[WIP] [Deepin-Kernel-SIG] [linux 6.6-y] [Deepin] Loongarch: optimize syscall reg save

[opsiff]

deepin inclusion
category: performance

It saves a st.d in the hot syscall path, and let
the compiler know to optimize it in asm,
and helps to improve the syscall performance little.

I have test in 3A6000 After patch:

Benchmark Run: 一 6月 16 2025 20:38:10 - 20:47:09
8 CPUs in system; running 1 parallel copy of tests

Dhrystone 2 using register variables 47066632.8 lps (10.0 s, 2 samples)
Double-Precision Whetstone 5036.1 MWIPS (10.0 s, 2 samples)
Execl Throughput 4484.2 lps (29.2 s, 1 samples)
File Copy 1024 bufsize 2000 maxblocks 656586.0 KBps (30.0 s, 1 samples)
File Copy 256 bufsize 500 maxblocks 175086.0 KBps (30.0 s, 1 samples)
File Copy 4096 bufsize 8000 maxblocks 1998702.0 KBps (30.0 s, 1 samples)
Pipe Throughput 1365130.7 lps (10.0 s, 2 samples)
Pipe-based Context Switching 126232.9 lps (10.0 s, 2 samples)
Process Creation 9202.7 lps (30.0 s, 1 samples)
Shell Scripts (1 concurrent) 12501.2 lpm (60.0 s, 1 samples)
Shell Scripts (8 concurrent) 4974.9 lpm (60.0 s, 1 samples)
System Call Overhead 1467021.7 lps (10.0 s, 2 samples)

System Benchmarks Index Values BASELINE RESULT INDEX
Dhrystone 2 using register variables 116700.0 47066632.8 4033.1
Double-Precision Whetstone 55.0 5036.1 915.7
Execl Throughput 43.0 4484.2 1042.8
File Copy 1024 bufsize 2000 maxblocks 3960.0 656586.0 1658.0
File Copy 256 bufsize 500 maxblocks 1655.0 175086.0 1057.9
File Copy 4096 bufsize 8000 maxblocks 5800.0 1998702.0 3446.0
Pipe Throughput 12440.0 1365130.7 1097.4
Pipe-based Context Switching 4000.0 126232.9 315.6
Process Creation 126.0 9202.7 730.4
Shell Scripts (1 concurrent) 42.4 12501.2 2948.4
Shell Scripts (8 concurrent) 6.0 4974.9 8291.5
System Call Overhead 15000.0 1467021.7 978.0
========
System Benchmarks Index Score 1510.2

------------------------------------------------------------------------ Benchmark Run: 一 6月 16 2025 20:47:09 - 20:56:08
8 CPUs in system; running 8 parallel copies of tests

Dhrystone 2 using register variables 221748966.2 lps (10.0 s, 2 samples)
Double-Precision Whetstone 37218.5 MWIPS (10.0 s, 2 samples)
Execl Throughput 24364.4 lps (29.0 s, 1 samples)
File Copy 1024 bufsize 2000 maxblocks 3681637.0 KBps (30.0 s, 1 samples)
File Copy 256 bufsize 500 maxblocks 1020033.0 KBps (30.0 s, 1 samples)
File Copy 4096 bufsize 8000 maxblocks 8054794.0 KBps (30.0 s, 1 samples)
Pipe Throughput 8209249.1 lps (10.0 s, 2 samples)
Pipe-based Context Switching 1058150.7 lps (10.0 s, 2 samples)
Process Creation 49636.4 lps (30.0 s, 1 samples)
Shell Scripts (1 concurrent) 43521.6 lpm (60.0 s, 1 samples)
Shell Scripts (8 concurrent) 5672.4 lpm (60.0 s, 1 samples)
System Call Overhead 9407101.4 lps (10.0 s, 2 samples)

System Benchmarks Index Values BASELINE RESULT INDEX
Dhrystone 2 using register variables 116700.0 221748966.2 19001.6
Double-Precision Whetstone 55.0 37218.5 6767.0
Execl Throughput 43.0 24364.4 5666.2
File Copy 1024 bufsize 2000 maxblocks 3960.0 3681637.0 9297.1
File Copy 256 bufsize 500 maxblocks 1655.0 1020033.0 6163.3
File Copy 4096 bufsize 8000 maxblocks 5800.0 8054794.0 13887.6
Pipe Throughput 12440.0 8209249.1 6599.1
Pipe-based Context Switching 4000.0 1058150.7 2645.4
Process Creation 126.0 49636.4 3939.4
Shell Scripts (1 concurrent) 42.4 43521.6 10264.5
Shell Scripts (8 concurrent) 6.0 5672.4 9454.0
System Call Overhead 15000.0 9407101.4 6271.4
========
System Benchmarks Index Score 7335.3

Before patch:
Benchmark Run: 一 6月 16 2025 22:58:12 - 23:07:11
8 CPUs in system; running 1 parallel copy of tests

Dhrystone 2 using register variables 41001790.5 lps (10.0 s, 2 samples)
Double-Precision Whetstone 5036.1 MWIPS (10.0 s, 2 samples)
Execl Throughput 4482.0 lps (29.6 s, 1 samples)
File Copy 1024 bufsize 2000 maxblocks 654904.0 KBps (30.0 s, 1 samples)
File Copy 256 bufsize 500 maxblocks 173158.0 KBps (30.0 s, 1 samples)
File Copy 4096 bufsize 8000 maxblocks 2008222.0 KBps (30.0 s, 1 samples)
Pipe Throughput 1370314.7 lps (10.0 s, 2 samples)
Pipe-based Context Switching 126314.0 lps (10.0 s, 2 samples)
Process Creation 9063.9 lps (30.0 s, 1 samples)
Shell Scripts (1 concurrent) 12506.3 lpm (60.0 s, 1 samples)
Shell Scripts (8 concurrent) 4972.7 lpm (60.0 s, 1 samples)
System Call Overhead 1448942.6 lps (10.0 s, 2 samples)

System Benchmarks Index Values BASELINE RESULT INDEX
Dhrystone 2 using register variables 116700.0 41001790.5 3513.4
Double-Precision Whetstone 55.0 5036.1 915.7
Execl Throughput 43.0 4482.0 1042.3
File Copy 1024 bufsize 2000 maxblocks 3960.0 654904.0 1653.8
File Copy 256 bufsize 500 maxblocks 1655.0 173158.0 1046.3
File Copy 4096 bufsize 8000 maxblocks 5800.0 2008222.0 3462.5
Pipe Throughput 12440.0 1370314.7 1101.5
Pipe-based Context Switching 4000.0 126314.0 315.8
Process Creation 126.0 9063.9 719.4
Shell Scripts (1 concurrent) 42.4 12506.3 2949.6
Shell Scripts (8 concurrent) 6.0 4972.7 8287.8
System Call Overhead 15000.0 1448942.6 966.0
========
System Benchmarks Index Score 1488.9

------------------------------------------------------------------------ Benchmark Run: 一 6月 16 2025 23:07:11 - 23:16:11
8 CPUs in system; running 8 parallel copies of tests

Dhrystone 2 using register variables 221753204.3 lps (10.0 s, 2 samples)
Double-Precision Whetstone 37215.6 MWIPS (10.0 s, 2 samples)
Execl Throughput 24319.0 lps (30.0 s, 1 samples)
File Copy 1024 bufsize 2000 maxblocks 3656936.0 KBps (30.0 s, 1 samples)
File Copy 256 bufsize 500 maxblocks 1016886.0 KBps (30.0 s, 1 samples)
File Copy 4096 bufsize 8000 maxblocks 7966493.0 KBps (30.0 s, 1 samples)
Pipe Throughput 8211487.8 lps (10.0 s, 2 samples)
Pipe-based Context Switching 1066013.7 lps (10.0 s, 2 samples)
Process Creation 50743.5 lps (30.0 s, 1 samples)
Shell Scripts (1 concurrent) 43664.4 lpm (60.0 s, 1 samples)
Shell Scripts (8 concurrent) 5674.7 lpm (60.0 s, 1 samples)
System Call Overhead 9320000.0 lps (10.0 s, 2 samples)

System Benchmarks Index Values BASELINE RESULT INDEX
Dhrystone 2 using register variables 116700.0 221753204.3 19002.0
Double-Precision Whetstone 55.0 37215.6 6766.5
Execl Throughput 43.0 24319.0 5655.6
File Copy 1024 bufsize 2000 maxblocks 3960.0 3656936.0 9234.7
File Copy 256 bufsize 500 maxblocks 1655.0 1016886.0 6144.3
File Copy 4096 bufsize 8000 maxblocks 5800.0 7966493.0 13735.3
Pipe Throughput 12440.0 8211487.8 6600.9
Pipe-based Context Switching 4000.0 1066013.7 2665.0
Process Creation 126.0 50743.5 4027.3
Shell Scripts (1 concurrent) 42.4 43664.4 10298.2
Shell Scripts (8 concurrent) 6.0 5674.7 9457.8
System Call Overhead 15000.0 9320000.0 6213.3
========
System Benchmarks Index Score 7336.1

Summary by Sourcery

Optimize the LoongArch syscall path by removing a redundant assembly store and moving zero-initialization of the return register into the C-level do_syscall, reducing a memory write and boosting syscall performance.

Enhancements:

• Remove redundant st.d in the assembly syscall entry.
• Move initialization of regs->regs[0] into do_syscall to save a memory write.

[sourcery-ai]

Reviewer's Guide

This patch optimizes the Loongarch system-call hot path by eliminating an unnecessary assembly-level store of the zero register and relocating that initialization into C, thus saving one memory write and enabling better compiler optimization for faster syscalls.

Sequence diagram for optimized syscall register initialization in Loongarch

sequenceDiagram
    participant User as actor User Process
    participant Kernel as Kernel (entry.S)
    participant SyscallC as do_syscall (syscall.c)
    User->>Kernel: Trigger syscall (trap)
    Kernel->>Kernel: Save registers (except PT_R0)
    Kernel->>SyscallC: Call do_syscall(regs)
    SyscallC->>SyscallC: regs->regs[0] = 0 (now in C, not asm)
    SyscallC->>SyscallC: if (nr < NR_syscalls) regs->regs[0] = nr + 1
    SyscallC-->>Kernel: Return from do_syscall
    Kernel-->>User: Return to user mode

Loading

Class diagram for pt_regs and syscall register handling changes

classDiagram
    class pt_regs {
        +unsigned long regs[32]
        +unsigned long prmd
        +unsigned long crmd
        ...
    }
    class do_syscall {
        +void do_syscall(struct pt_regs *regs)
        // now sets regs->regs[0] = 0 in C
    }
    pt_regs <.. do_syscall : uses

Loading

File-Level Changes

Change	Details	Files
Relocate zero-register initialization to C to remove assembly store	Comment out the st.d zero store in arch/loongarch/kernel/entry.S Add regs->regs[0] = 0 at the start of do_syscall in syscall.c before restart logic	arch/loongarch/kernel/entry.S arch/loongarch/kernel/syscall.c

---

Tips and commands

Interacting with Sourcery

• Trigger a new review: Comment @sourcery-ai review on the pull request.
• Continue discussions: Reply directly to Sourcery's review comments.
• Generate a GitHub issue from a review comment: Ask Sourcery to create an
  issue from a review comment by replying to it. You can also reply to a
  review comment with @sourcery-ai issue to create an issue from it.
• Generate a pull request title: Write @sourcery-ai anywhere in the pull
  request title to generate a title at any time. You can also comment
  @sourcery-ai title on the pull request to (re-)generate the title at any time.
• Generate a pull request summary: Write @sourcery-ai summary anywhere in
  the pull request body to generate a PR summary at any time exactly where you
  want it. You can also comment @sourcery-ai summary on the pull request to
  (re-)generate the summary at any time.
• Generate reviewer's guide: Comment @sourcery-ai guide on the pull
  request to (re-)generate the reviewer's guide at any time.
• Resolve all Sourcery comments: Comment @sourcery-ai resolve on the
  pull request to resolve all Sourcery comments. Useful if you've already
  addressed all the comments and don't want to see them anymore.
• Dismiss all Sourcery reviews: Comment @sourcery-ai dismiss on the pull
  request to dismiss all existing Sourcery reviews. Especially useful if you
  want to start fresh with a new review - don't forget to comment
  @sourcery-ai review to trigger a new review!

Customizing Your Experience

Access your dashboard to:

• Enable or disable review features such as the Sourcery-generated pull request
  summary, the reviewer's guide, and others.
• Change the review language.
• Add, remove or edit custom review instructions.
• Adjust other review settings.

Getting Help

• Contact our support team for questions or feedback.
• Visit our documentation for detailed guides and information.
• Keep in touch with the Sourcery team by following us on X/Twitter, LinkedIn or GitHub.

[deepin-ci-robot]

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by:
Once this PR has been reviewed and has the lgtm label, please ask for approval from opsiff. For more information see the Code Review Process.

The full list of commands accepted by this bot can be found here.

Needs approval from an approver in each of these files:

• deepin/OWNERS

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[deepin-ci-robot]

deepin pr auto review

代码审查意见：

1. 在entry.S文件中，注释掉了st.d zero, sp, PT_R0指令，但是没有提供替代的实现。如果这个指令被注释掉，那么regs->regs[0]的值将不会被设置为0，这可能会导致系统调用处理的不一致。建议在do_syscall函数中明确设置regs->regs[0]为0，而不是注释掉st.d zero, sp, PT_R0指令。

2. 在syscall.c文件中，do_syscall函数中添加了对regs->regs[0]的赋值操作。这个改动是合理的，因为它确保了系统调用的编号被正确地传递给内核。但是，如果handle_syscall宏中已经包含了设置regs->regs[0]为0的逻辑，那么这个改动可能是多余的。

3. 在syscall.c文件中，注释// Move from handle_syscall macro to save a memio可能是不必要的，因为handle_syscall宏中的操作已经被移到do_syscall函数中，并且已经得到了适当的处理。

4. 代码中没有明显的语法或逻辑错误。

5. 代码的性能影响应该很小，因为只是对寄存器进行了一次赋值操作。

6. 代码的安全性影响不大，因为这是一个内核级别的操作，通常不会受到外部攻击。

综上所述，建议在do_syscall函数中明确设置regs->regs[0]为0，而不是注释掉entry.S文件中的st.d zero, sp, PT_R0指令。同时，如果handle_syscall宏中已经包含了设置regs->regs[0]为0的逻辑，那么syscall.c文件中的改动可能是多余的。

[Copilot]

Pull Request Overview

This PR optimizes the LoongArch syscall path by eliminating a redundant zero store in the assembly entry and moving the zero-initialization of the return register into the C-level do_syscall, reducing a memory write in the hot path.

• Initialize regs->regs[0] to 0 in do_syscall instead of in assembly
• Comment out the redundant st.d zero, sp, PT_R0 in the assembly entry

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
arch/loongarch/kernel/syscall.c	Add regs->regs[0] = 0; in do_syscall and a comment
arch/loongarch/kernel/entry.S	Comment out the redundant zero store in handle_syscall

Comments suppressed due to low confidence (2)

arch/loongarch/kernel/syscall.c:47

• [nitpick] Consider clarifying this comment to explain why moving the zero-initialization here improves performance, e.g., "Initialize regs->regs[0] to 0 in C to eliminate a redundant store in the assembly entry and reduce memory access."

// Move from handle_syscall macro to save a memio

arch/loongarch/kernel/entry.S:34

• [nitpick] Remove the commented-out st.d zero, sp, PT_R0 instead of leaving dead code in the assembly file, to keep the entry path concise.

#	st.d		zero, sp, PT_R0

[sourcery-ai]

Hey @opsiff - I've reviewed your changes and they look great!

---

Sourcery is free for open source - if you like our reviews please consider sharing them ✨

• X
• Mastodon
• LinkedIn
• Facebook

_{Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.}