From review (shenmintao/aic8800d80#115): the default log keeps enough to read a first bug report.
- The patch table: each entry at LOGDEBUG, as before this series' first change; one line at LOGINFO (the default)
with the file and the number of entries written; a failure at LOGERROR, as before.
- 4-address frames dropped by a build without CONFIG_SUPPORT_4ADDR: counted per device and said at LOGINFO, the
first at once and then at most once a minute; each frame and its dump at LOGDEBUG, by aicwf_dbg_level, in place
of dev_dbg() and print_hex_dump_debug(), which need dynamic debug or DEBUG and were compiled out otherwise.
- The README says how to turn the detail on and off, at run time and before the driver loads (the patch table is
written when the adapter starts).
rwnx_cfg80211_scan() stores the request in rwnx_hw->scan_request before sending SCANU_START_REQ. When the send
fails, the error goes back to cfg80211, which frees the request, but scan_request still points at it. A late
SCANU_START_CFM (or the next scan's completion) then calls cfg80211_scan_done() on freed memory.
Seen on an AIC8800DC with a command timeout injected during a scan: the kernel panicked in cfg80211_scan_done(),
caught twice on the serial console.
Without WIPHY_FLAG_NETNS_OK, 'iw phy <phy> set netns' is refused, so the
interface can't be used from a namespace (as test rigs do to pair two radios
in one box). The driver keeps nothing per namespace: its only uses of
init_net are its own netlink control socket and an optional bridge lookup.
nomdetom