sched/hrtimer: Fix functional correctness issues and improve the performance.

[Fix-Point]

Summary

This PR is part of the #17675, it introduces significant improvements to the high-resolution timer (hrtimer) subsystem and its integration with the watchdog timer system. The key changes include:

1. Guard Timer Implementation: Added a guard timer (g_hrtimer_guard) that always remains in the timer queue with maximum expiration time. This simplifies empty queue checks, reduces branch conditions, and provides health monitoring capabilities for time acquisition errors.

2. Watchdog-HRTimer Integration: Replaced the separate scheduler tick timer with direct hrtimer integration for watchdog processing. The watchdog now uses its own hrtimer (g_wdog_hrtimer) instead of a separate tick-based mechanism, simplifying the architecture.

3. Red-Black Tree Optimization: Fixed left-most node detection in the hrtimer RB-tree by introducing g_cached_first caching, eliminating incorrect NULL checks on left child nodes.

4. API Clarifications: Updated function comments and signatures for better documentation, particularly for hrtimer_cancel() and hrtimer_cancel_sync() to clarify ownership semantics and concurrency considerations.

5. Code Simplification: Removed redundant timer processing logic, inlined hrtimer_start() for performance, and consolidated timer expiration handling between tick and tickless modes.

Impact

• Performance: Reduced branch checking through guard timer improves hrtimer processing efficiency. Inlined hrtimer_start() eliminates at least 1 branch.
• Memory Footprint: Guard timer uses minimal additional memory while providing monitoring and performance benefits.
• Correctness: Fixed RB-tree left-most node detection and timer cancellation edge cases prevent missed expirations and hardware timer leaks.
• Architecture: Simplified timer subsystem by eliminating separate scheduler tick timer code (sched_timer.c) and unifying watchdog processing under hrtimer.
• Build System: Updated CMakeLists.txt and Make.defs to reflect source file changes (removed sched_timer.c, renamed sched_tickexpiration.c to sched_processtickless.c).
• API Compatibility: Maintains backward compatibility while clarifying documentation for safer usage patterns.
• Error Handling: Guard timer enables system recovery from time acquisition errors through configurable callback mechanisms.
• Concurrency: Improved documentation for cancel operations helps prevent race conditions in multi-CPU environments.

Test

Tested on rv-virt:smp, ostest passed, parallel stressed hrtimer test passed.

[wangchdo]

I see lots of update that is not necessary, do you have performance or function comparison between the current implementation and your change?

[Fix-Point]

I see lots of update that is not necessary, do you have performance or function comparison between the current implementation and your change?

I have spent a significant amount of time testing and fixing functional correctness issues in this implementation. These issues should have been fixed during the earlier testing phase or can be avoided by the implementation #17675. Please don't waste our time anymore.

As you suggested, I won't waste time "showing off" the performance improvements. If anyone else is interested, I'll provide the performance test data. If you don't believe me, you can test it yourself.

[xiaoxiang781216]

do we still need g_wdtimernested

[Fix-Point]

The in_expiration parameter was added because we can not restart the hrtimer within its callback function; instead, we should return the nanoseconds of the next delay. Its semantics are different from g_wdtimernested.