timer-r0drv-linux.c@ 44824

Last change on this file since 44824 was 44528, checked in by vboxsync, 12 years ago
header (C) fixes
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 54.9 KB

Line
1	/* $Id: timer-r0drv-linux.c 44528 2013-02-04 14:27:54Z vboxsync $ */
2	/** @file
3	* IPRT - Timers, Ring-0 Driver, Linux.
4	*/
5
6	/*
7	* Copyright (C) 2006-2012 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*
17	* The contents of this file may alternatively be used under the terms
18	* of the Common Development and Distribution License Version 1.0
19	* (CDDL) only, as it comes in the "COPYING.CDDL" file of the
20	* VirtualBox OSE distribution, in which case the provisions of the
21	* CDDL are applicable instead of those of the GPL.
22	*
23	* You may elect to license modified versions of this file under the
24	* terms and conditions of either the GPL or the CDDL or both.
25	*/
26
27
28	/*******************************************************************************
29	* Header Files *
30	*******************************************************************************/
31	#include "the-linux-kernel.h"
32	#include "internal/iprt.h"
33
34	#include <iprt/timer.h>
35	#include <iprt/time.h>
36	#include <iprt/mp.h>
37	#include <iprt/cpuset.h>
38	#include <iprt/spinlock.h>
39	#include <iprt/err.h>
40	#include <iprt/asm.h>
41	#include <iprt/assert.h>
42	#include <iprt/alloc.h>
43
44	#include "internal/magics.h"
45
46	/** @def RTTIMER_LINUX_WITH_HRTIMER
47	* Whether to use high resolution timers. */
48	#if !defined(RTTIMER_LINUX_WITH_HRTIMER) \
49	&& defined(IPRT_LINUX_HAS_HRTIMER)
50	# define RTTIMER_LINUX_WITH_HRTIMER
51	#endif
52
53	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 31)
54	# define mod_timer_pinned mod_timer
55	# define HRTIMER_MODE_ABS_PINNED HRTIMER_MODE_ABS
56	#endif
57
58
59	/*******************************************************************************
60	* Structures and Typedefs *
61	*******************************************************************************/
62	/**
63	* Timer state machine.
64	*
65	* This is used to try handle the issues with MP events and
66	* timers that runs on all CPUs. It's relatively nasty :-/
67	*/
68	typedef enum RTTIMERLNXSTATE
69	{
70	/** Stopped. */
71	RTTIMERLNXSTATE_STOPPED = 0,
72	/** Transient state; next ACTIVE. */
73	RTTIMERLNXSTATE_STARTING,
74	/** Transient state; next ACTIVE. (not really necessary) */
75	RTTIMERLNXSTATE_MP_STARTING,
76	/** Active. */
77	RTTIMERLNXSTATE_ACTIVE,
78	/** Active and in callback; next ACTIVE, STOPPED or CALLBACK_DESTROYING. */
79	RTTIMERLNXSTATE_CALLBACK,
80	/** Stopped while in the callback; next STOPPED. */
81	RTTIMERLNXSTATE_CB_STOPPING,
82	/** Restarted while in the callback; next ACTIVE, STOPPED, DESTROYING. */
83	RTTIMERLNXSTATE_CB_RESTARTING,
84	/** The callback shall destroy the timer; next STOPPED. */
85	RTTIMERLNXSTATE_CB_DESTROYING,
86	/** Transient state; next STOPPED. */
87	RTTIMERLNXSTATE_STOPPING,
88	/** Transient state; next STOPPED. */
89	RTTIMERLNXSTATE_MP_STOPPING,
90	/** The usual 32-bit hack. */
91	RTTIMERLNXSTATE_32BIT_HACK = 0x7fffffff
92	} RTTIMERLNXSTATE;
93
94
95	/**
96	* A Linux sub-timer.
97	*/
98	typedef struct RTTIMERLNXSUBTIMER
99	{
100	/** Timer specific data. */
101	union
102	{
103	#if defined(RTTIMER_LINUX_WITH_HRTIMER)
104	/** High resolution timer. */
105	struct
106	{
107	/** The linux timer structure. */
108	struct hrtimer LnxTimer;
109	} Hr;
110	#endif
111	/** Standard timer. */
112	struct
113	{
114	/** The linux timer structure. */
115	struct timer_list LnxTimer;
116	/** The start of the current run (ns).
117	* This is used to calculate when the timer ought to fire the next time. */
118	uint64_t u64NextTS;
119	/** The u64NextTS in jiffies. */
120	unsigned long ulNextJiffies;
121	/** Set when starting or changing the timer so that u64StartTs
122	* and u64NextTS gets reinitialized (eliminating some jitter). */
123	bool volatile fFirstAfterChg;
124	} Std;
125	} u;
126	/** The current tick number. */
127	uint64_t iTick;
128	/** Restart the single shot timer at this specific time.
129	* Used when a single shot timer is restarted from the callback. */
130	uint64_t volatile uNsRestartAt;
131	/** Pointer to the parent timer. */
132	PRTTIMER pParent;
133	/** The current sub-timer state. */
134	RTTIMERLNXSTATE volatile enmState;
135	} RTTIMERLNXSUBTIMER;
136	/** Pointer to a linux sub-timer. */
137	typedef RTTIMERLNXSUBTIMER *PRTTIMERLNXSUBTIMER;
138
139
140	/**
141	* The internal representation of an Linux timer handle.
142	*/
143	typedef struct RTTIMER
144	{
145	/** Magic.
146	* This is RTTIMER_MAGIC, but changes to something else before the timer
147	* is destroyed to indicate clearly that thread should exit. */
148	uint32_t volatile u32Magic;
149	/** Spinlock synchronizing the fSuspended and MP event handling.
150	* This is NIL_RTSPINLOCK if cCpus == 1. */
151	RTSPINLOCK hSpinlock;
152	/** Flag indicating that the timer is suspended. */
153	bool volatile fSuspended;
154	/** Whether the timer must run on one specific CPU or not. */
155	bool fSpecificCpu;
156	#ifdef CONFIG_SMP
157	/** Whether the timer must run on all CPUs or not. */
158	bool fAllCpus;
159	#endif /* else: All -> specific on non-SMP kernels */
160	/** Whether it is a high resolution timer or a standard one. */
161	bool fHighRes;
162	/** The id of the CPU it must run on if fSpecificCpu is set. */
163	RTCPUID idCpu;
164	/** The number of CPUs this timer should run on. */
165	RTCPUID cCpus;
166	/** Callback. */
167	PFNRTTIMER pfnTimer;
168	/** User argument. */
169	void *pvUser;
170	/** The timer interval. 0 if one-shot. */
171	uint64_t volatile u64NanoInterval;
172	/** This is set to the number of jiffies between ticks if the interval is
173	* an exact number of jiffies. (Standard timers only.) */
174	unsigned long volatile cJiffies;
175	/** The change interval spinlock for standard timers only. */
176	spinlock_t ChgIntLock;
177	/** Workqueue item for delayed destruction. */
178	RTR0LNXWORKQUEUEITEM DtorWorkqueueItem;
179	/** Sub-timers.
180	* Normally there is just one, but for RTTIMER_FLAGS_CPU_ALL this will contain
181	* an entry for all possible cpus. In that case the index will be the same as
182	* for the RTCpuSet. */
183	RTTIMERLNXSUBTIMER aSubTimers[1];
184	} RTTIMER;
185
186
187	/**
188	* A rtTimerLinuxStartOnCpu and rtTimerLinuxStartOnCpu argument package.
189	*/
190	typedef struct RTTIMERLINUXSTARTONCPUARGS
191	{
192	/** The current time (RTTimeSystemNanoTS). */
193	uint64_t u64Now;
194	/** When to start firing (delta). */
195	uint64_t u64First;
196	} RTTIMERLINUXSTARTONCPUARGS;
197	/** Pointer to a rtTimerLinuxStartOnCpu argument package. */
198	typedef RTTIMERLINUXSTARTONCPUARGS *PRTTIMERLINUXSTARTONCPUARGS;
199
200
201	/*******************************************************************************
202	* Internal Functions *
203	*******************************************************************************/
204	#ifdef CONFIG_SMP
205	static DECLCALLBACK(void) rtTimerLinuxMpEvent(RTMPEVENT enmEvent, RTCPUID idCpu, void *pvUser);
206	#endif
207
208	#if 0
209	#define DEBUG_HACKING
210	#include <iprt/string.h>
211	#include <iprt/asm-amd64-x86.h>
212	static void myLogBackdoorPrintf(const char *pszFormat, ...)
213	{
214	char szTmp[256];
215	va_list args;
216	size_t cb;
217
218	cb = RTStrPrintf(szTmp, sizeof(szTmp) - 10, "%d: ", RTMpCpuId());
219	va_start(args, pszFormat);
220	cb += RTStrPrintfV(&szTmp[cb], sizeof(szTmp) - cb, pszFormat, args);
221	va_end(args);
222
223	ASMOutStrU8(0x504, (uint8_t *)&szTmp[0], cb);
224	}
225	# define RTAssertMsg1Weak(pszExpr, uLine, pszFile, pszFunction) \
226	myLogBackdoorPrintf("\n!!Guest Assertion failed!!\n%s(%d) %s\n%s\n", uLine, pszFile, pszFunction, (pszExpr))
227	# define RTAssertMsg2Weak myLogBackdoorPrintf
228	# define RTTIMERLNX_LOG(a) myLogBackdoorPrintf a
229	#else
230	# define RTTIMERLNX_LOG(a) do { } while (0)
231	#endif
232
233	/**
234	* Sets the state.
235	*/
236	DECLINLINE(void) rtTimerLnxSetState(RTTIMERLNXSTATE volatile *penmState, RTTIMERLNXSTATE enmNewState)
237	{
238	#ifdef DEBUG_HACKING
239	RTTIMERLNX_LOG(("set %d -> %d\n", *penmState, enmNewState));
240	#endif
241	ASMAtomicWriteU32((uint32_t volatile *)penmState, enmNewState);
242	}
243
244
245	/**
246	* Sets the state if it has a certain value.
247	*
248	* @return true if xchg was done.
249	* @return false if xchg wasn't done.
250	*/
251	#ifdef DEBUG_HACKING
252	#define rtTimerLnxCmpXchgState(penmState, enmNewState, enmCurState) rtTimerLnxCmpXchgStateDebug(penmState, enmNewState, enmCurState, __LINE__)
253	static bool rtTimerLnxCmpXchgStateDebug(RTTIMERLNXSTATE volatile *penmState, RTTIMERLNXSTATE enmNewState,
254	RTTIMERLNXSTATE enmCurState, uint32_t uLine)
255	{
256	RTTIMERLNXSTATE enmOldState = enmCurState;
257	bool fRc = ASMAtomicCmpXchgExU32((uint32_t volatile )penmState, enmNewState, enmCurState, (uint32_t )&enmOldState);
258	RTTIMERLNX_LOG(("cxg %d -> %d - %d at %u\n", enmOldState, enmNewState, fRc, uLine));
259	return fRc;
260	}
261	#else
262	DECLINLINE(bool) rtTimerLnxCmpXchgState(RTTIMERLNXSTATE volatile *penmState, RTTIMERLNXSTATE enmNewState,
263	RTTIMERLNXSTATE enmCurState)
264	{
265	return ASMAtomicCmpXchgU32((uint32_t volatile *)penmState, enmNewState, enmCurState);
266	}
267	#endif
268
269
270	/**
271	* Gets the state.
272	*/
273	DECLINLINE(RTTIMERLNXSTATE) rtTimerLnxGetState(RTTIMERLNXSTATE volatile *penmState)
274	{
275	return (RTTIMERLNXSTATE)ASMAtomicUoReadU32((uint32_t volatile *)penmState);
276	}
277
278	#ifdef RTTIMER_LINUX_WITH_HRTIMER
279
280	/**
281	* Converts a nano second time stamp to ktime_t.
282	*
283	* ASSUMES RTTimeSystemNanoTS() is implemented using ktime_get_ts().
284	*
285	* @returns ktime_t.
286	* @param cNanoSecs Nanoseconds.
287	*/
288	DECLINLINE(ktime_t) rtTimerLnxNanoToKt(uint64_t cNanoSecs)
289	{
290	/* With some luck the compiler optimizes the division out of this... (Bet it doesn't.) */
291	return ktime_set(cNanoSecs / 1000000000, cNanoSecs % 1000000000);
292	}
293
294	/**
295	* Converts ktime_t to a nano second time stamp.
296	*
297	* ASSUMES RTTimeSystemNanoTS() is implemented using ktime_get_ts().
298	*
299	* @returns nano second time stamp.
300	* @param Kt ktime_t.
301	*/
302	DECLINLINE(uint64_t) rtTimerLnxKtToNano(ktime_t Kt)
303	{
304	return ktime_to_ns(Kt);
305	}
306
307	#endif /* RTTIMER_LINUX_WITH_HRTIMER */
308
309	/**
310	* Converts a nano second interval to jiffies.
311	*
312	* @returns Jiffies.
313	* @param cNanoSecs Nanoseconds.
314	*/
315	DECLINLINE(unsigned long) rtTimerLnxNanoToJiffies(uint64_t cNanoSecs)
316	{
317	/* this can be made even better... */
318	if (cNanoSecs > (uint64_t)TICK_NSEC * MAX_JIFFY_OFFSET)
319	return MAX_JIFFY_OFFSET;
320	# if ARCH_BITS == 32
321	if (RT_LIKELY(cNanoSecs <= UINT32_MAX))
322	return ((uint32_t)cNanoSecs + (TICK_NSEC-1)) / TICK_NSEC;
323	# endif
324	return (cNanoSecs + (TICK_NSEC-1)) / TICK_NSEC;
325	}
326
327
328	/**
329	* Starts a sub-timer (RTTimerStart).
330	*
331	* @param pSubTimer The sub-timer to start.
332	* @param u64Now The current timestamp (RTTimeSystemNanoTS()).
333	* @param u64First The interval from u64Now to the first time the timer should fire.
334	* @param fPinned true = timer pinned to a specific CPU,
335	* false = timer can migrate between CPUs
336	* @param fHighRes Whether the user requested a high resolution timer or not.
337	* @param enmOldState The old timer state.
338	*/
339	static void rtTimerLnxStartSubTimer(PRTTIMERLNXSUBTIMER pSubTimer, uint64_t u64Now, uint64_t u64First,
340	bool fPinned, bool fHighRes)
341	{
342	/*
343	* Calc when it should start firing.
344	*/
345	uint64_t u64NextTS = u64Now + u64First;
346	if (!fHighRes)
347	pSubTimer->u.Std.u64NextTS = u64NextTS;
348	RTTIMERLNX_LOG(("startsubtimer %p\n", pSubTimer->pParent));
349
350	pSubTimer->iTick = 0;
351
352	#ifdef RTTIMER_LINUX_WITH_HRTIMER
353	if (fHighRes)
354	hrtimer_start(&pSubTimer->u.Hr.LnxTimer, rtTimerLnxNanoToKt(u64NextTS),
355	fPinned ? HRTIMER_MODE_ABS_PINNED : HRTIMER_MODE_ABS);
356	else
357	#endif
358	{
359	unsigned long cJiffies = !u64First ? 0 : rtTimerLnxNanoToJiffies(u64First);
360	pSubTimer->u.Std.ulNextJiffies = jiffies + cJiffies;
361	pSubTimer->u.Std.fFirstAfterChg = true;
362	#ifdef CONFIG_SMP
363	if (fPinned)
364	mod_timer_pinned(&pSubTimer->u.Std.LnxTimer, pSubTimer->u.Std.ulNextJiffies);
365	else
366	#endif
367	mod_timer(&pSubTimer->u.Std.LnxTimer, pSubTimer->u.Std.ulNextJiffies);
368	}
369
370	/* Be a bit careful here since we could be racing the callback. */
371	if (!rtTimerLnxCmpXchgState(&pSubTimer->enmState, RTTIMERLNXSTATE_ACTIVE, RTTIMERLNXSTATE_STARTING))
372	rtTimerLnxCmpXchgState(&pSubTimer->enmState, RTTIMERLNXSTATE_ACTIVE, RTTIMERLNXSTATE_MP_STARTING);
373	}
374
375
376	/**
377	* Stops a sub-timer (RTTimerStart and rtTimerLinuxMpEvent()).
378	*
379	* The caller has already changed the state, so we will not be in a callback
380	* situation wrt to the calling thread.
381	*
382	* @param pSubTimer The sub-timer.
383	* @param fHighRes Whether the user requested a high resolution timer or not.
384	*/
385	static void rtTimerLnxStopSubTimer(PRTTIMERLNXSUBTIMER pSubTimer, bool fHighRes)
386	{
387	RTTIMERLNX_LOG(("stopsubtimer %p %d\n", pSubTimer->pParent, fHighRes));
388	#ifdef RTTIMER_LINUX_WITH_HRTIMER
389	if (fHighRes)
390	{
391	/* There is no equivalent to del_timer in the hrtimer API,
392	hrtimer_cancel() == del_timer_sync(). Just like the WARN_ON in
393	del_timer_sync() asserts, waiting for a timer callback to complete
394	is deadlock prone, so don't do it. */
395	int rc = hrtimer_try_to_cancel(&pSubTimer->u.Hr.LnxTimer);
396	if (rc < 0)
397	{
398	hrtimer_start(&pSubTimer->u.Hr.LnxTimer, ktime_set(KTIME_SEC_MAX, 0), HRTIMER_MODE_ABS);
399	hrtimer_try_to_cancel(&pSubTimer->u.Hr.LnxTimer);
400	}
401	}
402	else
403	#endif
404	del_timer(&pSubTimer->u.Std.LnxTimer);
405
406	rtTimerLnxSetState(&pSubTimer->enmState, RTTIMERLNXSTATE_STOPPED);
407	}
408
409
410	/**
411	* Used by RTTimerDestroy and rtTimerLnxCallbackDestroy to do the actual work.
412	*
413	* @param pTimer The timer in question.
414	*/
415	static void rtTimerLnxDestroyIt(PRTTIMER pTimer)
416	{
417	RTSPINLOCK hSpinlock = pTimer->hSpinlock;
418	RTCPUID iCpu;
419	Assert(pTimer->fSuspended);
420	RTTIMERLNX_LOG(("destroyit %p\n", pTimer));
421
422	/*
423	* Remove the MP notifications first because it'll reduce the risk of
424	* us overtaking any MP event that might theoretically be racing us here.
425	*/
426	#ifdef CONFIG_SMP
427	if ( pTimer->cCpus > 1
428	&& hSpinlock != NIL_RTSPINLOCK)
429	{
430	int rc = RTMpNotificationDeregister(rtTimerLinuxMpEvent, pTimer);
431	AssertRC(rc);
432	}
433	#endif /* CONFIG_SMP */
434
435	/*
436	* Invalidate the handle.
437	*/
438	ASMAtomicWriteU32(&pTimer->u32Magic, ~RTTIMER_MAGIC);
439
440	/*
441	* Make sure all timers have stopped executing since we're stopping them in
442	* an asynchronous manner up in rtTimerLnxStopSubTimer.
443	*/
444	iCpu = pTimer->cCpus;
445	while (iCpu-- > 0)
446	{
447	#ifdef RTTIMER_LINUX_WITH_HRTIMER
448	if (pTimer->fHighRes)
449	hrtimer_cancel(&pTimer->aSubTimers[iCpu].u.Hr.LnxTimer);
450	else
451	#endif
452	del_timer_sync(&pTimer->aSubTimers[iCpu].u.Std.LnxTimer);
453	}
454
455	/*
456	* Finally, free the resources.
457	*/
458	RTMemFreeEx(pTimer, RT_OFFSETOF(RTTIMER, aSubTimers[pTimer->cCpus]));
459	if (hSpinlock != NIL_RTSPINLOCK)
460	RTSpinlockDestroy(hSpinlock);
461	}
462
463
464	/**
465	* Workqueue callback (no DECLCALLBACK!) for deferred destruction.
466	*
467	* @param pWork Pointer to the DtorWorkqueueItem member of our timer
468	* structure.
469	*/
470	static void rtTimerLnxDestroyDeferred(RTR0LNXWORKQUEUEITEM *pWork)
471	{
472	PRTTIMER pTimer = RT_FROM_MEMBER(pWork, RTTIMER, DtorWorkqueueItem);
473	rtTimerLnxDestroyIt(pTimer);
474	}
475
476
477	/**
478	* Called when the timer was destroyed by the callback function.
479	*
480	* @param pTimer The timer.
481	* @param pSubTimer The sub-timer which we're handling, the state of this
482	* will be RTTIMERLNXSTATE_CALLBACK_DESTROYING.
483	*/
484	static void rtTimerLnxCallbackDestroy(PRTTIMER pTimer, PRTTIMERLNXSUBTIMER pSubTimer)
485	{
486	/*
487	* If it's an omni timer, the last dude does the destroying.
488	*/
489	if (pTimer->cCpus > 1)
490	{
491	uint32_t iCpu = pTimer->cCpus;
492	RTSpinlockAcquire(pTimer->hSpinlock);
493
494	Assert(pSubTimer->enmState == RTTIMERLNXSTATE_CB_DESTROYING);
495	rtTimerLnxSetState(&pSubTimer->enmState, RTTIMERLNXSTATE_STOPPED);
496
497	while (iCpu-- > 0)
498	if (rtTimerLnxGetState(&pTimer->aSubTimers[iCpu].enmState) != RTTIMERLNXSTATE_STOPPED)
499	{
500	RTSpinlockRelease(pTimer->hSpinlock);
501	return;
502	}
503
504	RTSpinlockRelease(pTimer->hSpinlock);
505	}
506
507	/*
508	* Destroying a timer from the callback is unsafe since the callout code
509	* might be touching the timer structure upon return (hrtimer does!). So,
510	* we have to defer the actual destruction to the IRPT workqueue.
511	*/
512	rtR0LnxWorkqueuePush(&pTimer->DtorWorkqueueItem, rtTimerLnxDestroyDeferred);
513	}
514
515
516	#ifdef CONFIG_SMP
517	/**
518	* Deal with a sub-timer that has migrated.
519	*
520	* @param pTimer The timer.
521	* @param pSubTimer The sub-timer.
522	*/
523	static void rtTimerLnxCallbackHandleMigration(PRTTIMER pTimer, PRTTIMERLNXSUBTIMER pSubTimer)
524	{
525	RTTIMERLNXSTATE enmState;
526	if (pTimer->cCpus > 1)
527	RTSpinlockAcquire(pTimer->hSpinlock);
528
529	do
530	{
531	enmState = rtTimerLnxGetState(&pSubTimer->enmState);
532	switch (enmState)
533	{
534	case RTTIMERLNXSTATE_STOPPING:
535	case RTTIMERLNXSTATE_MP_STOPPING:
536	enmState = RTTIMERLNXSTATE_STOPPED;
537	case RTTIMERLNXSTATE_STOPPED:
538	break;
539
540	default:
541	AssertMsgFailed(("%d\n", enmState));
542	case RTTIMERLNXSTATE_STARTING:
543	case RTTIMERLNXSTATE_MP_STARTING:
544	case RTTIMERLNXSTATE_ACTIVE:
545	case RTTIMERLNXSTATE_CALLBACK:
546	case RTTIMERLNXSTATE_CB_STOPPING:
547	case RTTIMERLNXSTATE_CB_RESTARTING:
548	if (rtTimerLnxCmpXchgState(&pSubTimer->enmState, RTTIMERLNXSTATE_STOPPED, enmState))
549	enmState = RTTIMERLNXSTATE_STOPPED;
550	break;
551
552	case RTTIMERLNXSTATE_CB_DESTROYING:
553	{
554	if (pTimer->cCpus > 1)
555	RTSpinlockRelease(pTimer->hSpinlock);
556
557	rtTimerLnxCallbackDestroy(pTimer, pSubTimer);
558	return;
559	}
560	}
561	} while (enmState != RTTIMERLNXSTATE_STOPPED);
562
563	if (pTimer->cCpus > 1)
564	RTSpinlockRelease(pTimer->hSpinlock);
565	}
566	#endif /* CONFIG_SMP */
567
568
569	/**
570	* The slow path of rtTimerLnxChangeToCallbackState.
571	*
572	* @returns true if changed successfully, false if not.
573	* @param pSubTimer The sub-timer.
574	*/
575	static bool rtTimerLnxChangeToCallbackStateSlow(PRTTIMERLNXSUBTIMER pSubTimer)
576	{
577	for (;;)
578	{
579	RTTIMERLNXSTATE enmState = rtTimerLnxGetState(&pSubTimer->enmState);
580	switch (enmState)
581	{
582	case RTTIMERLNXSTATE_ACTIVE:
583	case RTTIMERLNXSTATE_STARTING:
584	case RTTIMERLNXSTATE_MP_STARTING:
585	if (rtTimerLnxCmpXchgState(&pSubTimer->enmState, RTTIMERLNXSTATE_CALLBACK, enmState))
586	return true;
587	break;
588
589	case RTTIMERLNXSTATE_CALLBACK:
590	case RTTIMERLNXSTATE_CB_STOPPING:
591	case RTTIMERLNXSTATE_CB_RESTARTING:
592	case RTTIMERLNXSTATE_CB_DESTROYING:
593	AssertMsgFailed(("%d\n", enmState));
594	default:
595	return false;
596	}
597	ASMNopPause();
598	}
599	}
600
601
602	/**
603	* Tries to change the sub-timer state to 'callback'.
604	*
605	* @returns true if changed successfully, false if not.
606	* @param pSubTimer The sub-timer.
607	*/
608	DECLINLINE(bool) rtTimerLnxChangeToCallbackState(PRTTIMERLNXSUBTIMER pSubTimer)
609	{
610	if (RT_LIKELY(rtTimerLnxCmpXchgState(&pSubTimer->enmState, RTTIMERLNXSTATE_CALLBACK, RTTIMERLNXSTATE_ACTIVE)))
611	return true;
612	return rtTimerLnxChangeToCallbackStateSlow(pSubTimer);
613	}
614
615
616	#ifdef RTTIMER_LINUX_WITH_HRTIMER
617	/**
618	* Timer callback function for high resolution timers.
619	*
620	* @returns HRTIMER_NORESTART or HRTIMER_RESTART depending on whether it's a
621	* one-shot or interval timer.
622	* @param pHrTimer Pointer to the sub-timer structure.
623	*/
624	static enum hrtimer_restart rtTimerLinuxHrCallback(struct hrtimer *pHrTimer)
625	{
626	PRTTIMERLNXSUBTIMER pSubTimer = RT_FROM_MEMBER(pHrTimer, RTTIMERLNXSUBTIMER, u.Hr.LnxTimer);
627	PRTTIMER pTimer = pSubTimer->pParent;
628
629
630	RTTIMERLNX_LOG(("hrcallback %p\n", pTimer));
631	if (RT_UNLIKELY(!rtTimerLnxChangeToCallbackState(pSubTimer)))
632	return HRTIMER_NORESTART;
633
634	#ifdef CONFIG_SMP
635	/*
636	* Check for unwanted migration.
637	*/
638	if (pTimer->fAllCpus \|\| pTimer->fSpecificCpu)
639	{
640	RTCPUID idCpu = RTMpCpuId();
641	if (RT_UNLIKELY( pTimer->fAllCpus
642	? (RTCPUID)(pSubTimer - &pTimer->aSubTimers[0]) != idCpu
643	: pTimer->idCpu != idCpu))
644	{
645	rtTimerLnxCallbackHandleMigration(pTimer, pSubTimer);
646	return HRTIMER_NORESTART;
647	}
648	}
649	#endif
650
651	if (pTimer->u64NanoInterval)
652	{
653	/*
654	* Periodic timer, run it and update the native timer afterwards so
655	* we can handle RTTimerStop and RTTimerChangeInterval from the
656	* callback as well as a racing control thread.
657	*/
658	pTimer->pfnTimer(pTimer, pTimer->pvUser, ++pSubTimer->iTick);
659	hrtimer_add_expires_ns(&pSubTimer->u.Hr.LnxTimer, ASMAtomicReadU64(&pTimer->u64NanoInterval));
660	if (RT_LIKELY(rtTimerLnxCmpXchgState(&pSubTimer->enmState, RTTIMERLNXSTATE_ACTIVE, RTTIMERLNXSTATE_CALLBACK)))
661	return HRTIMER_RESTART;
662	}
663	else
664	{
665	/*
666	* One shot timer (no omni), stop it before dispatching it.
667	* Allow RTTimerStart as well as RTTimerDestroy to be called from
668	* the callback.
669	*/
670	ASMAtomicWriteBool(&pTimer->fSuspended, true);
671	pTimer->pfnTimer(pTimer, pTimer->pvUser, ++pSubTimer->iTick);
672	if (RT_LIKELY(rtTimerLnxCmpXchgState(&pSubTimer->enmState, RTTIMERLNXSTATE_STOPPED, RTTIMERLNXSTATE_CALLBACK)))
673	return HRTIMER_NORESTART;
674	}
675
676	/*
677	* Some state change occurred while we were in the callback routine.
678	*/
679	for (;;)
680	{
681	RTTIMERLNXSTATE enmState = rtTimerLnxGetState(&pSubTimer->enmState);
682	switch (enmState)
683	{
684	case RTTIMERLNXSTATE_CB_DESTROYING:
685	rtTimerLnxCallbackDestroy(pTimer, pSubTimer);
686	return HRTIMER_NORESTART;
687
688	case RTTIMERLNXSTATE_CB_STOPPING:
689	if (rtTimerLnxCmpXchgState(&pSubTimer->enmState, RTTIMERLNXSTATE_STOPPED, RTTIMERLNXSTATE_CB_STOPPING))
690	return HRTIMER_NORESTART;
691	break;
692
693	case RTTIMERLNXSTATE_CB_RESTARTING:
694	if (rtTimerLnxCmpXchgState(&pSubTimer->enmState, RTTIMERLNXSTATE_ACTIVE, RTTIMERLNXSTATE_CB_RESTARTING))
695	{
696	pSubTimer->iTick = 0;
697	hrtimer_set_expires(&pSubTimer->u.Hr.LnxTimer, rtTimerLnxNanoToKt(pSubTimer->uNsRestartAt));
698	return HRTIMER_RESTART;
699	}
700	break;
701
702	default:
703	AssertMsgFailed(("%d\n", enmState));
704	return HRTIMER_NORESTART;
705	}
706	ASMNopPause();
707	}
708	}
709	#endif /* RTTIMER_LINUX_WITH_HRTIMER */
710
711
712	/**
713	* Timer callback function for standard timers.
714	*
715	* @param ulUser Address of the sub-timer structure.
716	*/
717	static void rtTimerLinuxStdCallback(unsigned long ulUser)
718	{
719	PRTTIMERLNXSUBTIMER pSubTimer = (PRTTIMERLNXSUBTIMER)ulUser;
720	PRTTIMER pTimer = pSubTimer->pParent;
721
722	RTTIMERLNX_LOG(("stdcallback %p\n", pTimer));
723	if (RT_UNLIKELY(!rtTimerLnxChangeToCallbackState(pSubTimer)))
724	return;
725
726	#ifdef CONFIG_SMP
727	/*
728	* Check for unwanted migration.
729	*/
730	if (pTimer->fAllCpus \|\| pTimer->fSpecificCpu)
731	{
732	RTCPUID idCpu = RTMpCpuId();
733	if (RT_UNLIKELY( pTimer->fAllCpus
734	? (RTCPUID)(pSubTimer - &pTimer->aSubTimers[0]) != idCpu
735	: pTimer->idCpu != idCpu))
736	{
737	rtTimerLnxCallbackHandleMigration(pTimer, pSubTimer);
738	return;
739	}
740	}
741	#endif
742
743	if (pTimer->u64NanoInterval)
744	{
745	/*
746	* Interval timer, calculate the next timeout.
747	*
748	* The first time around, we'll re-adjust the u.Std.u64NextTS to
749	* try prevent some jittering if we were started at a bad time.
750	*/
751	const uint64_t iTick = ++pSubTimer->iTick;
752	uint64_t u64NanoInterval;
753	unsigned long cJiffies;
754	unsigned long flFlags;
755
756	spin_lock_irqsave(&pTimer->ChgIntLock, flFlags);
757	u64NanoInterval = pTimer->u64NanoInterval;
758	cJiffies = pTimer->cJiffies;
759	if (RT_UNLIKELY(pSubTimer->u.Std.fFirstAfterChg))
760	{
761	pSubTimer->u.Std.fFirstAfterChg = false;
762	pSubTimer->u.Std.u64NextTS = RTTimeSystemNanoTS();
763	pSubTimer->u.Std.ulNextJiffies = jiffies;
764	}
765	spin_unlock_irqrestore(&pTimer->ChgIntLock, flFlags);
766
767	pSubTimer->u.Std.u64NextTS += u64NanoInterval;
768	if (cJiffies)
769	{
770	pSubTimer->u.Std.ulNextJiffies += cJiffies;
771	/* Prevent overflows when the jiffies counter wraps around.
772	* Special thanks to Ken Preslan for helping debugging! */
773	while (time_before(pSubTimer->u.Std.ulNextJiffies, jiffies))
774	{
775	pSubTimer->u.Std.ulNextJiffies += cJiffies;
776	pSubTimer->u.Std.u64NextTS += u64NanoInterval;
777	}
778	}
779	else
780	{
781	const uint64_t u64NanoTS = RTTimeSystemNanoTS();
782	while (pSubTimer->u.Std.u64NextTS < u64NanoTS)
783	pSubTimer->u.Std.u64NextTS += u64NanoInterval;
784	pSubTimer->u.Std.ulNextJiffies = jiffies + rtTimerLnxNanoToJiffies(pSubTimer->u.Std.u64NextTS - u64NanoTS);
785	}
786
787	/*
788	* Run the timer and re-arm it unless the state changed .
789	* .
790	* We must re-arm it afterwards as we're not in a position to undo this .
791	* operation if for instance someone stopped or destroyed us while we .
792	* were in the callback. (Linux takes care of any races here.)
793	*/
794	pTimer->pfnTimer(pTimer, pTimer->pvUser, iTick);
795	if (RT_LIKELY(rtTimerLnxCmpXchgState(&pSubTimer->enmState, RTTIMERLNXSTATE_ACTIVE, RTTIMERLNXSTATE_CALLBACK)))
796	{
797	#ifdef CONFIG_SMP
798	if (pTimer->fSpecificCpu \|\| pTimer->fAllCpus)
799	mod_timer_pinned(&pSubTimer->u.Std.LnxTimer, pSubTimer->u.Std.ulNextJiffies);
800	else
801	#endif
802	mod_timer(&pSubTimer->u.Std.LnxTimer, pSubTimer->u.Std.ulNextJiffies);
803	return;
804	}
805	}
806	else
807	{
808	/*
809	* One shot timer, stop it before dispatching it.
810	* Allow RTTimerStart as well as RTTimerDestroy to be called from
811	* the callback.
812	*/
813	ASMAtomicWriteBool(&pTimer->fSuspended, true);
814	pTimer->pfnTimer(pTimer, pTimer->pvUser, ++pSubTimer->iTick);
815	if (RT_LIKELY(rtTimerLnxCmpXchgState(&pSubTimer->enmState, RTTIMERLNXSTATE_STOPPED, RTTIMERLNXSTATE_CALLBACK)))
816	return;
817	}
818
819	/*
820	* Some state change occurred while we were in the callback routine.
821	*/
822	for (;;)
823	{
824	RTTIMERLNXSTATE enmState = rtTimerLnxGetState(&pSubTimer->enmState);
825	switch (enmState)
826	{
827	case RTTIMERLNXSTATE_CB_DESTROYING:
828	rtTimerLnxCallbackDestroy(pTimer, pSubTimer);
829	return;
830
831	case RTTIMERLNXSTATE_CB_STOPPING:
832	if (rtTimerLnxCmpXchgState(&pSubTimer->enmState, RTTIMERLNXSTATE_STOPPED, RTTIMERLNXSTATE_CB_STOPPING))
833	return;
834	break;
835
836	case RTTIMERLNXSTATE_CB_RESTARTING:
837	if (rtTimerLnxCmpXchgState(&pSubTimer->enmState, RTTIMERLNXSTATE_ACTIVE, RTTIMERLNXSTATE_CB_RESTARTING))
838	{
839	uint64_t u64NanoTS;
840	uint64_t u64NextTS;
841	unsigned long flFlags;
842
843	spin_lock_irqsave(&pTimer->ChgIntLock, flFlags);
844	u64NextTS = pSubTimer->uNsRestartAt;
845	u64NanoTS = RTTimeSystemNanoTS();
846	pSubTimer->iTick = 0;
847	pSubTimer->u.Std.u64NextTS = u64NextTS;
848	pSubTimer->u.Std.fFirstAfterChg = true;
849	pSubTimer->u.Std.ulNextJiffies = u64NextTS > u64NanoTS
850	? jiffies + rtTimerLnxNanoToJiffies(u64NextTS - u64NanoTS)
851	: jiffies;
852	spin_unlock_irqrestore(&pTimer->ChgIntLock, flFlags);
853
854	#ifdef CONFIG_SMP
855	if (pTimer->fSpecificCpu \|\| pTimer->fAllCpus)
856	mod_timer_pinned(&pSubTimer->u.Std.LnxTimer, pSubTimer->u.Std.ulNextJiffies);
857	else
858	#endif
859	mod_timer(&pSubTimer->u.Std.LnxTimer, pSubTimer->u.Std.ulNextJiffies);
860	return;
861	}
862	break;
863
864	default:
865	AssertMsgFailed(("%d\n", enmState));
866	return;
867	}
868	ASMNopPause();
869	}
870	}
871
872
873	#ifdef CONFIG_SMP
874
875	/**
876	* Per-cpu callback function (RTMpOnAll/RTMpOnSpecific).
877	*
878	* @param idCpu The current CPU.
879	* @param pvUser1 Pointer to the timer.
880	* @param pvUser2 Pointer to the argument structure.
881	*/
882	static DECLCALLBACK(void) rtTimerLnxStartAllOnCpu(RTCPUID idCpu, void pvUser1, void pvUser2)
883	{
884	PRTTIMERLINUXSTARTONCPUARGS pArgs = (PRTTIMERLINUXSTARTONCPUARGS)pvUser2;
885	PRTTIMER pTimer = (PRTTIMER)pvUser1;
886	Assert(idCpu < pTimer->cCpus);
887	rtTimerLnxStartSubTimer(&pTimer->aSubTimers[idCpu], pArgs->u64Now, pArgs->u64First, true /fPinned/, pTimer->fHighRes);
888	}
889
890
891	/**
892	* Worker for RTTimerStart() that takes care of the ugly bits.
893	*
894	* @returns RTTimerStart() return value.
895	* @param pTimer The timer.
896	* @param pArgs The argument structure.
897	*/
898	static int rtTimerLnxOmniStart(PRTTIMER pTimer, PRTTIMERLINUXSTARTONCPUARGS pArgs)
899	{
900	RTCPUID iCpu;
901	RTCPUSET OnlineSet;
902	RTCPUSET OnlineSet2;
903	int rc2;
904
905	/*
906	* Prepare all the sub-timers for the startup and then flag the timer
907	* as a whole as non-suspended, make sure we get them all before
908	* clearing fSuspended as the MP handler will be waiting on this
909	* should something happen while we're looping.
910	*/
911	RTSpinlockAcquire(pTimer->hSpinlock);
912
913	/* Just make it a omni timer restriction that no stop/start races are allowed. */
914	for (iCpu = 0; iCpu < pTimer->cCpus; iCpu++)
915	if (rtTimerLnxGetState(&pTimer->aSubTimers[iCpu].enmState) != RTTIMERLNXSTATE_STOPPED)
916	{
917	RTSpinlockRelease(pTimer->hSpinlock);
918	return VERR_TIMER_BUSY;
919	}
920
921	do
922	{
923	RTMpGetOnlineSet(&OnlineSet);
924	for (iCpu = 0; iCpu < pTimer->cCpus; iCpu++)
925	{
926	Assert(pTimer->aSubTimers[iCpu].enmState != RTTIMERLNXSTATE_MP_STOPPING);
927	rtTimerLnxSetState(&pTimer->aSubTimers[iCpu].enmState,
928	RTCpuSetIsMember(&OnlineSet, iCpu)
929	? RTTIMERLNXSTATE_STARTING
930	: RTTIMERLNXSTATE_STOPPED);
931	}
932	} while (!RTCpuSetIsEqual(&OnlineSet, RTMpGetOnlineSet(&OnlineSet2)));
933
934	ASMAtomicWriteBool(&pTimer->fSuspended, false);
935
936	RTSpinlockRelease(pTimer->hSpinlock);
937
938	/*
939	* Start them (can't find any exported function that allows me to
940	* do this without the cross calls).
941	*/
942	pArgs->u64Now = RTTimeSystemNanoTS();
943	rc2 = RTMpOnAll(rtTimerLnxStartAllOnCpu, pTimer, pArgs);
944	AssertRC(rc2); /* screw this if it fails. */
945
946	/*
947	* Reset the sub-timers who didn't start up (ALL CPUs case).
948	*/
949	RTSpinlockAcquire(pTimer->hSpinlock);
950
951	for (iCpu = 0; iCpu < pTimer->cCpus; iCpu++)
952	if (rtTimerLnxCmpXchgState(&pTimer->aSubTimers[iCpu].enmState, RTTIMERLNXSTATE_STOPPED, RTTIMERLNXSTATE_STARTING))
953	{
954	/** @todo very odd case for a rainy day. Cpus that temporarily went offline while
955	* we were between calls needs to nudged as the MP handler will ignore events for
956	* them because of the STARTING state. This is an extremely unlikely case - not that
957	* that means anything in my experience... ;-) */
958	RTTIMERLNX_LOG(("what!? iCpu=%u -> didn't start\n", iCpu));
959	}
960
961	RTSpinlockRelease(pTimer->hSpinlock);
962
963	return VINF_SUCCESS;
964	}
965
966
967	/**
968	* Worker for RTTimerStop() that takes care of the ugly SMP bits.
969	*
970	* @returns true if there was any active callbacks, false if not.
971	* @param pTimer The timer (valid).
972	* @param fForDestroy Whether this is for RTTimerDestroy or not.
973	*/
974	static bool rtTimerLnxOmniStop(PRTTIMER pTimer, bool fForDestroy)
975	{
976	bool fActiveCallbacks = false;
977	RTCPUID iCpu;
978	RTTIMERLNXSTATE enmState;
979
980
981	/*
982	* Mark the timer as suspended and flag all timers as stopping, except
983	* for those being stopped by an MP event.
984	*/
985	RTSpinlockAcquire(pTimer->hSpinlock);
986
987	ASMAtomicWriteBool(&pTimer->fSuspended, true);
988	for (iCpu = 0; iCpu < pTimer->cCpus; iCpu++)
989	{
990	for (;;)
991	{
992	enmState = rtTimerLnxGetState(&pTimer->aSubTimers[iCpu].enmState);
993	if ( enmState == RTTIMERLNXSTATE_STOPPED
994	\|\| enmState == RTTIMERLNXSTATE_MP_STOPPING)
995	break;
996	if ( enmState == RTTIMERLNXSTATE_CALLBACK
997	\|\| enmState == RTTIMERLNXSTATE_CB_STOPPING
998	\|\| enmState == RTTIMERLNXSTATE_CB_RESTARTING)
999	{
1000	Assert(enmState != RTTIMERLNXSTATE_CB_STOPPING \|\| fForDestroy);
1001	if (rtTimerLnxCmpXchgState(&pTimer->aSubTimers[iCpu].enmState,
1002	!fForDestroy ? RTTIMERLNXSTATE_CB_STOPPING : RTTIMERLNXSTATE_CB_DESTROYING,
1003	enmState))
1004	{
1005	fActiveCallbacks = true;
1006	break;
1007	}
1008	}
1009	else
1010	{
1011	Assert(enmState == RTTIMERLNXSTATE_ACTIVE);
1012	if (rtTimerLnxCmpXchgState(&pTimer->aSubTimers[iCpu].enmState, RTTIMERLNXSTATE_STOPPING, enmState))
1013	break;
1014	}
1015	ASMNopPause();
1016	}
1017	}
1018
1019	RTSpinlockRelease(pTimer->hSpinlock);
1020
1021	/*
1022	* Do the actual stopping. Fortunately, this doesn't require any IPIs.
1023	* Unfortunately it cannot be done synchronously.
1024	*/
1025	for (iCpu = 0; iCpu < pTimer->cCpus; iCpu++)
1026	if (rtTimerLnxGetState(&pTimer->aSubTimers[iCpu].enmState) == RTTIMERLNXSTATE_STOPPING)
1027	rtTimerLnxStopSubTimer(&pTimer->aSubTimers[iCpu], pTimer->fHighRes);
1028
1029	return fActiveCallbacks;
1030	}
1031
1032
1033	/**
1034	* Per-cpu callback function (RTMpOnSpecific) used by rtTimerLinuxMpEvent()
1035	* to start a sub-timer on a cpu that just have come online.
1036	*
1037	* @param idCpu The current CPU.
1038	* @param pvUser1 Pointer to the timer.
1039	* @param pvUser2 Pointer to the argument structure.
1040	*/
1041	static DECLCALLBACK(void) rtTimerLinuxMpStartOnCpu(RTCPUID idCpu, void pvUser1, void pvUser2)
1042	{
1043	PRTTIMERLINUXSTARTONCPUARGS pArgs = (PRTTIMERLINUXSTARTONCPUARGS)pvUser2;
1044	PRTTIMER pTimer = (PRTTIMER)pvUser1;
1045	RTSPINLOCK hSpinlock;
1046	Assert(idCpu < pTimer->cCpus);
1047
1048	/*
1049	* We have to be kind of careful here as we might be racing RTTimerStop
1050	* (and/or RTTimerDestroy, thus the paranoia.
1051	*/
1052	hSpinlock = pTimer->hSpinlock;
1053	if ( hSpinlock != NIL_RTSPINLOCK
1054	&& pTimer->u32Magic == RTTIMER_MAGIC)
1055	{
1056	RTSpinlockAcquire(hSpinlock);
1057
1058	if ( !ASMAtomicUoReadBool(&pTimer->fSuspended)
1059	&& pTimer->u32Magic == RTTIMER_MAGIC)
1060	{
1061	/* We're sane and the timer is not suspended yet. */
1062	PRTTIMERLNXSUBTIMER pSubTimer = &pTimer->aSubTimers[idCpu];
1063	if (rtTimerLnxCmpXchgState(&pSubTimer->enmState, RTTIMERLNXSTATE_MP_STARTING, RTTIMERLNXSTATE_STOPPED))
1064	rtTimerLnxStartSubTimer(pSubTimer, pArgs->u64Now, pArgs->u64First, true /fPinned/, pTimer->fHighRes);
1065	}
1066
1067	RTSpinlockRelease(hSpinlock);
1068	}
1069	}
1070
1071
1072	/**
1073	* MP event notification callback.
1074	*
1075	* @param enmEvent The event.
1076	* @param idCpu The cpu it applies to.
1077	* @param pvUser The timer.
1078	*/
1079	static DECLCALLBACK(void) rtTimerLinuxMpEvent(RTMPEVENT enmEvent, RTCPUID idCpu, void *pvUser)
1080	{
1081	PRTTIMER pTimer = (PRTTIMER)pvUser;
1082	PRTTIMERLNXSUBTIMER pSubTimer = &pTimer->aSubTimers[idCpu];
1083	RTSPINLOCK hSpinlock;
1084
1085	Assert(idCpu < pTimer->cCpus);
1086
1087	/*
1088	* Some initial paranoia.
1089	*/
1090	if (pTimer->u32Magic != RTTIMER_MAGIC)
1091	return;
1092	hSpinlock = pTimer->hSpinlock;
1093	if (hSpinlock == NIL_RTSPINLOCK)
1094	return;
1095
1096	RTSpinlockAcquire(hSpinlock);
1097
1098	/* Is it active? */
1099	if ( !ASMAtomicUoReadBool(&pTimer->fSuspended)
1100	&& pTimer->u32Magic == RTTIMER_MAGIC)
1101	{
1102	switch (enmEvent)
1103	{
1104	/*
1105	* Try do it without leaving the spin lock, but if we have to, retake it
1106	* when we're on the right cpu.
1107	*/
1108	case RTMPEVENT_ONLINE:
1109	if (rtTimerLnxCmpXchgState(&pSubTimer->enmState, RTTIMERLNXSTATE_MP_STARTING, RTTIMERLNXSTATE_STOPPED))
1110	{
1111	RTTIMERLINUXSTARTONCPUARGS Args;
1112	Args.u64Now = RTTimeSystemNanoTS();
1113	Args.u64First = 0;
1114
1115	if (RTMpCpuId() == idCpu)
1116	rtTimerLnxStartSubTimer(pSubTimer, Args.u64Now, Args.u64First, true /fPinned/, pTimer->fHighRes);
1117	else
1118	{
1119	rtTimerLnxSetState(&pSubTimer->enmState, RTTIMERLNXSTATE_STOPPED); /* we'll recheck it. */
1120	RTSpinlockRelease(hSpinlock);
1121
1122	RTMpOnSpecific(idCpu, rtTimerLinuxMpStartOnCpu, pTimer, &Args);
1123	return; /* we've left the spinlock */
1124	}
1125	}
1126	break;
1127
1128	/*
1129	* The CPU is (going) offline, make sure the sub-timer is stopped.
1130	*
1131	* Linux will migrate it to a different CPU, but we don't want this. The
1132	* timer function is checking for this.
1133	*/
1134	case RTMPEVENT_OFFLINE:
1135	{
1136	RTTIMERLNXSTATE enmState;
1137	while ( (enmState = rtTimerLnxGetState(&pSubTimer->enmState)) == RTTIMERLNXSTATE_ACTIVE
1138	\|\| enmState == RTTIMERLNXSTATE_CALLBACK
1139	\|\| enmState == RTTIMERLNXSTATE_CB_RESTARTING)
1140	{
1141	if (enmState == RTTIMERLNXSTATE_ACTIVE)
1142	{
1143	if (rtTimerLnxCmpXchgState(&pSubTimer->enmState, RTTIMERLNXSTATE_MP_STOPPING, RTTIMERLNXSTATE_ACTIVE))
1144	{
1145	RTSpinlockRelease(hSpinlock);
1146
1147	rtTimerLnxStopSubTimer(pSubTimer, pTimer->fHighRes);
1148	return; /* we've left the spinlock */
1149	}
1150	}
1151	else if (rtTimerLnxCmpXchgState(&pSubTimer->enmState, RTTIMERLNXSTATE_CB_STOPPING, enmState))
1152	break;
1153
1154	/* State not stable, try again. */
1155	ASMNopPause();
1156	}
1157	break;
1158	}
1159	}
1160	}
1161
1162	RTSpinlockRelease(hSpinlock);
1163	}
1164
1165	#endif /* CONFIG_SMP */
1166
1167
1168	/**
1169	* Callback function use by RTTimerStart via RTMpOnSpecific to start a timer
1170	* running on a specific CPU.
1171	*
1172	* @param idCpu The current CPU.
1173	* @param pvUser1 Pointer to the timer.
1174	* @param pvUser2 Pointer to the argument structure.
1175	*/
1176	static DECLCALLBACK(void) rtTimerLnxStartOnSpecificCpu(RTCPUID idCpu, void pvUser1, void pvUser2)
1177	{
1178	PRTTIMERLINUXSTARTONCPUARGS pArgs = (PRTTIMERLINUXSTARTONCPUARGS)pvUser2;
1179	PRTTIMER pTimer = (PRTTIMER)pvUser1;
1180	rtTimerLnxStartSubTimer(&pTimer->aSubTimers[0], pArgs->u64Now, pArgs->u64First, true /fPinned/, pTimer->fHighRes);
1181	}
1182
1183
1184	RTDECL(int) RTTimerStart(PRTTIMER pTimer, uint64_t u64First)
1185	{
1186	RTTIMERLINUXSTARTONCPUARGS Args;
1187	int rc2;
1188
1189	/*
1190	* Validate.
1191	*/
1192	AssertPtrReturn(pTimer, VERR_INVALID_HANDLE);
1193	AssertReturn(pTimer->u32Magic == RTTIMER_MAGIC, VERR_INVALID_HANDLE);
1194
1195	if (!ASMAtomicUoReadBool(&pTimer->fSuspended))
1196	return VERR_TIMER_ACTIVE;
1197	RTTIMERLNX_LOG(("start %p cCpus=%d\n", pTimer, pTimer->cCpus));
1198
1199	Args.u64First = u64First;
1200	#ifdef CONFIG_SMP
1201	/*
1202	* Omni timer?
1203	*/
1204	if (pTimer->fAllCpus)
1205	return rtTimerLnxOmniStart(pTimer, &Args);
1206	#endif
1207
1208	/*
1209	* Simple timer - Pretty straight forward if it wasn't for restarting.
1210	*/
1211	Args.u64Now = RTTimeSystemNanoTS();
1212	ASMAtomicWriteU64(&pTimer->aSubTimers[0].uNsRestartAt, Args.u64Now + u64First);
1213	for (;;)
1214	{
1215	RTTIMERLNXSTATE enmState = rtTimerLnxGetState(&pTimer->aSubTimers[0].enmState);
1216	switch (enmState)
1217	{
1218	case RTTIMERLNXSTATE_STOPPED:
1219	if (rtTimerLnxCmpXchgState(&pTimer->aSubTimers[0].enmState, RTTIMERLNXSTATE_STARTING, RTTIMERLNXSTATE_STOPPED))
1220	{
1221	ASMAtomicWriteBool(&pTimer->fSuspended, false);
1222	if (!pTimer->fSpecificCpu)
1223	rtTimerLnxStartSubTimer(&pTimer->aSubTimers[0], Args.u64Now, Args.u64First,
1224	false /fPinned/, pTimer->fHighRes);
1225	else
1226	{
1227	rc2 = RTMpOnSpecific(pTimer->idCpu, rtTimerLnxStartOnSpecificCpu, pTimer, &Args);
1228	if (RT_FAILURE(rc2))
1229	{
1230	/* Suspend it, the cpu id is probably invalid or offline. */
1231	ASMAtomicWriteBool(&pTimer->fSuspended, true);
1232	rtTimerLnxSetState(&pTimer->aSubTimers[0].enmState, RTTIMERLNXSTATE_STOPPED);
1233	return rc2;
1234	}
1235	}
1236	return VINF_SUCCESS;
1237	}
1238	break;
1239
1240	case RTTIMERLNXSTATE_CALLBACK:
1241	case RTTIMERLNXSTATE_CB_STOPPING:
1242	if (rtTimerLnxCmpXchgState(&pTimer->aSubTimers[0].enmState, RTTIMERLNXSTATE_CB_RESTARTING, enmState))
1243	{
1244	ASMAtomicWriteBool(&pTimer->fSuspended, false);
1245	return VINF_SUCCESS;
1246	}
1247	break;
1248
1249	default:
1250	AssertMsgFailed(("%d\n", enmState));
1251	return VERR_INTERNAL_ERROR_4;
1252	}
1253	ASMNopPause();
1254	}
1255	}
1256	RT_EXPORT_SYMBOL(RTTimerStart);
1257
1258
1259	/**
1260	* Common worker for RTTimerStop and RTTimerDestroy.
1261	*
1262	* @returns true if there was any active callbacks, false if not.
1263	* @param pTimer The timer to stop.
1264	* @param fForDestroy Whether it's RTTimerDestroy calling or not.
1265	*/
1266	static bool rtTimerLnxStop(PRTTIMER pTimer, bool fForDestroy)
1267	{
1268	RTTIMERLNX_LOG(("lnxstop %p %d\n", pTimer, fForDestroy));
1269	#ifdef CONFIG_SMP
1270	/*
1271	* Omni timer?
1272	*/
1273	if (pTimer->fAllCpus)
1274	return rtTimerLnxOmniStop(pTimer, fForDestroy);
1275	#endif
1276
1277	/*
1278	* Simple timer.
1279	*/
1280	ASMAtomicWriteBool(&pTimer->fSuspended, true);
1281	for (;;)
1282	{
1283	RTTIMERLNXSTATE enmState = rtTimerLnxGetState(&pTimer->aSubTimers[0].enmState);
1284	switch (enmState)
1285	{
1286	case RTTIMERLNXSTATE_ACTIVE:
1287	if (rtTimerLnxCmpXchgState(&pTimer->aSubTimers[0].enmState, RTTIMERLNXSTATE_STOPPING, RTTIMERLNXSTATE_ACTIVE))
1288	{
1289	rtTimerLnxStopSubTimer(&pTimer->aSubTimers[0], pTimer->fHighRes);
1290	return false;
1291	}
1292	break;
1293
1294	case RTTIMERLNXSTATE_CALLBACK:
1295	case RTTIMERLNXSTATE_CB_RESTARTING:
1296	case RTTIMERLNXSTATE_CB_STOPPING:
1297	Assert(enmState != RTTIMERLNXSTATE_CB_STOPPING \|\| fForDestroy);
1298	if (rtTimerLnxCmpXchgState(&pTimer->aSubTimers[0].enmState,
1299	!fForDestroy ? RTTIMERLNXSTATE_CB_STOPPING : RTTIMERLNXSTATE_CB_DESTROYING,
1300	enmState))
1301	return true;
1302	break;
1303
1304	case RTTIMERLNXSTATE_STOPPED:
1305	return VINF_SUCCESS;
1306
1307	case RTTIMERLNXSTATE_CB_DESTROYING:
1308	AssertMsgFailed(("enmState=%d pTimer=%p\n", enmState, pTimer));
1309	return true;
1310
1311	default:
1312	case RTTIMERLNXSTATE_STARTING:
1313	case RTTIMERLNXSTATE_MP_STARTING:
1314	case RTTIMERLNXSTATE_STOPPING:
1315	case RTTIMERLNXSTATE_MP_STOPPING:
1316	AssertMsgFailed(("enmState=%d pTimer=%p\n", enmState, pTimer));
1317	return false;
1318	}
1319
1320	/* State not stable, try again. */
1321	ASMNopPause();
1322	}
1323	}
1324
1325
1326	RTDECL(int) RTTimerStop(PRTTIMER pTimer)
1327	{
1328	/*
1329	* Validate.
1330	*/
1331	AssertPtrReturn(pTimer, VERR_INVALID_HANDLE);
1332	AssertReturn(pTimer->u32Magic == RTTIMER_MAGIC, VERR_INVALID_HANDLE);
1333	RTTIMERLNX_LOG(("stop %p\n", pTimer));
1334
1335	if (ASMAtomicUoReadBool(&pTimer->fSuspended))
1336	return VERR_TIMER_SUSPENDED;
1337
1338	rtTimerLnxStop(pTimer, false /fForDestroy/);
1339	return VINF_SUCCESS;
1340	}
1341	RT_EXPORT_SYMBOL(RTTimerStop);
1342
1343
1344	RTDECL(int) RTTimerChangeInterval(PRTTIMER pTimer, uint64_t u64NanoInterval)
1345	{
1346	unsigned long cJiffies;
1347	unsigned long flFlags;
1348
1349	/*
1350	* Validate.
1351	*/
1352	AssertPtrReturn(pTimer, VERR_INVALID_HANDLE);
1353	AssertReturn(pTimer->u32Magic == RTTIMER_MAGIC, VERR_INVALID_HANDLE);
1354	AssertReturn(u64NanoInterval, VERR_INVALID_PARAMETER);
1355	RTTIMERLNX_LOG(("change %p %llu\n", pTimer, u64NanoInterval));
1356
1357	#ifdef RTTIMER_LINUX_WITH_HRTIMER
1358	/*
1359	* For the high resolution timers it is easy since we don't care so much
1360	* about when it is applied to the sub-timers.
1361	*/
1362	if (pTimer->fHighRes)
1363	{
1364	ASMAtomicWriteU64(&pTimer->u64NanoInterval, u64NanoInterval);
1365	return VINF_SUCCESS;
1366	}
1367	#endif
1368
1369	/*
1370	* Standard timers have a bit more complicated way of calculating
1371	* their interval and such. So, forget omni timers for now.
1372	*/
1373	if (pTimer->cCpus > 1)
1374	return VERR_NOT_SUPPORTED;
1375
1376	cJiffies = u64NanoInterval / RTTimerGetSystemGranularity();
1377	if (cJiffies * RTTimerGetSystemGranularity() != u64NanoInterval)
1378	cJiffies = 0;
1379
1380	spin_lock_irqsave(&pTimer->ChgIntLock, flFlags);
1381	pTimer->aSubTimers[0].u.Std.fFirstAfterChg = true;
1382	pTimer->cJiffies = cJiffies;
1383	ASMAtomicWriteU64(&pTimer->u64NanoInterval, u64NanoInterval);
1384	spin_unlock_irqrestore(&pTimer->ChgIntLock, flFlags);
1385	return VINF_SUCCESS;
1386	}
1387	RT_EXPORT_SYMBOL(RTTimerChangeInterval);
1388
1389
1390	RTDECL(int) RTTimerDestroy(PRTTIMER pTimer)
1391	{
1392	bool fCanDestroy;
1393
1394	/*
1395	* Validate. It's ok to pass NULL pointer.
1396	*/
1397	if (pTimer == /NIL_RTTIMER/ NULL)
1398	return VINF_SUCCESS;
1399	AssertPtrReturn(pTimer, VERR_INVALID_HANDLE);
1400	AssertReturn(pTimer->u32Magic == RTTIMER_MAGIC, VERR_INVALID_HANDLE);
1401	RTTIMERLNX_LOG(("destroy %p\n", pTimer));
1402	/** @todo We should invalidate the magic here! */
1403
1404	/*
1405	* Stop the timer if it's still active, then destroy it if we can.
1406	*/
1407	if (!ASMAtomicUoReadBool(&pTimer->fSuspended))
1408	fCanDestroy = rtTimerLnxStop(pTimer, true /fForDestroy/);
1409	else
1410	{
1411	uint32_t iCpu = pTimer->cCpus;
1412	if (pTimer->cCpus > 1)
1413	RTSpinlockAcquire(pTimer->hSpinlock);
1414
1415	fCanDestroy = true;
1416	while (iCpu-- > 0)
1417	{
1418	for (;;)
1419	{
1420	RTTIMERLNXSTATE enmState = rtTimerLnxGetState(&pTimer->aSubTimers[iCpu].enmState);
1421	switch (enmState)
1422	{
1423	case RTTIMERLNXSTATE_CALLBACK:
1424	case RTTIMERLNXSTATE_CB_RESTARTING:
1425	case RTTIMERLNXSTATE_CB_STOPPING:
1426	if (!rtTimerLnxCmpXchgState(&pTimer->aSubTimers[iCpu].enmState, RTTIMERLNXSTATE_CB_DESTROYING, enmState))
1427	continue;
1428	fCanDestroy = false;
1429	break;
1430
1431	case RTTIMERLNXSTATE_CB_DESTROYING:
1432	AssertMsgFailed(("%d\n", enmState));
1433	fCanDestroy = false;
1434	break;
1435	default:
1436	break;
1437	}
1438	break;
1439	}
1440	}
1441
1442	if (pTimer->cCpus > 1)
1443	RTSpinlockReleaseNoInts(pTimer->hSpinlock);
1444	}
1445
1446	if (fCanDestroy)
1447	{
1448	/* For paranoid reasons, defer actually destroying the semaphore when
1449	in atomic or interrupt context. */
1450	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 32)
1451	if (in_atomic() \|\| in_interrupt())
1452	#else
1453	if (in_interrupt())
1454	#endif
1455	rtR0LnxWorkqueuePush(&pTimer->DtorWorkqueueItem, rtTimerLnxDestroyDeferred);
1456	else
1457	rtTimerLnxDestroyIt(pTimer);
1458	}
1459
1460	return VINF_SUCCESS;
1461	}
1462	RT_EXPORT_SYMBOL(RTTimerDestroy);
1463
1464
1465	RTDECL(int) RTTimerCreateEx(PRTTIMER ppTimer, uint64_t u64NanoInterval, uint32_t fFlags, PFNRTTIMER pfnTimer, void pvUser)
1466	{
1467	PRTTIMER pTimer;
1468	RTCPUID iCpu;
1469	unsigned cCpus;
1470	int rc;
1471
1472	rtR0LnxWorkqueueFlush(); /* for 2.4 */
1473	*ppTimer = NULL;
1474
1475	/*
1476	* Validate flags.
1477	*/
1478	if (!RTTIMER_FLAGS_ARE_VALID(fFlags))
1479	return VERR_INVALID_PARAMETER;
1480	if ( (fFlags & RTTIMER_FLAGS_CPU_SPECIFIC)
1481	&& (fFlags & RTTIMER_FLAGS_CPU_ALL) != RTTIMER_FLAGS_CPU_ALL
1482	&& !RTMpIsCpuPossible(RTMpCpuIdFromSetIndex(fFlags & RTTIMER_FLAGS_CPU_MASK)))
1483	return VERR_CPU_NOT_FOUND;
1484
1485	/*
1486	* Allocate the timer handler.
1487	*/
1488	cCpus = 1;
1489	#ifdef CONFIG_SMP
1490	if ((fFlags & RTTIMER_FLAGS_CPU_ALL) == RTTIMER_FLAGS_CPU_ALL)
1491	{
1492	cCpus = RTMpGetMaxCpuId() + 1;
1493	Assert(cCpus <= RTCPUSET_MAX_CPUS); /* On linux we have a 1:1 relationship between cpuid and set index. */
1494	AssertReturn(u64NanoInterval, VERR_NOT_IMPLEMENTED); /* We don't implement single shot on all cpus, sorry. */
1495	}
1496	#endif
1497
1498	rc = RTMemAllocEx(RT_OFFSETOF(RTTIMER, aSubTimers[cCpus]), 0,
1499	RTMEMALLOCEX_FLAGS_ZEROED \| RTMEMALLOCEX_FLAGS_ANY_CTX_FREE, (void **)&pTimer);
1500	if (RT_FAILURE(rc))
1501	return rc;
1502
1503	/*
1504	* Initialize it.
1505	*/
1506	pTimer->u32Magic = RTTIMER_MAGIC;
1507	pTimer->hSpinlock = NIL_RTSPINLOCK;
1508	pTimer->fSuspended = true;
1509	pTimer->fHighRes = !!(fFlags & RTTIMER_FLAGS_HIGH_RES);
1510	#ifdef CONFIG_SMP
1511	pTimer->fSpecificCpu = (fFlags & RTTIMER_FLAGS_CPU_SPECIFIC) && (fFlags & RTTIMER_FLAGS_CPU_ALL) != RTTIMER_FLAGS_CPU_ALL;
1512	pTimer->fAllCpus = (fFlags & RTTIMER_FLAGS_CPU_ALL) == RTTIMER_FLAGS_CPU_ALL;
1513	pTimer->idCpu = pTimer->fSpecificCpu
1514	? RTMpCpuIdFromSetIndex(fFlags & RTTIMER_FLAGS_CPU_MASK)
1515	: NIL_RTCPUID;
1516	#else
1517	pTimer->fSpecificCpu = !!(fFlags & RTTIMER_FLAGS_CPU_SPECIFIC);
1518	pTimer->idCpu = RTMpCpuId();
1519	#endif
1520	pTimer->cCpus = cCpus;
1521	pTimer->pfnTimer = pfnTimer;
1522	pTimer->pvUser = pvUser;
1523	pTimer->u64NanoInterval = u64NanoInterval;
1524	pTimer->cJiffies = u64NanoInterval / RTTimerGetSystemGranularity();
1525	if (pTimer->cJiffies * RTTimerGetSystemGranularity() != u64NanoInterval)
1526	pTimer->cJiffies = 0;
1527	spin_lock_init(&pTimer->ChgIntLock);
1528
1529	for (iCpu = 0; iCpu < cCpus; iCpu++)
1530	{
1531	#ifdef RTTIMER_LINUX_WITH_HRTIMER
1532	if (pTimer->fHighRes)
1533	{
1534	hrtimer_init(&pTimer->aSubTimers[iCpu].u.Hr.LnxTimer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
1535	pTimer->aSubTimers[iCpu].u.Hr.LnxTimer.function = rtTimerLinuxHrCallback;
1536	}
1537	else
1538	#endif
1539	{
1540	init_timer(&pTimer->aSubTimers[iCpu].u.Std.LnxTimer);
1541	pTimer->aSubTimers[iCpu].u.Std.LnxTimer.data = (unsigned long)&pTimer->aSubTimers[iCpu];
1542	pTimer->aSubTimers[iCpu].u.Std.LnxTimer.function = rtTimerLinuxStdCallback;
1543	pTimer->aSubTimers[iCpu].u.Std.LnxTimer.expires = jiffies;
1544	pTimer->aSubTimers[iCpu].u.Std.u64NextTS = 0;
1545	}
1546	pTimer->aSubTimers[iCpu].iTick = 0;
1547	pTimer->aSubTimers[iCpu].pParent = pTimer;
1548	pTimer->aSubTimers[iCpu].enmState = RTTIMERLNXSTATE_STOPPED;
1549	}
1550
1551	#ifdef CONFIG_SMP
1552	/*
1553	* If this is running on ALL cpus, we'll have to register a callback
1554	* for MP events (so timers can be started/stopped on cpus going
1555	* online/offline). We also create the spinlock for synchronizing
1556	* stop/start/mp-event.
1557	*/
1558	if (cCpus > 1)
1559	{
1560	int rc = RTSpinlockCreate(&pTimer->hSpinlock, RTSPINLOCK_FLAGS_INTERRUPT_SAFE, "RTTimerLnx");
1561	if (RT_SUCCESS(rc))
1562	rc = RTMpNotificationRegister(rtTimerLinuxMpEvent, pTimer);
1563	else
1564	pTimer->hSpinlock = NIL_RTSPINLOCK;
1565	if (RT_FAILURE(rc))
1566	{
1567	RTTimerDestroy(pTimer);
1568	return rc;
1569	}
1570	}
1571	#endif /* CONFIG_SMP */
1572
1573	RTTIMERLNX_LOG(("create %p hires=%d fFlags=%#x cCpus=%u\n", pTimer, pTimer->fHighRes, fFlags, cCpus));
1574	*ppTimer = pTimer;
1575	return VINF_SUCCESS;
1576	}
1577	RT_EXPORT_SYMBOL(RTTimerCreateEx);
1578
1579
1580	RTDECL(uint32_t) RTTimerGetSystemGranularity(void)
1581	{
1582	#if 0 /** @todo Not sure if this is what we want or not... Add new API for
1583	* querying the resolution of the high res timers? */
1584	struct timespec Ts;
1585	int rc = hrtimer_get_res(CLOCK_MONOTONIC, &Ts);
1586	if (!rc)
1587	{
1588	Assert(!Ts.tv_sec);
1589	return Ts.tv_nsec;
1590	}
1591	#endif
1592	return 1000000000 / HZ; /* ns */
1593	}
1594	RT_EXPORT_SYMBOL(RTTimerGetSystemGranularity);
1595
1596
1597	RTDECL(int) RTTimerRequestSystemGranularity(uint32_t u32Request, uint32_t *pu32Granted)
1598	{
1599	return VERR_NOT_SUPPORTED;
1600	}
1601	RT_EXPORT_SYMBOL(RTTimerRequestSystemGranularity);
1602
1603
1604	RTDECL(int) RTTimerReleaseSystemGranularity(uint32_t u32Granted)
1605	{
1606	return VERR_NOT_SUPPORTED;
1607	}
1608	RT_EXPORT_SYMBOL(RTTimerReleaseSystemGranularity);
1609
1610
1611	RTDECL(bool) RTTimerCanDoHighResolution(void)
1612	{
1613	#ifdef RTTIMER_LINUX_WITH_HRTIMER
1614	return true;
1615	#else
1616	return false;
1617	#endif
1618	}
1619	RT_EXPORT_SYMBOL(RTTimerCanDoHighResolution);
1620

Note: See TracBrowser for help on using the repository browser.

source: vbox/trunk/src/VBox/Runtime/r0drv/linux/timer-r0drv-linux.c@ 44824

Download in other formats: