change the SMP scheduling policy from synchronous to asynchronous

[hujun260]

Summary

Currently, if we need to schedule a task to another CPU, we have to completely halt the other CPU,
manipulate the scheduling linked list, and then resume the operation of that CPU. This process is both time-consuming and unnecessary.

During this process, both the current CPU and the target CPU are inevitably subjected to busyloop.

The improved strategy is to simply send a cross-core interrupt to the target CPU.
The current CPU continues to run while the target CPU responds to the interrupt, eliminating the certainty of a busyloop occurring.

Impact

all smp arch

Testing

We performed performance testing on the Cortex-R82 ARM64 SMP chip. Sending messages through a message queue which causes context switching. We recorded the time from sending a message to receiving it, with the time unit being cycles.

we also use monkey test in armv7-a

[nuttxpr]

[Experimental Bot, please feedback here]

NuttX PR Requirements Review

This PR appears to meet most NuttX requirements, but some clarifications are needed:

Strengths:

• Clear Summary: The summary explains the problem (inefficient cross-core task scheduling), the solution (cross-core interrupts), and the benefits (reduced overhead, no busyloops).
• Impact Well-Defined: The impact on SMP architectures is noted, but more details are needed for each "Impact" sub-point (User, Build, Hardware, Documentation, Security, Compatibility). For example:
  • User: Will applications need to change how they schedule tasks across cores?
  • Build: Are any new configuration options introduced?
  • Hardware: Are specific interrupt mechanisms required on target platforms?
• Testing Provided: Performance testing results are included, demonstrating improvements on a specific platform.

Areas for Improvement:

• Complete Impact Assessment: Provide specific answers for each impact sub-point. Even if the answer is "NO," briefly explain why.
• Testing Details:
  • Environment: Specify the OS, CPU, compiler used for the build host in the "Testing" section.
  • Target Details: Provide the full board:config used for the Cortex-R82 tests.
  • Monkey Test: Briefly describe the "monkey test" and its significance.
  • Before/After Logs: The "Testing logs" sections are empty. Include relevant log snippets showing the improvement.

Conciseness: The provided information is concise and well-written.

Overall:

This PR demonstrates a promising improvement, but requires a more thorough impact assessment and additional testing details before it can be considered complete.

[masayuki2009]

all smp arch

@hujun260
Which configurations did you test with this PR?

[hujun260]

all smp arch

@hujun260 Which configurations did you test with this PR?

qemu-armv7a:smp
qemu-armv8a:nsh_smp

[masayuki2009]

all smp arch

@hujun260 Which configurations did you test with this PR?

qemu-armv7a:smp qemu-armv8a:nsh_smp

Please test esp32-devkitc:smp, rv-virt:smp, rv-virt:smp64, rv-virt:ksmp64 and sim:smp, since the PR affects these architectures.
Aslo, if you have spresense and raspberrypi-pico, please test the boards as well.

[hujun260]

all smp arch

@hujun260 Which configurations did you test with this PR?

qemu-armv7a:smp qemu-armv8a:nsh_smp

Please test esp32-devkitc:smp, rv-virt:smp, rv-virt:smp64, rv-virt:ksmp64 and sim:smp, since the PR affects these architectures. Aslo, if you have spresense and raspberrypi-pico, please test the boards as well.

esp32-devkitc:smp ostest test ok
sim:smp can't boot, not caused by this pr
rv-virt:smp ostest test ok
rv-virt:smp64 ostest test ok
rv-virt:ksmp64 ostest test ok (except vfork break introdced by #12812)
raspberrypi-pico:smp ostest test ok

[masayuki2009]

@hujun260
After merging this PR, I noticed that the hello app hangs with qemu-intel64:smp.

NuttShell (NSH) NuttX-12.0.0
nsh>  
nsh> uname -a
NuttX 12.0.0 d4707646d5 Sep 30 2024 09:52:36 x86_64 qemu-intel64
nsh> mount
  /proc type procfs
nsh> ps
  PID GROUP CPU PRI POLICY   TYPE    NPX STATE    EVENT     SIGMASK           STACK COMMAND
    0     0   0   0 FIFO     Kthread   - Assigned           0000000000000000 4194280 CPU0 IDLE
    1     0   1   0 FIFO     Kthread   - Running            0000000000000000 4194272 CPU1 IDLE
    2     2   0 100 FIFO     Task      - Running            0000000000000000 4194240 nsh_main
nsh> free
                 total       used       free    maxused    maxfree  nused  nfree
      Umem:  241659896    4201320  237458576    4202112  237458576     34      1
nsh> kill -4 0
nsh> kill -4 1
nsh> hello
Hello, World!!

Also,

sim:smp can't boot, not caused by this pr

Hmm, but you modified the sim-related code as well.
This means that you've never tested this code with the sim:smp but pushed the code.

[hujun260]

@hujun260 After merging this PR, I noticed that the hello app hangs with qemu-intel64:smp.

NuttShell (NSH) NuttX-12.0.0
nsh>  
nsh> uname -a
NuttX 12.0.0 d4707646d5 Sep 30 2024 09:52:36 x86_64 qemu-intel64
nsh> mount
  /proc type procfs
nsh> ps
  PID GROUP CPU PRI POLICY   TYPE    NPX STATE    EVENT     SIGMASK           STACK COMMAND
    0     0   0   0 FIFO     Kthread   - Assigned           0000000000000000 4194280 CPU0 IDLE
    1     0   1   0 FIFO     Kthread   - Running            0000000000000000 4194272 CPU1 IDLE
    2     2   0 100 FIFO     Task      - Running            0000000000000000 4194240 nsh_main
nsh> free
                 total       used       free    maxused    maxfree  nused  nfree
      Umem:  241659896    4201320  237458576    4202112  237458576     34      1
nsh> kill -4 0
nsh> kill -4 1
nsh> hello
Hello, World!!

Also,

sim:smp can't boot, not caused by this pr

Hmm, but you modified the sim-related code as well. This means that you've never tested this code with the sim:smp but pushed the code.

qemu-intel64:smp, i'll fix it as soon as possible #13749
sim fix #13753