SUPDrv-linux.c@ 1666

Last change on this file since 1666 was 1666, checked in by vboxsync, 18 years ago
make VBOX_VERSION visible with modinfo
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 50.1 KB

Line
1	/** @file
2	* The VirtualBox Support Driver - Linux hosts.
3	*/
4
5	/*
6	* Copyright (C) 2006 InnoTek Systemberatung 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
21	/*******************************************************************************
22	* Header Files *
23	*******************************************************************************/
24	#include "SUPDRV.h"
25	#include "version-generated.h"
26
27	#include <iprt/assert.h>
28	#include <iprt/spinlock.h>
29	#include <iprt/semaphore.h>
30	#include <iprt/initterm.h>
31	#include <iprt/process.h>
32	#include <iprt/err.h>
33	#include <iprt/mem.h>
34
35	#include <linux/module.h>
36	#include <linux/kernel.h>
37	#include <linux/init.h>
38	#include <linux/fs.h>
39	#include <linux/mm.h>
40	#include <linux/pagemap.h>
41	#include <linux/slab.h>
42	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
43	# include <linux/jiffies.h>
44	#endif
45	#include <asm/mman.h>
46	#include <asm/io.h>
47	#include <asm/uaccess.h>
48	#ifdef CONFIG_DEVFS_FS
49	# include <linux/devfs_fs_kernel.h>
50	#endif
51	#ifdef CONFIG_VBOXDRV_AS_MISC
52	# include <linux/miscdevice.h>
53	#endif
54	#ifdef CONFIG_X86_LOCAL_APIC
55	# include <asm/apic.h>
56	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
57	# include <asm/nmi.h>
58	# endif
59	#endif
60
61	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0)
62	# ifndef page_to_pfn
63	# define page_to_pfn(page) ((page) - mem_map)
64	# endif
65	# include <asm/pgtable.h>
66	# define global_flush_tlb __flush_tlb_global
67	#endif
68
69	/* devfs defines */
70	#if defined(CONFIG_DEVFS_FS) && !defined(CONFIG_VBOXDRV_AS_MISC)
71	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
72
73	# define VBOX_REGISTER_DEVFS() \
74	({ \
75	void *rc = NULL; \
76	if (devfs_mk_cdev(MKDEV(DEVICE_MAJOR, 0), \
77	S_IFCHR \| S_IRUGO \| S_IWUGO, \
78	DEVICE_NAME) == 0) \
79	rc = (void )' '; / return not NULL */ \
80	rc; \
81	})
82
83	# define VBOX_UNREGISTER_DEVFS(handle) \
84	devfs_remove(DEVICE_NAME);
85
86	# else /* < 2.6.0 */
87
88	# define VBOX_REGISTER_DEVFS() \
89	devfs_register(NULL, DEVICE_NAME, DEVFS_FL_DEFAULT, \
90	DEVICE_MAJOR, 0, \
91	S_IFCHR \| S_IRUGO \| S_IWUGO, \
92	&gFileOpsVBoxDrv, NULL)
93
94	# define VBOX_UNREGISTER_DEVFS(handle) \
95	if (handle != NULL) \
96	devfs_unregister(handle)
97
98	# endif /* < 2.6.0 */
99	#endif /* CONFIG_DEV_FS && !CONFIG_VBOXDEV_AS_MISC */
100
101	#ifndef CONFIG_VBOXDRV_AS_MISC
102	# if defined(CONFIG_DEVFS_FS) && LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 0)
103	# define VBOX_REGISTER_DEVICE(a,b,c) devfs_register_chrdev(a,b,c)
104	# define VBOX_UNREGISTER_DEVICE(a,b) devfs_unregister_chrdev(a,b)
105	# else
106	# define VBOX_REGISTER_DEVICE(a,b,c) register_chrdev(a,b,c)
107	# define VBOX_UNREGISTER_DEVICE(a,b) unregister_chrdev(a,b)
108	# endif
109	#endif /* !CONFIG_VBOXDRV_AS_MISC */
110
111
112	#ifdef CONFIG_X86_HIGH_ENTRY
113	# error "CONFIG_X86_HIGH_ENTRY is not supported by VBoxDrv at this time."
114	#endif
115
116	/*
117	* This sucks soooo badly on x86! Why don't they export __PAGE_KERNEL_EXEC so PAGE_KERNEL_EXEC would be usable?
118	*/
119	#if defined(__AMD64__)
120	# define MY_PAGE_KERNEL_EXEC PAGE_KERNEL_EXEC
121	#elif defined(PAGE_KERNEL_EXEC) && defined(CONFIG_X86_PAE)
122	# define MY_PAGE_KERNEL_EXEC __pgprot(cpu_has_pge ? _PAGE_KERNEL_EXEC \| _PAGE_GLOBAL : _PAGE_KERNEL_EXEC)
123	#else
124	# define MY_PAGE_KERNEL_EXEC PAGE_KERNEL
125	#endif
126
127	/*
128	* The redhat hack section.
129	* - The current hacks are for 2.4.21-15.EL only.
130	*/
131	#ifndef NO_REDHAT_HACKS
132	/* accounting. */
133	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0)
134	# ifdef VM_ACCOUNT
135	# define MY_DO_MUNMAP(a,b,c) do_munmap(a, b, c, 0) /* should it be 1 or 0? */
136	# endif
137	# endif
138
139	/* backported remap_page_range. */
140	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0)
141	# include <asm/tlb.h>
142	# ifdef tlb_vma /* probably not good enough... */
143	# define HAVE_26_STYLE_REMAP_PAGE_RANGE 1
144	# endif
145	# endif
146
147	# ifndef __AMD64__
148	/* In 2.6.9-22.ELsmp we have to call change_page_attr() twice when changing
149	* the page attributes from PAGE_KERNEL to something else, because there appears
150	* to be a bug in one of the many patches that redhat applied.
151	* It should be safe to do this on less buggy linux kernels too. ;-)
152	*/
153	# define MY_CHANGE_PAGE_ATTR(pPages, cPages, prot) \
154	do { \
155	if (pgprot_val(prot) != pgprot_val(PAGE_KERNEL)) \
156	change_page_attr(pPages, cPages, prot); \
157	change_page_attr(pPages, cPages, prot); \
158	} while (0)
159	# endif
160	#endif /* !NO_REDHAT_HACKS */
161
162
163	#ifndef MY_DO_MUNMAP
164	# define MY_DO_MUNMAP(a,b,c) do_munmap(a, b, c)
165	#endif
166
167	#ifndef MY_CHANGE_PAGE_ATTR
168	# ifdef __AMD64__ /** @todo This is a cheap hack, but it'll get around that 'else BUG();' in __change_page_attr(). */
169	# define MY_CHANGE_PAGE_ATTR(pPages, cPages, prot) \
170	do { \
171	change_page_attr(pPages, cPages, PAGE_KERNEL_NOCACHE); \
172	change_page_attr(pPages, cPages, prot); \
173	} while (0)
174	# else
175	# define MY_CHANGE_PAGE_ATTR(pPages, cPages, prot) change_page_attr(pPages, cPages, prot)
176	# endif
177	#endif
178
179
180	/** @def ONE_MSEC_IN_JIFFIES
181	* The number of jiffies that make up 1 millisecond. This is only actually used
182	* when HZ is > 1000. */
183	#if HZ <= 1000
184	# define ONE_MSEC_IN_JIFFIES 0
185	#elif !(HZ % 1000)
186	# define ONE_MSEC_IN_JIFFIES (HZ / 1000)
187	#else
188	# define ONE_MSEC_IN_JIFFIES ((HZ + 999) / 1000)
189	# error "HZ is not a multiple of 1000, the GIP stuff won't work right!"
190	#endif
191
192
193	/*******************************************************************************
194	* Defined Constants And Macros *
195	*******************************************************************************/
196	/**
197	* Device extention & session data association structure.
198	*/
199	static SUPDRVDEVEXT g_DevExt;
200
201	/** Timer structure for the GIP update. */
202	static struct timer_list g_GipTimer;
203	/** Pointer to the page structure for the GIP. */
204	struct page *g_pGipPage;
205
206	/** Registered devfs device handle. */
207	#if defined(CONFIG_DEVFS_FS) && !defined(CONFIG_VBOXDRV_AS_MISC)
208	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
209	static void *g_hDevFsVBoxDrv = NULL;
210	# else
211	static devfs_handle_t g_hDevFsVBoxDrv = NULL;
212	# endif
213	#endif
214
215	#ifndef CONFIG_VBOXDRV_AS_MISC
216	/** Module major number */
217	#define DEVICE_MAJOR 234
218	/** Saved major device number */
219	static int g_iModuleMajor;
220	#endif /* !CONFIG_VBOXDRV_AS_MISC */
221
222	/** The module name. */
223	#define DEVICE_NAME "vboxdrv"
224
225	#ifdef __AMD64__
226	/**
227	* Memory for the executable memory heap (in IPRT).
228	*/
229	extern uint8_t g_abExecMemory[1572864]; /* 1.5 MB */
230	__asm__(".section execmemory, \"awx\", @progbits\n\t"
231	".align 32\n\t"
232	".globl g_abExecMemory\n"
233	"g_abExecMemory:\n\t"
234	".zero 1572864\n\t"
235	".type g_abExecMemory, @object\n\t"
236	".size g_abExecMemory, 1572864\n\t"
237	".text\n\t");
238	#endif
239
240
241	/*******************************************************************************
242	* Internal Functions *
243	*******************************************************************************/
244	static int VBoxSupDrvInit(void);
245	static void VBoxSupDrvUnload(void);
246	static int VBoxSupDrvCreate(struct inode pInode, struct file pFilp);
247	static int VBoxSupDrvClose(struct inode pInode, struct file pFilp);
248	static int VBoxSupDrvDeviceControl(struct inode pInode, struct file pFilp,
249	unsigned int IOCmd, unsigned long IOArg);
250	static RTR3PTR VBoxSupDrvMapUser(struct page **papPages, unsigned cPages, unsigned fProt, pgprot_t pgFlags);
251	static int VBoxSupDrvInitGip(PSUPDRVDEVEXT pDevExt);
252	static int VBoxSupDrvTermGip(PSUPDRVDEVEXT pDevExt);
253	static void VBoxSupGipTimer(unsigned long ulUser);
254	#ifdef CONFIG_SMP
255	static void VBoxSupGipTimerPerCpu(unsigned long ulUser);
256	static void VBoxSupGipResumePerCpu(void *pvUser);
257	#endif
258	static int VBoxSupDrvOrder(unsigned long size);
259	static int VBoxSupDrvErr2LinuxErr(int);
260
261
262	/** The file_operations structure. */
263	static struct file_operations gFileOpsVBoxDrv =
264	{
265	owner: THIS_MODULE,
266	open: VBoxSupDrvCreate,
267	release: VBoxSupDrvClose,
268	ioctl: VBoxSupDrvDeviceControl,
269	};
270
271	#ifdef CONFIG_VBOXDRV_AS_MISC
272	/** The miscdevice structure. */
273	static struct miscdevice gMiscDevice =
274	{
275	minor: MISC_DYNAMIC_MINOR,
276	name: DEVICE_NAME,
277	fops: &gFileOpsVBoxDrv,
278	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) && \
279	LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 17)
280	devfs_name: DEVICE_NAME,
281	# endif
282	};
283	#endif
284
285
286	/**
287	* Initialize module.
288	*
289	* @returns appropriate status code.
290	*/
291	static int __init VBoxSupDrvInit(void)
292	{
293	int rc;
294
295	dprintf(("VBoxDrv::ModuleInit\n"));
296
297	#ifdef CONFIG_X86_LOCAL_APIC
298	/*
299	* If an NMI occurs while we are inside the world switcher the macine will crash.
300	* The Linux NMI watchdog generates periodic NMIs increasing a counter which is
301	* compared with another counter increased in the timer interrupt handler. Therefore
302	* we don't allow to setup an NMI watchdog.
303	*/
304	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
305	/*
306	* First test: NMI actiated? Works only works with Linux 2.6 -- 2.4 does not export
307	* the nmi_watchdog variable.
308	*/
309	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
310	/*
311	* Permanent IO_APIC mode active? No way to handle this!
312	*/
313	if (nmi_watchdog == NMI_IO_APIC)
314	{
315	printk(KERN_ERR DEVICE_NAME
316	": NMI watchdog in IO_APIC mode active -- refused to load the kernel module!\n"
317	DEVICE_NAME
318	": Please disable the NMI watchdog by specifying 'nmi_watchdog=0' at kernel\n"
319	DEVICE_NAME
320	": command line.\n");
321	return -EINVAL;
322	}
323
324	/*
325	* See arch/i386/kernel/nmi.c on >= 2.6.19: -1 means it can never enabled again
326	*/
327	atomic_set(&nmi_active, -1);
328	printk(KERN_INFO DEVICE_NAME ": Trying to deactivate NMI watchdog permanently...\n");
329
330	/*
331	* Now fall through and see if it actually was enabled before. If so, fail
332	* as we cannot deactivate it cleanly from here.
333	*/
334	# else /* < 2.6.19 */
335	/*
336	* Older 2.6 kernels: nmi_watchdog is not initalized by default
337	*/
338	if (nmi_watchdog != NMI_NONE)
339	goto nmi_activated;
340	# endif
341	# endif /* >= 2.6.0 */
342
343	/*
344	* Second test: Interrupt generated by performance counter not masked and can
345	* generate an NMI. Works also with Linux 2.4.
346	*/
347	{
348	unsigned int v, ver, maxlvt;
349
350	v = apic_read(APIC_LVR);
351	ver = GET_APIC_VERSION(v);
352	/* 82489DXs do not report # of LVT entries. */
353	maxlvt = APIC_INTEGRATED(ver) ? GET_APIC_MAXLVT(v) : 2;
354	if (maxlvt >= 4)
355	{
356	/* Read status of performance counter IRQ vector */
357	v = apic_read(APIC_LVTPC);
358
359	/* performance counter generates NMI and is not masked? */
360	if ((GET_APIC_DELIVERY_MODE(v) == APIC_MODE_NMI) && !(v & APIC_LVT_MASKED))
361	{
362	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
363	printk(KERN_ERR DEVICE_NAME
364	": NMI watchdog either active or at least initialized. Please disable the NMI\n"
365	DEVICE_NAME
366	": watchdog by specifying 'nmi_watchdog=0' at kernel command line.\n");
367	return -EINVAL;
368	# else /* < 2.6.19 */
369	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
370	nmi_activated:
371	# endif
372	printk(KERN_ERR DEVICE_NAME
373	": NMI watchdog active -- refused to load the kernel module! Please disable\n"
374	DEVICE_NAME
375	": the NMI watchdog by specifying 'nmi_watchdog=0' at kernel command line.\n");
376	return -EINVAL;
377	# endif /* >= 2.6.19 */
378	}
379	}
380	}
381	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
382	printk(KERN_INFO DEVICE_NAME ": Successfully done.\n");
383	# endif /* >= 2.6.19 */
384	#endif /* CONFIG_X86_LOCAL_APIC */
385
386	#ifdef CONFIG_VBOXDRV_AS_MISC
387	rc = misc_register(&gMiscDevice);
388	if (rc)
389	{
390	printk(KERN_ERR DEVICE_NAME ": Can't register misc device! rc=%d\n", rc);
391	return rc;
392	}
393	#else /* !CONFIG_VBOXDRV_AS_MISC */
394	/*
395	* Register character device.
396	*/
397	g_iModuleMajor = DEVICE_MAJOR;
398	rc = VBOX_REGISTER_DEVICE((dev_t)g_iModuleMajor, DEVICE_NAME, &gFileOpsVBoxDrv);
399	if (rc < 0)
400	{
401	dprintf(("VBOX_REGISTER_DEVICE failed with rc=%#x!\n", rc));
402	return rc;
403	}
404
405	/*
406	* Save returned module major number
407	*/
408	if (DEVICE_MAJOR != 0)
409	g_iModuleMajor = DEVICE_MAJOR;
410	else
411	g_iModuleMajor = rc;
412	rc = 0;
413
414	#ifdef CONFIG_DEVFS_FS
415	/*
416	* Register a device entry
417	*/
418	g_hDevFsVBoxDrv = VBOX_REGISTER_DEVFS();
419	if (g_hDevFsVBoxDrv == NULL)
420	{
421	dprintf(("devfs_register failed!\n"));
422	rc = -EINVAL;
423	}
424	#endif
425	#endif /* !CONFIG_VBOXDRV_AS_MISC */
426	if (!rc)
427	{
428	/*
429	* Initialize the runtime.
430	* On AMD64 we'll have to donate the high rwx memory block to the exec allocator.
431	*/
432	rc = RTR0Init(0);
433	if (RT_SUCCESS(rc))
434	{
435	#ifdef __AMD64__
436	rc = RTR0MemExecDonate(&g_abExecMemory[0], sizeof(g_abExecMemory));
437	#endif
438	/*
439	* Initialize the device extension.
440	*/
441	if (RT_SUCCESS(rc))
442	rc = supdrvInitDevExt(&g_DevExt);
443	if (!rc)
444	{
445	/*
446	* Create the GIP page.
447	*/
448	rc = VBoxSupDrvInitGip(&g_DevExt);
449	if (!rc)
450	{
451	dprintf(("VBoxDrv::ModuleInit returning %#x\n", rc));
452	return rc;
453	}
454
455	supdrvDeleteDevExt(&g_DevExt);
456	}
457	else
458	rc = -EINVAL;
459	RTR0Term();
460	}
461	else
462	rc = -EINVAL;
463
464	/*
465	* Failed, cleanup and return the error code.
466	*/
467	#if defined(CONFIG_DEVFS_FS) && !defined(CONFIG_VBOXDRV_AS_MISC)
468	VBOX_UNREGISTER_DEVFS(g_hDevFsVBoxDrv);
469	#endif
470	}
471	#ifdef CONFIG_VBOXDRV_AS_MISC
472	misc_deregister(&gMiscDevice);
473	dprintf(("VBoxDrv::ModuleInit returning %#x (minor:%d)\n", rc, gMiscDevice.minor));
474	#else
475	VBOX_UNREGISTER_DEVICE(g_iModuleMajor, DEVICE_NAME);
476	dprintf(("VBoxDrv::ModuleInit returning %#x (major:%d)\n", rc, g_iModuleMajor));
477	#endif
478	return rc;
479	}
480
481
482	/**
483	* Unload the module.
484	*/
485	static void __exit VBoxSupDrvUnload(void)
486	{
487	int rc;
488	dprintf(("VBoxSupDrvUnload\n"));
489
490	/*
491	* I Don't think it's possible to unload a driver which processes have
492	* opened, at least we'll blindly assume that here.
493	*/
494	#ifdef CONFIG_VBOXDRV_AS_MISC
495	rc = misc_deregister(&gMiscDevice);
496	if (rc < 0)
497	{
498	dprintf(("misc_deregister failed with rc=%#x\n", rc));
499	}
500	#else /* !CONFIG_VBOXDRV_AS_MISC */
501	#ifdef CONFIG_DEVFS_FS
502	/*
503	* Unregister a device entry
504	*/
505	VBOX_UNREGISTER_DEVFS(g_hDevFsVBoxDrv);
506	#endif // devfs
507	rc = VBOX_UNREGISTER_DEVICE(g_iModuleMajor, DEVICE_NAME);
508	if (rc < 0)
509	{
510	dprintf(("unregister_chrdev failed with rc=%#x (major:%d)\n", rc, g_iModuleMajor));
511	}
512	#endif /* !CONFIG_VBOXDRV_AS_MISC */
513
514	/*
515	* Destroy GIP, delete the device extension and terminate IPRT.
516	*/
517	VBoxSupDrvTermGip(&g_DevExt);
518	supdrvDeleteDevExt(&g_DevExt);
519	RTR0Term();
520	}
521
522
523	/**
524	* Device open. Called on open /dev/vboxdrv
525	*
526	* @param pInode Pointer to inode info structure.
527	* @param pFilp Associated file pointer.
528	*/
529	static int VBoxSupDrvCreate(struct inode pInode, struct file pFilp)
530	{
531	int rc;
532	PSUPDRVSESSION pSession;
533	dprintf(("VBoxSupDrvCreate: pFilp=%p\n", pFilp));
534
535	/*
536	* Call common code for the rest.
537	*/
538	rc = supdrvCreateSession(&g_DevExt, (PSUPDRVSESSION *)&pSession);
539	if (!rc)
540	{
541	pSession->Uid = current->euid;
542	pSession->Gid = current->egid;
543	pSession->Process = RTProcSelf();
544	pSession->R0Process = RTR0ProcHandleSelf();
545	}
546
547	dprintf(("VBoxSupDrvCreate: g_DevExt=%p pSession=%p rc=%d\n", &g_DevExt, pSession, rc));
548	pFilp->private_data = pSession;
549
550	return VBoxSupDrvErr2LinuxErr(rc);
551	}
552
553
554	/**
555	* Close device.
556	*
557	* @param pInode Pointer to inode info structure.
558	* @param pFilp Associated file pointer.
559	*/
560	static int VBoxSupDrvClose(struct inode pInode, struct file pFilp)
561	{
562	dprintf(("VBoxSupDrvClose: pFilp=%p private_data=%p\n", pFilp, pFilp->private_data));
563	supdrvCloseSession(&g_DevExt, (PSUPDRVSESSION)pFilp->private_data);
564	pFilp->private_data = NULL;
565	return 0;
566	}
567
568
569	/**
570	* Device I/O Control entry point.
571	*
572	* @param pInode Pointer to inode info structure.
573	* @param pFilp Associated file pointer.
574	* @param IOCmd The function specified to ioctl().
575	* @param IOArg The argument specified to ioctl().
576	*/
577	static int VBoxSupDrvDeviceControl(struct inode pInode, struct file pFilp,
578	unsigned int IOCmd, unsigned long IOArg)
579	{
580	int rc;
581	SUPDRVIOCTLDATA Args;
582	void *pvBuf = NULL;
583	int cbBuf = 0;
584	unsigned cbOut = 0;
585
586	dprintf2(("VBoxSupDrvDeviceControl: pFilp=%p IOCmd=%x IOArg=%p\n", pFilp, IOCmd, (void *)IOArg));
587
588	/*
589	* Copy ioctl data structure from user space.
590	*/
591	if (_IOC_SIZE(IOCmd) != sizeof(SUPDRVIOCTLDATA))
592	{
593	dprintf(("VBoxSupDrvDeviceControl: incorrect input length! cbArgs=%d\n", _IOC_SIZE(IOCmd)));
594	return -EINVAL;
595	}
596	if (copy_from_user(&Args, (void *)IOArg, _IOC_SIZE(IOCmd)))
597	{
598	dprintf(("VBoxSupDrvDeviceControl: copy_from_user(&Args) failed.\n"));
599	return -EFAULT;
600	}
601
602	/*
603	* Allocate and copy user space input data buffer to kernel space.
604	*/
605	if (Args.cbIn > 0 \|\| Args.cbOut > 0)
606	{
607	cbBuf = max(Args.cbIn, Args.cbOut);
608	pvBuf = vmalloc(cbBuf);
609	if (pvBuf == NULL)
610	{
611	dprintf(("VBoxSupDrvDeviceControl: failed to allocate buffer of %d bytes.\n", cbBuf));
612	return -ENOMEM;
613	}
614
615	if (copy_from_user(pvBuf, (void *)Args.pvIn, Args.cbIn))
616	{
617	dprintf(("VBoxSupDrvDeviceControl: copy_from_user(pvBuf) failed.\n"));
618	vfree(pvBuf);
619	return -EFAULT;
620	}
621	}
622
623	/*
624	* Process the IOCtl.
625	*/
626	rc = supdrvIOCtl(IOCmd, &g_DevExt, (PSUPDRVSESSION)pFilp->private_data,
627	pvBuf, Args.cbIn, pvBuf, Args.cbOut, &cbOut);
628
629	/*
630	* Copy ioctl data and output buffer back to user space.
631	*/
632	if (rc)
633	{
634	dprintf(("VBoxSupDrvDeviceControl: pFilp=%p IOCmd=%x IOArg=%p failed, rc=%d (linux rc=%d)\n",
635	pFilp, IOCmd, (void *)IOArg, rc, VBoxSupDrvErr2LinuxErr(rc)));
636	rc = VBoxSupDrvErr2LinuxErr(rc);
637	}
638	else if (cbOut > 0)
639	{
640	if (pvBuf != NULL && cbOut <= cbBuf)
641	{
642	if (copy_to_user((void *)Args.pvOut, pvBuf, cbOut))
643	{
644	dprintf(("copy_to_user failed.\n"));
645	rc = -EFAULT;
646	}
647	}
648	else
649	{
650	dprintf(("WHAT!?! supdrvIOCtl messed up! cbOut=%d cbBuf=%d pvBuf=%p\n", cbOut, cbBuf, pvBuf));
651	rc = -EPERM;
652	}
653	}
654
655	if (pvBuf)
656	vfree(pvBuf);
657
658	dprintf2(("VBoxSupDrvDeviceControl: returns %d\n", rc));
659	return rc;
660	}
661
662
663	/**
664	* Initializes any OS specific object creator fields.
665	*/
666	void VBOXCALL supdrvOSObjInitCreator(PSUPDRVOBJ pObj, PSUPDRVSESSION pSession)
667	{
668	NOREF(pObj);
669	NOREF(pSession);
670	}
671
672
673	/**
674	* Checks if the session can access the object.
675	*
676	* @returns true if a decision has been made.
677	* @returns false if the default access policy should be applied.
678	*
679	* @param pObj The object in question.
680	* @param pSession The session wanting to access the object.
681	* @param pszObjName The object name, can be NULL.
682	* @param prc Where to store the result when returning true.
683	*/
684	bool VBOXCALL supdrvOSObjCanAccess(PSUPDRVOBJ pObj, PSUPDRVSESSION pSession, const char pszObjName, int prc)
685	{
686	NOREF(pObj);
687	NOREF(pSession);
688	NOREF(pszObjName);
689	NOREF(prc);
690	return false;
691	}
692
693
694	/**
695	* Compute order. Some functions allocate 2^order pages.
696	*
697	* @returns order.
698	* @param cPages Number of pages.
699	*/
700	static int VBoxSupDrvOrder(unsigned long cPages)
701	{
702	int iOrder;
703	unsigned long cTmp;
704
705	for (iOrder = 0, cTmp = cPages; cTmp >>= 1; ++iOrder)
706	;
707	if (cPages & ~(1 << iOrder))
708	++iOrder;
709
710	return iOrder;
711	}
712
713
714	/**
715	* OS Specific code for locking down memory.
716	*
717	* @returns 0 on success.
718	* @returns SUPDRV_ERR_* on failure.
719	* @param pMem Pointer to memory.
720	* This is not linked in anywhere.
721	* @param paPages Array which should be filled with the address of the physical pages.
722	*
723	* @remark See sgl_map_user_pages() for an example of an similar function.
724	*/
725	int VBOXCALL supdrvOSLockMemOne(PSUPDRVMEMREF pMem, PSUPPAGE paPages)
726	{
727	int rc;
728	struct page **papPages;
729	unsigned iPage;
730	unsigned cPages = pMem->cb >> PAGE_SHIFT;
731	unsigned long pv = (unsigned long)pMem->pvR3;
732	struct vm_area_struct **papVMAs;
733
734	/*
735	* Allocate page pointer array.
736	*/
737	papPages = vmalloc(cPages * sizeof(*papPages));
738	if (!papPages)
739	return SUPDRV_ERR_NO_MEMORY;
740
741	/*
742	* Allocate the VMA pointer array.
743	*/
744	papVMAs = vmalloc(cPages * sizeof(*papVMAs));
745	if (!papVMAs)
746	return SUPDRV_ERR_NO_MEMORY;
747
748	/*
749	* Get user pages.
750	*/
751	down_read(&current->mm->mmap_sem);
752	rc = get_user_pages(current, /* Task for fault acounting. */
753	current->mm, /* Whose pages. */
754	(unsigned long)pv, /* Where from. */
755	cPages, /* How many pages. */
756	1, /* Write to memory. */
757	0, /* force. */
758	papPages, /* Page array. */
759	papVMAs); /* vmas */
760	if (rc != cPages)
761	{
762	up_read(&current->mm->mmap_sem);
763	dprintf(("supdrvOSLockMemOne: get_user_pages failed. rc=%d\n", rc));
764	return SUPDRV_ERR_LOCK_FAILED;
765	}
766
767	for (iPage = 0; iPage < cPages; iPage++)
768	flush_dcache_page(papPages[iPage]);
769	up_read(&current->mm->mmap_sem);
770
771	pMem->u.locked.papPages = papPages;
772	pMem->u.locked.cPages = cPages;
773
774	/*
775	* Get addresses, protect against fork()
776	*/
777	for (iPage = 0; iPage < cPages; iPage++)
778	{
779	paPages[iPage].Phys = page_to_phys(papPages[iPage]);
780	paPages[iPage].uReserved = 0;
781	papVMAs[iPage]->vm_flags \|= VM_DONTCOPY;
782	}
783
784	vfree(papVMAs);
785
786	dprintf2(("supdrvOSLockMemOne: pvR3=%p cb=%d papPages=%p\n",
787	pMem->pvR3, pMem->cb, pMem->u.locked.papPages));
788	return 0;
789	}
790
791
792	/**
793	* Unlocks the memory pointed to by pv.
794	*
795	* @param pv Memory to unlock.
796	* @param cb Size of the memory (debug).
797	*
798	* @remark See sgl_unmap_user_pages() for an example of an similar function.
799	*/
800	void VBOXCALL supdrvOSUnlockMemOne(PSUPDRVMEMREF pMem)
801	{
802	unsigned iPage;
803	dprintf2(("supdrvOSUnlockMemOne: pvR3=%p cb=%d papPages=%p\n",
804	pMem->pvR3, pMem->cb, pMem->u.locked.papPages));
805
806	/*
807	* Loop thru the pages and release them.
808	*/
809	for (iPage = 0; iPage < pMem->u.locked.cPages; iPage++)
810	{
811	if (!PageReserved(pMem->u.locked.papPages[iPage]))
812	SetPageDirty(pMem->u.locked.papPages[iPage]);
813	page_cache_release(pMem->u.locked.papPages[iPage]);
814	}
815
816	/* free the page array */
817	vfree(pMem->u.locked.papPages);
818	pMem->u.locked.cPages = 0;
819	}
820
821
822	/**
823	* OS Specific code for allocating page aligned memory with continuous fixed
824	* physical paged backing.
825	*
826	* @returns 0 on success.
827	* @returns SUPDRV_ERR_* on failure.
828	* @param pMem Memory reference record of the memory to be allocated.
829	* (This is not linked in anywhere.)
830	* @param ppvR0 Where to store the virtual address of the ring-0 mapping. (optional)
831	* @param ppvR3 Where to store the virtual address of the ring-3 mapping.
832	* @param pHCPhys Where to store the physical address.
833	*/
834	int VBOXCALL supdrvOSContAllocOne(PSUPDRVMEMREF pMem, PRTR0PTR ppvR0, PRTR3PTR ppvR3, PRTHCPHYS pHCPhys)
835	{
836	struct page *paPages;
837	unsigned iPage;
838	unsigned cbAligned = RT_ALIGN(pMem->cb, PAGE_SIZE);
839	unsigned cPages = cbAligned >> PAGE_SHIFT;
840	unsigned cOrder = VBoxSupDrvOrder(cPages);
841	unsigned long ulAddr;
842	dma_addr_t HCPhys;
843	int rc = 0;
844	pgprot_t pgFlags;
845	pgprot_val(pgFlags) = _PAGE_PRESENT \| _PAGE_RW \| _PAGE_USER;
846
847	Assert(ppvR3);
848	Assert(pHCPhys);
849
850	/*
851	* Allocate page pointer array.
852	*/
853	#ifdef __AMD64__ /** @todo check out if there is a correct way of getting memory below 4GB (physically). */
854	paPages = alloc_pages(GFP_DMA, cOrder);
855	#else
856	paPages = alloc_pages(GFP_USER, cOrder);
857	#endif
858	if (!paPages)
859	return SUPDRV_ERR_NO_MEMORY;
860
861	/*
862	* Lock the pages.
863	*/
864	for (iPage = 0; iPage < cPages; iPage++)
865	{
866	SetPageReserved(&paPages[iPage]);
867	if (!PageHighMem(&paPages[iPage]) && pgprot_val(MY_PAGE_KERNEL_EXEC) != pgprot_val(PAGE_KERNEL))
868	MY_CHANGE_PAGE_ATTR(&paPages[iPage], 1, MY_PAGE_KERNEL_EXEC);
869	#ifdef DEBUG
870	if (iPage + 1 < cPages && (page_to_phys((&paPages[iPage])) + 0x1000) != page_to_phys((&paPages[iPage + 1])))
871	{
872	dprintf(("supdrvOSContAllocOne: Pages are not continuous!!!! iPage=%d phys=%llx physnext=%llx\n",
873	iPage, (long long)page_to_phys((&paPages[iPage])), (long long)page_to_phys((&paPages[iPage + 1]))));
874	BUG();
875	}
876	#endif
877	}
878	HCPhys = page_to_phys(paPages);
879
880	/*
881	* Allocate user space mapping and put the physical pages into it.
882	*/
883	down_write(&current->mm->mmap_sem);
884	ulAddr = do_mmap(NULL, 0, cbAligned, PROT_READ \| PROT_WRITE \| PROT_EXEC, MAP_SHARED \| MAP_ANONYMOUS, 0);
885	if (!(ulAddr & ~PAGE_MASK))
886	{
887	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0) && !defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
888	int rc2 = remap_page_range(ulAddr, HCPhys, cbAligned, pgFlags);
889	#else
890	int rc2 = 0;
891	struct vm_area_struct *vma = find_vma(current->mm, ulAddr);
892	if (vma)
893	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11)
894	rc2 = remap_page_range(vma, ulAddr, HCPhys, cbAligned, pgFlags);
895	#else
896	rc2 = remap_pfn_range(vma, ulAddr, HCPhys >> PAGE_SHIFT, cbAligned, pgFlags);
897	#endif
898	else
899	{
900	rc = SUPDRV_ERR_NO_MEMORY;
901	dprintf(("supdrvOSContAllocOne: no vma found for ulAddr=%#lx!\n", ulAddr));
902	}
903	#endif
904	if (rc2)
905	{
906	rc = SUPDRV_ERR_NO_MEMORY;
907	dprintf(("supdrvOSContAllocOne: remap_page_range failed rc2=%d\n", rc2));
908	}
909	}
910	else
911	{
912	dprintf(("supdrvOSContAllocOne: do_mmap failed ulAddr=%#lx\n", ulAddr));
913	rc = SUPDRV_ERR_NO_MEMORY;
914	}
915	up_write(&current->mm->mmap_sem); /* not quite sure when to give this up. */
916
917	/*
918	* Success?
919	*/
920	if (!rc)
921	{
922	*pHCPhys = HCPhys;
923	*ppvR3 = ulAddr;
924	if (ppvR0)
925	ppvR0 = (void )ulAddr;
926	pMem->pvR3 = ulAddr;
927	pMem->pvR0 = NULL;
928	pMem->u.cont.paPages = paPages;
929	pMem->u.cont.cPages = cPages;
930	pMem->cb = cbAligned;
931
932	dprintf2(("supdrvOSContAllocOne: pvR0=%p pvR3=%p cb=%d paPages=%p pHCPhys=%lx ppvR0=*ppvR3=%p\n",
933	pMem->pvR0, pMem->pvR3, pMem->cb, paPages, (unsigned long)pHCPhys, ppvR3));
934	global_flush_tlb();
935	return 0;
936	}
937
938	/*
939	* Failure, cleanup and be gone.
940	*/
941	down_write(&current->mm->mmap_sem);
942	if (ulAddr & ~PAGE_MASK)
943	MY_DO_MUNMAP(current->mm, ulAddr, pMem->cb);
944	for (iPage = 0; iPage < cPages; iPage++)
945	{
946	ClearPageReserved(&paPages[iPage]);
947	if (!PageHighMem(&paPages[iPage]) && pgprot_val(MY_PAGE_KERNEL_EXEC) != pgprot_val(PAGE_KERNEL))
948	MY_CHANGE_PAGE_ATTR(&paPages[iPage], 1, PAGE_KERNEL);
949	}
950	up_write(&current->mm->mmap_sem); /* check when we can leave this. */
951	__free_pages(paPages, cOrder);
952
953	global_flush_tlb();
954	return rc;
955	}
956
957
958	/**
959	* Frees contiguous memory.
960	*
961	* @param pMem Memory reference record of the memory to be freed.
962	*/
963	void VBOXCALL supdrvOSContFreeOne(PSUPDRVMEMREF pMem)
964	{
965	unsigned iPage;
966
967	dprintf2(("supdrvOSContFreeOne: pvR0=%p pvR3=%p cb=%d paPages=%p\n",
968	pMem->pvR0, pMem->pvR3, pMem->cb, pMem->u.cont.paPages));
969
970	/*
971	* do_exit() destroys the mm before closing files.
972	* I really hope it cleans up our stuff properly...
973	*/
974	if (current->mm)
975	{
976	down_write(&current->mm->mmap_sem);
977	MY_DO_MUNMAP(current->mm, (unsigned long)pMem->pvR3, pMem->cb);
978	up_write(&current->mm->mmap_sem); /* check when we can leave this. */
979	}
980
981	/*
982	* Change page attributes freeing the pages.
983	*/
984	for (iPage = 0; iPage < pMem->u.cont.cPages; iPage++)
985	{
986	ClearPageReserved(&pMem->u.cont.paPages[iPage]);
987	if (!PageHighMem(&pMem->u.cont.paPages[iPage]) && pgprot_val(MY_PAGE_KERNEL_EXEC) != pgprot_val(PAGE_KERNEL))
988	MY_CHANGE_PAGE_ATTR(&pMem->u.cont.paPages[iPage], 1, PAGE_KERNEL);
989	}
990	__free_pages(pMem->u.cont.paPages, VBoxSupDrvOrder(pMem->u.cont.cPages));
991
992	pMem->u.cont.cPages = 0;
993	}
994
995
996	/**
997	* Allocates memory which mapped into both kernel and user space.
998	* The returned memory is page aligned and so is the allocation.
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 address of the Ring-0 mapping.
1005	* @param ppvR3 Where to store the address of the Ring-3 mapping.
1006	*/
1007	int VBOXCALL supdrvOSMemAllocOne(PSUPDRVMEMREF pMem, PRTR0PTR ppvR0, PRTR3PTR ppvR3)
1008	{
1009	const unsigned cbAligned = RT_ALIGN(pMem->cb, PAGE_SIZE);
1010	const unsigned cPages = cbAligned >> PAGE_SHIFT;
1011	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 4, 22)
1012	unsigned cOrder = VBoxSupDrvOrder(cPages);
1013	struct page *paPages;
1014	#endif
1015	struct page **papPages;
1016	unsigned iPage;
1017	pgprot_t pgFlags;
1018	pgprot_val(pgFlags) = _PAGE_PRESENT \| _PAGE_RW \| _PAGE_USER;
1019
1020	/*
1021	* Allocate array with page pointers.
1022	*/
1023	pMem->u.mem.cPages = 0;
1024	pMem->u.mem.papPages = papPages = kmalloc(sizeof(papPages[0]) * cPages, GFP_KERNEL);
1025	if (!papPages)
1026	return SUPDRV_ERR_NO_MEMORY;
1027
1028	/*
1029	* Allocate the pages.
1030	*/
1031	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1032	for (iPage = 0; iPage < cPages; iPage++)
1033	{
1034	papPages[iPage] = alloc_page(GFP_HIGHUSER);
1035	if (!papPages[iPage])
1036	{
1037	pMem->u.mem.cPages = iPage;
1038	supdrvOSMemFreeOne(pMem);
1039	return SUPDRV_ERR_NO_MEMORY;
1040	}
1041	}
1042
1043	#else /* < 2.4.22 */
1044	paPages = alloc_pages(GFP_USER, cOrder);
1045	if (!paPages)
1046	{
1047	supdrvOSMemFreeOne(pMem);
1048	return SUPDRV_ERR_NO_MEMORY;
1049	}
1050	for (iPage = 0; iPage < cPages; iPage++)
1051	{
1052	papPages[iPage] = &paPages[iPage];
1053	if (pgprot_val(MY_PAGE_KERNEL_EXEC) != pgprot_val(PAGE_KERNEL))
1054	MY_CHANGE_PAGE_ATTR(papPages[iPage], 1, MY_PAGE_KERNEL_EXEC);
1055	if (PageHighMem(papPages[iPage]))
1056	BUG();
1057	}
1058	#endif
1059	pMem->u.mem.cPages = cPages;
1060
1061	/*
1062	* Reserve the pages.
1063	*/
1064	for (iPage = 0; iPage < cPages; iPage++)
1065	SetPageReserved(papPages[iPage]);
1066
1067	/*
1068	* Create the Ring-0 mapping.
1069	*/
1070	if (ppvR0)
1071	{
1072	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1073	# ifdef VM_MAP
1074	*ppvR0 = pMem->pvR0 = vmap(papPages, cPages, VM_MAP, pgFlags);
1075	# else
1076	*ppvR0 = pMem->pvR0 = vmap(papPages, cPages, VM_ALLOC, pgFlags);
1077	# endif
1078	#else
1079	*ppvR0 = pMem->pvR0 = phys_to_virt(page_to_phys(papPages[0]));
1080	#endif
1081	}
1082	if (pMem->pvR0 \|\| !ppvR0)
1083	{
1084	/*
1085	* Create the ring3 mapping.
1086	*/
1087	if (ppvR3)
1088	*ppvR3 = pMem->pvR3 = VBoxSupDrvMapUser(papPages, cPages, PROT_READ \| PROT_WRITE \| PROT_EXEC, pgFlags);
1089	if (pMem->pvR3 \|\| !ppvR3)
1090	return 0;
1091	dprintf(("supdrvOSMemAllocOne: failed to map into r3! cPages=%u\n", cPages));
1092	}
1093	else
1094	dprintf(("supdrvOSMemAllocOne: failed to map into r0! cPages=%u\n", cPages));
1095
1096	supdrvOSMemFreeOne(pMem);
1097	return SUPDRV_ERR_NO_MEMORY;
1098	}
1099
1100
1101	/**
1102	* Get the physical addresses of the pages in the allocation.
1103	* This is called while inside bundle the spinlock.
1104	*
1105	* @param pMem Memory reference record of the memory.
1106	* @param paPages Where to store the page addresses.
1107	*/
1108	void VBOXCALL supdrvOSMemGetPages(PSUPDRVMEMREF pMem, PSUPPAGE paPages)
1109	{
1110	unsigned iPage;
1111	for (iPage = 0; iPage < pMem->u.mem.cPages; iPage++)
1112	{
1113	paPages[iPage].Phys = page_to_phys(pMem->u.mem.papPages[iPage]);
1114	paPages[iPage].uReserved = 0;
1115	}
1116	}
1117
1118
1119	/**
1120	* Frees memory allocated by supdrvOSMemAllocOne().
1121	*
1122	* @param pMem Memory reference record of the memory to be free.
1123	*/
1124	void VBOXCALL supdrvOSMemFreeOne(PSUPDRVMEMREF pMem)
1125	{
1126	dprintf2(("supdrvOSMemFreeOne: pvR0=%p pvR3=%p cb=%d cPages=%d papPages=%p\n",
1127	pMem->pvR0, pMem->pvR3, pMem->cb, pMem->u.mem.cPages, pMem->u.mem.papPages));
1128
1129	/*
1130	* Unmap the user mapping (if any).
1131	* do_exit() destroys the mm before closing files.
1132	*/
1133	if (pMem->pvR3 && current->mm)
1134	{
1135	down_write(&current->mm->mmap_sem);
1136	MY_DO_MUNMAP(current->mm, (unsigned long)pMem->pvR3, RT_ALIGN(pMem->cb, PAGE_SIZE));
1137	up_write(&current->mm->mmap_sem); /* check when we can leave this. */
1138	}
1139	pMem->pvR3 = NIL_RTR3PTR;
1140
1141	/*
1142	* Unmap the kernel mapping (if any).
1143	*/
1144	if (pMem->pvR0)
1145	{
1146	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1147	vunmap(pMem->pvR0);
1148	#endif
1149	pMem->pvR0 = NULL;
1150	}
1151
1152	/*
1153	* Free the physical pages.
1154	*/
1155	if (pMem->u.mem.papPages)
1156	{
1157	struct page **papPages = pMem->u.mem.papPages;
1158	const unsigned cPages = pMem->u.mem.cPages;
1159	unsigned iPage;
1160
1161	/* Restore the page flags. */
1162	for (iPage = 0; iPage < cPages; iPage++)
1163	{
1164	ClearPageReserved(papPages[iPage]);
1165	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 4, 22)
1166	if (pgprot_val(MY_PAGE_KERNEL_EXEC) != pgprot_val(PAGE_KERNEL))
1167	MY_CHANGE_PAGE_ATTR(papPages[iPage], 1, PAGE_KERNEL);
1168	#endif
1169	}
1170
1171	/* Free the pages. */
1172	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1173	for (iPage = 0; iPage < pMem->u.cont.cPages; iPage++)
1174	__free_page(papPages[iPage]);
1175	#else
1176	if (cPages > 0)
1177	__free_pages(papPages[0], VBoxSupDrvOrder(cPages));
1178	#endif
1179	/* Free the page pointer array. */
1180	kfree(papPages);
1181	pMem->u.mem.papPages = NULL;
1182	}
1183	pMem->u.mem.cPages = 0;
1184	}
1185
1186
1187	/**
1188	* Maps a range of pages into user space.
1189	*
1190	* @returns Pointer to the user space mapping on success.
1191	* @returns NULL on failure.
1192	* @param papPages Array of the pages to map.
1193	* @param cPages Number of pages to map.
1194	* @param fProt The mapping protection.
1195	* @param pgFlags The page level protection.
1196	*/
1197	static RTR3PTR VBoxSupDrvMapUser(struct page **papPages, unsigned cPages, unsigned fProt, pgprot_t pgFlags)
1198	{
1199	int rc = SUPDRV_ERR_NO_MEMORY;
1200	unsigned long ulAddr;
1201
1202	/*
1203	* Allocate user space mapping.
1204	*/
1205	down_write(&current->mm->mmap_sem);
1206	ulAddr = do_mmap(NULL, 0, cPages * PAGE_SIZE, fProt, MAP_SHARED \| MAP_ANONYMOUS, 0);
1207	if (!(ulAddr & ~PAGE_MASK))
1208	{
1209	/*
1210	* Map page by page into the mmap area.
1211	* This is generic, paranoid and not very efficient.
1212	*/
1213	int rc = 0;
1214	unsigned long ulAddrCur = ulAddr;
1215	unsigned iPage;
1216	for (iPage = 0; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE)
1217	{
1218	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1219	struct vm_area_struct *vma = find_vma(current->mm, ulAddrCur);
1220	if (!vma)
1221	break;
1222	#endif
1223
1224	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
1225	rc = remap_pfn_range(vma, ulAddrCur, page_to_pfn(papPages[iPage]), PAGE_SIZE, pgFlags);
1226	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1227	rc = remap_page_range(vma, ulAddrCur, page_to_phys(papPages[iPage]), PAGE_SIZE, pgFlags);
1228	#else /* 2.4 */
1229	rc = remap_page_range(ulAddrCur, page_to_phys(papPages[iPage]), PAGE_SIZE, pgFlags);
1230	#endif
1231	if (rc)
1232	break;
1233	}
1234
1235	/*
1236	* Successful?
1237	*/
1238	if (iPage >= cPages)
1239	{
1240	up_write(&current->mm->mmap_sem);
1241	return ulAddr;
1242	}
1243
1244	/* no, cleanup! */
1245	if (rc)
1246	dprintf(("VBoxSupDrvMapUser: remap_[page\|pfn]_range failed! rc=%d\n", rc));
1247	else
1248	dprintf(("VBoxSupDrvMapUser: find_vma failed!\n"));
1249
1250	MY_DO_MUNMAP(current->mm, ulAddr, cPages * PAGE_SIZE);
1251	}
1252	else
1253	{
1254	dprintf(("supdrvOSContAllocOne: do_mmap failed ulAddr=%#lx\n", ulAddr));
1255	rc = SUPDRV_ERR_NO_MEMORY;
1256	}
1257	up_write(&current->mm->mmap_sem);
1258
1259	return NIL_RTR3PTR;
1260	}
1261
1262
1263	/**
1264	* Initializes the GIP.
1265	*
1266	* @returns negative errno.
1267	* @param pDevExt Instance data. GIP stuff may be updated.
1268	*/
1269	static int VBoxSupDrvInitGip(PSUPDRVDEVEXT pDevExt)
1270	{
1271	struct page *pPage;
1272	dma_addr_t HCPhys;
1273	PSUPGLOBALINFOPAGE pGip;
1274	#ifdef CONFIG_SMP
1275	unsigned i;
1276	#endif
1277	dprintf(("VBoxSupDrvInitGip:\n"));
1278
1279	/*
1280	* Allocate the page.
1281	*/
1282	pPage = alloc_pages(GFP_USER, 0);
1283	if (!pPage)
1284	{
1285	dprintf(("VBoxSupDrvInitGip: failed to allocate the GIP page\n"));
1286	return -ENOMEM;
1287	}
1288
1289	/*
1290	* Lock the page.
1291	*/
1292	SetPageReserved(pPage);
1293	g_pGipPage = pPage;
1294
1295	/*
1296	* Call common initialization routine.
1297	*/
1298	HCPhys = page_to_phys(pPage);
1299	pGip = (PSUPGLOBALINFOPAGE)page_address(pPage);
1300	pDevExt->ulLastJiffies = jiffies;
1301	#ifdef TICK_NSEC
1302	pDevExt->u64LastMonotime = (uint64_t)pDevExt->ulLastJiffies * TICK_NSEC;
1303	dprintf(("VBoxSupDrvInitGIP: TICK_NSEC=%ld HZ=%d jiffies=%ld now=%lld\n",
1304	TICK_NSEC, HZ, pDevExt->ulLastJiffies, pDevExt->u64LastMonotime));
1305	#else
1306	pDevExt->u64LastMonotime = (uint64_t)pDevExt->ulLastJiffies * (1000000 / HZ);
1307	dprintf(("VBoxSupDrvInitGIP: TICK_NSEC=%d HZ=%d jiffies=%ld now=%lld\n",
1308	(int)(1000000 / HZ), HZ, pDevExt->ulLastJiffies, pDevExt->u64LastMonotime));
1309	#endif
1310	supdrvGipInit(pDevExt, pGip, HCPhys, pDevExt->u64LastMonotime,
1311	HZ <= 1000 ? HZ : 1000);
1312
1313	/*
1314	* Initialize the timer.
1315	*/
1316	init_timer(&g_GipTimer);
1317	g_GipTimer.data = (unsigned long)pDevExt;
1318	g_GipTimer.function = VBoxSupGipTimer;
1319	g_GipTimer.expires = jiffies;
1320	#ifdef CONFIG_SMP
1321	for (i = 0; i < RT_ELEMENTS(pDevExt->aCPUs); i++)
1322	{
1323	init_timer(&pDevExt->aCPUs[i].Timer);
1324	pDevExt->aCPUs[i].Timer.data = (unsigned long)pDevExt;
1325	pDevExt->aCPUs[i].Timer.function = VBoxSupGipTimerPerCpu;
1326	pDevExt->aCPUs[i].Timer.expires = jiffies;
1327	}
1328	#endif
1329
1330	return 0;
1331	}
1332
1333
1334	/**
1335	* Terminates the GIP.
1336	*
1337	* @returns negative errno.
1338	* @param pDevExt Instance data. GIP stuff may be updated.
1339	*/
1340	static int VBoxSupDrvTermGip(PSUPDRVDEVEXT pDevExt)
1341	{
1342	struct page *pPage;
1343	PSUPGLOBALINFOPAGE pGip;
1344	#ifdef CONFIG_SMP
1345	unsigned i;
1346	#endif
1347	dprintf(("VBoxSupDrvTermGip:\n"));
1348
1349	/*
1350	* Delete the timer if it's pending.
1351	*/
1352	if (timer_pending(&g_GipTimer))
1353	del_timer_sync(&g_GipTimer);
1354	#ifdef CONFIG_SMP
1355	for (i = 0; i < RT_ELEMENTS(pDevExt->aCPUs); i++)
1356	if (timer_pending(&pDevExt->aCPUs[i].Timer))
1357	del_timer_sync(&pDevExt->aCPUs[i].Timer);
1358	#endif
1359
1360	/*
1361	* Uninitialize the content.
1362	*/
1363	pGip = pDevExt->pGip;
1364	pDevExt->pGip = NULL;
1365	if (pGip)
1366	supdrvGipTerm(pGip);
1367
1368	/*
1369	* Free the page.
1370	*/
1371	pPage = g_pGipPage;
1372	g_pGipPage = NULL;
1373	if (pPage)
1374	{
1375	ClearPageReserved(pPage);
1376	__free_pages(pPage, 0);
1377	}
1378
1379	return 0;
1380	}
1381
1382	/**
1383	* Timer callback function.
1384	*
1385	* In ASYNC TSC mode this is called on the primary CPU, and we're
1386	* assuming that the CPU remains online.
1387	*
1388	* @param ulUser The device extension pointer.
1389	*/
1390	static void VBoxSupGipTimer(unsigned long ulUser)
1391	{
1392	PSUPDRVDEVEXT pDevExt = (PSUPDRVDEVEXT)ulUser;
1393	PSUPGLOBALINFOPAGE pGip = pDevExt->pGip;
1394	unsigned long ulNow = jiffies;
1395	unsigned long ulDiff = ulNow - pDevExt->ulLastJiffies;
1396	uint64_t u64Monotime;
1397	pDevExt->ulLastJiffies = ulNow;
1398	#ifdef TICK_NSEC
1399	u64Monotime = pDevExt->u64LastMonotime + ulDiff * TICK_NSEC;
1400	#else
1401	u64Monotime = pDevExt->u64LastMonotime + ulDiff * (1000000 / HZ);
1402	#endif
1403	ASMAtomicXchgU64(&pDevExt->u64LastMonotime, u64Monotime);
1404	if (RT_LIKELY(pGip))
1405	supdrvGipUpdate(pDevExt->pGip, u64Monotime);
1406	mod_timer(&g_GipTimer, jiffies + (HZ <= 1000 ? 0 : ONE_MSEC_IN_JIFFIES));
1407	}
1408
1409
1410	#ifdef CONFIG_SMP
1411	/**
1412	* Timer callback function for the other CPUs.
1413	*
1414	* @param ulUser The device extension pointer.
1415	*/
1416	static void VBoxSupGipTimerPerCpu(unsigned long ulUser)
1417	{
1418	PSUPDRVDEVEXT pDevExt = (PSUPDRVDEVEXT)ulUser;
1419	PSUPGLOBALINFOPAGE pGip = pDevExt->pGip;
1420	unsigned long ulNow = jiffies;
1421	unsigned long ulDiff = ulNow - pDevExt->ulLastJiffies;
1422	uint64_t u64Monotime;
1423	uint8_t iCPU = ASMGetApicId();
1424
1425	if (RT_UNLIKELY(iCPU >= RT_ELEMENTS(pGip->aCPUs)))
1426	{
1427	printk("vboxdrv: error: apicid=%d max=%d cpuid=%d\n",
1428	iCPU, RT_ELEMENTS(pGip->aCPUs), smp_processor_id());
1429	return;
1430	}
1431
1432	pDevExt->aCPUs[iCPU].ulLastJiffies = ulNow;
1433	#ifdef TICK_NSEC
1434	u64Monotime = pDevExt->aCPUs[iCPU].u64LastMonotime + ulDiff * TICK_NSEC;
1435	#else
1436	u64Monotime = pDevExt->aCPUs[iCPU].u64LastMonotime + ulDiff * (1000000 / HZ);
1437	#endif
1438	ASMAtomicXchgU64(&pDevExt->aCPUs[iCPU].u64LastMonotime, u64Monotime);
1439	if (RT_LIKELY(pGip))
1440	supdrvGipUpdatePerCpu(pGip, pDevExt->aCPUs[iCPU].u64LastMonotime, iCPU);
1441	mod_timer(&pDevExt->aCPUs[iCPU].Timer, jiffies + (HZ <= 1000 ? 0 : ONE_MSEC_IN_JIFFIES));
1442	}
1443	#endif /* CONFIG_SMP */
1444
1445
1446	/**
1447	* Maps the GIP into user space.
1448	*
1449	* @returns negative errno.
1450	* @param pDevExt Instance data.
1451	*/
1452	int VBOXCALL supdrvOSGipMap(PSUPDRVDEVEXT pDevExt, PCSUPGLOBALINFOPAGE *ppGip)
1453	{
1454	int rc = 0;
1455	unsigned long ulAddr;
1456	unsigned long HCPhys = pDevExt->HCPhysGip;
1457	pgprot_t pgFlags;
1458	pgprot_val(pgFlags) = _PAGE_PRESENT \| _PAGE_USER;
1459	dprintf2(("supdrvOSGipMap: ppGip=%p\n", ppGip));
1460
1461	/*
1462	* Allocate user space mapping and put the physical pages into it.
1463	*/
1464	down_write(&current->mm->mmap_sem);
1465	ulAddr = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ, MAP_SHARED \| MAP_ANONYMOUS, 0);
1466	if (!(ulAddr & ~PAGE_MASK))
1467	{
1468	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0) && !defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1469	int rc2 = remap_page_range(ulAddr, HCPhys, PAGE_SIZE, pgFlags);
1470	#else
1471	int rc2 = 0;
1472	struct vm_area_struct *vma = find_vma(current->mm, ulAddr);
1473	if (vma)
1474	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11)
1475	rc2 = remap_page_range(vma, ulAddr, HCPhys, PAGE_SIZE, pgFlags);
1476	#else
1477	rc2 = remap_pfn_range(vma, ulAddr, HCPhys >> PAGE_SHIFT, PAGE_SIZE, pgFlags);
1478	#endif
1479	else
1480	{
1481	rc = SUPDRV_ERR_NO_MEMORY;
1482	dprintf(("supdrvOSGipMap: no vma found for ulAddr=%#lx!\n", ulAddr));
1483	}
1484	#endif
1485	if (rc2)
1486	{
1487	rc = SUPDRV_ERR_NO_MEMORY;
1488	dprintf(("supdrvOSGipMap: remap_page_range failed rc2=%d\n", rc2));
1489	}
1490	}
1491	else
1492	{
1493	dprintf(("supdrvOSGipMap: do_mmap failed ulAddr=%#lx\n", ulAddr));
1494	rc = SUPDRV_ERR_NO_MEMORY;
1495	}
1496	up_write(&current->mm->mmap_sem); /* not quite sure when to give this up. */
1497
1498	/*
1499	* Success?
1500	*/
1501	if (!rc)
1502	{
1503	*ppGip = (PCSUPGLOBALINFOPAGE)ulAddr;
1504	dprintf2(("supdrvOSGipMap: ppGip=%p\n", *ppGip));
1505	return 0;
1506	}
1507
1508	/*
1509	* Failure, cleanup and be gone.
1510	*/
1511	if (ulAddr & ~PAGE_MASK)
1512	{
1513	down_write(&current->mm->mmap_sem);
1514	MY_DO_MUNMAP(current->mm, ulAddr, PAGE_SIZE);
1515	up_write(&current->mm->mmap_sem);
1516	}
1517
1518	dprintf2(("supdrvOSGipMap: returns %d\n", rc));
1519	return rc;
1520	}
1521
1522
1523	/**
1524	* Maps the GIP into user space.
1525	*
1526	* @returns negative errno.
1527	* @param pDevExt Instance data.
1528	*/
1529	int VBOXCALL supdrvOSGipUnmap(PSUPDRVDEVEXT pDevExt, PCSUPGLOBALINFOPAGE pGip)
1530	{
1531	dprintf2(("supdrvOSGipUnmap: pGip=%p\n", pGip));
1532	if (current->mm)
1533	{
1534	down_write(&current->mm->mmap_sem);
1535	MY_DO_MUNMAP(current->mm, (unsigned long)pGip, PAGE_SIZE);
1536	up_write(&current->mm->mmap_sem);
1537	}
1538	dprintf2(("supdrvOSGipUnmap: returns 0\n"));
1539	return 0;
1540	}
1541
1542
1543	/**
1544	* Resumes the GIP updating.
1545	*
1546	* @param pDevExt Instance data.
1547	*/
1548	void VBOXCALL supdrvOSGipResume(PSUPDRVDEVEXT pDevExt)
1549	{
1550	dprintf2(("supdrvOSGipResume:\n"));
1551	#ifdef CONFIG_SMP
1552	if (pDevExt->pGip->u32Mode != SUPGIPMODE_ASYNC_TSC)
1553	#endif
1554	mod_timer(&g_GipTimer, jiffies);
1555	#ifdef CONFIG_SMP
1556	else
1557	{
1558	mod_timer(&g_GipTimer, jiffies);
1559	smp_call_function(VBoxSupGipResumePerCpu, pDevExt, 0 /* retry /, 1 / wait */);
1560	}
1561	#endif
1562	}
1563
1564
1565	#ifdef CONFIG_SMP
1566	/**
1567	* Callback for resuming GIP updating on the other CPUs.
1568	*
1569	* This is only used when the GIP is in async tsc mode.
1570	*
1571	* @param pvUser Pointer to the device instance.
1572	*/
1573	static void VBoxSupGipResumePerCpu(void *pvUser)
1574	{
1575	PSUPDRVDEVEXT pDevExt = (PSUPDRVDEVEXT)pvUser;
1576	uint8_t iCPU = ASMGetApicId();
1577
1578
1579	if (RT_UNLIKELY(iCPU >= RT_ELEMENTS(pDevExt->pGip->aCPUs)))
1580	{
1581	printk("vboxdrv: error: apicid=%d max=%d cpuid=%d\n",
1582	iCPU, RT_ELEMENTS(pDevExt->pGip->aCPUs), smp_processor_id());
1583	return;
1584	}
1585
1586	mod_timer(&pDevExt->aCPUs[iCPU].Timer, jiffies);
1587	}
1588	#endif /* CONFIG_SMP */
1589
1590
1591	/**
1592	* Suspends the GIP updating.
1593	*
1594	* @param pDevExt Instance data.
1595	*/
1596	void VBOXCALL supdrvOSGipSuspend(PSUPDRVDEVEXT pDevExt)
1597	{
1598	#ifdef CONFIG_SMP
1599	unsigned i;
1600	#endif
1601	dprintf2(("supdrvOSGipSuspend:\n"));
1602
1603	if (timer_pending(&g_GipTimer))
1604	del_timer_sync(&g_GipTimer);
1605	#ifdef CONFIG_SMP
1606	for (i = 0; i < RT_ELEMENTS(pDevExt->aCPUs); i++)
1607	if (timer_pending(&pDevExt->aCPUs[i].Timer))
1608	del_timer_sync(&pDevExt->aCPUs[i].Timer);
1609	#endif
1610	}
1611
1612
1613	/**
1614	* Get the current CPU count.
1615	* @returns Number of cpus.
1616	*/
1617	unsigned VBOXCALL supdrvOSGetCPUCount(void)
1618	{
1619	#ifdef CONFIG_SMP
1620	# ifdef num_present_cpus
1621	return num_present_cpus();
1622	# else
1623	return smp_num_cpus;
1624	# endif
1625	#else
1626	return 1;
1627	#endif
1628	}
1629
1630
1631	/**
1632	* Converts a supdrv error code to an linux error code.
1633	*
1634	* @returns corresponding linux error code.
1635	* @param rc supdrv error code (SUPDRV_ERR_* defines).
1636	*/
1637	static int VBoxSupDrvErr2LinuxErr(int rc)
1638	{
1639	switch (rc)
1640	{
1641	case 0: return 0;
1642	case SUPDRV_ERR_GENERAL_FAILURE: return -EACCES;
1643	case SUPDRV_ERR_INVALID_PARAM: return -EINVAL;
1644	case SUPDRV_ERR_INVALID_MAGIC: return -EILSEQ;
1645	case SUPDRV_ERR_INVALID_HANDLE: return -ENXIO;
1646	case SUPDRV_ERR_INVALID_POINTER: return -EFAULT;
1647	case SUPDRV_ERR_LOCK_FAILED: return -ENOLCK;
1648	case SUPDRV_ERR_ALREADY_LOADED: return -EEXIST;
1649	case SUPDRV_ERR_PERMISSION_DENIED: return -EPERM;
1650	case SUPDRV_ERR_VERSION_MISMATCH: return -ENOSYS;
1651	}
1652
1653	return -EPERM;
1654	}
1655
1656
1657	RTDECL(int) SUPR0Printf(const char *pszFormat, ...)
1658	{
1659	#if 1
1660	va_list args;
1661	char szMsg[512];
1662
1663	va_start(args, pszFormat);
1664	vsnprintf(szMsg, sizeof(szMsg) - 1, pszFormat, args);
1665	szMsg[sizeof(szMsg) - 1] = '\0';
1666	printk("%s", szMsg);
1667	va_end(args);
1668	#else
1669	/* forward to printf - needs some more GCC hacking to fix ebp... */
1670	__asm__ __volatile__ ("mov %0, %esp\n\t"
1671	"jmp %1\n\t",
1672	:: "r" ((uintptr_t)&pszFormat - 4),
1673	"m" (printk));
1674	#endif
1675	return 0;
1676	}
1677
1678
1679	/** Runtime assert implementation for Linux Ring-0. */
1680	RTDECL(void) AssertMsg1(const char pszExpr, unsigned uLine, const char pszFile, const char *pszFunction)
1681	{
1682	printk("!!Assertion Failed!!\n"
1683	"Expression: %s\n"
1684	"Location : %s(%d) %s\n",
1685	pszExpr, pszFile, uLine, pszFunction);
1686	}
1687
1688
1689	/** Runtime assert implementation for Linux Ring-0. */
1690	RTDECL(void) AssertMsg2(const char *pszFormat, ...)
1691	{ /* forwarder. */
1692	va_list ap;
1693	char msg[256];
1694
1695	va_start(ap, pszFormat);
1696	vsnprintf(msg, sizeof(msg) - 1, pszFormat, ap);
1697	msg[sizeof(msg) - 1] = '\0';
1698	printk("%s", msg);
1699	va_end(ap);
1700	}
1701
1702
1703	/* GCC C++ hack. */
1704	unsigned __gxx_personality_v0 = 0xcccccccc;
1705
1706
1707	module_init(VBoxSupDrvInit);
1708	module_exit(VBoxSupDrvUnload);
1709
1710	MODULE_AUTHOR("InnoTek Systemberatung GmbH");
1711	MODULE_DESCRIPTION("VirtualBox Support Driver");
1712	MODULE_LICENSE("GPL");
1713	#ifdef MODULE_VERSION
1714	#define xstr(s) str(s)
1715	#define str(s) #s
1716	MODULE_VERSION(VBOX_VERSION_STRING " (" xstr(SUPDRVIOC_VERSION) ")");
1717	#endif

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

source: vbox/trunk/src/VBox/HostDrivers/Support/linux/SUPDrv-linux.c@ 1666

Download in other formats: