* all: clean up code with Go 1.24+ in mind
* all: more old go cleanup
* all: even remove rand_fastrand64
* runtime: revert rand changes
* runtime: re-drop rand_fastrand64
A SHT_NOBITS section was unconditionally treated as .bss (RAM) unless it
was the stack. The ESP linker scripts emit non-writable SHT_NOBITS
sections (.irom_dummy / .rodata_dummy) that merely reserve the
flash-mapped XIP virtual address ranges and occupy no RAM. Counting them
inflated the reported bss/RAM usage.
Signed-off-by: deadprogram <ron@hybridgroup.com>
Having them as methods on reflectlite.Value makes them
visible to the user.
Also move new functions out of all_test.go so it can stay
closer to upstream (except for comments).
Generalize the shared esp32xx I2C implementation to also cover the
ESP32-C6 and add the chip-specific code.
Signed-off-by: deadprogram <ron@hybridgroup.com>
I think it's much nicer to have the test output inline in the source
file, that way it's much easier to review any changes. For example, when
escape analysis is improved this is visible with removed `// OUT` lines.
This is similar to how LLVM writes its tests, and I like that style.
Instead of referring to an unused global, use a constant value. This is
safe even when using `-gc=none` (since no actual memory gets allocated)
which wasn't the case before. It should also reduce binary size by a few
bytes for most programs.
This adds a minimal esp32c6 implementation, currently only
supporting the examples/serial and examples/blinky1 programs.
It does correctly output the expected "Hello, World" via the
serial port, as well as blink the onboard LED.
In addition, it adds support for the PLIC based IRQ handling
as used on the ESP32C6 processor.
Some parts of this code are loosely based on PR #5252 and #5248
Signed-off-by: deadprogram <ron@hybridgroup.com>
I think this failed in the past, but presumably those failures have been
fixed by now. So these intrinsics can now be used.
Using these intrinsics instead of the native Go implementations helps
LLVM to reason about them: it can for example evaluate the value at
compile time or do optimizations like convert
`float32(math.Sin(float64(x)))` into a 32-bit sin operation. If the math
operation is not available on the target platform the C library
implementation will be used instead.
When the interpreter encountered a load that was out of bounds of the
object, it panicked with "interp: load out of bounds", crashing the
compiler. This can happen for valid Go programs, for example when
dereferencing the pointer returned by unsafe.SliceData on a
zero-capacity slice, which points to a zero-sized object:
package main
import "unsafe"
var p = unsafe.SliceData([]int{})
var v = *p
func main() {}
Return nil for an out-of-bounds load, the same as for an external
global, so the caller defers the load to runtime instead of crashing.
This matches what regular Go does, where the load reads from the
runtime zero-base at runtime.
Fixes#4214
This converts the heap allocation to a statically allocated string.
This is useful for me, since it reduces binary size and makes it
possible to use `-gc=none` (which would previously result in a linker
error).
Track whether an allocation reaches a callee return separately from the
instruction where it escapes. The extra result state is needed because LLVM's
returned parameter attribute only covers scalar returns: a slice helper returns
its data pointer inside a {ptr, len, cap} aggregate, so the alias is only
visible by walking insertvalue and ret uses in the callee.
This lets OptimizeAllocs keep the backing array for returnIntSlice(s) on the
stack when the returned slice is ignored, matching gc's escape decision for the
same pattern. The golden output still keeps the escaping returned-slice case on
the heap.
Follow returned pointer aliases when deciding whether runtime.alloc calls can
be lowered to stack allocations. A returned parameter is not the same as
nocapture: it still flows back to the caller and must be checked as an alias
of the original allocation.
Keep the analysis conservative for recursive returned-parameter chains and
unknown operands. The existing golden test now shows that the non-escaping
returned pointer cases no longer require heap allocation.
Add allocation diagnostics coverage for pointer-returning helpers, aggregate
slice returns, and conditional pointer returns before changing the allocation
optimizer.
The golden files record the current heap-allocation behavior so later commits
show exactly which diagnostics each optimizer improvement removes.
This moves the objHeader from before an object body to after it.
On 32-bit systems with 16-byte alignment requirements (x86, ARM, RISC-V), we previously padded the header to a whole block.
This wastes up to 12 bytes, as on -gc=conservative the header is a single pointer.
With this change, no padding is required (beyond that from rounding the size up).
The "head" block in the metadata was moved to the end of the range to match the header location.
This changed the block loop directions throughout the GC logic.
The bit hacks used by sweep no longer work because there is no equivalent of addition that carries downwards.
However it is now possible to merge the sweep and free range list rebuild passes because their loop directions match.
There are two other places where we rebuilt the free ranges list: when initializing or growing the heap.
The former can be easily replaced with a single hardcoded range containing the entire heap.
In the latter case, I opted to only add the new space to the existing list.
These replacements allowed me to fully remove the buildFreeRanges function.
PR #5220 changed -print-allocs output to the go coverage tool format, which
replaced the original human-readable explanation of why each object had to be
heap allocated. That explanation is useful on its own, so this restores it as
the default behavior of -print-allocs and moves the coverage format behind a
-print-allocs-cover flag.
Signed-off-by: Piotr Bocheński <piotr@bochen.ski>
Implement SetStallEPIn, SetStallEPOut, ClearStallEPIn, and ClearStallEPOut methods on the nRF52840 USBDevice struct using the hardware EPSTALL register to support MSC driver requirements. Also, correct the handleUSBIRQ loops to iterate over physical endpoint numbers rather than dynamic configuration entries to prevent missed or incorrectly routed endpoint interrupts.
Remap CDC and MSC endpoints to share physical endpoint numbers. This change separates IN/OUT directional states for RP2 endpoints to prevent cross-talk, ensures SAMD21 and SAMD51 endpoint configuration merging does not overwrite bidirectionally shared registers, implements missing SAMD endpoint stall and clear methods, and corrects SAMD interrupt loop bounds to iterate over physical endpoint numbers rather than dynamic configuration entries.
Replace static endpoint variables with EndpointIN and EndpointOUT
functions. Allows flexible endpoint remapping across USB configs.
Map CDC, HID, MIDI and MSC USB devices to bidirectional endpoints consistently across different configurations.
* gba: add bios interrupt flags
Adds a new GBA register for interrupt flags, the register is equivalent
to the IF register, but is intended for BIOS functions.
Acknowledging the interrupts with this register allows us to use BIOS
halt functions such as VBlankIntrWait and IntrWait, which we can use to
get rid of busy loops in the GBA code.
* fix formatting
---------
Co-authored-by: zoey <git@zoey.si>
Implements `putchar` when `mgbadebug` build tag is set which outputs
text through the mGBA debugging output interface.
mGBA allows the games running in the emulator to output debug text the
process to do so is:
1. set 0x4FFF780 to value of 0xC0DE
2. write text in memory 0x4FFF600 to 0x4FFF700
3. set 0x4FFF700 to the desired log level value from 0x100 (fatal) to
0x104 (debug)
Once this is done mGBA will output the text in its logs view as well as
through stdout. (Setting the log level to fatal also produces a dialog
box with the text)
The text output in the CLI is prefixed with `[DEBUG] GBA Debug: ` so I
modified the regex used for address matching in panic messages to be
able to read the address when the output line doesn't start with
`panic`.
* compiler: make getTypeCodeName a method on compilerContext
* testdata: add regression tests for function-local named types
* compiler: disambiguate function-local named types
* Some PR feedback
Register a vectored exception handler at startup so Windows hardware
exceptions can be translated into Go panics. Access violations become
nil pointer panics, and integer divide-by-zero exceptions become divide
by zero panics, allowing defer/recover to handle them like ordinary
runtime panics.
Modify the Unix signal handler to redirect execution to a Go sigpanic
function instead of printing an error and re-raising the signal.
The C signal handler modifies the ucontext to make the faulting
instruction appear to have called tinygo_sigpanic, which then calls
runtimePanic with the appropriate message.
Supported on all architectures TinyGo targets on Linux and Darwin:
x86_64, i386, aarch64, ARM, and MIPS. Also registers SIGFPE, which
was previously not handled at all.
Change hashmap set operations and channel send/close from
createRuntimeCall to createRuntimeInvoke. This places a setjmp
checkpoint before each call, allowing runtimePanicAt to safely
longjmp when these operations panic.
Emit fault checkpoints around compiler-generated runtime assertions so
runtimePanicAt can unwind through the existing defer/recover machinery
instead of aborting. This lets panics from bounds checks, type checks,
and other compiler-inserted runtime checks be recovered by deferred
functions.
Mark functions that call recover as noinline. Inlining such a function
into a deferred closure can make recover observe the wrong call context
and report success when it should return nil.
Addresses tinygo-org/tinygo issues 2759 and 3510.
The optimization to skip cloning the destination object on partial
stores when the buffer is already owned by the current memory view
mutated obj.buffer.buf in place. The previous v.buf clone only handled
the case where the store value itself aliased that buffer; it did not
cover the more common case where an earlier partial load had returned
a slice into the same buffer. The in-place store would then corrupt
the previously loaded value, breaking code such as cloudflare/bn256
where the gfP arithmetic loads, computes, and stores within the same
global during package initialization.
Fix this in load instead: when the source object is owned by the
current memory view, copy the loaded slice so the returned value is
independent of the live buffer. The v.buf clone in store is no longer
needed and is removed.
Updates the regression test added in the previous commit to reflect
the corrected behavior.
Adds a third init function to the store testdata: an initial partial
store puts the buffer into the current memory view, then a partial load
is followed by another partial store at the same offset, and finally the
loaded value is written to a separate global.
The expected output captures the current (incorrect) behavior of the
in-place partial store optimization: the loaded value gets corrupted by
the subsequent in-place store. The next commit fixes this.
memoryView.store cloned the entire destination object on every partial
store. Compiling a program that pulls in golang.org/x/text/collate hit
this hard: its generated tables are initialized one element at a time,
and each element store copied the whole global, making interp quadratic
in the table size.
Switch to copy-on-first-write per memory view: clone the object the
first time this view writes to it, then mutate that private copy in
place on later partial stores. Rollback is unaffected because revert
still discards the view's objects wholesale.
Two extra safety tweaks fall out of this:
- Full overwrites now clone the incoming value, so the stored buffer
can never alias state held by the caller.
- Partial in-place stores snapshot the source buffer, so a store
whose source is a load from the same object (overlapping or not)
still sees the pre-store bytes.
Add an interp golden test exercising both the overlapping load/store
case and a cross-object aliased load/store case.
This has the following effects:
* It fixes issue #5414.
* It bumps the minimum Go version to Go 1.24 (from 1.23).
I looked, and there doesn't seem to be a version of golang.org/x/tools
that supports the improved `new` builtin without also bumping the
minimum Go version.
OpenOCD can be somewhat outdated. `probe-rs` is a more modern tool
written in Rust with many interesting features, like built-in RTT
logging support.
This commit just adds basic support, to be able to flash devices with
probe-rs and debug them.
- Uses PLL1 to boost the system clock from the 4 MHz MSIS default to 160 MHz.
- Sets VOS to Range 1 (1.2V) and enables the EPOD booster for higher frequency support.
- Configures flash latency (4 wait states) and enables prefetch for 160 MHz operation.
- Updates CPU and APB timer frequencies in the machine package accordingly.
- Fixes LPUART baud rate divisor computation by using 64-bit arithmetic to prevent
overflow with the newly increased 160 MHz clock.
Signed-off-by: deadprogram <ron@hybridgroup.com>
This fixes several issues in the STM32 UART implementation:
- `writeByte` now waits for the transmit register to be empty before writing, preventing data loss by avoiding overwriting the shift register.
- `flush` now correctly waits for the Transmission Complete (TC) flag.
- The interrupt handler now clears all error flags (ORE, NE, FE, PE) to prevent interrupt storms, rather than just ORE.
- Extracted `SetBaudRate` so it can be cleanly overridden by specific MCU families.
It also introduces specific fixes for the STM32U5 family:
- Enforces a minimum BRR divisor of 16 in `getBaudRateDivisor` to prevent undefined hardware behavior and CPU starvation.
- Overrides `SetBaudRate` for STM32U585 to momentarily disable the USART (UE=0) before updating the BRR register, which is read-only when enabled.
- Adds a readback after enabling the USART1 clock to ensure the clock is active before register access.
Signed-off-by: deadprogram <ron@hybridgroup.com>
Realized this while looking through the code: there is no way to use
this safely when the scheduler is involved. It can only be used safely
with a `wfe` or similar (part of scheduler code).
* runtime,syscall,internal/poll,os: wasip1 poll_oneoff scheduler integration + net.FileListener
On wasip1 today every syscall.Read/Write blocks the entire wasm module — the
cooperative scheduler invokes poll_oneoff only for sleep/timer wakeups, and
there's no path from the net package to a working TCP server. This change
fixes both: it threads poll_oneoff through the scheduler's idle path so a
goroutine doing FD I/O parks instead of blocking the module, and it provides
enough internal/poll / os / syscall surface that upstream Go's
net.FileListener / net.FileConn works on a host-pre-opened TCP socket.
* runtime: keep scheduler_cooperative idle-wait calls direct
TestBinarySize/hifive1b/examples/echo regressed by 32 bytes after the
previous commit routed the scheduler's idle wait through a
schedulerIdleWait helper. The extra call frame + branch landed on every
non-wasip1 cooperative target, where the original direct sleepTicks /
waitForEvents calls compile to a single inlined call.
This means runtime.alloc isn't allocated every time but instead only
when it is actually needed. It also just cleans up the compiler code a
little bit, and makes the next commit a bit nicer.
* compiler, runtime, reflect: generate type-specific hash/equal for composite map keys
For map keys that are not trivially binary-comparable, the compiler now
generates type-specific hash and equal functions as LLVM IR instead of
going through the interface+reflection path. This covers comparable
types: strings, floats, complex numbers, interfaces, channels, and
composites containing any mix of these.
Previously, maps with composite keys containing strings or floats
converted the key to interface{}, hashed via reflection, and compared
through interface equality. Now the compiler walks struct fields and
array elements directly, dispatching to the right runtime helper for
each field type and storing keys at their actual type.
Struct keys are always handled field-by-field so padding bytes do not
affect equality or hashing. Blank fields are ignored, matching Go
equality. Generated hash/equal function names use canonical underlying
type structure so structurally identical key types can share generated
functions. Padding zeroing before map operations is no longer needed
because structs no longer use the binary key path.
Also fix reflect map iteration for interface-keyed maps: MapIter.Key
returns an interface Value for map[interface{}] keys instead of
unpacking to the concrete key kind.
* compiler: generate loops for array map key hash/equal
Previously, array key hash and equal functions were unrolled at compile
time, generating one block of IR per element. For large arrays like
[1000]int inside a struct with non-binary fields, this caused code
explosion.
Now, binary-element arrays dispatch directly to hash32/memequal for the
whole array. Non-binary-element arrays generate an LLVM IR loop. The
equal loop short-circuits on the first mismatch.
Small arrays are still unrolled instead of looping, keeping the simple
cases compact.
* reflect: fix at-runtime map issues from review, and more found locally
Maps created through reflect.MakeMap need hash/equal behavior that
matches compiler-created maps. Add hashmapMakeReflect for composite key
types, using runtime closures that reconstruct interface{} values from
raw key bytes and delegate to the interface hash and equality paths.
Interface-keyed maps are already stored as interface values, so use the
existing interface hash/equal helpers directly for those. This keeps
reflect insert, lookup, delete, and compiled lookup paths consistent.
Also fix addressable small values used as interface map keys or
interface map values. loadSmallValue puts small indirect values back in
the pointer-sized interface data field the same way valueInterfaceUnsafe
does.
* compiler, interp, reflect: fix pointer map literals; remove interface fallback
Package-level map literals with pointer keys (both *T and
unsafe.Pointer) crash the compiler: the interp pass panics when trying
to hash pointer data as raw bytes, because pointer values in the interp
memory model are symbolic identities that do not fit in a byte.
Fix this by setting a recoverable error flag instead of panicking. The
interp detects the error after each instruction and defers the map
insert to runtime init code, where real addresses are available for
hashing. This matches how the interp already handles other operations
it cannot evaluate at compile time.
With this fix, unsafe.Pointer can also be classified as a binary map
key, which was the last type requiring the interface-based fallback.
Since all comparable types now use either the binary or the
compiler-generated hash/equal path, remove the interface fallback from
the compiler and reflect packages.
* compiler, transform: always pass hash/equal function pointers to hashmapMakeGeneric
The compiler now always resolves the hash and equal functions at compile
time and passes them directly to hashmapMakeGeneric, instead of passing
an algorithm enum to hashmapMake and resolving at runtime. For string
keys, the runtime hashmapStringPtrHash/hashmapStringEqual functions are
referenced directly. For binary keys, hash32/memequal are referenced.
The old hashmapMake with alg enum is retained for reflect, which still
needs runtime resolution when creating maps dynamically.
The OptimizeMaps transform pass is updated to handle both hashmapMake
and hashmapMakeGeneric, and to recognize hashmapGenericSet in addition
to hashmapBinarySet and hashmapStringSet. The now-unused
hashmapCanGenerateHashEqual helper is removed.
* runtime: store large map keys and values indirectly
When a map key or value exceeds 128 bytes, the bucket now stores a
pointer to separately allocated memory instead of the data inline. This
matches Go's MapMaxKeyBytes/MapMaxElemBytes threshold and prevents
bucket sizes from exploding for large key/value types.
For example, map[[256]byte]int previously used 2128 bytes per bucket
(16 header + 256*8 keys + 8*8 values); now it uses 144 bytes per bucket
(16 header + 8*8 pointers + 8*8 values).
The indirection is fully encapsulated in the runtime via helper
functions. Store the computed key and value slot sizes on the hashmap so
all runtime and reflect paths use the same bucket layout, including
non-indirect keys and values.
Add big-key golden coverage and benchmarks. Make the benchmark vary
enough key bytes to exercise hashing.
* interp: bail out of loops that iterate too many times
The existing loop guard (errLoopUnrolled) only fires when a loop body
emits runtime instructions. Loops that are fully evaluable at compile
time, such as inserting thousands of entries into a map, were not
caught and could hang the compiler.
Add a per-basic-block iteration counter that triggers a recoverable
error (errLoopTooLong) when any block is entered more than 1000 times
in a single function call. This defers the init function to runtime,
which is the same behavior as other interp bailouts.
Profiling showed that 83% of CPU time was spent in GC, caused by
allocation pressure from the interp memory cloning on each map
mutation. The iteration limit avoids this entirely by bailing out
before the quadratic cost becomes significant.
Performance on the reproducer from #2090 (map init with strconv.Itoa):
entries before after
5,000 7.4s 2.1s
10,000 17.5s 2.1s
20,000 48.0s 2.8s
65,536 >180s (OOM) 3.2s
* Add -interp-loop-limit
I was seeing bad error messages... but only some of the time. So I went
digging. Eventually I traced it to this one bug, probably introduced in
https://github.com/tinygo-org/tinygo/pull/5290.
This is not just a visual issue, this is actually a bug in TinyGo that
will result in non-reproducible binaries and random behavior
differences.
These methods were added in Go 1.20 but were missing from TinyGo's
sync.Map implementation, causing compilation failures for code that
uses them.
The implementation follows the same lock-based approach as the rest
of TinyGo's sync.Map.
Several places used the (*[1 << N]T)(ptr)[:len:len] pattern to convert
C pointers to Go slices. This has a hardcoded size limit that can panic
if exceeded; RunTool hit this with >1024 linker arguments when many
files are embedded.
Replace all instances with unsafe.Slice, which handles any size.
When a string variable already had a source-level default value, the -ldflags
"-X" flag was silently ignored and the variable remained the default value
at runtime.
Fix by stripping the InitOrder entry for each -X variable before
LoadSSA() is called. With no entry in InitOrder, go/ssa emits no init
store, so the global stays zero-valued in the IR. makeGlobalsModule
still injects the actual -X values at final link time.
The -X values remain out of the per-package build cache with only the
variable names appearing in the cache key.
Signed-off-by: deadprogram <ron@hybridgroup.com>
Rename the "arduino" target to "arduino-uno" to better reflect the
board it represents. The "arduino" name is kept as an alias that
inherits from "arduino-uno" for backward compatibility.
- Rename targets/arduino.json to targets/arduino-uno.json
- Create targets/arduino.json as alias inheriting from arduino-uno
- Rename board_arduino.go to board_arduino_uno.go
- Update build tag from "arduino" to "arduino_uno"
- Rename corresponding example files
Signed-off-by: deadprogram <ron@hybridgroup.com>
Move STATUS_W1TC clear to before the callback loop so that new GPIO
events arriving during handler execution generate a fresh edge on the
CPU interrupt line and are not lost. Matches the same fix applied to
ESP32-S3.
Signed-off-by: deadprogram <ron@hybridgroup.com>
When SPI is configured via the GPIO Matrix, SPI signal transitions set
GPIO.STATUS bits on the routed pins. With a level-triggered CPU interrupt
(line 8), the ISR re-enters continuously as long as any STATUS bit is
asserted — causing user GPIO callbacks to fire spuriously.
Switch cpuInterruptFromPin to CPU interrupt 10, which is edge-triggered
(level 1) on the Xtensa LX7. This ensures the ISR fires once per GPIO
event rather than looping while SPI is active.
Also move STATUS_W1TC clears to before callback dispatch so that new
GPIO events arriving during handler execution generate a fresh edge, and
add writeINTCLEAR(active) in handleInterrupt to properly acknowledge
edge-triggered CPU interrupt pending bits via the INTCLEAR register.
Fixes GPIO interrupts firing constantly when SPI and pin interrupts are
used together.
Signed-off-by: deadprogram <ron@hybridgroup.com>
SPIConfig.CS has a zero value of Pin(0) (GPIO0), but NoPin is Pin(0xff).
When the user does not set CS, the SPI driver sees Pin(0) != NoPin and
configures GPIO0 as the chip select output, hijacking whatever function
that pin was serving such as a button interrupt.
Default config.CS to NoPin when it is the zero value, so an omitted CS
field correctly means "no hardware CS pin". Applied to both ESP32-S3 and
ESP32-C3 SPI drivers.
Signed-off-by: deadprogram <ron@hybridgroup.com>
Ticker.Stop does not drain already-buffered ticks from the channel,
so on slow CI runners a tick may already be queued before Stop takes
effect, causing a spurious "fail: ticker should have stopped!" failure.
Drain the channel immediately after Stop before checking that no new
ticks arrive.
Signed-off-by: deadprogram <ron@hybridgroup.com>
* reflect: implement method-set based AssignableTo and Implements
Based on the design from #4376 by aykevl.
Fixes#4277, fixes#3580.
Co-authored-by: Ayke van Laethem <aykevanlaethem@gmail.com>
* builder: update expected binary sizes for reflect changes
* Make interface checks similar to invoke, allowing typeImplementsMethodSet and method info to be dropped when reflect is not present
* Add more tests that BigGo reflect tests
* Even more pruning
* Add go/token and net/url to passing tests
* Prune even further, I am less happy with this, though
* Update size test now that we are smaller
* Skip some tests
* elide method lists
* format, oops
* fix tests
* Add a panic, pull out constant to keep in sync
* Add debug info
* Remove code that was leftover from a previous refactor
---------
Co-authored-by: Ayke van Laethem <aykevanlaethem@gmail.com>
This fixes issue #5037. A few more wasip1 syscalls needed to be present,
and in particular fd_prestat_get needed to return EBADF on invalid file
descriptors.
Add support for rising, falling, and toggle edge interrupts on all
ESP32-S3 GPIO pins (0-48). The GPIO peripheral interrupt is routed
to CPU interrupt 8 via the interrupt matrix. The ISR reads both
STATUS and STATUS1 registers to cover the full 49-pin range.
Signed-off-by: deadprogram <ron@hybridgroup.com>
The UART handleInterrupt handler unconditionally read RDR on every
interrupt without checking which flag triggered it. On newer STM32
USART peripherals (U5, L4, L5, L0, G0, F7, WL), RXNEIE enables
interrupts for both RXFNE (data ready) and ORE (overrun error).
Unlike older families (F1, F4), ORE is not cleared by reading the
data register, it must be explicitly cleared via the ICR register.
When an overrun occurred (e.g. serial data arriving while ADC
busy-waits in Get()), ORE would trigger the interrupt, the handler
would fire without clearing it, and the interrupt would re-trigger
immediately, causing an infinite interrupt storm that locks up the
CPU.
Fix by:
- Checking RXFNE/RXNE (bit 5) before reading data from RDR
- Clearing ORE (bit 3) via ICR on newer peripherals when set
- Adding errClearReg field to UART struct, set to &Bus.ICR in
setRegisters() for all ICR-capable families
- Preserving the SR+DR clearing sequence for older F1/F4 families
Signed-off-by: deadprogram <ron@hybridgroup.com>
When no USB host is reading, flushAndWait() spins 50K iterations per
FIFO-full event. With putchar calling WriteByte per byte, the cumulative
delay starves I2C and other peripherals, freezing displays.
Add a txStalled flag: the first FIFO-full triggers one flushAndWait
attempt. If it fails (no host), txStalled is set and all subsequent
writes return immediately with no spin — just a register read and a
bool check. When a host reconnects, SERIAL_IN_EP_DATA_FREE goes back
to 1, bypassing the stall path and clearing the flag automatically.
The -march flag for compiling C libraries (compiler-rt, picolibc) was
hardcoded to rv32imac for all riscv32 targets. This is incorrect for
targets that lack the atomic extension, such as ESP32-C3 (rv32imc) and
TKey (rv32iczmmul).
Extract the -march value from the target's cflags when available,
falling back to the previous defaults (rv32imac, rv64gc) for targets
that don't override it.
Fixes#5114
Signed-off-by: deadprogram <ron@hybridgroup.com>
When no USB host is reading (e.g. board not connected to a serial
monitor), WriteByte and Write would spin for up to 200k iterations
per byte waiting for the FIFO to drain. This stalled the entire
application, freezing unrelated peripherals like I2C displays.
Reduce flushTimeout from 200,000 to 50,000 iterations (~3ms) which
is enough for 2-3 USB frames when a host is connected, but short
enough that serial output won't freeze the application when no host
is reading. Serial output is best-effort; callers like putchar
already ignore write errors.
Applies to both ESP32-S3 and ESP32-C3 which share the same USB
Serial/JTAG controller design.
Add a new flash method that uses adb to flash boards over Android Debug
Bridge. The method supports running pre-flash shell commands via
"adb shell", pushing the firmware binary via "adb push", and running
post-flash shell commands with a {remote} token for the remote path.
New target JSON fields:
- adb-pre-commands: shell commands to run before pushing firmware
- adb-push-remote: remote device path for adb push
- adb-post-commands: shell commands to run after pushing firmware
Switch arduino-uno-q target to use the new adb flash method with
openocd running on the remote device.
Signed-off-by: deadprogram <ron@hybridgroup.com>
Go 1.26 replaced the individual Darwin syscall entry points (syscall,
syscallX, syscallPtr, syscall6, syscall6X) with two variadic functions:
syscalln and rawsyscalln. The old wrappers now have Go bodies that call
these, then use errno/errnoX/errnoPtr to interpret the result.
This caused two bugs in our rawsyscalln implementation:
1. Return value truncation: We used call_syscall/call_syscall6 which
return int32, truncating 64-bit results (pointers from fdopendir,
offsets from lseek, addresses from mmap). For example, a DIR* pointer
returned by fdopendir would lose its upper 32 bits on arm64, causing
SIGSEGV when later accessed.
2. Lost 4th argument: The case 4/5/6 fallthrough had 'a3 = args[3]'
instead of 'a4 = args[3]', and a3 was immediately overwritten by
'a1, a2, a3 = args[0], args[1], args[2]'. This lost the 4th syscall
argument entirely, breaking pread (offset=0) and causing wrong data
to be read from files.
Fix by using call_syscallX/call_syscall6X (returning full uintptr),
always reading errno (letting the Go wrappers decide what to do with
it), and correcting the argument assignment.
Go 1.26 changed Darwin's rawSyscall/Syscall from assembly stubs to Go
wrappers that call variadic rawsyscalln/syscalln. When the interp tries
to evaluate syscall.init() at compile time (which calls Getrlimit →
rawSyscall → rawsyscalln), it encounters a ptrtoint of a function
pointer and fails with 'ptrtoint integer size does not equal pointer
size'.
Make this a recoverable error so the interp reverts the init
interpretation and defers it to runtime, matching the behavior for other
uninterpretable init functions.
Go 1.26 added syscall.runtimeClearenv (called by syscall.Clearenv) which
must be provided by the runtime via go:linkname. Without it, the os
package fails to link.
Go 1.26's crypto/internal/sysrand uses internal/syscall/unix.GetRandom
on Linux-like targets (including wasip2 which sets GOOS=linux). The
existing stub panicked with 'todo: unix.GetRandom', causing crypto/ecdsa
tests to fail on wasip2.
Implement GetRandom on WASI targets (wasip1, wasip2) by calling the
arc4random_buf libc function that TinyGo's runtime already provides.
For other TinyGo targets, return ENOSYS so sysrand can fall back to
/dev/urandom.
Go 1.26 added a CPU jitter-based SP 800-90B entropy source for FIPS 140-3
compliance (crypto/internal/entropy/v1.0.0). It declares a 32 MiB global
ScratchBuffer ([1<<25]byte) used for memory access timing noise. On systems
with virtual memory this stays in .noptrbss and is lazily paged, but on
baremetal targets it becomes static RAM, causing fatal overflow (e.g.
pybadge with 192 KB SRAM reports 33 MB overflow).
Since FIPS jitter entropy is never used on TinyGo targets (fips140.Enabled
is always false, so drbg.Read falls through to sysrand.Read), provide a
TinyGo overlay that replaces ScratchBuffer with [0]byte and stubs out all
entropy functions with panics.
Go 1.26 added SWAR optimizations to unicode/utf8 that use:
const ptrSize = 4 << (^uintptr(0) >> 63)
const hiBits = 0x8080808080808080 >> (64 - 8*ptrSize)
This formula only distinguishes 32-bit and 64-bit architectures.
On AVR (16-bit uintptr), ptrSize computes as 4, and hiBits becomes
0x80808080 (2155905152) which overflows the 16-bit uintptr type.
Fix by providing a patched unicode/utf8 overlay for Go 1.26+ that
uses a ptrSize formula handling all three sizes (16/32/64-bit):
const ptrSize = 1 << (^uintptr(0)>>15&1 + ^uintptr(0)>>31&1 + ^uintptr(0)>>63&1)
This evaluates to 2 on AVR, 4 on 32-bit, and 8 on 64-bit. The
word() helper also gains a ptrSize==2 case for 16-bit loads.
The previous approach of erasing basic blocks one-by-one from
duplicate function definitions could crash with a segfault. When
a function has complex control flow (branches, switches, PHI nodes),
deleting a basic block that is still referenced by instructions in
other blocks of the same function causes use-after-free in LLVM.
LLVM's C++ Function::deleteBody() avoids this by calling
dropAllReferences() on all blocks first, but that API is not
exposed in the Go LLVM bindings.
Fix this by using a completely different approach: instead of
manually deleting the function body, set the duplicate function's
linkage to LinkOnceODRLinkage. This tells the LLVM linker that the
definition can be merged with another copy, causing it to prefer
the already-linked ExternalLinkage definition in the destination
module. The result is the same (the runtime's definition wins) but
without any unsafe block manipulation.
Go 1.26 changed all Windows syscall wrappers in zsyscall_windows.go
to use SyscallN instead of fixed-argument Syscall/Syscall6/etc. The
SyscallN function now has a body that calls an unexported syscalln
function (provided by runtime via //go:linkname).
TinyGo's existing createSyscall compiler builtin used call.Args[2:]
to extract syscall arguments, but for variadic SyscallN the SSA
representation passes args as a slice value (not individual args),
causing call.Args[2:] to be empty -- resulting in zero arguments
being passed to Windows API calls and 0xc0000005 access violations.
Fix this by:
1. Excluding syscall.SyscallN from builtin interception, letting
Go 1.26's function body compile normally (it calls syscalln)
2. Adding a new createSyscalln compiler builtin that intercepts
syscall.syscalln and correctly handles the variadic slice:
- Generates a switch on the arg count n (0-18 cases)
- Each case loads args from the slice via GEP/Load
- Wraps calls with SetLastError(0)/GetLastError() as before
- Handles i386 stdcall conventions
3. Adding runtime stubs for both Go versions:
- go1.26: syscall.syscalln stub (body intercepted by compiler)
- pre-go1.26: syscall.SyscallN stub (linker satisfaction)
Go 1.26 changed syscall.loadlibrary, syscall.loadsystemlibrary, and
syscall.getprocaddress from declarations to definitions in
syscall/dll_windows.go. TinyGo's runtime also defines these via
//go:linkname, causing "symbol multiply defined!" during LLVM module
linking.
Resolve this by detecting duplicate function definitions before calling
llvm.LinkModules and turning the incoming duplicate into a declaration,
so the runtime's version wins. TinyGo's implementations must take
precedence because Go 1.26's versions depend on //go:cgo_import_dynamic,
which TinyGo does not support.
Also fix the signature of syscall_loadsystemlibrary to match the
standard library (remove unused absoluteFilepath parameter).
Signed-off-by: deadprogram <ron@hybridgroup.com>
The previous implementation was a bare tail-jump to tinygo_scanstack
without spilling any registers or passing an sp argument. On Xtensa
windowed ABI, heap pointers held in physical registers were invisible
to the conservative GC, causing it to collect live objects and leading
to nil pointer dereferences under allocation pressure.
Flush all register windows to the stack using recursive call4 (15
levels for NAREG=64), then pass the current sp to tinygo_scanstack so
the GC scan from sp to stackTop covers every live value. Interrupts
are briefly masked during the spill to prevent window-overflow
exceptions from interfering.
Fixes crashes on ESP32-S3 observed when serving concurrent HTTP
requests.
Signed-off-by: deadprogram <ron@hybridgroup.com>
Rewrite kernel and double exception handlers to save EXCCAUSE/EPC1 to
RTC STORE registers before triggering a software reset, replacing the
LED-blink diagnostic with post-mortem debug info that survives reset.
Add user exception dispatch in the level-1 handler with a weak
espradio_user_exception symbol so programs without espradio still link.
Implement procPin/procUnpin for Xtensa using RSIL/WSR PS to properly
disable interrupts during atomic operations. Fix abort() to use a
waiti loop instead of bare spin.
Add --wrap ldflags for malloc/calloc/free/realloc/ppCheckTxConnTrafficIdle
to support espradio WiFi blob integration.
Signed-off-by: deadprogram <ron@hybridgroup.com>
Replace the ROTW-based register flush with a recursive call4 approach
that properly triggers hardware window-overflow exceptions. ROTW only
modifies WindowBase without saving registers, causing corruption when
switching goroutines. The recursive call4 correctly spills all 15 window
panes. Also clear WindowStart after the stack switch to prevent stale
overflow of garbage register values.
Add tinygo_task_current export for C interop.
Signed-off-by: deadprogram <ron@hybridgroup.com>
Extend the linker script with proper IROM/DROM section layout for flash
execute-in-place. Update the boot assembly MMU init to dynamically map
all required flash pages based on _irom_end/_drom_end symbols instead
of hardcoding a single page.
Signed-off-by: deadprogram <ron@hybridgroup.com>
The ESP32-S3 ROM bootloader loads IRAM/DRAM into SRAM but does not
configure the flash cache or MMU. Previously the target incorrectly
reused the ESP32 boot assembly (esp32.S) which lacks flash XIP support.
Add a dedicated esp32s3.S boot assembly that:
- Sets up windowed-ABI registers, stack, and FPU
- Disables all watchdog timers (RTC, TIMG0, TIMG1, Super WDT)
- Configures VECBASE and clears PS.EXCM before any callx4
- Calls ROM functions to configure cache modes:
rom_config_instruction_cache_mode (16KB, 8-way, 32B line)
rom_config_data_cache_mode (32KB, 8-way, 32B line)
- Initializes MMU, maps flash page 0 for IROM and DROM,
clears bus-shut bits, and enables both caches
- Jumps to runtime.main in IROM (flash)
Update the linker script (esp32s3.ld) to place .text and .rodata in
flash-mapped regions (IROM/DROM) with proper alignment for the MMU
page size. Update esp32s3-interrupts.S with proper exception vector
handlers. Point esp32s3.json at the new esp32s3.S instead of esp32.S.
Signed-off-by: deadprogram <ron@hybridgroup.com>
This allows for changes to the 'short-enums' flag when
using CGo since TinyGo already does this, and it is
needed for the esp32s3 processor.
Signed-off-by: deadprogram <ron@hybridgroup.com>
The "all:" prefix in //go:embed directives instructs the compiler to
include hidden files (starting with "." or "_") when embedding a
directory. Previously, the prefix was passed literally to path.Match,
which would never match.
Introduce an embedPattern struct that parses and validates the pattern
at construction time via newEmbedPattern(). The "all:" prefix is
stripped once and stored along with the glob pattern, avoiding repeated
string prefix checks. Pattern validation is absorbed into the
constructor, guaranteeing that any embedPattern value is valid and
eliminating the need for separate validation loops.
Fixes embedding with patterns like "//go:embed all:static".
Replace the busy-wait sleepTicks with an interrupt-driven version that
sets a TIMG0 timer alarm and waits for the interrupt to fire. The timer
alarm handler disables INT_ENA at the peripheral level to prevent
level-triggered re-assertion; sleepTicks re-enables it after each wake.
This avoids burning CPU cycles during time.Sleep and similar delays.
Replace the minimal inline ISR (which only disabled INTENABLE) with a
full level-1 interrupt handler that saves/restores the interrupted
context and dispatches to Go's handleInterrupt.
The handler uses callx4 (not callx0) to call into Go code because:
- callx0 does not set PS.CALLINC, so the Go function's entry
instruction uses stale CALLINC from the interrupted code, causing
wrong window rotation and a garbage stack pointer.
- callx4 explicitly sets CALLINC=1, and our frame pointer (a1) is
outside the callee's register window so it is preserved.
Also updates the USB Serial/JTAG ISR to disable INT_ENA (peripheral
level) instead of relying on INTENABLE, and adds signalInterrupt to
the dispatcher so sleepTicks can be woken by any interrupt.
* feat: add inheritable-only field to hide processor-level targets from listing
Processor-level targets like esp32, rp2040, etc. have flash-method set
so they leak through the existing heuristic filter in GetTargetSpecs and
appear in `tinygo targets` output. Add an "inheritable-only" JSON field
that is checked on raw JSON before inheritance resolution, preventing
these non-board targets from being listed while keeping them loadable
for direct builds and inheritance.
Ref: tinygo-org/tinygo#5178
* fix: prevent inheritable-only from propagating to child targets
The InheritableOnly bool field was propagating from parent to child
targets through overrideProperties, which would hide board targets
from GetTargetSpecs. Fix by preserving/restoring the field across
resolveInherits. Also simplify GetTargetSpecs by removing the raw
JSON pre-check workaround and remove nonexistent esp32c6 from tests.
* prevent build commands to use inheritable only targets
* esp32s3: add interrupt support
This finally adds the long awaited support for interrupts on the
Xtensa arch. Initially just for the ESP32-S3 but then others.
Signed-off-by: deadprogram <ron@hybridgroup.com>
* esp32s3: get interrupts working correctly
There were a number of needed changes in order to get interrupts correctly working
on the esp32s3 processor:
- PS.UM=1 in interruptInit() - routed interrupts to user exception vector (0x340)
instead of kernel (0x300)
- Inline ISR in the vector slot - external handlers via j/call0 crashed (likely
clang Xtensa literal pool issue with large movi constants in separate sections)
- Disable INTENABLE (not just INT_CLR) - the USB RX interrupt is level-triggered;
clearing INT_CLR alone causes infinite re-entry since data is still in the FIFO
- Buffered() re-enables INTENABLE after draining the hardware FIFO
Signed-off-by: deadprogram <ron@hybridgroup.com>
---------
Signed-off-by: deadprogram <ron@hybridgroup.com>
This fixes an "index out of range" panic that occurs when calculating
binary sizes. The strings.SplitN operation in findPackagePath can sometimes
return a slice with only one element, so we now verify the length of the
slice before attempting to access the second element.
Signed-off-by: deadprogram <ron@hybridgroup.com>
Adds `.coexiram*`, `.wifiorslpiram*`, and `.iram1*` to the main `.iram`
segment in `esp32c3.ld`. This resolves the "Invalid image block, can't
boot" error on the ESP32-C3 by ensuring these precompiled ESP-IDF blob
sections are correctly aligned in memory.
Signed-off-by: deadprogram <ron@hybridgroup.com>
- Add `machine_esp32c3_flash.go` to implement the `BlockDevice` interface for ESP32-C3.
- Map internal ESP ROM SPI flash and cache invalidation functions via CGo.
- Update `targets/esp32c3.ld` to expose `__flash_data_start` and `__flash_data_end` linker variables.
- Add `esp32c3` to build tags in `src/machine/flash.go`.
- Ensure atomic flash operations by disabling interrupts and invalidating cache to prevent stale reads.
Signed-off-by: deadprogram <ron@hybridgroup.com>
The SVD-generated TIM constants in the STM32 device files have all per-channel
fields shifted by one channel position (e.g., CC1E is at the hardware position
of CC2E, OC1M_Pos is at the OC2M hardware position). This caused Set(), Unset(),
SetInverting(), and interrupt handlers to write to wrong bit positions.
Signed-off-by: deadprogram <ron@hybridgroup.com>
This changes the default stack size for the esp32-c3 and esp32-s3
to 8kb. Since they have more onboard memory and a larger stack size is
needed for any networking etc this default seems like a more sensible
one, in similar way to the rp2040/rp2350 defaults.
Signed-off-by: deadprogram <ron@hybridgroup.com>
Signed-off-by: deadprogram <ron@hybridgroup.com>
machine/stm32u585: fix PWR peripheral clock was never enabled
On STM32U5, PWR is on AHB3 and requires RCC.AHB3ENR.PWREN -
unlike some other STM32 families where PWR is always clocked.
Without this, all writes to PWR registers (including IO2SV for
VDDIO2) were silently dropped, so GPIOG pins could never drive.
Signed-off-by: deadprogram <ron@hybridgroup.com>
target: correct I2C pin mapping for Arduino UNO Q
Signed-off-by: deadprogram <ron@hybridgroup.com>
The complex 160MHz PLL initialization was hanging the MCU.
The fix: Replaced the entire PLL-based clock init with the MSI 4MHz default
Signed-off-by: deadprogram <ron@hybridgroup.com>
Add SPIv2 driver for STM32U5 using TXDR/RXDR and CFG1/CFG2 registers,
which differ from the classic SPI peripheral on older STM32 families.
Implements Configure() and single-byte Transfer().
Add SPI1 peripheral and pin definitions to arduino-uno-q board.
Exclude stm32u5 from the shared classic SPI build tag.
Add I2C timing values for STM32U585 at 160MHz PCLK1 (10/100/400/500 KHz).
Add I2C1 peripheral and pin definitions to arduino-uno-q board.
Update i2c_revb build tag to include stm32u5.
Add machine-level support for STM32U5 family: GPIO ports A-I with clock
enables via AHB2ENR1, EXTI pin interrupts (individual EXTI0-15 IRQs),
timer definitions (TIM1-8, TIM15-17), RNG, and alternate function clock
enables for all peripherals.
Add UART support with STM32U585 baud rate and register configuration.
Add arduino-uno-q board pin definitions and USART2 serial.
Update shared build tags: gpio_revb, gpio_reva, exti_exti, exti_syscfg.
Add processor target (cortex-m33), linker script (2048K flash, 768K RAM),
and arduino-uno-q board target. Runtime initializes 160MHz clock via PLL1
from HSI16, with PWR Range 1 and EPOD booster.
Also add psctype abstraction for timer PSC register width, needed because
the U5 SVD defines PSC as a 16-bit register.
This improves the ESP32-C3 USBDevice implementation by using interrupts
to properly handle data RX/TX.
It also stubs out USB interfaces for HID functions, since those cannot
be implemented on ESP32-C3 due to using a JTAG-USB interface.
Signed-off-by: deadprogram <ron@hybridgroup.com>
MCAUSE was never being cleared after handling an interrupt.
On RISC-V, mret does NOT zero MCAUSE — it retains the last
trap cause. Every other TinyGo RISC-V target (FE310, K210,
QEMU) explicitly does riscv.MCAUSE.Set(0) after handling.
The ESP32-C3 was missing this.
Signed-off-by: deadprogram <ron@hybridgroup.com>
This adds a new flash method 'esp32jtag' to explicitly define when this
reset method is needed. When flashing esp32c3/esp32s3 boards on Windows
this method of board reset is required, otherwise the reset does not
take place and the board cannot be flashed.
Signed-off-by: deadprogram <ron@hybridgroup.com>
This refactors and corrects the SPI implentation for the
ESP32C3 and ESP32S3 processors. There was a lot of duplicated
code, as well as some errors such as incorrectly calculating
speed on the esp32c3 implementation.
This will also be helpful when adding additional processors
that use very similar peripheral registers.
Signed-off-by: deadprogram <ron@hybridgroup.com>
This switches the rp2040 and rp2350 to use the tasks scheduler
by default instead of using the cores scheduler. Too many race
conditions at present, we need to look into exactly why. In
the meantime, going back to the tasks as default will address
a lot of these intermittent problems.
Signed-off-by: deadprogram <ron@hybridgroup.com>
This refactoring reduces code duplication from the esp32c3/esp32s3 ADC
implementation, by reusing the register/efuse calibration code since the
same basic procedures are used by both processors.
Signed-off-by: deadprogram <ron@hybridgroup.com>
* begin adding ring512 implementation
* refactor to make operation driven fuzz test
* refactor USBCDC.Read to use ring512
* try txhandler separate to flush
* working USBCDC with large packets
* fix binary size
* remove comment
* documentation improvements
This adds usage of the new espflash package to perform flashing on
ESP32, ESP32S3, ESP32C3, & ESP8266 boards. This means that you no longer
have to install esptool.py in order to flash ESP32 based boards.
Signed-off-by: deadprogram <ron@hybridgroup.com>
Recent changes in stm32-svd result in a change from previous
16-bit register access to 32-bit access.
Both access types are allowed, according to the register manuals.
Previously, there was no specific stm32f4r5.svd in lib/stm32-svd,
just stm32f4x5.svd was used; now, both files are present. This means
that the existing build-tag stm32f4r5 will include the
device/stm32/stm32f4r5.go file, and the additional build-tag stm32f4x5
would include the device/stm32/stm32f4x5.go file as well, resulting
in build conflicts. Renaming just the tag, which is used in src/machine,
and src/runtime, into stm32f4y5 solves this issue.
* tools/gen-device-svd: orderPeripherals: prevent skipping base peripherals derived by name
Recent SVDs from stm32-rs, like the one for stm32u595, define derivedFrom
attributes that do not refer to a peripheral group name, but to a
peripheral name. For instance, the peripheral I2C5 may be derived from
"I2C1", not from "I2C". The previous algorithm records, in case the
group name is non-empty, only the group name in the
knownBasePeripherals map, not the name of the peripheral itself.
So in case of the base peripheral I2C1 with group name I2C: although
the peripheral gets added to the sorted list, it would be added to
knownBasePeripherals with the group name "I2C" as key, not with "I2C1".
A following peripheral, I2C5, derived from I2C1, with an empty group
name, would be recorded as known with key "I2C5", but omitted from the
sorted list, because "I2C1" is not recognized as known.
The following peripheral SEC_I2C5, derived from I2C5, with empty group
name, would be added to both the knownPeripherals map and the sorted list.
So if, later, the sorted list is examined, it would find SEC_I2C5
earlier than its base peripheral I2C5, which would be missing from
"peripheralDict", resulting in a nil pointer access.
This patch makes sure that, to stay with the example, that
"I2C1" is recorded as known too (not only the group name "I2C"),
so that "I2C5" won't be skipped anymore, preventing the program from
crashing.
* tools/gen-device-svd: orderPeripherals: ensure ordered content of missingBasePeripherals
After the first run, missingBasePeripherals may contain peripherals
with dependencies that are not guaranteed to be in proper order.
This change implements additional loop runs that try to reduce the
size of the missingBasePeripherals as far as possible.
[With recent SVDs from stm32-rs this change will not produce different
results, though (since these source files contain already properly
ordered peripherals).]
* tools/gen-device-svd: Register: move dim array decoding to utility type dimArray
This allows encoding of dim increment and array indices to be re-used
by other elements supporting dim arrays.
This change just restructures parts of register specific code,
it does not change the output of the program.
* tools/gen-device-svd: parseBitfields: support field dim arrays
Patched SVD files from stm32-rs recently contain many fields with
dim array parameters and names containing %s (like "CC%sIF").
This change adjusts parseBitfields so that these field elements
get resolved.
* tools/gen-device-svd: SVDField: allow multiple enumeratedValues
In recent patched SVD files from stm32-rs there may be two
enumeratedValues elements per SVDField, not just one.
The SVD specification allows up to two entries (they may be used
to define different enums for read and write access).
This change extends SVDField and parseBitfields so that
two enumeratedValues are processed like a single one.
* tools/gen-device-svd: orderPeripherals: sort peripherals of same group with larger number of registers/bitfields first
In group "TIM" there may be general purpose timers like TIM16 and
advanced timers like TIM1. The advanced peripheral may contain a larger
number of registers than the general purpose ones.
TIM1 may contain CCR1..CCR4, SMCR and OR1, while TIM16 only knows
about CCR1.
Unfortunately in some SVDs, like the one for stm32g031, TIM16 is defined
before TIM1. Since register and bitfield constants are generated taking
only the first peripheral of a group into account, the resulting .go
file may lack definitions for e.g. CCR2..CCR4, SMCR, and OR1.
This change adjusts orderPeripherals so that, to stay with the example,
a peripheral like TIM1 will be moved in front of TIM16, resulting in an
output file containing the larger set of definitions.
* tools/gen-device-svd: SVDEnumeration: support isDefault
Recent SVD files created by stm32-rs use "isDefault" in enumeratedValue
elements for purposes like the Div1 enum for clock prescaler registers
without specifying a specific value.
Previously, these enumeratedValues would be skipped because of the
enumEl.Value == 0 condition, and the corresponding const definitions
like "RCC_CFGR2_PPRE2_Div1 = 0x0" would be missing from the
resulting .go files, so existing code relying on these constants would
not compile anymore.
This change adds a utility type enumDefaultResolver that helps
finding an actual value that is unused by the enumeratedValues that
are defined for the field. More examples for values marked
as "isDefault", along with their resolved values:
DAC_CR_WAVE1_Triangle => 2
IWDG_PR_PR_DivideBy256 => 6
DAC_CR_MAMP2_Amp4095 => 0xb
* tools/gen-device-svd: support derivedFrom attribute at field level
This ensures that some more constants are included in the .go files
that would otherwise be skipped (like e.g. ADC_SMPR2_SMP1_Cycles*
of some STM32 devices), which prevented compilation of some programs.
To avoid extending a lot of func argument lists, and since there
is no context.Context in use yet, this change introduces a
global derivationContext.
* tools/gen-device-svd: tweak: stm32: ensure USART_ISR_TXE/TXFNF are present
* tools/gen-device-svd: stm32: ensure CCMR*_Output alternate registers are sorted first
* tools/gen-device-svd: stm32: add IWDG peripheral alias if SVD defines IWDG1
The compiler may generate calls to fminimum/fmaximum on some platforms.
Neither of the libm implementations we statically link against have these functions yet.
Implement them ourselves.
This changes the order for initialization of the random number
seed generation on wasm platforms until after the heap has been
initialized. Should fix#5198
Signed-off-by: deadprogram <ron@hybridgroup.com>
* machine/attiny85: add USI-based SPI support
Implement SPI communication for ATTiny85 using the USI (Universal Serial
Interface) hardware in three-wire mode. The ATTiny85 lacks dedicated SPI
hardware but can emulate SPI using the USI module with software clock
strobing.
Implementation details:
- Configure USI in three-wire mode for SPI operation
- Use clock strobing technique to shift data in/out
- Pin mapping: PB2 (SCK), PB1 (MOSI/DO), PB0 (MISO/DI)
- Support both Transfer() and Tx() methods
The implementation uses the USI control register (USICR) to toggle the
clock pin, which triggers automatic bit shifting in hardware. This is
more efficient than pure software bit-banging.
Current limitations:
- Frequency configuration not yet implemented (runs at max software speed)
- Only SPI Mode 0 (CPOL=0, CPHA=0) supported
- Only MSB-first bit order supported
* machine/attiny85: add SPI frequency configuration support
Add software-based frequency control for USI SPI. The ATtiny85 USI lacks
hardware prescalers, so frequency is controlled via delay loops between
clock toggles.
- Calculate delay cycles based on requested frequency and CPU clock
- Fast path (no delay) when frequency is 0 or max speed requested
- Delay loop uses nop instructions for timing control
* machine/attiny85: add SPI mode configuration support
Add support for all 4 SPI modes (Mode 0-3) using USI hardware:
- Mode 0 (CPOL=0, CPHA=0): Clock idle low, sample on rising edge
- Mode 1 (CPOL=0, CPHA=1): Clock idle low, sample on falling edge
- Mode 2 (CPOL=1, CPHA=0): Clock idle high, sample on falling edge
- Mode 3 (CPOL=1, CPHA=1): Clock idle high, sample on rising edge
CPOL is controlled by setting the clock pin idle state.
CPHA is controlled via the USICS0 bit in USICR.
* machine/attiny85: add LSB-first bit order support
Add software-based LSB-first support for USI SPI. The USI hardware only
supports MSB-first, so bit reversal is done in software before sending
and after receiving.
Uses an efficient parallel bit swap algorithm (3 operations) to reverse
the byte.
* GNUmakefile: add mcp3008 SPI example to digispark smoketest
Test the USI-based SPI implementation for ATtiny85/digispark.
* machine/attiny85: minimize SPI RAM footprint
Reduce SPI struct from ~14 bytes to 1 byte to fit in ATtiny85's limited
512 bytes of RAM.
Changes:
- Remove register pointers (use avr.USIDR/USISR/USICR directly)
- Remove pin fields (USI pins are fixed: PB0/PB1/PB2)
- Remove CS pin management (user must handle CS)
- Remove frequency control (runs at max speed)
- Remove LSBFirst support
The SPI struct now only stores the USICR configuration byte.
* Revert "machine/attiny85: minimize SPI RAM footprint"
This reverts commit 387ccad494.
* machine/attiny85: reduce SPI RAM usage by 10 bytes
Remove unnecessary fields from SPI struct while keeping all functionality:
- Remove register pointers (use avr.USIDR/USISR/USICR directly)
- Remove pin fields (USI pins are fixed: PB0/PB1/PB2)
- Remove CS pin (user must manage it, standard practice)
Kept functional fields:
- delayCycles for frequency control
- usicrValue for SPI mode support
- lsbFirst for bit order support
SPI struct reduced from 14 bytes to 4 bytes.
---------
This simplifies the process of constructing and encoding layout bitmaps.
Instead of creating big integers and merging them, we can create a pre-sized bitmap and set positions within it.
This also changes the encoding logic to allow larger layouts to be encoded inline.
We would previously not encode a layout inline unless the size was less than the width of the data field.
This is overly conservative.
A layout can be encoded inline as long as:
1. The size fits within the size field.
2. All set bits in the bitmap fit into the data field.
* machine/attiny85: add PWM support for Timer0 and Timer1
Add complete PWM implementation for ATtiny85, supporting both Timer0
and Timer1 with their respective output channels:
- Timer0: 8-bit timer for pins PB0 (OC0A) and PB1 (OC0B)
- Timer1: 8-bit high-speed timer for pins PB1 (OC1A) and PB4 (OC1B)
Timer1 provides more flexible period control with configurable top value
(OCR1C) and extended prescaler options (1-16384), making it well-suited
for LED PWM control and other applications requiring variable frequencies.
Implements full PWM interface including Configure, SetPeriod, Channel,
Set, SetInverting, Top, Counter, and Period methods.
* machine/digispark: document PWM support on pins
Add documentation to the Digispark board file indicating which pins
support PWM output:
- P0 (PB0): Timer0 channel A
- P1 (PB1): Timer0 channel B or Timer1 channel A
- P4 (PB4): Timer1 channel B
Includes package comment explaining Timer0 vs Timer1 capabilities,
with Timer1 recommended for more flexible frequency control.
* machine/attiny85: optimize PWM prescaler lookups
Replace verbose switch statements with more efficient implementations:
- SetPeriod: Use bit shift (top >>= prescaler-1) instead of 15-case
switch for dividing uint64 by power-of-2 prescaler values
- Period: Replace switch statements with compact uint16 lookup tables
for both Timer0 and Timer1, casting to uint64 only when needed
This addresses review feedback about inefficient switch-based lookups.
On AVR, this approach is significantly smaller:
- Bit shifts for uint64 division: ~34 bytes vs ~140 bytes
- uint16 tables: 22 bytes code + 32/16 bytes data vs ~140 bytes
- Total savings: ~190 bytes (68% reduction)
* examples/pwm: add digispark support and smoketest
Add digispark.go configuration for PWM example using Timer1 with pins P1 (LED) and P4. Also add digispark PWM example to GNUmakefile smoketests.
---------
* feat: Implement Lchown function for changing file ownership
- Added Lchown function to change the numeric uid and gid of a file or symbolic link.
- Included error handling to return *PathError for failed operations.
- Added unit tests for Lchown to verify behavior on different error scenarios.
* fix: fix chmod tests
Make timeoffset atomic to be able to handle changing the system
time. Otherwise the scheduler can gets rather confused if you call
AdjustTimeOffset when there are multiple goroutines already running.
Signed-off-by: deadprogram <ron@hybridgroup.com>
* - Fix SPI initialization by disabling interface before configuration.
- Ensure proper 8-bit configuration for STM32G0 SPI data register access.
- Add default pin initialization for SPI.
* Add Window Watchdog (WWDG) and ADC support for STM32G0
* Comment cleanup
* Refactored STM32G0 ADC sampling time configuration to use `switch` for improved readability
* Add STM32G0B1 target support
Introduce support for STM32G0B1 microcontrollers, including target-specific JSON files, linker scripts, and runtime initialization. This update adds hardware support for GPIO, UART, SPI, I2C, timers, and additional board-specific configurations like Nucleo-G0B1RE.
* Update STM32G0 clock initialization to 64MHz and adjust related configurations
Reconfigure STM32G0 to use a 64MHz system clock via PLL with HSI16 as the source. Update flash latency, prescaler settings, and I2C timing values to reflect the new frequency.
* Cleanup
* Cleanup
* Add STM32G0-specific UART implementation
Introduce a new UART implementation for the STM32G0 series with chip-specific setup and configuration methods. Update the generic STM32 UART code to exclude STM32G0.
* Refactor STM32G0 runtime and machine code to utilize chip-specific register access functions
Simplify and standardize register operations with dedicated setter methods in the STM32G0 runtime and machine code and cleanup redundant syntax.
* Remove redundant commented-out APBENR1 register operations in STM32G0 machine code
* Introduce FDCAN support for STM32G0B1 series
Add FDCAN peripheral implementation targeting STM32G0B1, including support for standard, extended identifiers, and bit rate configuration. Update board files to include FDCAN pins, instances, and clock configuration for Nucleo-G0B1RE and Amken Trio boards.
The GC bitmap length is measured in multiples of the pointer alignment.
This is equal to the pointer size on all architectures except AVR.
Replace the hardcoded lengths with lengths that are computed naturally.
The compiler now implements the copy builtin directly instead of calling sliceCopy.
The length is calculated with the llvm.umax.* intrinsics, and the move is performed by llvm.memmove.*.
Both of these operations are easily understood by LLVM's optimization passes.
The type's alignment is also provided to llvm.memmove.*, which is useful when rewriting the move.
Interp no longer needs to reimplement sliceCopy.
Some edge case handling was implemented by sliceCopy but not llvm.memmove.*/llvm.memcpy.*.
I copied this over, so copies of external slices should work now.
Volatile moves/copies are now run at runtime by interp.
There is a 4-byte size increase due to some confusing length logic in sendUSBPacket.
I will look at sendUSBPacket in a future PR.
This fixes two edge cases for min/max:
min/max(+0.0, -0.0) = -0.0, +0.0
min/max(number, NaN) = NaN, NaN
The compare and select method does not work here.
I switched to the llvm.minimum/llvm.maximum intrinsics which match the intended behavior.
The integer min/max were also swapped over to using intrinsics.
The compare and select path is now only used by strings.
Remove the sendUSBPacket maxLen param because this greatly confused the compiler.
It also fixes a bug where the length provided to the hardware may not match the length of the packet.
sendUSBPacket now panics if the sent packet is too big.
I also fixed some of the string descriptor logic where we could create a packet without fully populating it.
RP2* systems might require some more work since they are implemented very differently?
I don't have any of those to test with yet, so maybe someone can deal with them in a seperate PR?
The reason is that allocating without a heap is usually more visible
than using an uninitialized random generator which is subtle and may
lead to security vulnerabilities.
Based on suggestion from @eliasnaur
Signed-off-by: deadprogram <ron@hybridgroup.com>
The data to write is specififed in words, not in bytes.
This fix is needed for correctness. Without it, a
`machine.Flash.WriteAt` call can result in **data loss** since it will
overwrite more data than it should.
Now that the drivers repo and friends are intended to target the most
recent release version of TinyGo instead of the development version,
this trigger does not serve any puppose to trigger a build that will have
already been tested.
Signed-off-by: deadprogram <ron@hybridgroup.com>
ahocorasick no longer times out during interp. I don't believe
this is because we fixed a bug in interp but rather because "something"
(garbage collector fixes?) has moved the tree building to runtime.
Instead of looping over each block, we can use bit hacks to operate on an entire state byte.
I deinterleaved the state bits in order to enable these tricks.
Sweep used to count free/freed allocations/blocks.
I managed to move/remove all of these counters:
- The free space is now calculated in buildFreeRanges by adding the range lengths.
- ReadMemStats counts freed objects by subtracting live objects from allocated objects.
- gcFreedBlocks was never necessary because MemStats.HeapAlloc is the same as MemStats.HeapInUse.
The allocator originally just looped through the blocks until it found a sufficiently-long range.
This is simple, but it fragments very easily and can degrade to a full heap scan for long requests.
Instead, we now maintain a sorted nested list of free ranges by size.
The allocator will select the shortest sufficient-length range, generally reducing fragmentation.
This data structure can find a range in time directly proportional to the requested length.
This updates the stack slot pass to include callers of external functions which may access non-argument memory.
Wiithout this change, a use-after-free could occur on WASM when calling a reentrant function or switching to another goroutine.
Enabling USB consumes a lot of power. So it's better to keep it disabled
by default when the serial port is set to something other than "usb"
(USB-CDC).
On my nice!nano clone, it reduces current consumption from 1mA to
0.13mA (and most of the rest is probably consumed by a connected SCD41
sensor).
This change might break some expectations, when a board uses USB but not
USB-CDC (I think this is rare, but I didn't check specifically).
The setjmp code sets r24 to 0 and checks if it is still 0 when it finishes.
This usually works because tinygo_longjmp stores the lower half of the stack pointer into r24.
However, r24 will be set to 0 if the stack is aligned to 256 bytes when entering the setjmp.
Manually set r24 to 1 in tinygo_longjmp to fix this issue.
Loop over valid pointer locations in heap objects instead of checking if each location is valid.
The conservative scanning code is now shared between markRoots and the heap scan.
This also removes the ending alignment requirement from markRoots, since the new scan* functions do not require an aligned length.
This requirement was occasionally violated by the linux global marking code.
This saves some code space and has negligible impact on performance.
The blocks GC originally used a fixed-size stack to hold objects to scan.
When this stack overflowed, the GC would fully rescan all marked objects.
This could cause the GC to degrade to O(n^2) when scanning large linked data structures.
Instead of using a fixed-size stack, we now add a pointer field to the start of each object.
This pointer field is used to implement an unbounded linked stack.
This also consolidates the heap object scanning into one place, which simplifies the process.
This comes at the cost of introducing a pointer field to the start of the object, plus the cost of aligning the result.
This translates to:
- 16 bytes of overhead on x86/arm64 with the conservative collector
- 0 bytes of overhead on x86/arm64 with the precise collector (the layout field cost gets aligned up to 16 bytes anyway)
- 8 bytes of overhead on other 64-bit systems
- 4 bytes of overhead on 32-bit systems
- 2 bytes of overhead on AVR
VDDH is typically connected to USB (~5V) or directly to a battery.
Measuring the voltage on VDDH is a way to measure the current battery
voltage, which in turn can be useful to determine how much power is left
in the battery.
I've special cased the nrf52840 in a somewhat unusual way, maybe there
is a better way, feel free to comment!
In order to scan stacks, the GC preempts all other threads and has them scan their own stack.
This is somewhat expensive since all of these threads have to fight over a single lock.
Instead, save the stack bounds and let the GC thread perform the scan.
This also fixes a few other bugs I ran into:
1. The GC starts scanning before the world stops. This can cause it to miss some objects (and mistakenly free them) if memory is modified while stopping.
2. The GC does not wait for threads to resume. This can cause notifications to be misinterpreted due to signal nesting if the GC is re-run before all threads wake.
This fixes a bug where runtime.alloc would not clear the padding from rounding the allocation up to a multiple of the block size.
The GC still has to scan this, and this could result in permanent false positives (and therefore memory leaks).
It also handles overflow in the size calculation.
Flash operations must go through the SoftDevice if it is enabled,
otherwise they will crash the chip. (And even then the SoftDevice
doesn't guarantee they'll succeed, depending on advertisement frequency
etc). But this patch makes sure to call the appropriate APIs so that
flash is usable while using Bluetooth.
Ideally we'd use something like
https://github.com/tinygo-org/tinygo/pull/5016 but that's a bit more
involved. As a quick improvement, call gosched() instead.
Example where I use this: a custom WS2812 driver that uses SPI to
transfer the data. It's useful to be able to switch back to the main
goroutine during the transfer to render the next LED update.
Writing the pointer of a buffer to memory-mapped I/O will normally cause
it to escape, which forces the compiler to heap-allocate the buffer. But
we do know how long the value stays alive, so we can tell the compiler
to keep it alive exactly until it is not needed anymore - and tell it to
not treat the pointer-to-uintptr cast as escaping.
This fixes the "gc: world already stopped" check by moving it before gcMarkReachable.
Previously the check did not work because gcMarkReachable would hang while waiting for activeTaskLock.
Previously, we created joinable threads.
As a result, all threads would be kept alive after completion so that pthread_join could wait for them.
We never call pthread_join, so threads would never get cleaned up.
Additionally, the thread creation error check was performed after waiting for the thread to start.
If pthread_create failed, the caller would get stuck waiting for a thread that was never created.
The compiler needs to know whether a defer is in a loop to determine whether to allocate stack or heap memory.
Previously, this performed a DFS of the CFG every time a defer was found.
This resulted in time complexity jointly proportional to the number of defers and the number of blocks in the function.
Now, the compiler will instead use Tarjan's strongly connected components algorithm to find cycles in linear time.
The search is performed lazily, so this has minimal performance impact on functions without defers.
In order to implement Tarjan's SCC algorithm, additional state needed to be attached to the blocks.
I chose to merge all of the per-block state into a single slice to simplify memory management.
See: https://github.com/tinygo-org/tinygo/issues/4959
This makes sure to emit an error instead of generating inline assembly,
which leads to hard-to-debug linker errors. Instead, it will now produce
errors like the following:
# golang.org/x/sys/unix
../../../../pkg/mod/golang.org/x/sys@v0.0.0-20220722155257-8c9f86f7a55f/unix/affinity_linux.go:20:23: system calls are not supported: target emulates a linux/arm system on wasm32
../../../../pkg/mod/golang.org/x/sys@v0.0.0-20220722155257-8c9f86f7a55f/unix/fcntl.go:17:29: system calls are not supported: target emulates a linux/arm system on wasm32
../../../../pkg/mod/golang.org/x/sys@v0.0.0-20220722155257-8c9f86f7a55f/unix/fcntl.go:32:24: system calls are not supported: target emulates a linux/arm system on wasm32
These chips have a larger upper limit for the DMA transfer than the
nrf52832. For best performance, we should be splitting the transfer in
as large blocks as possible on the given hardware.
- Do not use make([]byte, ...) to allocate, instead call the allocator
directly with a nil (undefined) layout. This makes sure the precise
GC will scan the contents of the allocation, since C could very well
put pointers in there.
- Simplify the map to use the pointer as the key and the size as the
value, instead of storing the slices directly in the map.
Long sleep durations weren't implemented correctly, by simply casting to
a smaller number of bits. What we want is a saturating downcast, which
is what I've used here instead.
This should fix https://github.com/tinygo-org/tinygo/issues/3356.
* Create "pico2-ice" target board
This board has an rp2350b chip on it, as well as a Lattice Semiconductor
iCE40UP5K FPGA. More details of this open hardware board here:
https://pico2-ice.tinyvision.ai/
Tested on a pico2-ice board with:
~/go/bin/tinygo flash -target=pico2-ice src/examples/blinky1/blinky1.go
which blinks the GREEN LED (connected to GPIO0) on this board.
Signed-off-by: Tinkerer <tinkerer@zappem.net>
* More silkscreen labels for pico2-ice board
Reading the schematic and the rev2 board viewer:
https://raw.githubusercontent.com/tinyvision-ai-inc/pico2-ice/refs/heads/main/Board/Rev2/bom/ibom.html
concluded that the RP2350B GPIO pins are not labeled with these
pin numbers in the silkscreen. Instead, the silkscreen refers to
uses of the GPIOs or the ICE numbered pins. RP2340B devoted pins
map to the A1..4 B1..4 pins on the RP PMOD connector. Also
silkscreen "~0".."~6" are labeled as pins N0..N6.
Signed-off-by: Tinkerer <tinkerer@zappem.net>
* Added a smoketest for pico2-ice board and tidied up GPIO defs
Addresses review comments from aykevl.
Signed-off-by: Tinkerer <tinkerer@zappem.net>
Strings are readonly, but the compiler doesn't always know this. Marking
them as readonly in the frontend allows the compiler to optimize based
on this knowledge.
This provides some small code size benefits. I didn't measure running
speed.
Incorrect factors are calculated for baudrates which are a bit
larger than integer multiples of 4194304.
For example for baudrates of 8_400_000 or 58_800_000.
Fixed the same way in newer versions of the RPI SDK.
By disabling some configuration options (and updating the library), the
library becomes a lot smaller. With `-no-debug`, binaries become ~18kB
smaller on Linux, ~4kB smaller on MacOS, ~9kB smaller on Windows, and
~12kB smaller on WebAssembly.
This function is called when a hard fault occurs. Hard faults happen
when something really bad happens - like writing to unwritable memory or
an unaligned memory access on Cortex-M0. It is not generally possible to
recover from these.
This commit optimizes the code size overhead of hard fault handling:
* It removes the stack overflow checking code.
This may seem like a bad thing, but the only thing this could check
were stack overflows outside goroutines. In practice, this could
only really happen on a stack overflow in the scheduler (unlikely),
or in interrupt code (possible, but interrupts are small so still
unlikely). Most stack overflows happen in regular goroutines, and
weren't caught in the HardFault.
* It makes the panic message similar to a regular panic. This has two
advantages:
* It reduces code size, because the string can be reused between
the HardFault handler and the runtime panic function.
* Using the same pattern automatically makes `-monitor` print the
source address for the hard fault. Not a big benefit as we could
trivially add any other pattern but a nice benefit nonetheless.
Result:
$ tinygo flash -target=microbit -size=short -programmer=openocd -monitor examples/serial
code data bss | flash ram
3036 8 2256 | 3044 2264
[...snip]
Connected to /dev/ttyACM0. Press Ctrl-C to exit.
panic: runtime error at 0x00000344: HardFault with sp=0x200007d0
[tinygo: panic at /home/ayke/src/tinygo/tinygo/src/internal/task/task_stack_cortexm.go:48:4]
(This is with https://github.com/tinygo-org/tinygo/pull/3680 not yet
fixed and some local changes to configure the UART so I can actually see
the panic).
For atsamd21/nrf51 chips this results in a binary size reduction of
around 100 bytes. For other Cortex-M chips it's around 24 bytes but I
hope to change this in the future because a lot of the fault decoding in
runtime_cortexm_hardfault_debug.go should IMHO be done by the TinyGo
monitor instead (I estimate that this would save around 800 bytes on
these chips).
When debugging is enabled for interp, print better errors in a specific
case.
Before:
!! revert because of error: interp: unsupported instruction (to be emitted at runtime)
After:
!! revert because of error: /usr/local/go1.24.0/src/regexp/syntax/parse.go:927:27: interp: unsupported instruction (to be emitted at runtime)
So this adds error location information, which can be quite useful.
the test-macos-homebrew job, and it conflicts with the actual build
that we want, which is macOS 14 for backwards-compatibility.
Signed-off-by: deadprogram <ron@hybridgroup.com>
GitHub org. The git server for git.musl-libc.org is having troubles,
and it also seems like a safer bet to have our own mirror just in case.
Signed-off-by: deadprogram <ron@hybridgroup.com>
- do not install cmake, instead use the version already installed
- add macOS 15 to the CI builds
- update a could of GH actions to latest release
- update Go version being use to Go 1.25.1
Signed-off-by: deadprogram <ron@hybridgroup.com>
This switches the Espressif fork from LLVM 19 to LLVM 20, so we can use
the improvements made between those LLVM versions. It also better aligns
with the system-LLVM build method, which currently also defaults to LLVM
20.
Note that this disables the machine outliner for RISC-V. It appears
there's a bug in there somewhere, with the machine outliner enabled the
crypto/elliptic package tests fail with -target=riscv-qemu.
This should ideally be investigated and reported upstream.
The revive command seems to have had a syntax error in the file
input glob. It appears to have been broken in a way that did not
result in a return code being set. This change uses 'find' to
build the input to the linter.
Note that it is expected to fail the CI script, because it is
uncovering some existing lint issues that were not being caught.
With these flags, the TinyGo binary gets 18.8MB (11.6%) smaller. That
seems like a quite useful win for such a small change!
This is only for Linux for now. MacOS and Windows can be tested later,
the flags for those probably need to be modified.
Originally inspired by:
https://discourse.llvm.org/t/state-of-the-art-for-reducing-executable-size-with-heavily-optimized-program/87952/18
There are some other flags like -Wl,--pack-dyn-relocs=relr that did not
shrink binary size in my testing, so I've left them out.
This also switches the linker to prefer mold or lld over the default
linker, since the system linker is usually ld.bfd which is very slow.
(Also, for some reason mold produces smaller binaries than lld).
In some cases, e.g nothing connected on the bus, repeated resume-stop sequences can lead to the bus never reaching the stop state, hanging Tx.
This change ensures the resume-stop sequence is submitted once on error. It also moves the error code read to before the sequence to ensure it's valid.
Fixes: #4998
Writing to the UART takes time and that may not be a good idea inside an
interrupt, but it is essential for debugging sometimes (especially since
USB-CDC typically doesn't work inside an interrupt).
This fixes UART support in interrupts for the RP2040 at least. You can
test it with `-serial=uart` and connecting a USB-UART adapter to the
right pins.
Account for the sleep queue base time in the computation of the wakeup
time.
Tested with the following program on pico2.
func main() {
go func() {
for i := range 60 {
const delay = 20 * time.Millisecond
before := time.Now()
time.Sleep(delay)
if d := time.Since(before); true || d < delay {
log.Println(i, "actual", d, "delay", delay)
}
}
}()
time.Sleep(500 * time.Millisecond)
log.Println("******** done sleeping ********")
select {}
}
Without this change, the program would print lines such as:
17 actual 15.494ms delay 20ms
18 actual 15.49ms delay 20ms
19 actual 15.585ms delay 20ms
20 actual 15.493ms delay 20ms
21 actual 15.494ms delay 20ms
22 actual 15.487ms delay 20ms
23 actual 15.498ms delay 20ms
******** done sleeping ********
24 actual 15.548ms delay 20ms
25 actual 20.011ms delay 20ms
26 actual 20.01ms delay 20ms
27 actual 20.011ms delay 20ms
28 actual 20.015ms delay 20ms
Note that while more than one sleeping goroutine is in the timer queue,
the sleep duration is 5ms short.
* Add -o flag support to flash command. Fixes#4937
* Remove empty outpath check in validateOutputFormat
---------
Co-authored-by: rdon <you@example.com>
This gets the path package tests to pass, so we can move ahead with Go
1.25. It should be implemented in the future at some point (that, or
we'll use the upstream testing package instead).
Since we switched to OS threads for goroutines, the
testdata/goroutines.go test has been flaky. Not surprising, if threads
need to be scheduled within 1ms on a busy CI system. This PR makes the
test a bit less flaky (hopefully) by increasing the time to 100ms.
Found this bug while trying to use the upstream testing package instead
of our own. The io/fs package wasn't passing, because the test was run
in a separate goroutine (and therefore a separate thread, with its own
stack) instead of all in the same thread with our own stack
creation/switching implementation.
This makes sure system calls like read don't return EINTR but instead
restart the call on an interrupt. This is by far the more sensible
option, the default POSIX behavior of returning EINTR is extremely
error-prone.
Found this bug while trying to use the upstream testing package instead
of our own.
Running tests against an installed TinyGoRoot fails with:
```
=== CONT TestBuild/WebAssembly/gc.go-boehm
main_test.go:441: wasi-libc: did not find any files for pattern builder.filePattern{glob:"dlmalloc/src/dlmalloc.c", exclude:[]string(nil)}
```
because these files are not there, and are required with Boehm GC.
The previous versions calculated at init() prevented `interp` from running
in many cases, increasing compile times due to the increased need to revert
the partially interpreted results and also increasing binary runtime because
fewer optimizations had happened during interp.
Similar to PWM, I2C can only be used on some pins. To automatically
generate this information per board, we need to add extra comments that
can then be interpreted by doc-gen for the tinygo.org website.
This fixes/improves a few issues with I2C support:
* Validate I2C pins, so only pins that are supported by the hardware
can be used (similar to how it's done with PWM).
* Add address to Tx API (without it, the simulator can't really
simulate I2C).
* Add frequency when configuring. Not currently used, but might be
useful in the future and adding it now avoids possibly breaking
changes.
This is a breaking change, but since the simulator doesn't support I2C
yet that seems fine to me. (It does in my local changes, but those need
to be cleaned up before I can push them).
* Use diskutil on macOS to extract volume name and path for FAT mounts
* Improve FAT mount detection on macOS by parsing diskutil output into a map
* Simplify conditional check
This should avoid a deadlock when trying to print inside an interrupt,
if the interrupted code is also printing (and therefore has the print
lock taken).
The GC shouldn't try to interrupt other cores before they are started.
For example, it would be possible for the GC to run in a package
initializer (which is currently run on a single core). That would
suggest questionable program design, but it is something that should
work. So this commit makes sure the GC only tries to scan the stack of
other cores when those other cores have in fact started.
This adds support for `-gc=boehm` on `-target=wasip1` and `-target=wasm`
(in a browser or NodeJS). Notably it does *not* add Boehm GC support for
`-target=wasip2`, since that target doesn't have a real libc.
This commit adds support for a scheduler that runs a scheduler on all
available cores. It is meant to be used on baremetal systems with a
fixed number of cores, such as the RP2040.
The initial implementation adds support for multicore scheduling to the
riscv-qemu target as a convenient testing target. This means that this
new multicore scheduler is tested in CI, including a bunch of standard
library tests (`make tinygo-test-baremetal`). This should ensure the new
scheduler is reasonably well tested before trying to use it on
harder-to-debug targets like the RP2040.
The system stack is only needed when we're not on it. So we can directly
call task.SystemStack() without problems.
This also saves a tiny bit of binary size.
This type can be used to jump back to a previous position in a program
from inside an interrupt. This is useful for baremetal systems that
implement wfi but not wfe, and therefore have no easy (race-free) way to
wait until a flag gets changed inside an interrupt. This is an issue on
RISC-V, where this is racy (the interrupt might happen after the check
but before the wfi instruction):
configureInterrupt()
for flag.Load() != 0 {
riscv.Asm("wfi")
}
Accessing the same variable from multiple goroutines is unsafe, and will
fail with parallelism. A lightweight way to avoid issues is by using
atomic variables.
* machine: add support for BTT SKR Pico
Adds support for the BigTreeTech SKR Pico 3D-printer mainboard.
This board uses the RP2040.
* Fix build tag
* Add I2C defaults
* Run UART test instead of blinky1
* Use NoPin for I2C and SPI on BTT SKR Pico
* Cleanup comments
* Don't use ADC pin names
* Fix DMA to SPI transfers on RP2350
DMA DREQ line numbers for "flow control" between SPI bus and DMA channels on RP2350 differ from RP2040.
Tested with st7789 driver for Pico-1.14-LCD from Waveshare on Pico 2W. Without this fix func st7789.tx() blocks indefinitely while attempting to use DMA to SPI transfers.
* Add definitions for DMA DREQ "handshake" lines
Specific for RP2350, missing in generated src/device/rp/rp2350.go
* Add definitions for DMA DREQ "handshake" lines
Specific for RP2040, missing in generated src/device/rp/rp2040.go
* Complete table
* Complete table
* Remove redundant DMA_ prefix
* Correct name of Datasheet
* Correct name of Datasheet
* Refactor
Move global definitions to device/rp/
* Refactor
* Refactor
* Refacture
* Refacture
* Fix comments
* go fmt
* rename new non-generated files
This commit changes signal handling in a few ways:
* It stubs signals for all wasm targets (not just wasi) and baremetal,
since none of those have traditional POSIX signals. And moves the
code for that into a single file, instead of duplicating it.
* It removes the stub for signal_ignored since the value `false` might
be wrong in some cases and it doesn't usually seem to be called (it
is not called in tsgo). Should be trivial to re-add if it is shown
to be needed.
* It adds a stub for `os/signal.signalWaitUntilIdle` which _is_ called
by tsgo.
With `-opt=2`, WriteByte gets inlined everywhere a println statement
exists. This blows up binary size for very little gain. In my case, the
binary size roughly doubled. Instead, don't inline it so that the binary
size remains somewhat reasonable. This might slow down WriteByte a tiny
bit, but likely not by any significant amount.
Errors are part of API, and the exported rp2 errors seemed arbitrary.
For example, the very particular ErrRP2040I2CDisable was exported, but
errI2CWriteTimeout (which is defined on all platforms) is not.
While here, remove "RP2040" from an error name and make the messages
consistent and idiomatic.
The `gosched` call introduce arbitrary long delays in general, and in
TinyGo particular because the goroutine scheduler is cooperative and
doesn't preempt busy (e.g. compute-heavy) goroutines.
Before this change, the timeout logic would read, simplified:
deadline := now() + timeout
startTX()
for !txDone() {
if now() > deadline { return timeoutError }
gosched() // (1)
}
startRx() // (2)
for !rxDone() {
// (3)
if now() > deadline { return timeoutError }
gosched()
}
What could happen in a busy system is:
- The gosched marked (1) would push now() to be > than deadline.
- startRx is called (2), but the call to rxDone immediately after would
report it not yet done.
- The check marked (3) would fail, even though only a miniscule amount
of time has passed between startRx and the check.
This change ensures that the timeout clock discounts time spent in
`gosched`. The logic now reads, around every call to `gosched`:
deadline := now() + timeout
startTX()
for !txDone() {
if now() > deadline { return timeoutError }
before := now()
gosched()
deadline += now() - before
}
I tested this change by simulating a busy goroutine:
go func() {
for {
// Busy.
before := time.Now()
for time.Since(before) < 100*time.Millisecond {
}
// Sleep.
time.Sleep(100 * time.Millisecond)
}
}()
and testing that I2C transfers would no longer time out.
Fixes#4884 by upgrading the ssa package.
The fix is in v0.26.0, which also bumps the minimum Go to 1.22. The
latest x/tools module still depending on 1.22 is v0.30.0.
I only discovered this issue after a while. All the baremetal PWM
implementations use pointers to a PWM instance, instead of the PWM
instance itself. For consistency (and because it's a better idea in
general), the simulated PWMs need to work the same.
This is useful for the TinyGo Playground: with this change it can show
the error span like in an IDE, instead of just the (single) error
position. Errors become a bit more readable as a result.
This changes the -json flag for build/run/flash/gdb etc commands to not
print `compileopts.Config` but instead match upstream Go and print build
errors in a structured way.
For more details, see the proposal:
https://github.com/golang/go/issues/62067
This adds support for the 'go install' case, where we link against the
system LLVM (Homebrew, Linux distro version, etc). This does not have an
effect on `make`, which still uses LLVM 19 for now.
The main reason for doing this is because newer distros like Fedora 42
and Homebrew have switched to LLVM 20, so switching to this newer
version helps those people. (I'm one of those people, I'm on Fedora 42).
There are two small changes for WebAssembly included:
* nontrapping-fptoint was enabled in the wasmbuiltin library used for
wasm-unknown. This matches wasm-unknown which already had this
enabled, so it doesn't add any new instructions.
* bulk-memory was enabled in wasi-libc. This was previously enabled in
all the other WebAssembly targets (wasm, wasip1, wasip2) so this
does't add any new instructions. I think it was also enabled in the
wasi-libc build before
https://github.com/tinygo-org/tinygo/pull/4820 but I don't know for
sure.
Without this change, a pending interrupt would spuriously trigger
immediately after enabling. This happens if an interrupt is triggered
during flashing (e.g. by DMA), which survives the subsequent reset.
This behaviour matches e.g. `machine.irqSet` in machine_rp2_rp2350.go.
See 7f970a45, whose symptoms were likely caused by spurious interrupts.
For the threads scheduler, it makes sense to have NumCPU available.
For all other schedulers, the number of available CPUs is practically
limited to one by the scheduler (even though the system might have more
CPUs).
A race condition was possible because the 'acquire' goroutine might not
have started in 4 milliseconds which changed the ordering of the test.
This patch fixes it by making sure the goroutine has started (and locked
the mutex) before continuing with the test.
This variable is only necessary on the cooperative and none scheduler.
It is not used on the threads scheduler.
The reason for moving is that the upcoming multicore baremetal scheduler
also needs mainExited but of a different type: an atomic variable
instead of a plain boolean.
This is more descriptive: the call is to exit a task, not to pause it.
This also makes it more obvious that there's an optimization
opportunity: to free the stack explicitly after the goroutine returns
(or to keep it as a cache for the next stack allocation).
This is not a scheduler in the runtime, instead every goroutine is
mapped to a single OS thread - meaning 1:1 scheduling.
While this may not perform well (or at all) for large numbers of
threads, it greatly simplifies many things in the runtime. For example,
blocking syscalls can be called directly instead of having to use epoll
or similar. Also, we don't need to do anything special to call C code -
the default stack is all we need.
Somewhat surprisingly, this results in smaller code than the old code
with the cooperative (tasks) scheduler. Probably because the new RWMutex
is also simpler.
Using a global lock may be slow, but it is certainly simple and safe.
If this global lock becomes a bottleneck, we can of course look into
making the GC truly support multithreading.
This may be a bit of a weird place to put the library path, but
otherwise it's difficult to get the pathname for the Config.CFlags
function. So I've put it here.
This should help to avoid stale library caches. The idea is to increment
it every time something changes to a library that means it needs to be
recompiled. It's a manual process.
Instead of just incrementing the timestamp, this causes the system to
actually sleep when calling time.Sleep. The direct effect is that this
works as expected:
$ tinygo run -target=riscv-qemu examples/serial
hello world!
hello world!
hello world!
[..etc]
This commit also adds a bare bones handler for exceptions (such as
invalid memory writes), since we're adding an interrupt handler anyway.
While this patch doesn't add that much functionality, having interrupt
support is going to be needed for multicore support on riscv-qemu. My
plan is to first add this support to riscv-qemu (based on the earlier
work I did for the RP2040 and demoed at FOSDEM 2025) and once the basics
are in place and fully tested we can extend this support to the RP2040.
Writing for QEMU first makes it much easier to debug any issues that
will come up.
Instead of relying on a build when TinyGo is being built, do it like all
other libraries when it is needed.
This brings a few benefits:
* No more running `make wasi-libc` on the command line as a special
case for WebAssembly. The generic `git submodule update --init` step
is now enough for wasi-libc.
* It becomes much easier to customize the build per system. For
example: include/exclude malloc as needed, disable/enable bulk
memory operations per target, etc.
Header files are built immediately, not in a separate job, so no
dependency is needed here. All we need is a flag whether to add flags
for that given libc.
This *should* not be needed. Using the right `go` binary should do the
job. It might break if users have explicitly set GOROOT (in which case
they probably need to fix that).
This avoids a subprocess call inside `make`.
Instead of calling it 5 times, call it only once on Unix-like systems
(Linux, MacOS) and avoid it entirely on Windows.
The main benefit of this is that it avoids a ton of overhead doing
subprocess calls, which are very slow on Windows (~0.2s on my system).
Most make commands don't need to check for the NodeJS version, so only
do it when needed.
This avoids the following error on Windows for example when NodeJS is
not installed:
/usr/bin/sh: line 1: node: command not found
which: no node in (/c/Users/Ayke/bin:/clangarm64/bin:/usr/local/bin:/usr/bin:/usr/bin:/mingw64/bin:/usr/bin:/c/Users/Ayke/bin:/c/Users/Ayke/.vscode-server/cli/servers/Stable-e54c774e0add60467559eb0d1e229c6452cf8447/server/bin/remote-cli:/c/WINDOWS/system32:/c/WINDOWS:/c/WINDOWS/System32/Wbem:/c/WINDOWS/System32/WindowsPowerShell/v1.0:/c/WINDOWS/System32/OpenSSH:/cmd:/c/Users/Ayke/scoop/apps/mingw-mstorsjo-llvm-ucrt/current/bin:/c/Users/Ayke/scoop/apps/python/current/Scripts:/c/Users/Ayke/scoop/apps/python/current:/c/Users/Ayke/go/bin:/c/Users/Ayke/scoop/shims:/c/Users/Ayke/AppData/Local/Microsoft/WindowsApps:/usr/bin/vendor_perl:/usr/bin/core_perl)
Also, some users have been confused by the error message even though in
most cases it is harmless and can be ignored.
This directive caused the code to be put in a non-executable area on
Windows which caused a segmentation fault. This patch fixes the issue by
removing `.section` directives, fixing windows/arm64 support.
A few small changes were needed to make this work. In particular, I
found a critical bug (see the previous commit) that needed to be fixed
to make this work on Windows.
I'm surprised this worked as long as it did, since it looks like the
goroutine stack did not get scanned. Or maybe the RCX register contained
the stack pointer by accident. In any case, it now uses the correct
register (RCX instead of RDI on Windows) for passing the stack pointer
as the first parameter.
Support for GOARCH=wasm was only available for `tinygo build`, not for
any of the other subcommands (`test`, `run`, etc). With this PR, these
subcommands should work again for supported values of GOOS.
This fixes the following error for example:
$ make tinygo-test-wasip1
GOOS=wasip1 GOARCH=wasm /home/ayke/bin/tinygo test cmp compress/lzw compress/zlib [...etc]
cannot resolve packages: GOARCH=wasm but GOOS is unset. Please set GOOS to wasm, wasip1, or wasip2.
make: *** [GNUmakefile:489: tinygo-test-wasip1] Error 1
This adds support for the well-known Boehm GC. It's significantly faster
than our own naive GC and could be used as an alternative on bigger
systems.
In the future, this GC might also be supported on WebAssembly with some
extra work. Right now it's Linux only (though Windows/MacOS shouldn't be
too difficult to add).
For example, with -gc=none and -gc=leaking, no heap needs to be
allocated when initializing the runtime. And some GCs (like -gc=custom)
are responsible for allocating the heap themselves.
Strings that are set via the command line (as in -ldflags="-X ...") are
now created in a separate module to avoid type renaming issues. LLVM
sometimes renames types or merges types that are structurally the same,
and putting these strings in a separate module avoids this issue (and
lets llvm.LinkModules deal with the difference).
This fixes https://github.com/tinygo-org/tinygo/issues/4810.
* machine/rp2350: add flash support for rp2350
* combine duplicate files
* clean things up and group by source file
* add stubbed out xip cache clean func if needed in the future
* update flash_enable_xip_via_boot2
* remove unused macros and fix inconsistent formatting
* make flash size configurable like rp2040
* add missing flash size configs
* retain big Go CGo compatibility per #4103
* clarify CS0_SIZE source and remove single-use typedef
Don't block forever if there's nothing to receive. On the other hand,
process the entire FIFO, not just a single byte.
This fixes an issue where the rp2350 would hang after programming through
openocd, where the UART0 interrupt would be spuriously pending.
Run a range of tests in CI, to make sure browser wasm and baremetal
don't regress too badly.
I have intentionally filtered out tests, so that newly added tests to
TEST_PACKAGES_FAST will be added here as well (and can be excluded if
needed).
It looks like we didn't have any tests for wasm. Having one tests is not
much, but it proves that the infrastructure works and it actually
verifies a fix to https://github.com/tinygo-org/tinygo/issues/4777.
We should add more packages to this list in the future.
Apparently this package imports `runtime.getRandomData` from the gojs
module. This is not yet implemented, but simply allowing this package to
do such imports gets crypto/sha256 to compile.
Right now it doesn't compile, with errors like the following:
# machine
/app/tinygo/src/machine/board_pico.go:7:13: undefined: GPIO0
/app/tinygo/src/machine/board_pico.go:8:13: undefined: GPIO1
/app/tinygo/src/machine/board_pico.go:9:13: undefined: GPIO2
/app/tinygo/src/machine/board_pico.go:10:13: undefined: GPIO3
[...etc...]
This patch should fix that.
* machine: add support for core voltage adjustments to rp2040
In preparation for bumping the core frequency of the rp2040, this
change implements the required core voltage adjustment logic.
* machine: bump rp2040 to 200MHz
Follow-up to #4728 which implemented the algorithm for finding the
dividers.
The calculation is computed at compile time by interp, as verified by
building example/blinky1 for -target pico.
The upstream one assumes it's running on a Unix system (which makes
sense), but this package is also used on baremetal. So replace it on
systems that need a replaced syscall package.
We can't use the bytes package in Go 1.24 since it would result in an
import cycle. Therefore, use bytealg.Index instead.
(I'm not sure this code is correct - just searching for a range of bytes
seems brittle. But at least this commit shouldn't change the code).
This mangles CGo identifier names to something like "_Cgo_foo" instead
of using literal identifiers like "C.foo". This works around
https://github.com/golang/go/issues/71777.
I don't like this solution, but I hope we'll find a better solution in
the future. In that case we can revert this commit.
older behavior for wasi modules to not return an exit code as if they were reactors.
See #4726 for some details on what this is intended to address.
Signed-off-by: deadprogram <ron@hybridgroup.com>
This ensures:
1. The xorshift state is initialized during interp.
2. The xorshift state gets initialized to a real random number on
hardware that supports it at runtime.
This fixes a big binary size regression from the previous commit. It's
still not perfect: most programs increase binary size by a few bytes.
But it's not nearly as bad as before.
Previously, if a function couldn't be interpreted in the interp pass, it
would just be run at runtime and the parameters would be marked as
potentially being modified. However, this is incorrect: the function
itself can also modify memory. So the function itself also needs to be
marked (recursively).
This fixes a number of regressions while upgrading to Go 1.24.
This results in a significant size regression that is mostly worked
around in the next commit.
Modify the ID used for looking up the reference, based on suggestion made by @prochac
Also use console.error in the caase that the reference is not found, since it is now actually
known to be an error.
`tinygo build -o /path/to/out .` expected `/path/to/out` to already exist, which is different behaviour to `go build`. This change ensures tinygo creates any directories needed to be able to build to the specified output dir.
The mutex implementation needs a different implementation once support
for threading lands. This implementation just moves code to the
internal/task package to centralize these algorithms.
The latest golang.org/x/net websocket package v0.33.0 needs
Dialer.DailContext in crypto/tls. This PR adds it.
Apps using golang.org/x/net are encouraged to use v0.33.0 to address:
CVE-2024-45338: Non-linear parsing of case-insensitive content in golang.org/x/net/html
CVE-2023-45288: net/http, x/net/http2: close connections when receiving too many headers
Tested with examples/net/websocket/dial.
If `ld.lld` is a version-specific binary (e.g., `ld.lld-18`), then its
error messages include the version. The parsing previously incorrectly
assumed it would be unversioned.
Show which files cause a binary size increase. This makes it easier to
see where the size is going: for example, this makes it easy to see how
much the GC contributes to code size compared to other runtime parts.
This is not a big change over the existing size report with -size=full,
but it is a bit more readable.
More information will be added in subsequent commits.
* initial implementation for Tillitis TKey device
* add UART implementation for TKey
* add Pin interface implementation for TKey touch sensor
* add RNG interface implementation for TKey
* add helpful machine package functions to return identifiers such as name and version for TKey
* use built-in timer for sleep timing on TKey
* modify UART implementation for TKey to implement Serialer interface
* implement BLAKE2s ROM function call for TKey device
* handle abort by triggering TKey device fault using illegal instruction to halt CPU
* simplify TKey implementation by inheriting from existing riscv32 target
* return error for trying to configure invalid baudrates on UART
* add tkey to builder test
* be very specific for features passed to LLVM for specific config in use for TKey
* handle feedback items from TKey device code review
Signed-off-by: deadprogram <ron@hybridgroup.com>
This actually simplifies the code and avoids a heap allocation in the
call to Wait. Instead, it uses the Data field of the task to store
information on whether a task was signalled early.
See the code comments for details. But in short, this implements a futex
for the cooperative scheduler (that is single threaded and
non-reentrant). Similar implementations can be made for basically every
other operating system, and even WebAssembly with the threading
(actually: atomics) proposal.
Using this basic futex implementation means we can use the same
implementation for synchronisation primitives on cooperative and
multicore systems.
For more information on futex across operating systems:
https://outerproduct.net/futex-dictionary.html
This adds some non-atomic types that have the same interface as the ones
in sync/atomic.
We currently don't need these to be atomic (because the scheduler is
entirely cooperative), but once we add support for a scheduler with
multiple threads and/or preemptive scheduling we can trivially add some
type aliases under a different build tag in the future when real
atomicity is needed for threading.
I don't think this is used anywhere right now, and it would need to be
updated to work with multithreading. So instead of fixing it, I think we
can remove it.
My original intention was to have something like this that could be used
in the machine package, but since this is in the runtime package (and
the runtime package imports the machine package on baremetal) it can't
actually be used that way.
I checked the TinyGo repo and the drivers repo, and `runtime.Cond` isn't
used anywhere except in that one test.
Previously the assembler was reordering this code:
jal tinygo_scanstack
move $a0, $sp
Into this:
jal tinygo_scanstack
nop
move $a0, $sp
So it was "helpfully" inserting a branch delay slot, even though this
was already being taken care of.
Somehow this didn't break, but it does break in the WIP threading branch
(https://github.com/tinygo-org/tinygo/pull/4559) where this bug leads to
a crash.
This ensures that calls to print/println happening in different threads
are not interleaved. It's a task.PMutex, so this should only change
things when threading is used.
This matches the Go compiler, which does the same thing:
https://godbolt.org/z/na5KzE7en
The locks are not recursive, which means that we need to be careful to
not call `print` or `println` inside a runtime.print* implementation,
inside putchar (recursively), and inside signal handlers. Making them
recursive might be useful to do in the future, but it's not really
necessary.
This uses uint32 instead of uint64. The reason for this is that uint64
atomic operations aren't universally available (especially on 32-bit
architectures). We could also use uintptr, but that seems needlessly
complicated: it's unlikely real-world programs will use more than a
billion select states (2^30).
This uses the task.PMutex parallel-only-mutex type to make the leaking
GC parallelism safe. The task.PMutex type is currently a no-op but will
become a real mutex once we add true parallelism.
Use the alignment from the align attribute of the runtime.alloc call.
This is going to be a more accurate alignment, and is typically smaller
than the default.
This requires an API introduced in MacOS 11. I think that's fine, since
the version before that (MacOS 10.15) is EOL since 2022. Though if
needed, we could certainly work around it by using an older and slightly
less nice API.
This really is the only sane way to register a signal. If this flag is
not set, many syscalls will return EINTR (and not complete their
operation) which will be a massive source of hard-to-debug bugs.
This rewrite simplifies the channel implementation considerably, with
34% less LOC. Perhaps the most important change is the removal of the
channel state, which made sense when we had only send and receive
operations but only makes things more compliated when multiple select
operations can be pending on a single channel.
I did this rewrite originally to make it possible to make channels
parallelism-safe. The current implementation is not parallelism-safe,
but it will be easy to make it so (the main additions will be a channel
lock, a global select lock, and an atomic compare-and-swap in
chanQueue.pop).
This only works on declarations, not definitions. This is intentional:
it follows the upstream Go implemetation.
However, we might want to loosen this requirement at some point: TinyGo
sometimes stores pointers in memory mapped I/O knowing they won't
actually escape, but the compiler doesn't know about this.
This moves all scheduler code into a separate file that is only compiled
when there's a scheduler in use (the tasks or asyncify scheduler, which
are both cooperative). The main goal of this change is to make it easier
to add a new "scheduler" based on OS threads.
It also fixes a few subtle issues with `-gc=none`:
- Gosched() panicked. This is now fixed to just return immediately
(the only logical thing to do when there's only one goroutine).
- Timers aren't supported without a scheduler, but the relevant code
was still present and would happily add a timer to the queue. It
just never ran. So now it exits with a runtime error, similar to any
blocking operation.
Making this work on all targets was interesting but there's now a test
in place to make sure this works on all targets that have the CGo test
enabled (which is almost all targets).
This fixes a bug where output would not actually be written to stdout
before exiting the process, leading to a flaky Windows CI. Exiting using
`exit(0)` appears to fix this.
For some background, see: https://github.com/tinygo-org/tinygo/pull/4589
* Move environment functions to their own files.
* Rewrite the WASIp2 version of environment variables to be much
simpler (don't go through C functions).
These should make sure basic functionality is still working.
Using the `-short` flag to avoid taking too long to run all tests (and
to install all the necessary emulators), and because some targets might
not work in older Go/LLVM versions (such as WASI).
This does _not_ run tests and checks against expected IR, because LLVM
IR changes a lot across versions.
It looks like this breaks on Windows:
https://github.com/tinygo-org/tinygo/issues/4557
I haven't confirmed this is indeed the problem, but it would make sense.
And passing through the temporary directory seems like a good idea
regardless, there's not much that could break due to that.
Use a single package for certain constants that must be the same between
the compiler and the runtime.
While just using the same values in both places works, this is much more
obvious and harder to mess up. It also avoids the need for comments
pointing to the other location the constant is defined. And having it in
code makes it possible for IDEs to analyze the source.
In the future, more such constants and maybe algorithms can be added.
This is similar to https://github.com/tinygo-org/tinygo/pull/3899, but
smaller and hopefully just as efficient.
Thanks to @HattoriHanzo031 for starting this work, benchmarking, and for
improving the performance of the code even further.
The git hash that's part of `tinygo version` is now read from the binary
itself instead of embedding it with a `-ldflags` flag. This means it is
also present when building TinyGo using `go build` or `go install`.
* internal/wasm-tools, internal/cm: update wasm-tools-go to v0.3.1 and regenerate bindings
* syscall: use new (cm.Result).Result() method instead of OK() and Err()
The alignment wasn't set, so defaulted to 4 (for a 32-bit int). LLVM saw
this, and therefore assumed that a ptrtoint of the pointer would have
had the lowest bits unset. That's an entirely valid optimization, except
that we are using these globals for arbitrary values (and aren't
actually using these globals).
Fixed by setting alignment to 1. It works, though long-term we should
maybe find a different solution for this.
Volatile loads/stors are only useful for communication with interrupts
or for memory-mapped I/O. They do not provide any sort of safety for
sync.Mutex, while making it *appear* as if it is more safe.
* `sync.Mutex` cannot be used safely inside interrupts, because any
blocking calls (including `Lock`) will cause a runtime panic.
* For multithreading, `volatile` is also the wrong choice. Atomic
operations should be used instead, and the current code would not
work for multithreaded programs anyway.
Every time we overflow the stack, we have to do a full rescan of the heap. Making this larger
means fewer overflows and thus fewer secondary+ heap scans.
With a few small modifications, all the problems with `-gc=precise` in
WebAssembly seem to have been fixed.
I didn't do any performance measurements, but this is supposed to
improve GC performance.
compiler: align with current wasm types proposal
https://github.com/golang/go/issues/66984
- Remove int and uint as allowed types in params, results, pointers, or struct fields
- Only allow small integers in pointers, arrays, or struct fields
- enforce structs.HostLayout usage per wasm types proposal
https://github.com/golang/go/issues/66984
- require go1.23 for structs.HostLayout
- use an interface to check if GoVersion() exists
This permits TinyGo to compile with Go 1.21.
- use goenv.Compare instead of WantGoVersion
- testdata/wasmexport: use int32 instead of int
- compiler/testdata: add structs.HostLayout
- compiler/testdata: improve tests for structs.HostLayout
goenv: parse patch version, add func Compare to compare two Go version strings
* Parse tests
* add Compare function to compare two Go version strings
* goenv, builder: parse patch version in Go version string
While there are some browser tests, Node.js is just a lot better for
testing this kind of stuff because it's much faster and we don't need a
browser for this.
This issue was originally reported here:
https://github.com/NixOS/nixpkgs/pull/341170#issuecomment-2359237471
The fix here isn't a great fix, it turns the error message from this:
# runtime/interrupt
into this:
# runtime/interrupt
package requires newer Go version go1.23
...so not great, because it doesn't show the real error message (which
is that TinyGo wasn't compiled with the right Go version). But at least
it gives a hint in the right direction.
It's difficult to test for this specific case, so I've left out testing
in this case (boo!)
This often doesn't work because there might not be a current task to
push the defer frame to. It will instead show an unhelpful nil pointer
dereference panic.
We could make this work with a separate defer stack for interrupts (as
if they were newly started goroutines) but that is difficult with
multiple interrupts happening at the same time (we shouldn't jump to a
previous interrupt in `panic()`!). So instead, disable defer altogether
in interrupts and adjust panic/recover accordingly.
This is a problem on Guix, which sets C_INCLUDE_PATH that is not
affected by `-nostdlibinc`. And therefore it affects the build in
unintended ways.
Removing all environmental variables fixes this issue, and perhaps also
other issues caused by Clang being affected by environment variables.
This adds support for //go:wasmexport with `-target=wasm` (in the
browser). This follows the //go:wasmexport proposal, meaning that
blocking functions are not allowed.
Both `-buildmode=default` and `-buildmode=c-shared` are supported. The
latter allows calling exported functions after `go.run()` has returned.
This should fix some CI issues we're currently having.
Thanks to @sylv-io for discovering that we need to remove LLVM to avoid
an Xtensa backend linker error.
This adds support for the `//go:wasmexport` pragma as proposed here:
https://github.com/golang/go/issues/65199
It is currently implemented only for wasip1 and wasm-unknown, but it is
certainly possible to extend it to other targets like GOOS=js and
wasip2.
Don't rename them when -work is set, instead update result.Binary each
time and leave result.Executable be the linker output (as intended:
result.Executable should be the unmodified linker output).
This matches upstream Go. Example:
$ go test -ldflags='-extldflags=-foobar' os
# os.test
/usr/local/go1.23.1/pkg/tool/linux_arm64/link: running gcc failed: exit status 1
/usr/bin/gcc -s -o $WORK/b001/os.test -rdynamic /tmp/go-link-914594215/go.o /tmp/go-link-914594215/000000.o /tmp/go-link-914594215/000001.o /tmp/go-link-914594215/000002.o /tmp/go-link-914594215/000003.o /tmp/go-link-914594215/000004.o /tmp/go-link-914594215/000005.o /tmp/go-link-914594215/000006.o /tmp/go-link-914594215/000007.o /tmp/go-link-914594215/000008.o /tmp/go-link-914594215/000009.o /tmp/go-link-914594215/000010.o /tmp/go-link-914594215/000011.o /tmp/go-link-914594215/000012.o /tmp/go-link-914594215/000013.o /tmp/go-link-914594215/000014.o /tmp/go-link-914594215/000015.o /tmp/go-link-914594215/000016.o /tmp/go-link-914594215/000017.o /tmp/go-link-914594215/000018.o /tmp/go-link-914594215/000019.o /tmp/go-link-914594215/000020.o /tmp/go-link-914594215/000021.o -O2 -g -O2 -g -lresolv -O2 -g -lpthread -foobar
gcc: error: unrecognized command-line option ‘-foobar’
FAIL os [build failed]
FAIL
And TinyGo, with this patch:
$ tinygo test -ldflags='-extldflags=-foobar' os
FAIL os 0.000s
ld.lld: error: unknown argument '-foobar'
Also note that Go doesn't support the `-extldflags` directly (which was
previously the case with TinyGo):
$ go test -extldflags='-foobar' os
flag provided but not defined: -extldflags
[...]
The values were stored in the passed object as the values itself (not
expanded like is common in the calling convention), and read back after
assuming they were expanded. This often works for simple parameters
(int, pointer, etc), but not for more complex parameters. Especially
when there's padding.
Found this while working on `//go:wasmexport`.
The reflect package needs to know the endianness of the system in a few
places. Before this patch, it assumed little-endian systems. But with
GOARCH=mips we now have a big-endian system which also needs to be
supported. So this patch fixes the reflect package to work on big-endian
systems.
Also, I've updated the tests for MIPS: instead of running the
little-endian tests, I've changed it to run the big-endian tests
instead. The two are very similar except for endianness so this should
be fine. To be sure we won't accidentally break little-endian support,
I've kept a single MIPS little-endian test (the CGo test, which doesn't
yet work on big-endian systems anyway).
The interp package was assuming that all targets were little-endian. But
that's not true: we now have a big-endian target (GOARCH=mips).
This fixes the interp package to use the appropriate byte order for a
given target.
This should widen compatibility a bit, so that older CPUs can also
execute programs built by TinyGo. The performance may be lower, if
that's an issue we can look into implementing the proposal here:
https://github.com/golang/go/issues/60072
This still wouldn't make programs usable on MIPS II CPUs, I suppose we
can lower compatiblity down to that CPU if needed.
I tried setting the -cpu flag in the QEMU command line to be able to
test this, but it looks like there are no QEMU CPU models that are
mips32r1 and have a FPU. So it's difficult to test this.
This required a few compiler and runtime tricks to work, but I ran a
bunch of tests and it seems fine. (CI will of course do more exhaustive
testing).
The main benefit here is that we don't need to maintain the darwin
version of the syscall package, and reduce extra risks for bugs (because
we reuse the well-tested syscall package). For example, Go 1.23 needed a
bunch of new constants in the syscall package. That would have been
avoided if we had used the native syscall package on MacOS.
This was still at jammy (22.04), while the CI container was noble
(24.04). Somehow this didn't break, but it certainly isn't ideal to
install packages across Ubuntu versions!
This is needed for Go 1.23 support.
These functions should ideally get implemented. Until that's done, just
panic here. Apparently panicking in internal/abi.FuncPCABI0 is enough to
fix all linker errors for mime/quotedprintable.
This package can never be a full version as seen in upstream Go, because
TinyGo is very different. But it is necessary to define so that no code
can accidentally use this package (now or in the future).
It currently defines:
- NoEscape which is needed by strings.Builder since Go 1.23.
- FuncPCABI* which is needed by internal/syscall/unix on MacOS.
This version probably isn't as fast as the upstream version, but it is
good enough for now. It also doesn't free unreferenced handles like the
upstream version.
I tried implementing enough CGo support to get the native os/user
package to work. But I hit a few bugs, probably in CGo itself. Then I
realized I could just as well set the osusergo build tag to disable CGo
for this specific case.
This actually gets the os/user package to work correctly on Linux (I
confirmed it returns my name/uid/homedir etc). On other systems, it
probably just returns an error if it can't determine these kinds of
things. But that's no worse than the current behavior which just doesn't
do anything at all.
Print a message for SIGBUS, SIGSEGV, and SIGILL when they happen.
These signals are always fatal, but it's very useful to know which of
them happened.
Also, it prints the location in the binary which can then be parsed by
`tinygo run` (see https://github.com/tinygo-org/tinygo/pull/4383).
While this does add some extra binary size, it's for Linux and MacOS
(systems that typically have plenty of RAM/storage) and could be very
useful when debugging some low-level crash such as a runtime bug.
This adds softfloat support to GOARM, as proposed here:
https://github.com/golang/go/issues/61588
This is similar to https://github.com/tinygo-org/tinygo/pull/4189 but
with a few differences:
* It is based on the work to support MIPS softfloat.
* It fixes the issue that the default changed to softfloat everywhere.
This PR defaults to softfloat on ARMv5 and hardfloat on ARMv6 and
ARMv7.
* It also compiles the C libraries (compiler-rt, musl) using the same
soft/hard floating point support. This is important because
otherwise a GOARM=7,softfloat binary could still contain hardware
floating point instructions.
This more accurately describes the libc we are using.
This also adds the environment to _all_ Linux systems, not just ARM. And
selects the correct ABI (soft/hardfloat) that's in use.
I don't know whether it has any practical implications, but LLVM appears
to be using this information so it seems wise to use the correct values.
I jumped through quite a few hoops to get test-llvm15-go119 to work, but this last hoop seems not worth jumping through:
cd internal/tools && go generate -tags tools ./
/go/pkg/mod/github.com/golangci/misspell@v0.6.0/mime.go:10:2: package slices is not in GOROOT (/usr/local/go/src/slices)
tools.go:12: running "go": exit status 1
make: *** [GNUmakefile:952: tools] Error 1
I mean, we could patch misspell to not use slices, or to use the prerelease slices, but...
TODO: Remove the go.mod/go.sum in internal/tools once doing so doesn't break CI (e.g. once we drop support for go 1.19)
* builder/cc1as.h: fix typo found by 'make spell'
* GNUmakefile: remove exception for inbetween, fix instance now found by 'make spell'
* GNUmakefile: remove exception for programmmer, fix instance now found by 'make spell'
* go.mod: use updated misspell. GNUmakefile: add spellfix target, use it.
* ignore directories properly when invoking spellchecker.
* make spell: give internal/tools its own go.mod, as misspell requires newer go
* make lint: depend on tools and run the installed revive
(which was perhaps implied by the change that added revive to internal/tools,
but not required in GNUmakefile until we gave internal/tools its own temporary go.mod)
* .github: now that 'make spell' works well, run it from CI
* GNUmakefile: make spell now aborts if it finds misspelt words, so what it finds doesn't get lost in CI logs
* GNUmakefile: tools: avoid -C option on go generate to make test-llvm15-go119 circleci job happy, see
https://cs.opensource.google/go/go/+/2af48cbb7d85e5fdc635e75b99f949010c607786
* internal/tools/go.mod: fix format of go version to leave out patchlevel, else go complains.
Example output:
$ make help
clean Remove build directory
fmt Reformat source
fmt-check Warn if any source needs reformatting
gen-device Generate microcontroller-specific sources
llvm-source Get LLVM sources
llvm-build Build LLVM
lint Lint source tree
spell Spellcheck source tree
Might even work on windows, since git for windows comes with awk.
This was needed in the past because LLVM used typed pointers and there
was a mismatch between pointer types. But we've dropped support for
typed pointers a while ago so now we can remove these wrappers.
This is just a cleanup, it shouldn't have any practical effect.
Previously, the compiler would default to hardfloat. This is not
supported by some MIPS CPUs.
This took me much longer than it should have because of a quirk in the
LLVM Mips backend: if the target-features string is not set (like during
LTO), the Mips backend picks the first function in the module and uses
that. Unfortunately, in the case of TinyGo this first function is
`llvm.dbg.value`, which is an LLVM intrinsic and doesn't have the
target-features string. I fixed it by adding a `-mllvm -mattr=` flag to
the linker.
Move triple calculation into defaultTarget. It was separate for historic
reasons that no longer apply. Merging the code makes future changes
easier (in particular, softfloat support).
The new code is actually shorter than the old code even though it has
more comments.
This shows nicely formatted error messages for missing symbol names and
for out-of-flash, out-of-RAM conditions (on microcontrollers with
limited flash/RAM).
Unfortunately the missing symbol name errors aren't available on Windows
and WebAssembly because the linker doesn't report source locations yet.
This is something that I could perhaps improve in LLD.
The C printf function is sometimes needed for C files included using
CGo. This commit makes sure they're available on all systems where CGo
is fully supported (that is, everywhere except on AVR).
For baremetal systems using picolibc, I've picked the integer-only
version of printf to save on flash size. We might want to consider
providing a way to pick the floating point version instead, if needed.
It's possible to detect the architecture from the target triple, but
there are a number of exceptions that make it unpleasant to use for this
purpose. There are just too many weird exceptions (like mips vs mipsel,
and armv6m vs thumv6m vs arm64 vs aarch64) so it's better to centralize
these to canonical architecture names.
I picked the architecture names that happen to match the musl
architecture names, because those seem the most natural to me.
This adds the same panic locations that are already present for
`tinygo flash -monitor`, but for `tinygo run` and `tinygo test`.
For example, this is the output that I get while working on some GC
code. It now shows the source location instead of just an address:
$ tinygo test -v archive/zip
=== RUN TestReader
=== RUN TestReader/test.zip
panic: runtime error at 0x000000000024d9b4: goroutine stack overflow
[tinygo: panic at /home/ayke/src/tinygo/tinygo/src/internal/task/task_stack.go:58:15]
FAIL archive/zip 0.139s
(This particular location isn't all that useful, but it shows that the
feature works).
These libraries will be automatically built when needed and cached.
The main reason these were needed is for play.tinygo.org, but I've now
prebuilt them there directly (so they don't need to be built for every
tarball).
* reflect: rawFieldByNameFunc: copy index slice to avoid later overwrites
* runtime: Simplify slice growing/appending code
Refactor the slice appending function to rely on the slice growing
function, and remove branches/loops to use a branchfree variant.
Signed-off-by: L. Pereira <l.pereira@fastly.com>
* runtime: Remove one branch in sliceAppend()
Both branches were equivalent, so guard the overall logic in
sliceAppend() with the more general condition.
Signed-off-by: L. Pereira <l.pereira@fastly.com>
* runtime: Simplify slice growing calculation
Use `bits.Len()` rather than `32 - bits.LeadingZeros32()`. They're
equivalent, but the Len version is a bit easier to read.
Signed-off-by: L. Pereira <l.pereira@fastly.com>
* reflect: Always call sliceGrow() in extendSlice()
sliceGrow() will return the old slice if its capacity is large enough.
Signed-off-by: L. Pereira <l.pereira@fastly.com>
---------
Signed-off-by: L. Pereira <l.pereira@fastly.com>
Co-authored-by: Damian Gryski <damian@gryski.com>
For some reason, the type kind name is "unsafe.Pointer" while the
.Name() method just returns "Pointer".
Not sure why this difference exists, but to be able to test .Name() in
testdata/reflect.go it needs to match.
This improves compilation performance by about 5% in my quick test,
while increasing binary size on average by 0.13% when comparing the
smoke tests in the drivers repo (and about two thirds of that 0.13% is
actually caused by a single smoke test).
I think this is a good idea because it aligns the TinyGo optimization
sequence with what ThinLTO expects.
This adds linux/mipsle (little endian Mips) support to TinyGo.
It also adds experimental linux/mips (big-endian) support. It doesn't
quite work yet, some parts of the standard library (like the reflect
package) currently seem to assume a little-endian system.
Previously, the error messages could be shown out of order.
With this patch, the errors are sorted before being shown to the user.
This makes it easier to read the error messages, and matches what the
`go` toolchain does.
This is a refactor, which should (in theory) not change the behavior of
the compiler. But since this is a pretty large change, there is a chance
there will be some regressions. For that reason, the previous commits
added a bunch of tests to make sure most error messages will not be
changed due to this refactor.
This is just one optimizer pass that's easy to test for. There are
others, but I'm ignoring them for now.
The one other that would be reasonable to test is when starting a
goroutine with -gc=none, but I'm leaving that one for another time.
Match the error messages in testdata/errors/*.go like LLVM FileCheck,
which includes regular expressions.
I believe this is much more flexible, and LLVM uses it in nearly all of
their tests so it must work quite well for compilers.
- add internal/wasm-tools/go.mod file to depend on wasm-tools-go
- copy package cm into src/internal/cm
- remove wasm-tools-go "vendor" submodule
internal/tools: fix typo
go.{mod,sum}, internal/tools: add wit-bindgen-go to tools
GNUmakefile: use go run for wit-bindgen-go
GNUmakefile: add tools target to go:generate tools binaries in internal/tools
GNUmakefile: add .PHONY for lint and spell
GNUmakefile, internal/cm: vendor package cm into internal/cm
go.{mod,sum}: update wasm-tools-go to v0.1.4
internal/wasi: use internal/cm package
remove submodule src/vendor/github.com/ydnar/wasm-tools-go
GNUmakefile: add comment documenting what wasi-cm target does
go.{mod,sum}: remove toolchain; go mod tidy
go.mod: revert to Go 1.19
go.mod: go 1.19
go.{mod,sum}, internal/{tools,wasm-tools}: revert root go.mod file to go1.19
Create a wasm-tools specific module that can require go1.22 for wasm-tools-go.
Compilers like GCC keep adding new checks that produce new warnings.
Sometimes it can be false positives.
Do not treat such warnings in binaryen library as errors. tinygo won't
be able to provide zero warnings in its dependencies.
Closes#4332
This is done at a later time anyway, so doesn't need to be done in the
cgo package. In fact, _not_ doing it there makes it easier to print
correct relative packages.
This is needed for some improvements I'm going to make next.
This commit also refactors error handling slightly to make it more
easily testable, this should hopefully not result in any actual changes
in behavior.
Go code might sometimes want to use preprocessor macros that were passed
on the command line. This wasn't working before and resulted in the
following error:
internal error: could not find file where macro is defined
This is now supported, though location information isn't available
(which makes sense: the command line is not a file).
I had to use the `clang_tokenize` API for this and reconstruct the
original source location. Apparently this is the only way to do it:
https://stackoverflow.com/a/19074846/559350
In the future we could consider replacing our own tokenization with the
tokenizer that's built into Clang directly. This should reduce the
possibility of bugs a bit.
Uses a vendored "internal/diff" from Big Go to show the differences
between the expected and actual outputs when tests fail.
Signed-off-by: L. Pereira <l.pereira@fastly.com>
Wasm linear memory is always initialized to zero by definition,
so there's no need to waste time zeroing out this allocation. This
is the case for freshly-obtained memory from mmap().
Signed-off-by: L. Pereira <l.pereira@fastly.com>
* internal/wasi: update to wasm-tools-go@v0.1.1
See https://github.com/ydnar/wasm-tools-go/releases/tag/v0.1.1 for additional information.
* internal/wasi: update to wasm-tools-go@v0.1.2
* internal/wasi: regenerate with wasm-tools-go@v0.1.3
It assumed the maximum alignment was equal to sizeof(void*), which is
definitely not the case. So this only worked more or less by accident
previously.
It now uses the alignment as specified by the frontend, or else
`unsafe.Alignof(complex128)` which is typically the maximum alignment of
a given platform (though this shouldn't really happen in practice: the
optimizer should keep the 'align' attribute in place).
This adds something like 'align 4' to the runtime.alloc calls, so that
the compiler knows the alignment of the allocation and can optimize
accordingly.
The optimization is very small, only 0.01% in my small test. However, my
plan is to read the value in the heap-to-stack transform pass to make it
more correct and let it produce slightly more optimal code.
This means it's possible to test just a particular OS/arch with a
command like this:
go test -run=Build -target=linux/arm
I found it useful while working on MIPS support.
See: https://github.com/tinygo-org/tinygo/issues/4314
The os package isn't particularly useful on wasm-unknown, but just like
on baremetal systems it's imported by a lot of packages so it should at
least be possible to import this package.
The encoding/binary package in Go 1.23 imports the slices package, which
results in circular import when imported from the machine package.
Therefore, encoding/binary cannot be used in the machine package.
This commit fixes that by introducing a new internal/binary package that
is just plain Go code without dependencies. It can be safely used
anywhere (including the runtime if needed).
We previously picked a work-in-progress patch, but this is the proper
fix for this race condition. I think we should use that instead of
relying on the previous work-in-progress patch.
The machine package wasn't tested for every board. Therefore, add a new
serial-like test that also tries to import the machine package. This
should highlight potential issues in the future.
Go 1.18 has been unsupported for quite a while now (the oldest supported
version is Go 1.21). But more importantly, the golang.org/x/tools module
now requires Go 1.19 or later. So we'll drop this older version.
There's no need to keep looping if one of the uses makes it impossible
to convert a call to `runtime.stringToBytes()` with a raw pointer.
Signed-off-by: L. Pereira <l.pereira@fastly.com>
They should, but we weren't testing this.
I discovered this while working on
https://github.com/tinygo-org/macos-minimal-sdk/pull/4 which will likely
make math.h not work anymore. So I wanted to make sure we have a test in
place before we update that dependency.
Some devices, such as Seeed Studio XIAO with Adafruit bootloader, use complex names for UF2 device. With this fix applied, /proc/mounts should be parsed correctly
The current list of targets does not build wasm-ld.
wasm-ld is a symlink created in ./llvm-build/bin pointing to ./lld.
Add the "lld" build target to get wasm-ld into ./llvm-build/bin.
Fixes a build failure where wasm-ld is not found.
Signed-off-by: Christian Stewart <christian@aperture.us>
- Add ReadRegister and WriteRegister (because they're still used by
some packages).
- Fix out-of-bounds panic in I2C.Tx when either w or r has a length of
zero (or is nil).
Both of my development boards exhibit stability problems when
programming at the default 4000 kHz speed of the target/stmf32d4x
OpenOCD configuration.
Note that the speed limit must be set by an event handler, because it
is hard-coded by a similar event handler in stmf32f4x.cfg:
$_TARGETNAME configure -event reset-init {
# Configure PLL to boost clock to HSI x 4 (64 MHz)
...
# Boost JTAG frequency
adapter speed 8000 <-- resolves to 4000 kHz by the SWD interface
}
While here, replace the reference to the deprecated "stlink-v2"
configuration.
Appending to a slice can lead to a race condition if the capacity of the
slice is larger than the length (and therefore the returned slice will
overwrite some fields in the underlying array).
This fixes that race condition. I don't know how severe this particular
one is. It shouldn't be that severe, but it is a bug and it makes it
easier to hunt for possibly more serious race conditions.
sleepTicks calls machineLightSleep with the duration cast to an unsigned
integer. This underflow for negative durations.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
* lint: expand to src/{os,reflect}, fix or suppress what it found
* internal/tools/tools.go: the tools idiom requires a build tag guard to avoid go test complaints
The openocd-commands option cannot refer to commands defined by the
target configuration. Lift this restriction by moving the commands to
after target loading.
See https://github.com/tinygo-org/tinygo/issues/4225
Runs in both circleci and github, circleci is run on branch push, github is run on PR
Revive builds so fast, don't bother installing it; saves us wondering which one we get
Uses tools.go idiom to give control over linter versions to go.mod.
Also pacifies linter re AppendToGlobal as a token first fix.
TODO: gradually expand the number of directories that are linted,
uncomment more entries in revive.toml, and fix or suppress the
warnings lint finds.
TODO: add linters "go vet" and staticcheck
NOT TODO: don't add metalinters like golangci-lint that pull in
lots of new of dependencies; we'd rather not clutter go.mod that
much, let alone open ourselves up to the additional attack surface.
This is much, _much_ faster than __tinygo_spi_transfer which can only
transfer a single byte at a time. This is especially important for
simulated displays.
I've already implemented the browser side of this on the playground and
have used this patch for local testing where it massively speeds up
display operations.
This makes all rp2040 boards available for simulation using
-tags=<board_name>. Importantly, this includes the Gopher Badge which
I'm working on to add to the TinyGo Playground.
This is makes it easier to use these in simulation, plus it avoids
duplication. The only downside I see is that more UARTs get defined than
a board supports, but no existing code should break (no UARTs get
renamed for example).
This eliminates the 'wasi' build tag in favor of 'GOOS=wasip1', introduced in Go 1.21.
For backwards compatablity, -target=wasi is a synonym for -target=wasip1.
It's not generally needed. It was added in
https://github.com/tinygo-org/tinygo/pull/3958 to fix an issue with
binaryen that has since been fixed in a different way, so we don't need
the googletest dependency anymore.
This way you can for example run `make test GOTESTPKGS=./builder` to
only test the builder package.
I've often done this by manually modifying the Makefile, so having a
make parameter available would make this much easier.
Support for `-panic=trap` was previously a pass in the optimization
pipeline. This change moves it to the compiler and runtime, which in my
opinion is a much better place.
As a side effect, it also fixes
https://github.com/tinygo-org/tinygo/issues/4161 by trapping inside
runtime.runtimePanicAt and not just runtime.runtimePanic.
This change also adds a test for the list of imported functions. This is
a more generic test where it's easy to add more tests for WebAssembly
file properties, such as exported functions.
getelementptr offsets are signed, not unsigned. Yet they were used as
unsigned integers in interp.
Somehow this worked most of the time, until finally there was some code
that did a getelementptr with a negative index.
This reduces the size of the Docker image from 13GB to 1.71GB, and
should therefore make CI of the drivers package much faster. It should
hopefully also avoid the out-of-space problem we currently have when
building the smoke tests for the drivers repo.
This size reduction is done by using multistage builds and only copying
the necessary files in the final stage.
This updates lib/cmsis-svd, pulling in the updates to the Espressif SVD
files here:
https://github.com/cmsis-svd/cmsis-svd-data/pull/3
This is needed for wifi/BLE support on the ESP32-C3 (the older SVD files
were missing some necessary interrupts).
CGo is needed for the rp2040 and for macOS (GOOS=darwin), without it
these targets just won't work. And there really isn't a benefit from
disabling CGo: we don't need any external linkers for example.
This avoids a somewhat common issue of people having CGO_ENABLED=0
somewhere in their environment and not understanding why things don't
work. See for example: https://github.com/tinygo-org/tinygo/issues/3450
I managed to get CGo sort-of working in VSCode (meaning that it will
typecheck code in the IDE itself) using a few crude hacks, but it
requires a few minor changes to the TinyGo standard library.
I intend to eventually add this support in the TinyGo extension for
VSCode directly, but for now I've manually updated .vscode/settings.json
to get IDE support. In any case, it would be nice to have this for when
I hopefully add this to the TinyGo extension eventually.
Node.js 16 is no longer supported, so we can drop support for it as
well.
This also means updating a whole lot of GitHub Actions versions, because
they were updated to work on Node.js 20 instead. For most actions this
should be a relatively small change, but the upload-aftifact action has
had some major changes (which should generally improve things a lot).
This means we can finally release native arm64 builds of TinyGo on
macOS!
Also update from macOS 11 to macOS 12, because macOS 11 is not supported
anymore.
Without this change, my tinygo builds crash with
SIGSEGV: segmentation violation
PC=0x10884f87c m=16 sigcode=2
signal arrived during cgo execution
goroutine 378 [syscall]:
runtime.cgocall(0x100631e44, 0x1401755be88)
/nix/store/4hf287252ilsdf2w36mfm8fa0rvbf33w-go-1.21.4/share/go/src/runtime/cgocall.go:157 +0x44 fp=0x1401755be50 sp=0x1401755be10 pc=0x1002a5084
tinygo.org/x/go-llvm._Cfunc_LLVMGoWriteThinLTOBitcodeToMemoryBuffer(0x123614160)
_cgo_gotypes.go:6078 +0x34 fp=0x1401755be80 sp=0x1401755be50 pc=0x1004b66b4
...
and reports using LLVM 16:
$ tinygo version
tinygo version 0.31.0-dev darwin/arm64 (using go version go1.21.4 and LLVM version 16.0.6)
I don't know why libcxx-16 is being included, but explicitly specifying
llvmPackages_17.libcxx fixes the crash and version skew.
These functions are used by the binary blobs that implement wifi/BLE on
the ESP32-C3. While this is a lot of lines of linker script, I think
it's a good idea to add them here (instead of in a separate library)
because they're part of the ESP32-C3 hardware.
The IsZero function was changed slightly in Go 1.22, so we'll need to
update it.
While this could in theory break existing programs that rely on the old
behavior, they'd also break with the imminent Go 1.22 release so I don't
think this is a real problem.
This fixes the new loop variable behavior in Go 1.22.
Specifically:
* The compiler (actually, the x/tools/go/ssa package) now correctly
picks up the Go version.
* If a module doesn't specify the Go version, the current Go version
(from the `go` tool and standard library) is used. This fixes
`go run`.
* The tests in testdata/ that use a separate directory are now
actually run in that directory. This makes it possible to use a
go.mod file there.
* There is a test to make sure old Go modules still work with the old
Go behavior, even on a newer Go version.
This function is needed for Go 1.22, and is used from various packages
like math/rand.
When there is no random number generator available, it falls back to a
static sequence of numbers. I think this is fine, because as far as I
can see it is only used for non-cryptographic needs.
A few non-generic functions like HashStrBytes have been kept for
backwards compatibility, so this package should continue to work with
older Go versions (as long as they support generics).
The right-hand grove port on the wioterminal can be used as UART, using
D0/D1 pins. The pins D0/D1 are tied to SERCOM4, so this patch exposes a
UART4 for sercom4 access.
RX = A0/D0 = PB08/SERCOM4.0 (port 4 pad 0)
TX = A1/D1 = PB09/SERCOM4.1 (port 4 pad 1)
Tested with Lora E5 UART.
uart : = machine.UART4
tx := machine.D0
rx := machine.D1
Note: must also cross Tx/Rx wires in grove cable. See
https://www.lucadentella.it/en/2022/01/29/wio-terminal-porta-grove-di-destra-e-moduli-uart/
* stub out more types/funcs to compile against golang.org/x/net/internal/socket
These are changes need to compile github.com/domainr/dnsr/ with TinyGo.
See issue https://github.com/tinygo-org/net/issues/14.
These change are mostly to fix missing symbols in src/crypto/tls and
src/net. Missing types and functions are cut-and-pasted from go1.21.4.
Functions are stubbed out returning errors.New("not implemented").
DNRS is compiled by running tinygo test:
sfeldma@nuc:~/work/dnsr$ tinygo test -target=wasi
With this patch, and a corresponding patch for tinygo-org/net to fixup
src/net, you should get a clean compile.
Thanks to @aykevl for actually finding and providing this fix, I really just
reported the problem and tested the fix.
Signed-off-by: deadprogram <ron@hybridgroup.com>
This PR adds a network device driver model called netdev. There will be a companion PR for TinyGo drivers to update the netdev drivers and network examples. This PR covers the core "net" package.
An RFC for the work is here: #tinygo-org/drivers#487. Some things have changed from the RFC, but nothing major.
The "net" package is a partial port of Go's "net" package, version 1.19.3. The src/net/README file has details on what is modified from Go's "net" package.
Most "net" features are working as they would in normal Go. TCP/UDP/TLS protocol support is there. As well as HTTP client and server support. Standard Go network packages such as golang.org/x/net/websockets and Paho MQTT client work as-is. Other packages are likely to work as-is.
Testing results are here (https://docs.google.com/spreadsheets/d/e/2PACX-1vT0cCjBvwXf9HJf6aJV2Sw198F2ief02gmbMV0sQocKT4y4RpfKv3dh6Jyew8lQW64FouZ8GwA2yjxI/pubhtml?gid=1013173032&single=true).
The default on MacOS is `md5`, while Nix only has `md5sum` available.
Therefore, make it possible to override the variable via the environment
so that flake.nix can set the correct binary name.
Hopefully this won't break anybody: while all tests still pass, there
could in theory be systems where not supplying those libraries leads to
linker errors.
When TinyGo is installed using `go install` or `go build`, it uses the
Clang resource directory from the host. This works for Linux, but
doesn't work anymore on macOS with a recent change I made.
This re-adds support for Darwin in that case (with much, much simpler
code than there used to be).
This adds a flake.nix file that makes it possible to quickly create a
development environment.
You can download Nix here, for use on your Linux or macOS system:
https://nixos.org/download.html
After you have installed Nix, you can enter the development environment
as follows:
nix develop
This drops you into a bash shell, where you can install TinyGo simply
using the following command:
go install
That's all! Assuming you've set up your $PATH correctly, you can now use
the tinygo command as usual:
tinygo version
You can also do many other things from this environment. Building and
flashing should work as you're used to: it's not a VM or container so
there are no access restrictions.
Set -resource-dir in a central place instead of passing the header path
around everywhere and adding it using the `-I` flag. I believe this is
closer to how Clang is intended to be used.
This change was inspired by my attempt to add a Nix flake file to
TinyGo.
This allows us to test and use LLVM 17, now that it is available in
Homebrew.
Full support for LLVM 17 (including using it by default) will have to
wait until Espressif rebases their Xtensa fork of LLVM.
These parts aren't critical and lead to crashes on small chips without
long jumps (like the attiny85) with LLVM 17. (Older LLVM versions would
emit long jumps regardless, even if the chip didn't support those).
For more information, see: https://github.com/llvm/llvm-project/issues/67042
This happens with `tinygo test` for example when testing multiple
packages at the same time. I found this because the compiler crashed in
`make tinygo-test-fast`:
fatal error: concurrent map writes
fatal error: concurrent map read and map write
goroutine 15 [running]:
github.com/tinygo-org/tinygo/builder.Build({0x40002d0be0, 0xa}, {0x0, 0x0}, {0x4000398048, 0x14}, 0x40003b0000)
/home/ayke/src/tinygo/tinygo/builder/build.go:131 +0x388
main.buildAndRun({0x40002d0be0, 0xa}, 0x40003b0000, {0x8bc178, 0x40003a4090}, {0x0, 0x0, 0x0}, {0x0, 0x0, ...}, ...)
/home/ayke/src/tinygo/tinygo/main.go:856 +0x45c
main.Test({0x40002d0be0, 0xa}, {0x8bc118, 0x40000b3a08}, {0x0?, 0x0?}, 0x40001e6000, {0x0, 0x0})
/home/ayke/src/tinygo/tinygo/main.go:270 +0x758
main.main.func3()
/home/ayke/src/tinygo/tinygo/main.go:1718 +0xe4
created by main.main in goroutine 1
/home/ayke/src/tinygo/tinygo/main.go:1712 +0x34ec
My solution is essentially to copy the map over instead of modifying it
directly.
I've also moved the code up a little, because I think that's a more
sensible place (out of the way of the whole package compile logic).
Previously all (except one!) usage of goenv.Version manually added the
git sha1 hash, leading to duplicate code. I've moved this to do it all
in one place, to avoid this duplication.
The old LLVM pass manager is deprecated and should not be used anymore.
Moreover, the pass manager builder (which we used to set up a pass
pipeline) is actually removed from LLVM entirely in LLVM 17:
https://reviews.llvm.org/D145387https://reviews.llvm.org/D145835
The new pass manager does change the binary size in many cases: both
growing and shrinking it. However, on average the binary size remains
more or less the same.
This is needed as a preparation for LLVM 17.
This is a big change: apart from removing LLVM 14 it also removes typed
pointer support (which was only fully supported in LLVM up to version
14). This removes about 200 lines of code, but more importantly removes
a ton of special cases for LLVM 14.
Prior to this commit, the build instructions were encoded in Makefiles
that contained GNU extensions that are unique to GNU Make and
incompatible with older implementations.
Thie commit renames all Makefiles to GNUmakefile which clearly denotes
that the file contains GNU extensions.
GNU Make actually first looks for a GNUmakefile, then makefile, THEN
Makefile so in addition to simply being more correct and portable, it
saves a few unnecessary failed attempts to open the correct build
file.
Prior to this commit, our Makefiles assumed the name of the make program
was simply "make".
Since we use GNU make extensions, this isn't always the case --
operating systems like FreeBSD come with their own implementation named
make which can be incompatible with the extensions used in by tinygo.
To run a compatible make on some of these systems, we have explicitly
call GNU make by running gmake.
This commit changes references to the command "make" to "$(MAKE)" which
is a variable that contains the name of the executable invoked to
process the Makefile.
This allow the Makefiles to be uniformly processed by the same make
program.
This fixes https://github.com/tinygo-org/tinygo/issues/3926.
While working on this I've found another bug: if C.bool is referenced
from within Go, it isn't available anymore on the C side. This is an
existing bug that also applies to float and double, but may be less
likely to be triggered there.
This bug is something to be fixed at a later time (it has something to
do with preprocessor defines).
Previously, this pass would convert any read-only use of a
runtime.stringToBytes call to use the original string buffer instead.
This is incorrect: if there are any writes to the resulting buffer, none
of the slice buffer pointers can be converted to use the original
read-only string buffer.
This commit fixes that bug and adds a test to prove the new (correct)
behavior.
The old traceback would look like this:
# internal/godebug
/usr/local/go/src/internal/godebug/godebug.go:101:11: interp: test
call <2> 0 <3> 0
traceback:
/usr/local/go/src/internal/godebug/godebug.go:101:11:
call <2> 0 <3> 0
/usr/local/go/src/internal/godebug:
call <1> 0
With this patch, it looks like this:
# io/fs
/usr/local/go/src/io/fs/fs.go:144:45: interp: test
%0 = load %runtime._interface, ptr @"internal/oserror.ErrInvalid", align 8, !dbg !316
traceback:
/usr/local/go/src/io/fs/fs.go:144:45:
%0 = load %runtime._interface, ptr @"internal/oserror.ErrInvalid", align 8, !dbg !316
/usr/local/go/src/io/fs/fs.go:137:28:
%0 = call %runtime._interface @"io/fs.errInvalid"(ptr undef), !dbg !317
For developers (like me) who are familiar with LLVM, this is probably
easier to read. For users, I'm not sure: the instructions have quite a
lot of distracting fluff in them. But at least it contains the function
names that are called (which are not currently present in the old
traceback).
...that said, having the LLVM instructions in a bug report is probably
going to be easier for people who are familar with LLVM.
This is a small release just in time for GopherCon US.
Some of the big changes of this release are:
- switch to LLVM 16
- fixes for two separate hard-to-reproduce compiler crashes
- added support for the Adafruit Gemma M0
This usually works by chance, but leads to crashes. So we should never
ever do this.
I'm pretty sure this is the crash behind this issue: https://github.com/tinygo-org/tinygo/issues/3894
It may also have caused this crash: https://github.com/tinygo-org/tinygo/issues/3874
I have a suspicion this is also behind the rather crash-prone CircleCI
jobs, that we haven't been able to find the source of. But we'll find
out soon enough once this fix is merged.
To avoid hitting this issue again in the future, I've created a PR to
remove these dangerous functions altogether from the go-llvm API:
https://github.com/tinygo-org/go-llvm/pull/54
This commit adds support for LLVM 16 and switches to it by default. That
means three LLVM versions are supported at the same time: LLVM 14, 15,
and 16.
This commit includes work by QuLogic:
* Part of this work was based on a PR by QuLogic:
https://github.com/tinygo-org/tinygo/pull/3649
But I also had parts of this already implemented in an old branch I
already made for LLVM 16.
* QuLogic also provided a CGo fix here, which is also incorporated in
this commit:
https://github.com/tinygo-org/tinygo/pull/3869
The difference with the original PR by QuLogic is that this commit is
more complete:
* It switches to LLVM 16 by default.
* It updates some things to also make it work with a self-built LLVM.
* It fixes the CGo bug in a slightly different way, and also fixes
another one not included in the original PR.
* It does not keep compiler tests passing on older LLVM versions. I
have found this to be quite burdensome and therefore don't generally
do this - the smoke tests should hopefully catch most regressions.
This was bug https://github.com/tinygo-org/tinygo/issues/3890.
See the diff for details. Essentially, it means that `copy` in the
interpreter was copying data that wasn't known yet, or copying data into
a slice that could be read externally after the interp pass.
such as program changes and pitch bend. Also add error handling for invalid
parameter values such as MIDI channel. This however makes a somewhat breaking
change to the current implementation, in that we now use the typical MIDI user
system of counting MIDI channels from 1-16 instead of from 0-15 as the lower
level USB-MIDI API itself expects.
Also add constant values for continuous controller messages, rename SendCC
function, and add SysEx function.
Signed-off-by: deadprogram <ron@hybridgroup.com>
Instead of markGlobals calling markRoots unconditionally (which doesn't
make sense for -gc=none and -gc=leaking), provide markRoots as a
callback function.
This is in preparation for -gc=boehm, where the previous design is even
more awkward and a callback makes far more sense.
I've tested the size impact using `make smoketest XTENSA=0`. There is
none, except for two cases:
* One with `-opt=0` so const-propagation for the callback didn't take
place.
* One other on AVR, I don't know why but as it's only 16 bytes in a
very specific case I'm going to assume it's just a random change in
compiler output that caused a size difference.
This adds true GOOS=wasip1 support in addition to our existing
-target=wasi support. The old support for WASI isn't removed, but should
be treated as deprecated and will likely be removed eventually to reduce
the test burden.
It looks like this on my system, for example:
{
"target": {
"llvm-target": "aarch64-unknown-linux",
"cpu": "generic",
"features": "+neon",
"goos": "linux",
"goarch": "arm64",
"build-tags": [
"linux",
"arm64"
],
"gc": "precise",
"scheduler": "tasks",
"linker": "ld.lld",
"rtlib": "compiler-rt",
"libc": "musl",
"default-stack-size": 65536,
"ldflags": [
"--gc-sections"
],
"extra-files": [
"src/runtime/asm_arm64.S",
"src/internal/task/task_stack_arm64.S"
],
"gdb": [
"gdb"
],
"flash-1200-bps-reset": "false"
},
"goroot": "/home/ayke/.cache/tinygo/goroot-23c311bcaa05f188affa3c42310aba343acc82562d5e5f04dea9d5b79ac35f7e",
"goos": "linux",
"goarch": "arm64",
"goarm": "6",
...
}
This can be very useful while working on the automatically generated
target object for example (in my case, GOOS=wasip1).
This should not happen under normal circumstances. It can still happen
when there is a mismatch between TinyGo version and the associated
runtime, or while developing the compiler package.
When a function is exported using //export, but also had a
//go:wasm-module pragma, the //export name was ignored. The
//go:wasm-module doesn't actually do anything besides breaking the
export (exported functions don't have a module name).
I've refactored and cleaned up the code, and in the process removed this
weird edge case.
This commit adds I2C timeouts for nrf51 and nrf52 (but not yet for
others like nrf52840).
Tested on the PineTime, where I now got a timeout instead of hanging and
resetting due to a watchdog reset.
* add gosched calls to UART
* add UART.flush() stubs for all supported architectures
* add comment un uart.go on flush functionality
* uart.writeByte as base of UART usage
* fix NXP having duplicate WriteByte
* fix writeByte not returning error on some platforms
* add flush method for fe310 device
* check for error in WriteByte call to writeByte
This replaces our own manual detection of various variables (GOROOT,
GOPATH, Go version) with a simple call to `go env`.
If the `go` command is not found:
error: could not find 'go' command: executable file not found in $PATH
If the Go version is too old:
error: requires go version 1.18 through 1.20, got go1.17
If the Go tool itself outputs an error (using GOROOT=foobar here):
go: cannot find GOROOT directory: foobar
This does break the case where `go` wasn't available in $PATH but we
would detect it anyway (via some hardcoded OS-dependent paths). I'm not
sure we want to fix that: I think it's better to tell users "make sure
`go version` prints the right value" than to do some automagic detection
of Go binary locations.
This is a rewrite of how filesystems are detected. Specifically, it
fixes an issue on Linux where the location of the FAT filesystem can
vary between distributions (for example, we supported most distros by
checking two different paths, but NixOS uses a different path): it now
uses the data in /proc/mounts instead which should be universal.
This adds FieldByNameFunc, which some libraries like reflect2 need.
For my usecase I could also just stub FieldByNameFunc to panic, but
figured that it would work OK to just make it work. I'm not sure if
the overhead to FieldByName using a closure is acceptable.
Signed-off-by: Tyler Rockwood <rockwood@redpanda.com>
Found while working on the PineTime. For some reason it still kind of
works in most cases, but I was hitting this issue when interacting with
two different I2C devices (the touch sensor and the BMA421).
Browsers previously didn't support the WebAssembly i64 type, so we had
to work around that limitation by converting the LLVM i64 type to
something else. Some people used a pair of i32 values, but we used a
pointer to a stack allocated i64.
Now however, all major browsers and Node.js do support WebAssembly
BigInt integration so that i64 values can be passed back and forth
between WebAssembly and JavaScript easily. Therefore, I think the time
has come to drop support for this workaround.
For more information: https://v8.dev/features/wasm-bigint (note that
TinyGo has used a slightly different way of passing i64 values between
JS and Wasm).
For information on browser support: https://webassembly.org/roadmap/
The Linux artifacts have clear names (linux-amd64-double-zipped etc),
but the MacOS and Windows ones didn't. This patch renames these artifact
names to be more readable, especially when downloading the artifacts.
These are some major or breaking changes:
- Reflect support got improved a lot.
- Interrupts became more strict: heap allocations an blocking
operations, which have always been undefined behavior, now result in
a panic.
- `//go:wasmimport` was added
- various new flags were added to `tinygo test`
- the source location for panics is printed in the `-monitor` output
The regular port access is around 4 cycles, instead of the usual 2
cycles for a store instruction on Cortex-M0+. The IOBUS however is
faster, I didn't measure exactly but I guess it's 2 cycles as expected.
This fixes a bug in the WS2812 driver that only happens on samd21 chips:
https://github.com/tinygo-org/drivers/issues/540
I didn't add this method in the initial PR.
Also, I found that a few of my assumptions were incorrect. I've changed
the code that configures the pin to make input (floating and pullup)
actually work. These chips really are quite different from all the older
AVRs.
If a pointer value was xor'ed with a value other than 0, it would not
have been run at runtime but instead would fall through to the generic
integer operations. This would likely result in a "cannot convert
pointer to integer" panic.
This commit fixes this subtle case.
This is just support for the chip, no boards are currently supported.
However, you can use this target on a custom board.
Notes:
- This required a new runtime and machine implementation, because the
hardware is actually very different (and much nicer than older
AVRs!).
- I had to update gen-device-avr to support this chip. This also
affects the generated output of other AVRs, but I checked all chips
we support and there shouldn't be any backwards incompatible
changes.
- I did not implement peripherals like UART, I2C, SPI, etc because I
don't need them. That is left to do in the future.
You can flash these chips with only a UART and a 1kOhm resistor, which
is really nice (no special hardware needed). Here is the program I've
used for this purpose: https://pypi.org/project/pymcuprog/
This refactors gen-device-avr to output two different formats: one for
all the existing AVR chips (that don't really have the concept of a
peripheral, just a bunch of registers), and one for all the new chips
like the ATtiny1616 (tinyAVR 1-series and 2-series) that have
peripherals like the Cortex-M chips with type structs and instances.
I checked the generated code for all the AVR chips we have support for
(atmega1280, atmega1284p, atmega2560, atmega328p, atmega32u4, attiny85)
and while the generated Go code did change, it looks safe to me.
This is a small change that's not really important in itself, but it
avoids duplicate errors in a future commit that adds error messages to
//go:wasmimport.
This basically reverts https://github.com/tinygo-org/tinygo/pull/3357
and replaces it with a different mechanism to get to the same goal.
I do not think filtering tags like this is a good idea: it's the wrong
part of the compiler to be concerned with such tags (that part sets
tags, but doesn't modify existing tags). Instead, I've written the
//go:build lines in such a way that it has the same effect: WASI
defaults to leveldb, everything else defaults to fnv, and it's possible
to override the default using build tags.
It could be expensive to call Size() three times, and it is unnecessary.
Instead, do it only once.
This results in a very small reduction of binary size if Zero() is used.
This makes reviewing PRs a lot easier because I don't have to run this
myself :)
This only uses the drivers repo so far, which is a good starting point
but doesn't include binary size changes for WebAssembly for example. A
future change could add some real-world programs to get a better idea of
the real-world impact.
To be clear: the intention is not to just look at the number at the
bottom. It is important to look at the actual size difference to see the
overall pattern (like, the difference may be due to a few outlier).
This avoids a dependency on nrfutil. I have verified that it creates
equivalent zip files to a wasp-os DFU zip file I downloaded here:
https://github.com/wasp-os/wasp-os/releases/
I have also tested that it produces valid DFU files that can be uploaded
using the dfu.py program here to my PineTime:
https://github.com/wasp-os/ota-dfu-python/tree/3d6fd30d33c2b20bc86ff6b9269fddf4a1d4c7c6
There are some minor differences in the generated file that should not
matter in practice (JSON whitespace, firmware file name, zip
compression).
* Initialize the ADC in Configure() (instead of in Get()).
* Do not set all channels to "not connected" - that's already the
reset value.
* Don't disable the ADC after use. It's not necessary to disable
(current consumption appears to remain the same whether enabled or
disabled).
* Rename pinetime-devkit0 to pinetime because the production device is
almost the same hardware (the only noticeable difference is a
different accelerometer, which isn't part of the board file).
* Remove the UART and set serial to none. The UART uses a lot of
current by default, so it seems better to disable it.
This is a breaking change, but honestly I think I'm the only one who has
ever actually used TinyGo for the PineTime and I'm fine with this
change :)
This tool can be very useful to compare binary sizes as output by
`-size=short`.
This is a tool I wrote a while ago. It's not perfect (we should probably
use a geomean) but it works well enough to get a good idea on the binary
size impact of a change.
The old code was broken and led to a HardFault in a rather convoluted
way:
1. The CFA offset was incorrect, in fact it was not aligned (the stack
is supposed to always be aligned to 4 bytes at least).
2. This unaligned size was then used for stack size calculations.
3. A stack that wasn't a multiple of 4 was allocated.
4. The calleeSavedRegs struct (in `(internal/task.state).archInit`) was
not correctly aligned.
5. Writing to this struct resulted in a HardFault.
The previous commit started printing the instruction address for runtime
panics. This commit starts using this address to print the source
location.
Here is an example where this feature is very useful. There is a heap
allocation in the Bluetooth package, but we don't know where exactly.
Printing the instruction address of the panic is already useful, but
what is even more useful is looking up this address in the DWARF debug
information that's part of the binary:
$ tinygo flash -target=circuitplay-bluefruit -monitor ./examples/heartrate
Connected to /dev/ttyACM0. Press Ctrl-C to exit.
tick 00:00.810
tick 00:01.587
tick 00:02.387
tick 00:03.244
panic: runtime error at 0x00027c4d: alloc in interrupt
[tinygo: panic at /home/ayke/src/tinygo/bluetooth/adapter_sd.go:74:4]
To be clear, this path isn't stored on the microcontroller. It's stored
as part of the build, and `-monitor` just looks up the path from the
panic message.
Possible enhancements:
- Print such an address for regular panics as well. I'm not sure
that's so useful, as it's usually a lot easier to look up panics
just by their message.
- Use runtimePanicAt (instead of runtimePanic) in other locations, if
that proves to be beneficial.
- Print the TinyGo-generated output in some other color, to
distinguish it from the regular console output.
- Print more details when panicking (registers, stack values), and
print an actual backtrace.
This is not very useful in itself, but makes it possible to detect this
address in the output. See the next commit.
This adds around 50 bytes to each binary (except for AVR and wasm). This
is unfortunate, but I think this feature is quite useful still.
A future enhancement might be to create a build tag for extended panic
information that's not set by default.
Multisampling/averaging (using the Samples configuration property) was
returning incorrect values. When I investigated this, I found that the
samd51 gives erratic values when using multisampling together with fewer
than 16 bits resolution.
I fixed this by forcing 16 bit resolution when multisampling, and
adjusting the output to account for multisampling.
Found while reading the battery value on a pybadge, which gave
non-sensible values with Samples set to a value larger than 1.
This improves slightly. It also is some groundwork for better DMA
support in TinyGo in the future.
I'm not entirely sure why it improves performance (in theory the old
code should already saturate the SPI bus) but it does, so 🤷
The SPI frequency was rounded up, not rounded down. This meant that if
you wanted to configure 15MHz for example, it would pick the next
available frequency (24MHz). That's unsafe, the safe option is to round
down and the SPI support for most other chips also rounds down for this
reason.
In addition, I've improved SPI clock selection so that it will pick the
best clock of the two, widening the available frequencies. See the
comments in the patch for details.
I find myself consistently running tests, seeing them panic, and then
immediately running them again with this environment variable set. It's
easier to just have tinygo do this for me.
The only thing that's different between all these chips is the flash
size, which can easily be passed as a linker flag instead. This removes
a bunch of duplicate code in an uncommon language (linker script).
I've also fixed a few boards with incorrect flash sizes:
* nano-rp2040 has 16MB instead of 2MB
* macropad-rp2040 has 8MB instead of 2MB
* gopher-badge has 8MB instead of 1MB
These functions can be implemented more efficiently using LLVM
intrinsics. That makes them the Go equivalent of functions like
__builtin_clz which are also implemented using these LLVM intrinsics.
I believe the Go compiler does something very similar: IIRC it converts
calls to these functions into optimal instructions for the given
architecture.
I tested these by running `tinygo test math/bits` after uncommenting the
tests that would always fail (the *PanicZero and *PanicOverflow tests).
As discussed on Slack, I believe this property does more harm than good:
* I don't think it's used anywhere. None of the drivers use it.
* It is not fully implemented. While values <= 8 might work fine,
values larger than 8 result in extra zero bits (instead of anything
sensible).
* Worse, it doesn't return an error when it's out of range. This is
not an optional property: if the SPI peripheral doesn't support a
particular number of bits, it should return an error instead of
silently limiting the number of bits. This will be confusing to
users.
Therefore, I propose we drop it. Maybe there are good uses for it
(perhaps for displays that use big endian 16-bit values?), but without a
good use case like a driver in tinygo.org/x/drivers, I think it's more
trouble than it's worth.
The runtime.stringFromBytesTyped and runtime.stringToBytesTyped
functions aren't really necessary, because they have the same LLVM IR
signature. Therefore, remove them and link directly to the functions
that the compiler uses internally.
This gives a small improvement now, and is needed to be able to use the
Heap2Stack transform that's available in the Attributor pass. This
Heap2Stack transform could replace our custom OptimizeAllocs pass.
Most of the changes are just IR that changed, the actual change is
relatively small.
To give an example of why this is useful, here is the code size before
this change:
$ tinygo build -o test -size=short ./testdata/stdlib.go
code data bss | flash ram
95620 1812 968 | 97432 2780
$ tinygo build -o test -size=short ./testdata/stdlib.go
code data bss | flash ram
95380 1812 968 | 97192 2780
That's a 0.25% reduction. Not a whole lot, but nice for such a small
patch.
There was a very subtle bug in the ADC read code: it stores a pointer to
a variable in a register, waits for the hardware to complete the read,
and then reads the value again from the local variable. Unfortunately,
the compiler doesn't know there is some form of synchronization
happening in between.
This can be fixed in roughly two ways:
* Introduce some sort of synchronization.
* Do a volatile read from the variable.
I chose the second one as it is probably the least intrusive. We
certainly don't need atomic instructions (the chip is single threaded),
we just need to tell the compiler the value could have changed by making
the read volatile.
It is possible to create function-local named types:
func foo() any {
type named int
return named(0)
}
This patch makes sure they don't alias with named types declared at the
package scope.
Bug originally found by Damian Gryski while working on reflect support.
Previously we only supported recursive types in structs. But there can
be other kinds of recursive types, like slices:
type RecursiveSlice []RecursiveSlice
This doesn't involve structs, so it led to infinite recursion in the
compiler. This fix avoids recursion at the proper level: at the place
where the named type is defined.
This test only applies when using the built-in LLVM version. This way,
we have a stable LLVM version to test against. Distribution versions of
LLVM (especially Debian) tend to be patched in a way that affect the
results.
This has two benefits:
1. It attributes these bytes to the internal/task package (in
-size=full), instead of (unknown).
2. It makes it possible to print the stack sizes variable in GDB.
This is what it might look like in GDB:
(gdb) p 'internal/task.stackSizes'
$13 = {344, 120, 80, 2048, 360, 112, 80, 120, 2048, 2048}
Previously we were using a really weird calculation to determine the
alignment in bits - written by me (no idea what I was thinking at the
time, it's obviously incorrect). Just replace it with the much more
obviously correct multiplication by 8 to get bits from bytes.
Found while working on properly dealing with alignment in `-size=full`.
The compiler now uses DW_OP_plus_uconst in the address attribute of
DWARF variables. To be able to understand these addresses, we need to be
able to parse this opcode.
This commit should also improve the reliability of address parsing a
bit.
Previously the code size of the compiler-rt library might be displayed
like this:
```
1050 0 0 0 | 1050 0 | /home/ayke/src/tinygo/tinygo/llvm-project/compiler-rt/lib/builtins
146 0 0 0 | 146 0 | /home/ayke/src/tinygo/tinygo/llvm-project/compiler-rt/lib/builtins/arm
```
With this change, it is displayed nicely like this:
```
1196 0 0 0 | 1196 0 | C compiler-rt
```
This should fix a number of concurrency/threading issues.
I had to force-disable concurrency in the linker using a hack. I'm not
entirely sure what the cause is, possibly the MinGW version (version 12
appears to work for me, while version 11 as used on the GitHub runner
image seems to be broken).
There are a few ways to fix this in a better way:
* Fix the underlying cause (possibly by upgrading to MinGW-w64 12).
* Add the `--threads` flag to the LLD MinGW linker, so we can use a
regular parameter instead of this hack.
This is an attempt to figure out why the Windows CI keeps crashing. It
should be removed before the next release, by which point we should know
whether this has helped or not.
There was a mostly benign race condition in the compiler. The issue was
that there is a check for type aliases (which can alias types in another
function), but this check was _after_ accessing a property of the
function that might not have been completed.
I don't think this can have any serious effects, as the function is
skipped anyway, but such bugs should certainly be fixed.
If there is no source location, at least use the file (but not line) for
debug information.
This attributes almost 1kB of string data for ./testdata/stdlib.go when
compiling for WASI.
This is helpful for WebAssembly: it makes it possible to attribute many
more data to locations in the source code, which makes `-size=full` more
useful.
Removes usage of AsmFull which required an optimization pass to remove the map parameter passed into it. This caused issues when compiling with -opt=0 where memory for the map was being allocated as an unintended side-effect of using AsmFull with no optimizations enabled.
I have checked this conversion is not needed anymore after the previous
commit, by running various smoke tests of which none triggered this
optimization. The only case where the optimization would have kicked in
is in syscall/syscall_windows.go:76 of the Go standard library.
Therefore, I prefer to remove it to reduce code complexity.
We have an optimization for this specific pattern, but it's really just
a hack. With the addition of unsafe.Add in Go 1.17 we can directly
specify the intent instead and eventually remove this special case.
The code is also easier to read.
This metadata is emitted by Clang and I found it is important for source
level debugging on MacOS. This patch does not get source level debugging
to work yet (for that, it seems like packages need to be built
separately), but it is a step in the right direction.
machine/stm32, nrf: implement machine.Flash
Implements the machine.Flash interface using the same definition as the tinyfs BlockDevice.
This implementation covers the stm32f4, stm32l4, stm32wlx, nrf51, nrf52, and nrf528xx processors.
Blocking inside an interrupt is always unsafe and will lead to all kinds
of bad behavior (even if it might appear to work sometimes). So disallow
it, just like allocating heap memory inside an interrupt is not allowed.
I suspect this will usually be caused by channel sends, like this:
ch <- someValue
The easy workaround is to make it a non-blocking send instead:
select {
case ch <- someValue:
default:
}
This does mean the application might become a bit more complex to be
able to deal with this case, but the alternative (undefined behavior) is
IMHO much worse.
Add the timers test because they now work correctly on AVR, probably as
a result of the reflect refactor: https://github.com/tinygo-org/tinygo/pull/2640
I've also updated a few of the other tests to indicate the new status
and why they don't work. It's no longer because of compiler errors, but
because of linker or runtime errors (which is at least some progress).
For example, I found that testdata/reflect.go works if you disable
`testAppendSlice` and increase the stack size.
It can be difficult to find what went wrong in a test. Omitting -v
should make it easier to see the failing tests and the output for them
(note that output is still printed for tests that fail).
2023-02-20 00:23:52 +01:00
1366 changed files with 95556 additions and 20346 deletions
TinyGo is a Go compiler intended for use in small places such as microcontrollers, WebAssembly (Wasm), and command-line tools.
TinyGo is a Go compiler intended for use in small places such as microcontrollers, WebAssembly (wasm/wasi), and command-line tools.
It reuses libraries used by the [Go language tools](https://golang.org/pkg/go/) alongside [LLVM](http://llvm.org) to provide an alternative way to compile programs written in the Go programming language.
> [!IMPORTANT]
> You can help TinyGo with a financial contribution using OpenCollective. Please see https://opencollective.com/tinygo for more information. Thank you!
## Embedded
Here is an example program that blinks the built-in LED when run directly on any supported board with onboard LED:
```go
@@ -29,121 +34,70 @@ func main() {
}
```
The above program can be compiled and run without modification on an Arduino Uno, an Adafruit ItsyBitsy M0, or any of the supported boards that have a built-in LED, just by setting the correct TinyGo compiler target. For example, this compiles and flashes an Arduino Uno:
The above program can be compiled and run without modification on an [Arduino Uno](https://tinygo.org/docs/reference/microcontrollers/boards/arduino-uno), an [Adafruit Circuit Playground Express](https://tinygo.org/docs/reference/microcontrollers/featured/circuitplay-express), a [Seeed Studio XIAO-ESP32S3](https://tinygo.org/docs/reference/microcontrollers/featured/xiao-esp32s3) or any of the many supported boards that have a built-in LED, just by setting the correct TinyGo compiler target. For example, this compiles and flashes an Arduino Uno:
```shell
tinygo flash -target arduino examples/blinky1
tinygo flash -target arduino-uno examples/blinky1
```
## WebAssembly
TinyGo is very useful for compiling programs both for use in browsers (WASM) as well as for use on servers and other edge devices (WASI).
TinyGo programs can run in [Fastly Compute](https://www.fastly.com/documentation/guides/compute/go/), [Fermyon Spin](https://developer.fermyon.com/spin/go-components), [wazero](https://wazero.io/languages/tinygo/) and many other WebAssembly runtimes.
Here is a small TinyGo program for use by a WASI host application:
```go
packagemain
//go:wasmexport add
funcadd(x,yuint32)uint32{
returnx+y
}
```
This compiles the above TinyGo program for use on any WASI Preview 1 runtime:
See the [getting started instructions](https://tinygo.org/getting-started/) for information on how to install TinyGo, as well as how to run the TinyGo compiler using our Docker container.
## Supported boards/targets
## Supported targets
You can compile TinyGo programs for microcontrollers, WebAssembly and Linux.
### Embedded
The following 94 microcontroller boards are currently supported:
You can compile TinyGo programs for over 150 different microcontroller boards.
returnnil,fmt.Errorf("cannot compile with Go toolchain version go%d.%d (TinyGo was built using toolchain version %s)",gorootMajor,gorootMinor,runtime.Version())
<p>How much space is used by Go packages, C libraries, and other bits to set up the program environment.</p>
<ul>
<li><strong>Code</strong> is the actual program code (machine code instructions).</li>
<li><strong>Read-only data</strong> are read-only global variables. On most microcontrollers, these are stored in flash and do not take up any RAM.</li>
<li><strong>Data</strong> are writable global variables with a non-zero initializer. On microcontrollers, they are copied from flash to RAM on reset.</li>
<li><strong>BSS</strong> are writable global variables that are zero initialized. They do not take up any space in the binary, but do take up RAM. On microcontrollers, this area is zeroed on reset.</li>
</ul>
<p>The binary size consists of code, read-only data, and data. On microcontrollers, this is exactly the size of the firmware image. On other systems, there is some extra overhead: binary metadata (headers of the ELF/MachO/COFF file), debug information, exception tables, symbol names, etc. Using <code>-no-debug</code> strips most of those.</p>
<h2>Program breakdown</h2>
<p>You can click on the rows below to see which files contribute to the binary size.</p>
matches:=regexp.MustCompile(`referenced by .* \(((.*):([0-9]+))\)`).FindStringSubmatch(line)
ifmatches!=nil{
parsedError=true
line,_:=strconv.Atoi(matches[3])
msg:="linker could not find symbol "+symbolName
switchsymbolName{
case"runtime.alloc":
msg="object allocated on the heap with -gc=none"
case"runtime.alloc_noheap":
msg="object allocated on the heap in //go:noheap function"
}
linkErrors=append(linkErrors,scanner.Error{
Pos:token.Position{
Filename:matches[2],
Line:line,
},
Msg:msg,
})
}
}
}
// Check for flash/RAM overflow.
ifmatches:=regexp.MustCompile(`^ld.lld(-[0-9]+)?: error: section '(.*?)' will not fit in region '(.*?)': overflowed by ([0-9]+) bytes$`).FindStringSubmatch(message);matches!=nil{
region:=matches[3]
n,err:=strconv.ParseUint(matches[4],10,64)
iferr!=nil{
// Should not happen at all (unless it overflows an uint64 for some reason).
continue
}
// Check which area overflowed.
// Some chips use differently named memory areas, but these are by
// far the most common.
switchregion{
case"FLASH_TEXT":
ifn>flashOverflow{
flashOverflow=n
}
parsedError=true
case"RAM":
ifn>ramOverflow{
ramOverflow=n
}
parsedError=true
}
}
// If we couldn't parse the linker error: show the error as-is to
Msg:fmt.Sprintf("program too large for this chip (flash overflowed by %d bytes)\n\toptimization guide: https://tinygo.org/docs/guides/optimizing-binaries/",flashOverflow),
})
}
iframOverflow>0{
linkErrors=append(linkErrors,LinkerError{
Msg:fmt.Sprintf("program uses too much static RAM on this chip (RAM overflowed by %d bytes)",ramOverflow),
})
}
returnnewMultiError(linkErrors,"")
}
// LLD linker error that could not be parsed or doesn't refer to a source
// Fix some warnings on Windows ARM. Without the __declspec(dllexport), it gives warnings like this:
// In file included from _cgo_export.c:4:
// cgo-gcc-export-header-prolog:49:34: warning: redeclaration of 'tinygo_clang_globals_visitor' should not add 'dllexport' attribute [-Wdll-attribute-on-redeclaration]
// libclang.go:68:5: note: previous declaration is here
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Patricio Whittingslow
Jake Bailey
rdon(あーるどん)
rdon
deadprogram
Damian Gryski
Ayke van Laethem
iamrajiv
Nia Waldvogel
Marco Feltmann
Konstantin Sharlaimov
Piotr Bocheński
zowhoey
sago35
Matthew Hiles
zoey
Matthew Mets
Kenneth Bell
Juraj Michálek
Achille
felipegenef
pedramktb
Joel Wetzell
MaximAL
Abhishek Goyal
Carlos Henrique Guardão Gandarez
Carlos Henrique Guardão Gandarez
Dima Jolkin
Dima
Michael Teichgräber
Micah Cowell
Elias Naur
Yaj
robo
Jesús Espino
Itxaka
Amken USA
gram
MakKi (makki_d)
Michael Smith
cjpeterson
sivchari
Ben Krieger
tinkerator
Olaf Flebbe
张之阳
Daniel Esteban
Ben Krieger
Matt Mets
luoliwoshang
Thomas Kohler
Michael Smith
Dmitrii Sharshakov
Elliott Sales de Andrade
diwamoto
emricks
micchie
Ömer Faruk IRMAK
Ella Fox
Mateusz Nowak
Rolf Sommerhalder
Caleb Champlin
Randy Reddig
Egawa Takashi
Scott Feldman
Laurent Demailly
leongross
JP Hastings-Spital
m00874241
HattoriHanzo031
Yurii Soldak
milkpirate
vaaski
Roman Grudzinski
Volodymyr Pobochii
Thomas Legris
Joonas Bergius
BCG
archie2x
Warren Guy
Dan Kegel
Unrud
dkegel-fastly
Kobayashi Shunta
L. Pereira
Matthew Hiles
Tai Groot
Anatol Pomozov
Dmitry Shemin
leongross
diamondburned
frenkel26
Johann Freymuth
Marco Manino
Christian Stewart
hongkuang
Patrick Ting
Patrick Lindsay
Jonathan Böcker
Anders Savill
Akshay Pai
BCG
Xudong Zheng
=
Dmitriy
Rado M
Shane O'Donovan
Randy Reddig
Christian Ege
Flavio Castelli
ginglis13
Anuraag Agrawal
ayan george
Timothy Rule
Pertti Erkkilä
Pierre Constantineau
Charlie Haley
Mansour Behabadi
Tyler Rockwood
Stepan Rakitin
ivoszz
Rajat Jindal
Daz Wilkin
Edoardo Vacchi
cui fliter
waj334
Justin A. Wilson
shivay
Damian Gryski
Christian Stewart
Federico G. Schwindt
Bjoern Poetzschke
John Clark