memobj-r0drv-linux.c@ 60372

Last change on this file since 60372 was 60372, checked in by vboxsync, 9 years ago
Runtime/r0drv/linux: bugref:8315: don't set the PG_reserved flag on potential compound pages. Play safe and do this change only on Linux 4.5 and later which would panic in that case (forward-port of r106453)
Property svn:eol-style set to `native` Property svn:keywords set to `Id Rev Revision`
File size: 55.4 KB

Line
1	/* $Id: memobj-r0drv-linux.c 60372 2016-04-07 14:12:17Z vboxsync $ */
2	/** @file
3	* IPRT - Ring-0 Memory Objects, Linux.
4	*/
5
6	/*
7	* Copyright (C) 2006-2015 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*
17	* The contents of this file may alternatively be used under the terms
18	* of the Common Development and Distribution License Version 1.0
19	* (CDDL) only, as it comes in the "COPYING.CDDL" file of the
20	* VirtualBox OSE distribution, in which case the provisions of the
21	* CDDL are applicable instead of those of the GPL.
22	*
23	* You may elect to license modified versions of this file under the
24	* terms and conditions of either the GPL or the CDDL or both.
25	*/
26
27
28	/*********************************************************************************************************************************
29	* Header Files *
30	*********************************************************************************************************************************/
31	#include "the-linux-kernel.h"
32
33	#include <iprt/memobj.h>
34	#include <iprt/alloc.h>
35	#include <iprt/assert.h>
36	#include <iprt/log.h>
37	#include <iprt/process.h>
38	#include <iprt/string.h>
39	#include "internal/memobj.h"
40
41
42	/*********************************************************************************************************************************
43	* Defined Constants And Macros *
44	*********************************************************************************************************************************/
45	/* early 2.6 kernels */
46	#ifndef PAGE_SHARED_EXEC
47	# define PAGE_SHARED_EXEC PAGE_SHARED
48	#endif
49	#ifndef PAGE_READONLY_EXEC
50	# define PAGE_READONLY_EXEC PAGE_READONLY
51	#endif
52
53	/*
54	* 2.6.29+ kernels don't work with remap_pfn_range() anymore because
55	* track_pfn_vma_new() is apparently not defined for non-RAM pages.
56	* It should be safe to use vm_insert_page() older kernels as well.
57	*/
58	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 23)
59	# define VBOX_USE_INSERT_PAGE
60	#endif
61	#if defined(CONFIG_X86_PAE) \
62	&& ( defined(HAVE_26_STYLE_REMAP_PAGE_RANGE) \
63	\|\| ( LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \
64	&& LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11)))
65	# define VBOX_USE_PAE_HACK
66	#endif
67
68
69	/*********************************************************************************************************************************
70	* Structures and Typedefs *
71	*********************************************************************************************************************************/
72	/**
73	* The Darwin version of the memory object structure.
74	*/
75	typedef struct RTR0MEMOBJLNX
76	{
77	/** The core structure. */
78	RTR0MEMOBJINTERNAL Core;
79	/** Set if the allocation is contiguous.
80	* This means it has to be given back as one chunk. */
81	bool fContiguous;
82	/** Set if we've vmap'ed the memory into ring-0. */
83	bool fMappedToRing0;
84	/** The pages in the apPages array. */
85	size_t cPages;
86	/** Array of struct page pointers. (variable size) */
87	struct page *apPages[1];
88	} RTR0MEMOBJLNX, *PRTR0MEMOBJLNX;
89
90
91	static void rtR0MemObjLinuxFreePages(PRTR0MEMOBJLNX pMemLnx);
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	static struct task_struct *rtR0ProcessToLinuxTask(RTR0PROCESS R0Process)
101	{
102	/** @todo fix rtR0ProcessToLinuxTask!! */
103	/** @todo many (all?) callers currently assume that we return 'current'! */
104	return R0Process == RTR0ProcHandleSelf() ? current : NULL;
105	}
106
107
108	/**
109	* Compute order. Some functions allocate 2^order pages.
110	*
111	* @returns order.
112	* @param cPages Number of pages.
113	*/
114	static int rtR0MemObjLinuxOrder(size_t cPages)
115	{
116	int iOrder;
117	size_t cTmp;
118
119	for (iOrder = 0, cTmp = cPages; cTmp >>= 1; ++iOrder)
120	;
121	if (cPages & ~((size_t)1 << iOrder))
122	++iOrder;
123
124	return iOrder;
125	}
126
127
128	/**
129	* Converts from RTMEM_PROT_* to Linux PAGE_*.
130	*
131	* @returns Linux page protection constant.
132	* @param fProt The IPRT protection mask.
133	* @param fKernel Whether it applies to kernel or user space.
134	*/
135	static pgprot_t rtR0MemObjLinuxConvertProt(unsigned fProt, bool fKernel)
136	{
137	switch (fProt)
138	{
139	default:
140	AssertMsgFailed(("%#x %d\n", fProt, fKernel));
141	case RTMEM_PROT_NONE:
142	return PAGE_NONE;
143
144	case RTMEM_PROT_READ:
145	return fKernel ? PAGE_KERNEL_RO : PAGE_READONLY;
146
147	case RTMEM_PROT_WRITE:
148	case RTMEM_PROT_WRITE \| RTMEM_PROT_READ:
149	return fKernel ? PAGE_KERNEL : PAGE_SHARED;
150
151	case RTMEM_PROT_EXEC:
152	case RTMEM_PROT_EXEC \| RTMEM_PROT_READ:
153	#if defined(RT_ARCH_X86) \|\| defined(RT_ARCH_AMD64)
154	if (fKernel)
155	{
156	pgprot_t fPg = MY_PAGE_KERNEL_EXEC;
157	pgprot_val(fPg) &= ~_PAGE_RW;
158	return fPg;
159	}
160	return PAGE_READONLY_EXEC;
161	#else
162	return fKernel ? MY_PAGE_KERNEL_EXEC : PAGE_READONLY_EXEC;
163	#endif
164
165	case RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC:
166	case RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC \| RTMEM_PROT_READ:
167	return fKernel ? MY_PAGE_KERNEL_EXEC : PAGE_SHARED_EXEC;
168	}
169	}
170
171
172	/**
173	* Worker for rtR0MemObjNativeReserveUser and rtR0MemObjNativerMapUser that creates
174	* an empty user space mapping.
175	*
176	* We acquire the mmap_sem of the task!
177	*
178	* @returns Pointer to the mapping.
179	* (void *)-1 on failure.
180	* @param R3PtrFixed (RTR3PTR)-1 if anywhere, otherwise a specific location.
181	* @param cb The size of the mapping.
182	* @param uAlignment The alignment of the mapping.
183	* @param pTask The Linux task to create this mapping in.
184	* @param fProt The RTMEM_PROT_* mask.
185	*/
186	static void rtR0MemObjLinuxDoMmap(RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment, struct task_struct pTask, unsigned fProt)
187	{
188	unsigned fLnxProt;
189	unsigned long ulAddr;
190
191	Assert((pTask == current)); /* do_mmap */
192
193	/*
194	* Convert from IPRT protection to mman.h PROT_ and call do_mmap.
195	*/
196	fProt &= (RTMEM_PROT_NONE \| RTMEM_PROT_READ \| RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC);
197	if (fProt == RTMEM_PROT_NONE)
198	fLnxProt = PROT_NONE;
199	else
200	{
201	fLnxProt = 0;
202	if (fProt & RTMEM_PROT_READ)
203	fLnxProt \|= PROT_READ;
204	if (fProt & RTMEM_PROT_WRITE)
205	fLnxProt \|= PROT_WRITE;
206	if (fProt & RTMEM_PROT_EXEC)
207	fLnxProt \|= PROT_EXEC;
208	}
209
210	if (R3PtrFixed != (RTR3PTR)-1)
211	{
212	#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 5, 0)
213	ulAddr = vm_mmap(NULL, R3PtrFixed, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS \| MAP_FIXED, 0);
214	#else
215	down_write(&pTask->mm->mmap_sem);
216	ulAddr = do_mmap(NULL, R3PtrFixed, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS \| MAP_FIXED, 0);
217	up_write(&pTask->mm->mmap_sem);
218	#endif
219	}
220	else
221	{
222	#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 5, 0)
223	ulAddr = vm_mmap(NULL, 0, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS, 0);
224	#else
225	down_write(&pTask->mm->mmap_sem);
226	ulAddr = do_mmap(NULL, 0, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS, 0);
227	up_write(&pTask->mm->mmap_sem);
228	#endif
229	if ( !(ulAddr & ~PAGE_MASK)
230	&& (ulAddr & (uAlignment - 1)))
231	{
232	/** @todo implement uAlignment properly... We'll probably need to make some dummy mappings to fill
233	* up alignment gaps. This is of course complicated by fragmentation (which we might have cause
234	* ourselves) and further by there begin two mmap strategies (top / bottom). */
235	/* For now, just ignore uAlignment requirements... */
236	}
237	}
238
239
240	if (ulAddr & ~PAGE_MASK) /* ~PAGE_MASK == PAGE_OFFSET_MASK */
241	return (void *)-1;
242	return (void *)ulAddr;
243	}
244
245
246	/**
247	* Worker that destroys a user space mapping.
248	* Undoes what rtR0MemObjLinuxDoMmap did.
249	*
250	* We acquire the mmap_sem of the task!
251	*
252	* @param pv The ring-3 mapping.
253	* @param cb The size of the mapping.
254	* @param pTask The Linux task to destroy this mapping in.
255	*/
256	static void rtR0MemObjLinuxDoMunmap(void pv, size_t cb, struct task_struct pTask)
257	{
258	#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 5, 0)
259	Assert(pTask == current);
260	vm_munmap((unsigned long)pv, cb);
261	#elif defined(USE_RHEL4_MUNMAP)
262	down_write(&pTask->mm->mmap_sem);
263	do_munmap(pTask->mm, (unsigned long)pv, cb, 0); /* should it be 1 or 0? */
264	up_write(&pTask->mm->mmap_sem);
265	#else
266	down_write(&pTask->mm->mmap_sem);
267	do_munmap(pTask->mm, (unsigned long)pv, cb);
268	up_write(&pTask->mm->mmap_sem);
269	#endif
270	}
271
272
273	/**
274	* Internal worker that allocates physical pages and creates the memory object for them.
275	*
276	* @returns IPRT status code.
277	* @param ppMemLnx Where to store the memory object pointer.
278	* @param enmType The object type.
279	* @param cb The number of bytes to allocate.
280	* @param uAlignment The alignment of the physical memory.
281	* Only valid if fContiguous == true, ignored otherwise.
282	* @param fFlagsLnx The page allocation flags (GPFs).
283	* @param fContiguous Whether the allocation must be contiguous.
284	* @param rcNoMem What to return when we're out of pages.
285	*/
286	static int rtR0MemObjLinuxAllocPages(PRTR0MEMOBJLNX *ppMemLnx, RTR0MEMOBJTYPE enmType, size_t cb,
287	size_t uAlignment, unsigned fFlagsLnx, bool fContiguous, int rcNoMem)
288	{
289	size_t iPage;
290	size_t const cPages = cb >> PAGE_SHIFT;
291	struct page *paPages;
292
293	/*
294	* Allocate a memory object structure that's large enough to contain
295	* the page pointer array.
296	*/
297	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_OFFSETOF(RTR0MEMOBJLNX, apPages[cPages]), enmType, NULL, cb);
298	if (!pMemLnx)
299	return VERR_NO_MEMORY;
300	pMemLnx->cPages = cPages;
301
302	if (cPages > 255)
303	{
304	# ifdef __GFP_REPEAT
305	/* Try hard to allocate the memory, but the allocation attempt might fail. */
306	fFlagsLnx \|= __GFP_REPEAT;
307	# endif
308	# ifdef __GFP_NOMEMALLOC
309	/* Introduced with Linux 2.6.12: Don't use emergency reserves */
310	fFlagsLnx \|= __GFP_NOMEMALLOC;
311	# endif
312	}
313
314	/*
315	* Allocate the pages.
316	* For small allocations we'll try contiguous first and then fall back on page by page.
317	*/
318	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
319	if ( fContiguous
320	\|\| cb <= PAGE_SIZE * 2)
321	{
322	# ifdef VBOX_USE_INSERT_PAGE
323	paPages = alloc_pages(fFlagsLnx \| __GFP_COMP \| __GFP_NOWARN, rtR0MemObjLinuxOrder(cPages));
324	# else
325	paPages = alloc_pages(fFlagsLnx \| __GFP_NOWARN, rtR0MemObjLinuxOrder(cPages));
326	# endif
327	if (paPages)
328	{
329	fContiguous = true;
330	for (iPage = 0; iPage < cPages; iPage++)
331	pMemLnx->apPages[iPage] = &paPages[iPage];
332	}
333	else if (fContiguous)
334	{
335	rtR0MemObjDelete(&pMemLnx->Core);
336	return rcNoMem;
337	}
338	}
339
340	if (!fContiguous)
341	{
342	for (iPage = 0; iPage < cPages; iPage++)
343	{
344	pMemLnx->apPages[iPage] = alloc_page(fFlagsLnx \| __GFP_NOWARN);
345	if (RT_UNLIKELY(!pMemLnx->apPages[iPage]))
346	{
347	while (iPage-- > 0)
348	__free_page(pMemLnx->apPages[iPage]);
349	rtR0MemObjDelete(&pMemLnx->Core);
350	return rcNoMem;
351	}
352	}
353	}
354
355	#else /* < 2.4.22 */
356	/** @todo figure out why we didn't allocate page-by-page on 2.4.21 and older... */
357	paPages = alloc_pages(fFlagsLnx, rtR0MemObjLinuxOrder(cPages));
358	if (!paPages)
359	{
360	rtR0MemObjDelete(&pMemLnx->Core);
361	return rcNoMem;
362	}
363	for (iPage = 0; iPage < cPages; iPage++)
364	{
365	pMemLnx->apPages[iPage] = &paPages[iPage];
366	MY_SET_PAGES_EXEC(pMemLnx->apPages[iPage], 1);
367	if (PageHighMem(pMemLnx->apPages[iPage]))
368	BUG();
369	}
370
371	fContiguous = true;
372	#endif /* < 2.4.22 */
373	pMemLnx->fContiguous = fContiguous;
374
375	#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 5, 0)
376	/*
377	* Reserve the pages.
378	*
379	* Linux >= 4.5 with CONFIG_DEBUG_VM panics when setting PG_reserved on compound
380	* pages. According to Michal Hocko this shouldn't be necessary anyway because
381	* as pages which are not on the LRU list are never evictable.
382	*/
383	for (iPage = 0; iPage < cPages; iPage++)
384	SetPageReserved(pMemLnx->apPages[iPage]);
385	#endif
386
387	/*
388	* Note that the physical address of memory allocated with alloc_pages(flags, order)
389	* is always 2^(PAGE_SHIFT+order)-aligned.
390	*/
391	if ( fContiguous
392	&& uAlignment > PAGE_SIZE)
393	{
394	/*
395	* Check for alignment constraints. The physical address of memory allocated with
396	* alloc_pages(flags, order) is always 2^(PAGE_SHIFT+order)-aligned.
397	*/
398	if (RT_UNLIKELY(page_to_phys(pMemLnx->apPages[0]) & (uAlignment - 1)))
399	{
400	/*
401	* This should never happen!
402	*/
403	printk("rtR0MemObjLinuxAllocPages(cb=0x%lx, uAlignment=0x%lx): alloc_pages(..., %d) returned physical memory at 0x%lx!\n",
404	(unsigned long)cb, (unsigned long)uAlignment, rtR0MemObjLinuxOrder(cPages), (unsigned long)page_to_phys(pMemLnx->apPages[0]));
405	rtR0MemObjLinuxFreePages(pMemLnx);
406	return rcNoMem;
407	}
408	}
409
410	*ppMemLnx = pMemLnx;
411	return VINF_SUCCESS;
412	}
413
414
415	/**
416	* Frees the physical pages allocated by the rtR0MemObjLinuxAllocPages() call.
417	*
418	* This method does NOT free the object.
419	*
420	* @param pMemLnx The object which physical pages should be freed.
421	*/
422	static void rtR0MemObjLinuxFreePages(PRTR0MEMOBJLNX pMemLnx)
423	{
424	size_t iPage = pMemLnx->cPages;
425	if (iPage > 0)
426	{
427	/*
428	* Restore the page flags.
429	*/
430	while (iPage-- > 0)
431	{
432	#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 5, 0)
433	/*
434	* See SetPageReserved() in rtR0MemObjLinuxAllocPages()
435	*/
436	ClearPageReserved(pMemLnx->apPages[iPage]);
437	#endif
438	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
439	#else
440	MY_SET_PAGES_NOEXEC(pMemLnx->apPages[iPage], 1);
441	#endif
442	}
443
444	/*
445	* Free the pages.
446	*/
447	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
448	if (!pMemLnx->fContiguous)
449	{
450	iPage = pMemLnx->cPages;
451	while (iPage-- > 0)
452	__free_page(pMemLnx->apPages[iPage]);
453	}
454	else
455	#endif
456	__free_pages(pMemLnx->apPages[0], rtR0MemObjLinuxOrder(pMemLnx->cPages));
457
458	pMemLnx->cPages = 0;
459	}
460	}
461
462
463	/**
464	* Maps the allocation into ring-0.
465	*
466	* This will update the RTR0MEMOBJLNX::Core.pv and RTR0MEMOBJ::fMappedToRing0 members.
467	*
468	* Contiguous mappings that isn't in 'high' memory will already be mapped into kernel
469	* space, so we'll use that mapping if possible. If execute access is required, we'll
470	* play safe and do our own mapping.
471	*
472	* @returns IPRT status code.
473	* @param pMemLnx The linux memory object to map.
474	* @param fExecutable Whether execute access is required.
475	*/
476	static int rtR0MemObjLinuxVMap(PRTR0MEMOBJLNX pMemLnx, bool fExecutable)
477	{
478	int rc = VINF_SUCCESS;
479
480	/*
481	* Choose mapping strategy.
482	*/
483	bool fMustMap = fExecutable
484	\|\| !pMemLnx->fContiguous;
485	if (!fMustMap)
486	{
487	size_t iPage = pMemLnx->cPages;
488	while (iPage-- > 0)
489	if (PageHighMem(pMemLnx->apPages[iPage]))
490	{
491	fMustMap = true;
492	break;
493	}
494	}
495
496	Assert(!pMemLnx->Core.pv);
497	Assert(!pMemLnx->fMappedToRing0);
498
499	if (fMustMap)
500	{
501	/*
502	* Use vmap - 2.4.22 and later.
503	*/
504	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
505	pgprot_t fPg;
506	pgprot_val(fPg) = _PAGE_PRESENT \| _PAGE_RW;
507	# ifdef _PAGE_NX
508	if (!fExecutable)
509	pgprot_val(fPg) \|= _PAGE_NX;
510	# endif
511
512	# ifdef VM_MAP
513	pMemLnx->Core.pv = vmap(&pMemLnx->apPages[0], pMemLnx->cPages, VM_MAP, fPg);
514	# else
515	pMemLnx->Core.pv = vmap(&pMemLnx->apPages[0], pMemLnx->cPages, VM_ALLOC, fPg);
516	# endif
517	if (pMemLnx->Core.pv)
518	pMemLnx->fMappedToRing0 = true;
519	else
520	rc = VERR_MAP_FAILED;
521	#else /* < 2.4.22 */
522	rc = VERR_NOT_SUPPORTED;
523	#endif
524	}
525	else
526	{
527	/*
528	* Use the kernel RAM mapping.
529	*/
530	pMemLnx->Core.pv = phys_to_virt(page_to_phys(pMemLnx->apPages[0]));
531	Assert(pMemLnx->Core.pv);
532	}
533
534	return rc;
535	}
536
537
538	/**
539	* Undoes what rtR0MemObjLinuxVMap() did.
540	*
541	* @param pMemLnx The linux memory object.
542	*/
543	static void rtR0MemObjLinuxVUnmap(PRTR0MEMOBJLNX pMemLnx)
544	{
545	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
546	if (pMemLnx->fMappedToRing0)
547	{
548	Assert(pMemLnx->Core.pv);
549	vunmap(pMemLnx->Core.pv);
550	pMemLnx->fMappedToRing0 = false;
551	}
552	#else /* < 2.4.22 */
553	Assert(!pMemLnx->fMappedToRing0);
554	#endif
555	pMemLnx->Core.pv = NULL;
556	}
557
558
559	DECLHIDDEN(int) rtR0MemObjNativeFree(RTR0MEMOBJ pMem)
560	{
561	IPRT_LINUX_SAVE_EFL_AC();
562	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
563
564	/*
565	* Release any memory that we've allocated or locked.
566	*/
567	switch (pMemLnx->Core.enmType)
568	{
569	case RTR0MEMOBJTYPE_LOW:
570	case RTR0MEMOBJTYPE_PAGE:
571	case RTR0MEMOBJTYPE_CONT:
572	case RTR0MEMOBJTYPE_PHYS:
573	case RTR0MEMOBJTYPE_PHYS_NC:
574	rtR0MemObjLinuxVUnmap(pMemLnx);
575	rtR0MemObjLinuxFreePages(pMemLnx);
576	break;
577
578	case RTR0MEMOBJTYPE_LOCK:
579	if (pMemLnx->Core.u.Lock.R0Process != NIL_RTR0PROCESS)
580	{
581	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
582	size_t iPage;
583	Assert(pTask);
584	if (pTask && pTask->mm)
585	down_read(&pTask->mm->mmap_sem);
586
587	iPage = pMemLnx->cPages;
588	while (iPage-- > 0)
589	{
590	if (!PageReserved(pMemLnx->apPages[iPage]))
591	SetPageDirty(pMemLnx->apPages[iPage]);
592	page_cache_release(pMemLnx->apPages[iPage]);
593	}
594
595	if (pTask && pTask->mm)
596	up_read(&pTask->mm->mmap_sem);
597	}
598	/* else: kernel memory - nothing to do here. */
599	break;
600
601	case RTR0MEMOBJTYPE_RES_VIRT:
602	Assert(pMemLnx->Core.pv);
603	if (pMemLnx->Core.u.ResVirt.R0Process != NIL_RTR0PROCESS)
604	{
605	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
606	Assert(pTask);
607	if (pTask && pTask->mm)
608	rtR0MemObjLinuxDoMunmap(pMemLnx->Core.pv, pMemLnx->Core.cb, pTask);
609	}
610	else
611	{
612	vunmap(pMemLnx->Core.pv);
613
614	Assert(pMemLnx->cPages == 1 && pMemLnx->apPages[0] != NULL);
615	__free_page(pMemLnx->apPages[0]);
616	pMemLnx->apPages[0] = NULL;
617	pMemLnx->cPages = 0;
618	}
619	pMemLnx->Core.pv = NULL;
620	break;
621
622	case RTR0MEMOBJTYPE_MAPPING:
623	Assert(pMemLnx->cPages == 0); Assert(pMemLnx->Core.pv);
624	if (pMemLnx->Core.u.ResVirt.R0Process != NIL_RTR0PROCESS)
625	{
626	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
627	Assert(pTask);
628	if (pTask && pTask->mm)
629	rtR0MemObjLinuxDoMunmap(pMemLnx->Core.pv, pMemLnx->Core.cb, pTask);
630	}
631	else
632	vunmap(pMemLnx->Core.pv);
633	pMemLnx->Core.pv = NULL;
634	break;
635
636	default:
637	AssertMsgFailed(("enmType=%d\n", pMemLnx->Core.enmType));
638	return VERR_INTERNAL_ERROR;
639	}
640	IPRT_LINUX_RESTORE_EFL_ONLY_AC();
641	return VINF_SUCCESS;
642	}
643
644
645	DECLHIDDEN(int) rtR0MemObjNativeAllocPage(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable)
646	{
647	IPRT_LINUX_SAVE_EFL_AC();
648	PRTR0MEMOBJLNX pMemLnx;
649	int rc;
650
651	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
652	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_PAGE, cb, PAGE_SIZE, GFP_HIGHUSER,
653	false /* non-contiguous */, VERR_NO_MEMORY);
654	#else
655	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_PAGE, cb, PAGE_SIZE, GFP_USER,
656	false /* non-contiguous */, VERR_NO_MEMORY);
657	#endif
658	if (RT_SUCCESS(rc))
659	{
660	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
661	if (RT_SUCCESS(rc))
662	{
663	*ppMem = &pMemLnx->Core;
664	IPRT_LINUX_RESTORE_EFL_AC();
665	return rc;
666	}
667
668	rtR0MemObjLinuxFreePages(pMemLnx);
669	rtR0MemObjDelete(&pMemLnx->Core);
670	}
671
672	IPRT_LINUX_RESTORE_EFL_AC();
673	return rc;
674	}
675
676
677	DECLHIDDEN(int) rtR0MemObjNativeAllocLow(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable)
678	{
679	IPRT_LINUX_SAVE_EFL_AC();
680	PRTR0MEMOBJLNX pMemLnx;
681	int rc;
682
683	/* Try to avoid GFP_DMA. GFM_DMA32 was introduced with Linux 2.6.15. */
684	#if (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
685	/* ZONE_DMA32: 0-4GB */
686	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_DMA32,
687	false /* non-contiguous */, VERR_NO_LOW_MEMORY);
688	if (RT_FAILURE(rc))
689	#endif
690	#ifdef RT_ARCH_AMD64
691	/* ZONE_DMA: 0-16MB */
692	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_DMA,
693	false /* non-contiguous */, VERR_NO_LOW_MEMORY);
694	#else
695	# ifdef CONFIG_X86_PAE
696	# endif
697	/* ZONE_NORMAL: 0-896MB */
698	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_USER,
699	false /* non-contiguous */, VERR_NO_LOW_MEMORY);
700	#endif
701	if (RT_SUCCESS(rc))
702	{
703	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
704	if (RT_SUCCESS(rc))
705	{
706	*ppMem = &pMemLnx->Core;
707	IPRT_LINUX_RESTORE_EFL_AC();
708	return rc;
709	}
710
711	rtR0MemObjLinuxFreePages(pMemLnx);
712	rtR0MemObjDelete(&pMemLnx->Core);
713	}
714
715	IPRT_LINUX_RESTORE_EFL_AC();
716	return rc;
717	}
718
719
720	DECLHIDDEN(int) rtR0MemObjNativeAllocCont(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable)
721	{
722	IPRT_LINUX_SAVE_EFL_AC();
723	PRTR0MEMOBJLNX pMemLnx;
724	int rc;
725
726	#if (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
727	/* ZONE_DMA32: 0-4GB */
728	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, PAGE_SIZE, GFP_DMA32,
729	true /* contiguous */, VERR_NO_CONT_MEMORY);
730	if (RT_FAILURE(rc))
731	#endif
732	#ifdef RT_ARCH_AMD64
733	/* ZONE_DMA: 0-16MB */
734	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, PAGE_SIZE, GFP_DMA,
735	true /* contiguous */, VERR_NO_CONT_MEMORY);
736	#else
737	/* ZONE_NORMAL (32-bit hosts): 0-896MB */
738	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, PAGE_SIZE, GFP_USER,
739	true /* contiguous */, VERR_NO_CONT_MEMORY);
740	#endif
741	if (RT_SUCCESS(rc))
742	{
743	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
744	if (RT_SUCCESS(rc))
745	{
746	#if defined(RT_STRICT) && (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_HIGHMEM64G))
747	size_t iPage = pMemLnx->cPages;
748	while (iPage-- > 0)
749	Assert(page_to_phys(pMemLnx->apPages[iPage]) < _4G);
750	#endif
751	pMemLnx->Core.u.Cont.Phys = page_to_phys(pMemLnx->apPages[0]);
752	*ppMem = &pMemLnx->Core;
753	IPRT_LINUX_RESTORE_EFL_AC();
754	return rc;
755	}
756
757	rtR0MemObjLinuxFreePages(pMemLnx);
758	rtR0MemObjDelete(&pMemLnx->Core);
759	}
760
761	IPRT_LINUX_RESTORE_EFL_AC();
762	return rc;
763	}
764
765
766	/**
767	* Worker for rtR0MemObjLinuxAllocPhysSub that tries one allocation strategy.
768	*
769	* @returns IPRT status code.
770	* @param ppMemLnx Where to
771	* @param enmType The object type.
772	* @param cb The size of the allocation.
773	* @param uAlignment The alignment of the physical memory.
774	* Only valid for fContiguous == true, ignored otherwise.
775	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
776	* @param fGfp The Linux GFP flags to use for the allocation.
777	*/
778	static int rtR0MemObjLinuxAllocPhysSub2(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType,
779	size_t cb, size_t uAlignment, RTHCPHYS PhysHighest, unsigned fGfp)
780	{
781	PRTR0MEMOBJLNX pMemLnx;
782	int rc;
783
784	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, enmType, cb, uAlignment, fGfp,
785	enmType == RTR0MEMOBJTYPE_PHYS /* contiguous / non-contiguous */,
786	VERR_NO_PHYS_MEMORY);
787	if (RT_FAILURE(rc))
788	return rc;
789
790	/*
791	* Check the addresses if necessary. (Can be optimized a bit for PHYS.)
792	*/
793	if (PhysHighest != NIL_RTHCPHYS)
794	{
795	size_t iPage = pMemLnx->cPages;
796	while (iPage-- > 0)
797	if (page_to_phys(pMemLnx->apPages[iPage]) > PhysHighest)
798	{
799	rtR0MemObjLinuxFreePages(pMemLnx);
800	rtR0MemObjDelete(&pMemLnx->Core);
801	return VERR_NO_MEMORY;
802	}
803	}
804
805	/*
806	* Complete the object.
807	*/
808	if (enmType == RTR0MEMOBJTYPE_PHYS)
809	{
810	pMemLnx->Core.u.Phys.PhysBase = page_to_phys(pMemLnx->apPages[0]);
811	pMemLnx->Core.u.Phys.fAllocated = true;
812	}
813	*ppMem = &pMemLnx->Core;
814	return rc;
815	}
816
817
818	/**
819	* Worker for rtR0MemObjNativeAllocPhys and rtR0MemObjNativeAllocPhysNC.
820	*
821	* @returns IPRT status code.
822	* @param ppMem Where to store the memory object pointer on success.
823	* @param enmType The object type.
824	* @param cb The size of the allocation.
825	* @param uAlignment The alignment of the physical memory.
826	* Only valid for enmType == RTR0MEMOBJTYPE_PHYS, ignored otherwise.
827	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
828	*/
829	static int rtR0MemObjLinuxAllocPhysSub(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType,
830	size_t cb, size_t uAlignment, RTHCPHYS PhysHighest)
831	{
832	int rc;
833	IPRT_LINUX_SAVE_EFL_AC();
834
835	/*
836	* There are two clear cases and that's the <=16MB and anything-goes ones.
837	* When the physical address limit is somewhere in-between those two we'll
838	* just have to try, starting with HIGHUSER and working our way thru the
839	* different types, hoping we'll get lucky.
840	*
841	* We should probably move this physical address restriction logic up to
842	* the page alloc function as it would be more efficient there. But since
843	* we don't expect this to be a performance issue just yet it can wait.
844	*/
845	if (PhysHighest == NIL_RTHCPHYS)
846	/* ZONE_HIGHMEM: the whole physical memory */
847	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_HIGHUSER);
848	else if (PhysHighest <= _1M * 16)
849	/* ZONE_DMA: 0-16MB */
850	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_DMA);
851	else
852	{
853	rc = VERR_NO_MEMORY;
854	if (RT_FAILURE(rc))
855	/* ZONE_HIGHMEM: the whole physical memory */
856	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_HIGHUSER);
857	if (RT_FAILURE(rc))
858	/* ZONE_NORMAL: 0-896MB */
859	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_USER);
860	#ifdef GFP_DMA32
861	if (RT_FAILURE(rc))
862	/* ZONE_DMA32: 0-4GB */
863	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_DMA32);
864	#endif
865	if (RT_FAILURE(rc))
866	/* ZONE_DMA: 0-16MB */
867	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_DMA);
868	}
869	IPRT_LINUX_RESTORE_EFL_AC();
870	return rc;
871	}
872
873
874	/**
875	* Translates a kernel virtual address to a linux page structure by walking the
876	* page tables.
877	*
878	* @note We do assume that the page tables will not change as we are walking
879	* them. This assumption is rather forced by the fact that I could not
880	* immediately see any way of preventing this from happening. So, we
881	* take some extra care when accessing them.
882	*
883	* Because of this, we don't want to use this function on memory where
884	* attribute changes to nearby pages is likely to cause large pages to
885	* be used or split up. So, don't use this for the linear mapping of
886	* physical memory.
887	*
888	* @returns Pointer to the page structur or NULL if it could not be found.
889	* @param pv The kernel virtual address.
890	*/
891	static struct page rtR0MemObjLinuxVirtToPage(void pv)
892	{
893	unsigned long ulAddr = (unsigned long)pv;
894	unsigned long pfn;
895	struct page *pPage;
896	pte_t *pEntry;
897	union
898	{
899	pgd_t Global;
900	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
901	pud_t Upper;
902	#endif
903	pmd_t Middle;
904	pte_t Entry;
905	} u;
906
907	/* Should this happen in a situation this code will be called in? And if
908	* so, can it change under our feet? See also
909	* "Documentation/vm/active_mm.txt" in the kernel sources. */
910	if (RT_UNLIKELY(!current->active_mm))
911	return NULL;
912	u.Global = *pgd_offset(current->active_mm, ulAddr);
913	if (RT_UNLIKELY(pgd_none(u.Global)))
914	return NULL;
915
916	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
917	u.Upper = *pud_offset(&u.Global, ulAddr);
918	if (RT_UNLIKELY(pud_none(u.Upper)))
919	return NULL;
920	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25)
921	if (pud_large(u.Upper))
922	{
923	pPage = pud_page(u.Upper);
924	AssertReturn(pPage, NULL);
925	pfn = page_to_pfn(pPage); /* doing the safe way... */
926	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (PUD_SHIFT - PAGE_SHIFT)) - 1);
927	return pfn_to_page(pfn);
928	}
929	# endif
930
931	u.Middle = *pmd_offset(&u.Upper, ulAddr);
932	#else /* < 2.6.11 */
933	u.Middle = *pmd_offset(&u.Global, ulAddr);
934	#endif /* < 2.6.11 */
935	if (RT_UNLIKELY(pmd_none(u.Middle)))
936	return NULL;
937	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
938	if (pmd_large(u.Middle))
939	{
940	pPage = pmd_page(u.Middle);
941	AssertReturn(pPage, NULL);
942	pfn = page_to_pfn(pPage); /* doing the safe way... */
943	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (PMD_SHIFT - PAGE_SHIFT)) - 1);
944	return pfn_to_page(pfn);
945	}
946	#endif
947
948	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 5) \|\| defined(pte_offset_map) /* As usual, RHEL 3 had pte_offset_map earlier. */
949	pEntry = pte_offset_map(&u.Middle, ulAddr);
950	#else
951	pEntry = pte_offset(&u.Middle, ulAddr);
952	#endif
953	if (RT_UNLIKELY(!pEntry))
954	return NULL;
955	u.Entry = *pEntry;
956	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 5) \|\| defined(pte_offset_map)
957	pte_unmap(pEntry);
958	#endif
959
960	if (RT_UNLIKELY(!pte_present(u.Entry)))
961	return NULL;
962	return pte_page(u.Entry);
963	}
964
965
966	DECLHIDDEN(int) rtR0MemObjNativeAllocPhys(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest, size_t uAlignment)
967	{
968	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS, cb, uAlignment, PhysHighest);
969	}
970
971
972	DECLHIDDEN(int) rtR0MemObjNativeAllocPhysNC(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest)
973	{
974	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS_NC, cb, PAGE_SIZE, PhysHighest);
975	}
976
977
978	DECLHIDDEN(int) rtR0MemObjNativeEnterPhys(PPRTR0MEMOBJINTERNAL ppMem, RTHCPHYS Phys, size_t cb, uint32_t uCachePolicy)
979	{
980	IPRT_LINUX_SAVE_EFL_AC();
981
982	/*
983	* All we need to do here is to validate that we can use
984	* ioremap on the specified address (32/64-bit dma_addr_t).
985	*/
986	PRTR0MEMOBJLNX pMemLnx;
987	dma_addr_t PhysAddr = Phys;
988	AssertMsgReturn(PhysAddr == Phys, ("%#llx\n", (unsigned long long)Phys), VERR_ADDRESS_TOO_BIG);
989
990	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_PHYS, NULL, cb);
991	if (!pMemLnx)
992	{
993	IPRT_LINUX_RESTORE_EFL_AC();
994	return VERR_NO_MEMORY;
995	}
996
997	pMemLnx->Core.u.Phys.PhysBase = PhysAddr;
998	pMemLnx->Core.u.Phys.fAllocated = false;
999	pMemLnx->Core.u.Phys.uCachePolicy = uCachePolicy;
1000	Assert(!pMemLnx->cPages);
1001	*ppMem = &pMemLnx->Core;
1002	IPRT_LINUX_RESTORE_EFL_AC();
1003	return VINF_SUCCESS;
1004	}
1005
1006
1007	DECLHIDDEN(int) rtR0MemObjNativeLockUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3Ptr, size_t cb, uint32_t fAccess, RTR0PROCESS R0Process)
1008	{
1009	IPRT_LINUX_SAVE_EFL_AC();
1010	const int cPages = cb >> PAGE_SHIFT;
1011	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1012	struct vm_area_struct **papVMAs;
1013	PRTR0MEMOBJLNX pMemLnx;
1014	int rc = VERR_NO_MEMORY;
1015	int const fWrite = fAccess & RTMEM_PROT_WRITE ? 1 : 0;
1016
1017	/*
1018	* Check for valid task and size overflows.
1019	*/
1020	if (!pTask)
1021	return VERR_NOT_SUPPORTED;
1022	if (((size_t)cPages << PAGE_SHIFT) != cb)
1023	return VERR_OUT_OF_RANGE;
1024
1025	/*
1026	* Allocate the memory object and a temporary buffer for the VMAs.
1027	*/
1028	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_OFFSETOF(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK, (void *)R3Ptr, cb);
1029	if (!pMemLnx)
1030	{
1031	IPRT_LINUX_RESTORE_EFL_AC();
1032	return VERR_NO_MEMORY;
1033	}
1034
1035	papVMAs = (struct vm_area_struct *)RTMemAlloc(sizeof(papVMAs) * cPages);
1036	if (papVMAs)
1037	{
1038	down_read(&pTask->mm->mmap_sem);
1039
1040	/*
1041	* Get user pages.
1042	*/
1043	rc = get_user_pages(pTask, /* Task for fault accounting. */
1044	pTask->mm, /* Whose pages. */
1045	R3Ptr, /* Where from. */
1046	cPages, /* How many pages. */
1047	fWrite, /* Write to memory. */
1048	fWrite, /* force write access. */
1049	&pMemLnx->apPages[0], /* Page array. */
1050	papVMAs); /* vmas */
1051	if (rc == cPages)
1052	{
1053	/*
1054	* Flush dcache (required?), protect against fork and _really_ pin the page
1055	* table entries. get_user_pages() will protect against swapping out the
1056	* pages but it will NOT protect against removing page table entries. This
1057	* can be achieved with
1058	* - using mlock / mmap(..., MAP_LOCKED, ...) from userland. This requires
1059	* an appropriate limit set up with setrlimit(..., RLIMIT_MEMLOCK, ...).
1060	* Usual Linux distributions support only a limited size of locked pages
1061	* (e.g. 32KB).
1062	* - setting the PageReserved bit (as we do in rtR0MemObjLinuxAllocPages()
1063	* or by
1064	* - setting the VM_LOCKED flag. This is the same as doing mlock() without
1065	* a range check.
1066	*/
1067	/** @todo The Linux fork() protection will require more work if this API
1068	* is to be used for anything but locking VM pages. */
1069	while (rc-- > 0)
1070	{
1071	flush_dcache_page(pMemLnx->apPages[rc]);
1072	papVMAs[rc]->vm_flags \|= (VM_DONTCOPY \| VM_LOCKED);
1073	}
1074
1075	up_read(&pTask->mm->mmap_sem);
1076
1077	RTMemFree(papVMAs);
1078
1079	pMemLnx->Core.u.Lock.R0Process = R0Process;
1080	pMemLnx->cPages = cPages;
1081	Assert(!pMemLnx->fMappedToRing0);
1082	*ppMem = &pMemLnx->Core;
1083
1084	IPRT_LINUX_RESTORE_EFL_AC();
1085	return VINF_SUCCESS;
1086	}
1087
1088	/*
1089	* Failed - we need to unlock any pages that we succeeded to lock.
1090	*/
1091	while (rc-- > 0)
1092	{
1093	if (!PageReserved(pMemLnx->apPages[rc]))
1094	SetPageDirty(pMemLnx->apPages[rc]);
1095	page_cache_release(pMemLnx->apPages[rc]);
1096	}
1097
1098	up_read(&pTask->mm->mmap_sem);
1099
1100	RTMemFree(papVMAs);
1101	rc = VERR_LOCK_FAILED;
1102	}
1103
1104	rtR0MemObjDelete(&pMemLnx->Core);
1105	IPRT_LINUX_RESTORE_EFL_AC();
1106	return rc;
1107	}
1108
1109
1110	DECLHIDDEN(int) rtR0MemObjNativeLockKernel(PPRTR0MEMOBJINTERNAL ppMem, void *pv, size_t cb, uint32_t fAccess)
1111	{
1112	IPRT_LINUX_SAVE_EFL_AC();
1113	void pvLast = (uint8_t )pv + cb - 1;
1114	size_t const cPages = cb >> PAGE_SHIFT;
1115	PRTR0MEMOBJLNX pMemLnx;
1116	bool fLinearMapping;
1117	int rc;
1118	uint8_t *pbPage;
1119	size_t iPage;
1120	NOREF(fAccess);
1121
1122	if ( !RTR0MemKernelIsValidAddr(pv)
1123	\|\| !RTR0MemKernelIsValidAddr(pv + cb))
1124	return VERR_INVALID_PARAMETER;
1125
1126	/*
1127	* The lower part of the kernel memory has a linear mapping between
1128	* physical and virtual addresses. So we take a short cut here. This is
1129	* assumed to be the cleanest way to handle those addresses (and the code
1130	* is well tested, though the test for determining it is not very nice).
1131	* If we ever decide it isn't we can still remove it.
1132	*/
1133	#if 0
1134	fLinearMapping = (unsigned long)pvLast < VMALLOC_START;
1135	#else
1136	fLinearMapping = (unsigned long)pv >= (unsigned long)__va(0)
1137	&& (unsigned long)pvLast < (unsigned long)high_memory;
1138	#endif
1139
1140	/*
1141	* Allocate the memory object.
1142	*/
1143	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_OFFSETOF(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK, pv, cb);
1144	if (!pMemLnx)
1145	{
1146	IPRT_LINUX_RESTORE_EFL_AC();
1147	return VERR_NO_MEMORY;
1148	}
1149
1150	/*
1151	* Gather the pages.
1152	* We ASSUME all kernel pages are non-swappable and non-movable.
1153	*/
1154	rc = VINF_SUCCESS;
1155	pbPage = (uint8_t *)pvLast;
1156	iPage = cPages;
1157	if (!fLinearMapping)
1158	{
1159	while (iPage-- > 0)
1160	{
1161	struct page *pPage = rtR0MemObjLinuxVirtToPage(pbPage);
1162	if (RT_UNLIKELY(!pPage))
1163	{
1164	rc = VERR_LOCK_FAILED;
1165	break;
1166	}
1167	pMemLnx->apPages[iPage] = pPage;
1168	pbPage -= PAGE_SIZE;
1169	}
1170	}
1171	else
1172	{
1173	while (iPage-- > 0)
1174	{
1175	pMemLnx->apPages[iPage] = virt_to_page(pbPage);
1176	pbPage -= PAGE_SIZE;
1177	}
1178	}
1179	if (RT_SUCCESS(rc))
1180	{
1181	/*
1182	* Complete the memory object and return.
1183	*/
1184	pMemLnx->Core.u.Lock.R0Process = NIL_RTR0PROCESS;
1185	pMemLnx->cPages = cPages;
1186	Assert(!pMemLnx->fMappedToRing0);
1187	*ppMem = &pMemLnx->Core;
1188
1189	IPRT_LINUX_RESTORE_EFL_AC();
1190	return VINF_SUCCESS;
1191	}
1192
1193	rtR0MemObjDelete(&pMemLnx->Core);
1194	IPRT_LINUX_RESTORE_EFL_AC();
1195	return rc;
1196	}
1197
1198
1199	DECLHIDDEN(int) rtR0MemObjNativeReserveKernel(PPRTR0MEMOBJINTERNAL ppMem, void *pvFixed, size_t cb, size_t uAlignment)
1200	{
1201	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1202	IPRT_LINUX_SAVE_EFL_AC();
1203	const size_t cPages = cb >> PAGE_SHIFT;
1204	struct page *pDummyPage;
1205	struct page **papPages;
1206
1207	/* check for unsupported stuff. */
1208	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
1209	if (uAlignment > PAGE_SIZE)
1210	return VERR_NOT_SUPPORTED;
1211
1212	/*
1213	* Allocate a dummy page and create a page pointer array for vmap such that
1214	* the dummy page is mapped all over the reserved area.
1215	*/
1216	pDummyPage = alloc_page(GFP_HIGHUSER \| __GFP_NOWARN);
1217	if (pDummyPage)
1218	{
1219	papPages = RTMemAlloc(sizeof(papPages) cPages);
1220	if (papPages)
1221	{
1222	void *pv;
1223	size_t iPage = cPages;
1224	while (iPage-- > 0)
1225	papPages[iPage] = pDummyPage;
1226	# ifdef VM_MAP
1227	pv = vmap(papPages, cPages, VM_MAP, PAGE_KERNEL_RO);
1228	# else
1229	pv = vmap(papPages, cPages, VM_ALLOC, PAGE_KERNEL_RO);
1230	# endif
1231	RTMemFree(papPages);
1232	if (pv)
1233	{
1234	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb);
1235	if (pMemLnx)
1236	{
1237	pMemLnx->Core.u.ResVirt.R0Process = NIL_RTR0PROCESS;
1238	pMemLnx->cPages = 1;
1239	pMemLnx->apPages[0] = pDummyPage;
1240	*ppMem = &pMemLnx->Core;
1241	IPRT_LINUX_RESTORE_EFL_AC();
1242	return VINF_SUCCESS;
1243	}
1244	vunmap(pv);
1245	}
1246	}
1247	__free_page(pDummyPage);
1248	}
1249	IPRT_LINUX_RESTORE_EFL_AC();
1250	return VERR_NO_MEMORY;
1251
1252	#else /* < 2.4.22 */
1253	/*
1254	* Could probably use ioremap here, but the caller is in a better position than us
1255	* to select some safe physical memory.
1256	*/
1257	return VERR_NOT_SUPPORTED;
1258	#endif
1259	}
1260
1261
1262	DECLHIDDEN(int) rtR0MemObjNativeReserveUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment, RTR0PROCESS R0Process)
1263	{
1264	IPRT_LINUX_SAVE_EFL_AC();
1265	PRTR0MEMOBJLNX pMemLnx;
1266	void *pv;
1267	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1268	if (!pTask)
1269	return VERR_NOT_SUPPORTED;
1270
1271	/*
1272	* Check that the specified alignment is supported.
1273	*/
1274	if (uAlignment > PAGE_SIZE)
1275	return VERR_NOT_SUPPORTED;
1276
1277	/*
1278	* Let rtR0MemObjLinuxDoMmap do the difficult bits.
1279	*/
1280	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, cb, uAlignment, pTask, RTMEM_PROT_NONE);
1281	if (pv == (void *)-1)
1282	{
1283	IPRT_LINUX_RESTORE_EFL_AC();
1284	return VERR_NO_MEMORY;
1285	}
1286
1287	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb);
1288	if (!pMemLnx)
1289	{
1290	rtR0MemObjLinuxDoMunmap(pv, cb, pTask);
1291	IPRT_LINUX_RESTORE_EFL_AC();
1292	return VERR_NO_MEMORY;
1293	}
1294
1295	pMemLnx->Core.u.ResVirt.R0Process = R0Process;
1296	*ppMem = &pMemLnx->Core;
1297	IPRT_LINUX_RESTORE_EFL_AC();
1298	return VINF_SUCCESS;
1299	}
1300
1301
1302	DECLHIDDEN(int) rtR0MemObjNativeMapKernel(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap,
1303	void *pvFixed, size_t uAlignment,
1304	unsigned fProt, size_t offSub, size_t cbSub)
1305	{
1306	int rc = VERR_NO_MEMORY;
1307	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1308	PRTR0MEMOBJLNX pMemLnx;
1309	IPRT_LINUX_SAVE_EFL_AC();
1310
1311	/* Fail if requested to do something we can't. */
1312	AssertMsgReturn(!offSub && !cbSub, ("%#x %#x\n", offSub, cbSub), VERR_NOT_SUPPORTED);
1313	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
1314	if (uAlignment > PAGE_SIZE)
1315	return VERR_NOT_SUPPORTED;
1316
1317	/*
1318	* Create the IPRT memory object.
1319	*/
1320	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, pMemLnxToMap->Core.cb);
1321	if (pMemLnx)
1322	{
1323	if (pMemLnxToMap->cPages)
1324	{
1325	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1326	/*
1327	* Use vmap - 2.4.22 and later.
1328	*/
1329	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, true /* kernel */);
1330	# ifdef VM_MAP
1331	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[0], pMemLnxToMap->cPages, VM_MAP, fPg);
1332	# else
1333	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[0], pMemLnxToMap->cPages, VM_ALLOC, fPg);
1334	# endif
1335	if (pMemLnx->Core.pv)
1336	{
1337	pMemLnx->fMappedToRing0 = true;
1338	rc = VINF_SUCCESS;
1339	}
1340	else
1341	rc = VERR_MAP_FAILED;
1342
1343	#else /* < 2.4.22 */
1344	/*
1345	* Only option here is to share mappings if possible and forget about fProt.
1346	*/
1347	if (rtR0MemObjIsRing3(pMemToMap))
1348	rc = VERR_NOT_SUPPORTED;
1349	else
1350	{
1351	rc = VINF_SUCCESS;
1352	if (!pMemLnxToMap->Core.pv)
1353	rc = rtR0MemObjLinuxVMap(pMemLnxToMap, !!(fProt & RTMEM_PROT_EXEC));
1354	if (RT_SUCCESS(rc))
1355	{
1356	Assert(pMemLnxToMap->Core.pv);
1357	pMemLnx->Core.pv = pMemLnxToMap->Core.pv;
1358	}
1359	}
1360	#endif
1361	}
1362	else
1363	{
1364	/*
1365	* MMIO / physical memory.
1366	*/
1367	Assert(pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS && !pMemLnxToMap->Core.u.Phys.fAllocated);
1368	pMemLnx->Core.pv = pMemLnxToMap->Core.u.Phys.uCachePolicy == RTMEM_CACHE_POLICY_MMIO
1369	? ioremap_nocache(pMemLnxToMap->Core.u.Phys.PhysBase, pMemLnxToMap->Core.cb)
1370	: ioremap(pMemLnxToMap->Core.u.Phys.PhysBase, pMemLnxToMap->Core.cb);
1371	if (pMemLnx->Core.pv)
1372	{
1373	/** @todo fix protection. */
1374	rc = VINF_SUCCESS;
1375	}
1376	}
1377	if (RT_SUCCESS(rc))
1378	{
1379	pMemLnx->Core.u.Mapping.R0Process = NIL_RTR0PROCESS;
1380	*ppMem = &pMemLnx->Core;
1381	IPRT_LINUX_RESTORE_EFL_AC();
1382	return VINF_SUCCESS;
1383	}
1384	rtR0MemObjDelete(&pMemLnx->Core);
1385	}
1386
1387	IPRT_LINUX_RESTORE_EFL_AC();
1388	return rc;
1389	}
1390
1391
1392	#ifdef VBOX_USE_PAE_HACK
1393	/**
1394	* Replace the PFN of a PTE with the address of the actual page.
1395	*
1396	* The caller maps a reserved dummy page at the address with the desired access
1397	* and flags.
1398	*
1399	* This hack is required for older Linux kernels which don't provide
1400	* remap_pfn_range().
1401	*
1402	* @returns 0 on success, -ENOMEM on failure.
1403	* @param mm The memory context.
1404	* @param ulAddr The mapping address.
1405	* @param Phys The physical address of the page to map.
1406	*/
1407	static int rtR0MemObjLinuxFixPte(struct mm_struct *mm, unsigned long ulAddr, RTHCPHYS Phys)
1408	{
1409	int rc = -ENOMEM;
1410	pgd_t *pgd;
1411
1412	spin_lock(&mm->page_table_lock);
1413
1414	pgd = pgd_offset(mm, ulAddr);
1415	if (!pgd_none(pgd) && !pgd_bad(pgd))
1416	{
1417	pmd_t *pmd = pmd_offset(pgd, ulAddr);
1418	if (!pmd_none(*pmd))
1419	{
1420	pte_t *ptep = pte_offset_map(pmd, ulAddr);
1421	if (ptep)
1422	{
1423	pte_t pte = *ptep;
1424	pte.pte_high &= 0xfff00000;
1425	pte.pte_high \|= ((Phys >> 32) & 0x000fffff);
1426	pte.pte_low &= 0x00000fff;
1427	pte.pte_low \|= (Phys & 0xfffff000);
1428	set_pte(ptep, pte);
1429	pte_unmap(ptep);
1430	rc = 0;
1431	}
1432	}
1433	}
1434
1435	spin_unlock(&mm->page_table_lock);
1436	return rc;
1437	}
1438	#endif /* VBOX_USE_PAE_HACK */
1439
1440
1441	DECLHIDDEN(int) rtR0MemObjNativeMapUser(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap, RTR3PTR R3PtrFixed,
1442	size_t uAlignment, unsigned fProt, RTR0PROCESS R0Process)
1443	{
1444	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1445	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1446	int rc = VERR_NO_MEMORY;
1447	PRTR0MEMOBJLNX pMemLnx;
1448	#ifdef VBOX_USE_PAE_HACK
1449	struct page *pDummyPage;
1450	RTHCPHYS DummyPhys;
1451	#endif
1452	IPRT_LINUX_SAVE_EFL_AC();
1453
1454	/*
1455	* Check for restrictions.
1456	*/
1457	if (!pTask)
1458	return VERR_NOT_SUPPORTED;
1459	if (uAlignment > PAGE_SIZE)
1460	return VERR_NOT_SUPPORTED;
1461
1462	#ifdef VBOX_USE_PAE_HACK
1463	/*
1464	* Allocate a dummy page for use when mapping the memory.
1465	*/
1466	pDummyPage = alloc_page(GFP_USER \| __GFP_NOWARN);
1467	if (!pDummyPage)
1468	{
1469	IPRT_LINUX_RESTORE_EFL_AC();
1470	return VERR_NO_MEMORY;
1471	}
1472	SetPageReserved(pDummyPage);
1473	DummyPhys = page_to_phys(pDummyPage);
1474	#endif
1475
1476	/*
1477	* Create the IPRT memory object.
1478	*/
1479	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, pMemLnxToMap->Core.cb);
1480	if (pMemLnx)
1481	{
1482	/*
1483	* Allocate user space mapping.
1484	*/
1485	void *pv;
1486	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, pMemLnxToMap->Core.cb, uAlignment, pTask, fProt);
1487	if (pv != (void *)-1)
1488	{
1489	/*
1490	* Map page by page into the mmap area.
1491	* This is generic, paranoid and not very efficient.
1492	*/
1493	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, false /* user */);
1494	unsigned long ulAddrCur = (unsigned long)pv;
1495	const size_t cPages = pMemLnxToMap->Core.cb >> PAGE_SHIFT;
1496	size_t iPage;
1497
1498	down_write(&pTask->mm->mmap_sem);
1499
1500	rc = VINF_SUCCESS;
1501	if (pMemLnxToMap->cPages)
1502	{
1503	for (iPage = 0; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE)
1504	{
1505	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11)
1506	RTHCPHYS Phys = page_to_phys(pMemLnxToMap->apPages[iPage]);
1507	#endif
1508	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1509	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
1510	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
1511	#endif
1512	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0) && defined(RT_ARCH_X86)
1513	/* remap_page_range() limitation on x86 */
1514	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
1515	#endif
1516
1517	#if defined(VBOX_USE_INSERT_PAGE) && LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1518	rc = vm_insert_page(vma, ulAddrCur, pMemLnxToMap->apPages[iPage]);
1519	/* Thes flags help making 100% sure some bad stuff wont happen (swap, core, ++).
1520	* See remap_pfn_range() in mm/memory.c */
1521	#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 7, 0)
1522	vma->vm_flags \|= VM_DONTEXPAND \| VM_DONTDUMP;
1523	#else
1524	vma->vm_flags \|= VM_RESERVED;
1525	#endif
1526	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
1527	rc = remap_pfn_range(vma, ulAddrCur, page_to_pfn(pMemLnxToMap->apPages[iPage]), PAGE_SIZE, fPg);
1528	#elif defined(VBOX_USE_PAE_HACK)
1529	rc = remap_page_range(vma, ulAddrCur, DummyPhys, PAGE_SIZE, fPg);
1530	if (!rc)
1531	rc = rtR0MemObjLinuxFixPte(pTask->mm, ulAddrCur, Phys);
1532	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1533	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1534	#else /* 2.4 */
1535	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
1536	#endif
1537	if (rc)
1538	{
1539	rc = VERR_NO_MEMORY;
1540	break;
1541	}
1542	}
1543	}
1544	else
1545	{
1546	RTHCPHYS Phys;
1547	if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS)
1548	Phys = pMemLnxToMap->Core.u.Phys.PhysBase;
1549	else if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_CONT)
1550	Phys = pMemLnxToMap->Core.u.Cont.Phys;
1551	else
1552	{
1553	AssertMsgFailed(("%d\n", pMemLnxToMap->Core.enmType));
1554	Phys = NIL_RTHCPHYS;
1555	}
1556	if (Phys != NIL_RTHCPHYS)
1557	{
1558	for (iPage = 0; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE, Phys += PAGE_SIZE)
1559	{
1560	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1561	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
1562	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
1563	#endif
1564	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0) && defined(RT_ARCH_X86)
1565	/* remap_page_range() limitation on x86 */
1566	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
1567	#endif
1568
1569	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
1570	rc = remap_pfn_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1571	#elif defined(VBOX_USE_PAE_HACK)
1572	rc = remap_page_range(vma, ulAddrCur, DummyPhys, PAGE_SIZE, fPg);
1573	if (!rc)
1574	rc = rtR0MemObjLinuxFixPte(pTask->mm, ulAddrCur, Phys);
1575	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1576	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1577	#else /* 2.4 */
1578	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
1579	#endif
1580	if (rc)
1581	{
1582	rc = VERR_NO_MEMORY;
1583	break;
1584	}
1585	}
1586	}
1587	}
1588
1589	#ifdef CONFIG_NUMA_BALANCING
1590	# if LINUX_VERSION_CODE < KERNEL_VERSION(3, 13, 0)
1591	# ifdef RHEL_RELEASE_CODE
1592	# if RHEL_RELEASE_CODE < RHEL_RELEASE_VERSION(7, 0)
1593	# define VBOX_NUMA_HACK_OLD
1594	# endif
1595	# endif
1596	# endif
1597	if (RT_SUCCESS(rc))
1598	{
1599	/** @todo Ugly hack! But right now we have no other means to
1600	* disable automatic NUMA page balancing. */
1601	# ifdef RT_OS_X86
1602	# ifdef VBOX_NUMA_HACK_OLD
1603	pTask->mm->numa_next_reset = jiffies + 0x7fffffffUL;
1604	# endif
1605	pTask->mm->numa_next_scan = jiffies + 0x7fffffffUL;
1606	# else
1607	# ifdef VBOX_NUMA_HACK_OLD
1608	pTask->mm->numa_next_reset = jiffies + 0x7fffffffffffffffUL;
1609	# endif
1610	pTask->mm->numa_next_scan = jiffies + 0x7fffffffffffffffUL;
1611	# endif
1612	}
1613	#endif /* CONFIG_NUMA_BALANCING */
1614
1615	up_write(&pTask->mm->mmap_sem);
1616
1617	if (RT_SUCCESS(rc))
1618	{
1619	#ifdef VBOX_USE_PAE_HACK
1620	__free_page(pDummyPage);
1621	#endif
1622	pMemLnx->Core.pv = pv;
1623	pMemLnx->Core.u.Mapping.R0Process = R0Process;
1624	*ppMem = &pMemLnx->Core;
1625	IPRT_LINUX_RESTORE_EFL_AC();
1626	return VINF_SUCCESS;
1627	}
1628
1629	/*
1630	* Bail out.
1631	*/
1632	rtR0MemObjLinuxDoMunmap(pv, pMemLnxToMap->Core.cb, pTask);
1633	}
1634	rtR0MemObjDelete(&pMemLnx->Core);
1635	}
1636	#ifdef VBOX_USE_PAE_HACK
1637	__free_page(pDummyPage);
1638	#endif
1639
1640	IPRT_LINUX_RESTORE_EFL_AC();
1641	return rc;
1642	}
1643
1644
1645	DECLHIDDEN(int) rtR0MemObjNativeProtect(PRTR0MEMOBJINTERNAL pMem, size_t offSub, size_t cbSub, uint32_t fProt)
1646	{
1647	NOREF(pMem);
1648	NOREF(offSub);
1649	NOREF(cbSub);
1650	NOREF(fProt);
1651	return VERR_NOT_SUPPORTED;
1652	}
1653
1654
1655	DECLHIDDEN(RTHCPHYS) rtR0MemObjNativeGetPagePhysAddr(PRTR0MEMOBJINTERNAL pMem, size_t iPage)
1656	{
1657	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
1658
1659	if (pMemLnx->cPages)
1660	return page_to_phys(pMemLnx->apPages[iPage]);
1661
1662	switch (pMemLnx->Core.enmType)
1663	{
1664	case RTR0MEMOBJTYPE_CONT:
1665	return pMemLnx->Core.u.Cont.Phys + (iPage << PAGE_SHIFT);
1666
1667	case RTR0MEMOBJTYPE_PHYS:
1668	return pMemLnx->Core.u.Phys.PhysBase + (iPage << PAGE_SHIFT);
1669
1670	/* the parent knows */
1671	case RTR0MEMOBJTYPE_MAPPING:
1672	return rtR0MemObjNativeGetPagePhysAddr(pMemLnx->Core.uRel.Child.pParent, iPage);
1673
1674	/* cPages > 0 */
1675	case RTR0MEMOBJTYPE_LOW:
1676	case RTR0MEMOBJTYPE_LOCK:
1677	case RTR0MEMOBJTYPE_PHYS_NC:
1678	case RTR0MEMOBJTYPE_PAGE:
1679	default:
1680	AssertMsgFailed(("%d\n", pMemLnx->Core.enmType));
1681	/* fall thru */
1682
1683	case RTR0MEMOBJTYPE_RES_VIRT:
1684	return NIL_RTHCPHYS;
1685	}
1686	}
1687

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

source: vbox/trunk/src/VBox/Runtime/r0drv/linux/memobj-r0drv-linux.c@ 60372

Download in other formats: