Linux hrtimer的實現

1. Linux hrtimer的實現方案

Linux hrtimer的實現是依賴硬件(通過可編程定時器來實現)的支持的，而且此定時器有自己的專用寄存器， 硬中斷和頻率。比如我的板子上的對應參數如下：

Timer at Vir:0xE0100200 = Phy:0xE0100200, using Irq:27, at Freq:250000000，由此可見，其頻率為250MHz，所以其精度為:1/250000000=4ns，比系統時鐘jiffy(HZ=100,精度為10ms)的精度高得太多了。可是支持此高精度timer是需要付出硬件成本的。即它是一個硬件時鐘。這裡所說的硬件時鐘特指的是硬件計時器時鐘。

2. 硬件時鐘 數據結構
　　和硬件計時器（本文又稱作硬件時鐘，區別於軟件時鐘）相關的數據結構主要有兩個：
　　struct clocksource ：對硬件設備的抽象，描述時鐘源信息

1. struct clocksource {
2. /*
3. * First part of structure is read mostly
4. */
5. char *name;
6. struct list_head list;
7. int rating;
8. cycle_t (*read)(struct clocksource *cs);
9. int (*enable)(struct clocksource *cs);
10. void (*disable)(struct clocksource *cs);
11. cycle_t mask;
12. u32 mult;
13. u32 shift;
14. u64 max_idle_ns;
15. unsigned long flags;
16. cycle_t (*vread)(void);
17. void (*suspend)(struct clocksource *cs);
18. void (*resume)(struct clocksource *cs);
19. #ifdef CONFIG_IA64
20. void *fsys_mmio; /* used by fsyscall asm code */
21. #define CLKSRC_FSYS_MMIO_SET(mmio, addr) ((mmio) = (addr))
22. #else
23. #define CLKSRC_FSYS_MMIO_SET(mmio, addr) do { } while (0)
24. #endif
26. /*
27. * Second part is written at each timer interrupt
28. * Keep it in a different cache line to dirty no
29. * more than one cache line.
30. */
31. cycle_t cycle_last ____cacheline_aligned_in_smp;
33. #ifdef CONFIG_CLOCKSOURCE_WATCHDOG
34. /* Watchdog related data, used by the framework */
35. struct list_head wd_list;
36. cycle_t wd_last;
37. #endif
38. };

struct clock_event_device ：時鐘的事件信息，包括當硬件時鐘中斷發生時要執行那些操作（實際上保存了相應函數的指針）。本文將該結構稱作為“時鐘事件設備”。

1. /**
2. * struct clock_event_device - clock event device descriptor
3. * @name: ptr to clock event name
4. * @features: features
5. * @max_delta_ns: maximum delta value in ns
6. * @min_delta_ns: minimum delta value in ns
7. * @mult: nanosecond to cycles multiplier
8. * @shift: nanoseconds to cycles divisor (power of two)
9. * @rating: variable to rate clock event devices
10. * @irq: IRQ number (only for non CPU local devices)
11. * @cpumask: cpumask to indicate for which CPUs this device works
12. * @set_next_event: set next event function
13. * @set_mode: set mode function
14. * @event_handler: Assigned by the framework to be called by the low
15. * level handler of the event source
16. * @broadcast: function to broadcast events
17. * @list: list head for the management code
18. * @mode: operating mode assigned by the management code
19. * @next_event: local storage for the next event in oneshot mode
20. * @retries: number of forced programming retries
21. */
22. struct clock_event_device {
23. const char *name;
24. unsigned int features;
25. u64 max_delta_ns;
26. u64 min_delta_ns;
27. u32 mult;
28. u32 shift;
29. int rating;
30. int irq;
31. const struct cpumask *cpumask;
32. int (*set_next_event)(unsigned long evt,
33. struct clock_event_device *);
34. void (*set_mode)(enum clock_event_mode mode,
35. struct clock_event_device *);
36. void (*event_handler)(struct clock_event_device *);
37. void (*broadcast)(const struct cpumask *mask);
38. struct list_head list;
39. enum clock_event_mode mode;
40. ktime_t next_event;
41. unsigned long retries;
42. };

上述兩個結構內核源代碼中有較詳細的注解，分別位於文件 clocksource.h 和 clockchips.h 中。需要特別注意的是結構 clock_event_device 的成員 event_handler ，它指定了當硬件時鐘中斷發生時，內核應該執行那些操作，也就是真正的時鐘中斷處理函數。

Linux 內核維護了兩個鏈表，分別存儲了系統中所有時鐘源的信息和時鐘事件設備的信息。這兩個鏈表的表頭在內核中分別是 clocksource_list 和 clockevent_devices 。