Add ZoeRax HG0402XG V1.1 to supported devices

The ZoeRax HG0402XG is an unmanaged 4 2.5GBit Ethernet, 2 SFP+
device.

.dockerignore

@@ -3,6 +3,7 @@
 .gitattributes
 .github/
 output/
+toolchain/
 installer/output/
 *.bin
 html_data.c

.github/workflows/build.yml

@@ -3,6 +3,12 @@ name: Build firmware
 on:
   push:
     branches: ['**']
+  pull_request:
+
+permissions:
+  contents: read
+  pull-requests: write
+  issues: write
 
 jobs:
   build:
@@ -12,12 +18,41 @@ jobs:
 
     steps:
       - uses: actions/checkout@v6
+        with:
+          path: branch
+
+      - uses: actions/checkout@v6
+        with:
+          path: main
+          ref: main
 
       - name: Install dependencies
         run: |
           apt update
-          apt install make gcc sdcc xxd python-is-python3 libjson-c-dev -y
-      - name: Check if machine.c can be compiled for all machines
-        run: make machine_check
-      - name: Make project
-        run: make MACHINE="KP_9000_6XHML_X2"
+          apt install make gcc g++ bison flex libboost-dev patch bzip2 curl ca-certificates xxd python-is-python3 libjson-c-dev zlib1g-dev -y
+      - name: Cache the patched SDCC
+        uses: actions/cache@v4
+        with:
+          path: branch/toolchain
+          key: sdcc-${{ hashFiles('branch/sdcc/*') }}
+      - name: Run the host-side unit tests
+        run: cd branch && make -C test
+      - name: Check that every machine definition can be compiled
+        run: cd branch && make machine_check
+      - name: Make project (4x2/RTL8372)
+        run: cd branch && make CI=1 MACHINE="KP_9000_6XHML_X2" && cp output/KP_9000_6XHML_X2/rtlplayground.map /tmp/branch.map
+      - name: Make project (8x1/RTL8373)
+        run: cd branch && make CI=1 MACHINE="KP_9000_9XHML_X_V3_1"
+      - name: Make project with the health instrumentation
+        run: cd branch && make clean && make CI=1 MACHINE="KP_9000_6XHML_X2" HEALTH=1
+      - name: Make main
+        run: |
+          ln -s ../branch/toolchain main/toolchain
+          PATH="$(echo "$PWD"/branch/toolchain/sdcc-*/bin):$PATH"
+          cd main && make CI=1 MACHINE="KP_9000_6XHML_X2" && cp output/KP_9000_6XHML_X2/rtlplayground.map /tmp/main.map
+      - name: Calculate bank growth
+        run: |
+          sed -rn 's/^([XCD]SEG|(BANK[0-9])).*=\s+([0-9]+)\. bytes.*$/MAIN_\1=\3/p' /tmp/main.map | sort -u >> "$GITHUB_ENV"
+          sed -rn 's/^([XCD]SEG|(BANK[0-9])).*=\s+([0-9]+)\. bytes.*$/BRANCH_\1=\3/p' /tmp/branch.map | sort -u >> "$GITHUB_ENV"
+          du -sb main/output/KP_9000_6XHML_X2/html_min | awk '{print "MAIN_WEB="$1}' >> "$GITHUB_ENV"
+          du -sb branch/output/KP_9000_6XHML_X2/html_min | awk '{print "BRANCH_WEB="$1}' >>"$GITHUB_ENV"

.gitignore

@@ -1,9 +1,17 @@
 .gitignore
 .idea/
 output/
+output-main/
+toolchain/
 html_data.c
 html_data.h
 version.h
 tools/httpd_sim
 tools/injector
 tools/fileadder
+build/
+
+# host-test stand-ins for the two generated headers
+!test/stub/html_data.h
+!test/stub/version.h
+.vscode

CHANGELOG.md

@@ -4,13 +4,55 @@
 
 ## Added
 
+- Devices
+  - Support the hongyavision HYWS-SGT0204S (PCB marking `2G5F4_10G2_V1.01`), an
+    RTL8372 based 4x2.5G + 2xSFP+ unmanaged switch with a 512 KB Bright Moon
+    `T25S40` flash. It is a sibling of the KP-9000-6XH-X2 (2M-PCB43-V1.1) with
+    the two SFP+ cages wired the other way round (physical 5 is on SDS1,
+    physical 6 on SDS0), no reset button fitted, wrong silkscreen labels on the
+    serial header, `LED_GLB_IO_EN` bit 6 has to be set for the port LEDs to
+    work, and the fitted T25S40 cannot be read in DIO mode on this board, so the
+    target selects single-IO flash reads (`flash_sio`). New machine target
+    `MACHINE_2G5F4_10G2_V1_01`, see `doc/devices/2G5F4_10G2_V1.01.md`.
+
 - Web UI
   - Compress the embedded assets (minify + gzip) and switch to a single-page layout. #315
+  - Replace the multi-page UI with a themed single-page app: light, dark and Selenized themes following the browser by default,
+    English, Japanese and Chinese, save to flash merges the command log into the startup config and verifies the write,
+    firmware images are checked in the browser before upload. #429
 
 ## Changed
 
 ## Fixed
 
+- Time
+  - `rtl8372_init()` sets SDS 0 to SGMII as the stock firmware does. The masks were 0, so on RTL8372
+    boards without 10G ports both SerDes stayed off, and with SDS 0 off `RTL837X_REG_SEC_COUNTER`
+    does not hold a written value: NTP never retried after a failed first attempt at boot and
+    froze the clock after a sync.
+
+- Web UI
+  - Firefox no longer fills a saved login into the new admin password fields. It also put the
+    user name into the NTP poll interval, which marked the Time card as edited, so it kept its
+    defaults (NTP off, no server, UTC-12:00) instead of the switch's settings.
+  - Accept firmware uploads regardless of the file part's MIME type, including
+    `application/macbinary` sent by Chrome on macOS for `.bin` files.
+
+- Health
+  - `HEALTH=1` builds no longer write past the phase timing arrays at the LLDP checkpoint, which
+    corrupted the `link` slow counter and the DHCP client state. #658
+- Config
+  - A startup configuration line longer than the command buffer no longer stops the replay:
+    the offending line is skipped and the following lines are still applied.
+  - The web UI and the configuration upload now refuse a line the replay cannot take,
+    instead of writing it to flash.
+
+- Startup
+  - Initialised and zero-initialised globals above xdata 0x4000 now get their values at boot.
+    Hardware reset leaves SFR 0x97 at 1, which maps xdata 0x4000-0xbfff to the flash image,
+    and it was cleared only in `main()`, after the C startup code had copied and cleared them.
+    This showed as a random web session timeout and LLDP blocked on some ports without config.
+
 ## Breaking changes
 
 - Config

Dockerfile

@@ -3,11 +3,19 @@ FROM debian:13-slim
 RUN apt-get update && apt-get install -y \
     make \
     gcc \
-    sdcc \
+    g++ \
+    bison \
+    flex \
+    libboost-dev \
+    patch \
+    bzip2 \
+    curl \
+    ca-certificates \
     xxd \
     python3 \
     libjson-c-dev \
     git \
+    zlib1g-dev \
     && rm -rf /var/lib/apt/lists/*
 
 # git safe.directory for mounted repos (Makefile uses git describe)

Makefile

@@ -4,23 +4,63 @@ DEFAULT_CONFIG_LOCATION = 454656
 CONFIG_LOCATION = 458752
 HTML_LOCATION = 262144
 
+# Startup configuration embedded in the image. Override it to build a specific
+# machine with its own configuration: make MACHINE=<name> CONFIG=<file>
+CONFIG ?= config.txt
+
+SDCC_TOOLCHAIN = sdcc-$(shell cat sdcc/* | cksum | cut -d' ' -f1)
+SDCC_PREFIX = toolchain/$(SDCC_TOOLCHAIN)
 ifeq ($(origin CC),default)
-CC = sdcc
+SDCC = $(SDCC_PREFIX)/bin/sdcc
+CC = $(SDCC)
+ASM = $(SDCC_PREFIX)/bin/sdas8051
+SDCC_REV := $(shell git log -1 --format=%h -- sdcc 2>/dev/null)
+ifneq ($(SDCC_REV),)
+ifneq ($(shell git status --porcelain --untracked-files=no -- sdcc 2>/dev/null),)
+SDCC_REV := $(SDCC_REV)-dirty
 endif
-CC_FLAGS = -mmcs51 -I. -Ihttpd -Iuip
+endif
+SDCC_BANNER = $(if $(SDCC_REV),sdcc/ at $(SDCC_REV),$(SDCC_TOOLCHAIN)) ($(words $(wildcard sdcc/*.patch)) patches)
+else
+SDCC_BANNER = system
+endif
+CC_FLAGS = -mmcs51 -I. -Imachine -Ihttpd -Iuip
 ASM ?= sdas8051
 AFLAGS= -plosgff
 
 SUBDIRS := tools
 SUBDIRSCLEAN=$(addsuffix clean,$(SUBDIRS))
 
-ifeq ($(MACHINE),)
-	MACHINE:= $(shell grep "^\s*#define MACHINE_" machine.h | sed "s/^\s*#define MACHINE_//")
-else
-	CC_FLAGS += -DMACHINE_$(MACHINE)
+# $MACHINE selects machine/$(MACHINE).c as the board definition and is passed
+# on as -DMACHINE_NAME, which machine.h uses to include machine/$(MACHINE).h.
+# Board selection itself needs no MACHINE_* symbol for #if chains, so the
+# definition is all the compiler has to know.
+ifneq ($(MACHINE),)
+	CC_FLAGS += -DMACHINE_NAME=$(MACHINE)
+endif
+# Health instrumentation and the "health" console command: HEALTH=1 gmake ...
+ifneq ($(HEALTH),)
+	CC_FLAGS += -DHEALTH
 endif
 
-BUILDDIR = output/$(MACHINE)
+ifeq ($(CI),1)
+	CC_FLAGS += --Werror
+endif
+
+# The board definition being built: one file per MACHINE_* symbol in machine.h.
+MACHINE_SRC = machine/$(MACHINE).c
+
+MACHINE_INIT_DEFAULT := machine/default_machine_init.c
+MACHINE_INIT_SRC := $(wildcard machine/$(MACHINE)_init.c)
+ifeq ($(MACHINE_INIT_SRC),)
+	MACHINE_INIT_SRC := $(MACHINE_INIT_DEFAULT)
+endif
+
+ifndef BUILDDIR
+	BUILDDIR = output/$(MACHINE)
+endif
+
+CONFIG_STAMP := $(BUILDDIR)/.config
 VERSION_HEADER := version.h
 
 GIT_VERSION := $(shell git rev-parse --short HEAD)
@@ -43,36 +83,74 @@ BUILD_DATE := $(shell date -u -d @$(SOURCE_DATE_EPOCH) +"%Y-%m-%d %H:%M:%S" 2>/d
 	|| date -u -r $(SOURCE_DATE_EPOCH) +"%Y-%m-%d %H:%M:%S")
 endif
 
-all: create_build_dir $(VERSION_HEADER) $(SUBDIRS) $(BUILDDIR)/rtlplayground-$(FILENAME_EXTENSION).bin
+# $(MACHINE_SRC) and $(CONFIG) are listed here so an unknown $MACHINE or a
+# missing config file fails before anything else is built, including the tools/
+# subdirectory.
+all: create_build_dir $(VERSION_HEADER) $(MACHINE_SRC) $(CONFIG) $(SUBDIRS) $(BUILDDIR)/rtlplayground-$(FILENAME_EXTENSION).bin
 
-create_build_dir:
+bank_size:
+	tools/bank_occupancy.sh "$(MACHINE)"
+
+create_build_dir: $(SDCC)
 	mkdir -p "$(BUILDDIR)"
 	mkdir -p "$(BUILDDIR)/uip"
 	mkdir -p "$(BUILDDIR)/httpd"
+	mkdir -p "$(BUILDDIR)/machine"
 
-# Keep machine.c in first position to fail immediately on invalid $MACHINE value
+# Every machine is a pair machine/<MACHINE>.c + machine/<MACHINE>.h. Selecting
+# them by name means no #if/#elif chain, so an unknown or empty $MACHINE is
+# caught here instead of by a #error further down the build.
+$(MACHINE_SRC):
+	@echo "ERROR: MACHINE='$(MACHINE)' has no definition (expected $(MACHINE_SRC))." >&2
+	@echo "       Every machine is one file pair under machine/, named after the" >&2
+	@echo "       MACHINE_* symbol it defines; see machine/*.c for the full list." >&2
+	@echo "       Example: make MACHINE=KP_9000_6XHML_X2" >&2
+	@false
+
+# Fail with a clear message when CONFIG names a file that does not exist.
+$(CONFIG):
+	@echo "ERROR: CONFIG='$(CONFIG)' does not exist." >&2
+	@echo "       Point CONFIG at a configuration file, or omit it to use config.txt." >&2
+	@false
+
+# Keep the per-machine definition in first position to fail immediately on an
+# invalid $MACHINE value
 SRCS = \
-	machine.c \
-	machine_init.c \
+	$(MACHINE_SRC) \
+	$(MACHINE_INIT_SRC) \
 	cmd_editor.c \
 	cmd_parser.c \
 	dhcp.c \
 	html_data.c \
 	rtlplayground.c \
+	boot.c \
+	sfp.c \
+	dns.c \
+	dns_cli.c \
+	sessions.c \
+	ntp.c \
+	ntp_cli.c \
 	syslog.c \
 	udp_apps.c
 
 # RTL837x
 SRCS += \
 	rtl837x_bandwidth.c \
+	rtl837x_cabletest.c \
 	rtl837x_flash.c \
 	rtl837x_igmp.c \
 	rtl837x_init.c \
+	rtl837x_nic.c \
 	rtl837x_leds.c \
+	rtl837x_leds_dump.c \
 	rtl837x_phy.c \
 	rtl837x_pins.c\
 	rtl837x_port.c \
-	rtl837x_stp.c
+	rtl837x_stp.c \
+	rtl837x_lldp.c \
+	rtl837x_storm.c \
+	rtl837x_lacp.c \
+	rtl837x_trap.c
 SRCS += \
 	httpd/httpd.c \
 	httpd/page_impl.c
@@ -92,71 +170,120 @@ HTML := $(shell find html -name '*.js' -or -name '*.html' -or -name '*.svg' -or
 # Minified copy of the web UI sources, used as the fileadder input.
 # The raw html/ sources stay untouched for development; the minified
 # copy is a build artifact under output/.
-HTML_MIN := output/html_min
+HTML_MIN := $(BUILDDIR)/html_min
+HTML_MIN_LIST := $(BUILDDIR)/html_min.list
+HTML_MIN_STAMP := $(BUILDDIR)/html_min.stamp
 .PHONY: html_min
-html_min: $(HTML)
+html_min: $(HTML_MIN_STAMP)
+
+$(HTML_MIN_LIST): FORCE
+	@mkdir -p $(BUILDDIR)
+	@echo $(sort $(HTML)) > $@.tmp
+	@cmp -s $@.tmp $@ && rm -f $@.tmp || mv $@.tmp $@
+
+$(HTML_MIN_STAMP): $(HTML) $(HTML_MIN_LIST) tools/minify.py $(if $(wildcard $(HTML_MIN)),,FORCE)
 	rm -rf $(HTML_MIN)
 	mkdir -p $(HTML_MIN)
 	@for f in $(HTML); do python3 tools/minify.py $$f $(HTML_MIN)/$$(basename $$f) || exit 1; done
+	@touch $@
 
-html_data.c html_data.h &: $(HTML) | tools html_min
-	tools/output/fileadder -a $(HTML_LOCATION) -s $(IMAGESIZE) -b BANK1 -z -d $(HTML_MIN) -p html_data
+html_data.c html_data.h &: $(HTML_MIN_STAMP) | tools
+	tools/$(BUILDDIR)/fileadder -a $(HTML_LOCATION) -s $(IMAGESIZE) -b BANK1 -z -d $(HTML_MIN) -p html_data
 
-$(VERSION_HEADER):
+$(VERSION_HEADER): FORCE
 	@printf '%s\n' "#ifndef VERSION_H" "#define VERSION_H" \
 		"#define VERSION_SW \"$(VERSION_EXTENSION)\"" \
 		"#define BUILD_DATE \"$(BUILD_DATE)\"" \
-		"#endif" > $(VERSION_HEADER)
+		"#define SDCC_TOOLCHAIN \"$(SDCC_BANNER)\"" \
+		"#endif" > $(VERSION_HEADER).tmp
+	@cmp -s $(VERSION_HEADER).tmp $(VERSION_HEADER) \
+		&& rm -f $(VERSION_HEADER).tmp \
+		|| mv $(VERSION_HEADER).tmp $(VERSION_HEADER)
 
 httpd: html_data.h
 
 $(SUBDIRS):
-	$(MAKE) -C $@
+	$(MAKE) -C $@ BUILDDIR=$(BUILDDIR)
 
 clean: $(SUBDIRSCLEAN)
-	-rm -f html_data.c html_data.h $(VERSION_HEADER)
+	-rm -f html_data.c html_data.h $(VERSION_HEADER) $(HTML_MIN_LIST) $(HTML_MIN_STAMP)
 	-if [ -d $(BUILDDIR) ]; then find $(BUILDDIR) -type f ! -name "*.bin" -delete; fi
 
 distclean: $(SUBDIRSCLEAN)
-	-rm -f html_data.c html_data.h $(VERSION_HEADER)
+	-rm -f html_data.c html_data.h $(VERSION_HEADER) $(HTML_MIN_LIST) $(HTML_MIN_STAMP)
 	-rm -rf $(BUILDDIR)
+	-rm -rf toolchain
+
+ifneq ($(SDCC),)
+$(SDCC):
+	sh sdcc/build.sh $(abspath $(SDCC_PREFIX))
+endif
 
 $(SUBDIRSCLEAN):
 	$(MAKE) -C $(@:clean=) clean
 
-$(BUILDDIR)/%.rel: %.c | create_build_dir html_data.h
+# Objects depend on the flags they were built with, through a stamp file that
+# is rewritten only when CC_FLAGS actually changes.
+CCFLAGS_STAMP := $(BUILDDIR)/.ccflags
+
+.PHONY: FORCE
+FORCE:
+
+$(CCFLAGS_STAMP): FORCE | create_build_dir
+	@echo '$(CC_FLAGS)' | cmp -s - $@ 2>/dev/null || echo '$(CC_FLAGS)' > $@
+
+# The same for $(CONFIG): its path is stamped so switching to a config file
+# older than the image still rebuilds it, while $(CONFIG) itself as a normal
+# prerequisite covers edits to the selected file.
+$(CONFIG_STAMP): FORCE | create_build_dir
+	@echo '$(CONFIG)' | cmp -s - $@ 2>/dev/null || echo '$(CONFIG)' > $@
+
+$(BUILDDIR)/%.rel: %.c $(CCFLAGS_STAMP) | create_build_dir html_data.h
 	$(CC) -MMD $(CC_FLAGS) -o $@ -c $<
 
-$(BUILDDIR)/%.rel: %.asm | create_build_dir
+$(BUILDDIR)/%.rel: %.asm $(CCFLAGS_STAMP) | create_build_dir
 	${ASM} ${AFLAGS} -o $@ $<
 #	mv -f $(addprefix $(basename $^), .lst .rel .sym) .
 
 $(BUILDDIR)/rtlplayground.ihx: $(OBJS) $(BUILDDIR)/crtbank.rel $(BUILDDIR)/crc16.rel
-	$(CC) $(CC_FLAGS) -Wl-bHOME=0x00000 -Wl-bBANK1=0x14000 -Wl-bBANK2=0x24000 -Wl-r -o $@ $^
+	$(CC) $(CC_FLAGS) --xram-size 49151 -Wl-bHOME=0x00000 -Wl-bBANK1=0x14000 -Wl-bBANK2=0x24000 -Wl-bBANK3=0x34000 -Wl-r -o $@ $^
 
 $(BUILDDIR)/rtlplayground.img: $(BUILDDIR)/rtlplayground.ihx
 	objcopy --input-target=ihex -O binary $< $@
 
-$(BUILDDIR)/rtlplayground-$(FILENAME_EXTENSION).bin: $(BUILDDIR)/rtlplayground.img | tools
+$(BUILDDIR)/rtlplayground-$(FILENAME_EXTENSION).bin: $(BUILDDIR)/rtlplayground.img $(CONFIG) $(CONFIG_STAMP) | tools
 	if [ -e $@ ]; then rm $@; fi
-	tools/output/imagebuilder -i $^ $@
-	tools/output/fileadder -a $(DEFAULT_CONFIG_LOCATION) -s $(IMAGESIZE) -d config.txt $@
-	tools/output/fileadder -a $(CONFIG_LOCATION) -s $(IMAGESIZE) -d config.txt $@
-	tools/output/fileadder -a $(HTML_LOCATION) -s $(IMAGESIZE) -z -d $(HTML_MIN) -p html_data -b BANK1 $@
-	tools/output/crc_calculator -u $@
-	ln -sf $(MACHINE)/rtlplayground-$(FILENAME_EXTENSION).bin output/rtlplayground.bin
+	tools/$(BUILDDIR)/imagebuilder -i $(BUILDDIR)/rtlplayground.img $@
+	tools/$(BUILDDIR)/fileadder -a $(DEFAULT_CONFIG_LOCATION) -s $(IMAGESIZE) -d "$(CONFIG)" $@
+	tools/$(BUILDDIR)/fileadder -a $(CONFIG_LOCATION) -s $(IMAGESIZE) -d "$(CONFIG)" $@
+	tools/$(BUILDDIR)/fileadder -a $(HTML_LOCATION) -s $(IMAGESIZE) -z -d $(HTML_MIN) -b BANK1 $@
+	tools/$(BUILDDIR)/crc_calculator -u $@
+	ln -sf $(MACHINE)/rtlplayground-$(FILENAME_EXTENSION).bin $(BUILDDIR)/../rtlplayground.bin
 
-.PHONY: clean distclean all $(SUBDIRS) $(SUBDIRSCLEAN) $(VERSION_HEADER) create_build_dir
+.PHONY: clean distclean all $(SUBDIRS) $(SUBDIRSCLEAN) create_build_dir
 
-.PHONY:
-machine_check:
+# Every machine/ definition is compiled, so a machine added to machine/ is
+# covered without registering it anywhere. A board only counts once both its
+# <machine>.c and its <machine>.h are present: the header is what the rest of
+# the build includes, so a .c without one is an incomplete board. The _init.c
+# files are excluded here: they are the boot hooks, compiled separately below
+# (the board's own if it exists, otherwise default_machine_init.c). The
+# machine-specific defines are filtered out of $(CC_FLAGS) and supplied per
+# iteration instead, so this also works as "make machine_check MACHINE=...".
+MACHINE_DEFS := $(sort $(filter $(patsubst machine/%.h,%,$(wildcard machine/*.h)),$(patsubst machine/%.c,%,$(filter-out machine/%_init.c,$(wildcard machine/*.c)))))
+MACHINE_LESS_CC_FLAGS = $(filter-out -DMACHINE_%,$(CC_FLAGS))
+
+.PHONY: machine_check
+machine_check: $(SDCC)
 	@mkdir -p $(BUILDDIR)/tmp
 	@set -eo pipefail; \
-	for MACHINE in `grep -E '^[[:space:]]*(//[[:space:]]*)?#define MACHINE_' machine.h | sed -E 's%^[[:space:]]*(//[[:space:]]*)?#define MACHINE_%%' | awk '{print $$1}' | sort -u`; \
+	for MACHINE in $(MACHINE_DEFS); \
 	do \
 	echo "Checking $${MACHINE}"; \
-	$(CC) $(CC_FLAGS) -DMACHINE_$${MACHINE} -MMD -o $(BUILDDIR)/tmp/machine_check -c machine.c; \
-	$(CC) $(CC_FLAGS) -DMACHINE_$${MACHINE} -MMD -o $(BUILDDIR)/tmp/machine_check -c machine_init.c; \
+	$(CC) $(MACHINE_LESS_CC_FLAGS) -DMACHINE_NAME=$${MACHINE} -MMD -o $(BUILDDIR)/tmp/machine_check -c machine/$${MACHINE}.c; \
+	INIT=machine/$${MACHINE}_init.c; \
+	if [ ! -e $$INIT ] && [ ! -L $$INIT ]; then INIT=$(MACHINE_INIT_DEFAULT); fi; \
+	$(CC) $(MACHINE_LESS_CC_FLAGS) -DMACHINE_NAME=$${MACHINE} -MMD -o $(BUILDDIR)/tmp/machine_check -c $$INIT; \
 	done
 	@rm -rf $(BUILDDIR)/tmp
 

README.md

@@ -2,15 +2,15 @@
 A Playground for Firmware development for advanced user of RTL8372/RTL8373 based 2.5GBit Switches.
 
 For each hardware configuration of these devices, there is usually a managed and an
-umanaged version sold, with mostly identical hardware. The aim is to provide management
+unmanaged version sold, with mostly identical hardware. The aim is to provide management
 features also for unmanaged devices with additional features such as Management VLAN,
-dhcp servers, multi-language support, IPv6 and TLS-encrypted web-pages. At present, however
+DHCP servers, multi-language support, IPv6 and TLS-encrypted web-pages. At present, however
 only the following features are provided:
 - A modern web-interface with mouse-over to display further information
 - A serial console interface to configure all features
 - IGMP to configure Multicast streaming
-- Port configuration showing detailed informtion about own and Link-partner advertised
-  Speed settins and configuration of these settings on the local side
+- Port configuration showing detailed information about own and Link-partner advertised
+  Speed settings and configuration of these settings on the local side
 - Per-port configuration of frame sizes (MTUs) for Jumbo-Frame support or limiting MTUs
   for particular devices
 - EEE (Energy Efficient Ethernet) can be configured per-port. Detailed information is
@@ -25,7 +25,7 @@ only the following features are provided:
 - Firmware updates via the web
 - Installation as a firmware upgrade from the original web-interface
 
-<img width="1420" height="623" alt="GUI" src="doc/images/gui.png" />
+<img width="1673" height="977" alt="GUI" src="doc/images/gui.png" />
 
 While the firmware provides already considerable improvements over the original managed firmware,
 the firmware still lacks support for the proprietary loop prevention
@@ -37,7 +37,7 @@ a backup of the original flash content via a SOIC clamp such as also used for BI
 backups and can re-install that firmware in case something is wrong. For this no soldering
 skills are necessary.
 
-The firmware supports all hardware featues of devices with
+The firmware supports all hardware features of devices with
 - 4 2.5GBit ports + 2 SFP+ ports
 - 5 2.5GBIT + 1 SFP+ port
 - 8 2.5GBit + 1 SFP+ port
@@ -56,12 +56,19 @@ device simulator is provided, which runs entirely under Linux as a local webserv
 
 ## (0) Compiling Requirements
 
-Install the following particular build requisites (Debian 12/13), note that Ubuntu 24.04
-still has an older version of sdcc, but you will need sdcc version 4.5 for the code to compile:
+Install the following particular build requisites (Debian 12/13):
 ```
-sudo apt install make gcc sdcc xxd python-is-python3 libjson-c-dev
+sudo apt install make gcc g++ bison flex libboost-dev patch bzip2 curl xxd python-is-python3 libjson-c-dev zlib1g-dev
 ```
 
+The first `make` downloads SDCC 4.6.0, applies the fixes in `sdcc/` and builds it into
+`toolchain/`, which takes a few minutes. Later builds reuse it. The released SDCC 4.5.0
+and 4.6.0 miscompile parts of this code, see [SDCC bug #4153](https://sourceforge.net/p/sdcc/bugs/4153/)
+and [#4154](https://sourceforge.net/p/sdcc/bugs/4154/); both fixes are in SDCC trunk since
+r17039. The third patch is for [#4070](https://sourceforge.net/p/sdcc/bugs/4070/), where the
+assembler drops the top byte of a `__sfr32` address list; it is proposed upstream and lets
+`SFR_DATA_U32` be used again. To use an SDCC you installed yourself instead, run `make CC=sdcc ...`.
+
 <details>
 <summary>If using Docker (click to expand)</summary>
 
@@ -103,24 +110,36 @@ docker run --rm -p 8080:8080 -v $(pwd):/workspace rtlplayground-dev \
   tools/output/httpd_sim /workspace/html
 ```
 
-Edit `machine.h` or `config.txt` on your host, then re-run `make` — the
+Edit `config.txt` on your host, then re-run `make` — the
 source directory is mounted into the container, so changes take effect
-immediately. To build for a different machine, pass `MACHINE=...`.
+immediately. To build for a different machine, pass `MACHINE=<name>`, which
+selects `machine/<name>.c` (see [Adding a new device](doc/adding-a-device.md)).
 
 </details>
 
 ## (1) Compiling for direct chip flashing AND upgrading an existing RTLPlayground running device
 
-Edit machine.h with an editor like vi or nano. Select the correct machine the firmware should build for.
+Pass the machine that matches your device, for example:
+```
+make MACHINE=DEFAULT_8C_1SFP
+```
+The name selects the board definition `machine/<name>.c`; `make` without a
+`MACHINE` value stops with an error listing what is expected. If your board is
+not in the list yet, see [Adding a new device](doc/adding-a-device.md).
+
+To embed a machine-specific startup configuration, pass `CONFIG`:
+```
+make MACHINE=DEFAULT_8C_1SFP CONFIG=configs/DEFAULT_8C_1SFP.txt
+```
+`CONFIG` defaults to `config.txt`; the selected file is embedded in the image
+and changing it triggers a rebuild.
 
 > [!TIP]
 > You can write configuration parameters in config.txt (see below) in order your switch to get
 > straight at the first boot, a correct IP configuration.
 
-Now, building the firmware image should work:
-```
-make 
-```
+Now, building the firmware image should work.
+
 Note, that the image generated ends in .bin, not .img, in order to make IMSProg happy.
 
 image location is stored in `RTLPlayground/output/rtlplayground_version_machine.bin`
@@ -131,7 +150,7 @@ rtlplayground-v0.1.0-12c98ba-dirty-LIANGUO_ZX_SWTGW215AS.bin
 
 > [!CAUTION]
 > This image can be flashed directly to the chip OR through the firmware update/upgrade
-> interface of RTLPlaygound interface
+> interface of RTLPlayground interface
 
 ## (2) Compiling for OEM running device with management options (web upgrade)
 
@@ -158,9 +177,9 @@ example of compilation console output
 RTLPlayground/installer$ make
 mkdir -p output
 gcc updatebuilder.c -o output/updatebuilder
-sdas8051 -plosgff -o output/crtstart.rel crtstart.asm
-sdcc -mmcs51 --code-loc 0x1000 -o output/installer.rel -c installer.c
-sdcc -mmcs51 -Wl-bHOME=0x1100 -Wl-r -o output/rtlinstaller.ihx output/crtstart.rel output/installer.rel
+../toolchain/sdcc-3390765201/bin/sdas8051 -plosgff -o output/crtstart.rel crtstart.asm
+../toolchain/sdcc-3390765201/bin/sdcc -mmcs51 --code-loc 0x1000 -o output/installer.rel -c installer.c
+../toolchain/sdcc-3390765201/bin/sdcc -mmcs51 -Wl-bHOME=0x1100 -Wl-r -o output/rtlinstaller.ihx output/crtstart.rel output/installer.rel
 ./output/updatebuilder -i output/rtlinstaller.ihx -o output/rtlplayground_oem_upgrade.bin ../output/rtlplayground.bin
 Input file size: 524288
 Bytes read: 524288
@@ -175,22 +194,22 @@ Header checksum is: 0x5a1
 ## (3) Sandbox Usage with Ghidra (optional)
 
 You can play with the image using ghidra or flash real Switch Hardware. For
-ghidra see this information about [Ghidra images](ghidra.md).
+ghidra see this information about [Ghidra images](doc/ghidra.md).
 
 ## (4) Installation through the Web interface (software way)
 
-Managed switches (OEM firmware of RTLplaygroud firmware) can be upgraded via the web interface.
+Managed switches (OEM firmware of RTLplayground firmware) can be upgraded via the web interface.
 Unmanaged switch cannot be flashed this way (see 5).
 
 Go to "Firmware update" tab, select the correct file.
 
 > [!IMPORTANT]
 > If your device already runs RTLPlayground, you must upload the binary file /RTLPlayground/output/rtlplayground_Version_Machine.bin
-> If your device is OEM, you must upload the binary file /RTLPlayground/installer/outputrtlplayground_oem_upgrade.bin
+> If your device is OEM, you must upload the binary file /RTLPlayground/installer/output/rtlplayground_oem_upgrade.bin
 
 > [!CAUTION]
 > Check one more time that your device matches the machine type before flashing.
-> Be shure you have a backup of the original firmware before diving in RTLPlaygroung.
+> Be sure you have a backup of the original firmware before diving in RTLPlayground.
 
 Finally, push the Upload File Button and you're done !
 
@@ -198,7 +217,7 @@ Finally, push the Upload File Button and you're done !
 ## (5) Flashing the ROM directly (hardware way, but also only way to rescue)
 
 This procedure is the only way to flash unmanaged switches, if the ROM chip is large enough.
-This is also the only way to unbrick your device if something went wroong.
+This is also the only way to unbrick your device if something went wrong.
 
 > [!IMPORTANT]
 > You need a SOIC-8 clip to flash the ROM chip directly onboard.
@@ -218,7 +237,7 @@ This is also the only way to unbrick your device if something went wroong.
 - ERASE THE ROM (BLANK) !
 - Load the firmware into IMSProg.
 - Flash is to the ROM chip.
-- Disconect the clip from the ROM chip.
+- Disconnect the clip from the ROM chip.
 - You're done, ready for the first boot.
 
 ## (6) Connecting a serial interface (optional)
@@ -233,7 +252,7 @@ When you power up the switch, the device will perform some examples and provide
 ## (8) The web-interface
 
 The web-interface can be reached under the [default 192.168.10.247](http://192.168.10.247) unless you
-specified an IP adress in the config.txt before compilation.
+specified an IP address in the config.txt before compilation.
 
 > [!TIP]
 > The default password is `1234`.
@@ -333,17 +352,18 @@ While in compilation part, you might write directly to config.txt file before ma
 nano config.txt
 ```
 
-If you want to modify settings after the flash is done, go to the Advanced Settings tab in System Settings
+If you want to modify settings after the flash is done, go to the System tab and find the Startup Configuration.
 
-<img width="1085" height="646" alt="ADVANCED SETTINGS" src="doc/images/advanced_settings.png" />
+<img width="1673" height="978" alt="ADVANCED SETTINGS" src="doc/images/advanced_settings.png" />
 
 ```
-ip xxx.xxx.xxx.xxx      = IP adress of the switch
-gw yyy.yyy.yyy.yyy      = IP adress of the gateway
+ip xxx.xxx.xxx.xxx      = IP address of the switch
+gw yyy.yyy.yyy.yyy      = IP address of the gateway
 netmask zzz.zzz.zzz.zzz = Network mask of the switch 
 port x name xxx         = Name xxx the port number x
 port z 1g               = Set 1g speed for port z
 igmp on/off             = Turn IGMP on or off
+session xxxx            = Web session timeout in seconds (default 200)
 ```
 [To be continue]
 
@@ -352,11 +372,14 @@ Enjoy playing!
 ## (11) Other documents
 
 The following documents give further documentation on specific features of the RTL837x SoCs:
+- [Adding a new device](doc/adding-a-device.md)
 - [RTL8372/3 Feature support](doc/hardware.md)
 - [CPU Port](doc/CpuPort.md)
 - [L2 learning](doc/l2.md) 
 - [IGMP (IP-MC streaming)](doc/igmp.md)
 - [SFP+ ports](doc/sfp.md) 
-- [Trunking aka. port aggregation](doc/trunking.md)
+- [Trunking aka. port aggregation](doc/link_aggregation.md)
+- [LACP (802.3ad link aggregation)](doc/lacp.md)
 - [VLAN](doc/vlan.md)
+- [Storm control](doc/storm_control.md)
 - [Modifications and Flash replacement](doc/mods.md)

boot.c

@@ -0,0 +1,483 @@
+#pragma codeseg BANK3
+#pragma constseg BANK3
+
+#include <8051.h>
+#include <stdint.h>
+#include <stdbool.h>
+
+#include "rtl837x_sfr.h"
+#include "rtl837x_regs.h"
+#include "rtl837x_common.h"
+#include "rtl837x_flash.h"
+#include "rtl837x_pins.h"
+#include "rtl837x_phy.h"
+#include "rtl837x_port.h"
+#include "rtl837x_init.h"
+#include "machine.h"
+#include "phy.h"
+#include "boot.h"
+
+extern __code const struct machine machine;
+extern __xdata struct machine_runtime machine_detected;
+extern __xdata uint8_t sfr_data[4];
+extern volatile __xdata uint32_t ticks;
+__xdata uint32_t rtl8224_release_tick;
+extern __xdata uint8_t tx_seq;
+extern __code const uint8_t * __code const hex;
+
+
+
+void sds_config_mac(uint8_t sds, uint8_t mode) __banked
+{
+	reg_read_m(RTL837X_REG_SDS_MODES);
+	sfr_data[0] = 0;
+	sfr_data[1] = 0;
+	switch (sds) {
+	case 0:
+		sfr_mask_data(0, 0x1f, mode);
+		break;
+	case 1:
+		sfr_mask_data(0, 0xe0, mode << 5);
+		sfr_mask_data(1, 0x03, mode >> 3);
+		break;
+	case 2:
+		sfr_mask_data(1, 0xfc, 0x02 << 2);
+	}
+	if (machine_detected.isRTL8373) // Set 3rd SERDES Mode to 0x2 for RTL8224
+		sfr_mask_data(1, 0xfc, 0x02 << 2);
+	else
+		sfr_data[2] &= 0x03;
+	reg_write_m(RTL837X_REG_SDS_MODES);
+	print_string("\nRTL837X_REG_SDS_MODES: ");
+	print_reg(RTL837X_REG_SDS_MODES);
+	print_string("\n");
+}
+
+
+
+void early_boot_handle_button(void) __banked
+{
+	if (machine.reset_pin == GPIO_NA)
+		return;
+
+	gpio_input_setup(machine.reset_pin);
+
+	// Debounce after init
+	delay(100);
+	// If the button is not already held at boot, continue normally.
+	if (gpio_pin_test(machine.reset_pin))
+		return;
+
+	set_sys_led_state(SYS_LED_FAST);
+	print_string("\n[Reset button held at boot]\n");
+
+	if (gpio_pin_test(machine.reset_pin))
+		return;
+
+	const __xdata uint32_t min_hold_ticks = 10UL * SYS_TICK_HZ;
+	const __xdata uint32_t max_hold_ticks = 30UL * SYS_TICK_HZ;
+	const __xdata uint32_t blink_ticks = SYS_TICK_HZ / 10;      // 100 ms
+	const __xdata uint32_t pause_ticks = SYS_TICK_HZ / 2;       // 500 ms
+	__xdata uint32_t start_ticks = ticks;
+	__xdata uint32_t last_blink_step = start_ticks;
+	__xdata uint8_t blink_step = 0;
+
+	set_sys_led_state(SYS_LED_ON);
+
+	while (!gpio_pin_test(machine.reset_pin)) {
+		__xdata uint32_t held_ticks = ticks - start_ticks;
+
+		if (held_ticks > max_hold_ticks) {
+			print_string("[Button held >30s at boot; continuing normal boot]\n");
+			return;
+		}
+
+		// Double blink pattern while button is held:
+		// ON (100ms), OFF (100ms), ON (100ms), OFF (500ms)
+		__xdata uint32_t step_ticks = (blink_step == 3) ? pause_ticks : blink_ticks;
+		if ((ticks - last_blink_step) >= step_ticks) {
+			blink_step = (blink_step + 1) & 0x3;
+			set_sys_led_state((blink_step == 0 || blink_step == 2) ? SYS_LED_ON : SYS_LED_OFF);
+			last_blink_step = ticks;
+		}
+
+		PCON |= 1;
+	}
+
+	set_sys_led_state(SYS_LED_ON);
+
+	if ((ticks - start_ticks) >= min_hold_ticks) {
+		print_string("[Button held 10s-30s at boot; restoring default config]\n");
+		set_sys_led_state(SYS_LED_FAST);
+		flash_default_config();
+		delay(3UL * SYS_TICK_HZ);
+	}
+}
+
+
+/*
+ * Configure the SerDes of the SoC for a particular mode
+ * to connect to an SFP module or a PHY
+ * Valid modes are SDS_10GR, SDS_QXGMII, SDS_HISGMII, SDS_HSG, SDS_SGMII and SDS_1000BX_FIBER
+ * The SerDes ID may be 0 or 1 for RTL8272 and 0-2 for RTL8373
+ * SDS_QXGMII is used for 10G Fiber, RTL8224 and RTL8261BE
+ */
+void sds_config(uint8_t sds, uint8_t mode) __banked
+{
+	print_string("sds_config sds: "); print_byte(sds); print_string(", mode: "); print_byte(mode); write_char('\n');
+
+	uint8_t page = 0;
+	uint16_t v = 0;
+
+	switch (mode) {
+	case SDS_SGMII:
+	case SDS_1000BX_FIBER:
+		v = 0x0300;
+		page = 0x24;
+		break;
+	case SDS_HISGMII:
+	case SDS_HSG:
+		v = 0x0200;
+		page = 0x28;
+		break;
+	case SDS_10GR:
+	case SDS_QXGMII:
+		v = 0x0200;
+		page = 0x2e;
+		break;
+	case SDS_100FX:
+		v = 0x0200;
+		page = 0x26;
+		break;
+	default:
+		print_string("Error in SDS Mode\n");
+		return;
+	}
+
+	sds_config_mac(sds, mode);
+
+	if (mode == SDS_10GR || mode == SDS_QXGMII)
+		sds_write_v(sds, 0x21, 0x10, 0x4480); // Q002110:6480
+	else
+		sds_write_v(sds, 0x21, 0x10, 0x6480); // Q002110:6480
+
+	sds_write_v(sds, 0x21, 0x13, 0x0400); // Q002113:0400
+	sds_write_v(sds, 0x21, 0x18, 0x6d02); // Q002118:6d02
+	sds_write_v(sds, 0x21, 0x1b, 0x424e); // Q00211b:424e
+	sds_write_v(sds, 0x21, 0x1d, 0x0002); // Q00211d:0002
+	sds_write_v(sds, 0x36, 0x1c, 0x1390); // Q00361c:1390
+	sds_write_v(sds, 0x36, 0x14, 0x003f); // Q003614:003f
+
+	sds_write_v(sds, 0x36, 0x10, v); // Q003610:0200
+
+	if (page == 0x2e) {  // 10G Fiber / SDS_QXGMII
+		sds_write_v(sds, page, 0x04, 0x0080); // Q012e04:0080
+		sds_write_v(sds, page, 0x06, 0x0408); // Q012e06:0408
+		sds_write_v(sds, page, 0x07, 0x020d); // Q012e07:020d
+		sds_write_v(sds, page, 0x09, 0x0601); // Q012e09:0601
+		sds_write_v(sds, page, 0x0b, 0x222c); // Q012e0b:222c
+		sds_write_v(sds, page, 0x0c, 0xa217); // Q012e0c:a217
+		sds_write_v(sds, page, 0x0d, 0xfe40); // Q012e0d:fe40
+		sds_write_v(sds, page, 0x15, 0xf5c1); // Q012e15:f5c1
+	} else {
+		sds_write_v(sds, page, 0x04, 0x0080); // Q002804:0080
+		sds_write_v(sds, page, 0x07, 0x1201); // Q002807:1201
+		sds_write_v(sds, page, 0x09, 0x0601); // Q002809:0601
+		sds_write_v(sds, page, 0x0b, 0x232c); // Q00280b:232c
+		sds_write_v(sds, page, 0x0c, 0x9217); // Q00280c:9217
+		sds_write_v(sds, page, 0x0f, 0x5b50); // Q00280f:5b50
+		sds_write_v(sds, page, 0x15, 0xe7c1); // Q002815:e7f1 BUG !
+	}
+
+	sds_write_v(sds, page, 0x16, 0x0443); // Q002816:0443 / Q012e16:0443
+	sds_write_v(sds, page, 0x1d, 0xabb0); // Q00281d:abb0 / Q012e1d:abb0
+
+	sds_write_v(sds, 0x06, 0x12, 0x5078); // Q000612:5078
+	sds_write_v(sds, 0x07, 0x06, 0x9401); // Q000706:9401
+	sds_write_v(sds, 0x07, 0x08, 0x9401); // Q000708:9401
+	sds_write_v(sds, 0x07, 0x0a, 0x9401); // Q00070a:9401
+	sds_write_v(sds, 0x07, 0x0c, 0x9401); // Q00070c:9401
+	sds_write_v(sds, 0x1f, 0x0b, 0x0003); // Q001f0b:0003
+	sds_write_v(sds, 0x06, 0x03, 0xc45c); // Q000603:c45c
+
+	// RTL8261BE
+	if (machine.n_10g && mode == SDS_QXGMII) {
+		sds_write_v(sds, 0x06, 0x1f, 0x2100); // Q00061f:2100
+		sds_write_v(sds, 0x07, 0x11, 0x054f); // Q000711:054f
+		sds_write_v(sds, 0x20, 0x00, 0x0030); // Q002000:0030
+		sds_write_v(sds, 0x20, 0x00, 0x0010); // Q002000:0010
+		sds_write_v(sds, 0x20, 0x00, 0x0050); // Q002000:0050
+		sds_write_v(sds, 0x20, 0x00, 0x00d0); // Q002000:00d0
+		sds_write_v(sds, 0x20, 0x00, 0x0cd0); // Q002000:0cd0
+		sds_write_v(sds, 0x20, 0x00, 0x04d0); // Q002000:04d0
+		sds_write_v(sds, 0x20, 0x00, 0x04d0); // Q002000:04d0
+		sds_write_v(sds, 0x20, 0x00, 0x0cd0); // Q002000:0cd0
+		sds_write_v(sds, 0x20, 0x00, 0x00d0); // Q002000:00d0
+		sds_write_v(sds, 0x20, 0x00, 0x00d0); // Q002000:00d0
+		sds_write_v(sds, 0x20, 0x00, 0x0050); // Q002000:0050
+		sds_write_v(sds, 0x20, 0x00, 0x0010); // Q002000:0010
+		sds_write_v(sds, 0x20, 0x00, 0x0010); // Q002000:0010
+		sds_write_v(sds, 0x20, 0x00, 0x0030); // Q002000:0030
+		sds_write_v(sds, 0x20, 0x00, 0x0000); // Q002000:0000
+		sds_write_v(sds, 0x1f, 0x00, 0x000b); // Q001f00:000b
+		sds_write_v(sds, 0x1f, 0x00, 0x0000); // Q001f00:0000
+		return;
+	}
+	if (mode != SDS_QXGMII)
+		sds_write_v(sds, 0x06, 0x1f, 0x2100); // Q00061f:2100
+
+	if (mode == SDS_1000BX_FIBER) {
+		sds_write_v(sds, 0x02, 0x04, 0x0020); 	// Q000204:0020
+		sds_write_v(sds, 0x00, 0x02, 0x73d0); 	// Q000002:73d0
+		sds_write_v(sds, 0x00, 0x04, 0x074d); 	// Q000004:074d
+		sds_write_v(sds, 0x20, 0x04, 0x0000); 	// Q002000:0000
+		sds_write_v(sds, 0x1f, 0x00, 0x0000); 	// Q001f00:0000
+	}
+}
+
+
+
+
+void rtl8224_enable(void) __banked
+{
+	// Set Pin 4 low
+	reg_bit_clear(RTL837X_REG_GPIO_32_63_OUTPUT, 4);
+	// Configure Pin as output
+	reg_bit_set(RTL837X_REG_GPIO_32_63_DIRECTION, 4);
+	delay(4);
+	// Set pin 4 high
+	reg_bit_set(RTL837X_REG_GPIO_32_63_OUTPUT, 4);
+	rtl8224_release_tick = ticks;
+}
+
+
+
+void nic_setup(void) __banked
+{
+	// Enable NIC
+	// r6040:00000100 R6040-00001100
+	reg_bit_set(RTL837X_REG_HW_CONF, 0xc);
+
+	// This sets the size of the RX buffer, the filling level is in 0x7874
+	// R7848-000004ff
+	REG_SET(RTL837X_REG_NIC_RXBUFF_RX, RTL837X_NIC_RX_LAST_SLOT);
+
+	// R7844-000007fe
+	REG_SET(RTL837X_REG_NIC_BUFFSIZE_TX, 0x7fe);
+
+	// Configure NIC RX to receive various types of packets
+	// RTL837X_REG_RX_CTRL: Set bits 24-31 to 0x4, clear bits 16/17
+	reg_read_m(RTL837X_REG_RX_CTRL);
+	sfr_mask_data(3, 0xff, 0x04);
+	sfr_mask_data(2, 0x03, 0);
+	reg_write_m(RTL837X_REG_RX_CTRL);
+
+	// Enable NIC TX (set bit 0)
+	reg_bit_set(RTL837X_REG_TX_CTRL, 0);
+
+	// Enable NIC RX (set bit 0)
+	reg_bit_set(RTL837X_REG_RX_CTRL, 0);
+
+	// Drop packets with invalid CRC
+	reg_bit_clear(RTL837X_REG_RX_CTRL, 2);
+
+	// R603c-00000200
+	// CPU-port is CPU-Tag aware (bit 9)
+	REG_SET(RTL837X_REG_CPU_TAG_AWARE_PMASK, 0x200);
+
+	// Insert CPU-tag for internally received packets (bit 0), MODE is 0, i.e. ALL packets (bits 8-9)
+	reg_read_m(RTL837X_REG_CPU_TAG);
+	sfr_mask_data(0, 1, 1);
+	sfr_mask_data(1, 3, 0);
+	reg_write_m(RTL837X_REG_CPU_TAG);
+
+	// Force MAC mode of the CPU port (port 9)
+	// r6368:00000194 R6368-00000197
+	reg_read_m(RTL837X_REG_MAC_FORCE_MODE + 9 * 4);
+	sfr_mask_data(0, 0, 3); // Set bits 0, 1: Force link
+	reg_write_m(RTL837X_REG_MAC_FORCE_MODE+ 9 * 4);
+
+	// Sequence number of TX packets
+	tx_seq = 0;
+	REG_SET(RTL837X_NIC_INT_MSK, NIC_INT_RXIE);
+}
+
+
+
+void rtl8373_revision(void) __banked
+{
+	reg_read_m(RTL837X_REG_CHIP_INFO);
+	sfr_mask_data(2, 0x0a, 0x0a); 	// Enable reading version
+	reg_write_m(RTL837X_REG_CHIP_INFO);
+	delay(50);
+
+	reg_read_m(RTL837X_REG_CHIP_INFO);
+	print_string("CPU revision: "); print_byte(sfr_data[2]); print_byte(sfr_data[2]); write_char('\n');
+	sfr_mask_data(2, 0x0a, 0x00); 	// Enable reading version
+	reg_write_m(RTL837X_REG_CHIP_INFO);
+}
+
+
+/*
+ * The SoC manages Link-State for steering the LEDs and can set PHY-settings
+ * automatically through Realtek's SMI (Simple Managagement) Interface, a
+ * proprietary version of MDIO which for example allows for more PHYs on the same
+ * bus.
+ * Configure polling via SMI and the interface setup during boot.
+ */
+void init_smi(void) __banked
+{
+	print_string("\ninit_switch called\n");
+
+	/* Set the SMI(i.e.I2C) type for PHY polling, 0b01 is 2.5/10G PHY. Disable (0b00) for the SFP-ports
+	 * which are at port 8 and additionally at port 3 for a dual SFP device
+	 */
+
+	// Default: 0x00005555
+	// Workaround for SDCC BUG 4070: SFR_DATA_U32 = 0x00005555;
+	SFR_DATA_U16_UPPER = 0x0000;
+	SFR_DATA_U16 = 0x5555;
+	if (machine.n_10g == 2) {
+		// 0x00015555, only change the bytes that differs from the default.
+		SFR_DATA_16 = 0x01;
+	} else if (machine.n_sfp == 2)
+		// 0x00005515
+		SFR_DATA_0 = 0x15;
+	reg_write(RTL837X_REG_SMI_MAC_TYPE);
+
+	// Configure polling of all PHYs by the MAC to detect link-state changes
+	// Default: 0x000000ff
+	// Workaround for SDCC BUG 4070: SFR_DATA_U32 = 0x000000ff;
+	SFR_DATA_U16_UPPER = 0x0000;
+	SFR_DATA_U16 = 0x00ff;
+	if (!machine_detected.isRTL8373) {
+		if (machine.n_sfp == 2) {
+			// 0x000000f0, only change the bytes that differs from the default.
+			SFR_DATA_0 = 0xf0;
+		} else {
+			// 0x000001f8, only change the bytes that differs from the default.
+			SFR_DATA_8 = 0x01;
+			SFR_DATA_0 = 0xf8;
+		}
+	}
+	reg_write(RTL837X_REG_SMI_PORT_POLLING);
+	// Enable MDC
+	reg_read_m(RTL837X_REG_SMI_CTRL);
+	sfr_mask_data(1, 0, 0x70); 	// Set bits 12-14 to enable MDC for SMI0-SMI2
+	reg_write_m(RTL837X_REG_SMI_CTRL);
+	delay(50);
+
+	if (!machine_detected.isRTL8373) {
+		// Change I2C addresses for SMI of the non-existent PHYs
+		// r6450:000020e6 R6450-000000e6
+		reg_read_m(RTL837X_REG_SMI_PORT6_9_ADDR);
+		sfr_mask_data(1, 0x7c, 0);
+		reg_write_m(RTL837X_REG_SMI_PORT6_9_ADDR);
+
+		// r644c:0a418820 R644c-0a400820
+		reg_read_m(RTL837X_REG_SMI_PORT0_5_ADDR);
+		sfr_mask_data(2, 0x0f, 0);
+		sfr_mask_data(1, 0x80, 0);
+		reg_write_m(RTL837X_REG_SMI_PORT0_5_ADDR);
+	}
+
+	if (machine.n_10g == 2) {
+		// Set address of second external PHY on port 8
+		REG_SET(RTL837X_REG_SMI_PORT6_9_ADDR, 0x000040e6);
+	}
+}
+
+
+
+
+void setup_i2c(void) __banked
+{
+	REG_SET(RTL837X_REG_I2C_MST_IF_CTRL, 0);
+	// Configure SFP EEPROM address (0x50) as I2C device address
+	// Configure SFP readings address (0x51) as I2C device address
+	REG_WRITE(RTL837X_REG_I2C_CTRL, 0x00, 0x1 << (I2C_MEM_ADDR_WIDTH-16),  0x50 >> 5, (0x50 << 3) & 0xff);
+
+	REG_SET(RTL837X_REG_I2C_CTRL2, 0);
+
+	// HW Control register, enable I2C depending on PIN configuration
+	reg_read_m(RTL837X_PIN_MUX_1);
+	for (uint8_t sfp = 0; sfp < machine.n_sfp; sfp++) {
+		uint8_t i2c = machine.sfp_port[sfp].i2c;
+		uint8_t scl_bus = (i2c >> RTL837X_REG_I2C_SCL_SHIFT) & RTL837X_REG_I2C_SCL_MASK;
+		uint8_t sda_bus = (i2c >> RTL837X_REG_I2C_SDA_SHIFT) & RTL837X_REG_I2C_SDA_MASK;
+		print_string("Configuring I2C for SFP idx="); print_byte(sfp); print_string(" SCL="); print_byte(scl_bus); print_string(", SDA="); print_byte(sda_bus); write_char('\n');
+		// `& RTL837X_REG_I2C_SCL_MASK` is needed to silens a compiler warning 110.
+		switch (scl_bus & RTL837X_REG_I2C_SCL_MASK) {
+			case 3:
+				// Bit 5-6 0b10 -> SCL (implies enabled SDA on bus 3)
+				sfr_mask_data(0, 0x60, 0x40);
+				break;
+			case 2: 
+				// Bit 15-16 0b01 -> SCL
+				sfr_mask_data(1, 0x80, 0x80);
+				sfr_mask_data(2, 0x01, 0x00);
+				break;
+			case 1:
+				// Bit 11-12 0b01 -> SCL
+				sfr_mask_data(1, 0x18, 0x08);
+				break;
+			case 0:
+				// Bit 7-8 0b01 -> SCL
+				sfr_mask_data(0, 0x80, 0x80);
+				sfr_mask_data(1, 0x01, 0x00);
+				break;
+		}
+
+		switch (sda_bus) {
+			case 4:
+				// Bit 29 0b0 -> SDA
+				sfr_mask_data(3, 0x20, 0x00);
+				break;
+			case 3:
+				// Bit 5-6 0b10 -> SDA (implies enabled SCL on bus 3)
+				sfr_mask_data(0, 0x60, 0x40);
+				break;
+			case 2:
+				// Bit 17-18 0b01 -> SDA
+				sfr_mask_data(2, 0x06, 0x02);
+				break;
+			case 1:
+				// Bit 13-14 0b01 -> SDA
+				sfr_mask_data(1, 0x60, 0x20);
+				break;
+			case 0:
+				// Bit 9-10 0b01 -> SDA
+				sfr_mask_data(1, 0x06, 0x02);
+				break;
+		}
+	}
+	reg_write_m(RTL837X_PIN_MUX_1);	
+}
+
+
+/* Give the switch a name carrying the tail of its MAC, so several of them on
+ * one network are distinguishable out of the box. Called after the startup
+ * config has been replayed and returns at once if that config already set a
+ * name, so a configured switch does no work for it (suggested in review).
+ *
+ * Written without a loop on purpose. Locals - counters and pointers alike -
+ * land in the 8051's internal-RAM overlay, and on an image with LACP and STP
+ * both enabled that overlay is exhausted: a loop here makes the linker fail
+ * with "Could not get 8 consecutive bytes in internal RAM for area OSEG".
+ * Moving the code into its own function does not help; the overlay is shared
+ * across the whole image. Hoisting the locals to xdata does not help either,
+ * because itohex() is inline and brings its own frame. */
+void set_hostname_default(void) __banked
+{
+	if (hostname[0] != NUL)
+		return;
+
+	strcpy((__xdata uint8_t *)hostname, "RTLPlayground-");
+	hostname[14] = hex[uip_ethaddr.addr[3] >> 4];
+	hostname[15] = hex[uip_ethaddr.addr[3] & 0xf];
+	hostname[16] = hex[uip_ethaddr.addr[4] >> 4];
+	hostname[17] = hex[uip_ethaddr.addr[4] & 0xf];
+	hostname[18] = hex[uip_ethaddr.addr[5] >> 4];
+	hostname[19] = hex[uip_ethaddr.addr[5] & 0xf];
+	hostname[20] = NUL;
+}

boot.h

@@ -0,0 +1,14 @@
+#ifndef BOOT_H
+#define BOOT_H
+
+void sds_config_mac(uint8_t sds, uint8_t mode) __banked;
+void sds_config(uint8_t sds, uint8_t mode) __banked;
+void early_boot_handle_button(void) __banked;
+void rtl8224_enable(void) __banked;
+void nic_setup(void) __banked;
+void rtl8373_revision(void) __banked;
+void init_smi(void) __banked;
+void setup_i2c(void) __banked;
+void set_hostname_default(void) __banked;
+
+#endif

cmd_parser.h

@@ -6,14 +6,42 @@
 #include "rtl837x_common.h"
 
 extern __xdata uint8_t cmd_buffer[CMD_BUF_SIZE];
-extern __xdata uint8_t cmd_available;
+extern __bit __at(0x03) cmd_available;
 extern __xdata uint8_t err_status;
 
+/* Health instrumentation, built with HEALTH=1. Without it the hooks below
+ * compile to nothing and the "health" command does not exist. */
+#ifdef HEALTH
+/* Phase ids, in main-loop order */
+#define HEALTH_PH_LINK	0
+#define HEALTH_PH_SFP	1
+#define HEALTH_PH_RX	2
+#define HEALTH_PH_TX	3
+#define HEALTH_PH_STP	4
+#define HEALTH_PH_CMD	5
+#define HEALTH_PH_LLDP	6
+#define HEALTH_PHASES	(HEALTH_PH_LLDP + 1)	/* last id + 1: sizes the arrays in cmd_parser.c */
+
+extern __xdata uint16_t health_rx_frames;
+#define health_rx_frame()	health_rx_frames++
+
+void health_loop_start(void) __banked;	/* once per idle() pass, resets the phase baseline */
+void health_phase(uint8_t id) __banked;	/* after each phase; accounts ticks since last checkpoint */
+void health_stack_paint(void) __banked;	/* paint the unused stack area once at boot */
+void health_show(void) __banked;	/* "health" CLI dump */
+#else
+#define health_rx_frame()
+#define health_loop_start()
+#define health_phase(id)
+#define health_stack_paint()
+#endif
+
 void cmd_tokenize(void) __banked;
 void cmd_parser(void) __banked;
 void execute_config(void) __banked;
 void execute_commands(__xdata uint8_t *p) __banked;
-void print_ip(__xdata uint8_t *ptr);
+void print_ip(__xdata uint8_t *ptr) __banked;
+void print_mac(__xdata uint8_t *ptr) __banked;
 void print_sw_version(void) __banked;
 void clear_command_history(void) __banked;
 

dhcp.c

@@ -50,8 +50,8 @@ __xdata uip_ipaddr_t server;
 #define DHCP_PARAM_DNS		6
 #define DHCP_END		255
 
-#pragma codeseg BANK2
-#pragma constseg BANK2
+#pragma codeseg BANK3
+#pragma constseg BANK3
 
 struct dhcp_pkt {
 	uint8_t type;
@@ -139,7 +139,6 @@ void dhcp_addopt_request_ip(void)
 	DHCP_OPT[dhcp_state.opt_ptr++] = dhcp_state.current_ip[1];
 	DHCP_OPT[dhcp_state.opt_ptr++] = dhcp_state.current_ip[2];
 	DHCP_OPT[dhcp_state.opt_ptr++] = dhcp_state.current_ip[3];
-	memcpy(&DHCP_OPT[dhcp_state.opt_ptr], uip_ethaddr.addr, 4);
 }
 
 
@@ -151,13 +150,11 @@ void dhcp_addopt_server_id(void)
 	DHCP_OPT[dhcp_state.opt_ptr++] = dhcp_state.server[1];
 	DHCP_OPT[dhcp_state.opt_ptr++] = dhcp_state.server[2];
 	DHCP_OPT[dhcp_state.opt_ptr++] = dhcp_state.server[3];
-	memcpy(&DHCP_OPT[dhcp_state.opt_ptr], uip_ethaddr.addr, 4);
 }
 
 
 void dhcp_send_discover(void)
 {
-	print_string("dhcp_send_discover called\n");
 	dhcp_prepare_request();
 
 	dhcp_state.opt_ptr = 0;
@@ -193,7 +190,6 @@ void dhcp_send_discover(void)
 
 void dhcp_send_request(void)
 {
-	print_string("dhcp_send_request called\n");
 	dhcp_prepare_request();
 
 	dhcp_state.opt_ptr = 0;
@@ -232,19 +228,21 @@ void ip_opt(__xdata uint8_t * ip)
 {
 	dhcp_state.opt_ptr++;
 	uint8_t len = DHCP_OPT[dhcp_state.opt_ptr++];
-	*ip++ = DHCP_OPT[dhcp_state.opt_ptr++];
-	*ip++ = DHCP_OPT[dhcp_state.opt_ptr++];
-	*ip++ = DHCP_OPT[dhcp_state.opt_ptr++];
-	*ip++ = DHCP_OPT[dhcp_state.opt_ptr++];
+	if (len >= 4)
+		memcpy(ip, &DHCP_OPT[dhcp_state.opt_ptr], 4);
 	 // There may be more than one IP option, such as 2 DNS servers advertised
-	dhcp_state.opt_ptr += len - 4;
+	dhcp_state.opt_ptr += len;
 }
 
 
-void long_opt(void)
+uint8_t long_opt(void)
 {
 	dhcp_state.opt_ptr++;
-	dhcp_state.opt_ptr++;
+	uint8_t len = DHCP_OPT[dhcp_state.opt_ptr++];
+	if (len < 4) {
+		dhcp_state.opt_ptr += len;
+		return 0;
+	}
 	long_value =  DHCP_OPT[dhcp_state.opt_ptr++];
 	long_value <<= 8;
 	long_value |= DHCP_OPT[dhcp_state.opt_ptr++];
@@ -252,6 +250,8 @@ void long_opt(void)
 	long_value |= DHCP_OPT[dhcp_state.opt_ptr++];
 	long_value <<= 8;
 	long_value |= DHCP_OPT[dhcp_state.opt_ptr++];
+	dhcp_state.opt_ptr += len - 4;
+	return 1;
 }
 
 
@@ -275,16 +275,16 @@ void parse_opts(void)
 			ip_opt(&dhcp_state.broadcast[0]);
 			break;
 		case DHCP_LEASE:
-			long_opt();
-			dhcp_state.lease = long_value;
+			if (long_opt())
+				dhcp_state.lease = long_value;
 			break;
 		case DHCP_REBIND:
-			long_opt();
-			dhcp_state.rebind = long_value;
+			if (long_opt())
+				dhcp_state.rebind = long_value;
 			break;
 		case DHCP_RENEWAL:
-			long_opt();
-			dhcp_state.renewal = long_value;
+			if (long_opt())
+				dhcp_state.renewal = long_value;
 			break;
 		case DHCP_END:
 			break;
@@ -376,13 +376,11 @@ void dhcp_start(void) __banked
 	*tid++ = SFR_DATA_8;
 	*tid = SFR_DATA_0;
 	dhcp_state.state = DHCP_START;
-	print_string("dhcp_start done\n");
 }
 
 
 void dhcp_stop(void) __banked
 {
-	print_string("dhcp_stop called\n");
 	uip_udp_remove(dhcp_state.conn);
 	dhcp_state.state = DHCP_OFF;
 }

dns.c

@@ -0,0 +1,238 @@
+/*
+ * DNS resolver for the RTL837x platform
+ * This code is in the Public Domain
+ */
+
+#include <stdint.h>
+#include "rtl837x_common.h"
+#include "cmd_parser.h"
+#include "uip/uip.h"
+#include "dhcp.h"
+#include "dns.h"
+
+#pragma codeseg BANK3
+#pragma constseg BANK3
+
+#define DNS_PORT	53
+#define DNS_TRIES	6
+#define DNS_TIMEOUT	(2 * SYS_TICK_HZ)
+
+extern __xdata struct dhcp_state dhcp_state;
+extern volatile __xdata uint32_t ticks;
+
+__xdata struct dns_state dns_state;
+__xdata uip_ipaddr_t dns_ip;
+uint8_t * __xdata dns_cand[3];
+
+
+static uint8_t dns_used(__xdata uint8_t *a) __reentrant
+{
+	return a[0] | a[1] | a[2] | a[3];
+}
+
+
+static uint8_t dns_name_is_ip(void) __reentrant
+{
+	__xdata char *s = dns_state.name;
+	uint8_t i, n;
+	uint16_t v;
+
+	for (i = 0; i < 4; i++) {
+		v = 0;
+		n = 0;
+		while (*s >= '0' && *s <= '9') {
+			v = v * 10 + (*s++ - '0');
+			if (++n > 3 || v > 255)
+				return 0;
+		}
+		if (!n)
+			return 0;
+		dns_state.addr[i] = v;
+		if (i < 3 && *s++ != '.')
+			return 0;
+	}
+	return !*s;
+}
+
+
+static void dns_next(void) __reentrant
+{
+	uint8_t k = 0;
+
+	if (dns_state.conn) {
+		uip_udp_remove(dns_state.conn);
+		dns_state.conn = 0;
+	}
+	if (dns_used(dns_state.server[0]))
+		dns_cand[k++] = dns_state.server[0];
+	if (dns_used(dns_state.server[1]))
+		dns_cand[k++] = dns_state.server[1];
+	if (dns_used(dhcp_state.dns))
+		dns_cand[k++] = dhcp_state.dns;
+	if (!k || dns_state.tries >= DNS_TRIES) {
+		dns_state.status = DNS_FAILED;
+		if (dns_state.verbose)
+			print_string(k ? "DNS: no answer\n" : "DNS: no server\n");
+		return;
+	}
+	k = dns_state.tries % k;
+	uip_ipaddr(dns_ip, dns_cand[k][0], dns_cand[k][1], dns_cand[k][2], dns_cand[k][3]);
+	dns_state.conn = uip_udp_new(&dns_ip, HTONS(DNS_PORT));
+	if (!dns_state.conn) {
+		dns_state.status = DNS_FAILED;
+		print_string("DNS: no free UDP connection\n");
+		return;
+	}
+	dns_state.send = 1;
+}
+
+
+static void dns_query(void) __reentrant
+{
+	__xdata uint8_t *p = (__xdata uint8_t *)uip_appdata;
+	__xdata char *n = dns_state.name;
+	__xdata uint8_t *label;
+	uint16_t len = 12;
+
+	dns_state.id++;
+	p[0] = dns_state.id >> 8;
+	p[1] = dns_state.id;
+	p[2] = 0x01;
+	p[3] = 0x00;
+	p[4] = 0;
+	p[5] = 1;
+	p[6] = p[7] = p[8] = p[9] = p[10] = p[11] = 0;
+	while (*n) {
+		label = p + len++;
+		*label = 0;
+		while (*n && *n != '.') {
+			p[len++] = *n++;
+			(*label)++;
+		}
+		if (*n)
+			n++;
+	}
+	p[len++] = 0;
+	p[len++] = 0;
+	p[len++] = 1;
+	p[len++] = 0;
+	p[len++] = 1;
+	uip_udp_send(len);
+}
+
+
+static uint16_t dns_skip_name(__xdata uint8_t *p, uint16_t i, uint16_t len) __reentrant
+{
+	while (i < len) {
+		if (!p[i])
+			return i + 1;
+		if ((p[i] & 0xc0) == 0xc0)
+			return i + 2;
+		i += p[i] + 1;
+	}
+	return len;
+}
+
+
+static void dns_answer(void) __reentrant
+{
+	__xdata uint8_t *p = (__xdata uint8_t *)uip_appdata;
+	uint16_t len = uip_datalen();
+	uint16_t i, an;
+
+	if (dns_state.status != DNS_PENDING || len < 12 || !(p[2] & 0x80)
+	    || p[0] != (uint8_t)(dns_state.id >> 8) || p[1] != (uint8_t)dns_state.id)
+		return;
+	an = ((uint16_t)p[6] << 8) | p[7];
+	i = dns_skip_name(p, 12, len) + 4;
+	while (!(p[3] & 0x0f) && an-- && i + 10 <= len) {
+		i = dns_skip_name(p, i, len);
+		if (i + 10 > len)
+			break;
+		if (!p[i] && p[i + 1] == 1 && !p[i + 8] && p[i + 9] == 4 && i + 14 <= len) {
+			memcpy(dns_state.addr, p + i + 10, 4);
+			dns_state.status = DNS_DONE;
+			uip_udp_remove(dns_state.conn);
+			dns_state.conn = 0;
+			if (dns_state.verbose) {
+				print_string("DNS: ");
+				print_string_x(dns_state.name);
+				print_string(" is ");
+				print_ip(dns_state.addr);
+				write_char('\n');
+			}
+			return;
+		}
+		i += 10 + (((uint16_t)p[i + 8] << 8) | p[i + 9]);
+	}
+	dns_state.tries = DNS_TRIES;
+	dns_next();
+}
+
+
+void dns_init(void) __banked
+{
+	memset(dns_state.server, 0, sizeof(dns_state.server));
+	dns_state.name[0] = 0;
+	dns_state.status = DNS_IDLE;
+	dns_state.conn = 0;
+}
+
+
+void dns_lookup(void) __banked
+{
+	if (dns_name_is_ip()) {
+		dns_state.status = DNS_DONE;
+		return;
+	}
+	dns_state.tries = 0;
+	dns_state.status = DNS_PENDING;
+	dns_next();
+}
+
+
+void dns_show(void) __banked
+{
+	print_string("DNS servers ");
+	print_ip(dns_state.server[0]);
+	write_char(' ');
+	print_ip(dns_state.server[1]);
+	print_string(", from DHCP ");
+	print_ip(dhcp_state.dns);
+	write_char('\n');
+	if (!dns_state.name[0])
+		return;
+	print_string("Last lookup ");
+	print_string_x(dns_state.name);
+	print_string(": ");
+	if (dns_state.status == DNS_DONE)
+		print_ip(dns_state.addr);
+	else if (dns_state.status == DNS_PENDING)
+		print_string("pending");
+	else
+		print_string("failed");
+	write_char('\n');
+}
+
+
+void dns_callback(uint16_t lport) __banked __reentrant
+{
+	if (!dns_state.conn || lport != dns_state.conn->lport)
+		return;
+	if (uip_newdata()) {
+		dns_answer();
+		return;
+	}
+	if (dns_state.status != DNS_PENDING)
+		return;
+	if (dns_state.send) {
+		dns_state.send = 0;
+		dns_state.sent = ticks;
+		dns_query();
+		return;
+	}
+	if (ticks - dns_state.sent >= DNS_TIMEOUT) {
+		dns_state.tries++;
+		dns_next();
+	}
+}

dns.h

@@ -0,0 +1,36 @@
+#ifndef _DNS_H_
+#define _DNS_H_
+
+#include <stdint.h>
+
+#define DNS_NAME_LEN	48
+
+#define DNS_IDLE	0
+#define DNS_PENDING	1
+#define DNS_DONE	2
+#define DNS_FAILED	3
+
+struct uip_udp_conn;
+
+struct dns_state {
+	uint8_t server[2][4];	/* configured servers, 0.0.0.0 = unused */
+	char name[DNS_NAME_LEN];	/* name of the current or last lookup */
+	uint8_t addr[4];	/* address found by the last lookup */
+	uint8_t status;		/* DNS_* */
+	uint8_t verbose;	/* print the result of the lookup */
+	uint8_t tries;		/* queries sent for the current lookup */
+	uint8_t send;		/* a query is due on the connection */
+	uint16_t id;
+	uint32_t sent;		/* ticks when the last query went out */
+	struct uip_udp_conn *conn;
+};
+
+extern __xdata struct dns_state dns_state;
+
+void dns_init(void) __banked;
+void dns_lookup(void) __banked;
+void dns_show(void) __banked;
+void dns_callback(uint16_t lport) __banked __reentrant;
+void dns_parse(void) __banked __reentrant;
+
+#endif

dns_cli.c

@@ -0,0 +1,63 @@
+/*
+ * Command line interface of the DNS resolver for the RTL837x platform
+ * This code is in the Public Domain
+ */
+
+#include <stdint.h>
+#include "rtl837x_common.h"
+#include "cmd_parser.h"
+#include "dns.h"
+
+#pragma codeseg BANK2
+#pragma constseg BANK2
+
+extern __xdata uint8_t cmd_buffer[CMD_BUF_SIZE];
+extern __xdata uint8_t cmd_words_len;
+extern __xdata uint8_t cmd_words_b[15];
+extern __xdata uint8_t ip[4];
+uint8_t cmd_compare(uint8_t start, __code const uint8_t * cmd);
+uint8_t parse_ip(uint8_t idx);
+
+
+void dns_parse(void) __banked __reentrant
+{
+	__xdata uint8_t *s;
+	uint8_t n;
+
+	if (cmd_words_len < 2) {
+		dns_show();
+		return;
+	}
+	if (cmd_compare(1, "server") && cmd_words_len >= 3 && parse_ip(cmd_words_b[2])) {
+		memcpy(dns_state.server[0], ip, 4);
+		memset(dns_state.server[1], 0, 4);
+		if (cmd_words_len >= 4) {
+			if (!parse_ip(cmd_words_b[3]))
+				goto err;
+			memcpy(dns_state.server[1], ip, 4);
+		}
+		return;
+	}
+	if (cmd_compare(1, "lookup") && cmd_words_len >= 3) {
+		if (dns_state.status == DNS_PENDING) {
+			print_string("DNS: a lookup is already running\n");
+			return;
+		}
+		s = &cmd_buffer[cmd_words_b[2]];
+		for (n = 0; s[n] && s[n] != ' '; n++) {
+			if (n == DNS_NAME_LEN - 1)
+				goto err;
+			dns_state.name[n] = s[n];
+		}
+		dns_state.name[n] = 0;
+		dns_state.verbose = 1;
+		dns_lookup();
+		if (dns_state.status == DNS_DONE) {
+			print_ip(dns_state.addr);
+			write_char('\n');
+		}
+		return;
+	}
+err:
+	print_string("Error: dns [server <ip> [<ip>]|lookup <name>]\n");
+}

doc/CpuPort.md

@@ -34,6 +34,21 @@ Next, transfer the actual packet over by repeating the SFR function with a
 pointer to the frame on the ASIC directly after the frame header and a
 length as given by the length in the frame header + 7, again divided by 8.
 
+The RX ring's inclusive last slot is configured as `0x4ff`, giving a byte
+range of `0x0000` through `0x27ff`. If the descriptor occupies the last slot
+at `0x27f8`, its payload starts at `0x0000`, not `0x2800`. The RX copy code
+normalizes that starting address before DMA. It relies on the hardware to
+wrap transfers that start inside the ring and cross its end, as observed
+on SWTGW218AS. Programming an initial address of `0x2800` does not wrap
+and can instead read zeros or stale TX data outside the RX ring.
+
+Frame lengths are checked against `UIP_CONF_BUFFER_SIZE` before copying.
+DMA works in whole eight-byte units. `UIP_CONF_BUFFER_ALLOC_SIZE` rounds
+the backing storage up to the next complete word: 1,560 bytes for the
+1,556-byte logical frame limit. This padding allows a maximum-size frame
+to be transferred directly without a separate tail copy. The logical
+frame limit and advertised MSS are unchanged.
+
 The received frame will have an RTL proprietary Ethernet frame type of
 0x8899 (RRPC) where normally the frame type 0x0800 for IPv4 would be located.
 Further 6 bytes follow describing the frame, before the normal IPv4 data

doc/adding-a-device.md

@@ -0,0 +1,220 @@
+# Adding a new device
+
+A device is supported by adding one board header `.h` and board definition `.c` file
+under `machine/`. 
+the build selects the definition by name, and CI compiles every file it finds.
+
+To compile for your added device, do:
+```
+make MACHINE=<NAME>
+```
+
+## What a new device consists of
+
+| File | Purpose |
+|------|---------|
+| `machine/<NAME>.c` | the `struct machine machine` board definition |
+| `machine/<NAME>.h` | `MACHINE_MODEL` (the model string shown in the UI and the CLI) and `MACHINE_CHIPSET` (`RTL8372` or `RTL8373`) |
+| `machine/<NAME>_init.c` | optional `machine_custom_init()` boot hook, for registers the machine definition cannot express |
+| `doc/devices/<name>.md` + `doc/supported_devices.md` | (only if the board is verified and you want to support it publicly) user-facing documentation |
+
+`<NAME>` is the machine symbol without its `MACHINE_` prefix, so `MACHINE` is
+passed exactly as spelled on the command line: `make MACHINE=MY_BOARD_1_0`
+needs `machine/MY_BOARD_1_0.c` and `machine/MY_BOARD_1_0.h`.
+
+## Step 1: the board header
+
+`machine/<NAME>.h` carries the product string and the SoC family. Use the
+existing files as the template:
+
+```c
+#ifndef _MACHINE_MY_BOARD_1_0_H_
+#define _MACHINE_MY_BOARD_1_0_H_
+
+#define MACHINE_MODEL "Brand Model V1.0"
+#define MACHINE_CHIPSET RTL8373
+
+#endif
+```
+
+`MACHINE_CHIPSET` is `RTL8373` or `RTL8372`; a build without it will fail. It selects the default
+logical port range: `MACHINE_MIN_PORT` is `0` for RTL8373 and `3` for RTL8372, while
+`MACHINE_MAX_PORT` is `8`, the highest logical port index. A board with a different range defines
+`MACHINE_MIN_PORT` and/or `MACHINE_MAX_PORT` in its header to override the
+defaults.
+
+The device product string is what the CLI or the Web-Interface will print to identify the device; `machine.h` includes it by name.
+
+> [!NOTE]
+> `machine_name` is a `char[30]` and `MACHINE_MODEL` is copied into it at
+> compile time. A longer model string still builds, but SDCC emits
+> `warning 147: excess elements in array of chars initializer` and the name is
+> truncated in the UI. Keep it to 29 characters or less.
+
+## Step 2: the board definition
+
+`machine/<NAME>.c` starts with the shared prelude and then defines the single
+`machine` object. Copy the closest existing board and adjust it — do not start
+from a blank file, the field meanings are easier to see in a real example.
+The best choice is usually a device that has the same number and type of Ethernet
+ports and the same number of SFP+ cages.
+
+```c
+#include "machine_defs.h"
+
+__code const struct machine machine = {
+	MACHINE_COMMON_SETTING,
+	.n_sfp = 1,
+	.log_to_phys_port = {1, 2, 3, 4, 5, 6, 7, 8, 9},
+	.phys_to_log_port = {0, 1, 2, 3, 4, 5, 6, 7, 8},
+	.is_sfp = {0, 0, 0, 0, 0, 0, 0, 0, 1},
+
+	.sfp_port[0].pin_detect = GPIO30_ACL_BIT3_EN,
+	.sfp_port[0].pin_los = GPIO37,
+	.sfp_port[0].pin_tx_disable = GPIO_NA,
+	.sfp_port[0].sds = 1,
+	.sfp_port[0].i2c = I2CBUS( GPIO39_I2C_SDA4, GPIO40_I2C_SCL3_MDC1 ),
+	.reset_pin = GPIO_NA,
+
+	.high_leds = { .mux = LED_27 | LED_29, .enable = LED_28_SYS | LED_29 },
+	.port_led_set = { 0, 0, 0, 0, 0, 0, 0, 0, 1 },
+	.led_sets = {
+		{   /* RJ45 ports */
+			LEDS_2G5 | LEDS_LINK,
+			LEDS_2G5 | LEDS_1G | LEDS_100M | LEDS_10M | LEDS_LINK | LEDS_ACT,
+			0,
+			0
+		}, { /* SFP port, single green LED */
+			LEDS_10G | LEDS_2G5 | LEDS_1G | LEDS_100M | LEDS_10M | LEDS_LINK | LEDS_ACT,
+			0,
+			0,
+			0
+		}
+	},
+};
+```
+
+`MACHINE_COMMON_SETTING` (from `machine_defs.h`) initializes `machine_name` and
+`isRTL8373` from the header's `MACHINE_MODEL` and `MACHINE_CHIPSET`. The logical
+port range comes from `MACHINE_MIN_PORT` and `MACHINE_MAX_PORT` in `machine.h`.
+Keep it as the first initializer and list the board-specific fields after it.
+
+### Fields
+
+| Field | Meaning |
+|-------|---------|
+| `MACHINE_COMMON_SETTING` | expands to `machine_name` (`MACHINE_MODEL`) and `isRTL8373` (`1` for `RTL8373`, `0` for `RTL8372`); both are derived from the header's `MACHINE_CHIPSET` (see Step 1) |
+| `n_sfp` | number of SFP+ cages |
+| `n_10g` | SerDes 10G capability used by the RTL837xN init: `0` none, `1` one, `2` two 10G-capable SerDes |
+| `log_to_phys_port[9]` | logical port (0-based index) → physical front-panel port number |
+| `phys_to_log_port[9]` | the inverse, physical port 1 is index 0, unused slots are 0 |
+| `is_sfp[9]` | `0` for copper, `1` for left slot and `2` for right slot |
+| `sfp_port[2]` | per-cage wiring, `sfp_port[0]` is the leftmost cage |
+| `reset_pin` | GPIO of the reset button, `GPIO_NA` if unpopulated |
+| `high_leds` | mux/enable for the LED pins 27–29 (`LED_27`, `LED_28_SYS`, `LED_29`) |
+| `port_led_set[9]` | which LED set (0–3) each logical port uses |
+| `led_sets[4][4]` | Each port can have up to 4 GPIO output pins assigned. 4 sets of LEDs with 4 LED-conditions per set; the conditions define which GPIO pins become active for a port given a port link-state. Which LEDs are then enabled depends on the wiring of the pins to up to 3 LEDs for that port, which can make a logical combination of the 4 LED output-pins. |
+| `led_mux_custom` / `led_mux[28]` | `1` plus the 28 mux values when the stock LED muxing does not fit the board |
+| `mac_flash_offset` | flash offset of the factory MAC address, omit or `0` if the board does not store one |
+
+Each `struct sfp_port` entry holds `pin_detect`, `pin_los`, `pin_tx_disable`,
+`sds` (which SerDes drives the cage) and `i2c` (built with `I2CBUS(sda, scl)`,
+or `GPIO_NA`-based values where a cage has no such pin).
+
+`I2CBUS(...)` only accepts the pin pairs listed in `machine_defs.h`. A wrong
+pair is caught at compile time: SDCC reports
+`error 176: sizeof applied to an incomplete type` pointing at the offending
+`.i2c` line, because the macro deliberately turns an invalid bus number into an
+array of size 0.
+
+### Reusing an existing definition
+
+If the new board is electrically identical to one that already exists — same
+port mapping, same SFP wiring, same LEDs — do not copy the file. Make the new
+`.c` a symlink and keep only a distinct `.h`, so the two boards cannot drift
+apart:
+
+```
+ln -s KP_9000_6XHML_X2.c machine/MY_BOARD_1_0.c
+```
+
+The new board still needs `machine/MY_BOARD_1_0.h` (pick the version number on
+the PCB) with its own `MACHINE_MODEL` and `MACHINE_CHIPSET`: `MACHINE_MODEL` is
+what distinguishes the products, and `MACHINE_CHIPSET` must match the SoC of
+the board definition being reused.
+Existing examples: `machine/PCB_K0402WS_V3.c`, `machine/KP_9000_9XHML_X_V3_2.c`
+and the six `KP_9000_*` revision variants.
+
+Two files that differ only in whitespace or comments are duplicates and should
+be collapsed this way.
+
+### Optional custom init hook
+
+If the board needs custom initialization, add
+`machine/<NAME>_init.c` defining `machine_custom_init()`. The Makefile picks the
+file up by name and compiles it.
+
+Boards that need the same hook share one file: keep the code in a single
+`machine/<NAME>_init.c` whose `machine/<NAME>.c` is a real file (not a symlink),
+and make the other boards' `_init.c` symlinks to it, for example:
+
+```
+ln -s KP_9000_6XH_X2_init.c machine/MY_BOARD_1_0_init.c
+```
+
+## Step 3: build and verify
+
+```
+make MACHINE=MY_BOARD_1_0
+```
+
+The generated flash image appears as `output/rtlplayground-<version>-MY_BOARD_1_0.bin`. An
+unknown or empty `MACHINE` fails immediately with a message naming the expected
+file, before anything is compiled.
+
+Then run the checks CI runs:
+
+```
+make machine_check    # compiles machine/<NAME>.c and its optional _init.c for every machine
+make -C test          # host-side unit tests
+```
+
+`make machine_check` discovers the definitions with
+`$(wildcard machine/*.c)`, so a new file is covered without registering it
+anywhere. CI (`.github/workflows/build.yml`) runs it on every push and pull
+request, alongside full builds for a 4+2 and an 8+1 board.
+
+A successful compile only proves the definition is syntactically valid. The
+actual behaviour — port numbering, SFP detection, LEDs — has to be confirmed on
+hardware, and the console commands (`port`, `stat`, `sfp`, `leds`) are the
+tools for that.
+
+## Step 4: documentation
+
+Once the board has been tested on hardware:
+
+1. Add a row to the table in [Supported devices](supported_devices.md).
+2. Add `doc/devices/<name>.md` describing the board, the build target
+   (`make MACHINE=MY_BOARD_1_0`), what works and what is unverified, plus
+   photos and the PCB silkscreen marking.
+3. If the board shares a PCB with other revisions, follow
+   `doc/devices/KP-9000-6XH-X2.md` and document which `MACHINE` to pick for
+   which silkscreen.
+
+### Identifying the CPU variant
+
+For RTL8373-based devices, use the read-only console command below to distinguish
+the N and non-N CPU variants:
+
+```text
+regget 4
+```
+
+| Register 0x0004 value | CPU |
+|---|---|
+| `0x83730000` | RTL8373 (non-N) |
+| `0x83737000` | RTL8373N |
+
+Record the identified CPU in the device documentation. These values identify
+only the CPU variant; document PHY and flash-chip models separately when verified.
+

doc/devices/2G5F4_10G2_V1.01.md

@@ -0,0 +1,178 @@
+hongyavision HYWS-SGT0204S
+
+RTL8372-based 4×2.5G + 2×SFP+ unmanaged switch.
+
+Also marked on the PCB as `2G5F4_10G2_V1.01`.
+
+Stock firmware runs a factory test console on UART at 9600 baud 8N1 (banner
+`RTL8371B start`, prompt `RTL8371B:`). There is no management CLI and no Web
+interface.
+
+Using SPI clamp in-board is the only method for initial installation and update.
+Erase the whole chip and verify the write (`flashrom -v`) before powering the
+board, so an incomplete write is caught before the board is powered up.
+
+Stock flash chip is 512 KB (Bright Moon `T25S40`, 4 Mbit, SOP-8), which the
+RTLPlayground image fills exactly.
+
+### Label specifications
+
+Case, top / front / nameplate:
+
+<img src="photos/2G5F4_10G2_V1.01/assembled-top.jpg" width="300" />
+<img src="photos/2G5F4_10G2_V1.01/assembled-front.jpg" width="300" />
+<img src="photos/2G5F4_10G2_V1.01/assembled-bottom.jpg" width="300" />
+
+- **Manufacturer**: hongyavision (Shenzhen Hongyavision Electronic Technology
+  Co., Ltd.)
+- **Model**: HYWS-SGT0204S — the nameplate is printed in Chinese and reads
+  "10/100/1000/2500M fibre switch" and "4x 2.5G copper port + 2x 10G SFP+
+  optical port"
+- **Ports**:
+  - 4 × RJ45: 10/100/1000/2500 Mbps
+  - 2 × SFP+: 1G / 2.5G / 10G
+- **Power**: 12 V DC, 2 A
+- **Serial number**: `S/N-YYYYMMDDNNNN`, e.g. `S/N-202312010028`
+
+The top of the case only carries generic printing ("2.5G+10G High Speed Fiber
+Transceiver / 4-Port 2.5G Ethernet Switch+10G Fiber"), so the nameplate on the
+bottom is what identifies the device.
+
+### What works
+
+- All four 2.5GBASE-T RJ45 ports at 10/100/1000/2500 Mbps
+- Both SFP+ ports supporting 1G, 2.5G and 10G modules
+- Copper LEDs: white (2.5G) and blue (link/activity)
+- SFP LEDs: one single LED per port (link/activity)
+- SFP EEPROM and DDM readable on both cages, verified with a 10G module
+  (`Rate: 0x67`, module identity and DDM values via the `sfp` console command)
+- Serial console at 115200 8N1 (RTLPlayground) / 9600 8N1 (stock)
+
+### Not verified
+
+- The reset button footprint is unpopulated and the case has no opening for it,
+  so it could not be tested.
+
+### PCB overview
+
+**Board markings**
+- Top silkscreen: `2G5F4_10G2_V1.01`
+
+**Key components**
+- Main switch: RTL8372 (RTL8371B family), reported as `RTL8372` (non-N) by
+  RTLPlayground. The chip sits under a heatsink and its package marking is not
+  readable.
+- U1: SPI NOR flash, Bright Moon `T25S40`, 4 Mbit / 512 KB, SOP-8
+- A 3-hole reset button footprint is present but unpopulated, and the case has
+  no opening for it, so there is no usable reset button
+
+Top side:
+
+<img src="photos/2G5F4_10G2_V1.01/pcb_top.jpg" width="600" />
+
+Bottom:
+
+<img src="photos/2G5F4_10G2_V1.01/pcb_bottom.jpg" width="600" />
+
+The stock flash chip, Bright Moon `T25S40` (SOP-8, 4 Mbit / 512 KB):
+
+<img src="photos/2G5F4_10G2_V1.01/flash_t25s40.jpg" width="300" />
+
+### Port layout
+
+| Front panel position | Logical port | Physical port | Type    |
+|----------------------|--------------|---------------|---------|
+| SFP left             | 8            | 5             | SFP+    |
+| RJ45 1               | 4            | 1             | Copper  |
+| RJ45 2               | 5            | 2             | Copper  |
+| RJ45 3               | 6            | 3             | Copper  |
+| RJ45 4               | 7            | 4             | Copper  |
+| SFP right            | 3            | 6             | SFP+    |
+
+Copper ports 1–4 match the KP-9000-6XH-X2 (2M-PCB43-V1.1) layout, which this
+board is a sibling of. The two SFP+ cages are wired the other way round
+compared to that board: physical port 5 is on SDS1 and physical port 6 is on
+SDS0.
+
+### Serial console
+
+The board has three through-holes used as the serial console, next to the
+heatsink in the middle of the board and visible in the top side photo above (the
+silkscreen labels are printed rotated by 90°). **The silkscreen labels are
+wrong**: they are shifted relative to the actual signals.
+
+| Silkscreen label | Actual signal |
+|------------------|---------------|
+| TX               | GND           |
+| RX               | TX            |
+| GND              | RX            |
+
+Wiring (following the measured signals, not the labels):
+
+| USB-TTL adapter | Board pad            |
+|-----------------|----------------------|
+| GND             | pad labelled `TX`    |
+| RX              | pad labelled `RX`    |
+| TX              | pad labelled `GND`   |
+
+- **Settings**: 115200 baud / 8N1 / 3.3V TTL (RTLPlayground), 9600 baud for the
+  stock firmware
+- The stock firmware only prints a few bytes and a factory test menu
+  (`?`, `rst`, `f`, `r`, `10`, `100`, `g1`, `sd`, `eye`, …); it is not an
+  interactive management console
+
+### LED configuration
+
+Case silkscreen, below the RJ45 ports:
+
+```
+White : 2.5G    Blue : 1000M/2500M L/A
+```
+
+The front panel also carries a `PWR` LED and one LED per SFP cage (`SFP1`,
+`SFP2`) on the right hand side.
+
+| Port       | LED          | Linked, idle | With traffic |
+|------------|--------------|--------------|--------------|
+| Copper     | white        | 2.5G only    | same         |
+| Copper     | blue         | solid        | blinks (L/A) |
+| SFP        | single LED   | solid        | blinks (L/A) |
+
+Measured LED-ID mapping (from the RTL837x LED set registers):
+
+| LED-ID | Function                        |
+|--------|---------------------------------|
+| LED-ID0| copper blue LED                 |
+| LED-ID1| copper white LED                |
+| LED-ID2| SFP single LED                  |
+
+**Important**: `RTL837X_REG_LED_GLB_IO_EN` bit 6 must be set in
+`machine_custom_init()`. Without it the RTL837x "solid" conditions (LINK/speed)
+do not drive the LEDs at all and only `ACT` does, which makes the white LED
+glow faintly at non-2.5G speeds.
+
+### SFP GPIO assignments
+
+| SFP               | pin\_detect (ModAbs)          | pin\_los         | SerDes | I2C SDA                 | I2C SCL                  |
+|-------------------|-------------------------------|------------------|--------|-------------------------|--------------------------|
+| Left (logical 8, physical 5) | GPIO30\_ACL\_BIT3\_EN | GPIO37   | SDS1   | GPIO39\_I2C\_SDA4       | GPIO40\_I2C\_SCL3\_MDC1  |
+| Right (logical 3, physical 6)| GPIO50\_I2C\_SCL2\_UART1\_TX | GPIO10\_LED10 | SDS0 | GPIO41\_I2C\_SDA3\_MDIO1 | GPIO40\_I2C\_SCL3\_MDC1 |
+
+`pin_tx_disable` is GPIO\_NA on both ports. Both I2C buses are confirmed
+working: with the same 10G module inserted, the `sfp` console command reads its
+vendor and model from the left cage (`Slot 1`) and from the right cage
+(`Slot 2`), DDM values included. The insertion also shows which SerDes drives
+each cage - the left cage reports `sds_config sds: 01` and the right cage
+`sds: 00`, which matches the table above.
+
+### Machine target
+
+Build for this device with `make MACHINE=2G5F4_10G2_V1_01`, which selects the
+`machine/2G5F4_10G2_V1_01.c` board definition. It is derived from
+`machine/KP_9000_6XHML_X2.c` with the SFP cage order and the reset pin changed.
+The fitted T25S40 cannot be read in DIO mode on this board: an image that leaves
+DIO enabled does not come up at all, both at the default read clock divider of 4
+and at 8 (two boards tested), while the same image with single-IO boots and runs
+normally. The board definition therefore sets `flash_sio`, which makes
+`flash_init()` use `CMD_FREAD` at divider 8. Only the management plane is
+slower, the ASIC keeps forwarding in hardware.

doc/devices/2M-PCB23-V3_2.md

@@ -0,0 +1,40 @@
+### 2M-PCB23-V3.2
+
+## Brands
+|Brand|Type|Managed|PCB|Flash|Chip RTL|
+|---|---|---|---|---|---|
+| keepLINK | KP-9000-9XHML-X | Yes|  2M-PCB23-V3.2 | 2M (Winbond 25Q) | |
+
+The original web interface reports this board as hardware version V3.2
+(original firmware V100.9.25, January 2026).
+
+## PCB
+
+<img src="photos/2M-PCB23-V3.2-managed/2M-PCB23-V3.2-top.jpeg" width="300" />
+
+## Compared to V3.1
+
+The component placement matches [2M-PCB23-V3.1](2M-PCB23-V3_1.md): same
+positions for the flash (U8), the T7/T8/T9/T10 headers, the SFP cage and its
+LED, and the unpopulated U7/U9 footprints. Build with
+`make MACHINE=KP_9000_9XHML_X_V3_2`, whose board file is a symlink to the V3.1
+definition and only differs in the reported name.
+
+## Tested
+
+Installed from the original firmware through its web interface with
+`installer/output/rtlplayground_oem_upgrade.bin`, then:
+
+- RJ45 ports link at 2.5G; LEDs: orange at 2.5G, green below
+- SFP port: module detected with vendor/model/serial and DOM readings
+  (temperature, Vcc, TX bias, TX/RX power), link up at 10G
+- factory MAC read from flash (0x1FC000)
+- reset button (GPIO48): held for more than 10 s it restores the default config
+
+## Flash access
+
+Reading the flash in-circuit with a SOIC-8 clip did not work: the chip ID
+comes back as `EF FF FF` (only the manufacturer byte is right) and reads
+return 0xff, the same as on a V3.1 board. The switch chip is powered through
+the clip and keeps driving the SPI lines. Make sure a way back is acceptable
+before installing, or hold the switch chip in reset while reading.

doc/devices/2M-PCB43-V1.1.md

@@ -15,7 +15,7 @@ and sold by Mokerlink.
 The device is fully supported:
 - All 4 2.5GBASE-T RJ45 ports work at 10/100/1000/2500 Mbps
 - The SFP+ port supports 1G, 2.5G and 10G modules 
-- LEDs work with the same indiciations as the OEM firmware (use KP_9000_6XHML_X2_V1_1 in machine.h if building yourself or the corresponding pre-compiled binary)
+- LEDs work with the same indiciations as the OEM firmware (build with `make MACHINE=KP_9000_6XHML_X2_V1_1` if building yourself, or use the corresponding pre-compiled binary)
 - untested due to missing Hardware: SFP+ ports equipped with 1G or 2.5G SFPs.
 
 ### Hardware overview

doc/devices/2M-PCB51-ML-V1_1.md

@@ -0,0 +1,34 @@
+# MokerLink POE-2G080110GS
+
+The `POE-2G080110GS` is an unmanaged 8x2.5G RJ45 and 1xSFP+ PoE switch sold by Mokerlink. There is also a `POE-2G080110GSM` managed version of the switch which may be similar but this has not been verified on actual hardware.
+The`KP-9000-9XHP-X` is the KeepLink rebrand for same product.
+
+## Brands
+| Brand  | Type             |Managed| PCB              | Flash           | Chip RTL      |
+|--------|------------------|-------|------------------|-----------------|---------------|
+| Mokerlink | POE-2G080110GS | No | 2M-PCB51-ML-V1.1 | 4MB (W25Q32JV) | 8373N + 8224N |
+| KeepLink  | KP-9000-9XHP-X | No | 2M-PCB51-ML-V1.1 | 4MB (W25Q32JV) | 8373N + 8224N |
+
+## Hardware overview
+Front
+
+<img src="photos/POE-2G080110GS/POE-2G080110GS-front.jpg" width="300" />
+
+Label
+
+<img src="photos/POE-2G080110GS/POE-2G080110GS-label.jpg" width="300" />
+
+The stock firmware from a POE-2G080110GS device had a sha256sum of `4c280853465eaad80e1772075c0f2d29c11fb8c561547f06353d3a5443c388e9`.
+
+### PCB
+
+Top silkscreen is marked `2M-PCB51-ML-V1.1`. PoE is provided by RTL8238C. Flash chip is 4 MB Winbond 25Q32JV (U8).
+
+<img src="photos/POE-2G080110GS/POE-2G080110GS-pcb.jpg" width="300" />
+
+> [!CAUTION]
+> This device operates on mains voltage, which can be a dangerous or potentially lethal hazard. Even after the device is unplugged, high-voltage capacitors can remain charged afterward and may still shock you. Do not attempt to open/disassemble a mains powered device unless you understand how to do so safely. Also, opening the device may void your warranty.
+
+## Notes
+
+The stock firmware blinks both the 2.5G and 1G LEDs on activity, despite the front of the device indicating that each LED should be independent. The `POE_2G080110GS` machine config behaves as the front of the device indicates (and what seems more logical) -- each port LED turns on and blinks independently.

doc/devices/F7008-2.5.md

@@ -0,0 +1,70 @@
+# F7008-2.5
+
+Documentation for the unmanaged 8 + 1 switch sold as `7018-2.5G` by Guangnianwei
+(光年微), board marking `F7008-2.5`, PCB `F7008-2.5GPOE-V1.2`.
+
+### Brands
+
+| Brand | Type | Managed | PCB | Flash | Chip RTL |
+|-------|------|---------|-----|-------|----------|
+| Guangnianwei (光年微) | 7018-2.5G | No | F7008-2.5GPOE-V1.2 | W25Q64JVSSIQ | 8373N + 8224N |
+
+### What works
+
+- All eight 2.5GBASE-T RJ45 ports at 10/100/1000/2500 Mbps
+- SFP+ port with 1G/2.5G/10G modules
+- LEDs of the eight RJ45 ports
+
+### Not verified
+
+- The LED pads for the SFP+ cage are not mapped yet. The SFP+ lamps are left
+  undriven, so an inserted module lights nothing.
+- The UART header footprint is present but unpopulated, and no boot log could be
+  captured from it, so the baud rate is unknown. There are most likely
+  unpopulated 0-Ohm resistors next to the header; bridging those should bring
+  the console to life.
+
+### Build target
+
+```
+make MACHINE=F7008_2_5
+```
+
+### LED wiring
+
+Measured pad to lamp map (`led_mux` index -> lamp):
+
+| Port | green lamp | yellow lamp |
+|------|-----------|-------------|
+| 1 | pad 0  | pad 1  |
+| 2 | pad 2  | pad 3  |
+| 3 | pad 4  | pad 5  |
+| 4 | pad 6  | pad 8  |
+| 5 | pad 9  | pad 10 |
+| 6 | pad 12 | pad 15 |
+| 7 | pad 18 | pad 21 |
+| 8 | pad 24 | pad 25 |
+
+Pads 7, 11, 13, 14, 16, 17, 19, 20, 22, 23, 26 and 27 do not drive an RJ45 lamp.
+
+The front panel is silkscreened `2.5G: Green / 10/100M: Yellow`, and the target
+follows it: the green lamp is on while the port is linked at 2.5G, the yellow
+lamp blinks while data is flowing.
+
+### PCB overview
+
+The board is marked `F7008-2.5GPOE-V1.2`.
+
+Top side:
+
+<img src="photos/F7008-2.5/PCB-top.jpg" width="600" />
+
+Bottom side:
+
+<img src="photos/F7008-2.5/PCB-bottom.jpg" width="600" />
+
+### Hardware notes
+
+- 4 x `RT4802SNE` magnetics
+- Unpopulated UART header footprint, next to the magnetics
+- Power input 12V 1A, 180 x 84 x 29 mm metal case

doc/devices/HASIVO_S1100WP_8GT_1SX_SE.md

@@ -0,0 +1,75 @@
+# Hasivo S1100WP-8GT-1SX-SE
+
+RTL8373-based 8×2.5G PoE + 1×SFP switch.
+
+### Label specifications
+
+- **Manufacturer**: Hasivo
+- **Model**: S1100WP-8GT-1SX-SE
+- **Ports**:
+  - 8 × RJ45: 10/100/1000/2500 Mbps with PoE
+  - 1 × SFP: 1G / 2.5G / 10G
+- **Power input**: DC 52V 2.5A
+- **Stock firmware defaults**: IP 192.168.0.1, user/password `admin`/`admin`
+
+### What works
+
+- All eight 2.5GBASE-T RJ45 ports at 10/100/1000/2500 Mbps (PoE is not configurable via RTLPlayground)
+- SFP port supporting 1G, 2.5G and 10G modules
+- LEDs: green (2.5G) and amber (1G/100M/10M) per copper port; combined link/act on the SFP port
+
+### Photos
+
+Front panel:
+
+<img src="photos/HASIVO_S1100WP-8GT-1SX-SE/front.jpg" width="600" />
+
+Label:
+
+<img src="photos/HASIVO_S1100WP-8GT-1SX-SE/label.jpg" width="600" />
+
+### PCB overview
+
+Top side:
+
+<img src="photos/HASIVO_S1100WP-8GT-1SX-SE/pcb_top.jpg" width="600" />
+
+Bottom side:
+
+<img src="photos/HASIVO_S1100WP-8GT-1SX-SE/pcb_bottom.jpg" width="600" />
+
+### Port layout
+
+| Logical port | Physical port | Type   |
+|--------------|---------------|--------|
+| 1-8          | 1-8           | Copper |
+| 9            | 9             | SFP    |
+
+### LED configuration
+
+Copper ports use LED SET0, the SFP port uses LED SET1.
+
+| SET  | LED0                                          | LED2                                                 |
+|------|-----------------------------------------------|------------------------------------------------------|
+| SET0 | Green — lights on 2.5G link with activity     | Amber — lights on 1G / 100M / 10M link with activity |
+| SET1 | All speeds — lights on any link with activity | —                                                    |
+
+### SFP GPIO assignments
+
+| SFP (logical 9) | pin_detect (ModAbs) | pin_los | pin_tx_disable | SerDes | I2C SDA         | I2C SCL             |
+|-----------------|---------------------|---------|----------------|--------|-----------------|---------------------|
+| SFP             | GPIO30_ACL_BIT3_EN  | GPIO37  | GPIO_NA        | SDS1   | GPIO39_I2C_SDA4 | GPIO40_I2C_SCL3_MDC1 |
+
+### Machine configuration
+
+- `reset_pin`: GPIO54_ACL_BIT2_EN
+- High LEDs (pads 27-29): mux = `LED_27 | LED_28_SYS | LED_29`, enable = `LED_28_SYS | LED_29`
+
+### PoE
+
+The 8 copper ports support PoE powered by a DC 52V 2.5A input. Power management is handled by two HS104PTI chipsets over I2C (GPIO47_I2C_SDA0, GPIO46_I2C_SCL0):
+
+* Chip 1 (I2C address 0x1A): 2.5G Ports 1–4
+* Chip 2 (I2C address 0x2A): 2.5G Ports 5–8
+
+Note: The current implementation does not yet support PoE management. Consequently, PoE currently operates in standalone mode without software control.

doc/devices/HYWS-SGT0108S.md

@@ -0,0 +1,64 @@
+# Lianguo HYWS-SGT0108S
+
+Following is documentation for unmanaged switch marked as `HYWS-SGT0108S`.
+
+Original software is running UART on 9600 baud rate. Output is very minimal.
+
+Using SPI clamp in-board is the only method for initial installation and update.
+
+Stock flash chip is 512KB in size. Consider replacing for update through Web UI.
+
+- Header for uart is clearly identified.
+- The red LED act as a powered-on LED.
+- There is no SYS LED.
+- There is a footprint for a Reset button. Unpopulated. Unclear if routed back to RTL8373.
+
+### Label specifications
+
+- **Name**: 2.5G Ethernet Switch 8+1
+- **Model**: HYWS-SGT0108S
+- **Ports**:  
+  - 8 × RJ45: 10/100/1000/2500 Mbps  
+  - 1 × SFP: 1000 / 2500 / 10000 Mbps  
+
+### What works
+
+- All eight 2.5GBASE-T RJ45 ports at 10/100/1000/2500 Mbps  
+- SFP port supporting 1G, 2.5G and 10G modules 
+- RJ45 LEDs (Blue+White)
+- SFP Single LED (Blue)
+
+### Assembled unit
+
+Top side
+
+<img src="photos/HYWS-SGT0108S-unmanaged\Assembled-top.jpg" width="300" />
+
+Front
+
+<img src="photos/HYWS-SGT0108S-unmanaged\Assembled-front.jpg" width="300" />
+
+Bottom
+
+<img src="photos/HYWS-SGT0108S-unmanaged\Assembled-bottom.jpg" width="300" />
+
+### PCB overview
+
+**Board markings**  
+- Top silkscreen: 2G5F_20G_V1.01 / 2023-09-28  
+
+Top side
+
+<img src="photos/HYWS-SGT0108S-unmanaged\PCB-top.jpg" width="300" />
+
+Bottom
+
+<img src="photos/HYWS-SGT0108S-unmanaged\PCB-bot.jpg" width="300" />
+
+Original SPI Flash Chip
+
+<img src="photos/HYWS-SGT0108S-unmanaged\og_spi_flash.jpg" width="300" />
+
+## Power supply
+
+Input power is delivered via barell plug, `12V 1A` adapter was provided.

doc/devices/Hasivo-S1100P-8GT-1SX-S.md

@@ -0,0 +1,41 @@
+# Hasivo S1100P-8GT-1SX-S
+
+RTL8373N switch with 8 × 2.5G RJ45 ports and 1 × SFP port. GD25Q16 2 MiB flash.
+
+## Port mapping
+
+| Front port | ASIC MAC | Interface |
+| --- | --- | --- |
+| 1–8 | 0–7 | RJ45 |
+| 9 | 8 | SFP (SDS1) |
+
+## Board signals
+
+| Function | GPIO |
+| --- | --- |
+| SFP detect | 38 |
+| SFP LOS | 37 |
+| SFP TX disable | Not connected (`GPIO_NA`) |
+| SFP I2C SDA / SCL | 39 / 40 |
+| Reset | 54 |
+
+PoE is managed by an RTL8238B and a separate STC8H8K64U controller, not directly by the RTL8373N. RTLPlayground does not control PoE on this board.
+
+## Build
+
+`make MACHINE=HASIVO_S1100P_8GT_1SX_S CI=1`
+
+## Photos
+
+Front panel:
+
+![Front panel](photos/Hasivo-S1100P-8GT-1SX-S/front.jpg)
+
+PCB top:
+
+![PCB top](photos/Hasivo-S1100P-8GT-1SX-S/pcb_top.jpg)
+
+PCB bottom:
+
+![PCB bottom](photos/Hasivo-S1100P-8GT-1SX-S/pcb_bottom.jpg)
+

doc/devices/ONT-S207CW-62TS-SE.md

@@ -0,0 +1,124 @@
+# ONT-S207CW-62TS-SE / Binardat 2G06-04210GSM / Binardat 2G06-04210GS
+
+## Overview
+
+ONT-S207CW-62TS-SE, Binardat 2G06-04210GSM (managed), and Binardat 2G06-04210GS (unmanaged) are **switches with identical hardware**:
+- **CPU:** RTL8372N (confirmed via serial: "Detecting CPU: RTL8372N")
+- **Flash:** GD25Q128E (16MB)
+- **Ports:** 4x 2.5G RJ45 + 2x 10G SFP+
+- **Console:** 115200 baud (RTLPlayground firmware) / 9600 baud (stock firmware)
+
+**Note:** These devices are **RTL8372N-based** with **LAN ports mounted upside-down on the PCB**, requiring a custom configuration. The `MACHINE_ONT_S207CW_62TS_SE` definition uses a **custom configuration** with:
+- Corrected port mapping (physical ports 1-4 = RJ45, 5-6 = SFP)
+- Optimized LED sets for different port wiring (RJ45 Ports 1-2 have reversed LED wiring)
+- Custom LED mux configuration
+
+## Device Photos
+
+### Binardat 2G06-04210GSM
+
+**Front View:**
+<img src="photos/ONT-S207CW-Binardat-2G06-04210GSM/binardat-2g06-04210gsm-front.jpg" width="400" />
+
+**Back View:**
+<img src="photos/ONT-S207CW-Binardat-2G06-04210GSM/binardat-2g06-04210gsm-back.jpg" width="400" />
+
+**PCB View:**
+<img src="photos/ONT-S207CW-Binardat-2G06-04210GSM/binardat-2g06-04210gsm-pcb.jpg" width="400" />
+
+*Photos courtesy of [andreas5232](https://github.com/andreas5232) from [emesix/ONT-S207CW-62TS-SE#1](https://github.com/emesix/ONT-S207CW-62TS-SE/issues/1)*
+
+## Hardware Specification
+
+| Feature | Value |
+|---|---|
+| **CPU** | RTL8372N |
+| **Flash** | GD25Q128E (16MB) |
+| **RAM** | Integrated in RTL8372N |
+| **RJ45 Ports** | 4x 2.5GBase-T (Ports 1-4) |
+| **SFP+ Ports** | 2x 10G (Ports 5-6) |
+| **Console** | UART (115200 baud for RTLPlayground, 9600 baud for stock) |
+| **Stock Web UI IP** | 192.168.2.1 |
+| **Loader Mode IP** | 192.168.10.247 |
+
+## Port Layout
+
+```
+┌─────────────────────────────────────────────┐
+│  ┌─────┐ ┌─────┐ ┌─────┐ ┌─────┐   ┌─────┐ ┌─────┐  │
+│  │RJ45 │ │RJ45 │ │RJ45 │ │RJ45 │   │SFP+ │ │SFP+ │  │
+│  │  1  │ │  2  │ │  3  │ │  4  │   │  5  │ │  6  │  │
+│  └─────┘ └─────┘ └─────┘ └─────┘   └─────┘ └─────┘  │
+│                                                     │
+│  [System LEDs]                                      │
+└─────────────────────────────────────────────┘
+```
+
+## LED Behavior
+
+### RJ45 Ports (Physical 1-4)
+
+| LED Color | Speed |
+|---|---|
+| **Green** | 2.5G |
+| **Orange** | 1G / 100M / 10M |
+| **Off** | No link |
+
+### SFP+ Ports (Physical 5-6)
+
+| LED Color | Speed |
+|---|---|
+| **Green** | 10G |
+| **Orange** | 2.5G / 1G |
+| **Off** | No link / No SFP module |
+
+## Port to Logical Mapping
+
+Custom mapping for ONT-S207CW-62TS-SE:
+
+| Physical Port | Logical Port | Type | LED Set |
+|---|---|---|---|
+| 1 | 7 | RJ45 | 2 |
+| 2 | 6 | RJ45 | 2 |
+| 3 | 5 | RJ45 | 0 |
+| 4 | 4 | RJ45 | 0 |
+| 5 | 3 | SFP+ (SDS0) | 3 |
+| 6 | 8 | SFP+ (SDS1) | 3 |
+
+**Note:** Ports 1-2 (RJ45) have reversed LED wiring and use LED-Set 2. SFP Ports 5-6 use LED-Set 3 to account for wiring differences.
+
+## Flashing Instructions
+
+See the [README.md](../../README.md) for general flashing instructions.
+For compilation, use command `make MACHINE=ONT_S207CW_62TS_SE` to specify the machine.
+If you are using managed version of the switch, you can then also compile the installer and flash using GUI - run command `cd installer` and finally `make`.
+
+## Stock Firmware Information
+
+| Property | Value |
+|---|---|
+| **Default IP** | 192.168.2.1 (ONT) / 192.168.10.247 (Loader Mode) |
+| **Default Login** | admin/admin (may vary by OEM) |
+| **Console Baud** | 9600 (stock) / 115200 (RTLPlayground) |
+| **CPU String** | "RTL8372N" |
+| **Loader Mode IP** | 192.168.10.247 |
+
+## Configuration Details
+
+- **CPU:** RTL8372N
+- **SFP:** SDS0 on logical port 3 (GPIO37), SDS1 on logical port 8 (GPIO38)
+- **Port Mapping:** Physical 1-4 → Logical 7-4, Physical 5-6 → Logical 3,8
+- **LED Wiring:** RJ45 Ports 1-2 and SFP Port 6 have reversed LED pins (orange/green swapped)
+- **LED Sets:** Set 0 for RJ45 Ports 3-4, Set 1 unused, Set 2 for RJ45 Ports 1-2, Set 3 for SFP Ports 5-6
+
+## Notes
+
+- ONT-S207CW-62TS-SE, Binardat 2G06-04210GSM (managed), and Binardat 2G06-04210GS (unmanaged) are **RTL8372N-based** with **LAN ports mounted upside-down on the PCB**
+- Web UI shows correct physical port order (1-2-3-4 for RJ45, 5-6 for SFP)
+- Loader Mode always remains accessible at `192.168.10.247` for recovery
+- Serial console: 9600 baud (stock) → 115200 baud (RTLPlayground)
+
+## References
+
+- [GitHub Issue: emesix/ONT-S207CW-62TS-SE#1](https://github.com/emesix/ONT-S207CW-62TS-SE/issues/1) - Original photos and discussion
+- [PR: logicog/RTLPlayground#465](https://github.com/logicog/RTLPlayground/pull/465) - Feedback and improvements

doc/devices/PB-2132.md

@@ -0,0 +1,53 @@
+# IG108
+
+Following is documentation for unmanaged switch marked as `IG108`.
+
+Original software is running UART on 9600 baud rate.
+
+Using SPI clamp in-board is the only method for initial installation.
+
+### Brands
+
+| Brand   | Type       | Managed | PCB        | Flash       | Chip RTL      |
+|---------|------------|---------|------------|-------------|---------------|
+| Goalake | IG108      | No      | PB-2132    | cFeon Q32A  | 8373N + 8224N |
+| Netcore | GS9        | No      | PB-2132    | cFeon Q32A  | 8373N + 8224N |
+
+### What works
+
+- All eight 2.5GBASE-T RJ45 ports at 10/100/1000/2500 Mbps
+- SFP port with 1G/2.5G/10G modules
+- LEDs
+
+### PCB overview
+
+**Board markings**
+
+- Top silkscreen: PB-2132
+
+Front Goalake:
+
+<img src="photos/PB-2132/case.jpg" width="600" />
+
+Label Goalake: 
+
+<img src="photos/PB-2132/label.jpg" width="600" />
+
+Top side:
+
+<img src="photos/PB-2132/PCB-top.jpg" width="600" />
+
+Bottom:
+
+<img src="photos/PB-2132/PCB-bottom.jpg" width="600" />
+
+### Serial console
+
+The PCB has unpopulated through-holes near the SFP Port, with description of each pin, works out of the box.
+
+- **Settings**: 115200 baud / 8N1 / 3.3V TTL
+
+### Power supply
+
+Input power is delivered via barell plug, `12V 1A` adapter was provided.
+

doc/devices/PCB-AFGW_SWTG218ANS_V2.0.2.md

@@ -0,0 +1,93 @@
+# PCB-AFGW SWTG218ANS V2.0.2
+
+RTL8373-based 8×2.5G PoE + 1×10G SFP+ web-managed switch, sold as the
+**RealHD PSW8-25MG**.
+
+### Label specifications
+
+- **Manufacturer**: RealHD
+- **Model**: PSW8-25MG
+- **PCB marking**: `PCB-AFGW SWTG218ANS_V2.0.2` (silkscreened as `PCB-AFGW SWTG218ANS_V2.0.2_23308`)
+- **Ports**:
+  - 8 × RJ45: 10/100/1000/2500 Mbps with PoE
+  - 1 × SFP+: 1G / 2.5G / 10G (front panel marking `9(10G)`)
+- **Power input**: internal AC PSU, `100-240V~ 50/60Hz 2A MAX`, DC output `53.5V 2.8A 150W`
+- **Front panel LEDs**: `PWR`, `SYS`, per RJ45 port `Link/Act` and `PoE`, per SFP port 2.5G (green) and 10G (blue)
+- **Stock firmware defaults**: web-managed PoE switch
+
+### What works
+
+- All eight 2.5GBASE-T RJ45 ports at 10/100/1000/2500 Mbps (PoE is not configurable via RTLPlayground)
+- SFP+ port supporting 1G, 2.5G and 10G modules
+- LEDs: green (2.5G) and amber (1G/100M/10M) per RJ45 port with activity, green (1G/2.5G) and blue (10G) on the SFP port, plus the `SYS` LED
+
+### Photos
+
+Front panel:
+
+<img src="photos/PCB-AFGW_SWTG218ANS_V2.0.2/device_front.jpg" width="800" />
+
+Board mounted in the case, with the internal PSU above it:
+
+<img src="photos/PCB-AFGW_SWTG218ANS_V2.0.2/dev_top_w_pcb.jpg" width="800" />
+
+### PCB overview
+
+Top side:
+
+<img src="photos/PCB-AFGW_SWTG218ANS_V2.0.2/pcb_front.jpg" width="800" />
+
+Bottom side:
+
+<img src="photos/PCB-AFGW_SWTG218ANS_V2.0.2/pcb_back.jpg" width="800" />
+
+A 4-pad UART header with `RXD` / `TXD` silkscreened sits next to the right heatsink. 
+
+### Port layout
+
+| Logical port | Physical port | Type   |
+|--------------|---------------|--------|
+| 1-8          | 1-8           | RJ45   |
+| 9            | 9             | SFP+   |
+
+### LED configuration
+
+RJ45 ports use LED SET0, the SFP+ port uses LED SET1. The board needs the custom
+LED mux (`led_mux_custom = 1`), and `RTL837X_REG_LED_GLB_IO_EN` is written to
+`0x7f24977f` in `machine_custom_init()` so the LED pins are enabled as wired here.
+
+| SET  | LED pin 0                                     | LED pin 1                                             |
+|------|-----------------------------------------------|-------------------------------------------------------|
+| SET0 | Green — lights on 2.5G link with activity     | Amber — lights on 1G / 100M / 10M link with activity  |
+| SET1 | All speeds — lights on any link with activity | 10G — lights on 10G link with activity                |
+
+On the front panel this gives one bi-color `Link/Act` LED per RJ45 port (green at
+2.5G, amber at 10/100/1000M, blinking with activity) and one `PoE` LED per RJ45
+port, which is driven by the PoE hardware and not by the switch. The SFP+ port has
+a green 2.5G LED and a blue 10G LED.
+
+### SFP GPIO assignments
+
+| SFP (logical 9) | pin_detect (ModAbs) | pin_los | pin_tx_disable | SerDes | I2C SDA          | I2C SCL              |
+|-----------------|---------------------|---------|----------------|--------|------------------|----------------------|
+| SFP             | GPIO38              | GPIO_NA | GPIO_NA        | SDS1   | GPIO39_I2C_SDA4  | GPIO40_I2C_SCL3_MDC1 |
+
+### Machine configuration
+
+```
+make MACHINE=REALHD_PSW8_25MG
+```
+
+### Programming
+
+The serial console runs at 8N1 on 115200 baud with the RTLPlayground firmware
+(see the UART header note under *PCB overview*). The stock firmware baud rate 57600. The flash chip was 2MB.
+
+### PoE
+
+The 8 RJ45 ports support PoE, powered by the internal 150W AC PSU (DC output
+53.5V 2.8A) mounted above the main board.
+
+Note: RTLPlayground does not provide PoE management for this board. The PoE
+hardware runs with its own configuration, the same as on the other PoE boards in
+this project.

doc/devices/PCB-K0402WS-V1.0.md

@@ -0,0 +1,72 @@
+# PCB-K0402WS-V1.0
+
+This is the first revision of the PCB-K0402WS series of boards; it's very similar to 2.0 in terms of layout/wiring.
+
+The board does have footprints for a VLAN modeswitch at the front and a second flash chip on the bottom of the PCB, but these have been seen to ship unpopulated.
+
+### Brands
+
+* YuanLey YS25-0402 <sup><sub>Note: there are [multiple revisions](PCB-K0402WS-V3.0.md) of the YS25-0402. This first revision lacks the "Vlan" switch cutout on the front of the enclosure.</sub></sup>
+
+### Programming
+
+Use a SOIC-8 clamp to the flash chip (U1 on silkscreen) at the top of the board, leaving it otherwise unpowered.
+
+The board has been seen to ship with a single [BoyaMicro BY25Q16BSSIG](https://www.lcsc.com/datasheet/C382744.pdf) 16Mbit (2MiB) chip, though exact BOM may vary.
+
+<a href="photos/PCB-K0402WS-V1.0/flash.jpg"><img src="photos/PCB-K0402WS-V1.0/flash.jpg" width="300" /></a>
+
+### What works
+
+- All four 2.5GBASE-T RJ45 ports at 10/100/1000/2500 Mbps
+- Both SFP ports supporting 1G, 2.5G and 10G modules
+- LEDs (Power LED is always on)
+
+### PCB overview
+
+#### Board markings
+- Top silkscreen: PCB-K0402W-V1.0 / DIP-K0402WS-V1.0
+
+#### Key components
+- U1: SPI NOR flash, 2 MiB
+- S1: unpopulated slide switch footprint (three through-holes used as serial console)
+
+Front panel
+
+<a href="photos/PCB-K0402WS-V1.0/ports.jpg"><img src="photos/PCB-K0402WS-V1.0/ports.jpg" width="300" /></a>
+
+PCB top
+
+<a href="photos/PCB-K0402WS-V1.0/pcb-top.jpg"><img src="photos/PCB-K0402WS-V1.0/pcb-top.jpg" width="300" /></a>
+
+PCB bottom
+
+<a href="photos/PCB-K0402WS-V1.0/pcb-bottom.jpg"><img src="photos/PCB-K0402WS-V1.0/pcb-bottom.jpg" width="300" /></a>
+
+Bottom and rear
+
+<a href="photos/PCB-K0402WS-V1.0/bottom.jpg"><img src="photos/PCB-K0402WS-V1.0/bottom.jpg" width="300" /></a>
+
+### Power supply
+
+Input power is delivered via a standard DC barrel jack (5.5x2.1mm) with an included 12V 1A adapter.
+
+There is an optional grounding screw next to the DC input.
+
+### Serial console
+
+The PCB has a slide switch footprint (unpopulated) at the front right that is connected to the UART. It is labeled as "S1" on the silkscreen.
+
+Facing the top front (port) side of the board, the pinout from left to right is:
+
+| Position | Signal | GPIO              |
+|----------|--------|-------------------|
+| 1        | RX     | GPIO32\_UART0\_RX |
+| 2        | GND    | N/A               |
+| 3        | TX     | GPIO31\_UART0\_TX |
+
+Standard 0.1" / 2.54mm pin headers will not fit, so wires or clips may be used instead.
+
+The image below shows some oversized clips clumsily attached.
+
+<a href="photos/PCB-K0402WS-V1.0/uart.jpg"><img src="photos/PCB-K0402WS-V1.0/uart.jpg" width="300" /></a>

doc/devices/PCB-K0402WS-V3.0.md

@@ -19,9 +19,14 @@ of the device just above the big label, might be covered by a QC sticker.
 
 <img src="photos/PCB-K0402WS-V3.0/Ztyuav_Z-QWYT0402_label.jpg" width="300" />
 
-* Siscolink SL-G0402F
+* SICSOLINK SL-G0402F
+
+<img src="photos/PCB-K0402WS-V3.0/SICSOLINK SL-G0402F.jpg" width="300"/>
+
+* YuanLey YS25-0402 <sup><sub>Note: there are [multiple revisions](PCB-K0402WS-V1.0.md) of the YS25-0402. This is the newer revision that has a "Vlan" switch on the front.</sub></sup>
+
+<img src="photos/PCB-K0402WS-V3.0/YuanLey_YS25-0402.jpg" width="300"/>
 
-<img src="photos/PCB-K0402WS-V3.0/Siscolink SL-G0402F.jpg" width="300"/>
 ### Programming
 
 Using SPI clamp in-board is the only method for initial installation.
@@ -43,7 +48,7 @@ For the initial flash (at least with flashrom), the bin file produced by the bui
 ### PCB overview
 
 **Board markings**  
-- Top silkscreen: PCB-KO4022W-V3.0 / DIP-KO4022WS-V3.0  
+- Top silkscreen: PCB-K0402W-V3.0 / DIP-K0402WS-V3.0  
 
 Top side
 

doc/devices/SIRIVISION_S25G2206FP_V1_1.md

@@ -0,0 +1,32 @@
+### SIRIVISION_S25G2206FP_V1_1
+
+## Brands
+|Brand|Type|Managed|PCB|Flash|Chip RTL|
+|---|---|---|---|---|---|
+| YuanLey | YS25-0402M | Yes | SIRIVISION_S25G2206FP_V1_1 | 2M | 8372 |
+
+### Label specifications
+
+- **Name**:
+- **Ports**:
+  - 4 × RJ45: 10/100/1000/2500 Mbps
+  - 2 × SFP+: 1000 / 2500 / 10000 Mbps
+- **Power**: 12V DC, 1A  5525 connector
+
+### What works
+The device is fully supported:
+- ALL 2.5GBASE-T RJ45 ports work at 10/100/1000/2500 Mbps
+- ALL SFP+ ports support 1G, 2.5G and 10G modules
+- LEDs work with the same indiciations as the OEM firmware
+### PCB overview
+
+**Board markings**
+- Top silkscreen: SIRIVISION S25G2206FP V1.1
+
+Top side
+
+<img src="photos/SIRIVISION_S25G2206FP_V1_1/dev_pcb_top.jpg" width="300" />
+
+## Power supply
+
+Input power is delivered via barell plug, `12V 1A` adapter was provided.

doc/devices/SWTG018AS_A_V2_0.md

@@ -1,9 +1,11 @@
 # SWTG018AS-A V2.0
 ## Brands
-| Brand  | Type             |Managed| PCB              | Flash           | Chip RTL      |
-|--------|------------------|-------|------------------|-----------------|---------------|
-| Ampcom | SWTG018AS-A V2.0 | No    | SWTG018AS-A V2.0 | 2MB             | 8273N + 8224N |
-| Horaco | HC-SWTGW218AS-A  | Yes   | SWTG018AS-A V2.0 | 2MB(25Q16JVSIQ) | 8273N + 8224N |
+| Brand   | Type             |Managed| PCB              | Flash           | Chip RTL      |
+|---------|------------------|-------|------------------|-----------------|---------------|
+| Ampcom  | SWTG018AS-A V2.0 | No    | SWTG018AS-A V2.0 | 2MB             | 8273N + 8224N |
+| Horaco  | HC-SWTGW218AS-A  | Yes   | SWTG018AS-A V2.0 | 2MB(25Q16JVSIQ) | 8273N + 8224N |
+| GoodTop | GT-ST018M        | Yes   | SWTG018AS-A V2.0 | 2MB             | 8273N + 8224N |
+| MokerLink | 2G08110GSN     | No    | SWTG018AS-A V2.0 | 512KB(25Q40)    | 8273N + 8224N |
 
 
 The following is a documentation for the unmanaged switch marked as `SWTG018AS-A V2.0`.
@@ -13,7 +15,7 @@ The original software is running UART on 9600 baud rate.
 
 Using a SOIC clamp in-board is the only method for initial installation.
 
-The board has a single flash chips `BS` with 4M-bit size. The front switch, switches between the two flash chips.
+The board has a single flash chips `BS` with 2M-byte size. The front switch, switches between the two flash chips.
 These can be programed independently by using said switch - so it is e.g. possible to run the original and new firmware in parallel.
 
 ### Label specifications

doc/devices/SWTG024AS-A-V2.0.1_4C_2SFP.md

@@ -2,9 +2,11 @@
 It is highly similar to SWTG024AS-V2.0, with the only difference being the GPIO configuration for the SFP port.
 
 ## Brands
-|Brand|Type|Managed|PCB|Flash|Chip RTL|
-|---|---|---|---|---|---|
-| Horaco | ZX-SG4T2 |  | PCB-SWTG024AS-A-V2.0.1_19650 | P25D40SH | ??? |
+|Brand    |Type          |Managed|PCB                           |Flash           |Chip RTL|
+|---------|--------------|-------|------------------------------|----------------|--------|
+| Horaco  | ZX-SG4T2     | No    | PCB-SWTG024AS-A-V2.0.1_19650 | 512K (P25D40SH)| 8372N  |
+| Horaco  | HC-SWTG124AS | Yes   | PCB-SWTG024AS-A-V2.0.1_19650 | 2M   (W25Q16JV)| 8372N  |
+| GoodTop | GT-ST024M    | Yes   | PCB-SWTG024AS-A-V2.0.1       | 2M             | 8372N  |
 
 ### Label specifications
 
@@ -33,6 +35,8 @@ Bottom
 
 <img src="photos/SWTG024AS-A-V2_0_1_19650/horaco-zx-sg4t2-pcb-bottom.jpg" width="300" />
 
+On Horaco HC-SWTG124AS with board PCB-SWTG024AS-A-V2.0.1_19650 the reset button is soldered at `SW1` (on the top left, GPIO48).
+
 ### J1, serial console
 
 | `J1` pin | Signal      |

doc/devices/SWTG024AS-V2.0.md

@@ -46,4 +46,4 @@ Note,`R52`、`R53`may not be installed.You need to bridge them using either sold
 
 ## Power supply
 
-Input power is delivered via barell plug, `12V 1A` adapter was provided.
+Input power is delivered via barell plug, `12V 1A` adapter was provided.

doc/devices/SWTGW215AS.md

@@ -9,7 +9,7 @@
 
 ## RTLPlayground target
 
-Use machine target `MACHINE_LIANGUO_ZX_SWTGW215AS` for this device.
+Build for this device with `make MACHINE=LIANGUO_ZX_SWTGW215AS`, which selects the `machine/LIANGUO_ZX_SWTGW215AS.c` board definition.
 
 Physical hardware verification: 5x RJ45 ports + 1x SFP port. 
 Port 5 RJ45 is interfaced through a RTL8221B IC.

doc/devices/SWTGW218AS.md

@@ -5,9 +5,15 @@
 |---|---|---|---|---|---|
 | Mokerlink | ZX-SWTGW218AS | Yes|  SWTG118AS-V2.0-16029 | 2MB (FM25Q16A)| 8273N + 8224N |
 | Sodola | | | | | |
-| Horaco | | | | | |
+| Horaco | HC-SWTGW218AS | Yes | SWTG118AS-V2.1 | 2MB | 8373 (non-N) |
 | XikeStor | SKS3200-8E1X | Yes | SWTG118AS-V2.1-17462 | 2MB (25Q16JVSIQ) | |
 
+All boards in this table use `MACHINE=SWTGW218AS`. The Horaco HC-SWTGW218AS is also sold with the
+[SWTG018AS-A V2.0](SWTG018AS_A_V2_0.md) board, which needs a different firmware. Check the marking
+on the PCB before flashing.
+
+The PHY and flash-chip models for the tested Horaco units remain unverified.
+
 
 ## Photos
 

doc/devices/ZOERAX_HG0402XG_V1.1.md

@@ -0,0 +1,36 @@
+### HG0402XG V1.1
+
+## Brands
+|Brand|Type|Managed|PCB|Flash|Chip RTL|
+|---|---|---|---|---|---|
+| ZoeRax | HG0402XG | No|  HG0402XG V1.1 | 1M (Winbond 25Q) | |
+
+### Label specifications
+
+- **Name**: ZOERAX 2.5G 2SFP 4-Port Gigabit Switch
+- **Model**: HG0402XG-EU
+- **Ports**:
+  - 4 × RJ45: 10/100/1000/2500 Mbps
+  - 2 × SFP+: 1000 / 2500 / 10000 Mbps
+<img src="photos/HG0402XG_V1_1/HG0402XG_V1_1_bottom.jpg" width="300" />
+
+## PCB
+
+<img src="photos/HG0402XG_V1_1/HG0402XG_V1_1_top.jpg" width="300" />
+<img src="photos/HG0402XG_V1_1/HG0402XG_V1_1_bottom.jpg" width="300" />
+
+Be aware that the UART header is metric, the pitch is 2mm. The JST PH2.0mm header shown was addeded after purchase
+
+The slider switch controls microcontroller U4 in order to configure the VLAN on/off settings. U4 can reset the SoC
+and also write into its registers to configure VLAN. With RTLPlayground the slider cannot be used for anything else
+than triggering reset. Any change in the position will trigger this reset.
+
+### Installation
+
+Using SPI clamp on-board is the only method for initial installation.
+
+### What works
+
+- All four 2.5GBASE-T RJ45 ports at 10/100/1000/2500 Mbps
+- SFP ports with various 1G/10G Fiber and Ethernet modules
+- LEDs

doc/dns.md

@@ -0,0 +1,23 @@
+# DNS
+
+The switch can resolve host names to IPv4 addresses, for the NTP client and
+for anything else that is given a name instead of an address.
+
+```
+dns server <ip> [<ip>]   # one or two servers, 0.0.0.0 removes them
+dns lookup <name>        # resolve a name and print the address
+dns                      # show the servers and the last lookup
+```
+
+The answer to a lookup arrives after the command has returned, so it is
+printed on the serial console and in the syslog. In the console of the web
+interface, run `dns` afterwards to see it.
+
+Without a configured server the resolver uses the one handed out by DHCP.
+Only address records are looked up; a query goes to the servers in turn,
+every two seconds, and gives up after six tries. A name that already is a
+dotted address is taken as it is, without a query.
+
+The DNS servers are set in the Network card of the System page as well.
+Settings made there or on the console are kept only once the configuration is
+saved.

doc/health.md

@@ -0,0 +1,80 @@
+# 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
+lldp max 00 slow 0x0000
+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, the LLDP timer 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.

doc/igmp.md

@@ -69,18 +69,26 @@ The code currently provides the following functions:
 void igmp_setup(void) __banked;
 void igmp_enable(void) __banked;
 void igmp_router_port_set(uint16_t pmask) __banked;
-void igmp_packet_handler(void) __banked;
+uint8_t igmp_packet_handler(void) __banked;
 void igmp_show(void) __banked;
 ```c
 `igmp_setup()` is called at boot-time and configures flooding of all IP-MC packets by
-default, as otherwise no IP-MC would be possible in the network.
+default, as otherwise no IP-MC would be possible in the network. `igmp off` calls it as
+well; groups learned while IGMP was on stay in the table, but are no longer looked up.
 
 `igmp_enable()`starts IGMP which cause IGMP packets to be handled by the CPU and forwarding
 of IP-MC packets to be limited to only subscribed ports.
 
 `igmp_router_port_set()`configures forwarding ports for IGMP messages.
 
-`igmp_packet_handler()` implements handling of trapped IGMP packets by the CPU.
+`igmp_packet_handler()` implements handling of trapped IGMP packets by the CPU. It is called
+for frames to 01:00:5e:xx:xx:xx and returns 0 when the frame is not a trapped IGMP packet,
+so that it is passed on to the IP stack. IGMPv1/v2 reports (0x12, 0x16) join the group given
+in the IGMP header, IGMPv2 leaves (0x17) leave it, IGMPv3 reports use the first group record
+(type 4 joins, type 3 leaves). Trapped IGMP frames arrive without a VLAN tag.
+Reports for groups in 224.0.0.0/24 are ignored: the switch floods that block
+even with IGMP on, and an entry for a group such as 224.0.0.251 (mDNS) would
+otherwise limit it to the ports that reported it, as RFC 4541 warns against.
 
 `igmp_show()` prints out the IGMP configuration on the CLI.
 

doc/l2.md

@@ -73,13 +73,14 @@ without being flooded to the other ports. No bridge relays these frames:
 their addresses are in the set that 802.1D-2004 clause 7.12.6 forbids a
 bridge to forward, and what travels the network is the information, with
 every bridge regenerating BPDUs of its own on its designated ports. The reserved-multicast *trap* action
-cannot do that on this hardware, because its destination is an external CPU
-attached to a physical port, which these boards do not populate. The protocol
-modules therefore leave the reserved-multicast action at *forward* and constrain
-the egress with a static L2 multicast entry instead: the lookup hits the entry's
-own port mask rather than the VLAN flood mask. Verified on a SWTGW218AS both
-ways, with the CPU bit cleared, where delivery stops, and with the CPU bit alone,
-where nothing egresses.
+does that: it delivers to the port named in `EXT_CPU_CTRL` (0x6724), which
+`trap_init()` sets to 9, the CPU port, at startup; with the reset value 0xf it
+names no port and the trapped frame is dropped. STP traps its address while it
+runs (see stp.md). Where the action is left at *forward*, the egress can be
+constrained with a static L2 multicast entry instead: the lookup hits the
+entry's own port mask rather than the VLAN flood mask. Verified on a SWTGW218AS
+both ways, with the CPU bit cleared, where delivery stops, and with the CPU bit
+alone, where nothing egresses.
 
 `port_l2mc_set()` writes one such entry. The SMI layout is the L2 multicast
 variant of the table entry:

doc/lacp.md

@@ -0,0 +1,155 @@
+# LACP (Link Aggregation Control Protocol, IEEE 802.3ad)
+
+A static link aggregation group works only if both ends are configured for the
+same links, and nothing checks that they are. LACP lets the two ends agree on
+which links reach the same partner, aggregates only those, and takes a link out
+of the group when the partner stops answering.
+
+The switch runs LACP in software on the 8051. LACP frames belong to the Slow
+Protocols, with destination `01:80:C2:00:00:02` and ethertype `0x8809`; the
+subtype of LACP frames is 1. The implementation has the Receive, Selection and
+Mux machines of 802.3ad clause 43 and does not aggregate links whose partners
+differ. Up to `LACP_NUM_LAGS` (4) groups can run LACP; the others stay static.
+
+## Getting the frames to the CPU
+
+While LACP runs, the reserved multicast action for `01:80:C2:00:00:02` is
+*trap*, the same mechanism STP uses for its BPDUs, see [stp.md](stp.md). A
+trapped LACPDU has no VLAN tag after the RTL tag, and `TRAP_RX_BODY()` gives the
+offset of the LACPDU in the received frame. When the last group leaves LACP,
+the action goes back to *drop*.
+
+## What has to match across the members
+
+802.3ad requires that every link in an aggregate be full duplex and run at the
+same speed. Beyond the standard, anything that changes how a frame is treated
+has to be identical across members, because the hash picks the member per frame
+and a difference would make the behaviour depend on that choice:
+
+* must match: speed and duplex, VLAN membership and pvid, ingress filtering,
+  port isolation
+* may differ: EEE, LED configuration, counters, cable diagnostics
+
+None of this is enforced yet — `port_lag_members_set()` writes a member mask and
+a hash setting and validates nothing. See #377 for the discussion of whether a
+mismatch should be rejected or applied to the whole group.
+
+## Protocol timing
+
+The timer values are in units of the LACP timer, which steps every four ticks
+of the system clock, 50 times per second:
+
+| Constant             | Value | Time |
+|----------------------|-------|------|
+| `LACP_FAST_PERIODIC` | 50    | 1 s  |
+| `LACP_SLOW_PERIODIC` | 1500  | 30 s |
+| `LACP_SHORT_TIMEOUT` | 200   | 4 s  |
+
+802.3ad specifies a fast periodic time of 1 s with a short timeout of 3 s, and a
+slow periodic time of 30 s with a long timeout of 90 s. The switch transmits at
+the rate the partner asks for with its TIMEOUT bit. Its own timeout is always
+the short one, one second above the standard, so a partner that stops sending
+while its link stays up leaves the trunk after 4 s.
+
+The Partner block of a transmitted LACPDU repeats the partner's actor
+information as it was last received: system priority, system, key, port
+priority, port and state. A Linux partner sets our SYNC only when this block
+matches what it sent.
+
+## LACP API
+
+```
+void lacp_init(void) __banked;
+void lacp_off(void) __banked;
+void lacp_lag_set(uint8_t lag, __xdata uint16_t ports) __banked;
+void lacp_in(void) __banked;
+void lacp_timers(void) __banked;
+void lacp_show(void) __banked;
+```
+
+`lacp_init()` has to run before the startup configuration replays, because xdata
+is not zeroed and the port-to-LAG map uses `0xff` as "no LAG"; a stray zero would
+make a port look like it belongs to LAG 0.
+
+## LACP configuration on the serial console
+
+A group is put under LACP by adding `lacp` to the `lag` command:
+
+```
+lag 1 lacp 7 8      # ports 7 and 8 become LACP candidates of group 1
+lag 1 lacp off      # group 1 leaves LACP, its trunk is cleared (so does "lag 1 lacp")
+lag 1 d             # group 1 disabled, whether it ran LACP or not
+lag 1 7 8           # plain static aggregation, no protocol
+```
+
+A group put under LACP gives up the static trunk it had: the hardware trunk is
+cleared and rebuilt from the ports that converge. The other way round, `lag 1 d`
+and a static member list take the group out of LACP before they touch the
+trunk. A port named in another LACP group moves to the new one. The protocol
+starts with the first group put under LACP and stops when the last one is taken
+out; there is no separate switch for it. Two commands act on the whole engine:
+
+```
+lacp off            # every group leaves LACP, protocol stopped
+lacp show
+```
+
+`lacp show` prints the engine state, then one line per group with its candidate
+ports, the elected aggregator and the trunk members actually programmed into the
+hardware, then one line per port with the actor and partner state bytes and the
+receive-machine state.
+
+## LACP configuration via the Web Interface
+
+Each group on the LAG page has a Static/LACP mode next to its member ports;
+for an LACP group it shows the elected aggregator and the active members. Below
+the groups a table has one row per LACP port with its group, actor and partner
+state, receive state, LACPDU count and partner system. The page reads
+`/lacp.json` every three seconds while it is open. The member checkboxes of a
+static group come from the hardware registers in `/lag.json`.
+
+## A test with a Linux 802.3ad bond
+
+Two ports of the switch to two interfaces of one host:
+
+```
+ip link add bond0 type bond mode 802.3ad
+ip link set enp1s0 down && ip link set enp1s0 master bond0
+ip link set enp2s0 down && ip link set enp2s0 master bond0
+ip link set bond0 up
+```
+
+On the switch, with those two ports as the group:
+
+```
+lag 1 lacp 7 8
+```
+
+`lacp show` should then give both ports an actor state of `0x3f` (ACTIVITY,
+TIMEOUT, AGGREGATION, SYNC, COLLECTING and DISTRIBUTING), a partner state with
+SYNC set, and the partner's system MAC. `lag show` lists the trunk members as
+programmed in the hardware and should show both ports.
+
+On the host, `/proc/net/bonding/bond0` should show the same `Aggregator ID` for
+both slaves; different IDs mean that each link is in a group of its own. Both
+churn states should read `none`.
+
+Further checks:
+
+* `tcpdump -i enp1s0 ether proto 0x8809` on each slave shows the LACPDUs of the
+  switch and of that slave, never those of the other slave.
+* The host log has no `illegal loopback`.
+* The three reserved bytes after the actor state in a LACPDU from the switch
+  are zero.
+
+When one slave goes down, the group narrows to the remaining member, and the
+slave rejoins after it comes back.
+
+## Known limits
+
+* No Churn Detection machines and no Marker protocol.
+* Coupled Mux: collecting and distributing are enabled together.
+* One aggregator per group.
+* Selection compares the partner's system only, not its key: two aggregates
+  that one partner system offers on the same group land in one trunk here.
+* Member compatibility is not validated; see #377.

doc/link_aggregation.md

@@ -227,3 +227,7 @@ iperf Done.
 ```
 As you can see, the total throughput was 4.71 GBit/sec which is close to the
 maximum possible with a single 5GBit link.
+
+## LACP
+
+Dynamic aggregation with LACP is described in [lacp.md](lacp.md).

doc/ntp.md

@@ -0,0 +1,33 @@
+# Time
+
+The switch has no battery backed clock. It counts seconds from power on, and
+it can take the date and time from an NTP server once the network is up. The
+time is lost at every restart and set again at the first synchronisation.
+
+```
+ntp server <name|ip>     # default pool.ntp.org
+ntp interval <1-1440>    # minutes between two synchronisations, default 60
+ntp timezone <+hh:mm>    # offset from UTC, -12:00 to +14:00
+ntp dst off|eu|us        # daylight saving rule
+ntp on
+ntp off
+ntp                      # settings and time
+time                     # local date and time
+```
+
+The client speaks SNTP version 4 and synchronises at once when turned on,
+then after every interval. A server given by name is resolved through the DNS
+resolver before every synchronisation. An answer from an unsynchronised
+server is ignored, and a failed attempt is repeated after 30 seconds.
+
+The local time is UTC plus the time zone offset, plus one hour while daylight
+saving time applies:
+
+* `eu`: from 01:00 UTC on the last Sunday of March to 01:00 UTC on the last
+  Sunday of October;
+* `us`: from 02:00 local time on the second Sunday of March to 02:00 local
+  time on the first Sunday of November.
+
+The same settings, with the current local time and the last synchronisation,
+are in the Time card of the System page. Settings made there or on the
+console are kept only once the configuration is saved.

doc/storm_control.md

@@ -0,0 +1,54 @@
+# Storm Control
+Storm control limits the rate at which a port admits broadcast, multicast,
+unknown unicast and unknown multicast frames. Frames of a type above its limit
+are dropped at ingress, so a loop or a misbehaving host cannot flood the rest
+of the network through that port. Known unicast traffic is not affected.
+
+## Commands
+```
+storm <port> bcast|mcast|ucast|umcast <rate> pps
+storm <port> bcast|mcast|ucast|umcast <rate> kbps
+storm <port> bcast|mcast|ucast|umcast off
+storm
+```
+`ucast` is unicast to a MAC address the switch has not learned yet, and
+`umcast` is multicast without an entry in the address table. Multicast groups
+get their entries from IGMP snooping, so without `igmp on` all multicast is
+unknown and only the `umcast` limit applies to it; `mcast` covers the groups
+IGMP has learned. A limit in
+packets per second goes from 1 to 1048575, a limit in kbit/s from 1 to
+10000000. `storm` without arguments prints the limits of every port. The
+limits are kept in the configuration like any other setting, and the Bandwidth
+page of the web interface has a Storm control card to set them.
+
+## Registers
+Each port and type has an enable bit and a meter index, and each limit uses
+one of the 64 shared meters of the switch. RTLPlayground gives every port and
+type its own meter, index `24 + port * 4 + type`, so one port's storm does not
+eat into the limit of another. Storm control uses meters 24 to 63 and leaves
+0 to 23 to the ACL. The limit of a meter is in `RTL837X_METER_RATE + 4 *
+meter`, and bit `meter % 32` of `RTL837X_METER_MODE + 4 * (meter / 32)`
+selects pps instead of kbit/s.
+```
+#define RTL837X_STORM_CTRL		0x54e4	/* + 4 * type, bit = port */
+#define RTL837X_STORM_MIDX		0x54f4	/* + 8 * type + 4 * (port / 5), 6 bits per port */
+#define RTL837X_METER_RATE		0x5cf0	/* + 4 * meter, 24 bits, kbit/s or pps */
+#define RTL837X_METER_MODE		0x5ef0	/* + 4 * (meter / 32), bit set = pps */
+```
+In kbit/s mode one step of the rate is 1 kbit/s. Measured with 100 and 1000 byte
+frames, the meter counts a frame with its FCS, but without preamble and
+interframe gap. Every meter has a bit that adds them, in `0x5f08 + 4 * (meter
+/ 32)`, which storm control leaves cleared. `IGBW_INC_IFG` and `EGBW_INC_IFG`
+are the matching bits of the bandwidth control.
+
+An ingress bandwidth limit on the same port works on the same arriving frames
+as the storm limit rather than after it, and the two together pass far less
+than either one: a 1 Mbit/s `umcast` limit with a 10 Mbit/s ingress limit let
+0.12 to 0.16 Mbit/s of a 1.4 Gbit/s stream through, where each alone passed
+1.17 and 10.2 Mbit/s.
+
+The type is 0 for broadcast, 1 for multicast, 2 for unknown unicast and 3 for
+unknown multicast. The burst size of a meter (0x5df0 + 4 * meter) counts packets
+in pps mode; its default of 0x2000 lets 8192 frames through before a pps limit
+bites, so a pps limit sets the burst to one second of traffic. A kbit/s limit
+puts back the default.

doc/stp.md

@@ -35,28 +35,60 @@ BPDUs are addressed to `01:80:C2:00:00:00`, a reserved link-local group. The
 ASIC's Reserved-Multicast action for that address decides what happens to the
 frame.
 
-Forwarding to the CPU port works normally: the 8051 sits behind an ordinary
-port of the internal switch and is an ordinary member of a forwarding mask.
-The *trap* action does not deliver to it. Its destination is an external CPU
-attached to a physical port (`cpuTag_externalCpuPort_set`, `EXT_CPU_CTRL` in
-the vendor SDK), which these boards do not populate. The ACL trap and
-redirect actions do not deliver to the 8051 either.
+While STP runs, the action is *trap*. A trap delivers to the port named in
+`EXT_CPU_CTRL` (0x6724), whose reset value 0xf names no port, and with that
+value the trapped frame is dropped at ingress. `trap_init()` sets it to 9, the
+port the 8051 sits behind, once at startup. The trap needs no per-VLAN L2
+entries, so it does not depend on which VLANs exist or on the PVIDs, and turning
+STP on no longer walks the whole VLAN table. BPDUs reach the CPU on a port that
+STP holds in blocking with either action: measured on a SWTGW218AS by sending
+BPDUs into a blocked port with forward and with trap, both ports stopped being
+edge and waited out the forward delay. The chip exempts this address from the
+ingress state check; the slow-protocols address used by LACP is not exempt.
 
-Delivery therefore uses the *forward* action, constrained to the CPU port
-by a static L2 multicast entry (`port_l2mc_set()`), one per VLAN in use:
+A trapped frame reaches the CPU in the format it came in with: unlike a
+forwarded one it has no VLAN tag inserted after the RTL tag, and the RTL tag
+carries the trap reason. `stp_in()` reads the BPDU from the offset
+`TRAP_RX_BODY()` returns, which works for both.
 
-* while STP runs, the entry's member mask is the CPU port only — BPDUs reach
-  the CPU and are not flooded to other ports, as a participating bridge
-  requires;
-* with STP off, the same entries are retargeted to all ports, restoring the
-  transparency an unmanaged switch is expected to have, so a surrounding
-  spanning tree can span *through* this device.
+With STP off, the action goes back to *forward* and a static L2 multicast
+entry per VLAN in use (`port_l2mc_set()`) targets all ports, restoring the
+transparency an unmanaged switch is expected to have, so a surrounding
+spanning tree can span *through* this device.
 
-A BPDU delivered this way is an ordinary frame to the port's ingress logic
-and passes through its acceptable-frame-type filter. BPDUs are untagged, so a
-port set to admit tagged frames only (`ingress <port>t`) never delivers one
-to the CPU. `stp_setup()` prints a warning for every STP-enabled port in that
-state.
+The port's acceptable-frame-type filter treats the two actions differently.
+BPDUs are untagged, so a port set to admit tagged frames only
+(`ingress <port>t`) drops a forwarded BPDU before it reaches the CPU, while a
+trapped one passes. Both were measured on a SWTGW218AS with `ingress 3t`:
+LACPDUs, delivered by forward, stopped arriving on port 3, and BPDUs sent into
+the same port, delivered by trap, kept arriving. A port that admits tagged
+frames only therefore takes part in STP.
+
+## Link aggregation
+
+A LAG (Link Aggregation Group) is handled like an additional port by the STP
+protocol, with its own timers and states. The switch devices support up to 4
+LAGs, which are shown alongside the physical ports in the STP status. A group
+carries its own path cost, priority, edge and guard settings, and gets its own
+row on the Spanning Tree page.
+
+```
+stp lag 1 cost 10000    # the group decides, not its members
+stp lag 1 edge off
+```
+
+LAGs are configured via the `lag` command. Once a port is a member of a LAG it
+can no longer be configured individually for STP, so `stp port <n>` on a member
+tells you which group to configure instead. Membership is re-read from the
+aggregation registers once a second, so a group changing under LACP is picked
+up without any coordination between the two.
+
+The switch hardware does not handle STP state for a LAG as a whole. State
+changes for the members have to be done by the firmware, updating every member
+port, which is possible using a single register write. BPDUs go out through the
+lowest member and carry the group's own port id. Losing one member of a live
+LAG is not a topology change; the logical port only goes down with its last
+link.
 
 Port states live in `RTL837X_MSTP_STATES (0x5310)`, two bits per port:
 `00` disabled, `01` blocking, `10` learning, `11` forwarding. A port in

doc/support-multi-language.md

@@ -1,135 +1,107 @@
 # Supporting Multiple Languages in the Web UI
 
-The firmware uses a client-side i18n approach
-all translations are stored in a single JavaScript dictionary embedded in the firmware.
-No server-side changes are needed.
+The firmware uses a client-side i18n approach: all translations are stored
+in one JavaScript dictionary embedded in the firmware. No server-side
+changes are needed.
 
 ## Architecture
 
-All translation logic lives in `html/i18n.js`. The file contains:
+All translation logic lives at the top of `html/app.js`:
 
-- A `LANG` object with one sub-object per language (`en`, `ja`, ...)
-- Language auto-detection (browser language → `localStorage` override)
-- `t(key)` — look up a translated string
-- `setLang(lang)` — switch language and update the page
-- `applyTranslation(el)` — apply translation to one DOM element
+- A `LANG` object with one sub-object per language (`en`, `ja`, `zh`)
+- Language auto-detection (browser language, overridden by the
+  `rtl_lang` key in `localStorage`, which the language selector in the
+  System section writes)
+- `t(key, vars)` looks up a translated string; `{name}` placeholders in
+  the string are replaced from `vars`, e.g. `t("v_del_q", {n: vid})`
+- `i18nApply()` applies the dictionary to the static markup once at load
 
-Translation keys are **flat strings** (no nesting). The English keys in `LANG.en` also serve as the fallback when a key is missing in another language.
+Translation keys are flat strings. The English entry is the fallback when
+a key is missing in another language, and the key itself is the fallback
+when it is missing everywhere.
+
+Changing the language stores the choice and reloads the page, so every
+dynamically built table is rebuilt in the new language.
+
+`login.html` is served before authentication and does not load `app.js`;
+it carries its own five strings per language in an inline table.
 
 ## How to Add a New Language
 
-### 1. Add a dictionary entry in `html/i18n.js`
+### 1. Add a dictionary entry in `html/app.js`
 
-Append a new sub-object to the `LANG` object. Every key from `LANG.en` must be present:
+Append a sub-object to `LANG`. Every key from `LANG.en` should be present:
 
 ```js
-var LANG = {
-  en: {
-    nav_overview: 'Overview',
-    nav_port_config: 'Port Configuration',
-    // ... all keys for English
-  },
-  ja: {
-    nav_overview: '概要',
-    nav_port_config: 'ポート設定',
-    // ... all keys for Japanese
-  },
-  LANGCODE: {            // ← add your language here
-    nav_overview: '...',
-    nav_port_config: '...',
-    // ... translate every key
-  },
+var LANG={
+en:{
+nav_dash:"Dashboard",
+...
+},
+LANGCODE:{            // add your language here
+nav_dash:"...",
+...
+}
 };
 ```
 
-### 2. Add the language to the navigation sidebar
+### 2. Add the language to the selector in `html/index.html`
 
-In `html/navigation.js`, add an `<option>` to the language selector:
-
-```js
-+ "<option value='en'>English</option><option value='ja'>日本語</option>"
+```html
+<select class="ctl" id="langSel" data-i18n-t="sy_lang">
+  <option value="en">English</option>
+  <option value="ja">日本語</option>
+  <option value="zh">中文</option>
+  <option value="LANGCODE">Native name</option>
+</select>
 ```
 
-Replace with:
+### 3. Add the login strings in `html/login.html`
 
-```js
-+ "<option value='en'>English</option><option value='ja'>日本語</option><option value='LANGCODE'>Native Name</option>"
-```
+The inline table there holds, per language: page title, subtitle,
+password label, button label, wrong-password message.
 
-### 3. Verify auto-detection
+### 4. Verify auto-detection
 
-The language detection code in `i18n.js` reads `navigator.language` and normalises it to the first two characters:
-
-```js
-var browser = (navigator.language || navigator.userLanguage || 'en').substring(0, 2);
-return LANG[browser] ? browser : 'en';
-```
-
-If the two-letter code matches a key in `LANG`, it will be auto-selected. No changes needed here.
+The detection normalises `navigator.language` to its first two
+characters and selects that language if `LANG` has it, otherwise English.
 
 ## Two Translation Mechanisms
 
-### (A) `data-i18n` attribute (declarative — for HTML)
-
-Add `data-i18n="key_name"` to any HTML element. The English text goes in the element content as a fallback:
+### (A) `data-i18n` attributes (declarative, for HTML)
 
 ```html
-<h1 data-i18n="port_heading">Port Configuration</h1>
-<input type="button" data-i18n="port_apply" value="Apply">
-<option data-i18n="port_auto">Auto</option>
-<title data-i18n="port_title">Port Configuration</title>
+<h2 data-i18n="pt_title">Port configuration</h2>
+<button class="ctl" data-i18n="c_apply" data-i18n-t="sy_dhcp_t">Apply</button>
+<input class="in" data-i18n-p="l2_filter">
 ```
 
-On page load, `applyTranslation()` sets:
+- `data-i18n` sets the element's text content
+- `data-i18n-t` sets the `title` attribute (tooltip)
+- `data-i18n-p` sets the `placeholder` attribute
 
-- `el.value` for `<input type="submit|button">`
-- `el.textContent` for `<option>`, `<title>`
-- `el.innerHTML` for everything else
+The English text stays in the markup as the fallback and as the source
+of truth for what the key means.
 
-### (B) `t('key')` call (imperative — for JavaScript strings)
-
-When generating HTML or text in JavaScript, wrap translatable strings with `t()`:
+### (B) `t("key")` calls (imperative, for JavaScript)
 
 ```js
-td.appendChild(document.createTextNode(t('common_port') + i));
-td.innerHTML = t('port_auto');
-iHTML += "<tr><td>" + t('port_vendor') + "</td></tr>";
+tr.insertCell().textContent=t("c_port")+" "+p;
+toast(t("v_loaded",{n:vid}),"ok");
 ```
 
-### Special Case: Link Speed Display
-
-The `linkS` array in `html/main.js` maps numeric link states to display strings. The first two entries (`speed_disabled`, `speed_down`) use `t()` for translation; the remaining entries are static literals (they are the same in all languages):
-
-```js
-const linkS = [
-  function(){return t('speed_disabled')},
-  function(){return t('speed_down')},
-  "10M", "100M", "1000M", "500M", "10G", "2.5G", "5G"
-];
-function linkText(idx) { var v = linkS[idx]; return typeof v === 'function' ? v() : v; }
-```
-
-Always use `linkText(idx)` (not `linkS[idx]`) to read these values.
+Strings that are identical in every language (`10M`, `2.5G`, `MAC`,
+`VLAN`, `CPU`, `RSTP`) are literals, not dictionary entries.
 
 ## Size Considerations
 
-- `html/i18n.js` is embedded in the firmware filesystem (~14 KB for two languages)
-- Each new language adds roughly the same number of bytes as the English dictionary (~3–4 KB)
-- The firmware binary is padded to 512 KiB, so a few extra KB do not change the flash footprint
-- Values that are identical in all languages should be inlined as literals rather than added to the dictionary (e.g., `"10M"`, `"2.5G"`, `"MAC"`, `"VLAN"`, `"CPU"`)
+- Each language adds roughly 3 KB to `app.js` before compression; the
+  build gzips the file, so the flash cost of a language is closer to 1 KB
+- The largest embedded file must stay below 64 KB after compression
+  (`uint16_t` length in the file table); `app.js` with three languages
+  is about 29 KB compressed
 
 ## Build
 
-No special flags are needed. The `html/` directory is embedded by `fileadder` during the build.
-
-## Script Load Order
-
-`i18n.js` must be loaded after `main.js` (which defines `t()`'s dependencies like `LANG`) but before any page-specific JS that calls `t()`:
-
-```html
-<script src="/main.js"></script>
-<script src="/i18n.js"></script>
-<script src="/eee.js"></script>    <!-- uses t() -->
-```
-
-The `navigation.js` script is loaded last (bottom of `<body>`).
+No special flags are needed. `html/` is minified and embedded by the
+normal build (see `webui-compression.md`).

doc/supported_devices.md

@@ -4,34 +4,55 @@ The following devices have been tested and are fully working:
 | Brand    | Type            | Managed | PCB                                                                       | Flash | Ports |
 |----------|-----------------|---------|---------------------------------------------------------------------------|-------|-------|
 | Ampcom   | WAM902-SWTG018AS| No      | [SWTG018AS-A V2.0](devices/SWTG018AS_A_V2_0.md)                           |       | 8 + 1 |
+| Binardat | 2G06-04210GSM   | No      | [PCB-SWTG024AS-A-2.0.1](devices/ONT-S207CW-62TS-SE.md)                    | 0.5M  | 4 + 2 |
 | Davuaz   | Da-K6501W       | No      | [PCB-K0501W-V2.0](devices/K0501W_V2_0.md)                                 |       | 5 + 1 |
 | FOXNEO   | FNS-1200P       | No      | [PCB-K0402W-U13-V2.0](devices/FNS-1200P.md)                               | 2M    | 4 + 2 |
-| Hisource | Hi-K0402WS      | No      | [PCB-K0402WS-V3.0](devices/PCB-K0402WS-V3.0.md)                           |       | 4 + 2 |
+| Goalake  | IG108           | No      | [PB-2132](devices/PB-2132.md)                                             | 4M    | 8 + 1 |
+| GoodTop  | GT-ST018M       | Yes     | [SWTG018AS-A V2.0](devices/SWTG018AS_A_V2_0.md)                           | 2M    | 8 + 1 |
+| GoodTop  | GT-ST024M       | Yes     | [SWTG024AS-A V2.0.1](devices/SWTG024AS-A-V2.0.1_4C_2SFP.md)               | 2M    | 4 + 2 |
+| Guangnianwei | 7018-2.5G   | No      | [F7008-2.5GPOE-V1.2](devices/F7008-2.5.md)                                | 8M    | 8 + 1 |
+| Hasivo   | S1100WP_8GT_1SX_SE | Yes  | [S1100WP_8GT_1SX_SE-V1.05](devices/HASIVO_S1100WP_8GT_1SX_SE.md)          | 2M    | 8 + 1 |
 | Hisource | Hi-K0402WS      | No      | [PCB-K0402WS-V2.0](devices/PCB-K0402WS-V2.0.md)                           | 2M    | 4 + 2 |
+| Hisource | Hi-K0402WS      | No      | [PCB-K0402WS-V3.0](devices/PCB-K0402WS-V3.0.md)                           |       | 4 + 2 |
 | Hisource | Hi-K0801WS      | No      | [PCB-KO801W-V2.0](devices/HI-K0801WS.md)                                  |       | 8 + 1 |
+| hongyavision | HYWS-SGT0204S | No    | [2G5F4_10G2_V1.01](devices/2G5F4_10G2_V1.01.md)                           | 0.5M  | 4 + 2 |
 | hongyavision | LG-SG5T1    | No      | [PCB-SWTG024AS-V2.0_16895](devices/SWTG024AS-V2.0.md)                     | 0.5M  | 5 + 1 |
+| Horaco   | HC-SWTG124AS    | Yes     | [SWTG024AS-A-V2.0.1](devices/SWTG024AS-A-V2.0.1_4C_2SFP.md)               | 2M    | 4 + 2 |
 | Horaco   | HC-SWTGW215AS   | Yes     | [SWTG024AS-A-V2.0.1_19650 5C 1SFP](devices/SWTG024AS-A-V2.0.1_5C_1SFP.md) | ?     | 5 + 1 |
 | Horaco   | HC-SWTGW218AS   | Yes     | [SWTG018AS-A V2.0](devices/SWTG018AS_A_V2_0.md)                           |       | 8 + 1 |
+| Horaco   | HC-SWTGW218AS   | Yes     | [SWTG118AS-V2.1](devices/SWTGW218AS.md)                                   | 2M    | 8 + 1 |
 | Horaco   | ZX310S-4T2XH    | Yes     | [PCB-SL310S-4T1T1X-V1.0.1-24107](devices/ZX310S-4T2XH.md)                 | 2M    | 5 + 1 |
 | Horaco   | ZX310S-4T2XT    | Yes     | [PCB-SL310S-4T2XT-V1.0.0-22273](devices/ZX310S-4T2XT.md)                  | 2M    | 6     |
 | Horaco   | ZX-SG4T2        | No      | [SWTG024AS-A-V2.0.1_19650_4C_2SFP](devices/SWTG024AS-A-V2.0.1_4C_2SFP.md) | 0.5M  | 4 + 2 |
 | Horaco   | ZX-SWTG124AS    | Yes     | [SWTG024AS-v2.0](devices/SWTG024AS.md)                                    |       | 4 + 2 |
-| Keeplink | KP-9000-6XH-X2 / KP-9000-6XHML-X2 | No/Yes | [2M-PCB43-V1.2 / V2.1](devices/KP-9000-6XH-X2.md) |       | 4 + 2 |
+| keepLINK | KP-9000-6XH-X2 / KP-9000-6XHML-X2 | No/Yes | [2M-PCB43-V1.2 / V2.1](devices/KP-9000-6XH-X2.md)        |       | 4 + 2 |
 | keepLINK | KP-9000-9XHML-X | Yes     | [2M-PCB23-V2.2](devices/2M-PCB23-V2_2.md)                                 | 2M    | 8 + 1 |
 | keepLINK | KP-9000-9XHML-X | Yes     | [2M-PCB23-V3.1](devices/2M-PCB23-V3_1.md)                                 | 2M    | 8 + 1 |
+| keepLINK | KP-9000-9XHML-X | Yes     | [2M-PCB23-V3.2](devices/2M-PCB23-V3_2.md)                                 | 2M    | 8 + 1 |
+| keepLINK | KP-9000-9XHP-X  | No      | [2M-PCB51-ML-V1.1](devices/2M-PCB51-ML-V1_1.md)                           | 4M    | 8 + 1 |
+| LIANGUO  | HYWS-SGT0108S   | No      | [2G5F_20G_V1.01](devices/HYWS-SGT0108S.md)                                | 0.5M  | 8 + 1 |
 | LIANGUO  | SWTG024AS       | No      | [SWTG024AS-v2.0-17452](devices/SWTG024AS.md)                              | 0.5M  | 4 + 2 |
-| Lianguo  | ZX-SWTGW215AS   | Yes     | [PCB-SWTG115AS-V2.0 / PCB-SWTG115AS-V2.1](devices/SWTGW215AS.md)          | 2M    | 5 + 1 |
-| Mokerlink| 2G040210GSM     | Yes     | [2M-PCB43-V1.1](devices/2M-PCB43-V1.1.md)                                 |       | 4 + 2 |
-| Mokerlink| ZX-SWTGW218AS   | Yes     | [SWTG118AS-V2.0-16029](devices/SWTGW218AS.md)                             | 2M    | 8 + 1 |
+| LIANGUO  | ZX-SWTGW215AS   | Yes     | [PCB-SWTG115AS-V2.0 / PCB-SWTG115AS-V2.1](devices/SWTGW215AS.md)          | 2M    | 5 + 1 |
+| MokerLink| 2G040210GSM     | Yes     | [2M-PCB43-V1.1](devices/2M-PCB43-V1.1.md)                                 |       | 4 + 2 |
+| MokerLink| POE-2G080110GS  | No      | [2M-PCB51-ML-V1.1](devices/2M-PCB51-ML-V1_1.md)                           | 4M    | 8 + 1 |
+| MokerLink| ZX-SWTGW218AS   | Yes     | [SWTG118AS-V2.0-16029](devices/SWTGW218AS.md)                             | 2M    | 8 + 1 |
+| MokerLink| 2G08110GSN      | No      | [SWTG018AS-A V2.0](devices/SWTG018AS_A_V2_0.md)                           | 0.5M  | 8 + 1 |
+| Netcore  | GS9             | No      | [PB-2132](devices/PB-2132.md)                                             | 4M    | 8 + 1 |
+| ONT      | ONT-S207CW-62TS-SE | No   | [PCB-SWTG024AS-A-2.0.1](devices/ONT-S207CW-62TS-SE.md)                    | 0.5M  | 4 + 2 |
+| RealHD   | PSW8-25MG       | Yes     | [PCB-AFGW SWTG218ANS V2.0.2](devices/PCB-AFGW_SWTG218ANS_V2.0.2.md)       | 2M    | 8 + 1 |
 | Ruiying  | RY-4GT-2SX      | No      | [FG-4GT-2SX_V2.0](devices/FG-4GT-2SX_V2.0.md)                             | 4M    | 4 + 2 |
 | Ruiying  | RY-8GT-1SX      | No      | [FG-8GT-1SX](devices/FG-8GT-1SX.md)                                       | 1M    | 8 + 1 |
-| SiscoLink | SL-G0402F      | No      | [PCB-K0402WS-V3.0](devices/PCB-K0402WS-V3.0.md)                           |       | 4 + 2 |
+| SICSOLINK| SL-G0402F       | No      | [PCB-K0402WS-V3.0](devices/PCB-K0402WS-V3.0.md)                           |       | 4 + 2 |
 | Sodola   | SL-SWTG124AS-D  | Yes     | [SWTG024AS-v2.0-17452](devices/SWTG024AS.md)                              | 2M    | 4 + 2 |
 | Steamemo | IG204-V1        | No      | [PB-2131](devices/STEAMEMO_IG204_V1.md)                                   |       | 4 + 2 |
-| TrendNet | TEG-S562        | No      | [TEG-S563/EU H/W: V1.0R](devices/TEG-S562.md)                             | 2M    | 4 + 2 |
-| Xikestore| SKS3200M-4GPY2XF| Yes     | [SWTG024AS-v1.0](devices/SWTG024AS.md)                                    |       | 4 + 2 |
+| TRENDnet | TEG-S562        | No      | [TEG-S563/EU H/W: V1.0R](devices/TEG-S562.md)                             | 2M    | 4 + 2 |
+| XikeStor | SKS3200M-4GPY2XF| Yes     | [SWTG024AS-v1.0](devices/SWTG024AS.md)                                    |       | 4 + 2 |
 | XikeStor | SKS3200-8E1X    | Yes     | [SWTG118AS-V2.1-17462](devices/SWTGW218AS.md)                             | 2M    | 8 + 1 |
+| YuanLey  | YS25-0402       | No      | [PCB-K0402WS-V1.0](devices/PCB-K0402WS-V1.0.md)                           | 2M    | 4 + 2 |
+| YuanLey  | YS25-0402       | No      | [PCB-K0402WS-V3.0](devices/PCB-K0402WS-V3.0.md)                           | 2M    | 4 + 2 |
+| YuanLey | YS25-0402M       | Yes     | [SIRIVISION_S25G2206FP_V1_1](devices/SIRIVISION_S25G2206FP_V1_1.md)       | 2M    | 4 + 2 |
 | Ztyuav   | Z-QWYT0402      | No      | [PCB-K0402WS-V3.0](devices/PCB-K0402WS-V3.0.md)                           |       | 4 + 2 |
+| ZoeRax   | HG0402XG        | No      | [HG0402XG_V1.1](devices/ZOERAX_HG0402XG_V1.1.md)                          | 1M    | 4 + 2 |
 
 For KP-9000-6XH-X2 / KP-9000-6XHML-X2 / Mokerlink 2G040210GSM devices, select
 the machine target by PCB revision. The ML/non-ML or managed/unmanaged label

doc/webui-compression.md

@@ -19,16 +19,17 @@ html/  --(minify.py)-->  output/html_min/  --(fileadder -z)-->  flash image
   served bytes, so minification alone already reduces the transfer size
   from 109,814 to 97,730 bytes, independent of gzip.
 - The Web UI is a single-page application: `index.html` holds all pages
-  as sections (`#/ports`, `#/vlan`, ...) that the sidebar switches
+  as sections (`#ports`, `#vlan`, ...) that the sidebar switches
   between via the URL hash.  `login.html` stays separate because it is
-  the authentication gate.  Only these two pages plus `main.js`,
-  `style.css`, the three SVG port images and the favicon are embedded.
-- All JavaScript (i18n, shared helpers, navigation and the per-section
-  scripts) lives in one `html/main.js` bundle that the pages load as
-  their only script, so the browser fetches it once and gzip can
-  compress across all of it.  A section's initialisation runs when the
-  section is shown, its polling intervals run only while it is visible,
-  and re-entering a section refetches the data.
+  the authentication gate and is self-contained (its own styles and
+  strings), so nothing else is served before login.  Only these two
+  pages plus `app.js` and the favicon are embedded; the stylesheet and
+  the theme bootstrap are inline in `index.html`.
+- All JavaScript (translation table, shared helpers, navigation and the
+  per-section code) lives in one `html/app.js` that the page loads as
+  its only script, so the browser fetches it once and gzip can compress
+  across all of it.  A section starts its pollers when it is shown and
+  stops them when it is left; re-entering a section refetches the data.
 - `fileadder -z` gzip-compresses every file (zlib, gzip format,
   `Z_BEST_COMPRESSION`) when it generates the file table
   (`html_data.c`/`html_data.h`) and when it embeds the files into the
@@ -71,14 +72,18 @@ JSON polling of the visible section).
 ## Notes
 
 - The files stay well below the `uint16_t` size limit of the file table
-  (largest gzip output: 17.2 KB for the merged `main.js`).
+  (largest gzip output: about 29 KB for `app.js`, three languages
+  included).
 - `fileadder` terminates the embedded files with a NUL directly after the
   content (previously at `data_read + 1`), so the `strlen()`-based size
   computation no longer depends on uninitialised buffer content.
 - `version.h` is now written by a single atomic `printf` (change from
   upstream) instead of five separate `echo`s, so a parallel build
-  (`make -j`) can no longer read a half-written copy.  It stays in
-  `.PHONY` so it is regenerated on every build.
+  (`make -j`) can no longer read a half-written copy.  It depends on
+  `FORCE`, so its recipe runs on every build, but the file is only
+  rewritten when its content changes.  The minified copy in
+  `output/html_min/` is redone only when a web file, the list of web
+  files or `tools/minify.py` changes.
 
 ## Verification