SUPDrv-linux.c@ 3968

Last change on this file since 3968 was 3672, checked in by vboxsync, 18 years ago
RT_OS_* and RT_ARCH_* for Runtime/ and Support/
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 56.7 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 as published by the Free Software Foundation,
12	* in version 2 as it comes in the "COPYING" file of the VirtualBox OSE
13	* distribution. VirtualBox OSE is distributed in the hope that it will
14	* be useful, but WITHOUT ANY WARRANTY of any kind.
15	*
16	* If you received this file as part of a commercial VirtualBox
17	* distribution, then only the terms of your commercial VirtualBox
18	* license agreement apply instead of the previous paragraph.
19	*
20	* Some lines of code to disable the local APIC on x86_64 machines taken
21	* from a Mandriva patch by Gwenole Beauchesne <[email protected]>.
22	*/
23
24	/*******************************************************************************
25	* Header Files *
26	*******************************************************************************/
27	#include "SUPDRV.h"
28	#include "version-generated.h"
29
30	#include <iprt/assert.h>
31	#include <iprt/spinlock.h>
32	#include <iprt/semaphore.h>
33	#include <iprt/initterm.h>
34	#include <iprt/process.h>
35	#include <iprt/err.h>
36	#include <iprt/mem.h>
37
38	#include <linux/module.h>
39	#include <linux/kernel.h>
40	#include <linux/init.h>
41	#include <linux/fs.h>
42	#include <linux/mm.h>
43	#include <linux/pagemap.h>
44	#include <linux/sched.h>
45	#include <linux/slab.h>
46	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
47	# include <linux/jiffies.h>
48	#endif
49	#include <asm/mman.h>
50	#include <asm/io.h>
51	#include <asm/uaccess.h>
52	#ifdef CONFIG_DEVFS_FS
53	# include <linux/devfs_fs_kernel.h>
54	#endif
55	#ifdef CONFIG_VBOXDRV_AS_MISC
56	# include <linux/miscdevice.h>
57	#endif
58	#ifdef CONFIG_X86_LOCAL_APIC
59	# include <asm/apic.h>
60	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
61	# include <asm/nmi.h>
62	# endif
63	#endif
64
65	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0)
66	# ifndef page_to_pfn
67	# define page_to_pfn(page) ((page) - mem_map)
68	# endif
69	# include <asm/pgtable.h>
70	# define global_flush_tlb __flush_tlb_global
71	#endif
72
73	/* devfs defines */
74	#if defined(CONFIG_DEVFS_FS) && !defined(CONFIG_VBOXDRV_AS_MISC)
75	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
76
77	# define VBOX_REGISTER_DEVFS() \
78	({ \
79	void *rc = NULL; \
80	if (devfs_mk_cdev(MKDEV(DEVICE_MAJOR, 0), \
81	S_IFCHR \| S_IRUGO \| S_IWUGO, \
82	DEVICE_NAME) == 0) \
83	rc = (void )' '; / return not NULL */ \
84	rc; \
85	})
86
87	# define VBOX_UNREGISTER_DEVFS(handle) \
88	devfs_remove(DEVICE_NAME);
89
90	# else /* < 2.6.0 */
91
92	# define VBOX_REGISTER_DEVFS() \
93	devfs_register(NULL, DEVICE_NAME, DEVFS_FL_DEFAULT, \
94	DEVICE_MAJOR, 0, \
95	S_IFCHR \| S_IRUGO \| S_IWUGO, \
96	&gFileOpsVBoxDrv, NULL)
97
98	# define VBOX_UNREGISTER_DEVFS(handle) \
99	if (handle != NULL) \
100	devfs_unregister(handle)
101
102	# endif /* < 2.6.0 */
103	#endif /* CONFIG_DEV_FS && !CONFIG_VBOXDEV_AS_MISC */
104
105	#ifndef CONFIG_VBOXDRV_AS_MISC
106	# if defined(CONFIG_DEVFS_FS) && LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 0)
107	# define VBOX_REGISTER_DEVICE(a,b,c) devfs_register_chrdev(a,b,c)
108	# define VBOX_UNREGISTER_DEVICE(a,b) devfs_unregister_chrdev(a,b)
109	# else
110	# define VBOX_REGISTER_DEVICE(a,b,c) register_chrdev(a,b,c)
111	# define VBOX_UNREGISTER_DEVICE(a,b) unregister_chrdev(a,b)
112	# endif
113	#endif /* !CONFIG_VBOXDRV_AS_MISC */
114
115
116	#ifdef CONFIG_X86_HIGH_ENTRY
117	# error "CONFIG_X86_HIGH_ENTRY is not supported by VBoxDrv at this time."
118	#endif
119
120	/*
121	* This sucks soooo badly on x86! Why don't they export __PAGE_KERNEL_EXEC so PAGE_KERNEL_EXEC would be usable?
122	*/
123	#if defined(RT_ARCH_AMD64)
124	# define MY_PAGE_KERNEL_EXEC PAGE_KERNEL_EXEC
125	#elif defined(PAGE_KERNEL_EXEC) && defined(CONFIG_X86_PAE)
126	# define MY_PAGE_KERNEL_EXEC __pgprot(cpu_has_pge ? _PAGE_KERNEL_EXEC \| _PAGE_GLOBAL : _PAGE_KERNEL_EXEC)
127	#else
128	# define MY_PAGE_KERNEL_EXEC PAGE_KERNEL
129	#endif
130
131	/*
132	* The redhat hack section.
133	* - The current hacks are for 2.4.21-15.EL only.
134	*/
135	#ifndef NO_REDHAT_HACKS
136	/* accounting. */
137	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0)
138	# ifdef VM_ACCOUNT
139	# define MY_DO_MUNMAP(a,b,c) do_munmap(a, b, c, 0) /* should it be 1 or 0? */
140	# endif
141	# endif
142
143	/* backported remap_page_range. */
144	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0)
145	# include <asm/tlb.h>
146	# ifdef tlb_vma /* probably not good enough... */
147	# define HAVE_26_STYLE_REMAP_PAGE_RANGE 1
148	# endif
149	# endif
150
151	# ifndef RT_ARCH_AMD64
152	/* In 2.6.9-22.ELsmp we have to call change_page_attr() twice when changing
153	* the page attributes from PAGE_KERNEL to something else, because there appears
154	* to be a bug in one of the many patches that redhat applied.
155	* It should be safe to do this on less buggy linux kernels too. ;-)
156	*/
157	# define MY_CHANGE_PAGE_ATTR(pPages, cPages, prot) \
158	do { \
159	if (pgprot_val(prot) != pgprot_val(PAGE_KERNEL)) \
160	change_page_attr(pPages, cPages, prot); \
161	change_page_attr(pPages, cPages, prot); \
162	} while (0)
163	# endif
164	#endif /* !NO_REDHAT_HACKS */
165
166
167	#ifndef MY_DO_MUNMAP
168	# define MY_DO_MUNMAP(a,b,c) do_munmap(a, b, c)
169	#endif
170
171	#ifndef MY_CHANGE_PAGE_ATTR
172	# ifdef RT_ARCH_AMD64 /** @todo This is a cheap hack, but it'll get around that 'else BUG();' in __change_page_attr(). */
173	# define MY_CHANGE_PAGE_ATTR(pPages, cPages, prot) \
174	do { \
175	change_page_attr(pPages, cPages, PAGE_KERNEL_NOCACHE); \
176	change_page_attr(pPages, cPages, prot); \
177	} while (0)
178	# else
179	# define MY_CHANGE_PAGE_ATTR(pPages, cPages, prot) change_page_attr(pPages, cPages, prot)
180	# endif
181	#endif
182
183
184	/** @def ONE_MSEC_IN_JIFFIES
185	* The number of jiffies that make up 1 millisecond. This is only actually used
186	* when HZ is > 1000. */
187	#if HZ <= 1000
188	# define ONE_MSEC_IN_JIFFIES 0
189	#elif !(HZ % 1000)
190	# define ONE_MSEC_IN_JIFFIES (HZ / 1000)
191	#else
192	# define ONE_MSEC_IN_JIFFIES ((HZ + 999) / 1000)
193	# error "HZ is not a multiple of 1000, the GIP stuff won't work right!"
194	#endif
195
196	#ifdef CONFIG_X86_LOCAL_APIC
197
198	/* If an NMI occurs while we are inside the world switcher the machine will
199	* crash. The Linux NMI watchdog generates periodic NMIs increasing a counter
200	* which is compared with another counter increased in the timer interrupt
201	* handler. We disable the NMI watchdog.
202	*
203	* - Linux >= 2.6.21: The watchdog is disabled by default on i386 and x86_64.
204	* - Linux < 2.6.21: The watchdog is normally enabled by default on x86_64
205	* and disabled on i386.
206	*/
207	# if defined(RT_ARCH_AMD64)
208	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 21)
209	# define DO_DISABLE_NMI 1
210	# endif
211	# endif
212
213	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 19)
214	extern int nmi_active;
215	# define nmi_atomic_read(P) *(P)
216	# define nmi_atomic_set(P, V) *(P) = (V)
217	# define nmi_atomic_dec(P) nmi_atomic_set(P, 0)
218	# else
219	# define nmi_atomic_read(P) atomic_read(P)
220	# define nmi_atomic_set(P, V) atomic_set(P, V)
221	# define nmi_atomic_dec(P) atomic_dec(P)
222	# endif
223
224	# ifndef X86_FEATURE_ARCH_PERFMON
225	# define X86_FEATURE_ARCH_PERFMON (332+9) / Intel Architectural PerfMon */
226	# endif
227	# ifndef MSR_ARCH_PERFMON_EVENTSEL0
228	# define MSR_ARCH_PERFMON_EVENTSEL0 0x186
229	# endif
230	# ifndef ARCH_PERFMON_UNHALTED_CORE_CYCLES_PRESENT
231	# define ARCH_PERFMON_UNHALTED_CORE_CYCLES_PRESENT (1 << 0)
232	# endif
233
234	#endif /* CONFIG_X86_LOCAL_APIC */
235
236
237	/*******************************************************************************
238	* Defined Constants And Macros *
239	*******************************************************************************/
240	/**
241	* Device extention & session data association structure.
242	*/
243	static SUPDRVDEVEXT g_DevExt;
244
245	/** Timer structure for the GIP update. */
246	static struct timer_list g_GipTimer;
247	/** Pointer to the page structure for the GIP. */
248	struct page *g_pGipPage;
249
250	/** Registered devfs device handle. */
251	#if defined(CONFIG_DEVFS_FS) && !defined(CONFIG_VBOXDRV_AS_MISC)
252	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
253	static void *g_hDevFsVBoxDrv = NULL;
254	# else
255	static devfs_handle_t g_hDevFsVBoxDrv = NULL;
256	# endif
257	#endif
258
259	#ifndef CONFIG_VBOXDRV_AS_MISC
260	/** Module major number */
261	#define DEVICE_MAJOR 234
262	/** Saved major device number */
263	static int g_iModuleMajor;
264	#endif /* !CONFIG_VBOXDRV_AS_MISC */
265
266	/** The module name. */
267	#define DEVICE_NAME "vboxdrv"
268
269	#ifdef RT_ARCH_AMD64
270	/**
271	* Memory for the executable memory heap (in IPRT).
272	*/
273	extern uint8_t g_abExecMemory[1572864]; /* 1.5 MB */
274	__asm__(".section execmemory, \"awx\", @progbits\n\t"
275	".align 32\n\t"
276	".globl g_abExecMemory\n"
277	"g_abExecMemory:\n\t"
278	".zero 1572864\n\t"
279	".type g_abExecMemory, @object\n\t"
280	".size g_abExecMemory, 1572864\n\t"
281	".text\n\t");
282	#endif
283
284
285	/*******************************************************************************
286	* Internal Functions *
287	*******************************************************************************/
288	static int VBoxSupDrvInit(void);
289	static void VBoxSupDrvUnload(void);
290	static int VBoxSupDrvCreate(struct inode pInode, struct file pFilp);
291	static int VBoxSupDrvClose(struct inode pInode, struct file pFilp);
292	static int VBoxSupDrvDeviceControl(struct inode pInode, struct file pFilp,
293	unsigned int IOCmd, unsigned long IOArg);
294	static RTR3PTR VBoxSupDrvMapUser(struct page **papPages, unsigned cPages, unsigned fProt, pgprot_t pgFlags);
295	static int VBoxSupDrvInitGip(PSUPDRVDEVEXT pDevExt);
296	static int VBoxSupDrvTermGip(PSUPDRVDEVEXT pDevExt);
297	static void VBoxSupGipTimer(unsigned long ulUser);
298	#ifdef CONFIG_SMP
299	static void VBoxSupGipTimerPerCpu(unsigned long ulUser);
300	static void VBoxSupGipResumePerCpu(void *pvUser);
301	#endif
302	static int VBoxSupDrvOrder(unsigned long size);
303	static int VBoxSupDrvErr2LinuxErr(int);
304
305
306	/** The file_operations structure. */
307	static struct file_operations gFileOpsVBoxDrv =
308	{
309	owner: THIS_MODULE,
310	open: VBoxSupDrvCreate,
311	release: VBoxSupDrvClose,
312	ioctl: VBoxSupDrvDeviceControl,
313	};
314
315	#ifdef CONFIG_VBOXDRV_AS_MISC
316	/** The miscdevice structure. */
317	static struct miscdevice gMiscDevice =
318	{
319	minor: MISC_DYNAMIC_MINOR,
320	name: DEVICE_NAME,
321	fops: &gFileOpsVBoxDrv,
322	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) && \
323	LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 17)
324	devfs_name: DEVICE_NAME,
325	# endif
326	};
327	#endif
328
329	#ifdef CONFIG_X86_LOCAL_APIC
330	# ifdef DO_DISABLE_NMI
331
332	/** Stop AMD NMI watchdog (x86_64 only). */
333	static int stop_k7_watchdog(void)
334	{
335	wrmsr(MSR_K7_EVNTSEL0, 0, 0);
336	return 1;
337	}
338
339	/** Stop Intel P4 NMI watchdog (x86_64 only). */
340	static int stop_p4_watchdog(void)
341	{
342	wrmsr(MSR_P4_IQ_CCCR0, 0, 0);
343	wrmsr(MSR_P4_IQ_CCCR1, 0, 0);
344	wrmsr(MSR_P4_CRU_ESCR0, 0, 0);
345	return 1;
346	}
347
348	/** The new method of detecting the event counter */
349	static int stop_intel_arch_watchdog(void)
350	{
351	unsigned ebx;
352
353	ebx = cpuid_ebx(10);
354	if (!(ebx & ARCH_PERFMON_UNHALTED_CORE_CYCLES_PRESENT))
355	wrmsr(MSR_ARCH_PERFMON_EVENTSEL0, 0, 0);
356	return 1;
357	}
358
359	/** Stop NMI watchdog. */
360	static void vbox_stop_apic_nmi_watchdog(void *unused)
361	{
362	int stopped = 0;
363
364	/* only support LOCAL and IO APICs for now */
365	if ((nmi_watchdog != NMI_LOCAL_APIC) &&
366	(nmi_watchdog != NMI_IO_APIC))
367	return;
368
369	if (nmi_watchdog == NMI_LOCAL_APIC)
370	{
371	switch (boot_cpu_data.x86_vendor)
372	{
373	case X86_VENDOR_AMD:
374	if (strstr(boot_cpu_data.x86_model_id, "Screwdriver"))
375	return;
376	stopped = stop_k7_watchdog();
377	break;
378	case X86_VENDOR_INTEL:
379	if (cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON))
380	{
381	stopped = stop_intel_arch_watchdog();
382	break;
383	}
384	stopped = stop_p4_watchdog();
385	break;
386	default:
387	return;
388	}
389	}
390
391	if (stopped)
392	nmi_atomic_dec(&nmi_active);
393	}
394
395	/** Disable LAPIC NMI watchdog. */
396	static void disable_lapic_nmi_watchdog(void)
397	{
398	BUG_ON(nmi_watchdog != NMI_LOCAL_APIC);
399
400	if (nmi_atomic_read(&nmi_active) <= 0)
401	return;
402
403	on_each_cpu(vbox_stop_apic_nmi_watchdog, NULL, 1, 1);
404
405	BUG_ON(nmi_atomic_read(&nmi_active) != 0);
406
407	/* tell do_nmi() and others that we're not active any more */
408	nmi_watchdog = NMI_NONE;
409	}
410
411	/** Shutdown NMI. */
412	static void nmi_cpu_shutdown(void * dummy)
413	{
414	unsigned int vERR, vPC;
415
416	vPC = apic_read(APIC_LVTPC);
417
418	if ((GET_APIC_DELIVERY_MODE(vPC) == APIC_MODE_NMI) && !(vPC & APIC_LVT_MASKED))
419	{
420	vERR = apic_read(APIC_LVTERR);
421	apic_write(APIC_LVTERR, vERR \| APIC_LVT_MASKED);
422	apic_write(APIC_LVTPC, vPC \| APIC_LVT_MASKED);
423	apic_write(APIC_LVTERR, vERR);
424	}
425	}
426
427	static void nmi_shutdown(void)
428	{
429	on_each_cpu(nmi_cpu_shutdown, NULL, 0, 1);
430	}
431	# endif /* DO_DISABLE_NMI */
432	#endif /* CONFIG_X86_LOCAL_APIC */
433
434	/**
435	* Initialize module.
436	*
437	* @returns appropriate status code.
438	*/
439	static int __init VBoxSupDrvInit(void)
440	{
441	int rc;
442
443	dprintf(("VBoxDrv::ModuleInit\n"));
444
445	#ifdef CONFIG_X86_LOCAL_APIC
446	/*
447	* If an NMI occurs while we are inside the world switcher the macine will crash.
448	* The Linux NMI watchdog generates periodic NMIs increasing a counter which is
449	* compared with another counter increased in the timer interrupt handler. Therefore
450	* we don't allow to setup an NMI watchdog.
451	*/
452	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
453	/*
454	* First test: NMI actiated? Works only works with Linux 2.6 -- 2.4 does not export
455	* the nmi_watchdog variable.
456	*/
457	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19) \|\| \
458	(defined CONFIG_X86_64 && LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0))
459	# ifdef DO_DISABLE_NMI
460	if (nmi_atomic_read(&nmi_active) > 0)
461	{
462	printk(KERN_INFO DEVICE_NAME ": Trying to deactivate the NMI watchdog...\n");
463
464	switch (nmi_watchdog)
465	{
466	case NMI_LOCAL_APIC:
467	disable_lapic_nmi_watchdog();
468	break;
469	case NMI_NONE:
470	nmi_atomic_dec(&nmi_active);
471	break;
472	}
473
474	if (nmi_atomic_read(&nmi_active) == 0)
475	{
476	nmi_shutdown();
477	printk(KERN_INFO DEVICE_NAME ": Successfully done.\n");
478	}
479	else
480	printk(KERN_INFO DEVICE_NAME ": Failed!\n");
481	}
482	# endif /* DO_DISABLE_NMI */
483
484	/*
485	* Permanent IO_APIC mode active? No way to handle this!
486	*/
487	if (nmi_watchdog == NMI_IO_APIC)
488	{
489	printk(KERN_ERR DEVICE_NAME
490	": NMI watchdog in IO_APIC mode active -- refused to load the kernel module!\n"
491	DEVICE_NAME
492	": Please disable the NMI watchdog by specifying 'nmi_watchdog=0' at kernel\n"
493	DEVICE_NAME
494	": command line.\n");
495	return -EINVAL;
496	}
497
498	/*
499	* See arch/i386/kernel/nmi.c on >= 2.6.19: -1 means it can never enabled again
500	*/
501	nmi_atomic_set(&nmi_active, -1);
502	printk(KERN_INFO DEVICE_NAME ": Trying to deactivate the NMI watchdog permanently...\n");
503
504	/*
505	* Now fall through and see if it actually was enabled before. If so, fail
506	* as we cannot deactivate it cleanly from here.
507	*/
508	# else /* < 2.6.19 */
509	/*
510	* Older 2.6 kernels: nmi_watchdog is not initalized by default
511	*/
512	if (nmi_watchdog != NMI_NONE)
513	goto nmi_activated;
514	# endif
515	# endif /* >= 2.6.0 */
516
517	/*
518	* Second test: Interrupt generated by performance counter not masked and can
519	* generate an NMI. Works also with Linux 2.4.
520	*/
521	{
522	unsigned int v, ver, maxlvt;
523
524	v = apic_read(APIC_LVR);
525	ver = GET_APIC_VERSION(v);
526	/* 82489DXs do not report # of LVT entries. */
527	maxlvt = APIC_INTEGRATED(ver) ? GET_APIC_MAXLVT(v) : 2;
528	if (maxlvt >= 4)
529	{
530	/* Read status of performance counter IRQ vector */
531	v = apic_read(APIC_LVTPC);
532
533	/* performance counter generates NMI and is not masked? */
534	if ((GET_APIC_DELIVERY_MODE(v) == APIC_MODE_NMI) && !(v & APIC_LVT_MASKED))
535	{
536	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19) \|\| \
537	(defined CONFIG_X86_64 && LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0))
538	printk(KERN_ERR DEVICE_NAME
539	": NMI watchdog either active or at least initialized. Please disable the NMI\n"
540	DEVICE_NAME
541	": watchdog by specifying 'nmi_watchdog=0' at kernel command line.\n");
542	return -EINVAL;
543	# else /* < 2.6.19 */
544	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
545	nmi_activated:
546	# endif
547	printk(KERN_ERR DEVICE_NAME
548	": NMI watchdog active -- refused to load the kernel module! Please disable\n"
549	DEVICE_NAME
550	": the NMI watchdog by specifying 'nmi_watchdog=0' at kernel command line.\n");
551	return -EINVAL;
552	# endif /* >= 2.6.19 */
553	}
554	}
555	}
556	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
557	printk(KERN_INFO DEVICE_NAME ": Successfully done.\n");
558	# endif /* >= 2.6.19 */
559	#endif /* CONFIG_X86_LOCAL_APIC */
560
561	#ifdef CONFIG_VBOXDRV_AS_MISC
562	rc = misc_register(&gMiscDevice);
563	if (rc)
564	{
565	printk(KERN_ERR DEVICE_NAME ": Can't register misc device! rc=%d\n", rc);
566	return rc;
567	}
568	#else /* !CONFIG_VBOXDRV_AS_MISC */
569	/*
570	* Register character device.
571	*/
572	g_iModuleMajor = DEVICE_MAJOR;
573	rc = VBOX_REGISTER_DEVICE((dev_t)g_iModuleMajor, DEVICE_NAME, &gFileOpsVBoxDrv);
574	if (rc < 0)
575	{
576	dprintf(("VBOX_REGISTER_DEVICE failed with rc=%#x!\n", rc));
577	return rc;
578	}
579
580	/*
581	* Save returned module major number
582	*/
583	if (DEVICE_MAJOR != 0)
584	g_iModuleMajor = DEVICE_MAJOR;
585	else
586	g_iModuleMajor = rc;
587	rc = 0;
588
589	#ifdef CONFIG_DEVFS_FS
590	/*
591	* Register a device entry
592	*/
593	g_hDevFsVBoxDrv = VBOX_REGISTER_DEVFS();
594	if (g_hDevFsVBoxDrv == NULL)
595	{
596	dprintf(("devfs_register failed!\n"));
597	rc = -EINVAL;
598	}
599	#endif
600	#endif /* !CONFIG_VBOXDRV_AS_MISC */
601	if (!rc)
602	{
603	/*
604	* Initialize the runtime.
605	* On AMD64 we'll have to donate the high rwx memory block to the exec allocator.
606	*/
607	rc = RTR0Init(0);
608	if (RT_SUCCESS(rc))
609	{
610	#ifdef RT_ARCH_AMD64
611	rc = RTR0MemExecDonate(&g_abExecMemory[0], sizeof(g_abExecMemory));
612	#endif
613	/*
614	* Initialize the device extension.
615	*/
616	if (RT_SUCCESS(rc))
617	rc = supdrvInitDevExt(&g_DevExt);
618	if (!rc)
619	{
620	/*
621	* Create the GIP page.
622	*/
623	rc = VBoxSupDrvInitGip(&g_DevExt);
624	if (!rc)
625	{
626	dprintf(("VBoxDrv::ModuleInit returning %#x\n", rc));
627	return rc;
628	}
629
630	supdrvDeleteDevExt(&g_DevExt);
631	}
632	else
633	rc = -EINVAL;
634	RTR0Term();
635	}
636	else
637	rc = -EINVAL;
638
639	/*
640	* Failed, cleanup and return the error code.
641	*/
642	#if defined(CONFIG_DEVFS_FS) && !defined(CONFIG_VBOXDRV_AS_MISC)
643	VBOX_UNREGISTER_DEVFS(g_hDevFsVBoxDrv);
644	#endif
645	}
646	#ifdef CONFIG_VBOXDRV_AS_MISC
647	misc_deregister(&gMiscDevice);
648	dprintf(("VBoxDrv::ModuleInit returning %#x (minor:%d)\n", rc, gMiscDevice.minor));
649	#else
650	VBOX_UNREGISTER_DEVICE(g_iModuleMajor, DEVICE_NAME);
651	dprintf(("VBoxDrv::ModuleInit returning %#x (major:%d)\n", rc, g_iModuleMajor));
652	#endif
653	return rc;
654	}
655
656
657	/**
658	* Unload the module.
659	*/
660	static void __exit VBoxSupDrvUnload(void)
661	{
662	int rc;
663	dprintf(("VBoxSupDrvUnload\n"));
664
665	/*
666	* I Don't think it's possible to unload a driver which processes have
667	* opened, at least we'll blindly assume that here.
668	*/
669	#ifdef CONFIG_VBOXDRV_AS_MISC
670	rc = misc_deregister(&gMiscDevice);
671	if (rc < 0)
672	{
673	dprintf(("misc_deregister failed with rc=%#x\n", rc));
674	}
675	#else /* !CONFIG_VBOXDRV_AS_MISC */
676	#ifdef CONFIG_DEVFS_FS
677	/*
678	* Unregister a device entry
679	*/
680	VBOX_UNREGISTER_DEVFS(g_hDevFsVBoxDrv);
681	#endif // devfs
682	rc = VBOX_UNREGISTER_DEVICE(g_iModuleMajor, DEVICE_NAME);
683	if (rc < 0)
684	{
685	dprintf(("unregister_chrdev failed with rc=%#x (major:%d)\n", rc, g_iModuleMajor));
686	}
687	#endif /* !CONFIG_VBOXDRV_AS_MISC */
688
689	/*
690	* Destroy GIP, delete the device extension and terminate IPRT.
691	*/
692	VBoxSupDrvTermGip(&g_DevExt);
693	supdrvDeleteDevExt(&g_DevExt);
694	RTR0Term();
695	}
696
697
698	/**
699	* Device open. Called on open /dev/vboxdrv
700	*
701	* @param pInode Pointer to inode info structure.
702	* @param pFilp Associated file pointer.
703	*/
704	static int VBoxSupDrvCreate(struct inode pInode, struct file pFilp)
705	{
706	int rc;
707	PSUPDRVSESSION pSession;
708	dprintf(("VBoxSupDrvCreate: pFilp=%p\n", pFilp));
709
710	/*
711	* Call common code for the rest.
712	*/
713	rc = supdrvCreateSession(&g_DevExt, (PSUPDRVSESSION *)&pSession);
714	if (!rc)
715	{
716	pSession->Uid = current->euid;
717	pSession->Gid = current->egid;
718	pSession->Process = RTProcSelf();
719	pSession->R0Process = RTR0ProcHandleSelf();
720	}
721
722	dprintf(("VBoxSupDrvCreate: g_DevExt=%p pSession=%p rc=%d\n", &g_DevExt, pSession, rc));
723	pFilp->private_data = pSession;
724
725	return VBoxSupDrvErr2LinuxErr(rc);
726	}
727
728
729	/**
730	* Close device.
731	*
732	* @param pInode Pointer to inode info structure.
733	* @param pFilp Associated file pointer.
734	*/
735	static int VBoxSupDrvClose(struct inode pInode, struct file pFilp)
736	{
737	dprintf(("VBoxSupDrvClose: pFilp=%p private_data=%p\n", pFilp, pFilp->private_data));
738	supdrvCloseSession(&g_DevExt, (PSUPDRVSESSION)pFilp->private_data);
739	pFilp->private_data = NULL;
740	return 0;
741	}
742
743
744	/**
745	* Device I/O Control entry point.
746	*
747	* @param pInode Pointer to inode info structure.
748	* @param pFilp Associated file pointer.
749	* @param IOCmd The function specified to ioctl().
750	* @param IOArg The argument specified to ioctl().
751	*/
752	static int VBoxSupDrvDeviceControl(struct inode pInode, struct file pFilp,
753	unsigned int IOCmd, unsigned long IOArg)
754	{
755	int rc;
756	SUPDRVIOCTLDATA Args;
757	void *pvBuf = NULL;
758	int cbBuf = 0;
759	unsigned cbOut = 0;
760
761	dprintf2(("VBoxSupDrvDeviceControl: pFilp=%p IOCmd=%x IOArg=%p\n", pFilp, IOCmd, (void *)IOArg));
762
763	/*
764	* Copy ioctl data structure from user space.
765	*/
766	if (_IOC_SIZE(IOCmd) != sizeof(SUPDRVIOCTLDATA))
767	{
768	dprintf(("VBoxSupDrvDeviceControl: incorrect input length! cbArgs=%d\n", _IOC_SIZE(IOCmd)));
769	return -EINVAL;
770	}
771	if (copy_from_user(&Args, (void *)IOArg, _IOC_SIZE(IOCmd)))
772	{
773	dprintf(("VBoxSupDrvDeviceControl: copy_from_user(&Args) failed.\n"));
774	return -EFAULT;
775	}
776
777	/*
778	* Allocate and copy user space input data buffer to kernel space.
779	*/
780	if (Args.cbIn > 0 \|\| Args.cbOut > 0)
781	{
782	cbBuf = max(Args.cbIn, Args.cbOut);
783	pvBuf = vmalloc(cbBuf);
784	if (pvBuf == NULL)
785	{
786	dprintf(("VBoxSupDrvDeviceControl: failed to allocate buffer of %d bytes.\n", cbBuf));
787	return -ENOMEM;
788	}
789
790	if (copy_from_user(pvBuf, (void *)Args.pvIn, Args.cbIn))
791	{
792	dprintf(("VBoxSupDrvDeviceControl: copy_from_user(pvBuf) failed.\n"));
793	vfree(pvBuf);
794	return -EFAULT;
795	}
796	}
797
798	/*
799	* Process the IOCtl.
800	*/
801	rc = supdrvIOCtl(IOCmd, &g_DevExt, (PSUPDRVSESSION)pFilp->private_data,
802	pvBuf, Args.cbIn, pvBuf, Args.cbOut, &cbOut);
803
804	/*
805	* Copy ioctl data and output buffer back to user space.
806	*/
807	if (rc)
808	{
809	dprintf(("VBoxSupDrvDeviceControl: pFilp=%p IOCmd=%x IOArg=%p failed, rc=%d (linux rc=%d)\n",
810	pFilp, IOCmd, (void *)IOArg, rc, VBoxSupDrvErr2LinuxErr(rc)));
811	rc = VBoxSupDrvErr2LinuxErr(rc);
812	}
813	else if (cbOut > 0)
814	{
815	if (pvBuf != NULL && cbOut <= cbBuf)
816	{
817	if (copy_to_user((void *)Args.pvOut, pvBuf, cbOut))
818	{
819	dprintf(("copy_to_user failed.\n"));
820	rc = -EFAULT;
821	}
822	}
823	else
824	{
825	dprintf(("WHAT!?! supdrvIOCtl messed up! cbOut=%d cbBuf=%d pvBuf=%p\n", cbOut, cbBuf, pvBuf));
826	rc = -EPERM;
827	}
828	}
829
830	if (pvBuf)
831	vfree(pvBuf);
832
833	dprintf2(("VBoxSupDrvDeviceControl: returns %d\n", rc));
834	return rc;
835	}
836
837
838	/**
839	* Initializes any OS specific object creator fields.
840	*/
841	void VBOXCALL supdrvOSObjInitCreator(PSUPDRVOBJ pObj, PSUPDRVSESSION pSession)
842	{
843	NOREF(pObj);
844	NOREF(pSession);
845	}
846
847
848	/**
849	* Checks if the session can access the object.
850	*
851	* @returns true if a decision has been made.
852	* @returns false if the default access policy should be applied.
853	*
854	* @param pObj The object in question.
855	* @param pSession The session wanting to access the object.
856	* @param pszObjName The object name, can be NULL.
857	* @param prc Where to store the result when returning true.
858	*/
859	bool VBOXCALL supdrvOSObjCanAccess(PSUPDRVOBJ pObj, PSUPDRVSESSION pSession, const char pszObjName, int prc)
860	{
861	NOREF(pObj);
862	NOREF(pSession);
863	NOREF(pszObjName);
864	NOREF(prc);
865	return false;
866	}
867
868
869	/**
870	* Compute order. Some functions allocate 2^order pages.
871	*
872	* @returns order.
873	* @param cPages Number of pages.
874	*/
875	static int VBoxSupDrvOrder(unsigned long cPages)
876	{
877	int iOrder;
878	unsigned long cTmp;
879
880	for (iOrder = 0, cTmp = cPages; cTmp >>= 1; ++iOrder)
881	;
882	if (cPages & ~(1 << iOrder))
883	++iOrder;
884
885	return iOrder;
886	}
887
888
889	/**
890	* OS Specific code for locking down memory.
891	*
892	* @returns 0 on success.
893	* @returns SUPDRV_ERR_* on failure.
894	* @param pMem Pointer to memory.
895	* This is not linked in anywhere.
896	* @param paPages Array which should be filled with the address of the physical pages.
897	*
898	* @remark See sgl_map_user_pages() for an example of an similar function.
899	*/
900	int VBOXCALL supdrvOSLockMemOne(PSUPDRVMEMREF pMem, PSUPPAGE paPages)
901	{
902	int rc;
903	struct page **papPages;
904	unsigned iPage;
905	unsigned cPages = pMem->cb >> PAGE_SHIFT;
906	unsigned long pv = (unsigned long)pMem->pvR3;
907	struct vm_area_struct **papVMAs;
908
909	/*
910	* Allocate page pointer array.
911	*/
912	papPages = vmalloc(cPages * sizeof(*papPages));
913	if (!papPages)
914	return SUPDRV_ERR_NO_MEMORY;
915
916	/*
917	* Allocate the VMA pointer array.
918	*/
919	papVMAs = vmalloc(cPages * sizeof(*papVMAs));
920	if (!papVMAs)
921	return SUPDRV_ERR_NO_MEMORY;
922
923	/*
924	* Get user pages.
925	*/
926	down_read(&current->mm->mmap_sem);
927	rc = get_user_pages(current, /* Task for fault acounting. */
928	current->mm, /* Whose pages. */
929	(unsigned long)pv, /* Where from. */
930	cPages, /* How many pages. */
931	1, /* Write to memory. */
932	0, /* force. */
933	papPages, /* Page array. */
934	papVMAs); /* vmas */
935	if (rc != cPages)
936	{
937	up_read(&current->mm->mmap_sem);
938	dprintf(("supdrvOSLockMemOne: get_user_pages failed. rc=%d\n", rc));
939	return SUPDRV_ERR_LOCK_FAILED;
940	}
941
942	for (iPage = 0; iPage < cPages; iPage++)
943	flush_dcache_page(papPages[iPage]);
944	up_read(&current->mm->mmap_sem);
945
946	pMem->u.locked.papPages = papPages;
947	pMem->u.locked.cPages = cPages;
948
949	/*
950	* Get addresses, protect against fork()
951	*/
952	for (iPage = 0; iPage < cPages; iPage++)
953	{
954	paPages[iPage].Phys = page_to_phys(papPages[iPage]);
955	paPages[iPage].uReserved = 0;
956	papVMAs[iPage]->vm_flags \|= VM_DONTCOPY;
957	}
958
959	vfree(papVMAs);
960
961	dprintf2(("supdrvOSLockMemOne: pvR3=%p cb=%d papPages=%p\n",
962	pMem->pvR3, pMem->cb, pMem->u.locked.papPages));
963	return 0;
964	}
965
966
967	/**
968	* Unlocks the memory pointed to by pv.
969	*
970	* @param pMem Pointer to memory to unlock.
971	*
972	* @remark See sgl_unmap_user_pages() for an example of an similar function.
973	*/
974	void VBOXCALL supdrvOSUnlockMemOne(PSUPDRVMEMREF pMem)
975	{
976	unsigned iPage;
977	dprintf2(("supdrvOSUnlockMemOne: pvR3=%p cb=%d papPages=%p\n",
978	pMem->pvR3, pMem->cb, pMem->u.locked.papPages));
979
980	/*
981	* Loop thru the pages and release them.
982	*/
983	for (iPage = 0; iPage < pMem->u.locked.cPages; iPage++)
984	{
985	if (!PageReserved(pMem->u.locked.papPages[iPage]))
986	SetPageDirty(pMem->u.locked.papPages[iPage]);
987	page_cache_release(pMem->u.locked.papPages[iPage]);
988	}
989
990	/* free the page array */
991	vfree(pMem->u.locked.papPages);
992	pMem->u.locked.cPages = 0;
993	}
994
995
996	/**
997	* OS Specific code for allocating page aligned memory with continuous fixed
998	* physical paged backing.
999	*
1000	* @returns 0 on success.
1001	* @returns SUPDRV_ERR_* on failure.
1002	* @param pMem Memory reference record of the memory to be allocated.
1003	* (This is not linked in anywhere.)
1004	* @param ppvR0 Where to store the virtual address of the ring-0 mapping. (optional)
1005	* @param ppvR3 Where to store the virtual address of the ring-3 mapping.
1006	* @param pHCPhys Where to store the physical address.
1007	*/
1008	int VBOXCALL supdrvOSContAllocOne(PSUPDRVMEMREF pMem, PRTR0PTR ppvR0, PRTR3PTR ppvR3, PRTHCPHYS pHCPhys)
1009	{
1010	struct page *paPages;
1011	unsigned iPage;
1012	unsigned cbAligned = RT_ALIGN(pMem->cb, PAGE_SIZE);
1013	unsigned cPages = cbAligned >> PAGE_SHIFT;
1014	unsigned cOrder = VBoxSupDrvOrder(cPages);
1015	unsigned long ulAddr;
1016	dma_addr_t HCPhys;
1017	int rc = 0;
1018	pgprot_t pgFlags;
1019	pgprot_val(pgFlags) = _PAGE_PRESENT \| _PAGE_RW \| _PAGE_USER;
1020
1021	Assert(ppvR3);
1022	Assert(pHCPhys);
1023
1024	/*
1025	* Allocate page pointer array.
1026	*/
1027	#ifdef RT_ARCH_AMD64 /** @todo check out if there is a correct way of getting memory below 4GB (physically). */
1028	paPages = alloc_pages(GFP_DMA, cOrder);
1029	#else
1030	paPages = alloc_pages(GFP_USER, cOrder);
1031	#endif
1032	if (!paPages)
1033	return SUPDRV_ERR_NO_MEMORY;
1034
1035	/*
1036	* Lock the pages.
1037	*/
1038	for (iPage = 0; iPage < cPages; iPage++)
1039	{
1040	SetPageReserved(&paPages[iPage]);
1041	if (!PageHighMem(&paPages[iPage]) && pgprot_val(MY_PAGE_KERNEL_EXEC) != pgprot_val(PAGE_KERNEL))
1042	MY_CHANGE_PAGE_ATTR(&paPages[iPage], 1, MY_PAGE_KERNEL_EXEC);
1043	#ifdef DEBUG
1044	if (iPage + 1 < cPages && (page_to_phys((&paPages[iPage])) + 0x1000) != page_to_phys((&paPages[iPage + 1])))
1045	{
1046	dprintf(("supdrvOSContAllocOne: Pages are not continuous!!!! iPage=%d phys=%llx physnext=%llx\n",
1047	iPage, (long long)page_to_phys((&paPages[iPage])), (long long)page_to_phys((&paPages[iPage + 1]))));
1048	BUG();
1049	}
1050	#endif
1051	}
1052	HCPhys = page_to_phys(paPages);
1053
1054	/*
1055	* Allocate user space mapping and put the physical pages into it.
1056	*/
1057	down_write(&current->mm->mmap_sem);
1058	ulAddr = do_mmap(NULL, 0, cbAligned, PROT_READ \| PROT_WRITE \| PROT_EXEC, MAP_SHARED \| MAP_ANONYMOUS, 0);
1059	if (!(ulAddr & ~PAGE_MASK))
1060	{
1061	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0) && !defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1062	int rc2 = remap_page_range(ulAddr, HCPhys, cbAligned, pgFlags);
1063	#else
1064	int rc2 = 0;
1065	struct vm_area_struct *vma = find_vma(current->mm, ulAddr);
1066	if (vma)
1067	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11)
1068	rc2 = remap_page_range(vma, ulAddr, HCPhys, cbAligned, pgFlags);
1069	#else
1070	rc2 = remap_pfn_range(vma, ulAddr, HCPhys >> PAGE_SHIFT, cbAligned, pgFlags);
1071	#endif
1072	else
1073	{
1074	rc = SUPDRV_ERR_NO_MEMORY;
1075	dprintf(("supdrvOSContAllocOne: no vma found for ulAddr=%#lx!\n", ulAddr));
1076	}
1077	#endif
1078	if (rc2)
1079	{
1080	rc = SUPDRV_ERR_NO_MEMORY;
1081	dprintf(("supdrvOSContAllocOne: remap_page_range failed rc2=%d\n", rc2));
1082	}
1083	}
1084	else
1085	{
1086	dprintf(("supdrvOSContAllocOne: do_mmap failed ulAddr=%#lx\n", ulAddr));
1087	rc = SUPDRV_ERR_NO_MEMORY;
1088	}
1089	up_write(&current->mm->mmap_sem); /* not quite sure when to give this up. */
1090
1091	/*
1092	* Success?
1093	*/
1094	if (!rc)
1095	{
1096	*pHCPhys = HCPhys;
1097	*ppvR3 = ulAddr;
1098	if (ppvR0)
1099	ppvR0 = (void )ulAddr;
1100	pMem->pvR3 = ulAddr;
1101	pMem->pvR0 = NULL;
1102	pMem->u.cont.paPages = paPages;
1103	pMem->u.cont.cPages = cPages;
1104	pMem->cb = cbAligned;
1105
1106	dprintf2(("supdrvOSContAllocOne: pvR0=%p pvR3=%p cb=%d paPages=%p pHCPhys=%lx ppvR0=*ppvR3=%p\n",
1107	pMem->pvR0, pMem->pvR3, pMem->cb, paPages, (unsigned long)pHCPhys, ppvR3));
1108	global_flush_tlb();
1109	return 0;
1110	}
1111
1112	/*
1113	* Failure, cleanup and be gone.
1114	*/
1115	down_write(&current->mm->mmap_sem);
1116	if (ulAddr & ~PAGE_MASK)
1117	MY_DO_MUNMAP(current->mm, ulAddr, pMem->cb);
1118	for (iPage = 0; iPage < cPages; iPage++)
1119	{
1120	ClearPageReserved(&paPages[iPage]);
1121	if (!PageHighMem(&paPages[iPage]) && pgprot_val(MY_PAGE_KERNEL_EXEC) != pgprot_val(PAGE_KERNEL))
1122	MY_CHANGE_PAGE_ATTR(&paPages[iPage], 1, PAGE_KERNEL);
1123	}
1124	up_write(&current->mm->mmap_sem); /* check when we can leave this. */
1125	__free_pages(paPages, cOrder);
1126
1127	global_flush_tlb();
1128	return rc;
1129	}
1130
1131
1132	/**
1133	* Frees contiguous memory.
1134	*
1135	* @param pMem Memory reference record of the memory to be freed.
1136	*/
1137	void VBOXCALL supdrvOSContFreeOne(PSUPDRVMEMREF pMem)
1138	{
1139	unsigned iPage;
1140
1141	dprintf2(("supdrvOSContFreeOne: pvR0=%p pvR3=%p cb=%d paPages=%p\n",
1142	pMem->pvR0, pMem->pvR3, pMem->cb, pMem->u.cont.paPages));
1143
1144	/*
1145	* do_exit() destroys the mm before closing files.
1146	* I really hope it cleans up our stuff properly...
1147	*/
1148	if (current->mm)
1149	{
1150	down_write(&current->mm->mmap_sem);
1151	MY_DO_MUNMAP(current->mm, (unsigned long)pMem->pvR3, pMem->cb);
1152	up_write(&current->mm->mmap_sem); /* check when we can leave this. */
1153	}
1154
1155	/*
1156	* Change page attributes freeing the pages.
1157	*/
1158	for (iPage = 0; iPage < pMem->u.cont.cPages; iPage++)
1159	{
1160	ClearPageReserved(&pMem->u.cont.paPages[iPage]);
1161	if (!PageHighMem(&pMem->u.cont.paPages[iPage]) && pgprot_val(MY_PAGE_KERNEL_EXEC) != pgprot_val(PAGE_KERNEL))
1162	MY_CHANGE_PAGE_ATTR(&pMem->u.cont.paPages[iPage], 1, PAGE_KERNEL);
1163	}
1164	__free_pages(pMem->u.cont.paPages, VBoxSupDrvOrder(pMem->u.cont.cPages));
1165
1166	pMem->u.cont.cPages = 0;
1167	}
1168
1169
1170	/**
1171	* Allocates memory which mapped into both kernel and user space.
1172	* The returned memory is page aligned and so is the allocation.
1173	*
1174	* @returns 0 on success.
1175	* @returns SUPDRV_ERR_* on failure.
1176	* @param pMem Memory reference record of the memory to be allocated.
1177	* (This is not linked in anywhere.)
1178	* @param ppvR0 Where to store the address of the Ring-0 mapping.
1179	* @param ppvR3 Where to store the address of the Ring-3 mapping.
1180	*/
1181	int VBOXCALL supdrvOSMemAllocOne(PSUPDRVMEMREF pMem, PRTR0PTR ppvR0, PRTR3PTR ppvR3)
1182	{
1183	const unsigned cbAligned = RT_ALIGN(pMem->cb, PAGE_SIZE);
1184	const unsigned cPages = cbAligned >> PAGE_SHIFT;
1185	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 4, 22)
1186	unsigned cOrder = VBoxSupDrvOrder(cPages);
1187	struct page *paPages;
1188	#endif
1189	struct page **papPages;
1190	unsigned iPage;
1191	pgprot_t pgFlags;
1192	pgprot_val(pgFlags) = _PAGE_PRESENT \| _PAGE_RW \| _PAGE_USER;
1193
1194	/*
1195	* Allocate array with page pointers.
1196	*/
1197	pMem->u.mem.cPages = 0;
1198	pMem->u.mem.papPages = papPages = kmalloc(sizeof(papPages[0]) * cPages, GFP_KERNEL);
1199	if (!papPages)
1200	return SUPDRV_ERR_NO_MEMORY;
1201
1202	/*
1203	* Allocate the pages.
1204	*/
1205	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1206	for (iPage = 0; iPage < cPages; iPage++)
1207	{
1208	papPages[iPage] = alloc_page(GFP_HIGHUSER);
1209	if (!papPages[iPage])
1210	{
1211	pMem->u.mem.cPages = iPage;
1212	supdrvOSMemFreeOne(pMem);
1213	return SUPDRV_ERR_NO_MEMORY;
1214	}
1215	}
1216
1217	#else /* < 2.4.22 */
1218	paPages = alloc_pages(GFP_USER, cOrder);
1219	if (!paPages)
1220	{
1221	supdrvOSMemFreeOne(pMem);
1222	return SUPDRV_ERR_NO_MEMORY;
1223	}
1224	for (iPage = 0; iPage < cPages; iPage++)
1225	{
1226	papPages[iPage] = &paPages[iPage];
1227	if (pgprot_val(MY_PAGE_KERNEL_EXEC) != pgprot_val(PAGE_KERNEL))
1228	MY_CHANGE_PAGE_ATTR(papPages[iPage], 1, MY_PAGE_KERNEL_EXEC);
1229	if (PageHighMem(papPages[iPage]))
1230	BUG();
1231	}
1232	#endif
1233	pMem->u.mem.cPages = cPages;
1234
1235	/*
1236	* Reserve the pages.
1237	*/
1238	for (iPage = 0; iPage < cPages; iPage++)
1239	SetPageReserved(papPages[iPage]);
1240
1241	/*
1242	* Create the Ring-0 mapping.
1243	*/
1244	if (ppvR0)
1245	{
1246	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1247	# ifdef VM_MAP
1248	*ppvR0 = pMem->pvR0 = vmap(papPages, cPages, VM_MAP, pgFlags);
1249	# else
1250	*ppvR0 = pMem->pvR0 = vmap(papPages, cPages, VM_ALLOC, pgFlags);
1251	# endif
1252	#else
1253	*ppvR0 = pMem->pvR0 = phys_to_virt(page_to_phys(papPages[0]));
1254	#endif
1255	}
1256	if (pMem->pvR0 \|\| !ppvR0)
1257	{
1258	/*
1259	* Create the ring3 mapping.
1260	*/
1261	if (ppvR3)
1262	*ppvR3 = pMem->pvR3 = VBoxSupDrvMapUser(papPages, cPages, PROT_READ \| PROT_WRITE \| PROT_EXEC, pgFlags);
1263	if (pMem->pvR3 \|\| !ppvR3)
1264	return 0;
1265	dprintf(("supdrvOSMemAllocOne: failed to map into r3! cPages=%u\n", cPages));
1266	}
1267	else
1268	dprintf(("supdrvOSMemAllocOne: failed to map into r0! cPages=%u\n", cPages));
1269
1270	supdrvOSMemFreeOne(pMem);
1271	return SUPDRV_ERR_NO_MEMORY;
1272	}
1273
1274
1275	/**
1276	* Get the physical addresses of the pages in the allocation.
1277	* This is called while inside bundle the spinlock.
1278	*
1279	* @param pMem Memory reference record of the memory.
1280	* @param paPages Where to store the page addresses.
1281	*/
1282	void VBOXCALL supdrvOSMemGetPages(PSUPDRVMEMREF pMem, PSUPPAGE paPages)
1283	{
1284	unsigned iPage;
1285	for (iPage = 0; iPage < pMem->u.mem.cPages; iPage++)
1286	{
1287	paPages[iPage].Phys = page_to_phys(pMem->u.mem.papPages[iPage]);
1288	paPages[iPage].uReserved = 0;
1289	}
1290	}
1291
1292
1293	/**
1294	* Frees memory allocated by supdrvOSMemAllocOne().
1295	*
1296	* @param pMem Memory reference record of the memory to be free.
1297	*/
1298	void VBOXCALL supdrvOSMemFreeOne(PSUPDRVMEMREF pMem)
1299	{
1300	dprintf2(("supdrvOSMemFreeOne: pvR0=%p pvR3=%p cb=%d cPages=%d papPages=%p\n",
1301	pMem->pvR0, pMem->pvR3, pMem->cb, pMem->u.mem.cPages, pMem->u.mem.papPages));
1302
1303	/*
1304	* Unmap the user mapping (if any).
1305	* do_exit() destroys the mm before closing files.
1306	*/
1307	if (pMem->pvR3 && current->mm)
1308	{
1309	down_write(&current->mm->mmap_sem);
1310	MY_DO_MUNMAP(current->mm, (unsigned long)pMem->pvR3, RT_ALIGN(pMem->cb, PAGE_SIZE));
1311	up_write(&current->mm->mmap_sem); /* check when we can leave this. */
1312	}
1313	pMem->pvR3 = NIL_RTR3PTR;
1314
1315	/*
1316	* Unmap the kernel mapping (if any).
1317	*/
1318	if (pMem->pvR0)
1319	{
1320	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1321	vunmap(pMem->pvR0);
1322	#endif
1323	pMem->pvR0 = NULL;
1324	}
1325
1326	/*
1327	* Free the physical pages.
1328	*/
1329	if (pMem->u.mem.papPages)
1330	{
1331	struct page **papPages = pMem->u.mem.papPages;
1332	const unsigned cPages = pMem->u.mem.cPages;
1333	unsigned iPage;
1334
1335	/* Restore the page flags. */
1336	for (iPage = 0; iPage < cPages; iPage++)
1337	{
1338	ClearPageReserved(papPages[iPage]);
1339	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 4, 22)
1340	if (pgprot_val(MY_PAGE_KERNEL_EXEC) != pgprot_val(PAGE_KERNEL))
1341	MY_CHANGE_PAGE_ATTR(papPages[iPage], 1, PAGE_KERNEL);
1342	#endif
1343	}
1344
1345	/* Free the pages. */
1346	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1347	for (iPage = 0; iPage < pMem->u.cont.cPages; iPage++)
1348	__free_page(papPages[iPage]);
1349	#else
1350	if (cPages > 0)
1351	__free_pages(papPages[0], VBoxSupDrvOrder(cPages));
1352	#endif
1353	/* Free the page pointer array. */
1354	kfree(papPages);
1355	pMem->u.mem.papPages = NULL;
1356	}
1357	pMem->u.mem.cPages = 0;
1358	}
1359
1360
1361	/**
1362	* Maps a range of pages into user space.
1363	*
1364	* @returns Pointer to the user space mapping on success.
1365	* @returns NULL on failure.
1366	* @param papPages Array of the pages to map.
1367	* @param cPages Number of pages to map.
1368	* @param fProt The mapping protection.
1369	* @param pgFlags The page level protection.
1370	*/
1371	static RTR3PTR VBoxSupDrvMapUser(struct page **papPages, unsigned cPages, unsigned fProt, pgprot_t pgFlags)
1372	{
1373	int rc = SUPDRV_ERR_NO_MEMORY;
1374	unsigned long ulAddr;
1375
1376	/*
1377	* Allocate user space mapping.
1378	*/
1379	down_write(&current->mm->mmap_sem);
1380	ulAddr = do_mmap(NULL, 0, cPages * PAGE_SIZE, fProt, MAP_SHARED \| MAP_ANONYMOUS, 0);
1381	if (!(ulAddr & ~PAGE_MASK))
1382	{
1383	/*
1384	* Map page by page into the mmap area.
1385	* This is generic, paranoid and not very efficient.
1386	*/
1387	int rc = 0;
1388	unsigned long ulAddrCur = ulAddr;
1389	unsigned iPage;
1390	for (iPage = 0; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE)
1391	{
1392	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1393	struct vm_area_struct *vma = find_vma(current->mm, ulAddrCur);
1394	if (!vma)
1395	break;
1396	#endif
1397
1398	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
1399	rc = remap_pfn_range(vma, ulAddrCur, page_to_pfn(papPages[iPage]), PAGE_SIZE, pgFlags);
1400	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1401	rc = remap_page_range(vma, ulAddrCur, page_to_phys(papPages[iPage]), PAGE_SIZE, pgFlags);
1402	#else /* 2.4 */
1403	rc = remap_page_range(ulAddrCur, page_to_phys(papPages[iPage]), PAGE_SIZE, pgFlags);
1404	#endif
1405	if (rc)
1406	break;
1407	}
1408
1409	/*
1410	* Successful?
1411	*/
1412	if (iPage >= cPages)
1413	{
1414	up_write(&current->mm->mmap_sem);
1415	return ulAddr;
1416	}
1417
1418	/* no, cleanup! */
1419	if (rc)
1420	dprintf(("VBoxSupDrvMapUser: remap_[page\|pfn]_range failed! rc=%d\n", rc));
1421	else
1422	dprintf(("VBoxSupDrvMapUser: find_vma failed!\n"));
1423
1424	MY_DO_MUNMAP(current->mm, ulAddr, cPages << PAGE_SHIFT);
1425	}
1426	else
1427	{
1428	dprintf(("supdrvOSContAllocOne: do_mmap failed ulAddr=%#lx\n", ulAddr));
1429	rc = SUPDRV_ERR_NO_MEMORY;
1430	}
1431	up_write(&current->mm->mmap_sem);
1432
1433	return NIL_RTR3PTR;
1434	}
1435
1436
1437	/**
1438	* Initializes the GIP.
1439	*
1440	* @returns negative errno.
1441	* @param pDevExt Instance data. GIP stuff may be updated.
1442	*/
1443	static int VBoxSupDrvInitGip(PSUPDRVDEVEXT pDevExt)
1444	{
1445	struct page *pPage;
1446	dma_addr_t HCPhys;
1447	PSUPGLOBALINFOPAGE pGip;
1448	#ifdef CONFIG_SMP
1449	unsigned i;
1450	#endif
1451	dprintf(("VBoxSupDrvInitGip:\n"));
1452
1453	/*
1454	* Allocate the page.
1455	*/
1456	pPage = alloc_pages(GFP_USER, 0);
1457	if (!pPage)
1458	{
1459	dprintf(("VBoxSupDrvInitGip: failed to allocate the GIP page\n"));
1460	return -ENOMEM;
1461	}
1462
1463	/*
1464	* Lock the page.
1465	*/
1466	SetPageReserved(pPage);
1467	g_pGipPage = pPage;
1468
1469	/*
1470	* Call common initialization routine.
1471	*/
1472	HCPhys = page_to_phys(pPage);
1473	pGip = (PSUPGLOBALINFOPAGE)page_address(pPage);
1474	pDevExt->ulLastJiffies = jiffies;
1475	#ifdef TICK_NSEC
1476	pDevExt->u64LastMonotime = (uint64_t)pDevExt->ulLastJiffies * TICK_NSEC;
1477	dprintf(("VBoxSupDrvInitGIP: TICK_NSEC=%ld HZ=%d jiffies=%ld now=%lld\n",
1478	TICK_NSEC, HZ, pDevExt->ulLastJiffies, pDevExt->u64LastMonotime));
1479	#else
1480	pDevExt->u64LastMonotime = (uint64_t)pDevExt->ulLastJiffies * (1000000 / HZ);
1481	dprintf(("VBoxSupDrvInitGIP: TICK_NSEC=%d HZ=%d jiffies=%ld now=%lld\n",
1482	(int)(1000000 / HZ), HZ, pDevExt->ulLastJiffies, pDevExt->u64LastMonotime));
1483	#endif
1484	supdrvGipInit(pDevExt, pGip, HCPhys, pDevExt->u64LastMonotime,
1485	HZ <= 1000 ? HZ : 1000);
1486
1487	/*
1488	* Initialize the timer.
1489	*/
1490	init_timer(&g_GipTimer);
1491	g_GipTimer.data = (unsigned long)pDevExt;
1492	g_GipTimer.function = VBoxSupGipTimer;
1493	g_GipTimer.expires = jiffies;
1494	#ifdef CONFIG_SMP
1495	for (i = 0; i < RT_ELEMENTS(pDevExt->aCPUs); i++)
1496	{
1497	pDevExt->aCPUs[i].u64LastMonotime = pDevExt->u64LastMonotime;
1498	pDevExt->aCPUs[i].ulLastJiffies = pDevExt->ulLastJiffies;
1499	pDevExt->aCPUs[i].iSmpProcessorId = -512;
1500	init_timer(&pDevExt->aCPUs[i].Timer);
1501	pDevExt->aCPUs[i].Timer.data = i;
1502	pDevExt->aCPUs[i].Timer.function = VBoxSupGipTimerPerCpu;
1503	pDevExt->aCPUs[i].Timer.expires = jiffies;
1504	}
1505	#endif
1506
1507	return 0;
1508	}
1509
1510
1511	/**
1512	* Terminates the GIP.
1513	*
1514	* @returns negative errno.
1515	* @param pDevExt Instance data. GIP stuff may be updated.
1516	*/
1517	static int VBoxSupDrvTermGip(PSUPDRVDEVEXT pDevExt)
1518	{
1519	struct page *pPage;
1520	PSUPGLOBALINFOPAGE pGip;
1521	#ifdef CONFIG_SMP
1522	unsigned i;
1523	#endif
1524	dprintf(("VBoxSupDrvTermGip:\n"));
1525
1526	/*
1527	* Delete the timer if it's pending.
1528	*/
1529	if (timer_pending(&g_GipTimer))
1530	del_timer_sync(&g_GipTimer);
1531	#ifdef CONFIG_SMP
1532	for (i = 0; i < RT_ELEMENTS(pDevExt->aCPUs); i++)
1533	if (timer_pending(&pDevExt->aCPUs[i].Timer))
1534	del_timer_sync(&pDevExt->aCPUs[i].Timer);
1535	#endif
1536
1537	/*
1538	* Uninitialize the content.
1539	*/
1540	pGip = pDevExt->pGip;
1541	pDevExt->pGip = NULL;
1542	if (pGip)
1543	supdrvGipTerm(pGip);
1544
1545	/*
1546	* Free the page.
1547	*/
1548	pPage = g_pGipPage;
1549	g_pGipPage = NULL;
1550	if (pPage)
1551	{
1552	ClearPageReserved(pPage);
1553	__free_pages(pPage, 0);
1554	}
1555
1556	return 0;
1557	}
1558
1559	/**
1560	* Timer callback function.
1561	*
1562	* In ASYNC TSC mode this is called on the primary CPU, and we're
1563	* assuming that the CPU remains online.
1564	*
1565	* @param ulUser The device extension pointer.
1566	*/
1567	static void VBoxSupGipTimer(unsigned long ulUser)
1568	{
1569	PSUPDRVDEVEXT pDevExt;
1570	PSUPGLOBALINFOPAGE pGip;
1571	unsigned long ulNow;
1572	unsigned long ulDiff;
1573	uint64_t u64Monotime;
1574	unsigned long SavedFlags;
1575
1576	local_irq_save(SavedFlags);
1577
1578	pDevExt = (PSUPDRVDEVEXT)ulUser;
1579	pGip = pDevExt->pGip;
1580	ulNow = jiffies;
1581
1582	#ifdef CONFIG_SMP
1583	if (pGip && pGip->u32Mode == SUPGIPMODE_ASYNC_TSC)
1584	{
1585	uint8_t iCPU = ASMGetApicId();
1586	ulDiff = ulNow - pDevExt->aCPUs[iCPU].ulLastJiffies;
1587	pDevExt->aCPUs[iCPU].ulLastJiffies = ulNow;
1588	#ifdef TICK_NSEC
1589	u64Monotime = pDevExt->aCPUs[iCPU].u64LastMonotime + ulDiff * TICK_NSEC;
1590	#else
1591	u64Monotime = pDevExt->aCPUs[iCPU].u64LastMonotime + ulDiff * (1000000 / HZ);
1592	#endif
1593	pDevExt->aCPUs[iCPU].u64LastMonotime = u64Monotime;
1594	}
1595	else
1596	#endif /* CONFIG_SMP */
1597	{
1598	ulDiff = ulNow - pDevExt->ulLastJiffies;
1599	pDevExt->ulLastJiffies = ulNow;
1600	#ifdef TICK_NSEC
1601	u64Monotime = pDevExt->u64LastMonotime + ulDiff * TICK_NSEC;
1602	#else
1603	u64Monotime = pDevExt->u64LastMonotime + ulDiff * (1000000 / HZ);
1604	#endif
1605	pDevExt->u64LastMonotime = u64Monotime;
1606	}
1607	if (RT_LIKELY(pGip))
1608	supdrvGipUpdate(pDevExt->pGip, u64Monotime);
1609	if (RT_LIKELY(!pDevExt->fGIPSuspended))
1610	mod_timer(&g_GipTimer, ulNow + (HZ <= 1000 ? 0 : ONE_MSEC_IN_JIFFIES));
1611
1612	local_irq_restore(SavedFlags);
1613	}
1614
1615
1616	#ifdef CONFIG_SMP
1617	/**
1618	* Timer callback function for the other CPUs.
1619	*
1620	* @param iTimerCPU The APIC ID of this timer.
1621	*/
1622	static void VBoxSupGipTimerPerCpu(unsigned long iTimerCPU)
1623	{
1624	PSUPDRVDEVEXT pDevExt;
1625	PSUPGLOBALINFOPAGE pGip;
1626	uint8_t iCPU;
1627	uint64_t u64Monotime;
1628	unsigned long SavedFlags;
1629
1630	local_irq_save(SavedFlags);
1631
1632	pDevExt = &g_DevExt;
1633	pGip = pDevExt->pGip;
1634	iCPU = ASMGetApicId();
1635
1636	if (RT_LIKELY(iCPU < RT_ELEMENTS(pGip->aCPUs)))
1637	{
1638	if (RT_LIKELY(iTimerCPU == iCPU))
1639	{
1640	unsigned long ulNow = jiffies;
1641	unsigned long ulDiff = ulNow - pDevExt->aCPUs[iCPU].ulLastJiffies;
1642	pDevExt->aCPUs[iCPU].ulLastJiffies = ulNow;
1643	#ifdef TICK_NSEC
1644	u64Monotime = pDevExt->aCPUs[iCPU].u64LastMonotime + ulDiff * TICK_NSEC;
1645	#else
1646	u64Monotime = pDevExt->aCPUs[iCPU].u64LastMonotime + ulDiff * (1000000 / HZ);
1647	#endif
1648	pDevExt->aCPUs[iCPU].u64LastMonotime = u64Monotime;
1649	if (RT_LIKELY(pGip))
1650	supdrvGipUpdatePerCpu(pGip, u64Monotime, iCPU);
1651	if (RT_LIKELY(!pDevExt->fGIPSuspended))
1652	mod_timer(&pDevExt->aCPUs[iCPU].Timer, ulNow + (HZ <= 1000 ? 0 : ONE_MSEC_IN_JIFFIES));
1653	}
1654	else
1655	printk("vboxdrv: error: GIP CPU update timer executing on the wrong CPU: apicid=%d != timer-apicid=%ld (cpuid=%d !=? timer-cpuid=%d)\n",
1656	iCPU, iTimerCPU, smp_processor_id(), pDevExt->aCPUs[iTimerCPU].iSmpProcessorId);
1657	}
1658	else
1659	printk("vboxdrv: error: APIC ID is bogus (GIP CPU update): apicid=%d max=%lu cpuid=%d\n",
1660	iCPU, (unsigned long)RT_ELEMENTS(pGip->aCPUs), smp_processor_id());
1661
1662	local_irq_restore(SavedFlags);
1663	}
1664	#endif /* CONFIG_SMP */
1665
1666
1667	/**
1668	* Maps the GIP into user space.
1669	*
1670	* @returns negative errno.
1671	* @param pDevExt Instance data.
1672	*/
1673	int VBOXCALL supdrvOSGipMap(PSUPDRVDEVEXT pDevExt, PSUPGLOBALINFOPAGE *ppGip)
1674	{
1675	int rc = 0;
1676	unsigned long ulAddr;
1677	unsigned long HCPhys = pDevExt->HCPhysGip;
1678	pgprot_t pgFlags;
1679	pgprot_val(pgFlags) = _PAGE_PRESENT \| _PAGE_USER;
1680	dprintf2(("supdrvOSGipMap: ppGip=%p\n", ppGip));
1681
1682	/*
1683	* Allocate user space mapping and put the physical pages into it.
1684	*/
1685	down_write(&current->mm->mmap_sem);
1686	ulAddr = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ, MAP_SHARED \| MAP_ANONYMOUS, 0);
1687	if (!(ulAddr & ~PAGE_MASK))
1688	{
1689	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0) && !defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1690	int rc2 = remap_page_range(ulAddr, HCPhys, PAGE_SIZE, pgFlags);
1691	#else
1692	int rc2 = 0;
1693	struct vm_area_struct *vma = find_vma(current->mm, ulAddr);
1694	if (vma)
1695	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11)
1696	rc2 = remap_page_range(vma, ulAddr, HCPhys, PAGE_SIZE, pgFlags);
1697	#else
1698	rc2 = remap_pfn_range(vma, ulAddr, HCPhys >> PAGE_SHIFT, PAGE_SIZE, pgFlags);
1699	#endif
1700	else
1701	{
1702	rc = SUPDRV_ERR_NO_MEMORY;
1703	dprintf(("supdrvOSGipMap: no vma found for ulAddr=%#lx!\n", ulAddr));
1704	}
1705	#endif
1706	if (rc2)
1707	{
1708	rc = SUPDRV_ERR_NO_MEMORY;
1709	dprintf(("supdrvOSGipMap: remap_page_range failed rc2=%d\n", rc2));
1710	}
1711	}
1712	else
1713	{
1714	dprintf(("supdrvOSGipMap: do_mmap failed ulAddr=%#lx\n", ulAddr));
1715	rc = SUPDRV_ERR_NO_MEMORY;
1716	}
1717	up_write(&current->mm->mmap_sem); /* not quite sure when to give this up. */
1718
1719	/*
1720	* Success?
1721	*/
1722	if (!rc)
1723	{
1724	*ppGip = (PSUPGLOBALINFOPAGE)ulAddr;
1725	dprintf2(("supdrvOSGipMap: ppGip=%p\n", *ppGip));
1726	return 0;
1727	}
1728
1729	/*
1730	* Failure, cleanup and be gone.
1731	*/
1732	if (ulAddr & ~PAGE_MASK)
1733	{
1734	down_write(&current->mm->mmap_sem);
1735	MY_DO_MUNMAP(current->mm, ulAddr, PAGE_SIZE);
1736	up_write(&current->mm->mmap_sem);
1737	}
1738
1739	dprintf2(("supdrvOSGipMap: returns %d\n", rc));
1740	return rc;
1741	}
1742
1743
1744	/**
1745	* Maps the GIP into user space.
1746	*
1747	* @returns negative errno.
1748	* @param pDevExt Instance data.
1749	*/
1750	int VBOXCALL supdrvOSGipUnmap(PSUPDRVDEVEXT pDevExt, PSUPGLOBALINFOPAGE pGip)
1751	{
1752	dprintf2(("supdrvOSGipUnmap: pGip=%p\n", pGip));
1753	if (current->mm)
1754	{
1755	down_write(&current->mm->mmap_sem);
1756	MY_DO_MUNMAP(current->mm, (unsigned long)pGip, PAGE_SIZE);
1757	up_write(&current->mm->mmap_sem);
1758	}
1759	dprintf2(("supdrvOSGipUnmap: returns 0\n"));
1760	return 0;
1761	}
1762
1763
1764	/**
1765	* Resumes the GIP updating.
1766	*
1767	* @param pDevExt Instance data.
1768	*/
1769	void VBOXCALL supdrvOSGipResume(PSUPDRVDEVEXT pDevExt)
1770	{
1771	dprintf2(("supdrvOSGipResume:\n"));
1772	ASMAtomicXchgU8(&pDevExt->fGIPSuspended, false);
1773	#ifdef CONFIG_SMP
1774	if (pDevExt->pGip->u32Mode != SUPGIPMODE_ASYNC_TSC)
1775	#endif
1776	mod_timer(&g_GipTimer, jiffies);
1777	#ifdef CONFIG_SMP
1778	else
1779	{
1780	mod_timer(&g_GipTimer, jiffies);
1781	smp_call_function(VBoxSupGipResumePerCpu, pDevExt, 0 /* retry /, 1 / wait */);
1782	}
1783	#endif
1784	}
1785
1786
1787	#ifdef CONFIG_SMP
1788	/**
1789	* Callback for resuming GIP updating on the other CPUs.
1790	*
1791	* This is only used when the GIP is in async tsc mode.
1792	*
1793	* @param pvUser Pointer to the device instance.
1794	*/
1795	static void VBoxSupGipResumePerCpu(void *pvUser)
1796	{
1797	PSUPDRVDEVEXT pDevExt = (PSUPDRVDEVEXT)pvUser;
1798	uint8_t iCPU = ASMGetApicId();
1799
1800	if (RT_UNLIKELY(iCPU >= RT_ELEMENTS(pDevExt->pGip->aCPUs)))
1801	{
1802	printk("vboxdrv: error: apicid=%d max=%lu cpuid=%d\n",
1803	iCPU, (unsigned long)RT_ELEMENTS(pDevExt->pGip->aCPUs), smp_processor_id());
1804	return;
1805	}
1806
1807	pDevExt->aCPUs[iCPU].iSmpProcessorId = smp_processor_id();
1808	mod_timer(&pDevExt->aCPUs[iCPU].Timer, jiffies);
1809	}
1810	#endif /* CONFIG_SMP */
1811
1812
1813	/**
1814	* Suspends the GIP updating.
1815	*
1816	* @param pDevExt Instance data.
1817	*/
1818	void VBOXCALL supdrvOSGipSuspend(PSUPDRVDEVEXT pDevExt)
1819	{
1820	#ifdef CONFIG_SMP
1821	unsigned i;
1822	#endif
1823	dprintf2(("supdrvOSGipSuspend:\n"));
1824	ASMAtomicXchgU8(&pDevExt->fGIPSuspended, true);
1825
1826	if (timer_pending(&g_GipTimer))
1827	del_timer_sync(&g_GipTimer);
1828	#ifdef CONFIG_SMP
1829	for (i = 0; i < RT_ELEMENTS(pDevExt->aCPUs); i++)
1830	if (timer_pending(&pDevExt->aCPUs[i].Timer))
1831	del_timer_sync(&pDevExt->aCPUs[i].Timer);
1832	#endif
1833	}
1834
1835
1836	/**
1837	* Get the current CPU count.
1838	* @returns Number of cpus.
1839	*/
1840	unsigned VBOXCALL supdrvOSGetCPUCount(void)
1841	{
1842	#ifdef CONFIG_SMP
1843	# ifdef num_present_cpus
1844	return num_present_cpus();
1845	# else
1846	return smp_num_cpus;
1847	# endif
1848	#else
1849	return 1;
1850	#endif
1851	}
1852
1853	/**
1854	* Force async tsc mode.
1855	* @todo add a module argument for this.
1856	*/
1857	bool VBOXCALL supdrvOSGetForcedAsyncTscMode(void)
1858	{
1859	return false;
1860	}
1861
1862
1863	/**
1864	* Converts a supdrv error code to an linux error code.
1865	*
1866	* @returns corresponding linux error code.
1867	* @param rc supdrv error code (SUPDRV_ERR_* defines).
1868	*/
1869	static int VBoxSupDrvErr2LinuxErr(int rc)
1870	{
1871	switch (rc)
1872	{
1873	case 0: return 0;
1874	case SUPDRV_ERR_GENERAL_FAILURE: return -EACCES;
1875	case SUPDRV_ERR_INVALID_PARAM: return -EINVAL;
1876	case SUPDRV_ERR_INVALID_MAGIC: return -EILSEQ;
1877	case SUPDRV_ERR_INVALID_HANDLE: return -ENXIO;
1878	case SUPDRV_ERR_INVALID_POINTER: return -EFAULT;
1879	case SUPDRV_ERR_LOCK_FAILED: return -ENOLCK;
1880	case SUPDRV_ERR_ALREADY_LOADED: return -EEXIST;
1881	case SUPDRV_ERR_PERMISSION_DENIED: return -EPERM;
1882	case SUPDRV_ERR_VERSION_MISMATCH: return -ENOSYS;
1883	case SUPDRV_ERR_IDT_FAILED: return -1000;
1884	}
1885
1886	return -EPERM;
1887	}
1888
1889
1890	RTDECL(int) SUPR0Printf(const char *pszFormat, ...)
1891	{
1892	#if 1
1893	va_list args;
1894	char szMsg[512];
1895
1896	va_start(args, pszFormat);
1897	vsnprintf(szMsg, sizeof(szMsg) - 1, pszFormat, args);
1898	szMsg[sizeof(szMsg) - 1] = '\0';
1899	printk("%s", szMsg);
1900	va_end(args);
1901	#else
1902	/* forward to printf - needs some more GCC hacking to fix ebp... */
1903	__asm__ __volatile__ ("mov %0, %esp\n\t"
1904	"jmp %1\n\t",
1905	:: "r" ((uintptr_t)&pszFormat - 4),
1906	"m" (printk));
1907	#endif
1908	return 0;
1909	}
1910
1911
1912	/** Runtime assert implementation for Linux Ring-0. */
1913	RTDECL(void) AssertMsg1(const char pszExpr, unsigned uLine, const char pszFile, const char *pszFunction)
1914	{
1915	printk("!!Assertion Failed!!\n"
1916	"Expression: %s\n"
1917	"Location : %s(%d) %s\n",
1918	pszExpr, pszFile, uLine, pszFunction);
1919	}
1920
1921
1922	/** Runtime assert implementation for Linux Ring-0. */
1923	RTDECL(void) AssertMsg2(const char *pszFormat, ...)
1924	{ /* forwarder. */
1925	va_list ap;
1926	char msg[256];
1927
1928	va_start(ap, pszFormat);
1929	vsnprintf(msg, sizeof(msg) - 1, pszFormat, ap);
1930	msg[sizeof(msg) - 1] = '\0';
1931	printk("%s", msg);
1932	va_end(ap);
1933	}
1934
1935
1936	/* GCC C++ hack. */
1937	unsigned __gxx_personality_v0 = 0xcccccccc;
1938
1939
1940	module_init(VBoxSupDrvInit);
1941	module_exit(VBoxSupDrvUnload);
1942
1943	MODULE_AUTHOR("innotek GmbH");
1944	MODULE_DESCRIPTION("VirtualBox Support Driver");
1945	MODULE_LICENSE("GPL");
1946	#ifdef MODULE_VERSION
1947	#define xstr(s) str(s)
1948	#define str(s) #s
1949	MODULE_VERSION(VBOX_VERSION_STRING " (" xstr(SUPDRVIOC_VERSION) ")");
1950	#endif

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

source: vbox/trunk/src/VBox/HostDrivers/Support/linux/SUPDrv-linux.c@ 3968

Download in other formats: