memobj-r0drv-linux.c@ 81587

Last change on this file since 81587 was 81587, checked in by vboxsync, 5 years ago
IPRT/memobj-r0drv-linux.c: Don't mark allocations as executable unless requested (linux < 2.4.22 only). Made the MY_SET_PAGES_EXEC macros be undefined on 5.4.0 and later since they should not be engaged there. (untested) ticketref:18945
Property svn:eol-style set to `native` Property svn:keywords set to `Id Rev Revision`
File size: 60.0 KB

Line
1	/* $Id: memobj-r0drv-linux.c 81587 2019-10-30 13:07:15Z vboxsync $ */
2	/** @file
3	* IPRT - Ring-0 Memory Objects, Linux.
4	*/
5
6	/*
7	* Copyright (C) 2006-2019 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/assert.h>
35	#include <iprt/err.h>
36	#include <iprt/log.h>
37	#include <iprt/mem.h>
38	#include <iprt/process.h>
39	#include <iprt/string.h>
40	#include "internal/memobj.h"
41	#include "internal/iprt.h"
42
43
44	/*********************************************************************************************************************************
45	* Defined Constants And Macros *
46	*********************************************************************************************************************************/
47	/* early 2.6 kernels */
48	#ifndef PAGE_SHARED_EXEC
49	# define PAGE_SHARED_EXEC PAGE_SHARED
50	#endif
51	#ifndef PAGE_READONLY_EXEC
52	# define PAGE_READONLY_EXEC PAGE_READONLY
53	#endif
54
55	/*
56	* 2.6.29+ kernels don't work with remap_pfn_range() anymore because
57	* track_pfn_vma_new() is apparently not defined for non-RAM pages.
58	* It should be safe to use vm_insert_page() older kernels as well.
59	*/
60	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 23)
61	# define VBOX_USE_INSERT_PAGE
62	#endif
63	#if defined(CONFIG_X86_PAE) \
64	&& ( defined(HAVE_26_STYLE_REMAP_PAGE_RANGE) \
65	\|\| ( LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \
66	&& LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11)))
67	# define VBOX_USE_PAE_HACK
68	#endif
69
70	/* gfp_t was introduced in 2.6.14, define it for earlier. */
71	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 14)
72	# define gfp_t unsigned
73	#endif
74
75
76	/*********************************************************************************************************************************
77	* Structures and Typedefs *
78	*********************************************************************************************************************************/
79	/**
80	* The Darwin version of the memory object structure.
81	*/
82	typedef struct RTR0MEMOBJLNX
83	{
84	/** The core structure. */
85	RTR0MEMOBJINTERNAL Core;
86	/** Set if the allocation is contiguous.
87	* This means it has to be given back as one chunk. */
88	bool fContiguous;
89	/** Set if executable allocation. */
90	bool fExecutable;
91	/** Set if we've vmap'ed the memory into ring-0. */
92	bool fMappedToRing0;
93	/** The pages in the apPages array. */
94	size_t cPages;
95	/** Array of struct page pointers. (variable size) */
96	struct page *apPages[1];
97	} RTR0MEMOBJLNX, *PRTR0MEMOBJLNX;
98
99
100	static void rtR0MemObjLinuxFreePages(PRTR0MEMOBJLNX pMemLnx);
101
102
103	/**
104	* Helper that converts from a RTR0PROCESS handle to a linux task.
105	*
106	* @returns The corresponding Linux task.
107	* @param R0Process IPRT ring-0 process handle.
108	*/
109	static struct task_struct *rtR0ProcessToLinuxTask(RTR0PROCESS R0Process)
110	{
111	/** @todo fix rtR0ProcessToLinuxTask!! */
112	/** @todo many (all?) callers currently assume that we return 'current'! */
113	return R0Process == RTR0ProcHandleSelf() ? current : NULL;
114	}
115
116
117	/**
118	* Compute order. Some functions allocate 2^order pages.
119	*
120	* @returns order.
121	* @param cPages Number of pages.
122	*/
123	static int rtR0MemObjLinuxOrder(size_t cPages)
124	{
125	int iOrder;
126	size_t cTmp;
127
128	for (iOrder = 0, cTmp = cPages; cTmp >>= 1; ++iOrder)
129	;
130	if (cPages & ~((size_t)1 << iOrder))
131	++iOrder;
132
133	return iOrder;
134	}
135
136
137	/**
138	* Converts from RTMEM_PROT_* to Linux PAGE_*.
139	*
140	* @returns Linux page protection constant.
141	* @param fProt The IPRT protection mask.
142	* @param fKernel Whether it applies to kernel or user space.
143	*/
144	static pgprot_t rtR0MemObjLinuxConvertProt(unsigned fProt, bool fKernel)
145	{
146	switch (fProt)
147	{
148	default:
149	AssertMsgFailed(("%#x %d\n", fProt, fKernel)); RT_FALL_THRU();
150	case RTMEM_PROT_NONE:
151	return PAGE_NONE;
152
153	case RTMEM_PROT_READ:
154	return fKernel ? PAGE_KERNEL_RO : PAGE_READONLY;
155
156	case RTMEM_PROT_WRITE:
157	case RTMEM_PROT_WRITE \| RTMEM_PROT_READ:
158	return fKernel ? PAGE_KERNEL : PAGE_SHARED;
159
160	case RTMEM_PROT_EXEC:
161	case RTMEM_PROT_EXEC \| RTMEM_PROT_READ:
162	#if defined(RT_ARCH_X86) \|\| defined(RT_ARCH_AMD64)
163	if (fKernel)
164	{
165	pgprot_t fPg = MY_PAGE_KERNEL_EXEC;
166	pgprot_val(fPg) &= ~_PAGE_RW;
167	return fPg;
168	}
169	return PAGE_READONLY_EXEC;
170	#else
171	return fKernel ? MY_PAGE_KERNEL_EXEC : PAGE_READONLY_EXEC;
172	#endif
173
174	case RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC:
175	case RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC \| RTMEM_PROT_READ:
176	return fKernel ? MY_PAGE_KERNEL_EXEC : PAGE_SHARED_EXEC;
177	}
178	}
179
180
181	/**
182	* Worker for rtR0MemObjNativeReserveUser and rtR0MemObjNativerMapUser that creates
183	* an empty user space mapping.
184	*
185	* We acquire the mmap_sem of the task!
186	*
187	* @returns Pointer to the mapping.
188	* (void *)-1 on failure.
189	* @param R3PtrFixed (RTR3PTR)-1 if anywhere, otherwise a specific location.
190	* @param cb The size of the mapping.
191	* @param uAlignment The alignment of the mapping.
192	* @param pTask The Linux task to create this mapping in.
193	* @param fProt The RTMEM_PROT_* mask.
194	*/
195	static void rtR0MemObjLinuxDoMmap(RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment, struct task_struct pTask, unsigned fProt)
196	{
197	unsigned fLnxProt;
198	unsigned long ulAddr;
199
200	Assert(pTask == current); /* do_mmap */
201	RT_NOREF_PV(pTask);
202
203	/*
204	* Convert from IPRT protection to mman.h PROT_ and call do_mmap.
205	*/
206	fProt &= (RTMEM_PROT_NONE \| RTMEM_PROT_READ \| RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC);
207	if (fProt == RTMEM_PROT_NONE)
208	fLnxProt = PROT_NONE;
209	else
210	{
211	fLnxProt = 0;
212	if (fProt & RTMEM_PROT_READ)
213	fLnxProt \|= PROT_READ;
214	if (fProt & RTMEM_PROT_WRITE)
215	fLnxProt \|= PROT_WRITE;
216	if (fProt & RTMEM_PROT_EXEC)
217	fLnxProt \|= PROT_EXEC;
218	}
219
220	if (R3PtrFixed != (RTR3PTR)-1)
221	{
222	#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 5, 0)
223	ulAddr = vm_mmap(NULL, R3PtrFixed, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS \| MAP_FIXED, 0);
224	#else
225	down_write(&pTask->mm->mmap_sem);
226	ulAddr = do_mmap(NULL, R3PtrFixed, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS \| MAP_FIXED, 0);
227	up_write(&pTask->mm->mmap_sem);
228	#endif
229	}
230	else
231	{
232	#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 5, 0)
233	ulAddr = vm_mmap(NULL, 0, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS, 0);
234	#else
235	down_write(&pTask->mm->mmap_sem);
236	ulAddr = do_mmap(NULL, 0, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS, 0);
237	up_write(&pTask->mm->mmap_sem);
238	#endif
239	if ( !(ulAddr & ~PAGE_MASK)
240	&& (ulAddr & (uAlignment - 1)))
241	{
242	/** @todo implement uAlignment properly... We'll probably need to make some dummy mappings to fill
243	* up alignment gaps. This is of course complicated by fragmentation (which we might have cause
244	* ourselves) and further by there begin two mmap strategies (top / bottom). */
245	/* For now, just ignore uAlignment requirements... */
246	}
247	}
248
249
250	if (ulAddr & ~PAGE_MASK) /* ~PAGE_MASK == PAGE_OFFSET_MASK */
251	return (void *)-1;
252	return (void *)ulAddr;
253	}
254
255
256	/**
257	* Worker that destroys a user space mapping.
258	* Undoes what rtR0MemObjLinuxDoMmap did.
259	*
260	* We acquire the mmap_sem of the task!
261	*
262	* @param pv The ring-3 mapping.
263	* @param cb The size of the mapping.
264	* @param pTask The Linux task to destroy this mapping in.
265	*/
266	static void rtR0MemObjLinuxDoMunmap(void pv, size_t cb, struct task_struct pTask)
267	{
268	#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 5, 0)
269	Assert(pTask == current); RT_NOREF_PV(pTask);
270	vm_munmap((unsigned long)pv, cb);
271	#elif defined(USE_RHEL4_MUNMAP)
272	down_write(&pTask->mm->mmap_sem);
273	do_munmap(pTask->mm, (unsigned long)pv, cb, 0); /* should it be 1 or 0? */
274	up_write(&pTask->mm->mmap_sem);
275	#else
276	down_write(&pTask->mm->mmap_sem);
277	do_munmap(pTask->mm, (unsigned long)pv, cb);
278	up_write(&pTask->mm->mmap_sem);
279	#endif
280	}
281
282
283	/**
284	* Internal worker that allocates physical pages and creates the memory object for them.
285	*
286	* @returns IPRT status code.
287	* @param ppMemLnx Where to store the memory object pointer.
288	* @param enmType The object type.
289	* @param cb The number of bytes to allocate.
290	* @param uAlignment The alignment of the physical memory.
291	* Only valid if fContiguous == true, ignored otherwise.
292	* @param fFlagsLnx The page allocation flags (GPFs).
293	* @param fContiguous Whether the allocation must be contiguous.
294	* @param fExecutable Whether the memory must be executable.
295	* @param rcNoMem What to return when we're out of pages.
296	*/
297	static int rtR0MemObjLinuxAllocPages(PRTR0MEMOBJLNX *ppMemLnx, RTR0MEMOBJTYPE enmType, size_t cb,
298	size_t uAlignment, gfp_t fFlagsLnx, bool fContiguous, bool fExecutable, int rcNoMem)
299	{
300	size_t iPage;
301	size_t const cPages = cb >> PAGE_SHIFT;
302	struct page *paPages;
303
304	/*
305	* Allocate a memory object structure that's large enough to contain
306	* the page pointer array.
307	*/
308	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_UOFFSETOF_DYN(RTR0MEMOBJLNX, apPages[cPages]), enmType, NULL, cb);
309	if (!pMemLnx)
310	return VERR_NO_MEMORY;
311	pMemLnx->cPages = cPages;
312
313	if (cPages > 255)
314	{
315	# ifdef __GFP_REPEAT
316	/* Try hard to allocate the memory, but the allocation attempt might fail. */
317	fFlagsLnx \|= __GFP_REPEAT;
318	# endif
319	# ifdef __GFP_NOMEMALLOC
320	/* Introduced with Linux 2.6.12: Don't use emergency reserves */
321	fFlagsLnx \|= __GFP_NOMEMALLOC;
322	# endif
323	}
324
325	/*
326	* Allocate the pages.
327	* For small allocations we'll try contiguous first and then fall back on page by page.
328	*/
329	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
330	if ( fContiguous
331	\|\| cb <= PAGE_SIZE * 2)
332	{
333	# ifdef VBOX_USE_INSERT_PAGE
334	paPages = alloc_pages(fFlagsLnx \| __GFP_COMP \| __GFP_NOWARN, rtR0MemObjLinuxOrder(cPages));
335	# else
336	paPages = alloc_pages(fFlagsLnx \| __GFP_NOWARN, rtR0MemObjLinuxOrder(cPages));
337	# endif
338	if (paPages)
339	{
340	fContiguous = true;
341	for (iPage = 0; iPage < cPages; iPage++)
342	pMemLnx->apPages[iPage] = &paPages[iPage];
343	}
344	else if (fContiguous)
345	{
346	rtR0MemObjDelete(&pMemLnx->Core);
347	return rcNoMem;
348	}
349	}
350
351	if (!fContiguous)
352	{
353	for (iPage = 0; iPage < cPages; iPage++)
354	{
355	pMemLnx->apPages[iPage] = alloc_page(fFlagsLnx \| __GFP_NOWARN);
356	if (RT_UNLIKELY(!pMemLnx->apPages[iPage]))
357	{
358	while (iPage-- > 0)
359	__free_page(pMemLnx->apPages[iPage]);
360	rtR0MemObjDelete(&pMemLnx->Core);
361	return rcNoMem;
362	}
363	}
364	}
365
366	#else /* < 2.4.22 */
367	/** @todo figure out why we didn't allocate page-by-page on 2.4.21 and older... */
368	paPages = alloc_pages(fFlagsLnx, rtR0MemObjLinuxOrder(cPages));
369	if (!paPages)
370	{
371	rtR0MemObjDelete(&pMemLnx->Core);
372	return rcNoMem;
373	}
374	for (iPage = 0; iPage < cPages; iPage++)
375	{
376	pMemLnx->apPages[iPage] = &paPages[iPage];
377	if (fExecutable)
378	MY_SET_PAGES_EXEC(pMemLnx->apPages[iPage], 1);
379	if (PageHighMem(pMemLnx->apPages[iPage]))
380	BUG();
381	}
382
383	fContiguous = true;
384	#endif /* < 2.4.22 */
385	pMemLnx->fContiguous = fContiguous;
386	pMemLnx->fExecutable = fExecutable;
387
388	#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 5, 0)
389	/*
390	* Reserve the pages.
391	*
392	* Linux >= 4.5 with CONFIG_DEBUG_VM panics when setting PG_reserved on compound
393	* pages. According to Michal Hocko this shouldn't be necessary anyway because
394	* as pages which are not on the LRU list are never evictable.
395	*/
396	for (iPage = 0; iPage < cPages; iPage++)
397	SetPageReserved(pMemLnx->apPages[iPage]);
398	#endif
399
400	/*
401	* Note that the physical address of memory allocated with alloc_pages(flags, order)
402	* is always 2^(PAGE_SHIFT+order)-aligned.
403	*/
404	if ( fContiguous
405	&& uAlignment > PAGE_SIZE)
406	{
407	/*
408	* Check for alignment constraints. The physical address of memory allocated with
409	* alloc_pages(flags, order) is always 2^(PAGE_SHIFT+order)-aligned.
410	*/
411	if (RT_UNLIKELY(page_to_phys(pMemLnx->apPages[0]) & (uAlignment - 1)))
412	{
413	/*
414	* This should never happen!
415	*/
416	printk("rtR0MemObjLinuxAllocPages(cb=0x%lx, uAlignment=0x%lx): alloc_pages(..., %d) returned physical memory at 0x%lx!\n",
417	(unsigned long)cb, (unsigned long)uAlignment, rtR0MemObjLinuxOrder(cPages), (unsigned long)page_to_phys(pMemLnx->apPages[0]));
418	rtR0MemObjLinuxFreePages(pMemLnx);
419	return rcNoMem;
420	}
421	}
422
423	*ppMemLnx = pMemLnx;
424	return VINF_SUCCESS;
425	}
426
427
428	/**
429	* Frees the physical pages allocated by the rtR0MemObjLinuxAllocPages() call.
430	*
431	* This method does NOT free the object.
432	*
433	* @param pMemLnx The object which physical pages should be freed.
434	*/
435	static void rtR0MemObjLinuxFreePages(PRTR0MEMOBJLNX pMemLnx)
436	{
437	size_t iPage = pMemLnx->cPages;
438	if (iPage > 0)
439	{
440	/*
441	* Restore the page flags.
442	*/
443	while (iPage-- > 0)
444	{
445	#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 5, 0)
446	/* See SetPageReserved() in rtR0MemObjLinuxAllocPages() */
447	ClearPageReserved(pMemLnx->apPages[iPage]);
448	#endif
449	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 4, 22)
450	if (pMemLnx->fExecutable)
451	MY_SET_PAGES_NOEXEC(pMemLnx->apPages[iPage], 1);
452	#endif
453	}
454
455	/*
456	* Free the pages.
457	*/
458	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
459	if (!pMemLnx->fContiguous)
460	{
461	iPage = pMemLnx->cPages;
462	while (iPage-- > 0)
463	__free_page(pMemLnx->apPages[iPage]);
464	}
465	else
466	#endif
467	__free_pages(pMemLnx->apPages[0], rtR0MemObjLinuxOrder(pMemLnx->cPages));
468
469	pMemLnx->cPages = 0;
470	}
471	}
472
473
474	/**
475	* Maps the allocation into ring-0.
476	*
477	* This will update the RTR0MEMOBJLNX::Core.pv and RTR0MEMOBJ::fMappedToRing0 members.
478	*
479	* Contiguous mappings that isn't in 'high' memory will already be mapped into kernel
480	* space, so we'll use that mapping if possible. If execute access is required, we'll
481	* play safe and do our own mapping.
482	*
483	* @returns IPRT status code.
484	* @param pMemLnx The linux memory object to map.
485	* @param fExecutable Whether execute access is required.
486	*/
487	static int rtR0MemObjLinuxVMap(PRTR0MEMOBJLNX pMemLnx, bool fExecutable)
488	{
489	int rc = VINF_SUCCESS;
490
491	/*
492	* Choose mapping strategy.
493	*/
494	bool fMustMap = fExecutable
495	\|\| !pMemLnx->fContiguous;
496	if (!fMustMap)
497	{
498	size_t iPage = pMemLnx->cPages;
499	while (iPage-- > 0)
500	if (PageHighMem(pMemLnx->apPages[iPage]))
501	{
502	fMustMap = true;
503	break;
504	}
505	}
506
507	Assert(!pMemLnx->Core.pv);
508	Assert(!pMemLnx->fMappedToRing0);
509
510	if (fMustMap)
511	{
512	/*
513	* Use vmap - 2.4.22 and later.
514	*/
515	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
516	pgprot_t fPg;
517	pgprot_val(fPg) = _PAGE_PRESENT \| _PAGE_RW;
518	# ifdef _PAGE_NX
519	if (!fExecutable)
520	pgprot_val(fPg) \|= _PAGE_NX;
521	# endif
522
523	# ifdef VM_MAP
524	pMemLnx->Core.pv = vmap(&pMemLnx->apPages[0], pMemLnx->cPages, VM_MAP, fPg);
525	# else
526	pMemLnx->Core.pv = vmap(&pMemLnx->apPages[0], pMemLnx->cPages, VM_ALLOC, fPg);
527	# endif
528	if (pMemLnx->Core.pv)
529	pMemLnx->fMappedToRing0 = true;
530	else
531	rc = VERR_MAP_FAILED;
532	#else /* < 2.4.22 */
533	rc = VERR_NOT_SUPPORTED;
534	#endif
535	}
536	else
537	{
538	/*
539	* Use the kernel RAM mapping.
540	*/
541	pMemLnx->Core.pv = phys_to_virt(page_to_phys(pMemLnx->apPages[0]));
542	Assert(pMemLnx->Core.pv);
543	}
544
545	return rc;
546	}
547
548
549	/**
550	* Undoes what rtR0MemObjLinuxVMap() did.
551	*
552	* @param pMemLnx The linux memory object.
553	*/
554	static void rtR0MemObjLinuxVUnmap(PRTR0MEMOBJLNX pMemLnx)
555	{
556	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
557	if (pMemLnx->fMappedToRing0)
558	{
559	Assert(pMemLnx->Core.pv);
560	vunmap(pMemLnx->Core.pv);
561	pMemLnx->fMappedToRing0 = false;
562	}
563	#else /* < 2.4.22 */
564	Assert(!pMemLnx->fMappedToRing0);
565	#endif
566	pMemLnx->Core.pv = NULL;
567	}
568
569
570	DECLHIDDEN(int) rtR0MemObjNativeFree(RTR0MEMOBJ pMem)
571	{
572	IPRT_LINUX_SAVE_EFL_AC();
573	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
574
575	/*
576	* Release any memory that we've allocated or locked.
577	*/
578	switch (pMemLnx->Core.enmType)
579	{
580	case RTR0MEMOBJTYPE_LOW:
581	case RTR0MEMOBJTYPE_PAGE:
582	case RTR0MEMOBJTYPE_CONT:
583	case RTR0MEMOBJTYPE_PHYS:
584	case RTR0MEMOBJTYPE_PHYS_NC:
585	rtR0MemObjLinuxVUnmap(pMemLnx);
586	rtR0MemObjLinuxFreePages(pMemLnx);
587	break;
588
589	case RTR0MEMOBJTYPE_LOCK:
590	if (pMemLnx->Core.u.Lock.R0Process != NIL_RTR0PROCESS)
591	{
592	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
593	size_t iPage;
594	Assert(pTask);
595	if (pTask && pTask->mm)
596	down_read(&pTask->mm->mmap_sem);
597
598	iPage = pMemLnx->cPages;
599	while (iPage-- > 0)
600	{
601	if (!PageReserved(pMemLnx->apPages[iPage]))
602	SetPageDirty(pMemLnx->apPages[iPage]);
603	#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 6, 0)
604	put_page(pMemLnx->apPages[iPage]);
605	#else
606	page_cache_release(pMemLnx->apPages[iPage]);
607	#endif
608	}
609
610	if (pTask && pTask->mm)
611	up_read(&pTask->mm->mmap_sem);
612	}
613	/* else: kernel memory - nothing to do here. */
614	break;
615
616	case RTR0MEMOBJTYPE_RES_VIRT:
617	Assert(pMemLnx->Core.pv);
618	if (pMemLnx->Core.u.ResVirt.R0Process != NIL_RTR0PROCESS)
619	{
620	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
621	Assert(pTask);
622	if (pTask && pTask->mm)
623	rtR0MemObjLinuxDoMunmap(pMemLnx->Core.pv, pMemLnx->Core.cb, pTask);
624	}
625	else
626	{
627	vunmap(pMemLnx->Core.pv);
628
629	Assert(pMemLnx->cPages == 1 && pMemLnx->apPages[0] != NULL);
630	__free_page(pMemLnx->apPages[0]);
631	pMemLnx->apPages[0] = NULL;
632	pMemLnx->cPages = 0;
633	}
634	pMemLnx->Core.pv = NULL;
635	break;
636
637	case RTR0MEMOBJTYPE_MAPPING:
638	Assert(pMemLnx->cPages == 0); Assert(pMemLnx->Core.pv);
639	if (pMemLnx->Core.u.ResVirt.R0Process != NIL_RTR0PROCESS)
640	{
641	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
642	Assert(pTask);
643	if (pTask && pTask->mm)
644	rtR0MemObjLinuxDoMunmap(pMemLnx->Core.pv, pMemLnx->Core.cb, pTask);
645	}
646	else
647	vunmap(pMemLnx->Core.pv);
648	pMemLnx->Core.pv = NULL;
649	break;
650
651	default:
652	AssertMsgFailed(("enmType=%d\n", pMemLnx->Core.enmType));
653	return VERR_INTERNAL_ERROR;
654	}
655	IPRT_LINUX_RESTORE_EFL_ONLY_AC();
656	return VINF_SUCCESS;
657	}
658
659
660	DECLHIDDEN(int) rtR0MemObjNativeAllocPage(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable)
661	{
662	IPRT_LINUX_SAVE_EFL_AC();
663	PRTR0MEMOBJLNX pMemLnx;
664	int rc;
665
666	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
667	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_PAGE, cb, PAGE_SIZE, GFP_HIGHUSER,
668	false /* non-contiguous */, fExecutable, VERR_NO_MEMORY);
669	#else
670	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_PAGE, cb, PAGE_SIZE, GFP_USER,
671	false /* non-contiguous */, fExecutable, VERR_NO_MEMORY);
672	#endif
673	if (RT_SUCCESS(rc))
674	{
675	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
676	if (RT_SUCCESS(rc))
677	{
678	*ppMem = &pMemLnx->Core;
679	IPRT_LINUX_RESTORE_EFL_AC();
680	return rc;
681	}
682
683	rtR0MemObjLinuxFreePages(pMemLnx);
684	rtR0MemObjDelete(&pMemLnx->Core);
685	}
686
687	IPRT_LINUX_RESTORE_EFL_AC();
688	return rc;
689	}
690
691
692	DECLHIDDEN(int) rtR0MemObjNativeAllocLow(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable)
693	{
694	IPRT_LINUX_SAVE_EFL_AC();
695	PRTR0MEMOBJLNX pMemLnx;
696	int rc;
697
698	/* Try to avoid GFP_DMA. GFM_DMA32 was introduced with Linux 2.6.15. */
699	#if (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
700	/* ZONE_DMA32: 0-4GB */
701	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_DMA32,
702	false /* non-contiguous */, fExecutable, VERR_NO_LOW_MEMORY);
703	if (RT_FAILURE(rc))
704	#endif
705	#ifdef RT_ARCH_AMD64
706	/* ZONE_DMA: 0-16MB */
707	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_DMA,
708	false /* non-contiguous */, fExecutable, VERR_NO_LOW_MEMORY);
709	#else
710	# ifdef CONFIG_X86_PAE
711	# endif
712	/* ZONE_NORMAL: 0-896MB */
713	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_USER,
714	false /* non-contiguous */, fExecutable, VERR_NO_LOW_MEMORY);
715	#endif
716	if (RT_SUCCESS(rc))
717	{
718	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
719	if (RT_SUCCESS(rc))
720	{
721	*ppMem = &pMemLnx->Core;
722	IPRT_LINUX_RESTORE_EFL_AC();
723	return rc;
724	}
725
726	rtR0MemObjLinuxFreePages(pMemLnx);
727	rtR0MemObjDelete(&pMemLnx->Core);
728	}
729
730	IPRT_LINUX_RESTORE_EFL_AC();
731	return rc;
732	}
733
734
735	DECLHIDDEN(int) rtR0MemObjNativeAllocCont(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable)
736	{
737	IPRT_LINUX_SAVE_EFL_AC();
738	PRTR0MEMOBJLNX pMemLnx;
739	int rc;
740
741	#if (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
742	/* ZONE_DMA32: 0-4GB */
743	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, PAGE_SIZE, GFP_DMA32,
744	true /* contiguous */, fExecutable, VERR_NO_CONT_MEMORY);
745	if (RT_FAILURE(rc))
746	#endif
747	#ifdef RT_ARCH_AMD64
748	/* ZONE_DMA: 0-16MB */
749	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, PAGE_SIZE, GFP_DMA,
750	true /* contiguous */, fExecutable, VERR_NO_CONT_MEMORY);
751	#else
752	/* ZONE_NORMAL (32-bit hosts): 0-896MB */
753	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, PAGE_SIZE, GFP_USER,
754	true /* contiguous */, fExecutable, VERR_NO_CONT_MEMORY);
755	#endif
756	if (RT_SUCCESS(rc))
757	{
758	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
759	if (RT_SUCCESS(rc))
760	{
761	#if defined(RT_STRICT) && (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_HIGHMEM64G))
762	size_t iPage = pMemLnx->cPages;
763	while (iPage-- > 0)
764	Assert(page_to_phys(pMemLnx->apPages[iPage]) < _4G);
765	#endif
766	pMemLnx->Core.u.Cont.Phys = page_to_phys(pMemLnx->apPages[0]);
767	*ppMem = &pMemLnx->Core;
768	IPRT_LINUX_RESTORE_EFL_AC();
769	return rc;
770	}
771
772	rtR0MemObjLinuxFreePages(pMemLnx);
773	rtR0MemObjDelete(&pMemLnx->Core);
774	}
775
776	IPRT_LINUX_RESTORE_EFL_AC();
777	return rc;
778	}
779
780
781	/**
782	* Worker for rtR0MemObjLinuxAllocPhysSub that tries one allocation strategy.
783	*
784	* @returns IPRT status code.
785	* @param ppMemLnx Where to
786	* @param enmType The object type.
787	* @param cb The size of the allocation.
788	* @param uAlignment The alignment of the physical memory.
789	* Only valid for fContiguous == true, ignored otherwise.
790	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
791	* @param fGfp The Linux GFP flags to use for the allocation.
792	*/
793	static int rtR0MemObjLinuxAllocPhysSub2(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType,
794	size_t cb, size_t uAlignment, RTHCPHYS PhysHighest, gfp_t fGfp)
795	{
796	PRTR0MEMOBJLNX pMemLnx;
797	int rc;
798
799	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, enmType, cb, uAlignment, fGfp,
800	enmType == RTR0MEMOBJTYPE_PHYS /* contiguous / non-contiguous */,
801	false /fExecutable/, VERR_NO_PHYS_MEMORY);
802	if (RT_FAILURE(rc))
803	return rc;
804
805	/*
806	* Check the addresses if necessary. (Can be optimized a bit for PHYS.)
807	*/
808	if (PhysHighest != NIL_RTHCPHYS)
809	{
810	size_t iPage = pMemLnx->cPages;
811	while (iPage-- > 0)
812	if (page_to_phys(pMemLnx->apPages[iPage]) > PhysHighest)
813	{
814	rtR0MemObjLinuxFreePages(pMemLnx);
815	rtR0MemObjDelete(&pMemLnx->Core);
816	return VERR_NO_MEMORY;
817	}
818	}
819
820	/*
821	* Complete the object.
822	*/
823	if (enmType == RTR0MEMOBJTYPE_PHYS)
824	{
825	pMemLnx->Core.u.Phys.PhysBase = page_to_phys(pMemLnx->apPages[0]);
826	pMemLnx->Core.u.Phys.fAllocated = true;
827	}
828	*ppMem = &pMemLnx->Core;
829	return rc;
830	}
831
832
833	/**
834	* Worker for rtR0MemObjNativeAllocPhys and rtR0MemObjNativeAllocPhysNC.
835	*
836	* @returns IPRT status code.
837	* @param ppMem Where to store the memory object pointer on success.
838	* @param enmType The object type.
839	* @param cb The size of the allocation.
840	* @param uAlignment The alignment of the physical memory.
841	* Only valid for enmType == RTR0MEMOBJTYPE_PHYS, ignored otherwise.
842	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
843	*/
844	static int rtR0MemObjLinuxAllocPhysSub(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType,
845	size_t cb, size_t uAlignment, RTHCPHYS PhysHighest)
846	{
847	int rc;
848	IPRT_LINUX_SAVE_EFL_AC();
849
850	/*
851	* There are two clear cases and that's the <=16MB and anything-goes ones.
852	* When the physical address limit is somewhere in-between those two we'll
853	* just have to try, starting with HIGHUSER and working our way thru the
854	* different types, hoping we'll get lucky.
855	*
856	* We should probably move this physical address restriction logic up to
857	* the page alloc function as it would be more efficient there. But since
858	* we don't expect this to be a performance issue just yet it can wait.
859	*/
860	if (PhysHighest == NIL_RTHCPHYS)
861	/* ZONE_HIGHMEM: the whole physical memory */
862	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_HIGHUSER);
863	else if (PhysHighest <= _1M * 16)
864	/* ZONE_DMA: 0-16MB */
865	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_DMA);
866	else
867	{
868	rc = VERR_NO_MEMORY;
869	if (RT_FAILURE(rc))
870	/* ZONE_HIGHMEM: the whole physical memory */
871	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_HIGHUSER);
872	if (RT_FAILURE(rc))
873	/* ZONE_NORMAL: 0-896MB */
874	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_USER);
875	#ifdef GFP_DMA32
876	if (RT_FAILURE(rc))
877	/* ZONE_DMA32: 0-4GB */
878	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_DMA32);
879	#endif
880	if (RT_FAILURE(rc))
881	/* ZONE_DMA: 0-16MB */
882	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_DMA);
883	}
884	IPRT_LINUX_RESTORE_EFL_AC();
885	return rc;
886	}
887
888
889	/**
890	* Translates a kernel virtual address to a linux page structure by walking the
891	* page tables.
892	*
893	* @note We do assume that the page tables will not change as we are walking
894	* them. This assumption is rather forced by the fact that I could not
895	* immediately see any way of preventing this from happening. So, we
896	* take some extra care when accessing them.
897	*
898	* Because of this, we don't want to use this function on memory where
899	* attribute changes to nearby pages is likely to cause large pages to
900	* be used or split up. So, don't use this for the linear mapping of
901	* physical memory.
902	*
903	* @returns Pointer to the page structur or NULL if it could not be found.
904	* @param pv The kernel virtual address.
905	*/
906	RTDECL(struct page ) rtR0MemObjLinuxVirtToPage(void pv)
907	{
908	unsigned long ulAddr = (unsigned long)pv;
909	unsigned long pfn;
910	struct page *pPage;
911	pte_t *pEntry;
912	union
913	{
914	pgd_t Global;
915	#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 12, 0)
916	p4d_t Four;
917	#endif
918	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
919	pud_t Upper;
920	#endif
921	pmd_t Middle;
922	pte_t Entry;
923	} u;
924
925	/* Should this happen in a situation this code will be called in? And if
926	* so, can it change under our feet? See also
927	* "Documentation/vm/active_mm.txt" in the kernel sources. */
928	if (RT_UNLIKELY(!current->active_mm))
929	return NULL;
930	u.Global = *pgd_offset(current->active_mm, ulAddr);
931	if (RT_UNLIKELY(pgd_none(u.Global)))
932	return NULL;
933	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
934	# if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 12, 0)
935	u.Four = *p4d_offset(&u.Global, ulAddr);
936	if (RT_UNLIKELY(p4d_none(u.Four)))
937	return NULL;
938	if (p4d_large(u.Four))
939	{
940	pPage = p4d_page(u.Four);
941	AssertReturn(pPage, NULL);
942	pfn = page_to_pfn(pPage); /* doing the safe way... */
943	AssertCompile(P4D_SHIFT - PAGE_SHIFT < 31);
944	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (P4D_SHIFT - PAGE_SHIFT)) - 1);
945	return pfn_to_page(pfn);
946	}
947	u.Upper = *pud_offset(&u.Four, ulAddr);
948	# else /* < 4.12 */
949	u.Upper = *pud_offset(&u.Global, ulAddr);
950	# endif /* < 4.12 */
951	if (RT_UNLIKELY(pud_none(u.Upper)))
952	return NULL;
953	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25)
954	if (pud_large(u.Upper))
955	{
956	pPage = pud_page(u.Upper);
957	AssertReturn(pPage, NULL);
958	pfn = page_to_pfn(pPage); /* doing the safe way... */
959	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (PUD_SHIFT - PAGE_SHIFT)) - 1);
960	return pfn_to_page(pfn);
961	}
962	# endif
963	u.Middle = *pmd_offset(&u.Upper, ulAddr);
964	#else /* < 2.6.11 */
965	u.Middle = *pmd_offset(&u.Global, ulAddr);
966	#endif /* < 2.6.11 */
967	if (RT_UNLIKELY(pmd_none(u.Middle)))
968	return NULL;
969	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
970	if (pmd_large(u.Middle))
971	{
972	pPage = pmd_page(u.Middle);
973	AssertReturn(pPage, NULL);
974	pfn = page_to_pfn(pPage); /* doing the safe way... */
975	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (PMD_SHIFT - PAGE_SHIFT)) - 1);
976	return pfn_to_page(pfn);
977	}
978	#endif
979
980	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 5) \|\| defined(pte_offset_map) /* As usual, RHEL 3 had pte_offset_map earlier. */
981	pEntry = pte_offset_map(&u.Middle, ulAddr);
982	#else
983	pEntry = pte_offset(&u.Middle, ulAddr);
984	#endif
985	if (RT_UNLIKELY(!pEntry))
986	return NULL;
987	u.Entry = *pEntry;
988	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 5) \|\| defined(pte_offset_map)
989	pte_unmap(pEntry);
990	#endif
991
992	if (RT_UNLIKELY(!pte_present(u.Entry)))
993	return NULL;
994	return pte_page(u.Entry);
995	}
996	RT_EXPORT_SYMBOL(rtR0MemObjLinuxVirtToPage);
997
998
999	DECLHIDDEN(int) rtR0MemObjNativeAllocPhys(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest, size_t uAlignment)
1000	{
1001	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS, cb, uAlignment, PhysHighest);
1002	}
1003
1004
1005	DECLHIDDEN(int) rtR0MemObjNativeAllocPhysNC(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest)
1006	{
1007	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS_NC, cb, PAGE_SIZE, PhysHighest);
1008	}
1009
1010
1011	DECLHIDDEN(int) rtR0MemObjNativeEnterPhys(PPRTR0MEMOBJINTERNAL ppMem, RTHCPHYS Phys, size_t cb, uint32_t uCachePolicy)
1012	{
1013	IPRT_LINUX_SAVE_EFL_AC();
1014
1015	/*
1016	* All we need to do here is to validate that we can use
1017	* ioremap on the specified address (32/64-bit dma_addr_t).
1018	*/
1019	PRTR0MEMOBJLNX pMemLnx;
1020	dma_addr_t PhysAddr = Phys;
1021	AssertMsgReturn(PhysAddr == Phys, ("%#llx\n", (unsigned long long)Phys), VERR_ADDRESS_TOO_BIG);
1022
1023	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_PHYS, NULL, cb);
1024	if (!pMemLnx)
1025	{
1026	IPRT_LINUX_RESTORE_EFL_AC();
1027	return VERR_NO_MEMORY;
1028	}
1029
1030	pMemLnx->Core.u.Phys.PhysBase = PhysAddr;
1031	pMemLnx->Core.u.Phys.fAllocated = false;
1032	pMemLnx->Core.u.Phys.uCachePolicy = uCachePolicy;
1033	Assert(!pMemLnx->cPages);
1034	*ppMem = &pMemLnx->Core;
1035	IPRT_LINUX_RESTORE_EFL_AC();
1036	return VINF_SUCCESS;
1037	}
1038
1039	/* openSUSE Leap 42.3 detection :-/ */
1040	#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 4, 0) \
1041	&& LINUX_VERSION_CODE < KERNEL_VERSION(4, 6, 0) \
1042	&& defined(FAULT_FLAG_REMOTE)
1043	# define GET_USER_PAGES_API KERNEL_VERSION(4, 10, 0) /* no typo! */
1044	#else
1045	# define GET_USER_PAGES_API LINUX_VERSION_CODE
1046	#endif
1047
1048	DECLHIDDEN(int) rtR0MemObjNativeLockUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3Ptr, size_t cb, uint32_t fAccess, RTR0PROCESS R0Process)
1049	{
1050	IPRT_LINUX_SAVE_EFL_AC();
1051	const int cPages = cb >> PAGE_SHIFT;
1052	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1053	struct vm_area_struct **papVMAs;
1054	PRTR0MEMOBJLNX pMemLnx;
1055	int rc = VERR_NO_MEMORY;
1056	int const fWrite = fAccess & RTMEM_PROT_WRITE ? 1 : 0;
1057
1058	/*
1059	* Check for valid task and size overflows.
1060	*/
1061	if (!pTask)
1062	return VERR_NOT_SUPPORTED;
1063	if (((size_t)cPages << PAGE_SHIFT) != cb)
1064	return VERR_OUT_OF_RANGE;
1065
1066	/*
1067	* Allocate the memory object and a temporary buffer for the VMAs.
1068	*/
1069	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_UOFFSETOF_DYN(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK, (void *)R3Ptr, cb);
1070	if (!pMemLnx)
1071	{
1072	IPRT_LINUX_RESTORE_EFL_AC();
1073	return VERR_NO_MEMORY;
1074	}
1075
1076	papVMAs = (struct vm_area_struct *)RTMemAlloc(sizeof(papVMAs) * cPages);
1077	if (papVMAs)
1078	{
1079	down_read(&pTask->mm->mmap_sem);
1080
1081	/*
1082	* Get user pages.
1083	*/
1084	/** @todo r=bird: Should we not force read access too? */
1085	#if GET_USER_PAGES_API >= KERNEL_VERSION(4, 6, 0)
1086	if (R0Process == RTR0ProcHandleSelf())
1087	rc = get_user_pages(R3Ptr, /* Where from. */
1088	cPages, /* How many pages. */
1089	# if GET_USER_PAGES_API >= KERNEL_VERSION(4, 9, 0)
1090	fWrite ? FOLL_WRITE \| /* Write to memory. */
1091	FOLL_FORCE /* force write access. */
1092	: 0, /* Write to memory. */
1093	# else
1094	fWrite, /* Write to memory. */
1095	fWrite, /* force write access. */
1096	# endif
1097	&pMemLnx->apPages[0], /* Page array. */
1098	papVMAs); /* vmas */
1099	/*
1100	* Actually this should not happen at the moment as call this function
1101	* only for our own process.
1102	*/
1103	else
1104	rc = get_user_pages_remote(
1105	pTask, /* Task for fault accounting. */
1106	pTask->mm, /* Whose pages. */
1107	R3Ptr, /* Where from. */
1108	cPages, /* How many pages. */
1109	# if GET_USER_PAGES_API >= KERNEL_VERSION(4, 9, 0)
1110	fWrite ? FOLL_WRITE \| /* Write to memory. */
1111	FOLL_FORCE /* force write access. */
1112	: 0, /* Write to memory. */
1113	# else
1114	fWrite, /* Write to memory. */
1115	fWrite, /* force write access. */
1116	# endif
1117	&pMemLnx->apPages[0], /* Page array. */
1118	papVMAs /* vmas */
1119	# if GET_USER_PAGES_API >= KERNEL_VERSION(4, 10, 0)
1120	, NULL /* locked */
1121	# endif
1122	);
1123	#else /* GET_USER_PAGES_API < KERNEL_VERSION(4, 6, 0) */
1124	rc = get_user_pages(pTask, /* Task for fault accounting. */
1125	pTask->mm, /* Whose pages. */
1126	R3Ptr, /* Where from. */
1127	cPages, /* How many pages. */
1128	/* The get_user_pages API change was back-ported to 4.4.168. */
1129	# if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 4, 168) \
1130	&& LINUX_VERSION_CODE < KERNEL_VERSION(4, 5, 0)
1131	fWrite ? FOLL_WRITE \| /* Write to memory. */
1132	FOLL_FORCE /* force write access. */
1133	: 0, /* Write to memory. */
1134	# else
1135	fWrite, /* Write to memory. */
1136	fWrite, /* force write access. */
1137	# endif
1138	&pMemLnx->apPages[0], /* Page array. */
1139	papVMAs); /* vmas */
1140	#endif /* GET_USER_PAGES_API < KERNEL_VERSION(4, 6, 0) */
1141	if (rc == cPages)
1142	{
1143	/*
1144	* Flush dcache (required?), protect against fork and _really_ pin the page
1145	* table entries. get_user_pages() will protect against swapping out the
1146	* pages but it will NOT protect against removing page table entries. This
1147	* can be achieved with
1148	* - using mlock / mmap(..., MAP_LOCKED, ...) from userland. This requires
1149	* an appropriate limit set up with setrlimit(..., RLIMIT_MEMLOCK, ...).
1150	* Usual Linux distributions support only a limited size of locked pages
1151	* (e.g. 32KB).
1152	* - setting the PageReserved bit (as we do in rtR0MemObjLinuxAllocPages()
1153	* or by
1154	* - setting the VM_LOCKED flag. This is the same as doing mlock() without
1155	* a range check.
1156	*/
1157	/** @todo The Linux fork() protection will require more work if this API
1158	* is to be used for anything but locking VM pages. */
1159	while (rc-- > 0)
1160	{
1161	flush_dcache_page(pMemLnx->apPages[rc]);
1162	papVMAs[rc]->vm_flags \|= VM_DONTCOPY \| VM_LOCKED;
1163	}
1164
1165	up_read(&pTask->mm->mmap_sem);
1166
1167	RTMemFree(papVMAs);
1168
1169	pMemLnx->Core.u.Lock.R0Process = R0Process;
1170	pMemLnx->cPages = cPages;
1171	Assert(!pMemLnx->fMappedToRing0);
1172	*ppMem = &pMemLnx->Core;
1173
1174	IPRT_LINUX_RESTORE_EFL_AC();
1175	return VINF_SUCCESS;
1176	}
1177
1178	/*
1179	* Failed - we need to unlock any pages that we succeeded to lock.
1180	*/
1181	while (rc-- > 0)
1182	{
1183	if (!PageReserved(pMemLnx->apPages[rc]))
1184	SetPageDirty(pMemLnx->apPages[rc]);
1185	#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 6, 0)
1186	put_page(pMemLnx->apPages[rc]);
1187	#else
1188	page_cache_release(pMemLnx->apPages[rc]);
1189	#endif
1190	}
1191
1192	up_read(&pTask->mm->mmap_sem);
1193
1194	RTMemFree(papVMAs);
1195	rc = VERR_LOCK_FAILED;
1196	}
1197
1198	rtR0MemObjDelete(&pMemLnx->Core);
1199	IPRT_LINUX_RESTORE_EFL_AC();
1200	return rc;
1201	}
1202
1203
1204	DECLHIDDEN(int) rtR0MemObjNativeLockKernel(PPRTR0MEMOBJINTERNAL ppMem, void *pv, size_t cb, uint32_t fAccess)
1205	{
1206	IPRT_LINUX_SAVE_EFL_AC();
1207	void pvLast = (uint8_t )pv + cb - 1;
1208	size_t const cPages = cb >> PAGE_SHIFT;
1209	PRTR0MEMOBJLNX pMemLnx;
1210	bool fLinearMapping;
1211	int rc;
1212	uint8_t *pbPage;
1213	size_t iPage;
1214	NOREF(fAccess);
1215
1216	if ( !RTR0MemKernelIsValidAddr(pv)
1217	\|\| !RTR0MemKernelIsValidAddr(pv + cb))
1218	return VERR_INVALID_PARAMETER;
1219
1220	/*
1221	* The lower part of the kernel memory has a linear mapping between
1222	* physical and virtual addresses. So we take a short cut here. This is
1223	* assumed to be the cleanest way to handle those addresses (and the code
1224	* is well tested, though the test for determining it is not very nice).
1225	* If we ever decide it isn't we can still remove it.
1226	*/
1227	#if 0
1228	fLinearMapping = (unsigned long)pvLast < VMALLOC_START;
1229	#else
1230	fLinearMapping = (unsigned long)pv >= (unsigned long)__va(0)
1231	&& (unsigned long)pvLast < (unsigned long)high_memory;
1232	#endif
1233
1234	/*
1235	* Allocate the memory object.
1236	*/
1237	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_UOFFSETOF_DYN(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK, pv, cb);
1238	if (!pMemLnx)
1239	{
1240	IPRT_LINUX_RESTORE_EFL_AC();
1241	return VERR_NO_MEMORY;
1242	}
1243
1244	/*
1245	* Gather the pages.
1246	* We ASSUME all kernel pages are non-swappable and non-movable.
1247	*/
1248	rc = VINF_SUCCESS;
1249	pbPage = (uint8_t *)pvLast;
1250	iPage = cPages;
1251	if (!fLinearMapping)
1252	{
1253	while (iPage-- > 0)
1254	{
1255	struct page *pPage = rtR0MemObjLinuxVirtToPage(pbPage);
1256	if (RT_UNLIKELY(!pPage))
1257	{
1258	rc = VERR_LOCK_FAILED;
1259	break;
1260	}
1261	pMemLnx->apPages[iPage] = pPage;
1262	pbPage -= PAGE_SIZE;
1263	}
1264	}
1265	else
1266	{
1267	while (iPage-- > 0)
1268	{
1269	pMemLnx->apPages[iPage] = virt_to_page(pbPage);
1270	pbPage -= PAGE_SIZE;
1271	}
1272	}
1273	if (RT_SUCCESS(rc))
1274	{
1275	/*
1276	* Complete the memory object and return.
1277	*/
1278	pMemLnx->Core.u.Lock.R0Process = NIL_RTR0PROCESS;
1279	pMemLnx->cPages = cPages;
1280	Assert(!pMemLnx->fMappedToRing0);
1281	*ppMem = &pMemLnx->Core;
1282
1283	IPRT_LINUX_RESTORE_EFL_AC();
1284	return VINF_SUCCESS;
1285	}
1286
1287	rtR0MemObjDelete(&pMemLnx->Core);
1288	IPRT_LINUX_RESTORE_EFL_AC();
1289	return rc;
1290	}
1291
1292
1293	DECLHIDDEN(int) rtR0MemObjNativeReserveKernel(PPRTR0MEMOBJINTERNAL ppMem, void *pvFixed, size_t cb, size_t uAlignment)
1294	{
1295	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1296	IPRT_LINUX_SAVE_EFL_AC();
1297	const size_t cPages = cb >> PAGE_SHIFT;
1298	struct page *pDummyPage;
1299	struct page **papPages;
1300
1301	/* check for unsupported stuff. */
1302	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
1303	if (uAlignment > PAGE_SIZE)
1304	return VERR_NOT_SUPPORTED;
1305
1306	/*
1307	* Allocate a dummy page and create a page pointer array for vmap such that
1308	* the dummy page is mapped all over the reserved area.
1309	*/
1310	pDummyPage = alloc_page(GFP_HIGHUSER \| __GFP_NOWARN);
1311	if (pDummyPage)
1312	{
1313	papPages = RTMemAlloc(sizeof(papPages) cPages);
1314	if (papPages)
1315	{
1316	void *pv;
1317	size_t iPage = cPages;
1318	while (iPage-- > 0)
1319	papPages[iPage] = pDummyPage;
1320	# ifdef VM_MAP
1321	pv = vmap(papPages, cPages, VM_MAP, PAGE_KERNEL_RO);
1322	# else
1323	pv = vmap(papPages, cPages, VM_ALLOC, PAGE_KERNEL_RO);
1324	# endif
1325	RTMemFree(papPages);
1326	if (pv)
1327	{
1328	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb);
1329	if (pMemLnx)
1330	{
1331	pMemLnx->Core.u.ResVirt.R0Process = NIL_RTR0PROCESS;
1332	pMemLnx->cPages = 1;
1333	pMemLnx->apPages[0] = pDummyPage;
1334	*ppMem = &pMemLnx->Core;
1335	IPRT_LINUX_RESTORE_EFL_AC();
1336	return VINF_SUCCESS;
1337	}
1338	vunmap(pv);
1339	}
1340	}
1341	__free_page(pDummyPage);
1342	}
1343	IPRT_LINUX_RESTORE_EFL_AC();
1344	return VERR_NO_MEMORY;
1345
1346	#else /* < 2.4.22 */
1347	/*
1348	* Could probably use ioremap here, but the caller is in a better position than us
1349	* to select some safe physical memory.
1350	*/
1351	return VERR_NOT_SUPPORTED;
1352	#endif
1353	}
1354
1355
1356	DECLHIDDEN(int) rtR0MemObjNativeReserveUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment, RTR0PROCESS R0Process)
1357	{
1358	IPRT_LINUX_SAVE_EFL_AC();
1359	PRTR0MEMOBJLNX pMemLnx;
1360	void *pv;
1361	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1362	if (!pTask)
1363	return VERR_NOT_SUPPORTED;
1364
1365	/*
1366	* Check that the specified alignment is supported.
1367	*/
1368	if (uAlignment > PAGE_SIZE)
1369	return VERR_NOT_SUPPORTED;
1370
1371	/*
1372	* Let rtR0MemObjLinuxDoMmap do the difficult bits.
1373	*/
1374	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, cb, uAlignment, pTask, RTMEM_PROT_NONE);
1375	if (pv == (void *)-1)
1376	{
1377	IPRT_LINUX_RESTORE_EFL_AC();
1378	return VERR_NO_MEMORY;
1379	}
1380
1381	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb);
1382	if (!pMemLnx)
1383	{
1384	rtR0MemObjLinuxDoMunmap(pv, cb, pTask);
1385	IPRT_LINUX_RESTORE_EFL_AC();
1386	return VERR_NO_MEMORY;
1387	}
1388
1389	pMemLnx->Core.u.ResVirt.R0Process = R0Process;
1390	*ppMem = &pMemLnx->Core;
1391	IPRT_LINUX_RESTORE_EFL_AC();
1392	return VINF_SUCCESS;
1393	}
1394
1395
1396	DECLHIDDEN(int) rtR0MemObjNativeMapKernel(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap,
1397	void *pvFixed, size_t uAlignment,
1398	unsigned fProt, size_t offSub, size_t cbSub)
1399	{
1400	int rc = VERR_NO_MEMORY;
1401	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1402	PRTR0MEMOBJLNX pMemLnx;
1403	IPRT_LINUX_SAVE_EFL_AC();
1404
1405	/* Fail if requested to do something we can't. */
1406	AssertMsgReturn(!offSub && !cbSub, ("%#x %#x\n", offSub, cbSub), VERR_NOT_SUPPORTED);
1407	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
1408	if (uAlignment > PAGE_SIZE)
1409	return VERR_NOT_SUPPORTED;
1410
1411	/*
1412	* Create the IPRT memory object.
1413	*/
1414	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, pMemLnxToMap->Core.cb);
1415	if (pMemLnx)
1416	{
1417	if (pMemLnxToMap->cPages)
1418	{
1419	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1420	/*
1421	* Use vmap - 2.4.22 and later.
1422	*/
1423	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, true /* kernel */);
1424	# ifdef VM_MAP
1425	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[0], pMemLnxToMap->cPages, VM_MAP, fPg);
1426	# else
1427	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[0], pMemLnxToMap->cPages, VM_ALLOC, fPg);
1428	# endif
1429	if (pMemLnx->Core.pv)
1430	{
1431	pMemLnx->fMappedToRing0 = true;
1432	rc = VINF_SUCCESS;
1433	}
1434	else
1435	rc = VERR_MAP_FAILED;
1436
1437	#else /* < 2.4.22 */
1438	/*
1439	* Only option here is to share mappings if possible and forget about fProt.
1440	*/
1441	if (rtR0MemObjIsRing3(pMemToMap))
1442	rc = VERR_NOT_SUPPORTED;
1443	else
1444	{
1445	rc = VINF_SUCCESS;
1446	if (!pMemLnxToMap->Core.pv)
1447	rc = rtR0MemObjLinuxVMap(pMemLnxToMap, !!(fProt & RTMEM_PROT_EXEC));
1448	if (RT_SUCCESS(rc))
1449	{
1450	Assert(pMemLnxToMap->Core.pv);
1451	pMemLnx->Core.pv = pMemLnxToMap->Core.pv;
1452	}
1453	}
1454	#endif
1455	}
1456	else
1457	{
1458	/*
1459	* MMIO / physical memory.
1460	*/
1461	Assert(pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS && !pMemLnxToMap->Core.u.Phys.fAllocated);
1462	pMemLnx->Core.pv = pMemLnxToMap->Core.u.Phys.uCachePolicy == RTMEM_CACHE_POLICY_MMIO
1463	? ioremap_nocache(pMemLnxToMap->Core.u.Phys.PhysBase, pMemLnxToMap->Core.cb)
1464	: ioremap(pMemLnxToMap->Core.u.Phys.PhysBase, pMemLnxToMap->Core.cb);
1465	if (pMemLnx->Core.pv)
1466	{
1467	/** @todo fix protection. */
1468	rc = VINF_SUCCESS;
1469	}
1470	}
1471	if (RT_SUCCESS(rc))
1472	{
1473	pMemLnx->Core.u.Mapping.R0Process = NIL_RTR0PROCESS;
1474	*ppMem = &pMemLnx->Core;
1475	IPRT_LINUX_RESTORE_EFL_AC();
1476	return VINF_SUCCESS;
1477	}
1478	rtR0MemObjDelete(&pMemLnx->Core);
1479	}
1480
1481	IPRT_LINUX_RESTORE_EFL_AC();
1482	return rc;
1483	}
1484
1485
1486	#ifdef VBOX_USE_PAE_HACK
1487	/**
1488	* Replace the PFN of a PTE with the address of the actual page.
1489	*
1490	* The caller maps a reserved dummy page at the address with the desired access
1491	* and flags.
1492	*
1493	* This hack is required for older Linux kernels which don't provide
1494	* remap_pfn_range().
1495	*
1496	* @returns 0 on success, -ENOMEM on failure.
1497	* @param mm The memory context.
1498	* @param ulAddr The mapping address.
1499	* @param Phys The physical address of the page to map.
1500	*/
1501	static int rtR0MemObjLinuxFixPte(struct mm_struct *mm, unsigned long ulAddr, RTHCPHYS Phys)
1502	{
1503	int rc = -ENOMEM;
1504	pgd_t *pgd;
1505
1506	spin_lock(&mm->page_table_lock);
1507
1508	pgd = pgd_offset(mm, ulAddr);
1509	if (!pgd_none(pgd) && !pgd_bad(pgd))
1510	{
1511	pmd_t *pmd = pmd_offset(pgd, ulAddr);
1512	if (!pmd_none(*pmd))
1513	{
1514	pte_t *ptep = pte_offset_map(pmd, ulAddr);
1515	if (ptep)
1516	{
1517	pte_t pte = *ptep;
1518	pte.pte_high &= 0xfff00000;
1519	pte.pte_high \|= ((Phys >> 32) & 0x000fffff);
1520	pte.pte_low &= 0x00000fff;
1521	pte.pte_low \|= (Phys & 0xfffff000);
1522	set_pte(ptep, pte);
1523	pte_unmap(ptep);
1524	rc = 0;
1525	}
1526	}
1527	}
1528
1529	spin_unlock(&mm->page_table_lock);
1530	return rc;
1531	}
1532	#endif /* VBOX_USE_PAE_HACK */
1533
1534
1535	DECLHIDDEN(int) rtR0MemObjNativeMapUser(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap, RTR3PTR R3PtrFixed, size_t uAlignment,
1536	unsigned fProt, RTR0PROCESS R0Process, size_t offSub, size_t cbSub)
1537	{
1538	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1539	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1540	int rc = VERR_NO_MEMORY;
1541	PRTR0MEMOBJLNX pMemLnx;
1542	#ifdef VBOX_USE_PAE_HACK
1543	struct page *pDummyPage;
1544	RTHCPHYS DummyPhys;
1545	#endif
1546	IPRT_LINUX_SAVE_EFL_AC();
1547
1548	/*
1549	* Check for restrictions.
1550	*/
1551	if (!pTask)
1552	return VERR_NOT_SUPPORTED;
1553	if (uAlignment > PAGE_SIZE)
1554	return VERR_NOT_SUPPORTED;
1555
1556	#ifdef VBOX_USE_PAE_HACK
1557	/*
1558	* Allocate a dummy page for use when mapping the memory.
1559	*/
1560	pDummyPage = alloc_page(GFP_USER \| __GFP_NOWARN);
1561	if (!pDummyPage)
1562	{
1563	IPRT_LINUX_RESTORE_EFL_AC();
1564	return VERR_NO_MEMORY;
1565	}
1566	SetPageReserved(pDummyPage);
1567	DummyPhys = page_to_phys(pDummyPage);
1568	#endif
1569
1570	/*
1571	* Create the IPRT memory object.
1572	*/
1573	Assert(!offSub \|\| cbSub);
1574	if (cbSub == 0)
1575	cbSub = pMemLnxToMap->Core.cb;
1576	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, cbSub);
1577	if (pMemLnx)
1578	{
1579	/*
1580	* Allocate user space mapping.
1581	*/
1582	void *pv;
1583	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, cbSub, uAlignment, pTask, fProt);
1584	if (pv != (void *)-1)
1585	{
1586	/*
1587	* Map page by page into the mmap area.
1588	* This is generic, paranoid and not very efficient.
1589	*/
1590	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, false /* user */);
1591	unsigned long ulAddrCur = (unsigned long)pv;
1592	const size_t cPages = (offSub + cbSub) >> PAGE_SHIFT;
1593	size_t iPage;
1594
1595	down_write(&pTask->mm->mmap_sem);
1596
1597	rc = VINF_SUCCESS;
1598	if (pMemLnxToMap->cPages)
1599	{
1600	for (iPage = offSub >> PAGE_SHIFT; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE)
1601	{
1602	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11)
1603	RTHCPHYS Phys = page_to_phys(pMemLnxToMap->apPages[iPage]);
1604	#endif
1605	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1606	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
1607	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
1608	#endif
1609	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0) && defined(RT_ARCH_X86)
1610	/* remap_page_range() limitation on x86 */
1611	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
1612	#endif
1613
1614	#if defined(VBOX_USE_INSERT_PAGE) && LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1615	rc = vm_insert_page(vma, ulAddrCur, pMemLnxToMap->apPages[iPage]);
1616	/* Thes flags help making 100% sure some bad stuff wont happen (swap, core, ++).
1617	* See remap_pfn_range() in mm/memory.c */
1618	#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 7, 0)
1619	vma->vm_flags \|= VM_DONTEXPAND \| VM_DONTDUMP;
1620	#else
1621	vma->vm_flags \|= VM_RESERVED;
1622	#endif
1623	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
1624	rc = remap_pfn_range(vma, ulAddrCur, page_to_pfn(pMemLnxToMap->apPages[iPage]), PAGE_SIZE, fPg);
1625	#elif defined(VBOX_USE_PAE_HACK)
1626	rc = remap_page_range(vma, ulAddrCur, DummyPhys, PAGE_SIZE, fPg);
1627	if (!rc)
1628	rc = rtR0MemObjLinuxFixPte(pTask->mm, ulAddrCur, Phys);
1629	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1630	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1631	#else /* 2.4 */
1632	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
1633	#endif
1634	if (rc)
1635	{
1636	rc = VERR_NO_MEMORY;
1637	break;
1638	}
1639	}
1640	}
1641	else
1642	{
1643	RTHCPHYS Phys;
1644	if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS)
1645	Phys = pMemLnxToMap->Core.u.Phys.PhysBase;
1646	else if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_CONT)
1647	Phys = pMemLnxToMap->Core.u.Cont.Phys;
1648	else
1649	{
1650	AssertMsgFailed(("%d\n", pMemLnxToMap->Core.enmType));
1651	Phys = NIL_RTHCPHYS;
1652	}
1653	if (Phys != NIL_RTHCPHYS)
1654	{
1655	for (iPage = offSub >> PAGE_SHIFT; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE, Phys += PAGE_SIZE)
1656	{
1657	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1658	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
1659	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
1660	#endif
1661	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0) && defined(RT_ARCH_X86)
1662	/* remap_page_range() limitation on x86 */
1663	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
1664	#endif
1665
1666	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
1667	rc = remap_pfn_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1668	#elif defined(VBOX_USE_PAE_HACK)
1669	rc = remap_page_range(vma, ulAddrCur, DummyPhys, PAGE_SIZE, fPg);
1670	if (!rc)
1671	rc = rtR0MemObjLinuxFixPte(pTask->mm, ulAddrCur, Phys);
1672	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1673	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1674	#else /* 2.4 */
1675	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
1676	#endif
1677	if (rc)
1678	{
1679	rc = VERR_NO_MEMORY;
1680	break;
1681	}
1682	}
1683	}
1684	}
1685
1686	#ifdef CONFIG_NUMA_BALANCING
1687	# if LINUX_VERSION_CODE < KERNEL_VERSION(3, 13, 0)
1688	# ifdef RHEL_RELEASE_CODE
1689	# if RHEL_RELEASE_CODE < RHEL_RELEASE_VERSION(7, 0)
1690	# define VBOX_NUMA_HACK_OLD
1691	# endif
1692	# endif
1693	# endif
1694	if (RT_SUCCESS(rc))
1695	{
1696	/** @todo Ugly hack! But right now we have no other means to
1697	* disable automatic NUMA page balancing. */
1698	# ifdef RT_OS_X86
1699	# ifdef VBOX_NUMA_HACK_OLD
1700	pTask->mm->numa_next_reset = jiffies + 0x7fffffffUL;
1701	# endif
1702	pTask->mm->numa_next_scan = jiffies + 0x7fffffffUL;
1703	# else
1704	# ifdef VBOX_NUMA_HACK_OLD
1705	pTask->mm->numa_next_reset = jiffies + 0x7fffffffffffffffUL;
1706	# endif
1707	pTask->mm->numa_next_scan = jiffies + 0x7fffffffffffffffUL;
1708	# endif
1709	}
1710	#endif /* CONFIG_NUMA_BALANCING */
1711
1712	up_write(&pTask->mm->mmap_sem);
1713
1714	if (RT_SUCCESS(rc))
1715	{
1716	#ifdef VBOX_USE_PAE_HACK
1717	__free_page(pDummyPage);
1718	#endif
1719	pMemLnx->Core.pv = pv;
1720	pMemLnx->Core.u.Mapping.R0Process = R0Process;
1721	*ppMem = &pMemLnx->Core;
1722	IPRT_LINUX_RESTORE_EFL_AC();
1723	return VINF_SUCCESS;
1724	}
1725
1726	/*
1727	* Bail out.
1728	*/
1729	rtR0MemObjLinuxDoMunmap(pv, cbSub, pTask);
1730	}
1731	rtR0MemObjDelete(&pMemLnx->Core);
1732	}
1733	#ifdef VBOX_USE_PAE_HACK
1734	__free_page(pDummyPage);
1735	#endif
1736
1737	IPRT_LINUX_RESTORE_EFL_AC();
1738	return rc;
1739	}
1740
1741
1742	DECLHIDDEN(int) rtR0MemObjNativeProtect(PRTR0MEMOBJINTERNAL pMem, size_t offSub, size_t cbSub, uint32_t fProt)
1743	{
1744	NOREF(pMem);
1745	NOREF(offSub);
1746	NOREF(cbSub);
1747	NOREF(fProt);
1748	return VERR_NOT_SUPPORTED;
1749	}
1750
1751
1752	DECLHIDDEN(RTHCPHYS) rtR0MemObjNativeGetPagePhysAddr(PRTR0MEMOBJINTERNAL pMem, size_t iPage)
1753	{
1754	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
1755
1756	if (pMemLnx->cPages)
1757	return page_to_phys(pMemLnx->apPages[iPage]);
1758
1759	switch (pMemLnx->Core.enmType)
1760	{
1761	case RTR0MEMOBJTYPE_CONT:
1762	return pMemLnx->Core.u.Cont.Phys + (iPage << PAGE_SHIFT);
1763
1764	case RTR0MEMOBJTYPE_PHYS:
1765	return pMemLnx->Core.u.Phys.PhysBase + (iPage << PAGE_SHIFT);
1766
1767	/* the parent knows */
1768	case RTR0MEMOBJTYPE_MAPPING:
1769	return rtR0MemObjNativeGetPagePhysAddr(pMemLnx->Core.uRel.Child.pParent, iPage);
1770
1771	/* cPages > 0 */
1772	case RTR0MEMOBJTYPE_LOW:
1773	case RTR0MEMOBJTYPE_LOCK:
1774	case RTR0MEMOBJTYPE_PHYS_NC:
1775	case RTR0MEMOBJTYPE_PAGE:
1776	default:
1777	AssertMsgFailed(("%d\n", pMemLnx->Core.enmType));
1778	/* fall thru */
1779
1780	case RTR0MEMOBJTYPE_RES_VIRT:
1781	return NIL_RTHCPHYS;
1782	}
1783	}
1784

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

source: vbox/trunk/src/VBox/Runtime/r0drv/linux/memobj-r0drv-linux.c@ 81587

Download in other formats: