BlackMark
996453c2a1
Two findings from the flash sweep, both measured and neither taken, because each is a decision rather than a defect. Narrowing uptime::millis() and its three timestamps to 32 bits is 608 B, 7.6 % of the image, and 16 B of RAM. The reason it is not a free win is not the display: a 32-bit millisecond counter wraps every 49.7 days on a board that runs continuously, and the interval tests would have to become subtractions, since `now - last >= interval` survives a wrap where `now >= last + interval` does not - which is how terminal.hpp's monitor tick is written today. The 64-bit counter is what puts the wrap out of reach, so this is a question about a wrapping uptime display, and the answer decides 608 B. The 1 kHz tick's own prologue is the smaller one. The handler increments 64 bits, so it calls __adddi3_s8, and a call in a signal handler decides the prologue - twelve push/pop pairs for what the helper might clobber. Two 32-bit halves remove the call and keep the range, taking the handler from ~47 instructions to ~26 with the carry running once every 49.7 days. Not taken either: it costs 66 B of flash to buy about 0.4 % of the CPU, and nothing here is timing-critical. This repo had nowhere to record work, so it has a tracker now. Docs only; the image is unchanged at 8004 B and the suite is green. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
fantemp
v2.2. Temperature-controlled fan firmware (ATmega328P, 16 MHz), rewritten on
libavr: thermistor on ADC0 sampled
free-running and averaged over 1000 conversions, fan on OC0B at 50 kHz,
115200 Bd serial console (help lists the commands), temperature
histogram persisted to EEPROM, and a direct jump into a boot-section
bootloader at 0x7e00.
The EEPROM format is the legacy firmware's, unchanged: 100 little-endian
uint32 buckets at address 0, one per °C. A board carrying years of history
from FanTemp 1.8b keeps every count — verified on hardware, all 67 non-empty
buckets byte-identical across the conversion.
The console
Commands may be abbreviated to any unambiguous-by-order prefix, as the legacy
firmware allowed: up is uptime, st is statistics, sa is save. The
table order resolves ties, so s is show — and reset is deliberately the one
command that cannot be abbreviated, because r should not be able to wipe the
histogram. save (new) forces a writeback, which otherwise happens every 30
minutes and on the way into the bootloader.
show, statistics and the histogram print one value per line behind a dotted
label, the way the original did — a run-on line is fine for one reading and
unreadable when monitor emits one a second. curve walks every whole degree
from 10 to 60 with a bar, because the curve is a cubic and five-degree samples
without a graph show none of its shape.
Ctrl+C abandons a half-typed line and gives a fresh prompt, echoing ^C, and
it is what stops monitor. Stopping on any byte, which is what the port did
first, reads well right up until a host sends a line ending: monitor\r\n then
stopped itself on the \n it arrived with, one reading in.
Reaching the bootloader
bootloader jumps; it does not reset. That is not a style choice:
- pureboot hands straight back on WDRF, by design — an unattended board that watchdog-resets in a loop must not sit in a loader. So the legacy watchdog-reset hand-over arrives and opens no window at all, and on a board with no reset line that is a board that cannot be reflashed.
- The address is
0x7e00, the top 512 bytes. The legacy firmware used0x7800, a 2 KB boot section's base, which on a board with a 512-byte boot section reads erased — so itsbootloadercommand silently never arrived anywhere. UCSR0Bis cleared first. WhileTXEN0is set the USART owns PD1, so a loader that bit-bangs the same pin receives perfectly and answers into nothing.
ctest reads all three back out of the emitted image (test/check_reachability.py),
because none of them is visible from the source alone and the failure mode is an
unreflashable board. Both the address and the watchdog checks are red-proven
against the legacy behaviour they exist to catch.
The Steinhart–Hart math of the legacy firmware (runtime doubles + libm log) is gone: the Beta equation and the cubic fan curve are evaluated consteval into flash tables — the firmware itself never touches floating point.
libavr rides as the libavr/ submodule, pinned to the commit this firmware
builds against; LIBAVR_ROOT (cache or environment) overrides it for
development against a working tree:
git submodule update --init libavr
cmake --preset atmega328p-generated
cmake --build --preset atmega328p-generated
The firmware is 8004 B of flash, byte-identical between the generated and
reflect modes, and ctest holds it to that number.
Atmel Studio
master carries a Studio solution, so this branch does too: ide/fantemp.atsln
builds the same firmware — byte-identical .text and .data to the CMake
build — from the same sources, with the flags mirrored by hand.
Studio finds libavr in the submodule, at
$(MSBuildProjectDirectory)\..\libavr\include — correct by construction, and
anchored to the project rather than written relative to the generated makefile,
which runs from the configuration's output directory and would need a different
number of ... Unlike the CMake build there is no LIBAVR_ROOT to point
elsewhere: an environment variable set in a shell is not visible to Studio
launched from the Start menu — which is what the submodule answers.
It also needs a GCC 16.1 toolchain registered as flavour avr-g++-16.1.0;
nothing older can compile -std=c++26.
One generated file is required before the project will load, and one command
checks the flags have not drifted (both from libavr's tools/atmelstudio/):
python ../libavr/tools/atmelstudio/componentinfo.py \
ide/fantemp.componentinfo.xml --device ATmega328P
python ../libavr/tools/atmelstudio/check-flags.py --solution ide/fantemp.atsln \
--compile-commands build/atmega328p-generated/compile_commands.json \
--log build/atmelstudio.log
CMake remains the build system; the solution is there so the project opens in
Studio as its predecessor did. Only the Release configuration is gated against
CMake — the presets define no debug build — and Debug carries the -Og -gdwarf-4 pair libavr's own debug preset uses.
Legacy (yazoalfa submodules) stays on master.