mirror of
https://github.com/zephyrproject-rtos/zephyr
synced 2025-09-10 01:52:00 +00:00
e7af4b8947
When zephyr receives a 15.4 frames it allocates 1 RX buf (nbuf) and 1 DATA (buf), and it gets queued on net stack for post processing. Problem: receiving a lot of frames is just completely nominal and very fast (timing is critical in radio, and the rx thread of the radio driver has a high priority, much higher than net stack's rx post processing). So guess what happens when CONFIG_NET_NBUF_RX_COUNT=10 though let's say 12 packets are coming in? The RX buffer pools gets empty and soon the system hangs on getting a newly allocated one. This proper CONFIG_NET_NBUF_RX_COUNT tweak fixes the issue, however some future work will reduce the need of so many RX meta buf: - l2's recv could be called right away at net_recv_data, before queuing the buffer for post-processing, but this could affects reception timing on radio side, which is critical. - An l2 pre-processing (in the middle of what current l2 recv does) could be implemented, it would be technology depenedent (aka: not mandatory on all l2) and could quickly decide what to do and at best drop the whole, or at least drop the RX meta buf (frag would still be kept). Let's see. Jira: ZEP-1427 Change-Id: Ic75b02efa36e35b450c02b3e6439a8d73f03f839 Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com> |
||
|---|---|---|
| .known-issues | ||
| arch | ||
| boards | ||
| doc | ||
| drivers | ||
| ext | ||
| include | ||
| kernel | ||
| lib | ||
| misc | ||
| samples | ||
| scripts | ||
| subsys | ||
| tests | ||
| .checkpatch.conf | ||
| .gitattributes | ||
| .gitignore | ||
| .gitreview | ||
| .mailmap | ||
| defaults.tc | ||
| Kbuild | ||
| Kconfig | ||
| Kconfig.zephyr | ||
| LICENSE | ||
| LICENSING.rst | ||
| MAINTAINERS | ||
| Makefile | ||
| Makefile.inc | ||
| release-notes.rst | ||
| zephyr-env.sh | ||