casaroli opened a new pull request, #4100:
URL: https://github.com/apache/nuttx-apps/pull/4100

   ## Summary
   
   depends-on: apache/nuttx/pull/22131
   
   A new case of `cmocka_net_tcp`, `test_tcp_send_limit_wakeup`: for 10 
seconds, the test thread sends 16 KiB messages over 127.0.0.1, and a second 
thread reads each message and answers one byte. With a send limit smaller than 
a message (`CONFIG_NET_SEND_BUFSIZE`, or `SO_SNDBUF`), `send()` waits on the 
limit in each round, and the ACK of the last data must wake it. Each round must 
end within 2 seconds. The sockets have 5 s time-outs, so a lost wakeup fails 
the case; it does not hang.
   
   The ACK is often handled on another CPU while the sender is about to wait, 
so the case finds the lost wakeup that apache/nuttx#22131 fixes best with 
`CONFIG_SMP`. Without a send limit (the default before the default limit of 
apache/nuttx#22089) the case passes: `send()` does not wait on a limit.
   
   ## Impact
   
   `testing/nettest` only: one more case in `cmocka_net_tcp`, built with 
`CONFIG_NET_IPv4` and `CONFIG_NET_SOCKOPTS`. Its list entries in the Makefile, 
`CMakeLists.txt`, `test_tcp.c` and `test_tcp.h` are on other lines than those 
of the open nettest PRs (#4086, #4096, #4099), so the PRs do not conflict.
   
   ## Testing
   
   qemu-armv8a:knsh under HVF, 96 IOBs of 196 bytes, `IOB_THROTTLE` 8, 
`CONFIG_NET_TCP_WRITE_BUFFERS`, `CONFIG_NET_SEND_BUFSIZE=2156` and 
`CONFIG_NET_RECV_BUFSIZE=2156`; nuttx master 5e4ecd1d603 with 
apache/nuttx#22025 (without it, the TIME_WAIT connections of the earlier cases 
can use all devif callbacks), with and without apache/nuttx#22131. One 
`cmocka_net_tcp` per boot; 3 of the 5 boots ran an earlier version of this 
commit that differs only in the place of the list entries and in the failure 
messages.
   
   | `test_tcp_send_limit_wakeup` | master | with apache/nuttx#22131 |
   |---|---|---|
   | knsh, 4 CPUs, 5 boots | FAIL 5 of 5: a round waits 10 s or 15 s | PASS 5 
of 5 |
   | knsh, 1 CPU | | PASS |
   | flat qemu-armv8a:nsh, 4 CPUs, 1 boot | PASS (the race was not hit) | PASS |
   | knsh, 4 CPUs, `CONFIG_NET_SEND_BUFSIZE` 0 | | PASS |
   | knsh, 1 and 4 CPUs, flat nsh, with the default limit of apache/nuttx#22089 
| | PASS |
   
   `test_tcp_connect_ipv4` fails in this configuration on all versions 
(`ECONNREFUSED` from the group server).
   
   How CI would run it: rv-virt:citest builds `cmocka_net_tcp` 
(`CONFIG_TESTING_NET_TEST`), but no NTFC case runs `cmocka_net_tcp` yet, and 
the CI configurations with networking have no send limit and no SMP. A run in 
CI needs an NTFC case for `cmocka_net_tcp` on an SMP configuration with 
`CONFIG_NET_TCP_WRITE_BUFFERS` and a send limit (`CONFIG_NET_SEND_BUFSIZE > 0`, 
or the default limit of apache/nuttx#22089).
   
   Builds in the NuttX CI image (`EXTRAFLAGS="-Wno-cpp -Werror"`) with 
`cmocka_net_tcp`: qemu-armv8a:netnsh with TCP write buffers (Make and CMake), 
without TCP write buffers, with the default limit of apache/nuttx#22089, and 
sim:tcpblaster: all OK, 0 warnings, the case is linked in each.
   
   `tools/checkpatch.sh -c -u -m -g` passes. Documentation: the nettest cases 
are not listed in Documentation/; no change.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to