feat: add support for restartable system calls on RISC-V64#10520
feat: add support for restartable system calls on RISC-V64#10520eatvector wants to merge 1 commit intoRT-Thread:masterfrom
Conversation
📌 Code Review Assignment🏷️ Tag: componentsReviewers: Maihuanyi Changed Files (Click to expand)
🏷️ Tag: components_lwpReviewers: xu18838022837 Changed Files (Click to expand)
🏷️ Tag: libcpu_riscvReviewers: Yaochenger Changed Files (Click to expand)
📊 Current Review Status (Last Updated: 2025-08-15 08:38 CST)
📝 Review Instructions
|
github-actions
bot
added
RT-Smart
eatvector
force-pushed
the
feature/riscv64-syscall-restart
branch
10 times, most recently
from
c3e2204 to
662b3f1
Compare
eatvector
force-pushed
the
feature/riscv64-syscall-restart
branch
4 times, most recently
from
c9e2878 to
30ef0db
Compare
|
请给出更多说明，包括对系统调用部分代码的更改，谢谢。 |
好的，我这几天改改 |
eatvector
force-pushed
the
feature/riscv64-syscall-restart
branch
from
30ef0db to
68859d2
Compare
eatvector
force-pushed
the
feature/riscv64-syscall-restart
branch
from
68859d2 to
05aa0a6
Compare
|
@BernardXiong 您好，我已补充了相关说明信息。如果您需要更多细节，或对现有信息有任何疑问，请随时告知，我将尽快提供进一步说明 |
eatvector
force-pushed
the
feature/riscv64-syscall-restart
branch
from
05aa0a6 to
e1e1be0
Compare
eatvector
force-pushed
the
feature/riscv64-syscall-restart
branch
from
e1e1be0 to
49382b0
Compare
There was a problem hiding this comment.
Pull Request Overview
This PR implements support for restartable system calls on RISC-V64 architecture, following the implementation pattern used in AArch64. The implementation allows system calls that are interrupted by signals to be automatically restarted when the signal handler completes.
Key Changes
- Implements system call restart mechanism by preserving original parameters and restoring execution context
- Adds signal handling integration to detect and restart interrupted system calls
- Provides architecture-specific assembly routines for context restoration
Reviewed Changes
Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.
| File | Description |
|---|---|
| libcpu/risc-v/common64/syscall_c.c | Preserves original a0 register value in t0 and removes a7 clearing to enable restart |
| components/lwp/arch/risc-v/rv64/lwp_gcc.S | Adds arch_syscall_restart assembly function and updates arch_signal_quit to pass kernel stack pointer |
| components/lwp/arch/risc-v/rv64/lwp_arch.h | Updates arch_signal_ucontext_restore function signature to include kernel_sp parameter |
| components/lwp/arch/risc-v/rv64/lwp_arch.c | Implements core restart logic including error checking, context restoration, and signal post-action handling |
| arch_syscall_restart(exp_frame, ksp); | ||
| } | ||
|
|
||
| return; |
There was a problem hiding this comment.
Redundant return statement at the end of a void function. This return statement can be removed for cleaner code.
在void函数末尾的冗余return语句。可以移除此return语句使代码更简洁。
| return; |
| long sys_id = exp_frame->a7; | ||
|
|
||
| (void)sys_id; | ||
|
|
There was a problem hiding this comment.
The sys_id variable is assigned but only used to suppress unused variable warnings. Consider removing the variable assignment if it's truly unused, or use it in the LOG_D statements that follow.
sys_id变量被赋值但仅用于抑制未使用变量警告。如果确实未使用，考虑移除变量赋值，或在后续的LOG_D语句中使用它。
| /* adjust for epc auto-increment in syscall_handler */ | ||
| exp_frame->epc -= 4; | ||
| /* copy exception frame from user stack to kernel stack */ | ||
| lwp_memcpy(ksp - sizeof(*exp_frame), exp_frame, sizeof(*exp_frame)); |
There was a problem hiding this comment.
[nitpick] Extra space between function arguments. Remove the extra space after the comma for consistent formatting.
函数参数间有多余空格。移除逗号后的多余空格以保持格式一致。
| lwp_memcpy(ksp - sizeof(*exp_frame), exp_frame, sizeof(*exp_frame)); | |
| lwp_memcpy(ksp - sizeof(*exp_frame), exp_frame, sizeof(*exp_frame)); |
3 participants
拉取/合并请求描述：(PR description)
参考aarch64架构下的系统调用重启实现riscv64架构下的系统调用重启,关联问题#9761
[
为什么提交这份PR (why to submit this PR)
你的解决方案是什么 (what is your solution)
可重启的系统调用如果由于接受到信号中断导致执行失败，则在信号退出处理时，根据用户栈保存的exception frame 恢复寄存器（包括epc以及存储系统调用号和系统调用参数的寄存器等），并跳转到 trap_entry，以重新启动系统调用。
系统调用功能函数执行前阶段
系统调用功能函数执行后阶段
移除对 a7（系统调用号寄存器）的清零操作，保留其原始值以供重启使用。
在系统调用执行被信号中断退出syscall_handler后，进行信号处理时如果判定需要进行系统调用重启，则通过 arch_syscall_set_errno 将 用户上下文中的a0 设为 -ERESTART，作为需重启的标记。之后将异常帧复制到用户栈，返回用户态执行中断处理函数（如果有的话)。
信号退出处理阶段（在这一步完成系统调用重启）
通过ecall再次陷入内核，执行arch_signal_quit
arch_signal_quit 调用 arch_signal_ucontext_restore，并传递两个关键参数。其一为用户栈中保存的异常帧的基地址，其二为新增参数，表示内核栈中，arch_signal_quit 用户上下文信息的基地址加上 CTX_REG_NR * REGBYTES，设为ksp。
通过调用链 arch_signal_ucontext_restore-> arch_signal_post_action-> arch_signal_check_erestart 执行到核心函数 arch_signal_check_erestart 。
该函数根据保存的异常帧信息判断是否需要重启，如果需要，将异常帧中的t0赋值给a0，即恢复系统调用第一个参数的值。随后将异常帧中的epc减去4进行恢复，因为系统调用在被信号中断后在 syscall_handler对epc进行了加4操作，现在需要将其恢复到跟第一次调用系统调用前一样。之后将异常帧拷贝到内核栈中原来存储 arch_signal_quit 用户上下文信息的位置。并调用汇编函数arch_syscall_restart。
arch_syscall_restart从内核栈中恢复刚刚拷贝进来的异常帧信息到寄存器（包括系统调用参数寄存器a0-a6,以及系统调用号寄存器a7,epc也被恢复指向用户空间的ecall指令地址）。此外，sscratch 会被设置为ksp。之后跳转 trap_entry以重新执行系统调用。
请提供验证的bsp和config (provide the config and bsp)
]
当前拉取/合并请求的状态 Intent for your PR
必须选择一项 Choose one (Mandatory):
代码质量 Code Quality：
我在这个拉取/合并请求中已经考虑了 As part of this pull request, I've considered the following:
#if 0代码，不包含已经被注释了的代码 All redundant code is removed and cleaned up