memobj-r0drv-linux.c@ 26864

Last change on this file since 26864 was 26864, checked in by vboxsync, 15 years ago
Runtime/r0drv/memobj-linux: alignment for rtR0MemObjNativeAllocPhys()
Property svn:eol-style set to `native` Property svn:keywords set to `Id Rev`
File size: 45.2 KB

Line
1	/* $Revision: 26864 $ */
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, size_t uAlignment)
703	{
704	/*
705	* Contiguous physical memory: Check for the alignment of the physical address.
706	* Use the knowledge that we are using alloc_pages(flags, order) for allocating
707	* contiguous physical memory which is always aligned at 2^(PAGE_SHIFT+order).
708	*/
709	if (uAlignment > cb)
710	return VERR_NOT_SUPPORTED;
711
712	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS, cb, PhysHighest);
713	}
714
715
716	int rtR0MemObjNativeAllocPhysNC(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest)
717	{
718	/*
719	* Non-contiguous memory.
720	*/
721	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS_NC, cb, PhysHighest);
722	}
723
724
725	int rtR0MemObjNativeEnterPhys(PPRTR0MEMOBJINTERNAL ppMem, RTHCPHYS Phys, size_t cb)
726	{
727	/*
728	* All we need to do here is to validate that we can use
729	* ioremap on the specified address (32/64-bit dma_addr_t).
730	*/
731	PRTR0MEMOBJLNX pMemLnx;
732	dma_addr_t PhysAddr = Phys;
733	AssertMsgReturn(PhysAddr == Phys, ("%#llx\n", (unsigned long long)Phys), VERR_ADDRESS_TOO_BIG);
734
735	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_PHYS, NULL, cb);
736	if (!pMemLnx)
737	return VERR_NO_MEMORY;
738
739	pMemLnx->Core.u.Phys.PhysBase = PhysAddr;
740	pMemLnx->Core.u.Phys.fAllocated = false;
741	Assert(!pMemLnx->cPages);
742	*ppMem = &pMemLnx->Core;
743	return VINF_SUCCESS;
744	}
745
746
747	int rtR0MemObjNativeLockUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3Ptr, size_t cb, uint32_t fAccess, RTR0PROCESS R0Process)
748	{
749	const int cPages = cb >> PAGE_SHIFT;
750	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
751	struct vm_area_struct **papVMAs;
752	PRTR0MEMOBJLNX pMemLnx;
753	int rc = VERR_NO_MEMORY;
754	NOREF(fAccess);
755
756	/*
757	* Check for valid task and size overflows.
758	*/
759	if (!pTask)
760	return VERR_NOT_SUPPORTED;
761	if (((size_t)cPages << PAGE_SHIFT) != cb)
762	return VERR_OUT_OF_RANGE;
763
764	/*
765	* Allocate the memory object and a temporary buffer for the VMAs.
766	*/
767	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_OFFSETOF(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK, (void *)R3Ptr, cb);
768	if (!pMemLnx)
769	return VERR_NO_MEMORY;
770
771	papVMAs = (struct vm_area_struct *)RTMemAlloc(sizeof(papVMAs) * cPages);
772	if (papVMAs)
773	{
774	down_read(&pTask->mm->mmap_sem);
775
776	/*
777	* Get user pages.
778	*/
779	rc = get_user_pages(pTask, /* Task for fault acounting. */
780	pTask->mm, /* Whose pages. */
781	R3Ptr, /* Where from. */
782	cPages, /* How many pages. */
783	1, /* Write to memory. */
784	0, /* force. */
785	&pMemLnx->apPages[0], /* Page array. */
786	papVMAs); /* vmas */
787	if (rc == cPages)
788	{
789	/*
790	* Flush dcache (required?), protect against fork and _really_ pin the page
791	* table entries. get_user_pages() will protect against swapping out the
792	* pages but it will NOT protect against removing page table entries. This
793	* can be achieved with
794	* - using mlock / mmap(..., MAP_LOCKED, ...) from userland. This requires
795	* an appropriate limit set up with setrlimit(..., RLIMIT_MEMLOCK, ...).
796	* Usual Linux distributions support only a limited size of locked pages
797	* (e.g. 32KB).
798	* - setting the PageReserved bit (as we do in rtR0MemObjLinuxAllocPages()
799	* or by
800	* - setting the VM_LOCKED flag. This is the same as doing mlock() without
801	* a range check.
802	*/
803	/** @todo The Linux fork() protection will require more work if this API
804	* is to be used for anything but locking VM pages. */
805	while (rc-- > 0)
806	{
807	flush_dcache_page(pMemLnx->apPages[rc]);
808	papVMAs[rc]->vm_flags \|= (VM_DONTCOPY \| VM_LOCKED);
809	}
810
811	up_read(&pTask->mm->mmap_sem);
812
813	RTMemFree(papVMAs);
814
815	pMemLnx->Core.u.Lock.R0Process = R0Process;
816	pMemLnx->cPages = cPages;
817	Assert(!pMemLnx->fMappedToRing0);
818	*ppMem = &pMemLnx->Core;
819
820	return VINF_SUCCESS;
821	}
822
823	/*
824	* Failed - we need to unlock any pages that we succeeded to lock.
825	*/
826	while (rc-- > 0)
827	{
828	if (!PageReserved(pMemLnx->apPages[rc]))
829	SetPageDirty(pMemLnx->apPages[rc]);
830	page_cache_release(pMemLnx->apPages[rc]);
831	}
832
833	up_read(&pTask->mm->mmap_sem);
834
835	RTMemFree(papVMAs);
836	rc = VERR_LOCK_FAILED;
837	}
838
839	rtR0MemObjDelete(&pMemLnx->Core);
840	return rc;
841	}
842
843
844	int rtR0MemObjNativeLockKernel(PPRTR0MEMOBJINTERNAL ppMem, void *pv, size_t cb, uint32_t fAccess)
845	{
846	void pvLast = (uint8_t )pv + cb - 1;
847	size_t const cPages = cb >> PAGE_SHIFT;
848	PRTR0MEMOBJLNX pMemLnx;
849	bool fLinearMapping;
850	int rc;
851	uint8_t *pbPage;
852	size_t iPage;
853	NOREF(fAccess);
854
855	/*
856	* Classify the memory and check that we can deal with it.
857	*/
858	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
859	fLinearMapping = virt_addr_valid(pvLast) && virt_addr_valid(pv);
860	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 0)
861	fLinearMapping = VALID_PAGE(virt_to_page(pvLast)) && VALID_PAGE(virt_to_page(pv));
862	#else
863	# error "not supported"
864	#endif
865	if (!fLinearMapping)
866	{
867	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 19)
868	if ( !RTR0MemKernelIsValidAddr(pv)
869	\|\| !RTR0MemKernelIsValidAddr(pv + cb))
870	#endif
871	return VERR_INVALID_PARAMETER;
872	}
873
874	/*
875	* Allocate the memory object.
876	*/
877	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_OFFSETOF(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK, pv, cb);
878	if (!pMemLnx)
879	return VERR_NO_MEMORY;
880
881	/*
882	* Gather the pages.
883	* We ASSUME all kernel pages are non-swappable.
884	*/
885	rc = VINF_SUCCESS;
886	pbPage = (uint8_t *)pvLast;
887	iPage = cPages;
888	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 19)
889	if (!fLinearMapping)
890	{
891	while (iPage-- > 0)
892	{
893	struct page *pPage = vmalloc_to_page(pbPage);
894	if (RT_UNLIKELY(!pPage))
895	{
896	rc = VERR_LOCK_FAILED;
897	break;
898	}
899	pMemLnx->apPages[iPage] = pPage;
900	pbPage -= PAGE_SIZE;
901	}
902	}
903	else
904	#endif
905	{
906	while (iPage-- > 0)
907	{
908	pMemLnx->apPages[iPage] = virt_to_page(pbPage);
909	pbPage -= PAGE_SIZE;
910	}
911	}
912	if (RT_SUCCESS(rc))
913	{
914	/*
915	* Complete the memory object and return.
916	*/
917	pMemLnx->Core.u.Lock.R0Process = NIL_RTR0PROCESS;
918	pMemLnx->cPages = cPages;
919	Assert(!pMemLnx->fMappedToRing0);
920	*ppMem = &pMemLnx->Core;
921
922	return VINF_SUCCESS;
923	}
924
925	rtR0MemObjDelete(&pMemLnx->Core);
926	return rc;
927	}
928
929
930	int rtR0MemObjNativeReserveKernel(PPRTR0MEMOBJINTERNAL ppMem, void *pvFixed, size_t cb, size_t uAlignment)
931	{
932	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
933	const size_t cPages = cb >> PAGE_SHIFT;
934	struct page *pDummyPage;
935	struct page **papPages;
936
937	/* check for unsupported stuff. */
938	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
939	if (uAlignment > PAGE_SIZE)
940	return VERR_NOT_SUPPORTED;
941
942	/*
943	* Allocate a dummy page and create a page pointer array for vmap such that
944	* the dummy page is mapped all over the reserved area.
945	*/
946	pDummyPage = alloc_page(GFP_HIGHUSER);
947	if (!pDummyPage)
948	return VERR_NO_MEMORY;
949	papPages = RTMemAlloc(sizeof(papPages) cPages);
950	if (papPages)
951	{
952	void *pv;
953	size_t iPage = cPages;
954	while (iPage-- > 0)
955	papPages[iPage] = pDummyPage;
956	# ifdef VM_MAP
957	pv = vmap(papPages, cPages, VM_MAP, PAGE_KERNEL_RO);
958	# else
959	pv = vmap(papPages, cPages, VM_ALLOC, PAGE_KERNEL_RO);
960	# endif
961	RTMemFree(papPages);
962	if (pv)
963	{
964	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb);
965	if (pMemLnx)
966	{
967	pMemLnx->Core.u.ResVirt.R0Process = NIL_RTR0PROCESS;
968	pMemLnx->cPages = 1;
969	pMemLnx->apPages[0] = pDummyPage;
970	*ppMem = &pMemLnx->Core;
971	return VINF_SUCCESS;
972	}
973	vunmap(pv);
974	}
975	}
976	__free_page(pDummyPage);
977	return VERR_NO_MEMORY;
978
979	#else /* < 2.4.22 */
980	/*
981	* Could probably use ioremap here, but the caller is in a better position than us
982	* to select some safe physical memory.
983	*/
984	return VERR_NOT_SUPPORTED;
985	#endif
986	}
987
988
989	/**
990	* Worker for rtR0MemObjNativeReserveUser and rtR0MemObjNativerMapUser that creates
991	* an empty user space mapping.
992	*
993	* The caller takes care of acquiring the mmap_sem of the task.
994	*
995	* @returns Pointer to the mapping.
996	* (void *)-1 on failure.
997	* @param R3PtrFixed (RTR3PTR)-1 if anywhere, otherwise a specific location.
998	* @param cb The size of the mapping.
999	* @param uAlignment The alignment of the mapping.
1000	* @param pTask The Linux task to create this mapping in.
1001	* @param fProt The RTMEM_PROT_* mask.
1002	*/
1003	static void rtR0MemObjLinuxDoMmap(RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment, struct task_struct pTask, unsigned fProt)
1004	{
1005	unsigned fLnxProt;
1006	unsigned long ulAddr;
1007
1008	/*
1009	* Convert from IPRT protection to mman.h PROT_ and call do_mmap.
1010	*/
1011	fProt &= (RTMEM_PROT_NONE \| RTMEM_PROT_READ \| RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC);
1012	if (fProt == RTMEM_PROT_NONE)
1013	fLnxProt = PROT_NONE;
1014	else
1015	{
1016	fLnxProt = 0;
1017	if (fProt & RTMEM_PROT_READ)
1018	fLnxProt \|= PROT_READ;
1019	if (fProt & RTMEM_PROT_WRITE)
1020	fLnxProt \|= PROT_WRITE;
1021	if (fProt & RTMEM_PROT_EXEC)
1022	fLnxProt \|= PROT_EXEC;
1023	}
1024
1025	if (R3PtrFixed != (RTR3PTR)-1)
1026	ulAddr = do_mmap(NULL, R3PtrFixed, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS \| MAP_FIXED, 0);
1027	else
1028	{
1029	ulAddr = do_mmap(NULL, 0, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS, 0);
1030	if ( !(ulAddr & ~PAGE_MASK)
1031	&& (ulAddr & (uAlignment - 1)))
1032	{
1033	/** @todo implement uAlignment properly... We'll probably need to make some dummy mappings to fill
1034	* up alignment gaps. This is of course complicated by fragmentation (which we might have cause
1035	* ourselves) and further by there begin two mmap strategies (top / bottom). */
1036	/* For now, just ignore uAlignment requirements... */
1037	}
1038	}
1039	if (ulAddr & ~PAGE_MASK) /* ~PAGE_MASK == PAGE_OFFSET_MASK */
1040	return (void *)-1;
1041	return (void *)ulAddr;
1042	}
1043
1044
1045	int rtR0MemObjNativeReserveUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment, RTR0PROCESS R0Process)
1046	{
1047	PRTR0MEMOBJLNX pMemLnx;
1048	void *pv;
1049	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1050	if (!pTask)
1051	return VERR_NOT_SUPPORTED;
1052
1053	/*
1054	* Check that the specified alignment is supported.
1055	*/
1056	if (uAlignment > PAGE_SIZE)
1057	return VERR_NOT_SUPPORTED;
1058
1059	/*
1060	* Let rtR0MemObjLinuxDoMmap do the difficult bits.
1061	*/
1062	down_write(&pTask->mm->mmap_sem);
1063	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, cb, uAlignment, pTask, RTMEM_PROT_NONE);
1064	up_write(&pTask->mm->mmap_sem);
1065	if (pv == (void *)-1)
1066	return VERR_NO_MEMORY;
1067
1068	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb);
1069	if (!pMemLnx)
1070	{
1071	down_write(&pTask->mm->mmap_sem);
1072	MY_DO_MUNMAP(pTask->mm, (unsigned long)pv, cb);
1073	up_write(&pTask->mm->mmap_sem);
1074	return VERR_NO_MEMORY;
1075	}
1076
1077	pMemLnx->Core.u.ResVirt.R0Process = R0Process;
1078	*ppMem = &pMemLnx->Core;
1079	return VINF_SUCCESS;
1080	}
1081
1082
1083	int rtR0MemObjNativeMapKernel(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap, void *pvFixed, size_t uAlignment,
1084	unsigned fProt, size_t offSub, size_t cbSub)
1085	{
1086	int rc = VERR_NO_MEMORY;
1087	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1088	PRTR0MEMOBJLNX pMemLnx;
1089
1090	/* Fail if requested to do something we can't. */
1091	AssertMsgReturn(!offSub && !cbSub, ("%#x %#x\n", offSub, cbSub), VERR_NOT_SUPPORTED);
1092	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
1093	if (uAlignment > PAGE_SIZE)
1094	return VERR_NOT_SUPPORTED;
1095
1096	/*
1097	* Create the IPRT memory object.
1098	*/
1099	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, pMemLnxToMap->Core.cb);
1100	if (pMemLnx)
1101	{
1102	if (pMemLnxToMap->cPages)
1103	{
1104	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1105	/*
1106	* Use vmap - 2.4.22 and later.
1107	*/
1108	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, true /* kernel */);
1109	# ifdef VM_MAP
1110	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[0], pMemLnxToMap->cPages, VM_MAP, fPg);
1111	# else
1112	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[0], pMemLnxToMap->cPages, VM_ALLOC, fPg);
1113	# endif
1114	if (pMemLnx->Core.pv)
1115	{
1116	pMemLnx->fMappedToRing0 = true;
1117	rc = VINF_SUCCESS;
1118	}
1119	else
1120	rc = VERR_MAP_FAILED;
1121
1122	#else /* < 2.4.22 */
1123	/*
1124	* Only option here is to share mappings if possible and forget about fProt.
1125	*/
1126	if (rtR0MemObjIsRing3(pMemToMap))
1127	rc = VERR_NOT_SUPPORTED;
1128	else
1129	{
1130	rc = VINF_SUCCESS;
1131	if (!pMemLnxToMap->Core.pv)
1132	rc = rtR0MemObjLinuxVMap(pMemLnxToMap, !!(fProt & RTMEM_PROT_EXEC));
1133	if (RT_SUCCESS(rc))
1134	{
1135	Assert(pMemLnxToMap->Core.pv);
1136	pMemLnx->Core.pv = pMemLnxToMap->Core.pv;
1137	}
1138	}
1139	#endif
1140	}
1141	else
1142	{
1143	/*
1144	* MMIO / physical memory.
1145	*/
1146	Assert(pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS && !pMemLnxToMap->Core.u.Phys.fAllocated);
1147	pMemLnx->Core.pv = ioremap(pMemLnxToMap->Core.u.Phys.PhysBase, pMemLnxToMap->Core.cb);
1148	if (pMemLnx->Core.pv)
1149	{
1150	/** @todo fix protection. */
1151	rc = VINF_SUCCESS;
1152	}
1153	}
1154	if (RT_SUCCESS(rc))
1155	{
1156	pMemLnx->Core.u.Mapping.R0Process = NIL_RTR0PROCESS;
1157	*ppMem = &pMemLnx->Core;
1158	return VINF_SUCCESS;
1159	}
1160	rtR0MemObjDelete(&pMemLnx->Core);
1161	}
1162
1163	return rc;
1164	}
1165
1166
1167	#ifdef VBOX_USE_PAE_HACK
1168	/**
1169	* Replace the PFN of a PTE with the address of the actual page.
1170	*
1171	* The caller maps a reserved dummy page at the address with the desired access
1172	* and flags.
1173	*
1174	* This hack is required for older Linux kernels which don't provide
1175	* remap_pfn_range().
1176	*
1177	* @returns 0 on success, -ENOMEM on failure.
1178	* @param mm The memory context.
1179	* @param ulAddr The mapping address.
1180	* @param Phys The physical address of the page to map.
1181	*/
1182	static int rtR0MemObjLinuxFixPte(struct mm_struct *mm, unsigned long ulAddr, RTHCPHYS Phys)
1183	{
1184	int rc = -ENOMEM;
1185	pgd_t *pgd;
1186
1187	spin_lock(&mm->page_table_lock);
1188
1189	pgd = pgd_offset(mm, ulAddr);
1190	if (!pgd_none(pgd) && !pgd_bad(pgd))
1191	{
1192	pmd_t *pmd = pmd_offset(pgd, ulAddr);
1193	if (!pmd_none(*pmd))
1194	{
1195	pte_t *ptep = pte_offset_map(pmd, ulAddr);
1196	if (ptep)
1197	{
1198	pte_t pte = *ptep;
1199	pte.pte_high &= 0xfff00000;
1200	pte.pte_high \|= ((Phys >> 32) & 0x000fffff);
1201	pte.pte_low &= 0x00000fff;
1202	pte.pte_low \|= (Phys & 0xfffff000);
1203	set_pte(ptep, pte);
1204	pte_unmap(ptep);
1205	rc = 0;
1206	}
1207	}
1208	}
1209
1210	spin_unlock(&mm->page_table_lock);
1211	return rc;
1212	}
1213	#endif /* VBOX_USE_PAE_HACK */
1214
1215
1216	int rtR0MemObjNativeMapUser(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap, RTR3PTR R3PtrFixed, size_t uAlignment, unsigned fProt, RTR0PROCESS R0Process)
1217	{
1218	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1219	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1220	int rc = VERR_NO_MEMORY;
1221	PRTR0MEMOBJLNX pMemLnx;
1222	#ifdef VBOX_USE_PAE_HACK
1223	struct page *pDummyPage;
1224	RTHCPHYS DummyPhys;
1225	#endif
1226
1227	/*
1228	* Check for restrictions.
1229	*/
1230	if (!pTask)
1231	return VERR_NOT_SUPPORTED;
1232	if (uAlignment > PAGE_SIZE)
1233	return VERR_NOT_SUPPORTED;
1234
1235	#ifdef VBOX_USE_PAE_HACK
1236	/*
1237	* Allocate a dummy page for use when mapping the memory.
1238	*/
1239	pDummyPage = alloc_page(GFP_USER);
1240	if (!pDummyPage)
1241	return VERR_NO_MEMORY;
1242	SetPageReserved(pDummyPage);
1243	DummyPhys = page_to_phys(pDummyPage);
1244	#endif
1245
1246	/*
1247	* Create the IPRT memory object.
1248	*/
1249	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, pMemLnxToMap->Core.cb);
1250	if (pMemLnx)
1251	{
1252	/*
1253	* Allocate user space mapping.
1254	*/
1255	void *pv;
1256	down_write(&pTask->mm->mmap_sem);
1257	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, pMemLnxToMap->Core.cb, uAlignment, pTask, fProt);
1258	if (pv != (void *)-1)
1259	{
1260	/*
1261	* Map page by page into the mmap area.
1262	* This is generic, paranoid and not very efficient.
1263	*/
1264	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, false /* user */);
1265	unsigned long ulAddrCur = (unsigned long)pv;
1266	const size_t cPages = pMemLnxToMap->Core.cb >> PAGE_SHIFT;
1267	size_t iPage;
1268
1269	rc = 0;
1270	if (pMemLnxToMap->cPages)
1271	{
1272	for (iPage = 0; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE)
1273	{
1274	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11)
1275	RTHCPHYS Phys = page_to_phys(pMemLnxToMap->apPages[iPage]);
1276	#endif
1277	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1278	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
1279	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
1280	#endif
1281	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0) && defined(RT_ARCH_X86)
1282	/* remap_page_range() limitation on x86 */
1283	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
1284	#endif
1285
1286	#if defined(VBOX_USE_INSERT_PAGE) && LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1287	rc = vm_insert_page(vma, ulAddrCur, pMemLnxToMap->apPages[iPage]);
1288	vma->vm_flags \|= VM_RESERVED; /* This flag helps making 100% sure some bad stuff wont happen (swap, core, ++). */
1289	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
1290	rc = remap_pfn_range(vma, ulAddrCur, page_to_pfn(pMemLnxToMap->apPages[iPage]), PAGE_SIZE, fPg);
1291	#elif defined(VBOX_USE_PAE_HACK)
1292	rc = remap_page_range(vma, ulAddrCur, DummyPhys, PAGE_SIZE, fPg);
1293	if (!rc)
1294	rc = rtR0MemObjLinuxFixPte(pTask->mm, ulAddrCur, Phys);
1295	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1296	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1297	#else /* 2.4 */
1298	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
1299	#endif
1300	if (rc)
1301	{
1302	rc = VERR_NO_MEMORY;
1303	break;
1304	}
1305	}
1306	}
1307	else
1308	{
1309	RTHCPHYS Phys;
1310	if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS)
1311	Phys = pMemLnxToMap->Core.u.Phys.PhysBase;
1312	else if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_CONT)
1313	Phys = pMemLnxToMap->Core.u.Cont.Phys;
1314	else
1315	{
1316	AssertMsgFailed(("%d\n", pMemLnxToMap->Core.enmType));
1317	Phys = NIL_RTHCPHYS;
1318	}
1319	if (Phys != NIL_RTHCPHYS)
1320	{
1321	for (iPage = 0; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE, Phys += PAGE_SIZE)
1322	{
1323	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1324	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
1325	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
1326	#endif
1327	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0) && defined(RT_ARCH_X86)
1328	/* remap_page_range() limitation on x86 */
1329	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
1330	#endif
1331
1332	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
1333	rc = remap_pfn_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1334	#elif defined(VBOX_USE_PAE_HACK)
1335	rc = remap_page_range(vma, ulAddrCur, DummyPhys, PAGE_SIZE, fPg);
1336	if (!rc)
1337	rc = rtR0MemObjLinuxFixPte(pTask->mm, ulAddrCur, Phys);
1338	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1339	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1340	#else /* 2.4 */
1341	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
1342	#endif
1343	if (rc)
1344	{
1345	rc = VERR_NO_MEMORY;
1346	break;
1347	}
1348	}
1349	}
1350	}
1351	if (!rc)
1352	{
1353	up_write(&pTask->mm->mmap_sem);
1354	#ifdef VBOX_USE_PAE_HACK
1355	__free_page(pDummyPage);
1356	#endif
1357
1358	pMemLnx->Core.pv = pv;
1359	pMemLnx->Core.u.Mapping.R0Process = R0Process;
1360	*ppMem = &pMemLnx->Core;
1361	return VINF_SUCCESS;
1362	}
1363
1364	/*
1365	* Bail out.
1366	*/
1367	MY_DO_MUNMAP(pTask->mm, (unsigned long)pv, pMemLnxToMap->Core.cb);
1368	}
1369	up_write(&pTask->mm->mmap_sem);
1370	rtR0MemObjDelete(&pMemLnx->Core);
1371	}
1372	#ifdef VBOX_USE_PAE_HACK
1373	__free_page(pDummyPage);
1374	#endif
1375
1376	return rc;
1377	}
1378
1379
1380	int rtR0MemObjNativeProtect(PRTR0MEMOBJINTERNAL pMem, size_t offSub, size_t cbSub, uint32_t fProt)
1381	{
1382	NOREF(pMem);
1383	NOREF(offSub);
1384	NOREF(cbSub);
1385	NOREF(fProt);
1386	return VERR_NOT_SUPPORTED;
1387	}
1388
1389
1390	RTHCPHYS rtR0MemObjNativeGetPagePhysAddr(PRTR0MEMOBJINTERNAL pMem, size_t iPage)
1391	{
1392	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
1393
1394	if (pMemLnx->cPages)
1395	return page_to_phys(pMemLnx->apPages[iPage]);
1396
1397	switch (pMemLnx->Core.enmType)
1398	{
1399	case RTR0MEMOBJTYPE_CONT:
1400	return pMemLnx->Core.u.Cont.Phys + (iPage << PAGE_SHIFT);
1401
1402	case RTR0MEMOBJTYPE_PHYS:
1403	return pMemLnx->Core.u.Phys.PhysBase + (iPage << PAGE_SHIFT);
1404
1405	/* the parent knows */
1406	case RTR0MEMOBJTYPE_MAPPING:
1407	return rtR0MemObjNativeGetPagePhysAddr(pMemLnx->Core.uRel.Child.pParent, iPage);
1408
1409	/* cPages > 0 */
1410	case RTR0MEMOBJTYPE_LOW:
1411	case RTR0MEMOBJTYPE_LOCK:
1412	case RTR0MEMOBJTYPE_PHYS_NC:
1413	case RTR0MEMOBJTYPE_PAGE:
1414	default:
1415	AssertMsgFailed(("%d\n", pMemLnx->Core.enmType));
1416	/* fall thru */
1417
1418	case RTR0MEMOBJTYPE_RES_VIRT:
1419	return NIL_RTHCPHYS;
1420	}
1421	}
1422

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

source: vbox/trunk/src/VBox/Runtime/r0drv/linux/memobj-r0drv-linux.c@ 26864

Download in other formats: