hujun260 opened a new pull request, #17959:
URL: https://github.com/apache/nuttx/pull/17959
## Summary
Replace global critical section protection with fine-grained spinlock
protection for signal management structures. Signal action queue
(tg_sigactionq) and signal pending queue (tg_sigpendingq) are now protected
using `spin_lock_irqsave/spin_unlock_irqrestore(&group->tg_lock)` instead of
`enter/leave_critical_section()`. This improves interrupt latency and SMP
scalability while maintaining thread-safe access to signal data structures.
## Changes
- **sched/signal/sig_action.c**:
- Add spinlock protection around signal action queue modifications (sq_rem
and sq_addlast)
- Use `spin_lock_irqsave/spin_unlock_irqrestore(&group->tg_lock)` around
queue operations
- Move `irqstate_t flags` declaration to function scope for consistency
- **sched/signal/sig_cleanup.c**:
- Add `#include <nuttx/spinlock.h>` header
- Add spinlock protection in nxsig_release() for signal action and pending
signal queue cleanup
- Acquire/release lock around each dequeue operation to prevent deadlock
during cleanup
- **sched/signal/sig_dispatch.c**:
- Add spinlock protection in nxsig_find_pendingsignal() to protect pending
signal list traversal
- Add spinlock protection in nxsig_add_pendingsignal() around pending
signal queue insertion
- Include comment explaining why lock is needed for interrupt-level signal
additions
- **sched/signal/sig_pending.c**:
- Replace `enter_critical_section()` with
`spin_lock_irqsave(&group->tg_lock)`
- Replace `leave_critical_section()` with
`spin_unlock_irqrestore(&group->tg_lock, flags)`
- Maintains same semantics with reduced interrupt latency
- **sched/signal/sig_removependingsignal.c**:
- Add `#include <nuttx/spinlock.h>` header
- Replace global critical section with spinlock on signal pending queue
access
- Protect pending signal removal operations with group-level lock
## Benefits & Technical Details
- **Reduced interrupt latency**: Spinlock-based synchronization keeps
interrupts enabled during most operations, allowing faster interrupt response
- **SMP scalability**: Fine-grained per-group locking enables parallel
signal operations on different task groups
- **Interrupt-safe**: Spinlock protects against simultaneous access from
interrupt handlers and task context
- **Lock scope optimization**: In cleanup path, lock is released between
queue element processing to prevent excessive lock hold times
- **Consistency**: All signal queue accesses now use uniform spinlock
protection pattern
## Testing
- Verified signal delivery and pending signal operations work correctly in
single-threaded context
- Confirmed proper synchronization when signals are sent from multiple
threads in same group
- Tested interrupt-safe pending signal addition from IRQ handlers
- Validated on SMP systems that signal queues are accessed safely without
global interrupt disabling
- Confirmed no performance regression in signal dispatch path
- Verified signal cleanup during task group termination works correctly with
spinlock protection
## Impact
- **Performance**: Reduced interrupt latency and improved context switching
speed on SMP systems
- **Real-time**: Better response time for external interrupts during signal
operations
- **Robustness**: Proper synchronization for concurrent signal operations
from interrupt and task contexts
- **Compatibility**: No API changes, fully backward compatible
- **Scope**: Affects all signal delivery, pending signal management, and
signal action installation operations
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
