The download link is dead: bitbucket.org/nuttx/buildroot is gone, and the
buildroot that still carries ldnxflat is github.com/patacongo/buildroot.
The instructions were also more than is needed. mknxflat came in tree with
PR #19600, so only ldnxflat has to be built, and an ordinary arm-none-eabi
GCC compiles and links NXFLAT modules: a board does not have to select
CONFIG_ARM_TOOLCHAIN_BUILDROOT to use them. What ldnxflat does need is a
binutils source and build tree, because it reads its input through libbfd.
The CI test list said mknxflat is what the container lacks. It is in tree
now; ldnxflat is the one that is missing.
Assisted-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Signed-off-by: Marco Casaroli <marco.casaroli@gmail.com>
mknxflat is built from tools/nxflat by the NuttX build itself, so the
toolchain section no longer sends the reader to buildroot for it; only
ldnxflat still comes from there. Bring the mknxflat usage text in line
with the tool, note that MKNXFLAT and LDNXFLAT are supplied by the ARM
Toolchain.defs, and correct the r10 references left in the PIC
descriptions.
Describe the module ABI marker, so that a user whose prebuilt module
starts failing exec() with ENOEXEC finds out that the loader refuses a
module whose import table does not name __nxflat_abi_v2, and that
rebuilding the module is the fix.
Assisted-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Signed-off-by: Marco Casaroli <marco.casaroli@gmail.com>
ARM PIC has used r10 as the base register, but the tree has never been
consistent about it. Toolchain.defs gives CONFIG_BUILD_PIC
-mpic-register=r9 and CONFIG_PIC -mpic-register=r10, twenty-five lines
apart, and arm_initialstate.c sets REG_R9 from inline assembly under one
and REG_PIC under the other, with a comment reading "Set the PIC base
register (probably R10)". This settles it on r9 for all of PIC: NXFLAT,
ELF PIC and CONFIG_BUILD_PIC alike.
r9 is the right choice rather than an arbitrary one. It is the AAPCS
platform register, the "static base", and it is what GCC itself picks
for -msingle-pic-base on an EABI target; r10 is the non-EABI default.
It also removes a combination that cannot build today. Stack checking
adds -ffixed-r10 in armv7-m/Toolchain.defs and armv8-m/Toolchain.defs,
while CONFIG_PIC adds -mpic-register=r10, and GCC rejects the pair with
"unable to use 'r10' for PIC register". The comment above REG_PIC has
always said the register "can be R9 if stack checking is enabled", but
the definition was unconditionally REG_R10, so it would have named the
wrong register even had the build succeeded.
The thunk generator moves with the firmware. NXFLAT import stubs had
the register baked in as "add ip,ip,sl", so a module built for r9 would
load and then branch to a wild address on its first call out. The stubs
now come from NXFLAT_PIC_REG in the in-tree tool, which is built only
when CONFIG_NXFLAT is set, following the
CONFIG_BOARD_ETC_ROMFS_PASSWD_ENABLE precedent in tools/Unix.mk.
That leaves modules built before this change, and they are the reason
for the ABI marker. The NXFLAT header cannot carry a version: h_magic
is written by ldnxflat, which is GPL, derived from elf2flt, and stays
out of this repository, so it can never be changed in step with the
loader. The import table can, because both of its ends are in-tree --
mknxflat emits it and nxflat_bindimports() reads it -- and ldnxflat
passes it through untouched. So every module now imports
__nxflat_abi_v2, the base firmware defines it, and a module that does
not import it is refused.
Making the marker a real exported symbol rather than a name the loader
special-cases is what keeps it out of the build system's way: a board's
symbol table picks it up exactly as it picks up printf, so mksymtab.sh
and its equivalents need no change. It also gives the reverse direction
a diagnosis for free -- a module built against a newer ABI than its
firmware fails with "Exported symbol __nxflat_abi_v2 not found".
Most of the remaining churn is boards restating a default. ARCHPICFLAGS
is a "?=" default so that a board only speaks up when it differs, and
twenty-six were assigning the value the default already had. MKNXFLAT
gets the same treatment: thirteen boards named the same tool, and the
only thing that varies is ARM versus Thumb-2, which falls out of
CONFIG_ARM_THUMB. LDNXFLAT gains a default too -- it stays an
out-of-tree PATH lookup, but naming it centrally fixes boards that never
assigned it, where it expanded to nothing and handed make a recipe
beginning "-e", whose leading dash make ate as "ignore errors".
The non-ARM boards carrying -mpic-register=r10 lose it: it is an
ARM-only option, reachable only through CPICFLAGS, which is only used to
build NXFLAT modules, and no non-ARM board enables NXFLAT.
Boards keep nothing about PIC flags any more. ARCHPICFLAGS was set by
sixty-three of them and only ever fed CPICFLAGS, which is only used to
build NXFLAT modules; no board outside arch/arm enables NXFLAT, so every
non-ARM copy was setting a variable nothing read. Those are removed
rather than moved somewhere more central, which would only make dead
text look load-bearing. LDNXFLAT goes the same way as MKNXFLAT, for the
same reason: thirteen boards named the same tool that Toolchain.defs now
names once.
One of them was not merely redundant. am67/t3-gem-o1 asked for
"-mpic-register=r10 -ffixed-r10", which GCC refuses outright with
"unable to use 'r10' for PIC register" -- the very combination the
filter-out machinery in Toolchain.defs exists to prevent. It has
survived because that board does not build NXFLAT modules, so the flags
are never handed to a compiler. Renaming the register would have
carried the fault forward unchanged, so the line goes.
Tested on lm3s6965-ek:qemu-nxflat under QEMU, configured and built with
no overrides. The nxflat example runs the errno, hello and struct
modules with output identical to the same config built from master.
Built with the old out-of-tree thunk generator instead, the same
firmware refuses all three with ENOEXEC rather than locking up in a
HardFault, which is what this change is for. mps3-an547:picostest,
which is CONFIG_PIC without CONFIG_NXFLAT, builds clean and does not
build the thunk generator.
The .def files pick up two cosmetic changes here alongside the register:
a "Dyanamic" typo that codespell rejects, and a reworded comment in each
thunk_*.c. Neither appears in the emitted thunk -- both are in C
comments -- so the generated text is still what the upstream tool
produces, modulo the register itself.
BREAKING CHANGE: ARM PIC moves from r10 to r9. An NXFLAT module built
before this change has r10 baked into its import stubs and will not run
against a firmware carrying it; the two cannot be mixed. The module is
refused with ENOEXEC rather than branching to a wild address, by way of the
__nxflat_abi_v2 marker described below.
Quick fix: rebuild the module against this tree. Its source needs no
change. A board that reserved r10 by hand, or that assigned ARCHPICFLAGS
or MKNXFLAT to restate a default, should drop those assignments; nothing
else is affected, and CONFIG_PIC without CONFIG_NXFLAT needs no action.
Assisted-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Signed-off-by: Marco Casaroli <marco.casaroli@gmail.com>
Describe the write-once usage model, the strict in-place mmap and what
MAP_XIP_STRICT is for, extent pinning, manual defragmentation and how to read
its result, the power-loss ordering, the on-media layout, and the
limitations.
The NXFLAT page said ROMFS was the only file system able to serve the XIP
mappings its loader needs. That is now one of two, so point at both, and at
what the writable one adds: a module can arrive at run time instead of being
baked into a host-built image.
Assisted-by: Claude Code:claude-opus-5
Signed-off-by: Marco Casaroli <marco.casaroli@gmail.com>
Toolchain.defs adds --fixed-r10 to CFLAGS under CONFIG_PIC, but nearly
every board Make.defs includes that file and then assigns
CFLAGS := $(ARCHCFLAGS) $(ARCHOPTIMIZATION) $(ARCHCPUFLAGS) ...
with ':=', which discards it. 266 of the 269 ARM board files assign
CFLAGS that way; the remaining three delegate to a shared makefile that
does the same thing.
The flag is what keeps the base firmware from allocating r10, the
register a PIC module reaches its own data through. Losing it is silent
and the symptom is remote from the cause: the build succeeds, and only a
callback from base firmware into module code -- qsort() with a module
comparison function is the standard case -- misbehaves, reading its data
through a register the firmware has since felt free to reuse.
Adding it to ARCHCFLAGS puts it on the far side of that ':=', which
re-expands ARCHCFLAGS, so every board picks it up with no board changes
at all.
A module is the other side of the --fixed-r10 contract: it gets r10 via
-mpic-register=r10, and GCC rejects both on one command line with
"unable to use 'r10' for PIC register". Every ARM board carried the
same three lines to set that up:
ARCHPICFLAGS = -fpic -msingle-pic-base -mpic-register=r10
CPICFLAGS = $(ARCHPICFLAGS) $(CFLAGS)
CXXPICFLAGS = $(ARCHPICFLAGS) $(CXXFLAGS)
They move to arch/arm/src/common/Toolchain.defs, which also filters
--fixed-r10 back out of CPICFLAGS, CXXPICFLAGS, CELFFLAGS and
CXXELFFLAGS in one place instead of each board having to know about the
interaction. ARCHPICFLAGS uses '?=' and the derived flags use deferred
'=', so a board can still override or append after including the file,
and CFLAGS is whatever the board finally set it to.
Two boards keep a definition because they genuinely differ: am67 adds
-ffixed-r10 and tiva conditionally adds -mno-pic-data-is-text-relative.
tlsr82 previously appended -fpic to an unset variable, so only -fpic
ever reached its compiler; it now inherits the standard set, verified
against tc32-elf-gcc 4.5.1.tc32-elf-1.5 (Telink TC32 v2.0), whose
ARM-derived backend honors -msingle-pic-base and -mpic-register=r10.
The mps2, mps3, qemu-armv7a, qemu-armv7r, fvp and mcx-nxxx families
referenced ARCHPICFLAGS without ever defining it, so their CPICFLAGS
carried no PIC flags at all; they get the standard set too.
Also adds the missing space in
CXXELFFLAGS = $(CXXFLAGS)-fvisibility=hidden -mlong-calls
which ran the last token of CXXFLAGS into -fvisibility=hidden, yielding
a single malformed token such as -DNDEBUG-fvisibility=hidden.
Also exempts the pre-existing "*.siz" gsize comment in Toolchain.defs
from codespell, which reads "siz" as a misspelling and fires for any
patch touching the file, since checkpatch scans whole files rather than
changed lines.
Before, with CONFIG_PIC=y: CFLAGS has --fixed-r10: NO
After: CFLAGS has --fixed-r10: YES
CPICFLAGS/CELFFLAGS: filtered out
Assisted-by: Claude Code:claude-opus-4-8
Assisted-by: Claude Code:claude-fable-5
Signed-off-by: Marco Casaroli <marco.casaroli@gmail.com>
