Allow for alternative fio job layout.
"group" (default) will create X fio job sections for
X assigned CPUs and distribute available NVMes between
sections. (e.g. 4 CPU test with 24 NVMes will result in
4 fio job sections with 6 NVMes each and scaled iodepth)
"split" will create a separate job section for each
available NVMe, but still pin it to desired CPU thread
using cpus_allowed option.
This allows for alternative way of testing io_uring
which fails to run some cases with fixedbufs=true and
high iodepth values (caused by system-wide default
UIO_MAXIOV=1024).
Signed-off-by: Karol Latecki <karol.latecki@intel.com>
Change-Id: Iab835596a59f6437b067d1117208e5abd6c8d36b
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/5727
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Michal Berger <michalx.berger@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Signed-off-by: Maciej Wawryk <maciejx.wawryk@intel.com>
Change-Id: I00998ce36b46902687065c5fa4684da9b7dd26d4
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/5541
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Reviewed-by: Karol Latecki <karol.latecki@intel.com>
Save latency log only when needed (i.e. latency tests).
Otherwise in some condition (high IO tests) this may stress
fio too much and cause a crash or fio might just refuse
to run the job.
Signed-off-by: Karol Latecki <karol.latecki@intel.com>
Change-Id: I61993d05ba91e88e9b6580d8ee39f255e1b09076
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/5476
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Community-CI: Mellanox Build Bot
io_uring engine benefits a lot from these options
for better results.
Signed-off-by: Karol Latecki <karol.latecki@intel.com>
Change-Id: I1fbc62b88e1726f4ea1569c6ee6997164a9d09a6
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/5475
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: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Community-CI: Mellanox Build Bot
Similarly replace PCI_BLACKLIST with PCI_BLOCKED.
Use of ALLOWED/BLOCKED matches similar changes made
in DPDK.
While here, replace use of term "blacklist" with "blocked"
in one of the nvme perf scripts. The usage there was
associated with how devices are blocked by using the
environment variables that are changed by this patch.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I720d99118ba5e050f436612c9fd415db44294a63
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/5275
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: Changpeng Liu <changpeng.liu@intel.com>
POSIX defines PRId64/PRIu64/PRIx64 for printing 64-bit values in a
portable way. Replace a few references to %ld to remove the assumption
about the size of a long. Similarly, use %z with size_t arguments.
Where the value being printed is an unsigned 64-bit value, use PRIu64
instead of %ld.
Add a cast when comparing an enum to an integer value.
Add an explicit uint64_t cast to the definition of BLOCK_CNT to prevent
value truncation.
Signed-off-by: Nick Connolly <nick.connolly@mayadata.io>
Change-Id: I093cffe7e79bc04eba012f4ffaf33062a642d062
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/5137
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>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
This fixes link issues with clang on some systems.
Fixes another part of issue #1613.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Icd51351b8e38043f267918b85e8ed1c3daab2db7
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/4860
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Need to select proper path for passing libs into the VM
in case we're building with custom DPDK.
Signed-off-by: Karol Latecki <karol.latecki@intel.com>
Change-Id: I97d301c70adee31b727c6b6673eadac3cbde9817
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/4975
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
We want to replace few qcow2 images with one universal.
This commit contains:
- change password in autotest.sh
- change image path
- change image name
- use snapshot mode in hotplug.sh instead of copying base image
Signed-off-by: Maciej Wawryk <maciejx.wawryk@intel.com>
Change-Id: I75c457fe75f005b0ab43ca909be7886529ed115b
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/4551
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Reviewed-by: Pawel Piatek <pawelx.piatek@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Use spdk_nvme_detach_async() and spdk_nvme_detach_poll_async() with
a local variable detach_ctx to detach multiple controllers.
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: I77cdc07ddfcd97569e31eeb245cd4e5a26289dbd
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/4441
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
From the log we can see that it reports ns_ctx was freed before it
was used. In unregister_controller, it still need ns_ctx to do io_complete.
And now this app works as a signle worker to do something.
Just need to change the order, execute unregister_worker after
unregister_controllers.
Fixes issue #1626
tested on clearlinux.
Signed-off-by: yidong0635 <dongx.yi@intel.com>
Change-Id: I392ff168abb862ba236a24b775fb31bb341e2157
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/4761
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
This function does need to handle legacy configs and
it's single use did not request any wait_for_spdk_bdev.
Remove those arguments.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: Ic988762c7ca622ff538aeb7bd2bb529e6e6d56e4
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/4616
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>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Log flags could also report on SPDK_LOG_INFO level by using
SPDK_INFOLOG() macro. Yet this did not result in additional
log due to check for debug build.
This patch allows release builds to use that flag.
Meanwhile the -L option besides turning on particular log flag
changes print log level to SPDK_LOG_DEBUG.
Applied changes to serveral applications to follow the
behaviour of event framework.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: Ib0d6cce33263c350d6d83300d60e7d15bdfe4b64
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/4557
Reviewed-by: Jacek Kalwas <jacek.kalwas@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
AIO not working because the check for $OS==Linux
is before the $OS is set. Therefore, the binary
build does not include libaio
Change-Id: Iab862d0113d637134a5275d46252c60fe27dde5e
Signed-off-by: John Kariuki <John.K.Kariuki@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/4516
Reviewed-by: Mellanox Build Bot
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Community-CI: Broadcom CI
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Using TAILQ instead of next pointer will make the object relationship
cleaner and the asynchronous detach opearation easier to implement.
Besides, do the following minor clean up.
Rename g_workers by g_worker and register_workers() by register_worker()
because it holds only a single instance, and clean up references to g_worker.
To fix memory leak at cleanup, add unregister_worker() to free a worker
and its associated contexts.
The prefix, spdk_reset, was not necessary as function name. Rename
spdk_reset_free_tasks() by free_tasks().
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: I0da11108222491e3290ffa5e405eb3ebe70a91bc
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/4435
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
This will make the object relationship cleaner and the asynchronous
detach operation easier to implement.
Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Change-Id: I3095d59b632ea2fc29fc44f8da330cb98c50bed1
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/4434
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: Ben Walker <benjamin.walker@intel.com>
Here code returns but context isn't freed.
And we can change to break here also.
Signed-off-by: yidong0635 <dongx.yi@intel.com>
Change-Id: I8f2173ae3599c13c99c7d70c19e2f8ce0c023f38
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/4396
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>
Print bdev.conf we're going to use in test
after generating it.
Signed-off-by: Karol Latecki <karol.latecki@intel.com>
Change-Id: I8b74b8afb9c4fca200b961d0d0d9818578ef1307
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/4281
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Maciej Wawryk <maciejx.wawryk@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Setting should be "-1" for polling, otherwise "0"
works as hybrid-polling.
Signed-off-by: Karol Latecki <karol.latecki@intel.com>
Change-Id: I5f1fb602f750162a7c164703859f2e74c9092146
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/4280
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Maciej Wawryk <maciejx.wawryk@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
These were not really needed and parsing functions had to
be unnecessarily called multiple times.
Signed-off-by: Karol Latecki <karol.latecki@intel.com>
Change-Id: I277125bbb3d598a348164e8834843d0d04f1434c
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/4098
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Michal Berger <michalx.berger@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Calculating kernel fio results did not take
number of workload retries into account for latency
stats.
Signed-off-by: Karol Latecki <karol.latecki@intel.com>
Change-Id: I418845d2f6be328827e4217c2eca761b0c211504
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/4096
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Maciej Wawryk <maciejx.wawryk@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Michal Berger <michalx.berger@intel.com>
Using only Bash for calculation in this script
results in operating only on integers, which lead
to wrong results being saved after test run.
For example - submission latency of 0.175usec was
saved as 0usec.
Change-Id: I6f9924d12189631ab6d0bdc5385e642b6c2aa34f
Signed-off-by: Karol Latecki <karol.latecki@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/4013
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Michal Berger <michalx.berger@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
This removes all the sleeps called after setup.sh reset. These
sleeps were meant to wait for devices given tests depend on.
Change-Id: Ibb86c9f6f8d5f1b05d165e84d9019530af84f9ea
Signed-off-by: Michal Berger <michalx.berger@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/4035
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
nblocks is zero based, so read path was missing the increment.
NVMe device that cuse represents can be of any block_size,
so rather than hardcoding 512 - actually verify it.
Both paths didn't request enough of a buffer from cuse.
Reported-by: Niklas Cassel
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I228dc2572bc94ecbcb913e950d912a7ab5be9434
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/4037
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>
When opal_revert_and_init() is interrupted for some reason,
the spdk_tgt still exists, but it should be killed at the same time.
Change-Id: I8546d3b0b4d6a0fda1687558a664decb535ef2b4
Signed-off-by: WANGHAILIANG <hailiangx.e.wang@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/3830
Community-CI: Mellanox Build Bot
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@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>
Originally the idea was to disable error checking,
to match output from Kernel and SPDK NVMe cuse.
This includes passing test commands and failures.
Any discrepancy would be caught by log output diff
at the end.
Flaw in this logic is that test command itself might
be incorrect. We shouldn't depend on that, nor
attempt to cover up some of the failures even if
they occur on both interfaces.
Most probable cause for this at all, was NVMe emulated
in QEMU not really working with all the nvme-cli commands
from this test.
Since the original creation of this test, CUSE executes
on physical devices (to be able to support namespace management).
The behavior there is predictable and works with current
test commands, thus the test exits on any error with this patch.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I086faf38b2cbbb6225935cc50d4fad14e81f1972
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/3032
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>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Karol Latecki <karol.latecki@intel.com>
We can do better than "sleep 5". Instead use waitforfile(),
that will wait until the namespace is available.
Moved around the variables with names and checks for the ctrlr/ns
existence after all the tests complete.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: I8326c2cec0d0cf5aec88b5f80c1126c81eb09436
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/2951
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Karol Latecki <karol.latecki@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
If intel_pstate is disabled in boot options, allow to
change CPU frequency using "userspace" cpu governor.
Change-Id: I390492c7be28df408f3cbb774ba5360859649fe0
Signed-off-by: Karol Latecki <karol.latecki@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/3376
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>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
For some reason the routine for generating fio
job/filename sections was different for kernel and
spdk fio modes. This made fio results hard to compare
because even configuration files themselves were not
comparable.
The function was modified to create the same fio
configuration files for kernel and spdk.
While at it, iodepth calculation has been slightly
modified as well. It is now being calculated for each
[ ] job section, so that each "filename" gets the
actual iodepth value specified in test run parameters.
While generating the config file the script will add
NUMA node information as a comment for "cpus_allowed"
and "filename" sections, so that it's clearly visible
how resources are aligned. In case of cross numa
configuration additional warnings will be printed to
screen as well.
For example, a configuration file for running test with
7 NVMe drives, 3 CPUs and effective (per filename) queue
depth of 32 would look like this:
run_perf.sh --cpu-allowed=0,1,2 --iodepth=32
--driver=kernel-classic-polling [...other params...]
--------
[filename0]
iodepth=96
cpus_allowed=0 #CPU NUMA Node X
filename=/dev/nvme0n1 #NVMe NUMA Node X
filename=/dev/nvme1n1 #NVMe NUMA Node X
filename=/dev/nvme2n1 #NVMe NUMA Node X
[filename1]
iodepth=64
cpus_allowed=1 #CPU NUMA Node X
filename=/dev/nvme3n1 #NVMe NUMA Node X
filename=/dev/nvme4n1 #NVMe NUMA Node X
[filename2]
iodepth=64
cpus_allowed=2 #CPU NUMA Node X
filename=/dev/nvme5n1 #NVMe NUMA Node X
filename=/dev/nvme6n1 #NVMe NUMA Node X
--------
Change-Id: Ida2f781fbb93c4a8c62154e711151152843ab997
Signed-off-by: Karol Latecki <karol.latecki@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/3370
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Reviewed-by: Michal Berger <michalx.berger@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Configuration file generation was split into multiple
places/functions in scripts. Move it into a single
function. "time_based" fio parameter has been moved to
fio config template file as it is not configurable.
There is still a lot of room for improvement here, because
create_fio_config() function uses mixed local variables
(as function arguments) and global variables.
Change-Id: Ica3717d0d28692d24c861c2d32e729761c5c8647
Signed-off-by: Karol Latecki <karol.latecki@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/3345
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>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
A workaround for kernel deadlocks surfaced in #1275.
DPDK basically offers two APIs for hotplugging all PCI devices:
rte_bus_scan() and rte_bus_probe(). Scan iterates through
/sys/bus/pci/devices/* and creates corresponding rte_pci_device-s,
then rte_bus_probe() tries to initialize each device with the
supporting driver.
