SUPDrv-linux.c@ 7521

Last change on this file since 7521 was 7521, checked in by vboxsync, 17 years ago
vboxdrv: made it compile with Linux 2.6.24 if CONFIG_VBOXDRV_FIXEDMAJOR is defined
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 42.6 KB

Line
1	/** @file
2	* The VirtualBox Support Driver - Linux hosts.
3	*/
4
5	/*
6	* Copyright (C) 2006-2007 innotek GmbH
7	*
8	* This file is part of VirtualBox Open Source Edition (OSE), as
9	* available from http://www.virtualbox.org. This file is free software;
10	* you can redistribute it and/or modify it under the terms of the GNU
11	* General Public License (GPL) as published by the Free Software
12	* Foundation, in version 2 as it comes in the "COPYING" file of the
13	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
14	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
15	*
16	* The contents of this file may alternatively be used under the terms
17	* of the Common Development and Distribution License Version 1.0
18	* (CDDL) only, as it comes in the "COPYING.CDDL" file of the
19	* VirtualBox OSE distribution, in which case the provisions of the
20	* CDDL are applicable instead of those of the GPL.
21	*
22	* You may elect to license modified versions of this file under the
23	* terms and conditions of either the GPL or the CDDL or both.
24	* Some lines of code to disable the local APIC on x86_64 machines taken
25	* from a Mandriva patch by Gwenole Beauchesne <[email protected]>.
26	*/
27
28	/*******************************************************************************
29	* Header Files *
30	*******************************************************************************/
31	#include "SUPDRV.h"
32	#include "the-linux-kernel.h"
33	#include "version-generated.h"
34
35	#include <iprt/assert.h>
36	#include <iprt/spinlock.h>
37	#include <iprt/semaphore.h>
38	#include <iprt/initterm.h>
39	#include <iprt/process.h>
40	#include <iprt/err.h>
41	#include <iprt/mem.h>
42	#include <iprt/log.h>
43
44	#include <linux/sched.h>
45	#ifdef CONFIG_DEVFS_FS
46	# include <linux/devfs_fs_kernel.h>
47	#endif
48	#ifdef CONFIG_VBOXDRV_AS_MISC
49	# include <linux/miscdevice.h>
50	#endif
51	#ifdef CONFIG_X86_LOCAL_APIC
52	# include <asm/apic.h>
53	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
54	# include <asm/nmi.h>
55	# endif
56	#endif
57
58	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0)
59	# include <asm/pgtable.h>
60	# define global_flush_tlb __flush_tlb_global
61	#endif
62
63	#include <iprt/mem.h>
64
65
66	/* devfs defines */
67	#if defined(CONFIG_DEVFS_FS) && !defined(CONFIG_VBOXDRV_AS_MISC)
68	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
69
70	# define VBOX_REGISTER_DEVFS() \
71	({ \
72	void *rc = NULL; \
73	if (devfs_mk_cdev(MKDEV(DEVICE_MAJOR, 0), \
74	S_IFCHR \| S_IRUGO \| S_IWUGO, \
75	DEVICE_NAME) == 0) \
76	rc = (void )' '; / return not NULL */ \
77	rc; \
78	})
79
80	# define VBOX_UNREGISTER_DEVFS(handle) \
81	devfs_remove(DEVICE_NAME);
82
83	# else /* < 2.6.0 */
84
85	# define VBOX_REGISTER_DEVFS() \
86	devfs_register(NULL, DEVICE_NAME, DEVFS_FL_DEFAULT, \
87	DEVICE_MAJOR, 0, \
88	S_IFCHR \| S_IRUGO \| S_IWUGO, \
89	&gFileOpsVBoxDrv, NULL)
90
91	# define VBOX_UNREGISTER_DEVFS(handle) \
92	if (handle != NULL) \
93	devfs_unregister(handle)
94
95	# endif /* < 2.6.0 */
96	#endif /* CONFIG_DEV_FS && !CONFIG_VBOXDEV_AS_MISC */
97
98	#ifndef CONFIG_VBOXDRV_AS_MISC
99	# if defined(CONFIG_DEVFS_FS) && LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 0)
100	# define VBOX_REGISTER_DEVICE(a,b,c) devfs_register_chrdev(a,b,c)
101	# define VBOX_UNREGISTER_DEVICE(a,b) devfs_unregister_chrdev(a,b)
102	# else
103	# define VBOX_REGISTER_DEVICE(a,b,c) register_chrdev(a,b,c)
104	# define VBOX_UNREGISTER_DEVICE(a,b) unregister_chrdev(a,b)
105	# endif
106	#endif /* !CONFIG_VBOXDRV_AS_MISC */
107
108
109	#ifdef CONFIG_X86_HIGH_ENTRY
110	# error "CONFIG_X86_HIGH_ENTRY is not supported by VBoxDrv at this time."
111	#endif
112
113	/*
114	* This sucks soooo badly on x86! Why don't they export __PAGE_KERNEL_EXEC so PAGE_KERNEL_EXEC would be usable?
115	*/
116	#if defined(RT_ARCH_AMD64)
117	# define MY_PAGE_KERNEL_EXEC PAGE_KERNEL_EXEC
118	#elif defined(PAGE_KERNEL_EXEC) && defined(CONFIG_X86_PAE)
119	# define MY_PAGE_KERNEL_EXEC __pgprot(cpu_has_pge ? _PAGE_KERNEL_EXEC \| _PAGE_GLOBAL : _PAGE_KERNEL_EXEC)
120	#else
121	# define MY_PAGE_KERNEL_EXEC PAGE_KERNEL
122	#endif
123
124	/*
125	* The redhat hack section.
126	* - The current hacks are for 2.4.21-15.EL only.
127	*/
128	#ifndef NO_REDHAT_HACKS
129	/* accounting. */
130	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0)
131	# ifdef VM_ACCOUNT
132	# define MY_DO_MUNMAP(a,b,c) do_munmap(a, b, c, 0) /* should it be 1 or 0? */
133	# endif
134	# endif
135
136	/* backported remap_page_range. */
137	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0)
138	# include <asm/tlb.h>
139	# ifdef tlb_vma /* probably not good enough... */
140	# define HAVE_26_STYLE_REMAP_PAGE_RANGE 1
141	# endif
142	# endif
143
144	#endif /* !NO_REDHAT_HACKS */
145
146
147	#ifndef MY_DO_MUNMAP
148	# define MY_DO_MUNMAP(a,b,c) do_munmap(a, b, c)
149	#endif
150
151
152	/** @def ONE_MSEC_IN_JIFFIES
153	* The number of jiffies that make up 1 millisecond. Must be at least 1! */
154	#if HZ <= 1000
155	# define ONE_MSEC_IN_JIFFIES 1
156	#elif !(HZ % 1000)
157	# define ONE_MSEC_IN_JIFFIES (HZ / 1000)
158	#else
159	# define ONE_MSEC_IN_JIFFIES ((HZ + 999) / 1000)
160	# error "HZ is not a multiple of 1000, the GIP stuff won't work right!"
161	#endif
162
163	/** @def TICK_NSEC
164	* The time between ticks in nsec */
165	#ifndef TICK_NSEC
166	# define TICK_NSEC (1000000UL / HZ)
167	#endif
168
169	#ifdef CONFIG_X86_LOCAL_APIC
170
171	/* If an NMI occurs while we are inside the world switcher the machine will
172	* crash. The Linux NMI watchdog generates periodic NMIs increasing a counter
173	* which is compared with another counter increased in the timer interrupt
174	* handler. We disable the NMI watchdog.
175	*
176	* - Linux >= 2.6.21: The watchdog is disabled by default on i386 and x86_64.
177	* - Linux < 2.6.21: The watchdog is normally enabled by default on x86_64
178	* and disabled on i386.
179	*/
180	# if defined(RT_ARCH_AMD64)
181	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 21) && !defined(VBOX_REDHAT_KABI)
182	# define DO_DISABLE_NMI 1
183	# endif
184	# endif
185
186	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 19)
187	extern int nmi_active;
188	# define nmi_atomic_read(P) *(P)
189	# define nmi_atomic_set(P, V) *(P) = (V)
190	# define nmi_atomic_dec(P) nmi_atomic_set(P, 0)
191	# else
192	# define nmi_atomic_read(P) atomic_read(P)
193	# define nmi_atomic_set(P, V) atomic_set(P, V)
194	# define nmi_atomic_dec(P) atomic_dec(P)
195	# endif
196
197	# ifndef X86_FEATURE_ARCH_PERFMON
198	# define X86_FEATURE_ARCH_PERFMON (332+9) / Intel Architectural PerfMon */
199	# endif
200	# ifndef MSR_ARCH_PERFMON_EVENTSEL0
201	# define MSR_ARCH_PERFMON_EVENTSEL0 0x186
202	# endif
203	# ifndef ARCH_PERFMON_UNHALTED_CORE_CYCLES_PRESENT
204	# define ARCH_PERFMON_UNHALTED_CORE_CYCLES_PRESENT (1 << 0)
205	# endif
206
207	#endif /* CONFIG_X86_LOCAL_APIC */
208
209	#define xstr(s) str(s)
210	#define str(s) #s
211
212	/*******************************************************************************
213	* Defined Constants And Macros *
214	*******************************************************************************/
215	/**
216	* Device extention & session data association structure.
217	*/
218	static SUPDRVDEVEXT g_DevExt;
219
220	/** Timer structure for the GIP update. */
221	static VBOXKTIMER g_GipTimer;
222	/** Pointer to the page structure for the GIP. */
223	struct page *g_pGipPage;
224
225	/** Registered devfs device handle. */
226	#if defined(CONFIG_DEVFS_FS) && !defined(CONFIG_VBOXDRV_AS_MISC)
227	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
228	static void *g_hDevFsVBoxDrv = NULL;
229	# else
230	static devfs_handle_t g_hDevFsVBoxDrv = NULL;
231	# endif
232	#endif
233
234	#ifndef CONFIG_VBOXDRV_AS_MISC
235	/** Module major number */
236	#define DEVICE_MAJOR 234
237	/** Saved major device number */
238	static int g_iModuleMajor;
239	#endif /* !CONFIG_VBOXDRV_AS_MISC */
240
241	/** Module parameter.
242	* Not prefixed because the name is used by macros and the end of this file. */
243	static int force_async_tsc = 0;
244
245	/** The module name. */
246	#define DEVICE_NAME "vboxdrv"
247
248	#ifdef RT_ARCH_AMD64
249	/**
250	* Memory for the executable memory heap (in IPRT).
251	*/
252	extern uint8_t g_abExecMemory[1572864]; /* 1.5 MB */
253	__asm__(".section execmemory, \"awx\", @progbits\n\t"
254	".align 32\n\t"
255	".globl g_abExecMemory\n"
256	"g_abExecMemory:\n\t"
257	".zero 1572864\n\t"
258	".type g_abExecMemory, @object\n\t"
259	".size g_abExecMemory, 1572864\n\t"
260	".text\n\t");
261	#endif
262
263
264	/*******************************************************************************
265	* Internal Functions *
266	*******************************************************************************/
267	#ifdef VBOX_HRTIMER
268	typedef enum hrtimer_restart (PFNVBOXKTIMER)(struct hrtimer );
269	#else
270	typedef void (*PFNVBOXKTIMER)(unsigned long);
271	#endif
272
273	static int VBoxDrvLinuxInit(void);
274	static void VBoxDrvLinuxUnload(void);
275	static int VBoxDrvLinuxCreate(struct inode pInode, struct file pFilp);
276	static int VBoxDrvLinuxClose(struct inode pInode, struct file pFilp);
277	#ifdef HAVE_UNLOCKED_IOCTL
278	static long VBoxDrvLinuxIOCtl(struct file *pFilp, unsigned int uCmd, unsigned long ulArg);
279	#else
280	static int VBoxDrvLinuxIOCtl(struct inode pInode, struct file pFilp, unsigned int uCmd, unsigned long ulArg);
281	#endif
282	static int VBoxDrvLinuxIOCtlSlow(struct file *pFilp, unsigned int uCmd, unsigned long ulArg);
283	static int VBoxDrvLinuxInitGip(PSUPDRVDEVEXT pDevExt);
284	static int VBoxDrvLinuxTermGip(PSUPDRVDEVEXT pDevExt);
285	#ifdef VBOX_HRTIMER
286	static enum hrtimer_restart VBoxDrvLinuxGipTimer(struct hrtimer *pTimer);
287	#else
288	static void VBoxDrvLinuxGipTimer(unsigned long ulUser);
289	#endif
290	#ifdef CONFIG_SMP
291	# ifdef VBOX_HRTIMER
292	static enum hrtimer_restart VBoxDrvLinuxGipTimerPerCpu(struct hrtimer *pTimer);
293	# else
294	static void VBoxDrvLinuxGipTimerPerCpu(unsigned long ulUser);
295	# endif
296	static void VBoxDrvLinuxGipResumePerCpu(void *pvUser);
297	#endif
298	static int VBoxDrvLinuxErr2LinuxErr(int);
299
300
301	/** The file_operations structure. */
302	static struct file_operations gFileOpsVBoxDrv =
303	{
304	owner: THIS_MODULE,
305	open: VBoxDrvLinuxCreate,
306	release: VBoxDrvLinuxClose,
307	#ifdef HAVE_UNLOCKED_IOCTL
308	unlocked_ioctl: VBoxDrvLinuxIOCtl,
309	#else
310	ioctl: VBoxDrvLinuxIOCtl,
311	#endif
312	};
313
314	#ifdef CONFIG_VBOXDRV_AS_MISC
315	/** The miscdevice structure. */
316	static struct miscdevice gMiscDevice =
317	{
318	minor: MISC_DYNAMIC_MINOR,
319	name: DEVICE_NAME,
320	fops: &gFileOpsVBoxDrv,
321	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) && \
322	LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 17)
323	devfs_name: DEVICE_NAME,
324	# endif
325	};
326	#endif
327
328	static inline void vbox_ktimer_init(PVBOXKTIMER pTimer, PFNVBOXKTIMER pfnFunction, unsigned long ulData)
329	{
330	#ifdef VBOX_HRTIMER
331	hrtimer_init(pTimer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
332	pTimer->function = pfnFunction;
333	#else
334	init_timer(pTimer);
335	pTimer->data = ulData;
336	pTimer->function = pfnFunction;
337	pTimer->expires = jiffies;
338	#endif
339	}
340
341	static inline void vbox_ktimer_start(PVBOXKTIMER pTimer)
342	{
343	#ifdef VBOX_HRTIMER
344	hrtimer_start(pTimer, ktime_add_ns(ktime_get(), 1000000), HRTIMER_MODE_ABS);
345	#else
346	mod_timer(pTimer, jiffies);
347	#endif
348	}
349
350	static inline void vbox_ktimer_stop(PVBOXKTIMER pTimer)
351	{
352	#ifdef VBOX_HRTIMER
353	hrtimer_cancel(pTimer);
354	#else
355	if (timer_pending(pTimer))
356	del_timer_sync(pTimer);
357	#endif
358	}
359
360	#ifdef CONFIG_X86_LOCAL_APIC
361	# ifdef DO_DISABLE_NMI
362
363	/** Stop AMD NMI watchdog (x86_64 only). */
364	static int stop_k7_watchdog(void)
365	{
366	wrmsr(MSR_K7_EVNTSEL0, 0, 0);
367	return 1;
368	}
369
370	/** Stop Intel P4 NMI watchdog (x86_64 only). */
371	static int stop_p4_watchdog(void)
372	{
373	wrmsr(MSR_P4_IQ_CCCR0, 0, 0);
374	wrmsr(MSR_P4_IQ_CCCR1, 0, 0);
375	wrmsr(MSR_P4_CRU_ESCR0, 0, 0);
376	return 1;
377	}
378
379	/** The new method of detecting the event counter */
380	static int stop_intel_arch_watchdog(void)
381	{
382	unsigned ebx;
383
384	ebx = cpuid_ebx(10);
385	if (!(ebx & ARCH_PERFMON_UNHALTED_CORE_CYCLES_PRESENT))
386	wrmsr(MSR_ARCH_PERFMON_EVENTSEL0, 0, 0);
387	return 1;
388	}
389
390	/** Stop NMI watchdog. */
391	static void vbox_stop_apic_nmi_watchdog(void *unused)
392	{
393	int stopped = 0;
394
395	/* only support LOCAL and IO APICs for now */
396	if ((nmi_watchdog != NMI_LOCAL_APIC) &&
397	(nmi_watchdog != NMI_IO_APIC))
398	return;
399
400	if (nmi_watchdog == NMI_LOCAL_APIC)
401	{
402	switch (boot_cpu_data.x86_vendor)
403	{
404	case X86_VENDOR_AMD:
405	if (strstr(boot_cpu_data.x86_model_id, "Screwdriver"))
406	return;
407	stopped = stop_k7_watchdog();
408	break;
409	case X86_VENDOR_INTEL:
410	if (cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON))
411	{
412	stopped = stop_intel_arch_watchdog();
413	break;
414	}
415	stopped = stop_p4_watchdog();
416	break;
417	default:
418	return;
419	}
420	}
421
422	if (stopped)
423	nmi_atomic_dec(&nmi_active);
424	}
425
426	/** Disable LAPIC NMI watchdog. */
427	static void disable_lapic_nmi_watchdog(void)
428	{
429	BUG_ON(nmi_watchdog != NMI_LOCAL_APIC);
430
431	if (nmi_atomic_read(&nmi_active) <= 0)
432	return;
433
434	on_each_cpu(vbox_stop_apic_nmi_watchdog, NULL, 1, 1);
435
436	BUG_ON(nmi_atomic_read(&nmi_active) != 0);
437
438	/* tell do_nmi() and others that we're not active any more */
439	nmi_watchdog = NMI_NONE;
440	}
441
442	/** Shutdown NMI. */
443	static void nmi_cpu_shutdown(void * dummy)
444	{
445	unsigned int vERR, vPC;
446
447	vPC = apic_read(APIC_LVTPC);
448
449	if ((GET_APIC_DELIVERY_MODE(vPC) == APIC_MODE_NMI) && !(vPC & APIC_LVT_MASKED))
450	{
451	vERR = apic_read(APIC_LVTERR);
452	apic_write(APIC_LVTERR, vERR \| APIC_LVT_MASKED);
453	apic_write(APIC_LVTPC, vPC \| APIC_LVT_MASKED);
454	apic_write(APIC_LVTERR, vERR);
455	}
456	}
457
458	static void nmi_shutdown(void)
459	{
460	on_each_cpu(nmi_cpu_shutdown, NULL, 0, 1);
461	}
462	# endif /* DO_DISABLE_NMI */
463	#endif /* CONFIG_X86_LOCAL_APIC */
464
465	/**
466	* Initialize module.
467	*
468	* @returns appropriate status code.
469	*/
470	static int __init VBoxDrvLinuxInit(void)
471	{
472	int rc;
473
474	dprintf(("VBoxDrv::ModuleInit\n"));
475
476	#ifdef CONFIG_X86_LOCAL_APIC
477	/*
478	* If an NMI occurs while we are inside the world switcher the macine will crash.
479	* The Linux NMI watchdog generates periodic NMIs increasing a counter which is
480	* compared with another counter increased in the timer interrupt handler. Therefore
481	* we don't allow to setup an NMI watchdog.
482	*/
483	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) && !defined(VBOX_REDHAT_KABI)
484	/*
485	* First test: NMI actiated? Works only works with Linux 2.6 -- 2.4 does not export
486	* the nmi_watchdog variable.
487	*/
488	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19) \|\| \
489	(defined CONFIG_X86_64 && LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0))
490	# ifdef DO_DISABLE_NMI
491	if (nmi_atomic_read(&nmi_active) > 0)
492	{
493	printk(KERN_DEBUG DEVICE_NAME ": Trying to deactivate the NMI watchdog...\n");
494
495	switch (nmi_watchdog)
496	{
497	case NMI_LOCAL_APIC:
498	disable_lapic_nmi_watchdog();
499	break;
500	case NMI_NONE:
501	nmi_atomic_dec(&nmi_active);
502	break;
503	}
504
505	if (nmi_atomic_read(&nmi_active) == 0)
506	{
507	nmi_shutdown();
508	printk(KERN_DEBUG DEVICE_NAME ": Successfully done.\n");
509	}
510	else
511	printk(KERN_DEBUG DEVICE_NAME ": Failed!\n");
512	}
513	# endif /* DO_DISABLE_NMI */
514
515	/*
516	* Permanent IO_APIC mode active? No way to handle this!
517	*/
518	if (nmi_watchdog == NMI_IO_APIC)
519	{
520	printk(KERN_ERR DEVICE_NAME
521	": NMI watchdog in IO_APIC mode active -- refused to load the kernel module!\n"
522	DEVICE_NAME
523	": Please disable the NMI watchdog by specifying 'nmi_watchdog=0' at kernel\n"
524	DEVICE_NAME
525	": command line.\n");
526	return -EINVAL;
527	}
528
529	/*
530	* See arch/i386/kernel/nmi.c on >= 2.6.19: -1 means it can never enabled again
531	*/
532	nmi_atomic_set(&nmi_active, -1);
533	printk(KERN_DEBUG DEVICE_NAME ": Trying to deactivate the NMI watchdog permanently...\n");
534
535	/*
536	* Now fall through and see if it actually was enabled before. If so, fail
537	* as we cannot deactivate it cleanly from here.
538	*/
539	# else /* < 2.6.19 */
540	/*
541	* Older 2.6 kernels: nmi_watchdog is not initalized by default
542	*/
543	if (nmi_watchdog != NMI_NONE)
544	goto nmi_activated;
545	# endif
546	# endif /* >= 2.6.0 && !defined(VBOX_REDHAT_KABI) */
547
548	/*
549	* Second test: Interrupt generated by performance counter not masked and can
550	* generate an NMI. Works also with Linux 2.4.
551	*/
552	{
553	unsigned int v, ver, maxlvt;
554
555	v = apic_read(APIC_LVR);
556	ver = GET_APIC_VERSION(v);
557	/* 82489DXs do not report # of LVT entries. */
558	maxlvt = APIC_INTEGRATED(ver) ? GET_APIC_MAXLVT(v) : 2;
559	if (maxlvt >= 4)
560	{
561	/* Read status of performance counter IRQ vector */
562	v = apic_read(APIC_LVTPC);
563
564	/* performance counter generates NMI and is not masked? */
565	if ((GET_APIC_DELIVERY_MODE(v) == APIC_MODE_NMI) && !(v & APIC_LVT_MASKED))
566	{
567	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19) \|\| \
568	(defined CONFIG_X86_64 && LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0))
569	printk(KERN_ERR DEVICE_NAME
570	": NMI watchdog either active or at least initialized. Please disable the NMI\n"
571	DEVICE_NAME
572	": watchdog by specifying 'nmi_watchdog=0' at kernel command line.\n");
573	return -EINVAL;
574	# else /* < 2.6.19 */
575	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) && !defined(VBOX_REDHAT_KABI)
576	nmi_activated:
577	# endif
578	printk(KERN_ERR DEVICE_NAME
579	": NMI watchdog active -- refused to load the kernel module! Please disable\n"
580	DEVICE_NAME
581	": the NMI watchdog by specifying 'nmi_watchdog=0' at kernel command line.\n");
582	return -EINVAL;
583	# endif /* >= 2.6.19 */
584	}
585	}
586	}
587	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
588	printk(KERN_DEBUG DEVICE_NAME ": Successfully done.\n");
589	# endif /* >= 2.6.19 */
590	#endif /* CONFIG_X86_LOCAL_APIC */
591
592	#ifdef CONFIG_VBOXDRV_AS_MISC
593	rc = misc_register(&gMiscDevice);
594	if (rc)
595	{
596	printk(KERN_ERR DEVICE_NAME ": Can't register misc device! rc=%d\n", rc);
597	return rc;
598	}
599	#else /* !CONFIG_VBOXDRV_AS_MISC */
600	/*
601	* Register character device.
602	*/
603	g_iModuleMajor = DEVICE_MAJOR;
604	rc = VBOX_REGISTER_DEVICE((dev_t)g_iModuleMajor, DEVICE_NAME, &gFileOpsVBoxDrv);
605	if (rc < 0)
606	{
607	dprintf(("VBOX_REGISTER_DEVICE failed with rc=%#x!\n", rc));
608	return rc;
609	}
610
611	/*
612	* Save returned module major number
613	*/
614	if (DEVICE_MAJOR != 0)
615	g_iModuleMajor = DEVICE_MAJOR;
616	else
617	g_iModuleMajor = rc;
618	rc = 0;
619
620	#ifdef CONFIG_DEVFS_FS
621	/*
622	* Register a device entry
623	*/
624	g_hDevFsVBoxDrv = VBOX_REGISTER_DEVFS();
625	if (g_hDevFsVBoxDrv == NULL)
626	{
627	dprintf(("devfs_register failed!\n"));
628	rc = -EINVAL;
629	}
630	#endif
631	#endif /* !CONFIG_VBOXDRV_AS_MISC */
632	if (!rc)
633	{
634	/*
635	* Initialize the runtime.
636	* On AMD64 we'll have to donate the high rwx memory block to the exec allocator.
637	*/
638	rc = RTR0Init(0);
639	if (RT_SUCCESS(rc))
640	{
641	#ifdef RT_ARCH_AMD64
642	rc = RTR0MemExecDonate(&g_abExecMemory[0], sizeof(g_abExecMemory));
643	#endif
644	/*
645	* Initialize the device extension.
646	*/
647	if (RT_SUCCESS(rc))
648	rc = supdrvInitDevExt(&g_DevExt);
649	if (!rc)
650	{
651	/*
652	* Create the GIP page.
653	*/
654	rc = VBoxDrvLinuxInitGip(&g_DevExt);
655	if (!rc)
656	{
657	printk(KERN_INFO DEVICE_NAME ": TSC mode is %s, kernel timer mode is "
658	#ifdef VBOX_HRTIMER
659	"'high-res'"
660	#else
661	"'normal'"
662	#endif
663	".\n",
664	g_DevExt.pGip->u32Mode == SUPGIPMODE_SYNC_TSC ? "'synchronous'" : "'asynchronous'");
665	LogFlow(("VBoxDrv::ModuleInit returning %#x\n", rc));
666	printk(KERN_DEBUG DEVICE_NAME ": Successfully loaded version "
667	VBOX_VERSION_STRING " (interface " xstr(SUPDRVIOC_VERSION) ").\n");
668	return rc;
669	}
670
671	supdrvDeleteDevExt(&g_DevExt);
672	}
673	else
674	rc = -EINVAL;
675	RTR0Term();
676	}
677	else
678	rc = -EINVAL;
679
680	/*
681	* Failed, cleanup and return the error code.
682	*/
683	#if defined(CONFIG_DEVFS_FS) && !defined(CONFIG_VBOXDRV_AS_MISC)
684	VBOX_UNREGISTER_DEVFS(g_hDevFsVBoxDrv);
685	#endif
686	}
687	#ifdef CONFIG_VBOXDRV_AS_MISC
688	misc_deregister(&gMiscDevice);
689	dprintf(("VBoxDrv::ModuleInit returning %#x (minor:%d)\n", rc, gMiscDevice.minor));
690	#else
691	VBOX_UNREGISTER_DEVICE(g_iModuleMajor, DEVICE_NAME);
692	dprintf(("VBoxDrv::ModuleInit returning %#x (major:%d)\n", rc, g_iModuleMajor));
693	#endif
694	return rc;
695	}
696
697
698	/**
699	* Unload the module.
700	*/
701	static void __exit VBoxDrvLinuxUnload(void)
702	{
703	int rc;
704	dprintf(("VBoxDrvLinuxUnload\n"));
705	NOREF(rc);
706
707	/*
708	* I Don't think it's possible to unload a driver which processes have
709	* opened, at least we'll blindly assume that here.
710	*/
711	#ifdef CONFIG_VBOXDRV_AS_MISC
712	rc = misc_deregister(&gMiscDevice);
713	if (rc < 0)
714	{
715	dprintf(("misc_deregister failed with rc=%#x\n", rc));
716	}
717	#else /* !CONFIG_VBOXDRV_AS_MISC */
718	# ifdef CONFIG_DEVFS_FS
719	/*
720	* Unregister a device entry
721	*/
722	VBOX_UNREGISTER_DEVFS(g_hDevFsVBoxDrv);
723	# endif /* devfs */
724	VBOX_UNREGISTER_DEVICE(g_iModuleMajor, DEVICE_NAME);
725	#endif /* !CONFIG_VBOXDRV_AS_MISC */
726
727	/*
728	* Destroy GIP, delete the device extension and terminate IPRT.
729	*/
730	VBoxDrvLinuxTermGip(&g_DevExt);
731	supdrvDeleteDevExt(&g_DevExt);
732	RTR0Term();
733	}
734
735
736	/**
737	* Device open. Called on open /dev/vboxdrv
738	*
739	* @param pInode Pointer to inode info structure.
740	* @param pFilp Associated file pointer.
741	*/
742	static int VBoxDrvLinuxCreate(struct inode pInode, struct file pFilp)
743	{
744	int rc;
745	PSUPDRVSESSION pSession;
746	Log(("VBoxDrvLinuxCreate: pFilp=%p pid=%d/%d %s\n", pFilp, RTProcSelf(), current->pid, current->comm));
747
748	/*
749	* Call common code for the rest.
750	*/
751	rc = supdrvCreateSession(&g_DevExt, (PSUPDRVSESSION *)&pSession);
752	if (!rc)
753	{
754	pSession->Uid = current->euid;
755	pSession->Gid = current->egid;
756	pSession->Process = RTProcSelf();
757	pSession->R0Process = RTR0ProcHandleSelf();
758	}
759
760	pFilp->private_data = pSession;
761
762	Log(("VBoxDrvLinuxCreate: g_DevExt=%p pSession=%p rc=%d/%d (pid=%d/%d %s)\n",
763	&g_DevExt, pSession, rc, VBoxDrvLinuxErr2LinuxErr(rc),
764	RTProcSelf(), current->pid, current->comm));
765	return VBoxDrvLinuxErr2LinuxErr(rc);
766	}
767
768
769	/**
770	* Close device.
771	*
772	* @param pInode Pointer to inode info structure.
773	* @param pFilp Associated file pointer.
774	*/
775	static int VBoxDrvLinuxClose(struct inode pInode, struct file pFilp)
776	{
777	Log(("VBoxDrvLinuxClose: pFilp=%p pSession=%p pid=%d/%d %s\n",
778	pFilp, pFilp->private_data, RTProcSelf(), current->pid, current->comm));
779	supdrvCloseSession(&g_DevExt, (PSUPDRVSESSION)pFilp->private_data);
780	pFilp->private_data = NULL;
781	return 0;
782	}
783
784
785	/**
786	* Device I/O Control entry point.
787	*
788	* @param pFilp Associated file pointer.
789	* @param uCmd The function specified to ioctl().
790	* @param ulArg The argument specified to ioctl().
791	*/
792	#ifdef HAVE_UNLOCKED_IOCTL
793	static long VBoxDrvLinuxIOCtl(struct file *pFilp, unsigned int uCmd, unsigned long ulArg)
794	#else
795	static int VBoxDrvLinuxIOCtl(struct inode pInode, struct file pFilp, unsigned int uCmd, unsigned long ulArg)
796	#endif
797	{
798	/*
799	* Deal with the two high-speed IOCtl that takes it's arguments from
800	* the session and iCmd, and only returns a VBox status code.
801	*/
802	#ifdef HAVE_UNLOCKED_IOCTL
803	if (RT_LIKELY( uCmd == SUP_IOCTL_FAST_DO_RAW_RUN
804	\|\| uCmd == SUP_IOCTL_FAST_DO_HWACC_RUN
805	\|\| uCmd == SUP_IOCTL_FAST_DO_NOP))
806	return supdrvIOCtlFast(uCmd, &g_DevExt, (PSUPDRVSESSION)pFilp->private_data);
807	return VBoxDrvLinuxIOCtlSlow(pFilp, uCmd, ulArg);
808
809	#else /* !HAVE_UNLOCKED_IOCTL */
810
811	int rc;
812	unlock_kernel();
813	if (RT_LIKELY( uCmd == SUP_IOCTL_FAST_DO_RAW_RUN
814	\|\| uCmd == SUP_IOCTL_FAST_DO_HWACC_RUN
815	\|\| uCmd == SUP_IOCTL_FAST_DO_NOP))
816	rc = supdrvIOCtlFast(uCmd, &g_DevExt, (PSUPDRVSESSION)pFilp->private_data);
817	else
818	rc = VBoxDrvLinuxIOCtlSlow(pFilp, uCmd, ulArg);
819	lock_kernel();
820	return rc;
821	#endif /* !HAVE_UNLOCKED_IOCTL */
822	}
823
824
825	/**
826	* Device I/O Control entry point.
827	*
828	* @param pFilp Associated file pointer.
829	* @param uCmd The function specified to ioctl().
830	* @param ulArg The argument specified to ioctl().
831	*/
832	static int VBoxDrvLinuxIOCtlSlow(struct file *pFilp, unsigned int uCmd, unsigned long ulArg)
833	{
834	int rc;
835	SUPREQHDR Hdr;
836	PSUPREQHDR pHdr;
837	uint32_t cbBuf;
838
839	Log6(("VBoxDrvLinuxIOCtl: pFilp=%p uCmd=%#x ulArg=%p pid=%d/%d\n", pFilp, uCmd, (void *)ulArg, RTProcSelf(), current->pid));
840
841	/*
842	* Read the header.
843	*/
844	if (RT_UNLIKELY(copy_from_user(&Hdr, (void *)ulArg, sizeof(Hdr))))
845	{
846	Log(("VBoxDrvLinuxIOCtl: copy_from_user(,%#lx,) failed; uCmd=%#x.\n", ulArg, uCmd));
847	return -EFAULT;
848	}
849	if (RT_UNLIKELY((Hdr.fFlags & SUPREQHDR_FLAGS_MAGIC_MASK) != SUPREQHDR_FLAGS_MAGIC))
850	{
851	Log(("VBoxDrvLinuxIOCtl: bad header magic %#x; uCmd=%#x\n", Hdr.fFlags & SUPREQHDR_FLAGS_MAGIC_MASK, uCmd));
852	return -EINVAL;
853	}
854
855	/*
856	* Buffer the request.
857	*/
858	cbBuf = RT_MAX(Hdr.cbIn, Hdr.cbOut);
859	if (RT_UNLIKELY(cbBuf > _1M*16))
860	{
861	Log(("VBoxDrvLinuxIOCtl: too big cbBuf=%#x; uCmd=%#x\n", cbBuf, uCmd));
862	return -E2BIG;
863	}
864	if (RT_UNLIKELY(cbBuf != _IOC_SIZE(uCmd) && _IOC_SIZE(uCmd)))
865	{
866	Log(("VBoxDrvLinuxIOCtl: bad ioctl cbBuf=%#x _IOC_SIZE=%#x; uCmd=%#x.\n", cbBuf, _IOC_SIZE(uCmd), uCmd));
867	return -EINVAL;
868	}
869	pHdr = RTMemAlloc(cbBuf);
870	if (RT_UNLIKELY(!pHdr))
871	{
872	OSDBGPRINT(("VBoxDrvLinuxIOCtl: failed to allocate buffer of %d bytes for uCmd=%#x.\n", cbBuf, uCmd));
873	return -ENOMEM;
874	}
875	if (RT_UNLIKELY(copy_from_user(pHdr, (void *)ulArg, Hdr.cbIn)))
876	{
877	Log(("VBoxDrvLinuxIOCtl: copy_from_user(,%#lx, %#x) failed; uCmd=%#x.\n", ulArg, Hdr.cbIn, uCmd));
878	RTMemFree(pHdr);
879	return -EFAULT;
880	}
881
882	/*
883	* Process the IOCtl.
884	*/
885	rc = supdrvIOCtl(uCmd, &g_DevExt, (PSUPDRVSESSION)pFilp->private_data, pHdr);
886
887	/*
888	* Copy ioctl data and output buffer back to user space.
889	*/
890	if (RT_LIKELY(!rc))
891	{
892	uint32_t cbOut = pHdr->cbOut;
893	if (RT_UNLIKELY(cbOut > cbBuf))
894	{
895	OSDBGPRINT(("VBoxDrvLinuxIOCtl: too much output! %#x > %#x; uCmd=%#x!\n", cbOut, cbBuf, uCmd));
896	cbOut = cbBuf;
897	}
898	if (RT_UNLIKELY(copy_to_user((void *)ulArg, pHdr, cbOut)))
899	{
900	/* this is really bad! */
901	OSDBGPRINT(("VBoxDrvLinuxIOCtl: copy_to_user(%#lx,,%#x); uCmd=%#x!\n", ulArg, cbOut, uCmd));
902	rc = -EFAULT;
903	}
904	}
905	else
906	{
907	Log(("VBoxDrvLinuxIOCtl: pFilp=%p uCmd=%#x ulArg=%p failed, rc=%d\n", pFilp, uCmd, (void *)ulArg, rc));
908	rc = -EINVAL;
909	}
910	RTMemFree(pHdr);
911
912	Log6(("VBoxDrvLinuxIOCtl: returns %d (pid=%d/%d)\n", rc, RTProcSelf(), current->pid));
913	return rc;
914	}
915
916
917	/**
918	* Initializes any OS specific object creator fields.
919	*/
920	void VBOXCALL supdrvOSObjInitCreator(PSUPDRVOBJ pObj, PSUPDRVSESSION pSession)
921	{
922	NOREF(pObj);
923	NOREF(pSession);
924	}
925
926
927	/**
928	* Checks if the session can access the object.
929	*
930	* @returns true if a decision has been made.
931	* @returns false if the default access policy should be applied.
932	*
933	* @param pObj The object in question.
934	* @param pSession The session wanting to access the object.
935	* @param pszObjName The object name, can be NULL.
936	* @param prc Where to store the result when returning true.
937	*/
938	bool VBOXCALL supdrvOSObjCanAccess(PSUPDRVOBJ pObj, PSUPDRVSESSION pSession, const char pszObjName, int prc)
939	{
940	NOREF(pObj);
941	NOREF(pSession);
942	NOREF(pszObjName);
943	NOREF(prc);
944	return false;
945	}
946
947
948	/**
949	* Initializes the GIP.
950	*
951	* @returns negative errno.
952	* @param pDevExt Instance data. GIP stuff may be updated.
953	*/
954	static int VBoxDrvLinuxInitGip(PSUPDRVDEVEXT pDevExt)
955	{
956	struct page *pPage;
957	dma_addr_t HCPhys;
958	PSUPGLOBALINFOPAGE pGip;
959	#ifdef CONFIG_SMP
960	unsigned i;
961	#endif
962	LogFlow(("VBoxDrvLinuxInitGip:\n"));
963
964	/*
965	* Allocate the page.
966	*/
967	pPage = alloc_pages(GFP_USER, 0);
968	if (!pPage)
969	{
970	Log(("VBoxDrvLinuxInitGip: failed to allocate the GIP page\n"));
971	return -ENOMEM;
972	}
973
974	/*
975	* Lock the page.
976	*/
977	SetPageReserved(pPage);
978	g_pGipPage = pPage;
979
980	/*
981	* Call common initialization routine.
982	*/
983	HCPhys = page_to_phys(pPage);
984	pGip = (PSUPGLOBALINFOPAGE)page_address(pPage);
985	pDevExt->ulLastJiffies = jiffies;
986	pDevExt->u64LastMonotime = (uint64_t)pDevExt->ulLastJiffies * TICK_NSEC;
987	Log(("VBoxDrvInitGIP: TICK_NSEC=%ld HZ=%d jiffies=%ld now=%lld\n",
988	TICK_NSEC, HZ, pDevExt->ulLastJiffies, pDevExt->u64LastMonotime));
989	supdrvGipInit(pDevExt, pGip, HCPhys, pDevExt->u64LastMonotime,
990	HZ <= 1000 ? HZ : 1000);
991
992	/*
993	* Initialize the timer.
994	*/
995	vbox_ktimer_init(&g_GipTimer, VBoxDrvLinuxGipTimer, (unsigned long)pDevExt);
996	#ifdef CONFIG_SMP
997	for (i = 0; i < RT_ELEMENTS(pDevExt->aCPUs); i++)
998	{
999	pDevExt->aCPUs[i].u64LastMonotime = pDevExt->u64LastMonotime;
1000	pDevExt->aCPUs[i].ulLastJiffies = pDevExt->ulLastJiffies;
1001	pDevExt->aCPUs[i].iSmpProcessorId = -512;
1002	vbox_ktimer_init(&pDevExt->aCPUs[i].Timer, VBoxDrvLinuxGipTimerPerCpu, i);
1003	}
1004	#endif
1005
1006	return 0;
1007	}
1008
1009
1010	/**
1011	* Terminates the GIP.
1012	*
1013	* @returns negative errno.
1014	* @param pDevExt Instance data. GIP stuff may be updated.
1015	*/
1016	static int VBoxDrvLinuxTermGip(PSUPDRVDEVEXT pDevExt)
1017	{
1018	struct page *pPage;
1019	PSUPGLOBALINFOPAGE pGip;
1020	#ifdef CONFIG_SMP
1021	unsigned i;
1022	#endif
1023	LogFlow(("VBoxDrvLinuxTermGip:\n"));
1024
1025	/*
1026	* Delete the timer if it's pending.
1027	*/
1028	vbox_ktimer_stop(&g_GipTimer);
1029	#ifdef CONFIG_SMP
1030	for (i = 0; i < RT_ELEMENTS(pDevExt->aCPUs); i++)
1031	vbox_ktimer_stop(&pDevExt->aCPUs[i].Timer);
1032	#endif
1033
1034	/*
1035	* Uninitialize the content.
1036	*/
1037	pGip = pDevExt->pGip;
1038	pDevExt->pGip = NULL;
1039	if (pGip)
1040	supdrvGipTerm(pGip);
1041
1042	/*
1043	* Free the page.
1044	*/
1045	pPage = g_pGipPage;
1046	g_pGipPage = NULL;
1047	if (pPage)
1048	{
1049	ClearPageReserved(pPage);
1050	__free_pages(pPage, 0);
1051	}
1052
1053	return 0;
1054	}
1055
1056	/**
1057	* Timer callback function.
1058	*
1059	* In ASYNC TSC mode this is called on the primary CPU, and we're
1060	* assuming that the CPU remains online.
1061	*
1062	* @param ulUser The device extension pointer.
1063	*/
1064	#ifdef VBOX_HRTIMER
1065	static enum hrtimer_restart VBoxDrvLinuxGipTimer(struct hrtimer *pTimer)
1066	#else
1067	static void VBoxDrvLinuxGipTimer(unsigned long ulUser)
1068	#endif
1069	{
1070	PSUPDRVDEVEXT pDevExt;
1071	PSUPGLOBALINFOPAGE pGip;
1072	unsigned long ulNow;
1073	unsigned long ulDiff;
1074	uint64_t u64Monotime;
1075	unsigned long SavedFlags;
1076	#ifdef VBOX_HRTIMER
1077	ktime_t KtNow;
1078	#endif
1079
1080	local_irq_save(SavedFlags);
1081
1082	ulNow = jiffies;
1083	#ifdef VBOX_HRTIMER
1084	KtNow = ktime_get();
1085	pDevExt = &g_DevExt;
1086	#else
1087	pDevExt = (PSUPDRVDEVEXT)ulUser;
1088	#endif
1089	pGip = pDevExt->pGip;
1090
1091	#ifdef CONFIG_SMP
1092	if (pGip && pGip->u32Mode == SUPGIPMODE_ASYNC_TSC)
1093	{
1094	uint8_t iCPU = ASMGetApicId();
1095	ulDiff = ulNow - pDevExt->aCPUs[iCPU].ulLastJiffies;
1096	pDevExt->aCPUs[iCPU].ulLastJiffies = ulNow;
1097	u64Monotime = pDevExt->aCPUs[iCPU].u64LastMonotime + ulDiff * TICK_NSEC;
1098	pDevExt->aCPUs[iCPU].u64LastMonotime = u64Monotime;
1099	}
1100	else
1101	#endif /* CONFIG_SMP */
1102	{
1103	ulDiff = ulNow - pDevExt->ulLastJiffies;
1104	pDevExt->ulLastJiffies = ulNow;
1105	u64Monotime = pDevExt->u64LastMonotime + ulDiff * TICK_NSEC;
1106	pDevExt->u64LastMonotime = u64Monotime;
1107	}
1108	if (RT_LIKELY(pGip))
1109	supdrvGipUpdate(pDevExt->pGip, u64Monotime);
1110	if (RT_LIKELY(!pDevExt->fGIPSuspended))
1111	{
1112	#ifdef VBOX_HRTIMER
1113	hrtimer_forward(&g_GipTimer, KtNow, ktime_set(0, 1000000));
1114	#else
1115	mod_timer(&g_GipTimer, ulNow + ONE_MSEC_IN_JIFFIES);
1116	#endif
1117	}
1118
1119	local_irq_restore(SavedFlags);
1120
1121	#ifdef VBOX_HRTIMER
1122	return pDevExt->fGIPSuspended ? HRTIMER_NORESTART : HRTIMER_RESTART;
1123	#endif
1124	}
1125
1126
1127	#ifdef CONFIG_SMP
1128	/**
1129	* Timer callback function for the other CPUs.
1130	*
1131	* @param iTimerCPU The APIC ID of this timer.
1132	*/
1133	#ifdef VBOX_HRTIMER
1134	static enum hrtimer_restart VBoxDrvLinuxGipTimerPerCpu(struct hrtimer *pTimer)
1135	#else
1136	static void VBoxDrvLinuxGipTimerPerCpu(unsigned long iTimerCPU)
1137	#endif
1138	{
1139	PSUPDRVDEVEXT pDevExt;
1140	PSUPGLOBALINFOPAGE pGip;
1141	uint8_t iCPU;
1142	uint64_t u64Monotime;
1143	unsigned long SavedFlags;
1144	unsigned long ulNow;
1145	# ifdef VBOX_HRTIMER
1146	unsigned long iTimerCPU;
1147	ktime_t KtNow;
1148	# endif
1149
1150	local_irq_save(SavedFlags);
1151
1152	ulNow = jiffies;
1153	pDevExt = &g_DevExt;
1154	pGip = pDevExt->pGip;
1155	iCPU = ASMGetApicId();
1156	# ifdef VBOX_HRTIMER
1157	iTimerCPU = iCPU; /* XXX hrtimer does not support a 'data' field */
1158	KtNow = ktime_get();
1159	# endif
1160
1161	if (RT_LIKELY(iCPU < RT_ELEMENTS(pGip->aCPUs)))
1162	{
1163	if (RT_LIKELY(iTimerCPU == iCPU))
1164	{
1165	unsigned long ulDiff = ulNow - pDevExt->aCPUs[iCPU].ulLastJiffies;
1166	pDevExt->aCPUs[iCPU].ulLastJiffies = ulNow;
1167	u64Monotime = pDevExt->aCPUs[iCPU].u64LastMonotime + ulDiff * TICK_NSEC;
1168	pDevExt->aCPUs[iCPU].u64LastMonotime = u64Monotime;
1169	if (RT_LIKELY(pGip))
1170	supdrvGipUpdatePerCpu(pGip, u64Monotime, iCPU);
1171	if (RT_LIKELY(!pDevExt->fGIPSuspended))
1172	{
1173	# ifdef VBOX_HRTIMER
1174	hrtimer_forward(&pDevExt->aCPUs[iCPU].Timer, KtNow, ktime_set(0, 1000000));
1175	# else
1176	mod_timer(&pDevExt->aCPUs[iCPU].Timer, ulNow + ONE_MSEC_IN_JIFFIES);
1177	# endif
1178	}
1179	}
1180	else
1181	printk("vboxdrv: error: GIP CPU update timer executing on the wrong CPU: apicid=%d != timer-apicid=%ld (cpuid=%d !=? timer-cpuid=%d)\n",
1182	iCPU, iTimerCPU, smp_processor_id(), pDevExt->aCPUs[iTimerCPU].iSmpProcessorId);
1183	}
1184	else
1185	printk("vboxdrv: error: APIC ID is bogus (GIP CPU update): apicid=%d max=%lu cpuid=%d\n",
1186	iCPU, (unsigned long)RT_ELEMENTS(pGip->aCPUs), smp_processor_id());
1187
1188	local_irq_restore(SavedFlags);
1189
1190	# ifdef VBOX_HRTIMER
1191	return pDevExt->fGIPSuspended ? HRTIMER_NORESTART : HRTIMER_RESTART;
1192	# endif
1193	}
1194	#endif /* CONFIG_SMP */
1195
1196
1197	/**
1198	* Maps the GIP into user space.
1199	*
1200	* @returns negative errno.
1201	* @param pDevExt Instance data.
1202	*/
1203	int VBOXCALL supdrvOSGipMap(PSUPDRVDEVEXT pDevExt, PSUPGLOBALINFOPAGE *ppGip)
1204	{
1205	int rc = 0;
1206	unsigned long ulAddr;
1207	unsigned long HCPhys = pDevExt->HCPhysGip;
1208	pgprot_t pgFlags;
1209	pgprot_val(pgFlags) = _PAGE_PRESENT \| _PAGE_USER;
1210	LogFlow(("supdrvOSGipMap: ppGip=%p\n", ppGip));
1211
1212	/*
1213	* Allocate user space mapping and put the physical pages into it.
1214	*/
1215	down_write(&current->mm->mmap_sem);
1216	ulAddr = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ, MAP_SHARED \| MAP_ANONYMOUS, 0);
1217	if (!(ulAddr & ~PAGE_MASK))
1218	{
1219	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0) && !defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1220	int rc2 = remap_page_range(ulAddr, HCPhys, PAGE_SIZE, pgFlags);
1221	#else
1222	int rc2 = 0;
1223	struct vm_area_struct *vma = find_vma(current->mm, ulAddr);
1224	if (vma)
1225	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11)
1226	rc2 = remap_page_range(vma, ulAddr, HCPhys, PAGE_SIZE, pgFlags);
1227	#else
1228	rc2 = remap_pfn_range(vma, ulAddr, HCPhys >> PAGE_SHIFT, PAGE_SIZE, pgFlags);
1229	#endif
1230	else
1231	{
1232	rc = SUPDRV_ERR_NO_MEMORY;
1233	Log(("supdrvOSGipMap: no vma found for ulAddr=%#lx!\n", ulAddr));
1234	}
1235	#endif
1236	if (rc2)
1237	{
1238	rc = SUPDRV_ERR_NO_MEMORY;
1239	Log(("supdrvOSGipMap: remap_page_range failed rc2=%d\n", rc2));
1240	}
1241	}
1242	else
1243	{
1244	Log(("supdrvOSGipMap: do_mmap failed ulAddr=%#lx\n", ulAddr));
1245	rc = SUPDRV_ERR_NO_MEMORY;
1246	}
1247	up_write(&current->mm->mmap_sem); /* not quite sure when to give this up. */
1248
1249	/*
1250	* Success?
1251	*/
1252	if (!rc)
1253	{
1254	*ppGip = (PSUPGLOBALINFOPAGE)ulAddr;
1255	LogFlow(("supdrvOSGipMap: ppGip=%p\n", *ppGip));
1256	return 0;
1257	}
1258
1259	/*
1260	* Failure, cleanup and be gone.
1261	*/
1262	if (ulAddr & ~PAGE_MASK)
1263	{
1264	down_write(&current->mm->mmap_sem);
1265	MY_DO_MUNMAP(current->mm, ulAddr, PAGE_SIZE);
1266	up_write(&current->mm->mmap_sem);
1267	}
1268
1269	LogFlow(("supdrvOSGipMap: returns %d\n", rc));
1270	return rc;
1271	}
1272
1273
1274	/**
1275	* Maps the GIP into user space.
1276	*
1277	* @returns negative errno.
1278	* @param pDevExt Instance data.
1279	*/
1280	int VBOXCALL supdrvOSGipUnmap(PSUPDRVDEVEXT pDevExt, PSUPGLOBALINFOPAGE pGip)
1281	{
1282	LogFlow(("supdrvOSGipUnmap: pGip=%p\n", pGip));
1283	if (current->mm)
1284	{
1285	down_write(&current->mm->mmap_sem);
1286	MY_DO_MUNMAP(current->mm, (unsigned long)pGip, PAGE_SIZE);
1287	up_write(&current->mm->mmap_sem);
1288	}
1289	LogFlow(("supdrvOSGipUnmap: returns 0\n"));
1290	return 0;
1291	}
1292
1293
1294	/**
1295	* Resumes the GIP updating.
1296	*
1297	* @param pDevExt Instance data.
1298	*/
1299	void VBOXCALL supdrvOSGipResume(PSUPDRVDEVEXT pDevExt)
1300	{
1301	LogFlow(("supdrvOSGipResume:\n"));
1302	ASMAtomicXchgU8(&pDevExt->fGIPSuspended, false);
1303	#ifdef CONFIG_SMP
1304	if (pDevExt->pGip->u32Mode != SUPGIPMODE_ASYNC_TSC)
1305	{
1306	#endif
1307	vbox_ktimer_start(&g_GipTimer);
1308	#ifdef CONFIG_SMP
1309	}
1310	else
1311	{
1312	vbox_ktimer_start(&g_GipTimer);
1313	smp_call_function(VBoxDrvLinuxGipResumePerCpu, pDevExt, 0 /* retry /, 1 / wait */);
1314	}
1315	#endif
1316	}
1317
1318
1319	#ifdef CONFIG_SMP
1320	/**
1321	* Callback for resuming GIP updating on the other CPUs.
1322	*
1323	* This is only used when the GIP is in async tsc mode.
1324	*
1325	* @param pvUser Pointer to the device instance.
1326	*/
1327	static void VBoxDrvLinuxGipResumePerCpu(void *pvUser)
1328	{
1329	PSUPDRVDEVEXT pDevExt = (PSUPDRVDEVEXT)pvUser;
1330	uint8_t iCPU = ASMGetApicId();
1331
1332	if (RT_UNLIKELY(iCPU >= RT_ELEMENTS(pDevExt->pGip->aCPUs)))
1333	{
1334	printk("vboxdrv: error: apicid=%d max=%lu cpuid=%d\n",
1335	iCPU, (unsigned long)RT_ELEMENTS(pDevExt->pGip->aCPUs), smp_processor_id());
1336	return;
1337	}
1338
1339	pDevExt->aCPUs[iCPU].iSmpProcessorId = smp_processor_id();
1340	vbox_ktimer_start(&pDevExt->aCPUs[iCPU].Timer);
1341	}
1342	#endif /* CONFIG_SMP */
1343
1344
1345	/**
1346	* Suspends the GIP updating.
1347	*
1348	* @param pDevExt Instance data.
1349	*/
1350	void VBOXCALL supdrvOSGipSuspend(PSUPDRVDEVEXT pDevExt)
1351	{
1352	#ifdef CONFIG_SMP
1353	unsigned i;
1354	#endif
1355	LogFlow(("supdrvOSGipSuspend:\n"));
1356	ASMAtomicXchgU8(&pDevExt->fGIPSuspended, true);
1357
1358	vbox_ktimer_stop(&g_GipTimer);
1359	#ifdef CONFIG_SMP
1360	for (i = 0; i < RT_ELEMENTS(pDevExt->aCPUs); i++)
1361	vbox_ktimer_stop(&pDevExt->aCPUs[i].Timer);
1362	#endif
1363	}
1364
1365
1366	/**
1367	* Get the current CPU count.
1368	* @returns Number of cpus.
1369	*/
1370	unsigned VBOXCALL supdrvOSGetCPUCount(void)
1371	{
1372	#ifdef CONFIG_SMP
1373	# if defined(num_present_cpus) && !defined(VBOX_REDHAT_KABI)
1374	return num_present_cpus();
1375	# elif defined(num_possible_cpus)
1376	return num_possible_cpus();
1377	# else
1378	return smp_num_cpus;
1379	# endif
1380	#else
1381	return 1;
1382	#endif
1383	}
1384
1385	/**
1386	* Force async tsc mode.
1387	* @todo add a module argument for this.
1388	*/
1389	bool VBOXCALL supdrvOSGetForcedAsyncTscMode(void)
1390	{
1391	return force_async_tsc != 0;
1392	}
1393
1394
1395	/**
1396	* Converts a supdrv error code to an linux error code.
1397	*
1398	* @returns corresponding linux error code.
1399	* @param rc supdrv error code (SUPDRV_ERR_* defines).
1400	*/
1401	static int VBoxDrvLinuxErr2LinuxErr(int rc)
1402	{
1403	switch (rc)
1404	{
1405	case 0: return 0;
1406	case SUPDRV_ERR_GENERAL_FAILURE: return -EACCES;
1407	case SUPDRV_ERR_INVALID_PARAM: return -EINVAL;
1408	case SUPDRV_ERR_INVALID_MAGIC: return -EILSEQ;
1409	case SUPDRV_ERR_INVALID_HANDLE: return -ENXIO;
1410	case SUPDRV_ERR_INVALID_POINTER: return -EFAULT;
1411	case SUPDRV_ERR_LOCK_FAILED: return -ENOLCK;
1412	case SUPDRV_ERR_ALREADY_LOADED: return -EEXIST;
1413	case SUPDRV_ERR_PERMISSION_DENIED: return -EPERM;
1414	case SUPDRV_ERR_VERSION_MISMATCH: return -ENOSYS;
1415	case SUPDRV_ERR_IDT_FAILED: return -1000;
1416	}
1417
1418	return -EPERM;
1419	}
1420
1421
1422	RTDECL(int) SUPR0Printf(const char *pszFormat, ...)
1423	{
1424	#if 1
1425	va_list args;
1426	char szMsg[512];
1427
1428	va_start(args, pszFormat);
1429	vsnprintf(szMsg, sizeof(szMsg) - 1, pszFormat, args);
1430	szMsg[sizeof(szMsg) - 1] = '\0';
1431	printk("%s", szMsg);
1432	va_end(args);
1433	#else
1434	/* forward to printf - needs some more GCC hacking to fix ebp... */
1435	__asm__ __volatile__ ("mov %0, %esp\n\t"
1436	"jmp %1\n\t",
1437	:: "r" ((uintptr_t)&pszFormat - 4),
1438	"m" (printk));
1439	#endif
1440	return 0;
1441	}
1442
1443
1444	/** Runtime assert implementation for Linux Ring-0. */
1445	RTDECL(bool) RTAssertDoBreakpoint(void)
1446	{
1447	return true;
1448	}
1449
1450
1451	/** Runtime assert implementation for Linux Ring-0. */
1452	RTDECL(void) AssertMsg1(const char pszExpr, unsigned uLine, const char pszFile, const char *pszFunction)
1453	{
1454	printk("!!Assertion Failed!!\n"
1455	"Expression: %s\n"
1456	"Location : %s(%d) %s\n",
1457	pszExpr, pszFile, uLine, pszFunction);
1458	}
1459
1460
1461	/** Runtime assert implementation for Linux Ring-0. */
1462	RTDECL(void) AssertMsg2(const char *pszFormat, ...)
1463	{ /* forwarder. */
1464	va_list ap;
1465	char msg[256];
1466
1467	va_start(ap, pszFormat);
1468	vsnprintf(msg, sizeof(msg) - 1, pszFormat, ap);
1469	msg[sizeof(msg) - 1] = '\0';
1470	printk("%s", msg);
1471	va_end(ap);
1472	}
1473
1474
1475	/* GCC C++ hack. */
1476	unsigned __gxx_personality_v0 = 0xcccccccc;
1477
1478
1479	module_init(VBoxDrvLinuxInit);
1480	module_exit(VBoxDrvLinuxUnload);
1481
1482	MODULE_AUTHOR("innotek GmbH");
1483	MODULE_DESCRIPTION("VirtualBox Support Driver");
1484	MODULE_LICENSE("GPL");
1485	#ifdef MODULE_VERSION
1486	MODULE_VERSION(VBOX_VERSION_STRING " (" xstr(SUPDRVIOC_VERSION) ")");
1487	#endif
1488
1489	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
1490	module_param(force_async_tsc, int, 0444);
1491	#else
1492	MODULE_PARM(force_async_tsc, "i");
1493	#endif
1494	MODULE_PARM_DESC(force_async_tsc, "force the asynchronous TSC mode");
1495

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

source: vbox/trunk/src/VBox/HostDrivers/Support/linux/SUPDrv-linux.c@ 7521

Download in other formats: