Implement ticket spinlock

[TaiJuWu]

Summary

A proposal to slove #1488
Implement ticket spinlock.

Impact

spinlock

Testing

./tools/configure.sh -l qemu-armv8a:nsh_smp
pass ostest

[mu578]

You should try to stress it more to see if you don't hit any dead-locking on one cpu, process timing is important, so try to process heavy / greedy calls in each context, like allocating, read-write file, socket streams (triggering nasty interrupts in the middle), leaking memory on purpose, calls to getpid and alike, anything that is known to be resource system intensive.

[TaiJuWu]

You should try to stress it more to see if you don't hit any dead-locking on one cpu, process timing is important, so try to process heavy / greedy calls in each context, like allocating, read-write file, socket streams (triggering nasty interrupts in the middle), leaking memory on purpose, calls to getpid and alike, anything that is known to be resource system intensive.

I have two question:

1. mem allocating and read-write file seems to use mutex, is it means that I have to modify these to use fair spinlock ?
2. How can I test performance, can I test it on qemu? I have no physical board.

Could you explain more detail?

[acassis]

You should try to stress it more to see if you don't hit any dead-locking on one cpu, process timing is important, so try to process heavy / greedy calls in each context, like allocating, read-write file, socket streams (triggering nasty interrupts in the middle), leaking memory on purpose, calls to getpid and alike, anything that is known to be resource system intensive.

I have two question:

1. mem allocating and  read-write file seems to use mutex, is it means that I have to modify these to use fair spinlock ?

2. How can I test performance, can I test it on qemu? I have no physical board.

Could you explain more detail?

I think you can use QEMU or even the SIM on Linux, but you will need to add some instrumentation to confirm everything is working as expected.

[TaiJuWu]

You should try to stress it more to see if you don't hit any dead-locking on one cpu, process timing is important, so try to process heavy / greedy calls in each context, like allocating, read-write file, socket streams (triggering nasty interrupts in the middle), leaking memory on purpose, calls to getpid and alike, anything that is known to be resource system intensive.

I have two question:

1. mem allocating and  read-write file seems to use mutex, is it means that I have to modify these to use fair spinlock ?

2. How can I test performance, can I test it on qemu? I have no physical board.

Could you explain more detail?

I think you can use QEMU or even the SIM on Linux, but you will need to add some instrumentation to confirm everything is working as expected.

I try following doc to enable INSTRUMENTATION.
But the result shows

This result is different from what I excepted.
Is the reason is encode my shell and ash?

[mu578]

You should try to stress it more to see if you don't hit any dead-locking on one cpu, process timing is important, so try to process heavy / greedy calls in each context, like allocating, read-write file, socket streams (triggering nasty interrupts in the middle), leaking memory on purpose, calls to getpid and alike, anything that is known to be resource system intensive.

I have two question:

1. mem allocating and read-write file seems to use mutex, is it means that I have to modify these to use fair spinlock ?
2. How can I test performance, can I test it on qemu? I have no physical board.

Could you explain more detail?

no there are good calls as is, that's because they will block and will forcibly trigger all sort of events in the middle, such in a unpredictable order, so context switch will happen and a lot, so if you wish: producing mess on purpose.

Qemu on a good host is ok if host and slave share the same arch, meanwhile having a physical board is a must, real clock.

[xiaoxiang781216]

You should try to stress it more to see if you don't hit any dead-locking on one cpu, process timing is important, so try to process heavy / greedy calls in each context, like allocating, read-write file, socket streams (triggering nasty interrupts in the middle), leaking memory on purpose, calls to getpid and alike, anything that is known to be resource system intensive.

I have two question:

1. mem allocating and  read-write file seems to use mutex, is it means that I have to modify these to use fair spinlock ?

2. How can I test performance, can I test it on qemu? I have no physical board.

Could you explain more detail?

I think you can use QEMU or even the SIM on Linux, but you will need to add some instrumentation to confirm everything is working as expected.

I try following doc to enable INSTRUMENTATION. But the result shows

This result is different from what I excepted. Is the reason is encode my shell and ash?

You need redirect the output to file and open it with https://ui.perfetto.dev/ or https://eclipse.dev/tracecompass/.

[xiaoxiang781216]

please run './tools/refresh.sh qemu-armv8a/nsh_smp' to fix the ci error.

[TaiJuWu]

please run './tools/refresh.sh qemu-armv8a/nsh_smp' to fix the ci error.

Done.

[xiaoxiang781216]

The style need fix too:

../nuttx/tools/checkpatch.sh -u -m -g 57bf9d44d2d93619616a4c87ce96b85986f96c65..HEAD
Error: /home/runner/work/nuttx/nuttx/nuttx/sched/semaphore/spinlock.c:167:6: error: Bad left brace alignment
Error: /home/runner/work/nuttx/nuttx/nuttx/sched/semaphore/spinlock.c:169:6: error: Bad right brace alignment
Error: /home/runner/work/nuttx/nuttx/nuttx/sched/semaphore/spinlock.c:174:6: error: Bad left brace alignment
Error: /home/runner/work/nuttx/nuttx/nuttx/sched/semaphore/spinlock.c:176:6: error: Bad right brace alignment
Error: /home/runner/work/nuttx/nuttx/nuttx/sched/semaphore/spinlock.c:[23](https://github.com/apache/nuttx/actions/runs/6427966184/job/17454400895?pr=10605#step:3:24)1:6: error: Bad left brace alignment
Error: /home/runner/work/nuttx/nuttx/nuttx/sched/semaphore/spinlock.c:233:6: error: Bad right brace alignment
Error: /home/runner/work/nuttx/nuttx/nuttx/sched/semaphore/spinlock.c:238:6: error: Bad left brace alignment
Error: /home/runner/work/nuttx/nuttx/nuttx/sched/semaphore/spinlock.c:[24](https://github.com/apache/nuttx/actions/runs/6427966184/job/17454400895?pr=10605#step:3:25)0:6: error: Bad right brace alignment

you can verify locally by:

./tools/checkpatch.sh -g HEAD~...HEAD

[TaiJuWu]

The style need fix too:

../nuttx/tools/checkpatch.sh -u -m -g 57bf9d44d2d93619616a4c87ce96b85986f96c65..HEAD
Error: /home/runner/work/nuttx/nuttx/nuttx/sched/semaphore/spinlock.c:167:6: error: Bad left brace alignment
Error: /home/runner/work/nuttx/nuttx/nuttx/sched/semaphore/spinlock.c:169:6: error: Bad right brace alignment
Error: /home/runner/work/nuttx/nuttx/nuttx/sched/semaphore/spinlock.c:174:6: error: Bad left brace alignment
Error: /home/runner/work/nuttx/nuttx/nuttx/sched/semaphore/spinlock.c:176:6: error: Bad right brace alignment
Error: /home/runner/work/nuttx/nuttx/nuttx/sched/semaphore/spinlock.c:[23](https://github.com/apache/nuttx/actions/runs/6427966184/job/17454400895?pr=10605#step:3:24)1:6: error: Bad left brace alignment
Error: /home/runner/work/nuttx/nuttx/nuttx/sched/semaphore/spinlock.c:233:6: error: Bad right brace alignment
Error: /home/runner/work/nuttx/nuttx/nuttx/sched/semaphore/spinlock.c:238:6: error: Bad left brace alignment
Error: /home/runner/work/nuttx/nuttx/nuttx/sched/semaphore/spinlock.c:[24](https://github.com/apache/nuttx/actions/runs/6427966184/job/17454400895?pr=10605#step:3:25)0:6: error: Bad right brace alignment

you can verify locally by:

./tools/checkpatch.sh -g HEAD~...HEAD

This is nxstyle issue. The formatting tool seems not to consider the case of nest curly brackets.
I can't find any reasonable format about this case.
The style like this.