Previously we did scan and probe together, one after another, now
we'll have an intermediate step. After scanning the bus, we'll
iterate through all rte_pci_device-s and temporarily blacklist any
newly detected devices. We'll use devargs->data field to a store
a timeout value (integer) after which the device can be un-blacklisted
and initialized. devargs->data is documented in DPDK as "Device
string storage" and it's a char*, but it's not referenced anywhere
in DPDK. rte_bus_probe() respects the blacklist and doesn't do
absolutely anything with blacklisted ones.
The timeout value is 2 seconds, which should be plenty enough
for an NVMe device to reset, leave the critical lock sections in
kernel, and let us initialize it safely.
Note that direct attach by BDF doesn't respect the blacklist,
so an NVMe attach RPC won't be delayed in any way, it will continue
to work as it always did. Only the automatic discovery & enumeration
is deferred.
Change-Id: I62b719271bd0755bc2882331ea33f69897b1e5e5
Signed-off-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1733
Community-CI: Mellanox Build Bot
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>
There's no reason to keep usage and getopts
outside of the main script.
Other small changes while at it:
- use small letters for all "dir" variables
(consistent with most of other test scripts)
- try to move default & global variables into
one place.
- remove long comment from top of the main script.
It was basically the same as "help" for getopts.
Change-Id: Ic99f4bb1808b55dea80b534f7b4e0a65595ed053
Signed-off-by: Karol Latecki <karol.latecki@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/3335
Community-CI: Mellanox Build Bot
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Read CPU list to use in test directly from option
value of from a file if it's an existing path.
This will make running tests in CI a bit easier.
Signed-off-by: Karol Latecki <karol.latecki@intel.com>
Change-Id: I174a38ab615c2f2cd2c3b4a95e0ebf886c478a51
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/3333
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Allow to set custom size of bdev IO cache and pool
sizes.
Change-Id: I574d1e949f705d8f834539d5879f0a1e72865e00
Signed-off-by: Karol Latecki <karol.latecki@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/3297
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Use a simple config file to select NVMes for test
instead of selecting them all via gen_nvme.sh.
HW platforms may differ significantly in setup and
capabilities, so using all disks in sequential order
is not always a good way.
(e.g. using "first" four available NVMes for test
may yield not optimal results if it turns out that
NVMes are connected to oversubscibed PCIe switch)
Change-Id: I55dd85799a5859c4764e94e1d8058e01ad9b84f0
Signed-off-by: Karol Latecki <karol.latecki@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/3296
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Script was written in such a way that tests would be
performed starting with all found NVMe disks in configuration.
Then the number of disks for the workload would be reduced in
loop until reaching 1. This is the scenario used in test
case 1, described in SPDK NVMe Performance Benchmark document.
--disk-no allowed to avoid running test in loop, and specify
just a single value for the number of disks to be tested.
Both loop logic and --disk-no are no longer required as
tests are scheduled and run via CI system which picks the
parameters for each test run.
Change-Id: Iad0f4d0d259ac64680fd5bb69d7cbb7e0afff79c
Signed-off-by: Karol Latecki <karol.latecki@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/3295
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
- Change divisor in nsec to usec calculation to "1000".
- Ignore MIX param when rw mode is set to 100% read or
100% write.
- Use MIX param as number in 0.0-1.0 range for calculating
summary read+write latency.
- Use jq "//" alternative operator when querying for fields
which might not exist in JSON object. Assume "0" for
non-existent fields, otherwise jq reports "null", which
in turn results in script errors.
Change-Id: Ibc8f91f058d9ae31ee1b60f4b253c42b743ae22f
Signed-off-by: Karol Latecki <karol.latecki@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/3294
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Previous changes moved some of the fio CLI parameters into
a bash heredoc to save into a text file, but it was put in
wrong place in the script.
Move global fio parameters a couple lines up so that the
resulting fio config file has the correct form of:
[global]
global_params
[testjob]
testjob_params
Signed-off-by: Karol Latecki <karol.latecki@intel.com>
Change-Id: I3a958fe5ce880ab63a079c345e6b24fd1d3bae32
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/3293
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Michal Berger <michalx.berger@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
