memobj-r0drv-linux.c@ 60453

Last change on this file since 60453 was 60453, checked in by vboxsync, 9 years ago
Runtime/Additions: Linux 4.6 fixes
Property svn:eol-style set to `native` Property svn:keywords set to `Id Rev Revision`
File size: 55.6 KB

Line
1	/* $Id: memobj-r0drv-linux.c 60453 2016-04-12 13:04:28Z 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	#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 6, 0)
593	put_page(pMemLnx->apPages[iPage]);
594	#else
595	page_cache_release(pMemLnx->apPages[iPage]);
596	#endif
597	}
598
599	if (pTask && pTask->mm)
600	up_read(&pTask->mm->mmap_sem);
601	}
602	/* else: kernel memory - nothing to do here. */
603	break;
604
605	case RTR0MEMOBJTYPE_RES_VIRT:
606	Assert(pMemLnx->Core.pv);
607	if (pMemLnx->Core.u.ResVirt.R0Process != NIL_RTR0PROCESS)
608	{
609	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
610	Assert(pTask);
611	if (pTask && pTask->mm)
612	rtR0MemObjLinuxDoMunmap(pMemLnx->Core.pv, pMemLnx->Core.cb, pTask);
613	}
614	else
615	{
616	vunmap(pMemLnx->Core.pv);
617
618	Assert(pMemLnx->cPages == 1 && pMemLnx->apPages[0] != NULL);
619	__free_page(pMemLnx->apPages[0]);
620	pMemLnx->apPages[0] = NULL;
621	pMemLnx->cPages = 0;
622	}
623	pMemLnx->Core.pv = NULL;
624	break;
625
626	case RTR0MEMOBJTYPE_MAPPING:
627	Assert(pMemLnx->cPages == 0); Assert(pMemLnx->Core.pv);
628	if (pMemLnx->Core.u.ResVirt.R0Process != NIL_RTR0PROCESS)
629	{
630	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
631	Assert(pTask);
632	if (pTask && pTask->mm)
633	rtR0MemObjLinuxDoMunmap(pMemLnx->Core.pv, pMemLnx->Core.cb, pTask);
634	}
635	else
636	vunmap(pMemLnx->Core.pv);
637	pMemLnx->Core.pv = NULL;
638	break;
639
640	default:
641	AssertMsgFailed(("enmType=%d\n", pMemLnx->Core.enmType));
642	return VERR_INTERNAL_ERROR;
643	}
644	IPRT_LINUX_RESTORE_EFL_ONLY_AC();
645	return VINF_SUCCESS;
646	}
647
648
649	DECLHIDDEN(int) rtR0MemObjNativeAllocPage(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable)
650	{
651	IPRT_LINUX_SAVE_EFL_AC();
652	PRTR0MEMOBJLNX pMemLnx;
653	int rc;
654
655	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
656	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_PAGE, cb, PAGE_SIZE, GFP_HIGHUSER,
657	false /* non-contiguous */, VERR_NO_MEMORY);
658	#else
659	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_PAGE, cb, PAGE_SIZE, GFP_USER,
660	false /* non-contiguous */, VERR_NO_MEMORY);
661	#endif
662	if (RT_SUCCESS(rc))
663	{
664	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
665	if (RT_SUCCESS(rc))
666	{
667	*ppMem = &pMemLnx->Core;
668	IPRT_LINUX_RESTORE_EFL_AC();
669	return rc;
670	}
671
672	rtR0MemObjLinuxFreePages(pMemLnx);
673	rtR0MemObjDelete(&pMemLnx->Core);
674	}
675
676	IPRT_LINUX_RESTORE_EFL_AC();
677	return rc;
678	}
679
680
681	DECLHIDDEN(int) rtR0MemObjNativeAllocLow(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable)
682	{
683	IPRT_LINUX_SAVE_EFL_AC();
684	PRTR0MEMOBJLNX pMemLnx;
685	int rc;
686
687	/* Try to avoid GFP_DMA. GFM_DMA32 was introduced with Linux 2.6.15. */
688	#if (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
689	/* ZONE_DMA32: 0-4GB */
690	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_DMA32,
691	false /* non-contiguous */, VERR_NO_LOW_MEMORY);
692	if (RT_FAILURE(rc))
693	#endif
694	#ifdef RT_ARCH_AMD64
695	/* ZONE_DMA: 0-16MB */
696	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_DMA,
697	false /* non-contiguous */, VERR_NO_LOW_MEMORY);
698	#else
699	# ifdef CONFIG_X86_PAE
700	# endif
701	/* ZONE_NORMAL: 0-896MB */
702	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_USER,
703	false /* non-contiguous */, VERR_NO_LOW_MEMORY);
704	#endif
705	if (RT_SUCCESS(rc))
706	{
707	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
708	if (RT_SUCCESS(rc))
709	{
710	*ppMem = &pMemLnx->Core;
711	IPRT_LINUX_RESTORE_EFL_AC();
712	return rc;
713	}
714
715	rtR0MemObjLinuxFreePages(pMemLnx);
716	rtR0MemObjDelete(&pMemLnx->Core);
717	}
718
719	IPRT_LINUX_RESTORE_EFL_AC();
720	return rc;
721	}
722
723
724	DECLHIDDEN(int) rtR0MemObjNativeAllocCont(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable)
725	{
726	IPRT_LINUX_SAVE_EFL_AC();
727	PRTR0MEMOBJLNX pMemLnx;
728	int rc;
729
730	#if (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
731	/* ZONE_DMA32: 0-4GB */
732	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, PAGE_SIZE, GFP_DMA32,
733	true /* contiguous */, VERR_NO_CONT_MEMORY);
734	if (RT_FAILURE(rc))
735	#endif
736	#ifdef RT_ARCH_AMD64
737	/* ZONE_DMA: 0-16MB */
738	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, PAGE_SIZE, GFP_DMA,
739	true /* contiguous */, VERR_NO_CONT_MEMORY);
740	#else
741	/* ZONE_NORMAL (32-bit hosts): 0-896MB */
742	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, PAGE_SIZE, GFP_USER,
743	true /* contiguous */, VERR_NO_CONT_MEMORY);
744	#endif
745	if (RT_SUCCESS(rc))
746	{
747	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
748	if (RT_SUCCESS(rc))
749	{
750	#if defined(RT_STRICT) && (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_HIGHMEM64G))
751	size_t iPage = pMemLnx->cPages;
752	while (iPage-- > 0)
753	Assert(page_to_phys(pMemLnx->apPages[iPage]) < _4G);
754	#endif
755	pMemLnx->Core.u.Cont.Phys = page_to_phys(pMemLnx->apPages[0]);
756	*ppMem = &pMemLnx->Core;
757	IPRT_LINUX_RESTORE_EFL_AC();
758	return rc;
759	}
760
761	rtR0MemObjLinuxFreePages(pMemLnx);
762	rtR0MemObjDelete(&pMemLnx->Core);
763	}
764
765	IPRT_LINUX_RESTORE_EFL_AC();
766	return rc;
767	}
768
769
770	/**
771	* Worker for rtR0MemObjLinuxAllocPhysSub that tries one allocation strategy.
772	*
773	* @returns IPRT status code.
774	* @param ppMemLnx Where to
775	* @param enmType The object type.
776	* @param cb The size of the allocation.
777	* @param uAlignment The alignment of the physical memory.
778	* Only valid for fContiguous == true, ignored otherwise.
779	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
780	* @param fGfp The Linux GFP flags to use for the allocation.
781	*/
782	static int rtR0MemObjLinuxAllocPhysSub2(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType,
783	size_t cb, size_t uAlignment, RTHCPHYS PhysHighest, unsigned fGfp)
784	{
785	PRTR0MEMOBJLNX pMemLnx;
786	int rc;
787
788	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, enmType, cb, uAlignment, fGfp,
789	enmType == RTR0MEMOBJTYPE_PHYS /* contiguous / non-contiguous */,
790	VERR_NO_PHYS_MEMORY);
791	if (RT_FAILURE(rc))
792	return rc;
793
794	/*
795	* Check the addresses if necessary. (Can be optimized a bit for PHYS.)
796	*/
797	if (PhysHighest != NIL_RTHCPHYS)
798	{
799	size_t iPage = pMemLnx->cPages;
800	while (iPage-- > 0)
801	if (page_to_phys(pMemLnx->apPages[iPage]) > PhysHighest)
802	{
803	rtR0MemObjLinuxFreePages(pMemLnx);
804	rtR0MemObjDelete(&pMemLnx->Core);
805	return VERR_NO_MEMORY;
806	}
807	}
808
809	/*
810	* Complete the object.
811	*/
812	if (enmType == RTR0MEMOBJTYPE_PHYS)
813	{
814	pMemLnx->Core.u.Phys.PhysBase = page_to_phys(pMemLnx->apPages[0]);
815	pMemLnx->Core.u.Phys.fAllocated = true;
816	}
817	*ppMem = &pMemLnx->Core;
818	return rc;
819	}
820
821
822	/**
823	* Worker for rtR0MemObjNativeAllocPhys and rtR0MemObjNativeAllocPhysNC.
824	*
825	* @returns IPRT status code.
826	* @param ppMem Where to store the memory object pointer on success.
827	* @param enmType The object type.
828	* @param cb The size of the allocation.
829	* @param uAlignment The alignment of the physical memory.
830	* Only valid for enmType == RTR0MEMOBJTYPE_PHYS, ignored otherwise.
831	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
832	*/
833	static int rtR0MemObjLinuxAllocPhysSub(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType,
834	size_t cb, size_t uAlignment, RTHCPHYS PhysHighest)
835	{
836	int rc;
837	IPRT_LINUX_SAVE_EFL_AC();
838
839	/*
840	* There are two clear cases and that's the <=16MB and anything-goes ones.
841	* When the physical address limit is somewhere in-between those two we'll
842	* just have to try, starting with HIGHUSER and working our way thru the
843	* different types, hoping we'll get lucky.
844	*
845	* We should probably move this physical address restriction logic up to
846	* the page alloc function as it would be more efficient there. But since
847	* we don't expect this to be a performance issue just yet it can wait.
848	*/
849	if (PhysHighest == NIL_RTHCPHYS)
850	/* ZONE_HIGHMEM: the whole physical memory */
851	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_HIGHUSER);
852	else if (PhysHighest <= _1M * 16)
853	/* ZONE_DMA: 0-16MB */
854	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_DMA);
855	else
856	{
857	rc = VERR_NO_MEMORY;
858	if (RT_FAILURE(rc))
859	/* ZONE_HIGHMEM: the whole physical memory */
860	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_HIGHUSER);
861	if (RT_FAILURE(rc))
862	/* ZONE_NORMAL: 0-896MB */
863	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_USER);
864	#ifdef GFP_DMA32
865	if (RT_FAILURE(rc))
866	/* ZONE_DMA32: 0-4GB */
867	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_DMA32);
868	#endif
869	if (RT_FAILURE(rc))
870	/* ZONE_DMA: 0-16MB */
871	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_DMA);
872	}
873	IPRT_LINUX_RESTORE_EFL_AC();
874	return rc;
875	}
876
877
878	/**
879	* Translates a kernel virtual address to a linux page structure by walking the
880	* page tables.
881	*
882	* @note We do assume that the page tables will not change as we are walking
883	* them. This assumption is rather forced by the fact that I could not
884	* immediately see any way of preventing this from happening. So, we
885	* take some extra care when accessing them.
886	*
887	* Because of this, we don't want to use this function on memory where
888	* attribute changes to nearby pages is likely to cause large pages to
889	* be used or split up. So, don't use this for the linear mapping of
890	* physical memory.
891	*
892	* @returns Pointer to the page structur or NULL if it could not be found.
893	* @param pv The kernel virtual address.
894	*/
895	static struct page rtR0MemObjLinuxVirtToPage(void pv)
896	{
897	unsigned long ulAddr = (unsigned long)pv;
898	unsigned long pfn;
899	struct page *pPage;
900	pte_t *pEntry;
901	union
902	{
903	pgd_t Global;
904	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
905	pud_t Upper;
906	#endif
907	pmd_t Middle;
908	pte_t Entry;
909	} u;
910
911	/* Should this happen in a situation this code will be called in? And if
912	* so, can it change under our feet? See also
913	* "Documentation/vm/active_mm.txt" in the kernel sources. */
914	if (RT_UNLIKELY(!current->active_mm))
915	return NULL;
916	u.Global = *pgd_offset(current->active_mm, ulAddr);
917	if (RT_UNLIKELY(pgd_none(u.Global)))
918	return NULL;
919
920	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
921	u.Upper = *pud_offset(&u.Global, ulAddr);
922	if (RT_UNLIKELY(pud_none(u.Upper)))
923	return NULL;
924	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25)
925	if (pud_large(u.Upper))
926	{
927	pPage = pud_page(u.Upper);
928	AssertReturn(pPage, NULL);
929	pfn = page_to_pfn(pPage); /* doing the safe way... */
930	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (PUD_SHIFT - PAGE_SHIFT)) - 1);
931	return pfn_to_page(pfn);
932	}
933	# endif
934
935	u.Middle = *pmd_offset(&u.Upper, ulAddr);
936	#else /* < 2.6.11 */
937	u.Middle = *pmd_offset(&u.Global, ulAddr);
938	#endif /* < 2.6.11 */
939	if (RT_UNLIKELY(pmd_none(u.Middle)))
940	return NULL;
941	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
942	if (pmd_large(u.Middle))
943	{
944	pPage = pmd_page(u.Middle);
945	AssertReturn(pPage, NULL);
946	pfn = page_to_pfn(pPage); /* doing the safe way... */
947	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (PMD_SHIFT - PAGE_SHIFT)) - 1);
948	return pfn_to_page(pfn);
949	}
950	#endif
951
952	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 5) \|\| defined(pte_offset_map) /* As usual, RHEL 3 had pte_offset_map earlier. */
953	pEntry = pte_offset_map(&u.Middle, ulAddr);
954	#else
955	pEntry = pte_offset(&u.Middle, ulAddr);
956	#endif
957	if (RT_UNLIKELY(!pEntry))
958	return NULL;
959	u.Entry = *pEntry;
960	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 5) \|\| defined(pte_offset_map)
961	pte_unmap(pEntry);
962	#endif
963
964	if (RT_UNLIKELY(!pte_present(u.Entry)))
965	return NULL;
966	return pte_page(u.Entry);
967	}
968
969
970	DECLHIDDEN(int) rtR0MemObjNativeAllocPhys(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest, size_t uAlignment)
971	{
972	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS, cb, uAlignment, PhysHighest);
973	}
974
975
976	DECLHIDDEN(int) rtR0MemObjNativeAllocPhysNC(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest)
977	{
978	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS_NC, cb, PAGE_SIZE, PhysHighest);
979	}
980
981
982	DECLHIDDEN(int) rtR0MemObjNativeEnterPhys(PPRTR0MEMOBJINTERNAL ppMem, RTHCPHYS Phys, size_t cb, uint32_t uCachePolicy)
983	{
984	IPRT_LINUX_SAVE_EFL_AC();
985
986	/*
987	* All we need to do here is to validate that we can use
988	* ioremap on the specified address (32/64-bit dma_addr_t).
989	*/
990	PRTR0MEMOBJLNX pMemLnx;
991	dma_addr_t PhysAddr = Phys;
992	AssertMsgReturn(PhysAddr == Phys, ("%#llx\n", (unsigned long long)Phys), VERR_ADDRESS_TOO_BIG);
993
994	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_PHYS, NULL, cb);
995	if (!pMemLnx)
996	{
997	IPRT_LINUX_RESTORE_EFL_AC();
998	return VERR_NO_MEMORY;
999	}
1000
1001	pMemLnx->Core.u.Phys.PhysBase = PhysAddr;
1002	pMemLnx->Core.u.Phys.fAllocated = false;
1003	pMemLnx->Core.u.Phys.uCachePolicy = uCachePolicy;
1004	Assert(!pMemLnx->cPages);
1005	*ppMem = &pMemLnx->Core;
1006	IPRT_LINUX_RESTORE_EFL_AC();
1007	return VINF_SUCCESS;
1008	}
1009
1010
1011	DECLHIDDEN(int) rtR0MemObjNativeLockUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3Ptr, size_t cb, uint32_t fAccess, RTR0PROCESS R0Process)
1012	{
1013	IPRT_LINUX_SAVE_EFL_AC();
1014	const int cPages = cb >> PAGE_SHIFT;
1015	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1016	struct vm_area_struct **papVMAs;
1017	PRTR0MEMOBJLNX pMemLnx;
1018	int rc = VERR_NO_MEMORY;
1019	int const fWrite = fAccess & RTMEM_PROT_WRITE ? 1 : 0;
1020
1021	/*
1022	* Check for valid task and size overflows.
1023	*/
1024	if (!pTask)
1025	return VERR_NOT_SUPPORTED;
1026	if (((size_t)cPages << PAGE_SHIFT) != cb)
1027	return VERR_OUT_OF_RANGE;
1028
1029	/*
1030	* Allocate the memory object and a temporary buffer for the VMAs.
1031	*/
1032	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_OFFSETOF(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK, (void *)R3Ptr, cb);
1033	if (!pMemLnx)
1034	{
1035	IPRT_LINUX_RESTORE_EFL_AC();
1036	return VERR_NO_MEMORY;
1037	}
1038
1039	papVMAs = (struct vm_area_struct *)RTMemAlloc(sizeof(papVMAs) * cPages);
1040	if (papVMAs)
1041	{
1042	down_read(&pTask->mm->mmap_sem);
1043
1044	/*
1045	* Get user pages.
1046	*/
1047	rc = get_user_pages(pTask, /* Task for fault accounting. */
1048	pTask->mm, /* Whose pages. */
1049	R3Ptr, /* Where from. */
1050	cPages, /* How many pages. */
1051	fWrite, /* Write to memory. */
1052	fWrite, /* force write access. */
1053	&pMemLnx->apPages[0], /* Page array. */
1054	papVMAs); /* vmas */
1055	if (rc == cPages)
1056	{
1057	/*
1058	* Flush dcache (required?), protect against fork and _really_ pin the page
1059	* table entries. get_user_pages() will protect against swapping out the
1060	* pages but it will NOT protect against removing page table entries. This
1061	* can be achieved with
1062	* - using mlock / mmap(..., MAP_LOCKED, ...) from userland. This requires
1063	* an appropriate limit set up with setrlimit(..., RLIMIT_MEMLOCK, ...).
1064	* Usual Linux distributions support only a limited size of locked pages
1065	* (e.g. 32KB).
1066	* - setting the PageReserved bit (as we do in rtR0MemObjLinuxAllocPages()
1067	* or by
1068	* - setting the VM_LOCKED flag. This is the same as doing mlock() without
1069	* a range check.
1070	*/
1071	/** @todo The Linux fork() protection will require more work if this API
1072	* is to be used for anything but locking VM pages. */
1073	while (rc-- > 0)
1074	{
1075	flush_dcache_page(pMemLnx->apPages[rc]);
1076	papVMAs[rc]->vm_flags \|= (VM_DONTCOPY \| VM_LOCKED);
1077	}
1078
1079	up_read(&pTask->mm->mmap_sem);
1080
1081	RTMemFree(papVMAs);
1082
1083	pMemLnx->Core.u.Lock.R0Process = R0Process;
1084	pMemLnx->cPages = cPages;
1085	Assert(!pMemLnx->fMappedToRing0);
1086	*ppMem = &pMemLnx->Core;
1087
1088	IPRT_LINUX_RESTORE_EFL_AC();
1089	return VINF_SUCCESS;
1090	}
1091
1092	/*
1093	* Failed - we need to unlock any pages that we succeeded to lock.
1094	*/
1095	while (rc-- > 0)
1096	{
1097	if (!PageReserved(pMemLnx->apPages[rc]))
1098	SetPageDirty(pMemLnx->apPages[rc]);
1099	#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 6, 0)
1100	put_page(pMemLnx->apPages[rc]);
1101	#else
1102	page_cache_release(pMemLnx->apPages[rc]);
1103	#endif
1104	}
1105
1106	up_read(&pTask->mm->mmap_sem);
1107
1108	RTMemFree(papVMAs);
1109	rc = VERR_LOCK_FAILED;
1110	}
1111
1112	rtR0MemObjDelete(&pMemLnx->Core);
1113	IPRT_LINUX_RESTORE_EFL_AC();
1114	return rc;
1115	}
1116
1117
1118	DECLHIDDEN(int) rtR0MemObjNativeLockKernel(PPRTR0MEMOBJINTERNAL ppMem, void *pv, size_t cb, uint32_t fAccess)
1119	{
1120	IPRT_LINUX_SAVE_EFL_AC();
1121	void pvLast = (uint8_t )pv + cb - 1;
1122	size_t const cPages = cb >> PAGE_SHIFT;
1123	PRTR0MEMOBJLNX pMemLnx;
1124	bool fLinearMapping;
1125	int rc;
1126	uint8_t *pbPage;
1127	size_t iPage;
1128	NOREF(fAccess);
1129
1130	if ( !RTR0MemKernelIsValidAddr(pv)
1131	\|\| !RTR0MemKernelIsValidAddr(pv + cb))
1132	return VERR_INVALID_PARAMETER;
1133
1134	/*
1135	* The lower part of the kernel memory has a linear mapping between
1136	* physical and virtual addresses. So we take a short cut here. This is
1137	* assumed to be the cleanest way to handle those addresses (and the code
1138	* is well tested, though the test for determining it is not very nice).
1139	* If we ever decide it isn't we can still remove it.
1140	*/
1141	#if 0
1142	fLinearMapping = (unsigned long)pvLast < VMALLOC_START;
1143	#else
1144	fLinearMapping = (unsigned long)pv >= (unsigned long)__va(0)
1145	&& (unsigned long)pvLast < (unsigned long)high_memory;
1146	#endif
1147
1148	/*
1149	* Allocate the memory object.
1150	*/
1151	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_OFFSETOF(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK, pv, cb);
1152	if (!pMemLnx)
1153	{
1154	IPRT_LINUX_RESTORE_EFL_AC();
1155	return VERR_NO_MEMORY;
1156	}
1157
1158	/*
1159	* Gather the pages.
1160	* We ASSUME all kernel pages are non-swappable and non-movable.
1161	*/
1162	rc = VINF_SUCCESS;
1163	pbPage = (uint8_t *)pvLast;
1164	iPage = cPages;
1165	if (!fLinearMapping)
1166	{
1167	while (iPage-- > 0)
1168	{
1169	struct page *pPage = rtR0MemObjLinuxVirtToPage(pbPage);
1170	if (RT_UNLIKELY(!pPage))
1171	{
1172	rc = VERR_LOCK_FAILED;
1173	break;
1174	}
1175	pMemLnx->apPages[iPage] = pPage;
1176	pbPage -= PAGE_SIZE;
1177	}
1178	}
1179	else
1180	{
1181	while (iPage-- > 0)
1182	{
1183	pMemLnx->apPages[iPage] = virt_to_page(pbPage);
1184	pbPage -= PAGE_SIZE;
1185	}
1186	}
1187	if (RT_SUCCESS(rc))
1188	{
1189	/*
1190	* Complete the memory object and return.
1191	*/
1192	pMemLnx->Core.u.Lock.R0Process = NIL_RTR0PROCESS;
1193	pMemLnx->cPages = cPages;
1194	Assert(!pMemLnx->fMappedToRing0);
1195	*ppMem = &pMemLnx->Core;
1196
1197	IPRT_LINUX_RESTORE_EFL_AC();
1198	return VINF_SUCCESS;
1199	}
1200
1201	rtR0MemObjDelete(&pMemLnx->Core);
1202	IPRT_LINUX_RESTORE_EFL_AC();
1203	return rc;
1204	}
1205
1206
1207	DECLHIDDEN(int) rtR0MemObjNativeReserveKernel(PPRTR0MEMOBJINTERNAL ppMem, void *pvFixed, size_t cb, size_t uAlignment)
1208	{
1209	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1210	IPRT_LINUX_SAVE_EFL_AC();
1211	const size_t cPages = cb >> PAGE_SHIFT;
1212	struct page *pDummyPage;
1213	struct page **papPages;
1214
1215	/* check for unsupported stuff. */
1216	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
1217	if (uAlignment > PAGE_SIZE)
1218	return VERR_NOT_SUPPORTED;
1219
1220	/*
1221	* Allocate a dummy page and create a page pointer array for vmap such that
1222	* the dummy page is mapped all over the reserved area.
1223	*/
1224	pDummyPage = alloc_page(GFP_HIGHUSER \| __GFP_NOWARN);
1225	if (pDummyPage)
1226	{
1227	papPages = RTMemAlloc(sizeof(papPages) cPages);
1228	if (papPages)
1229	{
1230	void *pv;
1231	size_t iPage = cPages;
1232	while (iPage-- > 0)
1233	papPages[iPage] = pDummyPage;
1234	# ifdef VM_MAP
1235	pv = vmap(papPages, cPages, VM_MAP, PAGE_KERNEL_RO);
1236	# else
1237	pv = vmap(papPages, cPages, VM_ALLOC, PAGE_KERNEL_RO);
1238	# endif
1239	RTMemFree(papPages);
1240	if (pv)
1241	{
1242	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb);
1243	if (pMemLnx)
1244	{
1245	pMemLnx->Core.u.ResVirt.R0Process = NIL_RTR0PROCESS;
1246	pMemLnx->cPages = 1;
1247	pMemLnx->apPages[0] = pDummyPage;
1248	*ppMem = &pMemLnx->Core;
1249	IPRT_LINUX_RESTORE_EFL_AC();
1250	return VINF_SUCCESS;
1251	}
1252	vunmap(pv);
1253	}
1254	}
1255	__free_page(pDummyPage);
1256	}
1257	IPRT_LINUX_RESTORE_EFL_AC();
1258	return VERR_NO_MEMORY;
1259
1260	#else /* < 2.4.22 */
1261	/*
1262	* Could probably use ioremap here, but the caller is in a better position than us
1263	* to select some safe physical memory.
1264	*/
1265	return VERR_NOT_SUPPORTED;
1266	#endif
1267	}
1268
1269
1270	DECLHIDDEN(int) rtR0MemObjNativeReserveUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment, RTR0PROCESS R0Process)
1271	{
1272	IPRT_LINUX_SAVE_EFL_AC();
1273	PRTR0MEMOBJLNX pMemLnx;
1274	void *pv;
1275	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1276	if (!pTask)
1277	return VERR_NOT_SUPPORTED;
1278
1279	/*
1280	* Check that the specified alignment is supported.
1281	*/
1282	if (uAlignment > PAGE_SIZE)
1283	return VERR_NOT_SUPPORTED;
1284
1285	/*
1286	* Let rtR0MemObjLinuxDoMmap do the difficult bits.
1287	*/
1288	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, cb, uAlignment, pTask, RTMEM_PROT_NONE);
1289	if (pv == (void *)-1)
1290	{
1291	IPRT_LINUX_RESTORE_EFL_AC();
1292	return VERR_NO_MEMORY;
1293	}
1294
1295	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb);
1296	if (!pMemLnx)
1297	{
1298	rtR0MemObjLinuxDoMunmap(pv, cb, pTask);
1299	IPRT_LINUX_RESTORE_EFL_AC();
1300	return VERR_NO_MEMORY;
1301	}
1302
1303	pMemLnx->Core.u.ResVirt.R0Process = R0Process;
1304	*ppMem = &pMemLnx->Core;
1305	IPRT_LINUX_RESTORE_EFL_AC();
1306	return VINF_SUCCESS;
1307	}
1308
1309
1310	DECLHIDDEN(int) rtR0MemObjNativeMapKernel(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap,
1311	void *pvFixed, size_t uAlignment,
1312	unsigned fProt, size_t offSub, size_t cbSub)
1313	{
1314	int rc = VERR_NO_MEMORY;
1315	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1316	PRTR0MEMOBJLNX pMemLnx;
1317	IPRT_LINUX_SAVE_EFL_AC();
1318
1319	/* Fail if requested to do something we can't. */
1320	AssertMsgReturn(!offSub && !cbSub, ("%#x %#x\n", offSub, cbSub), VERR_NOT_SUPPORTED);
1321	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
1322	if (uAlignment > PAGE_SIZE)
1323	return VERR_NOT_SUPPORTED;
1324
1325	/*
1326	* Create the IPRT memory object.
1327	*/
1328	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, pMemLnxToMap->Core.cb);
1329	if (pMemLnx)
1330	{
1331	if (pMemLnxToMap->cPages)
1332	{
1333	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1334	/*
1335	* Use vmap - 2.4.22 and later.
1336	*/
1337	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, true /* kernel */);
1338	# ifdef VM_MAP
1339	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[0], pMemLnxToMap->cPages, VM_MAP, fPg);
1340	# else
1341	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[0], pMemLnxToMap->cPages, VM_ALLOC, fPg);
1342	# endif
1343	if (pMemLnx->Core.pv)
1344	{
1345	pMemLnx->fMappedToRing0 = true;
1346	rc = VINF_SUCCESS;
1347	}
1348	else
1349	rc = VERR_MAP_FAILED;
1350
1351	#else /* < 2.4.22 */
1352	/*
1353	* Only option here is to share mappings if possible and forget about fProt.
1354	*/
1355	if (rtR0MemObjIsRing3(pMemToMap))
1356	rc = VERR_NOT_SUPPORTED;
1357	else
1358	{
1359	rc = VINF_SUCCESS;
1360	if (!pMemLnxToMap->Core.pv)
1361	rc = rtR0MemObjLinuxVMap(pMemLnxToMap, !!(fProt & RTMEM_PROT_EXEC));
1362	if (RT_SUCCESS(rc))
1363	{
1364	Assert(pMemLnxToMap->Core.pv);
1365	pMemLnx->Core.pv = pMemLnxToMap->Core.pv;
1366	}
1367	}
1368	#endif
1369	}
1370	else
1371	{
1372	/*
1373	* MMIO / physical memory.
1374	*/
1375	Assert(pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS && !pMemLnxToMap->Core.u.Phys.fAllocated);
1376	pMemLnx->Core.pv = pMemLnxToMap->Core.u.Phys.uCachePolicy == RTMEM_CACHE_POLICY_MMIO
1377	? ioremap_nocache(pMemLnxToMap->Core.u.Phys.PhysBase, pMemLnxToMap->Core.cb)
1378	: ioremap(pMemLnxToMap->Core.u.Phys.PhysBase, pMemLnxToMap->Core.cb);
1379	if (pMemLnx->Core.pv)
1380	{
1381	/** @todo fix protection. */
1382	rc = VINF_SUCCESS;
1383	}
1384	}
1385	if (RT_SUCCESS(rc))
1386	{
1387	pMemLnx->Core.u.Mapping.R0Process = NIL_RTR0PROCESS;
1388	*ppMem = &pMemLnx->Core;
1389	IPRT_LINUX_RESTORE_EFL_AC();
1390	return VINF_SUCCESS;
1391	}
1392	rtR0MemObjDelete(&pMemLnx->Core);
1393	}
1394
1395	IPRT_LINUX_RESTORE_EFL_AC();
1396	return rc;
1397	}
1398
1399
1400	#ifdef VBOX_USE_PAE_HACK
1401	/**
1402	* Replace the PFN of a PTE with the address of the actual page.
1403	*
1404	* The caller maps a reserved dummy page at the address with the desired access
1405	* and flags.
1406	*
1407	* This hack is required for older Linux kernels which don't provide
1408	* remap_pfn_range().
1409	*
1410	* @returns 0 on success, -ENOMEM on failure.
1411	* @param mm The memory context.
1412	* @param ulAddr The mapping address.
1413	* @param Phys The physical address of the page to map.
1414	*/
1415	static int rtR0MemObjLinuxFixPte(struct mm_struct *mm, unsigned long ulAddr, RTHCPHYS Phys)
1416	{
1417	int rc = -ENOMEM;
1418	pgd_t *pgd;
1419
1420	spin_lock(&mm->page_table_lock);
1421
1422	pgd = pgd_offset(mm, ulAddr);
1423	if (!pgd_none(pgd) && !pgd_bad(pgd))
1424	{
1425	pmd_t *pmd = pmd_offset(pgd, ulAddr);
1426	if (!pmd_none(*pmd))
1427	{
1428	pte_t *ptep = pte_offset_map(pmd, ulAddr);
1429	if (ptep)
1430	{
1431	pte_t pte = *ptep;
1432	pte.pte_high &= 0xfff00000;
1433	pte.pte_high \|= ((Phys >> 32) & 0x000fffff);
1434	pte.pte_low &= 0x00000fff;
1435	pte.pte_low \|= (Phys & 0xfffff000);
1436	set_pte(ptep, pte);
1437	pte_unmap(ptep);
1438	rc = 0;
1439	}
1440	}
1441	}
1442
1443	spin_unlock(&mm->page_table_lock);
1444	return rc;
1445	}
1446	#endif /* VBOX_USE_PAE_HACK */
1447
1448
1449	DECLHIDDEN(int) rtR0MemObjNativeMapUser(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap, RTR3PTR R3PtrFixed,
1450	size_t uAlignment, unsigned fProt, RTR0PROCESS R0Process)
1451	{
1452	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1453	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1454	int rc = VERR_NO_MEMORY;
1455	PRTR0MEMOBJLNX pMemLnx;
1456	#ifdef VBOX_USE_PAE_HACK
1457	struct page *pDummyPage;
1458	RTHCPHYS DummyPhys;
1459	#endif
1460	IPRT_LINUX_SAVE_EFL_AC();
1461
1462	/*
1463	* Check for restrictions.
1464	*/
1465	if (!pTask)
1466	return VERR_NOT_SUPPORTED;
1467	if (uAlignment > PAGE_SIZE)
1468	return VERR_NOT_SUPPORTED;
1469
1470	#ifdef VBOX_USE_PAE_HACK
1471	/*
1472	* Allocate a dummy page for use when mapping the memory.
1473	*/
1474	pDummyPage = alloc_page(GFP_USER \| __GFP_NOWARN);
1475	if (!pDummyPage)
1476	{
1477	IPRT_LINUX_RESTORE_EFL_AC();
1478	return VERR_NO_MEMORY;
1479	}
1480	SetPageReserved(pDummyPage);
1481	DummyPhys = page_to_phys(pDummyPage);
1482	#endif
1483
1484	/*
1485	* Create the IPRT memory object.
1486	*/
1487	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, pMemLnxToMap->Core.cb);
1488	if (pMemLnx)
1489	{
1490	/*
1491	* Allocate user space mapping.
1492	*/
1493	void *pv;
1494	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, pMemLnxToMap->Core.cb, uAlignment, pTask, fProt);
1495	if (pv != (void *)-1)
1496	{
1497	/*
1498	* Map page by page into the mmap area.
1499	* This is generic, paranoid and not very efficient.
1500	*/
1501	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, false /* user */);
1502	unsigned long ulAddrCur = (unsigned long)pv;
1503	const size_t cPages = pMemLnxToMap->Core.cb >> PAGE_SHIFT;
1504	size_t iPage;
1505
1506	down_write(&pTask->mm->mmap_sem);
1507
1508	rc = VINF_SUCCESS;
1509	if (pMemLnxToMap->cPages)
1510	{
1511	for (iPage = 0; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE)
1512	{
1513	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11)
1514	RTHCPHYS Phys = page_to_phys(pMemLnxToMap->apPages[iPage]);
1515	#endif
1516	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1517	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
1518	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
1519	#endif
1520	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0) && defined(RT_ARCH_X86)
1521	/* remap_page_range() limitation on x86 */
1522	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
1523	#endif
1524
1525	#if defined(VBOX_USE_INSERT_PAGE) && LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1526	rc = vm_insert_page(vma, ulAddrCur, pMemLnxToMap->apPages[iPage]);
1527	/* Thes flags help making 100% sure some bad stuff wont happen (swap, core, ++).
1528	* See remap_pfn_range() in mm/memory.c */
1529	#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 7, 0)
1530	vma->vm_flags \|= VM_DONTEXPAND \| VM_DONTDUMP;
1531	#else
1532	vma->vm_flags \|= VM_RESERVED;
1533	#endif
1534	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
1535	rc = remap_pfn_range(vma, ulAddrCur, page_to_pfn(pMemLnxToMap->apPages[iPage]), PAGE_SIZE, fPg);
1536	#elif defined(VBOX_USE_PAE_HACK)
1537	rc = remap_page_range(vma, ulAddrCur, DummyPhys, PAGE_SIZE, fPg);
1538	if (!rc)
1539	rc = rtR0MemObjLinuxFixPte(pTask->mm, ulAddrCur, Phys);
1540	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1541	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1542	#else /* 2.4 */
1543	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
1544	#endif
1545	if (rc)
1546	{
1547	rc = VERR_NO_MEMORY;
1548	break;
1549	}
1550	}
1551	}
1552	else
1553	{
1554	RTHCPHYS Phys;
1555	if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS)
1556	Phys = pMemLnxToMap->Core.u.Phys.PhysBase;
1557	else if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_CONT)
1558	Phys = pMemLnxToMap->Core.u.Cont.Phys;
1559	else
1560	{
1561	AssertMsgFailed(("%d\n", pMemLnxToMap->Core.enmType));
1562	Phys = NIL_RTHCPHYS;
1563	}
1564	if (Phys != NIL_RTHCPHYS)
1565	{
1566	for (iPage = 0; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE, Phys += PAGE_SIZE)
1567	{
1568	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1569	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
1570	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
1571	#endif
1572	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0) && defined(RT_ARCH_X86)
1573	/* remap_page_range() limitation on x86 */
1574	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
1575	#endif
1576
1577	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
1578	rc = remap_pfn_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1579	#elif defined(VBOX_USE_PAE_HACK)
1580	rc = remap_page_range(vma, ulAddrCur, DummyPhys, PAGE_SIZE, fPg);
1581	if (!rc)
1582	rc = rtR0MemObjLinuxFixPte(pTask->mm, ulAddrCur, Phys);
1583	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1584	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1585	#else /* 2.4 */
1586	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
1587	#endif
1588	if (rc)
1589	{
1590	rc = VERR_NO_MEMORY;
1591	break;
1592	}
1593	}
1594	}
1595	}
1596
1597	#ifdef CONFIG_NUMA_BALANCING
1598	# if LINUX_VERSION_CODE < KERNEL_VERSION(3, 13, 0)
1599	# ifdef RHEL_RELEASE_CODE
1600	# if RHEL_RELEASE_CODE < RHEL_RELEASE_VERSION(7, 0)
1601	# define VBOX_NUMA_HACK_OLD
1602	# endif
1603	# endif
1604	# endif
1605	if (RT_SUCCESS(rc))
1606	{
1607	/** @todo Ugly hack! But right now we have no other means to
1608	* disable automatic NUMA page balancing. */
1609	# ifdef RT_OS_X86
1610	# ifdef VBOX_NUMA_HACK_OLD
1611	pTask->mm->numa_next_reset = jiffies + 0x7fffffffUL;
1612	# endif
1613	pTask->mm->numa_next_scan = jiffies + 0x7fffffffUL;
1614	# else
1615	# ifdef VBOX_NUMA_HACK_OLD
1616	pTask->mm->numa_next_reset = jiffies + 0x7fffffffffffffffUL;
1617	# endif
1618	pTask->mm->numa_next_scan = jiffies + 0x7fffffffffffffffUL;
1619	# endif
1620	}
1621	#endif /* CONFIG_NUMA_BALANCING */
1622
1623	up_write(&pTask->mm->mmap_sem);
1624
1625	if (RT_SUCCESS(rc))
1626	{
1627	#ifdef VBOX_USE_PAE_HACK
1628	__free_page(pDummyPage);
1629	#endif
1630	pMemLnx->Core.pv = pv;
1631	pMemLnx->Core.u.Mapping.R0Process = R0Process;
1632	*ppMem = &pMemLnx->Core;
1633	IPRT_LINUX_RESTORE_EFL_AC();
1634	return VINF_SUCCESS;
1635	}
1636
1637	/*
1638	* Bail out.
1639	*/
1640	rtR0MemObjLinuxDoMunmap(pv, pMemLnxToMap->Core.cb, pTask);
1641	}
1642	rtR0MemObjDelete(&pMemLnx->Core);
1643	}
1644	#ifdef VBOX_USE_PAE_HACK
1645	__free_page(pDummyPage);
1646	#endif
1647
1648	IPRT_LINUX_RESTORE_EFL_AC();
1649	return rc;
1650	}
1651
1652
1653	DECLHIDDEN(int) rtR0MemObjNativeProtect(PRTR0MEMOBJINTERNAL pMem, size_t offSub, size_t cbSub, uint32_t fProt)
1654	{
1655	NOREF(pMem);
1656	NOREF(offSub);
1657	NOREF(cbSub);
1658	NOREF(fProt);
1659	return VERR_NOT_SUPPORTED;
1660	}
1661
1662
1663	DECLHIDDEN(RTHCPHYS) rtR0MemObjNativeGetPagePhysAddr(PRTR0MEMOBJINTERNAL pMem, size_t iPage)
1664	{
1665	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
1666
1667	if (pMemLnx->cPages)
1668	return page_to_phys(pMemLnx->apPages[iPage]);
1669
1670	switch (pMemLnx->Core.enmType)
1671	{
1672	case RTR0MEMOBJTYPE_CONT:
1673	return pMemLnx->Core.u.Cont.Phys + (iPage << PAGE_SHIFT);
1674
1675	case RTR0MEMOBJTYPE_PHYS:
1676	return pMemLnx->Core.u.Phys.PhysBase + (iPage << PAGE_SHIFT);
1677
1678	/* the parent knows */
1679	case RTR0MEMOBJTYPE_MAPPING:
1680	return rtR0MemObjNativeGetPagePhysAddr(pMemLnx->Core.uRel.Child.pParent, iPage);
1681
1682	/* cPages > 0 */
1683	case RTR0MEMOBJTYPE_LOW:
1684	case RTR0MEMOBJTYPE_LOCK:
1685	case RTR0MEMOBJTYPE_PHYS_NC:
1686	case RTR0MEMOBJTYPE_PAGE:
1687	default:
1688	AssertMsgFailed(("%d\n", pMemLnx->Core.enmType));
1689	/* fall thru */
1690
1691	case RTR0MEMOBJTYPE_RES_VIRT:
1692	return NIL_RTHCPHYS;
1693	}
1694	}
1695

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

source: vbox/trunk/src/VBox/Runtime/r0drv/linux/memobj-r0drv-linux.c@ 60453

Download in other formats: