VBoxPci-linux.c@ 75599

Last change on this file since 75599 was 75599, checked in by vboxsync, 6 years ago
HostDrivers: fix some warnings on old Linux kernels and add missing functions in VBoxNetFlt.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 34.2 KB

Line
1	/* $Id: VBoxPci-linux.c 75599 2018-11-19 20:39:52Z vboxsync $ */
2	/** @file
3	* VBoxPci - PCI Driver (Host), Linux Specific Code.
4	*/
5
6	/*
7	* Copyright (C) 2011-2017 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*
17	* The contents of this file may alternatively be used under the terms
18	* of the Common Development and Distribution License Version 1.0
19	* (CDDL) only, as it comes in the "COPYING.CDDL" file of the
20	* VirtualBox OSE distribution, in which case the provisions of the
21	* CDDL are applicable instead of those of the GPL.
22	*
23	* You may elect to license modified versions of this file under the
24	* terms and conditions of either the GPL or the CDDL or both.
25	*/
26
27
28	/*********************************************************************************************************************************
29	* Header Files *
30	*********************************************************************************************************************************/
31	#include "the-linux-kernel.h"
32	#include "version-generated.h"
33	#include "revision-generated.h"
34	#include "product-generated.h"
35
36	#define LOG_GROUP LOG_GROUP_DEV_PCI_RAW
37	#include <VBox/log.h>
38	#include <VBox/err.h>
39	#include <iprt/process.h>
40	#include <iprt/initterm.h>
41	#include <iprt/string.h>
42	#include <iprt/mem.h>
43
44	#include "../VBoxPciInternal.h"
45
46	#ifdef VBOX_WITH_IOMMU
47	# include <linux/dmar.h>
48	# include <linux/intel-iommu.h>
49	# include <linux/pci.h>
50	# if LINUX_VERSION_CODE < KERNEL_VERSION(3, 1, 0) && \
51	(LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 41) \|\| LINUX_VERSION_CODE >= KERNEL_VERSION(3, 0, 0))
52	# include <asm/amd_iommu.h>
53	# else
54	# include <linux/amd-iommu.h>
55	# endif
56	# if LINUX_VERSION_CODE < KERNEL_VERSION(3, 2, 0)
57	# define IOMMU_PRESENT() iommu_found()
58	# define IOMMU_DOMAIN_ALLOC() iommu_domain_alloc()
59	# else
60	# define IOMMU_PRESENT() iommu_present(&pci_bus_type)
61	# define IOMMU_DOMAIN_ALLOC() iommu_domain_alloc(&pci_bus_type)
62	# endif
63	#endif /* VBOX_WITH_IOMMU */
64
65
66	/*********************************************************************************************************************************
67	* Internal Functions *
68	*********************************************************************************************************************************/
69	static int VBoxPciLinuxInit(void);
70	static void VBoxPciLinuxUnload(void);
71
72
73	/*********************************************************************************************************************************
74	* Global Variables *
75	*********************************************************************************************************************************/
76	static VBOXRAWPCIGLOBALS g_VBoxPciGlobals;
77
78	module_init(VBoxPciLinuxInit);
79	module_exit(VBoxPciLinuxUnload);
80
81	MODULE_AUTHOR(VBOX_VENDOR);
82	MODULE_DESCRIPTION(VBOX_PRODUCT " PCI access Driver");
83	MODULE_LICENSE("GPL");
84	#ifdef MODULE_VERSION
85	MODULE_VERSION(VBOX_VERSION_STRING " r" RT_XSTR(VBOX_SVN_REV));
86	#endif
87
88
89	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
90	# define PCI_DEV_GET(v,d,p) pci_get_device(v,d,p)
91	# define PCI_DEV_PUT(x) pci_dev_put(x)
92	#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 17, 0)
93	/* assume the domain number to be zero - exactly the same assumption of
94	* pci_get_bus_and_slot()
95	*/
96	# define PCI_DEV_GET_SLOT(bus, devfn) pci_get_domain_bus_and_slot(0, bus, devfn)
97	#else
98	# define PCI_DEV_GET_SLOT(bus, devfn) pci_get_bus_and_slot(bus, devfn)
99	#endif
100	#else
101	# define PCI_DEV_GET(v,d,p) pci_find_device(v,d,p)
102	# define PCI_DEV_PUT(x) do { } while (0)
103	# define PCI_DEV_GET_SLOT(bus, devfn) pci_find_slot(bus, devfn)
104	#endif
105
106	/**
107	* Name of module used to attach to the host PCI device, when
108	* PCI device passthrough is used.
109	*/
110	#define PCI_STUB_MODULE "pci-stub"
111	/* For some reasons my kernel names module for find_module() this way,
112	* while device name seems to be above one.
113	*/
114	#define PCI_STUB_MODULE_NAME "pci_stub"
115
116	/**
117	* Our driver name.
118	*/
119	#define DRIVER_NAME "vboxpci"
120
121	/*
122	* Currently we keep the device bound to pci stub driver, so
123	* dev_printk() &co would report that instead of our name. They also
124	* expect non-NULL dev pointer in older kernels.
125	*/
126	#define vbpci_printk(level, pdev, format, arg...) \
127	printk(level DRIVER_NAME "%s%s: " format, \
128	pdev ? " " : "", pdev ? pci_name(pdev) : "", \
129	## arg)
130
131
132	/**
133	* Initialize module.
134	*
135	* @returns appropriate status code.
136	*/
137	static int __init VBoxPciLinuxInit(void)
138	{
139	int rc;
140	/*
141	* Initialize IPRT.
142	*/
143	rc = RTR0Init(0);
144
145	if (RT_FAILURE(rc))
146	goto error;
147
148
149	LogRel(("VBoxPciLinuxInit\n"));
150
151	RT_ZERO(g_VBoxPciGlobals);
152
153	rc = vboxPciInit(&g_VBoxPciGlobals);
154	if (RT_FAILURE(rc))
155	{
156	LogRel(("cannot do VBoxPciInit: %Rc\n", rc));
157	goto error;
158	}
159
160	#if defined(CONFIG_PCI_STUB)
161	/* nothing to do, pci_stub module part of the kernel */
162	g_VBoxPciGlobals.fPciStubModuleAvail = true;
163
164	#elif defined(CONFIG_PCI_STUB_MODULE)
165	if (request_module(PCI_STUB_MODULE) == 0)
166	{
167	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 30)
168	/* find_module() is static before Linux 2.6.30 */
169	mutex_lock(&module_mutex);
170	g_VBoxPciGlobals.pciStubModule = find_module(PCI_STUB_MODULE_NAME);
171	mutex_unlock(&module_mutex);
172	if (g_VBoxPciGlobals.pciStubModule)
173	{
174	if (try_module_get(g_VBoxPciGlobals.pciStubModule))
175	g_VBoxPciGlobals.fPciStubModuleAvail = true;
176	}
177	else
178	printk(KERN_INFO "vboxpci: find_module %s failed\n", PCI_STUB_MODULE);
179	# endif
180	}
181	else
182	printk(KERN_INFO "vboxpci: cannot load %s\n", PCI_STUB_MODULE);
183
184	#else
185	printk(KERN_INFO "vboxpci: %s module not available, cannot detach PCI devices\n",
186	PCI_STUB_MODULE);
187	#endif
188
189	#ifdef VBOX_WITH_IOMMU
190	if (IOMMU_PRESENT())
191	printk(KERN_INFO "vboxpci: IOMMU found\n");
192	else
193	printk(KERN_INFO "vboxpci: IOMMU not found (not registered)\n");
194	#else
195	printk(KERN_INFO "vboxpci: IOMMU not found (not compiled)\n");
196	#endif
197
198	return 0;
199
200	error:
201	return -RTErrConvertToErrno(rc);
202	}
203
204	/**
205	* Unload the module.
206	*/
207	static void __exit VBoxPciLinuxUnload(void)
208	{
209	LogRel(("VBoxPciLinuxLinuxUnload\n"));
210
211	/*
212	* Undo the work done during start (in reverse order).
213	*/
214	vboxPciShutdown(&g_VBoxPciGlobals);
215
216	RTR0Term();
217
218	if (g_VBoxPciGlobals.pciStubModule)
219	{
220	module_put(g_VBoxPciGlobals.pciStubModule);
221	g_VBoxPciGlobals.pciStubModule = NULL;
222	}
223
224	Log(("VBoxPciLinuxUnload - done\n"));
225	}
226
227	static int vboxPciLinuxDevRegisterWithIommu(PVBOXRAWPCIINS pIns)
228	{
229	#ifdef VBOX_WITH_IOMMU
230	int rc = VINF_SUCCESS;
231	struct pci_dev *pPciDev = pIns->pPciDev;
232	PVBOXRAWPCIDRVVM pData = VBOX_DRV_VMDATA(pIns);
233	IPRT_LINUX_SAVE_EFL_AC();
234
235	if (RT_LIKELY(pData))
236	{
237	if (RT_LIKELY(pData->pIommuDomain))
238	{
239	/** @todo KVM checks IOMMU_CAP_CACHE_COHERENCY and sets
240	* flag IOMMU_CACHE later used when mapping physical
241	* addresses, which could improve performance.
242	*/
243	int rcLnx = iommu_attach_device(pData->pIommuDomain, &pPciDev->dev);
244	if (!rcLnx)
245	{
246	vbpci_printk(KERN_DEBUG, pPciDev, "attached to IOMMU\n");
247	pIns->fIommuUsed = true;
248	rc = VINF_SUCCESS;
249	}
250	else
251	{
252	vbpci_printk(KERN_DEBUG, pPciDev, "failed to attach to IOMMU, error %d\n", rcLnx);
253	rc = VERR_INTERNAL_ERROR;
254	}
255	}
256	else
257	{
258	vbpci_printk(KERN_DEBUG, pIns->pPciDev, "cannot attach to IOMMU, no domain\n");
259	rc = VERR_NOT_FOUND;
260	}
261	}
262	else
263	{
264	vbpci_printk(KERN_DEBUG, pPciDev, "cannot attach to IOMMU, no VM data\n");
265	rc = VERR_INVALID_PARAMETER;
266	}
267
268	IPRT_LINUX_RESTORE_EFL_AC();
269	return rc;
270	#else
271	return VERR_NOT_SUPPORTED;
272	#endif
273	}
274
275	static int vboxPciLinuxDevUnregisterWithIommu(PVBOXRAWPCIINS pIns)
276	{
277	#ifdef VBOX_WITH_IOMMU
278	int rc = VINF_SUCCESS;
279	struct pci_dev *pPciDev = pIns->pPciDev;
280	PVBOXRAWPCIDRVVM pData = VBOX_DRV_VMDATA(pIns);
281	IPRT_LINUX_SAVE_EFL_AC();
282
283	if (RT_LIKELY(pData))
284	{
285	if (RT_LIKELY(pData->pIommuDomain))
286	{
287	if (pIns->fIommuUsed)
288	{
289	iommu_detach_device(pData->pIommuDomain, &pIns->pPciDev->dev);
290	vbpci_printk(KERN_DEBUG, pPciDev, "detached from IOMMU\n");
291	pIns->fIommuUsed = false;
292	}
293	}
294	else
295	{
296	vbpci_printk(KERN_DEBUG, pPciDev,
297	"cannot detach from IOMMU, no domain\n");
298	rc = VERR_NOT_FOUND;
299	}
300	}
301	else
302	{
303	vbpci_printk(KERN_DEBUG, pPciDev,
304	"cannot detach from IOMMU, no VM data\n");
305	rc = VERR_INVALID_PARAMETER;
306	}
307
308	IPRT_LINUX_RESTORE_EFL_AC();
309	return rc;
310	#else
311	return VERR_NOT_SUPPORTED;
312	#endif
313	}
314
315	static int vboxPciLinuxDevReset(PVBOXRAWPCIINS pIns)
316	{
317	int rc = VINF_SUCCESS;
318	IPRT_LINUX_SAVE_EFL_AC();
319
320	if (RT_LIKELY(pIns->pPciDev))
321	{
322	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 28)
323	if (pci_reset_function(pIns->pPciDev))
324	{
325	vbpci_printk(KERN_DEBUG, pIns->pPciDev,
326	"pci_reset_function() failed\n");
327	rc = VERR_INTERNAL_ERROR;
328	}
329	#else
330	rc = VERR_NOT_SUPPORTED;
331	#endif
332	}
333	else
334	rc = VERR_INVALID_PARAMETER;
335
336	IPRT_LINUX_RESTORE_EFL_AC();
337	return rc;
338	}
339
340	static struct file* vboxPciFileOpen(const char* path, int flags)
341	{
342	struct file* filp = NULL;
343	int err = 0;
344
345	filp = filp_open(path, flags, 0);
346
347	if (IS_ERR(filp))
348	{
349	err = PTR_ERR(filp);
350	printk(KERN_DEBUG "vboxPciFileOpen: error %d\n", err);
351	return NULL;
352	}
353
354	if (!filp->f_op \|\| !filp->f_op->write)
355	{
356	printk(KERN_DEBUG "Not writable FS\n");
357	filp_close(filp, NULL);
358	return NULL;
359	}
360
361	return filp;
362	}
363
364	static void vboxPciFileClose(struct file* file)
365	{
366	filp_close(file, NULL);
367	}
368
369	static int vboxPciFileWrite(struct file* file, unsigned long long offset, unsigned char* data, unsigned int size)
370	{
371	int ret;
372	mm_segment_t fs_save;
373
374	fs_save = get_fs();
375	set_fs(get_ds());
376	#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 14, 0)
377	ret = kernel_write(file, data, size, &offset);
378	#else
379	ret = vfs_write(file, data, size, &offset);
380	#endif
381	set_fs(fs_save);
382	if (ret < 0)
383	printk(KERN_DEBUG "vboxPciFileWrite: error %d\n", ret);
384
385	return ret;
386	}
387
388	static int vboxPciLinuxDevDetachHostDriver(PVBOXRAWPCIINS pIns)
389	{
390	struct pci_dev *pPciDev = NULL;
391	uint8_t uBus = (pIns->HostPciAddress) >> 8;
392	uint8_t uDevFn = (pIns->HostPciAddress) & 0xff;
393	const char* currentDriver;
394	uint16_t uVendor, uDevice;
395	bool fDetach = 0;
396
397	if (!g_VBoxPciGlobals.fPciStubModuleAvail)
398	{
399	printk(KERN_INFO "vboxpci: stub module %s not detected: cannot detach\n",
400	PCI_STUB_MODULE);
401	return VERR_ACCESS_DENIED;
402	}
403
404	pPciDev = PCI_DEV_GET_SLOT(uBus, uDevFn);
405
406	if (!pPciDev)
407	{
408	printk(KERN_INFO "vboxpci: device at %02x:%02x.%d not found\n",
409	uBus, uDevFn>>3, uDevFn&7);
410	return VERR_NOT_FOUND;
411	}
412
413	uVendor = pPciDev->vendor;
414	uDevice = pPciDev->device;
415
416	currentDriver = pPciDev->driver ? pPciDev->driver->name : NULL;
417
418	printk(KERN_DEBUG "vboxpci: detected device: %04x:%04x at %02x:%02x.%d, driver %s\n",
419	uVendor, uDevice, uBus, uDevFn>>3, uDevFn&7,
420	currentDriver ? currentDriver : "<none>");
421
422	fDetach = (currentDriver == NULL \|\| (strcmp(currentDriver, PCI_STUB_MODULE) != 0));
423
424	/* Init previous driver data. */
425	pIns->szPrevDriver[0] = '\0';
426
427	if (fDetach && currentDriver)
428	{
429	/* Dangerous: if device name for some reasons contains slashes - arbitrary file could be written to. */
430	if (strchr(currentDriver, '/') != 0)
431	{
432	printk(KERN_DEBUG "vboxpci: ERROR: %s contains invalid symbols\n", currentDriver);
433	return VERR_ACCESS_DENIED;
434	}
435	/** @todo RTStrCopy not exported. */
436	strncpy(pIns->szPrevDriver, currentDriver, sizeof(pIns->szPrevDriver) - 1);
437	pIns->szPrevDriver[sizeof(pIns->szPrevDriver) - 1] = '\0';
438	}
439
440	PCI_DEV_PUT(pPciDev);
441	pPciDev = NULL;
442
443	if (fDetach)
444	{
445	char* szCmdBuf;
446	char* szFileBuf;
447	struct file* pFile;
448	int iCmdLen;
449	const int cMaxBuf = 128;
450	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
451	const struct cred *pOldCreds;
452	struct cred *pNewCreds;
453	#endif
454
455	/*
456	* Now perform kernel analog of:
457	*
458	* echo -n "10de 040a" > /sys/bus/pci/drivers/pci-stub/new_id
459	* echo -n 0000:03:00.0 > /sys/bus/pci/drivers/nvidia/unbind
460	* echo -n 0000:03:00.0 > /sys/bus/pci/drivers/pci-stub/bind
461	*
462	* We do this way, as this interface is presumingly more stable than
463	* in-kernel ones.
464	*/
465	szCmdBuf = kmalloc(cMaxBuf, GFP_KERNEL);
466	szFileBuf = kmalloc(cMaxBuf, GFP_KERNEL);
467	if (!szCmdBuf \|\| !szFileBuf)
468	goto done;
469
470	/* Somewhat ugly hack - override current credentials */
471	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
472	pNewCreds = prepare_creds();
473	if (!pNewCreds)
474	goto done;
475
476	# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 5, 0)
477	pNewCreds->fsuid = GLOBAL_ROOT_UID;
478	# else
479	pNewCreds->fsuid = 0;
480	# endif
481	pOldCreds = override_creds(pNewCreds);
482	#endif
483
484	RTStrPrintf(szFileBuf, cMaxBuf,
485	"/sys/bus/pci/drivers/%s/new_id",
486	PCI_STUB_MODULE);
487	pFile = vboxPciFileOpen(szFileBuf, O_WRONLY);
488	if (pFile)
489	{
490	iCmdLen = RTStrPrintf(szCmdBuf, cMaxBuf,
491	"%04x %04x",
492	uVendor, uDevice);
493	/* Don't write trailing \0 */
494	vboxPciFileWrite(pFile, 0, szCmdBuf, iCmdLen);
495	vboxPciFileClose(pFile);
496	}
497	else
498	printk(KERN_DEBUG "vboxpci: cannot open %s\n", szFileBuf);
499
500	iCmdLen = RTStrPrintf(szCmdBuf, cMaxBuf,
501	"0000:%02x:%02x.%d",
502	uBus, uDevFn>>3, uDevFn&7);
503
504	/* Unbind if bound to smth */
505	if (pIns->szPrevDriver[0])
506	{
507	RTStrPrintf(szFileBuf, cMaxBuf,
508	"/sys/bus/pci/drivers/%s/unbind",
509	pIns->szPrevDriver);
510	pFile = vboxPciFileOpen(szFileBuf, O_WRONLY);
511	if (pFile)
512	{
513
514	/* Don't write trailing \0 */
515	vboxPciFileWrite(pFile, 0, szCmdBuf, iCmdLen);
516	vboxPciFileClose(pFile);
517	}
518	else
519	printk(KERN_DEBUG "vboxpci: cannot open %s\n", szFileBuf);
520	}
521
522	RTStrPrintf(szFileBuf, cMaxBuf,
523	"/sys/bus/pci/drivers/%s/bind",
524	PCI_STUB_MODULE);
525	pFile = vboxPciFileOpen(szFileBuf, O_WRONLY);
526	if (pFile)
527	{
528	/* Don't write trailing \0 */
529	vboxPciFileWrite(pFile, 0, szCmdBuf, iCmdLen);
530	vboxPciFileClose(pFile);
531	}
532	else
533	printk(KERN_DEBUG "vboxpci: cannot open %s\n", szFileBuf);
534
535	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
536	revert_creds(pOldCreds);
537	put_cred(pNewCreds);
538	#endif
539
540	done:
541	kfree(szCmdBuf);
542	kfree(szFileBuf);
543	}
544
545	return 0;
546	}
547
548	static int vboxPciLinuxDevReattachHostDriver(PVBOXRAWPCIINS pIns)
549	{
550	struct pci_dev *pPciDev = pIns->pPciDev;
551
552	if (!pPciDev)
553	return VINF_SUCCESS;
554
555	if (pIns->szPrevDriver[0])
556	{
557	char* szCmdBuf;
558	char* szFileBuf;
559	struct file* pFile;
560	int iCmdLen;
561	const int cMaxBuf = 128;
562	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
563	const struct cred *pOldCreds;
564	struct cred *pNewCreds;
565	#endif
566	uint8_t uBus = (pIns->HostPciAddress) >> 8;
567	uint8_t uDevFn = (pIns->HostPciAddress) & 0xff;
568
569	vbpci_printk(KERN_DEBUG, pPciDev,
570	"reattaching old host driver %s\n", pIns->szPrevDriver);
571	/*
572	* Now perform kernel analog of:
573	*
574	* echo -n 0000:03:00.0 > /sys/bus/pci/drivers/pci-stub/unbind
575	* echo -n 0000:03:00.0 > /sys/bus/pci/drivers/nvidia/bind
576	*/
577	szCmdBuf = kmalloc(cMaxBuf, GFP_KERNEL);
578	szFileBuf = kmalloc(cMaxBuf, GFP_KERNEL);
579
580	if (!szCmdBuf \|\| !szFileBuf)
581	goto done;
582
583	iCmdLen = RTStrPrintf(szCmdBuf, cMaxBuf,
584	"0000:%02x:%02x.%d",
585	uBus, uDevFn>>3, uDevFn&7);
586
587	/* Somewhat ugly hack - override current credentials */
588	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
589	pNewCreds = prepare_creds();
590	if (!pNewCreds)
591	goto done;
592
593	# if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 5, 0)
594	pNewCreds->fsuid = GLOBAL_ROOT_UID;
595	# else
596	pNewCreds->fsuid = 0;
597	# endif
598	pOldCreds = override_creds(pNewCreds);
599	#endif
600	RTStrPrintf(szFileBuf, cMaxBuf,
601	"/sys/bus/pci/drivers/%s/unbind",
602	PCI_STUB_MODULE);
603	pFile = vboxPciFileOpen(szFileBuf, O_WRONLY);
604	if (pFile)
605	{
606
607	/* Don't write trailing \0 */
608	vboxPciFileWrite(pFile, 0, szCmdBuf, iCmdLen);
609	vboxPciFileClose(pFile);
610	}
611	else
612	printk(KERN_DEBUG "vboxpci: cannot open %s\n", szFileBuf);
613
614	RTStrPrintf(szFileBuf, cMaxBuf,
615	"/sys/bus/pci/drivers/%s/bind",
616	pIns->szPrevDriver);
617	pFile = vboxPciFileOpen(szFileBuf, O_WRONLY);
618	if (pFile)
619	{
620
621	/* Don't write trailing \0 */
622	vboxPciFileWrite(pFile, 0, szCmdBuf, iCmdLen);
623	vboxPciFileClose(pFile);
624	pIns->szPrevDriver[0] = '\0';
625	}
626	else
627	printk(KERN_DEBUG "vboxpci: cannot open %s\n", szFileBuf);
628
629	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 29)
630	revert_creds(pOldCreds);
631	put_cred(pNewCreds);
632	#endif
633
634	done:
635	kfree(szCmdBuf);
636	kfree(szFileBuf);
637	}
638
639	return VINF_SUCCESS;
640	}
641
642	DECLHIDDEN(int) vboxPciOsDevInit(PVBOXRAWPCIINS pIns, uint32_t fFlags)
643	{
644	struct pci_dev *pPciDev = NULL;
645	int rc = VINF_SUCCESS;
646	IPRT_LINUX_SAVE_EFL_AC();
647
648	if (fFlags & PCIRAWDRIVERRFLAG_DETACH_HOST_DRIVER)
649	{
650	rc = vboxPciLinuxDevDetachHostDriver(pIns);
651	if (RT_FAILURE(rc))
652	{
653	printk(KERN_DEBUG "Cannot detach host driver for device %x: %d\n",
654	pIns->HostPciAddress, rc);
655	}
656	}
657
658	if (RT_SUCCESS(rc))
659	{
660	pPciDev = PCI_DEV_GET_SLOT((pIns->HostPciAddress) >> 8,
661	(pIns->HostPciAddress) & 0xff);
662
663	if (RT_LIKELY(pPciDev))
664	{
665	int rcLnx = pci_enable_device(pPciDev);
666
667	if (!rcLnx)
668	{
669	pIns->pPciDev = pPciDev;
670	vbpci_printk(KERN_DEBUG, pPciDev, "%s\n", __func__);
671
672	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 1)
673	if (pci_enable_msi(pPciDev) == 0)
674	pIns->fMsiUsed = true;
675	#endif
676
677	/** @todo
678	* pci_enable_msix(pPciDev, entries, nvec)
679	*
680	* In fact, if device uses interrupts, and cannot be forced to use MSI or MSI-X
681	* we have to refuse using it, as we cannot work with shared PCI interrupts (unless we're lucky
682	* to grab unshared PCI interrupt).
683	*/
684	}
685	else
686	rc = RTErrConvertFromErrno(RT_ABS(rcLnx));
687	}
688	else
689	rc = VERR_NOT_FOUND;
690	}
691
692	IPRT_LINUX_RESTORE_EFL_AC();
693	return rc;
694	}
695
696	DECLHIDDEN(int) vboxPciOsDevDeinit(PVBOXRAWPCIINS pIns, uint32_t fFlags)
697	{
698	int rc = VINF_SUCCESS;
699	struct pci_dev *pPciDev = pIns->pPciDev;
700	IPRT_LINUX_SAVE_EFL_AC();
701
702	vbpci_printk(KERN_DEBUG, pPciDev, "%s\n", __func__);
703
704	if (RT_LIKELY(pPciDev))
705	{
706	int iRegion;
707	for (iRegion = 0; iRegion < 7; ++iRegion)
708	{
709	if (pIns->aRegionR0Mapping[iRegion])
710	{
711	iounmap(pIns->aRegionR0Mapping[iRegion]);
712	pIns->aRegionR0Mapping[iRegion] = 0;
713	pci_release_region(pPciDev, iRegion);
714	}
715	}
716
717	vboxPciLinuxDevUnregisterWithIommu(pIns);
718
719	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 1)
720	if (pIns->fMsiUsed)
721	pci_disable_msi(pPciDev);
722	#endif
723	// pci_disable_msix(pPciDev);
724	pci_disable_device(pPciDev);
725	vboxPciLinuxDevReattachHostDriver(pIns);
726
727	PCI_DEV_PUT(pPciDev);
728	pIns->pPciDev = NULL;
729	}
730	else
731	rc = VERR_INVALID_PARAMETER;
732
733	IPRT_LINUX_RESTORE_EFL_AC();
734	return rc;
735	}
736
737	DECLHIDDEN(int) vboxPciOsDevDestroy(PVBOXRAWPCIINS pIns)
738	{
739	return VINF_SUCCESS;
740	}
741
742	DECLHIDDEN(int) vboxPciOsDevGetRegionInfo(PVBOXRAWPCIINS pIns,
743	int32_t iRegion,
744	RTHCPHYS *pRegionStart,
745	uint64_t *pu64RegionSize,
746	bool *pfPresent,
747	uint32_t *pfFlags)
748	{
749	int rc = VINF_SUCCESS;
750	struct pci_dev *pPciDev = pIns->pPciDev;
751	IPRT_LINUX_SAVE_EFL_AC();
752
753	if (RT_LIKELY(pPciDev))
754	{
755	int fFlags = pci_resource_flags(pPciDev, iRegion);
756
757	if ( ((fFlags & (IORESOURCE_MEM \| IORESOURCE_IO)) == 0)
758	\|\| ((fFlags & IORESOURCE_DISABLED) != 0))
759	{
760	*pfPresent = false;
761	rc = VERR_INVALID_PARAMETER;
762	}
763	else
764	{
765	uint32_t fResFlags = 0;
766	*pfPresent = true;
767
768	if (fFlags & IORESOURCE_MEM)
769	fResFlags \|= PCIRAW_ADDRESS_SPACE_MEM;
770
771	if (fFlags & IORESOURCE_IO)
772	fResFlags \|= PCIRAW_ADDRESS_SPACE_IO;
773
774	#ifdef IORESOURCE_MEM_64
775	if (fFlags & IORESOURCE_MEM_64)
776	fResFlags \|= PCIRAW_ADDRESS_SPACE_BAR64;
777	#endif
778
779	if (fFlags & IORESOURCE_PREFETCH)
780	fResFlags \|= PCIRAW_ADDRESS_SPACE_MEM_PREFETCH;
781
782	*pfFlags = fResFlags;
783	*pRegionStart = pci_resource_start(pPciDev, iRegion);
784	*pu64RegionSize = pci_resource_len (pPciDev, iRegion);
785
786	vbpci_printk(KERN_DEBUG, pPciDev,
787	"region %d: %s %llx+%lld\n",
788	iRegion, (fFlags & IORESOURCE_MEM) ? "mmio" : "pio",
789	pRegionStart, pu64RegionSize);
790	}
791	}
792	else
793	{
794	*pfPresent = false;
795	rc = VERR_INVALID_PARAMETER;
796	}
797
798	IPRT_LINUX_RESTORE_EFL_AC();
799	return rc;
800	}
801
802	DECLHIDDEN(int) vboxPciOsDevMapRegion(PVBOXRAWPCIINS pIns,
803	int32_t iRegion,
804	RTHCPHYS RegionStart,
805	uint64_t u64RegionSize,
806	uint32_t fFlags,
807	RTR0PTR *pRegionBase)
808	{
809	int rc = VINF_SUCCESS;
810	struct pci_dev *pPciDev = pIns->pPciDev;
811	IPRT_LINUX_SAVE_EFL_AC();
812
813	if (!pPciDev \|\| iRegion < 0 \|\| iRegion > 0)
814	{
815	if (pPciDev)
816	vbpci_printk(KERN_DEBUG, pPciDev, "invalid region %d\n", iRegion);
817
818	IPRT_LINUX_RESTORE_EFL_AC();
819	return VERR_INVALID_PARAMETER;
820	}
821
822	vbpci_printk(KERN_DEBUG, pPciDev, "reg=%d start=%llx size=%lld\n",
823	iRegion, RegionStart, u64RegionSize);
824
825	if ( (pci_resource_flags(pPciDev, iRegion) & IORESOURCE_IO)
826	\|\| RegionStart != pci_resource_start(pPciDev, iRegion)
827	\|\| u64RegionSize != pci_resource_len(pPciDev, iRegion))
828	{
829	IPRT_LINUX_RESTORE_EFL_AC();
830	return VERR_INVALID_PARAMETER;
831	}
832
833	/*
834	* XXX: Current code never calls unmap. To avoid leaking mappings
835	* only request and map resources once.
836	*/
837	if (!pIns->aRegionR0Mapping[iRegion])
838	{
839	int rcLnx;
840	*pRegionBase = pIns->aRegionR0Mapping[iRegion];
841
842	rcLnx = pci_request_region(pPciDev, iRegion, "vboxpci");
843	if (!rcLnx)
844	{
845	/* For now no caching, try to optimize later. */
846	RTR0PTR R0PtrMapping = ioremap_nocache(pci_resource_start(pPciDev, iRegion),
847	pci_resource_len(pPciDev, iRegion));
848
849	if (R0PtrMapping != NIL_RTR0PTR)
850	pIns->aRegionR0Mapping[iRegion] = R0PtrMapping;
851	else
852	{
853	vbpci_printk(KERN_DEBUG, pPciDev, "ioremap_nocache() failed\n");
854	pci_release_region(pPciDev, iRegion);
855	rc = VERR_MAP_FAILED;
856	}
857	}
858	else
859	rc = VERR_RESOURCE_BUSY;
860	}
861
862	if (RT_SUCCESS(rc))
863	*pRegionBase = pIns->aRegionR0Mapping[iRegion];
864
865	IPRT_LINUX_RESTORE_EFL_AC();
866	return rc;
867	}
868
869	DECLHIDDEN(int) vboxPciOsDevUnmapRegion(PVBOXRAWPCIINS pIns,
870	int32_t iRegion,
871	RTHCPHYS RegionStart,
872	uint64_t u64RegionSize,
873	RTR0PTR RegionBase)
874	{
875	/* XXX: Current code never calls unmap. */
876	return VERR_NOT_IMPLEMENTED;
877	}
878
879	DECLHIDDEN(int) vboxPciOsDevPciCfgWrite(PVBOXRAWPCIINS pIns, uint32_t Register, PCIRAWMEMLOC *pValue)
880	{
881	struct pci_dev *pPciDev = pIns->pPciDev;
882	int rc = VINF_SUCCESS;
883	IPRT_LINUX_SAVE_EFL_AC();
884
885	if (RT_LIKELY(pPciDev))
886	{
887	switch (pValue->cb)
888	{
889	case 1:
890	pci_write_config_byte(pPciDev, Register, pValue->u.u8);
891	break;
892	case 2:
893	pci_write_config_word(pPciDev, Register, pValue->u.u16);
894	break;
895	case 4:
896	pci_write_config_dword(pPciDev, Register, pValue->u.u32);
897	break;
898	}
899	}
900	else
901	rc = VERR_INVALID_PARAMETER;
902
903	IPRT_LINUX_RESTORE_EFL_AC();
904	return rc;
905	}
906
907	DECLHIDDEN(int) vboxPciOsDevPciCfgRead(PVBOXRAWPCIINS pIns, uint32_t Register, PCIRAWMEMLOC *pValue)
908	{
909	struct pci_dev *pPciDev = pIns->pPciDev;
910	int rc = VINF_SUCCESS;
911
912	if (RT_LIKELY(pPciDev))
913	{
914	IPRT_LINUX_SAVE_EFL_AC();
915
916	switch (pValue->cb)
917	{
918	case 1:
919	pci_read_config_byte(pPciDev, Register, &pValue->u.u8);
920	break;
921	case 2:
922	pci_read_config_word(pPciDev, Register, &pValue->u.u16);
923	break;
924	case 4:
925	pci_read_config_dword(pPciDev, Register, &pValue->u.u32);
926	break;
927	}
928
929	IPRT_LINUX_RESTORE_EFL_AC();
930	}
931	else
932	rc = VERR_INVALID_PARAMETER;
933
934	return rc;
935	}
936
937	/**
938	* Interrupt service routine.
939	*
940	* @returns In 2.6 we indicate whether we've handled the IRQ or not.
941	*
942	* @param iIrq The IRQ number.
943	* @param pvDevId The device ID, a pointer to PVBOXRAWPCIINS.
944	* @param pRegs Register set. Removed in 2.6.19.
945	*/
946	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 19) && !defined(DOXYGEN_RUNNING)
947	static irqreturn_t vboxPciOsIrqHandler(int iIrq, void *pvDevId)
948	#else
949	static irqreturn_t vboxPciOsIrqHandler(int iIrq, void pvDevId, struct pt_regs pRegs)
950	#endif
951	{
952	PVBOXRAWPCIINS pIns = (PVBOXRAWPCIINS)pvDevId;
953	bool fTaken = true;
954
955	if (pIns && pIns->IrqHandler.pfnIrqHandler)
956	fTaken = pIns->IrqHandler.pfnIrqHandler(pIns->IrqHandler.pIrqContext, iIrq);
957	#ifndef VBOX_WITH_SHARED_PCI_INTERRUPTS
958	/* If we don't allow interrupts sharing, we consider all interrupts as non-shared, thus targetted to us. */
959	fTaken = true;
960	#endif
961
962	return fTaken;
963	}
964
965	DECLHIDDEN(int) vboxPciOsDevRegisterIrqHandler(PVBOXRAWPCIINS pIns, PFNRAWPCIISR pfnHandler, void* pIrqContext, int32_t *piHostIrq)
966	{
967	int rc;
968	int32_t iIrq = pIns->pPciDev->irq;
969	IPRT_LINUX_SAVE_EFL_AC();
970
971	if (iIrq == 0)
972	{
973	vbpci_printk(KERN_NOTICE, pIns->pPciDev, "no irq assigned\n");
974	IPRT_LINUX_RESTORE_EFL_AC();
975	return VERR_INVALID_PARAMETER;
976	}
977
978	rc = request_irq(iIrq,
979	vboxPciOsIrqHandler,
980	#ifdef VBOX_WITH_SHARED_PCI_INTERRUPTS
981	/* Allow interrupts sharing. */
982	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20)
983	IRQF_SHARED,
984	# else
985	SA_SHIRQ,
986	# endif
987
988	#else
989
990	/* We don't allow interrupts sharing */
991	/* XXX overhaul */
992	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 20) && LINUX_VERSION_CODE < KERNEL_VERSION(4, 1, 0)
993	IRQF_DISABLED, /* keep irqs disabled when calling the action handler */
994	# else
995	0,
996	# endif
997	#endif
998	DRIVER_NAME,
999	pIns);
1000	if (rc)
1001	{
1002	vbpci_printk(KERN_DEBUG, pIns->pPciDev,
1003	"could not request irq %d, error %d\n", iIrq, rc);
1004	IPRT_LINUX_RESTORE_EFL_AC();
1005	return VERR_RESOURCE_BUSY;
1006	}
1007
1008	vbpci_printk(KERN_DEBUG, pIns->pPciDev, "got irq %d\n", iIrq);
1009	*piHostIrq = iIrq;
1010
1011	IPRT_LINUX_RESTORE_EFL_AC();
1012	return VINF_SUCCESS;
1013	}
1014
1015	DECLHIDDEN(int) vboxPciOsDevUnregisterIrqHandler(PVBOXRAWPCIINS pIns, int32_t iHostIrq)
1016	{
1017	IPRT_LINUX_SAVE_EFL_AC();
1018
1019	vbpci_printk(KERN_DEBUG, pIns->pPciDev, "freeing irq %d\n", iHostIrq);
1020	free_irq(iHostIrq, pIns);
1021
1022	IPRT_LINUX_RESTORE_EFL_AC();
1023	return VINF_SUCCESS;
1024	}
1025
1026	DECLHIDDEN(int) vboxPciOsDevPowerStateChange(PVBOXRAWPCIINS pIns, PCIRAWPOWERSTATE aState)
1027	{
1028	int rc;
1029
1030	switch (aState)
1031	{
1032	case PCIRAW_POWER_ON:
1033	vbpci_printk(KERN_DEBUG, pIns->pPciDev, "PCIRAW_POWER_ON\n");
1034	/* Reset device, just in case. */
1035	vboxPciLinuxDevReset(pIns);
1036	/* register us with IOMMU */
1037	rc = vboxPciLinuxDevRegisterWithIommu(pIns);
1038	break;
1039	case PCIRAW_POWER_RESET:
1040	vbpci_printk(KERN_DEBUG, pIns->pPciDev, "PCIRAW_POWER_RESET\n");
1041	rc = vboxPciLinuxDevReset(pIns);
1042	break;
1043	case PCIRAW_POWER_OFF:
1044	vbpci_printk(KERN_DEBUG, pIns->pPciDev, "PCIRAW_POWER_OFF\n");
1045	/* unregister us from IOMMU */
1046	rc = vboxPciLinuxDevUnregisterWithIommu(pIns);
1047	break;
1048	case PCIRAW_POWER_SUSPEND:
1049	vbpci_printk(KERN_DEBUG, pIns->pPciDev, "PCIRAW_POWER_SUSPEND\n");
1050	rc = VINF_SUCCESS;
1051	/// @todo what do we do here?
1052	break;
1053	case PCIRAW_POWER_RESUME:
1054	vbpci_printk(KERN_DEBUG, pIns->pPciDev, "PCIRAW_POWER_RESUME\n");
1055	rc = VINF_SUCCESS;
1056	/// @todo what do we do here?
1057	break;
1058	default:
1059	vbpci_printk(KERN_DEBUG, pIns->pPciDev, "unknown power state %u\n", aState);
1060	/* to make compiler happy */
1061	rc = VERR_NOT_SUPPORTED;
1062	break;
1063	}
1064
1065	return rc;
1066	}
1067
1068
1069	#ifdef VBOX_WITH_IOMMU
1070	/** Callback for FNRAWPCICONTIGPHYSMEMINFO. */
1071	static DECLCALLBACK(int) vboxPciOsContigMemInfo(PRAWPCIPERVM pVmCtx, RTHCPHYS HostStart, RTGCPHYS GuestStart,
1072	uint64_t cMemSize, PCIRAWMEMINFOACTION Action)
1073	{
1074	struct iommu_domain* domain = ((PVBOXRAWPCIDRVVM)(pVmCtx->pDriverData))->pIommuDomain;
1075	int rc = VINF_SUCCESS;
1076	IPRT_LINUX_SAVE_EFL_AC();
1077
1078	switch (Action)
1079	{
1080	case PCIRAW_MEMINFO_MAP:
1081	{
1082	int flags, r;
1083
1084	if (iommu_iova_to_phys(domain, GuestStart))
1085	break;
1086
1087	flags = IOMMU_READ \| IOMMU_WRITE;
1088	/** @todo flags \|= IOMMU_CACHE; */
1089
1090	r = iommu_map(domain, GuestStart, HostStart, get_order(cMemSize), flags);
1091	if (r)
1092	{
1093	printk(KERN_ERR "vboxPciOsContigMemInfo:"
1094	"iommu failed to map pfn=%llx\n", HostStart);
1095	rc = VERR_GENERAL_FAILURE;
1096	break;
1097	}
1098	rc = VINF_SUCCESS;
1099	break;
1100	}
1101	case PCIRAW_MEMINFO_UNMAP:
1102	{
1103	int order;
1104	order = iommu_unmap(domain, GuestStart, get_order(cMemSize));
1105	NOREF(order);
1106	break;
1107	}
1108
1109	default:
1110	printk(KERN_DEBUG "Unsupported action: %d\n", (int)Action);
1111	rc = VERR_NOT_SUPPORTED;
1112	break;
1113	}
1114
1115	IPRT_LINUX_RESTORE_EFL_AC();
1116	return rc;
1117	}
1118	#endif
1119
1120	DECLHIDDEN(int) vboxPciOsInitVm(PVBOXRAWPCIDRVVM pThis, PVM pVM, PRAWPCIPERVM pVmData)
1121	{
1122	int rc = VINF_SUCCESS;
1123
1124	#ifdef VBOX_WITH_IOMMU
1125	IPRT_LINUX_SAVE_EFL_AC();
1126
1127	if (IOMMU_PRESENT())
1128	{
1129	pThis->pIommuDomain = IOMMU_DOMAIN_ALLOC();
1130	if (!pThis->pIommuDomain)
1131	{
1132	vbpci_printk(KERN_DEBUG, NULL, "cannot allocate IOMMU domain\n");
1133	rc = VERR_NO_MEMORY;
1134	}
1135	else
1136	{
1137	pVmData->pfnContigMemInfo = vboxPciOsContigMemInfo;
1138
1139	vbpci_printk(KERN_DEBUG, NULL, "created IOMMU domain %p\n",
1140	pThis->pIommuDomain);
1141	}
1142	}
1143
1144	IPRT_LINUX_RESTORE_EFL_AC();
1145	#endif
1146	return rc;
1147	}
1148
1149	DECLHIDDEN(void) vboxPciOsDeinitVm(PVBOXRAWPCIDRVVM pThis, PVM pVM)
1150	{
1151	#ifdef VBOX_WITH_IOMMU
1152	IPRT_LINUX_SAVE_EFL_AC();
1153
1154	if (pThis->pIommuDomain)
1155	{
1156	vbpci_printk(KERN_DEBUG, NULL, "freeing IOMMU domain %p\n",
1157	pThis->pIommuDomain);
1158	iommu_domain_free(pThis->pIommuDomain);
1159	pThis->pIommuDomain = NULL;
1160	}
1161
1162	IPRT_LINUX_RESTORE_EFL_AC();
1163	#endif
1164	}

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

source: vbox/trunk/src/VBox/HostDrivers/VBoxPci/linux/VBoxPci-linux.c@ 75599

Download in other formats: