memobj-r0drv-linux.c

Last change on this file was 107034, checked in by vboxsync, 13 days ago
iprt: Linux: Add initial support for RHEL 9.6 kernel (fix for pre 6.5 kernels), bugref:10482.
Property svn:eol-style set to `native` Property svn:keywords set to `Id Rev Revision`
File size: 76.0 KB

Line
1	/* $Id: memobj-r0drv-linux.c 107034 2024-11-18 21:19:23Z 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	# if RTLNX_VER_MIN(6,5,0)
659	pgprot_val(fPg) = _PAGE_KERNEL; /* (PROT_DEFAULT \| PTE_PXN \| PTE_UXN \| PTE_WRITE \| PTE_ATTRINDX(MT_NORMAL). */
660	# else /* < 6.5.0 */
661	pgprot_val(fPg) = PROT_NORMAL; /* (PROT_DEFAULT \| PTE_PXN \| PTE_UXN \| PTE_WRITE \| PTE_ATTRINDX(MT_NORMAL). */
662	# endif /* 6.5.0 */
663	# else /* !RT_ARCH_ARM64 */
664	pgprot_val(fPg) = _PAGE_PRESENT \| _PAGE_RW;
665	# ifdef _PAGE_NX
666	if (!fExecutable)
667	pgprot_val(fPg) \|= _PAGE_NX;
668	# endif
669	# endif /* RT_ARCH_ARM64 */
670
671	# ifdef IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
672	if (fExecutable)
673	{
674	# if RTLNX_VER_MIN(3,2,51)
675	pte_t papPtes = (pte_t )kmalloc_array(pMemLnx->cPages, sizeof(papPtes[0]), GFP_KERNEL);
676	# else
677	pte_t papPtes = (pte_t )kmalloc(pMemLnx->cPages * sizeof(papPtes[0]), GFP_KERNEL);
678	# endif
679	if (papPtes)
680	{
681	pMemLnx->pArea = alloc_vm_area(pMemLnx->Core.cb, papPtes); /* Note! pArea->nr_pages is not set. */
682	if (pMemLnx->pArea)
683	{
684	size_t i;
685	Assert(pMemLnx->pArea->size >= pMemLnx->Core.cb); /* Note! includes guard page. */
686	Assert(pMemLnx->pArea->addr);
687	# if !defined(RT_ARCH_ARM64) && defined(_PAGE_NX)
688	pgprot_val(fPg) \|= _PAGE_NX; /* Uses RTR0MemObjProtect to clear NX when memory ready, W^X fashion. */
689	# endif
690	pMemLnx->papPtesForArea = papPtes;
691	for (i = 0; i < pMemLnx->cPages; i++)
692	*papPtes[i] = mk_pte(pMemLnx->apPages[i], fPg);
693	pMemLnx->Core.pv = pMemLnx->pArea->addr;
694	pMemLnx->fMappedToRing0 = true;
695	}
696	else
697	{
698	kfree(papPtes);
699	rc = VERR_MAP_FAILED;
700	}
701	}
702	else
703	rc = VERR_MAP_FAILED;
704	}
705	else
706	# endif
707	{
708	# if !defined(RT_ARCH_ARM64) && defined(IPRT_USE_APPLY_TO_PAGE_RANGE_FOR_EXEC)
709	if (fExecutable)
710	pgprot_val(fPg) \|= _PAGE_NX; /* Uses RTR0MemObjProtect to clear NX when memory ready, W^X fashion. */
711	# endif
712
713	# ifdef VM_MAP
714	pMemLnx->Core.pv = vmap(&pMemLnx->apPages[0], pMemLnx->cPages, VM_MAP, fPg);
715	# else
716	pMemLnx->Core.pv = vmap(&pMemLnx->apPages[0], pMemLnx->cPages, VM_ALLOC, fPg);
717	# endif
718	if (pMemLnx->Core.pv)
719	pMemLnx->fMappedToRing0 = true;
720	else
721	rc = VERR_MAP_FAILED;
722	}
723	#else /* < 2.4.22 */
724	rc = VERR_NOT_SUPPORTED;
725	#endif
726	}
727	else
728	{
729	/*
730	* Use the kernel RAM mapping.
731	*/
732	pMemLnx->Core.pv = phys_to_virt(page_to_phys(pMemLnx->apPages[0]));
733	Assert(pMemLnx->Core.pv);
734	}
735
736	return rc;
737	}
738
739
740	/**
741	* Undoes what rtR0MemObjLinuxVMap() did.
742	*
743	* @param pMemLnx The linux memory object.
744	*/
745	static void rtR0MemObjLinuxVUnmap(PRTR0MEMOBJLNX pMemLnx)
746	{
747	#if RTLNX_VER_MIN(2,4,22)
748	# ifdef IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
749	if (pMemLnx->pArea)
750	{
751	# if 0
752	pte_t **papPtes = pMemLnx->papPtesForArea;
753	size_t i;
754	for (i = 0; i < pMemLnx->cPages; i++)
755	*papPtes[i] = 0;
756	# endif
757	free_vm_area(pMemLnx->pArea);
758	kfree(pMemLnx->papPtesForArea);
759	pMemLnx->pArea = NULL;
760	pMemLnx->papPtesForArea = NULL;
761	}
762	else
763	# endif
764	if (pMemLnx->fMappedToRing0)
765	{
766	Assert(pMemLnx->Core.pv);
767	vunmap(pMemLnx->Core.pv);
768	pMemLnx->fMappedToRing0 = false;
769	}
770	#else /* < 2.4.22 */
771	Assert(!pMemLnx->fMappedToRing0);
772	#endif
773	pMemLnx->Core.pv = NULL;
774	}
775
776
777	DECLHIDDEN(int) rtR0MemObjNativeFree(RTR0MEMOBJ pMem)
778	{
779	IPRT_LINUX_SAVE_EFL_AC();
780	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
781
782	/*
783	* Release any memory that we've allocated or locked.
784	*/
785	switch (pMemLnx->Core.enmType)
786	{
787	case RTR0MEMOBJTYPE_PAGE:
788	case RTR0MEMOBJTYPE_LOW:
789	case RTR0MEMOBJTYPE_CONT:
790	case RTR0MEMOBJTYPE_PHYS:
791	case RTR0MEMOBJTYPE_PHYS_NC:
792	rtR0MemObjLinuxVUnmap(pMemLnx);
793	rtR0MemObjLinuxFreePages(pMemLnx);
794	break;
795
796	case RTR0MEMOBJTYPE_LARGE_PAGE:
797	{
798	uint32_t const cLargePages = pMemLnx->Core.cb >> (pMemLnx->cLargePageOrder + PAGE_SHIFT);
799	uint32_t iLargePage;
800	for (iLargePage = 0; iLargePage < cLargePages; iLargePage++)
801	__free_pages(pMemLnx->apPages[iLargePage << pMemLnx->cLargePageOrder], pMemLnx->cLargePageOrder);
802	pMemLnx->cPages = 0;
803
804	#ifdef IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
805	Assert(!pMemLnx->pArea);
806	Assert(!pMemLnx->papPtesForArea);
807	#endif
808	break;
809	}
810
811	case RTR0MEMOBJTYPE_LOCK:
812	if (pMemLnx->Core.u.Lock.R0Process != NIL_RTR0PROCESS)
813	{
814	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
815	size_t iPage;
816	Assert(pTask);
817	if (pTask && pTask->mm)
818	LNX_MM_DOWN_READ(pTask->mm);
819
820	iPage = pMemLnx->cPages;
821	while (iPage-- > 0)
822	{
823	if (!PageReserved(pMemLnx->apPages[iPage]))
824	SetPageDirty(pMemLnx->apPages[iPage]);
825	#if RTLNX_VER_MIN(4,6,0)
826	put_page(pMemLnx->apPages[iPage]);
827	#else
828	page_cache_release(pMemLnx->apPages[iPage]);
829	#endif
830	}
831
832	if (pTask && pTask->mm)
833	LNX_MM_UP_READ(pTask->mm);
834	}
835	/* else: kernel memory - nothing to do here. */
836	break;
837
838	case RTR0MEMOBJTYPE_RES_VIRT:
839	Assert(pMemLnx->Core.pv);
840	if (pMemLnx->Core.u.ResVirt.R0Process != NIL_RTR0PROCESS)
841	{
842	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
843	Assert(pTask);
844	if (pTask && pTask->mm)
845	rtR0MemObjLinuxDoMunmap(pMemLnx->Core.pv, pMemLnx->Core.cb, pTask);
846	}
847	else
848	{
849	vunmap(pMemLnx->Core.pv);
850
851	Assert(pMemLnx->cPages == 1 && pMemLnx->apPages[0] != NULL);
852	__free_page(pMemLnx->apPages[0]);
853	pMemLnx->apPages[0] = NULL;
854	pMemLnx->cPages = 0;
855	}
856	pMemLnx->Core.pv = NULL;
857	break;
858
859	case RTR0MEMOBJTYPE_MAPPING:
860	Assert(pMemLnx->cPages == 0); Assert(pMemLnx->Core.pv);
861	if (pMemLnx->Core.u.ResVirt.R0Process != NIL_RTR0PROCESS)
862	{
863	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
864	Assert(pTask);
865	if (pTask && pTask->mm)
866	rtR0MemObjLinuxDoMunmap(pMemLnx->Core.pv, pMemLnx->Core.cb, pTask);
867	}
868	else
869	vunmap(pMemLnx->Core.pv);
870	pMemLnx->Core.pv = NULL;
871	break;
872
873	default:
874	AssertMsgFailed(("enmType=%d\n", pMemLnx->Core.enmType));
875	return VERR_INTERNAL_ERROR;
876	}
877	IPRT_LINUX_RESTORE_EFL_ONLY_AC();
878	return VINF_SUCCESS;
879	}
880
881
882	DECLHIDDEN(int) rtR0MemObjNativeAllocPage(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable, const char *pszTag)
883	{
884	IPRT_LINUX_SAVE_EFL_AC();
885	PRTR0MEMOBJLNX pMemLnx;
886	int rc;
887
888	#if RTLNX_VER_MIN(2,4,22)
889	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_PAGE, cb, PAGE_SIZE, GFP_HIGHUSER,
890	false /* non-contiguous */, fExecutable, VERR_NO_MEMORY, pszTag);
891	#else
892	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_PAGE, cb, PAGE_SIZE, GFP_USER,
893	false /* non-contiguous */, fExecutable, VERR_NO_MEMORY, pszTag);
894	#endif
895	if (RT_SUCCESS(rc))
896	{
897	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
898	if (RT_SUCCESS(rc))
899	{
900	*ppMem = &pMemLnx->Core;
901	IPRT_LINUX_RESTORE_EFL_AC();
902	return rc;
903	}
904
905	rtR0MemObjLinuxFreePages(pMemLnx);
906	rtR0MemObjDelete(&pMemLnx->Core);
907	}
908
909	IPRT_LINUX_RESTORE_EFL_AC();
910	return rc;
911	}
912
913
914	DECLHIDDEN(int) rtR0MemObjNativeAllocLarge(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, size_t cbLargePage, uint32_t fFlags,
915	const char *pszTag)
916	{
917	#ifdef GFP_TRANSHUGE
918	/*
919	* Allocate a memory object structure that's large enough to contain
920	* the page pointer array.
921	*/
922	# ifdef __GFP_MOVABLE
923	unsigned const fGfp = (GFP_TRANSHUGE \| __GFP_ZERO) & ~__GFP_MOVABLE;
924	# else
925	unsigned const fGfp = (GFP_TRANSHUGE \| __GFP_ZERO);
926	# endif
927	size_t const cPagesPerLarge = cbLargePage >> PAGE_SHIFT;
928	unsigned const cLargePageOrder = rtR0MemObjLinuxOrder(cPagesPerLarge);
929	size_t const cLargePages = cb >> (cLargePageOrder + PAGE_SHIFT);
930	size_t const cPages = cb >> PAGE_SHIFT;
931	PRTR0MEMOBJLNX pMemLnx;
932
933	Assert(RT_BIT_64(cLargePageOrder + PAGE_SHIFT) == cbLargePage);
934	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_UOFFSETOF_DYN(RTR0MEMOBJLNX, apPages[cPages]),
935	RTR0MEMOBJTYPE_LARGE_PAGE, NULL, cb, pszTag);
936	if (pMemLnx)
937	{
938	size_t iLargePage;
939
940	pMemLnx->Core.fFlags \|= RTR0MEMOBJ_FLAGS_ZERO_AT_ALLOC;
941	pMemLnx->cLargePageOrder = cLargePageOrder;
942	pMemLnx->cPages = cPages;
943
944	/*
945	* Allocate the requested number of large pages.
946	*/
947	for (iLargePage = 0; iLargePage < cLargePages; iLargePage++)
948	{
949	struct page *paPages = alloc_pages(fGfp, cLargePageOrder);
950	if (paPages)
951	{
952	size_t const iPageBase = iLargePage << cLargePageOrder;
953	size_t iPage = cPagesPerLarge;
954	while (iPage-- > 0)
955	pMemLnx->apPages[iPageBase + iPage] = &paPages[iPage];
956	}
957	else
958	{
959	/*Log(("rtR0MemObjNativeAllocLarge: cb=%#zx cPages=%#zx cLargePages=%#zx cLargePageOrder=%u cPagesPerLarge=%#zx iLargePage=%#zx -> failed!\n",
960	cb, cPages, cLargePages, cLargePageOrder, cPagesPerLarge, iLargePage, paPages));*/
961	while (iLargePage-- > 0)
962	__free_pages(pMemLnx->apPages[iLargePage << (cLargePageOrder - PAGE_SHIFT)], cLargePageOrder);
963	rtR0MemObjDelete(&pMemLnx->Core);
964	return VERR_NO_MEMORY;
965	}
966	}
967	*ppMem = &pMemLnx->Core;
968	return VINF_SUCCESS;
969	}
970	return VERR_NO_MEMORY;
971
972	#else
973	/*
974	* We don't call rtR0MemObjFallbackAllocLarge here as it can be a really
975	* bad idea to trigger the swap daemon and whatnot. So, just fail.
976	*/
977	RT_NOREF(ppMem, cb, cbLargePage, fFlags, pszTag);
978	return VERR_NOT_SUPPORTED;
979	#endif
980	}
981
982
983	DECLHIDDEN(int) rtR0MemObjNativeAllocLow(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable, const char *pszTag)
984	{
985	IPRT_LINUX_SAVE_EFL_AC();
986	PRTR0MEMOBJLNX pMemLnx;
987	int rc;
988
989	/* Try to avoid GFP_DMA. GFM_DMA32 was introduced with Linux 2.6.15. */
990	#if (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
991	/* ZONE_DMA32: 0-4GB */
992	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_DMA32,
993	false /* non-contiguous */, fExecutable, VERR_NO_LOW_MEMORY, pszTag);
994	if (RT_FAILURE(rc))
995	#endif
996	#ifdef RT_ARCH_AMD64
997	/* ZONE_DMA: 0-16MB */
998	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_DMA,
999	false /* non-contiguous */, fExecutable, VERR_NO_LOW_MEMORY, pszTag);
1000	#else
1001	# ifdef CONFIG_X86_PAE
1002	# endif
1003	/* ZONE_NORMAL: 0-896MB */
1004	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_USER,
1005	false /* non-contiguous */, fExecutable, VERR_NO_LOW_MEMORY, pszTag);
1006	#endif
1007	if (RT_SUCCESS(rc))
1008	{
1009	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
1010	if (RT_SUCCESS(rc))
1011	{
1012	*ppMem = &pMemLnx->Core;
1013	IPRT_LINUX_RESTORE_EFL_AC();
1014	return rc;
1015	}
1016
1017	rtR0MemObjLinuxFreePages(pMemLnx);
1018	rtR0MemObjDelete(&pMemLnx->Core);
1019	}
1020
1021	IPRT_LINUX_RESTORE_EFL_AC();
1022	return rc;
1023	}
1024
1025
1026	DECLHIDDEN(int) rtR0MemObjNativeAllocCont(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest,
1027	bool fExecutable, const char *pszTag)
1028	{
1029	IPRT_LINUX_SAVE_EFL_AC();
1030	PRTR0MEMOBJLNX pMemLnx;
1031	int rc;
1032	uint32_t idxZone;
1033
1034	/*
1035	* The last zone must be able to satisfy the PhysHighest requirement or there
1036	* will be no zone at all.
