In case of TCP upload error, zperf would leak a socket when running in
asynchronous mode. The upload work have to release the socket it
allocated in any case, regardless of the session result.
Signed-off-by: Robert Lubos <robert.lubos@nordicsemi.no>
During socket setup, if one of the setsockopt() calls failed, the
function would return an error w/o closing the socket. That's wrong, as
in case of errors the function should clean up any resources it
allocated, the socket file descriptor is lost otherwise and resources
are leaked.
Signed-off-by: Robert Lubos <robert.lubos@nordicsemi.no>
Fixes remote address for ping before upload. This caused the ping in zperf
upload to timeout as shown in the following output:
```
uart:~$ zperf udp upload 2001:db8::2 5001 10 50 1M
Remote port is 5001
Connecting to 2001:db8::2
Duration: 10.00 s
Packet size: 50 bytes
Rate: 1000 kbps
Starting...
ping 2001:db8::2 timeout
Rate: 1.00 Mbps
Packet duration 390 us
```
Fixes: https://github.com/zephyrproject-rtos/zephyr/issues/68674
Signed-off-by: Ayush Singh <ayushdevel1325@gmail.com>
Add option for zperf tcp upload that will enable periodic result reporting.
This is useful for monitoring performance swings during a longer session.
Signed-off-by: Adam Matus <adam.matus@nxp.com>
TCP nodelay option is part of zperf upload params, but was not being
set in common zperf_prepare_upload_sock function. Move it there
to align with how other options are set.
Signed-off-by: Adam Matus <adam.matus@nxp.com>
Minor improvements to usage of zperf_shell upload.
Added common defaults for the upload commands arguments.
Print defaults in cmd help.
Throw warning when user sets baud rate for TCP upload.
Signed-off-by: Adam Matus <adam.matus@nxp.com>
Zperf upload multicast always use default interface.
Zperf download multicast cannot receive packets from other than
224.0.0.1 which is default multicast group.
Add zperf upload/download option -I <interface name> for multicast.
So that user can select interface for multicast.
Add join multicast group for zperf download.
Use the "device list" command to get the interface name as
follows:
"- ua (READY)" #uAP interface name
"- ml (READY)" #STA interface name
Multicast traffic commands:
zperf udp upload -a -I ua 224.0.0.2 5001 10 1470 1M
zperf udp download -I ua 5001 224.0.0.3
Signed-off-by: Fengming Ye <frank.ye@nxp.com>
For command zperf udp download 5001 192.168.10.1,
zperf will bind both ipv4 and ipv6 sockets on ipv4 address.
But bind ipv6 socket will fail, thus command return fail.
Fix it by check ip address when zperf download.
For ipv4 address only bind ipv4 socket.
For ipv6 address only bind ipv6 socket.
For unspecific address bind both ipv4 and ipv6 sockets.
Signed-off-by: Fengming Ye <frank.ye@nxp.com>
In iperf2, for multicast data, if it is a client, it will not wait for
the server's AckFIN packet. Because the iperf2 server will not send an
AckFIN packet.
So in zperf_upload_fin(), an error will occur when the zperf client
waits for the server's AckFIN packet.
Multicast only send the negative sequence number packet and doesn't
wait for a server ack can fix this issue.
Signed-off-by: Gang Li <gang.li_1@nxp.com>
The zperf received or sent bytes length and duration are in 32bits,
if running long-duration zperf test more than 20min, the value will
overflow, and the test result is wrong. Change it to 64bits can fix
this issue.
Signed-off-by: Maochen Wang <maochen.wang@nxp.com>
When the zperf command is called with '-S' option which means IP_TOS
for IPv4 and IPV6_TCLASS for IPv6, an error is printed and the
setting does not work. The socket option handling was changed by
commit 77e522a5a243('net: context: Refactor option setters'), but the
callers of option setters were not changed. This causes the IP_TOS
or IPV6_TCLASS option failed to set. The fix is to use uint8_t to
store the value of the -S option.
Signed-off-by: Maochen Wang <maochen.wang@nxp.com>
We select sockets service API in Kconfig but should select also
sockets API so that user does not need to set the sockets API separately.
Signed-off-by: Jukka Rissanen <jukka.rissanen@nordicsemi.no>
In case zperf session was aborted by the user (by for instance stopping
it from shell), or practically in case of any other
communication-related error, the zperf session could end up in a state
other than NULL or COMPLETED, with no way to recover. This made the
session no longer usable and eventually could lead to zperf being not
able to start a new session anymore.
Fix this by introducing zperf_session_reset() function, which resets the
session state back to defaults. The function is called when the zperf
receiver service is stopped.
Signed-off-by: Robert Lubos <robert.lubos@nordicsemi.no>
The issues found for UDP receiver were also identified for TCP receiver,
this commit applies practically the same set of changes as in case of
UDP.
Signed-off-by: Robert Lubos <robert.lubos@nordicsemi.no>
This commit fixes restarting of UDP receiver service, along with some
other minor cleanups:
* The core issue was udp_server_running flag not being cleared when
service was stopped. Fix this by introducing udp_receiver_cleanup()
which does all of the required cleanups when receiver service is
stopped. The function is called either when the application stopped
the service with zperf_udp_download_stop(), or when the service was
stopped due to error.
* net_socket_service_unregister() was not called on
zperf_udp_download_stop(), but only from the service callback - that
would not work in case there's no active communication.
* at the same time, net_socket_service_unregister() would be called from
the service callback in case of errors. Fix this, by making
udp_recv_data() only return an error, and let the service callback to
do the cleanup.
* Remove no longer used udp_server_run semaphore
* Remove udp_server_stop - with socket services it seems no longer
needed.
* zperf_udp_receiver_init() now returns an error, so that we don't
mark the service as running in case of socket/services error.
Signed-off-by: Robert Lubos <robert.lubos@nordicsemi.no>
The ratio between mbps and kbps, kbps and bps should be 1000, instead of
1024, as common sense.
The wrong ratio will decrease the Zperf throughput result.
Signed-off-by: Fengming Ye <frank.ye@nxp.com>
Make sure we send the entire packet buffer before bumping the packet
counter, send() does not guarantee that all of the requested data will
be sent at once with STREAM socket.
Signed-off-by: Robert Lubos <robert.lubos@nordicsemi.no>
Default behaviour should only bind to port independent of IP, this
allows even multicast/broadcast L4 traffic to be received.
User can always specify a specific address to bind using shell or
Kconfig or API.
Signed-off-by: Chaitanya Tata <Chaitanya.Tata@nordicsemi.no>
Using a generic IP for address set failures is confusing, esp. two same
prints (one for v4 and the other for v6), so, use explicit version.
Signed-off-by: Chaitanya Tata <Chaitanya.Tata@nordicsemi.no>
The compiler emits a "null where non-null expected" warning unless the
argument of strlen is non-null at compile time.
Signed-off-by: Florian Grandel <fgrandel@code-for-humans.de>
The current zperf tcp/udp download command doesn't provide the option
to bind the server to a specific host address. If there is more than
one interface, it will not be possible to test each interface with zperf
tcp/udp download command without building the Zpehyr.
This patch will add support for zperf tcp/udp download command to bind
server to host interface address.
Signed-off-by: Rahul Singh <rahul.singh@arm.com>
The zperf shell sends a IPv6 ping at the start when working
with IPv6. Convert the sending of the ping to use the new API.
Signed-off-by: Jukka Rissanen <jukka.rissanen@nordicsemi.no>
This is handy in testing of setting priority directly rather than
deriving from DSCP. Please note ICMP doesn't use net context.
This is applicable for both shell and API.
Signed-off-by: Chaitanya Tata <Chaitanya.Tata@nordicsemi.no>
The initial goal was to remove sys_clock_timeout_end_calc(). However,
several related issues have been fixed as well.
First this:
int64_t print_interval = sys_clock_timeout_end_calc(K_SECONDS(1));
/* Print log every seconds */
int64_t print_info = print_interval - k_uptime_ticks();
if (print_info <= 0) {
[...]
}
The above condition will simply never be true.
Then there is lots of back-and-forth time conversions using expensive
base-10 divisions for each loop iterations which is likely to impact
performance.
Let's do the time conversion only once outside the loop and track
everything in terms of ticks within the loop. Also the various timeouts
are open-coded based on the absolute uptime tick so to sample it only
once per round. Using sys_timepoint_calc() and sys_timepoint_timeout()
would have introduced additional uptime tick sampling which implies the
overhead of a downstream lock each time for no gain. For those reasons,
open coding those timeouts bears more benefits in this particular case
compared to using the timepoint API.
Then this:
secs = k_ticks_to_ms_ceil32(loop_time) / 1000U;
usecs = k_ticks_to_us_ceil32(loop_time) - secs * USEC_PER_SEC;
The above should round down not up to work accurately. And the usecs
value will become garbage past 1.2 hour of runtime due to overflows.
And no need to clamp the wait period which is on the microsec scale
using the total duration argument being on the millisec scale. That's
yet more loop overhead that can be omitted. The actual duration is
recorded at the end anyway.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Compiling an application with CONFIG_NET_ZPERF=y leaving
CONFIG_ZPERF_WORK_Q_THREAD_PRIORITY at its default value would
systematically cause a kernel panic during thread initialization.
The Kconfig variable is NUM_PREEMPT_PRIORITIES by default. Application
threads may not define a priority lower than NUM_PREEMPT_PRIORITIES - 1,
though.
This change limits zperf's thread priority to a valid range. It does not
change the default value as it makes sense to default the thread
priority to the lowest possible value (which is NUM_PREEMPT_PRIORITIES)
but Kconfig does not allow for arithmentic. So the combination of
CLAMP() plus the Kconfig default will ensure min priority plus limit the
range to valid values no matter what has been defined as priority in
Kconfig.
Fixes: #59141
Signed-off-by: Florian Grandel <fgrandel@code-for-humans.de>
The init infrastructure, found in `init.h`, is currently used by:
- `SYS_INIT`: to call functions before `main`
- `DEVICE_*`: to initialize devices
They are all sorted according to an initialization level + a priority.
`SYS_INIT` calls are really orthogonal to devices, however, the required
function signature requires a `const struct device *dev` as a first
argument. The only reason for that is because the same init machinery is
used by devices, so we have something like:
```c
struct init_entry {
int (*init)(const struct device *dev);
/* only set by DEVICE_*, otherwise NULL */
const struct device *dev;
}
```
As a result, we end up with such weird/ugly pattern:
```c
static int my_init(const struct device *dev)
{
/* always NULL! add ARG_UNUSED to avoid compiler warning */
ARG_UNUSED(dev);
...
}
```
This is really a result of poor internals isolation. This patch proposes
a to make init entries more flexible so that they can accept sytem
initialization calls like this:
```c
static int my_init(void)
{
...
}
```
This is achieved using a union:
```c
union init_function {
/* for SYS_INIT, used when init_entry.dev == NULL */
int (*sys)(void);
/* for DEVICE*, used when init_entry.dev != NULL */
int (*dev)(const struct device *dev);
};
struct init_entry {
/* stores init function (either for SYS_INIT or DEVICE*)
union init_function init_fn;
/* stores device pointer for DEVICE*, NULL for SYS_INIT. Allows
* to know which union entry to call.
*/
const struct device *dev;
}
```
This solution **does not increase ROM usage**, and allows to offer clean
public APIs for both SYS_INIT and DEVICE*. Note that however, init
machinery keeps a coupling with devices.
**NOTE**: This is a breaking change! All `SYS_INIT` functions will need
to be converted to the new signature. See the script offered in the
following commit.
Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no>
init: convert SYS_INIT functions to the new signature
Conversion scripted using scripts/utils/migrate_sys_init.py.
Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no>
manifest: update projects for SYS_INIT changes
Update modules with updated SYS_INIT calls:
- hal_ti
- lvgl
- sof
- TraceRecorderSource
Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no>
tests: devicetree: devices: adjust test
Adjust test according to the recently introduced SYS_INIT
infrastructure.
Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no>
tests: kernel: threads: adjust SYS_INIT call
Adjust to the new signature: int (*init_fn)(void);
Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no>
C only permits labels on statements, not declarations. Separate the
declarations from the assignments so that the labels can target
statements instead.
Signed-off-by: Keith Packard <keithp@keithp.com>
Currently the zperf_tcp_receiver can handle only one TCP connection
each time, modify the code to poll and handle multiple connections.
Take the occasion to unify the bind and listen part of the code
between ipv4 and ipv6 part using a structure introduced to handle
the multiple connections.
Now in case the zsock_recv fails, we can't stop every connection
and fail through the error label, so just print the error message
and report the failure through the callback.
Signed-off-by: Luca Fancellu <luca.fancellu@arm.com>
Reduce the scope of in4_addr_my and in6_addr_my pointer variables
that are currently global, but they are used only inside
tcp_receiver_thread.
Take the occasion to fix a typo in one error message.
Signed-off-by: Luca Fancellu <luca.fancellu@arm.com>
Currently, in tcp_received function defined in zperf_tcp_receiver
module, the assignment of session->state to STATE_COMPLETED is
overwritten on the same path to STATE_NULL and a session is
considered free for both STATE_COMPLETED and STATE_NULL, so remove
the assignment to STATE_NULL.
Remove the break from the STATE_COMPLETED case handling so that it
can fallthrough, in case the same session is used after finish.
Remove also the STATE_LAST_PACKET_RECEIVED case because this state
is never reached.
Signed-off-by: Luca Fancellu <luca.fancellu@arm.com>
The modules zperf_udp_receiver and zperf_tcp_receiver use two
different functions to get a zperf session to store the
statistics, there is a TODO comment in the zperf_session module
suggesting to unify that part.
So delete the get_tcp_session function and use get_session for
both TCP and UDP receiver module.
Delete sock field from struct session because it's not used
anymore.
Signed-off-by: Luca Fancellu <luca.fancellu@arm.com>
Currently the maximum number of zperf sessions handled is hardcoded
to 4, create a Kconfig parameter, with default value 4, to make the
maximum number of sessions configurable.
Signed-off-by: Luca Fancellu <luca.fancellu@arm.com>
A variable was defined directly after a label in two case statements,
resulting in build warning with certain compilers.
Signed-off-by: Robert Lubos <robert.lubos@nordicsemi.no>
Now that we a proper API, shell is just optional, so, make the
dependency optional by refactoring the code.
Also, add a build test combination in twister.
Signed-off-by: Krishna T <krishna.t@nordicsemi.no>
Instead of calling zperf shell initialization routine on the first
command execution, initialize it during system boot, along with other
zperf submodules.
Remove redundant IP address configuration on an interface. The default
configuration relies on NET_CONFIG module, so there's no need to set the
address manually in zperf.
Signed-off-by: Robert Lubos <robert.lubos@nordicsemi.no>
Zperf shell functionality is now encapsuled within a single file,
therefore it no longer makes sense to have a separate shell_utils
file.
Signed-off-by: Robert Lubos <robert.lubos@nordicsemi.no>