memobj-r0drv-linux.c@ 25645

Last change on this file since 25645 was 23611, checked in by vboxsync, 15 years ago
typo
Property svn:eol-style set to `native` Property svn:keywords set to `Id Rev`
File size: 44.8 KB

Line
1	/* $Revision: 23611 $ */
2	/** @file
3	* IPRT - Ring-0 Memory Objects, Linux.
4	*/
5
6	/*
7	* Copyright (C) 2006-2007 Sun Microsystems, Inc.
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	* Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa
27	* Clara, CA 95054 USA or visit http://www.sun.com if you need
28	* additional information or have any questions.
29	*/
30
31
32	/*******************************************************************************
33	* Header Files *
34	*******************************************************************************/
35	#include "the-linux-kernel.h"
36
37	#include <iprt/memobj.h>
38	#include <iprt/alloc.h>
39	#include <iprt/assert.h>
40	#include <iprt/log.h>
41	#include <iprt/process.h>
42	#include <iprt/string.h>
43	#include "internal/memobj.h"
44
45
46	/*******************************************************************************
47	* Defined Constants And Macros *
48	*******************************************************************************/
49	/* early 2.6 kernels */
50	#ifndef PAGE_SHARED_EXEC
51	# define PAGE_SHARED_EXEC PAGE_SHARED
52	#endif
53	#ifndef PAGE_READONLY_EXEC
54	# define PAGE_READONLY_EXEC PAGE_READONLY
55	#endif
56
57	/*
58	* 2.6.29+ kernels don't work with remap_pfn_range() anymore because
59	* track_pfn_vma_new() is apparently not defined for non-RAM pages.
60	* It should be safe to use vm_insert_page() older kernels as well.
61	*/
62	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 23)
63	# define VBOX_USE_INSERT_PAGE
64	#endif
65	#if defined(CONFIG_X86_PAE) \
66	&& ( HAVE_26_STYLE_REMAP_PAGE_RANGE \
67	\|\| (LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) && LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11)))
68	# define VBOX_USE_PAE_HACK
69	#endif
70
71
72	/*******************************************************************************
73	* Structures and Typedefs *
74	*******************************************************************************/
75	/**
76	* The Darwin version of the memory object structure.
77	*/
78	typedef struct RTR0MEMOBJLNX
79	{
80	/** The core structure. */
81	RTR0MEMOBJINTERNAL Core;
82	/** Set if the allocation is contiguous.
83	* This means it has to be given back as one chunk. */
84	bool fContiguous;
85	/** Set if we've vmap'ed thed memory into ring-0. */
86	bool fMappedToRing0;
87	/** The pages in the apPages array. */
88	size_t cPages;
89	/** Array of struct page pointers. (variable size) */
90	struct page *apPages[1];
91	} RTR0MEMOBJLNX, *PRTR0MEMOBJLNX;
92
93
94	/**
95	* Helper that converts from a RTR0PROCESS handle to a linux task.
96	*
97	* @returns The corresponding Linux task.
98	* @param R0Process IPRT ring-0 process handle.
99	*/
100	struct task_struct *rtR0ProcessToLinuxTask(RTR0PROCESS R0Process)
101	{
102	/** @todo fix rtR0ProcessToLinuxTask!! */
103	return R0Process == RTR0ProcHandleSelf() ? current : NULL;
104	}
105
106
107	/**
108	* Compute order. Some functions allocate 2^order pages.
109	*
110	* @returns order.
111	* @param cPages Number of pages.
112	*/
113	static int rtR0MemObjLinuxOrder(size_t cPages)
114	{
115	int iOrder;
116	size_t cTmp;
117
118	for (iOrder = 0, cTmp = cPages; cTmp >>= 1; ++iOrder)
119	;
120	if (cPages & ~((size_t)1 << iOrder))
121	++iOrder;
122
123	return iOrder;
124	}
125
126
127	/**
128	* Converts from RTMEM_PROT_* to Linux PAGE_*.
129	*
130	* @returns Linux page protection constant.
131	* @param fProt The IPRT protection mask.
132	* @param fKernel Whether it applies to kernel or user space.
133	*/
134	static pgprot_t rtR0MemObjLinuxConvertProt(unsigned fProt, bool fKernel)
135	{
136	switch (fProt)
137	{
138	default:
139	AssertMsgFailed(("%#x %d\n", fProt, fKernel));
140	case RTMEM_PROT_NONE:
141	return PAGE_NONE;
142
143	case RTMEM_PROT_READ:
144	return fKernel ? PAGE_KERNEL_RO : PAGE_READONLY;
145
146	case RTMEM_PROT_WRITE:
147	case RTMEM_PROT_WRITE \| RTMEM_PROT_READ:
148	return fKernel ? PAGE_KERNEL : PAGE_SHARED;
149
150	case RTMEM_PROT_EXEC:
151	case RTMEM_PROT_EXEC \| RTMEM_PROT_READ:
152	#if defined(RT_ARCH_X86) \|\| defined(RT_ARCH_AMD64)
153	if (fKernel)
154	{
155	pgprot_t fPg = MY_PAGE_KERNEL_EXEC;
156	pgprot_val(fPg) &= ~_PAGE_RW;
157	return fPg;
158	}
159	return PAGE_READONLY_EXEC;
160	#else
161	return fKernel ? MY_PAGE_KERNEL_EXEC : PAGE_READONLY_EXEC;
162	#endif
163
164	case RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC:
165	case RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC \| RTMEM_PROT_READ:
166	return fKernel ? MY_PAGE_KERNEL_EXEC : PAGE_SHARED_EXEC;
167	}
168	}
169
170
171	/**
172	* Internal worker that allocates physical pages and creates the memory object for them.
173	*
174	* @returns IPRT status code.
175	* @param ppMemLnx Where to store the memory object pointer.
176	* @param enmType The object type.
177	* @param cb The number of bytes to allocate.
178	* @param fFlagsLnx The page allocation flags (GPFs).
179	* @param fContiguous Whether the allocation must be contiguous.
180	*/
181	static int rtR0MemObjLinuxAllocPages(PRTR0MEMOBJLNX *ppMemLnx, RTR0MEMOBJTYPE enmType, size_t cb, unsigned fFlagsLnx, bool fContiguous)
182	{
183	size_t iPage;
184	size_t cPages = cb >> PAGE_SHIFT;
185	struct page *paPages;
186
187	/*
188	* Allocate a memory object structure that's large enough to contain
189	* the page pointer array.
190	*/
191	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_OFFSETOF(RTR0MEMOBJLNX, apPages[cPages]), enmType, NULL, cb);
192	if (!pMemLnx)
193	return VERR_NO_MEMORY;
194	pMemLnx->cPages = cPages;
195
196	/*
197	* Allocate the pages.
198	* For small allocations we'll try contiguous first and then fall back on page by page.
199	*/
200	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
201	if ( fContiguous
202	\|\| cb <= PAGE_SIZE * 2)
203	{
204	#ifdef VBOX_USE_INSERT_PAGE
205	paPages = alloc_pages(fFlagsLnx \| __GFP_COMP, rtR0MemObjLinuxOrder(cb >> PAGE_SHIFT));
206	#else
207	paPages = alloc_pages(fFlagsLnx, rtR0MemObjLinuxOrder(cb >> PAGE_SHIFT));
208	#endif
209	if (paPages)
210	{
211	fContiguous = true;
212	for (iPage = 0; iPage < cPages; iPage++)
213	pMemLnx->apPages[iPage] = &paPages[iPage];
214	}
215	else if (fContiguous)
216	{
217	rtR0MemObjDelete(&pMemLnx->Core);
218	return VERR_NO_MEMORY;
219	}
220	}
221
222	if (!fContiguous)
223	{
224	for (iPage = 0; iPage < cPages; iPage++)
225	{
226	pMemLnx->apPages[iPage] = alloc_page(fFlagsLnx);
227	if (RT_UNLIKELY(!pMemLnx->apPages[iPage]))
228	{
229	while (iPage-- > 0)
230	__free_page(pMemLnx->apPages[iPage]);
231	rtR0MemObjDelete(&pMemLnx->Core);
232	return VERR_NO_MEMORY;
233	}
234	}
235	}
236
237	#else /* < 2.4.22 */
238	/** @todo figure out why we didn't allocate page-by-page on 2.4.21 and older... */
239	paPages = alloc_pages(fFlagsLnx, rtR0MemObjLinuxOrder(cb >> PAGE_SHIFT));
240	if (!paPages)
241	{
242	rtR0MemObjDelete(&pMemLnx->Core);
243	return VERR_NO_MEMORY;
244	}
245	for (iPage = 0; iPage < cPages; iPage++)
246	{
247	pMemLnx->apPages[iPage] = &paPages[iPage];
248	MY_SET_PAGES_EXEC(pMemLnx->apPages[iPage], 1);
249	if (PageHighMem(pMemLnx->apPages[iPage]))
250	BUG();
251	}
252
253	fContiguous = true;
254	#endif /* < 2.4.22 */
255	pMemLnx->fContiguous = fContiguous;
256
257	/*
258	* Reserve the pages.
259	*/
260	for (iPage = 0; iPage < cPages; iPage++)
261	SetPageReserved(pMemLnx->apPages[iPage]);
262
263	*ppMemLnx = pMemLnx;
264	return VINF_SUCCESS;
265	}
266
267
268	/**
269	* Frees the physical pages allocated by the rtR0MemObjLinuxAllocPages() call.
270	*
271	* This method does NOT free the object.
272	*
273	* @param pMemLnx The object which physical pages should be freed.
274	*/
275	static void rtR0MemObjLinuxFreePages(PRTR0MEMOBJLNX pMemLnx)
276	{
277	size_t iPage = pMemLnx->cPages;
278	if (iPage > 0)
279	{
280	/*
281	* Restore the page flags.
282	*/
283	while (iPage-- > 0)
284	{
285	ClearPageReserved(pMemLnx->apPages[iPage]);
286	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
287	#else
288	MY_SET_PAGES_NOEXEC(pMemLnx->apPages[iPage], 1);
289	#endif
290	}
291
292	/*
293	* Free the pages.
294	*/
295	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
296	if (!pMemLnx->fContiguous)
297	{
298	iPage = pMemLnx->cPages;
299	while (iPage-- > 0)
300	__free_page(pMemLnx->apPages[iPage]);
301	}
302	else
303	#endif
304	__free_pages(pMemLnx->apPages[0], rtR0MemObjLinuxOrder(pMemLnx->cPages));
305
306	pMemLnx->cPages = 0;
307	}
308	}
309
310
311	/**
312	* Maps the allocation into ring-0.
313	*
314	* This will update the RTR0MEMOBJLNX::Core.pv and RTR0MEMOBJ::fMappedToRing0 members.
315	*
316	* Contiguous mappings that isn't in 'high' memory will already be mapped into kernel
317	* space, so we'll use that mapping if possible. If execute access is required, we'll
318	* play safe and do our own mapping.
319	*
320	* @returns IPRT status code.
321	* @param pMemLnx The linux memory object to map.
322	* @param fExecutable Whether execute access is required.
323	*/
324	static int rtR0MemObjLinuxVMap(PRTR0MEMOBJLNX pMemLnx, bool fExecutable)
325	{
326	int rc = VINF_SUCCESS;
327
328	/*
329	* Choose mapping strategy.
330	*/
331	bool fMustMap = fExecutable
332	\|\| !pMemLnx->fContiguous;
333	if (!fMustMap)
334	{
335	size_t iPage = pMemLnx->cPages;
336	while (iPage-- > 0)
337	if (PageHighMem(pMemLnx->apPages[iPage]))
338	{
339	fMustMap = true;
340	break;
341	}
342	}
343
344	Assert(!pMemLnx->Core.pv);
345	Assert(!pMemLnx->fMappedToRing0);
346
347	if (fMustMap)
348	{
349	/*
350	* Use vmap - 2.4.22 and later.
351	*/
352	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
353	pgprot_t fPg;
354	pgprot_val(fPg) = _PAGE_PRESENT \| _PAGE_RW;
355	# ifdef _PAGE_NX
356	if (!fExecutable)
357	pgprot_val(fPg) \|= _PAGE_NX;
358	# endif
359
360	# ifdef VM_MAP
361	pMemLnx->Core.pv = vmap(&pMemLnx->apPages[0], pMemLnx->cPages, VM_MAP, fPg);
362	# else
363	pMemLnx->Core.pv = vmap(&pMemLnx->apPages[0], pMemLnx->cPages, VM_ALLOC, fPg);
364	# endif
365	if (pMemLnx->Core.pv)
366	pMemLnx->fMappedToRing0 = true;
367	else
368	rc = VERR_MAP_FAILED;
369	#else /* < 2.4.22 */
370	rc = VERR_NOT_SUPPORTED;
371	#endif
372	}
373	else
374	{
375	/*
376	* Use the kernel RAM mapping.
377	*/
378	pMemLnx->Core.pv = phys_to_virt(page_to_phys(pMemLnx->apPages[0]));
379	Assert(pMemLnx->Core.pv);
380	}
381
382	return rc;
383	}
384
385
386	/**
387	* Undos what rtR0MemObjLinuxVMap() did.
388	*
389	* @param pMemLnx The linux memory object.
390	*/
391	static void rtR0MemObjLinuxVUnmap(PRTR0MEMOBJLNX pMemLnx)
392	{
393	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
394	if (pMemLnx->fMappedToRing0)
395	{
396	Assert(pMemLnx->Core.pv);
397	vunmap(pMemLnx->Core.pv);
398	pMemLnx->fMappedToRing0 = false;
399	}
400	#else /* < 2.4.22 */
401	Assert(!pMemLnx->fMappedToRing0);
402	#endif
403	pMemLnx->Core.pv = NULL;
404	}
405
406
407	int rtR0MemObjNativeFree(RTR0MEMOBJ pMem)
408	{
409	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
410
411	/*
412	* Release any memory that we've allocated or locked.
413	*/
414	switch (pMemLnx->Core.enmType)
415	{
416	case RTR0MEMOBJTYPE_LOW:
417	case RTR0MEMOBJTYPE_PAGE:
418	case RTR0MEMOBJTYPE_CONT:
419	case RTR0MEMOBJTYPE_PHYS:
420	case RTR0MEMOBJTYPE_PHYS_NC:
421	rtR0MemObjLinuxVUnmap(pMemLnx);
422	rtR0MemObjLinuxFreePages(pMemLnx);
423	break;
424
425	case RTR0MEMOBJTYPE_LOCK:
426	if (pMemLnx->Core.u.Lock.R0Process != NIL_RTR0PROCESS)
427	{
428	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
429	size_t iPage;
430	Assert(pTask);
431	if (pTask && pTask->mm)
432	down_read(&pTask->mm->mmap_sem);
433
434	iPage = pMemLnx->cPages;
435	while (iPage-- > 0)
436	{
437	if (!PageReserved(pMemLnx->apPages[iPage]))
438	SetPageDirty(pMemLnx->apPages[iPage]);
439	page_cache_release(pMemLnx->apPages[iPage]);
440	}
441
442	if (pTask && pTask->mm)
443	up_read(&pTask->mm->mmap_sem);
444	}
445	/* else: kernel memory - nothing to do here. */
446	break;
447
448	case RTR0MEMOBJTYPE_RES_VIRT:
449	Assert(pMemLnx->Core.pv);
450	if (pMemLnx->Core.u.ResVirt.R0Process != NIL_RTR0PROCESS)
451	{
452	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
453	Assert(pTask);
454	if (pTask && pTask->mm)
455	{
456	down_write(&pTask->mm->mmap_sem);
457	MY_DO_MUNMAP(pTask->mm, (unsigned long)pMemLnx->Core.pv, pMemLnx->Core.cb);
458	up_write(&pTask->mm->mmap_sem);
459	}
460	}
461	else
462	{
463	vunmap(pMemLnx->Core.pv);
464
465	Assert(pMemLnx->cPages == 1 && pMemLnx->apPages[0] != NULL);
466	__free_page(pMemLnx->apPages[0]);
467	pMemLnx->apPages[0] = NULL;
468	pMemLnx->cPages = 0;
469	}
470	pMemLnx->Core.pv = NULL;
471	break;
472
473	case RTR0MEMOBJTYPE_MAPPING:
474	Assert(pMemLnx->cPages == 0); Assert(pMemLnx->Core.pv);
475	if (pMemLnx->Core.u.ResVirt.R0Process != NIL_RTR0PROCESS)
476	{
477	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
478	Assert(pTask);
479	if (pTask && pTask->mm)
480	{
481	down_write(&pTask->mm->mmap_sem);
482	MY_DO_MUNMAP(pTask->mm, (unsigned long)pMemLnx->Core.pv, pMemLnx->Core.cb);
483	up_write(&pTask->mm->mmap_sem);
484	}
485	}
486	else
487	vunmap(pMemLnx->Core.pv);
488	pMemLnx->Core.pv = NULL;
489	break;
490
491	default:
492	AssertMsgFailed(("enmType=%d\n", pMemLnx->Core.enmType));
493	return VERR_INTERNAL_ERROR;
494	}
495	return VINF_SUCCESS;
496	}
497
498
499	int rtR0MemObjNativeAllocPage(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable)
500	{
501	PRTR0MEMOBJLNX pMemLnx;
502	int rc;
503
504	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
505	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_PAGE, cb, GFP_HIGHUSER, false /* non-contiguous */);
506	#else
507	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_PAGE, cb, GFP_USER, false /* non-contiguous */);
508	#endif
509	if (RT_SUCCESS(rc))
510	{
511	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
512	if (RT_SUCCESS(rc))
513	{
514	*ppMem = &pMemLnx->Core;
515	return rc;
516	}
517
518	rtR0MemObjLinuxFreePages(pMemLnx);
519	rtR0MemObjDelete(&pMemLnx->Core);
520	}
521
522	return rc;
523	}
524
525
526	int rtR0MemObjNativeAllocLow(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable)
527	{
528	PRTR0MEMOBJLNX pMemLnx;
529	int rc;
530
531	/* Try to avoid GFP_DMA. GFM_DMA32 was introduced with Linux 2.6.15. */
532	#if (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
533	/* ZONE_DMA32: 0-4GB */
534	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, GFP_DMA32, false /* non-contiguous */);
535	if (RT_FAILURE(rc))
536	#endif
537	#ifdef RT_ARCH_AMD64
538	/* ZONE_DMA: 0-16MB */
539	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, GFP_DMA, false /* non-contiguous */);
540	#else
541	# ifdef CONFIG_X86_PAE
542	# endif
543	/* ZONE_NORMAL: 0-896MB */
544	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, GFP_USER, false /* non-contiguous */);
545	#endif
546	if (RT_SUCCESS(rc))
547	{
548	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
549	if (RT_SUCCESS(rc))
550	{
551	*ppMem = &pMemLnx->Core;
552	return rc;
553	}
554
555	rtR0MemObjLinuxFreePages(pMemLnx);
556	rtR0MemObjDelete(&pMemLnx->Core);
557	}
558
559	return rc;
560	}
561
562
563	int rtR0MemObjNativeAllocCont(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable)
564	{
565	PRTR0MEMOBJLNX pMemLnx;
566	int rc;
567
568	#if (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
569	/* ZONE_DMA32: 0-4GB */
570	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, GFP_DMA32, true /* contiguous */);
571	if (RT_FAILURE(rc))
572	#endif
573	#ifdef RT_ARCH_AMD64
574	/* ZONE_DMA: 0-16MB */
575	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, GFP_DMA, true /* contiguous */);
576	#else
577	/* ZONE_NORMAL (32-bit hosts): 0-896MB */
578	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, GFP_USER, true /* contiguous */);
579	#endif
580	if (RT_SUCCESS(rc))
581	{
582	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
583	if (RT_SUCCESS(rc))
584	{
585	#if defined(RT_STRICT) && (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_HIGHMEM64G))
586	size_t iPage = pMemLnx->cPages;
587	while (iPage-- > 0)
588	Assert(page_to_phys(pMemLnx->apPages[iPage]) < _4G);
589	#endif
590	pMemLnx->Core.u.Cont.Phys = page_to_phys(pMemLnx->apPages[0]);
591	*ppMem = &pMemLnx->Core;
592	return rc;
593	}
594
595	rtR0MemObjLinuxFreePages(pMemLnx);
596	rtR0MemObjDelete(&pMemLnx->Core);
597	}
598
599	return rc;
600	}
601
602
603	/**
604	* Worker for rtR0MemObjLinuxAllocPhysSub that tries one allocation strategy.
605	*
606	* @returns IPRT status.
607	* @param ppMemLnx Where to
608	* @param enmType The object type.
609	* @param cb The size of the allocation.
610	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
611	* @param fGfp The Linux GFP flags to use for the allocation.
612	*/
613	static int rtR0MemObjLinuxAllocPhysSub2(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType, size_t cb, RTHCPHYS PhysHighest, unsigned fGfp)
614	{
615	PRTR0MEMOBJLNX pMemLnx;
616	int rc;
617
618	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, enmType, cb, fGfp,
619	enmType == RTR0MEMOBJTYPE_PHYS /* contiguous / non-contiguous */);
620	if (RT_FAILURE(rc))
621	return rc;
622
623	/*
624	* Check the addresses if necessary. (Can be optimized a bit for PHYS.)
625	*/
626	if (PhysHighest != NIL_RTHCPHYS)
627	{
628	size_t iPage = pMemLnx->cPages;
629	while (iPage-- > 0)
630	if (page_to_phys(pMemLnx->apPages[iPage]) >= PhysHighest)
631	{
632	rtR0MemObjLinuxFreePages(pMemLnx);
633	rtR0MemObjDelete(&pMemLnx->Core);
634	return VERR_NO_MEMORY;
635	}
636	}
637
638	/*
639	* Complete the object.
640	*/
641	if (enmType == RTR0MEMOBJTYPE_PHYS)
642	{
643	pMemLnx->Core.u.Phys.PhysBase = page_to_phys(pMemLnx->apPages[0]);
644	pMemLnx->Core.u.Phys.fAllocated = true;
645	}
646	*ppMem = &pMemLnx->Core;
647	return rc;
648	}
649
650
651	/**
652	* Worker for rtR0MemObjNativeAllocPhys and rtR0MemObjNativeAllocPhysNC.
653	*
654	* @returns IPRT status.
655	* @param ppMem Where to store the memory object pointer on success.
656	* @param enmType The object type.
657	* @param cb The size of the allocation.
658	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
659	*/
660	static int rtR0MemObjLinuxAllocPhysSub(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType, size_t cb, RTHCPHYS PhysHighest)
661	{
662	int rc;
663
664	/*
665	* There are two clear cases and that's the <=16MB and anything-goes ones.
666	* When the physical address limit is somewhere inbetween those two we'll
667	* just have to try, starting with HIGHUSER and working our way thru the
668	* different types, hoping we'll get lucky.
669	*
670	* We should probably move this physical address restriction logic up to
671	* the page alloc function as it would be more efficient there. But since
672	* we don't expect this to be a performance issue just yet it can wait.
673	*/
674	if (PhysHighest == NIL_RTHCPHYS)
675	/* ZONE_HIGHMEM: the whole physical memory */
676	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, PhysHighest, GFP_HIGHUSER);
677	else if (PhysHighest <= _1M * 16)
678	/* ZONE_DMA: 0-16MB */
679	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, PhysHighest, GFP_DMA);
680	else
681	{
682	rc = VERR_NO_MEMORY;
683	if (RT_FAILURE(rc))
684	/* ZONE_HIGHMEM: the whole physical memory */
685	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, PhysHighest, GFP_HIGHUSER);
686	if (RT_FAILURE(rc))
687	/* ZONE_NORMAL: 0-896MB */
688	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, PhysHighest, GFP_USER);
689	#ifdef GFP_DMA32
690	if (RT_FAILURE(rc))
691	/* ZONE_DMA32: 0-4GB */
692	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, PhysHighest, GFP_DMA32);
693	#endif
694	if (RT_FAILURE(rc))
695	/* ZONE_DMA: 0-16MB */
696	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, PhysHighest, GFP_DMA);
697	}
698	return rc;
699	}
700
701
702	int rtR0MemObjNativeAllocPhys(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest)
703	{
704	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS, cb, PhysHighest);
705	}
706
707
708	int rtR0MemObjNativeAllocPhysNC(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest)
709	{
710	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS_NC, cb, PhysHighest);
711	}
712
713
714	int rtR0MemObjNativeEnterPhys(PPRTR0MEMOBJINTERNAL ppMem, RTHCPHYS Phys, size_t cb)
715	{
716	/*
717	* All we need to do here is to validate that we can use
718	* ioremap on the specified address (32/64-bit dma_addr_t).
719	*/
720	PRTR0MEMOBJLNX pMemLnx;
721	dma_addr_t PhysAddr = Phys;
722	AssertMsgReturn(PhysAddr == Phys, ("%#llx\n", (unsigned long long)Phys), VERR_ADDRESS_TOO_BIG);
723
724	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_PHYS, NULL, cb);
725	if (!pMemLnx)
726	return VERR_NO_MEMORY;
727
728	pMemLnx->Core.u.Phys.PhysBase = PhysAddr;
729	pMemLnx->Core.u.Phys.fAllocated = false;
730	Assert(!pMemLnx->cPages);
731	*ppMem = &pMemLnx->Core;
732	return VINF_SUCCESS;
733	}
734
735
736	int rtR0MemObjNativeLockUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3Ptr, size_t cb, uint32_t fAccess, RTR0PROCESS R0Process)
737	{
738	const int cPages = cb >> PAGE_SHIFT;
739	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
740	struct vm_area_struct **papVMAs;
741	PRTR0MEMOBJLNX pMemLnx;
742	int rc = VERR_NO_MEMORY;
743	NOREF(fAccess);
744
745	/*
746	* Check for valid task and size overflows.
747	*/
748	if (!pTask)
749	return VERR_NOT_SUPPORTED;
750	if (((size_t)cPages << PAGE_SHIFT) != cb)
751	return VERR_OUT_OF_RANGE;
752
753	/*
754	* Allocate the memory object and a temporary buffer for the VMAs.
755	*/
756	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_OFFSETOF(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK, (void *)R3Ptr, cb);
757	if (!pMemLnx)
758	return VERR_NO_MEMORY;
759
760	papVMAs = (struct vm_area_struct *)RTMemAlloc(sizeof(papVMAs) * cPages);
761	if (papVMAs)
762	{
763	down_read(&pTask->mm->mmap_sem);
764
765	/*
766	* Get user pages.
767	*/
768	rc = get_user_pages(pTask, /* Task for fault acounting. */
769	pTask->mm, /* Whose pages. */
770	R3Ptr, /* Where from. */
771	cPages, /* How many pages. */
772	1, /* Write to memory. */
773	0, /* force. */
774	&pMemLnx->apPages[0], /* Page array. */
775	papVMAs); /* vmas */
776	if (rc == cPages)
777	{
778	/*
779	* Flush dcache (required?), protect against fork and _really_ pin the page
780	* table entries. get_user_pages() will protect against swapping out the
781	* pages but it will NOT protect against removing page table entries. This
782	* can be achieved with
783	* - using mlock / mmap(..., MAP_LOCKED, ...) from userland. This requires
784	* an appropriate limit set up with setrlimit(..., RLIMIT_MEMLOCK, ...).
785	* Usual Linux distributions support only a limited size of locked pages
786	* (e.g. 32KB).
787	* - setting the PageReserved bit (as we do in rtR0MemObjLinuxAllocPages()
788	* or by
789	* - setting the VM_LOCKED flag. This is the same as doing mlock() without
790	* a range check.
791	*/
792	/** @todo The Linux fork() protection will require more work if this API
793	* is to be used for anything but locking VM pages. */
794	while (rc-- > 0)
795	{
796	flush_dcache_page(pMemLnx->apPages[rc]);
797	papVMAs[rc]->vm_flags \|= (VM_DONTCOPY \| VM_LOCKED);
798	}
799
800	up_read(&pTask->mm->mmap_sem);
801
802	RTMemFree(papVMAs);
803
804	pMemLnx->Core.u.Lock.R0Process = R0Process;
805	pMemLnx->cPages = cPages;
806	Assert(!pMemLnx->fMappedToRing0);
807	*ppMem = &pMemLnx->Core;
808
809	return VINF_SUCCESS;
810	}
811
812	/*
813	* Failed - we need to unlock any pages that we succeeded to lock.
814	*/
815	while (rc-- > 0)
816	{
817	if (!PageReserved(pMemLnx->apPages[rc]))
818	SetPageDirty(pMemLnx->apPages[rc]);
819	page_cache_release(pMemLnx->apPages[rc]);
820	}
821
822	up_read(&pTask->mm->mmap_sem);
823
824	RTMemFree(papVMAs);
825	rc = VERR_LOCK_FAILED;
826	}
827
828	rtR0MemObjDelete(&pMemLnx->Core);
829	return rc;
830	}
831
832
833	int rtR0MemObjNativeLockKernel(PPRTR0MEMOBJINTERNAL ppMem, void *pv, size_t cb, uint32_t fAccess)
834	{
835	void pvLast = (uint8_t )pv + cb - 1;
836	size_t const cPages = cb >> PAGE_SHIFT;
837	PRTR0MEMOBJLNX pMemLnx;
838	bool fLinearMapping;
839	int rc;
840	uint8_t *pbPage;
841	size_t iPage;
842	NOREF(fAccess);
843
844	/*
845	* Classify the memory and check that we can deal with it.
846	*/
847	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
848	fLinearMapping = virt_addr_valid(pvLast) && virt_addr_valid(pv);
849	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 0)
850	fLinearMapping = VALID_PAGE(virt_to_page(pvLast)) && VALID_PAGE(virt_to_page(pv));
851	#else
852	# error "not supported"
853	#endif
854	if (!fLinearMapping)
855	{
856	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 19)
857	if ( !RTR0MemKernelIsValidAddr(pv)
858	\|\| !RTR0MemKernelIsValidAddr(pv + cb))
859	#endif
860	return VERR_INVALID_PARAMETER;
861	}
862
863	/*
864	* Allocate the memory object.
865	*/
866	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_OFFSETOF(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK, pv, cb);
867	if (!pMemLnx)
868	return VERR_NO_MEMORY;
869
870	/*
871	* Gather the pages.
872	* We ASSUME all kernel pages are non-swappable.
873	*/
874	rc = VINF_SUCCESS;
875	pbPage = (uint8_t *)pvLast;
876	iPage = cPages;
877	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 19)
878	if (!fLinearMapping)
879	{
880	while (iPage-- > 0)
881	{
882	struct page *pPage = vmalloc_to_page(pbPage);
883	if (RT_UNLIKELY(!pPage))
884	{
885	rc = VERR_LOCK_FAILED;
886	break;
887	}
888	pMemLnx->apPages[iPage] = pPage;
889	pbPage -= PAGE_SIZE;
890	}
891	}
892	else
893	#endif
894	{
895	while (iPage-- > 0)
896	{
897	pMemLnx->apPages[iPage] = virt_to_page(pbPage);
898	pbPage -= PAGE_SIZE;
899	}
900	}
901	if (RT_SUCCESS(rc))
902	{
903	/*
904	* Complete the memory object and return.
905	*/
906	pMemLnx->Core.u.Lock.R0Process = NIL_RTR0PROCESS;
907	pMemLnx->cPages = cPages;
908	Assert(!pMemLnx->fMappedToRing0);
909	*ppMem = &pMemLnx->Core;
910
911	return VINF_SUCCESS;
912	}
913
914	rtR0MemObjDelete(&pMemLnx->Core);
915	return rc;
916	}
917
918
919	int rtR0MemObjNativeReserveKernel(PPRTR0MEMOBJINTERNAL ppMem, void *pvFixed, size_t cb, size_t uAlignment)
920	{
921	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
922	const size_t cPages = cb >> PAGE_SHIFT;
923	struct page *pDummyPage;
924	struct page **papPages;
925
926	/* check for unsupported stuff. */
927	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
928	if (uAlignment > PAGE_SIZE)
929	return VERR_NOT_SUPPORTED;
930
931	/*
932	* Allocate a dummy page and create a page pointer array for vmap such that
933	* the dummy page is mapped all over the reserved area.
934	*/
935	pDummyPage = alloc_page(GFP_HIGHUSER);
936	if (!pDummyPage)
937	return VERR_NO_MEMORY;
938	papPages = RTMemAlloc(sizeof(papPages) cPages);
939	if (papPages)
940	{
941	void *pv;
942	size_t iPage = cPages;
943	while (iPage-- > 0)
944	papPages[iPage] = pDummyPage;
945	# ifdef VM_MAP
946	pv = vmap(papPages, cPages, VM_MAP, PAGE_KERNEL_RO);
947	# else
948	pv = vmap(papPages, cPages, VM_ALLOC, PAGE_KERNEL_RO);
949	# endif
950	RTMemFree(papPages);
951	if (pv)
952	{
953	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb);
954	if (pMemLnx)
955	{
956	pMemLnx->Core.u.ResVirt.R0Process = NIL_RTR0PROCESS;
957	pMemLnx->cPages = 1;
958	pMemLnx->apPages[0] = pDummyPage;
959	*ppMem = &pMemLnx->Core;
960	return VINF_SUCCESS;
961	}
962	vunmap(pv);
963	}
964	}
965	__free_page(pDummyPage);
966	return VERR_NO_MEMORY;
967
968	#else /* < 2.4.22 */
969	/*
970	* Could probably use ioremap here, but the caller is in a better position than us
971	* to select some safe physical memory.
972	*/
973	return VERR_NOT_SUPPORTED;
974	#endif
975	}
976
977
978	/**
979	* Worker for rtR0MemObjNativeReserveUser and rtR0MemObjNativerMapUser that creates
980	* an empty user space mapping.
981	*
982	* The caller takes care of acquiring the mmap_sem of the task.
983	*
984	* @returns Pointer to the mapping.
985	* (void *)-1 on failure.
986	* @param R3PtrFixed (RTR3PTR)-1 if anywhere, otherwise a specific location.
987	* @param cb The size of the mapping.
988	* @param uAlignment The alignment of the mapping.
989	* @param pTask The Linux task to create this mapping in.
990	* @param fProt The RTMEM_PROT_* mask.
991	*/
992	static void rtR0MemObjLinuxDoMmap(RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment, struct task_struct pTask, unsigned fProt)
993	{
994	unsigned fLnxProt;
995	unsigned long ulAddr;
996
997	/*
998	* Convert from IPRT protection to mman.h PROT_ and call do_mmap.
999	*/
1000	fProt &= (RTMEM_PROT_NONE \| RTMEM_PROT_READ \| RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC);
1001	if (fProt == RTMEM_PROT_NONE)
1002	fLnxProt = PROT_NONE;
1003	else
1004	{
1005	fLnxProt = 0;
1006	if (fProt & RTMEM_PROT_READ)
1007	fLnxProt \|= PROT_READ;
1008	if (fProt & RTMEM_PROT_WRITE)
1009	fLnxProt \|= PROT_WRITE;
1010	if (fProt & RTMEM_PROT_EXEC)
1011	fLnxProt \|= PROT_EXEC;
1012	}
1013
1014	if (R3PtrFixed != (RTR3PTR)-1)
1015	ulAddr = do_mmap(NULL, R3PtrFixed, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS \| MAP_FIXED, 0);
1016	else
1017	{
1018	ulAddr = do_mmap(NULL, 0, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS, 0);
1019	if ( !(ulAddr & ~PAGE_MASK)
1020	&& (ulAddr & (uAlignment - 1)))
1021	{
1022	/** @todo implement uAlignment properly... We'll probably need to make some dummy mappings to fill
1023	* up alignment gaps. This is of course complicated by fragmentation (which we might have cause
1024	* ourselves) and further by there begin two mmap strategies (top / bottom). */
1025	/* For now, just ignore uAlignment requirements... */
1026	}
1027	}
1028	if (ulAddr & ~PAGE_MASK) /* ~PAGE_MASK == PAGE_OFFSET_MASK */
1029	return (void *)-1;
1030	return (void *)ulAddr;
1031	}
1032
1033
1034	int rtR0MemObjNativeReserveUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment, RTR0PROCESS R0Process)
1035	{
1036	PRTR0MEMOBJLNX pMemLnx;
1037	void *pv;
1038	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1039	if (!pTask)
1040	return VERR_NOT_SUPPORTED;
1041
1042	/*
1043	* Check that the specified alignment is supported.
1044	*/
1045	if (uAlignment > PAGE_SIZE)
1046	return VERR_NOT_SUPPORTED;
1047
1048	/*
1049	* Let rtR0MemObjLinuxDoMmap do the difficult bits.
1050	*/
1051	down_write(&pTask->mm->mmap_sem);
1052	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, cb, uAlignment, pTask, RTMEM_PROT_NONE);
1053	up_write(&pTask->mm->mmap_sem);
1054	if (pv == (void *)-1)
1055	return VERR_NO_MEMORY;
1056
1057	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb);
1058	if (!pMemLnx)
1059	{
1060	down_write(&pTask->mm->mmap_sem);
1061	MY_DO_MUNMAP(pTask->mm, (unsigned long)pv, cb);
1062	up_write(&pTask->mm->mmap_sem);
1063	return VERR_NO_MEMORY;
1064	}
1065
1066	pMemLnx->Core.u.ResVirt.R0Process = R0Process;
1067	*ppMem = &pMemLnx->Core;
1068	return VINF_SUCCESS;
1069	}
1070
1071
1072	int rtR0MemObjNativeMapKernel(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap, void *pvFixed, size_t uAlignment,
1073	unsigned fProt, size_t offSub, size_t cbSub)
1074	{
1075	int rc = VERR_NO_MEMORY;
1076	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1077	PRTR0MEMOBJLNX pMemLnx;
1078
1079	/* Fail if requested to do something we can't. */
1080	AssertMsgReturn(!offSub && !cbSub, ("%#x %#x\n", offSub, cbSub), VERR_NOT_SUPPORTED);
1081	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
1082	if (uAlignment > PAGE_SIZE)
1083	return VERR_NOT_SUPPORTED;
1084
1085	/*
1086	* Create the IPRT memory object.
1087	*/
1088	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, pMemLnxToMap->Core.cb);
1089	if (pMemLnx)
1090	{
1091	if (pMemLnxToMap->cPages)
1092	{
1093	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1094	/*
1095	* Use vmap - 2.4.22 and later.
1096	*/
1097	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, true /* kernel */);
1098	# ifdef VM_MAP
1099	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[0], pMemLnxToMap->cPages, VM_MAP, fPg);
1100	# else
1101	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[0], pMemLnxToMap->cPages, VM_ALLOC, fPg);
1102	# endif
1103	if (pMemLnx->Core.pv)
1104	{
1105	pMemLnx->fMappedToRing0 = true;
1106	rc = VINF_SUCCESS;
1107	}
1108	else
1109	rc = VERR_MAP_FAILED;
1110
1111	#else /* < 2.4.22 */
1112	/*
1113	* Only option here is to share mappings if possible and forget about fProt.
1114	*/
1115	if (rtR0MemObjIsRing3(pMemToMap))
1116	rc = VERR_NOT_SUPPORTED;
1117	else
1118	{
1119	rc = VINF_SUCCESS;
1120	if (!pMemLnxToMap->Core.pv)
1121	rc = rtR0MemObjLinuxVMap(pMemLnxToMap, !!(fProt & RTMEM_PROT_EXEC));
1122	if (RT_SUCCESS(rc))
1123	{
1124	Assert(pMemLnxToMap->Core.pv);
1125	pMemLnx->Core.pv = pMemLnxToMap->Core.pv;
1126	}
1127	}
1128	#endif
1129	}
1130	else
1131	{
1132	/*
1133	* MMIO / physical memory.
1134	*/
1135	Assert(pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS && !pMemLnxToMap->Core.u.Phys.fAllocated);
1136	pMemLnx->Core.pv = ioremap(pMemLnxToMap->Core.u.Phys.PhysBase, pMemLnxToMap->Core.cb);
1137	if (pMemLnx->Core.pv)
1138	{
1139	/** @todo fix protection. */
1140	rc = VINF_SUCCESS;
1141	}
1142	}
1143	if (RT_SUCCESS(rc))
1144	{
1145	pMemLnx->Core.u.Mapping.R0Process = NIL_RTR0PROCESS;
1146	*ppMem = &pMemLnx->Core;
1147	return VINF_SUCCESS;
1148	}
1149	rtR0MemObjDelete(&pMemLnx->Core);
1150	}
1151
1152	return rc;
1153	}
1154
1155
1156	#ifdef VBOX_USE_PAE_HACK
1157	/**
1158	* Replace the PFN of a PTE with the address of the actual page.
1159	*
1160	* The caller maps a reserved dummy page at the address with the desired access
1161	* and flags.
1162	*
1163	* This hack is required for older Linux kernels which don't provide
1164	* remap_pfn_range().
1165	*
1166	* @returns 0 on success, -ENOMEM on failure.
1167	* @param mm The memory context.
1168	* @param ulAddr The mapping address.
1169	* @param Phys The physical address of the page to map.
1170	*/
1171	static int rtR0MemObjLinuxFixPte(struct mm_struct *mm, unsigned long ulAddr, RTHCPHYS Phys)
1172	{
1173	int rc = -ENOMEM;
1174	pgd_t *pgd;
1175
1176	spin_lock(&mm->page_table_lock);
1177
1178	pgd = pgd_offset(mm, ulAddr);
1179	if (!pgd_none(pgd) && !pgd_bad(pgd))
1180	{
1181	pmd_t *pmd = pmd_offset(pgd, ulAddr);
1182	if (!pmd_none(*pmd))
1183	{
1184	pte_t *ptep = pte_offset_map(pmd, ulAddr);
1185	if (ptep)
1186	{
1187	pte_t pte = *ptep;
1188	pte.pte_high &= 0xfff00000;
1189	pte.pte_high \|= ((Phys >> 32) & 0x000fffff);
1190	pte.pte_low &= 0x00000fff;
1191	pte.pte_low \|= (Phys & 0xfffff000);
1192	set_pte(ptep, pte);
1193	pte_unmap(ptep);
1194	rc = 0;
1195	}
1196	}
1197	}
1198
1199	spin_unlock(&mm->page_table_lock);
1200	return rc;
1201	}
1202	#endif /* VBOX_USE_PAE_HACK */
1203
1204
1205	int rtR0MemObjNativeMapUser(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap, RTR3PTR R3PtrFixed, size_t uAlignment, unsigned fProt, RTR0PROCESS R0Process)
1206	{
1207	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1208	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1209	int rc = VERR_NO_MEMORY;
1210	PRTR0MEMOBJLNX pMemLnx;
1211	#ifdef VBOX_USE_PAE_HACK
1212	struct page *pDummyPage;
1213	RTHCPHYS DummyPhys;
1214	#endif
1215
1216	/*
1217	* Check for restrictions.
1218	*/
1219	if (!pTask)
1220	return VERR_NOT_SUPPORTED;
1221	if (uAlignment > PAGE_SIZE)
1222	return VERR_NOT_SUPPORTED;
1223
1224	#ifdef VBOX_USE_PAE_HACK
1225	/*
1226	* Allocate a dummy page for use when mapping the memory.
1227	*/
1228	pDummyPage = alloc_page(GFP_USER);
1229	if (!pDummyPage)
1230	return VERR_NO_MEMORY;
1231	SetPageReserved(pDummyPage);
1232	DummyPhys = page_to_phys(pDummyPage);
1233	#endif
1234
1235	/*
1236	* Create the IPRT memory object.
1237	*/
1238	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, pMemLnxToMap->Core.cb);
1239	if (pMemLnx)
1240	{
1241	/*
1242	* Allocate user space mapping.
1243	*/
1244	void *pv;
1245	down_write(&pTask->mm->mmap_sem);
1246	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, pMemLnxToMap->Core.cb, uAlignment, pTask, fProt);
1247	if (pv != (void *)-1)
1248	{
1249	/*
1250	* Map page by page into the mmap area.
1251	* This is generic, paranoid and not very efficient.
1252	*/
1253	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, false /* user */);
1254	unsigned long ulAddrCur = (unsigned long)pv;
1255	const size_t cPages = pMemLnxToMap->Core.cb >> PAGE_SHIFT;
1256	size_t iPage;
1257
1258	rc = 0;
1259	if (pMemLnxToMap->cPages)
1260	{
1261	for (iPage = 0; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE)
1262	{
1263	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11)
1264	RTHCPHYS Phys = page_to_phys(pMemLnxToMap->apPages[iPage]);
1265	#endif
1266	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1267	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
1268	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
1269	#endif
1270	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0) && defined(RT_ARCH_X86)
1271	/* remap_page_range() limitation on x86 */
1272	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
1273	#endif
1274
1275	#if defined(VBOX_USE_INSERT_PAGE) && LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1276	rc = vm_insert_page(vma, ulAddrCur, pMemLnxToMap->apPages[iPage]);
1277	vma->vm_flags \|= VM_RESERVED; /* This flag helps making 100% sure some bad stuff wont happen (swap, core, ++). */
1278	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
1279	rc = remap_pfn_range(vma, ulAddrCur, page_to_pfn(pMemLnxToMap->apPages[iPage]), PAGE_SIZE, fPg);
1280	#elif defined(VBOX_USE_PAE_HACK)
1281	rc = remap_page_range(vma, ulAddrCur, DummyPhys, PAGE_SIZE, fPg);
1282	if (!rc)
1283	rc = rtR0MemObjLinuxFixPte(pTask->mm, ulAddrCur, Phys);
1284	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1285	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1286	#else /* 2.4 */
1287	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
1288	#endif
1289	if (rc)
1290	{
1291	rc = VERR_NO_MEMORY;
1292	break;
1293	}
1294	}
1295	}
1296	else
1297	{
1298	RTHCPHYS Phys;
1299	if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS)
1300	Phys = pMemLnxToMap->Core.u.Phys.PhysBase;
1301	else if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_CONT)
1302	Phys = pMemLnxToMap->Core.u.Cont.Phys;
1303	else
1304	{
1305	AssertMsgFailed(("%d\n", pMemLnxToMap->Core.enmType));
1306	Phys = NIL_RTHCPHYS;
1307	}
1308	if (Phys != NIL_RTHCPHYS)
1309	{
1310	for (iPage = 0; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE, Phys += PAGE_SIZE)
1311	{
1312	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1313	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
1314	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
1315	#endif
1316	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0) && defined(RT_ARCH_X86)
1317	/* remap_page_range() limitation on x86 */
1318	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
1319	#endif
1320
1321	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
1322	rc = remap_pfn_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1323	#elif defined(VBOX_USE_PAE_HACK)
1324	rc = remap_page_range(vma, ulAddrCur, DummyPhys, PAGE_SIZE, fPg);
1325	if (!rc)
1326	rc = rtR0MemObjLinuxFixPte(pTask->mm, ulAddrCur, Phys);
1327	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1328	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1329	#else /* 2.4 */
1330	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
1331	#endif
1332	if (rc)
1333	{
1334	rc = VERR_NO_MEMORY;
1335	break;
1336	}
1337	}
1338	}
1339	}
1340	if (!rc)
1341	{
1342	up_write(&pTask->mm->mmap_sem);
1343	#ifdef VBOX_USE_PAE_HACK
1344	__free_page(pDummyPage);
1345	#endif
1346
1347	pMemLnx->Core.pv = pv;
1348	pMemLnx->Core.u.Mapping.R0Process = R0Process;
1349	*ppMem = &pMemLnx->Core;
1350	return VINF_SUCCESS;
1351	}
1352
1353	/*
1354	* Bail out.
1355	*/
1356	MY_DO_MUNMAP(pTask->mm, (unsigned long)pv, pMemLnxToMap->Core.cb);
1357	}
1358	up_write(&pTask->mm->mmap_sem);
1359	rtR0MemObjDelete(&pMemLnx->Core);
1360	}
1361	#ifdef VBOX_USE_PAE_HACK
1362	__free_page(pDummyPage);
1363	#endif
1364
1365	return rc;
1366	}
1367
1368
1369	int rtR0MemObjNativeProtect(PRTR0MEMOBJINTERNAL pMem, size_t offSub, size_t cbSub, uint32_t fProt)
1370	{
1371	NOREF(pMem);
1372	NOREF(offSub);
1373	NOREF(cbSub);
1374	NOREF(fProt);
1375	return VERR_NOT_SUPPORTED;
1376	}
1377
1378
1379	RTHCPHYS rtR0MemObjNativeGetPagePhysAddr(PRTR0MEMOBJINTERNAL pMem, size_t iPage)
1380	{
1381	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
1382
1383	if (pMemLnx->cPages)
1384	return page_to_phys(pMemLnx->apPages[iPage]);
1385
1386	switch (pMemLnx->Core.enmType)
1387	{
1388	case RTR0MEMOBJTYPE_CONT:
1389	return pMemLnx->Core.u.Cont.Phys + (iPage << PAGE_SHIFT);
1390
1391	case RTR0MEMOBJTYPE_PHYS:
1392	return pMemLnx->Core.u.Phys.PhysBase + (iPage << PAGE_SHIFT);
1393
1394	/* the parent knows */
1395	case RTR0MEMOBJTYPE_MAPPING:
1396	return rtR0MemObjNativeGetPagePhysAddr(pMemLnx->Core.uRel.Child.pParent, iPage);
1397
1398	/* cPages > 0 */
1399	case RTR0MEMOBJTYPE_LOW:
1400	case RTR0MEMOBJTYPE_LOCK:
1401	case RTR0MEMOBJTYPE_PHYS_NC:
1402	case RTR0MEMOBJTYPE_PAGE:
1403	default:
1404	AssertMsgFailed(("%d\n", pMemLnx->Core.enmType));
1405	/* fall thru */
1406
1407	case RTR0MEMOBJTYPE_RES_VIRT:
1408	return NIL_RTHCPHYS;
1409	}
1410	}
1411

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

source: vbox/trunk/src/VBox/Runtime/r0drv/linux/memobj-r0drv-linux.c@ 25645

Download in other formats: