test/stp/shim held hand-written copies of the headers the STP code
includes, and they had drifted: the uIP buffer was 600 bytes instead of
1556, struct machine had three fields, port_ingress_filter_get()
returned a plain uint8_t. The simulator now compiles rtl837x_stp.c and
rtl837x_stp_cli.c the way test/ compiles firmware sources, with the real
headers and sdcc_shim.h forced in, and with the same warnings, including
-Wno-pointer-sign for the string literals passed to cmd_compare(). It
stays on -std=c11: in gnu11 mode glibc defines LITTLE_ENDIAN, which
uip-conf.h names as the byte order, and HTONS() stops swapping. Its own
print_string() and cmd_compare() take the firmware's argument types.
* stp: select the root port by cost and block redundant links
* stp: move the protocol to BANK4 and keep the command line in BANK2
* stp: implement the RSTP state machines of 802.1D-2004 clause 17
* stp: counters, more settings and a fuller Spanning Tree page
* stp: make a link aggregation group one port of the RSTP tree
* web: serve the Spanning Tree JSON from BANK4 as well
Kept in BANK1, the larger Spanning Tree page would leave under 1 kB
there for the rest of the web server and every page still to come, so
send_stp() and send_stp_counters() move to httpd/page_stp.c in BANK4
with the protocol, as banked calls, with their own copies of the few output
helpers and of the JSON header, so nothing in BANK4 calls into BANK1
directly or hands it a pointer into its code space. BANK4 takes the
free flash between BANK3 and the web files.
KP_9000_6XHML_X2: BANK1 44336, BANK2 36969, BANK3 11942, BANK4 20965
bytes of 49152.
A commit hash can be looked up directly, a checksum cannot. Print the last commit that touched sdcc/, with -dirty for local changes, and fall back to the checksum when git is not available. The toolchain directory keeps the checksum.
The installer Makefile called whatever sdcc was on the PATH, so an
installer built by hand came out of the system compiler, without the
fixes in sdcc/. Use the same toolchain/sdcc-<cksum> as the main Makefile,
built on demand by sdcc/build.sh if a bare installer build gets there
first, and keep CC= on the command line as the way to pick another
compiler. The README example shows the new command lines.
The boot banner and the version command print the SDCC release and
revision from the compiler's own macros, and the name of the in-tree
toolchain the Makefile built, which is the checksum of sdcc/ and so
changes with every patch in it. A build with CC given on the command
line says "system" instead, since that compiler need not carry the
patches. A plain release and the same release with the patches report
the same revision, so the toolchain line is what tells them apart.
Costs 66 bytes of BANK2.
html_min was phony, so every make deleted output/html_min and minified
all of html/ again, even with nothing changed. A stamp file now stands for
the minified copy and depends on the sources and on minify.py. A list of
the source names, rewritten only when it changes, is a dependency too, so
adding or removing a file in html/ still redoes the copy and a removed
file does not stay behind in the image. If output/html_min is missing,
the stamp is remade in the same run.
The note on version.h in webui-compression.md still said it was phony.
Three more bench scenarios around the closing boundary. Every value
of the last image byte, uploaded whole and split right after the
payload, must end the part. A run of CR before the boundary of even
length passed before the fix and an odd one did not, so both are
checked. An image made of CR only, longer than a flash page, and one
made of repeated "\r\n-" exercise the broken matches back to back:
each one appends to the page buffer without a byte written, and the
page flush after both paths has to keep up.
stream_upload() matches the closing multipart boundary byte by byte.
When a byte breaks a partial match, the matched prefix goes to flash as
data, and so does the breaking byte, without being compared against the
start of the boundary.
The boundary starts with "\r\n--", so an image whose last byte is '\r'
never completes: that '\r' starts a match, the '\r' of the boundary
breaks it and is written as data, and the rest of the boundary no
longer matches. The browser waits for a reply that never comes.
Flush the matched prefix and look at the breaking byte again from the
start of the boundary. A boundary may not contain CR, so no longer
suffix of the matched prefix can start a new match.
The 8051 assembler sign-extends every value to 24 bits, so the top
byte of a __sfr32 address list like 0xa4a5a6a7 comes out as 0xff.
The patch keeps 32-bit arithmetic while leaving the 24-bit T line
addresses alone, so the relocations are unchanged. Proposed upstream
in the bug thread. With it SFR_DATA_U32 can be used again, which
saves about 240 bytes of code in REG_SET.
The workflow and the Dockerfile install what building SDCC needs instead
of the distribution's sdcc, and the workflow caches toolchain/ keyed on
the contents of sdcc/.
Released SDCC 4.5.0 and 4.6.0 miscompile some of our code in the mcs51
register allocator (SDCC bugs 4153 and 4154, fixed in trunk r17025 and
r17039). The first make now downloads the 4.6.0 sources, checks their
sha256, applies the two fixes from sdcc/ and installs the compiler into
toolchain/, named after a checksum of sdcc/, so a changed patch builds a
new one. make CC=sdcc still uses an installed compiler.
The strip no longer wraps. Ports shrink to fit the row and scroll
when they cannot, so the header on a phone stays at two rows. The
strip moves under the title only below 720 px, where it no longer
fits beside it. Drop the unused d_ports strings and the strip title.
The row of port icons lived in a card on the Dashboard, so link
state and speed were out of sight on every other tab. Move it into the
header between the title and the save button, a little smaller, and
poll /status.json whenever the page is open instead of only on the
Dashboard, Ports and Statistics tabs; the existing idle and hidden-tab
gating still pauses it. On narrow screens the strip wraps to its own
row under the title.
The column counts the addresses the switch learns from incoming frames,
so "Connected" suggested a connection it does not show. The translations
follow the words the L2 tab already uses for learned entries.
hexvalue[] holds four bytes, but atoi_hex() stored every digit it was
given, so a ninth digit was written past the array into whatever follows
it in xdata. More than eight digits now return 0, as any bad input does.
c - '0' < 10 is computed as an int, so characters below '0' such as '-',
'.' or '/' passed as digits and "-1" came out as 0xdf. The difference is
now taken as uint8_t, so they are rejected.
The switch keeps flooding 224.0.0.0/24 with IGMP on, but a report for a
group in that block still made an entry for it, and the entry limited
the group to the ports that reported it. A host announcing mDNS on
224.0.0.251 thereby cut mDNS off from every other port. Such reports
are now ignored, as RFC 4541 asks.
igmp_setup(), which `igmp off` also calls, enabled the lookup of IPv4
multicast in the L2 table. Groups learned while IGMP was on therefore
kept going only to the ports that had reported them, and with reports
no longer trapped nothing updated those entries until a reboot.
The lookup is now enabled in igmp_enable() and disabled in igmp_setup(),
so with IGMP off all IP multicast is flooded again.
Move port_bit() to rtl837x_common.h as PORT_BIT(), a uint16_t shift,
and MAC_ADDR_LEN from rtl837x_lldp.h next to it, and use MAC_ADDR_LEN
for every address in the LACP code instead of a literal 6. Define
lacp_mux_update() before its first caller and drop the forward
declaration, declare variables at the top of their blocks, put the
operator at the end of a continued line, drop the field value comments
of the LACPDU struct, and leave lacp_port_lag to lacp_init() instead of
initialising it in the test environment.
The trap is described with STP, so the LACP page only says that it is
used. The LACP sections leave link_aggregation.md, which now points to
lacp.md, and the timing section gives the values in seconds and the
802.3ad requirements without the test history.
The comments repeated the documentation, described tests or explained
the code to its author. What stays describes the LACPDU structures. The
timer constants are decimal, in units of the LACP timer.
The LACP test now compiles with the shared harness flags and
sdcc_shim.h against the firmware headers, so the copies under
test/lacp/shim are gone. test_port_tables already covers the trunk hash
seed that test/port_hash checked against its own register copies.
lacp_in() declares its locals at the top, which the stricter host build
needs.
The port and group of lacp_in(), the member mask passed to
lacp_lag_set() and the compiler temporaries both functions spilled took
six bytes of DSEG, and with them six bytes of stack. The locals and the
mask live in xdata now, the partner fields go through the scratch word,
and lacp_lag_set() records the new group of a port before it releases
the old one. One byte is left: the group number passed in.
lacp_in() always leaves uip_len at 0, so the tcpip_output() after it
never ran. The LAG page names the active members by front-panel port
instead of a logical mask. The partner key is not part of selection;
say so under Known limits.
Listing by next-address also returned the reserved-multicast entries STP
writes, shown as port 1 with a delete button, while l2_delete() still
looks up by next-unicast and so removed the unicast entry after that
index instead. The change was there to show the LACP steering entries;
delivery by trap writes none, so it goes.
Both paths wrote the trunk registers directly and left the protocol
running on the group, so "lag show" said disabled while the partner kept
both links in one aggregate, and re-sending the LACP line did not repair
it because the same-ports shortcut returns early. Each path now releases
the group from LACP first; for a group that never ran LACP that is a
no-op.
The receive machine armed current_while from the partner's TIMEOUT bit,
so against a partner on the slow rate a port whose LACPDUs stopped with
the link still up stayed in the trunk for the long timeout, two minutes.
802.3ad 43.4.12 starts the timer from the actor's own LACP_Timeout, and
we always advertise the short one: the partner sends every second and is
let go after the short timeout.
Sandbox T23 plays a slow-rate partner that goes quiet and fails on the
code before this.
Dynamic link aggregation on top of the static LAGG, for RTL8372/8373:
receive, selection and mux state machines (coupled mux, one aggregator
per group), with "lacp show" to look inside. Each of the four trunk
groups runs its own aggregator on the candidate ports assigned with
"lag <n> lacp <ports>"; the engine starts with the first such group and
stops with the last.
LACPDUs reach the CPU through the reserved-multicast trap, so delivery
needs no L2 entry and nothing written per VLAN or per PVID. The trap
acts on the address before the VLAN lookup and before the ingress
filters, so a LACPDU also arrives on a port that admits tagged frames
only and on a port STP holds in blocking, neither of which a forwarded
one reaches. A trapped frame carries no VLAN tag,
so lacp_in() takes the offset of the PDU body from TRAP_RX_BODY(). The
action is set with trap_rma_set(), which leaves the flags sharing that
register alone, and "lacp off" puts the slow-protocols drop default
back. doc/lacp.md and doc/link_aggregation.md record what was measured,
including why the trap looked like a dead end at first.
The web UI gets a Static/LACP mode per group on the LAG page with live
status, and the startup-config validator knows the per-group syntax. A
host-side sandbox for the state machines runs in the test harness
together with the LAG hash check. It drives both forms of an incoming
LACPDU, so the body offset of a trapped one is covered as well.
The L2 table page iterated with the "next unicast" read method, so
static multicast entries such as an IGMP group were invisible. It now
iterates with "next address", skips L3 entries (their layout carries IP
addresses where a MAC is expected), and reports the lowest member of a
multicast entry's port mask, so such an entry shows up as a static
entry on that port.
A port taken out of LACP forgets its partner, so "lacp show" after
"lacp off" no longer reports one. Applying the same candidate set to a
group again, as the LAG page does on every Apply, leaves the trunk
alone instead of tearing it down for a second or two. lacp_in() checks
the frame length and ethertype before reading a LACPDU, so a runt or a
foreign frame on the slow-protocols address cannot leave stale buffer
bytes as a partner. The sandbox covers these three in T20-T22. On the
LAG page, LACP mode with no ports checked sends "lag N lacp off", and
the startup-config validator wants at least one port after "lacp".
The document was renamed from doc/trunking.md in "Update LAG configuration
registers" and the entry in the README kept the old name, so it leads
nowhere.
Move port_bit() to rtl837x_common.h as PORT_BIT(), a uint16_t shift,
and MAC_ADDR_LEN from rtl837x_lldp.h next to it, and use MAC_ADDR_LEN
for every address in the LACP code instead of a literal 6. Define
lacp_mux_update() before its first caller and drop the forward
declaration, declare variables at the top of their blocks, put the
operator at the end of a continued line, drop the field value comments
of the LACPDU struct, and leave lacp_port_lag to lacp_init() instead of
initialising it in the test environment.
The simulator answered /status.json without the "macs" field the Ports page now reads. It counts the dynamic entries of its own L2 table for the port, so the column and /l2.json agree and the two static entries show the difference.
The column listed the MAC addresses of the L2 table entries of a port, which cost a chain of paged /l2.json requests every tenth status poll. The status JSON carries the count now, so the column reads it from there and the page does not touch the L2 table at all.
The register counts the dynamic entries, so the static ones the list used to include are not in the number.
The trap is described with STP, so the LACP page only says that it is
used. The LACP sections leave link_aggregation.md, which now points to
lacp.md, and the timing section gives the values in seconds and the
802.3ad requirements without the test history.
The comments repeated the documentation, described tests or explained
the code to its author. What stays describes the LACPDU structures. The
timer constants are decimal, in units of the LACP timer.
The LACP test now compiles with the shared harness flags and
sdcc_shim.h against the firmware headers, so the copies under
test/lacp/shim are gone. test_port_tables already covers the trunk hash
seed that test/port_hash checked against its own register copies.
lacp_in() declares its locals at the top, which the stricter host build
needs.
The port and group of lacp_in(), the member mask passed to
lacp_lag_set() and the compiler temporaries both functions spilled took
six bytes of DSEG, and with them six bytes of stack. The locals and the
mask live in xdata now, the partner fields go through the scratch word,
and lacp_lag_set() records the new group of a port before it releases
the old one. One byte is left: the group number passed in.
lacp_in() always leaves uip_len at 0, so the tcpip_output() after it
never ran. The LAG page names the active members by front-panel port
instead of a logical mask. The partner key is not part of selection;
say so under Known limits.
Listing by next-address also returned the reserved-multicast entries STP
writes, shown as port 1 with a delete button, while l2_delete() still
looks up by next-unicast and so removed the unicast entry after that
index instead. The change was there to show the LACP steering entries;
delivery by trap writes none, so it goes.
Both paths wrote the trunk registers directly and left the protocol
running on the group, so "lag show" said disabled while the partner kept
both links in one aggregate, and re-sending the LACP line did not repair
it because the same-ports shortcut returns early. Each path now releases
the group from LACP first; for a group that never ran LACP that is a
no-op.
The receive machine armed current_while from the partner's TIMEOUT bit,
so against a partner on the slow rate a port whose LACPDUs stopped with
the link still up stayed in the trunk for the long timeout, two minutes.
802.3ad 43.4.12 starts the timer from the actor's own LACP_Timeout, and
we always advertise the short one: the partner sends every second and is
let go after the short timeout.
Sandbox T23 plays a slow-rate partner that goes quiet and fails on the
code before this.
Dynamic link aggregation on top of the static LAGG, for RTL8372/8373:
receive, selection and mux state machines (coupled mux, one aggregator
per group), with "lacp show" to look inside. Each of the four trunk
groups runs its own aggregator on the candidate ports assigned with
"lag <n> lacp <ports>"; the engine starts with the first such group and
stops with the last.
LACPDUs reach the CPU through the reserved-multicast trap, so delivery
needs no L2 entry and nothing written per VLAN or per PVID. The trap
acts on the address before the VLAN lookup and before the ingress
filters, so a LACPDU also arrives on a port that admits tagged frames
only and on a port STP holds in blocking, neither of which a forwarded
one reaches. A trapped frame carries no VLAN tag,
so lacp_in() takes the offset of the PDU body from TRAP_RX_BODY(). The
action is set with trap_rma_set(), which leaves the flags sharing that
register alone, and "lacp off" puts the slow-protocols drop default
back. doc/lacp.md and doc/link_aggregation.md record what was measured,
including why the trap looked like a dead end at first.
The web UI gets a Static/LACP mode per group on the LAG page with live
status, and the startup-config validator knows the per-group syntax. A
host-side sandbox for the state machines runs in the test harness
together with the LAG hash check. It drives both forms of an incoming
LACPDU, so the body offset of a trapped one is covered as well.
The L2 table page iterated with the "next unicast" read method, so
static multicast entries such as an IGMP group were invisible. It now
iterates with "next address", skips L3 entries (their layout carries IP
addresses where a MAC is expected), and reports the lowest member of a
multicast entry's port mask, so such an entry shows up as a static
entry on that port.
A port taken out of LACP forgets its partner, so "lacp show" after
"lacp off" no longer reports one. Applying the same candidate set to a
group again, as the LAG page does on every Apply, leaves the trunk
alone instead of tearing it down for a second or two. lacp_in() checks
the frame length and ethertype before reading a LACPDU, so a runt or a
foreign frame on the slow-protocols address cannot leave stale buffer
bytes as a partner. The sandbox covers these three in T20-T22. On the
LAG page, LACP mode with no ports checked sends "lag N lacp off", and
the startup-config validator wants at least one port after "lacp".
A reserved-multicast trap goes to the port named in EXT_CPU_CTRL, which
names no port out of reset, so trapped frames were dropped at ingress.
trap_init() points it at the CPU port once at startup, and
trap_rma_set() changes only the action field of an RMA register.
A trapped frame arrives without the VLAN tag a forwarded one gets, so
TRAP_RX_BODY() gives the offset of the payload for both forms.
STP now traps its address while it runs instead of steering it with a
static L2 entry per VLAN, so turning it on no longer walks the VLAN
table and a pvid change no longer needs stp off and on. With STP off
the flooding entries are restored as before.
A trapped BPDU passes the tagged-only ingress filter that drops a
forwarded one, so the stp_setup() warning about such ports goes.
Opcode 0x48 outputs 8 dummy clocks between the address and the first data
byte. flash_configure_sio() sets DUMMYCYCLES to 0 and flash_read_security()
never raises it, so the controller captures the dummy slot as the first of
its four data bytes while the address still advances by four. The printed
output therefore shows 0xff in place of every fourth byte and skips one
content byte per transfer. flash_read_uid() already sets DUMMYCYCLES for
opcode 0x4b, which needs the same allowance.
The answer to a lookup arrives after the command has returned, so the
console of the web interface never shows it. Plain "dns" now prints the
name and the address, or whether it is still pending or failed.
igmp_setup(), which `igmp off` also calls, enabled the lookup of IPv4
multicast in the L2 table. Groups learned while IGMP was on therefore
kept going only to the ports that had reported them, and with reports
no longer trapped nothing updated those entries until a reboot.
The lookup is now enabled in igmp_enable() and disabled in igmp_setup(),
so with IGMP off all IP multicast is flooded again.
The switch keeps a count of the dynamic L2 entries learned on every
port in L2_LRN_PORT_CONSTRT_CNT (0x53ac + 4 * port, 13 bits). `stat`
prints it in a new MACs column and /status.json carries it as "macs".
On a SWTGW218AS the count matched the L2 table port by port, static
entries are not included, and taking a port down drops it to 0.
The switch keeps flooding 224.0.0.0/24 with IGMP on, but a report for a
group in that block still made an entry for it, and the entry limited
the group to the ports that reported it. A host announcing mDNS on
224.0.0.251 thereby cut mDNS off from every other port. Such reports
are now ignored, as RFC 4541 asks.
leds_dump() printed the pin mux, the LED enables and every LED set on each
boot, from BANK2. It now lives in its own file in BANK3, where there is
room, and runs only on the new "leds" command, next to gpio and sds, so it
stays at hand for bringing up a new board without costing BANK2 or boot
output.
A matrix of the ports, one row per ingress port, showing which ports a
frame entering there may leave through. The port number of a row turns
red when the port is not fully open, and the cell for the port itself is
marked since it is always allowed. Apply sends one "isolate" command per
changed row, and Allow all ticks the whole matrix.
The switch side is a new /isolation.json, which reports each port's
allowance mask read back from the hardware. The page is an Isolation
section of the single-page UI.
494320b removed strcpy() from rtl837x_flash.c but left its declaration
in rtl837x_common.h. Since then the linker has taken strcpy from the
SDCC library instead, which expects the destination as a generic pointer
in DPL, DPH and B. The callers pass an __xdata pointer and leave B as it
is, so the copy lands wherever B happens to point. boot.c only gets away
with it because B is zero there; a call with B set, as from the ChaCha20
test in #207, writes into internal RAM and stops the 8051.
Only frames to 01:00:5e:00:00:16 reached the IGMP handler, so v1/v2
reports, which go to the group address, were never seen. The handler
also took the group from the first v3 group record for every type.
Pass all IPv4 multicast to the handler, which takes the IGMP header
from the IP header length, the group from the IGMP header for v1/v2
reports and v2 leaves and from the first group record for v3, and
returns 0 for anything that is not IGMP so it goes on to the IP stack.
The message and record types have names now.
Storm control used meters 0 to 39, the same meters the ACL hands out,
so a storm limit and an acl meter could overwrite each other. It now
uses 24 + port * 4 + type, meters 24 to 63.
parse_storm() is no longer reentrant and uses plain locals, looks the
type up in a while loop and returns straight from the one-line cases.
storm_show() uses plain locals too.
The documentation says how IGMP decides between mcast and umcast, how
to get from a meter index to its limit, where the IFG bit of a meter
is, and what an ingress bandwidth limit on the same port does to the
storm limit.
The switch has no battery backed clock, only a seconds counter that starts
at zero on every boot. An SNTP client keeps UTC as the time of the last
answer plus the seconds counted since: "ntp server <name|ip>", "ntp interval
<minutes>", "ntp timezone <+hh:mm>", "ntp dst off|eu|us", "ntp on|off",
"ntp" and "time". A server given by name goes through the DNS resolver, an
answer from an unsynchronised server is ignored, and a failed attempt is
repeated after 30 seconds.
The local time adds the time zone offset and, while the European or United
States rule says so, one hour. The date is split with a shift and subtract
division and a walk over the years, so no 32-bit division from the library
ends up in the common area.
The System page gets the NTP settings and shows the local time.
A small resolver for IPv4 address records, so that clients such as NTP can
be given a host name: "dns server <ip> [<ip>]", "dns lookup <name>" and
"dns". Without a configured server it uses the one DHCP handed out. A query
goes to the servers in turn every two seconds and gives up after six tries;
answers with compressed names and CNAME records ahead of the address are
understood. A name that already is a dotted address is taken without a
query.
The resolver sits in BANK3, the command line in BANK2 next to the helpers it
uses. The System page shows and sets the servers.
The image rule no longer passes -p, so fileadder fell through to printing
html_data.h and html_data.c on stdout on every image build. The listing is
only useful when fileadder is run without an image, to see what it would
generate.
The .bin rule passes -p html_data to fileadder, which rewrites
html_data.c and html_data.h after the objects that include them have
already been compiled and linked. The next make therefore rebuilds
html_data.c, httpd.c and page_impl.c even when nothing changed, and
so does the one after that. The rule only needs to embed the files
into the image; the header comes from the html_data.c rule above.
Spotted by plaes in #358, which cannot be merged as it stands.
With this and the previous commit, a second make after a full build
compiles nothing, and the image is byte for byte the same as before.
Two things kept the same source from producing the same image.
fileadder walked the input directory with readdir(), which returns entries
in whatever order the filesystem hands them over. The same tree therefore
embedded the files in a different order on a different filesystem, and the
image differed byte for byte. scandir() with alphasort() fixes the order.
version.h was listed in .PHONY, which makes it permanently out of date and
propagates that to everything including it, so a plain build recompiled the
four largest translation units even when nothing had changed. Dropping it
from .PHONY alone would go too far the other way: with no prerequisites the
recipe would never run again and the version string would freeze at its
first value. A FORCE prerequisite keeps the recipe running every time, and
the generated header is only moved into place when its content actually
differs, so the timestamp survives an unchanged rebuild.
Measured on this tree: with FORCE the version follows GIT_VERSION and an
unchanged rebuild leaves version.h untouched; without it the header keeps
the value from the first build.
rtl837x_port.c prints "called" and "done" around mirroring, PVID, VLAN
and L2 setup, and dhcp.c around discover, request, start and stop. They
land on the console on every boot and config replay, and in the log
once syslog is running. The VLAN line keeps the VLAN, members and
tagged values, only without the function name in front.
On the KP_9000_6XHML_X2 this takes 359 bytes out of BANK1, 88 out of
BANK2 and 24 out of BANK3. BANK1 goes from 431 to 790 bytes free.
dhcp_addopt_request_ip() and dhcp_addopt_server_id() both end with a
copy of four bytes of our MAC to the current position, without moving
the position on. The next option always starts there, so those bytes
are written over before the packet goes out and never reach the wire.
The packet stays the same without the copy, and BANK3 is 102 bytes
smaller on the KP_9000_6XHML_X2.
ip_opt() copies four bytes whatever the option length says, so a
subnet, router, DNS, server id or broadcast option shorter than four
bytes stores the start of the next option as the address. long_opt()
does not look at the length at all and always moves on by six bytes, so
a lease, renewal or rebind option of another length puts the parser out
of step for the rest of the packet.
Both now read the length. An address or time shorter than four bytes is
skipped, a longer one is taken from its first four bytes and stepped
over whole, and a skipped time leaves the previous value alone.
Found in the code, not on the wire; no server here sends such options.
BANK3 is 47 bytes smaller on the KP_9000_6XHML_X2.
The main loop drains the NIC only after an RX interrupt, at most
RX_BUDGET frames, then clears the status and waits for the next one.
When frames are left and the buffer reaches the RX stop threshold, the
NIC takes no more frames and raises no further interrupt, so reception
stops for good.
Check the buffer after clearing the status and keep draining while it
holds data; the interrupt is re-enabled only once it is empty.
rtl8224_enable() waits 2.5 s after releasing the RTL8224 reset before
anything talks to the PHY. On an SWTGW218AS the chip id already reads
back 0x8224 at the first poll, about 265 ms after the release.
Remember when the reset was released and, once SMI is up, poll the chip
id until it answers, giving up after the same 2.5 s as before. The PHY
writes that follow read back the same as before.
The reset pulse before it goes from 0.5 s to 20 ms, twice the 10 ms the
related RTL8372 and RTL8221B need. The pulse does reset the chip: the
MDI reverse register (MMD 30, 0x0a90) still reads its power-on value
0x00f3 just before the init writes 0x00fc, as it does with 0.5 s.
Together the console banner comes 3 s sooner.
The SerDes write loop in phy_config_8224() waited on the busy bit with
no limit, so a PHY that never answers stopped the boot before the
network came up. Bound it and report a timeout.
The name branch read the fourth word without checking that the line had
one, and the tokenizer does not reset unused word slots, so "port 5 name"
took the name from wherever the previous command's fourth word had been:
"port 5 name hello" followed by "port 5 name" gave "hello" again, and
after a longer line it gave the tail of that line. The web UI never sends
the form, the console can. With no argument the name is now empty, which
also gives the console a way to remove a name.
The receive interrupt was wired up but never used: EX1 was enabled, the
handler re-enabled itself and did nothing, and the mask bit for it in the
NIC stayed clear, so idle() read the receive buffer register on every
pass to find out whether a packet had arrived, and slept until the next
tick when none had. A packet that arrived during that sleep waited for
the tick: 2.2 ms on average, 5 ms at worst, for every frame the CPU
handles.
Now the handler latches rx_irq and masks itself, and idle() sleeps
unless a receive, a link change, a console command or the tick is
pending. On rx_irq the loop drains the buffer, clears the status bit and
re-enables EX1, in that order: the NIC holds the line as long as the
status bit is set, and the bit does not clear while the buffer still
holds a frame, so a frame that lands between the last read and EX1 = 1
raises the interrupt again rather than getting lost. The other order,
clearing before draining, left the line asserted and produced an
interrupt on every pass. The mask is written after nic_setup(), which
resets it.
Measured on an RTL8373 board with the same counters in both images,
clocked by the tick counter, traffic from a host behind an LACP trunk:
```
polling interrupt
idle, register ops/s 788 79
25 pps, RTT avg/max 2.19/4.96 0.44/0.75 ms
500 pps, RTT avg/max 2.78/5.06 0.39/5.40 ms
500 pps, loss 0.08% 0.08%
```
The handler compiles to clr EX1 / setb rx_irq / reti.
The timer interrupt now raises a tick flag too, so idle() no longer works
out TICKS_DUE twice per pass, and cmd_available moves from xdata to a
bit. The four wake-up flags sit in the low bits of one byte at 0x20, so
the sleep test is a single read of that byte. tx_buf_empty deliberately
stays out: it is 1 when there is nothing to do, so including it would
keep the sleep test from ever firing.
idle() ran everything on every pass: the seconds block, the SFP pin
test, the TCP periodic timer behind a gate of its own, the STP timers
behind theirs. The timed work moves into handle_tick(), which idle()
calls once per system tick from a single gate; handle_tx() loses its
private one.
handle_sfp() ran on every pass as well, two GPIO reads per pass to learn
that nothing changed; it now runs every ten ticks. A module inserted is
still seen within 50 ms, and the read itself has been off the loop since
the wake-up wait moved to the tick.
No change in behaviour otherwise: the link, receive and console paths
stay where they were. SWTGW218AS: code +45 B, xdata +1 B, stack
unchanged.
The health checkpoints follow the work they measure: the seconds block,
SFP, TX and STP phases are taken inside handle_tick(), the link, RX and
command phases in idle().
The link-change interrupt was set up at boot (mode 0x5f84, mask 0x5f34)
but never enabled: the handler did nothing and did not clear the event in
the switch, so EX0 stayed commented out and idle() read the two link
registers on every pass instead, 400 reads a second on a quiet switch.
The handler now only latches a flag and masks itself; the register access
macros build a write in the shared SFR data bytes, so an interrupt that
touched the switch in the middle of a write from the main loop would
corrupt it. idle() clears the event register (write-one-to-clear, 0x3ff
for the ten ports), re-enables EX0 and then reads the link state, in that
order, so a change that lands during the read raises a new edge. The flag
is also set once a second from the seconds block, so a board whose INT0
line turns out not to be wired keeps detecting link changes, a second
late at worst. Boot drops the events pending from bringing the links up
before the mask is set.
Measured on an RTL8373 switch before this change: 0x5f88 held 0x3fd, nine
ports had reported a change nobody read. With the change a counter in the
loop went 2 -> 3 -> 4 on "port 5 off" / "port 5 on" and stayed at 4 over
three reads without events, so the event is delivered, cleared and
re-armed, and there is no flood.
SWTGW218AS: code +57 B, xdata 0, stack 151 -> 150 B; the flag is a bit
variable. The handler compiles to clr EX0 / setb link_irq / reti.
idle() also keeps in a bit what it saw in NIC_RX_BUFF_DATA when deciding
whether to sleep, and the first iteration of the receive loop uses it
instead of reading the register again a moment later. Between the two
only the link check and handle_sfp() run, and neither can make a queued
packet disappear. Every later iteration reads the register as before, so
the budget loop behaves as it did. This part was not measured on
hardware: one register read per pass needs a throughput run to show up.
SWTGW218AS: BANK1 and BANK2 +6 bytes each, BANK3 unchanged.
stp_timers() ran every fourth pass of the main loop and stp_clock counted
passes, which only measures time while a pass is one tick long. It never
quite was: a pass is as long as the packets it handles, and #482 now takes
up to four of them, so under an IPv4 stream a pass is about 16 ms and STP
time runs at a third of its rate, with hellos 6 s apart and a root aged out
after a minute. Since #483 the loop also stops sleeping while packets wait,
so under a flood of cheap frames passes come faster than ticks and STP time
runs fast instead, shortening the forward delay in exactly the situation a
loop produces. uip_periodic() had the same problem from the other side: it
ran once per pass, so the retransmit timer sped up with the loop.
Both now go by the tick counter the timer interrupt keeps. The STP block
catches up on ticks a long pass missed, at most eight steps at a time, and
handle_tx() runs its periodics once per tick at most. One byte of xdata
more, 48 bytes of code.
A host test drives the gate with one pass every 5 ms, every 16 ms and
every 0.6 ms for ten seconds each; the STP clock lands on 500 steps in
all three, where counting passes gave 156 and 4166. A pass of 1.3 s loses
time rather than firing 65 steps at once, and the arithmetic survives the
tick counter wrapping.
The advertisement written for a forced speed of 1G and above is 0x1001,
while auto and the forced 10M and 100M modes all carry bit 10, the PAUSE
capability. Forcing a port to 1g or faster therefore takes pause off the
link: the advertisement narrows and the far end has nothing left to
agree to.
Measured on a SWTGW218AS through phyget and physet, on a port with no
link so nothing was disturbed. Auto reads 0x15e1, port 1 1g reads
0x1001, port 1 auto reads 0x15e1 again, and writing 0x1401 by hand
sticks. That bit 10 is the pause capability follows from a second
reading: two ports advertise the same 0x15e1 from this side, and pause
is active only against the partner whose own advertisement carries the
bit.
Reported by feelfree69 in #489.
The .d files track headers, so adding or removing a flag on a tree that
was already built leaves every object that did not change for some other
reason compiled with the old flags, and the image ends up a mix of the
two.
How that shows depends on the flag. DIAG=1 fails the link on undefined
symbols, which at least stops the build. An opt-in whose hooks expand to
empty macros loses the calls instead, so the link succeeds and the
feature quietly measures nothing.
Objects now depend on a stamp holding CC_FLAGS, rewritten only when the
flags differ, so an unchanged flag set still does not force a rebuild.
Two per-port MAC status registers that nothing in the firmware reads:
0x6408 says which end is master on a BASE-T link, 0x63FC and 0x6400 say
whether pause was negotiated in each direction. port <n> show prints
both at the end.
The read sits in rtl837x_port.c rather than phy_show(), because these
are MAC registers and phy_show() is skipped for SFP ports. Master/slave
keys off machine.is_sfp[] so it stays off the fibre ports, where the
notion does not apply.
Bit N of each register is chip port N, which is not how
RTL837X_REG_LINKS_STS indexes its bytes a few lines above in the same
header.
edc469f made max_upload count, which bounds how many bytes a request may
carry. Nothing bounds where those bytes land. The two are not the same
check: the first watches the request, the second watches the flash address.
uptr walks forward one page at a time from FIRMWARE_UPLOAD_START and has no
ceiling. A body whose closing boundary never arrives keeps writing page after
page past the staging region, and on a flash exactly twice FIRMWARE_UPLOAD_START
the address wraps onto the running image at zero, so a bad upload takes the
working firmware with it. That commit's own message notes the related hazard:
flash_write_bytes() carries the address in three bytes and drops what does not
fit.
Bound it where uptr actually moves: refuse the next page once it would leave
the area check_and_flash_update_image() reads back, and answer 400. A partial
image left behind is harmless - boot rejects it on CRC and erases it.
This rebase drops the part of the original patch that removed max_upload;
the length limit and the address limit now sit side by side.
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.
The tcp slot dump printed the slot index behind `if (UIP_CONNS > 1)`, and
with one connection configured that is a constant-false condition around
live code: sdcc reported two "unreachable code" warnings for the body and
one "conditional flow changed by optimizer" for the loop that could only
run once.
Print the single slot directly and state the assumption as a build-time
check, so a build that raises UIP_CONNS fails loudly here instead of
quietly labelling every line as slot zero. Builds warning-free again.
What each line of the dump means, how two dumps give rates, and the
two things worth knowing before trusting the numbers: the command
inflates its own cmd phase counters by the cost of printing, and the
untouched figure is a high-water mark, not a live level.
Every probe of a misbehaving switch so far measured it from the
outside. This puts the witnesses inside.
Each idle() pass increments a loop counter and stamps a checkpoint
after every phase: link polling, SFP, RX, TX, STP and the command
dispatch. A phase spanning two or more system ticks counts as slow and
its worst time is kept, in 5 ms units. handle_rx counts frames, so the
dump separates a flooded CPU from a stalled loop.
At boot the free stack area is painted, preserving the interrupt state
around the painting, and the dump reports how much of the pattern
survives next to the live stack pointer. The TCP connection table is
printed whole: state, timers, retransmissions, ports and peer, next to
the httpd transfer state, so a permanently occupied slot names its
occupant. With more than one slot configured every slot is listed.
Everything is raw hex; two dumps a known time apart give the rates.
The implementation sits in cmd_parser next to the other command
handlers and needs no storage beyond its counters.
handle_sfp() called sfp_module_read() as soon as it saw a module appear, and
that path opened with delay(100). delay() spins on PCON until the tick counter
runs down, so the main loop stood still for 100 ticks - half a second at
SYS_TICK_HZ 200. Nothing else ran in that window: no STP, no LACP, no packet
handling. A diagnostic build measured the sfp phase at up to 107 ticks, the
longest stall of any phase in the loop.
Remember when the module showed up and let the loop continue. The I2C read
happens on a later pass, once the wake-up period has elapsed. Pulling a module
cancels a pending read, and a failed read still restores the detect bit, so the
existing retry-on-the-next-poll behaviour is unchanged.
The comparison is written out rather than using a helper so that this does not
depend on anything outside the current tree. The pending flag is an array
indexed by slot, like the other per-slot state in this file.
Cost on SWTGW218AS, measured against the branch point: BANK3 11173 -> 11264,
CSEG, BANK1 and BANK2 unchanged, plus 4 bytes of xdata.
idle() went to sleep before it looked at the receive buffer, so a packet that
arrived just after the loop dozed off waited for the next tick before anything
touched it. Every TCP round-trip paid that toll, which is why the web UI felt
sluggish even though no phase of the loop was anywhere near busy.
Check the receive buffer first and only sleep when it is empty. The read takes
its result straight from SFR_DATA_U16 rather than going through reg_read_m()
and the sfr_data[] copy, which is what the rest of the tree already does.
Measured on SWTGW218AS with this change alone:
web UI, median of 15 requests for the main page 88 ms -> 76 ms
CPU port under a one-way UDP stream 159 -> 230 packets/s
The loop still sleeps normally when there is nothing to do, so this does not
turn the core into a busy-wait: loops/ticks stays at 0.99 on an idle switch,
and rises above 1.0 only while packets are actually queued.
Costs 12 bytes of code and no extra XDATA.
handle_rx() picked up exactly one packet per iteration of the main loop. Since
the loop sleeps until the next tick, that capped the CPU port at roughly one
packet per tick no matter how little work it had to do. A burst of broadcast,
ARP or IGMP traffic aimed at the CPU queued up behind that and pushed out the
STP and LACP timers.
Drain up to RX_BUDGET packets in one pass instead. The budget matters: without
one a sustained flood would spin here and starve everything else in the loop.
The two register reads in the loop take their result straight from
SFR_DATA_U16 rather than going through reg_read_m() and the sfr_data[] copy,
which is what the rest of the tree already does.
Measured on SWTGW218AS with a one-way UDP stream, counting the packets the CPU
actually processed, with this change alone:
before 159 packets/s
RX_BUDGET 4 245 packets/s, loop iteration ~16 ms
RX_BUDGET 8 298 packets/s, loop iteration ~26 ms
Throughput is already most of the way there at 4, so the larger budget mostly
buys a longer worst-case delay for the other phases. STP stayed responsive in
both cases.
Costs 71 bytes of code and no extra XDATA.
The TODO next to the disabled TCP checksum code asks which bit of the RX tag
carries the result. It is bit 2 of byte 1: the hardware sets it when the L4
checksum is bad.
That answers the question for UDP as well, which was not being checked either
because UIP_CONF_UDP_CHECKSUMS is 0. Both paths now read the flag instead of
running a checksum over the payload on the 8051.
A zero UDP checksum, which is optional in IPv4, is not flagged by the hardware,
so DHCP keeps working and no extra condition is needed.
rx_headers lives in rtlplayground.c, which the host tests do not link, so
test_httpd_tx carries the storage itself the way it already carries uip_buf.
A zeroed descriptor reads as a good checksum, so the existing scenarios are
unaffected.
A new scenario sets the flag on a request and expects it counted as a checksum
error and left unanswered, then sends the same request with the flag clear and
expects it served. Built against uip.c without this change, those first two
checks fail and the two control checks still pass, so the scenario tells the
two versions apart rather than merely exercising the code.
Costs 156 bytes of code on SWTGW218AS and no extra XDATA.
The row sits between Hardware and Flash. The firmware sends the formatted string
rather than the raw register, since the System table prints its values as they
arrive.
Costs 186 bytes in BANK1.
The chip has a pair of thermal meters and nothing was reading them. This adds a
`temp` command that prints the current reading next to the one taken at power-on.
The result register at 0x340 splits a 19 bit value across two fields: bits 18:3
live in the low 16 bits, bits 2:0 in bits 18:16, with the sign at bit 18, degrees
in 17:10 and a fraction below. The nice part of that layout is that the degrees
fall out of a shift of the low half, so the conversion stays 16 bit.
The meter comes out of reset latched: the register keeps whatever was measured
when the board powered up and then never moves. Four samples 40 s apart came back
identical down to the raw ADC, and 0x340 matched the power-on register at 0x350.
Clearing bit 2 of the control register at 0x338 during init is what makes it live.
Measured on an SWTGW218AS: 58.7 to 59.5 C while running, with the power-on
register still showing the 49.2 C it started from. For what it is worth, the
scale lines up three ways: the threshold register reads -40 and +125, the slope
from warm-up works out at 4450 LSB/C, and the slope between the two meters at
4388 LSB/C.
We only read the meter that is up after boot. The other sits in reset with no
coefficients, so it would just repeat its power-on value.
Costs 231 bytes in BANK2.
The helper landed between that comment and the function it describes, so the
comment now reads as if it belonged to name_char(). Moving the helper up puts
the two back together. The generated code is unchanged.
DrDoof