When OpenSBI is built with a relatively new compiler (gcc-13 and greater)
I observed that GDB is unable to produce proper backtraces and some
variable values appear corrupted (even if the associated DWARF
location descriptor is correct).
Turns out that to properly work with debug information, debuggers often
need to unwind the stack. They generally rely on Call Frame Information
(CFI) records provided by the compiler to facilitate this task.
Currently, the GCC compiler offers two mechanisms:
- `.debug_frame` section (as described in the DWARF specification).
- `.eh_frame` sections (as described in LSB documents).
The latter (`.eh_frame`) supports stack unwinding at runtime, providing
a framework for C++ exceptions or enabling backtrace generation using
libraries like libunwind. However, the downside of this approach is that
these sections should be part of loadable segments.
The former (`.debug_frame`) is simply an ordinary debug section.
Starting from GCC 13, Linux targets enable the `-fasynchronous-unwind-tables`
and `-funwind-tables` flags by default. Relevant commit:
https://github.com/gcc-mirror/gcc/commit/3cd08f7168
When these flags are active, the compiler generates `.eh_frame` sections
instead of `.debug_frame`. Since OpenSBI is built using the **Linux
toolchain**, this behavior applies to OpenSBI as well.
The problem arises because the SBI build system uses `-Wl,--gc-sections`,
which discards the `.eh_frame` section.
Possible Fixes:
1. Enforce `.debug_frame` generation – Modify compiler flags to generate
`.debug_frame` instead of `.eh_frame`.
2. Preserve `.eh_frame` in the linker script – Add `KEEP(*(.eh_frame))`
to ensure the section is not discarded.
I chose Option 1 because it avoids any runtime overhead.
Signed-off-by: Parshintsev Anatoly <anatoly.parshintsev@syntacore.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
Link: https://lore.kernel.org/r/20250421124729.36364-1-anatoly.parshintsev@syntacore.com
Signed-off-by: Anup Patel <anup@brainfault.org>
Recursively expanded variables (defined with '=') are expanded at
evaluation time. These version information variables are evaluated
inside a recipe as part of GENFLAGS. As a result, the shell commands
are executed separately for each compiler invocation. Convert the
version information variables to be simply expanded, so the shell
commands are executed only once, at Makefile evaluation time. This
speeds up the build by as much as 75%.
A separate check is needed to maintain the behavior of preferring the
value of OPENSBI_BUILD_TIME_STAMP from the environment.
Signed-off-by: Samuel Holland <samuel.holland@sifive.com>
Reviewed-by: Xiang W <wxjstz@126.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
Link: https://lore.kernel.org/r/20250313035755.3796610-1-samuel.holland@sifive.com
Signed-off-by: Anup Patel <anup@brainfault.org>
OpenSBI is compiled with -fPIE, which generally implies dynamic linking.
This causes the compiler to generate GOT references for global symbols
in order to support runtime symbol interposition. However, OpenSBI does
not actually perform dynamic linking, so the GOT indirection just adds
unnecessary overhead.
The GOT references can be avoided by declaring global symbols with
hidden visibility, thus making them local to this dynamic object and
non-interposable. GCC/Clang's -fvisibility parameter is insufficient for
this purpose when referencing objects from other translation units;
either __attribute__((visibility(...)) or the pragma is required. Use
the pragma since it is easier to apply to every symbol. Additionally
clean up the one GOT reference from inline assembly.
With this change, a firmware linked with LLD does not contain either a
GOT or a PLT, and a firmware linked with BFD ld contains only a GOT with
a single (unreferenced, legacy) _GLOBAL_OFFSET_TABLE_ entry.
Signed-off-by: Samuel Holland <samuel.holland@sifive.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
If the script fails, we end up trying to build either
an empty or damaged .c file. Just remove it and let
gcc fail on non-existent file.
Signed-off-by: Ben Dooks <ben.dooks@codethink.co.uk>
Reviewed-by: Anup Patel <anup@brainfault.org>
Enabling V-extension using -march option causes OpenSBI boot-time
hang with LLVM compiler.
As a work-around, don't enable V-extension using -march option and
instead use a custom OpenSBI specific define inform availability of
V-extension to lib/sbi/sbi_trap_v_ldst.c.
Fixes: c2acc5e5b0 ("lib: sbi_misaligned_ldst: Add handling of vector load/store")
Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Samuel Holland <samuel.holland@sifive.com>
Add misaligned load/store handling for the vector extension
to the sbi_misaligned_ldst library.
This implementation is inspired from the misaligned_vec_ldst
implementation in the riscv-pk project.
Co-developed-by: Zong Li <zong.li@sifive.com>
Signed-off-by: Zong Li <zong.li@sifive.com>
Signed-off-by: Nylon Chen <nylon.chen@sifive.com>
Reviewed-by: Andy Chiu <andy.chiu@sifive.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
grep -e "-mstrict-align\|-mno-unaligned-access" makes use of GNU grep's
backslash-escaped alternation operator \| which is available in basic
regular expression syntax (BRE) mode.
However, in POSIX grep's BRE mode | is an ordinary character which, when
backslash-escaped, matches itself. Therefore, the search pattern becomes
a plain string '-mstrict-align|-mno-unaligned-access' which obviously
never matches the expected error and CC_SUPPORT_STRICT_ALIGN is always set
to y.
When cross-compiling with LLVM on amd64-unknown-openbsd7.6 host for
riscv64-unknown-elf target this results in a compilation error:
clang: error: unsupported option '-mno-unaligned-access' for target
'riscv64-unknown-elf'
Using multiple -e options for this case maintains consistent behaviour
across different grep implementations and fixes the issue.
Signed-off-by: Igor Melnikov <imel@purelymail.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
The --gc-sections option enables the linker to perform garbage
collection of unreferenced code and data, thereby reducing the binary
size.
The -ffunction-sections option will place each function into a separate
section, so it is necessary to add .text.* to the linker script.
Signed-off-by: Kele Zhang <zhangcola2003@gmail.com>
Signed-off-by: Yuan Tan <tanyuan@tinylab.org>
Signed-off-by: Zhangjin Wu <falcon@tinylab.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Xiang W <wxjstz@126.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
Force carray C source files to be regenerated when the script changes,
since their contents depend on the script's output.
Signed-off-by: Samuel Holland <samuel.holland@sifive.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
When building OpenSBI via a submodule, OPENSBI_VERSION_GIT can be left
unset in case '.git' isn't a dir. This is the case when building OpenSBI
as a QEMU submodule:
$ cat .git
gitdir: ../../.git/modules/roms/opensbi
As a result, building OpenSBI tag v1.5.1 in QEMU will result in a binary
that will have "OpenSBI v1.5" as a banner.
Use "git rev-parse --git-dir" instead of checking if '.git' is a dir to
detect if the current dir is a git repo.
Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
The rule included from auto.conf.cmd adds a dependency on every Kconfig
file, so these two Kconfig files do not need to be specified again here.
Signed-off-by: Samuel Holland <samuel.holland@sifive.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
Tested-by: Anup Patel <anup@brainfault.org>
The .config file may be manually edited or copied from another location.
Since genconfig.py is responsible for generating auto.conf (the Makefile
fragment) and autoconf.h (the C header) from .config, it must be run any
time .config changes, not just when running menuconfig.
Fixes: 662e631cce ("Makefile: Add initial kconfig support for each platform")
Signed-off-by: Samuel Holland <samuel.holland@sifive.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
Tested-by: Anup Patel <anup@brainfault.org>
While writing to the dynsym is futile, the --exclude-libs options is not
recognized by all linkers (e.g. riscv64-elf-ld.bfd).
Signed-off-by: Leon M. Busch-George <leon@georgemail.eu>
Reviewed-by: Anup Patel <anup@brainfault.org>
Support for that option will be added in LLVM 18:
23ce536840
Clang 17.0.6, however, will error when passed the
`-mstrict-align` flag.
We should only use the flag if it is supported.
Signed-off-by: Kalle Wachsmuth <kalle.wachsmuth@gmail.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
Reviewed-by: Xiang W <wxjstz@126.com>
Since everything is statically linked, we don't need to expose symbols
for dynamic linking.
For a default build this saves about 2 KiB of useless read only data in
.dynsym, .dynstr, .hash, .gnu.hash sections.
Signed-off-by: Vivian Wang <dramforever@live.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
grep (at least my version, grep-3.8-3.fc38.x86_64) warns with
"grep: warning: stray \ before -". Fix the warning by making
the command line input to grep less ambiguous.
Signed-off-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
Adds the `-L` flag (follow symlinks) to the `cp` commands used to
install `libsbi.a` and `include/sbi/*`.
This should make no difference in regular compilation. However,
it does make a difference when compiling with bazel. Namely,
bazel's sandboxing will turn all the source files into symlinks.
After installation with `cp` the destination files will be
symlinks pointing to the sandbox symlinks. As the sandbox files
are removed when compilation ends, the just-copied symlinks
become dangling symlinks.
The resulting include files will be
unusable due to the dangling symlink issues. Adding `-L` when
copying ensures that the files obtained by executing the `install`
targets are always dereferenced to files, rather than symlinks,
eliminating this issue.
Signed-off-by: Filip Filmar <fmil@google.com>
Reviewed-by: Xiang W <wxjstz@126.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
Add a new make command line option "make DEBUG=1" to prevent compiler
optimizations using -O2.
Signed-off-by: Bin Meng <bmeng@tinylab.org>
Reviewed-by: Xiang W <wxjstz@126.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
Since we don't currently create these, changes to fw_base.ldS do not
cause the preprocessed fw_*.elf.ld files to be rebuilt, and thus
incremental builds can end up failing with missing symbols if crossing
the recent commits that introduced _fw_rw_offset and then replaced it
with _fw_rw_start.
Reported-by: Ben Dooks <ben.dooks@sifive.com>
Signed-off-by: Jessica Clarke <jrtc27@jrtc27.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
-N option coalesce all sections into single LOAD segment which causes
data and other sections to have executable permission causing warning
with new binutils ld 2.39.
New ld emits warning when any segment have all three permissions RWX.
ld.bfd: warning: test.elf has a LOAD segment with RWX permissions
ld.bfd: warning: fw_dynamic.elf has a LOAD segment with RWX permissions
ld.bfd: warning: fw_jump.elf has a LOAD segment with RWX permissions
ld.bfd: warning: fw_payload.elf has a LOAD segment with RWX permissions
This option was added in below commit -
commit: eeab92f242 ("Makefile: Convert to a more standard format")
Removing -N option allows to have text and rodata into one LOAD
segment and other sections into separate LOAD segment which prevents
RWX permissions on single LOAD segment. Here X == E
Current
LOAD 0x0000000000000120 0x0000000080000000 0x0000000080000000
0x000000000001d4d0 0x0000000000032ed8 RWE 0x10
-N removed
LOAD 0x0000000000001000 0x0000000080000000 0x0000000080000000
0x00000000000198cc 0x00000000000198cc R E 0x1000
LOAD 0x000000000001b000 0x000000008001a000 0x000000008001a000
0x00000000000034d0 0x0000000000018ed8 RW 0x1000
Signed-off-by: Rahul Pathak <rpathak@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
Tested-by: Samuel Holland <samuel@sholland.org>
If the path where this repo is located contains the platform name on
it, the original Makefile replaced its occurrences from the path making
it an invalid path. This commit prevents this behavior replacing only
the last part of the path as intended.
Signed-off-by: Alejandro Cabrera Aldaya <aldaya@gmail.com>
Reviewed-by: Xiang W <wxjstz@126.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
I don't know why but `echo -n` didn't work for me. macOS supports
the `-n` option but it doesn't work in the makefile. What it does
instead is it literally writes `-n` to the file and then also
leaves a newline at the end.
I'm using GNU Make 4.4 (`gmake` from Homebrew).
Signed-off-by: KaDiWa <kalle.wachsmuth@gmail.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
Add back the missing rules needed to build carray files in lib/sbi. This
allows future usage of carray in lib/sbi.
Fixes: de80e9337d ("Makefile: Compile lib/utils sources separately for each platform")
Signed-off-by: Vivian Wang <dramforever@live.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
Add cscope support so that running `make cscope` will generate/update
cscope files used for source code browsing, while running `make
distclean` will remove the cscope files.
Also add entry in .gitignore to ignore generated cscope files.
Signed-off-by: Tan En De <ende.tan@linux.starfivetech.com>
Reviewed-by: Xiang W <wxjstz@126.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Currently, if same build directory is used to compile two different
platforms then lib/utils objects are shared for these platforms.
We will be having platform specific configs to enable/disable drivers
in lib/utils and select compile time options for lib/utils sources.
This means lib/utils sources will now be compiled in a platform
specific way.
To tackle above, we update top-level Makefile as follows:
1) Don't create libsbiutils.a anymore because this can't be shared
between platforms.
2) Compile lib/utils sources separately for each platform.
3) Add comments showing which make rules are for lib/sbi, lib/utils,
firmware, and platform sources.
Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Tested-by: Andrew Jones <ajones@ventanamicro.com>
Acked-by: Atish Patra <atishp@rivosinc.com>
Tested-by: Atish Patra <atishp@rivosinc.com>
We extend the top-level makefile to allow kconfig based configuration
for each platform where each platform has it's own set of configs with
"defconfig" being the default config.
Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Tested-by: Andrew Jones <ajones@ventanamicro.com>
Acked-by: Atish Patra <atishp@rivosinc.com>
Tested-by: Atish Patra <atishp@rivosinc.com>
Generating C array at compile time based on details provided by
objects.mk is a very useful feature which will help us compile
only a subset of drivers or modules.
We add a bash script (carray.sh) which takes array details and
object/variable list from command-line to generate a C source
containing array of object/variable pointers. We also extend
top-level makefile to use carray.sh whenever specified through
objects.mk.
Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>
The generated C source could be anywhere within build directory so
let us update the make rule to comple generated C source accordingly.
Signed-off-by: Anup Patel <apatel@ventanamicro.com>
Reviewed-by: Atish Patra <atishp@rivosinc.com>
From version 2.38, binutils default to ISA spec version 20191213. This
means that the csr read/write (csrr*/csrw*) instructions and fence.i
instruction has separated from the `I` extension, become two standalone
extensions: Zicsr and Zifencei. As the kernel uses those instruction,
this causes the following build failure:
CC lib/sbi/sbi_tlb.o
<<BUILDDIR>>/lib/sbi/sbi_tlb.c: Assembler messages:
<<BUILDDIR>>/lib/sbi/sbi_tlb.c:190: Error: unrecognized opcode `fence.i'
make: *** [Makefile:431: <<BUILDDIR>>/build/lib/sbi/sbi_tlb.o] Error 1
The fix is to specify those extensions explicitly in -march. However as
older binutils version do not support this, we first need to detect
that.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Alexandre Ghiti <alexandre.ghiti@canonical.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
The commit 69d7e53 disables the -m(no-)save-restore option for
clang, but clang11 supports this option. This patch uses the
output information of the compiler to check whether the compiler
supports this option.
Signed-off-by: Xiang W <wxjstz@126.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>
The riscv target of CLANG-10 (or lower) does not support the
-m(no-)save-restore option so we get compile warnings. This patch
fixes compile warning by using -m(no-)save-restore option only
for GCC.
Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Dong Du <Dd_nirvana@sjtu.edu.cn>
When we are doing opensbi development, we want to know the build time
and compiler info for debug purpose.
To enable this message, please add "BUILD_INFO=y", like:
```
make BUILD_INFO=y
```
NOTE: Using `BUILD_INFO=y` without specifying SOURCE_DATE_EPOCH will
violate "reproducible builds". So it's ONLY for development and debug
purpose, and should NOT be used in a product which follows "reproducible
builds".
Signed-off-by: Wei Fu <wefu@redhat.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
When generating code with -mno-relax, GCC puts .option norelax in
the generated assembly, and so doesn’t bother passing on -mno-relax
to the assembler. This has the unfortunate effect that, when using
GCC to assemble hand-written assembly, -mno-relax does nothing,
and we have to pass -Wa,-mno-relax to manually forward it to the
assembler.
This is an old GCC bug that was fixed [1] recently. For the time
being, let's pass "-Wa,-mno-relax" to ASFLAGS for the GCC + LLD
combination to work, e.g.:
$ make CC=riscv64-unknown-elf-gcc LLVM=1 PLATFORM=generic
[1] 3b0a7d624e
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>
We only need libgcc for 64-bit division on RV32. Whilst GCC toolchains
bundle libgcc, Clang toolchains tend not to ship libclang_rt.builtins
given every compiler is a cross-compiler for every target and so you
would need a silly number of builds of it, with only the native library
available; only vendor-provided Clang toolchains specifically for bare
metal cross-compiling are likely to provide it.
Thus, import part of FreeBSD's implementation of the division support
functions needed and stop linking against libgcc.
Signed-off-by: Jessica Clarke <jrtc27@jrtc27.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>
Tested-by: Anup Patel <anup.patel@wdc.com>
This is intended to mirror the Linux kernel. Building with CC=clang will
use Clang as the compiler but default to using the existing binutils.
Building with LLVM=1 will default to using Clang and LLVM binutils.
Whilst GCC will accept the -N linker option and forward it on to the
linker, Clang will not, and so in order to support both compilers we
must use -Wl, to forward it to the linker as is required for most other
linker options.
Note that there is currently a bug when using Clang as the compiler and
riscv64-linux-gnu-ld as the linker for FW_PIC=y. At first glance this
appears to be a bug in GNU binutils, but this could also be Clang or
OpenSBI at fault in some subtle way. Thus, for now, advise that this
combination be avoided.
Signed-off-by: Jessica Clarke <jrtc27@jrtc27.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>
Tested-by: Anup Patel <anup.patel@wdc.com>
Bare-metal GNU ld does not support PIE, so if using it this will result
in a failure to build. Instead, default to FW_PIC=n if not supported.
Note that an explicit FW_PIC=y is not overridden, to ensure the build
fails rather than silently producing a position-dependent binary.
Signed-off-by: Jessica Clarke <jrtc27@jrtc27.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>
GCC has already a predefined macro __ASSEMBLER__ therefore, it can be
used without the need to define a new flag with -D__ASSEMBLY__.
This is useful when adding the library to projects having a build
system such one can build without the need to make changes.
THe build system does not use the Makefile in the sources tree.
Signed-off-by: Marouene Boubakri <marouene.boubakri@nxp.com>
Signed-off-by: Anup Patel <anup.patel@wdc.com>
Enable OpenSBI to support position independent execution. Because the
position independent code will cause an additional GOT reference when
accessing the global variables, it will reduce performance a bit. Therefore,
the position independent execution is disabled by default. Users can
through specifying "FW_PIC=y" on the make command to enable this feature.
In theory, after enabling position-independent execution, the OpenSBI
can run at arbitrary address with appropriate alignment. Therefore, the
original relocation mechanism will be skipped. In other words, OpenSBI will
directly run at the load address without any code movement.
Signed-off-by: Vincent Chen <vincent.chen@sifive.com>
Reviewed-by: Anup Patel <anup.patel@wdc.com>
When PLATFORM_RISCV_ABI and PLATFORM_RISCV_ISA are not specified,
we force "-mabi=lp64 -march=rv64gc" for RV64 and force "-mabi=ilp32
-march=rv32gc" for RV32. This can prevent users from using the
toolchain default "-mabi" and "-march" options.
To allow using toolchain defaults, we add compile-time option
PLATFORM_RISCV_TOOLCHAIN_DEFAULT which when enabled forces the
top-level makefile to use toolchain default ABI and ISA string.
To enable the option, pass "PLATFORM_RISCV_TOOLCHAIN_DEFAULT=1"
to top-level make.
Reported-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
We extend d2c.sh to allow padding zeros in output C source when
converting DTB to C source. Using this feature, platforms can
create extra room for in-place FDT fixups on built-in DTBs.
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>
Currently, the make rules for ELF, BIN and LD script are little
generic to allowing ELF, BIN and LD script to be anywhere in the
build directory. For OpenSBI firmwares, the ELF, BIN, and LD script
are always platform specific so we update make rules accordingly.
Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>
Now that no platform is using FW_PAYLOAD_FDT mechanism, we
remove related code from Makefile and related documentation.
Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>
