memobj-r0drv-linux.c@ 22436

Last change on this file since 22436 was 22355, checked in by vboxsync, 16 years ago
Runtime/Linux: don't allow to pass pages above 4GB to remap_page_range(); some remarks about GFP_USER
Property svn:eol-style set to `native` Property svn:keywords set to `Id Rev`
File size: 41.5 KB

Line
1	/* $Revision: 22355 $ */
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	/** XXX Wrong: GFP_USER can return page frames above 4GB! */
534	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, GFP_USER, false /* non-contiguous */);
535	#endif
536	if (RT_SUCCESS(rc))
537	{
538	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
539	if (RT_SUCCESS(rc))
540	{
541	*ppMem = &pMemLnx->Core;
542	return rc;
543	}
544
545	rtR0MemObjLinuxFreePages(pMemLnx);
546	rtR0MemObjDelete(&pMemLnx->Core);
547	}
548
549	return rc;
550	}
551
552
553	int rtR0MemObjNativeAllocCont(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable)
554	{
555	PRTR0MEMOBJLNX pMemLnx;
556	int rc;
557
558	#ifdef RT_ARCH_AMD64
559	# ifdef GFP_DMA32
560	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, GFP_DMA32, true /* contiguous */);
561	if (RT_FAILURE(rc))
562	# endif
563	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, GFP_DMA, true /* contiguous */);
564	#else
565	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, GFP_USER, true /* contiguous */);
566	#endif
567	if (RT_SUCCESS(rc))
568	{
569	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
570	if (RT_SUCCESS(rc))
571	{
572	#if defined(RT_STRICT) && (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_HIGHMEM64G))
573	size_t iPage = pMemLnx->cPages;
574	while (iPage-- > 0)
575	Assert(page_to_phys(pMemLnx->apPages[iPage]) < _4G);
576	#endif
577	pMemLnx->Core.u.Cont.Phys = page_to_phys(pMemLnx->apPages[0]);
578	*ppMem = &pMemLnx->Core;
579	return rc;
580	}
581
582	rtR0MemObjLinuxFreePages(pMemLnx);
583	rtR0MemObjDelete(&pMemLnx->Core);
584	}
585
586	return rc;
587	}
588
589
590	/**
591	* Worker for rtR0MemObjLinuxAllocPhysSub that tries one allocation strategy.
592	*
593	* @returns IPRT status.
594	* @param ppMemLnx Where to
595	* @param enmType The object type.
596	* @param cb The size of the allocation.
597	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
598	* @param fGfp The Linux GFP flags to use for the allocation.
599	*/
600	static int rtR0MemObjLinuxAllocPhysSub2(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType, size_t cb, RTHCPHYS PhysHighest, unsigned fGfp)
601	{
602	PRTR0MEMOBJLNX pMemLnx;
603	int rc;
604
605	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, enmType, cb, fGfp,
606	enmType == RTR0MEMOBJTYPE_PHYS /* contiguous / non-contiguous */);
607	if (RT_FAILURE(rc))
608	return rc;
609
610	/*
611	* Check the addresses if necessary. (Can be optimized a bit for PHYS.)
612	*/
613	if (PhysHighest != NIL_RTHCPHYS)
614	{
615	size_t iPage = pMemLnx->cPages;
616	while (iPage-- > 0)
617	if (page_to_phys(pMemLnx->apPages[iPage]) >= PhysHighest)
618	{
619	rtR0MemObjLinuxFreePages(pMemLnx);
620	rtR0MemObjDelete(&pMemLnx->Core);
621	return VERR_NO_MEMORY;
622	}
623	}
624
625	/*
626	* Complete the object.
627	*/
628	if (enmType == RTR0MEMOBJTYPE_PHYS)
629	{
630	pMemLnx->Core.u.Phys.PhysBase = page_to_phys(pMemLnx->apPages[0]);
631	pMemLnx->Core.u.Phys.fAllocated = true;
632	}
633	*ppMem = &pMemLnx->Core;
634	return rc;
635	}
636
637
638	/**
639	* Worker for rtR0MemObjNativeAllocPhys and rtR0MemObjNativeAllocPhysNC.
640	*
641	* @returns IPRT status.
642	* @param ppMem Where to store the memory object pointer on success.
643	* @param enmType The object type.
644	* @param cb The size of the allocation.
645	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
646	*/
647	static int rtR0MemObjLinuxAllocPhysSub(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType, size_t cb, RTHCPHYS PhysHighest)
648	{
649	int rc;
650
651	/*
652	* There are two clear cases and that's the <=16MB and anything-goes ones.
653	* When the physical address limit is somewhere inbetween those two we'll
654	* just have to try, starting with HIGHUSER and working our way thru the
655	* different types, hoping we'll get lucky.
656	*
657	* We should probably move this physical address restriction logic up to
658	* the page alloc function as it would be more efficient there. But since
659	* we don't expect this to be a performance issue just yet it can wait.
660	*/
661	if (PhysHighest == NIL_RTHCPHYS)
662	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, PhysHighest, GFP_HIGHUSER);
663	else if (PhysHighest <= _1M * 16)
664	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, PhysHighest, GFP_DMA);
665	else
666	{
667	rc = VERR_NO_MEMORY;
668	if (RT_FAILURE(rc))
669	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, PhysHighest, GFP_HIGHUSER);
670	if (RT_FAILURE(rc))
671	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, PhysHighest, GFP_USER);
672	#ifdef GFP_DMA32
673	if (RT_FAILURE(rc))
674	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, PhysHighest, GFP_DMA32);
675	#endif
676	if (RT_FAILURE(rc))
677	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, PhysHighest, GFP_DMA);
678	}
679	return rc;
680	}
681
682
683	int rtR0MemObjNativeAllocPhys(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest)
684	{
685	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS, cb, PhysHighest);
686	}
687
688
689	int rtR0MemObjNativeAllocPhysNC(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest)
690	{
691	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS_NC, cb, PhysHighest);
692	}
693
694
695	int rtR0MemObjNativeEnterPhys(PPRTR0MEMOBJINTERNAL ppMem, RTHCPHYS Phys, size_t cb)
696	{
697	/*
698	* All we need to do here is to validate that we can use
699	* ioremap on the specified address (32/64-bit dma_addr_t).
700	*/
701	PRTR0MEMOBJLNX pMemLnx;
702	dma_addr_t PhysAddr = Phys;
703	AssertMsgReturn(PhysAddr == Phys, ("%#llx\n", (unsigned long long)Phys), VERR_ADDRESS_TOO_BIG);
704
705	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_PHYS, NULL, cb);
706	if (!pMemLnx)
707	return VERR_NO_MEMORY;
708
709	pMemLnx->Core.u.Phys.PhysBase = PhysAddr;
710	pMemLnx->Core.u.Phys.fAllocated = false;
711	Assert(!pMemLnx->cPages);
712	*ppMem = &pMemLnx->Core;
713	return VINF_SUCCESS;
714	}
715
716
717	int rtR0MemObjNativeLockUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3Ptr, size_t cb, RTR0PROCESS R0Process)
718	{
719	const int cPages = cb >> PAGE_SHIFT;
720	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
721	struct vm_area_struct **papVMAs;
722	PRTR0MEMOBJLNX pMemLnx;
723	int rc = VERR_NO_MEMORY;
724
725	/*
726	* Check for valid task and size overflows.
727	*/
728	if (!pTask)
729	return VERR_NOT_SUPPORTED;
730	if (((size_t)cPages << PAGE_SHIFT) != cb)
731	return VERR_OUT_OF_RANGE;
732
733	/*
734	* Allocate the memory object and a temporary buffer for the VMAs.
735	*/
736	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_OFFSETOF(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK, (void *)R3Ptr, cb);
737	if (!pMemLnx)
738	return VERR_NO_MEMORY;
739
740	papVMAs = (struct vm_area_struct *)RTMemAlloc(sizeof(papVMAs) * cPages);
741	if (papVMAs)
742	{
743	down_read(&pTask->mm->mmap_sem);
744
745	/*
746	* Get user pages.
747	*/
748	rc = get_user_pages(pTask, /* Task for fault acounting. */
749	pTask->mm, /* Whose pages. */
750	R3Ptr, /* Where from. */
751	cPages, /* How many pages. */
752	1, /* Write to memory. */
753	0, /* force. */
754	&pMemLnx->apPages[0], /* Page array. */
755	papVMAs); /* vmas */
756	if (rc == cPages)
757	{
758	/*
759	* Flush dcache (required?), protect against fork and _really_ pin the page
760	* table entries. get_user_pages() will protect against swapping out the
761	* pages but it will NOT protect against removing page table entries. This
762	* can be achieved with
763	* - using mlock / mmap(..., MAP_LOCKED, ...) from userland. This requires
764	* an appropriate limit set up with setrlimit(..., RLIMIT_MEMLOCK, ...).
765	* Usual Linux distributions support only a limited size of locked pages
766	* (e.g. 32KB).
767	* - setting the PageReserved bit (as we do in rtR0MemObjLinuxAllocPages()
768	* or by
769	* - setting the VM_LOCKED flag. This is the same as doing mlock() without
770	* a range check.
771	*/
772	/** @todo The Linux fork() protection will require more work if this API
773	* is to be used for anything but locking VM pages. */
774	while (rc-- > 0)
775	{
776	flush_dcache_page(pMemLnx->apPages[rc]);
777	papVMAs[rc]->vm_flags \|= (VM_DONTCOPY \| VM_LOCKED);
778	}
779
780	up_read(&pTask->mm->mmap_sem);
781
782	RTMemFree(papVMAs);
783
784	pMemLnx->Core.u.Lock.R0Process = R0Process;
785	pMemLnx->cPages = cPages;
786	Assert(!pMemLnx->fMappedToRing0);
787	*ppMem = &pMemLnx->Core;
788
789	return VINF_SUCCESS;
790	}
791
792	/*
793	* Failed - we need to unlock any pages that we succeeded to lock.
794	*/
795	while (rc-- > 0)
796	{
797	if (!PageReserved(pMemLnx->apPages[rc]))
798	SetPageDirty(pMemLnx->apPages[rc]);
799	page_cache_release(pMemLnx->apPages[rc]);
800	}
801
802	up_read(&pTask->mm->mmap_sem);
803
804	RTMemFree(papVMAs);
805	rc = VERR_LOCK_FAILED;
806	}
807
808	rtR0MemObjDelete(&pMemLnx->Core);
809	return rc;
810	}
811
812
813	int rtR0MemObjNativeLockKernel(PPRTR0MEMOBJINTERNAL ppMem, void *pv, size_t cb)
814	{
815	void pvLast = (uint8_t )pv + cb - 1;
816	size_t const cPages = cb >> PAGE_SHIFT;
817	PRTR0MEMOBJLNX pMemLnx;
818	bool fLinearMapping;
819	int rc;
820	uint8_t *pbPage;
821	size_t iPage;
822
823	/*
824	* Classify the memory and check that we can deal with it.
825	*/
826	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
827	fLinearMapping = virt_addr_valid(pvLast) && virt_addr_valid(pv);
828	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 0)
829	fLinearMapping = VALID_PAGE(virt_to_page(pvLast)) && VALID_PAGE(virt_to_page(pv));
830	#else
831	# error "not supported"
832	#endif
833	if (!fLinearMapping)
834	{
835	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 19)
836	if ( !RTR0MemKernelIsValidAddr(pv)
837	\|\| !RTR0MemKernelIsValidAddr(pv + cb))
838	#endif
839	return VERR_INVALID_PARAMETER;
840	}
841
842	/*
843	* Allocate the memory object.
844	*/
845	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_OFFSETOF(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK, pv, cb);
846	if (!pMemLnx)
847	return VERR_NO_MEMORY;
848
849	/*
850	* Gather the pages.
851	* We ASSUME all kernel pages are non-swappable.
852	*/
853	rc = VINF_SUCCESS;
854	pbPage = (uint8_t *)pvLast;
855	iPage = cPages;
856	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 19)
857	if (!fLinearMapping)
858	{
859	while (iPage-- > 0)
860	{
861	struct page *pPage = vmalloc_to_page(pbPage);
862	if (RT_UNLIKELY(!pPage))
863	{
864	rc = VERR_LOCK_FAILED;
865	break;
866	}
867	pMemLnx->apPages[iPage] = pPage;
868	pbPage -= PAGE_SIZE;
869	}
870	}
871	else
872	#endif
873	{
874	while (iPage-- > 0)
875	{
876	pMemLnx->apPages[iPage] = virt_to_page(pbPage);
877	pbPage -= PAGE_SIZE;
878	}
879	}
880	if (RT_SUCCESS(rc))
881	{
882	/*
883	* Complete the memory object and return.
884	*/
885	pMemLnx->Core.u.Lock.R0Process = NIL_RTR0PROCESS;
886	pMemLnx->cPages = cPages;
887	Assert(!pMemLnx->fMappedToRing0);
888	*ppMem = &pMemLnx->Core;
889
890	return VINF_SUCCESS;
891	}
892
893	rtR0MemObjDelete(&pMemLnx->Core);
894	return rc;
895	}
896
897
898	int rtR0MemObjNativeReserveKernel(PPRTR0MEMOBJINTERNAL ppMem, void *pvFixed, size_t cb, size_t uAlignment)
899	{
900	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
901	const size_t cPages = cb >> PAGE_SHIFT;
902	struct page *pDummyPage;
903	struct page **papPages;
904
905	/* check for unsupported stuff. */
906	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
907	if (uAlignment > PAGE_SIZE)
908	return VERR_NOT_SUPPORTED;
909
910	/*
911	* Allocate a dummy page and create a page pointer array for vmap such that
912	* the dummy page is mapped all over the reserved area.
913	*/
914	pDummyPage = alloc_page(GFP_HIGHUSER);
915	if (!pDummyPage)
916	return VERR_NO_MEMORY;
917	papPages = RTMemAlloc(sizeof(papPages) cPages);
918	if (papPages)
919	{
920	void *pv;
921	size_t iPage = cPages;
922	while (iPage-- > 0)
923	papPages[iPage] = pDummyPage;
924	# ifdef VM_MAP
925	pv = vmap(papPages, cPages, VM_MAP, PAGE_KERNEL_RO);
926	# else
927	pv = vmap(papPages, cPages, VM_ALLOC, PAGE_KERNEL_RO);
928	# endif
929	RTMemFree(papPages);
930	if (pv)
931	{
932	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb);
933	if (pMemLnx)
934	{
935	pMemLnx->Core.u.ResVirt.R0Process = NIL_RTR0PROCESS;
936	pMemLnx->cPages = 1;
937	pMemLnx->apPages[0] = pDummyPage;
938	*ppMem = &pMemLnx->Core;
939	return VINF_SUCCESS;
940	}
941	vunmap(pv);
942	}
943	}
944	__free_page(pDummyPage);
945	return VERR_NO_MEMORY;
946
947	#else /* < 2.4.22 */
948	/*
949	* Could probably use ioremap here, but the caller is in a better position than us
950	* to select some safe physical memory.
951	*/
952	return VERR_NOT_SUPPORTED;
953	#endif
954	}
955
956
957	/**
958	* Worker for rtR0MemObjNativeReserveUser and rtR0MemObjNativerMapUser that creates
959	* an empty user space mapping.
960	*
961	* The caller takes care of acquiring the mmap_sem of the task.
962	*
963	* @returns Pointer to the mapping.
964	* (void *)-1 on failure.
965	* @param R3PtrFixed (RTR3PTR)-1 if anywhere, otherwise a specific location.
966	* @param cb The size of the mapping.
967	* @param uAlignment The alignment of the mapping.
968	* @param pTask The Linux task to create this mapping in.
969	* @param fProt The RTMEM_PROT_* mask.
970	*/
971	static void rtR0MemObjLinuxDoMmap(RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment, struct task_struct pTask, unsigned fProt)
972	{
973	unsigned fLnxProt;
974	unsigned long ulAddr;
975
976	/*
977	* Convert from IPRT protection to mman.h PROT_ and call do_mmap.
978	*/
979	fProt &= (RTMEM_PROT_NONE \| RTMEM_PROT_READ \| RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC);
980	if (fProt == RTMEM_PROT_NONE)
981	fLnxProt = PROT_NONE;
982	else
983	{
984	fLnxProt = 0;
985	if (fProt & RTMEM_PROT_READ)
986	fLnxProt \|= PROT_READ;
987	if (fProt & RTMEM_PROT_WRITE)
988	fLnxProt \|= PROT_WRITE;
989	if (fProt & RTMEM_PROT_EXEC)
990	fLnxProt \|= PROT_EXEC;
991	}
992
993	if (R3PtrFixed != (RTR3PTR)-1)
994	ulAddr = do_mmap(NULL, R3PtrFixed, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS \| MAP_FIXED, 0);
995	else
996	{
997	ulAddr = do_mmap(NULL, 0, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS, 0);
998	if ( !(ulAddr & ~PAGE_MASK)
999	&& (ulAddr & (uAlignment - 1)))
1000	{
1001	/** @todo implement uAlignment properly... We'll probably need to make some dummy mappings to fill
1002	* up alignment gaps. This is of course complicated by fragmentation (which we might have cause
1003	* ourselves) and further by there begin two mmap strategies (top / bottom). */
1004	/* For now, just ignore uAlignment requirements... */
1005	}
1006	}
1007	if (ulAddr & ~PAGE_MASK) /* ~PAGE_MASK == PAGE_OFFSET_MASK */
1008	return (void *)-1;
1009	return (void *)ulAddr;
1010	}
1011
1012
1013	int rtR0MemObjNativeReserveUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment, RTR0PROCESS R0Process)
1014	{
1015	PRTR0MEMOBJLNX pMemLnx;
1016	void *pv;
1017	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1018	if (!pTask)
1019	return VERR_NOT_SUPPORTED;
1020
1021	/*
1022	* Check that the specified alignment is supported.
1023	*/
1024	if (uAlignment > PAGE_SIZE)
1025	return VERR_NOT_SUPPORTED;
1026
1027	/*
1028	* Let rtR0MemObjLinuxDoMmap do the difficult bits.
1029	*/
1030	down_write(&pTask->mm->mmap_sem);
1031	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, cb, uAlignment, pTask, RTMEM_PROT_NONE);
1032	up_write(&pTask->mm->mmap_sem);
1033	if (pv == (void *)-1)
1034	return VERR_NO_MEMORY;
1035
1036	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb);
1037	if (!pMemLnx)
1038	{
1039	down_write(&pTask->mm->mmap_sem);
1040	MY_DO_MUNMAP(pTask->mm, (unsigned long)pv, cb);
1041	up_write(&pTask->mm->mmap_sem);
1042	return VERR_NO_MEMORY;
1043	}
1044
1045	pMemLnx->Core.u.ResVirt.R0Process = R0Process;
1046	*ppMem = &pMemLnx->Core;
1047	return VINF_SUCCESS;
1048	}
1049
1050
1051	int rtR0MemObjNativeMapKernel(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap, void *pvFixed, size_t uAlignment,
1052	unsigned fProt, size_t offSub, size_t cbSub)
1053	{
1054	int rc = VERR_NO_MEMORY;
1055	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1056	PRTR0MEMOBJLNX pMemLnx;
1057
1058	/* Fail if requested to do something we can't. */
1059	AssertMsgReturn(!offSub && !cbSub, ("%#x %#x\n", offSub, cbSub), VERR_NOT_SUPPORTED);
1060	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
1061	if (uAlignment > PAGE_SIZE)
1062	return VERR_NOT_SUPPORTED;
1063
1064	/*
1065	* Create the IPRT memory object.
1066	*/
1067	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, pMemLnxToMap->Core.cb);
1068	if (pMemLnx)
1069	{
1070	if (pMemLnxToMap->cPages)
1071	{
1072	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1073	/*
1074	* Use vmap - 2.4.22 and later.
1075	*/
1076	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, true /* kernel */);
1077	# ifdef VM_MAP
1078	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[0], pMemLnxToMap->cPages, VM_MAP, fPg);
1079	# else
1080	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[0], pMemLnxToMap->cPages, VM_ALLOC, fPg);
1081	# endif
1082	if (pMemLnx->Core.pv)
1083	{
1084	pMemLnx->fMappedToRing0 = true;
1085	rc = VINF_SUCCESS;
1086	}
1087	else
1088	rc = VERR_MAP_FAILED;
1089
1090	#else /* < 2.4.22 */
1091	/*
1092	* Only option here is to share mappings if possible and forget about fProt.
1093	*/
1094	if (rtR0MemObjIsRing3(pMemToMap))
1095	rc = VERR_NOT_SUPPORTED;
1096	else
1097	{
1098	rc = VINF_SUCCESS;
1099	if (!pMemLnxToMap->Core.pv)
1100	rc = rtR0MemObjLinuxVMap(pMemLnxToMap, !!(fProt & RTMEM_PROT_EXEC));
1101	if (RT_SUCCESS(rc))
1102	{
1103	Assert(pMemLnxToMap->Core.pv);
1104	pMemLnx->Core.pv = pMemLnxToMap->Core.pv;
1105	}
1106	}
1107	#endif
1108	}
1109	else
1110	{
1111	/*
1112	* MMIO / physical memory.
1113	*/
1114	Assert(pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS && !pMemLnxToMap->Core.u.Phys.fAllocated);
1115	pMemLnx->Core.pv = ioremap(pMemLnxToMap->Core.u.Phys.PhysBase, pMemLnxToMap->Core.cb);
1116	if (pMemLnx->Core.pv)
1117	{
1118	/** @todo fix protection. */
1119	rc = VINF_SUCCESS;
1120	}
1121	}
1122	if (RT_SUCCESS(rc))
1123	{
1124	pMemLnx->Core.u.Mapping.R0Process = NIL_RTR0PROCESS;
1125	*ppMem = &pMemLnx->Core;
1126	return VINF_SUCCESS;
1127	}
1128	rtR0MemObjDelete(&pMemLnx->Core);
1129	}
1130
1131	return rc;
1132	}
1133
1134
1135	int rtR0MemObjNativeMapUser(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap, RTR3PTR R3PtrFixed, size_t uAlignment, unsigned fProt, RTR0PROCESS R0Process)
1136	{
1137	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1138	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1139	int rc = VERR_NO_MEMORY;
1140	PRTR0MEMOBJLNX pMemLnx;
1141
1142	/*
1143	* Check for restrictions.
1144	*/
1145	if (!pTask)
1146	return VERR_NOT_SUPPORTED;
1147	if (uAlignment > PAGE_SIZE)
1148	return VERR_NOT_SUPPORTED;
1149
1150	/*
1151	* Create the IPRT memory object.
1152	*/
1153	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, pMemLnxToMap->Core.cb);
1154	if (pMemLnx)
1155	{
1156	/*
1157	* Allocate user space mapping.
1158	*/
1159	void *pv;
1160	down_write(&pTask->mm->mmap_sem);
1161	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, pMemLnxToMap->Core.cb, uAlignment, pTask, fProt);
1162	if (pv != (void *)-1)
1163	{
1164	/*
1165	* Map page by page into the mmap area.
1166	* This is generic, paranoid and not very efficient.
1167	*/
1168	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, false /* user */);
1169	unsigned long ulAddrCur = (unsigned long)pv;
1170	const size_t cPages = pMemLnxToMap->Core.cb >> PAGE_SHIFT;
1171	size_t iPage;
1172	rc = 0;
1173	if (pMemLnxToMap->cPages)
1174	{
1175	for (iPage = 0; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE)
1176	{
1177	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11)
1178	uint64_t u64Phys = page_to_phys(pMemLnxToMap->apPages[iPage]);
1179	#endif
1180	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1181	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
1182	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
1183	#endif
1184	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11) && defined(RT_ARCH_X86)
1185	/* remap_page_range() limitation on x86 */
1186	AssertBreakStmt(u64Phys < _4G, rc = VERR_NO_MEMORY);
1187	#endif
1188
1189	#if defined(VBOX_USE_INSERT_PAGE) && LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1190	rc = vm_insert_page(vma, ulAddrCur, pMemLnxToMap->apPages[iPage]);
1191	vma->vm_flags \|= VM_RESERVED; /* This flag helps making 100% sure some bad stuff wont happen (swap, core, ++). */
1192	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
1193	rc = remap_pfn_range(vma, ulAddrCur, page_to_pfn(pMemLnxToMap->apPages[iPage]), PAGE_SIZE, fPg);
1194	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1195	rc = remap_page_range(vma, ulAddrCur, u64Phys, PAGE_SIZE, fPg);
1196	#else /* 2.4 */
1197	rc = remap_page_range(ulAddrCur, u64Phys, PAGE_SIZE, fPg);
1198	#endif
1199	if (rc)
1200	break;
1201	}
1202	}
1203	else
1204	{
1205	RTHCPHYS Phys;
1206	if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS)
1207	Phys = pMemLnxToMap->Core.u.Phys.PhysBase;
1208	else if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_CONT)
1209	Phys = pMemLnxToMap->Core.u.Cont.Phys;
1210	else
1211	{
1212	AssertMsgFailed(("%d\n", pMemLnxToMap->Core.enmType));
1213	Phys = NIL_RTHCPHYS;
1214	}
1215	if (Phys != NIL_RTHCPHYS)
1216	{
1217	for (iPage = 0; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE, Phys += PAGE_SIZE)
1218	{
1219	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1220	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
1221	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
1222	#endif
1223	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11) && defined(RT_ARCH_X86)
1224	/* remap_page_range() limitation on x86 */
1225	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
1226	#endif
1227
1228	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
1229	rc = remap_pfn_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1230	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1231	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1232	#else /* 2.4 */
1233	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
1234	#endif
1235	if (rc)
1236	break;
1237	}
1238	}
1239	}
1240	if (!rc)
1241	{
1242	up_write(&pTask->mm->mmap_sem);
1243
1244	pMemLnx->Core.pv = pv;
1245	pMemLnx->Core.u.Mapping.R0Process = R0Process;
1246	*ppMem = &pMemLnx->Core;
1247	return VINF_SUCCESS;
1248	}
1249
1250	/*
1251	* Bail out.
1252	*/
1253	MY_DO_MUNMAP(pTask->mm, (unsigned long)pv, pMemLnxToMap->Core.cb);
1254	if (rc != VERR_INTERNAL_ERROR)
1255	rc = VERR_NO_MEMORY;
1256	}
1257
1258	up_write(&pTask->mm->mmap_sem);
1259
1260	rtR0MemObjDelete(&pMemLnx->Core);
1261	}
1262
1263	return rc;
1264	}
1265
1266
1267	int rtR0MemObjNativeProtect(PRTR0MEMOBJINTERNAL pMem, size_t offSub, size_t cbSub, uint32_t fProt)
1268	{
1269	NOREF(pMem);
1270	NOREF(offSub);
1271	NOREF(cbSub);
1272	NOREF(fProt);
1273	return VERR_NOT_SUPPORTED;
1274	}
1275
1276
1277	RTHCPHYS rtR0MemObjNativeGetPagePhysAddr(PRTR0MEMOBJINTERNAL pMem, size_t iPage)
1278	{
1279	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
1280
1281	if (pMemLnx->cPages)
1282	return page_to_phys(pMemLnx->apPages[iPage]);
1283
1284	switch (pMemLnx->Core.enmType)
1285	{
1286	case RTR0MEMOBJTYPE_CONT:
1287	return pMemLnx->Core.u.Cont.Phys + (iPage << PAGE_SHIFT);
1288
1289	case RTR0MEMOBJTYPE_PHYS:
1290	return pMemLnx->Core.u.Phys.PhysBase + (iPage << PAGE_SHIFT);
1291
1292	/* the parent knows */
1293	case RTR0MEMOBJTYPE_MAPPING:
1294	return rtR0MemObjNativeGetPagePhysAddr(pMemLnx->Core.uRel.Child.pParent, iPage);
1295
1296	/* cPages > 0 */
1297	case RTR0MEMOBJTYPE_LOW:
1298	case RTR0MEMOBJTYPE_LOCK:
1299	case RTR0MEMOBJTYPE_PHYS_NC:
1300	case RTR0MEMOBJTYPE_PAGE:
1301	default:
1302	AssertMsgFailed(("%d\n", pMemLnx->Core.enmType));
1303	/* fall thru */
1304
1305	case RTR0MEMOBJTYPE_RES_VIRT:
1306	return NIL_RTHCPHYS;
1307	}
1308	}
1309

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

source: vbox/trunk/src/VBox/Runtime/r0drv/linux/memobj-r0drv-linux.c@ 22436

Download in other formats: