SUPDrv-linux.c@ 417

Last change on this file since 417 was 387, checked in by vboxsync, 18 years ago
Use RTR0ProcHandleSelf / RTProcSelf. Implemented clientDied() cleanup on darwin.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 46.1 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 <iprt/assert.h>
26	#include <iprt/spinlock.h>
27	#include <iprt/semaphore.h>
28	#include <iprt/initterm.h>
29	#include <iprt/process.h>
30	#include <iprt/err.h>
31	#include <iprt/mem.h>
32
33	#include <linux/module.h>
34	#include <linux/kernel.h>
35	#include <linux/init.h>
36	#include <linux/fs.h>
37	#include <linux/mm.h>
38	#include <linux/pagemap.h>
39	#include <linux/slab.h>
40	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
41	# include <linux/jiffies.h>
42	#endif
43	#include <asm/mman.h>
44	#include <asm/io.h>
45	#include <asm/uaccess.h>
46	#ifdef CONFIG_DEVFS_FS
47	# include <linux/devfs_fs_kernel.h>
48	#endif
49	#ifdef CONFIG_VBOXDRV_AS_MISC
50	# include <linux/miscdevice.h>
51	#endif
52	#ifdef CONFIG_X86_LOCAL_APIC
53	# include <asm/apic.h>
54	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
55	# include <asm/nmi.h>
56	# endif
57	#endif
58
59	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0)
60	# ifndef page_to_pfn
61	# define page_to_pfn(page) ((page) - mem_map)
62	# endif
63	# include <asm/pgtable.h>
64	# define global_flush_tlb __flush_tlb_global
65	#endif
66
67	/* devfs defines */
68	#if defined(CONFIG_DEVFS_FS) && !defined(CONFIG_VBOXDRV_AS_MISC)
69	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
70
71	# define VBOX_REGISTER_DEVFS() \
72	({ \
73	void *rc = NULL; \
74	if (devfs_mk_cdev(MKDEV(DEVICE_MAJOR, 0), \
75	S_IFCHR \| S_IRUGO \| S_IWUGO, \
76	DEVICE_NAME) == 0) \
77	rc = (void )' '; / return not NULL */ \
78	rc; \
79	})
80
81	# define VBOX_UNREGISTER_DEVFS(handle) \
82	devfs_remove(DEVICE_NAME);
83
84	# else /* < 2.6.0 */
85
86	# define VBOX_REGISTER_DEVFS() \
87	devfs_register(NULL, DEVICE_NAME, DEVFS_FL_DEFAULT, \
88	DEVICE_MAJOR, 0, \
89	S_IFCHR \| S_IRUGO \| S_IWUGO, \
90	&gFileOpsVBoxDrv, NULL)
91
92	# define VBOX_UNREGISTER_DEVFS(handle) \
93	if (handle != NULL) \
94	devfs_unregister(handle)
95
96	# endif /* < 2.6.0 */
97	#endif /* CONFIG_DEV_FS && !CONFIG_VBOXDEV_AS_MISC */
98
99	#ifndef CONFIG_VBOXDRV_AS_MISC
100	# if defined(CONFIG_DEVFS_FS) && LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 0)
101	# define VBOX_REGISTER_DEVICE(a,b,c) devfs_register_chrdev(a,b,c)
102	# define VBOX_UNREGISTER_DEVICE(a,b) devfs_unregister_chrdev(a,b)
103	# else
104	# define VBOX_REGISTER_DEVICE(a,b,c) register_chrdev(a,b,c)
105	# define VBOX_UNREGISTER_DEVICE(a,b) unregister_chrdev(a,b)
106	# endif
107	#endif /* !CONFIG_VBOXDRV_AS_MISC */
108
109
110	#ifdef CONFIG_X86_HIGH_ENTRY
111	# error "CONFIG_X86_HIGH_ENTRY is not supported by VBoxDrv at this time."
112	#endif
113
114	/*
115	* This sucks soooo badly on x86! Why don't they export __PAGE_KERNEL_EXEC so PAGE_KERNEL_EXEC would be usable?
116	*/
117	#if defined(__AMD64__)
118	# define MY_PAGE_KERNEL_EXEC PAGE_KERNEL_EXEC
119	#elif defined(PAGE_KERNEL_EXEC) && defined(CONFIG_X86_PAE)
120	# define MY_PAGE_KERNEL_EXEC __pgprot(cpu_has_pge ? _PAGE_KERNEL_EXEC \| _PAGE_GLOBAL : _PAGE_KERNEL_EXEC)
121	#else
122	# define MY_PAGE_KERNEL_EXEC PAGE_KERNEL
123	#endif
124
125	/*
126	* The redhat hack section.
127	* - The current hacks are for 2.4.21-15.EL only.
128	*/
129	#ifndef NO_REDHAT_HACKS
130	/* accounting. */
131	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0)
132	# ifdef VM_ACCOUNT
133	# define MY_DO_MUNMAP(a,b,c) do_munmap(a, b, c, 0) /* should it be 1 or 0? */
134	# endif
135	# endif
136
137	/* backported remap_page_range. */
138	# if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0)
139	# include <asm/tlb.h>
140	# ifdef tlb_vma /* probably not good enough... */
141	# define HAVE_26_STYLE_REMAP_PAGE_RANGE 1
142	# endif
143	# endif
144
145	# ifndef __AMD64__
146	/* In 2.6.9-22.ELsmp we have to call change_page_attr() twice when changing
147	* the page attributes from PAGE_KERNEL to something else, because there appears
148	* to be a bug in one of the many patches that redhat applied.
149	* It should be safe to do this on less buggy linux kernels too. ;-)
150	*/
151	# define MY_CHANGE_PAGE_ATTR(pPages, cPages, prot) \
152	do { \
153	if (pgprot_val(prot) != pgprot_val(PAGE_KERNEL)) \
154	change_page_attr(pPages, cPages, prot); \
155	change_page_attr(pPages, cPages, prot); \
156	} while (0)
157	# endif
158	#endif /* !NO_REDHAT_HACKS */
159
160
161	#ifndef MY_DO_MUNMAP
162	# define MY_DO_MUNMAP(a,b,c) do_munmap(a, b, c)
163	#endif
164
165	#ifndef MY_CHANGE_PAGE_ATTR
166	# ifdef __AMD64__ /** @todo This is a cheap hack, but it'll get around that 'else BUG();' in __change_page_attr(). */
167	# define MY_CHANGE_PAGE_ATTR(pPages, cPages, prot) \
168	do { \
169	change_page_attr(pPages, cPages, PAGE_KERNEL_NOCACHE); \
170	change_page_attr(pPages, cPages, prot); \
171	} while (0)
172	# else
173	# define MY_CHANGE_PAGE_ATTR(pPages, cPages, prot) change_page_attr(pPages, cPages, prot)
174	# endif
175	#endif
176
177
178	/** @def ONE_MSEC_IN_JIFFIES
179	* The number of jiffies that make up 1 millisecond. This is only actually used
180	* when HZ is > 1000. */
181	#if HZ <= 1000
182	# define ONE_MSEC_IN_JIFFIES 0
183	#elif !(HZ % 1000)
184	# define ONE_MSEC_IN_JIFFIES (HZ / 1000)
185	#else
186	# define ONE_MSEC_IN_JIFFIES ((HZ + 999) / 1000)
187	# error "HZ is not a multiple of 1000, the GIP stuff won't work right!"
188	#endif
189
190
191	/*******************************************************************************
192	* Defined Constants And Macros *
193	*******************************************************************************/
194	/**
195	* Device extention & session data association structure.
196	*/
197	static SUPDRVDEVEXT g_DevExt;
198
199	/** Timer structure for the GIP update. */
200	static struct timer_list g_GipTimer;
201	/** Pointer to the page structure for the GIP. */
202	struct page *g_pGipPage;
203
204	/** Registered devfs device handle. */
205	#if defined(CONFIG_DEVFS_FS) && !defined(CONFIG_VBOXDRV_AS_MISC)
206	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
207	static void *g_hDevFsVBoxDrv = NULL;
208	# else
209	static devfs_handle_t g_hDevFsVBoxDrv = NULL;
210	# endif
211	#endif
212
213	#ifndef CONFIG_VBOXDRV_AS_MISC
214	/** Module major number */
215	#define DEVICE_MAJOR 234
216	/** Saved major device number */
217	static int g_iModuleMajor;
218	#endif /* !CONFIG_VBOXDRV_AS_MISC */
219
220	/** The module name. */
221	#define DEVICE_NAME "vboxdrv"
222
223	#ifdef __AMD64__
224	/**
225	* Memory for the executable memory heap (in IPRT).
226	*/
227	extern uint8_t g_abExecMemory[1572864]; /* 1.5 MB */
228	__asm__(".section execmemory, \"awx\", @progbits\n\t"
229	".align 32\n\t"
230	".globl g_abExecMemory\n"
231	"g_abExecMemory:\n\t"
232	".zero 1572864\n\t"
233	".type g_abExecMemory, @object\n\t"
234	".size g_abExecMemory, 1572864\n\t"
235	".text\n\t");
236	#endif
237
238
239	/*******************************************************************************
240	* Internal Functions *
241	*******************************************************************************/
242	static int VBoxSupDrvInit(void);
243	static void VBoxSupDrvUnload(void);
244	static int VBoxSupDrvCreate(struct inode pInode, struct file pFilp);
245	static int VBoxSupDrvClose(struct inode pInode, struct file pFilp);
246	static int VBoxSupDrvDeviceControl(struct inode pInode, struct file pFilp,
247	unsigned int IOCmd, unsigned long IOArg);
248	static void VBoxSupDrvMapUser(struct page *papPages, unsigned cPages, unsigned fProt, pgprot_t pgFlags);
249	static int VBoxSupDrvInitGip(PSUPDRVDEVEXT pDevExt);
250	static int VBoxSupDrvTermGip(PSUPDRVDEVEXT pDevExt);
251	static void VBoxSupGipTimer(unsigned long ulUser);
252	static int VBoxSupDrvOrder(unsigned long size);
253	static int VBoxSupDrvErr2LinuxErr(int);
254
255
256	/** The file_operations structure. */
257	static struct file_operations gFileOpsVBoxDrv =
258	{
259	owner: THIS_MODULE,
260	open: VBoxSupDrvCreate,
261	release: VBoxSupDrvClose,
262	ioctl: VBoxSupDrvDeviceControl,
263	};
264
265	#ifdef CONFIG_VBOXDRV_AS_MISC
266	/** The miscdevice structure. */
267	static struct miscdevice gMiscDevice =
268	{
269	minor: MISC_DYNAMIC_MINOR,
270	name: DEVICE_NAME,
271	fops: &gFileOpsVBoxDrv,
272	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) && \
273	LINUX_VERSION_CODE <= KERNEL_VERSION(2, 6, 17)
274	devfs_name: DEVICE_NAME,
275	# endif
276	};
277	#endif
278
279
280	/**
281	* Initialize module.
282	*
283	* @returns appropritate status code.
284	*/
285	static int __init VBoxSupDrvInit(void)
286	{
287	int rc;
288
289	dprintf(("VBoxDrv::ModuleInit\n"));
290
291	#ifdef CONFIG_X86_LOCAL_APIC
292	/*
293	* If an NMI occurs while we are inside the world switcher the macine will crash.
294	* The Linux NMI watchdog generates periodic NMIs increasing a counter which is
295	* compared with another counter increased in the timer interrupt handler. Therefore
296	* we don't allow to setup an NMI watchdog.
297	*/
298	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
299	/*
300	* First test: NMI actiated? Works only works with Linux 2.6 -- 2.4 does not export
301	* the nmi_watchdog variable.
302	*/
303	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
304	/*
305	* Permanent IO_APIC mode active? No way to handle this!
306	*/
307	if (nmi_watchdog == NMI_IO_APIC)
308	{
309	printk(KERN_ERR DEVICE_NAME
310	": NMI watchdog in IO_APIC mode active -- refused to load the kernel module!\n"
311	DEVICE_NAME
312	": Please disable the NMI watchdog by specifying 'nmi_watchdog=0' at kernel\n"
313	DEVICE_NAME
314	": command line.\n");
315	return -EINVAL;
316	}
317
318	/*
319	* See arch/i386/kernel/nmi.c on >= 2.6.19: -1 means it can never enabled again
320	*/
321	atomic_set(&nmi_active, -1);
322	printk(KERN_INFO DEVICE_NAME ": Trying to deactivate NMI watchdog permanently...\n");
323
324	/*
325	* Now fall through and see if it actually was enabled before. If so, fail
326	* as we cannot deactivate it cleanly from here.
327	*/
328	# else /* < 2.6.19 */
329	/*
330	* Older 2.6 kernels: nmi_watchdog is not initalized by default
331	*/
332	if (nmi_watchdog != NMI_NONE)
333	goto nmi_activated;
334	# endif
335	# endif /* >= 2.6.0 */
336
337	/*
338	* Second test: Interrupt generated by performance counter not masked and can
339	* generate an NMI. Works also with Linux 2.4.
340	*/
341	{
342	unsigned int v, ver, maxlvt;
343
344	v = apic_read(APIC_LVR);
345	ver = GET_APIC_VERSION(v);
346	/* 82489DXs do not report # of LVT entries. */
347	maxlvt = APIC_INTEGRATED(ver) ? GET_APIC_MAXLVT(v) : 2;
348	if (maxlvt >= 4)
349	{
350	/* Read status of performance counter IRQ vector */
351	v = apic_read(APIC_LVTPC);
352
353	/* performance counter generates NMI and is not masked? */
354	if ((GET_APIC_DELIVERY_MODE(v) == APIC_MODE_NMI) && !(v & APIC_LVT_MASKED))
355	{
356	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
357	printk(KERN_ERR DEVICE_NAME
358	": NMI watchdog either active or at least initialized. Please disable the NMI\n"
359	DEVICE_NAME
360	": watchdog by specifying 'nmi_watchdog=0' at kernel command line.\n");
361	return -EINVAL;
362	# else /* < 2.6.19 */
363	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
364	nmi_activated:
365	# endif
366	printk(KERN_ERR DEVICE_NAME
367	": NMI watchdog active -- refused to load the kernel module! Please disable\n"
368	DEVICE_NAME
369	": the NMI watchdog by specifying 'nmi_watchdog=0' at kernel command line.\n");
370	return -EINVAL;
371	# endif /* >= 2.6.19 */
372	}
373	}
374	}
375	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19)
376	printk(KERN_INFO DEVICE_NAME ": Successfully done.\n");
377	# endif /* >= 2.6.19 */
378	#endif /* CONFIG_X86_LOCAL_APIC */
379
380	#ifdef CONFIG_VBOXDRV_AS_MISC
381	rc = misc_register(&gMiscDevice);
382	if (rc)
383	{
384	printk(KERN_ERR DEVICE_NAME ": Can't register misc device! rc=%d\n", rc);
385	return rc;
386	}
387	#else /* !CONFIG_VBOXDRV_AS_MISC */
388	/*
389	* Register character device.
390	*/
391	g_iModuleMajor = DEVICE_MAJOR;
392	rc = VBOX_REGISTER_DEVICE((dev_t)g_iModuleMajor, DEVICE_NAME, &gFileOpsVBoxDrv);
393	if (rc < 0)
394	{
395	dprintf(("VBOX_REGISTER_DEVICE failed with rc=%#x!\n", rc));
396	return rc;
397	}
398
399	/*
400	* Save returned module major number
401	*/
402	if (DEVICE_MAJOR != 0)
403	g_iModuleMajor = DEVICE_MAJOR;
404	else
405	g_iModuleMajor = rc;
406	rc = 0;
407
408	#ifdef CONFIG_DEVFS_FS
409	/*
410	* Register a device entry
411	*/
412	g_hDevFsVBoxDrv = VBOX_REGISTER_DEVFS();
413	if (g_hDevFsVBoxDrv == NULL)
414	{
415	dprintf(("devfs_register failed!\n"));
416	rc = -EINVAL;
417	}
418	#endif
419	#endif /* !CONFIG_VBOXDRV_AS_MISC */
420	if (!rc)
421	{
422	/*
423	* Initialize the runtime.
424	* On AMD64 we'll have to donate the high rwx memory block to the exec allocator.
425	*/
426	rc = RTR0Init(0);
427	if (RT_SUCCESS(rc))
428	{
429	#ifdef __AMD64__
430	rc = RTR0MemExecDonate(&g_abExecMemory[0], sizeof(g_abExecMemory));
431	#endif
432	/*
433	* Initialize the device extension.
434	*/
435	if (RT_SUCCESS(rc))
436	rc = supdrvInitDevExt(&g_DevExt);
437	if (!rc)
438	{
439	/*
440	* Create the GIP page.
441	*/
442	rc = VBoxSupDrvInitGip(&g_DevExt);
443	if (!rc)
444	{
445	dprintf(("VBoxDrv::ModuleInit returning %#x\n", rc));
446	return rc;
447	}
448
449	supdrvDeleteDevExt(&g_DevExt);
450	}
451	else
452	rc = -EINVAL;
453	RTR0Term();
454	}
455	else
456	rc = -EINVAL;
457
458	/*
459	* Failed, cleanup and return the error code.
460	*/
461	#if defined(CONFIG_DEVFS_FS) && !defined(CONFIG_VBOXDRV_AS_MISC)
462	VBOX_UNREGISTER_DEVFS(g_hDevFsVBoxDrv);
463	#endif
464	}
465	#ifdef CONFIG_VBOXDRV_AS_MISC
466	misc_deregister(&gMiscDevice);
467	dprintf(("VBoxDrv::ModuleInit returning %#x (minor:%d)\n", rc, gMiscDevice.minor));
468	#else
469	VBOX_UNREGISTER_DEVICE(g_iModuleMajor, DEVICE_NAME);
470	dprintf(("VBoxDrv::ModuleInit returning %#x (major:%d)\n", rc, g_iModuleMajor));
471	#endif
472	return rc;
473	}
474
475
476	/**
477	* Unload the module.
478	*/
479	static void __exit VBoxSupDrvUnload(void)
480	{
481	int rc;
482	dprintf(("VBoxSupDrvUnload\n"));
483
484	/*
485	* I Don't think it's possible to unload a driver which processes have
486	* opened, at least we'll blindly assume that here.
487	*/
488	#ifdef CONFIG_VBOXDRV_AS_MISC
489	rc = misc_deregister(&gMiscDevice);
490	if (rc < 0)
491	{
492	dprintf(("misc_deregister failed with rc=%#x\n", rc));
493	}
494	#else /* !CONFIG_VBOXDRV_AS_MISC */
495	#ifdef CONFIG_DEVFS_FS
496	/*
497	* Unregister a device entry
498	*/
499	VBOX_UNREGISTER_DEVFS(g_hDevFsVBoxDrv);
500	#endif // devfs
501	rc = VBOX_UNREGISTER_DEVICE(g_iModuleMajor, DEVICE_NAME);
502	if (rc < 0)
503	{
504	dprintf(("unregister_chrdev failed with rc=%#x (major:%d)\n", rc, g_iModuleMajor));
505	}
506	#endif /* !CONFIG_VBOXDRV_AS_MISC */
507
508	/*
509	* Destroy GIP, delete the device extension and terminate IPRT.
510	*/
511	VBoxSupDrvTermGip(&g_DevExt);
512	supdrvDeleteDevExt(&g_DevExt);
513	RTR0Term();
514	}
515
516
517	/**
518	* Device open. Called on open /dev/vboxdrv
519	*
520	* @param pInode Pointer to inode info structure.
521	* @param pFilp Associated file pointer.
522	*/
523	static int VBoxSupDrvCreate(struct inode pInode, struct file pFilp)
524	{
525	int rc;
526	PSUPDRVSESSION pSession;
527	dprintf(("VBoxSupDrvCreate: pFilp=%p\n", pFilp));
528
529	/*
530	* Call common code for the rest.
531	*/
532	rc = supdrvCreateSession(&g_DevExt, (PSUPDRVSESSION *)&pSession);
533	if (!rc)
534	{
535	pSession->Uid = current->euid;
536	pSession->Gid = current->egid;
537	pSession->Process = RTProcSelf();
538	pSession->R0Process = RTR0ProcHandleSelf();
539	}
540
541	dprintf(("VBoxSupDrvCreate: g_DevExt=%p pSession=%p rc=%d\n", &g_DevExt, pSession, rc));
542	pFilp->private_data = pSession;
543
544	return VBoxSupDrvErr2LinuxErr(rc);
545	}
546
547
548	/**
549	* Close device.
550	*
551	* @param pInode Pointer to inode info structure.
552	* @param pFilp Associated file pointer.
553	*/
554	static int VBoxSupDrvClose(struct inode pInode, struct file pFilp)
555	{
556	dprintf(("VBoxSupDrvClose: pFilp=%p private_data=%p\n", pFilp, pFilp->private_data));
557	supdrvCloseSession(&g_DevExt, (PSUPDRVSESSION)pFilp->private_data);
558	pFilp->private_data = NULL;
559	return 0;
560	}
561
562
563	/**
564	* Device I/O Control entry point.
565	*
566	* @param pInode Pointer to inode info structure.
567	* @param pFilp Associated file pointer.
568	* @param IOCmd The function specified to ioctl().
569	* @param IOArg The argument specified to ioctl().
570	*/
571	static int VBoxSupDrvDeviceControl(struct inode pInode, struct file pFilp,
572	unsigned int IOCmd, unsigned long IOArg)
573	{
574	int rc;
575	SUPDRVIOCTLDATA Args;
576	void *pvBuf = NULL;
577	int cbBuf = 0;
578	unsigned cbOut = 0;
579
580	dprintf2(("VBoxSupDrvDeviceControl: pFilp=%p IOCmd=%x IOArg=%p\n", pFilp, IOCmd, (void *)IOArg));
581
582	/*
583	* Copy ioctl data structure from user space.
584	*/
585	if (_IOC_SIZE(IOCmd) != sizeof(SUPDRVIOCTLDATA))
586	{
587	dprintf(("VBoxSupDrvDeviceControl: incorrect input length! cbArgs=%d\n", _IOC_SIZE(IOCmd)));
588	return -EINVAL;
589	}
590	if (copy_from_user(&Args, (void *)IOArg, _IOC_SIZE(IOCmd)))
591	{
592	dprintf(("VBoxSupDrvDeviceControl: copy_from_user(&Args) failed.\n"));
593	return -EFAULT;
594	}
595
596	/*
597	* Allocate and copy user space input data buffer to kernel space.
598	*/
599	if (Args.cbIn > 0 \|\| Args.cbOut > 0)
600	{
601	cbBuf = max(Args.cbIn, Args.cbOut);
602	pvBuf = vmalloc(cbBuf);
603	if (pvBuf == NULL)
604	{
605	dprintf(("VBoxSupDrvDeviceControl: failed to allocate buffer of %d bytes.\n", cbBuf));
606	return -ENOMEM;
607	}
608
609	if (copy_from_user(pvBuf, (void *)Args.pvIn, Args.cbIn))
610	{
611	dprintf(("VBoxSupDrvDeviceControl: copy_from_user(pvBuf) failed.\n"));
612	vfree(pvBuf);
613	return -EFAULT;
614	}
615	}
616
617	/*
618	* Process the IOCtl.
619	*/
620	rc = supdrvIOCtl(IOCmd, &g_DevExt, (PSUPDRVSESSION)pFilp->private_data,
621	pvBuf, Args.cbIn, pvBuf, Args.cbOut, &cbOut);
622
623	/*
624	* Copy ioctl data and output buffer back to user space.
625	*/
626	if (rc)
627	{
628	dprintf(("VBoxSupDrvDeviceControl: pFilp=%p IOCmd=%x IOArg=%p failed, rc=%d (linux rc=%d)\n",
629	pFilp, IOCmd, (void *)IOArg, rc, VBoxSupDrvErr2LinuxErr(rc)));
630	rc = VBoxSupDrvErr2LinuxErr(rc);
631	}
632	else if (cbOut > 0)
633	{
634	if (pvBuf != NULL && cbOut <= cbBuf)
635	{
636	if (copy_to_user((void *)Args.pvOut, pvBuf, cbOut))
637	{
638	dprintf(("copy_to_user failed.\n"));
639	rc = -EFAULT;
640	}
641	}
642	else
643	{
644	dprintf(("WHAT!?! supdrvIOCtl messed up! cbOut=%d cbBuf=%d pvBuf=%p\n", cbOut, cbBuf, pvBuf));
645	rc = -EPERM;
646	}
647	}
648
649	if (pvBuf)
650	vfree(pvBuf);
651
652	dprintf2(("VBoxSupDrvDeviceControl: returns %d\n", rc));
653	return rc;
654	}
655
656
657	/**
658	* Initializes any OS specific object creator fields.
659	*/
660	void VBOXCALL supdrvOSObjInitCreator(PSUPDRVOBJ pObj, PSUPDRVSESSION pSession)
661	{
662	NOREF(pObj);
663	NOREF(pSession);
664	}
665
666
667	/**
668	* Checks if the session can access the object.
669	*
670	* @returns true if a decision has been made.
671	* @returns false if the default access policy should be applied.
672	*
673	* @param pObj The object in question.
674	* @param pSession The session wanting to access the object.
675	* @param pszObjName The object name, can be NULL.
676	* @param prc Where to store the result when returning true.
677	*/
678	bool VBOXCALL supdrvOSObjCanAccess(PSUPDRVOBJ pObj, PSUPDRVSESSION pSession, const char pszObjName, int prc)
679	{
680	NOREF(pObj);
681	NOREF(pSession);
682	NOREF(pszObjName);
683	NOREF(prc);
684	return false;
685	}
686
687
688	/**
689	* Compute order. Some functions allocate 2^order pages.
690	*
691	* @returns order.
692	* @param cPages Number of pages.
693	*/
694	static int VBoxSupDrvOrder(unsigned long cPages)
695	{
696	int iOrder;
697	unsigned long cTmp;
698
699	for (iOrder = 0, cTmp = cPages; cTmp >>= 1; ++iOrder)
700	;
701	if (cPages & ~(1 << iOrder))
702	++iOrder;
703
704	return iOrder;
705	}
706
707
708	/**
709	* OS Specific code for locking down memory.
710	*
711	* @returns 0 on success.
712	* @returns SUPDRV_ERR_* on failure.
713	* @param pMem Pointer to memory.
714	* This is not linked in anywhere.
715	* @param paPages Array which should be filled with the address of the physical pages.
716	*
717	* @remark See sgl_map_user_pages() for an example of an similar function.
718	*/
719	int VBOXCALL supdrvOSLockMemOne(PSUPDRVMEMREF pMem, PSUPPAGE paPages)
720	{
721	int rc;
722	struct page **papPages;
723	unsigned iPage;
724	unsigned cPages = pMem->cb >> PAGE_SHIFT;
725	unsigned long pv = (unsigned long)pMem->pvR3;
726
727	/*
728	* Allocate page pointer array.
729	*/
730	papPages = vmalloc(cPages * sizeof(*papPages));
731	if (!papPages)
732	return SUPDRV_ERR_NO_MEMORY;
733
734	/*
735	* Get user pages.
736	*/
737	down_read(&current->mm->mmap_sem);
738	rc = get_user_pages(current, /* Task for fault acounting. */
739	current->mm, /* Whose pages. */
740	(unsigned long)pv, /* Where from. */
741	cPages, /* How many pages. */
742	1, /* Write to memory. */
743	0, /* force. */
744	papPages, /* Page array. */
745	NULL); /* vmas */
746	if (rc != cPages)
747	{
748	up_read(&current->mm->mmap_sem);
749	dprintf(("supdrvOSLockMemOne: get_user_pages failed. rc=%d\n", rc));
750	return SUPDRV_ERR_LOCK_FAILED;
751	}
752
753	for (iPage = 0; iPage < cPages; iPage++)
754	flush_dcache_page(papPages[iPage]);
755	up_read(&current->mm->mmap_sem);
756
757	pMem->u.locked.papPages = papPages;
758	pMem->u.locked.cPages = cPages;
759
760	/*
761	* Get addresses.
762	*/
763	for (iPage = 0; iPage < cPages; iPage++)
764	{
765	paPages[iPage].Phys = page_to_phys(papPages[iPage]);
766	paPages[iPage].uReserved = 0;
767	}
768
769	dprintf2(("supdrvOSLockMemOne: pvR3=%p cb=%d papPages=%p\n",
770	pMem->pvR3, pMem->cb, pMem->u.locked.papPages));
771	return 0;
772	}
773
774
775	/**
776	* Unlocks the memory pointed to by pv.
777	*
778	* @param pv Memory to unlock.
779	* @param cb Size of the memory (debug).
780	*
781	* @remark See sgl_unmap_user_pages() for an example of an similar function.
782	*/
783	void VBOXCALL supdrvOSUnlockMemOne(PSUPDRVMEMREF pMem)
784	{
785	unsigned iPage;
786	dprintf2(("supdrvOSUnlockMemOne: pvR3=%p cb=%d papPages=%p\n",
787	pMem->pvR3, pMem->cb, pMem->u.locked.papPages));
788
789	/*
790	* Loop thru the pages and release them.
791	*/
792	for (iPage = 0; iPage < pMem->u.locked.cPages; iPage++)
793	{
794	if (!PageReserved(pMem->u.locked.papPages[iPage]))
795	SetPageDirty(pMem->u.locked.papPages[iPage]);
796	page_cache_release(pMem->u.locked.papPages[iPage]);
797	}
798
799	/* free the page array */
800	vfree(pMem->u.locked.papPages);
801	pMem->u.locked.cPages = 0;
802	}
803
804
805	/**
806	* OS Specific code for allocating page aligned memory with continuous fixed
807	* physical paged backing.
808	*
809	* @returns 0 on success.
810	* @returns SUPDRV_ERR_* on failure.
811	* @param pMem Memory reference record of the memory to be allocated.
812	* (This is not linked in anywhere.)
813	* @param ppvR0 Where to store the virtual address of the ring-0 mapping. (optional)
814	* @param ppvR3 Where to store the virtual address of the ring-3 mapping.
815	* @param pHCPhys Where to store the physical address.
816	*/
817	int VBOXCALL supdrvOSContAllocOne(PSUPDRVMEMREF pMem, void ppvR0, void ppvR3, PRTHCPHYS pHCPhys)
818	{
819	struct page *paPages;
820	unsigned iPage;
821	unsigned cbAligned = RT_ALIGN(pMem->cb, PAGE_SIZE);
822	unsigned cPages = cbAligned >> PAGE_SHIFT;
823	unsigned cOrder = VBoxSupDrvOrder(cPages);
824	unsigned long ulAddr;
825	dma_addr_t HCPhys;
826	int rc = 0;
827	pgprot_t pgFlags;
828	pgprot_val(pgFlags) = _PAGE_PRESENT \| _PAGE_RW \| _PAGE_USER;
829
830	Assert(ppvR3);
831	Assert(pHCPhys);
832
833	/*
834	* Allocate page pointer array.
835	*/
836	#ifdef __AMD64__ /** @todo check out if there is a correct way of getting memory below 4GB (physically). */
837	paPages = alloc_pages(GFP_DMA, cOrder);
838	#else
839	paPages = alloc_pages(GFP_USER, cOrder);
840	#endif
841	if (!paPages)
842	return SUPDRV_ERR_NO_MEMORY;
843
844	/*
845	* Lock the pages.
846	*/
847	for (iPage = 0; iPage < cPages; iPage++)
848	{
849	SetPageReserved(&paPages[iPage]);
850	if (!PageHighMem(&paPages[iPage]) && pgprot_val(MY_PAGE_KERNEL_EXEC) != pgprot_val(PAGE_KERNEL))
851	MY_CHANGE_PAGE_ATTR(&paPages[iPage], 1, MY_PAGE_KERNEL_EXEC);
852	#ifdef DEBUG
853	if (iPage + 1 < cPages && (page_to_phys((&paPages[iPage])) + 0x1000) != page_to_phys((&paPages[iPage + 1])))
854	{
855	dprintf(("supdrvOSContAllocOne: Pages are not continuous!!!! iPage=%d phys=%llx physnext=%llx\n",
856	iPage, (long long)page_to_phys((&paPages[iPage])), (long long)page_to_phys((&paPages[iPage + 1]))));
857	BUG();
858	}
859	#endif
860	}
861	HCPhys = page_to_phys(paPages);
862
863	/*
864	* Allocate user space mapping and put the physical pages into it.
865	*/
866	down_write(&current->mm->mmap_sem);
867	ulAddr = do_mmap(NULL, 0, cbAligned, PROT_READ \| PROT_WRITE \| PROT_EXEC, MAP_SHARED \| MAP_ANONYMOUS, 0);
868	if (!(ulAddr & ~PAGE_MASK))
869	{
870	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0) && !defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
871	int rc2 = remap_page_range(ulAddr, HCPhys, cbAligned, pgFlags);
872	#else
873	int rc2 = 0;
874	struct vm_area_struct *vma = find_vma(current->mm, ulAddr);
875	if (vma)
876	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11)
877	rc2 = remap_page_range(vma, ulAddr, HCPhys, cbAligned, pgFlags);
878	#else
879	rc2 = remap_pfn_range(vma, ulAddr, HCPhys >> PAGE_SHIFT, cbAligned, pgFlags);
880	#endif
881	else
882	{
883	rc = SUPDRV_ERR_NO_MEMORY;
884	dprintf(("supdrvOSContAllocOne: no vma found for ulAddr=%#lx!\n", ulAddr));
885	}
886	#endif
887	if (rc2)
888	{
889	rc = SUPDRV_ERR_NO_MEMORY;
890	dprintf(("supdrvOSContAllocOne: remap_page_range failed rc2=%d\n", rc2));
891	}
892	}
893	else
894	{
895	dprintf(("supdrvOSContAllocOne: do_mmap failed ulAddr=%#lx\n", ulAddr));
896	rc = SUPDRV_ERR_NO_MEMORY;
897	}
898	up_write(&current->mm->mmap_sem); /* not quite sure when to give this up. */
899
900	/*
901	* Success?
902	*/
903	if (!rc)
904	{
905	*pHCPhys = HCPhys;
906	ppvR3 = (void )ulAddr;
907	if (ppvR0)
908	ppvR0 = (void )ulAddr;
909	pMem->pvR3 = (void *)ulAddr;
910	pMem->pvR0 = NULL;
911	pMem->u.cont.paPages = paPages;
912	pMem->u.cont.cPages = cPages;
913	pMem->cb = cbAligned;
914
915	dprintf2(("supdrvOSContAllocOne: pvR0=%p pvR3=%p cb=%d paPages=%p pHCPhys=%lx ppvR0=*ppvR3=%p\n",
916	pMem->pvR0, pMem->pvR3, pMem->cb, paPages, (unsigned long)pHCPhys, ppvR3));
917	global_flush_tlb();
918	return 0;
919	}
920
921	/*
922	* Failure, cleanup and be gone.
923	*/
924	down_write(&current->mm->mmap_sem);
925	if (ulAddr & ~PAGE_MASK)
926	MY_DO_MUNMAP(current->mm, ulAddr, pMem->cb);
927	for (iPage = 0; iPage < cPages; iPage++)
928	{
929	ClearPageReserved(&paPages[iPage]);
930	if (!PageHighMem(&paPages[iPage]) && pgprot_val(MY_PAGE_KERNEL_EXEC) != pgprot_val(PAGE_KERNEL))
931	MY_CHANGE_PAGE_ATTR(&paPages[iPage], 1, PAGE_KERNEL);
932	}
933	up_write(&current->mm->mmap_sem); /* check when we can leave this. */
934	__free_pages(paPages, cOrder);
935
936	global_flush_tlb();
937	return rc;
938	}
939
940
941	/**
942	* Frees contiguous memory.
943	*
944	* @param pMem Memory reference record of the memory to be freed.
945	*/
946	void VBOXCALL supdrvOSContFreeOne(PSUPDRVMEMREF pMem)
947	{
948	unsigned iPage;
949
950	dprintf2(("supdrvOSContFreeOne: pvR0=%p pvR3=%p cb=%d paPages=%p\n",
951	pMem->pvR0, pMem->pvR3, pMem->cb, pMem->u.cont.paPages));
952
953	/*
954	* do_exit() destroys the mm before closing files.
955	* I really hope it cleans up our stuff properly...
956	*/
957	if (current->mm)
958	{
959	down_write(&current->mm->mmap_sem);
960	MY_DO_MUNMAP(current->mm, (unsigned long)pMem->pvR3, pMem->cb);
961	up_write(&current->mm->mmap_sem); /* check when we can leave this. */
962	}
963
964	/*
965	* Change page attributes freeing the pages.
966	*/
967	for (iPage = 0; iPage < pMem->u.cont.cPages; iPage++)
968	{
969	ClearPageReserved(&pMem->u.cont.paPages[iPage]);
970	if (!PageHighMem(&pMem->u.cont.paPages[iPage]) && pgprot_val(MY_PAGE_KERNEL_EXEC) != pgprot_val(PAGE_KERNEL))
971	MY_CHANGE_PAGE_ATTR(&pMem->u.cont.paPages[iPage], 1, PAGE_KERNEL);
972	}
973	__free_pages(pMem->u.cont.paPages, VBoxSupDrvOrder(pMem->u.cont.cPages));
974
975	pMem->u.cont.cPages = 0;
976	}
977
978
979	/**
980	* Allocates memory which mapped into both kernel and user space.
981	* The returned memory is page aligned and so is the allocation.
982	*
983	* @returns 0 on success.
984	* @returns SUPDRV_ERR_* on failure.
985	* @param pMem Memory reference record of the memory to be allocated.
986	* (This is not linked in anywhere.)
987	* @param ppvR0 Where to store the address of the Ring-0 mapping.
988	* @param ppvR3 Where to store the address of the Ring-3 mapping.
989	*/
990	int VBOXCALL supdrvOSMemAllocOne(PSUPDRVMEMREF pMem, void ppvR0, void ppvR3)
991	{
992	const unsigned cbAligned = RT_ALIGN(pMem->cb, PAGE_SIZE);
993	const unsigned cPages = cbAligned >> PAGE_SHIFT;
994	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 4, 22)
995	unsigned cOrder = VBoxSupDrvOrder(cPages);
996	struct page *paPages;
997	#endif
998	struct page **papPages;
999	unsigned iPage;
1000	pgprot_t pgFlags;
1001	pgprot_val(pgFlags) = _PAGE_PRESENT \| _PAGE_RW \| _PAGE_USER;
1002
1003	/*
1004	* Allocate array with page pointers.
1005	*/
1006	pMem->u.mem.cPages = 0;
1007	pMem->u.mem.papPages = papPages = kmalloc(sizeof(papPages[0]) * cPages, GFP_KERNEL);
1008	if (!papPages)
1009	return SUPDRV_ERR_NO_MEMORY;
1010
1011	/*
1012	* Allocate the pages.
1013	*/
1014	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1015	for (iPage = 0; iPage < cPages; iPage++)
1016	{
1017	papPages[iPage] = alloc_page(GFP_HIGHUSER);
1018	if (!papPages[iPage])
1019	{
1020	pMem->u.mem.cPages = iPage;
1021	supdrvOSMemFreeOne(pMem);
1022	return SUPDRV_ERR_NO_MEMORY;
1023	}
1024	}
1025
1026	#else /* < 2.4.22 */
1027	paPages = alloc_pages(GFP_USER, cOrder);
1028	if (!paPages)
1029	{
1030	supdrvOSMemFreeOne(pMem);
1031	return SUPDRV_ERR_NO_MEMORY;
1032	}
1033	for (iPage = 0; iPage < cPages; iPage++)
1034	{
1035	papPages[iPage] = &paPages[iPage];
1036	if (pgprot_val(MY_PAGE_KERNEL_EXEC) != pgprot_val(PAGE_KERNEL))
1037	MY_CHANGE_PAGE_ATTR(papPages[iPage], 1, MY_PAGE_KERNEL_EXEC);
1038	if (PageHighMem(papPages[iPage]))
1039	BUG();
1040	}
1041	#endif
1042	pMem->u.mem.cPages = cPages;
1043
1044	/*
1045	* Reserve the pages.
1046	*/
1047	for (iPage = 0; iPage < cPages; iPage++)
1048	SetPageReserved(papPages[iPage]);
1049
1050	/*
1051	* Create the Ring-0 mapping.
1052	*/
1053	if (ppvR0)
1054	{
1055	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1056	# ifdef VM_MAP
1057	*ppvR0 = pMem->pvR0 = vmap(papPages, cPages, VM_MAP, pgFlags);
1058	# else
1059	*ppvR0 = pMem->pvR0 = vmap(papPages, cPages, VM_ALLOC, pgFlags);
1060	# endif
1061	#else
1062	*ppvR0 = pMem->pvR0 = phys_to_virt(page_to_phys(papPages[0]));
1063	#endif
1064	}
1065	if (pMem->pvR0 \|\| !ppvR0)
1066	{
1067	/*
1068	* Create the ring3 mapping.
1069	*/
1070	if (ppvR3)
1071	*ppvR3 = pMem->pvR3 = VBoxSupDrvMapUser(papPages, cPages, PROT_READ \| PROT_WRITE \| PROT_EXEC, pgFlags);
1072	if (pMem->pvR3 \|\| !ppvR3)
1073	return 0;
1074	dprintf(("supdrvOSMemAllocOne: failed to map into r3! cPages=%u\n", cPages));
1075	}
1076	else
1077	dprintf(("supdrvOSMemAllocOne: failed to map into r0! cPages=%u\n", cPages));
1078
1079	supdrvOSMemFreeOne(pMem);
1080	return SUPDRV_ERR_NO_MEMORY;
1081	}
1082
1083
1084	/**
1085	* Get the physical addresses of the pages in the allocation.
1086	* This is called while inside bundle the spinlock.
1087	*
1088	* @param pMem Memory reference record of the memory.
1089	* @param paPages Where to store the page addresses.
1090	*/
1091	void VBOXCALL supdrvOSMemGetPages(PSUPDRVMEMREF pMem, PSUPPAGE paPages)
1092	{
1093	unsigned iPage;
1094	for (iPage = 0; iPage < pMem->u.mem.cPages; iPage++)
1095	{
1096	paPages[iPage].Phys = page_to_phys(pMem->u.mem.papPages[iPage]);
1097	paPages[iPage].uReserved = 0;
1098	}
1099	}
1100
1101
1102	/**
1103	* Frees memory allocated by supdrvOSMemAllocOne().
1104	*
1105	* @param pMem Memory reference record of the memory to be free.
1106	*/
1107	void VBOXCALL supdrvOSMemFreeOne(PSUPDRVMEMREF pMem)
1108	{
1109	dprintf2(("supdrvOSMemFreeOne: pvR0=%p pvR3=%p cb=%d cPages=%d papPages=%p\n",
1110	pMem->pvR0, pMem->pvR3, pMem->cb, pMem->u.mem.cPages, pMem->u.mem.papPages));
1111
1112	/*
1113	* Unmap the user mapping (if any).
1114	* do_exit() destroys the mm before closing files.
1115	*/
1116	if (pMem->pvR3 && current->mm)
1117	{
1118	down_write(&current->mm->mmap_sem);
1119	MY_DO_MUNMAP(current->mm, (unsigned long)pMem->pvR3, RT_ALIGN(pMem->cb, PAGE_SIZE));
1120	up_write(&current->mm->mmap_sem); /* check when we can leave this. */
1121	}
1122	pMem->pvR3 = NULL;
1123
1124	/*
1125	* Unmap the kernel mapping (if any).
1126	*/
1127	if (pMem->pvR0)
1128	{
1129	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1130	vunmap(pMem->pvR0);
1131	#endif
1132	pMem->pvR0 = NULL;
1133	}
1134
1135	/*
1136	* Free the physical pages.
1137	*/
1138	if (pMem->u.mem.papPages)
1139	{
1140	struct page **papPages = pMem->u.mem.papPages;
1141	const unsigned cPages = pMem->u.mem.cPages;
1142	unsigned iPage;
1143
1144	/* Restore the page flags. */
1145	for (iPage = 0; iPage < cPages; iPage++)
1146	{
1147	ClearPageReserved(papPages[iPage]);
1148	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 4, 22)
1149	if (pgprot_val(MY_PAGE_KERNEL_EXEC) != pgprot_val(PAGE_KERNEL))
1150	MY_CHANGE_PAGE_ATTR(papPages[iPage], 1, PAGE_KERNEL);
1151	#endif
1152	}
1153
1154	/* Free the pages. */
1155	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1156	for (iPage = 0; iPage < pMem->u.cont.cPages; iPage++)
1157	__free_page(papPages[iPage]);
1158	#else
1159	if (cPages > 0)
1160	__free_pages(papPages[0], VBoxSupDrvOrder(cPages));
1161	#endif
1162	/* Free the page pointer array. */
1163	kfree(papPages);
1164	pMem->u.mem.papPages = NULL;
1165	}
1166	pMem->u.mem.cPages = 0;
1167	}
1168
1169
1170	/**
1171	* Maps a range of pages into user space.
1172	*
1173	* @returns Pointer to the user space mapping on success.
1174	* @returns NULL on failure.
1175	* @param papPages Array of the pages to map.
1176	* @param cPages Number of pages to map.
1177	* @param fProt The mapping protection.
1178	* @param pgFlags The page level protection.
1179	*/
1180	static void VBoxSupDrvMapUser(struct page *papPages, unsigned cPages, unsigned fProt, pgprot_t pgFlags)
1181	{
1182	int rc = SUPDRV_ERR_NO_MEMORY;
1183	unsigned long ulAddr;
1184
1185	/*
1186	* Allocate user space mapping.
1187	*/
1188	down_write(&current->mm->mmap_sem);
1189	ulAddr = do_mmap(NULL, 0, cPages * PAGE_SIZE, fProt, MAP_SHARED \| MAP_ANONYMOUS, 0);
1190	if (!(ulAddr & ~PAGE_MASK))
1191	{
1192	/*
1193	* Map page by page into the mmap area.
1194	* This is generic, paranoid and not very efficient.
1195	*/
1196	int rc = 0;
1197	unsigned long ulAddrCur = ulAddr;
1198	unsigned iPage;
1199	for (iPage = 0; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE)
1200	{
1201	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1202	struct vm_area_struct *vma = find_vma(current->mm, ulAddrCur);
1203	if (!vma)
1204	break;
1205	#endif
1206
1207	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
1208	rc = remap_pfn_range(vma, ulAddrCur, page_to_pfn(papPages[iPage]), PAGE_SIZE, pgFlags);
1209	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1210	rc = remap_page_range(vma, ulAddrCur, page_to_phys(papPages[iPage]), PAGE_SIZE, pgFlags);
1211	#else /* 2.4 */
1212	rc = remap_page_range(ulAddrCur, page_to_phys(papPages[iPage]), PAGE_SIZE, pgFlags);
1213	#endif
1214	if (rc)
1215	break;
1216	}
1217
1218	/*
1219	* Successful?
1220	*/
1221	if (iPage >= cPages)
1222	{
1223	up_write(&current->mm->mmap_sem);
1224	return (void *)ulAddr;
1225	}
1226
1227	/* no, cleanup! */
1228	if (rc)
1229	dprintf(("VBoxSupDrvMapUser: remap_[page\|pfn]_range failed! rc=%d\n", rc));
1230	else
1231	dprintf(("VBoxSupDrvMapUser: find_vma failed!\n"));
1232
1233	MY_DO_MUNMAP(current->mm, ulAddr, cPages * PAGE_SIZE);
1234	}
1235	else
1236	{
1237	dprintf(("supdrvOSContAllocOne: do_mmap failed ulAddr=%#lx\n", ulAddr));
1238	rc = SUPDRV_ERR_NO_MEMORY;
1239	}
1240	up_write(&current->mm->mmap_sem);
1241
1242	return NULL;
1243	}
1244
1245
1246	/**
1247	* Initializes the GIP.
1248	*
1249	* @returns negative errno.
1250	* @param pDevExt Instance data. GIP stuff may be updated.
1251	*/
1252	static int VBoxSupDrvInitGip(PSUPDRVDEVEXT pDevExt)
1253	{
1254	struct page *pPage;
1255	dma_addr_t HCPhys;
1256	PSUPGLOBALINFOPAGE pGip;
1257	dprintf(("VBoxSupDrvInitGip:\n"));
1258
1259	/*
1260	* Allocate the page.
1261	*/
1262	pPage = alloc_pages(GFP_USER, 0);
1263	if (!pPage)
1264	{
1265	dprintf(("VBoxSupDrvInitGip: failed to allocate the GIP page\n"));
1266	return -ENOMEM;
1267	}
1268
1269	/*
1270	* Lock the page.
1271	*/
1272	SetPageReserved(pPage);
1273	g_pGipPage = pPage;
1274
1275	/*
1276	* Call common initialization routine.
1277	*/
1278	HCPhys = page_to_phys(pPage);
1279	pGip = (PSUPGLOBALINFOPAGE)page_address(pPage);
1280	pDevExt->ulLastJiffies = jiffies;
1281	#ifdef TICK_NSEC
1282	pDevExt->u64LastMonotime = (uint64_t)pDevExt->ulLastJiffies * TICK_NSEC;
1283	dprintf(("VBoxSupDrvInitGIP: TICK_NSEC=%ld HZ=%d jiffies=%ld now=%lld\n",
1284	TICK_NSEC, HZ, pDevExt->ulLastJiffies, pDevExt->u64LastMonotime));
1285	#else
1286	pDevExt->u64LastMonotime = (uint64_t)pDevExt->ulLastJiffies * (1000000 / HZ);
1287	dprintf(("VBoxSupDrvInitGIP: TICK_NSEC=%d HZ=%d jiffies=%ld now=%lld\n",
1288	(int)(1000000 / HZ), HZ, pDevExt->ulLastJiffies, pDevExt->u64LastMonotime));
1289	#endif
1290	supdrvGipInit(pDevExt, pGip, HCPhys, pDevExt->u64LastMonotime,
1291	HZ <= 1000 ? HZ : 1000);
1292
1293	/*
1294	* Initialize the timer.
1295	*/
1296	init_timer(&g_GipTimer);
1297	g_GipTimer.data = (unsigned long)pDevExt;
1298	g_GipTimer.function = VBoxSupGipTimer;
1299	g_GipTimer.expires = jiffies;
1300
1301	return 0;
1302	}
1303
1304
1305	/**
1306	* Terminates the GIP.
1307	*
1308	* @returns negative errno.
1309	* @param pDevExt Instance data. GIP stuff may be updated.
1310	*/
1311	static int VBoxSupDrvTermGip(PSUPDRVDEVEXT pDevExt)
1312	{
1313	struct page *pPage;
1314	PSUPGLOBALINFOPAGE pGip;
1315	dprintf(("VBoxSupDrvTermGip:\n"));
1316
1317	/*
1318	* Delete the timer if it's pending.
1319	*/
1320	if (timer_pending(&g_GipTimer))
1321	del_timer(&g_GipTimer);
1322
1323	/*
1324	* Uninitialize the content.
1325	*/
1326	pGip = pDevExt->pGip;
1327	pDevExt->pGip = NULL;
1328	if (pGip)
1329	supdrvGipTerm(pGip);
1330
1331	/*
1332	* Free the page.
1333	*/
1334	pPage = g_pGipPage;
1335	g_pGipPage = NULL;
1336	if (pPage)
1337	{
1338	ClearPageReserved(pPage);
1339	__free_pages(pPage, 0);
1340	}
1341
1342	return 0;
1343	}
1344
1345	/**
1346	* Timer callback function.
1347	* The ulUser parameter is the device extension pointer.
1348	*/
1349	static void VBoxSupGipTimer(unsigned long ulUser)
1350	{
1351	PSUPDRVDEVEXT pDevExt = (PSUPDRVDEVEXT)ulUser;
1352	unsigned long ulNow = jiffies;
1353	unsigned long ulDiff = ulNow - pDevExt->ulLastJiffies;
1354	pDevExt->ulLastJiffies = ulNow;
1355	#ifdef TICK_NSEC
1356	pDevExt->u64LastMonotime += ulDiff * TICK_NSEC;
1357	#else
1358	pDevExt->u64LastMonotime += ulDiff * (1000000 / HZ);
1359	#endif
1360	supdrvGipUpdate(pDevExt->pGip, pDevExt->u64LastMonotime);
1361	mod_timer(&g_GipTimer, jiffies + (HZ <= 1000 ? 0 : ONE_MSEC_IN_JIFFIES));
1362	}
1363
1364
1365	/**
1366	* Maps the GIP into user space.
1367	*
1368	* @returns negative errno.
1369	* @param pDevExt Instance data.
1370	*/
1371	int VBOXCALL supdrvOSGipMap(PSUPDRVDEVEXT pDevExt, PCSUPGLOBALINFOPAGE *ppGip)
1372	{
1373	int rc = 0;
1374	unsigned long ulAddr;
1375	unsigned long HCPhys = pDevExt->HCPhysGip;
1376	pgprot_t pgFlags;
1377	pgprot_val(pgFlags) = _PAGE_PRESENT \| _PAGE_USER;
1378	dprintf2(("supdrvOSGipMap: ppGip=%p\n", ppGip));
1379
1380	/*
1381	* Allocate user space mapping and put the physical pages into it.
1382	*/
1383	down_write(&current->mm->mmap_sem);
1384	ulAddr = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ, MAP_SHARED \| MAP_ANONYMOUS, 0);
1385	if (!(ulAddr & ~PAGE_MASK))
1386	{
1387	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0) && !defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1388	int rc2 = remap_page_range(ulAddr, HCPhys, PAGE_SIZE, pgFlags);
1389	#else
1390	int rc2 = 0;
1391	struct vm_area_struct *vma = find_vma(current->mm, ulAddr);
1392	if (vma)
1393	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11)
1394	rc2 = remap_page_range(vma, ulAddr, HCPhys, PAGE_SIZE, pgFlags);
1395	#else
1396	rc2 = remap_pfn_range(vma, ulAddr, HCPhys >> PAGE_SHIFT, PAGE_SIZE, pgFlags);
1397	#endif
1398	else
1399	{
1400	rc = SUPDRV_ERR_NO_MEMORY;
1401	dprintf(("supdrvOSGipMap: no vma found for ulAddr=%#lx!\n", ulAddr));
1402	}
1403	#endif
1404	if (rc2)
1405	{
1406	rc = SUPDRV_ERR_NO_MEMORY;
1407	dprintf(("supdrvOSGipMap: remap_page_range failed rc2=%d\n", rc2));
1408	}
1409	}
1410	else
1411	{
1412	dprintf(("supdrvOSGipMap: do_mmap failed ulAddr=%#lx\n", ulAddr));
1413	rc = SUPDRV_ERR_NO_MEMORY;
1414	}
1415	up_write(&current->mm->mmap_sem); /* not quite sure when to give this up. */
1416
1417	/*
1418	* Success?
1419	*/
1420	if (!rc)
1421	{
1422	*ppGip = (PCSUPGLOBALINFOPAGE)ulAddr;
1423	dprintf2(("supdrvOSGipMap: ppGip=%p\n", *ppGip));
1424	return 0;
1425	}
1426
1427	/*
1428	* Failure, cleanup and be gone.
1429	*/
1430	if (ulAddr & ~PAGE_MASK)
1431	{
1432	down_write(&current->mm->mmap_sem);
1433	MY_DO_MUNMAP(current->mm, ulAddr, PAGE_SIZE);
1434	up_write(&current->mm->mmap_sem);
1435	}
1436
1437	dprintf2(("supdrvOSGipMap: returns %d\n", rc));
1438	return rc;
1439	}
1440
1441
1442	/**
1443	* Maps the GIP into user space.
1444	*
1445	* @returns negative errno.
1446	* @param pDevExt Instance data.
1447	*/
1448	int VBOXCALL supdrvOSGipUnmap(PSUPDRVDEVEXT pDevExt, PCSUPGLOBALINFOPAGE pGip)
1449	{
1450	dprintf2(("supdrvOSGipUnmap: pGip=%p\n", pGip));
1451	if (current->mm)
1452	{
1453	down_write(&current->mm->mmap_sem);
1454	MY_DO_MUNMAP(current->mm, (unsigned long)pGip, PAGE_SIZE);
1455	up_write(&current->mm->mmap_sem);
1456	}
1457	dprintf2(("supdrvOSGipUnmap: returns 0\n"));
1458	return 0;
1459	}
1460
1461
1462	/**
1463	* Resumes the GIP updating.
1464	*
1465	* @param pDevExt Instance data.
1466	*/
1467	void VBOXCALL supdrvOSGipResume(PSUPDRVDEVEXT pDevExt)
1468	{
1469	dprintf2(("supdrvOSGipResume:\n"));
1470	mod_timer(&g_GipTimer, jiffies);
1471	}
1472
1473
1474	/**
1475	* Suspends the GIP updating.
1476	*
1477	* @param pDevExt Instance data.
1478	*/
1479	void VBOXCALL supdrvOSGipSuspend(PSUPDRVDEVEXT pDevExt)
1480	{
1481	dprintf2(("supdrvOSGipSuspend:\n"));
1482	if (timer_pending(&g_GipTimer))
1483	del_timer(&g_GipTimer);
1484	}
1485
1486
1487	/**
1488	* Converts a supdrv error code to an linux error code.
1489	*
1490	* @returns corresponding linux error code.
1491	* @param rc supdrv error code (SUPDRV_ERR_* defines).
1492	*/
1493	static int VBoxSupDrvErr2LinuxErr(int rc)
1494	{
1495	switch (rc)
1496	{
1497	case 0: return 0;
1498	case SUPDRV_ERR_GENERAL_FAILURE: return -EACCES;
1499	case SUPDRV_ERR_INVALID_PARAM: return -EINVAL;
1500	case SUPDRV_ERR_INVALID_MAGIC: return -EILSEQ;
1501	case SUPDRV_ERR_INVALID_HANDLE: return -ENXIO;
1502	case SUPDRV_ERR_INVALID_POINTER: return -EFAULT;
1503	case SUPDRV_ERR_LOCK_FAILED: return -ENOLCK;
1504	case SUPDRV_ERR_ALREADY_LOADED: return -EEXIST;
1505	case SUPDRV_ERR_PERMISSION_DENIED: return -EPERM;
1506	case SUPDRV_ERR_VERSION_MISMATCH: return -ENOSYS;
1507	}
1508
1509	return -EPERM;
1510	}
1511
1512
1513	RTDECL(int) SUPR0Printf(const char *pszFormat, ...)
1514	{
1515	#if 1
1516	va_list args;
1517	char szMsg[512];
1518
1519	va_start(args, pszFormat);
1520	vsnprintf(szMsg, sizeof(szMsg) - 1, pszFormat, args);
1521	szMsg[sizeof(szMsg) - 1] = '\0';
1522	printk("%s", szMsg);
1523	va_end(args);
1524	#else
1525	/* forward to printf - needs some more GCC hacking to fix ebp... */
1526	__asm__ __volatile__ ("mov %0, %esp\n\t"
1527	"jmp %1\n\t",
1528	:: "r" ((uintptr_t)&pszFormat - 4),
1529	"m" (printk));
1530	#endif
1531	return 0;
1532	}
1533
1534
1535	/** Runtime assert implementation for Linux Ring-0. */
1536	RTDECL(void) AssertMsg1(const char pszExpr, unsigned uLine, const char pszFile, const char *pszFunction)
1537	{
1538	printk("!!Assertion Failed!!\n"
1539	"Expression: %s\n"
1540	"Location : %s(%d) %s\n",
1541	pszExpr, pszFile, uLine, pszFunction);
1542	}
1543
1544
1545	/** Runtime assert implementation for Linux Ring-0. */
1546	RTDECL(void) AssertMsg2(const char *pszFormat, ...)
1547	{ /* forwarder. */
1548	va_list ap;
1549	char msg[256];
1550
1551	va_start(ap, pszFormat);
1552	vsnprintf(msg, sizeof(msg) - 1, pszFormat, ap);
1553	msg[sizeof(msg) - 1] = '\0';
1554	printk("%s", msg);
1555	va_end(ap);
1556	}
1557
1558
1559	/* GCC C++ hack. */
1560	unsigned __gxx_personality_v0 = 0xcccccccc;
1561
1562
1563	module_init(VBoxSupDrvInit);
1564	module_exit(VBoxSupDrvUnload);
1565
1566	MODULE_AUTHOR("InnoTek Systemberatung GmbH");
1567	MODULE_DESCRIPTION("VirtualBox Support Driver");
1568	MODULE_LICENSE("GPL");

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

source: vbox/trunk/src/VBox/HostDrivers/Support/linux/SUPDrv-linux.c@ 417

Download in other formats: