spin_lock_irqsave+sched_lock #14578
Merged
spin_lock_irqsave+sched_lock #14578
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
github-actions
bot
added
Area: OS Components
github-actions
bot
added
the
Arch: xtensa
acassis
reviewed
Oct 31, 2024
anchao
requested changes
Nov 1, 2024
hujun260
force-pushed
the
apache_3
branch
7 times, most recently
from
157736b to
175819e
Compare
|
@patacongo please review this patch which fix the long issue about sched lock. |
anchao
reviewed
Nov 19, 2024
anchao
reviewed
Nov 19, 2024
hujun260
force-pushed
the
apache_3
branch
2 times, most recently
from
87f9b88 to
cde599d
Compare
github-actions
bot
added
the
Arch: risc-v
hujun260
force-pushed
the
apache_3
branch
2 times, most recently
from
5e196c5 to
9cf3e7f
Compare
… with the _wo_note suffix. Signed-off-by: hujun5 <[email protected]>
…ed_[un]lock reason: 1 Accelerated the implementation of sched_lock, remove enter_critical_section in sched_lock and only enter_critical_section when task scheduling is required. 2 we add sched_lock_wo_note/sched_unlock_wo_note and it does not perform instrumentation logic Signed-off-by: hujun5 <[email protected]>
reason: We aim to replace big locks with smaller ones. So we will use spin_lock_irqsave extensively to replace enter_critical_section in the subsequent process. We imitate the implementation of Linux by adding sched_lock to spin_lock_irqsave in order to address scenarios where sem_post occurs within spin_lock_irqsave, which can lead to spinlock failures and deadlocks. Signed-off-by: hujun5 <[email protected]>
xiaoxiang781216
approved these changes
Jan 23, 2025
|
@hujun260 Great work, I think we can prepare a report showing the benefits of SMP mode after all spinlock changes |
anchao
added a commit
to anchao/nuttx
that referenced
this pull request
Jan 24, 2025
after below change merge to kernel, spin_lock() will turn off preemption by default, but this change is not applicable to all scenarios. The locations in the kernel that use spin_lock() extensively only require short critical sections and do not trigger scheduling, which leads to serious performance degradation of NuttX in AMP mode. In this PR, I try to expose similar problems and hope that each subsystem will carefully check the code coverage apache#14578 |commit b69111d |Author: hujun5 <[email protected]> |Date: Thu Jan 23 16:14:18 2025 +0800 | | spinlock: add sched_lock to spin_lock_irqsave | | reason: | We aim to replace big locks with smaller ones. So we will use spin_lock_irqsave extensively to | replace enter_critical_section in the subsequent process. We imitate the implementation of Linux | by adding sched_lock to spin_lock_irqsave in order to address scenarios where sem_post occurs | within spin_lock_irqsave, which can lead to spinlock failures and deadlocks. | | Signed-off-by: hujun5 <[email protected]> Signed-off-by: chao an <[email protected]>
xiaoxiang781216
pushed a commit
that referenced
this pull request
Jan 24, 2025
after below change merge to kernel, spin_lock() will turn off preemption by default, but this change is not applicable to all scenarios. The locations in the kernel that use spin_lock() extensively only require short critical sections and do not trigger scheduling, which leads to serious performance degradation of NuttX in AMP mode. In this PR, I try to expose similar problems and hope that each subsystem will carefully check the code coverage #14578 |commit b69111d |Author: hujun5 <[email protected]> |Date: Thu Jan 23 16:14:18 2025 +0800 | | spinlock: add sched_lock to spin_lock_irqsave | | reason: | We aim to replace big locks with smaller ones. So we will use spin_lock_irqsave extensively to | replace enter_critical_section in the subsequent process. We imitate the implementation of Linux | by adding sched_lock to spin_lock_irqsave in order to address scenarios where sem_post occurs | within spin_lock_irqsave, which can lead to spinlock failures and deadlocks. | | Signed-off-by: hujun5 <[email protected]> Signed-off-by: chao an <[email protected]>
1 task
|
Hi, Seems this PR is crashing most of our devices. My local tests and internal tests seems failing due to Here is the open issue #15688 |
Is this affecting all Espressif devices, @eren-terzioglu ? Can you please take a look as soon as possible, @hujun260 ? |
As far as I saw failing defconfigs are: Xtensa
Risc-V
|
|
ok, i will look into this issue |
hujun260
added a commit
to hujun260/nuttx
that referenced
this pull request
Jan 25, 2025
…69111d of apache#14578 reason: Due to the addition of sched_lock in the spinlock, using a spinlock in the *cpustart file during the boot phase is quite special. CPU0 waits for CPU1 to start up, using a spinlock as a multi-core synchronization strategy. However, the matching calls are not made within the same task, resulting in a mismatch in the scheduler lock count and preventing the system from booting. The sequence is: CPU0 spin_lock, spin_lock, spin_unlock; CPU1 spin_unlock. CPU0 and CPU1 are running different tasks. Signed-off-by: hujun5 <[email protected]>
hujun260
added a commit
to hujun260/nuttx
that referenced
this pull request
Jan 25, 2025
…69111d of apache#14578 reason: Due to the addition of sched_lock in the spinlock, using a spinlock in the *cpustart file during the boot phase is quite special. CPU0 waits for CPU1 to start up, using a spinlock as a multi-core synchronization strategy. However, the matching calls are not made within the same task, resulting in a mismatch in the scheduler lock count and preventing the system from booting. The sequence is: CPU0 spin_lock, spin_lock, spin_unlock; CPU1 spin_unlock. CPU0 and CPU1 are running different tasks. Signed-off-by: hujun5 <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
1 Accelerated the implementation of sched_lock, remove enter_critical_section in sched_lock and
only enter_critical_section when task scheduling is required.
2 we add sched_lock_wo_note/sched_unlock_wo_note and it does not perform instrumentation logic
3 We aim to replace big locks with smaller ones. So we will use spin_lock_irqsave extensively to
replace enter_critical_section in the subsequent process. We imitate the implementation of Linux
by adding sched_lock to spin_lock_irqsave in order to address scenarios where sem_post occurs
within spin_lock_irqsave, which can lead to spinlock failures and deadlocks.
The entire implementation process includes:
1 spin_lock_irqsave + sched_lock
2 spin_lock/rw/spin_trylock + sched_lock
3 enter_critical_section + sched_lock
We are currently implementing the first step.
Impact
spinlock and sched_lock
Testing
Build Host:
Configuring NuttX and compile:
$ ./tools/configure.sh -l qemu-armv8a:nsh_smp
$ make
Running with qemu
$ qemu-system-aarch64 -cpu cortex-a53 -smp 4 -nographic
-machine virt,virtualization=on,gic-version=3
-net none -chardev stdio,id=con,mux=on -serial chardev:con
-mon chardev=con,mode=readline -kernel ./nuttx