RISC-V relocation R_RISCV_HI20 out of range with CONFIG_CMODEL_MEDLOW=y

[nathanchance]

The kernel test robot reported a build failure in the stable tree with a randconfig.

Initially, I thought this was related to CONFIG_MMU, as defconfig + CONFIG_MMU=n reproduces a similar error (also reported in #1409):

$ make -skj"$(nproc)" ARCH=riscv LLVM=1 distclean defconfig

$ scripts/config -d MMU

$ make -skj"$(nproc)" ARCH=riscv LLVM=1 olddefconfig all
ld.lld: error: kernel/built-in.a(kallsyms.o):(function kallsyms_lookup_name: .text+0x42): relocation R_RISCV_PCREL_HI20 out of range: -524432 is not in [-524288, 524287]; references kallsyms_num_syms
>>> referenced by kallsyms.c
>>> defined in kernel/built-in.a(kallsyms.o)

ld.lld: error: kernel/built-in.a(kallsyms.o):(function kallsyms_lookup_name: .text+0x56): relocation R_RISCV_PCREL_HI20 out of range: -524432 is not in [-524288, 524287]; references kallsyms_names
>>> referenced by kallsyms.c
>>> defined in kernel/built-in.a(kallsyms.o)

ld.lld: error: kernel/built-in.a(kallsyms.o):(function kallsyms_lookup_name: .text+0x5e): relocation R_RISCV_PCREL_HI20 out of range: -524432 is not in [-524288, 524287]; references kallsyms_token_index
>>> referenced by kallsyms.c
>>> defined in kernel/built-in.a(kallsyms.o)

ld.lld: error: kernel/built-in.a(kallsyms.o):(function kallsyms_lookup_name: .text+0x66): relocation R_RISCV_PCREL_HI20 out of range: -524432 is not in [-524288, 524287]; references kallsyms_token_table
>>> referenced by kallsyms.c
>>> defined in kernel/built-in.a(kallsyms.o
...

However, that one goes away when CONFIG_KALLSYMS gets disabled. To reproduce this exact build failure on v5.16-rc4, I disabled CONFIG_KALLSYMS and CONFIG_MMU then switched CONFIG_CMODEL_MEDANY for CONFIG_CMODEL_MEDLOW:

$ make -skj"$(nproc)" ARCH=riscv LLVM=1 distclean defconfig

$ scripts/config -d CMODEL_MEDANY -e CMODEL_MEDLOW -d KALLSYMS -d MMU

$ make -skj"$(nproc)" ARCH=riscv LLVM=1 olddefconfig all
ld.lld: error: init/built-in.a(initramfs.o):(function wait_for_initramfs: .text+0x8): relocation R_RISCV_HI20 out of range: 526574 is not in [-524288, 524287]
>>> referenced by initramfs.c

ld.lld: error: init/built-in.a(initramfs.o):(function wait_for_initramfs: .text+0x14): relocation R_RISCV_HI20 out of range: 526366 is not in [-524288, 524287]
>>> referenced by initramfs.c

ld.lld: error: init/built-in.a(initramfs.o):(function wait_for_initramfs: .text+0x26): relocation R_RISCV_HI20 out of range: 526547 is not in [-524288, 524287]
>>> referenced by initramfs.c

ld.lld: error: init/built-in.a(main.o):(function do_one_initcall: .text+0x10): relocation R_RISCV_HI20 out of range: 526574 is not in [-524288, 524287]; references __stack_chk_guard
>>> referenced by main.c
>>> defined in arch/riscv/built-in.a(kernel/process.o)

ld.lld: error: init/built-in.a(main.o):(function do_one_initcall: .text+0x34): relocation R_RISCV_HI20 out of range: 526574 is not in [-524288, 524287]; references initcall_debug
>>> referenced by main.c
>>> defined in init/built-in.a(main.o)
...

[nathanchance]

Per Arnd's recommendation during our weekly meeting, I can reproduce these same errors with GCC 10.2.0 and binutils 2.35.2 from Debian:

$ make -skj"$(nproc)" ARCH=riscv CROSS_COMPILE=riscv64-linux-gnu- distclean defconfig

$ scripts/config -d CMODEL_MEDANY -e CMODEL_MEDLOW -d KALLSYMS -d MMU

$ make -skj"$(nproc)" ARCH=riscv CROSS_COMPILE=riscv64-linux-gnu- olddefconfig all
init/main.o: in function `do_one_initcall':
main.c:(.text+0x12): relocation truncated to fit: R_RISCV_HI20 against symbol `__stack_chk_guard' defined in .sbss section in arch/riscv/kernel/process.o
init/main.o: in function `set_reset_devices':
main.c:(.init.text+0x8): relocation truncated to fit: R_RISCV_HI20 against symbol `reset_devices' defined in .sbss section in init/main.o
init/main.o: in function `run_init_process':
main.c:(.text.unlikely+0x2c): relocation truncated to fit: R_RISCV_HI20 against `.LC9'
init/main.o: in function `rest_init':
main.c:(.ref.text+0x12): relocation truncated to fit: R_RISCV_HI20 against `kernel_init'
init/do_mounts.o: in function `name_to_dev_t':
do_mounts.c:(.text+0x82): relocation truncated to fit: R_RISCV_HI20 against symbol `__stack_chk_guard' defined in .sbss section in arch/riscv/kernel/process.o
init/do_mounts.o: in function `rootwait_setup':
do_mounts.c:(.init.text+0x50): relocation truncated to fit: R_RISCV_HI20 against `root_wait'
init/do_mounts.o: in function `.L40':
do_mounts.c:(.text.unlikely+0x1a): relocation truncated to fit: R_RISCV_HI20 against `.LC1'
init/do_mounts_initrd.o: in function `no_initrd':
do_mounts_initrd.c:(.init.text+0x8): relocation truncated to fit: R_RISCV_HI20 against `.LANCHOR0'
init/initramfs.o: in function `wait_for_initramfs':
initramfs.c:(.text+0x8): relocation truncated to fit: R_RISCV_HI20 against `initramfs_cookie'
init/initramfs.o: in function `error':
initramfs.c:(.init.text+0x4): relocation truncated to fit: R_RISCV_HI20 against `.LANCHOR0'
init/initramfs.o: in function `panic_show_mem':
initramfs.c:(.text.unlikely+0xa): additional relocation overflows omitted from the output

[emojifreak]

Linux 5.18 nommu_qemu_defconfig in riscv64 + KALLSYM_ALL causes this error in ld.llld-14 while gcc-11+binutils 2.38 do not cause this error. Specifically,

xzcat linux-5.18.tar.xz | tar xf -
cd linux-5.18
make ARCH=riscv CROSS_COMPILE=riscv64-linux-gnu- nommu_virt_defconfig
cat >>.config <<EOF
CONFIG_KALLSYMS=y
CONFIG_KALLSYMS_ALL=y
CONFIG_KALLSYMS_BASE_RELATIVE=y
EOF
make ARCH=riscv CROSS_COMPILE=riscv64-linux-gnu- V=1 W=0 >error.txt 2>&1 &
tail -f error.txt

The above script does not cause errors with

ryutaroh@bookworm-amd64:~/nommu/gcctest$ riscv64-linux-gnu-ld --version
GNU ld (GNU Binutils for Debian) 2.38
Copyright (C) 2022 Free Software Foundation, Inc.
This program is free software; you may redistribute it under the terms of
the GNU General Public License version 3 or (at your option) a later version.
This program has absolutely no warranty.
ryutaroh@bookworm-amd64:~/nommu/gcctest$ riscv64-linux-gnu-gcc --version
riscv64-linux-gnu-gcc (Debian 11.2.0-9) 11.2.0
Copyright (C) 2021 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

But clang 14.0.4 + lld 14.0.4 on Debian unstable cause this runtime error as

ld.lld: error: kernel/built-in.a(kallsyms.o):(function update_iter: .text+0x94e): relocation R_RISCV_PCREL_HI20 out of range: -524390 is not in [-524288, 524287]; references kallsyms_relative_base
>>> referenced by kallsyms.c
>>> defined in kernel/built-in.a(kallsyms.o)

ld.lld: error: too many errors emitted, stopping now (use -error-limit=0 to see all errors)

A reproducing script is below:

xzcat linux-5.18.tar.xz | tar xf -
cd linux-5.18
make LLVM=1 LLVM_IAS=1 ARCH=riscv CROSS_COMPILE=riscv64-linux-gnu- nommu_virt_defconfig
cat >>.config <<EOF
CONFIG_KALLSYMS=y
CONFIG_KALLSYMS_ALL=y
CONFIG_KALLSYMS_BASE_RELATIVE=y
EOF
make LLVM=1 LLVM_IAS=1 ARCH=riscv CROSS_COMPILE=riscv64-linux-gnu- V=1 W=0 >error.txt 2>&1 &
tail -f error.txt

[nickdesaulniers]

Just grepping for R_RISCV_HI20 on lore, it looks like there's been lots of reports of this one.
https://lore.kernel.org/llvm/?q=R_RISCV_HI20