d00f
fa0eb91586
The counters, the phase hooks in idle() and the "health" command are now behind HEALTH, set from the command line with HEALTH=1. Without it the hooks are empty macros and the command does not exist, so a default build carries none of it: on SWTGW218AS that is 1065 bytes of bank 2, 101 of the common segment and 28 of xdata, plus seven banked calls per main-loop pass. It is a developer's tool. Reading the dump takes knowing the main loop, and a user who needs it for a report can build with the flag on, which is how the kernel treats its lock and latency instrumentation as well.
3.2 KiB
The health command
health prints a one-page snapshot of the firmware's vital signs on the
serial console. It exists for the situations where the switch misbehaves and
every probe from the network measures the network instead of the device: the
dump is taken from inside, costs one command, and works while HTTP does not.
> health
up 0000001f ticks 0x00001f4c loops 0x000016da rx 0x024e
link max 01 slow 0x0000
sfp max 74 slow 0x0001
rx max 00 slow 0x0000
tx max 00 slow 0x0000
stp max 04 slow 0x0001
cmd max 02 slow 0x0001
sp 80 untouched 62
tcp st 00 tmr 00 rtx 00 rto 00 len 0x0000 mss 0x05b4
lport 0x0000 rport 0x0000 rip 0x0000 0x0000
httpd left 0x0000 entry 00 stp 01 mvlan 0x0002
The instrumentation is not part of a default build. It costs about 1.1 kB of bank 2, a few bytes of xdata and seven banked calls per main-loop pass, which is nothing a user needs unless they are chasing a problem, so it is switched on at build time:
make MACHINE=SWTGW218AS HEALTH=1
Without it the hooks compile to nothing and health is not a command.
All figures are raw hex. The counters are cumulative: take two dumps a known time apart and the differences give the rates.
The first line
up is the seconds counter, ticks the system tick (200 per second),
loops the number of main-loop passes and rx the frames handle_rx has
taken off the NIC. A healthy loop completes one pass per tick, so the
difference in loops tracks the difference in ticks; a loop that falls
behind is being held up by one of the phases below. The rx rate separates
a flooded CPU port from a stalled loop, which look identical from outside.
Phases
Each line is one section of the main loop: the link poll, SFP handling,
frame reception, TCP transmission, the STP timers and the command dispatch.
max is the worst time that section has taken, in ticks of 5 ms; slow
counts the passes that took two ticks or more. A misbehaving subsystem
names itself here, which is the difference between knowing that the loop is
slow and knowing why.
A dump of its own output takes the console some milliseconds, so running
health repeatedly increments the cmd counters by itself. That is the
instrument observing itself, not a fault.
Stack
At boot the free stack area is painted with a pattern. sp is the live
stack pointer; untouched counts the painted bytes still intact at the top,
which is the closest thing to a high-water mark this hardware offers. A
shrinking untouched across dumps means something is reaching deeper than
anything before it.
The TCP slot
The connection table is printed whole, one slot per pair of lines: TCP state (3 is ESTABLISHED, 0 closed), retransmission timer and counter, bytes in flight, negotiated MSS, then the ports and the peer address. With a single slot serving the entire web interface, one glance answers the question every management outage starts with: who is holding it. A slot showing state 3 with nothing in flight and a peer that no longer answers is a connection whose owner went away.
httpd left and entry describe the file transfer the web server believes
it is in the middle of, and the trailing flags show whether STP is running
and which VLAN carries management.