1037	*/
1038	if (g_aZones[RT_ELEMENTS(g_aZones) - 1].PhysHighest > PhysHighest)
1039	{
1040	IPRT_LINUX_RESTORE_EFL_AC();
1041	AssertMsgFailedReturn(("No zone can satisfy PhysHighest=%RHp!\n", PhysHighest),
1042	VERR_NO_CONT_MEMORY);
1043	}
1044
1045	/* Find the first zone matching our PhysHighest requirement. */
1046	idxZone = 0;
1047	for (;;)
1048	{
1049	if (g_aZones[idxZone].PhysHighest <= PhysHighest)
1050	break; /* We found a zone satisfying the requirement. */
1051	idxZone++;
1052	}
1053
1054	/* Now try to allocate pages from all the left zones until one succeeds. */
1055	for (;;)
1056	{
1057	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, PAGE_SIZE, g_aZones[idxZone].fGfp,
1058	true /* contiguous */, fExecutable, VERR_NO_CONT_MEMORY, pszTag);
1059	idxZone++;
1060	if (RT_SUCCESS(rc) \|\| idxZone == RT_ELEMENTS(g_aZones))
1061	break;
1062	}
1063	if (RT_SUCCESS(rc))
1064	{
1065	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
1066	if (RT_SUCCESS(rc))
1067	{
1068	#if defined(RT_STRICT)
1069	size_t iPage = pMemLnx->cPages;
1070	while (iPage-- > 0)
1071	Assert(page_to_phys(pMemLnx->apPages[iPage]) < PhysHighest);
1072	#endif
1073	pMemLnx->Core.u.Cont.Phys = page_to_phys(pMemLnx->apPages[0]);
1074	*ppMem = &pMemLnx->Core;
1075	IPRT_LINUX_RESTORE_EFL_AC();
1076	return rc;
1077	}
1078
1079	rtR0MemObjLinuxFreePages(pMemLnx);
1080	rtR0MemObjDelete(&pMemLnx->Core);
1081	}
1082
1083	IPRT_LINUX_RESTORE_EFL_AC();
1084	return rc;
1085	}
1086
1087
1088	/**
1089	* Worker for rtR0MemObjLinuxAllocPhysSub that tries one allocation strategy.
1090	*
1091	* @returns IPRT status code.
1092	* @param ppMemLnx Where to
1093	* @param enmType The object type.
1094	* @param cb The size of the allocation.
1095	* @param uAlignment The alignment of the physical memory.
1096	* Only valid for fContiguous == true, ignored otherwise.
1097	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
1098	* @param pszTag Allocation tag used for statistics and such.
1099	* @param fGfp The Linux GFP flags to use for the allocation.
1100	*/
1101	static int rtR0MemObjLinuxAllocPhysSub2(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType,
1102	size_t cb, size_t uAlignment, RTHCPHYS PhysHighest, const char *pszTag, gfp_t fGfp)
1103	{
1104	PRTR0MEMOBJLNX pMemLnx;
1105	int rc = rtR0MemObjLinuxAllocPages(&pMemLnx, enmType, cb, uAlignment, fGfp,
1106	enmType == RTR0MEMOBJTYPE_PHYS /* contiguous / non-contiguous */,
1107	false /fExecutable/, VERR_NO_PHYS_MEMORY, pszTag);
1108	if (RT_FAILURE(rc))
1109	return rc;
1110
1111	/*
1112	* Check the addresses if necessary. (Can be optimized a bit for PHYS.)
1113	*/
1114	if (PhysHighest != NIL_RTHCPHYS)
1115	{
1116	size_t iPage = pMemLnx->cPages;
1117	while (iPage-- > 0)
1118	if (page_to_phys(pMemLnx->apPages[iPage]) > PhysHighest)
1119	{
1120	rtR0MemObjLinuxFreePages(pMemLnx);
1121	rtR0MemObjDelete(&pMemLnx->Core);
1122	return VERR_NO_MEMORY;
1123	}
1124	}
1125
1126	/*
1127	* Complete the object.
1128	*/
1129	if (enmType == RTR0MEMOBJTYPE_PHYS)
1130	{
1131	pMemLnx->Core.u.Phys.PhysBase = page_to_phys(pMemLnx->apPages[0]);
1132	pMemLnx->Core.u.Phys.fAllocated = true;
1133	}
1134	*ppMem = &pMemLnx->Core;
1135	return rc;
1136	}
1137
1138
1139	/**
1140	* Worker for rtR0MemObjNativeAllocPhys and rtR0MemObjNativeAllocPhysNC.
1141	*
1142	* @returns IPRT status code.
1143	* @param ppMem Where to store the memory object pointer on success.
1144	* @param enmType The object type.
1145	* @param cb The size of the allocation.
1146	* @param uAlignment The alignment of the physical memory.
1147	* Only valid for enmType == RTR0MEMOBJTYPE_PHYS, ignored otherwise.
1148	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
1149	* @param pszTag Allocation tag used for statistics and such.
1150	*/
1151	static int rtR0MemObjLinuxAllocPhysSub(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType,
1152	size_t cb, size_t uAlignment, RTHCPHYS PhysHighest, const char *pszTag)
1153	{
1154	int rc;
1155	IPRT_LINUX_SAVE_EFL_AC();
1156
1157	/*
1158	* There are two clear cases and that's the <=16MB and anything-goes ones.
1159	* When the physical address limit is somewhere in-between those two we'll
1160	* just have to try, starting with HIGHUSER and working our way thru the
1161	* different types, hoping we'll get lucky.
1162	*
1163	* We should probably move this physical address restriction logic up to
1164	* the page alloc function as it would be more efficient there. But since
1165	* we don't expect this to be a performance issue just yet it can wait.
1166	*/
1167	if (PhysHighest == NIL_RTHCPHYS)
1168	/* ZONE_HIGHMEM: the whole physical memory */
1169	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_HIGHUSER);
1170	else if (PhysHighest <= _1M * 16)
1171	/* ZONE_DMA: 0-16MB */
1172	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_DMA);
1173	else
1174	{
1175	rc = VERR_NO_MEMORY;
1176	if (RT_FAILURE(rc))
1177	/* ZONE_HIGHMEM: the whole physical memory */
1178	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_HIGHUSER);
1179	if (RT_FAILURE(rc))
1180	/* ZONE_NORMAL: 0-896MB */
1181	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_USER);
1182	#ifdef GFP_DMA32
1183	if (RT_FAILURE(rc))
1184	/* ZONE_DMA32: 0-4GB */
1185	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_DMA32);
1186	#endif
1187	if (RT_FAILURE(rc))
1188	/* ZONE_DMA: 0-16MB */
1189	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_DMA);
1190	}
1191	IPRT_LINUX_RESTORE_EFL_AC();
1192	return rc;
1193	}
1194
1195
1196	/**
1197	* Translates a kernel virtual address to a linux page structure by walking the
1198	* page tables.
1199	*
1200	* @note We do assume that the page tables will not change as we are walking
1201	* them. This assumption is rather forced by the fact that I could not
1202	* immediately see any way of preventing this from happening. So, we
1203	* take some extra care when accessing them.
1204	*
1205	* Because of this, we don't want to use this function on memory where
1206	* attribute changes to nearby pages is likely to cause large pages to
1207	* be used or split up. So, don't use this for the linear mapping of
1208	* physical memory.
1209	*
1210	* @returns Pointer to the page structur or NULL if it could not be found.
1211	* @param pv The kernel virtual address.
1212	*/
1213	RTDECL(struct page ) rtR0MemObjLinuxVirtToPage(void pv)
1214	{
1215	#if defined(RT_ARCH_AMD64) \|\| defined(RT_ARCH_X86)
1216	unsigned long ulAddr = (unsigned long)pv;
1217	unsigned long pfn;
1218	struct page *pPage;
1219	pte_t *pEntry;
1220	union
1221	{
1222	pgd_t Global;
1223	# if RTLNX_VER_MIN(4,12,0)
1224	p4d_t Four;
1225	# endif
1226	# if RTLNX_VER_MIN(2,6,11)
1227	pud_t Upper;
1228	# endif
1229	pmd_t Middle;
1230	pte_t Entry;
1231	} u;
1232
1233	/* Should this happen in a situation this code will be called in? And if
1234	* so, can it change under our feet? See also
1235	* "Documentation/vm/active_mm.txt" in the kernel sources. */
1236	if (RT_UNLIKELY(!current->active_mm))
1237	return NULL;
1238	u.Global = *pgd_offset(current->active_mm, ulAddr);
1239	if (RT_UNLIKELY(pgd_none(u.Global)))
1240	return NULL;
1241	# if RTLNX_VER_MIN(2,6,11)
1242	# if RTLNX_VER_MIN(4,12,0)
1243	u.Four = *p4d_offset(&u.Global, ulAddr);
1244	if (RT_UNLIKELY(p4d_none(u.Four)))
1245	return NULL;
1246	# if RTLNX_VER_MIN(5,6,0)
1247	if (p4d_leaf(u.Four))
1248	# else
1249	if (p4d_large(u.Four))
1250	# endif
1251	{
1252	pPage = p4d_page(u.Four);
1253	AssertReturn(pPage, NULL);
1254	pfn = page_to_pfn(pPage); /* doing the safe way... */
1255	AssertCompile(P4D_SHIFT - PAGE_SHIFT < 31);
1256	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (P4D_SHIFT - PAGE_SHIFT)) - 1);
1257	return pfn_to_page(pfn);
1258	}
1259	u.Upper = *pud_offset(&u.Four, ulAddr);
1260	# else /* < 4.12 */
1261	u.Upper = *pud_offset(&u.Global, ulAddr);
1262	# endif /* < 4.12 */
1263	if (RT_UNLIKELY(pud_none(u.Upper)))
1264	return NULL;
1265	# if RTLNX_VER_MIN(2,6,25)
1266	# if RTLNX_VER_MIN(5,6,0)
1267	if (pud_leaf(u.Upper))
1268	# else
1269	if (pud_large(u.Upper))
1270	# endif
1271	{
1272	pPage = pud_page(u.Upper);
1273	AssertReturn(pPage, NULL);
1274	pfn = page_to_pfn(pPage); /* doing the safe way... */
1275	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (PUD_SHIFT - PAGE_SHIFT)) - 1);
1276	return pfn_to_page(pfn);
1277	}
1278	# endif
1279	u.Middle = *pmd_offset(&u.Upper, ulAddr);
1280	# else /* < 2.6.11 */
1281	u.Middle = *pmd_offset(&u.Global, ulAddr);
1282	# endif /* < 2.6.11 */
1283	if (RT_UNLIKELY(pmd_none(u.Middle)))
1284	return NULL;
1285	# if RTLNX_VER_MIN(2,6,0)
1286	# if RTLNX_VER_MIN(5,6,0)
1287	if (pmd_leaf(u.Middle))
1288	# else
1289	if (pmd_large(u.Middle))
1290	# endif
1291	{
1292	pPage = pmd_page(u.Middle);
1293	AssertReturn(pPage, NULL);
1294	pfn = page_to_pfn(pPage); /* doing the safe way... */
1295	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (PMD_SHIFT - PAGE_SHIFT)) - 1);
1296	return pfn_to_page(pfn);
1297	}
1298	# endif
1299
1300	# if RTLNX_VER_MIN(6,5,0) \|\| RTLNX_RHEL_RANGE(9,4, 9,99)
1301	pEntry = __pte_map(&u.Middle, ulAddr);
1302	# elif RTLNX_VER_MIN(2,5,5) \|\| defined(pte_offset_map) /* As usual, RHEL 3 had pte_offset_map earlier. */
1303	pEntry = pte_offset_map(&u.Middle, ulAddr);
1304	# else
1305	pEntry = pte_offset(&u.Middle, ulAddr);
1306	# endif
1307	if (RT_UNLIKELY(!pEntry))
1308	return NULL;
1309	u.Entry = *pEntry;
1310	# if RTLNX_VER_MIN(2,5,5) \|\| defined(pte_offset_map)
1311	pte_unmap(pEntry);
1312	# endif
1313
1314	if (RT_UNLIKELY(!pte_present(u.Entry)))
1315	return NULL;
1316	return pte_page(u.Entry);
1317	#else /* !defined(RT_ARCH_AMD64) && !defined(RT_ARCH_X86) */
1318
1319	if (is_vmalloc_addr(pv))
1320	return vmalloc_to_page(pv);
1321
1322	return virt_to_page(pv);
1323	#endif
1324	}
1325	RT_EXPORT_SYMBOL(rtR0MemObjLinuxVirtToPage);
1326
1327
1328	DECLHIDDEN(int) rtR0MemObjNativeAllocPhys(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest, size_t uAlignment,
1329	const char *pszTag)
1330	{
1331	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS, cb, uAlignment, PhysHighest, pszTag);
1332	}
1333
1334
1335	DECLHIDDEN(int) rtR0MemObjNativeAllocPhysNC(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest, const char *pszTag)
1336	{
1337	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS_NC, cb, PAGE_SIZE, PhysHighest, pszTag);
1338	}
1339
1340
1341	DECLHIDDEN(int) rtR0MemObjNativeEnterPhys(PPRTR0MEMOBJINTERNAL ppMem, RTHCPHYS Phys, size_t cb, uint32_t uCachePolicy,
1342	const char *pszTag)
1343	{
1344	IPRT_LINUX_SAVE_EFL_AC();
1345
1346	/*
1347	* All we need to do here is to validate that we can use
1348	* ioremap on the specified address (32/64-bit dma_addr_t).
1349	*/
1350	PRTR0MEMOBJLNX pMemLnx;
1351	dma_addr_t PhysAddr = Phys;
1352	AssertMsgReturn(PhysAddr == Phys, ("%#llx\n", (unsigned long long)Phys), VERR_ADDRESS_TOO_BIG);
1353
1354	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_PHYS, NULL, cb, pszTag);
1355	if (!pMemLnx)
1356	{
1357	IPRT_LINUX_RESTORE_EFL_AC();
1358	return VERR_NO_MEMORY;
1359	}
1360
1361	pMemLnx->Core.u.Phys.PhysBase = PhysAddr;
1362	pMemLnx->Core.u.Phys.fAllocated = false;
1363	pMemLnx->Core.u.Phys.uCachePolicy = uCachePolicy;
1364	Assert(!pMemLnx->cPages);
1365	*ppMem = &pMemLnx->Core;
1366	IPRT_LINUX_RESTORE_EFL_AC();
1367	return VINF_SUCCESS;
1368	}
1369
1370	/* openSUSE Leap 42.3 detection :-/ */
1371	#if RTLNX_VER_RANGE(4,4,0, 4,6,0) && defined(FAULT_FLAG_REMOTE)
1372	# define GET_USER_PAGES_API KERNEL_VERSION(4, 10, 0) /* no typo! */
1373	#else
1374	# define GET_USER_PAGES_API LINUX_VERSION_CODE
1375	#endif
1376
1377	DECLHIDDEN(int) rtR0MemObjNativeLockUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3Ptr, size_t cb, uint32_t fAccess,
1378	RTR0PROCESS R0Process, const char *pszTag)
1379	{
1380	IPRT_LINUX_SAVE_EFL_AC();
1381	const int cPages = cb >> PAGE_SHIFT;
1382	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1383	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1384	struct vm_area_struct **papVMAs;
1385	# endif
1386	PRTR0MEMOBJLNX pMemLnx;
1387	int rc = VERR_NO_MEMORY;
1388	int const fWrite = fAccess & RTMEM_PROT_WRITE ? 1 : 0;
1389
1390	/*
1391	* Check for valid task and size overflows.
1392	*/
1393	if (!pTask)
1394	return VERR_NOT_SUPPORTED;
1395	if (((size_t)cPages << PAGE_SHIFT) != cb)
1396	return VERR_OUT_OF_RANGE;
1397
1398	/*
1399	* Allocate the memory object and a temporary buffer for the VMAs.
1400	*/
1401	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_UOFFSETOF_DYN(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK,
1402	(void *)R3Ptr, cb, pszTag);
1403	if (!pMemLnx)
1404	{
1405	IPRT_LINUX_RESTORE_EFL_AC();
1406	return VERR_NO_MEMORY;
1407	}
1408
1409	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1410	papVMAs = (struct vm_area_struct *)RTMemAlloc(sizeof(papVMAs) * cPages);
1411	if (papVMAs)
1412	{
1413	# endif
1414	LNX_MM_DOWN_READ(pTask->mm);
1415
1416	/*
1417	* Get user pages.
1418	*/
1419	/** @todo r=bird: Should we not force read access too? */
1420	#if GET_USER_PAGES_API >= KERNEL_VERSION(4, 6, 0)
1421	if (R0Process == RTR0ProcHandleSelf())
1422	rc = get_user_pages(R3Ptr, /* Where from. */
1423	cPages, /* How many pages. */
1424	# if GET_USER_PAGES_API >= KERNEL_VERSION(4, 9, 0)
1425	fWrite ? FOLL_WRITE \| /* Write to memory. */
1426	FOLL_FORCE /* force write access. */
1427	: 0, /* Write to memory. */
1428	# else
1429	fWrite, /* Write to memory. */
1430	fWrite, /* force write access. */
1431	# endif
1432	&pMemLnx->apPages[0] /* Page array. */
1433	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_SUSE_MAJ_PREREQ(15, 6) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1434	, papVMAs /* vmas */
1435	# endif
1436	);
1437	/*
1438	* Actually this should not happen at the moment as call this function
1439	* only for our own process.
1440	*/
1441	else
1442	rc = get_user_pages_remote(
1443	# if GET_USER_PAGES_API < KERNEL_VERSION(5, 9, 0)
1444	pTask, /* Task for fault accounting. */
1445	# endif
1446	pTask->mm, /* Whose pages. */
1447	R3Ptr, /* Where from. */
1448	cPages, /* How many pages. */
1449	# if GET_USER_PAGES_API >= KERNEL_VERSION(4, 9, 0)
1450	fWrite ? FOLL_WRITE \| /* Write to memory. */
1451	FOLL_FORCE /* force write access. */
1452	: 0, /* Write to memory. */
1453	# else
1454	fWrite, /* Write to memory. */
1455	fWrite, /* force write access. */
1456	# endif
1457	&pMemLnx->apPages[0] /* Page array. */
1458	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1459	, papVMAs /* vmas */
1460	# endif
1461	# if GET_USER_PAGES_API >= KERNEL_VERSION(4, 10, 0)
1462	, NULL /* locked */
1463	# endif
1464	);
1465	#else /* GET_USER_PAGES_API < KERNEL_VERSION(4, 6, 0) */
1466	rc = get_user_pages(pTask, /* Task for fault accounting. */
1467	pTask->mm, /* Whose pages. */
1468	R3Ptr, /* Where from. */
1469	cPages, /* How many pages. */
1470	/* The get_user_pages API change was back-ported to 4.4.168. */
1471	# if RTLNX_VER_RANGE(4,4,168, 4,5,0)
1472	fWrite ? FOLL_WRITE \| /* Write to memory. */
1473	FOLL_FORCE /* force write access. */
1474	: 0, /* Write to memory. */
1475	# else
1476	fWrite, /* Write to memory. */
1477	fWrite, /* force write access. */
1478	# endif
1479	&pMemLnx->apPages[0], /* Page array. */
1480	papVMAs /* vmas */
1481	);
1482	#endif /* GET_USER_PAGES_API < KERNEL_VERSION(4, 6, 0) */
1483	if (rc == cPages)
1484	{
1485	/*
1486	* Flush dcache (required?), protect against fork and _really_ pin the page
1487	* table entries. get_user_pages() will protect against swapping out the
1488	* pages but it will NOT protect against removing page table entries. This
1489	* can be achieved with
1490	* - using mlock / mmap(..., MAP_LOCKED, ...) from userland. This requires
1491	* an appropriate limit set up with setrlimit(..., RLIMIT_MEMLOCK, ...).
1492	* Usual Linux distributions support only a limited size of locked pages
1493	* (e.g. 32KB).
1494	* - setting the PageReserved bit (as we do in rtR0MemObjLinuxAllocPages()
1495	* or by
1496	* - setting the VM_LOCKED flag. This is the same as doing mlock() without
1497	* a range check.
1498	*/
1499	/** @todo The Linux fork() protection will require more work if this API
1500	* is to be used for anything but locking VM pages. */
1501	while (rc-- > 0)
1502	{
1503	flush_dcache_page(pMemLnx->apPages[rc]);
1504	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_SUSE_MAJ_PREREQ(15, 6) && !RTLNX_RHEL_RANGE(9,5, 9,99)
1505	# if RTLNX_VER_MIN(6,3,0)
1506	vm_flags_set(papVMAs[rc], VM_DONTCOPY \| VM_LOCKED);
1507	# else
1508	papVMAs[rc]->vm_flags \|= VM_DONTCOPY \| VM_LOCKED;
1509	# endif
1510	# endif
1511	}
1512
1513	LNX_MM_UP_READ(pTask->mm);
1514
1515	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1516	RTMemFree(papVMAs);
1517	# endif
1518
1519	pMemLnx->Core.u.Lock.R0Process = R0Process;
1520	pMemLnx->cPages = cPages;
1521	Assert(!pMemLnx->fMappedToRing0);
1522	*ppMem = &pMemLnx->Core;
1523
1524	IPRT_LINUX_RESTORE_EFL_AC();
1525	return VINF_SUCCESS;
1526	}
1527
1528	/*
1529	* Failed - we need to unlock any pages that we succeeded to lock.
1530	*/
1531	while (rc-- > 0)
1532	{
1533	if (!PageReserved(pMemLnx->apPages[rc]))
1534	SetPageDirty(pMemLnx->apPages[rc]);
1535	#if RTLNX_VER_MIN(4,6,0)
1536	put_page(pMemLnx->apPages[rc]);
1537	#else
1538	page_cache_release(pMemLnx->apPages[rc]);
1539	#endif
1540	}
1541
1542	LNX_MM_UP_READ(pTask->mm);
1543
1544	rc = VERR_LOCK_FAILED;
1545
1546	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1547	RTMemFree(papVMAs);
1548	}
1549	# endif
1550
1551	rtR0MemObjDelete(&pMemLnx->Core);
1552	IPRT_LINUX_RESTORE_EFL_AC();
1553	return rc;
1554	}
1555
1556
1557	DECLHIDDEN(int) rtR0MemObjNativeLockKernel(PPRTR0MEMOBJINTERNAL ppMem, void pv, size_t cb, uint32_t fAccess, const char pszTag)
1558	{
1559	IPRT_LINUX_SAVE_EFL_AC();
1560	void pvLast = (uint8_t )pv + cb - 1;
1561	size_t const cPages = cb >> PAGE_SHIFT;
1562	PRTR0MEMOBJLNX pMemLnx;
1563	bool fLinearMapping;
1564	int rc;
1565	uint8_t *pbPage;
1566	size_t iPage;
1567	NOREF(fAccess);
1568
1569	if ( !RTR0MemKernelIsValidAddr(pv)
1570	\|\| !RTR0MemKernelIsValidAddr(pv + cb))
1571	return VERR_INVALID_PARAMETER;
1572
1573	/*
1574	* The lower part of the kernel memory has a linear mapping between
1575	* physical and virtual addresses. So we take a short cut here. This is
1576	* assumed to be the cleanest way to handle those addresses (and the code
1577	* is well tested, though the test for determining it is not very nice).
1578	* If we ever decide it isn't we can still remove it.
1579	*/
1580	#if 0
1581	fLinearMapping = (unsigned long)pvLast < VMALLOC_START;
1582	#else
1583	fLinearMapping = (unsigned long)pv >= (unsigned long)__va(0)
1584	&& (unsigned long)pvLast < (unsigned long)high_memory;
1585	#endif
1586
1587	/*
1588	* Allocate the memory object.
1589	*/
1590	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_UOFFSETOF_DYN(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK,
1591	pv, cb, pszTag);
1592	if (!pMemLnx)
1593	{
1594	IPRT_LINUX_RESTORE_EFL_AC();
1595	return VERR_NO_MEMORY;
1596	}
1597
1598	/*
1599	* Gather the pages.
1600	* We ASSUME all kernel pages are non-swappable and non-movable.
1601	*/
1602	rc = VINF_SUCCESS;
1603	pbPage = (uint8_t *)pvLast;
1604	iPage = cPages;
1605	if (!fLinearMapping)
1606	{
1607	while (iPage-- > 0)
1608	{
1609	struct page *pPage = rtR0MemObjLinuxVirtToPage(pbPage);
1610	if (RT_UNLIKELY(!pPage))
1611	{
1612	rc = VERR_LOCK_FAILED;
1613	break;
1614	}
1615	pMemLnx->apPages[iPage] = pPage;
1616	pbPage -= PAGE_SIZE;
1617	}
1618	}
1619	else
1620	{
1621	while (iPage-- > 0)
1622	{
1623	pMemLnx->apPages[iPage] = virt_to_page(pbPage);
1624	pbPage -= PAGE_SIZE;
1625	}
1626	}
1627	if (RT_SUCCESS(rc))
1628	{
1629	/*
1630	* Complete the memory object and return.
1631	*/
1632	pMemLnx->Core.u.Lock.R0Process = NIL_RTR0PROCESS;
1633	pMemLnx->cPages = cPages;
1634	Assert(!pMemLnx->fMappedToRing0);
1635	*ppMem = &pMemLnx->Core;
1636
1637	IPRT_LINUX_RESTORE_EFL_AC();
1638	return VINF_SUCCESS;
1639	}
1640
1641	rtR0MemObjDelete(&pMemLnx->Core);
1642	IPRT_LINUX_RESTORE_EFL_AC();
1643	return rc;
1644	}
1645
1646
1647	DECLHIDDEN(int) rtR0MemObjNativeReserveKernel(PPRTR0MEMOBJINTERNAL ppMem, void *pvFixed, size_t cb, size_t uAlignment,
1648	const char *pszTag)
1649	{
1650	#if RTLNX_VER_MIN(2,4,22)
1651	IPRT_LINUX_SAVE_EFL_AC();
1652	const size_t cPages = cb >> PAGE_SHIFT;
1653	struct page *pDummyPage;
1654	struct page **papPages;
1655
1656	/* check for unsupported stuff. */
1657	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
1658	if (uAlignment > PAGE_SIZE)
1659	return VERR_NOT_SUPPORTED;
1660
1661	/*
1662	* Allocate a dummy page and create a page pointer array for vmap such that
1663	* the dummy page is mapped all over the reserved area.
1664	*/
1665	pDummyPage = alloc_page(GFP_HIGHUSER \| __GFP_NOWARN);
1666	if (pDummyPage)
1667	{
1668	papPages = RTMemAlloc(sizeof(papPages) cPages);
1669	if (papPages)
1670	{
1671	void *pv;
1672	size_t iPage = cPages;
1673	while (iPage-- > 0)
1674	papPages[iPage] = pDummyPage;
1675	# ifdef VM_MAP
1676	pv = vmap(papPages, cPages, VM_MAP, PAGE_KERNEL_RO);
1677	# else
1678	pv = vmap(papPages, cPages, VM_ALLOC, PAGE_KERNEL_RO);
1679	# endif
1680	RTMemFree(papPages);
1681	if (pv)
1682	{
1683	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb, pszTag);
1684	if (pMemLnx)
1685	{
1686	pMemLnx->Core.u.ResVirt.R0Process = NIL_RTR0PROCESS;
1687	pMemLnx->cPages = 1;
1688	pMemLnx->apPages[0] = pDummyPage;
1689	*ppMem = &pMemLnx->Core;
1690	IPRT_LINUX_RESTORE_EFL_AC();
1691	return VINF_SUCCESS;
1692	}
1693	vunmap(pv);
1694	}
1695	}
1696	__free_page(pDummyPage);
1697	}
1698	IPRT_LINUX_RESTORE_EFL_AC();
1699	return VERR_NO_MEMORY;
1700
1701	#else /* < 2.4.22 */
1702	/*
1703	* Could probably use ioremap here, but the caller is in a better position than us
1704	* to select some safe physical memory.
1705	*/
1706	return VERR_NOT_SUPPORTED;
1707	#endif
1708	}
1709
1710
1711	DECLHIDDEN(int) rtR0MemObjNativeReserveUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment,
1712	RTR0PROCESS R0Process, const char *pszTag)
1713	{
1714	IPRT_LINUX_SAVE_EFL_AC();
1715	PRTR0MEMOBJLNX pMemLnx;
1716	void *pv;
1717	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1718	if (!pTask)
1719	return VERR_NOT_SUPPORTED;
1720
1721	/*
1722	* Check that the specified alignment is supported.
1723	*/
1724	if (uAlignment > PAGE_SIZE)
1725	return VERR_NOT_SUPPORTED;
1726
1727	/*
1728	* Let rtR0MemObjLinuxDoMmap do the difficult bits.
1729	*/
1730	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, cb, uAlignment, pTask, RTMEM_PROT_NONE);
1731	if (pv == (void *)-1)
1732	{
1733	IPRT_LINUX_RESTORE_EFL_AC();
1734	return VERR_NO_MEMORY;
1735	}
1736
1737	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb, pszTag);
1738	if (!pMemLnx)
1739	{
1740	rtR0MemObjLinuxDoMunmap(pv, cb, pTask);
1741	IPRT_LINUX_RESTORE_EFL_AC();
1742	return VERR_NO_MEMORY;
1743	}
1744
1745	pMemLnx->Core.u.ResVirt.R0Process = R0Process;
1746	*ppMem = &pMemLnx->Core;
1747	IPRT_LINUX_RESTORE_EFL_AC();
1748	return VINF_SUCCESS;
1749	}
1750
1751
1752	DECLHIDDEN(int) rtR0MemObjNativeMapKernel(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap, void *pvFixed, size_t uAlignment,
1753	unsigned fProt, size_t offSub, size_t cbSub, const char *pszTag)
1754	{
1755	int rc = VERR_NO_MEMORY;
1756	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1757	PRTR0MEMOBJLNX pMemLnx;
1758	IPRT_LINUX_SAVE_EFL_AC();
1759
1760	/* Fail if requested to do something we can't. */
1761	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
1762	if (uAlignment > PAGE_SIZE)
1763	return VERR_NOT_SUPPORTED;
1764
1765	/*
1766	* Create the IPRT memory object.
1767	*/
1768	if (!cbSub)
1769	cbSub = pMemLnxToMap->Core.cb - offSub;
1770	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, cbSub, pszTag);
1771	if (pMemLnx)
1772	{
1773	if (pMemLnxToMap->cPages)
1774	{
1775	#if RTLNX_VER_MIN(2,4,22)
1776	/*
1777	* Use vmap - 2.4.22 and later.
1778	*/
1779	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, true /* kernel */);
1780	/** @todo We don't really care too much for EXEC here... 5.8 always adds NX. */
1781	Assert(((offSub + cbSub) >> PAGE_SHIFT) <= pMemLnxToMap->cPages);
1782	# ifdef VM_MAP
1783	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[offSub >> PAGE_SHIFT], cbSub >> PAGE_SHIFT, VM_MAP, fPg);
1784	# else
1785	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[offSub >> PAGE_SHIFT], cbSub >> PAGE_SHIFT, VM_ALLOC, fPg);
1786	# endif
1787	if (pMemLnx->Core.pv)
1788	{
1789	pMemLnx->fMappedToRing0 = true;
1790	rc = VINF_SUCCESS;
1791	}
1792	else
1793	rc = VERR_MAP_FAILED;
1794
1795	#else /* < 2.4.22 */
1796	/*
1797	* Only option here is to share mappings if possible and forget about fProt.
1798	*/
1799	if (rtR0MemObjIsRing3(pMemToMap))
1800	rc = VERR_NOT_SUPPORTED;
1801	else
1802	{
1803	rc = VINF_SUCCESS;
1804	if (!pMemLnxToMap->Core.pv)
1805	rc = rtR0MemObjLinuxVMap(pMemLnxToMap, !!(fProt & RTMEM_PROT_EXEC));
1806	if (RT_SUCCESS(rc))
1807	{
1808	Assert(pMemLnxToMap->Core.pv);
1809	pMemLnx->Core.pv = (uint8_t *)pMemLnxToMap->Core.pv + offSub;
1810	}
1811	}
1812	#endif
1813	}
1814	else
1815	{
1816	/*
1817	* MMIO / physical memory.
1818	*/
1819	Assert(pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS && !pMemLnxToMap->Core.u.Phys.fAllocated);
1820	#if RTLNX_VER_MIN(2,6,25)
1821	/*
1822	* ioremap() defaults to no caching since the 2.6 kernels.
1823	* ioremap_nocache() has been removed finally in 5.6-rc1.
1824	*/
1825	pMemLnx->Core.pv = pMemLnxToMap->Core.u.Phys.uCachePolicy == RTMEM_CACHE_POLICY_MMIO
1826	? ioremap(pMemLnxToMap->Core.u.Phys.PhysBase + offSub, cbSub)
1827	: ioremap_cache(pMemLnxToMap->Core.u.Phys.PhysBase + offSub, cbSub);
1828	#else /* KERNEL_VERSION < 2.6.25 */
1829	pMemLnx->Core.pv = pMemLnxToMap->Core.u.Phys.uCachePolicy == RTMEM_CACHE_POLICY_MMIO
1830	? ioremap_nocache(pMemLnxToMap->Core.u.Phys.PhysBase + offSub, cbSub)
1831	: ioremap(pMemLnxToMap->Core.u.Phys.PhysBase + offSub, cbSub);
1832	#endif /* KERNEL_VERSION < 2.6.25 */
1833	if (pMemLnx->Core.pv)
1834	{
1835	/** @todo fix protection. */
1836	rc = VINF_SUCCESS;
1837	}
1838	}
1839	if (RT_SUCCESS(rc))
1840	{
1841	pMemLnx->Core.u.Mapping.R0Process = NIL_RTR0PROCESS;
1842	*ppMem = &pMemLnx->Core;
1843	IPRT_LINUX_RESTORE_EFL_AC();
1844	return VINF_SUCCESS;
1845	}
1846	rtR0MemObjDelete(&pMemLnx->Core);
1847	}
1848
1849	IPRT_LINUX_RESTORE_EFL_AC();
1850	return rc;
1851	}
1852
1853
1854	#ifdef VBOX_USE_PAE_HACK
1855	/**
1856	* Replace the PFN of a PTE with the address of the actual page.
1857	*
1858	* The caller maps a reserved dummy page at the address with the desired access
1859	* and flags.
1860	*
1861	* This hack is required for older Linux kernels which don't provide
1862	* remap_pfn_range().
1863	*
1864	* @returns 0 on success, -ENOMEM on failure.
1865	* @param mm The memory context.
1866	* @param ulAddr The mapping address.
1867	* @param Phys The physical address of the page to map.
1868	*/
1869	static int rtR0MemObjLinuxFixPte(struct mm_struct *mm, unsigned long ulAddr, RTHCPHYS Phys)
1870	{
1871	int rc = -ENOMEM;
1872	pgd_t *pgd;
1873
1874	spin_lock(&mm->page_table_lock);
1875
1876	pgd = pgd_offset(mm, ulAddr);
1877	if (!pgd_none(pgd) && !pgd_bad(pgd))
1878	{
1879	pmd_t *pmd = pmd_offset(pgd, ulAddr);
1880	if (!pmd_none(*pmd))
1881	{
1882	pte_t *ptep = pte_offset_map(pmd, ulAddr);
1883	if (ptep)
1884	{
1885	pte_t pte = *ptep;
1886	pte.pte_high &= 0xfff00000;
1887	pte.pte_high \|= ((Phys >> 32) & 0x000fffff);
1888	pte.pte_low &= 0x00000fff;
1889	pte.pte_low \|= (Phys & 0xfffff000);
1890	set_pte(ptep, pte);
1891	pte_unmap(ptep);
1892	rc = 0;
1893	}
1894	}
1895	}
1896
1897	spin_unlock(&mm->page_table_lock);
1898	return rc;
1899	}
1900	#endif /* VBOX_USE_PAE_HACK */
1901
1902
1903	DECLHIDDEN(int) rtR0MemObjNativeMapUser(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap, RTR3PTR R3PtrFixed, size_t uAlignment,
1904	unsigned fProt, RTR0PROCESS R0Process, size_t offSub, size_t cbSub, const char *pszTag)
1905	{
1906	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1907	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1908	int rc = VERR_NO_MEMORY;
1909	PRTR0MEMOBJLNX pMemLnx;
1910	#ifdef VBOX_USE_PAE_HACK
1911	struct page *pDummyPage;
1912	RTHCPHYS DummyPhys;
1913	#endif
1914	IPRT_LINUX_SAVE_EFL_AC();
1915
1916	/*
1917	* Check for restrictions.
1918	*/
1919	if (!pTask)
1920	return VERR_NOT_SUPPORTED;
1921	if (uAlignment > PAGE_SIZE)
1922	return VERR_NOT_SUPPORTED;
1923
1924	#ifdef VBOX_USE_PAE_HACK
1925	/*
1926	* Allocate a dummy page for use when mapping the memory.
1927	*/
1928	pDummyPage = alloc_page(GFP_USER \| __GFP_NOWARN);
1929	if (!pDummyPage)
1930	{
1931	IPRT_LINUX_RESTORE_EFL_AC();
1932	return VERR_NO_MEMORY;
1933	}
1934	SetPageReserved(pDummyPage);
1935	DummyPhys = page_to_phys(pDummyPage);
1936	#endif
1937
1938	/*
1939	* Create the IPRT memory object.
1940	*/
1941	Assert(!offSub \|\| cbSub);
1942	if (cbSub == 0)
1943	cbSub = pMemLnxToMap->Core.cb;
1944	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, cbSub, pszTag);
1945	if (pMemLnx)
1946	{
1947	/*
1948	* Allocate user space mapping.
1949	*/
1950	void *pv;
1951	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, cbSub, uAlignment, pTask, fProt);
1952	if (pv != (void *)-1)
1953	{
1954	/*
1955	* Map page by page into the mmap area.
1956	* This is generic, paranoid and not very efficient.
1957	*/
1958	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, false /* user */);
1959	unsigned long ulAddrCur = (unsigned long)pv;
1960	const size_t cPages = (offSub + cbSub) >> PAGE_SHIFT;
1961	size_t iPage;
1962
1963	LNX_MM_DOWN_WRITE(pTask->mm);
1964
1965	rc = VINF_SUCCESS;
1966	if (pMemLnxToMap->cPages)
1967	{
1968	for (iPage = offSub >> PAGE_SHIFT; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE)
1969	{
1970	#if RTLNX_VER_MAX(2,6,11)
1971	RTHCPHYS Phys = page_to_phys(pMemLnxToMap->apPages[iPage]);
1972	#endif
1973	#if RTLNX_VER_MIN(2,6,0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1974	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
1975	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
1976	#endif
1977	#if RTLNX_VER_MAX(2,6,0) && defined(RT_ARCH_X86)
1978	/* remap_page_range() limitation on x86 */
1979	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
1980	#endif
1981
1982	#if defined(VBOX_USE_INSERT_PAGE) && RTLNX_VER_MIN(2,6,22)
1983	rc = vm_insert_page(vma, ulAddrCur, pMemLnxToMap->apPages[iPage]);
1984	/* Thes flags help making 100% sure some bad stuff wont happen (swap, core, ++).
1985	* See remap_pfn_range() in mm/memory.c */
1986
1987	#if RTLNX_VER_MIN(6,3,0) \|\| RTLNX_RHEL_RANGE(9,5, 9,99)
1988	vm_flags_set(vma, VM_DONTEXPAND \| VM_DONTDUMP);
1989	#elif RTLNX_VER_MIN(3,7,0)
1990	vma->vm_flags \|= VM_DONTEXPAND \| VM_DONTDUMP;
1991	#else
1992	vma->vm_flags \|= VM_RESERVED;
1993	#endif
1994	#elif RTLNX_VER_MIN(2,6,11)
1995	rc = remap_pfn_range(vma, ulAddrCur, page_to_pfn(pMemLnxToMap->apPages[iPage]), PAGE_SIZE, fPg);
1996	#elif defined(VBOX_USE_PAE_HACK)
1997	rc = remap_page_range(vma, ulAddrCur, DummyPhys, PAGE_SIZE, fPg);
1998	if (!rc)
1999	rc = rtR0MemObjLinuxFixPte(pTask->mm, ulAddrCur, Phys);
2000	#elif RTLNX_VER_MIN(2,6,0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
2001	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
2002	#else /* 2.4 */
2003	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
2004	#endif
2005	if (rc)
2006	{
2007	rc = VERR_NO_MEMORY;
2008	break;
2009	}
2010	}
2011	}
2012	else
2013	{
2014	RTHCPHYS Phys;
2015	if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS)
2016	Phys = pMemLnxToMap->Core.u.Phys.PhysBase;
2017	else if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_CONT)
2018	Phys = pMemLnxToMap->Core.u.Cont.Phys;
2019	else
2020	{
2021	AssertMsgFailed(("%d\n", pMemLnxToMap->Core.enmType));
2022	Phys = NIL_RTHCPHYS;
2023	}
2024	if (Phys != NIL_RTHCPHYS)
2025	{
2026	for (iPage = offSub >> PAGE_SHIFT; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE, Phys += PAGE_SIZE)
2027	{
2028	#if RTLNX_VER_MIN(2,6,0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
2029	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
2030	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
2031	#endif
2032	#if RTLNX_VER_MAX(2,6,0) && defined(RT_ARCH_X86)
2033	/* remap_page_range() limitation on x86 */
2034	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
2035	#endif
2036
2037	#if RTLNX_VER_MIN(2,6,11)
2038	rc = remap_pfn_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
2039	#elif defined(VBOX_USE_PAE_HACK)
2040	rc = remap_page_range(vma, ulAddrCur, DummyPhys, PAGE_SIZE, fPg);
2041	if (!rc)
2042	rc = rtR0MemObjLinuxFixPte(pTask->mm, ulAddrCur, Phys);
2043	#elif RTLNX_VER_MIN(2,6,0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
2044	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
2045	#else /* 2.4 */
2046	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
2047	#endif
2048	if (rc)
2049	{
2050	rc = VERR_NO_MEMORY;
2051	break;
2052	}
2053	}
2054	}
2055	}
2056
2057	#ifdef CONFIG_NUMA_BALANCING
2058	# if RTLNX_VER_MAX(3,13,0) && RTLNX_RHEL_MAX(7,0)
2059	# define VBOX_NUMA_HACK_OLD
2060	# endif
2061	if (RT_SUCCESS(rc))
2062	{
2063	/** @todo Ugly hack! But right now we have no other means to
2064	* disable automatic NUMA page balancing. */
2065	# ifdef RT_OS_X86
2066	# ifdef VBOX_NUMA_HACK_OLD
2067	pTask->mm->numa_next_reset = jiffies + 0x7fffffffUL;
2068	# endif
2069	pTask->mm->numa_next_scan = jiffies + 0x7fffffffUL;
2070	# else
2071	# ifdef VBOX_NUMA_HACK_OLD
2072	pTask->mm->numa_next_reset = jiffies + 0x7fffffffffffffffUL;
2073	# endif
2074	pTask->mm->numa_next_scan = jiffies + 0x7fffffffffffffffUL;
2075	# endif
2076	}
2077	#endif /* CONFIG_NUMA_BALANCING */
2078
2079	LNX_MM_UP_WRITE(pTask->mm);
2080
2081	if (RT_SUCCESS(rc))
2082	{
2083	#ifdef VBOX_USE_PAE_HACK
2084	__free_page(pDummyPage);
2085	#endif
2086	pMemLnx->Core.pv = pv;
2087	pMemLnx->Core.u.Mapping.R0Process = R0Process;
2088	*ppMem = &pMemLnx->Core;
2089	IPRT_LINUX_RESTORE_EFL_AC();
2090	return VINF_SUCCESS;
2091	}
2092
2093	/*
2094	* Bail out.
2095	*/
2096	rtR0MemObjLinuxDoMunmap(pv, cbSub, pTask);
2097	}
2098	rtR0MemObjDelete(&pMemLnx->Core);
2099	}
2100	#ifdef VBOX_USE_PAE_HACK
2101	__free_page(pDummyPage);
2102	#endif
2103
2104	IPRT_LINUX_RESTORE_EFL_AC();
2105	return rc;
2106	}
2107
2108
2109	DECLHIDDEN(int) rtR0MemObjNativeProtect(PRTR0MEMOBJINTERNAL pMem, size_t offSub, size_t cbSub, uint32_t fProt)
2110	{
2111	# ifdef IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
2112	/*
2113	* Currently only supported when we've got addresses PTEs from the kernel.
2114	*/
2115	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
2116	if (pMemLnx->pArea && pMemLnx->papPtesForArea)
2117	{
2118	pgprot_t const fPg = rtR0MemObjLinuxConvertProt(fProt, true /fKernel/);
2119	size_t const cPages = (offSub + cbSub) >> PAGE_SHIFT;
2120	pte_t **papPtes = pMemLnx->papPtesForArea;
2121	size_t i;
2122
2123	for (i = offSub >> PAGE_SHIFT; i < cPages; i++)
2124	{
2125	set_pte(papPtes[i], mk_pte(pMemLnx->apPages[i], fPg));
2126	}
2127	preempt_disable();
2128	__flush_tlb_all();
2129	preempt_enable();
2130	return VINF_SUCCESS;
2131	}
2132	# elif defined(IPRT_USE_APPLY_TO_PAGE_RANGE_FOR_EXEC)
2133	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
2134	if ( pMemLnx->fExecutable
2135	&& pMemLnx->fMappedToRing0)
2136	{
2137	LNXAPPLYPGRANGE Args;
2138	Args.pMemLnx = pMemLnx;
2139	Args.fPg = rtR0MemObjLinuxConvertProt(fProt, true /fKernel/);
2140	int rcLnx = apply_to_page_range(current->active_mm, (unsigned long)pMemLnx->Core.pv + offSub, cbSub,
2141	rtR0MemObjLinuxApplyPageRange, (void *)&Args);
2142	if (rcLnx)
2143	return VERR_NOT_SUPPORTED;
2144
2145	return VINF_SUCCESS;
2146	}
2147	# endif
2148
2149	NOREF(pMem);
2150	NOREF(offSub);
2151	NOREF(cbSub);
2152	NOREF(fProt);
2153	return VERR_NOT_SUPPORTED;
2154	}
2155
2156
2157	DECLHIDDEN(RTHCPHYS) rtR0MemObjNativeGetPagePhysAddr(PRTR0MEMOBJINTERNAL pMem, size_t iPage)
2158	{
2159	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
2160
2161	if (pMemLnx->cPages)
2162	return page_to_phys(pMemLnx->apPages[iPage]);
2163
2164	switch (pMemLnx->Core.enmType)
2165	{
2166	case RTR0MEMOBJTYPE_CONT:
2167	return pMemLnx->Core.u.Cont.Phys + (iPage << PAGE_SHIFT);
2168
2169	case RTR0MEMOBJTYPE_PHYS:
2170	return pMemLnx->Core.u.Phys.PhysBase + (iPage << PAGE_SHIFT);
2171
2172	/* the parent knows */
2173	case RTR0MEMOBJTYPE_MAPPING:
2174	return rtR0MemObjNativeGetPagePhysAddr(pMemLnx->Core.uRel.Child.pParent, iPage);
2175
2176	/* cPages > 0 */
2177	case RTR0MEMOBJTYPE_LOW:
2178	case RTR0MEMOBJTYPE_LOCK:
2179	case RTR0MEMOBJTYPE_PHYS_NC:
2180	case RTR0MEMOBJTYPE_PAGE:
2181	case RTR0MEMOBJTYPE_LARGE_PAGE:
2182	default:
2183	AssertMsgFailed(("%d\n", pMemLnx->Core.enmType));
2184	RT_FALL_THROUGH();
2185
2186	case RTR0MEMOBJTYPE_RES_VIRT:
2187	return NIL_RTHCPHYS;
2188	}
2189	}
2190
2191
2192	DECLHIDDEN(int) rtR0MemObjNativeZeroInitWithoutMapping(PRTR0MEMOBJINTERNAL pMem)
2193	{
2194	PRTR0MEMOBJLNX const pMemLnx = (PRTR0MEMOBJLNX)pMem;
2195	size_t const cPages = pMemLnx->Core.cb >> PAGE_SHIFT;
2196	size_t iPage;
2197	/** @todo optimize this. */
2198	for (iPage = 0; iPage < cPages; iPage++)
2199	{
2200	void *pvPage;
2201
2202	/* Get the physical address of the page. */
2203	RTHCPHYS const HCPhys = rtR0MemObjNativeGetPagePhysAddr(&pMemLnx->Core, iPage);
2204	AssertReturn(HCPhys != NIL_RTHCPHYS, VERR_INTERNAL_ERROR_3);
2205	Assert(!(HCPhys & PAGE_OFFSET_MASK));
2206
2207	/* Would've like to use valid_phys_addr_range for this test, but it isn't exported. */
2208	AssertReturn((HCPhys \| PAGE_OFFSET_MASK) < __pa(high_memory), VERR_INTERNAL_ERROR_3);
2209
2210	/* Map it. */
2211	pvPage = phys_to_virt(HCPhys);
2212	AssertPtrReturn(pvPage, VERR_INTERNAL_ERROR_3);
2213
2214	/* Zero it. */
2215	RT_BZERO(pvPage, PAGE_SIZE);
2216	}
2217	return VINF_SUCCESS;
2218	}
2219

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

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

Download in other formats: