memobj-r0drv-linux.c@ 77367

Last change on this file since 77367 was 77367, checked in by vboxsync, 6 years ago
Runtime/r0drv/linux: handle get_user_pages back-port to Linux 4.4.168. ticketref:18315:Kernel modules do not build with linux kernel 4.4.169.
Property svn:eol-style set to `native` Property svn:keywords set to `Id Rev Revision`
File size: 59.5 KB

Line
1	/* $Id: memobj-r0drv-linux.c 77367 2019-02-19 17:11:53Z 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
42
43	/*********************************************************************************************************************************
44	* Defined Constants And Macros *
45	*********************************************************************************************************************************/
46	/* early 2.6 kernels */
47	#ifndef PAGE_SHARED_EXEC
48	# define PAGE_SHARED_EXEC PAGE_SHARED
49	#endif
50	#ifndef PAGE_READONLY_EXEC
51	# define PAGE_READONLY_EXEC PAGE_READONLY
52	#endif
53
54	/*
55	* 2.6.29+ kernels don't work with remap_pfn_range() anymore because
56	* track_pfn_vma_new() is apparently not defined for non-RAM pages.
57	* It should be safe to use vm_insert_page() older kernels as well.
58	*/
59	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 23)
60	# define VBOX_USE_INSERT_PAGE
61	#endif
62	#if defined(CONFIG_X86_PAE) \
63	&& ( defined(HAVE_26_STYLE_REMAP_PAGE_RANGE) \
64	\|\| ( LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \
65	&& LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11)))
66	# define VBOX_USE_PAE_HACK
67	#endif
68
69
70	/*********************************************************************************************************************************
71	* Structures and Typedefs *
72	*********************************************************************************************************************************/
73	/**
74	* The Darwin version of the memory object structure.
75	*/
76	typedef struct RTR0MEMOBJLNX
77	{
78	/** The core structure. */
79	RTR0MEMOBJINTERNAL Core;
80	/** Set if the allocation is contiguous.
81	* This means it has to be given back as one chunk. */
82	bool fContiguous;
83	/** Set if we've vmap'ed the memory into ring-0. */
84	bool fMappedToRing0;
85	/** The pages in the apPages array. */
86	size_t cPages;
87	/** Array of struct page pointers. (variable size) */
88	struct page *apPages[1];
89	} RTR0MEMOBJLNX, *PRTR0MEMOBJLNX;
90
91
92	static void rtR0MemObjLinuxFreePages(PRTR0MEMOBJLNX pMemLnx);
93
94
95	/**
96	* Helper that converts from a RTR0PROCESS handle to a linux task.
97	*
98	* @returns The corresponding Linux task.
99	* @param R0Process IPRT ring-0 process handle.
100	*/
101	static struct task_struct *rtR0ProcessToLinuxTask(RTR0PROCESS R0Process)
102	{
103	/** @todo fix rtR0ProcessToLinuxTask!! */
104	/** @todo many (all?) callers currently assume that we return 'current'! */
105	return R0Process == RTR0ProcHandleSelf() ? current : NULL;
106	}
107
108
109	/**
110	* Compute order. Some functions allocate 2^order pages.
111	*
112	* @returns order.
113	* @param cPages Number of pages.
114	*/
115	static int rtR0MemObjLinuxOrder(size_t cPages)
116	{
117	int iOrder;
118	size_t cTmp;
119
120	for (iOrder = 0, cTmp = cPages; cTmp >>= 1; ++iOrder)
121	;
122	if (cPages & ~((size_t)1 << iOrder))
123	++iOrder;
124
125	return iOrder;
126	}
127
128
129	/**
130	* Converts from RTMEM_PROT_* to Linux PAGE_*.
131	*
132	* @returns Linux page protection constant.
133	* @param fProt The IPRT protection mask.
134	* @param fKernel Whether it applies to kernel or user space.
135	*/
136	static pgprot_t rtR0MemObjLinuxConvertProt(unsigned fProt, bool fKernel)
137	{
138	switch (fProt)
139	{
140	default:
141	AssertMsgFailed(("%#x %d\n", fProt, fKernel));
142	case RTMEM_PROT_NONE:
143	return PAGE_NONE;
144
145	case RTMEM_PROT_READ:
146	return fKernel ? PAGE_KERNEL_RO : PAGE_READONLY;
147
148	case RTMEM_PROT_WRITE:
149	case RTMEM_PROT_WRITE \| RTMEM_PROT_READ:
150	return fKernel ? PAGE_KERNEL : PAGE_SHARED;
151
152	case RTMEM_PROT_EXEC:
153	case RTMEM_PROT_EXEC \| RTMEM_PROT_READ:
154	#if defined(RT_ARCH_X86) \|\| defined(RT_ARCH_AMD64)
155	if (fKernel)
156	{
157	pgprot_t fPg = MY_PAGE_KERNEL_EXEC;
158	pgprot_val(fPg) &= ~_PAGE_RW;
159	return fPg;
160	}
161	return PAGE_READONLY_EXEC;
162	#else
163	return fKernel ? MY_PAGE_KERNEL_EXEC : PAGE_READONLY_EXEC;
164	#endif
165
166	case RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC:
167	case RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC \| RTMEM_PROT_READ:
168	return fKernel ? MY_PAGE_KERNEL_EXEC : PAGE_SHARED_EXEC;
169	}
170	}
171
172
173	/**
174	* Worker for rtR0MemObjNativeReserveUser and rtR0MemObjNativerMapUser that creates
175	* an empty user space mapping.
176	*
177	* We acquire the mmap_sem of the task!
178	*
179	* @returns Pointer to the mapping.
180	* (void *)-1 on failure.
181	* @param R3PtrFixed (RTR3PTR)-1 if anywhere, otherwise a specific location.
182	* @param cb The size of the mapping.
183	* @param uAlignment The alignment of the mapping.
184	* @param pTask The Linux task to create this mapping in.
185	* @param fProt The RTMEM_PROT_* mask.
186	*/
187	static void rtR0MemObjLinuxDoMmap(RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment, struct task_struct pTask, unsigned fProt)
188	{
189	unsigned fLnxProt;
190	unsigned long ulAddr;
191
192	Assert(pTask == current); /* do_mmap */
193	RT_NOREF_PV(pTask);
194
195	/*
196	* Convert from IPRT protection to mman.h PROT_ and call do_mmap.
197	*/
198	fProt &= (RTMEM_PROT_NONE \| RTMEM_PROT_READ \| RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC);
199	if (fProt == RTMEM_PROT_NONE)
200	fLnxProt = PROT_NONE;
201	else
202	{
203	fLnxProt = 0;
204	if (fProt & RTMEM_PROT_READ)
205	fLnxProt \|= PROT_READ;
206	if (fProt & RTMEM_PROT_WRITE)
207	fLnxProt \|= PROT_WRITE;
208	if (fProt & RTMEM_PROT_EXEC)
209	fLnxProt \|= PROT_EXEC;
210	}
211
212	if (R3PtrFixed != (RTR3PTR)-1)
213	{
214	#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 5, 0)
215	ulAddr = vm_mmap(NULL, R3PtrFixed, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS \| MAP_FIXED, 0);
216	#else
217	down_write(&pTask->mm->mmap_sem);
218	ulAddr = do_mmap(NULL, R3PtrFixed, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS \| MAP_FIXED, 0);
219	up_write(&pTask->mm->mmap_sem);
220	#endif
221	}
222	else
223	{
224	#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 5, 0)
225	ulAddr = vm_mmap(NULL, 0, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS, 0);
226	#else
227	down_write(&pTask->mm->mmap_sem);
228	ulAddr = do_mmap(NULL, 0, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS, 0);
229	up_write(&pTask->mm->mmap_sem);
230	#endif
231	if ( !(ulAddr & ~PAGE_MASK)
232	&& (ulAddr & (uAlignment - 1)))
233	{
234	/** @todo implement uAlignment properly... We'll probably need to make some dummy mappings to fill
235	* up alignment gaps. This is of course complicated by fragmentation (which we might have cause
236	* ourselves) and further by there begin two mmap strategies (top / bottom). */
237	/* For now, just ignore uAlignment requirements... */
238	}
239	}
240
241
242	if (ulAddr & ~PAGE_MASK) /* ~PAGE_MASK == PAGE_OFFSET_MASK */
243	return (void *)-1;
244	return (void *)ulAddr;
245	}
246
247
248	/**
249	* Worker that destroys a user space mapping.
250	* Undoes what rtR0MemObjLinuxDoMmap did.
251	*
252	* We acquire the mmap_sem of the task!
253	*
254	* @param pv The ring-3 mapping.
255	* @param cb The size of the mapping.
256	* @param pTask The Linux task to destroy this mapping in.
257	*/
258	static void rtR0MemObjLinuxDoMunmap(void pv, size_t cb, struct task_struct pTask)
259	{
260	#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 5, 0)
261	Assert(pTask == current); RT_NOREF_PV(pTask);
262	vm_munmap((unsigned long)pv, cb);
263	#elif defined(USE_RHEL4_MUNMAP)
264	down_write(&pTask->mm->mmap_sem);
265	do_munmap(pTask->mm, (unsigned long)pv, cb, 0); /* should it be 1 or 0? */
266	up_write(&pTask->mm->mmap_sem);
267	#else
268	down_write(&pTask->mm->mmap_sem);
269	do_munmap(pTask->mm, (unsigned long)pv, cb);
270	up_write(&pTask->mm->mmap_sem);
271	#endif
272	}
273
274
275	/**
276	* Internal worker that allocates physical pages and creates the memory object for them.
277	*
278	* @returns IPRT status code.
279	* @param ppMemLnx Where to store the memory object pointer.
280	* @param enmType The object type.
281	* @param cb The number of bytes to allocate.
282	* @param uAlignment The alignment of the physical memory.
283	* Only valid if fContiguous == true, ignored otherwise.
284	* @param fFlagsLnx The page allocation flags (GPFs).
285	* @param fContiguous Whether the allocation must be contiguous.
286	* @param rcNoMem What to return when we're out of pages.
287	*/
288	static int rtR0MemObjLinuxAllocPages(PRTR0MEMOBJLNX *ppMemLnx, RTR0MEMOBJTYPE enmType, size_t cb,
289	size_t uAlignment, unsigned fFlagsLnx, bool fContiguous, int rcNoMem)
290	{
291	size_t iPage;
292	size_t const cPages = cb >> PAGE_SHIFT;
293	struct page *paPages;
294
295	/*
296	* Allocate a memory object structure that's large enough to contain
297	* the page pointer array.
298	*/
299	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_UOFFSETOF_DYN(RTR0MEMOBJLNX, apPages[cPages]), enmType, NULL, cb);
300	if (!pMemLnx)
301	return VERR_NO_MEMORY;
302	pMemLnx->cPages = cPages;
303
304	if (cPages > 255)
305	{
306	# ifdef __GFP_REPEAT
307	/* Try hard to allocate the memory, but the allocation attempt might fail. */
308	fFlagsLnx \|= __GFP_REPEAT;
309	# endif
310	# ifdef __GFP_NOMEMALLOC
311	/* Introduced with Linux 2.6.12: Don't use emergency reserves */
312	fFlagsLnx \|= __GFP_NOMEMALLOC;
313	# endif
314	}
315
316	/*
317	* Allocate the pages.
318	* For small allocations we'll try contiguous first and then fall back on page by page.
319	*/
320	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
321	if ( fContiguous
322	\|\| cb <= PAGE_SIZE * 2)
323	{
324	# ifdef VBOX_USE_INSERT_PAGE
325	paPages = alloc_pages(fFlagsLnx \| __GFP_COMP \| __GFP_NOWARN, rtR0MemObjLinuxOrder(cPages));
326	# else
327	paPages = alloc_pages(fFlagsLnx \| __GFP_NOWARN, rtR0MemObjLinuxOrder(cPages));
328	# endif
329	if (paPages)
330	{
331	fContiguous = true;
332	for (iPage = 0; iPage < cPages; iPage++)
333	pMemLnx->apPages[iPage] = &paPages[iPage];
334	}
335	else if (fContiguous)
336	{
337	rtR0MemObjDelete(&pMemLnx->Core);
338	return rcNoMem;
339	}
340	}
341
342	if (!fContiguous)
343	{
344	for (iPage = 0; iPage < cPages; iPage++)
345	{
346	pMemLnx->apPages[iPage] = alloc_page(fFlagsLnx \| __GFP_NOWARN);
347	if (RT_UNLIKELY(!pMemLnx->apPages[iPage]))
348	{
349	while (iPage-- > 0)
350	__free_page(pMemLnx->apPages[iPage]);
351	rtR0MemObjDelete(&pMemLnx->Core);
352	return rcNoMem;
353	}
354	}
355	}
356
357	#else /* < 2.4.22 */
358	/** @todo figure out why we didn't allocate page-by-page on 2.4.21 and older... */
359	paPages = alloc_pages(fFlagsLnx, rtR0MemObjLinuxOrder(cPages));
360	if (!paPages)
361	{
362	rtR0MemObjDelete(&pMemLnx->Core);
363	return rcNoMem;
364	}
365	for (iPage = 0; iPage < cPages; iPage++)
366	{
367	pMemLnx->apPages[iPage] = &paPages[iPage];
368	MY_SET_PAGES_EXEC(pMemLnx->apPages[iPage], 1);
369	if (PageHighMem(pMemLnx->apPages[iPage]))
370	BUG();
371	}
372
373	fContiguous = true;
374	#endif /* < 2.4.22 */
375	pMemLnx->fContiguous = fContiguous;
376
377	#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 5, 0)
378	/*
379	* Reserve the pages.
380	*
381	* Linux >= 4.5 with CONFIG_DEBUG_VM panics when setting PG_reserved on compound
382	* pages. According to Michal Hocko this shouldn't be necessary anyway because
383	* as pages which are not on the LRU list are never evictable.
384	*/
385	for (iPage = 0; iPage < cPages; iPage++)
386	SetPageReserved(pMemLnx->apPages[iPage]);
387	#endif
388
389	/*
390	* Note that the physical address of memory allocated with alloc_pages(flags, order)
391	* is always 2^(PAGE_SHIFT+order)-aligned.
392	*/
393	if ( fContiguous
394	&& uAlignment > PAGE_SIZE)
395	{
396	/*
397	* Check for alignment constraints. The physical address of memory allocated with
398	* alloc_pages(flags, order) is always 2^(PAGE_SHIFT+order)-aligned.
399	*/
400	if (RT_UNLIKELY(page_to_phys(pMemLnx->apPages[0]) & (uAlignment - 1)))
401	{
402	/*
403	* This should never happen!
404	*/
405	printk("rtR0MemObjLinuxAllocPages(cb=0x%lx, uAlignment=0x%lx): alloc_pages(..., %d) returned physical memory at 0x%lx!\n",
406	(unsigned long)cb, (unsigned long)uAlignment, rtR0MemObjLinuxOrder(cPages), (unsigned long)page_to_phys(pMemLnx->apPages[0]));
407	rtR0MemObjLinuxFreePages(pMemLnx);
408	return rcNoMem;
409	}
410	}
411
412	*ppMemLnx = pMemLnx;
413	return VINF_SUCCESS;
414	}
415
416
417	/**
418	* Frees the physical pages allocated by the rtR0MemObjLinuxAllocPages() call.
419	*
420	* This method does NOT free the object.
421	*
422	* @param pMemLnx The object which physical pages should be freed.
423	*/
424	static void rtR0MemObjLinuxFreePages(PRTR0MEMOBJLNX pMemLnx)
425	{
426	size_t iPage = pMemLnx->cPages;
427	if (iPage > 0)
428	{
429	/*
430	* Restore the page flags.
431	*/
432	while (iPage-- > 0)
433	{
434	#if LINUX_VERSION_CODE < KERNEL_VERSION(4, 5, 0)
435	/*
436	* See SetPageReserved() in rtR0MemObjLinuxAllocPages()
437	*/
438	ClearPageReserved(pMemLnx->apPages[iPage]);
439	#endif
440	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
441	#else
442	MY_SET_PAGES_NOEXEC(pMemLnx->apPages[iPage], 1);
443	#endif
444	}
445
446	/*
447	* Free the pages.
448	*/
449	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
450	if (!pMemLnx->fContiguous)
451	{
452	iPage = pMemLnx->cPages;
453	while (iPage-- > 0)
454	__free_page(pMemLnx->apPages[iPage]);
455	}
456	else
457	#endif
458	__free_pages(pMemLnx->apPages[0], rtR0MemObjLinuxOrder(pMemLnx->cPages));
459
460	pMemLnx->cPages = 0;
461	}
462	}
463
464
465	/**
466	* Maps the allocation into ring-0.
467	*
468	* This will update the RTR0MEMOBJLNX::Core.pv and RTR0MEMOBJ::fMappedToRing0 members.
469	*
470	* Contiguous mappings that isn't in 'high' memory will already be mapped into kernel
471	* space, so we'll use that mapping if possible. If execute access is required, we'll
472	* play safe and do our own mapping.
473	*
474	* @returns IPRT status code.
475	* @param pMemLnx The linux memory object to map.
476	* @param fExecutable Whether execute access is required.
477	*/
478	static int rtR0MemObjLinuxVMap(PRTR0MEMOBJLNX pMemLnx, bool fExecutable)
479	{
480	int rc = VINF_SUCCESS;
481
482	/*
483	* Choose mapping strategy.
484	*/
485	bool fMustMap = fExecutable
486	\|\| !pMemLnx->fContiguous;
487	if (!fMustMap)
488	{
489	size_t iPage = pMemLnx->cPages;
490	while (iPage-- > 0)
491	if (PageHighMem(pMemLnx->apPages[iPage]))
492	{
493	fMustMap = true;
494	break;
495	}
496	}
497
498	Assert(!pMemLnx->Core.pv);
499	Assert(!pMemLnx->fMappedToRing0);
500
501	if (fMustMap)
502	{
503	/*
504	* Use vmap - 2.4.22 and later.
505	*/
506	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
507	pgprot_t fPg;
508	pgprot_val(fPg) = _PAGE_PRESENT \| _PAGE_RW;
509	# ifdef _PAGE_NX
510	if (!fExecutable)
511	pgprot_val(fPg) \|= _PAGE_NX;
512	# endif
513
514	# ifdef VM_MAP
515	pMemLnx->Core.pv = vmap(&pMemLnx->apPages[0], pMemLnx->cPages, VM_MAP, fPg);
516	# else
517	pMemLnx->Core.pv = vmap(&pMemLnx->apPages[0], pMemLnx->cPages, VM_ALLOC, fPg);
518	# endif
519	if (pMemLnx->Core.pv)
520	pMemLnx->fMappedToRing0 = true;
521	else
522	rc = VERR_MAP_FAILED;
523	#else /* < 2.4.22 */
524	rc = VERR_NOT_SUPPORTED;
525	#endif
526	}
527	else
528	{
529	/*
530	* Use the kernel RAM mapping.
531	*/
532	pMemLnx->Core.pv = phys_to_virt(page_to_phys(pMemLnx->apPages[0]));
533	Assert(pMemLnx->Core.pv);
534	}
535
536	return rc;
537	}
538
539
540	/**
541	* Undoes what rtR0MemObjLinuxVMap() did.
542	*
543	* @param pMemLnx The linux memory object.
544	*/
545	static void rtR0MemObjLinuxVUnmap(PRTR0MEMOBJLNX pMemLnx)
546	{
547	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
548	if (pMemLnx->fMappedToRing0)
549	{
550	Assert(pMemLnx->Core.pv);
551	vunmap(pMemLnx->Core.pv);
552	pMemLnx->fMappedToRing0 = false;
553	}
554	#else /* < 2.4.22 */
555	Assert(!pMemLnx->fMappedToRing0);
556	#endif
557	pMemLnx->Core.pv = NULL;
558	}
559
560
561	DECLHIDDEN(int) rtR0MemObjNativeFree(RTR0MEMOBJ pMem)
562	{
563	IPRT_LINUX_SAVE_EFL_AC();
564	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
565
566	/*
567	* Release any memory that we've allocated or locked.
568	*/
569	switch (pMemLnx->Core.enmType)
570	{
571	case RTR0MEMOBJTYPE_LOW:
572	case RTR0MEMOBJTYPE_PAGE:
573	case RTR0MEMOBJTYPE_CONT:
574	case RTR0MEMOBJTYPE_PHYS:
575	case RTR0MEMOBJTYPE_PHYS_NC:
576	rtR0MemObjLinuxVUnmap(pMemLnx);
577	rtR0MemObjLinuxFreePages(pMemLnx);
578	break;
579
580	case RTR0MEMOBJTYPE_LOCK:
581	if (pMemLnx->Core.u.Lock.R0Process != NIL_RTR0PROCESS)
582	{
583	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
584	size_t iPage;
585	Assert(pTask);
586	if (pTask && pTask->mm)
587	down_read(&pTask->mm->mmap_sem);
588
589	iPage = pMemLnx->cPages;
590	while (iPage-- > 0)
591	{
592	if (!PageReserved(pMemLnx->apPages[iPage]))
593	SetPageDirty(pMemLnx->apPages[iPage]);
594	#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 6, 0)
595	put_page(pMemLnx->apPages[iPage]);
596	#else
597	page_cache_release(pMemLnx->apPages[iPage]);
598	#endif
599	}
600
601	if (pTask && pTask->mm)
602	up_read(&pTask->mm->mmap_sem);
603	}
604	/* else: kernel memory - nothing to do here. */
605	break;
606
607	case RTR0MEMOBJTYPE_RES_VIRT:
608	Assert(pMemLnx->Core.pv);
609	if (pMemLnx->Core.u.ResVirt.R0Process != NIL_RTR0PROCESS)
610	{
611	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
612	Assert(pTask);
613	if (pTask && pTask->mm)
614	rtR0MemObjLinuxDoMunmap(pMemLnx->Core.pv, pMemLnx->Core.cb, pTask);
615	}
616	else
617	{
618	vunmap(pMemLnx->Core.pv);
619
620	Assert(pMemLnx->cPages == 1 && pMemLnx->apPages[0] != NULL);
621	__free_page(pMemLnx->apPages[0]);
622	pMemLnx->apPages[0] = NULL;
623	pMemLnx->cPages = 0;
624	}
625	pMemLnx->Core.pv = NULL;
626	break;
627
628	case RTR0MEMOBJTYPE_MAPPING:
629	Assert(pMemLnx->cPages == 0); Assert(pMemLnx->Core.pv);
630	if (pMemLnx->Core.u.ResVirt.R0Process != NIL_RTR0PROCESS)
631	{
632	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
633	Assert(pTask);
634	if (pTask && pTask->mm)
635	rtR0MemObjLinuxDoMunmap(pMemLnx->Core.pv, pMemLnx->Core.cb, pTask);
636	}
637	else
638	vunmap(pMemLnx->Core.pv);
639	pMemLnx->Core.pv = NULL;
640	break;
641
642	default:
643	AssertMsgFailed(("enmType=%d\n", pMemLnx->Core.enmType));
644	return VERR_INTERNAL_ERROR;
645	}
646	IPRT_LINUX_RESTORE_EFL_ONLY_AC();
647	return VINF_SUCCESS;
648	}
649
650
651	DECLHIDDEN(int) rtR0MemObjNativeAllocPage(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable)
652	{
653	IPRT_LINUX_SAVE_EFL_AC();
654	PRTR0MEMOBJLNX pMemLnx;
655	int rc;
656
657	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
658	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_PAGE, cb, PAGE_SIZE, GFP_HIGHUSER,
659	false /* non-contiguous */, VERR_NO_MEMORY);
660	#else
661	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_PAGE, cb, PAGE_SIZE, GFP_USER,
662	false /* non-contiguous */, VERR_NO_MEMORY);
663	#endif
664	if (RT_SUCCESS(rc))
665	{
666	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
667	if (RT_SUCCESS(rc))
668	{
669	*ppMem = &pMemLnx->Core;
670	IPRT_LINUX_RESTORE_EFL_AC();
671	return rc;
672	}
673
674	rtR0MemObjLinuxFreePages(pMemLnx);
675	rtR0MemObjDelete(&pMemLnx->Core);
676	}
677
678	IPRT_LINUX_RESTORE_EFL_AC();
679	return rc;
680	}
681
682
683	DECLHIDDEN(int) rtR0MemObjNativeAllocLow(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable)
684	{
685	IPRT_LINUX_SAVE_EFL_AC();
686	PRTR0MEMOBJLNX pMemLnx;
687	int rc;
688
689	/* Try to avoid GFP_DMA. GFM_DMA32 was introduced with Linux 2.6.15. */
690	#if (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
691	/* ZONE_DMA32: 0-4GB */
692	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_DMA32,
693	false /* non-contiguous */, VERR_NO_LOW_MEMORY);
694	if (RT_FAILURE(rc))
695	#endif
696	#ifdef RT_ARCH_AMD64
697	/* ZONE_DMA: 0-16MB */
698	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_DMA,
699	false /* non-contiguous */, VERR_NO_LOW_MEMORY);
700	#else
701	# ifdef CONFIG_X86_PAE
702	# endif
703	/* ZONE_NORMAL: 0-896MB */
704	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_USER,
705	false /* non-contiguous */, VERR_NO_LOW_MEMORY);
706	#endif
707	if (RT_SUCCESS(rc))
708	{
709	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
710	if (RT_SUCCESS(rc))
711	{
712	*ppMem = &pMemLnx->Core;
713	IPRT_LINUX_RESTORE_EFL_AC();
714	return rc;
715	}
716
717	rtR0MemObjLinuxFreePages(pMemLnx);
718	rtR0MemObjDelete(&pMemLnx->Core);
719	}
720
721	IPRT_LINUX_RESTORE_EFL_AC();
722	return rc;
723	}
724
725
726	DECLHIDDEN(int) rtR0MemObjNativeAllocCont(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable)
727	{
728	IPRT_LINUX_SAVE_EFL_AC();
729	PRTR0MEMOBJLNX pMemLnx;
730	int rc;
731
732	#if (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
733	/* ZONE_DMA32: 0-4GB */
734	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, PAGE_SIZE, GFP_DMA32,
735	true /* contiguous */, VERR_NO_CONT_MEMORY);
736	if (RT_FAILURE(rc))
737	#endif
738	#ifdef RT_ARCH_AMD64
739	/* ZONE_DMA: 0-16MB */
740	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, PAGE_SIZE, GFP_DMA,
741	true /* contiguous */, VERR_NO_CONT_MEMORY);
742	#else
743	/* ZONE_NORMAL (32-bit hosts): 0-896MB */
744	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, PAGE_SIZE, GFP_USER,
745	true /* contiguous */, VERR_NO_CONT_MEMORY);
746	#endif
747	if (RT_SUCCESS(rc))
748	{
749	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
750	if (RT_SUCCESS(rc))
751	{
752	#if defined(RT_STRICT) && (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_HIGHMEM64G))
753	size_t iPage = pMemLnx->cPages;
754	while (iPage-- > 0)
755	Assert(page_to_phys(pMemLnx->apPages[iPage]) < _4G);
756	#endif
757	pMemLnx->Core.u.Cont.Phys = page_to_phys(pMemLnx->apPages[0]);
758	*ppMem = &pMemLnx->Core;
759	IPRT_LINUX_RESTORE_EFL_AC();
760	return rc;
761	}
762
763	rtR0MemObjLinuxFreePages(pMemLnx);
764	rtR0MemObjDelete(&pMemLnx->Core);
765	}
766
767	IPRT_LINUX_RESTORE_EFL_AC();
768	return rc;
769	}
770
771
772	/**
773	* Worker for rtR0MemObjLinuxAllocPhysSub that tries one allocation strategy.
774	*
775	* @returns IPRT status code.
776	* @param ppMemLnx Where to
777	* @param enmType The object type.
778	* @param cb The size of the allocation.
779	* @param uAlignment The alignment of the physical memory.
780	* Only valid for fContiguous == true, ignored otherwise.
781	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
782	* @param fGfp The Linux GFP flags to use for the allocation.
783	*/
784	static int rtR0MemObjLinuxAllocPhysSub2(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType,
785	size_t cb, size_t uAlignment, RTHCPHYS PhysHighest, unsigned fGfp)
786	{
787	PRTR0MEMOBJLNX pMemLnx;
788	int rc;
789
790	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, enmType, cb, uAlignment, fGfp,
791	enmType == RTR0MEMOBJTYPE_PHYS /* contiguous / non-contiguous */,
792	VERR_NO_PHYS_MEMORY);
793	if (RT_FAILURE(rc))
794	return rc;
795
796	/*
797	* Check the addresses if necessary. (Can be optimized a bit for PHYS.)
798	*/
799	if (PhysHighest != NIL_RTHCPHYS)
800	{
801	size_t iPage = pMemLnx->cPages;
802	while (iPage-- > 0)
803	if (page_to_phys(pMemLnx->apPages[iPage]) > PhysHighest)
804	{
805	rtR0MemObjLinuxFreePages(pMemLnx);
806	rtR0MemObjDelete(&pMemLnx->Core);
807	return VERR_NO_MEMORY;
808	}
809	}
810
811	/*
812	* Complete the object.
813	*/
814	if (enmType == RTR0MEMOBJTYPE_PHYS)
815	{
816	pMemLnx->Core.u.Phys.PhysBase = page_to_phys(pMemLnx->apPages[0]);
817	pMemLnx->Core.u.Phys.fAllocated = true;
818	}
819	*ppMem = &pMemLnx->Core;
820	return rc;
821	}
822
823
824	/**
825	* Worker for rtR0MemObjNativeAllocPhys and rtR0MemObjNativeAllocPhysNC.
826	*
827	* @returns IPRT status code.
828	* @param ppMem Where to store the memory object pointer on success.
829	* @param enmType The object type.
830	* @param cb The size of the allocation.
831	* @param uAlignment The alignment of the physical memory.
832	* Only valid for enmType == RTR0MEMOBJTYPE_PHYS, ignored otherwise.
833	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
834	*/
835	static int rtR0MemObjLinuxAllocPhysSub(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType,
836	size_t cb, size_t uAlignment, RTHCPHYS PhysHighest)
837	{
838	int rc;
839	IPRT_LINUX_SAVE_EFL_AC();
840
841	/*
842	* There are two clear cases and that's the <=16MB and anything-goes ones.
843	* When the physical address limit is somewhere in-between those two we'll
844	* just have to try, starting with HIGHUSER and working our way thru the
845	* different types, hoping we'll get lucky.
846	*
847	* We should probably move this physical address restriction logic up to
848	* the page alloc function as it would be more efficient there. But since
849	* we don't expect this to be a performance issue just yet it can wait.
850	*/
851	if (PhysHighest == NIL_RTHCPHYS)
852	/* ZONE_HIGHMEM: the whole physical memory */
853	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_HIGHUSER);
854	else if (PhysHighest <= _1M * 16)
855	/* ZONE_DMA: 0-16MB */
856	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_DMA);
857	else
858	{
859	rc = VERR_NO_MEMORY;
860	if (RT_FAILURE(rc))
861	/* ZONE_HIGHMEM: the whole physical memory */
862	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_HIGHUSER);
863	if (RT_FAILURE(rc))
864	/* ZONE_NORMAL: 0-896MB */
865	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_USER);
866	#ifdef GFP_DMA32
867	if (RT_FAILURE(rc))
868	/* ZONE_DMA32: 0-4GB */
869	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_DMA32);
870	#endif
871	if (RT_FAILURE(rc))
872	/* ZONE_DMA: 0-16MB */
873	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, GFP_DMA);
874	}
875	IPRT_LINUX_RESTORE_EFL_AC();
876	return rc;
877	}
878
879
880	/**
881	* Translates a kernel virtual address to a linux page structure by walking the
882	* page tables.
883	*
884	* @note We do assume that the page tables will not change as we are walking
885	* them. This assumption is rather forced by the fact that I could not
886	* immediately see any way of preventing this from happening. So, we
887	* take some extra care when accessing them.
888	*
889	* Because of this, we don't want to use this function on memory where
890	* attribute changes to nearby pages is likely to cause large pages to
891	* be used or split up. So, don't use this for the linear mapping of
892	* physical memory.
893	*
894	* @returns Pointer to the page structur or NULL if it could not be found.
895	* @param pv The kernel virtual address.
896	*/
897	static struct page rtR0MemObjLinuxVirtToPage(void pv)
898	{
899	unsigned long ulAddr = (unsigned long)pv;
900	unsigned long pfn;
901	struct page *pPage;
902	pte_t *pEntry;
903	union
904	{
905	pgd_t Global;
906	#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 12, 0)
907	p4d_t Four;
908	#endif
909	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
910	pud_t Upper;
911	#endif
912	pmd_t Middle;
913	pte_t Entry;
914	} u;
915
916	/* Should this happen in a situation this code will be called in? And if
917	* so, can it change under our feet? See also
918	* "Documentation/vm/active_mm.txt" in the kernel sources. */
919	if (RT_UNLIKELY(!current->active_mm))
920	return NULL;
921	u.Global = *pgd_offset(current->active_mm, ulAddr);
922	if (RT_UNLIKELY(pgd_none(u.Global)))
923	return NULL;
924	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
925	# if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 12, 0)
926	u.Four = *p4d_offset(&u.Global, ulAddr);
927	if (RT_UNLIKELY(p4d_none(u.Four)))
928	return NULL;
929	if (p4d_large(u.Four))
930	{
931	pPage = p4d_page(u.Four);
932	AssertReturn(pPage, NULL);
933	pfn = page_to_pfn(pPage); /* doing the safe way... */
934	AssertCompile(P4D_SHIFT - PAGE_SHIFT < 31);
935	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (P4D_SHIFT - PAGE_SHIFT)) - 1);
936	return pfn_to_page(pfn);
937	}
938	u.Upper = *pud_offset(&u.Four, ulAddr);
939	# else /* < 4.12 */
940	u.Upper = *pud_offset(&u.Global, ulAddr);
941	# endif /* < 4.12 */
942	if (RT_UNLIKELY(pud_none(u.Upper)))
943	return NULL;
944	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 25)
945	if (pud_large(u.Upper))
946	{
947	pPage = pud_page(u.Upper);
948	AssertReturn(pPage, NULL);
949	pfn = page_to_pfn(pPage); /* doing the safe way... */
950	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (PUD_SHIFT - PAGE_SHIFT)) - 1);
951	return pfn_to_page(pfn);
952	}
953	# endif
954	u.Middle = *pmd_offset(&u.Upper, ulAddr);
955	#else /* < 2.6.11 */
956	u.Middle = *pmd_offset(&u.Global, ulAddr);
957	#endif /* < 2.6.11 */
958	if (RT_UNLIKELY(pmd_none(u.Middle)))
959	return NULL;
960	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0)
961	if (pmd_large(u.Middle))
962	{
963	pPage = pmd_page(u.Middle);
964	AssertReturn(pPage, NULL);
965	pfn = page_to_pfn(pPage); /* doing the safe way... */
966	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (PMD_SHIFT - PAGE_SHIFT)) - 1);
967	return pfn_to_page(pfn);
968	}
969	#endif
970
971	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 5) \|\| defined(pte_offset_map) /* As usual, RHEL 3 had pte_offset_map earlier. */
972	pEntry = pte_offset_map(&u.Middle, ulAddr);
973	#else
974	pEntry = pte_offset(&u.Middle, ulAddr);
975	#endif
976	if (RT_UNLIKELY(!pEntry))
977	return NULL;
978	u.Entry = *pEntry;
979	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 5, 5) \|\| defined(pte_offset_map)
980	pte_unmap(pEntry);
981	#endif
982
983	if (RT_UNLIKELY(!pte_present(u.Entry)))
984	return NULL;
985	return pte_page(u.Entry);
986	}
987
988
989	DECLHIDDEN(int) rtR0MemObjNativeAllocPhys(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest, size_t uAlignment)
990	{
991	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS, cb, uAlignment, PhysHighest);
992	}
993
994
995	DECLHIDDEN(int) rtR0MemObjNativeAllocPhysNC(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest)
996	{
997	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS_NC, cb, PAGE_SIZE, PhysHighest);
998	}
999
1000
1001	DECLHIDDEN(int) rtR0MemObjNativeEnterPhys(PPRTR0MEMOBJINTERNAL ppMem, RTHCPHYS Phys, size_t cb, uint32_t uCachePolicy)
1002	{
1003	IPRT_LINUX_SAVE_EFL_AC();
1004
1005	/*
1006	* All we need to do here is to validate that we can use
1007	* ioremap on the specified address (32/64-bit dma_addr_t).
1008	*/
1009	PRTR0MEMOBJLNX pMemLnx;
1010	dma_addr_t PhysAddr = Phys;
1011	AssertMsgReturn(PhysAddr == Phys, ("%#llx\n", (unsigned long long)Phys), VERR_ADDRESS_TOO_BIG);
1012
1013	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_PHYS, NULL, cb);
1014	if (!pMemLnx)
1015	{
1016	IPRT_LINUX_RESTORE_EFL_AC();
1017	return VERR_NO_MEMORY;
1018	}
1019
1020	pMemLnx->Core.u.Phys.PhysBase = PhysAddr;
1021	pMemLnx->Core.u.Phys.fAllocated = false;
1022	pMemLnx->Core.u.Phys.uCachePolicy = uCachePolicy;
1023	Assert(!pMemLnx->cPages);
1024	*ppMem = &pMemLnx->Core;
1025	IPRT_LINUX_RESTORE_EFL_AC();
1026	return VINF_SUCCESS;
1027	}
1028
1029	/* openSUSE Leap 42.3 detection :-/ */
1030	#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 4, 0) \
1031	&& LINUX_VERSION_CODE < KERNEL_VERSION(4, 6, 0) \
1032	&& defined(FAULT_FLAG_REMOTE)
1033	# define GET_USER_PAGES_API KERNEL_VERSION(4, 10, 0) /* no typo! */
1034	/* The get_user_pages API change was back-ported to 4.4.168. */
1035	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(4, 4, 168) \
1036	&& LINUX_VERSION_CODE < KERNEL_VERSION(4, 5, 0)
1037	# define GET_USER_PAGES_API KERNEL_VERSION(4, 10, 0) /* no typo! */
1038	#else
1039	# define GET_USER_PAGES_API LINUX_VERSION_CODE
1040	#endif
1041
1042	DECLHIDDEN(int) rtR0MemObjNativeLockUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3Ptr, size_t cb, uint32_t fAccess, RTR0PROCESS R0Process)
1043	{
1044	IPRT_LINUX_SAVE_EFL_AC();
1045	const int cPages = cb >> PAGE_SHIFT;
1046	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1047	struct vm_area_struct **papVMAs;
1048	PRTR0MEMOBJLNX pMemLnx;
1049	int rc = VERR_NO_MEMORY;
1050	int const fWrite = fAccess & RTMEM_PROT_WRITE ? 1 : 0;
1051
1052	/*
1053	* Check for valid task and size overflows.
1054	*/
1055	if (!pTask)
1056	return VERR_NOT_SUPPORTED;
1057	if (((size_t)cPages << PAGE_SHIFT) != cb)
1058	return VERR_OUT_OF_RANGE;
1059
1060	/*
1061	* Allocate the memory object and a temporary buffer for the VMAs.
1062	*/
1063	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_UOFFSETOF_DYN(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK, (void *)R3Ptr, cb);
1064	if (!pMemLnx)
1065	{
1066	IPRT_LINUX_RESTORE_EFL_AC();
1067	return VERR_NO_MEMORY;
1068	}
1069
1070	papVMAs = (struct vm_area_struct *)RTMemAlloc(sizeof(papVMAs) * cPages);
1071	if (papVMAs)
1072	{
1073	down_read(&pTask->mm->mmap_sem);
1074
1075	/*
1076	* Get user pages.
1077	*/
1078	/** @todo r=bird: Should we not force read access too? */
1079	#if GET_USER_PAGES_API >= KERNEL_VERSION(4, 6, 0)
1080	if (R0Process == RTR0ProcHandleSelf())
1081	rc = get_user_pages(R3Ptr, /* Where from. */
1082	cPages, /* How many pages. */
1083	# if GET_USER_PAGES_API >= KERNEL_VERSION(4, 9, 0)
1084	fWrite ? FOLL_WRITE \| /* Write to memory. */
1085	FOLL_FORCE /* force write access. */
1086	: 0, /* Write to memory. */
1087	# else
1088	fWrite, /* Write to memory. */
1089	fWrite, /* force write access. */
1090	# endif
1091	&pMemLnx->apPages[0], /* Page array. */
1092	papVMAs); /* vmas */
1093	/*
1094	* Actually this should not happen at the moment as call this function
1095	* only for our own process.
1096	*/
1097	else
1098	rc = get_user_pages_remote(
1099	pTask, /* Task for fault accounting. */
1100	pTask->mm, /* Whose pages. */
1101	R3Ptr, /* Where from. */
1102	cPages, /* How many pages. */
1103	# if GET_USER_PAGES_API >= KERNEL_VERSION(4, 9, 0)
1104	fWrite ? FOLL_WRITE \| /* Write to memory. */
1105	FOLL_FORCE /* force write access. */
1106	: 0, /* Write to memory. */
1107	# else
1108	fWrite, /* Write to memory. */
1109	fWrite, /* force write access. */
1110	# endif
1111	&pMemLnx->apPages[0], /* Page array. */
1112	papVMAs /* vmas */
1113	# if GET_USER_PAGES_API >= KERNEL_VERSION(4, 10, 0)
1114	, NULL /* locked */
1115	# endif
1116	);
1117	#else /* GET_USER_PAGES_API < KERNEL_VERSION(4, 6, 0) */
1118	rc = get_user_pages(pTask, /* Task for fault accounting. */
1119	pTask->mm, /* Whose pages. */
1120	R3Ptr, /* Where from. */
1121	cPages, /* How many pages. */
1122	# if GET_USER_PAGES_API >= KERNEL_VERSION(4, 9, 0)
1123	fWrite ? FOLL_WRITE \| /* Write to memory. */
1124	FOLL_FORCE /* force write access. */
1125	: 0, /* Write to memory. */
1126	# else
1127	fWrite, /* Write to memory. */
1128	fWrite, /* force write access. */
1129	# endif
1130	&pMemLnx->apPages[0], /* Page array. */
1131	papVMAs); /* vmas */
1132	#endif /* GET_USER_PAGES_API < KERNEL_VERSION(4, 6, 0) */
1133	if (rc == cPages)
1134	{
1135	/*
1136	* Flush dcache (required?), protect against fork and _really_ pin the page
1137	* table entries. get_user_pages() will protect against swapping out the
1138	* pages but it will NOT protect against removing page table entries. This
1139	* can be achieved with
1140	* - using mlock / mmap(..., MAP_LOCKED, ...) from userland. This requires
1141	* an appropriate limit set up with setrlimit(..., RLIMIT_MEMLOCK, ...).
1142	* Usual Linux distributions support only a limited size of locked pages
1143	* (e.g. 32KB).
1144	* - setting the PageReserved bit (as we do in rtR0MemObjLinuxAllocPages()
1145	* or by
1146	* - setting the VM_LOCKED flag. This is the same as doing mlock() without
1147	* a range check.
1148	*/
1149	/** @todo The Linux fork() protection will require more work if this API
1150	* is to be used for anything but locking VM pages. */
1151	while (rc-- > 0)
1152	{
1153	flush_dcache_page(pMemLnx->apPages[rc]);
1154	papVMAs[rc]->vm_flags \|= VM_DONTCOPY \| VM_LOCKED;
1155	}
1156
1157	up_read(&pTask->mm->mmap_sem);
1158
1159	RTMemFree(papVMAs);
1160
1161	pMemLnx->Core.u.Lock.R0Process = R0Process;
1162	pMemLnx->cPages = cPages;
1163	Assert(!pMemLnx->fMappedToRing0);
1164	*ppMem = &pMemLnx->Core;
1165
1166	IPRT_LINUX_RESTORE_EFL_AC();
1167	return VINF_SUCCESS;
1168	}
1169
1170	/*
1171	* Failed - we need to unlock any pages that we succeeded to lock.
1172	*/
1173	while (rc-- > 0)
1174	{
1175	if (!PageReserved(pMemLnx->apPages[rc]))
1176	SetPageDirty(pMemLnx->apPages[rc]);
1177	#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 6, 0)
1178	put_page(pMemLnx->apPages[rc]);
1179	#else
1180	page_cache_release(pMemLnx->apPages[rc]);
1181	#endif
1182	}
1183
1184	up_read(&pTask->mm->mmap_sem);
1185
1186	RTMemFree(papVMAs);
1187	rc = VERR_LOCK_FAILED;
1188	}
1189
1190	rtR0MemObjDelete(&pMemLnx->Core);
1191	IPRT_LINUX_RESTORE_EFL_AC();
1192	return rc;
1193	}
1194
1195
1196	DECLHIDDEN(int) rtR0MemObjNativeLockKernel(PPRTR0MEMOBJINTERNAL ppMem, void *pv, size_t cb, uint32_t fAccess)
1197	{
1198	IPRT_LINUX_SAVE_EFL_AC();
1199	void pvLast = (uint8_t )pv + cb - 1;
1200	size_t const cPages = cb >> PAGE_SHIFT;
1201	PRTR0MEMOBJLNX pMemLnx;
1202	bool fLinearMapping;
1203	int rc;
1204	uint8_t *pbPage;
1205	size_t iPage;
1206	NOREF(fAccess);
1207
1208	if ( !RTR0MemKernelIsValidAddr(pv)
1209	\|\| !RTR0MemKernelIsValidAddr(pv + cb))
1210	return VERR_INVALID_PARAMETER;
1211
1212	/*
1213	* The lower part of the kernel memory has a linear mapping between
1214	* physical and virtual addresses. So we take a short cut here. This is
1215	* assumed to be the cleanest way to handle those addresses (and the code
1216	* is well tested, though the test for determining it is not very nice).
1217	* If we ever decide it isn't we can still remove it.
1218	*/
1219	#if 0
1220	fLinearMapping = (unsigned long)pvLast < VMALLOC_START;
1221	#else
1222	fLinearMapping = (unsigned long)pv >= (unsigned long)__va(0)
1223	&& (unsigned long)pvLast < (unsigned long)high_memory;
1224	#endif
1225
1226	/*
1227	* Allocate the memory object.
1228	*/
1229	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_UOFFSETOF_DYN(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK, pv, cb);
1230	if (!pMemLnx)
1231	{
1232	IPRT_LINUX_RESTORE_EFL_AC();
1233	return VERR_NO_MEMORY;
1234	}
1235
1236	/*
1237	* Gather the pages.
1238	* We ASSUME all kernel pages are non-swappable and non-movable.
1239	*/
1240	rc = VINF_SUCCESS;
1241	pbPage = (uint8_t *)pvLast;
1242	iPage = cPages;
1243	if (!fLinearMapping)
1244	{
1245	while (iPage-- > 0)
1246	{
1247	struct page *pPage = rtR0MemObjLinuxVirtToPage(pbPage);
1248	if (RT_UNLIKELY(!pPage))
1249	{
1250	rc = VERR_LOCK_FAILED;
1251	break;
1252	}
1253	pMemLnx->apPages[iPage] = pPage;
1254	pbPage -= PAGE_SIZE;
1255	}
1256	}
1257	else
1258	{
1259	while (iPage-- > 0)
1260	{
1261	pMemLnx->apPages[iPage] = virt_to_page(pbPage);
1262	pbPage -= PAGE_SIZE;
1263	}
1264	}
1265	if (RT_SUCCESS(rc))
1266	{
1267	/*
1268	* Complete the memory object and return.
1269	*/
1270	pMemLnx->Core.u.Lock.R0Process = NIL_RTR0PROCESS;
1271	pMemLnx->cPages = cPages;
1272	Assert(!pMemLnx->fMappedToRing0);
1273	*ppMem = &pMemLnx->Core;
1274
1275	IPRT_LINUX_RESTORE_EFL_AC();
1276	return VINF_SUCCESS;
1277	}
1278
1279	rtR0MemObjDelete(&pMemLnx->Core);
1280	IPRT_LINUX_RESTORE_EFL_AC();
1281	return rc;
1282	}
1283
1284
1285	DECLHIDDEN(int) rtR0MemObjNativeReserveKernel(PPRTR0MEMOBJINTERNAL ppMem, void *pvFixed, size_t cb, size_t uAlignment)
1286	{
1287	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1288	IPRT_LINUX_SAVE_EFL_AC();
1289	const size_t cPages = cb >> PAGE_SHIFT;
1290	struct page *pDummyPage;
1291	struct page **papPages;
1292
1293	/* check for unsupported stuff. */
1294	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
1295	if (uAlignment > PAGE_SIZE)
1296	return VERR_NOT_SUPPORTED;
1297
1298	/*
1299	* Allocate a dummy page and create a page pointer array for vmap such that
1300	* the dummy page is mapped all over the reserved area.
1301	*/
1302	pDummyPage = alloc_page(GFP_HIGHUSER \| __GFP_NOWARN);
1303	if (pDummyPage)
1304	{
1305	papPages = RTMemAlloc(sizeof(papPages) cPages);
1306	if (papPages)
1307	{
1308	void *pv;
1309	size_t iPage = cPages;
1310	while (iPage-- > 0)
1311	papPages[iPage] = pDummyPage;
1312	# ifdef VM_MAP
1313	pv = vmap(papPages, cPages, VM_MAP, PAGE_KERNEL_RO);
1314	# else
1315	pv = vmap(papPages, cPages, VM_ALLOC, PAGE_KERNEL_RO);
1316	# endif
1317	RTMemFree(papPages);
1318	if (pv)
1319	{
1320	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb);
1321	if (pMemLnx)
1322	{
1323	pMemLnx->Core.u.ResVirt.R0Process = NIL_RTR0PROCESS;
1324	pMemLnx->cPages = 1;
1325	pMemLnx->apPages[0] = pDummyPage;
1326	*ppMem = &pMemLnx->Core;
1327	IPRT_LINUX_RESTORE_EFL_AC();
1328	return VINF_SUCCESS;
1329	}
1330	vunmap(pv);
1331	}
1332	}
1333	__free_page(pDummyPage);
1334	}
1335	IPRT_LINUX_RESTORE_EFL_AC();
1336	return VERR_NO_MEMORY;
1337
1338	#else /* < 2.4.22 */
1339	/*
1340	* Could probably use ioremap here, but the caller is in a better position than us
1341	* to select some safe physical memory.
1342	*/
1343	return VERR_NOT_SUPPORTED;
1344	#endif
1345	}
1346
1347
1348	DECLHIDDEN(int) rtR0MemObjNativeReserveUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment, RTR0PROCESS R0Process)
1349	{
1350	IPRT_LINUX_SAVE_EFL_AC();
1351	PRTR0MEMOBJLNX pMemLnx;
1352	void *pv;
1353	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1354	if (!pTask)
1355	return VERR_NOT_SUPPORTED;
1356
1357	/*
1358	* Check that the specified alignment is supported.
1359	*/
1360	if (uAlignment > PAGE_SIZE)
1361	return VERR_NOT_SUPPORTED;
1362
1363	/*
1364	* Let rtR0MemObjLinuxDoMmap do the difficult bits.
1365	*/
1366	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, cb, uAlignment, pTask, RTMEM_PROT_NONE);
1367	if (pv == (void *)-1)
1368	{
1369	IPRT_LINUX_RESTORE_EFL_AC();
1370	return VERR_NO_MEMORY;
1371	}
1372
1373	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb);
1374	if (!pMemLnx)
1375	{
1376	rtR0MemObjLinuxDoMunmap(pv, cb, pTask);
1377	IPRT_LINUX_RESTORE_EFL_AC();
1378	return VERR_NO_MEMORY;
1379	}
1380
1381	pMemLnx->Core.u.ResVirt.R0Process = R0Process;
1382	*ppMem = &pMemLnx->Core;
1383	IPRT_LINUX_RESTORE_EFL_AC();
1384	return VINF_SUCCESS;
1385	}
1386
1387
1388	DECLHIDDEN(int) rtR0MemObjNativeMapKernel(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap,
1389	void *pvFixed, size_t uAlignment,
1390	unsigned fProt, size_t offSub, size_t cbSub)
1391	{
1392	int rc = VERR_NO_MEMORY;
1393	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1394	PRTR0MEMOBJLNX pMemLnx;
1395	IPRT_LINUX_SAVE_EFL_AC();
1396
1397	/* Fail if requested to do something we can't. */
1398	AssertMsgReturn(!offSub && !cbSub, ("%#x %#x\n", offSub, cbSub), VERR_NOT_SUPPORTED);
1399	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
1400	if (uAlignment > PAGE_SIZE)
1401	return VERR_NOT_SUPPORTED;
1402
1403	/*
1404	* Create the IPRT memory object.
1405	*/
1406	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, pMemLnxToMap->Core.cb);
1407	if (pMemLnx)
1408	{
1409	if (pMemLnxToMap->cPages)
1410	{
1411	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 4, 22)
1412	/*
1413	* Use vmap - 2.4.22 and later.
1414	*/
1415	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, true /* kernel */);
1416	# ifdef VM_MAP
1417	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[0], pMemLnxToMap->cPages, VM_MAP, fPg);
1418	# else
1419	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[0], pMemLnxToMap->cPages, VM_ALLOC, fPg);
1420	# endif
1421	if (pMemLnx->Core.pv)
1422	{
1423	pMemLnx->fMappedToRing0 = true;
1424	rc = VINF_SUCCESS;
1425	}
1426	else
1427	rc = VERR_MAP_FAILED;
1428
1429	#else /* < 2.4.22 */
1430	/*
1431	* Only option here is to share mappings if possible and forget about fProt.
1432	*/
1433	if (rtR0MemObjIsRing3(pMemToMap))
1434	rc = VERR_NOT_SUPPORTED;
1435	else
1436	{
1437	rc = VINF_SUCCESS;
1438	if (!pMemLnxToMap->Core.pv)
1439	rc = rtR0MemObjLinuxVMap(pMemLnxToMap, !!(fProt & RTMEM_PROT_EXEC));
1440	if (RT_SUCCESS(rc))
1441	{
1442	Assert(pMemLnxToMap->Core.pv);
1443	pMemLnx->Core.pv = pMemLnxToMap->Core.pv;
1444	}
1445	}
1446	#endif
1447	}
1448	else
1449	{
1450	/*
1451	* MMIO / physical memory.
1452	*/
1453	Assert(pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS && !pMemLnxToMap->Core.u.Phys.fAllocated);
1454	pMemLnx->Core.pv = pMemLnxToMap->Core.u.Phys.uCachePolicy == RTMEM_CACHE_POLICY_MMIO
1455	? ioremap_nocache(pMemLnxToMap->Core.u.Phys.PhysBase, pMemLnxToMap->Core.cb)
1456	: ioremap(pMemLnxToMap->Core.u.Phys.PhysBase, pMemLnxToMap->Core.cb);
1457	if (pMemLnx->Core.pv)
1458	{
1459	/** @todo fix protection. */
1460	rc = VINF_SUCCESS;
1461	}
1462	}
1463	if (RT_SUCCESS(rc))
1464	{
1465	pMemLnx->Core.u.Mapping.R0Process = NIL_RTR0PROCESS;
1466	*ppMem = &pMemLnx->Core;
1467	IPRT_LINUX_RESTORE_EFL_AC();
1468	return VINF_SUCCESS;
1469	}
1470	rtR0MemObjDelete(&pMemLnx->Core);
1471	}
1472
1473	IPRT_LINUX_RESTORE_EFL_AC();
1474	return rc;
1475	}
1476
1477
1478	#ifdef VBOX_USE_PAE_HACK
1479	/**
1480	* Replace the PFN of a PTE with the address of the actual page.
1481	*
1482	* The caller maps a reserved dummy page at the address with the desired access
1483	* and flags.
1484	*
1485	* This hack is required for older Linux kernels which don't provide
1486	* remap_pfn_range().
1487	*
1488	* @returns 0 on success, -ENOMEM on failure.
1489	* @param mm The memory context.
1490	* @param ulAddr The mapping address.
1491	* @param Phys The physical address of the page to map.
1492	*/
1493	static int rtR0MemObjLinuxFixPte(struct mm_struct *mm, unsigned long ulAddr, RTHCPHYS Phys)
1494	{
1495	int rc = -ENOMEM;
1496	pgd_t *pgd;
1497
1498	spin_lock(&mm->page_table_lock);
1499
1500	pgd = pgd_offset(mm, ulAddr);
1501	if (!pgd_none(pgd) && !pgd_bad(pgd))
1502	{
1503	pmd_t *pmd = pmd_offset(pgd, ulAddr);
1504	if (!pmd_none(*pmd))
1505	{
1506	pte_t *ptep = pte_offset_map(pmd, ulAddr);
1507	if (ptep)
1508	{
1509	pte_t pte = *ptep;
1510	pte.pte_high &= 0xfff00000;
1511	pte.pte_high \|= ((Phys >> 32) & 0x000fffff);
1512	pte.pte_low &= 0x00000fff;
1513	pte.pte_low \|= (Phys & 0xfffff000);
1514	set_pte(ptep, pte);
1515	pte_unmap(ptep);
1516	rc = 0;
1517	}
1518	}
1519	}
1520
1521	spin_unlock(&mm->page_table_lock);
1522	return rc;
1523	}
1524	#endif /* VBOX_USE_PAE_HACK */
1525
1526
1527	DECLHIDDEN(int) rtR0MemObjNativeMapUser(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap, RTR3PTR R3PtrFixed,
1528	size_t uAlignment, unsigned fProt, RTR0PROCESS R0Process)
1529	{
1530	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1531	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1532	int rc = VERR_NO_MEMORY;
1533	PRTR0MEMOBJLNX pMemLnx;
1534	#ifdef VBOX_USE_PAE_HACK
1535	struct page *pDummyPage;
1536	RTHCPHYS DummyPhys;
1537	#endif
1538	IPRT_LINUX_SAVE_EFL_AC();
1539
1540	/*
1541	* Check for restrictions.
1542	*/
1543	if (!pTask)
1544	return VERR_NOT_SUPPORTED;
1545	if (uAlignment > PAGE_SIZE)
1546	return VERR_NOT_SUPPORTED;
1547
1548	#ifdef VBOX_USE_PAE_HACK
1549	/*
1550	* Allocate a dummy page for use when mapping the memory.
1551	*/
1552	pDummyPage = alloc_page(GFP_USER \| __GFP_NOWARN);
1553	if (!pDummyPage)
1554	{
1555	IPRT_LINUX_RESTORE_EFL_AC();
1556	return VERR_NO_MEMORY;
1557	}
1558	SetPageReserved(pDummyPage);
1559	DummyPhys = page_to_phys(pDummyPage);
1560	#endif
1561
1562	/*
1563	* Create the IPRT memory object.
1564	*/
1565	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, pMemLnxToMap->Core.cb);
1566	if (pMemLnx)
1567	{
1568	/*
1569	* Allocate user space mapping.
1570	*/
1571	void *pv;
1572	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, pMemLnxToMap->Core.cb, uAlignment, pTask, fProt);
1573	if (pv != (void *)-1)
1574	{
1575	/*
1576	* Map page by page into the mmap area.
1577	* This is generic, paranoid and not very efficient.
1578	*/
1579	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, false /* user */);
1580	unsigned long ulAddrCur = (unsigned long)pv;
1581	const size_t cPages = pMemLnxToMap->Core.cb >> PAGE_SHIFT;
1582	size_t iPage;
1583
1584	down_write(&pTask->mm->mmap_sem);
1585
1586	rc = VINF_SUCCESS;
1587	if (pMemLnxToMap->cPages)
1588	{
1589	for (iPage = 0; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE)
1590	{
1591	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 11)
1592	RTHCPHYS Phys = page_to_phys(pMemLnxToMap->apPages[iPage]);
1593	#endif
1594	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1595	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
1596	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
1597	#endif
1598	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0) && defined(RT_ARCH_X86)
1599	/* remap_page_range() limitation on x86 */
1600	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
1601	#endif
1602
1603	#if defined(VBOX_USE_INSERT_PAGE) && LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 22)
1604	rc = vm_insert_page(vma, ulAddrCur, pMemLnxToMap->apPages[iPage]);
1605	/* Thes flags help making 100% sure some bad stuff wont happen (swap, core, ++).
1606	* See remap_pfn_range() in mm/memory.c */
1607	#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 7, 0)
1608	vma->vm_flags \|= VM_DONTEXPAND \| VM_DONTDUMP;
1609	#else
1610	vma->vm_flags \|= VM_RESERVED;
1611	#endif
1612	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
1613	rc = remap_pfn_range(vma, ulAddrCur, page_to_pfn(pMemLnxToMap->apPages[iPage]), PAGE_SIZE, fPg);
1614	#elif defined(VBOX_USE_PAE_HACK)
1615	rc = remap_page_range(vma, ulAddrCur, DummyPhys, PAGE_SIZE, fPg);
1616	if (!rc)
1617	rc = rtR0MemObjLinuxFixPte(pTask->mm, ulAddrCur, Phys);
1618	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1619	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1620	#else /* 2.4 */
1621	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
1622	#endif
1623	if (rc)
1624	{
1625	rc = VERR_NO_MEMORY;
1626	break;
1627	}
1628	}
1629	}
1630	else
1631	{
1632	RTHCPHYS Phys;
1633	if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS)
1634	Phys = pMemLnxToMap->Core.u.Phys.PhysBase;
1635	else if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_CONT)
1636	Phys = pMemLnxToMap->Core.u.Cont.Phys;
1637	else
1638	{
1639	AssertMsgFailed(("%d\n", pMemLnxToMap->Core.enmType));
1640	Phys = NIL_RTHCPHYS;
1641	}
1642	if (Phys != NIL_RTHCPHYS)
1643	{
1644	for (iPage = 0; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE, Phys += PAGE_SIZE)
1645	{
1646	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1647	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
1648	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
1649	#endif
1650	#if LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 0) && defined(RT_ARCH_X86)
1651	/* remap_page_range() limitation on x86 */
1652	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
1653	#endif
1654
1655	#if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 11)
1656	rc = remap_pfn_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1657	#elif defined(VBOX_USE_PAE_HACK)
1658	rc = remap_page_range(vma, ulAddrCur, DummyPhys, PAGE_SIZE, fPg);
1659	if (!rc)
1660	rc = rtR0MemObjLinuxFixPte(pTask->mm, ulAddrCur, Phys);
1661	#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1662	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1663	#else /* 2.4 */
1664	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
1665	#endif
1666	if (rc)
1667	{
1668	rc = VERR_NO_MEMORY;
1669	break;
1670	}
1671	}
1672	}
1673	}
1674
1675	#ifdef CONFIG_NUMA_BALANCING
1676	# if LINUX_VERSION_CODE < KERNEL_VERSION(3, 13, 0)
1677	# ifdef RHEL_RELEASE_CODE
1678	# if RHEL_RELEASE_CODE < RHEL_RELEASE_VERSION(7, 0)
1679	# define VBOX_NUMA_HACK_OLD
1680	# endif
1681	# endif
1682	# endif
1683	if (RT_SUCCESS(rc))
1684	{
1685	/** @todo Ugly hack! But right now we have no other means to
1686	* disable automatic NUMA page balancing. */
1687	# ifdef RT_OS_X86
1688	# ifdef VBOX_NUMA_HACK_OLD
1689	pTask->mm->numa_next_reset = jiffies + 0x7fffffffUL;
1690	# endif
1691	pTask->mm->numa_next_scan = jiffies + 0x7fffffffUL;
1692	# else
1693	# ifdef VBOX_NUMA_HACK_OLD
1694	pTask->mm->numa_next_reset = jiffies + 0x7fffffffffffffffUL;
1695	# endif
1696	pTask->mm->numa_next_scan = jiffies + 0x7fffffffffffffffUL;
1697	# endif
1698	}
1699	#endif /* CONFIG_NUMA_BALANCING */
1700
1701	up_write(&pTask->mm->mmap_sem);
1702
1703	if (RT_SUCCESS(rc))
1704	{
1705	#ifdef VBOX_USE_PAE_HACK
1706	__free_page(pDummyPage);
1707	#endif
1708	pMemLnx->Core.pv = pv;
1709	pMemLnx->Core.u.Mapping.R0Process = R0Process;
1710	*ppMem = &pMemLnx->Core;
1711	IPRT_LINUX_RESTORE_EFL_AC();
1712	return VINF_SUCCESS;
1713	}
1714
1715	/*
1716	* Bail out.
1717	*/
1718	rtR0MemObjLinuxDoMunmap(pv, pMemLnxToMap->Core.cb, pTask);
1719	}
1720	rtR0MemObjDelete(&pMemLnx->Core);
1721	}
1722	#ifdef VBOX_USE_PAE_HACK
1723	__free_page(pDummyPage);
1724	#endif
1725
1726	IPRT_LINUX_RESTORE_EFL_AC();
1727	return rc;
1728	}
1729
1730
1731	DECLHIDDEN(int) rtR0MemObjNativeProtect(PRTR0MEMOBJINTERNAL pMem, size_t offSub, size_t cbSub, uint32_t fProt)
1732	{
1733	NOREF(pMem);
1734	NOREF(offSub);
1735	NOREF(cbSub);
1736	NOREF(fProt);
1737	return VERR_NOT_SUPPORTED;
1738	}
1739
1740
1741	DECLHIDDEN(RTHCPHYS) rtR0MemObjNativeGetPagePhysAddr(PRTR0MEMOBJINTERNAL pMem, size_t iPage)
1742	{
1743	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
1744
1745	if (pMemLnx->cPages)
1746	return page_to_phys(pMemLnx->apPages[iPage]);
1747
1748	switch (pMemLnx->Core.enmType)
1749	{
1750	case RTR0MEMOBJTYPE_CONT:
1751	return pMemLnx->Core.u.Cont.Phys + (iPage << PAGE_SHIFT);
1752
1753	case RTR0MEMOBJTYPE_PHYS:
1754	return pMemLnx->Core.u.Phys.PhysBase + (iPage << PAGE_SHIFT);
1755
1756	/* the parent knows */
1757	case RTR0MEMOBJTYPE_MAPPING:
1758	return rtR0MemObjNativeGetPagePhysAddr(pMemLnx->Core.uRel.Child.pParent, iPage);
1759
1760	/* cPages > 0 */
1761	case RTR0MEMOBJTYPE_LOW:
1762	case RTR0MEMOBJTYPE_LOCK:
1763	case RTR0MEMOBJTYPE_PHYS_NC:
1764	case RTR0MEMOBJTYPE_PAGE:
1765	default:
1766	AssertMsgFailed(("%d\n", pMemLnx->Core.enmType));
1767	/* fall thru */
1768
1769	case RTR0MEMOBJTYPE_RES_VIRT:
1770	return NIL_RTHCPHYS;
1771	}
1772	}
1773

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

source: vbox/trunk/src/VBox/Runtime/r0drv/linux/memobj-r0drv-linux.c@ 77367

Download in other formats: