Compare commits

...

22 Commits

Author SHA1 Message Date
78c9071f0c deps: the pins advance to the identity check that reads its own sibling
libavr 93d8b0e, pureboot 26a8621. The check is handed the sibling build tree now and asks its build system whether the image is current, so a comparison against a tree the run did not build skips rather than passing. Gated at the pin. Its own gate builds both modes, so the comparison still happens.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-03 08:23:59 +02:00
d6673db131 deps: the pins advance, and this repository gets a committed gate
libavr moves to its USI I2C rate accounting, the two override surfaces the
fan-controller port filed - a PWM solve that can be pinned, a capture edge that
is a value - and the shared preset loop's own fix. pureboot moves with it. This
firmware names none of them and no image moved.

`tools/check.sh` is new, and its absence was the defect: there was no committed
gate at all, so "run the suite" was a snippet somebody remembered - and the
order in a snippet is the one thing nobody re-derives. Every preset built
before any is tested is subtle and load-bearing, because the mode-identity
check reads the sibling mode's tree and a build-then-test-per-preset run holds
a fresh image against a stale sibling. The loop is libavr's
`tools/check-presets.sh`.

Two defaults stop being the editing machine's. `.vscode/settings.json` named
`D:/dev/libavr/local/toolchain/avr-gcc-16.1.0-mingw`, which is a path true of
one machine (libavr guidance rule 50) and, since the in-repo toolchain copies
went, true of none - `local/machine.cmake` is where a checkout says that and it
already did. And `.gitattributes` gains `* text=auto eol=lf`: naming the source
extensions left Markdown, Python, shell and CMake to whatever wrote them, which
on the Windows side is CRLF, and a rewrite there buries the change it made.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-03 03:53:43 +02:00
fb82f0f7a9 deps: the pins advance, and a missing interpreter skips rather than vanishes
libavr and pureboot move to the tips that register their checks everywhere.
fantemp.reachability stops being wrapped in `if(Python3_FOUND)`: it is
registered on every host now and skips where the interpreter is absent, so what
`ctest -N` counts is a fact about this repository rather than about the machine
that configured it.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-02 21:37:44 +02:00
88df1c1ada deps: libavr and pureboot to their tips, both still matching the board
libavr's commit is the compare-debug tool's Windows command line and its test;
pureboot's is that pin advancing. Re-checked against the hardware rather than
only against the previous build: the loader this repo produces is still
byte-for-byte the 386 B the board's slot holds, and the firmware still the
8004 B it is running.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-02 10:48:38 +02:00
fc65457de3 deps: pureboot to its tip, whose own libavr pin is what moved
Nothing in pureboot's device code changed - the commit is its libavr pin
advancing eleven, measured across its twenty-one loader variants as
byte-identical. Re-checked from this side too: the loader this repo builds is
still byte-for-byte the 386 B the board's slot holds, and the firmware still
the 8004 B it is running.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-02 10:18:50 +02:00
11da387ffa deps, build: the pins catch up, and the loader is built here now
Fifty-two libavr commits and twenty-seven of pureboot behind, which is far
enough that "it still builds" is not the interesting part. It builds, both
modes, byte-identical across them, five checks green - and the image is
**byte-for-byte the 8004 B the board is running**, so the catch-up costs this
deployment nothing and a redeploy was ruled out by comparison rather than
skipped by assumption.

The loader is the gap that mattered. pureboot rode as a submodule for its
geometry alone, so the commit that pinned this firmware did not build the one
image this board cannot be recovered without - the same hole tempmon had.
`pureboot_add_loader(pureboot)` closes it on nothing but pureboot's own
defaults for the chip, and the 386 B it produces is byte-for-byte what the
board's slot reads back.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-02 09:33:48 +02:00
491ff76447 build: pureboot rides as a pinned submodule, and the libavr pin advances
pureboot moved to its own repo, so the loader this board hands over to stopped
being reachable through a sibling checkout of the bootloader repo. It is a
submodule here now: this board has no reset line and no programming header, so
the resident loader is the only way in, and the commit naming the firmware
should name the loader it has to reach.

The reachability check stops carrying its own copy of where that loader is.
0x7e00 was a literal beside pureboot's own geometry, which the submodule
exports as PUREBOOT_BASE_HEX - one source for the fact now, and the check reads
whichever slot the pinned loader actually has. The boot-section bound stays a
literal, being a fuse fact rather than a loader one.

Built and tested at both pins on the bench: 5/5, cross-mode identity included,
and the image deployed to the board verifies byte-for-byte through its loader.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-23 22:38:19 +02:00
e56e8b977f docs: the millisecond clock stays 64 bits, owner-refused
608 B of flash, 7.6 % of the image, and it is not available: this board runs
continuously and an uptime that restarts every 49.7 days is not acceptable,
which is what a 32-bit millisecond counter wraps at.

The tick split goes with it. It only ever made sense beside the narrowing, and
on its own it costs 66 B to buy about 0.4 % of a CPU nothing here is competing
for.

Both are recorded rather than deleted, and that is the point of the change: the
numbers are real and re-measurable, so someone will find 608 B again and read
it as an opportunity. What they need with it is the reason it was refused.

Docs only; five tests green.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-23 16:31:54 +02:00
8f43584c5c build: the libavr pin advances to the i2c fix
e697920. The change is confined to the i2c thunks' argument list, which this
firmware does not reach - 8004 B either way, and the suite is green in both modes.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-23 06:23:49 +02:00
996453c2a1 docs: what the 64-bit millisecond clock costs, measured
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>
2026-08-23 05:33:52 +02:00
0f5e40510c perf: the console's constants move to flash, -378 B and -214 B of RAM
8382 -> 8004 B of flash, 725 -> 511 B of RAM on a part that has 2048.

The command names were a std::to_array of string_view, and on a Harvard
machine that is the worst of both: the characters land in .data and so does
the table's own pointer-and-length pair for each of them, so the firmware
carried 216 B of RAM for thirteen words that never change - and paid for them
in flash too, since .data is copied out of an initialiser image at startup.

They are one NUL-separated blob in flash now, walked with lpm. Separators
rather than an offset table, because an offset table is the RAM this exists to
give back; the names sit in it in match order, so the walk that finds a name
is the same walk that compares it and measures it. Flash falls further than
RAM does: the initialiser image and the two tables were 216 B of it, and the
blob is 87.

The header said the names "cannot" be in flash because they are matched at run
time. Being matched at run time is not a reason to be in RAM on a machine with
two address spaces - only being *written* is, and nothing writes these.

Two smaller things came with it. `reset`'s exact-match rule was a bool on
every entry to protect one; it is an index found by searching the list, so
reordering the commands cannot move the protection onto a different one. And
`version` was the last string_view left, holding its own characters and a
pointer to them.

The matching is now pinned rather than assumed: lookup() is constexpr and the
battery asserts the load-bearing order the README documents - `s` is show and
not statistics, `st` is statistics, no abbreviation of `reset` resolves, and
`helpful` is not `help`. Red-checked by claiming `s` is statistics.

Both modes byte-identical, ten tests green.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-23 05:29:01 +02:00
bf18f99635 build: the libavr pin advances, and one adiw appears in the ring
43fc479 -> aec9955, 8380 -> 8382 B. Two bytes, and they are all of e226340:
the uart ring now declares its indices before its storage.

The library measured that reorder at 0 B and it is +2 here, so the difference
is worth stating. The reorder exists for the 0..63 displacement window, and at
32 entries the indices were never outside it - ldd Z+32 and ld Z are both two
bytes, so moving them to the front buys nothing. What it does do is take
storage off offset zero, and storage is the member reached by a computed
index: pop() loaded storage[tail] as X = Z + tail with the base free, and now
adds the base with an adiw.

So it is free where the indices were out of the window and a loss where they
were in it, and which of those a consumer gets depends on its ring size and on
whether pop() is out of line - fantemp's is. Filed upstream with the
disassembly; nothing to work around here, and 8382 of 32768 is not a budget
question.

Everything else crossed is inert for this firmware: no i2c, no eeprom writer,
no spare vectors, and percent_t already reached through ::of().

Five tests green in both modes, cross-mode identity held.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-23 02:20:57 +02:00
5244953654 test: the README's cross-mode claim is now checked, not just stated
The pin advances to the commit that gives consumers the gate's own image
comparison, and the claim written into this README a few commits ago - 8380 B,
byte-identical between the two modes - is registered against it. It compares
clean over every loadable section.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-16 20:50:00 +02:00
cde7e60170 test: the image's size stops being a number nobody holds
Every other port in the fleet states its image size and is held to it; this
one stated none, so a library advance could move the firmware and no check
would say a word. 8380 B is what the presets build today, byte-identical
between the two modes, and the README now says so.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-16 19:21:29 +02:00
4c351c6c9d build: the libavr pin advances over the inlined vector
The ADC vector's handler was a lambda reached through a pointer, so it sat
out of line as a 140 B thunk; it is now inlined and the vector grows 136 B,
which is the whole of the image's 8384 -> 8380 B. The layout shift that
follows costs `statistics::record` an rcall widened to a call and refunds
the same two bytes at another site.

The prologue does not shrink here: 15 push/pop pairs before and after,
because the inlined body still calls `__udivmodsi4` and that clobbers the
caller-saved set the call used to. What this board gets is the removed
rcall and ret on every conversion, not a shorter frame.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-16 19:12:51 +02:00
5974a5aa90 chore: machine-local state has a home here too
Guidance rule 15 puts every machine-local artefact under a gitignored local/,
and this repo could only have committed one.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-16 03:32:53 +02:00
ba37ecea04 test: the format and ASCII rules stop being a habit, and the stated sizes stop drifting
libavr's guidance binds this repo too, and until now nothing here checked it -
`ctest` runs `libavr_format_test()` over this tree's own sources now (rules 11
and 33), skipping rather than passing where clang-format is absent. It caught
drift on its first run: a file written this week and edited after formatting.

Where the README states a measured size, `libavr_size_claim_test()` holds it to
the image and holds the image to the prose: advancing the library pin moved
three of these across the fleet with nothing saying so, and re-recording one
now requires the sentence that quotes it to move too.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-12 23:02:40 +02:00
0f99b7787c build: the libavr pin advances to the sweep's own record
Documentation only - the guideline sweep's condensed entry, the three measured
facts about class-type constants it produced, and the port filings it left
open. No header, tool or generated input moves, so every image is untouched.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-12 17:02:56 +02:00
3e95160b00 fix: the quarter-degree rounding was wrong below zero, in three copies
`(quarters + 2) / 4` is round-half-up only for positive values: C truncates a
negative quotient toward zero, so -3.00 C read as -2, -1.00 C as 0 and -0.75 C
as 0 - nine of fifteen negative quarter-values off by a whole degree, always
toward zero. `(quarters + 2) >> 2` is an arithmetic shift, which floors, and
is right across the whole range. It is also smaller: the shift skips the bias
correction signed division needs.

The formula was written three times - thermistor::celsius(), which nothing
called, and inline at both live call sites - so the defect had three homes and
so would its fix. One `thermistor::whole_degrees()` now, called from both, and
the dead entry point is gone (rules 6, 27).

test/consteval.cpp is new and is what should have caught it: whole_degrees
across zero including both ties and both table limits, the curve's start and
saturation points and its monotonicity, and the thermistor table anchored
where the Beta equation fixes it - the count at which the divider reads the
thermistor's nominal resistance must read the nominal temperature - plus both
clamps and the fall across every step. Red-green: four assertions fire against
the old division.

Beside it: the cubic's three coefficients are named rather than inlined and
restated in prose (rule 5), the consteval table builders take explicit 32-bit
types (rule 25), the curve's clamp reads the table's own size (rule 36), and
the serial override says what expects the rate rather than what the board has
always done (rules 12, 13).

And bootloader::handle_reset()'s watchdog diversion is gone. Its own comment
called it a leftover "kept only because it is free and cannot hurt", and it
did not: pureboot peeks WDRF without clearing it and hands back on purpose, so
a watchdog reset arrives here with the flag still set and the diversion jumped
into the loader with MCUSR already cleared - opening the activation window
that policy exists to close. Clearing MCUSR is the whole job and stays.

8206 -> 8168 bytes, byte-identical between generated and reflect.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-12 16:23:05 +02:00
837b832bc7 build: the libavr pin advances past the audit sweep, and the numbers get names
The pin crosses libavr's phase-6 close and the guideline sweep behind it;
the image is byte-identical in both modes at 8206 bytes.

The port's own sweep, against the same rules. Every mutable `static inline`
takes `m_` - uptime's counter, the sampler's window, the controller's five,
the statistics histogram and the terminal's line state (rule 46; a private
`static constexpr` is a constant rather than state and keeps its bare name).
The command table is `std::to_array` and the serial config breaks one member
per line (rules 36, 40). And three numbers get the name they already had
somewhere: duty goes through `percent_t::of()` rather than a hand-built
basis-point count, the ADC's top count is `thermistor::adc_full_scale`
instead of 1023 in four places, and the two `0xffffffff` are `open_circuit`
- which was already declared five lines away - and `never_written`, which
replaces a comment explaining the literal (rules 5, 6, 41).

Measured, not assumed: rendering `adc_full_scale` into the `show` line
instead of leaving it in the message string cost 6 bytes, so the display
text stays text.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-12 14:40:33 +02:00
1489f4c3a0 editor: clangd and cmake work from a committed vscode workspace
The three files libavr's consumers carry, in the leaf-application shape:
nothing vendors this repo, so .clangd names the one build tree an editor
should read (build/atmega328p-generated) beside the stand-ins clang needs
for GCC's AVR dialect; .vscode/settings.json passes --query-driver and names
the toolchain prefix for the window that opens this folder. The libavr pin
advances to the editor-audit fixes. One residue stays red and is the
documented frontend divergence, not a defect: thermistor.hpp's consteval
Beta-equation table folds __builtin_log under GCC, which clang refuses to
fold in a constant expression -- the same class as libavr's breathe example
and its __builtin_cos.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-08-09 16:48:46 +02:00
b1caf49522 build: the libavr pin advances past phase 6
The renames land (interrupt_guard, consume_reset_cause, set_duty), the
sampler binds its input in the new converter shape (the input pack plus
in<>::start() as free-running's one kick), and the console states
.allow_baud_error = true for the 115200-at-16-MHz this board has always
spoken - the receiver-tolerance table libavr now enforces is stricter
than the rate's own +2.1 %. The loader probe reads through
avr::flash_load instead of raw pgmspace, the terminal's line buffer is
std::array with backspace and delete named, the tree is reformatted
under InsertBraces, and the sources are ASCII.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-08-09 11:45:07 +02:00
24 changed files with 709 additions and 271 deletions

View File

@@ -1,13 +1,15 @@
---
BasedOnStyle: LLVM
Standard: Latest
ColumnLimit: 120
IndentWidth: 4
TabWidth: 4
UseTab: ForIndentation
AlignEscapedNewlines: DontAlign
AllowShortFunctionsOnASingleLine: Empty
AlwaysBreakTemplateDeclarations: true
BreakTemplateDeclarations: Yes
BreakBeforeBraces: Custom
BraceWrapping:
AfterFunction: true
InsertBraces: true
...

30
.clangd Normal file
View File

@@ -0,0 +1,30 @@
CompileFlags:
# Named here rather than in the editor's settings because nothing vendors
# this repo: the one build tree is the one an editor should read.
CompilationDatabase: build/atmega328p-generated
Add:
# clang has no 24-bit integer and GCC's are keywords, not macros, so the
# editor needs a stand-in for avr::uint24_t. The next width up is the only
# one available — clang rejects _BitInt(24) on this target.
- -D__uint24=unsigned long
- -D__int24=long
# clangd forwards the driver's system includes but not its own header
# directory, so <stdint.h> resolves to avr-libc's, which still gates the
# limit and constant macros on the C++98 opt-in.
- -D__STDC_LIMIT_MACROS
- -D__STDC_CONSTANT_MACROS
# isr::emit spells a vector number into [[gnu::signal(N)]], which clang
# rejects rather than ignores — enough of them in one TU to reach the
# default limit of 19 inside the headers and truncate the parse.
- -ferror-limit=0
Remove:
# The build promotes warnings for the compiler that has to be right about
# them; in the editor the flag paints a second frontend's opinions in the
# colour reserved for things that do not compile.
- -Werror
Diagnostics:
Suppress:
# clang's AVR `signal` attribute takes no arguments and it knows none of
# progmem, naked or OS_main. A misspelling is what the build is for.
- attribute_wrong_number_arguments
- unknown-attributes

13
.gitattributes vendored
View File

@@ -1,8 +1,11 @@
*.h eol=lf
*.hpp eol=lf
*.c eol=lf
*.cpp eol=lf
.git* eol=lf
# Line endings are the repository's, not the editing machine's: this checkout
# is reached from two hosts, and a file rewritten by a Windows tool comes back
# with every line changed unless something says otherwise. Naming the source
# extensions left Markdown, Python, shell and CMake to whatever the writing
# tool defaulted to, which is CRLF on one of the two.
* text=auto eol=lf
# Atmel Studio writes these and expects them back.
*.vcxproj* eol=crlf
*.cppproj eol=crlf
*.sln eol=crlf

1
.gitignore vendored
View File

@@ -1,4 +1,5 @@
build/
local/
.cache/
# Atmel Studio: generated per machine, and its build outputs

3
.gitmodules vendored
View File

@@ -1,3 +1,6 @@
[submodule "libavr"]
path = libavr
url = ../libavr.git
[submodule "pureboot"]
path = pureboot
url = ../pureboot.git

6
.vscode/extensions.json vendored Normal file
View File

@@ -0,0 +1,6 @@
{
"recommendations": [
"llvm-vs-code-extensions.vscode-clangd",
"ms-vscode.cmake-tools"
]
}

33
.vscode/settings.json vendored Normal file
View File

@@ -0,0 +1,33 @@
{
// clangd is the language server; the cpptools engine would parse every file
// a second time and disagree, since nothing tells it about a cross
// compiler. Which database to read is in .clangd.
"C_Cpp.intelliSenseEngine": "disabled",
// --query-driver lets clangd ask the cross compiler for its own system
// includes and target; without it every standard header is missing.
"clangd.arguments": [
"--query-driver=**avr-g++*",
"--header-insertion=never"
],
// The presets are the build interface, and the toolchain file inside the
// libavr submodule is the one place the compiler is chosen. **No prefix is
// named here**: a committed file may not name a path that is true of one
// machine (libavr guidance rule 50), so the gitignored local/machine.cmake
// at this repository's root is where a checkout says where its toolchain
// is - one file, and it answers for both hosts.
"cmake.useCMakePresets": "always",
"cmake.configureOnOpen": true,
"cmake.options.statusBarVisibility": "compact",
"files.watcherExclude": {
"**/build/**": true,
"**/libavr/**": true
},
"files.associations": {
".clangd": "yaml",
".clang-format": "yaml"
}
}

View File

@@ -12,9 +12,19 @@ if(NOT LIBAVR_ROOT)
set(LIBAVR_ROOT ${CMAKE_CURRENT_SOURCE_DIR}/libavr)
endif()
if(NOT EXISTS ${LIBAVR_ROOT}/CMakeLists.txt)
message(FATAL_ERROR "libavr not found at ${LIBAVR_ROOT} run: git submodule update --init libavr")
message(FATAL_ERROR "libavr not found at ${LIBAVR_ROOT} - run: git submodule update --init libavr")
endif()
# pureboot rides as a pinned submodule too: this board's only way in is its
# resident loader, so the commit that names this firmware names the loader it
# has to hand over to. Consumed for the geometry it exports - the loader links
# the libavr target above, so pureboot's own libavr submodule stays
# uninitialised.
if(NOT EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/pureboot/CMakeLists.txt)
message(FATAL_ERROR "pureboot not found - run: git submodule update --init pureboot")
endif()
add_subdirectory(${LIBAVR_ROOT} libavr-build)
add_subdirectory(pureboot pureboot-build)
include(${LIBAVR_ROOT}/cmake/checks.cmake)
add_executable(fantemp src/main.cpp)
target_link_libraries(fantemp PRIVATE libavr)
@@ -27,5 +37,15 @@ add_custom_command(TARGET fantemp POST_BUILD
COMMAND ${CMAKE_OBJCOPY} -O ihex -R .eeprom
$<TARGET_FILE:fantemp> $<TARGET_FILE_DIR:fantemp>/fantemp.hex)
# The board's loader, built here rather than named from memory: consuming
# pureboot for its geometry alone left the one image this deployment cannot be
# recovered without outside the repository that pins it. Every parameter is
# pureboot's own default for this chip - USART0 at 115200 on the 16 MHz crystal
# `src/board.hpp` declares - and that is measured rather than assumed, the
# image being byte-for-byte the 386 B the board's slot reads back. It goes on
# over the wire (`--update-loader`), which is why there is no flash target
# beside it: no programmer has ever been in this board's path.
pureboot_add_loader(pureboot)
enable_testing()
add_subdirectory(test)

View File

@@ -66,6 +66,9 @@ 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`

40
dev/tasks.md Normal file
View File

@@ -0,0 +1,40 @@
# Tasks
This file is current work. The repo is the sole task tracker (libavr guidance
rule 20).
## Open
Nothing.
## Decided against
Both of these are measured, and both are recorded here rather than deleted so
that the next person to measure them does not read a number as an opportunity
and re-propose work the owner has already refused.
### The millisecond clock stays 64 bits
Narrowing `uptime::millis()` and the three timestamps that hold its value from
`uint64_t` to `uint32_t` is **608 B of flash, 7.6 % of the image** (8004 B down
to 7396), plus 16 B of RAM.
**Refused, owner-stated: this board runs continuously and an uptime that
restarts every 49.7 days is not acceptable.** That is what a 32-bit
millisecond counter wraps at, and the display is not the only cost - the
interval tests would have to become subtractions, since `now - last >=
interval` survives a wrap where `now >= last + interval` does not, and
`terminal.hpp`'s monitor tick is written the second way. The 64-bit counter is
what puts the wrap out of reach, which is the property being bought.
### The 1 kHz tick keeps its 64-bit increment
The compare handler increments 64 bits, so it calls libgcc's `__adddi3_s8`, and
a call inside a signal handler decides the prologue - twelve push/pop pairs for
what the helper might clobber. Splitting the counter into two 32-bit halves
removes the call and keeps the full range, taking the handler from ~47
instructions to ~26 with the carry running once every 49.7 days.
**Refused: it costs 66 B of flash to buy about 0.4 % of the CPU**, and nothing
here is timing-critical. It was only ever worth considering alongside the
narrowing above, which is refused outright.

2
libavr

Submodule libavr updated: b719ed74d8...93d8b0e491

1
pureboot Submodule

Submodule pureboot added at 26a86213ca

View File

@@ -3,7 +3,7 @@
#include <libavr/libavr.hpp>
// The board composition: every peripheral of the fan controller in one
// place. ATmega328P at 16 MHz thermistor divider on ADC0 (PC0), fan on
// place. ATmega328P at 16 MHz - thermistor divider on ADC0 (PC0), fan on
// OC0B (PD5) at 50 kHz, console on the hardware UART.
namespace app {
@@ -13,56 +13,67 @@ using dev = avr::device<{.clock = 16_MHz}>;
// Millisecond uptime from timer2 CTC (the fan owns timer0).
class uptime {
static inline volatile std::uint64_t ms = 0;
static inline volatile std::uint64_t m_ms = 0;
public:
using ticker = dev::timer2<{.frequency = 1_kHz, .on_compare = [] { ms = ms + 1; }}>;
using ticker = dev::timer2<{.frequency = 1_kHz, .on_compare = [] { m_ms = m_ms + 1; }}>;
static std::uint64_t millis()
{
avr::irq::atomic_guard lock;
return ms;
avr::irq::interrupt_guard lock;
return m_ms;
}
};
// 1000-sample averaging window fed by the conversion interrupt.
class sampler {
static inline volatile std::uint32_t sum = 0;
static inline volatile std::uint16_t count = 0;
static inline volatile std::uint16_t window = 0;
static inline volatile bool ready = false;
static inline volatile std::uint32_t m_sum = 0;
static inline volatile std::uint16_t m_count = 0;
static inline volatile std::uint16_t m_window = 0;
static inline volatile bool m_ready = false;
static constexpr std::uint16_t samples = 1000;
public:
using input = dev::adc<{.input = avr::adc::input_pin(0),
.trigger = avr::adc::trigger::free_running,
.on_conversion = [](std::uint16_t value) {
sum = sum + value;
count = count + 1;
if (count >= samples) {
window = static_cast<std::uint16_t>(sum / samples);
sum = 0;
count = 0;
ready = true;
}
}}>;
using input = dev::adc<{.trigger = avr::adc::trigger::free_running,
.on_conversion =
[](std::uint16_t value) {
m_sum = m_sum + value;
m_count = m_count + 1;
if (m_count >= samples) {
m_window = static_cast<std::uint16_t>(m_sum / samples);
m_sum = 0;
m_count = 0;
m_ready = true;
}
}},
avr::adc::input<avr::adc::input_pin(0)>>;
// The bound input, whose start() is free-running's one kick.
using thermistor = input::in<avr::adc::input_pin(0)>;
// The finished average (raw 10-bit), once per window.
static bool take(std::uint16_t &value)
{
avr::irq::atomic_guard lock;
if (!ready)
avr::irq::interrupt_guard lock;
if (!m_ready) {
return false;
value = window;
ready = false;
}
value = m_window;
m_ready = false;
return true;
}
};
using fan = dev::pwm<avr::pd5, {.frequency = 50_kHz}>;
using serial_t = dev::uart0<{.baud = 115200_Bd, .rx_buffer = 32, .max_baud_error = 2.5_pct}>;
// 115200 at 16 MHz lands +2.1 % off, past the receiver-tolerance table the
// solver holds rates to. It is the rate the console on the other end of the
// cable expects, so the override states that the miss is meant.
using serial_t = dev::uart0<{
.baud = 115200_Bd,
.rx_buffer = 32,
.allow_baud_error = true,
}>;
inline constexpr serial_t serial{};
} // namespace app

View File

@@ -1,7 +1,5 @@
#pragma once
#include <avr/pgmspace.h>
#include <libavr/libavr.hpp>
#include "board.hpp"
@@ -11,7 +9,7 @@
// The legacy firmware did this with a watchdog reset: `bootloader` armed the
// watchdog and hung, and the next boot noticed WDRF and jumped to the boot
// section. That works for TinySafeBoot and **does not work for pureboot**, which
// deliberately hands straight back to the application on WDRF an unattended
// deliberately hands straight back to the application on WDRF - an unattended
// board that watchdog-resets in a loop must not sit in a loader instead of
// running. So a reset-based route into pureboot opens no window at all, and on a
// board whose only way in is the firmware that is a lockout.
@@ -22,7 +20,7 @@
// The address is this board's, and it is not the legacy one: the loader lives in
// the top 512 bytes at 0x7e00 (`hfuse d4` puts the boot section at 0x7c00 with
// pureboot's staging slot below its own slot). The legacy firmware probed 0x7800
// a 2 KB boot section's base which on this board reads erased, so its check
// - a 2 KB boot section's base - which on this board reads erased, so its check
// was always false and its `bootloader` command never actually arrived anywhere.
namespace app {
@@ -32,14 +30,14 @@ class bootloader {
using guard = dev::watchdog<{.timeout = 16_ms}>;
// The top 512 bytes. An erased slot reads 0xffff, which is not an
// instruction any loader begins with so this asks "is a loader installed"
// instruction any loader begins with - so this asks "is a loader installed"
// rather than "is it the one I expect", which is the check the legacy
// firmware got wrong in the other direction by testing one specific byte.
static constexpr std::uint16_t base = 0x7e00;
static bool present()
{
return pgm_read_word(base) != 0xffff;
return avr::flash_load(reinterpret_cast<const std::uint16_t *>(base)) != 0xffff;
}
// A function pointer holds a word address on AVR, so the byte address
@@ -52,24 +50,22 @@ class bootloader {
}
public:
// Call first thing in main. reset_cause() reads *and clears* MCUSR, which
// matters on its own: a lingering WDRF forces the watchdog back on at its
// shortest timeout. The diversion below is a leftover of the legacy route
// and is kept only because it is free and cannot hurt — with BOOTRST
// programmed the loader has already run before this line, so nothing
// normally reaches it.
// Call first thing in main. Clearing MCUSR is the whole job: a lingering
// WDRF forces the watchdog back on at its shortest timeout, which is a
// reset loop rather than a boot. The cause is not routed on - pureboot
// peeks WDRF without clearing it and hands back on purpose, so a watchdog
// reset arrives here with the flag still set, and diverting into the
// loader on it would reopen the very window that policy closes.
static void handle_reset()
{
auto cause = avr::power::reset_cause();
avr::power::consume_reset_cause();
guard::disable();
if (cause.watchdog && present())
call(reinterpret_cast<jump_fn>(base / 2));
}
// Hand over for real: no reset, so no WDRF for the loader to refuse.
[[noreturn]] static void enter()
{
// Interrupts first the receive vector and the timer live in this
// Interrupts first - the receive vector and the timer live in this
// application's vector table, and once the loader is running there is no
// application to vector into.
avr::irq::disable();
@@ -77,7 +73,7 @@ class bootloader {
// Release the USART. While TXEN0 is set the peripheral owns PD1, not the
// port register, so a loader that bit-bangs the same pin receives
// perfectly and answers into nothing mute, not deaf, and unverifiable
// perfectly and answers into nothing - mute, not deaf, and unverifiable
// from the host. pureboot clears this itself; TinySafeBoot, which is what
// this board still carries, does not. Four bytes make the hand-over work
// for either one, which is the only reason this route can be tested

View File

@@ -6,71 +6,73 @@
#include "curve.hpp"
#include "thermistor.hpp"
// Control loop: averaged thermistor samples temperature fan duty
// Control loop: averaged thermistor samples -> temperature -> fan duty
// through the curve table (auto) or a console-set value (manual).
namespace app {
class controller {
static inline std::uint16_t adc_average = 0;
static inline std::int16_t temp_quarters = 0;
static inline std::uint8_t percent = 100;
static inline bool auto_mode = true;
static inline bool have_data = false;
static inline std::uint16_t m_adc_average = 0;
static inline std::int16_t m_temp_quarters = 0;
static inline std::uint8_t m_percent = 100;
static inline bool m_auto_mode = true;
static inline bool m_have_data = false;
public:
static void init()
{
fan::duty(avr::percent_t{10000}); // full blast until the first reading
fan::set_duty(100_pct); // full blast until the first reading
}
static void poll()
{
std::uint16_t sample;
if (!sampler::take(sample))
if (!sampler::take(sample)) {
return;
adc_average = sample;
temp_quarters = thermistor::quarters(sample);
have_data = true;
if (auto_mode)
percent = curve::duty(static_cast<std::int8_t>((temp_quarters + 2) / 4));
fan::duty(avr::percent_t{static_cast<std::uint16_t>(percent * 100)});
}
m_adc_average = sample;
m_temp_quarters = thermistor::quarters(sample);
m_have_data = true;
if (m_auto_mode) {
m_percent = curve::duty(thermistor::whole_degrees(m_temp_quarters));
}
fan::set_duty(avr::percent_t::of(m_percent));
}
static void set_manual(std::uint8_t p)
{
auto_mode = false;
percent = p;
fan::duty(avr::percent_t{static_cast<std::uint16_t>(p * 100)});
m_auto_mode = false;
m_percent = p;
fan::set_duty(avr::percent_t::of(p));
}
static void set_automatic()
{
auto_mode = true;
m_auto_mode = true;
}
static bool automatic()
{
return auto_mode;
return m_auto_mode;
}
static bool data_available()
{
return have_data;
return m_have_data;
}
static std::int16_t temperature_quarters()
{
return temp_quarters;
return m_temp_quarters;
}
static std::uint16_t last_adc()
{
return adc_average;
return m_adc_average;
}
static std::uint8_t fan_percent()
{
return percent;
return m_percent;
}
};

View File

@@ -5,43 +5,55 @@
#include <libavr/flash.hpp>
// The auto-mode fan curve, tabulated at compile time: a cubic in °C
// (0.002246·x³ 0.09·x² + 0.91·x, zero below 20 °C) as a flash_table of
// duty percent per °C.
// The auto-mode fan curve, tabulated at compile time as a flash_table of duty
// percent per C: a cubic that is held at zero below the temperature the fan
// starts at.
namespace app::curve {
inline constexpr double cubic_term = 0.002246;
inline constexpr double square_term = -0.09;
inline constexpr double linear_term = 0.91;
inline constexpr std::int8_t start_celsius = 20;
namespace detail {
consteval std::uint8_t duty_entry(int celsius)
consteval std::uint8_t duty_entry(std::int32_t celsius)
{
double x = celsius;
if (x < 20)
if (x < start_celsius) {
return 0;
double duty = 0.002246 * x * x * x - 0.09 * x * x + 0.91 * x;
if (duty < 0)
}
double duty = cubic_term * x * x * x + square_term * x * x + linear_term * x;
if (duty < 0) {
duty = 0;
if (duty > 100)
}
if (duty > 100) {
duty = 100;
}
return static_cast<std::uint8_t>(duty + 0.5);
}
inline constexpr avr::flash_table<[] {
std::array<std::uint8_t, 100> out{};
for (int t = 0; t < 100; ++t)
for (std::int32_t t = 0; t < static_cast<std::int32_t>(out.size()); ++t) {
out[static_cast<std::size_t>(t)] = duty_entry(t);
}
return out;
}()>
table;
} // namespace detail
// Duty percent for a temperature (clamped to the 0..99 °C table window).
// Duty percent for a temperature, clamped to the table's own window.
inline std::uint8_t duty(std::int8_t celsius)
{
if (celsius < 0)
constexpr auto highest = static_cast<std::int8_t>(detail::table.size() - 1);
if (celsius < 0) {
celsius = 0;
if (celsius > 99)
celsius = 99;
}
if (celsius > highest) {
celsius = highest;
}
return detail::table[static_cast<std::uint8_t>(celsius)];
}

View File

@@ -5,6 +5,7 @@
#include "controller.hpp"
#include "statistics.hpp"
#include "terminal.hpp"
#include "thermistor.hpp"
using namespace app;
@@ -16,14 +17,15 @@ int main()
avr::init<uptime::ticker, sampler::input, fan, serial_t, statistics>();
avr::irq::enable();
sampler::input::start();
sampler::thermistor::start();
controller::init();
terminal::init();
while (true) {
controller::poll();
if (controller::data_available())
statistics::record(static_cast<std::int8_t>((controller::temperature_quarters() + 2) / 4));
if (controller::data_available()) {
statistics::record(thermistor::whole_degrees(controller::temperature_quarters()));
}
terminal::poll();
}
}

View File

@@ -7,9 +7,9 @@
#include "board.hpp"
// Temperature histogram: one uint32 bucket per °C 0..99, sampled once a
// Temperature histogram: one uint32 bucket per C 0..99, sampled once a
// second, written back to EEPROM every 30 minutes (update() only touches
// changed bytes). Erased EEPROM reads back as 0xffffffff — treated as 0.
// changed bytes).
namespace app {
class statistics {
@@ -17,11 +17,15 @@ class statistics {
static constexpr std::uint32_t sample_delay_ms = 1'000;
static constexpr std::uint32_t writeback_delay_ms = 1'800'000;
// What an erased cell reads back as, so a bucket nobody has written yet
// counts as no samples rather than four billion.
static constexpr std::uint32_t never_written = ~std::uint32_t{0};
using stored = avr::eeprom::var<std::array<std::uint32_t, range>, 0>;
static inline std::array<std::uint32_t, range> histogram{};
static inline std::uint64_t last_sample = 0;
static inline std::uint64_t last_writeback = 0;
static inline std::array<std::uint32_t, range> m_histogram{};
static inline std::uint64_t m_last_sample = 0;
static inline std::uint64_t m_last_writeback = 0;
static constexpr std::uint8_t clamp(std::int8_t t)
{
@@ -33,72 +37,81 @@ class statistics {
static void init()
{
histogram = stored::read();
for (auto &bucket : histogram)
if (bucket == 0xffffffff)
m_histogram = stored::read();
for (auto &bucket : m_histogram) {
if (bucket == never_written) {
bucket = 0;
}
}
}
static void record(std::int8_t celsius)
{
auto now = uptime::millis();
if (now >= last_sample + sample_delay_ms) {
++histogram[clamp(celsius)];
last_sample = now;
if (now >= m_last_sample + sample_delay_ms) {
++m_histogram[clamp(celsius)];
m_last_sample = now;
}
if (now >= last_writeback + writeback_delay_ms) {
if (now >= m_last_writeback + writeback_delay_ms) {
save();
last_writeback = now;
m_last_writeback = now;
}
}
static void save()
{
stored::update(histogram);
stored::update(m_histogram);
}
static void reset()
{
histogram = {};
stored::update(histogram);
m_histogram = {};
stored::update(m_histogram);
}
static std::uint8_t min_temperature()
{
for (std::uint8_t i = 0; i < range; ++i)
if (histogram[i])
for (std::uint8_t i = 0; i < range; ++i) {
if (m_histogram[i]) {
return i;
}
}
return range;
}
static std::uint8_t max_temperature()
{
for (std::uint8_t i = range; i > 0; --i)
if (histogram[i - 1])
for (std::uint8_t i = range; i > 0; --i) {
if (m_histogram[i - 1]) {
return i - 1;
}
}
return 0;
}
static std::uint64_t total_samples()
{
std::uint64_t total = 0;
for (auto bucket : histogram)
for (auto bucket : m_histogram) {
total += bucket;
}
return total;
}
static std::uint32_t highest_bucket()
{
std::uint32_t highest = 0;
for (auto bucket : histogram)
if (bucket > highest)
for (auto bucket : m_histogram) {
if (bucket > highest) {
highest = bucket;
}
}
return highest;
}
static std::uint32_t bucket(std::uint8_t celsius)
{
return histogram[clamp(static_cast<std::int8_t>(celsius))];
return m_histogram[clamp(static_cast<std::int8_t>(celsius))];
}
};

View File

@@ -18,44 +18,89 @@
// until Ctrl+C, and Ctrl+C abandons a half-typed line anywhere else.
namespace app {
// Commands, in the order they are matched - which is the order the original
// firmware matched them in, and that order is load-bearing. An abbreviation
// resolves to the *first* entry it prefixes, so `s` is show (not statistics,
// not set) exactly as it always was, and anything appended to this list cannot
// steal an abbreviation that already meant something else.
//
// A function rather than a variable, so the list exists only while the flash
// blob below is being built and never as storage of its own.
consteval auto command_names()
{
return std::to_array<std::string_view>({
"help",
"show",
"curve",
"monitor",
"bootloader",
"uptime",
"statistics",
"histogram",
"reset",
"set",
"auto",
"version",
"save",
});
}
class terminal {
static constexpr char ctrl_c = 0x03;
static constexpr std::uint8_t line_max = 24;
static inline char line[line_max]{};
static inline std::uint8_t at = 0;
static inline bool overflowed = false;
static inline bool monitoring = false;
static inline std::uint64_t last_monitor = 0;
static constexpr char backspace = 0x08;
static constexpr char del = 0x7f;
static inline std::array<char, 24> m_line{};
static inline std::uint8_t m_at = 0;
static inline bool m_overflowed = false;
static inline bool m_monitoring = false;
static inline std::uint64_t m_last_monitor = 0;
// Commands, in the order they are matched — which is the order the original
// firmware matched them in, and that order is load-bearing. An abbreviation
// resolves to the *first* entry it prefixes, so `s` is show (not statistics,
// not set) exactly as it always was, and anything appended to this list
// cannot steal an abbreviation that already meant something else.
struct command {
std::string_view name;
bool exact; // reset only: an abbreviation must not be able to wipe data
};
static constexpr std::uint8_t command_count = command_names().size();
static constexpr std::array<command, 13> commands{{
{"help", false},
{"show", false},
{"curve", false},
{"monitor", false},
{"bootloader", false},
{"uptime", false},
{"statistics", false},
{"histogram", false},
{"reset", true},
{"set", false},
{"auto", false},
{"version", false},
{"save", false},
}};
static constexpr std::size_t name_bytes = [] {
std::size_t total = 0;
for (auto one : command_names()) {
total += one.size() + 1;
}
return total;
}();
// The names as one NUL-separated blob in flash. Matched at run time, which
// is not a reason to sit in RAM on a machine with two address spaces: a
// table of `string_view` puts the characters *and* its own pointers there,
// and on this part that was 216 B of 2 KB spent on data that never changes.
// Separators rather than an offset table, because an offset table is the
// RAM this exists to give back.
static constexpr auto names_data = [] {
std::array<char, name_bytes> chars{};
std::size_t at = 0;
for (auto one : command_names()) {
for (char c : one) {
chars[at++] = c;
}
chars[at++] = '\0';
}
return chars;
}();
static constexpr avr::flash_table<names_data> names{};
// `reset` must be typed in full - an abbreviation must not be able to wipe
// data. Found in the list rather than written as an index, so reordering the
// commands cannot move the protection onto a different one.
static constexpr std::uint8_t exact_command = [] {
const auto list = command_names();
for (std::uint8_t i = 0; i < list.size(); ++i) {
if (list[i] == "reset") {
return i;
}
}
return static_cast<std::uint8_t>(list.size());
}();
// Column the descriptions' colons line up in, counted from the start of the
// name. The longest name is `bootloader` at 10, so 12 leaves it a space and
// one dot the original's layout exactly.
// one dot - the original's layout exactly.
static constexpr std::uint8_t help_column = 12;
static void prompt()
@@ -63,42 +108,45 @@ class terminal {
serial << "> "_P;
}
// `name ....: ` the dotted label the original used everywhere it printed a
// `name ....: ` - the dotted label the original used everywhere it printed a
// list of things, which is what makes a column of values readable without
// counting spaces. One renderer for all three users; only the column differs.
static void label(std::string_view text, std::uint8_t column)
{
serial << text << ' ';
for (auto i = text.size() + 1; i < column; ++i)
serial << '.';
serial << ": "_P;
}
// The same, for a label that is a literal rather than a command name: it
// stays in flash, and its width comes from the type, so the padding needs no
// hand-counted constant. The command names cannot use this — they are
// string_views because they are matched at run time.
// counting spaces. Its width comes from the type, so the padding needs no
// hand-counted constant.
template <typename Flash>
static void label(Flash text, std::uint8_t column)
{
serial << text << ' ';
for (auto i = Flash::size + 1; i < column; ++i)
for (auto i = Flash::size + 1; i < column; ++i) {
serial << '.';
}
serial << ": "_P;
}
static void help_row(std::string_view name)
// The same shape for a command name, whose width is not a type but the walk
// to its separator - so it is counted as it is printed rather than measured
// first, which would read the blob twice.
static void help_row(std::uint8_t which)
{
label(name, help_column);
auto at = name_start(which);
std::uint8_t width = 0;
for (char c; (c = names.at(at + width)) != '\0'; ++width) {
serial << c;
}
serial << ' ';
for (std::uint8_t i = width + 1; i < help_column; ++i) {
serial << '.';
}
serial << ": "_P;
}
// Quarter-°C as a signed decimal with a two-digit fraction. The sign is taken
// Quarter- C as a signed decimal with a two-digit fraction. The sign is taken
// off first: C++ gives a negative remainder for a negative dividend, so
// `(q % 4) * 25` on -40.25 C yields -25 and prints "-40.-25".
static void temperature(std::int16_t quarters)
{
if (quarters < 0)
if (quarters < 0) {
serial << '-';
}
auto magnitude = static_cast<std::uint16_t>(quarters < 0 ? -quarters : quarters);
serial << magnitude / 4 << '.' << avr::dec<{.width = 2, .fill = '0'}>((magnitude % 4) * 25);
}
@@ -106,31 +154,31 @@ class terminal {
static void help()
{
serial << "\r\nFanTemp "_P << version << " command overview\r\n"_P;
help_row(commands[0].name);
help_row(0);
serial << "prints this help message\r\n"_P;
help_row(commands[1].name);
help_row(1);
serial << "shows current temperature and fan speed\r\n"_P;
help_row(commands[2].name);
help_row(2);
serial << "shows mapping from temperature to fan speed\r\n"_P;
help_row(commands[3].name);
help_row(3);
serial << "loops the show command until Ctrl+C is pressed\r\n"_P;
help_row(commands[4].name);
help_row(4);
serial << "enters the bootloader\r\n"_P;
help_row(commands[5].name);
help_row(5);
serial << "shows system uptime\r\n"_P;
help_row(commands[6].name);
help_row(6);
serial << "prints overall statistics like min and max temp\r\n"_P;
help_row(commands[7].name);
help_row(7);
serial << "prints a histogram of the temperature\r\n"_P;
help_row(commands[8].name);
help_row(8);
serial << "resets statistics to 0 in EEPROM and RAM (no abbreviation)\r\n"_P;
help_row(commands[9].name);
help_row(9);
serial << "sets the fan speed to the provided value, 0-100\r\n"_P;
help_row(commands[10].name);
help_row(10);
serial << "turns on automatic fan control\r\n"_P;
help_row(commands[11].name);
help_row(11);
serial << "displays firmware version\r\n"_P;
help_row(commands[12].name);
help_row(12);
serial << "writes the statistics to EEPROM now\r\n"_P;
serial << "commands may be abbreviated: 'up' is uptime\r\n"_P;
}
@@ -142,9 +190,10 @@ class terminal {
static std::uint32_t resistance()
{
auto adc = controller::last_adc();
if (adc >= 1023)
return 0xffffffff; // open circuit: the divider has no solution
return static_cast<std::uint32_t>(thermistor::series_resistor) * adc / (1023u - adc);
if (adc >= thermistor::adc_full_scale) {
return open_circuit;
}
return static_cast<std::uint32_t>(thermistor::series_resistor) * adc / (thermistor::adc_full_scale - adc);
}
// One value per line behind a dotted label, as the original had it. A single
@@ -160,10 +209,11 @@ class terminal {
serial << controller::last_adc() << " / 1023\r\n"_P;
label("Resistance"_P, reading_column);
if (auto ohms = resistance(); ohms == open_circuit)
if (auto ohms = resistance(); ohms == open_circuit) {
serial << "open circuit\r\n"_P;
else
} else {
serial << ohms << " Ohm\r\n"_P;
}
label("Temperature"_P, reading_column);
temperature(controller::temperature_quarters());
@@ -171,10 +221,11 @@ class terminal {
label("Fan speed"_P, reading_column);
serial << controller::fan_percent() << "% "_P;
if (controller::automatic())
if (controller::automatic()) {
serial << "auto\r\n"_P;
else
} else {
serial << "manual\r\n"_P;
}
}
// Every whole degree from 10 to 60 with a bar, which is the original's and is
@@ -187,8 +238,9 @@ class terminal {
auto duty = curve::duty(static_cast<std::int8_t>(t));
serial << avr::dec<{.width = 2, .fill = '0'}>(t) << " C = "_P << avr::dec<{.width = 3, .fill = ' '}>(duty)
<< "% |"_P;
for (std::uint8_t i = 0; i < duty; ++i)
for (std::uint8_t i = 0; i < duty; ++i) {
serial << '#';
}
serial << "\r\n"_P;
}
}
@@ -204,16 +256,18 @@ class terminal {
{
auto empty = statistics::total_samples() == 0;
label("Minimum temperature"_P, stat_column);
if (empty)
if (empty) {
serial << "not available\r\n"_P;
else
} else {
serial << statistics::min_temperature() << " C\r\n"_P;
}
label("Maximum temperature"_P, stat_column);
if (empty)
if (empty) {
serial << "not available\r\n"_P;
else
} else {
serial << statistics::max_temperature() << " C\r\n"_P;
}
label("Total samples"_P, stat_column);
serial << static_cast<std::uint32_t>(statistics::total_samples()) << "\r\n"_P;
@@ -232,8 +286,9 @@ class terminal {
// neighbouring buckets, which on a distribution this narrow is the whole
// picture.
std::uint32_t factor = highest / bar_max > 1 ? highest / bar_max : 1;
while (highest / factor > bar_max)
while (highest / factor > bar_max) {
++factor;
}
for (std::uint8_t t = statistics::min_temperature(); t <= statistics::max_temperature(); ++t) {
auto count = statistics::bucket(t);
@@ -241,55 +296,87 @@ class terminal {
// instead of trailing off the ragged right-hand end of the bars.
serial << avr::dec<{.width = 2, .fill = '0'}>(t) << " C : "_P << avr::dec<{.width = 10, .fill = ' '}>(count)
<< " |"_P;
for (std::uint32_t i = 0; i < count / factor; ++i)
for (std::uint32_t i = 0; i < count / factor; ++i) {
serial << '#';
}
serial << "\r\n"_P;
}
}
// Abbreviations: the input matches a command when it is a non-empty prefix
// of it. `reset` is the exception and must be typed in full.
//
// starts_with, not substr: substr throws std::out_of_range, and one
// potentially-throwing call is enough to pull in std::terminate, which does
// not exist in a freestanding AVR build. The link fails rather than the
// firmware, so this is a build-time trap rather than a runtime one — but it
// is a trap, and the whole file avoids substr for that reason.
static bool matches(std::string_view input, const command &c)
// Where the given command's name begins in the blob. A walk over the
// separators, for the two callers that need to start at one name.
static constexpr std::uint8_t name_start(std::uint8_t which)
{
if (input.empty())
return false;
if (c.exact)
return input == c.name;
return c.name.starts_with(input);
std::uint8_t at = 0;
for (std::uint8_t i = 0; i < which; ++i) {
while (names.at(at) != '\0') {
++at;
}
++at;
}
return at;
}
public:
// Which command a word names, or `command_count` for none. The input
// matches when it is a non-empty prefix of a name; `reset` is the exception
// and must be typed in full.
//
// One pass over the blob answers both, because the names sit in it in match
// order: each name is compared as it is walked, and the walk to its
// separator is also what measures it.
//
// Public and `constexpr` so the battery can pin it: the match order is
// load-bearing, an abbreviation must never reach `reset`, and reading the
// names from flash must not have changed either.
static constexpr std::uint8_t lookup(std::string_view input)
{
if (input.empty()) {
return command_count;
}
std::uint8_t at = 0;
for (std::uint8_t i = 0; i < command_count; ++i) {
std::uint8_t length = 0;
bool prefix = true;
for (char c; (c = names.at(at + length)) != '\0'; ++length) {
if (length < input.size() && input[length] != c) {
prefix = false;
}
}
const bool whole = input.size() == length;
if (prefix && input.size() <= length && (i != exact_command || whole)) {
return i;
}
at += length + 1;
}
return command_count;
}
static constexpr std::uint8_t no_command = command_count;
private:
static void dispatch(std::string_view input)
{
// A line that overflowed the buffer is not a command it is the tail of
// A line that overflowed the buffer is not a command - it is the tail of
// one. Acting on it is how a truncated `reset` becomes a surprise.
if (overflowed) {
if (m_overflowed) {
serial << "input too long, ignored\r\n"_P;
overflowed = false;
m_overflowed = false;
return;
}
// Split on the first space with the (pointer, length) constructor rather
// than substr, which throws see matches().
// than substr, which throws - see matches().
const auto space = input.find(' ');
const auto word = space == std::string_view::npos ? input : std::string_view{input.data(), space};
const auto rest = space == std::string_view::npos
? std::string_view{}
: std::string_view{input.data() + space + 1, input.size() - space - 1};
if (word.empty())
if (word.empty()) {
return;
}
std::uint8_t which = commands.size();
for (std::uint8_t i = 0; i < commands.size(); ++i)
if (matches(word, commands[i])) {
which = i;
break;
}
const auto which = lookup(word);
switch (which) {
case 0:
@@ -302,7 +389,7 @@ class terminal {
print_curve();
return;
case 3:
monitoring = true;
m_monitoring = true;
return;
case 4:
serial << "entering bootloader\r\n"_P;
@@ -331,8 +418,9 @@ class terminal {
break;
}
percent = static_cast<std::uint16_t>(percent * 10 + (c - '0'));
if (percent > 100)
if (percent > 100) {
valid = false;
}
}
if (valid) {
controller::set_manual(static_cast<std::uint8_t>(percent));
@@ -360,7 +448,10 @@ class terminal {
}
public:
static constexpr std::string_view version = "v2.2";
// A flash string, like every other literal the console prints: as a
// `string_view` it was the last constant left in RAM, holding its own
// characters and a pointer to them.
static constexpr auto version = "v2.2"_P;
static constexpr std::uint8_t bar_max = 100;
// Columns the dotted labels' colons land in, and the curve's span. All four
@@ -381,17 +472,17 @@ class terminal {
static void poll()
{
if (monitoring) {
if (uptime::millis() >= last_monitor + 1000) {
if (m_monitoring) {
if (uptime::millis() >= m_last_monitor + 1000) {
show();
last_monitor = uptime::millis();
m_last_monitor = uptime::millis();
}
// Ctrl+C only, as the original had it. Stopping on *any* byte reads
// well until a host sends a line ending: `monitor\r\n` then stops
// itself on the `\n` it arrived with, one reading in.
if (auto in = serial_t::read(); in && *in == ctrl_c) {
serial << "^C\r\n"_P;
monitoring = false;
m_monitoring = false;
prompt();
}
return;
@@ -402,35 +493,35 @@ class terminal {
// Abandon whatever was typed and start a fresh line, which is
// what Ctrl+C means at every other prompt in the world.
serial << "^C\r\n"_P;
at = 0;
overflowed = false;
m_at = 0;
m_overflowed = false;
prompt();
} else if (c == '\r' || c == '\n') {
serial << "\r\n"_P;
if (at == 0 && !overflowed) {
if (m_at == 0 && !m_overflowed) {
prompt(); // a bare Enter just reprompts, no gap needed
continue;
}
dispatch(std::string_view{line, at});
at = 0;
if (!monitoring) {
dispatch(std::string_view{m_line.data(), m_at});
m_at = 0;
if (!m_monitoring) {
// A blank line between a command's output and the next
// prompt: without it the answer and the thing you type
// next run together and a screen of them is unreadable.
serial << "\r\n"_P;
prompt();
}
} else if (c == 0x7f || c == 0x08) {
if (at) {
--at;
} else if (c == del || c == backspace) {
if (m_at) {
--m_at;
serial << "\b \b"_P;
}
} else if (c >= ' ') {
if (at < line_max) {
line[at++] = c;
if (m_at < m_line.size()) {
m_line[m_at++] = c;
serial << c; // echo
} else {
overflowed = true; // reported when the line is submitted
m_overflowed = true; // reported when the line is submitted
}
}
}

View File

@@ -7,10 +7,14 @@
// NTC thermistor on a series divider, solved entirely at compile time:
// the Beta equation (logarithm and all) runs consteval into a libavr
// flash_table the firmware never does floating point. Raw 10-bit ADC
// counts map to quarter-°C with linear interpolation between table steps.
// flash_table - the firmware never does floating point. Raw 10-bit ADC
// counts map to quarter- C with linear interpolation between table steps.
namespace app::thermistor {
// The converter's top count: this board reads the divider at the ADC's
// full 10 bits, so a reading and the resistance it implies both scale by it.
inline constexpr std::uint16_t adc_full_scale = (1u << 10) - 1;
inline constexpr double series_resistor = 9951;
inline constexpr double nominal_resistance = 9270;
inline constexpr double beta = 3212;
@@ -20,42 +24,47 @@ namespace detail {
consteval double temperature_of(double adc)
{
double resistance = series_resistor * adc / (1023.0 - adc);
double resistance = series_resistor * adc / (adc_full_scale - adc);
// __builtin_log constant-folds on the AVR backend, so the table is
// built at compile time with no runtime libm.
double steinhart = __builtin_log(resistance / nominal_resistance) / beta + 1.0 / (nominal_temperature + 273.15);
return 1.0 / steinhart - 273.15;
}
// 256 entries over the 10-bit range (steps of 4 counts), quarter-°C,
// 256 entries over the 10-bit range (steps of 4 counts), quarter- C,
// clamped to a sane sensor window; entry 256 mirrors 255 so interpolation
// at full scale has a right neighbour.
consteval std::int16_t quarters_entry(int index)
consteval std::int16_t quarters_entry(std::int32_t index)
{
double adc = index * 4.0;
if (adc < 4)
if (adc < 4) {
adc = 4;
if (adc > 1019)
}
if (adc > 1019) {
adc = 1019;
}
double t = temperature_of(adc) * 4.0;
if (t < -40 * 4)
if (t < -40 * 4) {
t = -40 * 4;
if (t > 125 * 4)
}
if (t > 125 * 4) {
t = 125 * 4;
}
return static_cast<std::int16_t>(t < 0 ? t - 0.5 : t + 0.5);
}
inline constexpr avr::flash_table<[] {
std::array<std::int16_t, 257> out{};
for (int i = 0; i < 257; ++i)
for (std::int32_t i = 0; i < static_cast<std::int32_t>(out.size()); ++i) {
out[static_cast<std::size_t>(i)] = quarters_entry(i < 256 ? i : 255);
}
return out;
}()>
table;
} // namespace detail
// Temperature in quarter-°C from a raw (or averaged) 10-bit sample.
// Temperature in quarter- C from a raw (or averaged) 10-bit sample.
inline std::int16_t quarters(std::uint16_t adc)
{
std::uint16_t index = adc >> 2; // the 257th entry backs index+1 at full scale
@@ -65,9 +74,12 @@ inline std::int16_t quarters(std::uint16_t adc)
return static_cast<std::int16_t>(a + ((b - a) * frac) / 4);
}
inline std::int8_t celsius(std::uint16_t adc)
// Whole degrees from quarter-degrees, rounded to nearest. The shift is the
// point: C truncates a negative quotient toward zero, so `(q + 2) / 4` reads
// -3.00 C as -2, where an arithmetic shift floors and reads it as -3.
inline constexpr std::int8_t whole_degrees(std::int16_t quarters)
{
return static_cast<std::int8_t>((quarters(adc) + 2) / 4);
return static_cast<std::int8_t>((quarters + 2) >> 2);
}
} // namespace app::thermistor

View File

@@ -1,12 +1,34 @@
libavr_format_test()
# The image must stay below the boot section at 0x7e00, and the README states
# what it measures; a library advance that moves it says nothing on its own.
libavr_size_claim_test(fantemp 8004)
# The README says the two modes emit the same image, and only a tree with
# both built can say whether they do.
libavr_mode_identity_test(fantemp)
# The battery is a compile: a static_assert that fails is the failure. It is a
# target rather than only a ctest so a plain build catches a regression too.
add_library(consteval_tests OBJECT consteval.cpp)
target_include_directories(consteval_tests PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/../src)
target_link_libraries(consteval_tests PRIVATE libavr)
add_test(NAME fantemp.consteval
COMMAND ${CMAKE_COMMAND} --build ${CMAKE_BINARY_DIR} --target consteval_tests)
# The board has no reset line and no programming header, so the loader-entry
# route in the emitted image is the only thing standing between a firmware change
# and an unreflashable board. It has been wrong before — see the script.
find_package(Python3 COMPONENTS Interpreter)
if(Python3_FOUND)
add_test(NAME fantemp.reachability
COMMAND ${Python3_EXECUTABLE} ${CMAKE_CURRENT_SOURCE_DIR}/check_reachability.py
--objdump ${CMAKE_OBJDUMP} --elf $<TARGET_FILE:fantemp>
--image $<TARGET_FILE_DIR:fantemp>/fantemp.bin)
else()
message(STATUS "Python not found — the reachability check is skipped")
endif()
# and an unreflashable board.
# A check whose interpreter this host lacks is registered and *skips*, rather
# than leaving the suite a different size on a different machine - which is a
# suite nothing can be compared against. The launcher and the marker are
# libavr's (`cmake/checks.cmake`).
libavr_launcher(_python "${LIBAVR_NO_PYTHON}" ${Python3_EXECUTABLE})
add_test(NAME fantemp.reachability
COMMAND ${_python} ${CMAKE_CURRENT_SOURCE_DIR}/check_reachability.py
--objdump ${CMAKE_OBJDUMP} --elf $<TARGET_FILE:fantemp>
--image $<TARGET_FILE_DIR:fantemp>/fantemp.bin
--loader-base ${PUREBOOT_BASE_HEX})
libavr_skip_unverified()

View File

@@ -3,7 +3,7 @@
This board has no reset line and no programming header. The single route to the
bootloader is the running firmware's `bootloader` command, so a firmware that
gets that route wrong is a board that cannot be reflashed and the failure is
gets that route wrong is a board that cannot be reflashed - and the failure is
silent, because everything else still works.
It has been wrong before. The firmware this one replaces probed and jumped to
@@ -15,7 +15,7 @@ hardware, which is what this replaces.
Three properties, all read out of the disassembly rather than the source:
1. The image ends below the boot section. `hfuse d4` puts that at 0x7c00, so an
application reaching into it would be overwritten by the loader or worse,
application reaching into it would be overwritten by the loader - or worse,
executed at reset, since BOOTRST points there.
2. The hand-over targets the loader base. A word address of 0x3f00 is byte
0x7e00; anything else is the 0x7800 bug again.
@@ -35,8 +35,7 @@ import re
import subprocess
import sys
BOOT_SECTION = 0x7C00 # hfuse d4: BOOTSZ 512 words
LOADER_BASE = 0x7E00 # pureboot's 512-byte slot, at the top
BOOT_SECTION = 0x7C00 # hfuse d4: BOOTSZ 512 words - a fuse fact, not a loader one
WDTCSR = 0x60
@@ -45,7 +44,12 @@ def main() -> int:
parser.add_argument("--objdump", required=True)
parser.add_argument("--elf", type=pathlib.Path, required=True)
parser.add_argument("--image", type=pathlib.Path, required=True)
# pureboot's own geometry, from the pinned submodule that exports it, so the
# slot's base is stated once for the loader this firmware is deployed with.
parser.add_argument("--loader-base", required=True,
type=lambda v: int(v, 0), metavar="ADDR")
args = parser.parse_args()
loader_base = args.loader_base
failures = []
@@ -60,7 +64,7 @@ def main() -> int:
text = subprocess.run([args.objdump, "-d", str(args.elf)],
capture_output=True, text=True, check=True).stdout
# The address the hand-over actually targets, read at its call sites not
# The address the hand-over actually targets, read at its call sites - not
# "does the image contain this byte somewhere", which proves nothing: 0x3f is
# an ordinary constant that appears in the curve tables, so a check like that
# passes just as happily on the 0x7800 bug it is supposed to catch.
@@ -81,27 +85,27 @@ def main() -> int:
if "r24" in held and "r25" in held:
sites.append(held["r25"] << 8 | held["r24"])
want = LOADER_BASE // 2
want = loader_base // 2
if not sites:
failures.append("no call to bootloader::call with a loaded target the "
failures.append("no call to bootloader::call with a loaded target - the "
"hand-over could not be read out of the image")
elif wrong := [a for a in sites if a != want]:
failures.append(f"the hand-over targets word {[hex(a) for a in wrong]} "
f"(byte {[hex(a * 2) for a in wrong]}), not the loader at "
f"0x{LOADER_BASE:04x}")
f"0x{loader_base:04x}")
else:
print(f" ok all {len(sites)} hand-over site(s) target word 0x{want:04x} "
f"(byte 0x{LOADER_BASE:04x})")
f"(byte 0x{loader_base:04x})")
# An icall/ijmp has to exist for that address to be jumped to indirectly.
if not re.search(r"\b(icall|ijmp)\b", text):
failures.append("no icall/ijmp the hand-over cannot reach across flash")
failures.append("no icall/ijmp - the hand-over cannot reach across flash")
else:
print(" ok an indirect call exists (a relative one cannot reach)")
# What actually reaches WDTCSR, not what the image happens to load somewhere.
# A timed disable writes WDCE|WDE (0x18) and then zero. Arming writes WDE
# *without* WDCE including 0x08, a 16 ms timeout with every prescaler bit
# *without* WDCE - including 0x08, a 16 ms timeout with every prescaler bit
# clear, which is precisely what the legacy route used and is why this cannot
# be a check for "a prescaler is present".
WDCE, WDE = 0x10, 0x08
@@ -114,14 +118,14 @@ def main() -> int:
values.append(0 if reg == "r1" else held.get(reg))
armed = [v for v in values if v is not None and (v & WDE) and not (v & WDCE)]
if armed:
failures.append(f"WDTCSR is written {[hex(v) for v in armed]} WDE without "
failures.append(f"WDTCSR is written {[hex(v) for v in armed]} - WDE without "
f"WDCE is arming the watchdog, and a reset-based hand-over "
f"opens no pureboot window")
elif not values:
print(" ok the watchdog is never written")
else:
print(f" ok WDTCSR writes are {[hex(v) if v is not None else '?' for v in values]}"
f" unlock and clear, never an arm")
f" - unlock and clear, never an arm")
for line in failures:
print(f" FAIL {line}")

120
test/consteval.cpp Normal file
View File

@@ -0,0 +1,120 @@
// Compile-only battery, built with the cross compiler so the target's 16-bit
// int is exercised. Two of this firmware's three numeric surfaces are decided
// entirely at compile time - the fan curve and the thermistor table - and the
// third, quarter-degrees to whole ones, is the arithmetic between them.
#include <cstdint>
#include "curve.hpp"
#include "terminal.hpp"
#include "thermistor.hpp"
namespace {
using app::thermistor::whole_degrees;
// Rounding to nearest, across zero. A negative quotient truncates toward zero
// in C, so a `(q + 2) / 4` reads -3.00 C as -2 and every case below the tie
// with it; the arithmetic shift floors, which is what these hold it to.
static_assert(whole_degrees(0) == 0);
static_assert(whole_degrees(1) == 0); // +0.25
static_assert(whole_degrees(2) == 1); // +0.50, the tie
static_assert(whole_degrees(3) == 1); // +0.75
static_assert(whole_degrees(4) == 1); // +1.00
static_assert(whole_degrees(-1) == 0); // -0.25
static_assert(whole_degrees(-2) == 0); // -0.50, the tie, toward zero
static_assert(whole_degrees(-3) == -1);
static_assert(whole_degrees(-4) == -1);
static_assert(whole_degrees(-12) == -3);
static_assert(whole_degrees(-160) == -40); // the table's own floor
static_assert(whole_degrees(500) == 125); // and its ceiling
// The curve is held at zero below the temperature the fan starts at, and
// saturates inside the table's window rather than at its edge.
using app::curve::detail::duty_entry;
static_assert(duty_entry(app::curve::start_celsius - 1) == 0);
static_assert(duty_entry(app::curve::start_celsius) == 0);
static_assert(duty_entry(30) == 7);
static_assert(duty_entry(40) == 36);
static_assert(duty_entry(49) == 93);
static_assert(duty_entry(50) == 100);
static_assert(duty_entry(99) == 100);
// Monotone across the whole table: a warmer reading never asks for less air.
consteval bool curve_rises()
{
for (std::int32_t t = 1; t < 100; ++t) {
if (duty_entry(t) < duty_entry(t - 1)) {
return false;
}
}
return true;
}
static_assert(curve_rises());
// The thermistor table, anchored where the Beta equation fixes it rather than
// against numbers this file computed the same way: the divider reads the
// thermistor's nominal resistance at
// adc = full_scale * nominal / (series + nominal), and the equation's own
// definition puts that count at the nominal temperature.
using app::thermistor::detail::quarters_entry;
inline constexpr std::int32_t nominal_count =
static_cast<std::int32_t>(app::thermistor::adc_full_scale * app::thermistor::nominal_resistance /
(app::thermistor::series_resistor + app::thermistor::nominal_resistance));
static_assert(quarters_entry(nominal_count / 4) >= 100); // 25.00 C, in quarters
static_assert(quarters_entry(nominal_count / 4 + 1) < 100); // and the step below it
// Both clamps, at the ends the divider cannot leave.
static_assert(quarters_entry(0) == 125 * 4);
static_assert(quarters_entry(255) == -40 * 4);
// An NTC on this divider falls with the count: more counts is more resistance
// is a colder sensor, over every step of the table.
consteval bool thermistor_falls()
{
for (std::int32_t i = 1; i < 256; ++i) {
if (quarters_entry(i) > quarters_entry(i - 1)) {
return false;
}
}
return true;
}
static_assert(thermistor_falls());
// The console's one parsing rule, read out of the flash blob the names live
// in. Every case here is a property of the *order* the names sit in, so this
// is what an edit to that list has to answer to.
using app::terminal;
// Full names, and the ends of the list - a walk that miscounts a separator
// lands on a neighbour rather than failing, so both ends are named.
static_assert(terminal::lookup("help") == 0);
static_assert(terminal::lookup("save") == 12);
static_assert(terminal::lookup("bootloader") == 4);
// Abbreviations resolve to the first entry they prefix. `s` prefixes show,
// statistics, set and save, and show is first - which is the original's
// behaviour and the reason the list is ordered rather than sorted.
static_assert(terminal::lookup("s") == 1);
static_assert(terminal::lookup("st") == 6);
static_assert(terminal::lookup("se") == 9);
static_assert(terminal::lookup("sa") == 12);
static_assert(terminal::lookup("up") == 5);
static_assert(terminal::lookup("b") == 4);
// `reset` is the exception: it must be typed in full, so no abbreviation of it
// resolves - and none of its prefixes names anything else either.
static_assert(terminal::lookup("reset") == 8);
static_assert(terminal::lookup("rese") == terminal::no_command);
static_assert(terminal::lookup("res") == terminal::no_command);
static_assert(terminal::lookup("r") == terminal::no_command);
// Nothing, and nothing that matches.
static_assert(terminal::lookup("") == terminal::no_command);
static_assert(terminal::lookup("xyzzy") == terminal::no_command);
// Longer than the name it prefixes is not a match: `helpful` is not `help`.
static_assert(terminal::lookup("helpful") == terminal::no_command);
} // namespace

11
tools/check.sh Executable file
View File

@@ -0,0 +1,11 @@
#!/usr/bin/env bash
# The task gate: every configure preset built, and only then every one tested.
#
# The loop itself is libavr's (tools/check-presets.sh) because the order in it
# is subtle and was got wrong by hand more than once - the mode-identity checks
# read the sibling mode's tree, so a build-then-test-per-preset run compares a
# fresh image against a stale sibling. Anything this repository needs beyond
# the presets is written after the call, where every tree is built and tested.
set -euo pipefail
cd "$(dirname "${BASH_SOURCE[0]}")/.."
"${LIBAVR_ROOT:-libavr}/tools/check-presets.sh" "$@"