure RDMA ACK timeout according
to the formula 4.096 * 2^(transport_ack_timeout) usec.
The parameter should be in range 0..31 where 0 means use
driver-specific default value.
Change-Id: I0c8a5a636aa9d816bda5c1ba58f56a00a585b060
Signed-off-by: Alexey Marchuk <alexeymar@mellanox.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/502
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>
Add an new API spdk_thread_set_cpumask() and enum SPDK_THREAD_OP_RESCHED.
spdk_thread_set_cpumask() can be called only from the current thread
and requires SPDK thread operation supports reschedule operation.
spdk_thread_set_cpumask() updates the cpumask of the current thread to
the specified value, and then invokes framework's reschedule operation
to the thread.
If spdk_thread_set_cpumask() calls multiple times in a single
spdk_thread_poll() context, the last value will be used in the
reschedule operation.
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: I7808626b10269543c1e2cd86793a504daa4b6389
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/499
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>
Remove the prefix "spdk" from spdk_reactor_schedule_thread(), add
spdk_reactor_thread_op() and spdk_reactor_thread_op_supported().
For SPDK_THREAD_OP_NEW, spdk_reactor_thread_op() calls
_reactor_schedule_thread() and spdk_reactor_thread_op_supported()
returns true.
Then replace spdk_thread_lib_init() by spdk_thread_lib_init_ext()
with spdk_reactor_thread_op() and spdk_reactor_thread_op_supported().
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: I232a3b2c6bcaf4d86b0dd3cefacd3e47eadda6d8
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/968
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 enum spdk_thread_op and two function typedefs spdk_thread_op_fn
and spdk_thread_op_supported_fn.
The first operation type of enum spdk_thread_op is SPDK_THREAD_OP_NEW,
and it is used as an alternative to spdk_new_thread_fn.
Add global variables, g_thread_op_fn and g_thread_op_supported_fn, and
then add spdk_thread_lib_init_ext() to initialize these.
spdk_thread_lib_init() requires both of thread_op_fn and
thread_op_supported_fn are specified or not specified.
spdk_thread_create() calls g_thread_op_fn() with SPDK_THREAD_OP_NEW
if g_new_thread_fn is NULL, g_thread_op_supported_fn is not NULL,
and g_thread_op_supported_fn(SPDK_THREAD_OP_NEW) returns true.
Update unit test to test these addition.
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: I56db903f62437f6ff3198248ffc5dede396c22bc
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/967
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 an new API spdk_thread_get_by_id(). This will be used in the
subsequent patches to set the cpumask of the running thread to the
specified value.
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: Ib5d02b9d7b499477c43e6527cf8f603d8323e063
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/966
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Add an unique ID for each created SPDK thread. Use a single 64 bits
variable, g_thread_id, and guard its update by the global mutex
g_devlist_mutex. For our safety, further thread creation is not
allowed if g_thread_id rolls over, and request user to restart SPDK
application.
Besides, as a minor update, move the debug log down and add ID to it
in spdk_thread_create(), and ID is added to thread_get_stats RPC and
framework_get_reactors RPC.
The thread ID will be used to set the cpumask of the running thread
to the specified value in the subsequent patches.
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: Ic09f11d4c7175c3b89acba6a42e76063acd0d1a0
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/498
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>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
There is a warning triggered when holding ref to const obj and passing
to these getters.
Signed-off-by: Jacek Kalwas <jacek.kalwas@intel.com>
Change-Id: I2c7b4ea0d325d84d66923fc524273ea44a3a311b
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/997
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>
Users may only set the transport type, but for the actual probe
process, the trstring field is mandatory, so set the trstring
based on transport type at first. Also remove unnecessary
spdk_nvme_trid_populate_transport() call from each transport
module.
Fix#1228.
Change-Id: I2378065945cf725df4b1997293a737c101969e69
Signed-off-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1001
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
In the autotest, when calling kill_stub() function, there is error log
like this: "Device 0000:83:00.0 is still attached at shutdown!", so it's
better to detach the controller when exit the stub process.
But after call spdk_nvme_detach() in the stub process, there is another issue:
1. NVMe stub running as the primary process, and it will send 4 AERs.
2. Using NVMe reset tool as the secondary process.
When doing NVMe reset from the secondary process, it will abort all the
outstanding requests, so for the 4 AERs from the primary process, the 4
requests will be added to the active_proc->active_reqs list.
When calling spdk_nvme_detach() to detach a controller, there is a
assertion in the nvme_ctrlr_free_processes() at last to check the
active requests list of this active process data structure.
We can add a check before destructing the controller to poll the
completion queue, so that the active requests list can be flushed.
Change-Id: I0c473e935333a28d16f4c9fb443341fc47c5c24f
Signed-off-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/977
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>
Solve the problem that the /sys/block/vd../max_segments is always 1 in the virtual
machine,and avoid the problem of low sequential read and write performance caused
by this limitation in the general block device layer of some lower kernels.
Signed-off-by: suhua <suhua1@kingsoft.com>
Change-Id: I3289b4c6c74d9225f34e117f444f310bf8bc8ea2
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/975
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>
There are synchronous security send/receive APIs defined in nvme.h,
however, we still need the asynchronous APIs so that we can make the
OPAL library can be used in asynchronous way. As the asynchronous APIs
are already defined in nvme_ctrlr_cmd.c, so just export them to public
APIs.
Change-Id: I5646f342a4bf70faad37daa956476f05a1327bcc
Signed-off-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/675
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>
