memobj-r0drv-linux.c@ 21497

Last change on this file since 21497 was 21497, checked in by vboxsync, 16 years ago
RTR0MemObj*: correctly reject unsupported alignment requirements (VERR_NOT_SUPPORTED).
Property svn:eol-style set to `native` Property svn:keywords set to `Id Rev`
File size: 40.9 KB

Line
1	/* $Revision: 21497 $ */
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
66
67	/*******************************************************************************
68	* Structures and Typedefs *
69	*******************************************************************************/
70	/**
71	* The Darwin version of the memory object structure.
72	*/
73	typedef struct RTR0MEMOBJLNX
74	{
75	/** The core structure. */
76	RTR0MEMOBJINTERNAL Core;
77	/** Set if the allocation is contiguous.
78	* This means it has to be given back as one chunk. */
79	bool fContiguous;
80	/** Set if we've vmap'ed thed memory into ring-0. */
81	bool fMappedToRing0;
82	/** The pages in the apPages array. */
83	size_t cPages;
84	/** Array of struct page pointers. (variable size) */
85	struct page *apPages[1];
86	} RTR0MEMOBJLNX, *PRTR0MEMOBJLNX;
87
88
89	/**
90	* Helper that converts from a RTR0PROCESS handle to a linux task.
91	*
92	* @returns The corresponding Linux task.
93	* @param R0Process IPRT ring-0 process handle.
94	*/
95	struct task_struct *rtR0ProcessToLinuxTask(RTR0PROCESS R0Process)
96	{
97	/** @todo fix rtR0ProcessToLinuxTask!! */
98	return R0Process == RTR0ProcHandleSelf() ? current : NULL;
99	}
100
101
102	/**
103	* Compute order. Some functions allocate 2^order pages.
104	*
105	* @returns order.
106	* @param cPages Number of pages.
107	*/
108	static int rtR0MemObjLinuxOrder(size_t cPages)
109	{
110	int iOrder;
111	size_t cTmp;
112
113	for (iOrder = 0, cTmp = cPages; cTmp >>= 1; ++iOrder)
114	;
115	if (cPages & ~((size_t)1 << iOrder))
116	++iOrder;
117
118	return iOrder;
119	}
120
121
122	/**
123	* Converts from RTMEM_PROT_* to Linux PAGE_*.
124	*
125	* @returns Linux page protection constant.
126	* @param fProt The IPRT protection mask.
127	* @param fKernel Whether it applies to kernel or user space.
128	*/
129	static pgprot_t rtR0MemObjLinuxConvertProt(unsigned fProt, bool fKernel)
130	{
131	switch (fProt)
132	{
133	default:
134	AssertMsgFailed(("%#x %d\n", fProt, fKernel));
135	case RTMEM_PROT_NONE:
136	return PAGE_NONE;
137
138	case RTMEM_PROT_READ:
139	return fKernel ? PAGE_KERNEL_RO : PAGE_READONLY;
140
141	case RTMEM_PROT_WRITE:
142	case RTMEM_PROT_WRITE \| RTMEM_PROT_READ:
143	return fKernel ? PAGE_KERNEL : PAGE_SHARED;
144
145	case RTMEM_PROT_EXEC:
146	case RTMEM_PROT_EXEC \| RTMEM_PROT_READ:
147	#if defined(RT_ARCH_X86) \|\| defined(RT_ARCH_AMD64)
148	if (fKernel)
149	{
150	pgprot_t fPg = MY_PAGE_KERNEL_EXEC;
151	pgprot_val(fPg) &= ~_PAGE_RW;
152	return fPg;
153	}
154	return PAGE_READONLY_EXEC;
155	#else
156	return fKernel ? MY_PAGE_KERNEL_EXEC : PAGE_READONLY_EXEC;
157	#endif
158
159	case RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC:
160	case RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC \| RTMEM_PROT_READ:
161	return fKernel ? MY_PAGE_KERNEL_EXEC : PAGE_SHARED_EXEC;
162	}
163	}
164
165
166	/**
167	* Internal worker that allocates physical pages and creates the memory object for them.
168	*
169	* @returns IPRT status code.
170	* @param ppMemLnx Where to store the memory object pointer.
171	* @param enmType The object type.
172	* @param cb The number of bytes to allocate.
173	* @param fFlagsLnx The page allocation flags (GPFs).
174	* @param fContiguous Whether the allocation must be contiguous.
175	*/
176	static int rtR0MemObjLinuxAllocPages(PRTR0MEMOBJLNX *ppMemLnx, RTR0MEMOBJTYPE enmType, size_t cb, unsigned fFlagsLnx, bool fContiguous)
177	{
178	size_t iPage;
179	size_t cPages = cb >> PAGE_SHIFT;
180	struct page *paPages;
181
182	/*
183	* Allocate a memory object structure that's large enough to contain
184	* the page pointer array.
185	*/
186	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_OFFSETOF(RTR0MEMOBJLNX, apPages[cPages]), enmType, NULL, cb);
187	if (!pMemLnx)
188	return VERR_NO_MEMORY;
189	pMemLnx->cPages = cPages;
190
191	/*
192	* Allocate the pages.
193	* For small allocations we'll try contiguous first and then fall back on page by page.
194	*/
195	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
196	if ( fContiguous
197	\|\| cb <= PAGE_SIZE * 2)
198	{
199	#ifdef VBOX_USE_INSERT_PAGE
200	paPages = alloc_pages(fFlagsLnx \| __GFP_COMP, rtR0MemObjLinuxOrder(cb >> PAGE_SHIFT));
201	#else
202	paPages = alloc_pages(fFlagsLnx, rtR0MemObjLinuxOrder(cb >> PAGE_SHIFT));
203	#endif
204	if (paPages)
205	{
206	fContiguous = true;
207	for (iPage = 0; iPage < cPages; iPage++)
208	pMemLnx->apPages[iPage] = &paPages[iPage];
209	}
210	else if (fContiguous)
211	{
212	rtR0MemObjDelete(&pMemLnx->Core);
213	return VERR_NO_MEMORY;
214	}
215	}
216
217	if (!fContiguous)
218	{
219	for (iPage = 0; iPage < cPages; iPage++)
220	{
221	pMemLnx->apPages[iPage] = alloc_page(fFlagsLnx);
222	if (RT_UNLIKELY(!pMemLnx->apPages[iPage]))
223	{
224	while (iPage-- > 0)
225	__free_page(pMemLnx->apPages[iPage]);
226	rtR0MemObjDelete(&pMemLnx->Core);
227	return VERR_NO_MEMORY;
228	}
229	}
230	}
231
232	#else /* < 2.4.22 */
233	/** @todo figure out why we didn't allocate page-by-page on 2.4.21 and older... */
234	paPages = alloc_pages(fFlagsLnx, rtR0MemObjLinuxOrder(cb >> PAGE_SHIFT));
235	if (!paPages)
236	{
237	rtR0MemObjDelete(&pMemLnx->Core);
238	return VERR_NO_MEMORY;
239	}
240	for (iPage = 0; iPage < cPages; iPage++)
241	{
242	pMemLnx->apPages[iPage] = &paPages[iPage];
243	MY_SET_PAGES_EXEC(pMemLnx->apPages[iPage], 1);
244	if (PageHighMem(pMemLnx->apPages[iPage]))
245	BUG();
246	}
247
248	fContiguous = true;
249	#endif /* < 2.4.22 */
250	pMemLnx->fContiguous = fContiguous;
251
252	/*
253	* Reserve the pages.
254	*/
255	for (iPage = 0; iPage < cPages; iPage++)
256	SetPageReserved(pMemLnx->apPages[iPage]);
257
258	*ppMemLnx = pMemLnx;
259	return VINF_SUCCESS;
260	}
261
262
263	/**
264	* Frees the physical pages allocated by the rtR0MemObjLinuxAllocPages() call.
265	*
266	* This method does NOT free the object.
267	*
268	* @param pMemLnx The object which physical pages should be freed.
269	*/
270	static void rtR0MemObjLinuxFreePages(PRTR0MEMOBJLNX pMemLnx)
271	{
272	size_t iPage = pMemLnx->cPages;
273	if (iPage > 0)
274	{
275	/*
276	* Restore the page flags.
277	*/
278	while (iPage-- > 0)
279	{
280	ClearPageReserved(pMemLnx->apPages[iPage]);
281	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
282	#else
283	MY_SET_PAGES_NOEXEC(pMemLnx->apPages[iPage], 1);
284	#endif
285	}
286
287	/*
288	* Free the pages.
289	*/
290	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
291	if (!pMemLnx->fContiguous)
292	{
293	iPage = pMemLnx->cPages;
294	while (iPage-- > 0)
295	__free_page(pMemLnx->apPages[iPage]);
296	}
297	else
298	#endif
299	__free_pages(pMemLnx->apPages[0], rtR0MemObjLinuxOrder(pMemLnx->cPages));
300
301	pMemLnx->cPages = 0;
302	}
303	}
304
305
306	/**
307	* Maps the allocation into ring-0.
308	*
309	* This will update the RTR0MEMOBJLNX::Core.pv and RTR0MEMOBJ::fMappedToRing0 members.
310	*
311	* Contiguous mappings that isn't in 'high' memory will already be mapped into kernel
312	* space, so we'll use that mapping if possible. If execute access is required, we'll
313	* play safe and do our own mapping.
314	*
315	* @returns IPRT status code.
316	* @param pMemLnx The linux memory object to map.
317	* @param fExecutable Whether execute access is required.
318	*/
319	static int rtR0MemObjLinuxVMap(PRTR0MEMOBJLNX pMemLnx, bool fExecutable)
320	{
321	int rc = VINF_SUCCESS;
322
323	/*
324	* Choose mapping strategy.
325	*/
326	bool fMustMap = fExecutable
327	\|\| !pMemLnx->fContiguous;
328	if (!fMustMap)
329	{
330	size_t iPage = pMemLnx->cPages;
331	while (iPage-- > 0)
332	if (PageHighMem(pMemLnx->apPages[iPage]))
333	{
334	fMustMap = true;
335	break;
336	}
337	}
338
339	Assert(!pMemLnx->Core.pv);
340	Assert(!pMemLnx->fMappedToRing0);
341
342	if (fMustMap)
343	{
344	/*
345	* Use vmap - 2.4.22 and later.
346	*/
347	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
348	pgprot_t fPg;
349	pgprot_val(fPg) = _PAGE_PRESENT \| _PAGE_RW;
350	# ifdef _PAGE_NX
351	if (!fExecutable)
352	pgprot_val(fPg) \|= _PAGE_NX;
353	# endif
354
355	# ifdef VM_MAP
356	pMemLnx->Core.pv = vmap(&pMemLnx->apPages[0], pMemLnx->cPages, VM_MAP, fPg);
357	# else
358	pMemLnx->Core.pv = vmap(&pMemLnx->apPages[0], pMemLnx->cPages, VM_ALLOC, fPg);
359	# endif
360	if (pMemLnx->Core.pv)
361	pMemLnx->fMappedToRing0 = true;
362	else
363	rc = VERR_MAP_FAILED;
364	#else /* < 2.4.22 */
365	rc = VERR_NOT_SUPPORTED;
366	#endif
367	}
368	else
369	{
370	/*
371	* Use the kernel RAM mapping.
372	*/
373	pMemLnx->Core.pv = phys_to_virt(page_to_phys(pMemLnx->apPages[0]));
374	Assert(pMemLnx->Core.pv);
375	}
376
377	return rc;
378	}
379
380
381	/**
382	* Undos what rtR0MemObjLinuxVMap() did.
383	*
384	* @param pMemLnx The linux memory object.
385	*/
386	static void rtR0MemObjLinuxVUnmap(PRTR0MEMOBJLNX pMemLnx)
387	{
388	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
389	if (pMemLnx->fMappedToRing0)
390	{
391	Assert(pMemLnx->Core.pv);
392	vunmap(pMemLnx->Core.pv);
393	pMemLnx->fMappedToRing0 = false;
394	}
395	#else /* < 2.4.22 */
396	Assert(!pMemLnx->fMappedToRing0);
397	#endif
398	pMemLnx->Core.pv = NULL;
399	}
400
401
402	int rtR0MemObjNativeFree(RTR0MEMOBJ pMem)
403	{
404	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
405
406	/*
407	* Release any memory that we've allocated or locked.
408	*/
409	switch (pMemLnx->Core.enmType)
410	{
411	case RTR0MEMOBJTYPE_LOW:
412	case RTR0MEMOBJTYPE_PAGE:
413	case RTR0MEMOBJTYPE_CONT:
414	case RTR0MEMOBJTYPE_PHYS:
415	case RTR0MEMOBJTYPE_PHYS_NC:
416	rtR0MemObjLinuxVUnmap(pMemLnx);
417	rtR0MemObjLinuxFreePages(pMemLnx);
418	break;
419
420	case RTR0MEMOBJTYPE_LOCK:
421	if (pMemLnx->Core.u.Lock.R0Process != NIL_RTR0PROCESS)
422	{
423	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
424	size_t iPage;
425	Assert(pTask);
426	if (pTask && pTask->mm)
427	down_read(&pTask->mm->mmap_sem);
428
429	iPage = pMemLnx->cPages;
430	while (iPage-- > 0)
431	{
432	if (!PageReserved(pMemLnx->apPages[iPage]))
433	SetPageDirty(pMemLnx->apPages[iPage]);
434	page_cache_release(pMemLnx->apPages[iPage]);
435	}
436
437	if (pTask && pTask->mm)
438	up_read(&pTask->mm->mmap_sem);
439	}
440	/* else: kernel memory - nothing to do here. */
441	break;
442
443	case RTR0MEMOBJTYPE_RES_VIRT:
444	Assert(pMemLnx->Core.pv);
445	if (pMemLnx->Core.u.ResVirt.R0Process != NIL_RTR0PROCESS)
446	{
447	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
448	Assert(pTask);
449	if (pTask && pTask->mm)
450	{
451	down_write(&pTask->mm->mmap_sem);
452	MY_DO_MUNMAP(pTask->mm, (unsigned long)pMemLnx->Core.pv, pMemLnx->Core.cb);
453	up_write(&pTask->mm->mmap_sem);
454	}
455	}
456	else
457	{
458	vunmap(pMemLnx->Core.pv);
459
460	Assert(pMemLnx->cPages == 1 && pMemLnx->apPages[0] != NULL);
461	__free_page(pMemLnx->apPages[0]);
462	pMemLnx->apPages[0] = NULL;
463	pMemLnx->cPages = 0;
464	}
465	pMemLnx->Core.pv = NULL;
466	break;
467
468	case RTR0MEMOBJTYPE_MAPPING:
469	Assert(pMemLnx->cPages == 0); Assert(pMemLnx->Core.pv);
470	if (pMemLnx->Core.u.ResVirt.R0Process != NIL_RTR0PROCESS)
471	{
472	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
473	Assert(pTask);
474	if (pTask && pTask->mm)
475	{
476	down_write(&pTask->mm->mmap_sem);
477	MY_DO_MUNMAP(pTask->mm, (unsigned long)pMemLnx->Core.pv, pMemLnx->Core.cb);
478	up_write(&pTask->mm->mmap_sem);
479	}
480	}
481	else
482	vunmap(pMemLnx->Core.pv);
483	pMemLnx->Core.pv = NULL;
484	break;
485
486	default:
487	AssertMsgFailed(("enmType=%d\n", pMemLnx->Core.enmType));
488	return VERR_INTERNAL_ERROR;
489	}
490	return VINF_SUCCESS;
491	}
492
493
494	int rtR0MemObjNativeAllocPage(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable)
495	{
496	PRTR0MEMOBJLNX pMemLnx;
497	int rc;
498
499	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
500	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_PAGE, cb, GFP_HIGHUSER, false /* non-contiguous */);
501	#else
502	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_PAGE, cb, GFP_USER, false /* non-contiguous */);
503	#endif
504	if (RT_SUCCESS(rc))
505	{
506	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
507	if (RT_SUCCESS(rc))
508	{
509	*ppMem = &pMemLnx->Core;
510	return rc;
511	}
512
513	rtR0MemObjLinuxFreePages(pMemLnx);
514	rtR0MemObjDelete(&pMemLnx->Core);
515	}
516
517	return rc;
518	}
519
520
521	int rtR0MemObjNativeAllocLow(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable)
522	{
523	PRTR0MEMOBJLNX pMemLnx;
524	int rc;
525
526	#ifdef RT_ARCH_AMD64
527	# ifdef GFP_DMA32
528	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, GFP_DMA32, false /* non-contiguous */);
529	if (RT_FAILURE(rc))
530	# endif
531	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, GFP_DMA, false /* non-contiguous */);
532	#else
533	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, GFP_USER, false /* non-contiguous */);
534	#endif
535	if (RT_SUCCESS(rc))
536	{
537	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
538	if (RT_SUCCESS(rc))
539	{
540	*ppMem = &pMemLnx->Core;
541	return rc;
542	}
543
544	rtR0MemObjLinuxFreePages(pMemLnx);
545	rtR0MemObjDelete(&pMemLnx->Core);
546	}
547
548	return rc;
549	}
550
551
552	int rtR0MemObjNativeAllocCont(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable)
553	{
554	PRTR0MEMOBJLNX pMemLnx;
555	int rc;
556
557	#ifdef RT_ARCH_AMD64
558	# ifdef GFP_DMA32
559	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, GFP_DMA32, true /* contiguous */);
560	if (RT_FAILURE(rc))
561	# endif
562	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, GFP_DMA, true /* contiguous */);
563	#else
564	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, GFP_USER, true /* contiguous */);
565	#endif
566	if (RT_SUCCESS(rc))
567	{
568	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
569	if (RT_SUCCESS(rc))
570	{
571	#if defined(RT_STRICT) && (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_HIGHMEM64G))
572	size_t iPage = pMemLnx->cPages;
573	while (iPage-- > 0)
574	Assert(page_to_phys(pMemLnx->apPages[iPage]) < _4G);
575	#endif
576	pMemLnx->Core.u.Cont.Phys = page_to_phys(pMemLnx->apPages[0]);
577	*ppMem = &pMemLnx->Core;
578	return rc;
579	}
580
581	rtR0MemObjLinuxFreePages(pMemLnx);
582	rtR0MemObjDelete(&pMemLnx->Core);
583	}
584
585	return rc;
586	}
587
588
589	/**
590	* Worker for rtR0MemObjLinuxAllocPhysSub that tries one allocation strategy.
591	*
592	* @returns IPRT status.
593	* @param ppMemLnx Where to
594	* @param enmType The object type.
595	* @param cb The size of the allocation.
596	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
597	* @param fGfp The Linux GFP flags to use for the allocation.
598	*/
599	static int rtR0MemObjLinuxAllocPhysSub2(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType, size_t cb, RTHCPHYS PhysHighest, unsigned fGfp)
600	{
601	PRTR0MEMOBJLNX pMemLnx;
602	int rc;
603
604	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, enmType, cb, fGfp,
605	enmType == RTR0MEMOBJTYPE_PHYS /* contiguous / non-contiguous */);
606	if (RT_FAILURE(rc))
607	return rc;
608
609	/*
610	* Check the addresses if necessary. (Can be optimized a bit for PHYS.)
611	*/
612	if (PhysHighest != NIL_RTHCPHYS)
613	{
614	size_t iPage = pMemLnx->cPages;
615	while (iPage-- > 0)
616	if (page_to_phys(pMemLnx->apPages[iPage]) >= PhysHighest)
617	{
618	rtR0MemObjLinuxFreePages(pMemLnx);
619	rtR0MemObjDelete(&pMemLnx->Core);
620	return VERR_NO_MEMORY;
621	}
622	}
623
624	/*
625	* Complete the object.
626	*/
627	if (enmType == RTR0MEMOBJTYPE_PHYS)
628	{
629	pMemLnx->Core.u.Phys.PhysBase = page_to_phys(pMemLnx->apPages[0]);
630	pMemLnx->Core.u.Phys.fAllocated = true;
631	}
632	*ppMem = &pMemLnx->Core;
633	return rc;
634	}
635
636
637	/**
638	* Worker for rtR0MemObjNativeAllocPhys and rtR0MemObjNativeAllocPhysNC.
639	*
640	* @returns IPRT status.
641	* @param ppMem Where to store the memory object pointer on success.
642	* @param enmType The object type.
643	* @param cb The size of the allocation.
644	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
645	*/
646	static int rtR0MemObjLinuxAllocPhysSub(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType, size_t cb, RTHCPHYS PhysHighest)
647	{
648	int rc;
649
650	/*
651	* There are two clear cases and that's the <=16MB and anything-goes ones.
652	* When the physical address limit is somewhere inbetween those two we'll
653	* just have to try, starting with HIGHUSER and working our way thru the
654	* different types, hoping we'll get lucky.
655	*
656	* We should probably move this physical address restriction logic up to
657	* the page alloc function as it would be more efficient there. But since
658	* we don't expect this to be a performance issue just yet it can wait.
659	*/
660	if (PhysHighest == NIL_RTHCPHYS)
661	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, PhysHighest, GFP_HIGHUSER);
662	else if (PhysHighest <= _1M * 16)
663	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, PhysHighest, GFP_DMA);
664	else
665	{
666	rc = VERR_NO_MEMORY;
667	if (RT_FAILURE(rc))
668	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, PhysHighest, GFP_HIGHUSER);
669	if (RT_FAILURE(rc))
670	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, PhysHighest, GFP_USER);
671	#ifdef GFP_DMA32
672	if (RT_FAILURE(rc))
673	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, PhysHighest, GFP_DMA32);
674	#endif
675	if (RT_FAILURE(rc))
676	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, PhysHighest, GFP_DMA);
677	}
678	return rc;
679	}
680
681
682	int rtR0MemObjNativeAllocPhys(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest)
683	{
684	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS, cb, PhysHighest);
685	}
686
687
688	int rtR0MemObjNativeAllocPhysNC(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest)
689	{
690	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS_NC, cb, PhysHighest);
691	}
692
693
694	int rtR0MemObjNativeEnterPhys(PPRTR0MEMOBJINTERNAL ppMem, RTHCPHYS Phys, size_t cb)
695	{
696	/*
697	* All we need to do here is to validate that we can use
698	* ioremap on the specified address (32/64-bit dma_addr_t).
699	*/
700	PRTR0MEMOBJLNX pMemLnx;
701	dma_addr_t PhysAddr = Phys;
702	AssertMsgReturn(PhysAddr == Phys, ("%#llx\n", (unsigned long long)Phys), VERR_ADDRESS_TOO_BIG);
703
704	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_PHYS, NULL, cb);
705	if (!pMemLnx)
706	return VERR_NO_MEMORY;
707
708	pMemLnx->Core.u.Phys.PhysBase = PhysAddr;
709	pMemLnx->Core.u.Phys.fAllocated = false;
710	Assert(!pMemLnx->cPages);
711	*ppMem = &pMemLnx->Core;
712	return VINF_SUCCESS;
713	}
714
715
716	int rtR0MemObjNativeLockUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3Ptr, size_t cb, RTR0PROCESS R0Process)
717	{
718	const int cPages = cb >> PAGE_SHIFT;
719	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
720	struct vm_area_struct **papVMAs;
721	PRTR0MEMOBJLNX pMemLnx;
722	int rc = VERR_NO_MEMORY;
723
724	/*
725	* Check for valid task and size overflows.
726	*/
727	if (!pTask)
728	return VERR_NOT_SUPPORTED;
729	if (((size_t)cPages << PAGE_SHIFT) != cb)
730	return VERR_OUT_OF_RANGE;
731
732	/*
733	* Allocate the memory object and a temporary buffer for the VMAs.
734	*/
735	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_OFFSETOF(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK, (void *)R3Ptr, cb);
736	if (!pMemLnx)
737	return VERR_NO_MEMORY;
738
739	papVMAs = (struct vm_area_struct *)RTMemAlloc(sizeof(papVMAs) * cPages);
740	if (papVMAs)
741	{
742	down_read(&pTask->mm->mmap_sem);
743
744	/*
745	* Get user pages.
746	*/
747	rc = get_user_pages(pTask, /* Task for fault acounting. */
748	pTask->mm, /* Whose pages. */
749	R3Ptr, /* Where from. */
750	cPages, /* How many pages. */
751	1, /* Write to memory. */
752	0, /* force. */
753	&pMemLnx->apPages[0], /* Page array. */
754	papVMAs); /* vmas */
755	if (rc == cPages)
756	{
757	/*
758	* Flush dcache (required?), protect against fork and _really_ pin the page
759	* table entries. get_user_pages() will protect against swapping out the
760	* pages but it will NOT protect against removing page table entries. This
761	* can be achieved with
762	* - using mlock / mmap(..., MAP_LOCKED, ...) from userland. This requires
763	* an appropriate limit set up with setrlimit(..., RLIMIT_MEMLOCK, ...).
764	* Usual Linux distributions support only a limited size of locked pages
765	* (e.g. 32KB).
766	* - setting the PageReserved bit (as we do in rtR0MemObjLinuxAllocPages()
767	* or by
768	* - setting the VM_LOCKED flag. This is the same as doing mlock() without
769	* a range check.
770	*/
771	/** @todo The Linux fork() protection will require more work if this API
772	* is to be used for anything but locking VM pages. */
773	while (rc-- > 0)
774	{
775	flush_dcache_page(pMemLnx->apPages[rc]);
776	papVMAs[rc]->vm_flags \|= (VM_DONTCOPY \| VM_LOCKED);
777	}
778
779	up_read(&pTask->mm->mmap_sem);
780
781	RTMemFree(papVMAs);
782
783	pMemLnx->Core.u.Lock.R0Process = R0Process;
784	pMemLnx->cPages = cPages;
785	Assert(!pMemLnx->fMappedToRing0);
786	*ppMem = &pMemLnx->Core;
787
788	return VINF_SUCCESS;
789	}
790
791	/*
792	* Failed - we need to unlock any pages that we succeeded to lock.
793	*/
794	while (rc-- > 0)
795	{
796	if (!PageReserved(pMemLnx->apPages[rc]))
797	SetPageDirty(pMemLnx->apPages[rc]);
798	page_cache_release(pMemLnx->apPages[rc]);
799	}
800
801	up_read(&pTask->mm->mmap_sem);
802
803	RTMemFree(papVMAs);
804	rc = VERR_LOCK_FAILED;
805	}
806
807	rtR0MemObjDelete(&pMemLnx->Core);
808	return rc;
809	}
810
811
812	int rtR0MemObjNativeLockKernel(PPRTR0MEMOBJINTERNAL ppMem, void *pv, size_t cb)
813	{
814	void pvLast = (uint8_t )pv + cb - 1;
815	size_t const cPages = cb >> PAGE_SHIFT;
816	PRTR0MEMOBJLNX pMemLnx;
817	bool fLinearMapping;
818	int rc;
819	uint8_t *pbPage;
820	size_t iPage;
821
822	/*
823	* Classify the memory and check that we can deal with it.
824	*/
825	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
826	fLinearMapping = virt_addr_valid(pvLast) && virt_addr_valid(pv);
827	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 0)
828	fLinearMapping = VALID_PAGE(virt_to_page(pvLast)) && VALID_PAGE(virt_to_page(pv));
829	#else
830	# error "not supported"
831	#endif
832	if (!fLinearMapping)
833	{
834	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 19)
835	if ( !RTR0MemKernelIsValidAddr(pv)
836	\|\| !RTR0MemKernelIsValidAddr(pv + cb))
837	#endif
838	return VERR_INVALID_PARAMETER;
839	}
840
841	/*
842	* Allocate the memory object.
843	*/
844	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_OFFSETOF(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK, pv, cb);
845	if (!pMemLnx)
846	return VERR_NO_MEMORY;
847
848	/*
849	* Gather the pages.
850	* We ASSUME all kernel pages are non-swappable.
851	*/
852	rc = VINF_SUCCESS;
853	pbPage = (uint8_t *)pvLast;
854	iPage = cPages;
855	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 19)
856	if (!fLinearMapping)
857	{
858	while (iPage-- > 0)
859	{
860	struct page *pPage = vmalloc_to_page(pbPage);
861	if (RT_UNLIKELY(!pPage))
862	{
863	rc = VERR_LOCK_FAILED;
864	break;
865	}
866	pMemLnx->apPages[iPage] = pPage;
867	pbPage -= PAGE_SIZE;
868	}
869	}
870	else
871	#endif
872	{
873	while (iPage-- > 0)
874	{
875	pMemLnx->apPages[iPage] = virt_to_page(pbPage);
876	pbPage -= PAGE_SIZE;
877	}
878	}
879	if (RT_SUCCESS(rc))
880	{
881	/*
882	* Complete the memory object and return.
883	*/
884	pMemLnx->Core.u.Lock.R0Process = NIL_RTR0PROCESS;
885	pMemLnx->cPages = cPages;
886	Assert(!pMemLnx->fMappedToRing0);
887	*ppMem = &pMemLnx->Core;
888
889	return VINF_SUCCESS;
890	}
891
892	rtR0MemObjDelete(&pMemLnx->Core);
893	return rc;
894	}
895
896
897	int rtR0MemObjNativeReserveKernel(PPRTR0MEMOBJINTERNAL ppMem, void *pvFixed, size_t cb, size_t uAlignment)
898	{
899	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
900	const size_t cPages = cb >> PAGE_SHIFT;
901	struct page *pDummyPage;
902	struct page **papPages;
903
904	/* check for unsupported stuff. */
905	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
906	if (uAlignment > PAGE_SIZE)
907	return VERR_NOT_SUPPORTED;
908
909	/*
910	* Allocate a dummy page and create a page pointer array for vmap such that
911	* the dummy page is mapped all over the reserved area.
912	*/
913	pDummyPage = alloc_page(GFP_HIGHUSER);
914	if (!pDummyPage)
915	return VERR_NO_MEMORY;
916	papPages = RTMemAlloc(sizeof(papPages) cPages);
917	if (papPages)
918	{
919	void *pv;
920	size_t iPage = cPages;
921	while (iPage-- > 0)
922	papPages[iPage] = pDummyPage;
923	# ifdef VM_MAP
924	pv = vmap(papPages, cPages, VM_MAP, PAGE_KERNEL_RO);
925	# else
926	pv = vmap(papPages, cPages, VM_ALLOC, PAGE_KERNEL_RO);
927	# endif
928	RTMemFree(papPages);
929	if (pv)
930	{
931	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb);
932	if (pMemLnx)
933	{
934	pMemLnx->Core.u.ResVirt.R0Process = NIL_RTR0PROCESS;
935	pMemLnx->cPages = 1;
936	pMemLnx->apPages[0] = pDummyPage;
937	*ppMem = &pMemLnx->Core;
938	return VINF_SUCCESS;
939	}
940	vunmap(pv);
941	}
942	}
943	__free_page(pDummyPage);
944	return VERR_NO_MEMORY;
945
946	#else /* < 2.4.22 */
947	/*
948	* Could probably use ioremap here, but the caller is in a better position than us
949	* to select some safe physical memory.
950	*/
951	return VERR_NOT_SUPPORTED;
952	#endif
953	}
954
955
956	/**
957	* Worker for rtR0MemObjNativeReserveUser and rtR0MemObjNativerMapUser that creates
958	* an empty user space mapping.
959	*
960	* The caller takes care of acquiring the mmap_sem of the task.
961	*
962	* @returns Pointer to the mapping.
963	* (void *)-1 on failure.
964	* @param R3PtrFixed (RTR3PTR)-1 if anywhere, otherwise a specific location.
965	* @param cb The size of the mapping.
966	* @param uAlignment The alignment of the mapping.
967	* @param pTask The Linux task to create this mapping in.
968	* @param fProt The RTMEM_PROT_* mask.
969	*/
970	static void rtR0MemObjLinuxDoMmap(RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment, struct task_struct pTask, unsigned fProt)
971	{
972	unsigned fLnxProt;
973	unsigned long ulAddr;
974
975	/*
976	* Convert from IPRT protection to mman.h PROT_ and call do_mmap.
977	*/
978	fProt &= (RTMEM_PROT_NONE \| RTMEM_PROT_READ \| RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC);
979	if (fProt == RTMEM_PROT_NONE)
980	fLnxProt = PROT_NONE;
981	else
982	{
983	fLnxProt = 0;
984	if (fProt & RTMEM_PROT_READ)
985	fLnxProt \|= PROT_READ;
986	if (fProt & RTMEM_PROT_WRITE)
987	fLnxProt \|= PROT_WRITE;
988	if (fProt & RTMEM_PROT_EXEC)
989	fLnxProt \|= PROT_EXEC;
990	}
991
992	if (R3PtrFixed != (RTR3PTR)-1)
993	ulAddr = do_mmap(NULL, R3PtrFixed, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS \| MAP_FIXED, 0);
994	else
995	{
996	ulAddr = do_mmap(NULL, 0, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS, 0);
997	if ( !(ulAddr & ~PAGE_MASK)
998	&& (ulAddr & (uAlignment - 1)))
999	{
1000	/** @todo implement uAlignment properly... We'll probably need to make some dummy mappings to fill
1001	* up alignment gaps. This is of course complicated by fragmentation (which we might have cause
1002	* ourselves) and further by there begin two mmap strategies (top / bottom). */
1003	/* For now, just ignore uAlignment requirements... */
1004	}
1005	}
1006	if (ulAddr & ~PAGE_MASK) /* ~PAGE_MASK == PAGE_OFFSET_MASK */
1007	return (void *)-1;
1008	return (void *)ulAddr;
1009	}
1010
1011
1012	int rtR0MemObjNativeReserveUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment, RTR0PROCESS R0Process)
1013	{
1014	PRTR0MEMOBJLNX pMemLnx;
1015	void *pv;
1016	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1017	if (!pTask)
1018	return VERR_NOT_SUPPORTED;
1019
1020	/*
1021	* Check that the specified alignment is supported.
1022	*/
1023	if (uAlignment > PAGE_SIZE)
1024	return VERR_NOT_SUPPORTED;
1025
1026	/*
1027	* Let rtR0MemObjLinuxDoMmap do the difficult bits.
1028	*/
1029	down_write(&pTask->mm->mmap_sem);
1030	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, cb, uAlignment, pTask, RTMEM_PROT_NONE);
1031	up_write(&pTask->mm->mmap_sem);
1032	if (pv == (void *)-1)
1033	return VERR_NO_MEMORY;
1034
1035	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb);
1036	if (!pMemLnx)
1037	{
1038	down_write(&pTask->mm->mmap_sem);
1039	MY_DO_MUNMAP(pTask->mm, (unsigned long)pv, cb);
1040	up_write(&pTask->mm->mmap_sem);
1041	return VERR_NO_MEMORY;
1042	}
1043
1044	pMemLnx->Core.u.ResVirt.R0Process = R0Process;
1045	*ppMem = &pMemLnx->Core;
1046	return VINF_SUCCESS;
1047	}
1048
1049
1050	int rtR0MemObjNativeMapKernel(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap, void *pvFixed, size_t uAlignment,
1051	unsigned fProt, size_t offSub, size_t cbSub)
1052	{
1053	int rc = VERR_NO_MEMORY;
1054	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1055	PRTR0MEMOBJLNX pMemLnx;
1056
1057	/* Fail if requested to do something we can't. */
1058	AssertMsgReturn(!offSub && !cbSub, ("%#x %#x\n", offSub, cbSub), VERR_NOT_SUPPORTED);
1059	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
1060	if (uAlignment > PAGE_SIZE)
1061	return VERR_NOT_SUPPORTED;
1062
1063	/*
1064	* Create the IPRT memory object.
1065	*/
1066	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, pMemLnxToMap->Core.cb);
1067	if (pMemLnx)
1068	{
1069	if (pMemLnxToMap->cPages)
1070	{
1071	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1072	/*
1073	* Use vmap - 2.4.22 and later.
1074	*/
1075	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, true /* kernel */);
1076	# ifdef VM_MAP
1077	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[0], pMemLnxToMap->cPages, VM_MAP, fPg);
1078	# else
1079	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[0], pMemLnxToMap->cPages, VM_ALLOC, fPg);
1080	# endif
1081	if (pMemLnx->Core.pv)
1082	{
1083	pMemLnx->fMappedToRing0 = true;
1084	rc = VINF_SUCCESS;
1085	}
1086	else
1087	rc = VERR_MAP_FAILED;
1088
1089	#else /* < 2.4.22 */
1090	/*
1091	* Only option here is to share mappings if possible and forget about fProt.
1092	*/
1093	if (rtR0MemObjIsRing3(pMemToMap))
1094	rc = VERR_NOT_SUPPORTED;
1095	else
1096	{
1097	rc = VINF_SUCCESS;
1098	if (!pMemLnxToMap->Core.pv)
1099	rc = rtR0MemObjLinuxVMap(pMemLnxToMap, !!(fProt & RTMEM_PROT_EXEC));
1100	if (RT_SUCCESS(rc))
1101	{
1102	Assert(pMemLnxToMap->Core.pv);
1103	pMemLnx->Core.pv = pMemLnxToMap->Core.pv;
1104	}
1105	}
1106	#endif
1107	}
1108	else
1109	{
1110	/*
1111	* MMIO / physical memory.
1112	*/
1113	Assert(pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS && !pMemLnxToMap->Core.u.Phys.fAllocated);
1114	pMemLnx->Core.pv = ioremap(pMemLnxToMap->Core.u.Phys.PhysBase, pMemLnxToMap->Core.cb);
1115	if (pMemLnx->Core.pv)
1116	{
1117	/** @todo fix protection. */
1118	rc = VINF_SUCCESS;
1119	}
1120	}
1121	if (RT_SUCCESS(rc))
1122	{
1123	pMemLnx->Core.u.Mapping.R0Process = NIL_RTR0PROCESS;
1124	*ppMem = &pMemLnx->Core;
1125	return VINF_SUCCESS;
1126	}
1127	rtR0MemObjDelete(&pMemLnx->Core);
1128	}
1129
1130	return rc;
1131	}
1132
1133
1134	int rtR0MemObjNativeMapUser(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap, RTR3PTR R3PtrFixed, size_t uAlignment, unsigned fProt, RTR0PROCESS R0Process)
1135	{
1136	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1137	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1138	int rc = VERR_NO_MEMORY;
1139	PRTR0MEMOBJLNX pMemLnx;
1140
1141	/*
1142	* Check for restrictions.
1143	*/
1144	if (!pTask)
1145	return VERR_NOT_SUPPORTED;
1146	if (uAlignment > PAGE_SIZE)
1147	return VERR_NOT_SUPPORTED;
1148
1149	/*
1150	* Create the IPRT memory object.
1151	*/
1152	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, pMemLnxToMap->Core.cb);
1153	if (pMemLnx)
1154	{
1155	/*
1156	* Allocate user space mapping.
1157	*/
1158	void *pv;
1159	down_write(&pTask->mm->mmap_sem);
1160	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, pMemLnxToMap->Core.cb, uAlignment, pTask, fProt);
1161	if (pv != (void *)-1)
1162	{
1163	/*
1164	* Map page by page into the mmap area.
1165	* This is generic, paranoid and not very efficient.
1166	*/
1167	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, false /* user */);
1168	unsigned long ulAddrCur = (unsigned long)pv;
1169	const size_t cPages = pMemLnxToMap->Core.cb >> PAGE_SHIFT;
1170	size_t iPage;
1171	rc = 0;
1172	if (pMemLnxToMap->cPages)
1173	{
1174	for (iPage = 0; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE)
1175	{
1176	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1177	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
1178	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
1179	#endif
1180
1181	#if defined(VBOX_USE_INSERT_PAGE) && LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1182	rc = vm_insert_page(vma, ulAddrCur, pMemLnxToMap->apPages[iPage]);
1183	vma->vm_flags \|= VM_RESERVED; /* This flag helps making 100% sure some bad stuff wont happen (swap, core, ++). */
1184	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
1185	rc = remap_pfn_range(vma, ulAddrCur, page_to_pfn(pMemLnxToMap->apPages[iPage]), PAGE_SIZE, fPg);
1186	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1187	rc = remap_page_range(vma, ulAddrCur, page_to_phys(pMemLnxToMap->apPages[iPage]), PAGE_SIZE, fPg);
1188	#else /* 2.4 */
1189	rc = remap_page_range(ulAddrCur, page_to_phys(pMemLnxToMap->apPages[iPage]), PAGE_SIZE, fPg);
1190	#endif
1191	if (rc)
1192	break;
1193	}
1194	}
1195	else
1196	{
1197	RTHCPHYS Phys;
1198	if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS)
1199	Phys = pMemLnxToMap->Core.u.Phys.PhysBase;
1200	else if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_CONT)
1201	Phys = pMemLnxToMap->Core.u.Cont.Phys;
1202	else
1203	{
1204	AssertMsgFailed(("%d\n", pMemLnxToMap->Core.enmType));
1205	Phys = NIL_RTHCPHYS;
1206	}
1207	if (Phys != NIL_RTHCPHYS)
1208	{
1209	for (iPage = 0; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE, Phys += PAGE_SIZE)
1210	{
1211	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1212	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
1213	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
1214	#endif
1215
1216	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
1217	rc = remap_pfn_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1218	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1219	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1220	#else /* 2.4 */
1221	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
1222	#endif
1223	if (rc)
1224	break;
1225	}
1226	}
1227	}
1228	if (!rc)
1229	{
1230	up_write(&pTask->mm->mmap_sem);
1231
1232	pMemLnx->Core.pv = pv;
1233	pMemLnx->Core.u.Mapping.R0Process = R0Process;
1234	*ppMem = &pMemLnx->Core;
1235	return VINF_SUCCESS;
1236	}
1237
1238	/*
1239	* Bail out.
1240	*/
1241	MY_DO_MUNMAP(pTask->mm, (unsigned long)pv, pMemLnxToMap->Core.cb);
1242	if (rc != VERR_INTERNAL_ERROR)
1243	rc = VERR_NO_MEMORY;
1244	}
1245
1246	up_write(&pTask->mm->mmap_sem);
1247
1248	rtR0MemObjDelete(&pMemLnx->Core);
1249	}
1250
1251	return rc;
1252	}
1253
1254
1255	int rtR0MemObjNativeProtect(PRTR0MEMOBJINTERNAL pMem, size_t offSub, size_t cbSub, uint32_t fProt)
1256	{
1257	NOREF(pMem);
1258	NOREF(offSub);
1259	NOREF(cbSub);
1260	NOREF(fProt);
1261	return VERR_NOT_SUPPORTED;
1262	}
1263
1264
1265	RTHCPHYS rtR0MemObjNativeGetPagePhysAddr(PRTR0MEMOBJINTERNAL pMem, size_t iPage)
1266	{
1267	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
1268
1269	if (pMemLnx->cPages)
1270	return page_to_phys(pMemLnx->apPages[iPage]);
1271
1272	switch (pMemLnx->Core.enmType)
1273	{
1274	case RTR0MEMOBJTYPE_CONT:
1275	return pMemLnx->Core.u.Cont.Phys + (iPage << PAGE_SHIFT);
1276
1277	case RTR0MEMOBJTYPE_PHYS:
1278	return pMemLnx->Core.u.Phys.PhysBase + (iPage << PAGE_SHIFT);
1279
1280	/* the parent knows */
1281	case RTR0MEMOBJTYPE_MAPPING:
1282	return rtR0MemObjNativeGetPagePhysAddr(pMemLnx->Core.uRel.Child.pParent, iPage);
1283
1284	/* cPages > 0 */
1285	case RTR0MEMOBJTYPE_LOW:
1286	case RTR0MEMOBJTYPE_LOCK:
1287	case RTR0MEMOBJTYPE_PHYS_NC:
1288	case RTR0MEMOBJTYPE_PAGE:
1289	default:
1290	AssertMsgFailed(("%d\n", pMemLnx->Core.enmType));
1291	/* fall thru */
1292
1293	case RTR0MEMOBJTYPE_RES_VIRT:
1294	return NIL_RTHCPHYS;
1295	}
1296	}
1297

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

source: vbox/trunk/src/VBox/Runtime/r0drv/linux/memobj-r0drv-linux.c@ 21497

Download in other formats: