avr/bootloader
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

fix: --flash never ran the brick guard, because it never read the fuses

Browse Source 
check_walk_region() refuses an image writing into the span reset crosses to
reach the loader, and a plain --flash never ran it: the fuses were read for
--fuses and for --update-loader, so the tool read them to protect the loader
and never to protect the reset path. It cost an ssd1306 board - an application
grown through 0x7c00 on a 328P with hfuse 0xdc, reset landing mid-function,
an ICE the only way back.

The check now fetches its own input, so the operation that asks for no fuses
cannot skip it and neither can a direct API caller: pbdirty and pbmute call
op_flash() as a library and are guarded without a line changing in them.
Fuses that cannot be read are a refusal naming --assume-fuses and --force,
because unknown is not empty.

The rig had to stop lying first. The fuse read is an LPM diverted by BLBSET,
which simavr executes straight out of flash with no hook, so a fuse read
answered flash bytes and --fuses had been printing them on every chip. The
runner models the diversion at the SPMCSR write, -f states the profile - and
stores the register itself, since a registered handler replaces simavr's store
and would otherwise swallow every SPM command on the cores where nothing else
watches it.

pureboot.walk reproduces the brick: the unfixed tool writes 249 pages through
0x7c00 in silence, the fixed one refuses.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
BlackMark
2026-08-16 03:56:20 +02:00
parent 6ff407e7bb
commit 37f9747e26
7 changed files with 272 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
test/pbsim.py
View File

@@ -9,12 +9,16 @@ import subprocess
class Device:
def __init__(self, binary, elf, mcu, hz, base_hex, page, baud, dump, reset_hex=None, resume=None, link=None,
window=False):
window=False, fuses=None):
cmd = [binary]
if link:
cmd += ["-l", link]
if window:
cmd.append("-w") # report the first-transmit cycle, free-run idle
if fuses:
# What a fuse read answers, low,lock,extended,high. Unprogrammed
# otherwise, which is the profile every other test here runs.
cmd += ["-f", fuses]
cmd += [elf, mcu, hz, base_hex, str(page), str(baud), dump]
if reset_hex is not None or resume is not None:
# Chips without a hardware boot section - the tinies and the

119
test/pbwalk.py Normal file
View File

@@ -0,0 +1,119 @@
#!/usr/bin/env python3
"""The reset walk region, end to end - the brick this refusal exists for.
BOOTRST programmed on a boot-sectioned mega enters reset at the boot section's
base, and the loader is reached only by walking up across erased flash. An
application that grew into that span is executed by the reset instead: it
happened to an ssd1306 board on a 328P with hfuse 0xdc, and with no reset edge
brought out, an ICE was the only way back.
The plain --flash is the path under test. It asks for no fuses of its own,
which is exactly why it once skipped the check and wrote the image anyway.
The runner still resets into the loader - the walk itself is not modeled. What
is modeled is the fuse read, so the refusal is decided from the bytes silicon
would have answered (-f, test/pureboot_device.cpp).
Usage: pbwalk.py <device_bin> <pureboot_elf> <mcu> <hz> <base_hex> <page>
<baud> <app_bin> <tool_py> <workdir>
"""
import os
import sys
def fail(message):
print(f"FAIL: {message}")
sys.exit(1)
def intel_hex(chunks):
"""Intel HEX for {address: bytes}. An application that reaches the walk
region is sparse there - a .bin would have to carry every byte between."""
lines = []
for address, data in sorted(chunks.items()):
for at in range(0, len(data), 16):
row = data[at : at + 16]
here = address + at
record = bytes((len(row), here >> 8, here & 0xFF, 0)) + row
lines.append(":" + (record + bytes((-sum(record) & 0xFF,))).hex())
return "\n".join(lines + [":00000001FF"]) + "\n"
def main():
device_bin, elf, mcu, hz, base_hex, page, baud, app_bin, tool, workdir = sys.argv[1:]
page, baud = int(page), int(baud)
sys.path.insert(0, os.path.dirname(os.path.abspath(tool)))
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
import pbsim
import pureboot as pb
os.makedirs(workdir, exist_ok=True)
dump = os.path.join(workdir, "dump.bin")
# The ssd1306 board's own fuses, and the same boot section with reset left
# on the application - the one difference that decides the refusal.
bricking, harmless = "ffffffdc", "ffffffdd"
def session(fuses):
return pbsim.Device(device_bin, elf, mcu, hz, base_hex, page, baud, dump, fuses=fuses)
def connected(device):
port = pb.Port(device.pty, baud)
loader = pb.Loader(port)
loader.connect(25)
return port, loader
device = session(bricking)
try:
port, loader = connected(device)
info = loader.info
bootrst, bls_start = pb.mega_boot(info, bytes.fromhex(bricking))
if not bootrst or bls_start >= info.base:
fail(f"this profile cannot brick: BOOTRST {bootrst}, section at {bls_start:#06x}")
port.close()
# An application whose tail lands in the walk region, which is what the
# bricked board's dump showed: live bytes at the section base.
overgrown = os.path.join(workdir, "overgrown.hex")
with open(overgrown, "w") as image:
image.write(intel_hex({0: open(app_bin, "rb").read(), bls_start: bytes(page)}))
out = pbsim.run_tool(tool, device.pty, baud, "--fuses", "--stay")
if f"0x{bricking[6:]}" not in out:
fail(f"the device did not answer the fuses it was given ({bricking}):\n{out}")
try:
pbsim.run_tool(tool, device.pty, baud, "--flash", overgrown, "--stay")
except RuntimeError as refused:
if "walk region" not in str(refused):
fail(f"--flash failed, but not on the walk region: {refused}")
else:
fail("a plain --flash wrote an application into the reset walk region")
port, loader = connected(device)
if loader.read_flash(bls_start, page) != bytes((0xFF,)) * page:
fail(f"the refused image was written to {bls_start:#06x} anyway")
port.close()
# The override is the escape hatch, not the default.
pbsim.run_tool(tool, device.pty, baud, "--flash", overgrown, "--erase-flash", "--force", "--stay")
port, loader = connected(device)
if loader.read_flash(bls_start, page) != bytes(page):
fail(f"--force did not write {bls_start:#06x}")
port.close()
finally:
device.stop()
# An image that stays below the region flashes with no override at all,
# and so does the same overgrown image once reset boots the application.
for fuses, image in ((bricking, app_bin), (harmless, overgrown)):
device = session(fuses)
try:
pbsim.run_tool(tool, device.pty, baud, "--flash", image, "--erase-flash", "--stay")
finally:
device.stop()
print("pbwalk: the walk region is refused by default, overridden by --force, and left alone otherwise")
main()

82
test/pureboot_device.cpp
View File

@@ -18,12 +18,13 @@
// is a silent no-op (the mega's boot section has one, avr_flash). The
// missing module is supplied here: the SPM ioctl reads SPMCSR/Z/r1:r0 and
// implements buffer fill, page erase, page write, and CTPB, completing
// instantly. RFLB's LPM diversion (fuse readout) stays unmodeled, so the
// 'F' command answers with flash bytes - the tests assert transport only.
// instantly. The fuse readout's LPM diversion is missing from every core and
// is supplied too, so a fuse read answers fuses (-f) and not flash bytes.
//
// On exit (or SIGTERM) the flash and EEPROM are dumped to files for a
// ground-truth cross-check against what the host read back.
#include <array>
#include <charconv>
#include <csignal>
#include <cstdint>
#include <cstdio>
@@ -177,6 +178,58 @@ void fix_mega_flash_erase()
std::println(stderr, "device: no flash module to fix - SPM page erases may misalign");
}
// --------------------------------------------------------------- fuses ---
// The fuse and lock bytes answer an LPM, not an SPM: BLBSET|SELFPRGEN in
// SPMCSR diverts the next LPM to them, selected by Z (Atmel-8271 section
// 26.8.9). simavr executes LPM straight out of avr->flash and offers no hook
// on it, so the diversion is modeled where there is one - the SPMCSR write -
// by lending the four flash bytes Z can select to the fuses for as long as the
// hardware holds SELFPRGEN. Unprogrammed until -f says otherwise, as a part
// ships and as the erased flash above is.
auto fuses = std::to_array<std::uint8_t>({0xFF, 0xFF, 0xFF, 0xFF}); // Z order: low, lock, extended, high
decltype(fuses) lent;
int parse_fuses(std::string_view spec)
{
if (spec.size() != 2 * fuses.size()) {
return -1;
}
for (std::size_t at = 0; at < fuses.size(); at++) {
const auto digits = spec.substr(2 * at, 2);
if (std::from_chars(digits.data(), digits.data() + digits.size(), fuses[at], 16).ec != std::errc{}) {
return -1;
}
}
return 0;
}
avr_cycle_count_t end_fuse_read(avr_t *mcu, avr_cycle_count_t, void *)
{
std::memcpy(mcu->flash, lent.data(), lent.size());
return 0;
}
void spmcsr_written(avr_t *mcu, avr_io_addr_t at, std::uint8_t value, void *)
{
// A registered write handler *replaces* the store simavr would have done
// (sim_core.c), so performing it is this handler's job - on the tinies
// nothing else is watching the register, and swallowing the store would
// leave every SPM command unseen by the NVM model above.
avr_core_watch_write(mcu, at, value);
constexpr std::uint8_t read_fuse_command = 0x09; // BLBSET|SELFPRGEN
// The LPM must follow within three cycles of the arming store (Atmel-8271
// section 26.8.9), and simavr runs its cycle timers between instructions -
// so the loan is returned after the LPM that took it, never during.
constexpr avr_cycle_count_t selfprgen_window = 3;
if ((value & 0x1F) != read_fuse_command) {
return;
}
std::memcpy(lent.data(), mcu->flash, lent.size());
std::memcpy(mcu->flash, fuses.data(), fuses.size());
avr_cycle_timer_register(mcu, selfprgen_window, end_fuse_read, nullptr);
}
void request_reset(int)
{
reset_requested = 1;
@@ -184,6 +237,9 @@ void request_reset(int)
// ------------------------------------------------------------- tiny NVM ---
// Where SPMCSR sits on the cores that carry no flash module to name it.
constexpr avr_io_addr_t tiny_spmcsr = 0x57;
struct tiny_nvm_t {
avr_io_t io;
std::array<std::uint8_t, 128> buffer;
@@ -200,7 +256,7 @@ int nvm_ioctl(avr_io_t *io, std::uint32_t ctl, void *)
}
auto *n = reinterpret_cast<tiny_nvm_t *>(io);
avr_t *mcu = io->avr;
std::uint8_t command = mcu->data[0x57] & 0x1f; // SPMCSR, both tinies
std::uint8_t command = mcu->data[tiny_spmcsr] & 0x1f;
auto z = static_cast<std::uint16_t>(mcu->data[30] | (mcu->data[31] << 8));
std::uint32_t page_base = static_cast<std::uint32_t>(z & ~(n->page - 1)) % (mcu->flashend + 1);
if (command == 0x01) { // SPMEN alone: buffer fill from r1:r0
@@ -222,7 +278,7 @@ int nvm_ioctl(avr_io_t *io, std::uint32_t ctl, void *)
std::memset(n->buffer.data(), 0xff, n->page);
std::memset(n->used.data(), 0, n->page);
}
mcu->data[0x57] &= static_cast<std::uint8_t>(~0x1f); // the operation completes instantly
mcu->data[tiny_spmcsr] &= static_cast<std::uint8_t>(~0x1f); // the operation completes instantly
return 0;
}
@@ -494,11 +550,18 @@ void poll_pty()
int main(int argc, char *argv[])
{
bool link_given = false;
for (int opt; (opt = getopt(argc, argv, "l:w")) != -1;) {
for (int opt; (opt = getopt(argc, argv, "l:wf:")) != -1;) {
if (opt == 'w') {
window_report = true;
continue;
}
if (opt == 'f') {
if (parse_fuses(optarg) != 0) {
std::println(stderr, "device: -f takes 8 hex digits: low,lock,extended,high");
return 2;
}
continue;
}
if (opt != 'l' || parse_link(optarg) != 0) {
std::println(stderr, "device: bad link spec (usart0, usart1, sw, or sw:B0,B1 as RX,TX)");
return 2;
@@ -508,10 +571,12 @@ int main(int argc, char *argv[])
int args = argc - optind;
if (args < 7 || args > 9) {
std::print(stderr,
"usage: {} [-l link] [-w] <pureboot.elf> <mcu> <hz> <base_hex> <page> <baud> <flash_dump>"
" [reset_hex] [resume_flash]\n"
"usage: {} [-l link] [-w] [-f fuses] <pureboot.elf> <mcu> <hz> <base_hex> <page> <baud>"
" <flash_dump> [reset_hex] [resume_flash]\n"
" -l link: usart0 | usart1 | sw[:B0,B1[@0]] (RX,TX, then the USART owning\n"
" them); default: the chip's own\n"
" -f fuses: 8 hex digits, low,lock,extended,high - what a fuse read answers;\n"
" default: unprogrammed. reset_hex stays the reset target\n"
" -w: print PB_WINDOW_TX <cycle> at the first transmit activity and\n"
" free-run idle time (window measurement mode)\n"
" reset_hex: reset vector (default: base with a boot section, else 0)\n"
@@ -599,6 +664,9 @@ int main(int argc, char *argv[])
nvm.io.ioctl = nvm_ioctl;
avr_register_io(avr, &nvm.io);
}
// Where the fuse read is armed: the mega's flash module names its SPMCSR,
// and the tinies keep theirs at the address both those cores share.
avr_register_io_write(avr, mega_flash ? mega_flash->r_spm : tiny_spmcsr, spmcsr_written, nullptr);
if (!link_software) {
// POLL_SLEEP paces an idle-polling loader in host real time (a

9
test/test_planner.py
View File

@@ -243,7 +243,14 @@ def main():
pb.check_walk_region(deep, mega, fuses(0xFA), True)
pb.check_walk_region(deep, mega, fuses(0xFB), False) # BOOTRST unprogrammed
pb.check_walk_region({0x7800: bytes((0xFF,)) * 128}, mega, fuses(0xFA), False)
pb.check_walk_region(deep, mega, None, False) # fuses unknown: no check
# Fuses that could not be read leave the span unknown, and unknown is not
# empty: the check that did not run must refuse rather than pass, since the
# span it would have named is the one that costs the part.
expect_error("walk region without fuses",
lambda: pb.check_walk_region(deep, mega, None, False), "--assume-fuses", "--force")
pb.check_walk_region(deep, mega, None, True)
pb.check_walk_region(deep, tiny, None, False) # patched reset vector: no walk to guard
# The repairing verify: a mismatched page is rewritten rather than raised,
# bounded so a fault that is not self-clearing cannot spin.