memobj-r0drv-linux.c@ 106734

Last change on this file since 106734 was 106311, checked in by vboxsync, 7 months ago
iprt: Linux: Add initial support for RHEL 9.6 kernel, bugref:10482.
Property svn:eol-style set to `native` Property svn:keywords set to `Id Rev Revision`
File size: 75.8 KB

Line
1	/* $Id: memobj-r0drv-linux.c 106311 2024-10-14 16:17:54Z vboxsync $ */
2	/** @file
3	* IPRT - Ring-0 Memory Objects, Linux.
4	*/
5
6	/*
7	* Copyright (C) 2006-2024 Oracle and/or its affiliates.
8	*
9	* This file is part of VirtualBox base platform packages, as
10	* available from https://www.virtualbox.org.
11	*
12	* This program is free software; you can redistribute it and/or
13	* modify it under the terms of the GNU General Public License
14	* as published by the Free Software Foundation, in version 3 of the
15	* License.
16	*
17	* This program is distributed in the hope that it will be useful, but
18	* WITHOUT ANY WARRANTY; without even the implied warranty of
19	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20	* General Public License for more details.
21	*
22	* You should have received a copy of the GNU General Public License
23	* along with this program; if not, see <https://www.gnu.org/licenses>.
24	*
25	* The contents of this file may alternatively be used under the terms
26	* of the Common Development and Distribution License Version 1.0
27	* (CDDL), a copy of it is provided in the "COPYING.CDDL" file included
28	* in the VirtualBox distribution, in which case the provisions of the
29	* CDDL are applicable instead of those of the GPL.
30	*
31	* You may elect to license modified versions of this file under the
32	* terms and conditions of either the GPL or the CDDL or both.
33	*
34	* SPDX-License-Identifier: GPL-3.0-only OR CDDL-1.0
35	*/
36
37
38	/*********************************************************************************************************************************
39	* Header Files *
40	*********************************************************************************************************************************/
41	#include "the-linux-kernel.h"
42
43	#include <iprt/memobj.h>
44	#include <iprt/assert.h>
45	#include <iprt/err.h>
46	#include <iprt/log.h>
47	#include <iprt/mem.h>
48	#include <iprt/process.h>
49	#include <iprt/string.h>
50	#include "internal/memobj.h"
51	#include "internal/iprt.h"
52
53
54	/*********************************************************************************************************************************
55	* Defined Constants And Macros *
56	*********************************************************************************************************************************/
57	/* early 2.6 kernels */
58	#ifndef PAGE_SHARED_EXEC
59	# define PAGE_SHARED_EXEC PAGE_SHARED
60	#endif
61	#ifndef PAGE_READONLY_EXEC
62	# define PAGE_READONLY_EXEC PAGE_READONLY
63	#endif
64
65	/** @def IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
66	* Whether we use alloc_vm_area (3.2+) for executable memory.
67	* This is a must for 5.8+, but we enable it all the way back to 3.2.x for
68	* better W^R compliance (fExecutable flag). */
69	#if RTLNX_VER_RANGE(3,2,0, 5,10,0) \|\| defined(DOXYGEN_RUNNING)
70	# define IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
71	#endif
72	/** @def IPRT_USE_APPLY_TO_PAGE_RANGE_FOR_EXEC
73	* alloc_vm_area was removed with 5.10 so we have to resort to a different way
74	* to allocate executable memory.
75	* It would be possible to remove IPRT_USE_ALLOC_VM_AREA_FOR_EXEC and use
76	* this path execlusively for 3.2+ but no time to test it really works on every
77	* supported kernel, so better play safe for now.
78	*/
79	#if RTLNX_VER_MIN(5,10,0) \|\| defined(DOXYGEN_RUNNING)
80	# define IPRT_USE_APPLY_TO_PAGE_RANGE_FOR_EXEC
81	#endif
82
83	/*
84	* 2.6.29+ kernels don't work with remap_pfn_range() anymore because
85	* track_pfn_vma_new() is apparently not defined for non-RAM pages.
86	* It should be safe to use vm_insert_page() older kernels as well.
87	*/
88	#if RTLNX_VER_MIN(2,6,23)
89	# define VBOX_USE_INSERT_PAGE
90	#endif
91	#if defined(CONFIG_X86_PAE) \
92	&& ( defined(HAVE_26_STYLE_REMAP_PAGE_RANGE) \
93	\|\| RTLNX_VER_RANGE(2,6,0, 2,6,11) )
94	# define VBOX_USE_PAE_HACK
95	#endif
96
97	/* gfp_t was introduced in 2.6.14, define it for earlier. */
98	#if RTLNX_VER_MAX(2,6,14)
99	# define gfp_t unsigned
100	#endif
101
102	/*
103	* Wrappers around mmap_lock/mmap_sem difference.
104	*/
105	#if RTLNX_VER_MIN(5,8,0)
106	# define LNX_MM_DOWN_READ(a_pMm) down_read(&(a_pMm)->mmap_lock)
107	# define LNX_MM_UP_READ(a_pMm) up_read(&(a_pMm)->mmap_lock)
108	# define LNX_MM_DOWN_WRITE(a_pMm) down_write(&(a_pMm)->mmap_lock)
109	# define LNX_MM_UP_WRITE(a_pMm) up_write(&(a_pMm)->mmap_lock)
110	#else
111	# define LNX_MM_DOWN_READ(a_pMm) down_read(&(a_pMm)->mmap_sem)
112	# define LNX_MM_UP_READ(a_pMm) up_read(&(a_pMm)->mmap_sem)
113	# define LNX_MM_DOWN_WRITE(a_pMm) down_write(&(a_pMm)->mmap_sem)
114	# define LNX_MM_UP_WRITE(a_pMm) up_write(&(a_pMm)->mmap_sem)
115	#endif
116
117
118	/*********************************************************************************************************************************
119	* Structures and Typedefs *
120	*********************************************************************************************************************************/
121	/**
122	* The Linux version of the memory object structure.
123	*/
124	typedef struct RTR0MEMOBJLNX
125	{
126	/** The core structure. */
127	RTR0MEMOBJINTERNAL Core;
128	/** Set if the allocation is contiguous.
129	* This means it has to be given back as one chunk. */
130	bool fContiguous;
131	/** Set if executable allocation. */
132	bool fExecutable;
133	/** Set if we've vmap'ed the memory into ring-0. */
134	bool fMappedToRing0;
135	/** This is non-zero if large page allocation. */
136	uint8_t cLargePageOrder;
137	#ifdef IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
138	/** Return from alloc_vm_area() that we now need to use for executable
139	* memory. */
140	struct vm_struct *pArea;
141	/** PTE array that goes along with pArea (must be freed). */
142	pte_t **papPtesForArea;
143	#endif
144	/** The pages in the apPages array. */
145	size_t cPages;
146	/** Array of struct page pointers. (variable size) */
147	RT_FLEXIBLE_ARRAY_EXTENSION
148	struct page *apPages[RT_FLEXIBLE_ARRAY];
149	} RTR0MEMOBJLNX;
150	/** Pointer to the linux memory object. */
151	typedef RTR0MEMOBJLNX *PRTR0MEMOBJLNX;
152
153
154	/*********************************************************************************************************************************
155	* Global Variables *
156	*********************************************************************************************************************************/
157	/*
158	* Linux allows only a coarse selection of zones for
159	* allocations matching a particular maximum physical address.
160	*
161	* Sorted from high to low physical address!
162	*/
163	static const struct
164	{
165	RTHCPHYS PhysHighest;
166	gfp_t fGfp;
167	} g_aZones[] =
168	{
169	{ NIL_RTHCPHYS, GFP_KERNEL },
170	#if (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
171	{ _4G - 1, GFP_DMA32 }, /* ZONE_DMA32: 0-4GB */
172	#elif defined(RT_ARCH_ARM32) \|\| defined(RT_ARCH_ARM64)
173	{ _4G - 1, GFP_DMA }, /* ZONE_DMA: 0-4GB */
174	#endif
175	#if defined(RT_ARCH_AMD64)
176	{ _16M - 1, GFP_DMA }, /* ZONE_DMA: 0-16MB */
177	#elif defined(RT_ARCH_X86)
178	{ 896 * _1M - 1, GFP_USER }, /* ZONE_NORMAL (32-bit hosts): 0-896MB */
179	#endif
180	};
181
182
183	static void rtR0MemObjLinuxFreePages(PRTR0MEMOBJLNX pMemLnx);
184
185
186	/**
187	* Helper that converts from a RTR0PROCESS handle to a linux task.
188	*
189	* @returns The corresponding Linux task.
190	* @param R0Process IPRT ring-0 process handle.
191	*/
192	static struct task_struct *rtR0ProcessToLinuxTask(RTR0PROCESS R0Process)
193	{
194	/** @todo fix rtR0ProcessToLinuxTask!! */
195	/** @todo many (all?) callers currently assume that we return 'current'! */
196	return R0Process == RTR0ProcHandleSelf() ? current : NULL;
197	}
198
199
200	/**
201	* Compute order. Some functions allocate 2^order pages.
202	*
203	* @returns order.
204	* @param cPages Number of pages.
205	*/
206	static int rtR0MemObjLinuxOrder(size_t cPages)
207	{
208	int iOrder;
209	size_t cTmp;
210
211	for (iOrder = 0, cTmp = cPages; cTmp >>= 1; ++iOrder)
212	;
213	if (cPages & ~((size_t)1 << iOrder))
214	++iOrder;
215
216	return iOrder;
217	}
218
219
220	/**
221	* Converts from RTMEM_PROT_* to Linux PAGE_*.
222	*
223	* @returns Linux page protection constant.
224	* @param fProt The IPRT protection mask.
225	* @param fKernel Whether it applies to kernel or user space.
226	*/
227	static pgprot_t rtR0MemObjLinuxConvertProt(unsigned fProt, bool fKernel)
228	{
229	switch (fProt)
230	{
231	default:
232	AssertMsgFailed(("%#x %d\n", fProt, fKernel)); RT_FALL_THRU();
233	case RTMEM_PROT_NONE:
234	return PAGE_NONE;
235
236	case RTMEM_PROT_READ:
237	return fKernel ? PAGE_KERNEL_RO : PAGE_READONLY;
238
239	case RTMEM_PROT_WRITE:
240	case RTMEM_PROT_WRITE \| RTMEM_PROT_READ:
241	return fKernel ? PAGE_KERNEL : PAGE_SHARED;
242
243	case RTMEM_PROT_EXEC:
244	case RTMEM_PROT_EXEC \| RTMEM_PROT_READ:
245	#if defined(RT_ARCH_X86) \|\| defined(RT_ARCH_AMD64)
246	if (fKernel)
247	{
248	# if RTLNX_VER_MIN(6,6,0) \|\| RTLNX_SUSE_MAJ_PREREQ(15, 6)
249	/* In kernel 6.6 mk_pte() macro was fortified with additional
250	* check which does not allow to use our custom mask anymore
251	* (see kernel commit ae1f05a617dcbc0a732fbeba0893786cd009536c).
252	* For this particular mapping case, an existing mask PAGE_KERNEL_ROX
253	* can be used instead. PAGE_KERNEL_ROX was introduced in
254	* kernel 5.8, however, lets apply it for kernels 6.6 and newer
255	* to be on a safe side.
256	*/
257	return PAGE_KERNEL_ROX;
258	# else
259	pgprot_t fPg = MY_PAGE_KERNEL_EXEC;
260	pgprot_val(fPg) &= ~_PAGE_RW;
261	return fPg;
262	# endif
263	}
264	return PAGE_READONLY_EXEC;
265	#else
266	return fKernel ? MY_PAGE_KERNEL_EXEC : PAGE_READONLY_EXEC;
267	#endif
268
269	case RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC:
270	case RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC \| RTMEM_PROT_READ:
271	return fKernel ? MY_PAGE_KERNEL_EXEC : PAGE_SHARED_EXEC;
272	}
273	}
274
275
276	/**
277	* Worker for rtR0MemObjNativeReserveUser and rtR0MemObjNativerMapUser that creates
278	* an empty user space mapping.
279	*
280	* We acquire the mmap_sem/mmap_lock of the task!
281	*
282	* @returns Pointer to the mapping.
283	* (void *)-1 on failure.
284	* @param R3PtrFixed (RTR3PTR)-1 if anywhere, otherwise a specific location.
285	* @param cb The size of the mapping.
286	* @param uAlignment The alignment of the mapping.
287	* @param pTask The Linux task to create this mapping in.
288	* @param fProt The RTMEM_PROT_* mask.
289	*/
290	static void rtR0MemObjLinuxDoMmap(RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment, struct task_struct pTask, unsigned fProt)
291	{
292	unsigned fLnxProt;
293	unsigned long ulAddr;
294
295	Assert(pTask == current); /* do_mmap */
296	RT_NOREF_PV(pTask);
297
298	/*
299	* Convert from IPRT protection to mman.h PROT_ and call do_mmap.
300	*/
301	fProt &= (RTMEM_PROT_NONE \| RTMEM_PROT_READ \| RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC);
302	if (fProt == RTMEM_PROT_NONE)
303	fLnxProt = PROT_NONE;
304	else
305	{
306	fLnxProt = 0;
307	if (fProt & RTMEM_PROT_READ)
308	fLnxProt \|= PROT_READ;
309	if (fProt & RTMEM_PROT_WRITE)
310	fLnxProt \|= PROT_WRITE;
311	if (fProt & RTMEM_PROT_EXEC)
312	fLnxProt \|= PROT_EXEC;
313	}
314
315	if (R3PtrFixed != (RTR3PTR)-1)
316	{
317	#if RTLNX_VER_MIN(3,5,0)
318	ulAddr = vm_mmap(NULL, R3PtrFixed, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS \| MAP_FIXED, 0);
319	#else
320	LNX_MM_DOWN_WRITE(pTask->mm);
321	ulAddr = do_mmap(NULL, R3PtrFixed, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS \| MAP_FIXED, 0);
322	LNX_MM_UP_WRITE(pTask->mm);
323	#endif
324	}
325	else
326	{
327	#if RTLNX_VER_MIN(3,5,0)
328	ulAddr = vm_mmap(NULL, 0, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS, 0);
329	#else
330	LNX_MM_DOWN_WRITE(pTask->mm);
331	ulAddr = do_mmap(NULL, 0, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS, 0);
332	LNX_MM_UP_WRITE(pTask->mm);
333	#endif
334	if ( !(ulAddr & ~PAGE_MASK)
335	&& (ulAddr & (uAlignment - 1)))
336	{
337	/** @todo implement uAlignment properly... We'll probably need to make some dummy mappings to fill
338	* up alignment gaps. This is of course complicated by fragmentation (which we might have cause
339	* ourselves) and further by there begin two mmap strategies (top / bottom). */
340	/* For now, just ignore uAlignment requirements... */
341	}
342	}
343
344
345	if (ulAddr & ~PAGE_MASK) /* ~PAGE_MASK == PAGE_OFFSET_MASK */
346	return (void *)-1;
347	return (void *)ulAddr;
348	}
349
350
351	/**
352	* Worker that destroys a user space mapping.
353	* Undoes what rtR0MemObjLinuxDoMmap did.
354	*
355	* We acquire the mmap_sem/mmap_lock of the task!
356	*
357	* @param pv The ring-3 mapping.
358	* @param cb The size of the mapping.
359	* @param pTask The Linux task to destroy this mapping in.
360	*/
361	static void rtR0MemObjLinuxDoMunmap(void pv, size_t cb, struct task_struct pTask)
362	{
363	#if RTLNX_VER_MIN(3,5,0)
364	Assert(pTask == current); RT_NOREF_PV(pTask);
365	vm_munmap((unsigned long)pv, cb);
366	#elif defined(USE_RHEL4_MUNMAP)
367	LNX_MM_DOWN_WRITE(pTask->mm);
368	do_munmap(pTask->mm, (unsigned long)pv, cb, 0); /* should it be 1 or 0? */
369	LNX_MM_UP_WRITE(pTask->mm);
370	#else
371	LNX_MM_DOWN_WRITE(pTask->mm);
372	do_munmap(pTask->mm, (unsigned long)pv, cb);
373	LNX_MM_UP_WRITE(pTask->mm);
374	#endif
375	}
376
377
378	/**
379	* Internal worker that allocates physical pages and creates the memory object for them.
380	*
381	* @returns IPRT status code.
382	* @param ppMemLnx Where to store the memory object pointer.
383	* @param enmType The object type.
384	* @param cb The number of bytes to allocate.
385	* @param uAlignment The alignment of the physical memory.
386	* Only valid if fContiguous == true, ignored otherwise.
387	* @param fFlagsLnx The page allocation flags (GPFs).
388	* @param fContiguous Whether the allocation must be contiguous.
389	* @param fExecutable Whether the memory must be executable.
390	* @param rcNoMem What to return when we're out of pages.
391	* @param pszTag Allocation tag used for statistics and such.
392	*/
393	static int rtR0MemObjLinuxAllocPages(PRTR0MEMOBJLNX *ppMemLnx, RTR0MEMOBJTYPE enmType, size_t cb,
394	size_t uAlignment, gfp_t fFlagsLnx, bool fContiguous, bool fExecutable, int rcNoMem,
395	const char *pszTag)
396	{
397	size_t iPage;
398	size_t const cPages = cb >> PAGE_SHIFT;
399	struct page *paPages;
400
401	/*
402	* Allocate a memory object structure that's large enough to contain
403	* the page pointer array.
404	*/
405	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_UOFFSETOF_DYN(RTR0MEMOBJLNX, apPages[cPages]), enmType,
406	NULL, cb, pszTag);
407	if (!pMemLnx)
408	return VERR_NO_MEMORY;
409	pMemLnx->Core.fFlags \|= RTR0MEMOBJ_FLAGS_UNINITIALIZED_AT_ALLOC;
410	pMemLnx->cPages = cPages;
411
412	if (cPages > 255)
413	{
414	# ifdef __GFP_REPEAT
415	/* Try hard to allocate the memory, but the allocation attempt might fail. */
416	fFlagsLnx \|= __GFP_REPEAT;
417	# endif
418	# ifdef __GFP_NOMEMALLOC
419	/* Introduced with Linux 2.6.12: Don't use emergency reserves */
420	fFlagsLnx \|= __GFP_NOMEMALLOC;
421	# endif
422	}
423
424	/*
425	* Allocate the pages.
426	* For small allocations we'll try contiguous first and then fall back on page by page.
427	*/
428	#if RTLNX_VER_MIN(2,4,22)
429	if ( fContiguous
430	\|\| cb <= PAGE_SIZE * 2)
431	{
432	# ifdef VBOX_USE_INSERT_PAGE
433	paPages = alloc_pages(fFlagsLnx \| __GFP_COMP \| __GFP_NOWARN, rtR0MemObjLinuxOrder(cPages));
434	# else
435	paPages = alloc_pages(fFlagsLnx \| __GFP_NOWARN, rtR0MemObjLinuxOrder(cPages));
436	# endif
437	if (paPages)
438	{
439	fContiguous = true;
440	for (iPage = 0; iPage < cPages; iPage++)
441	pMemLnx->apPages[iPage] = &paPages[iPage];
442	}
443	else if (fContiguous)
444	{
445	rtR0MemObjDelete(&pMemLnx->Core);
446	return rcNoMem;
447	}
448	}
449
450	if (!fContiguous)
451	{
452	/** @todo Try use alloc_pages_bulk_array when available, it should be faster
453	* than a alloc_page loop. Put it in #ifdefs similar to
454	* IPRT_USE_APPLY_TO_PAGE_RANGE_FOR_EXEC. */
455	for (iPage = 0; iPage < cPages; iPage++)
456	{
457	pMemLnx->apPages[iPage] = alloc_page(fFlagsLnx \| __GFP_NOWARN);
458	if (RT_UNLIKELY(!pMemLnx->apPages[iPage]))
459	{
460	while (iPage-- > 0)
461	__free_page(pMemLnx->apPages[iPage]);
462	rtR0MemObjDelete(&pMemLnx->Core);
463	return rcNoMem;
464	}
465	}
466	}
467
468	#else /* < 2.4.22 */
469	/** @todo figure out why we didn't allocate page-by-page on 2.4.21 and older... */
470	paPages = alloc_pages(fFlagsLnx, rtR0MemObjLinuxOrder(cPages));
471	if (!paPages)
472	{
473	rtR0MemObjDelete(&pMemLnx->Core);
474	return rcNoMem;
475	}
476	for (iPage = 0; iPage < cPages; iPage++)
477	{
478	pMemLnx->apPages[iPage] = &paPages[iPage];
479	if (fExecutable)
480	MY_SET_PAGES_EXEC(pMemLnx->apPages[iPage], 1);
481	if (PageHighMem(pMemLnx->apPages[iPage]))
482	BUG();
483	}
484
485	fContiguous = true;
486	#endif /* < 2.4.22 */
487	pMemLnx->fContiguous = fContiguous;
488	pMemLnx->fExecutable = fExecutable;
489
490	#if RTLNX_VER_MAX(4,5,0)
491	/*
492	* Reserve the pages.
493	*
494	* Linux >= 4.5 with CONFIG_DEBUG_VM panics when setting PG_reserved on compound
495	* pages. According to Michal Hocko this shouldn't be necessary anyway because
496	* as pages which are not on the LRU list are never evictable.
497	*/
498	for (iPage = 0; iPage < cPages; iPage++)
499	SetPageReserved(pMemLnx->apPages[iPage]);
500	#endif
501
502	/*
503	* Note that the physical address of memory allocated with alloc_pages(flags, order)
504	* is always 2^(PAGE_SHIFT+order)-aligned.
505	*/
506	if ( fContiguous
507	&& uAlignment > PAGE_SIZE)
508	{
509	/*
510	* Check for alignment constraints. The physical address of memory allocated with
511	* alloc_pages(flags, order) is always 2^(PAGE_SHIFT+order)-aligned.
512	*/
513	if (RT_UNLIKELY(page_to_phys(pMemLnx->apPages[0]) & (uAlignment - 1)))
514	{
515	/*
516	* This should never happen!
517	*/
518	printk("rtR0MemObjLinuxAllocPages(cb=0x%lx, uAlignment=0x%lx): alloc_pages(..., %d) returned physical memory at 0x%lx!\n",
519	(unsigned long)cb, (unsigned long)uAlignment, rtR0MemObjLinuxOrder(cPages), (unsigned long)page_to_phys(pMemLnx->apPages[0]));
520	rtR0MemObjLinuxFreePages(pMemLnx);
521	return rcNoMem;
522	}
523	}
524
525	*ppMemLnx = pMemLnx;
526	return VINF_SUCCESS;
527	}
528
529
530	/**
531	* Frees the physical pages allocated by the rtR0MemObjLinuxAllocPages() call.
532	*
533	* This method does NOT free the object.
534	*
535	* @param pMemLnx The object which physical pages should be freed.
536	*/
537	static void rtR0MemObjLinuxFreePages(PRTR0MEMOBJLNX pMemLnx)
538	{
539	size_t iPage = pMemLnx->cPages;
540	if (iPage > 0)
541	{
542	/*
543	* Restore the page flags.
544	*/
545	while (iPage-- > 0)
546	{
547	#if RTLNX_VER_MAX(4,5,0)
548	/* See SetPageReserved() in rtR0MemObjLinuxAllocPages() */
549	ClearPageReserved(pMemLnx->apPages[iPage]);
550	#endif
551	#if RTLNX_VER_MAX(2,4,22)
552	if (pMemLnx->fExecutable)
553	MY_SET_PAGES_NOEXEC(pMemLnx->apPages[iPage], 1);
554	#endif
555	}
556
557	/*
558	* Free the pages.
559	*/
560	#if RTLNX_VER_MIN(2,4,22)
561	if (!pMemLnx->fContiguous)
562	{
563	iPage = pMemLnx->cPages;
564	while (iPage-- > 0)
565	__free_page(pMemLnx->apPages[iPage]);
566	}
567	else
568	#endif
569	__free_pages(pMemLnx->apPages[0], rtR0MemObjLinuxOrder(pMemLnx->cPages));
570
571	pMemLnx->cPages = 0;
572	}
573	}
574
575
576	#ifdef IPRT_USE_APPLY_TO_PAGE_RANGE_FOR_EXEC
577	/**
578	* User data passed to the apply_to_page_range() callback.
579	*/
580	typedef struct LNXAPPLYPGRANGE
581	{
582	/** Pointer to the memory object. */
583	PRTR0MEMOBJLNX pMemLnx;
584	/** The page protection flags to apply. */
585	pgprot_t fPg;
586	} LNXAPPLYPGRANGE;
587	/** Pointer to the user data. */
588	typedef LNXAPPLYPGRANGE *PLNXAPPLYPGRANGE;
589	/** Pointer to the const user data. */
590	typedef const LNXAPPLYPGRANGE *PCLNXAPPLYPGRANGE;
591
592	/**
593	* Callback called in apply_to_page_range().
594	*
595	* @returns Linux status code.
596	* @param pPte Pointer to the page table entry for the given address.
597	* @param uAddr The address to apply the new protection to.
598	* @param pvUser The opaque user data.
599	*/
600	static int rtR0MemObjLinuxApplyPageRange(pte_t pPte, unsigned long uAddr, void pvUser)
601	{
602	PCLNXAPPLYPGRANGE pArgs = (PCLNXAPPLYPGRANGE)pvUser;
603	PRTR0MEMOBJLNX pMemLnx = pArgs->pMemLnx;
604	size_t idxPg = (uAddr - (unsigned long)pMemLnx->Core.pv) >> PAGE_SHIFT;
605
606	set_pte(pPte, mk_pte(pMemLnx->apPages[idxPg], pArgs->fPg));
607	return 0;
608	}
609	#endif
610
611
612	/**
613	* Maps the allocation into ring-0.
614	*
615	* This will update the RTR0MEMOBJLNX::Core.pv and RTR0MEMOBJ::fMappedToRing0 members.
616	*
617	* Contiguous mappings that isn't in 'high' memory will already be mapped into kernel
618	* space, so we'll use that mapping if possible. If execute access is required, we'll
619	* play safe and do our own mapping.
620	*
621	* @returns IPRT status code.
622	* @param pMemLnx The linux memory object to map.
623	* @param fExecutable Whether execute access is required.
624	*/
625	static int rtR0MemObjLinuxVMap(PRTR0MEMOBJLNX pMemLnx, bool fExecutable)
626	{
627	int rc = VINF_SUCCESS;
628
629	/*
630	* Choose mapping strategy.
631	*/
632	bool fMustMap = fExecutable
633	\|\| !pMemLnx->fContiguous;
634	if (!fMustMap)
635	{
636	size_t iPage = pMemLnx->cPages;
637	while (iPage-- > 0)
638	if (PageHighMem(pMemLnx->apPages[iPage]))
639	{
640	fMustMap = true;
641	break;
642	}
643	}
644
645	Assert(!pMemLnx->Core.pv);
646	Assert(!pMemLnx->fMappedToRing0);
647
648	if (fMustMap)
649	{
650	/*
651	* Use vmap - 2.4.22 and later.
652	*/
653	#if RTLNX_VER_MIN(2,4,22) && (defined(RT_ARCH_AMD64) \|\| defined(RT_ARCH_X86) \|\| defined(RT_ARCH_ARM64))
654	pgprot_t fPg;
655	# if defined(RT_ARCH_ARM64)
656	/* ARM64 architecture has no _PAGE_NX, _PAGE_PRESENT and _PAGE_RW flags.
657	* Closest alternatives would be PTE_PXN, PTE_UXN, PROT_DEFAULT and PTE_WRITE. */
658	pgprot_val(fPg) = _PAGE_KERNEL; /* (PROT_DEFAULT \| PTE_PXN \| PTE_UXN \| PTE_WRITE \| PTE_ATTRINDX(MT_NORMAL). */
659	# else /* !RT_ARCH_ARM64 */
660	pgprot_val(fPg) = _PAGE_PRESENT \| _PAGE_RW;
661	# ifdef _PAGE_NX
662	if (!fExecutable)
663	pgprot_val(fPg) \|= _PAGE_NX;
664	# endif
665	# endif /* RT_ARCH_ARM64 */
666
667	# ifdef IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
668	if (fExecutable)
669	{
670	# if RTLNX_VER_MIN(3,2,51)
671	pte_t papPtes = (pte_t )kmalloc_array(pMemLnx->cPages, sizeof(papPtes[0]), GFP_KERNEL);
672	# else
673	pte_t papPtes = (pte_t )kmalloc(pMemLnx->cPages * sizeof(papPtes[0]), GFP_KERNEL);
674	# endif
675	if (papPtes)
676	{
677	pMemLnx->pArea = alloc_vm_area(pMemLnx->Core.cb, papPtes); /* Note! pArea->nr_pages is not set. */
678	if (pMemLnx->pArea)
679	{
680	size_t i;
681	Assert(pMemLnx->pArea->size >= pMemLnx->Core.cb); /* Note! includes guard page. */
682	Assert(pMemLnx->pArea->addr);
683	# if !defined(RT_ARCH_ARM64) && defined(_PAGE_NX)
684	pgprot_val(fPg) \|= _PAGE_NX; /* Uses RTR0MemObjProtect to clear NX when memory ready, W^X fashion. */
685	# endif
686	pMemLnx->papPtesForArea = papPtes;
687	for (i = 0; i < pMemLnx->cPages; i++)
688	*papPtes[i] = mk_pte(pMemLnx->apPages[i], fPg);
689	pMemLnx->Core.pv = pMemLnx->pArea->addr;
690	pMemLnx->fMappedToRing0 = true;
691	}
692	else
693	{
694	kfree(papPtes);
695	rc = VERR_MAP_FAILED;
696	}
697	}
698	else
699	rc = VERR_MAP_FAILED;
700	}
701	else
702	# endif
703	{
704	# if !defined(RT_ARCH_ARM64) && defined(IPRT_USE_APPLY_TO_PAGE_RANGE_FOR_EXEC)
705	if (fExecutable)
706	pgprot_val(fPg) \|= _PAGE_NX; /* Uses RTR0MemObjProtect to clear NX when memory ready, W^X fashion. */
707	# endif
708
709	# ifdef VM_MAP
710	pMemLnx->Core.pv = vmap(&pMemLnx->apPages[0], pMemLnx->cPages, VM_MAP, fPg);
711	# else
712	pMemLnx->Core.pv = vmap(&pMemLnx->apPages[0], pMemLnx->cPages, VM_ALLOC, fPg);
713	# endif
714	if (pMemLnx->Core.pv)
715	pMemLnx->fMappedToRing0 = true;
716	else
717	rc = VERR_MAP_FAILED;
718	}
719	#else /* < 2.4.22 */
720	rc = VERR_NOT_SUPPORTED;
721	#endif
722	}
723	else
724	{
725	/*
726	* Use the kernel RAM mapping.
727	*/
728	pMemLnx->Core.pv = phys_to_virt(page_to_phys(pMemLnx->apPages[0]));
729	Assert(pMemLnx->Core.pv);
730	}
731
732	return rc;
733	}
734
735
736	/**
737	* Undoes what rtR0MemObjLinuxVMap() did.
738	*
739	* @param pMemLnx The linux memory object.
740	*/
741	static void rtR0MemObjLinuxVUnmap(PRTR0MEMOBJLNX pMemLnx)
742	{
743	#if RTLNX_VER_MIN(2,4,22)
744	# ifdef IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
745	if (pMemLnx->pArea)
746	{
747	# if 0
748	pte_t **papPtes = pMemLnx->papPtesForArea;
749	size_t i;
750	for (i = 0; i < pMemLnx->cPages; i++)
751	*papPtes[i] = 0;
752	# endif
753	free_vm_area(pMemLnx->pArea);
754	kfree(pMemLnx->papPtesForArea);
755	pMemLnx->pArea = NULL;
756	pMemLnx->papPtesForArea = NULL;
757	}
758	else
759	# endif
760	if (pMemLnx->fMappedToRing0)
761	{
762	Assert(pMemLnx->Core.pv);
763	vunmap(pMemLnx->Core.pv);
764	pMemLnx->fMappedToRing0 = false;
765	}
766	#else /* < 2.4.22 */
767	Assert(!pMemLnx->fMappedToRing0);
768	#endif
769	pMemLnx->Core.pv = NULL;
770	}
771
772
773	DECLHIDDEN(int) rtR0MemObjNativeFree(RTR0MEMOBJ pMem)
774	{
775	IPRT_LINUX_SAVE_EFL_AC();
776	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
777
778	/*
779	* Release any memory that we've allocated or locked.
780	*/
781	switch (pMemLnx->Core.enmType)
782	{
783	case RTR0MEMOBJTYPE_PAGE:
784	case RTR0MEMOBJTYPE_LOW:
785	case RTR0MEMOBJTYPE_CONT:
786	case RTR0MEMOBJTYPE_PHYS:
787	case RTR0MEMOBJTYPE_PHYS_NC:
788	rtR0MemObjLinuxVUnmap(pMemLnx);
789	rtR0MemObjLinuxFreePages(pMemLnx);
790	break;
791
792	case RTR0MEMOBJTYPE_LARGE_PAGE:
793	{
794	uint32_t const cLargePages = pMemLnx->Core.cb >> (pMemLnx->cLargePageOrder + PAGE_SHIFT);
795	uint32_t iLargePage;
796	for (iLargePage = 0; iLargePage < cLargePages; iLargePage++)
797	__free_pages(pMemLnx->apPages[iLargePage << pMemLnx->cLargePageOrder], pMemLnx->cLargePageOrder);
798	pMemLnx->cPages = 0;
799
800	#ifdef IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
801	Assert(!pMemLnx->pArea);
802	Assert(!pMemLnx->papPtesForArea);
803	#endif
804	break;
805	}
806
807	case RTR0MEMOBJTYPE_LOCK:
808	if (pMemLnx->Core.u.Lock.R0Process != NIL_RTR0PROCESS)
809	{
810	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
811	size_t iPage;
812	Assert(pTask);
813	if (pTask && pTask->mm)
814	LNX_MM_DOWN_READ(pTask->mm);
815
816	iPage = pMemLnx->cPages;
817	while (iPage-- > 0)
818	{
819	if (!PageReserved(pMemLnx->apPages[iPage]))
820	SetPageDirty(pMemLnx->apPages[iPage]);
821	#if RTLNX_VER_MIN(4,6,0)
822	put_page(pMemLnx->apPages[iPage]);
823	#else
824	page_cache_release(pMemLnx->apPages[iPage]);
825	#endif
826	}
827
828	if (pTask && pTask->mm)
829	LNX_MM_UP_READ(pTask->mm);
830	}
831	/* else: kernel memory - nothing to do here. */
832	break;
833
834	case RTR0MEMOBJTYPE_RES_VIRT:
835	Assert(pMemLnx->Core.pv);
836	if (pMemLnx->Core.u.ResVirt.R0Process != NIL_RTR0PROCESS)
837	{
838	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
839	Assert(pTask);
840	if (pTask && pTask->mm)
841	rtR0MemObjLinuxDoMunmap(pMemLnx->Core.pv, pMemLnx->Core.cb, pTask);
842	}
843	else
844	{
845	vunmap(pMemLnx->Core.pv);
846
847	Assert(pMemLnx->cPages == 1 && pMemLnx->apPages[0] != NULL);
848	__free_page(pMemLnx->apPages[0]);
849	pMemLnx->apPages[0] = NULL;
850	pMemLnx->cPages = 0;
851	}
852	pMemLnx->Core.pv = NULL;
853	break;
854
855	case RTR0MEMOBJTYPE_MAPPING:
856	Assert(pMemLnx->cPages == 0); Assert(pMemLnx->Core.pv);
857	if (pMemLnx->Core.u.ResVirt.R0Process != NIL_RTR0PROCESS)
858	{
859	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
860	Assert(pTask);
861	if (pTask && pTask->mm)
862	rtR0MemObjLinuxDoMunmap(pMemLnx->Core.pv, pMemLnx->Core.cb, pTask);
863	}
864	else
865	vunmap(pMemLnx->Core.pv);
866	pMemLnx->Core.pv = NULL;
867	break;
868
869	default:
870	AssertMsgFailed(("enmType=%d\n", pMemLnx->Core.enmType));
871	return VERR_INTERNAL_ERROR;
872	}
873	IPRT_LINUX_RESTORE_EFL_ONLY_AC();
874	return VINF_SUCCESS;
875	}
876
877
878	DECLHIDDEN(int) rtR0MemObjNativeAllocPage(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable, const char *pszTag)
879	{
880	IPRT_LINUX_SAVE_EFL_AC();
881	PRTR0MEMOBJLNX pMemLnx;
882	int rc;
883
884	#if RTLNX_VER_MIN(2,4,22)
885	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_PAGE, cb, PAGE_SIZE, GFP_HIGHUSER,
886	false /* non-contiguous */, fExecutable, VERR_NO_MEMORY, pszTag);
887	#else
888	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_PAGE, cb, PAGE_SIZE, GFP_USER,
889	false /* non-contiguous */, fExecutable, VERR_NO_MEMORY, pszTag);
890	#endif
891	if (RT_SUCCESS(rc))
892	{
893	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
894	if (RT_SUCCESS(rc))
895	{
896	*ppMem = &pMemLnx->Core;
897	IPRT_LINUX_RESTORE_EFL_AC();
898	return rc;
899	}
900
901	rtR0MemObjLinuxFreePages(pMemLnx);
902	rtR0MemObjDelete(&pMemLnx->Core);
903	}
904
905	IPRT_LINUX_RESTORE_EFL_AC();
906	return rc;
907	}
908
909
910	DECLHIDDEN(int) rtR0MemObjNativeAllocLarge(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, size_t cbLargePage, uint32_t fFlags,
911	const char *pszTag)
912	{
913	#ifdef GFP_TRANSHUGE
914	/*
915	* Allocate a memory object structure that's large enough to contain
916	* the page pointer array.
917	*/
918	# ifdef __GFP_MOVABLE
919	unsigned const fGfp = (GFP_TRANSHUGE \| __GFP_ZERO) & ~__GFP_MOVABLE;
920	# else
921	unsigned const fGfp = (GFP_TRANSHUGE \| __GFP_ZERO);
922	# endif
923	size_t const cPagesPerLarge = cbLargePage >> PAGE_SHIFT;
924	unsigned const cLargePageOrder = rtR0MemObjLinuxOrder(cPagesPerLarge);
925	size_t const cLargePages = cb >> (cLargePageOrder + PAGE_SHIFT);
926	size_t const cPages = cb >> PAGE_SHIFT;
927	PRTR0MEMOBJLNX pMemLnx;
928
929	Assert(RT_BIT_64(cLargePageOrder + PAGE_SHIFT) == cbLargePage);
930	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_UOFFSETOF_DYN(RTR0MEMOBJLNX, apPages[cPages]),
931	RTR0MEMOBJTYPE_LARGE_PAGE, NULL, cb, pszTag);
932	if (pMemLnx)
933	{
934	size_t iLargePage;
935
936	pMemLnx->Core.fFlags \|= RTR0MEMOBJ_FLAGS_ZERO_AT_ALLOC;
937	pMemLnx->cLargePageOrder = cLargePageOrder;
938	pMemLnx->cPages = cPages;
939
940	/*
941	* Allocate the requested number of large pages.
942	*/
943	for (iLargePage = 0; iLargePage < cLargePages; iLargePage++)
944	{
945	struct page *paPages = alloc_pages(fGfp, cLargePageOrder);
946	if (paPages)
947	{
948	size_t const iPageBase = iLargePage << cLargePageOrder;
949	size_t iPage = cPagesPerLarge;
950	while (iPage-- > 0)
951	pMemLnx->apPages[iPageBase + iPage] = &paPages[iPage];
952	}
953	else
954	{
955	/*Log(("rtR0MemObjNativeAllocLarge: cb=%#zx cPages=%#zx cLargePages=%#zx cLargePageOrder=%u cPagesPerLarge=%#zx iLargePage=%#zx -> failed!\n",
956	cb, cPages, cLargePages, cLargePageOrder, cPagesPerLarge, iLargePage, paPages));*/
957	while (iLargePage-- > 0)
958	__free_pages(pMemLnx->apPages[iLargePage << (cLargePageOrder - PAGE_SHIFT)], cLargePageOrder);
959	rtR0MemObjDelete(&pMemLnx->Core);
960	return VERR_NO_MEMORY;
961	}
962	}
963	*ppMem = &pMemLnx->Core;
964	return VINF_SUCCESS;
965	}
966	return VERR_NO_MEMORY;
967
968	#else
969	/*
970	* We don't call rtR0MemObjFallbackAllocLarge here as it can be a really
971	* bad idea to trigger the swap daemon and whatnot. So, just fail.
972	*/
973	RT_NOREF(ppMem, cb, cbLargePage, fFlags, pszTag);
974	return VERR_NOT_SUPPORTED;
975	#endif
976	}
977
978
979	DECLHIDDEN(int) rtR0MemObjNativeAllocLow(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable, const char *pszTag)
980	{
981	IPRT_LINUX_SAVE_EFL_AC();
982	PRTR0MEMOBJLNX pMemLnx;
983	int rc;
984
985	/* Try to avoid GFP_DMA. GFM_DMA32 was introduced with Linux 2.6.15. */
986	#if (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
987	/* ZONE_DMA32: 0-4GB */
988	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_DMA32,
989	false /* non-contiguous */, fExecutable, VERR_NO_LOW_MEMORY, pszTag);
990	if (RT_FAILURE(rc))
991	#endif
992	#ifdef RT_ARCH_AMD64
993	/* ZONE_DMA: 0-16MB */
994	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_DMA,
995	false /* non-contiguous */, fExecutable, VERR_NO_LOW_MEMORY, pszTag);
996	#else
997	# ifdef CONFIG_X86_PAE
998	# endif
999	/* ZONE_NORMAL: 0-896MB */
1000	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_USER,
1001	false /* non-contiguous */, fExecutable, VERR_NO_LOW_MEMORY, pszTag);
1002	#endif
1003	if (RT_SUCCESS(rc))
1004	{
1005	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
1006	if (RT_SUCCESS(rc))
1007	{
1008	*ppMem = &pMemLnx->Core;
1009	IPRT_LINUX_RESTORE_EFL_AC();
1010	return rc;
1011	}
1012
1013	rtR0MemObjLinuxFreePages(pMemLnx);
1014	rtR0MemObjDelete(&pMemLnx->Core);
1015	}
1016
1017	IPRT_LINUX_RESTORE_EFL_AC();
1018	return rc;
1019	}
1020
1021
1022	DECLHIDDEN(int) rtR0MemObjNativeAllocCont(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest,
1023	bool fExecutable, const char *pszTag)
1024	{
1025	IPRT_LINUX_SAVE_EFL_AC();
1026	PRTR0MEMOBJLNX pMemLnx;
1027	int rc;
1028	uint32_t idxZone;
1029
1030	/*
1031	* The last zone must be able to satisfy the PhysHighest requirement or there
1032	* will be no zone at all.
1033	*/
1034	if (g_aZones[RT_ELEMENTS(g_aZones) - 1].PhysHighest > PhysHighest)
1035	{
1036	IPRT_LINUX_RESTORE_EFL_AC();
1037	AssertMsgFailedReturn(("No zone can satisfy PhysHighest=%RHp!\n", PhysHighest),
1038	VERR_NO_CONT_MEMORY);
1039	}
1040
1041	/* Find the first zone matching our PhysHighest requirement. */
1042	idxZone = 0;
1043	for (;;)
1044	{
1045	if (g_aZones[idxZone].PhysHighest <= PhysHighest)
1046	break; /* We found a zone satisfying the requirement. */
1047	idxZone++;
1048	}
1049
1050	/* Now try to allocate pages from all the left zones until one succeeds. */
1051	for (;;)
1052	{
1053	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, PAGE_SIZE, g_aZones[idxZone].fGfp,
1054	true /* contiguous */, fExecutable, VERR_NO_CONT_MEMORY, pszTag);
1055	idxZone++;
1056	if (RT_SUCCESS(rc) \|\| idxZone == RT_ELEMENTS(g_aZones))
1057	break;
1058	}
1059	if (RT_SUCCESS(rc))
1060	{
1061	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
1062	if (RT_SUCCESS(rc))
1063	{
1064	#if defined(RT_STRICT)
1065	size_t iPage = pMemLnx->cPages;
1066	while (iPage-- > 0)
1067	Assert(page_to_phys(pMemLnx->apPages[iPage]) < PhysHighest);
1068	#endif
1069	pMemLnx->Core.u.Cont.Phys = page_to_phys(pMemLnx->apPages[0]);
1070	*ppMem = &pMemLnx->Core;
1071	IPRT_LINUX_RESTORE_EFL_AC();
1072	return rc;
1073	}
1074
1075	rtR0MemObjLinuxFreePages(pMemLnx);
1076	rtR0MemObjDelete(&pMemLnx->Core);
1077	}
1078
1079	IPRT_LINUX_RESTORE_EFL_AC();
1080	return rc;
1081	}
1082
1083
1084	/**
1085	* Worker for rtR0MemObjLinuxAllocPhysSub that tries one allocation strategy.
1086	*
1087	* @returns IPRT status code.
1088	* @param ppMemLnx Where to
1089	* @param enmType The object type.
1090	* @param cb The size of the allocation.
1091	* @param uAlignment The alignment of the physical memory.
1092	* Only valid for fContiguous == true, ignored otherwise.
1093	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
1094	* @param pszTag Allocation tag used for statistics and such.
1095	* @param fGfp The Linux GFP flags to use for the allocation.
1096	*/
1097	static int rtR0MemObjLinuxAllocPhysSub2(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType,
1098	size_t cb, size_t uAlignment, RTHCPHYS PhysHighest, const char *pszTag, gfp_t fGfp)
1099	{
1100	PRTR0MEMOBJLNX pMemLnx;
1101	int rc = rtR0MemObjLinuxAllocPages(&pMemLnx, enmType, cb, uAlignment, fGfp,
1102	enmType == RTR0MEMOBJTYPE_PHYS /* contiguous / non-contiguous */,
1103	false /fExecutable/, VERR_NO_PHYS_MEMORY, pszTag);
1104	if (RT_FAILURE(rc))
1105	return rc;
1106
1107	/*
1108	* Check the addresses if necessary. (Can be optimized a bit for PHYS.)
1109	*/
1110	if (PhysHighest != NIL_RTHCPHYS)
1111	{
1112	size_t iPage = pMemLnx->cPages;
1113	while (iPage-- > 0)
1114	if (page_to_phys(pMemLnx->apPages[iPage]) > PhysHighest)
1115	{
1116	rtR0MemObjLinuxFreePages(pMemLnx);
1117	rtR0MemObjDelete(&pMemLnx->Core);
1118	return VERR_NO_MEMORY;
1119	}
1120	}
1121
1122	/*
1123	* Complete the object.
1124	*/
1125	if (enmType == RTR0MEMOBJTYPE_PHYS)
1126	{
1127	pMemLnx->Core.u.Phys.PhysBase = page_to_phys(pMemLnx->apPages[0]);
1128	pMemLnx->Core.u.Phys.fAllocated = true;
1129	}
1130	*ppMem = &pMemLnx->Core;
1131	return rc;
1132	}
1133
1134
1135	/**
1136	* Worker for rtR0MemObjNativeAllocPhys and rtR0MemObjNativeAllocPhysNC.
1137	*
1138	* @returns IPRT status code.
1139	* @param ppMem Where to store the memory object pointer on success.
1140	* @param enmType The object type.
1141	* @param cb The size of the allocation.
1142	* @param uAlignment The alignment of the physical memory.
1143	* Only valid for enmType == RTR0MEMOBJTYPE_PHYS, ignored otherwise.
1144	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
1145	* @param pszTag Allocation tag used for statistics and such.
1146	*/
1147	static int rtR0MemObjLinuxAllocPhysSub(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType,
1148	size_t cb, size_t uAlignment, RTHCPHYS PhysHighest, const char *pszTag)
1149	{
1150	int rc;
1151	IPRT_LINUX_SAVE_EFL_AC();
1152
1153	/*
1154	* There are two clear cases and that's the <=16MB and anything-goes ones.
1155	* When the physical address limit is somewhere in-between those two we'll
1156	* just have to try, starting with HIGHUSER and working our way thru the
1157	* different types, hoping we'll get lucky.
1158	*
1159	* We should probably move this physical address restriction logic up to
1160	* the page alloc function as it would be more efficient there. But since
1161	* we don't expect this to be a performance issue just yet it can wait.
1162	*/
1163	if (PhysHighest == NIL_RTHCPHYS)
1164	/* ZONE_HIGHMEM: the whole physical memory */
1165	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_HIGHUSER);
1166	else if (PhysHighest <= _1M * 16)
1167	/* ZONE_DMA: 0-16MB */
1168	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_DMA);
1169	else
1170	{
1171	rc = VERR_NO_MEMORY;
1172	if (RT_FAILURE(rc))
1173	/* ZONE_HIGHMEM: the whole physical memory */
1174	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_HIGHUSER);
1175	if (RT_FAILURE(rc))
1176	/* ZONE_NORMAL: 0-896MB */
1177	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_USER);
1178	#ifdef GFP_DMA32
1179	if (RT_FAILURE(rc))
1180	/* ZONE_DMA32: 0-4GB */
1181	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_DMA32);
1182	#endif
1183	if (RT_FAILURE(rc))
1184	/* ZONE_DMA: 0-16MB */
1185	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_DMA);
1186	}
1187	IPRT_LINUX_RESTORE_EFL_AC();
1188	return rc;
1189	}
1190
1191
1192	/**
1193	* Translates a kernel virtual address to a linux page structure by walking the
1194	* page tables.
1195	*
1196	* @note We do assume that the page tables will not change as we are walking
1197	* them. This assumption is rather forced by the fact that I could not
1198	* immediately see any way of preventing this from happening. So, we
1199	* take some extra care when accessing them.
1200	*
1201	* Because of this, we don't want to use this function on memory where
1202	* attribute changes to nearby pages is likely to cause large pages to
1203	* be used or split up. So, don't use this for the linear mapping of
1204	* physical memory.
1205	*
1206	* @returns Pointer to the page structur or NULL if it could not be found.
1207	* @param pv The kernel virtual address.
1208	*/
1209	RTDECL(struct page ) rtR0MemObjLinuxVirtToPage(void pv)
1210	{
1211	#if defined(RT_ARCH_AMD64) \|\| defined(RT_ARCH_X86)
1212	unsigned long ulAddr = (unsigned long)pv;
1213	unsigned long pfn;
1214	struct page *pPage;
1215	pte_t *pEntry;
1216	union
1217	{
1218	pgd_t Global;
1219	# if RTLNX_VER_MIN(4,12,0)
1220	p4d_t Four;
1221	# endif
1222	# if RTLNX_VER_MIN(2,6,11)
1223	pud_t Upper;
1224	# endif
1225	pmd_t Middle;
1226	pte_t Entry;
1227	} u;
1228
1229	/* Should this happen in a situation this code will be called in? And if
1230	* so, can it change under our feet? See also
1231	* "Documentation/vm/active_mm.txt" in the kernel sources. */
1232	if (RT_UNLIKELY(!current->active_mm))
1233	return NULL;
1234	u.Global = *pgd_offset(current->active_mm, ulAddr);
1235	if (RT_UNLIKELY(pgd_none(u.Global)))
1236	return NULL;
1237	# if RTLNX_VER_MIN(2,6,11)
1238	# if RTLNX_VER_MIN(4,12,0)
1239	u.Four = *p4d_offset(&u.Global, ulAddr);
1240	if (RT_UNLIKELY(p4d_none(u.Four)))
1241	return NULL;
1242	# if RTLNX_VER_MIN(5,6,0)
1243	if (p4d_leaf(u.Four))
1244	# else
1245	if (p4d_large(u.Four))
1246	# endif
1247	{
1248	pPage = p4d_page(u.Four);
1249	AssertReturn(pPage, NULL);
1250	pfn = page_to_pfn(pPage); /* doing the safe way... */
1251	AssertCompile(P4D_SHIFT - PAGE_SHIFT < 31);
1252	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (P4D_SHIFT - PAGE_SHIFT)) - 1);
1253	return pfn_to_page(pfn);
1254	}
1255	u.Upper = *pud_offset(&u.Four, ulAddr);
1256	# else /* < 4.12 */
1257	u.Upper = *pud_offset(&u.Global, ulAddr);
1258	# endif /* < 4.12 */
1259	if (RT_UNLIKELY(pud_none(u.Upper)))
1260	return NULL;
1261	# if RTLNX_VER_MIN(2,6,25)
1262	# if RTLNX_VER_MIN(5,6,0)
1263	if (pud_leaf(u.Upper))
1264	# else
1265	if (pud_large(u.Upper))
1266	# endif
1267	{
1268	pPage = pud_page(u.Upper);
1269	AssertReturn(pPage, NULL);
1270	pfn = page_to_pfn(pPage); /* doing the safe way... */
1271	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (PUD_SHIFT - PAGE_SHIFT)) - 1);
1272	return pfn_to_page(pfn);
1273	}
1274	# endif
1275	u.Middle = *pmd_offset(&u.Upper, ulAddr);
1276	# else /* < 2.6.11 */
1277	u.Middle = *pmd_offset(&u.Global, ulAddr);
1278	# endif /* < 2.6.11 */
1279	if (RT_UNLIKELY(pmd_none(u.Middle)))
1280	return NULL;
1281	# if RTLNX_VER_MIN(2,6,0)
1282	# if RTLNX_VER_MIN(5,6,0)
1283	if (pmd_leaf(u.Middle))
1284	# else
1285	if (pmd_large(u.Middle))
1286	# endif
1287	{
1288	pPage = pmd_page(u.Middle);
1289	AssertReturn(pPage, NULL);
1290	pfn = page_to_pfn(pPage); /* doing the safe way... */
1291	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (PMD_SHIFT - PAGE_SHIFT)) - 1);
1292	return pfn_to_page(pfn);
1293	}
1294	# endif
1295
1296	# if RTLNX_VER_MIN(6,5,0) \|\| RTLNX_RHEL_RANGE(9,4, 9,99)
1297	pEntry = __pte_map(&u.Middle, ulAddr);
1298	# elif RTLNX_VER_MIN(2,5,5) \|\| defined(pte_offset_map) /* As usual, RHEL 3 had pte_offset_map earlier. */
1299	pEntry = pte_offset_map(&u.Middle, ulAddr);
1300	# else
1301	pEntry = pte_offset(&u.Middle, ulAddr);
1302	# endif
1303	if (RT_UNLIKELY(!pEntry))
1304	return NULL;
1305	u.Entry = *pEntry;
1306	# if RTLNX_VER_MIN(2,5,5) \|\| defined(pte_offset_map)
1307	pte_unmap(pEntry);
1308	# endif
1309
1310	if (RT_UNLIKELY(!pte_present(u.Entry)))
1311	return NULL;
1312	return pte_page(u.Entry);
1313	#else /* !defined(RT_ARCH_AMD64) && !defined(RT_ARCH_X86) */
1314
1315	if (is_vmalloc_addr(pv))
1316	return vmalloc_to_page(pv);
1317
1318	return virt_to_page(pv);
1319	#endif
1320	}
1321	RT_EXPORT_SYMBOL(rtR0MemObjLinuxVirtToPage);
1322
1323
1324	DECLHIDDEN(int) rtR0MemObjNativeAllocPhys(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest, size_t uAlignment,
1325	const char *pszTag)
1326	{
1327	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS, cb, uAlignment, PhysHighest, pszTag);
1328	}
1329
1330
1331	DECLHIDDEN(int) rtR0MemObjNativeAllocPhysNC(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest, const char *pszTag)
1332	{
1333	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS_NC, cb, PAGE_SIZE, PhysHighest, pszTag);
1334	}
1335
1336
1337	DECLHIDDEN(int) rtR0MemObjNativeEnterPhys(PPRTR0MEMOBJINTERNAL ppMem, RTHCPHYS Phys, size_t cb, uint32_t uCachePolicy,
1338	const char *pszTag)
1339	{
1340	IPRT_LINUX_SAVE_EFL_AC();
1341
1342	/*
1343	* All we need to do here is to validate that we can use
1344	* ioremap on the specified address (32/64-bit dma_addr_t).
1345	*/
1346	PRTR0MEMOBJLNX pMemLnx;
1347	dma_addr_t PhysAddr = Phys;
1348	AssertMsgReturn(PhysAddr == Phys, ("%#llx\n", (unsigned long long)Phys), VERR_ADDRESS_TOO_BIG);
1349
1350	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_PHYS, NULL, cb, pszTag);
1351	if (!pMemLnx)
1352	{
1353	IPRT_LINUX_RESTORE_EFL_AC();
1354	return VERR_NO_MEMORY;
1355	}
1356
1357	pMemLnx->Core.u.Phys.PhysBase = PhysAddr;
1358	pMemLnx->Core.u.Phys.fAllocated = false;
1359	pMemLnx->Core.u.Phys.uCachePolicy = uCachePolicy;
1360	Assert(!pMemLnx->cPages);
1361	*ppMem = &pMemLnx->Core;
1362	IPRT_LINUX_RESTORE_EFL_AC();
1363	return VINF_SUCCESS;
1364	}
1365
1366	/* openSUSE Leap 42.3 detection :-/ */
1367	#if RTLNX_VER_RANGE(4,4,0, 4,6,0) && defined(FAULT_FLAG_REMOTE)
1368	# define GET_USER_PAGES_API KERNEL_VERSION(4, 10, 0) /* no typo! */
1369	#else
1370	# define GET_USER_PAGES_API LINUX_VERSION_CODE
1371	#endif
1372
1373	DECLHIDDEN(int) rtR0MemObjNativeLockUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3Ptr, size_t cb, uint32_t fAccess,
1374	RTR0PROCESS R0Process, const char *pszTag)
1375	{
1376	IPRT_LINUX_SAVE_EFL_AC();
1377	const int cPages = cb >> PAGE_SHIFT;
1378	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1379	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1380	struct vm_area_struct **papVMAs;
1381	# endif
1382	PRTR0MEMOBJLNX pMemLnx;
1383	int rc = VERR_NO_MEMORY;
1384	int const fWrite = fAccess & RTMEM_PROT_WRITE ? 1 : 0;
1385
1386	/*
1387	* Check for valid task and size overflows.
1388	*/
1389	if (!pTask)
1390	return VERR_NOT_SUPPORTED;
1391	if (((size_t)cPages << PAGE_SHIFT) != cb)
1392	return VERR_OUT_OF_RANGE;
1393
1394	/*
1395	* Allocate the memory object and a temporary buffer for the VMAs.
1396	*/
1397	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_UOFFSETOF_DYN(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK,
1398	(void *)R3Ptr, cb, pszTag);
1399	if (!pMemLnx)
1400	{
1401	IPRT_LINUX_RESTORE_EFL_AC();
1402	return VERR_NO_MEMORY;
1403	}
1404
1405	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1406	papVMAs = (struct vm_area_struct *)RTMemAlloc(sizeof(papVMAs) * cPages);
1407	if (papVMAs)
1408	{
1409	# endif
1410	LNX_MM_DOWN_READ(pTask->mm);
1411
1412	/*
1413	* Get user pages.
1414	*/
1415	/** @todo r=bird: Should we not force read access too? */
1416	#if GET_USER_PAGES_API >= KERNEL_VERSION(4, 6, 0)
1417	if (R0Process == RTR0ProcHandleSelf())
1418	rc = get_user_pages(R3Ptr, /* Where from. */
1419	cPages, /* How many pages. */
1420	# if GET_USER_PAGES_API >= KERNEL_VERSION(4, 9, 0)
1421	fWrite ? FOLL_WRITE \| /* Write to memory. */
1422	FOLL_FORCE /* force write access. */
1423	: 0, /* Write to memory. */
1424	# else
1425	fWrite, /* Write to memory. */
1426	fWrite, /* force write access. */
1427	# endif
1428	&pMemLnx->apPages[0] /* Page array. */
1429	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_SUSE_MAJ_PREREQ(15, 6) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1430	, papVMAs /* vmas */
1431	# endif
1432	);
1433	/*
1434	* Actually this should not happen at the moment as call this function
1435	* only for our own process.
1436	*/
1437	else
1438	rc = get_user_pages_remote(
1439	# if GET_USER_PAGES_API < KERNEL_VERSION(5, 9, 0)
1440	pTask, /* Task for fault accounting. */
1441	# endif
1442	pTask->mm, /* Whose pages. */
1443	R3Ptr, /* Where from. */
1444	cPages, /* How many pages. */
1445	# if GET_USER_PAGES_API >= KERNEL_VERSION(4, 9, 0)
1446	fWrite ? FOLL_WRITE \| /* Write to memory. */
1447	FOLL_FORCE /* force write access. */
1448	: 0, /* Write to memory. */
1449	# else
1450	fWrite, /* Write to memory. */
1451	fWrite, /* force write access. */
1452	# endif
1453	&pMemLnx->apPages[0] /* Page array. */
1454	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1455	, papVMAs /* vmas */
1456	# endif
1457	# if GET_USER_PAGES_API >= KERNEL_VERSION(4, 10, 0)
1458	, NULL /* locked */
1459	# endif
1460	);
1461	#else /* GET_USER_PAGES_API < KERNEL_VERSION(4, 6, 0) */
1462	rc = get_user_pages(pTask, /* Task for fault accounting. */
1463	pTask->mm, /* Whose pages. */
1464	R3Ptr, /* Where from. */
1465	cPages, /* How many pages. */
1466	/* The get_user_pages API change was back-ported to 4.4.168. */
1467	# if RTLNX_VER_RANGE(4,4,168, 4,5,0)
1468	fWrite ? FOLL_WRITE \| /* Write to memory. */
1469	FOLL_FORCE /* force write access. */
1470	: 0, /* Write to memory. */
1471	# else
1472	fWrite, /* Write to memory. */
1473	fWrite, /* force write access. */
1474	# endif
1475	&pMemLnx->apPages[0], /* Page array. */
1476	papVMAs /* vmas */
1477	);
1478	#endif /* GET_USER_PAGES_API < KERNEL_VERSION(4, 6, 0) */
1479	if (rc == cPages)
1480	{
1481	/*
1482	* Flush dcache (required?), protect against fork and _really_ pin the page
1483	* table entries. get_user_pages() will protect against swapping out the
1484	* pages but it will NOT protect against removing page table entries. This
1485	* can be achieved with
1486	* - using mlock / mmap(..., MAP_LOCKED, ...) from userland. This requires
1487	* an appropriate limit set up with setrlimit(..., RLIMIT_MEMLOCK, ...).
1488	* Usual Linux distributions support only a limited size of locked pages
1489	* (e.g. 32KB).
1490	* - setting the PageReserved bit (as we do in rtR0MemObjLinuxAllocPages()
1491	* or by
1492	* - setting the VM_LOCKED flag. This is the same as doing mlock() without
1493	* a range check.
1494	*/
1495	/** @todo The Linux fork() protection will require more work if this API
1496	* is to be used for anything but locking VM pages. */
1497	while (rc-- > 0)
1498	{
1499	flush_dcache_page(pMemLnx->apPages[rc]);
1500	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_SUSE_MAJ_PREREQ(15, 6) && !RTLNX_RHEL_RANGE(9,5, 9,99)
1501	# if RTLNX_VER_MIN(6,3,0)
1502	vm_flags_set(papVMAs[rc], VM_DONTCOPY \| VM_LOCKED);
1503	# else
1504	papVMAs[rc]->vm_flags \|= VM_DONTCOPY \| VM_LOCKED;
1505	# endif
1506	# endif
1507	}
1508
1509	LNX_MM_UP_READ(pTask->mm);
1510
1511	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1512	RTMemFree(papVMAs);
1513	# endif
1514
1515	pMemLnx->Core.u.Lock.R0Process = R0Process;
1516	pMemLnx->cPages = cPages;
1517	Assert(!pMemLnx->fMappedToRing0);
1518	*ppMem = &pMemLnx->Core;
1519
1520	IPRT_LINUX_RESTORE_EFL_AC();
1521	return VINF_SUCCESS;
1522	}
1523
1524	/*
1525	* Failed - we need to unlock any pages that we succeeded to lock.
1526	*/
1527	while (rc-- > 0)
1528	{
1529	if (!PageReserved(pMemLnx->apPages[rc]))
1530	SetPageDirty(pMemLnx->apPages[rc]);
1531	#if RTLNX_VER_MIN(4,6,0)
1532	put_page(pMemLnx->apPages[rc]);
1533	#else
1534	page_cache_release(pMemLnx->apPages[rc]);
1535	#endif
1536	}
1537
1538	LNX_MM_UP_READ(pTask->mm);
1539
1540	rc = VERR_LOCK_FAILED;
1541
1542	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1543	RTMemFree(papVMAs);
1544	}
1545	# endif
1546
1547	rtR0MemObjDelete(&pMemLnx->Core);
1548	IPRT_LINUX_RESTORE_EFL_AC();
1549	return rc;
1550	}
1551
1552
1553	DECLHIDDEN(int) rtR0MemObjNativeLockKernel(PPRTR0MEMOBJINTERNAL ppMem, void pv, size_t cb, uint32_t fAccess, const char pszTag)
1554	{
1555	IPRT_LINUX_SAVE_EFL_AC();
1556	void pvLast = (uint8_t )pv + cb - 1;
1557	size_t const cPages = cb >> PAGE_SHIFT;
1558	PRTR0MEMOBJLNX pMemLnx;
1559	bool fLinearMapping;
1560	int rc;
1561	uint8_t *pbPage;
1562	size_t iPage;
1563	NOREF(fAccess);
1564
1565	if ( !RTR0MemKernelIsValidAddr(pv)
1566	\|\| !RTR0MemKernelIsValidAddr(pv + cb))
1567	return VERR_INVALID_PARAMETER;
1568
1569	/*
1570	* The lower part of the kernel memory has a linear mapping between
1571	* physical and virtual addresses. So we take a short cut here. This is
1572	* assumed to be the cleanest way to handle those addresses (and the code
1573	* is well tested, though the test for determining it is not very nice).
1574	* If we ever decide it isn't we can still remove it.
1575	*/
1576	#if 0
1577	fLinearMapping = (unsigned long)pvLast < VMALLOC_START;
1578	#else
1579	fLinearMapping = (unsigned long)pv >= (unsigned long)__va(0)
1580	&& (unsigned long)pvLast < (unsigned long)high_memory;
1581	#endif
1582
1583	/*
1584	* Allocate the memory object.
1585	*/
1586	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_UOFFSETOF_DYN(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK,
1587	pv, cb, pszTag);
1588	if (!pMemLnx)
1589	{
1590	IPRT_LINUX_RESTORE_EFL_AC();
1591	return VERR_NO_MEMORY;
1592	}
1593
1594	/*
1595	* Gather the pages.
1596	* We ASSUME all kernel pages are non-swappable and non-movable.
1597	*/
1598	rc = VINF_SUCCESS;
1599	pbPage = (uint8_t *)pvLast;
1600	iPage = cPages;
1601	if (!fLinearMapping)
1602	{
1603	while (iPage-- > 0)
1604	{
1605	struct page *pPage = rtR0MemObjLinuxVirtToPage(pbPage);
1606	if (RT_UNLIKELY(!pPage))
1607	{
1608	rc = VERR_LOCK_FAILED;
1609	break;
1610	}
1611	pMemLnx->apPages[iPage] = pPage;
1612	pbPage -= PAGE_SIZE;
1613	}
1614	}
1615	else
1616	{
1617	while (iPage-- > 0)
1618	{
1619	pMemLnx->apPages[iPage] = virt_to_page(pbPage);
1620	pbPage -= PAGE_SIZE;
1621	}
1622	}
1623	if (RT_SUCCESS(rc))
1624	{
1625	/*
1626	* Complete the memory object and return.
1627	*/
1628	pMemLnx->Core.u.Lock.R0Process = NIL_RTR0PROCESS;
1629	pMemLnx->cPages = cPages;
1630	Assert(!pMemLnx->fMappedToRing0);
1631	*ppMem = &pMemLnx->Core;
1632
1633	IPRT_LINUX_RESTORE_EFL_AC();
1634	return VINF_SUCCESS;
1635	}
1636
1637	rtR0MemObjDelete(&pMemLnx->Core);
1638	IPRT_LINUX_RESTORE_EFL_AC();
1639	return rc;
1640	}
1641
1642
1643	DECLHIDDEN(int) rtR0MemObjNativeReserveKernel(PPRTR0MEMOBJINTERNAL ppMem, void *pvFixed, size_t cb, size_t uAlignment,
1644	const char *pszTag)
1645	{
1646	#if RTLNX_VER_MIN(2,4,22)
1647	IPRT_LINUX_SAVE_EFL_AC();
1648	const size_t cPages = cb >> PAGE_SHIFT;
1649	struct page *pDummyPage;
1650	struct page **papPages;
1651
1652	/* check for unsupported stuff. */
1653	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
1654	if (uAlignment > PAGE_SIZE)
1655	return VERR_NOT_SUPPORTED;
1656
1657	/*
1658	* Allocate a dummy page and create a page pointer array for vmap such that
1659	* the dummy page is mapped all over the reserved area.
1660	*/
1661	pDummyPage = alloc_page(GFP_HIGHUSER \| __GFP_NOWARN);
1662	if (pDummyPage)
1663	{
1664	papPages = RTMemAlloc(sizeof(papPages) cPages);
1665	if (papPages)
1666	{
1667	void *pv;
1668	size_t iPage = cPages;
1669	while (iPage-- > 0)
1670	papPages[iPage] = pDummyPage;
1671	# ifdef VM_MAP
1672	pv = vmap(papPages, cPages, VM_MAP, PAGE_KERNEL_RO);
1673	# else
1674	pv = vmap(papPages, cPages, VM_ALLOC, PAGE_KERNEL_RO);
1675	# endif
1676	RTMemFree(papPages);
1677	if (pv)
1678	{
1679	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb, pszTag);
1680	if (pMemLnx)
1681	{
1682	pMemLnx->Core.u.ResVirt.R0Process = NIL_RTR0PROCESS;
1683	pMemLnx->cPages = 1;
1684	pMemLnx->apPages[0] = pDummyPage;
1685	*ppMem = &pMemLnx->Core;
1686	IPRT_LINUX_RESTORE_EFL_AC();
1687	return VINF_SUCCESS;
1688	}
1689	vunmap(pv);
1690	}
1691	}
1692	__free_page(pDummyPage);
1693	}
1694	IPRT_LINUX_RESTORE_EFL_AC();
1695	return VERR_NO_MEMORY;
1696
1697	#else /* < 2.4.22 */
1698	/*
1699	* Could probably use ioremap here, but the caller is in a better position than us
1700	* to select some safe physical memory.
1701	*/
1702	return VERR_NOT_SUPPORTED;
1703	#endif
1704	}
1705
1706
1707	DECLHIDDEN(int) rtR0MemObjNativeReserveUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment,
1708	RTR0PROCESS R0Process, const char *pszTag)
1709	{
1710	IPRT_LINUX_SAVE_EFL_AC();
1711	PRTR0MEMOBJLNX pMemLnx;
1712	void *pv;
1713	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1714	if (!pTask)
1715	return VERR_NOT_SUPPORTED;
1716
1717	/*
1718	* Check that the specified alignment is supported.
1719	*/
1720	if (uAlignment > PAGE_SIZE)
1721	return VERR_NOT_SUPPORTED;
1722
1723	/*
1724	* Let rtR0MemObjLinuxDoMmap do the difficult bits.
1725	*/
1726	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, cb, uAlignment, pTask, RTMEM_PROT_NONE);
1727	if (pv == (void *)-1)
1728	{
1729	IPRT_LINUX_RESTORE_EFL_AC();
1730	return VERR_NO_MEMORY;
1731	}
1732
1733	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb, pszTag);
1734	if (!pMemLnx)
1735	{
1736	rtR0MemObjLinuxDoMunmap(pv, cb, pTask);
1737	IPRT_LINUX_RESTORE_EFL_AC();
1738	return VERR_NO_MEMORY;
1739	}
1740
1741	pMemLnx->Core.u.ResVirt.R0Process = R0Process;
1742	*ppMem = &pMemLnx->Core;
1743	IPRT_LINUX_RESTORE_EFL_AC();
1744	return VINF_SUCCESS;
1745	}
1746
1747
1748	DECLHIDDEN(int) rtR0MemObjNativeMapKernel(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap, void *pvFixed, size_t uAlignment,
1749	unsigned fProt, size_t offSub, size_t cbSub, const char *pszTag)
1750	{
1751	int rc = VERR_NO_MEMORY;
1752	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1753	PRTR0MEMOBJLNX pMemLnx;
1754	IPRT_LINUX_SAVE_EFL_AC();
1755
1756	/* Fail if requested to do something we can't. */
1757	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
1758	if (uAlignment > PAGE_SIZE)
1759	return VERR_NOT_SUPPORTED;
1760
1761	/*
1762	* Create the IPRT memory object.
1763	*/
1764	if (!cbSub)
1765	cbSub = pMemLnxToMap->Core.cb - offSub;
1766	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, cbSub, pszTag);
1767	if (pMemLnx)
1768	{
1769	if (pMemLnxToMap->cPages)
1770	{
1771	#if RTLNX_VER_MIN(2,4,22)
1772	/*
1773	* Use vmap - 2.4.22 and later.
1774	*/
1775	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, true /* kernel */);
1776	/** @todo We don't really care too much for EXEC here... 5.8 always adds NX. */
1777	Assert(((offSub + cbSub) >> PAGE_SHIFT) <= pMemLnxToMap->cPages);
1778	# ifdef VM_MAP
1779	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[offSub >> PAGE_SHIFT], cbSub >> PAGE_SHIFT, VM_MAP, fPg);
1780	# else
1781	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[offSub >> PAGE_SHIFT], cbSub >> PAGE_SHIFT, VM_ALLOC, fPg);
1782	# endif
1783	if (pMemLnx->Core.pv)
1784	{
1785	pMemLnx->fMappedToRing0 = true;
1786	rc = VINF_SUCCESS;
1787	}
1788	else
1789	rc = VERR_MAP_FAILED;
1790
1791	#else /* < 2.4.22 */
1792	/*
1793	* Only option here is to share mappings if possible and forget about fProt.
1794	*/
1795	if (rtR0MemObjIsRing3(pMemToMap))
1796	rc = VERR_NOT_SUPPORTED;
1797	else
1798	{
1799	rc = VINF_SUCCESS;
1800	if (!pMemLnxToMap->Core.pv)
1801	rc = rtR0MemObjLinuxVMap(pMemLnxToMap, !!(fProt & RTMEM_PROT_EXEC));
1802	if (RT_SUCCESS(rc))
1803	{
1804	Assert(pMemLnxToMap->Core.pv);
1805	pMemLnx->Core.pv = (uint8_t *)pMemLnxToMap->Core.pv + offSub;
1806	}
1807	}
1808	#endif
1809	}
1810	else
1811	{
1812	/*
1813	* MMIO / physical memory.
1814	*/
1815	Assert(pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS && !pMemLnxToMap->Core.u.Phys.fAllocated);
1816	#if RTLNX_VER_MIN(2,6,25)
1817	/*
1818	* ioremap() defaults to no caching since the 2.6 kernels.
1819	* ioremap_nocache() has been removed finally in 5.6-rc1.
1820	*/
1821	pMemLnx->Core.pv = pMemLnxToMap->Core.u.Phys.uCachePolicy == RTMEM_CACHE_POLICY_MMIO
1822	? ioremap(pMemLnxToMap->Core.u.Phys.PhysBase + offSub, cbSub)
1823	: ioremap_cache(pMemLnxToMap->Core.u.Phys.PhysBase + offSub, cbSub);
1824	#else /* KERNEL_VERSION < 2.6.25 */
1825	pMemLnx->Core.pv = pMemLnxToMap->Core.u.Phys.uCachePolicy == RTMEM_CACHE_POLICY_MMIO
1826	? ioremap_nocache(pMemLnxToMap->Core.u.Phys.PhysBase + offSub, cbSub)
1827	: ioremap(pMemLnxToMap->Core.u.Phys.PhysBase + offSub, cbSub);
1828	#endif /* KERNEL_VERSION < 2.6.25 */
1829	if (pMemLnx->Core.pv)
1830	{
1831	/** @todo fix protection. */
1832	rc = VINF_SUCCESS;
1833	}
1834	}
1835	if (RT_SUCCESS(rc))
1836	{
1837	pMemLnx->Core.u.Mapping.R0Process = NIL_RTR0PROCESS;
1838	*ppMem = &pMemLnx->Core;
1839	IPRT_LINUX_RESTORE_EFL_AC();
1840	return VINF_SUCCESS;
1841	}
1842	rtR0MemObjDelete(&pMemLnx->Core);
1843	}
1844
1845	IPRT_LINUX_RESTORE_EFL_AC();
1846	return rc;
1847	}
1848
1849
1850	#ifdef VBOX_USE_PAE_HACK
1851	/**
1852	* Replace the PFN of a PTE with the address of the actual page.
1853	*
1854	* The caller maps a reserved dummy page at the address with the desired access
1855	* and flags.
1856	*
1857	* This hack is required for older Linux kernels which don't provide
1858	* remap_pfn_range().
1859	*
1860	* @returns 0 on success, -ENOMEM on failure.
1861	* @param mm The memory context.
1862	* @param ulAddr The mapping address.
1863	* @param Phys The physical address of the page to map.
1864	*/
1865	static int rtR0MemObjLinuxFixPte(struct mm_struct *mm, unsigned long ulAddr, RTHCPHYS Phys)
1866	{
1867	int rc = -ENOMEM;
1868	pgd_t *pgd;
1869
1870	spin_lock(&mm->page_table_lock);
1871
1872	pgd = pgd_offset(mm, ulAddr);
1873	if (!pgd_none(pgd) && !pgd_bad(pgd))
1874	{
1875	pmd_t *pmd = pmd_offset(pgd, ulAddr);
1876	if (!pmd_none(*pmd))
1877	{
1878	pte_t *ptep = pte_offset_map(pmd, ulAddr);
1879	if (ptep)
1880	{
1881	pte_t pte = *ptep;
1882	pte.pte_high &= 0xfff00000;
1883	pte.pte_high \|= ((Phys >> 32) & 0x000fffff);
1884	pte.pte_low &= 0x00000fff;
1885	pte.pte_low \|= (Phys & 0xfffff000);
1886	set_pte(ptep, pte);
1887	pte_unmap(ptep);
1888	rc = 0;
1889	}
1890	}
1891	}
1892
1893	spin_unlock(&mm->page_table_lock);
1894	return rc;
1895	}
1896	#endif /* VBOX_USE_PAE_HACK */
1897
1898
1899	DECLHIDDEN(int) rtR0MemObjNativeMapUser(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap, RTR3PTR R3PtrFixed, size_t uAlignment,
1900	unsigned fProt, RTR0PROCESS R0Process, size_t offSub, size_t cbSub, const char *pszTag)
1901	{
1902	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1903	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1904	int rc = VERR_NO_MEMORY;
1905	PRTR0MEMOBJLNX pMemLnx;
1906	#ifdef VBOX_USE_PAE_HACK
1907	struct page *pDummyPage;
1908	RTHCPHYS DummyPhys;
1909	#endif
1910	IPRT_LINUX_SAVE_EFL_AC();
1911
1912	/*
1913	* Check for restrictions.
1914	*/
1915	if (!pTask)
1916	return VERR_NOT_SUPPORTED;
1917	if (uAlignment > PAGE_SIZE)
1918	return VERR_NOT_SUPPORTED;
1919
1920	#ifdef VBOX_USE_PAE_HACK
1921	/*
1922	* Allocate a dummy page for use when mapping the memory.
1923	*/
1924	pDummyPage = alloc_page(GFP_USER \| __GFP_NOWARN);
1925	if (!pDummyPage)
1926	{
1927	IPRT_LINUX_RESTORE_EFL_AC();
1928	return VERR_NO_MEMORY;
1929	}
1930	SetPageReserved(pDummyPage);
1931	DummyPhys = page_to_phys(pDummyPage);
1932	#endif
1933
1934	/*
1935	* Create the IPRT memory object.
1936	*/
1937	Assert(!offSub \|\| cbSub);
1938	if (cbSub == 0)
1939	cbSub = pMemLnxToMap->Core.cb;
1940	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, cbSub, pszTag);
1941	if (pMemLnx)
1942	{
1943	/*
1944	* Allocate user space mapping.
1945	*/
1946	void *pv;
1947	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, cbSub, uAlignment, pTask, fProt);
1948	if (pv != (void *)-1)
1949	{
1950	/*
1951	* Map page by page into the mmap area.
1952	* This is generic, paranoid and not very efficient.
1953	*/
1954	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, false /* user */);
1955	unsigned long ulAddrCur = (unsigned long)pv;
1956	const size_t cPages = (offSub + cbSub) >> PAGE_SHIFT;
1957	size_t iPage;
1958
1959	LNX_MM_DOWN_WRITE(pTask->mm);
1960
1961	rc = VINF_SUCCESS;
1962	if (pMemLnxToMap->cPages)
1963	{
1964	for (iPage = offSub >> PAGE_SHIFT; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE)
1965	{
1966	#if RTLNX_VER_MAX(2,6,11)
1967	RTHCPHYS Phys = page_to_phys(pMemLnxToMap->apPages[iPage]);
1968	#endif
1969	#if RTLNX_VER_MIN(2,6,0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1970	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
1971	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
1972	#endif
1973	#if RTLNX_VER_MAX(2,6,0) && defined(RT_ARCH_X86)
1974	/* remap_page_range() limitation on x86 */
1975	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
1976	#endif
1977
1978	#if defined(VBOX_USE_INSERT_PAGE) && RTLNX_VER_MIN(2,6,22)
1979	rc = vm_insert_page(vma, ulAddrCur, pMemLnxToMap->apPages[iPage]);
1980	/* Thes flags help making 100% sure some bad stuff wont happen (swap, core, ++).
1981	* See remap_pfn_range() in mm/memory.c */
1982
1983	#if RTLNX_VER_MIN(6,3,0) \|\| RTLNX_RHEL_RANGE(9,5, 9,99)
1984	vm_flags_set(vma, VM_DONTEXPAND \| VM_DONTDUMP);
1985	#elif RTLNX_VER_MIN(3,7,0)
1986	vma->vm_flags \|= VM_DONTEXPAND \| VM_DONTDUMP;
1987	#else
1988	vma->vm_flags \|= VM_RESERVED;
1989	#endif
1990	#elif RTLNX_VER_MIN(2,6,11)
1991	rc = remap_pfn_range(vma, ulAddrCur, page_to_pfn(pMemLnxToMap->apPages[iPage]), PAGE_SIZE, fPg);
1992	#elif defined(VBOX_USE_PAE_HACK)
1993	rc = remap_page_range(vma, ulAddrCur, DummyPhys, PAGE_SIZE, fPg);
1994	if (!rc)
1995	rc = rtR0MemObjLinuxFixPte(pTask->mm, ulAddrCur, Phys);
1996	#elif RTLNX_VER_MIN(2,6,0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1997	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
1998	#else /* 2.4 */
1999	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
2000	#endif
2001	if (rc)
2002	{
2003	rc = VERR_NO_MEMORY;
2004	break;
2005	}
2006	}
2007	}
2008	else
2009	{
2010	RTHCPHYS Phys;
2011	if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS)
2012	Phys = pMemLnxToMap->Core.u.Phys.PhysBase;
2013	else if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_CONT)
2014	Phys = pMemLnxToMap->Core.u.Cont.Phys;
2015	else
2016	{
2017	AssertMsgFailed(("%d\n", pMemLnxToMap->Core.enmType));
2018	Phys = NIL_RTHCPHYS;
2019	}
2020	if (Phys != NIL_RTHCPHYS)
2021	{
2022	for (iPage = offSub >> PAGE_SHIFT; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE, Phys += PAGE_SIZE)
2023	{
2024	#if RTLNX_VER_MIN(2,6,0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
2025	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
2026	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
2027	#endif
2028	#if RTLNX_VER_MAX(2,6,0) && defined(RT_ARCH_X86)
2029	/* remap_page_range() limitation on x86 */
2030	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
2031	#endif
2032
2033	#if RTLNX_VER_MIN(2,6,11)
2034	rc = remap_pfn_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
2035	#elif defined(VBOX_USE_PAE_HACK)
2036	rc = remap_page_range(vma, ulAddrCur, DummyPhys, PAGE_SIZE, fPg);
2037	if (!rc)
2038	rc = rtR0MemObjLinuxFixPte(pTask->mm, ulAddrCur, Phys);
2039	#elif RTLNX_VER_MIN(2,6,0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
2040	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
2041	#else /* 2.4 */
2042	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
2043	#endif
2044	if (rc)
2045	{
2046	rc = VERR_NO_MEMORY;
2047	break;
2048	}
2049	}
2050	}
2051	}
2052
2053	#ifdef CONFIG_NUMA_BALANCING
2054	# if RTLNX_VER_MAX(3,13,0) && RTLNX_RHEL_MAX(7,0)
2055	# define VBOX_NUMA_HACK_OLD
2056	# endif
2057	if (RT_SUCCESS(rc))
2058	{
2059	/** @todo Ugly hack! But right now we have no other means to
2060	* disable automatic NUMA page balancing. */
2061	# ifdef RT_OS_X86
2062	# ifdef VBOX_NUMA_HACK_OLD
2063	pTask->mm->numa_next_reset = jiffies + 0x7fffffffUL;
2064	# endif
2065	pTask->mm->numa_next_scan = jiffies + 0x7fffffffUL;
2066	# else
2067	# ifdef VBOX_NUMA_HACK_OLD
2068	pTask->mm->numa_next_reset = jiffies + 0x7fffffffffffffffUL;
2069	# endif
2070	pTask->mm->numa_next_scan = jiffies + 0x7fffffffffffffffUL;
2071	# endif
2072	}
2073	#endif /* CONFIG_NUMA_BALANCING */
2074
2075	LNX_MM_UP_WRITE(pTask->mm);
2076
2077	if (RT_SUCCESS(rc))
2078	{
2079	#ifdef VBOX_USE_PAE_HACK
2080	__free_page(pDummyPage);
2081	#endif
2082	pMemLnx->Core.pv = pv;
2083	pMemLnx->Core.u.Mapping.R0Process = R0Process;
2084	*ppMem = &pMemLnx->Core;
2085	IPRT_LINUX_RESTORE_EFL_AC();
2086	return VINF_SUCCESS;
2087	}
2088
2089	/*
2090	* Bail out.
2091	*/
2092	rtR0MemObjLinuxDoMunmap(pv, cbSub, pTask);
2093	}
2094	rtR0MemObjDelete(&pMemLnx->Core);
2095	}
2096	#ifdef VBOX_USE_PAE_HACK
2097	__free_page(pDummyPage);
2098	#endif
2099
2100	IPRT_LINUX_RESTORE_EFL_AC();
2101	return rc;
2102	}
2103
2104
2105	DECLHIDDEN(int) rtR0MemObjNativeProtect(PRTR0MEMOBJINTERNAL pMem, size_t offSub, size_t cbSub, uint32_t fProt)
2106	{
2107	# ifdef IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
2108	/*
2109	* Currently only supported when we've got addresses PTEs from the kernel.
2110	*/
2111	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
2112	if (pMemLnx->pArea && pMemLnx->papPtesForArea)
2113	{
2114	pgprot_t const fPg = rtR0MemObjLinuxConvertProt(fProt, true /fKernel/);
2115	size_t const cPages = (offSub + cbSub) >> PAGE_SHIFT;
2116	pte_t **papPtes = pMemLnx->papPtesForArea;
2117	size_t i;
2118
2119	for (i = offSub >> PAGE_SHIFT; i < cPages; i++)
2120	{
2121	set_pte(papPtes[i], mk_pte(pMemLnx->apPages[i], fPg));
2122	}
2123	preempt_disable();
2124	__flush_tlb_all();
2125	preempt_enable();
2126	return VINF_SUCCESS;
2127	}
2128	# elif defined(IPRT_USE_APPLY_TO_PAGE_RANGE_FOR_EXEC)
2129	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
2130	if ( pMemLnx->fExecutable
2131	&& pMemLnx->fMappedToRing0)
2132	{
2133	LNXAPPLYPGRANGE Args;
2134	Args.pMemLnx = pMemLnx;
2135	Args.fPg = rtR0MemObjLinuxConvertProt(fProt, true /fKernel/);
2136	int rcLnx = apply_to_page_range(current->active_mm, (unsigned long)pMemLnx->Core.pv + offSub, cbSub,
2137	rtR0MemObjLinuxApplyPageRange, (void *)&Args);
2138	if (rcLnx)
2139	return VERR_NOT_SUPPORTED;
2140
2141	return VINF_SUCCESS;
2142	}
2143	# endif
2144
2145	NOREF(pMem);
2146	NOREF(offSub);
2147	NOREF(cbSub);
2148	NOREF(fProt);
2149	return VERR_NOT_SUPPORTED;
2150	}
2151
2152
2153	DECLHIDDEN(RTHCPHYS) rtR0MemObjNativeGetPagePhysAddr(PRTR0MEMOBJINTERNAL pMem, size_t iPage)
2154	{
2155	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
2156
2157	if (pMemLnx->cPages)
2158	return page_to_phys(pMemLnx->apPages[iPage]);
2159
2160	switch (pMemLnx->Core.enmType)
2161	{
2162	case RTR0MEMOBJTYPE_CONT:
2163	return pMemLnx->Core.u.Cont.Phys + (iPage << PAGE_SHIFT);
2164
2165	case RTR0MEMOBJTYPE_PHYS:
2166	return pMemLnx->Core.u.Phys.PhysBase + (iPage << PAGE_SHIFT);
2167
2168	/* the parent knows */
2169	case RTR0MEMOBJTYPE_MAPPING:
2170	return rtR0MemObjNativeGetPagePhysAddr(pMemLnx->Core.uRel.Child.pParent, iPage);
2171
2172	/* cPages > 0 */
2173	case RTR0MEMOBJTYPE_LOW:
2174	case RTR0MEMOBJTYPE_LOCK:
2175	case RTR0MEMOBJTYPE_PHYS_NC:
2176	case RTR0MEMOBJTYPE_PAGE:
2177	case RTR0MEMOBJTYPE_LARGE_PAGE:
2178	default:
2179	AssertMsgFailed(("%d\n", pMemLnx->Core.enmType));
2180	RT_FALL_THROUGH();
2181
2182	case RTR0MEMOBJTYPE_RES_VIRT:
2183	return NIL_RTHCPHYS;
2184	}
2185	}
2186
2187
2188	DECLHIDDEN(int) rtR0MemObjNativeZeroInitWithoutMapping(PRTR0MEMOBJINTERNAL pMem)
2189	{
2190	PRTR0MEMOBJLNX const pMemLnx = (PRTR0MEMOBJLNX)pMem;
2191	size_t const cPages = pMemLnx->Core.cb >> PAGE_SHIFT;
2192	size_t iPage;
2193	/** @todo optimize this. */
2194	for (iPage = 0; iPage < cPages; iPage++)
2195	{
2196	void *pvPage;
2197
2198	/* Get the physical address of the page. */
2199	RTHCPHYS const HCPhys = rtR0MemObjNativeGetPagePhysAddr(&pMemLnx->Core, iPage);
2200	AssertReturn(HCPhys != NIL_RTHCPHYS, VERR_INTERNAL_ERROR_3);
2201	Assert(!(HCPhys & PAGE_OFFSET_MASK));
2202
2203	/* Would've like to use valid_phys_addr_range for this test, but it isn't exported. */
2204	AssertReturn((HCPhys \| PAGE_OFFSET_MASK) < __pa(high_memory), VERR_INTERNAL_ERROR_3);
2205
2206	/* Map it. */
2207	pvPage = phys_to_virt(HCPhys);
2208	AssertPtrReturn(pvPage, VERR_INTERNAL_ERROR_3);
2209
2210	/* Zero it. */
2211	RT_BZERO(pvPage, PAGE_SIZE);
2212	}
2213	return VINF_SUCCESS;
2214	}
2215

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

source: vbox/trunk/src/VBox/Runtime/r0drv/linux/memobj-r0drv-linux.c@ 106734

Download in other formats: