Fix module load issue for R_RISCV_32_PCREL

[xmzzz]

fixed: #310

Backport the upstream RISC-V module relocation support to allow loading
kernel modules built with newer binutils (e.g. 2.42) that emit
R_RISCV_32_PCREL and other relocation types.

This fixes the "Unknown relocation type 57" error observed when loading
modules such as cambricon_drv compiled with newer toolchains.

Included patches:

1. Revert "sg2042: riscv: kernel: Optimize apply_relocate_add()"
2. riscv: Avoid unaligned access when relocating modules
3. riscv: Add remaining module relocations
4. RISC-V: Don't rely on positional structure initialization
5. riscv: Safely remove entries from relocation list
6. riscv: Correct type casting in module loading
7. riscv: module: Allocate PLT entries for R_RISCV_PLT32
8. riscv: Fix module loading free order
9. riscv: Correctly free relocation hashtable on error
10. riscv: Fix relocation hashtable size
11. riscv: module: remove relocation_head rel_entry member allocation
12. riscv: Use kvmalloc_array on relocation_hashtable

Test:

• Built a minimal test module (pcrel_test.ko) that forces generation of
  an R_RISCV_32_PCREL relocation. Verified with readelf:

  $ riscv64-unknown-linux-gnu-readelf -r pcrel_test.ko | grep 32_PCREL
  000000000000 003600000039 R_RISCV_32_PCREL 0000000000000000 pcrel_target + 0

• Loaded the module successfully on LicheePi 4A board.
  dmesg output:

$ sudo insmod ./pcrel_test.ko 
[sudo] password for openeuler: 
[  688.973056][ T1250] pcrel_test: loading out-of-tree module taints kernel.
[  688.979904][ T1250] pcrel_test: module verification failed: signature and/or required key missing - tainting kernel
[  688.991394][ T1250] R_RISCV_32_PCREL OK: value=0xffffdff8 expected=0xffffdff8

补充分析, 基于上游主线 master （by @unicornx）：
解释为何要 backport 这些补丁，基本上是基于 “8fd6c5142395 riscv: Add remaining module relocations” 以及围绕它的后期的 fixes。

git log --oneline arch/riscv/kernel/module.c
189f164e573e Convert remaining multi-line kmalloc_obj/flex GFP_KERNEL uses
bf4afc53b77a Convert 'alloc_obj' family to use the new default GFP_KERNEL argument
69050f8d6d07 treewide: Replace kmalloc with kmalloc_obj for non-scalar types
0b4cce68efb9 riscv: module: Fix out-of-bounds relocation access
8578b2f7e1fb riscv: Use kvmalloc_array on relocation_hashtable         // backported by PR#344, Fixes: 8fd6c5142395 ("riscv: Add remaining module relocations")
03f0b548537f riscv: module: remove relocation_head rel_entry member    // backported by PR#344, Fixes: 8fd6c5142395 ("riscv: Add remaining module relocations")
3cc754c237e9 riscv: Use LIST_HEAD() to simplify code
0cc2dc4902f4 arch: make execmem setup available regardless of CONFIG_MODULES
4d7b321a9ce0 riscv: extend execmem_params for generated code allocations
f6bec26c0a73 mm/execmem, arch: convert simple overrides of module_alloc to execmem
080c4324fa5e riscv: optimize ELF relocation function in riscv
3a58275099b9 Merge patch series "riscv: modules: Fix module loading error handling"
a35551c7244d riscv: Fix relocation_hashtable size                      // backported by PR#344, Fixes: d8792a5734b0 ("riscv: Safely remove entries from relocation list")
4b38b36bfbd8 riscv: Correctly free relocation hashtable on error       // backported by PR#344, Fixes: d8792a5734b0 ("riscv: Safely remove entries from relocation list")
78996eee79eb riscv: Fix module loading free order                      // backported by PR#344, Fixes: d8792a5734b0 ("riscv: Safely remove entries from relocation list")
749b94b08005 riscv: Fix module_alloc() that did not reset the linear   // backported by 6.6.y as 8633e74002cd
4a92a87950c4 riscv: Correct type casting in module loading             // backported by PR#344, Fixes: 8fd6c5142395 ("riscv: Add remaining module relocations")
d8792a5734b0 riscv: Safely remove entries from relocation list         // backported by PR#344, Fixes: 8fd6c5142395 ("riscv: Add remaining module relocations")
28ea54bade76 RISC-V: Don't rely on positional structure initialization // backported by PR#344, Fixes: b51fc88cb35e ("Merge patch series "riscv: Add remaining module relocations and tests"")
8fd6c5142395 riscv: Add remaining module relocations				   // backported by PR#344, 这个是解决问题的主要补丁
8cbe0accc4a6 riscv: Avoid unaligned access when relocating modules     // backported by PR#344
1bc400ffb52b riscv: module: Add ADD16 and SUB16 rela types             // 6.6 包含的补丁到这里为止
......

[github-actions]

---

开始测试 log: https://github.com/RVCK-Project/rvck/actions/runs/29353724774

参数解析结果

args	value
repository	RVCK-Project/rvck
head ref	pull/334/head
base ref	rvck-6.6
LAVA repo	RVCK-Project/lavaci
LAVA hardware	['qemu']
LAVA Testcase path	lava-testcases/common-test/ltp/ltp.yaml
need run job	kunit-test,kernel-build,check-patch,lava-trigger

---

测试完成

详细结果:

check	result
kunit-test	success
kernel-build	success
check-patch	success
lava-trigger-qemu	success
lava-trigger-sg2042	skipped
lava-trigger-k1	skipped
lava-trigger-lpi4a	skipped

Kunit Test Result

[17:30:17] Testing complete. Ran 482 tests: passed: 465, skipped: 17

• detail log: https://github.com/RVCK-Project/rvck/actions/runs/29353724774/attempts/1/#summary-87156080909

Kernel Build Result

• Kernel build succeeded: RVCK-Project/rvck/334_29353724774_1/

• 93d57dc650e5a4ecefd3d9bd35b9bf84 /srv/guix_result/d182f33f45680fba26dad1fa22bda13fbc401a76/Image

• f137279a02ca176b89c5506e96004049 /root/initramfs.img

• detail log: https://github.com/RVCK-Project/rvck/actions/runs/29353724774/attempts/1/#summary-87156080818

Check Patch Result

Total Errors	0
Total Warnings	23

• detail log: https://github.com/RVCK-Project/rvck/actions/runs/29353724774/attempts/1/#summary-87156080761

LAVA Check (qemu)

args	value
testcase_repo	RVCK-Project/lavaci
lava_template	lava-job-template/qemu/qemu-ltp.yaml
testcase_path	lava-testcases/common-test/ltp/ltp.yaml
kernel_download_url	http://10.30.190.110/openEuler-RISC-V/RVCK/OERV-RVCI/RVCK-Project/rvck/334_29353724774_1/Image
initramfs_download_url	http://10.30.190.110/openEuler-RISC-V/RVCK/OERV-RVCI/RVCK-Project/rvck/334_29353724774_1/initramfs.img
rootfs_download_url	https://fast-mirror.isrc.ac.cn/openeuler-sig-riscv/openEuler-RISC-V/RVCK/openEuler24.03-LTS-SP1/openeuler-rootfs.img.zst
testcase_ref	main
testitem_name	RVCK-Project_rvck_pull_request_target_334__common-test_qemu

result: Lava check done!

• lava log: https://lava.oerv.ac.cn/scheduler/job/3456

• lava result count: [pass]: 10

• detail log: https://github.com/RVCK-Project/rvck/actions/runs/29353724774/attempts/1/#summary-87188658109

[unicornx]

@xmzzz

Test:

• Built a minimal test module (pcrel_test.ko) that forces generation of
  an R_RISCV_32_PCREL relocation. Verified with readelf:

请问如何制作出这种 ko？对工具链和编译选项有何要求，我这里想试一下。

$ riscv64-unknown-linux-gnu-readelf -r pcrel_test.ko | grep 32_PCREL
000000000000 003600000039 R_RISCV_32_PCREL 0000000000000000 pcrel_target + 0

• Loaded the module successfully on LicheePi 4A board.

这个问题和硬件应该没有关系吧，我想基于 QEMU 测试一下，ok？

dmesg output:

$ sudo insmod ./pcrel_test.ko 
[sudo] password for openeuler: 
[  688.973056][ T1250] pcrel_test: loading out-of-tree module taints kernel.
[  688.979904][ T1250] pcrel_test: module verification failed: signature and/or required key missing - tainting kernel
[  688.991394][ T1250] R_RISCV_32_PCREL OK: value=0xffffdff8 expected=0xffffdff8

fix: #310

[xmzzz]

我在下面贴一下测试模块源码，包括C代码和汇编

测试模块通过汇编伪指令 .reloc 强制生成一个 type=57 的重定位项，所以理论上不依赖
高版本工具链，只要 assembler 能识别 R_RISCV_32_PCREL（若不能试试用数字 57 代替）。

编译方式和普通内核模块一致：

make KDIR=<kernel_build> ARCH=riscv CROSS_COMPILE=riscv64-unknown-linux-gnu-

与硬件无关，qemu理论上可以测，因为是内核模块加载器对 ELF relocation 的处理逻辑。
(测试时如果遇到 module_layout 版本校验失败的问题，试下编译内核时关闭 CONFIG_MODVERSIONS)

~/test-module/pcrel_test_main.c

/* SPDX-License-Identifier: GPL-2.0 */
/*
 * Minimal kernel module to verify R_RISCV_32_PCREL relocation handling.
 */

#include <linux/module.h>
#include <linux/kernel.h>

extern u32 pcrel_value;
extern void pcrel_target(void);

static int __init pcrel_init(void)
{
        u32 expected = (u32)((unsigned long)&pcrel_target -
                             (unsigned long)&pcrel_value);

        if (pcrel_value == expected)
                pr_info("R_RISCV_32_PCREL OK: value=0x%08x expected=0x%08x\n",
                        pcrel_value, expected);
        else
                pr_err("R_RISCV_32_PCREL FAIL: value=0x%08x expected=0x%08x\n",
                        pcrel_value, expected);

        return 0;
}

static void __exit pcrel_exit(void)
{
        pr_info("pcrel_test unloaded\n");
}

module_init(pcrel_init);
module_exit(pcrel_exit);

MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("Test R_RISCV_32_PCREL relocation");

~/test-module/pcrel_test_asm.S

/* SPDX-License-Identifier: GPL-2.0 */
/*
 * Force generation of an R_RISCV_32_PCREL relocation.
 *
 * R_RISCV_32_PCREL (type 57) is used by newer binutils (>= 2.42) for
 * 32-bit PC-relative data references.
 */

.section .data
.align 2
.globl pcrel_value
pcrel_value:
        .reloc pcrel_value, R_RISCV_32_PCREL, pcrel_target
        .long 0

.section .text
.globl pcrel_target
pcrel_target:
        ret

[unicornx]

Tested-by: Chen Wang wangchen20@iscas.ac.cn

Test with

• QEMU with openEuler-24.03-LTS-SP3-RVA23-riscv64
  and
• Pioneerbox with https://mirror.iscas.ac.cn/openeuler/openEuler-24.03-LTS-SP3/embedded_image/riscv64/milkv/pioneer/

Without this Patchset:

[u@localhost ~]$ sudo insmod ./test_pcrel.ko 
[sudo] password for u: 
[  135.174302] test_pcrel: loading out-of-tree module taints kernel.
[  135.174718] test_pcrel: module verification failed: signature and/or required key missing - tainting kernel
[  135.175147] test_pcrel: Unknown relocation type 57
insmod: ERROR: could not insert module ./test_pcrel.ko: Invalid parameters

Applied this patchset:

[u@localhost ~]$ sudo insmod ./test_pcrel.ko 
[sudo] password for u: 
[   89.766152] test_pcrel: loading out-of-tree module taints kernel.
[   89.766656] test_pcrel: module verification failed: signature and/or required key missing - tainting kernel
[   89.779364] R_RISCV_32_PCREL OK: value=0xffffdff8 expected=0xffffdff8

The test module source:
test_pcrel.tar.gz

[unicornx]

针对 ”Revert "riscv: kernel: Optimize apply_relocate_add()"“ 这笔提交有个疑问：回退这个补丁会不会有什么副作用？当初加算能的这个补丁的改动已经被这次 pr 覆盖了吗？

[liuhj2023]

针对 ”Revert "riscv: kernel: Optimize apply_relocate_add()"“ 这笔提交有个疑问：回退这个补丁会不会有什么副作用？当初加算能的这个补丁的改动已经被这次 pr 覆盖了吗？

汪老师，riscv: kernel: Optimize apply_relocate_add() 这笔patch是我司同事为了加速模块加载所做的一次优化，回退不影响功能。

[unicornx]

LGTM

Reviewed-by: Chen Wang wangchen20@iscas.ac.cn
Tested-by: Chen Wang wangchen20@iscas.ac.cn