API tools faq
paste
Advertisement
Guest User

Untitled

a guest
Jan 17th, 2010
190
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 47.12 KB | None | 0 0
raw download clone embed print report
  1. cat ./linux/arch/x86/kernel/time_32-xen.c
  2. /*
  3. * Copyright (C) 1991, 1992, 1995 Linus Torvalds
  4. *
  5. * This file contains the PC-specific time handling details:
  6. * reading the RTC at bootup, etc..
  7. * 1994-07-02 Alan Modra
  8. * fixed set_rtc_mmss, fixed time.year for >= 2000, new mktime
  9. * 1995-03-26 Markus Kuhn
  10. * fixed 500 ms bug at call to set_rtc_mmss, fixed DS12887
  11. * precision CMOS clock update
  12. * 1996-05-03 Ingo Molnar
  13. * fixed time warps in do_[slow|fast]_gettimeoffset()
  14. * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
  15. * "A Kernel Model for Precision Timekeeping" by Dave Mills
  16. * 1998-09-05 (Various)
  17. * More robust do_fast_gettimeoffset() algorithm implemented
  18. * (works with APM, Cyrix 6x86MX and Centaur C6),
  19. * monotonic gettimeofday() with fast_get_timeoffset(),
  20. * drift-proof precision TSC calibration on boot
  21. * (C. Scott Ananian <cananian@alumni.princeton.edu>, Andrew D.
  22. * Balsa <andrebalsa@altern.org>, Philip Gladstone <philip@raptor.com>;
  23. * ported from 2.0.35 Jumbo-9 by Michael Krause <m.krause@tu-harburg.de>).
  24. * 1998-12-16 Andrea Arcangeli
  25. * Fixed Jumbo-9 code in 2.1.131: do_gettimeofday was missing 1 jiffy
  26. * because was not accounting lost_ticks.
  27. * 1998-12-24 Copyright (C) 1998 Andrea Arcangeli
  28. * Fixed a xtime SMP race (we need the xtime_lock rw spinlock to
  29. * serialize accesses to xtime/lost_ticks).
  30. */
  31.  
  32. #include <linux/init.h>
  33. #include <linux/interrupt.h>
  34. #include <linux/time.h>
  35. #include <linux/mca.h>
  36. #include <linux/sysctl.h>
  37. #include <linux/percpu.h>
  38. #include <linux/kernel_stat.h>
  39. #include <linux/posix-timers.h>
  40. #include <linux/cpufreq.h>
  41. #include <linux/clocksource.h>
  42. #include <linux/sysdev.h>
  43.  
  44. #include <asm/delay.h>
  45. #include <asm/time.h>
  46. #include <asm/timer.h>
  47.  
  48. #include <xen/evtchn.h>
  49. #include <xen/sysctl.h>
  50. #include <xen/interface/vcpu.h>
  51.  
  52. #include <asm/i8253.h>
  53. DEFINE_SPINLOCK(i8253_lock);
  54. EXPORT_SYMBOL(i8253_lock);
  55.  
  56. #ifdef CONFIG_X86_64
  57. #include <asm/vsyscall.h>
  58. volatile unsigned long __jiffies __section_jiffies = INITIAL_JIFFIES;
  59. #endif
  60.  
  61. #define XEN_SHIFT 22
  62.  
  63. unsigned int cpu_khz; /* Detected as we calibrate the TSC */
  64. EXPORT_SYMBOL(cpu_khz);
  65.  
  66. /* These are peridically updated in shared_info, and then copied here. */
  67. struct shadow_time_info {
  68. u64 tsc_timestamp; /* TSC at last update of time vals. */
  69. u64 system_timestamp; /* Time, in nanosecs, since boot. */
  70. u32 tsc_to_nsec_mul;
  71. u32 tsc_to_usec_mul;
  72. int tsc_shift;
  73. u32 version;
  74. };
  75. static DEFINE_PER_CPU(struct shadow_time_info, shadow_time);
  76. static struct timespec shadow_tv;
  77. static u32 shadow_tv_version;
  78.  
  79. /* Keep track of last time we did processing/updating of jiffies and xtime. */
  80. static u64 processed_system_time; /* System time (ns) at last processing. */
  81. static DEFINE_PER_CPU(u64, processed_system_time);
  82.  
  83. /* How much CPU time was spent blocked and how much was 'stolen'? */
  84. static DEFINE_PER_CPU(u64, processed_stolen_time);
  85. static DEFINE_PER_CPU(u64, processed_blocked_time);
  86.  
  87. /* Current runstate of each CPU (updated automatically by the hypervisor). */
  88. static DEFINE_PER_CPU(struct vcpu_runstate_info, runstate);
  89.  
  90. /* Must be signed, as it's compared with s64 quantities which can be -ve. */
  91. #define NS_PER_TICK (1000000000LL/HZ)
  92.  
  93. static struct vcpu_set_periodic_timer xen_set_periodic_tick = {
  94. .period_ns = NS_PER_TICK
  95. };
  96.  
  97. static void __clock_was_set(struct work_struct *unused)
  98. {
  99. clock_was_set();
  100. }
  101. static DECLARE_WORK(clock_was_set_work, __clock_was_set);
  102.  
  103. /*
  104. * GCC 4.3 can turn loops over an induction variable into division. We do
  105. * not support arbitrary 64-bit division, and so must break the induction.
  106. */
  107. #define clobber_induction_variable(v) asm ( "" : "+r" (v) )
  108.  
  109. static inline void __normalize_time(time_t *sec, s64 *nsec)
  110. {
  111. while (*nsec >= NSEC_PER_SEC) {
  112. clobber_induction_variable(*nsec);
  113. (*nsec) -= NSEC_PER_SEC;
  114. (*sec)++;
  115. }
  116. while (*nsec < 0) {
  117. clobber_induction_variable(*nsec);
  118. (*nsec) += NSEC_PER_SEC;
  119. (*sec)--;
  120. }
  121. }
  122.  
  123. /* Does this guest OS track Xen time, or set its wall clock independently? */
  124. static int independent_wallclock = 0;
  125. static int __init __independent_wallclock(char *str)
  126. {
  127. independent_wallclock = 1;
  128. return 1;
  129. }
  130. __setup("independent_wallclock", __independent_wallclock);
  131.  
  132. int xen_independent_wallclock(void)
  133. {
  134. return independent_wallclock;
  135. }
  136.  
  137. /* Permitted clock jitter, in nsecs, beyond which a warning will be printed. */
  138. static unsigned long permitted_clock_jitter = 10000000UL; /* 10ms */
  139. static int __init __permitted_clock_jitter(char *str)
  140. {
  141. permitted_clock_jitter = simple_strtoul(str, NULL, 0);
  142. return 1;
  143. }
  144. __setup("permitted_clock_jitter=", __permitted_clock_jitter);
  145.  
  146. /*
  147. * Scale a 64-bit delta by scaling and multiplying by a 32-bit fraction,
  148. * yielding a 64-bit result.
  149. */
  150. static inline u64 scale_delta(u64 delta, u32 mul_frac, int shift)
  151. {
  152. u64 product;
  153. #ifdef __i386__
  154. u32 tmp1, tmp2;
  155. #endif
  156.  
  157. if (shift < 0)
  158. delta >>= -shift;
  159. else
  160. delta <<= shift;
  161.  
  162. #ifdef __i386__
  163. __asm__ (
  164. "mul %5 ; "
  165. "mov %4,%%eax ; "
  166. "mov %%edx,%4 ; "
  167. "mul %5 ; "
  168. "xor %5,%5 ; "
  169. "add %4,%%eax ; "
  170. "adc %5,%%edx ; "
  171. : "=A" (product), "=r" (tmp1), "=r" (tmp2)
  172. : "a" ((u32)delta), "1" ((u32)(delta >> 32)), "2" (mul_frac) );
  173. #else
  174. __asm__ (
  175. "mul %%rdx ; shrd $32,%%rdx,%%rax"
  176. : "=a" (product) : "0" (delta), "d" ((u64)mul_frac) );
  177. #endif
  178.  
  179. return product;
  180. }
  181.  
  182. static inline u64 get64(volatile u64 *ptr)
  183. {
  184. #ifndef CONFIG_64BIT
  185. return cmpxchg64(ptr, 0, 0);
  186. #else
  187. return *ptr;
  188. #endif
  189. }
  190.  
  191. static inline u64 get64_local(volatile u64 *ptr)
  192. {
  193. #ifndef CONFIG_64BIT
  194. return cmpxchg64_local(ptr, 0, 0);
  195. #else
  196. return *ptr;
  197. #endif
  198. }
  199.  
  200. static void init_cpu_khz(void)
  201. {
  202. u64 __cpu_khz = 1000000ULL << 32;
  203. struct vcpu_time_info *info = &vcpu_info(0)->time;
  204. do_div(__cpu_khz, info->tsc_to_system_mul);
  205. if (info->tsc_shift < 0)
  206. cpu_khz = __cpu_khz << -info->tsc_shift;
  207. else
  208. cpu_khz = __cpu_khz >> info->tsc_shift;
  209. }
  210.  
  211. static u64 get_nsec_offset(struct shadow_time_info *shadow)
  212. {
  213. u64 now, delta;
  214. rdtscll(now);
  215. delta = now - shadow->tsc_timestamp;
  216. return scale_delta(delta, shadow->tsc_to_nsec_mul, shadow->tsc_shift);
  217. }
  218.  
  219. static void __update_wallclock(time_t sec, long nsec)
  220. {
  221. long wtm_nsec, xtime_nsec;
  222. time_t wtm_sec, xtime_sec;
  223. u64 tmp, wc_nsec;
  224.  
  225. /* Adjust wall-clock time base. */
  226. wc_nsec = processed_system_time;
  227. wc_nsec += sec * (u64)NSEC_PER_SEC;
  228. wc_nsec += nsec;
  229.  
  230. /* Split wallclock base into seconds and nanoseconds. */
  231. tmp = wc_nsec;
  232. xtime_nsec = do_div(tmp, 1000000000);
  233. xtime_sec = (time_t)tmp;
  234.  
  235. wtm_sec = wall_to_monotonic.tv_sec + (xtime.tv_sec - xtime_sec);
  236. wtm_nsec = wall_to_monotonic.tv_nsec + (xtime.tv_nsec - xtime_nsec);
  237.  
  238. set_normalized_timespec(&xtime, xtime_sec, xtime_nsec);
  239. set_normalized_timespec(&wall_to_monotonic, wtm_sec, wtm_nsec);
  240. }
  241.  
  242. static void update_wallclock(void)
  243. {
  244. shared_info_t *s = HYPERVISOR_shared_info;
  245.  
  246. do {
  247. shadow_tv_version = s->wc_version;
  248. rmb();
  249. shadow_tv.tv_sec = s->wc_sec;
  250. shadow_tv.tv_nsec = s->wc_nsec;
  251. rmb();
  252. } while ((s->wc_version & 1) | (shadow_tv_version ^ s->wc_version));
  253.  
  254. if (!independent_wallclock)
  255. __update_wallclock(shadow_tv.tv_sec, shadow_tv.tv_nsec);
  256. }
  257.  
  258. /*
  259. * Reads a consistent set of time-base values from Xen, into a shadow data
  260. * area.
  261. */
  262. static void get_time_values_from_xen(unsigned int cpu)
  263. {
  264. struct vcpu_time_info *src;
  265. struct shadow_time_info *dst;
  266. unsigned long flags;
  267. u32 pre_version, post_version;
  268.  
  269. src = &vcpu_info(cpu)->time;
  270. dst = &per_cpu(shadow_time, cpu);
  271.  
  272. local_irq_save(flags);
  273.  
  274. do {
  275. pre_version = dst->version = src->version;
  276. rmb();
  277. dst->tsc_timestamp = src->tsc_timestamp;
  278. dst->system_timestamp = src->system_time;
  279. dst->tsc_to_nsec_mul = src->tsc_to_system_mul;
  280. dst->tsc_shift = src->tsc_shift;
  281. rmb();
  282. post_version = src->version;
  283. } while ((pre_version & 1) | (pre_version ^ post_version));
  284.  
  285. dst->tsc_to_usec_mul = dst->tsc_to_nsec_mul / 1000;
  286.  
  287. local_irq_restore(flags);
  288. }
  289.  
  290. static inline int time_values_up_to_date(void)
  291. {
  292. rmb();
  293. return percpu_read(shadow_time.version) == vcpu_info_read(time.version);
  294. }
  295.  
  296. static void sync_xen_wallclock(unsigned long dummy);
  297. static DEFINE_TIMER(sync_xen_wallclock_timer, sync_xen_wallclock, 0, 0);
  298. static void sync_xen_wallclock(unsigned long dummy)
  299. {
  300. time_t sec;
  301. s64 nsec;
  302. struct xen_platform_op op;
  303.  
  304. BUG_ON(!is_initial_xendomain());
  305. if (!ntp_synced() || independent_wallclock)
  306. return;
  307.  
  308. write_seqlock_irq(&xtime_lock);
  309.  
  310. sec = xtime.tv_sec;
  311. nsec = xtime.tv_nsec;
  312. __normalize_time(&sec, &nsec);
  313.  
  314. op.cmd = XENPF_settime;
  315. op.u.settime.secs = sec;
  316. op.u.settime.nsecs = nsec;
  317. op.u.settime.system_time = processed_system_time;
  318. WARN_ON(HYPERVISOR_platform_op(&op));
  319.  
  320. update_wallclock();
  321.  
  322. write_sequnlock_irq(&xtime_lock);
  323.  
  324. /* Once per minute. */
  325. mod_timer(&sync_xen_wallclock_timer, jiffies + 60*HZ);
  326. }
  327.  
  328. static unsigned long long local_clock(void)
  329. {
  330. unsigned int cpu = get_cpu();
  331. struct shadow_time_info *shadow = &per_cpu(shadow_time, cpu);
  332. u64 time;
  333. u32 local_time_version;
  334.  
  335. do {
  336. local_time_version = shadow->version;
  337. rdtsc_barrier();
  338. time = shadow->system_timestamp + get_nsec_offset(shadow);
  339. if (!time_values_up_to_date())
  340. get_time_values_from_xen(cpu);
  341. barrier();
  342. } while (local_time_version != shadow->version);
  343.  
  344. put_cpu();
  345.  
  346. return time;
  347. }
  348.  
  349. /*
  350. * Runstate accounting
  351. */
  352. static void get_runstate_snapshot(struct vcpu_runstate_info *res)
  353. {
  354. u64 state_time;
  355. struct vcpu_runstate_info *state;
  356.  
  357. BUG_ON(preemptible());
  358.  
  359. state = &__get_cpu_var(runstate);
  360.  
  361. do {
  362. state_time = get64_local(&state->state_entry_time);
  363. *res = *state;
  364. } while (get64_local(&state->state_entry_time) != state_time);
  365.  
  366. WARN_ON_ONCE(res->state != RUNSTATE_running);
  367. }
  368.  
  369. /*
  370. * Xen sched_clock implementation. Returns the number of unstolen
  371. * nanoseconds, which is nanoseconds the VCPU spent in RUNNING+BLOCKED
  372. * states.
  373. */
  374. unsigned long long sched_clock(void)
  375. {
  376. struct vcpu_runstate_info runstate;
  377. cycle_t now;
  378. u64 ret;
  379. s64 offset;
  380.  
  381. /*
  382. * Ideally sched_clock should be called on a per-cpu basis
  383. * anyway, so preempt should already be disabled, but that's
  384. * not current practice at the moment.
  385. */
  386. preempt_disable();
  387.  
  388. now = local_clock();
  389.  
  390. get_runstate_snapshot(&runstate);
  391.  
  392. offset = now - runstate.state_entry_time;
  393. if (offset < 0)
  394. offset = 0;
  395.  
  396. ret = offset + runstate.time[RUNSTATE_running]
  397. + runstate.time[RUNSTATE_blocked];
  398.  
  399. preempt_enable();
  400.  
  401. return ret;
  402. }
  403.  
  404. unsigned long profile_pc(struct pt_regs *regs)
  405. {
  406. unsigned long pc = instruction_pointer(regs);
  407.  
  408. #if defined(CONFIG_SMP) || defined(__x86_64__)
  409. if (!user_mode_vm(regs) && in_lock_functions(pc)) {
  410. # ifdef CONFIG_FRAME_POINTER
  411. return *(unsigned long *)(regs->bp + sizeof(long));
  412. # else
  413. # ifdef __i386__
  414. unsigned long *sp = (unsigned long *)&regs->sp;
  415. # else
  416. unsigned long *sp = (unsigned long *)regs->sp;
  417. # endif
  418.  
  419. /* Return address is either directly at stack pointer
  420. or above a saved flags. Eflags has bits 22-31 zero,
  421. kernel addresses don't. */
  422. if (sp[0] >> 22)
  423. return sp[0];
  424. if (sp[1] >> 22)
  425. return sp[1];
  426. # endif
  427. }
  428. #endif
  429.  
  430. return pc;
  431. }
  432. EXPORT_SYMBOL(profile_pc);
  433.  
  434. /*
  435. * This is the same as the above, except we _also_ save the current
  436. * Time Stamp Counter value at the time of the timer interrupt, so that
  437. * we later on can estimate the time of day more exactly.
  438. */
  439. irqreturn_t timer_interrupt(int irq, void *dev_id)
  440. {
  441. s64 delta, delta_cpu, stolen, blocked;
  442. unsigned int i, cpu = smp_processor_id();
  443. struct shadow_time_info *shadow = &per_cpu(shadow_time, cpu);
  444. struct vcpu_runstate_info runstate;
  445.  
  446. /* Keep nmi watchdog up to date */
  447. inc_irq_stat(irq0_irqs);
  448.  
  449. /*
  450. * Here we are in the timer irq handler. We just have irqs locally
  451. * disabled but we don't know if the timer_bh is running on the other
  452. * CPU. We need to avoid to SMP race with it. NOTE: we don' t need
  453. * the irq version of write_lock because as just said we have irq
  454. * locally disabled. -arca
  455. */
  456. write_seqlock(&xtime_lock);
  457.  
  458. do {
  459. get_time_values_from_xen(cpu);
  460.  
  461. /* Obtain a consistent snapshot of elapsed wallclock cycles. */
  462. delta = delta_cpu =
  463. shadow->system_timestamp + get_nsec_offset(shadow);
  464. delta -= processed_system_time;
  465. delta_cpu -= per_cpu(processed_system_time, cpu);
  466.  
  467. get_runstate_snapshot(&runstate);
  468. } while (!time_values_up_to_date());
  469.  
  470. if ((unlikely(delta < -(s64)permitted_clock_jitter) ||
  471. unlikely(delta_cpu < -(s64)permitted_clock_jitter))
  472. && printk_ratelimit()) {
  473. printk("Timer ISR/%u: Time went backwards: "
  474. "delta=%lld delta_cpu=%lld shadow=%lld "
  475. "off=%lld processed=%lld cpu_processed=%lld\n",
  476. cpu, delta, delta_cpu, shadow->system_timestamp,
  477. (s64)get_nsec_offset(shadow),
  478. processed_system_time,
  479. per_cpu(processed_system_time, cpu));
  480. for (i = 0; i < num_online_cpus(); i++)
  481. printk(" %d: %lld\n", i,
  482. per_cpu(processed_system_time, i));
  483. }
  484.  
  485. /* System-wide jiffy work. */
  486. if (delta >= NS_PER_TICK) {
  487. do_div(delta, NS_PER_TICK);
  488. processed_system_time += delta * NS_PER_TICK;
  489. while (delta > HZ) {
  490. clobber_induction_variable(delta);
  491. do_timer(HZ);
  492. delta -= HZ;
  493. }
  494. do_timer(delta);
  495. }
  496.  
  497. if (shadow_tv_version != HYPERVISOR_shared_info->wc_version) {
  498. update_wallclock();
  499. if (keventd_up())
  500. schedule_work(&clock_was_set_work);
  501. }
  502.  
  503. write_sequnlock(&xtime_lock);
  504.  
  505. /*
  506. * Account stolen ticks.
  507. * ensures that the ticks are accounted as stolen.
  508. */
  509. stolen = runstate.time[RUNSTATE_runnable]
  510. + runstate.time[RUNSTATE_offline]
  511. - per_cpu(processed_stolen_time, cpu);
  512. if ((stolen > 0) && (delta_cpu > 0)) {
  513. delta_cpu -= stolen;
  514. if (unlikely(delta_cpu < 0))
  515. stolen += delta_cpu; /* clamp local-time progress */
  516. do_div(stolen, NS_PER_TICK);
  517. per_cpu(processed_stolen_time, cpu) += stolen * NS_PER_TICK;
  518. per_cpu(processed_system_time, cpu) += stolen * NS_PER_TICK;
  519. account_steal_time((cputime_t)stolen);
  520. }
  521.  
  522. /*
  523. * Account blocked ticks.
  524. * ensures that the ticks are accounted as idle/wait.
  525. */
  526. blocked = runstate.time[RUNSTATE_blocked]
  527. - per_cpu(processed_blocked_time, cpu);
  528. if ((blocked > 0) && (delta_cpu > 0)) {
  529. delta_cpu -= blocked;
  530. if (unlikely(delta_cpu < 0))
  531. blocked += delta_cpu; /* clamp local-time progress */
  532. do_div(blocked, NS_PER_TICK);
  533. per_cpu(processed_blocked_time, cpu) += blocked * NS_PER_TICK;
  534. per_cpu(processed_system_time, cpu) += blocked * NS_PER_TICK;
  535. account_idle_time((cputime_t)blocked);
  536. }
  537.  
  538. /* Account user/system ticks. */
  539. if (delta_cpu > 0) {
  540. do_div(delta_cpu, NS_PER_TICK);
  541. per_cpu(processed_system_time, cpu) += delta_cpu * NS_PER_TICK;
  542. if (user_mode_vm(get_irq_regs()))
  543. account_user_time(current, (cputime_t)delta_cpu,
  544. (cputime_t)delta_cpu);
  545. else if (current != idle_task(cpu))
  546. account_system_time(current, HARDIRQ_OFFSET,
  547. (cputime_t)delta_cpu,
  548. (cputime_t)delta_cpu);
  549. else
  550. account_idle_time((cputime_t)delta_cpu);
  551. }
  552.  
  553. /* Offlined for more than a few seconds? Avoid lockup warnings. */
  554. if (stolen > 5*HZ)
  555. touch_softlockup_watchdog();
  556.  
  557. /* Local timer processing (see update_process_times()). */
  558. run_local_timers();
  559. if (rcu_pending(cpu))
  560. rcu_check_callbacks(cpu, user_mode_vm(get_irq_regs()));
  561. printk_tick();
  562. scheduler_tick();
  563. run_posix_cpu_timers(current);
  564. profile_tick(CPU_PROFILING);
  565.  
  566. return IRQ_HANDLED;
  567. }
  568.  
  569. void mark_tsc_unstable(char *reason)
  570. {
  571. #ifndef CONFIG_XEN /* XXX Should tell the hypervisor about this fact. */
  572. tsc_unstable = 1;
  573. #endif
  574. }
  575. EXPORT_SYMBOL_GPL(mark_tsc_unstable);
  576.  
  577. static void init_missing_ticks_accounting(unsigned int cpu)
  578. {
  579. struct vcpu_register_runstate_memory_area area;
  580. struct vcpu_runstate_info *runstate = &per_cpu(runstate, cpu);
  581. int rc;
  582.  
  583. memset(runstate, 0, sizeof(*runstate));
  584.  
  585. area.addr.v = runstate;
  586. rc = HYPERVISOR_vcpu_op(VCPUOP_register_runstate_memory_area, cpu, &area);
  587. WARN_ON(rc && rc != -ENOSYS);
  588.  
  589. per_cpu(processed_blocked_time, cpu) =
  590. runstate->time[RUNSTATE_blocked];
  591. per_cpu(processed_stolen_time, cpu) =
  592. runstate->time[RUNSTATE_runnable] +
  593. runstate->time[RUNSTATE_offline];
  594. }
  595.  
  596. static cycle_t cs_last;
  597.  
  598. static cycle_t xen_clocksource_read(struct clocksource *cs)
  599. {
  600. #ifdef CONFIG_SMP
  601. cycle_t last = get64(&cs_last);
  602. cycle_t ret = local_clock();
  603.  
  604. if (unlikely((s64)(ret - last) < 0)) {
  605. if (last - ret > permitted_clock_jitter
  606. && printk_ratelimit()) {
  607. unsigned int cpu = get_cpu();
  608. struct shadow_time_info *shadow = &per_cpu(shadow_time, cpu);
  609.  
  610. printk(KERN_WARNING "clocksource/%u: "
  611. "Time went backwards: "
  612. "ret=%Lx delta=%Ld shadow=%Lx offset=%Lx\n",
  613. cpu, ret, ret - last, shadow->system_timestamp,
  614. get_nsec_offset(shadow));
  615. put_cpu();
  616. }
  617. return last;
  618. }
  619.  
  620. for (;;) {
  621. cycle_t cur = cmpxchg64(&cs_last, last, ret);
  622.  
  623. if (cur == last || (s64)(ret - cur) < 0)
  624. return ret;
  625. last = cur;
  626. }
  627. #else
  628. return local_clock();
  629. #endif
  630. }
  631.  
  632. /* No locking required. Interrupts are disabled on all CPUs. */
  633. static void xen_clocksource_resume(void)
  634. {
  635. unsigned int cpu;
  636.  
  637. init_cpu_khz();
  638.  
  639. for_each_online_cpu(cpu) {
  640. switch (HYPERVISOR_vcpu_op(VCPUOP_set_periodic_timer, cpu,
  641. &xen_set_periodic_tick)) {
  642. case 0:
  643. #if CONFIG_XEN_COMPAT <= 0x030004
  644. case -ENOSYS:
  645. #endif
  646. break;
  647. default:
  648. BUG();
  649. }
  650. get_time_values_from_xen(cpu);
  651. per_cpu(processed_system_time, cpu) =
  652. per_cpu(shadow_time, 0).system_timestamp;
  653. init_missing_ticks_accounting(cpu);
  654. }
  655.  
  656. processed_system_time = per_cpu(shadow_time, 0).system_timestamp;
  657.  
  658. cs_last = local_clock();
  659. }
  660.  
  661. static struct clocksource clocksource_xen = {
  662. .name = "xen",
  663. .rating = 400,
  664. .read = xen_clocksource_read,
  665. .mask = CLOCKSOURCE_MASK(64),
  666. .mult = 1 << XEN_SHIFT, /* time directly in nanoseconds */
  667. .shift = XEN_SHIFT,
  668. .flags = CLOCK_SOURCE_IS_CONTINUOUS,
  669. .resume = xen_clocksource_resume,
  670. };
  671.  
  672. unsigned long xen_read_persistent_clock(void)
  673. {
  674. const shared_info_t *s = HYPERVISOR_shared_info;
  675. u32 version, sec, nsec;
  676. u64 delta;
  677.  
  678. do {
  679. version = s->wc_version;
  680. rmb();
  681. sec = s->wc_sec;
  682. nsec = s->wc_nsec;
  683. rmb();
  684. } while ((s->wc_version & 1) | (version ^ s->wc_version));
  685.  
  686. delta = local_clock() + (u64)sec * NSEC_PER_SEC + nsec;
  687. do_div(delta, NSEC_PER_SEC);
  688.  
  689. return delta;
  690. }
  691.  
  692. int xen_update_persistent_clock(void)
  693. {
  694. if (!is_initial_xendomain())
  695. return -1;
  696. mod_timer(&sync_xen_wallclock_timer, jiffies + 1);
  697. return 0;
  698. }
  699.  
  700. /* Dynamically-mapped IRQ. */
  701. static int __read_mostly timer_irq = -1;
  702. static struct irqaction timer_action = {
  703. .handler = timer_interrupt,
  704. .flags = IRQF_DISABLED|IRQF_TIMER,
  705. .name = "timer"
  706. };
  707.  
  708. static void __init setup_cpu0_timer_irq(void)
  709. {
  710. timer_irq = bind_virq_to_irqaction(VIRQ_TIMER, 0, &timer_action);
  711. BUG_ON(timer_irq < 0);
  712. }
  713.  
  714. void __init time_init(void)
  715. {
  716. init_cpu_khz();
  717. printk(KERN_INFO "Xen reported: %u.%03u MHz processor.\n",
  718. cpu_khz / 1000, cpu_khz % 1000);
  719.  
  720. switch (HYPERVISOR_vcpu_op(VCPUOP_set_periodic_timer, 0,
  721. &xen_set_periodic_tick)) {
  722. case 0:
  723. #if CONFIG_XEN_COMPAT <= 0x030004
  724. case -ENOSYS:
  725. #endif
  726. break;
  727. default:
  728. BUG();
  729. }
  730.  
  731. get_time_values_from_xen(0);
  732.  
  733. processed_system_time = per_cpu(shadow_time, 0).system_timestamp;
  734. per_cpu(processed_system_time, 0) = processed_system_time;
  735. init_missing_ticks_accounting(0);
  736.  
  737. clocksource_register(&clocksource_xen);
  738.  
  739. update_wallclock();
  740.  
  741. use_tsc_delay();
  742.  
  743. /* Cannot request_irq() until kmem is initialised. */
  744. late_time_init = setup_cpu0_timer_irq;
  745. }
  746.  
  747. /* Convert jiffies to system time. */
  748. u64 jiffies_to_st(unsigned long j)
  749. {
  750. unsigned long seq;
  751. long delta;
  752. u64 st;
  753.  
  754. do {
  755. seq = read_seqbegin(&xtime_lock);
  756. delta = j - jiffies;
  757. if (delta < 1) {
  758. /* Triggers in some wrap-around cases, but that's okay:
  759. * we just end up with a shorter timeout. */
  760. st = processed_system_time + NS_PER_TICK;
  761. } else if (((unsigned long)delta >> (BITS_PER_LONG-3)) != 0) {
  762. /* Very long timeout means there is no pending timer.
  763. * We indicate this to Xen by passing zero timeout. */
  764. st = 0;
  765. } else {
  766. st = processed_system_time + delta * (u64)NS_PER_TICK;
  767. }
  768. } while (read_seqretry(&xtime_lock, seq));
  769.  
  770. return st;
  771. }
  772. EXPORT_SYMBOL(jiffies_to_st);
  773.  
  774. /*
  775. * stop_hz_timer / start_hz_timer - enter/exit 'tickless mode' on an idle cpu
  776. * These functions are based on implementations from arch/s390/kernel/time.c
  777. */
  778. static void stop_hz_timer(void)
  779. {
  780. struct vcpu_set_singleshot_timer singleshot;
  781. unsigned int cpu = smp_processor_id();
  782. unsigned long j;
  783. int rc;
  784.  
  785. cpumask_set_cpu(cpu, nohz_cpu_mask);
  786.  
  787. /* See matching smp_mb in rcu_start_batch in rcupdate.c. These mbs */
  788. /* ensure that if __rcu_pending (nested in rcu_needs_cpu) fetches a */
  789. /* value of rcp->cur that matches rdp->quiescbatch and allows us to */
  790. /* stop the hz timer then the cpumasks created for subsequent values */
  791. /* of cur in rcu_start_batch are guaranteed to pick up the updated */
  792. /* nohz_cpu_mask and so will not depend on this cpu. */
  793.  
  794. smp_mb();
  795.  
  796. /* Leave ourselves in tick mode if rcu or softirq or timer pending. */
  797. if (rcu_needs_cpu(cpu) || printk_needs_cpu(cpu) ||
  798. local_softirq_pending() ||
  799. (j = get_next_timer_interrupt(jiffies),
  800. time_before_eq(j, jiffies))) {
  801. cpumask_clear_cpu(cpu, nohz_cpu_mask);
  802. j = jiffies + 1;
  803. }
  804.  
  805. singleshot.timeout_abs_ns = jiffies_to_st(j) + NS_PER_TICK/2;
  806. singleshot.flags = 0;
  807. rc = HYPERVISOR_vcpu_op(VCPUOP_set_singleshot_timer, cpu, &singleshot);
  808. #if CONFIG_XEN_COMPAT <= 0x030004
  809. if (rc) {
  810. BUG_ON(rc != -ENOSYS);
  811. rc = HYPERVISOR_set_timer_op(singleshot.timeout_abs_ns);
  812. }
  813. #endif
  814. BUG_ON(rc);
  815. }
  816.  
  817. static void start_hz_timer(void)
  818. {
  819. cpumask_clear_cpu(smp_processor_id(), nohz_cpu_mask);
  820. }
  821.  
  822. void xen_safe_halt(void)
  823. {
  824. stop_hz_timer();
  825. /* Blocking includes an implicit local_irq_enable(). */
  826. HYPERVISOR_block();
  827. start_hz_timer();
  828. }
  829. EXPORT_SYMBOL(xen_safe_halt);
  830.  
  831. void xen_halt(void)
  832. {
  833. if (irqs_disabled())
  834. VOID(HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL));
  835. }
  836. EXPORT_SYMBOL(xen_halt);
  837.  
  838. #ifdef CONFIG_SMP
  839. int __cpuinit local_setup_timer(unsigned int cpu)
  840. {
  841. int seq, irq;
  842.  
  843. BUG_ON(cpu == 0);
  844.  
  845. switch (HYPERVISOR_vcpu_op(VCPUOP_set_periodic_timer, cpu,
  846. &xen_set_periodic_tick)) {
  847. case 0:
  848. #if CONFIG_XEN_COMPAT <= 0x030004
  849. case -ENOSYS:
  850. #endif
  851. break;
  852. default:
  853. BUG();
  854. }
  855.  
  856. do {
  857. seq = read_seqbegin(&xtime_lock);
  858. /* Use cpu0 timestamp: cpu's shadow is not initialised yet. */
  859. per_cpu(processed_system_time, cpu) =
  860. per_cpu(shadow_time, 0).system_timestamp;
  861. init_missing_ticks_accounting(cpu);
  862. } while (read_seqretry(&xtime_lock, seq));
  863.  
  864. irq = bind_virq_to_irqaction(VIRQ_TIMER, cpu, &timer_action);
  865. if (irq < 0)
  866. return irq;
  867. BUG_ON(timer_irq != irq);
  868.  
  869. return 0;
  870. }
  871.  
  872. void __cpuinit local_teardown_timer(unsigned int cpu)
  873. {
  874. BUG_ON(cpu == 0);
  875. unbind_from_per_cpu_irq(timer_irq, cpu, &timer_action);
  876. }
  877. #endif
  878.  
  879. #ifdef CONFIG_CPU_FREQ
  880. static int time_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
  881. void *data)
  882. {
  883. struct cpufreq_freqs *freq = data;
  884. struct xen_platform_op op;
  885.  
  886. if (cpu_has(&cpu_data(freq->cpu), X86_FEATURE_CONSTANT_TSC))
  887. return 0;
  888.  
  889. if (val == CPUFREQ_PRECHANGE)
  890. return 0;
  891.  
  892. op.cmd = XENPF_change_freq;
  893. op.u.change_freq.flags = 0;
  894. op.u.change_freq.cpu = freq->cpu;
  895. op.u.change_freq.freq = (u64)freq->new * 1000;
  896. WARN_ON(HYPERVISOR_platform_op(&op));
  897.  
  898. return 0;
  899. }
  900.  
  901. static struct notifier_block time_cpufreq_notifier_block = {
  902. .notifier_call = time_cpufreq_notifier
  903. };
  904.  
  905. static int __init cpufreq_time_setup(void)
  906. {
  907. if (!cpufreq_register_notifier(&time_cpufreq_notifier_block,
  908. CPUFREQ_TRANSITION_NOTIFIER)) {
  909. printk(KERN_ERR "failed to set up cpufreq notifier\n");
  910. return -ENODEV;
  911. }
  912. return 0;
  913. }
  914.  
  915. core_initcall(cpufreq_time_setup);
  916. #endif
  917.  
  918. /*
  919. * /proc/sys/xen: This really belongs in another file. It can stay here for
  920. * now however.
  921. */
  922. static ctl_table xen_subtable[] = {
  923. {
  924. .ctl_name = CTL_XEN_INDEPENDENT_WALLCLOCK,
  925. .procname = "independent_wallclock",
  926. .data = &independent_wallclock,
  927. .maxlen = sizeof(independent_wallclock),
  928. .mode = 0644,
  929. .strategy = sysctl_data,
  930. .proc_handler = proc_dointvec
  931. },
  932. {
  933. .ctl_name = CTL_XEN_PERMITTED_CLOCK_JITTER,
  934. .procname = "permitted_clock_jitter",
  935. .data = &permitted_clock_jitter,
  936. .maxlen = sizeof(permitted_clock_jitter),
  937. .mode = 0644,
  938. .strategy = sysctl_data,
  939. .proc_handler = proc_doulongvec_minmax
  940. },
  941. { }
  942. };
  943. static ctl_table xen_table[] = {
  944. {
  945. .ctl_name = CTL_XEN,
  946. .procname = "xen",
  947. .mode = 0555,
  948. .child = xen_subtable
  949. },
  950. { }
  951. };
  952. static int __init xen_sysctl_init(void)
  953. {
  954. (void)register_sysctl_table(xen_table);
  955. return 0;
  956. }
  957. __initcall(xen_sysctl_init);
  958.  
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement
Public Pastes
Advertisement
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy.  OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!