memobj-r0drv-linux.c

Last change on this file was 109076, checked in by vboxsync, 9 days ago
Runtime/r0drv/linux/memobj-r0drv-linux.c: Try using GFP_KERNEL to allow reclaiming memory from the page cache
Property svn:eol-style set to `native` Property svn:keywords set to `Id Rev Revision`
File size: 76.4 KB

Line
1	/* $Id: memobj-r0drv-linux.c 109076 2025-04-24 18:17:25Z 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	#if RTLNX_VER_MIN(2,6,0)
66	# ifdef __GFP_REPEAT
67	# define MY_GFP_REPEAT __GFP_REPEAT
68	# elif defined (__GFP_RETRY_MAYFAIL) /* Renamed in commit dcda9b04713c3f6ff0875652924844fae28286ea . */
69	# define MY_GFP_REPEAT __GFP_RETRY_MAYFAIL
70	# else /* This is to notice when the flags are renamed/moved around again. */
71	# error "Was this flag renamed again?"
72	# endif
73	# else
74	# define MY_GFP_REPEAT 0
75	#endif
76
77	/** @def IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
78	* Whether we use alloc_vm_area (3.2+) for executable memory.
79	* This is a must for 5.8+, but we enable it all the way back to 3.2.x for
80	* better W^R compliance (fExecutable flag). */
81	#if RTLNX_VER_RANGE(3,2,0, 5,10,0) \|\| defined(DOXYGEN_RUNNING)
82	# define IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
83	#endif
84	/** @def IPRT_USE_APPLY_TO_PAGE_RANGE_FOR_EXEC
85	* alloc_vm_area was removed with 5.10 so we have to resort to a different way
86	* to allocate executable memory.
87	* It would be possible to remove IPRT_USE_ALLOC_VM_AREA_FOR_EXEC and use
88	* this path execlusively for 3.2+ but no time to test it really works on every
89	* supported kernel, so better play safe for now.
90	*/
91	#if RTLNX_VER_MIN(5,10,0) \|\| defined(DOXYGEN_RUNNING)
92	# define IPRT_USE_APPLY_TO_PAGE_RANGE_FOR_EXEC
93	#endif
94
95	/*
96	* 2.6.29+ kernels don't work with remap_pfn_range() anymore because
97	* track_pfn_vma_new() is apparently not defined for non-RAM pages.
98	* It should be safe to use vm_insert_page() older kernels as well.
99	*/
100	#if RTLNX_VER_MIN(2,6,23)
101	# define VBOX_USE_INSERT_PAGE
102	#endif
103	#if defined(CONFIG_X86_PAE) \
104	&& ( defined(HAVE_26_STYLE_REMAP_PAGE_RANGE) \
105	\|\| RTLNX_VER_RANGE(2,6,0, 2,6,11) )
106	# define VBOX_USE_PAE_HACK
107	#endif
108
109	/* gfp_t was introduced in 2.6.14, define it for earlier. */
110	#if RTLNX_VER_MAX(2,6,14)
111	# define gfp_t unsigned
112	#endif
113
114	/*
115	* Wrappers around mmap_lock/mmap_sem difference.
116	*/
117	#if RTLNX_VER_MIN(5,8,0)
118	# define LNX_MM_DOWN_READ(a_pMm) down_read(&(a_pMm)->mmap_lock)
119	# define LNX_MM_UP_READ(a_pMm) up_read(&(a_pMm)->mmap_lock)
120	# define LNX_MM_DOWN_WRITE(a_pMm) down_write(&(a_pMm)->mmap_lock)
121	# define LNX_MM_UP_WRITE(a_pMm) up_write(&(a_pMm)->mmap_lock)
122	#else
123	# define LNX_MM_DOWN_READ(a_pMm) down_read(&(a_pMm)->mmap_sem)
124	# define LNX_MM_UP_READ(a_pMm) up_read(&(a_pMm)->mmap_sem)
125	# define LNX_MM_DOWN_WRITE(a_pMm) down_write(&(a_pMm)->mmap_sem)
126	# define LNX_MM_UP_WRITE(a_pMm) up_write(&(a_pMm)->mmap_sem)
127	#endif
128
129
130	/*********************************************************************************************************************************
131	* Structures and Typedefs *
132	*********************************************************************************************************************************/
133	/**
134	* The Linux version of the memory object structure.
135	*/
136	typedef struct RTR0MEMOBJLNX
137	{
138	/** The core structure. */
139	RTR0MEMOBJINTERNAL Core;
140	/** Set if the allocation is contiguous.
141	* This means it has to be given back as one chunk. */
142	bool fContiguous;
143	/** Set if executable allocation. */
144	bool fExecutable;
145	/** Set if we've vmap'ed the memory into ring-0. */
146	bool fMappedToRing0;
147	/** This is non-zero if large page allocation. */
148	uint8_t cLargePageOrder;
149	#ifdef IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
150	/** Return from alloc_vm_area() that we now need to use for executable
151	* memory. */
152	struct vm_struct *pArea;
153	/** PTE array that goes along with pArea (must be freed). */
154	pte_t **papPtesForArea;
155	#endif
156	/** The pages in the apPages array. */
157	size_t cPages;
158	/** Array of struct page pointers. (variable size) */
159	RT_FLEXIBLE_ARRAY_EXTENSION
160	struct page *apPages[RT_FLEXIBLE_ARRAY];
161	} RTR0MEMOBJLNX;
162	/** Pointer to the linux memory object. */
163	typedef RTR0MEMOBJLNX *PRTR0MEMOBJLNX;
164
165
166	/*********************************************************************************************************************************
167	* Global Variables *
168	*********************************************************************************************************************************/
169	/*
170	* Linux allows only a coarse selection of zones for
171	* allocations matching a particular maximum physical address.
172	*
173	* Sorted from high to low physical address!
174	*/
175	static const struct
176	{
177	RTHCPHYS PhysHighest;
178	gfp_t fGfp;
179	} g_aZones[] =
180	{
181	{ NIL_RTHCPHYS, GFP_KERNEL },
182	#if (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
183	{ _4G - 1, GFP_DMA32 }, /* ZONE_DMA32: 0-4GB */
184	#elif defined(RT_ARCH_ARM32) \|\| defined(RT_ARCH_ARM64)
185	{ _4G - 1, GFP_DMA }, /* ZONE_DMA: 0-4GB */
186	#endif
187	#if defined(RT_ARCH_AMD64)
188	{ _16M - 1, GFP_DMA }, /* ZONE_DMA: 0-16MB */
189	#elif defined(RT_ARCH_X86)
190	{ 896 * _1M - 1, GFP_USER }, /* ZONE_NORMAL (32-bit hosts): 0-896MB */
191	#endif
192	};
193
194
195	static void rtR0MemObjLinuxFreePages(PRTR0MEMOBJLNX pMemLnx);
196
197
198	/**
199	* Helper that converts from a RTR0PROCESS handle to a linux task.
200	*
201	* @returns The corresponding Linux task.
202	* @param R0Process IPRT ring-0 process handle.
203	*/
204	static struct task_struct *rtR0ProcessToLinuxTask(RTR0PROCESS R0Process)
205	{
206	/** @todo fix rtR0ProcessToLinuxTask!! */
207	/** @todo many (all?) callers currently assume that we return 'current'! */
208	return R0Process == RTR0ProcHandleSelf() ? current : NULL;
209	}
210
211
212	/**
213	* Compute order. Some functions allocate 2^order pages.
214	*
215	* @returns order.
216	* @param cPages Number of pages.
217	*/
218	static int rtR0MemObjLinuxOrder(size_t cPages)
219	{
220	int iOrder;
221	size_t cTmp;
222
223	for (iOrder = 0, cTmp = cPages; cTmp >>= 1; ++iOrder)
224	;
225	if (cPages & ~((size_t)1 << iOrder))
226	++iOrder;
227
228	return iOrder;
229	}
230
231
232	/**
233	* Converts from RTMEM_PROT_* to Linux PAGE_*.
234	*
235	* @returns Linux page protection constant.
236	* @param fProt The IPRT protection mask.
237	* @param fKernel Whether it applies to kernel or user space.
238	*/
239	static pgprot_t rtR0MemObjLinuxConvertProt(unsigned fProt, bool fKernel)
240	{
241	switch (fProt)
242	{
243	default:
244	AssertMsgFailed(("%#x %d\n", fProt, fKernel)); RT_FALL_THRU();
245	case RTMEM_PROT_NONE:
246	return PAGE_NONE;
247
248	case RTMEM_PROT_READ:
249	return fKernel ? PAGE_KERNEL_RO : PAGE_READONLY;
250
251	case RTMEM_PROT_WRITE:
252	case RTMEM_PROT_WRITE \| RTMEM_PROT_READ:
253	return fKernel ? PAGE_KERNEL : PAGE_SHARED;
254
255	case RTMEM_PROT_EXEC:
256	case RTMEM_PROT_EXEC \| RTMEM_PROT_READ:
257	#if defined(RT_ARCH_X86) \|\| defined(RT_ARCH_AMD64)
258	if (fKernel)
259	{
260	# if RTLNX_VER_MIN(6,6,0) \|\| RTLNX_SUSE_MAJ_PREREQ(15, 6)
261	/* In kernel 6.6 mk_pte() macro was fortified with additional
262	* check which does not allow to use our custom mask anymore
263	* (see kernel commit ae1f05a617dcbc0a732fbeba0893786cd009536c).
264	* For this particular mapping case, an existing mask PAGE_KERNEL_ROX
265	* can be used instead. PAGE_KERNEL_ROX was introduced in
266	* kernel 5.8, however, lets apply it for kernels 6.6 and newer
267	* to be on a safe side.
268	*/
269	return PAGE_KERNEL_ROX;
270	# else
271	pgprot_t fPg = MY_PAGE_KERNEL_EXEC;
272	pgprot_val(fPg) &= ~_PAGE_RW;
273	return fPg;
274	# endif
275	}
276	return PAGE_READONLY_EXEC;
277	#else
278	return fKernel ? MY_PAGE_KERNEL_EXEC : PAGE_READONLY_EXEC;
279	#endif
280
281	case RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC:
282	case RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC \| RTMEM_PROT_READ:
283	return fKernel ? MY_PAGE_KERNEL_EXEC : PAGE_SHARED_EXEC;
284	}
285	}
286
287
288	/**
289	* Worker for rtR0MemObjNativeReserveUser and rtR0MemObjNativerMapUser that creates
290	* an empty user space mapping.
291	*
292	* We acquire the mmap_sem/mmap_lock of the task!
293	*
294	* @returns Pointer to the mapping.
295	* (void *)-1 on failure.
296	* @param R3PtrFixed (RTR3PTR)-1 if anywhere, otherwise a specific location.
297	* @param cb The size of the mapping.
298	* @param uAlignment The alignment of the mapping.
299	* @param pTask The Linux task to create this mapping in.
300	* @param fProt The RTMEM_PROT_* mask.
301	*/
302	static void rtR0MemObjLinuxDoMmap(RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment, struct task_struct pTask, unsigned fProt)
303	{
304	unsigned fLnxProt;
305	unsigned long ulAddr;
306
307	Assert(pTask == current); /* do_mmap */
308	RT_NOREF_PV(pTask);
309
310	/*
311	* Convert from IPRT protection to mman.h PROT_ and call do_mmap.
312	*/
313	fProt &= (RTMEM_PROT_NONE \| RTMEM_PROT_READ \| RTMEM_PROT_WRITE \| RTMEM_PROT_EXEC);
314	if (fProt == RTMEM_PROT_NONE)
315	fLnxProt = PROT_NONE;
316	else
317	{
318	fLnxProt = 0;
319	if (fProt & RTMEM_PROT_READ)
320	fLnxProt \|= PROT_READ;
321	if (fProt & RTMEM_PROT_WRITE)
322	fLnxProt \|= PROT_WRITE;
323	if (fProt & RTMEM_PROT_EXEC)
324	fLnxProt \|= PROT_EXEC;
325	}
326
327	if (R3PtrFixed != (RTR3PTR)-1)
328	{
329	#if RTLNX_VER_MIN(3,5,0)
330	ulAddr = vm_mmap(NULL, R3PtrFixed, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS \| MAP_FIXED, 0);
331	#else
332	LNX_MM_DOWN_WRITE(pTask->mm);
333	ulAddr = do_mmap(NULL, R3PtrFixed, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS \| MAP_FIXED, 0);
334	LNX_MM_UP_WRITE(pTask->mm);
335	#endif
336	}
337	else
338	{
339	#if RTLNX_VER_MIN(3,5,0)
340	ulAddr = vm_mmap(NULL, 0, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS, 0);
341	#else
342	LNX_MM_DOWN_WRITE(pTask->mm);
343	ulAddr = do_mmap(NULL, 0, cb, fLnxProt, MAP_SHARED \| MAP_ANONYMOUS, 0);
344	LNX_MM_UP_WRITE(pTask->mm);
345	#endif
346	if ( !(ulAddr & ~PAGE_MASK)
347	&& (ulAddr & (uAlignment - 1)))
348	{
349	/** @todo implement uAlignment properly... We'll probably need to make some dummy mappings to fill
350	* up alignment gaps. This is of course complicated by fragmentation (which we might have cause
351	* ourselves) and further by there begin two mmap strategies (top / bottom). */
352	/* For now, just ignore uAlignment requirements... */
353	}
354	}
355
356
357	if (ulAddr & ~PAGE_MASK) /* ~PAGE_MASK == PAGE_OFFSET_MASK */
358	return (void *)-1;
359	return (void *)ulAddr;
360	}
361
362
363	/**
364	* Worker that destroys a user space mapping.
365	* Undoes what rtR0MemObjLinuxDoMmap did.
366	*
367	* We acquire the mmap_sem/mmap_lock of the task!
368	*
369	* @param pv The ring-3 mapping.
370	* @param cb The size of the mapping.
371	* @param pTask The Linux task to destroy this mapping in.
372	*/
373	static void rtR0MemObjLinuxDoMunmap(void pv, size_t cb, struct task_struct pTask)
374	{
375	#if RTLNX_VER_MIN(3,5,0)
376	Assert(pTask == current); RT_NOREF_PV(pTask);
377	vm_munmap((unsigned long)pv, cb);
378	#elif defined(USE_RHEL4_MUNMAP)
379	LNX_MM_DOWN_WRITE(pTask->mm);
380	do_munmap(pTask->mm, (unsigned long)pv, cb, 0); /* should it be 1 or 0? */
381	LNX_MM_UP_WRITE(pTask->mm);
382	#else
383	LNX_MM_DOWN_WRITE(pTask->mm);
384	do_munmap(pTask->mm, (unsigned long)pv, cb);
385	LNX_MM_UP_WRITE(pTask->mm);
386	#endif
387	}
388
389
390	/**
391	* Internal worker that allocates physical pages and creates the memory object for them.
392	*
393	* @returns IPRT status code.
394	* @param ppMemLnx Where to store the memory object pointer.
395	* @param enmType The object type.
396	* @param cb The number of bytes to allocate.
397	* @param uAlignment The alignment of the physical memory.
398	* Only valid if fContiguous == true, ignored otherwise.
399	* @param fFlagsLnx The page allocation flags (GPFs).
400	* @param fContiguous Whether the allocation must be contiguous.
401	* @param fExecutable Whether the memory must be executable.
402	* @param rcNoMem What to return when we're out of pages.
403	* @param pszTag Allocation tag used for statistics and such.
404	*/
405	static int rtR0MemObjLinuxAllocPages(PRTR0MEMOBJLNX *ppMemLnx, RTR0MEMOBJTYPE enmType, size_t cb,
406	size_t uAlignment, gfp_t fFlagsLnx, bool fContiguous, bool fExecutable, int rcNoMem,
407	const char *pszTag)
408	{
409	size_t iPage;
410	size_t const cPages = cb >> PAGE_SHIFT;
411	struct page *paPages;
412
413	/*
414	* Allocate a memory object structure that's large enough to contain
415	* the page pointer array.
416	*/
417	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_UOFFSETOF_DYN(RTR0MEMOBJLNX, apPages[cPages]), enmType,
418	NULL, cb, pszTag);
419	if (!pMemLnx)
420	return VERR_NO_MEMORY;
421	pMemLnx->Core.fFlags \|= RTR0MEMOBJ_FLAGS_UNINITIALIZED_AT_ALLOC;
422	pMemLnx->cPages = cPages;
423
424	if (cPages > 255)
425	{
426	/* Try hard to allocate the memory, but the allocation attempt might fail. */
427	fFlagsLnx \|= MY_GFP_REPEAT;
428	# ifdef __GFP_NOMEMALLOC
429	/* Introduced with Linux 2.6.12: Don't use emergency reserves */
430	fFlagsLnx \|= __GFP_NOMEMALLOC;
431	# endif
432	}
433
434	/*
435	* Allocate the pages.
436	* For small allocations we'll try contiguous first and then fall back on page by page.
437	*/
438	#if RTLNX_VER_MIN(2,4,22)
439	if ( fContiguous
440	\|\| cb <= PAGE_SIZE * 2)
441	{
442	# ifdef VBOX_USE_INSERT_PAGE
443	paPages = alloc_pages(fFlagsLnx \| __GFP_COMP \| __GFP_NOWARN, rtR0MemObjLinuxOrder(cPages));
444	# else
445	paPages = alloc_pages(fFlagsLnx \| __GFP_NOWARN, rtR0MemObjLinuxOrder(cPages));
446	# endif
447	if (paPages)
448	{
449	fContiguous = true;
450	for (iPage = 0; iPage < cPages; iPage++)
451	pMemLnx->apPages[iPage] = &paPages[iPage];
452	}
453	else if (fContiguous)
454	{
455	rtR0MemObjDelete(&pMemLnx->Core);
456	return rcNoMem;
457	}
458	}
459
460	if (!fContiguous)
461	{
462	/** @todo Try use alloc_pages_bulk_array when available, it should be faster
463	* than a alloc_page loop. Put it in #ifdefs similar to
464	* IPRT_USE_APPLY_TO_PAGE_RANGE_FOR_EXEC. */
465	for (iPage = 0; iPage < cPages; iPage++)
466	{
467	pMemLnx->apPages[iPage] = alloc_page(fFlagsLnx \| __GFP_NOWARN);
468	if (RT_UNLIKELY(!pMemLnx->apPages[iPage]))
469	{
470	while (iPage-- > 0)
471	__free_page(pMemLnx->apPages[iPage]);
472	rtR0MemObjDelete(&pMemLnx->Core);
473	return rcNoMem;
474	}
475	}
476	}
477
478	#else /* < 2.4.22 */
479	/** @todo figure out why we didn't allocate page-by-page on 2.4.21 and older... */
480	paPages = alloc_pages(fFlagsLnx, rtR0MemObjLinuxOrder(cPages));
481	if (!paPages)
482	{
483	rtR0MemObjDelete(&pMemLnx->Core);
484	return rcNoMem;
485	}
486	for (iPage = 0; iPage < cPages; iPage++)
487	{
488	pMemLnx->apPages[iPage] = &paPages[iPage];
489	if (fExecutable)
490	MY_SET_PAGES_EXEC(pMemLnx->apPages[iPage], 1);
491	if (PageHighMem(pMemLnx->apPages[iPage]))
492	BUG();
493	}
494
495	fContiguous = true;
496	#endif /* < 2.4.22 */
497	pMemLnx->fContiguous = fContiguous;
498	pMemLnx->fExecutable = fExecutable;
499
500	#if RTLNX_VER_MAX(4,5,0)
501	/*
502	* Reserve the pages.
503	*
504	* Linux >= 4.5 with CONFIG_DEBUG_VM panics when setting PG_reserved on compound
505	* pages. According to Michal Hocko this shouldn't be necessary anyway because
506	* as pages which are not on the LRU list are never evictable.
507	*/
508	for (iPage = 0; iPage < cPages; iPage++)
509	SetPageReserved(pMemLnx->apPages[iPage]);
510	#endif
511
512	/*
513	* Note that the physical address of memory allocated with alloc_pages(flags, order)
514	* is always 2^(PAGE_SHIFT+order)-aligned.
515	*/
516	if ( fContiguous
517	&& uAlignment > PAGE_SIZE)
518	{
519	/*
520	* Check for alignment constraints. The physical address of memory allocated with
521	* alloc_pages(flags, order) is always 2^(PAGE_SHIFT+order)-aligned.
522	*/
523	if (RT_UNLIKELY(page_to_phys(pMemLnx->apPages[0]) & (uAlignment - 1)))
524	{
525	/*
526	* This should never happen!
527	*/
528	printk("rtR0MemObjLinuxAllocPages(cb=0x%lx, uAlignment=0x%lx): alloc_pages(..., %d) returned physical memory at 0x%lx!\n",
529	(unsigned long)cb, (unsigned long)uAlignment, rtR0MemObjLinuxOrder(cPages), (unsigned long)page_to_phys(pMemLnx->apPages[0]));
530	rtR0MemObjLinuxFreePages(pMemLnx);
531	return rcNoMem;
532	}
533	}
534
535	*ppMemLnx = pMemLnx;
536	return VINF_SUCCESS;
537	}
538
539
540	/**
541	* Frees the physical pages allocated by the rtR0MemObjLinuxAllocPages() call.
542	*
543	* This method does NOT free the object.
544	*
545	* @param pMemLnx The object which physical pages should be freed.
546	*/
547	static void rtR0MemObjLinuxFreePages(PRTR0MEMOBJLNX pMemLnx)
548	{
549	size_t iPage = pMemLnx->cPages;
550	if (iPage > 0)
551	{
552	/*
553	* Restore the page flags.
554	*/
555	while (iPage-- > 0)
556	{
557	#if RTLNX_VER_MAX(4,5,0)
558	/* See SetPageReserved() in rtR0MemObjLinuxAllocPages() */
559	ClearPageReserved(pMemLnx->apPages[iPage]);
560	#endif
561	#if RTLNX_VER_MAX(2,4,22)
562	if (pMemLnx->fExecutable)
563	MY_SET_PAGES_NOEXEC(pMemLnx->apPages[iPage], 1);
564	#endif
565	}
566
567	/*
568	* Free the pages.
569	*/
570	#if RTLNX_VER_MIN(2,4,22)
571	if (!pMemLnx->fContiguous)
572	{
573	iPage = pMemLnx->cPages;
574	while (iPage-- > 0)
575	__free_page(pMemLnx->apPages[iPage]);
576	}
577	else
578	#endif
579	__free_pages(pMemLnx->apPages[0], rtR0MemObjLinuxOrder(pMemLnx->cPages));
580
581	pMemLnx->cPages = 0;
582	}
583	}
584
585
586	#ifdef IPRT_USE_APPLY_TO_PAGE_RANGE_FOR_EXEC
587	/**
588	* User data passed to the apply_to_page_range() callback.
589	*/
590	typedef struct LNXAPPLYPGRANGE
591	{
592	/** Pointer to the memory object. */
593	PRTR0MEMOBJLNX pMemLnx;
594	/** The page protection flags to apply. */
595	pgprot_t fPg;
596	} LNXAPPLYPGRANGE;
597	/** Pointer to the user data. */
598	typedef LNXAPPLYPGRANGE *PLNXAPPLYPGRANGE;
599	/** Pointer to the const user data. */
600	typedef const LNXAPPLYPGRANGE *PCLNXAPPLYPGRANGE;
601
602	/**
603	* Callback called in apply_to_page_range().
604	*
605	* @returns Linux status code.
606	* @param pPte Pointer to the page table entry for the given address.
607	* @param uAddr The address to apply the new protection to.
608	* @param pvUser The opaque user data.
609	*/
610	static int rtR0MemObjLinuxApplyPageRange(pte_t pPte, unsigned long uAddr, void pvUser)
611	{
612	PCLNXAPPLYPGRANGE pArgs = (PCLNXAPPLYPGRANGE)pvUser;
613	PRTR0MEMOBJLNX pMemLnx = pArgs->pMemLnx;
614	size_t idxPg = (uAddr - (unsigned long)pMemLnx->Core.pv) >> PAGE_SHIFT;
615
616	set_pte(pPte, mk_pte(pMemLnx->apPages[idxPg], pArgs->fPg));
617	return 0;
618	}
619	#endif
620
621
622	/**
623	* Maps the allocation into ring-0.
624	*
625	* This will update the RTR0MEMOBJLNX::Core.pv and RTR0MEMOBJ::fMappedToRing0 members.
626	*
627	* Contiguous mappings that isn't in 'high' memory will already be mapped into kernel
628	* space, so we'll use that mapping if possible. If execute access is required, we'll
629	* play safe and do our own mapping.
630	*
631	* @returns IPRT status code.
632	* @param pMemLnx The linux memory object to map.
633	* @param fExecutable Whether execute access is required.
634	*/
635	static int rtR0MemObjLinuxVMap(PRTR0MEMOBJLNX pMemLnx, bool fExecutable)
636	{
637	int rc = VINF_SUCCESS;
638
639	/*
640	* Choose mapping strategy.
641	*/
642	bool fMustMap = fExecutable
643	\|\| !pMemLnx->fContiguous;
644	if (!fMustMap)
645	{
646	size_t iPage = pMemLnx->cPages;
647	while (iPage-- > 0)
648	if (PageHighMem(pMemLnx->apPages[iPage]))
649	{
650	fMustMap = true;
651	break;
652	}
653	}
654
655	Assert(!pMemLnx->Core.pv);
656	Assert(!pMemLnx->fMappedToRing0);
657
658	if (fMustMap)
659	{
660	/*
661	* Use vmap - 2.4.22 and later.
662	*/
663	#if RTLNX_VER_MIN(2,4,22) && (defined(RT_ARCH_AMD64) \|\| defined(RT_ARCH_X86) \|\| defined(RT_ARCH_ARM64))
664	pgprot_t fPg;
665	# if defined(RT_ARCH_ARM64)
666	/* ARM64 architecture has no _PAGE_NX, _PAGE_PRESENT and _PAGE_RW flags.
667	* Closest alternatives would be PTE_PXN, PTE_UXN, PROT_DEFAULT and PTE_WRITE. */
668	# if RTLNX_VER_MIN(6,5,0)
669	pgprot_val(fPg) = _PAGE_KERNEL; /* (PROT_DEFAULT \| PTE_PXN \| PTE_UXN \| PTE_WRITE \| PTE_ATTRINDX(MT_NORMAL). */
670	# else /* < 6.5.0 */
671	pgprot_val(fPg) = PROT_NORMAL; /* (PROT_DEFAULT \| PTE_PXN \| PTE_UXN \| PTE_WRITE \| PTE_ATTRINDX(MT_NORMAL). */
672	# endif /* 6.5.0 */
673	# else /* !RT_ARCH_ARM64 */
674	pgprot_val(fPg) = _PAGE_PRESENT \| _PAGE_RW;
675	# ifdef _PAGE_NX
676	if (!fExecutable)
677	pgprot_val(fPg) \|= _PAGE_NX;
678	# endif
679	# endif /* RT_ARCH_ARM64 */
680
681	# ifdef IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
682	if (fExecutable)
683	{
684	# if RTLNX_VER_MIN(3,2,51)
685	pte_t papPtes = (pte_t )kmalloc_array(pMemLnx->cPages, sizeof(papPtes[0]), GFP_KERNEL);
686	# else
687	pte_t papPtes = (pte_t )kmalloc(pMemLnx->cPages * sizeof(papPtes[0]), GFP_KERNEL);
688	# endif
689	if (papPtes)
690	{
691	pMemLnx->pArea = alloc_vm_area(pMemLnx->Core.cb, papPtes); /* Note! pArea->nr_pages is not set. */
692	if (pMemLnx->pArea)
693	{
694	size_t i;
695	Assert(pMemLnx->pArea->size >= pMemLnx->Core.cb); /* Note! includes guard page. */
696	Assert(pMemLnx->pArea->addr);
697	# if !defined(RT_ARCH_ARM64) && defined(_PAGE_NX)
698	pgprot_val(fPg) \|= _PAGE_NX; /* Uses RTR0MemObjProtect to clear NX when memory ready, W^X fashion. */
699	# endif
700	pMemLnx->papPtesForArea = papPtes;
701	for (i = 0; i < pMemLnx->cPages; i++)
702	*papPtes[i] = mk_pte(pMemLnx->apPages[i], fPg);
703	pMemLnx->Core.pv = pMemLnx->pArea->addr;
704	pMemLnx->fMappedToRing0 = true;
705	}
706	else
707	{
708	kfree(papPtes);
709	rc = VERR_MAP_FAILED;
710	}
711	}
712	else
713	rc = VERR_MAP_FAILED;
714	}
715	else
716	# endif
717	{
718	# if !defined(RT_ARCH_ARM64) && defined(IPRT_USE_APPLY_TO_PAGE_RANGE_FOR_EXEC)
719	if (fExecutable)
720	pgprot_val(fPg) \|= _PAGE_NX; /* Uses RTR0MemObjProtect to clear NX when memory ready, W^X fashion. */
721	# endif
722
723	# ifdef VM_MAP
724	pMemLnx->Core.pv = vmap(&pMemLnx->apPages[0], pMemLnx->cPages, VM_MAP, fPg);
725	# else
726	pMemLnx->Core.pv = vmap(&pMemLnx->apPages[0], pMemLnx->cPages, VM_ALLOC, fPg);
727	# endif
728	if (pMemLnx->Core.pv)
729	pMemLnx->fMappedToRing0 = true;
730	else
731	rc = VERR_MAP_FAILED;
732	}
733	#else /* < 2.4.22 */
734	rc = VERR_NOT_SUPPORTED;
735	#endif
736	}
737	else
738	{
739	/*
740	* Use the kernel RAM mapping.
741	*/
742	pMemLnx->Core.pv = phys_to_virt(page_to_phys(pMemLnx->apPages[0]));
743	Assert(pMemLnx->Core.pv);
744	}
745
746	return rc;
747	}
748
749
750	/**
751	* Undoes what rtR0MemObjLinuxVMap() did.
752	*
753	* @param pMemLnx The linux memory object.
754	*/
755	static void rtR0MemObjLinuxVUnmap(PRTR0MEMOBJLNX pMemLnx)
756	{
757	#if RTLNX_VER_MIN(2,4,22)
758	# ifdef IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
759	if (pMemLnx->pArea)
760	{
761	# if 0
762	pte_t **papPtes = pMemLnx->papPtesForArea;
763	size_t i;
764	for (i = 0; i < pMemLnx->cPages; i++)
765	*papPtes[i] = 0;
766	# endif
767	free_vm_area(pMemLnx->pArea);
768	kfree(pMemLnx->papPtesForArea);
769	pMemLnx->pArea = NULL;
770	pMemLnx->papPtesForArea = NULL;
771	}
772	else
773	# endif
774	if (pMemLnx->fMappedToRing0)
775	{
776	Assert(pMemLnx->Core.pv);
777	vunmap(pMemLnx->Core.pv);
778	pMemLnx->fMappedToRing0 = false;
779	}
780	#else /* < 2.4.22 */
781	Assert(!pMemLnx->fMappedToRing0);
782	#endif
783	pMemLnx->Core.pv = NULL;
784	}
785
786
787	DECLHIDDEN(int) rtR0MemObjNativeFree(RTR0MEMOBJ pMem)
788	{
789	IPRT_LINUX_SAVE_EFL_AC();
790	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
791
792	/*
793	* Release any memory that we've allocated or locked.
794	*/
795	switch (pMemLnx->Core.enmType)
796	{
797	case RTR0MEMOBJTYPE_PAGE:
798	case RTR0MEMOBJTYPE_LOW:
799	case RTR0MEMOBJTYPE_CONT:
800	case RTR0MEMOBJTYPE_PHYS:
801	case RTR0MEMOBJTYPE_PHYS_NC:
802	rtR0MemObjLinuxVUnmap(pMemLnx);
803	rtR0MemObjLinuxFreePages(pMemLnx);
804	break;
805
806	case RTR0MEMOBJTYPE_LARGE_PAGE:
807	{
808	uint32_t const cLargePages = pMemLnx->Core.cb >> (pMemLnx->cLargePageOrder + PAGE_SHIFT);
809	uint32_t iLargePage;
810	for (iLargePage = 0; iLargePage < cLargePages; iLargePage++)
811	__free_pages(pMemLnx->apPages[iLargePage << pMemLnx->cLargePageOrder], pMemLnx->cLargePageOrder);
812	pMemLnx->cPages = 0;
813
814	#ifdef IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
815	Assert(!pMemLnx->pArea);
816	Assert(!pMemLnx->papPtesForArea);
817	#endif
818	break;
819	}
820
821	case RTR0MEMOBJTYPE_LOCK:
822	if (pMemLnx->Core.u.Lock.R0Process != NIL_RTR0PROCESS)
823	{
824	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
825	size_t iPage;
826	Assert(pTask);
827	if (pTask && pTask->mm)
828	LNX_MM_DOWN_READ(pTask->mm);
829
830	iPage = pMemLnx->cPages;
831	while (iPage-- > 0)
832	{
833	if (!PageReserved(pMemLnx->apPages[iPage]))
834	SetPageDirty(pMemLnx->apPages[iPage]);
835	#if RTLNX_VER_MIN(4,6,0)
836	put_page(pMemLnx->apPages[iPage]);
837	#else
838	page_cache_release(pMemLnx->apPages[iPage]);
839	#endif
840	}
841
842	if (pTask && pTask->mm)
843	LNX_MM_UP_READ(pTask->mm);
844	}
845	/* else: kernel memory - nothing to do here. */
846	break;
847
848	case RTR0MEMOBJTYPE_RES_VIRT:
849	Assert(pMemLnx->Core.pv);
850	if (pMemLnx->Core.u.ResVirt.R0Process != NIL_RTR0PROCESS)
851	{
852	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
853	Assert(pTask);
854	if (pTask && pTask->mm)
855	rtR0MemObjLinuxDoMunmap(pMemLnx->Core.pv, pMemLnx->Core.cb, pTask);
856	}
857	else
858	{
859	vunmap(pMemLnx->Core.pv);
860
861	Assert(pMemLnx->cPages == 1 && pMemLnx->apPages[0] != NULL);
862	__free_page(pMemLnx->apPages[0]);
863	pMemLnx->apPages[0] = NULL;
864	pMemLnx->cPages = 0;
865	}
866	pMemLnx->Core.pv = NULL;
867	break;
868
869	case RTR0MEMOBJTYPE_MAPPING:
870	Assert(pMemLnx->cPages == 0); Assert(pMemLnx->Core.pv);
871	if (pMemLnx->Core.u.ResVirt.R0Process != NIL_RTR0PROCESS)
872	{
873	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
874	Assert(pTask);
875	if (pTask && pTask->mm)
876	rtR0MemObjLinuxDoMunmap(pMemLnx->Core.pv, pMemLnx->Core.cb, pTask);
877	}
878	else
879	vunmap(pMemLnx->Core.pv);
880	pMemLnx->Core.pv = NULL;
881	break;
882
883	default:
884	AssertMsgFailed(("enmType=%d\n", pMemLnx->Core.enmType));
885	return VERR_INTERNAL_ERROR;
886	}
887	IPRT_LINUX_RESTORE_EFL_ONLY_AC();
888	return VINF_SUCCESS;
889	}
890
891
892	DECLHIDDEN(int) rtR0MemObjNativeAllocPage(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable, const char *pszTag)
893	{
894	IPRT_LINUX_SAVE_EFL_AC();
895	PRTR0MEMOBJLNX pMemLnx;
896	int rc;
897
898	#if RTLNX_VER_MIN(2,4,22)
899	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_PAGE, cb, PAGE_SIZE, GFP_HIGHUSER,
900	false /* non-contiguous */, fExecutable, VERR_NO_MEMORY, pszTag);
901	#else
902	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_PAGE, cb, PAGE_SIZE, GFP_USER,
903	false /* non-contiguous */, fExecutable, VERR_NO_MEMORY, pszTag);
904	#endif
905	if (RT_SUCCESS(rc))
906	{
907	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
908	if (RT_SUCCESS(rc))
909	{
910	*ppMem = &pMemLnx->Core;
911	IPRT_LINUX_RESTORE_EFL_AC();
912	return rc;
913	}
914
915	rtR0MemObjLinuxFreePages(pMemLnx);
916	rtR0MemObjDelete(&pMemLnx->Core);
917	}
918
919	IPRT_LINUX_RESTORE_EFL_AC();
920	return rc;
921	}
922
923
924	DECLHIDDEN(int) rtR0MemObjNativeAllocLarge(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, size_t cbLargePage, uint32_t fFlags,
925	const char *pszTag)
926	{
927	#ifdef GFP_TRANSHUGE
928	/*
929	* Allocate a memory object structure that's large enough to contain
930	* the page pointer array.
931	*/
932	# ifdef __GFP_MOVABLE
933	unsigned const fGfp = (GFP_TRANSHUGE \| __GFP_ZERO) & ~__GFP_MOVABLE;
934	# else
935	unsigned const fGfp = (GFP_TRANSHUGE \| __GFP_ZERO);
936	# endif
937	size_t const cPagesPerLarge = cbLargePage >> PAGE_SHIFT;
938	unsigned const cLargePageOrder = rtR0MemObjLinuxOrder(cPagesPerLarge);
939	size_t const cLargePages = cb >> (cLargePageOrder + PAGE_SHIFT);
940	size_t const cPages = cb >> PAGE_SHIFT;
941	PRTR0MEMOBJLNX pMemLnx;
942
943	Assert(RT_BIT_64(cLargePageOrder + PAGE_SHIFT) == cbLargePage);
944	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_UOFFSETOF_DYN(RTR0MEMOBJLNX, apPages[cPages]),
945	RTR0MEMOBJTYPE_LARGE_PAGE, NULL, cb, pszTag);
946	if (pMemLnx)
947	{
948	size_t iLargePage;
949
950	pMemLnx->Core.fFlags \|= RTR0MEMOBJ_FLAGS_ZERO_AT_ALLOC;
951	pMemLnx->cLargePageOrder = cLargePageOrder;
952	pMemLnx->cPages = cPages;
953
954	/*
955	* Allocate the requested number of large pages.
956	*/
957	for (iLargePage = 0; iLargePage < cLargePages; iLargePage++)
958	{
959	struct page *paPages = alloc_pages(fGfp, cLargePageOrder);
960	if (paPages)
961	{
962	size_t const iPageBase = iLargePage << cLargePageOrder;
963	size_t iPage = cPagesPerLarge;
964	while (iPage-- > 0)
965	pMemLnx->apPages[iPageBase + iPage] = &paPages[iPage];
966	}
967	else
968	{
969	/*Log(("rtR0MemObjNativeAllocLarge: cb=%#zx cPages=%#zx cLargePages=%#zx cLargePageOrder=%u cPagesPerLarge=%#zx iLargePage=%#zx -> failed!\n",
970	cb, cPages, cLargePages, cLargePageOrder, cPagesPerLarge, iLargePage, paPages));*/
971	while (iLargePage-- > 0)
972	__free_pages(pMemLnx->apPages[iLargePage << (cLargePageOrder - PAGE_SHIFT)], cLargePageOrder);
973	rtR0MemObjDelete(&pMemLnx->Core);
974	return VERR_NO_MEMORY;
975	}
976	}
977	*ppMem = &pMemLnx->Core;
978	return VINF_SUCCESS;
979	}
980	return VERR_NO_MEMORY;
981
982	#else
983	/*
984	* We don't call rtR0MemObjFallbackAllocLarge here as it can be a really
985	* bad idea to trigger the swap daemon and whatnot. So, just fail.
986	*/
987	RT_NOREF(ppMem, cb, cbLargePage, fFlags, pszTag);
988	return VERR_NOT_SUPPORTED;
989	#endif
990	}
991
992
993	DECLHIDDEN(int) rtR0MemObjNativeAllocLow(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable, const char *pszTag)
994	{
995	IPRT_LINUX_SAVE_EFL_AC();
996	PRTR0MEMOBJLNX pMemLnx;
997	int rc;
998
999	/* Try to avoid GFP_DMA. GFM_DMA32 was introduced with Linux 2.6.15. */
1000	#if (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
1001	/* ZONE_DMA32: 0-4GB */
1002	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_KERNEL \| GFP_DMA32,
1003	false /* non-contiguous */, fExecutable, VERR_NO_LOW_MEMORY, pszTag);
1004	if (RT_FAILURE(rc))
1005	#endif
1006	#ifdef RT_ARCH_AMD64
1007	/* ZONE_DMA: 0-16MB */
1008	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_KERNEL \| GFP_DMA,
1009	false /* non-contiguous */, fExecutable, VERR_NO_LOW_MEMORY, pszTag);
1010	#else
1011	# ifdef CONFIG_X86_PAE
1012	# endif
1013	/* ZONE_NORMAL: 0-896MB */
1014	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_USER,
1015	false /* non-contiguous */, fExecutable, VERR_NO_LOW_MEMORY, pszTag);
1016	#endif
1017	if (RT_SUCCESS(rc))
1018	{
1019	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
1020	if (RT_SUCCESS(rc))
1021	{
1022	*ppMem = &pMemLnx->Core;
1023	IPRT_LINUX_RESTORE_EFL_AC();
1024	return rc;
1025	}
1026
1027	rtR0MemObjLinuxFreePages(pMemLnx);
1028	rtR0MemObjDelete(&pMemLnx->Core);
1029	}
1030
1031	IPRT_LINUX_RESTORE_EFL_AC();
1032	return rc;
1033	}
1034
1035
1036	DECLHIDDEN(int) rtR0MemObjNativeAllocCont(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest,
1037	bool fExecutable, const char *pszTag)
1038	{
1039	IPRT_LINUX_SAVE_EFL_AC();
1040	PRTR0MEMOBJLNX pMemLnx;
1041	int rc;
1042	uint32_t idxZone;
1043
1044	/*
1045	* The last zone must be able to satisfy the PhysHighest requirement or there
1046	* will be no zone at all.
1047	*/
1048	if (g_aZones[RT_ELEMENTS(g_aZones) - 1].PhysHighest > PhysHighest)
1049	{
1050	IPRT_LINUX_RESTORE_EFL_AC();
1051	AssertMsgFailedReturn(("No zone can satisfy PhysHighest=%RHp!\n", PhysHighest),
1052	VERR_NO_CONT_MEMORY);
1053	}
1054
1055	/* Find the first zone matching our PhysHighest requirement. */
1056	idxZone = 0;
1057	for (;;)
1058	{
1059	if (g_aZones[idxZone].PhysHighest <= PhysHighest)
1060	break; /* We found a zone satisfying the requirement. */
1061	idxZone++;
1062	}
1063
1064	/* Now try to allocate pages from all the left zones until one succeeds. */
1065	for (;;)
1066	{
1067	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, PAGE_SIZE, g_aZones[idxZone].fGfp,
1068	true /* contiguous */, fExecutable, VERR_NO_CONT_MEMORY, pszTag);
1069	idxZone++;
1070	if (RT_SUCCESS(rc) \|\| idxZone == RT_ELEMENTS(g_aZones))
1071	break;
1072	}
1073	if (RT_SUCCESS(rc))
1074	{
1075	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
1076	if (RT_SUCCESS(rc))
1077	{
1078	#if defined(RT_STRICT)
1079	size_t iPage = pMemLnx->cPages;
1080	while (iPage-- > 0)
1081	Assert(page_to_phys(pMemLnx->apPages[iPage]) < PhysHighest);
1082	#endif
1083	pMemLnx->Core.u.Cont.Phys = page_to_phys(pMemLnx->apPages[0]);
1084	*ppMem = &pMemLnx->Core;
1085	IPRT_LINUX_RESTORE_EFL_AC();
1086	return rc;
1087	}
1088
1089	rtR0MemObjLinuxFreePages(pMemLnx);
1090	rtR0MemObjDelete(&pMemLnx->Core);
1091	}
1092
1093	IPRT_LINUX_RESTORE_EFL_AC();
1094	return rc;
1095	}
1096
1097
1098	/**
1099	* Worker for rtR0MemObjLinuxAllocPhysSub that tries one allocation strategy.
1100	*
1101	* @returns IPRT status code.
1102	* @param ppMemLnx Where to
1103	* @param enmType The object type.
1104	* @param cb The size of the allocation.
1105	* @param uAlignment The alignment of the physical memory.
1106	* Only valid for fContiguous == true, ignored otherwise.
1107	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
1108	* @param pszTag Allocation tag used for statistics and such.
1109	* @param fGfp The Linux GFP flags to use for the allocation.
1110	*/
1111	static int rtR0MemObjLinuxAllocPhysSub2(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType,
1112	size_t cb, size_t uAlignment, RTHCPHYS PhysHighest, const char *pszTag, gfp_t fGfp)
1113	{
1114	PRTR0MEMOBJLNX pMemLnx;
1115	int rc = rtR0MemObjLinuxAllocPages(&pMemLnx, enmType, cb, uAlignment, fGfp,
1116	enmType == RTR0MEMOBJTYPE_PHYS /* contiguous / non-contiguous */,
1117	false /fExecutable/, VERR_NO_PHYS_MEMORY, pszTag);
1118	if (RT_FAILURE(rc))
1119	return rc;
1120
1121	/*
1122	* Check the addresses if necessary. (Can be optimized a bit for PHYS.)
1123	*/
1124	if (PhysHighest != NIL_RTHCPHYS)
1125	{
1126	size_t iPage = pMemLnx->cPages;
1127	while (iPage-- > 0)
1128	if (page_to_phys(pMemLnx->apPages[iPage]) > PhysHighest)
1129	{
1130	rtR0MemObjLinuxFreePages(pMemLnx);
1131	rtR0MemObjDelete(&pMemLnx->Core);
1132	return VERR_NO_MEMORY;
1133	}
1134	}
1135
1136	/*
1137	* Complete the object.
1138	*/
1139	if (enmType == RTR0MEMOBJTYPE_PHYS)
1140	{
1141	pMemLnx->Core.u.Phys.PhysBase = page_to_phys(pMemLnx->apPages[0]);
1142	pMemLnx->Core.u.Phys.fAllocated = true;
1143	}
1144	*ppMem = &pMemLnx->Core;
1145	return rc;
1146	}
1147
1148
1149	/**
1150	* Worker for rtR0MemObjNativeAllocPhys and rtR0MemObjNativeAllocPhysNC.
1151	*
1152	* @returns IPRT status code.
1153	* @param ppMem Where to store the memory object pointer on success.
1154	* @param enmType The object type.
1155	* @param cb The size of the allocation.
1156	* @param uAlignment The alignment of the physical memory.
1157	* Only valid for enmType == RTR0MEMOBJTYPE_PHYS, ignored otherwise.
1158	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
1159	* @param pszTag Allocation tag used for statistics and such.
1160	*/
1161	static int rtR0MemObjLinuxAllocPhysSub(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType,
1162	size_t cb, size_t uAlignment, RTHCPHYS PhysHighest, const char *pszTag)
1163	{
1164	int rc;
1165	IPRT_LINUX_SAVE_EFL_AC();
1166
1167	/*
1168	* There are two clear cases and that's the <=16MB and anything-goes ones.
1169	* When the physical address limit is somewhere in-between those two we'll
1170	* just have to try, starting with HIGHUSER and working our way thru the
1171	* different types, hoping we'll get lucky.
1172	*
1173	* We should probably move this physical address restriction logic up to
1174	* the page alloc function as it would be more efficient there. But since
1175	* we don't expect this to be a performance issue just yet it can wait.
1176	*/
1177	if (PhysHighest == NIL_RTHCPHYS)
1178	/* ZONE_HIGHMEM: the whole physical memory */
1179	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_HIGHUSER);
1180	else if (PhysHighest <= _1M * 16)
1181	/* ZONE_DMA: 0-16MB */
1182	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_DMA);
1183	else
1184	{
1185	rc = VERR_NO_MEMORY;
1186	if (RT_FAILURE(rc))
1187	/* ZONE_HIGHMEM: the whole physical memory */
1188	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_HIGHUSER);
1189	if (RT_FAILURE(rc))
1190	/* ZONE_NORMAL: 0-896MB */
1191	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_USER);
1192	#ifdef GFP_DMA32
1193	if (RT_FAILURE(rc))
1194	/* ZONE_DMA32: 0-4GB */
1195	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_DMA32);
1196	#endif
1197	if (RT_FAILURE(rc))
1198	/* ZONE_DMA: 0-16MB */
1199	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_DMA);
1200	}
1201	IPRT_LINUX_RESTORE_EFL_AC();
1202	return rc;
1203	}
1204
1205
1206	/**
1207	* Translates a kernel virtual address to a linux page structure by walking the
1208	* page tables.
1209	*
1210	* @note We do assume that the page tables will not change as we are walking
1211	* them. This assumption is rather forced by the fact that I could not
1212	* immediately see any way of preventing this from happening. So, we
1213	* take some extra care when accessing them.
1214	*
1215	* Because of this, we don't want to use this function on memory where
1216	* attribute changes to nearby pages is likely to cause large pages to
1217	* be used or split up. So, don't use this for the linear mapping of
1218	* physical memory.
1219	*
1220	* @returns Pointer to the page structur or NULL if it could not be found.
1221	* @param pv The kernel virtual address.
1222	*/
1223	RTDECL(struct page ) rtR0MemObjLinuxVirtToPage(void pv)
1224	{
1225	#if defined(RT_ARCH_AMD64) \|\| defined(RT_ARCH_X86)
1226	unsigned long ulAddr = (unsigned long)pv;
1227	unsigned long pfn;
1228	struct page *pPage;
1229	pte_t *pEntry;
1230	union
1231	{
1232	pgd_t Global;
1233	# if RTLNX_VER_MIN(4,12,0)
1234	p4d_t Four;
1235	# endif
1236	# if RTLNX_VER_MIN(2,6,11)
1237	pud_t Upper;
1238	# endif
1239	pmd_t Middle;
1240	pte_t Entry;
1241	} u;
1242
1243	/* Should this happen in a situation this code will be called in? And if
1244	* so, can it change under our feet? See also
1245	* "Documentation/vm/active_mm.txt" in the kernel sources. */
1246	if (RT_UNLIKELY(!current->active_mm))
1247	return NULL;
1248	u.Global = *pgd_offset(current->active_mm, ulAddr);
1249	if (RT_UNLIKELY(pgd_none(u.Global)))
1250	return NULL;
1251	# if RTLNX_VER_MIN(2,6,11)
1252	# if RTLNX_VER_MIN(4,12,0)
1253	u.Four = *p4d_offset(&u.Global, ulAddr);
1254	if (RT_UNLIKELY(p4d_none(u.Four)))
1255	return NULL;
1256	# if RTLNX_VER_MIN(5,6,0)
1257	if (p4d_leaf(u.Four))
1258	# else
1259	if (p4d_large(u.Four))
1260	# endif
1261	{
1262	pPage = p4d_page(u.Four);
1263	AssertReturn(pPage, NULL);
1264	pfn = page_to_pfn(pPage); /* doing the safe way... */
1265	AssertCompile(P4D_SHIFT - PAGE_SHIFT < 31);
1266	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (P4D_SHIFT - PAGE_SHIFT)) - 1);
1267	return pfn_to_page(pfn);
1268	}
1269	u.Upper = *pud_offset(&u.Four, ulAddr);
1270	# else /* < 4.12 */
1271	u.Upper = *pud_offset(&u.Global, ulAddr);
1272	# endif /* < 4.12 */
1273	if (RT_UNLIKELY(pud_none(u.Upper)))
1274	return NULL;
1275	# if RTLNX_VER_MIN(2,6,25)
1276	# if RTLNX_VER_MIN(5,6,0)
1277	if (pud_leaf(u.Upper))
1278	# else
1279	if (pud_large(u.Upper))
1280	# endif
1281	{
1282	pPage = pud_page(u.Upper);
1283	AssertReturn(pPage, NULL);
1284	pfn = page_to_pfn(pPage); /* doing the safe way... */
1285	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (PUD_SHIFT - PAGE_SHIFT)) - 1);
1286	return pfn_to_page(pfn);
1287	}
1288	# endif
1289	u.Middle = *pmd_offset(&u.Upper, ulAddr);
1290	# else /* < 2.6.11 */
1291	u.Middle = *pmd_offset(&u.Global, ulAddr);
1292	# endif /* < 2.6.11 */
1293	if (RT_UNLIKELY(pmd_none(u.Middle)))
1294	return NULL;
1295	# if RTLNX_VER_MIN(2,6,0)
1296	# if RTLNX_VER_MIN(5,6,0)
1297	if (pmd_leaf(u.Middle))
1298	# else
1299	if (pmd_large(u.Middle))
1300	# endif
1301	{
1302	pPage = pmd_page(u.Middle);
1303	AssertReturn(pPage, NULL);
1304	pfn = page_to_pfn(pPage); /* doing the safe way... */
1305	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (PMD_SHIFT - PAGE_SHIFT)) - 1);
1306	return pfn_to_page(pfn);
1307	}
1308	# endif
1309
1310	# if RTLNX_VER_MIN(6,5,0) \|\| RTLNX_RHEL_RANGE(9,4, 9,99)
1311	pEntry = __pte_map(&u.Middle, ulAddr);
1312	# elif RTLNX_VER_MIN(2,5,5) \|\| defined(pte_offset_map) /* As usual, RHEL 3 had pte_offset_map earlier. */
1313	pEntry = pte_offset_map(&u.Middle, ulAddr);
1314	# else
1315	pEntry = pte_offset(&u.Middle, ulAddr);
1316	# endif
1317	if (RT_UNLIKELY(!pEntry))
1318	return NULL;
1319	u.Entry = *pEntry;
1320	# if RTLNX_VER_MIN(2,5,5) \|\| defined(pte_offset_map)
1321	pte_unmap(pEntry);
1322	# endif
1323
1324	if (RT_UNLIKELY(!pte_present(u.Entry)))
1325	return NULL;
1326	return pte_page(u.Entry);
1327	#else /* !defined(RT_ARCH_AMD64) && !defined(RT_ARCH_X86) */
1328
1329	if (is_vmalloc_addr(pv))
1330	return vmalloc_to_page(pv);
1331
1332	return virt_to_page(pv);
1333	#endif
1334	}
1335	RT_EXPORT_SYMBOL(rtR0MemObjLinuxVirtToPage);
1336
1337
1338	DECLHIDDEN(int) rtR0MemObjNativeAllocPhys(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest, size_t uAlignment,
1339	const char *pszTag)
1340	{
1341	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS, cb, uAlignment, PhysHighest, pszTag);
1342	}
1343
1344
1345	DECLHIDDEN(int) rtR0MemObjNativeAllocPhysNC(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest, const char *pszTag)
1346	{
1347	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS_NC, cb, PAGE_SIZE, PhysHighest, pszTag);
1348	}
1349
1350
1351	DECLHIDDEN(int) rtR0MemObjNativeEnterPhys(PPRTR0MEMOBJINTERNAL ppMem, RTHCPHYS Phys, size_t cb, uint32_t uCachePolicy,
1352	const char *pszTag)
1353	{
1354	IPRT_LINUX_SAVE_EFL_AC();
1355
1356	/*
1357	* All we need to do here is to validate that we can use
1358	* ioremap on the specified address (32/64-bit dma_addr_t).
1359	*/
1360	PRTR0MEMOBJLNX pMemLnx;
1361	dma_addr_t PhysAddr = Phys;
1362	AssertMsgReturn(PhysAddr == Phys, ("%#llx\n", (unsigned long long)Phys), VERR_ADDRESS_TOO_BIG);
1363
1364	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_PHYS, NULL, cb, pszTag);
1365	if (!pMemLnx)
1366	{
1367	IPRT_LINUX_RESTORE_EFL_AC();
1368	return VERR_NO_MEMORY;
1369	}
1370
1371	pMemLnx->Core.u.Phys.PhysBase = PhysAddr;
1372	pMemLnx->Core.u.Phys.fAllocated = false;
1373	pMemLnx->Core.u.Phys.uCachePolicy = uCachePolicy;
1374	Assert(!pMemLnx->cPages);
1375	*ppMem = &pMemLnx->Core;
1376	IPRT_LINUX_RESTORE_EFL_AC();
1377	return VINF_SUCCESS;
1378	}
1379
1380	/* openSUSE Leap 42.3 detection :-/ */
1381	#if RTLNX_VER_RANGE(4,4,0, 4,6,0) && defined(FAULT_FLAG_REMOTE)
1382	# define GET_USER_PAGES_API KERNEL_VERSION(4, 10, 0) /* no typo! */
1383	#else
1384	# define GET_USER_PAGES_API LINUX_VERSION_CODE
1385	#endif
1386
1387	DECLHIDDEN(int) rtR0MemObjNativeLockUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3Ptr, size_t cb, uint32_t fAccess,
1388	RTR0PROCESS R0Process, const char *pszTag)
1389	{
1390	IPRT_LINUX_SAVE_EFL_AC();
1391	const int cPages = cb >> PAGE_SHIFT;
1392	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1393	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1394	struct vm_area_struct **papVMAs;
1395	# endif
1396	PRTR0MEMOBJLNX pMemLnx;
1397	int rc = VERR_NO_MEMORY;
1398	int const fWrite = fAccess & RTMEM_PROT_WRITE ? 1 : 0;
1399
1400	/*
1401	* Check for valid task and size overflows.
1402	*/
1403	if (!pTask)
1404	return VERR_NOT_SUPPORTED;
1405	if (((size_t)cPages << PAGE_SHIFT) != cb)
1406	return VERR_OUT_OF_RANGE;
1407
1408	/*
1409	* Allocate the memory object and a temporary buffer for the VMAs.
1410	*/
1411	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_UOFFSETOF_DYN(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK,
1412	(void *)R3Ptr, cb, pszTag);
1413	if (!pMemLnx)
1414	{
1415	IPRT_LINUX_RESTORE_EFL_AC();
1416	return VERR_NO_MEMORY;
1417	}
1418
1419	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1420	papVMAs = (struct vm_area_struct *)RTMemAlloc(sizeof(papVMAs) * cPages);
1421	if (papVMAs)
1422	{
1423	# endif
1424	LNX_MM_DOWN_READ(pTask->mm);
1425
1426	/*
1427	* Get user pages.
1428	*/
1429	/** @todo r=bird: Should we not force read access too? */
1430	#if GET_USER_PAGES_API >= KERNEL_VERSION(4, 6, 0)
1431	if (R0Process == RTR0ProcHandleSelf())
1432	rc = get_user_pages(R3Ptr, /* Where from. */
1433	cPages, /* How many pages. */
1434	# if GET_USER_PAGES_API >= KERNEL_VERSION(4, 9, 0)
1435	fWrite ? FOLL_WRITE \| /* Write to memory. */
1436	FOLL_FORCE /* force write access. */
1437	: 0, /* Write to memory. */
1438	# else
1439	fWrite, /* Write to memory. */
1440	fWrite, /* force write access. */
1441	# endif
1442	&pMemLnx->apPages[0] /* Page array. */
1443	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_SUSE_MAJ_PREREQ(15, 6) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1444	, papVMAs /* vmas */
1445	# endif
1446	);
1447	/*
1448	* Actually this should not happen at the moment as call this function
1449	* only for our own process.
1450	*/
1451	else
1452	rc = get_user_pages_remote(
1453	# if GET_USER_PAGES_API < KERNEL_VERSION(5, 9, 0)
1454	pTask, /* Task for fault accounting. */
1455	# endif
1456	pTask->mm, /* Whose pages. */
1457	R3Ptr, /* Where from. */
1458	cPages, /* How many pages. */
1459	# if GET_USER_PAGES_API >= KERNEL_VERSION(4, 9, 0)
1460	fWrite ? FOLL_WRITE \| /* Write to memory. */
1461	FOLL_FORCE /* force write access. */
1462	: 0, /* Write to memory. */
1463	# else
1464	fWrite, /* Write to memory. */
1465	fWrite, /* force write access. */
1466	# endif
1467	&pMemLnx->apPages[0] /* Page array. */
1468	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1469	, papVMAs /* vmas */
1470	# endif
1471	# if GET_USER_PAGES_API >= KERNEL_VERSION(4, 10, 0)
1472	, NULL /* locked */
1473	# endif
1474	);
1475	#else /* GET_USER_PAGES_API < KERNEL_VERSION(4, 6, 0) */
1476	rc = get_user_pages(pTask, /* Task for fault accounting. */
1477	pTask->mm, /* Whose pages. */
1478	R3Ptr, /* Where from. */
1479	cPages, /* How many pages. */
1480	/* The get_user_pages API change was back-ported to 4.4.168. */
1481	# if RTLNX_VER_RANGE(4,4,168, 4,5,0)
1482	fWrite ? FOLL_WRITE \| /* Write to memory. */
1483	FOLL_FORCE /* force write access. */
1484	: 0, /* Write to memory. */
1485	# else
1486	fWrite, /* Write to memory. */
1487	fWrite, /* force write access. */
1488	# endif
1489	&pMemLnx->apPages[0], /* Page array. */
1490	papVMAs /* vmas */
1491	);
1492	#endif /* GET_USER_PAGES_API < KERNEL_VERSION(4, 6, 0) */
1493	if (rc == cPages)
1494	{
1495	/*
1496	* Flush dcache (required?), protect against fork and _really_ pin the page
1497	* table entries. get_user_pages() will protect against swapping out the
1498	* pages but it will NOT protect against removing page table entries. This
1499	* can be achieved with
1500	* - using mlock / mmap(..., MAP_LOCKED, ...) from userland. This requires
1501	* an appropriate limit set up with setrlimit(..., RLIMIT_MEMLOCK, ...).
1502	* Usual Linux distributions support only a limited size of locked pages
1503	* (e.g. 32KB).
1504	* - setting the PageReserved bit (as we do in rtR0MemObjLinuxAllocPages()
1505	* or by
1506	* - setting the VM_LOCKED flag. This is the same as doing mlock() without
1507	* a range check.
1508	*/
1509	/** @todo The Linux fork() protection will require more work if this API
1510	* is to be used for anything but locking VM pages. */
1511	while (rc-- > 0)
1512	{
1513	flush_dcache_page(pMemLnx->apPages[rc]);
1514	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_SUSE_MAJ_PREREQ(15, 6) && !RTLNX_RHEL_RANGE(9,5, 9,99)
1515	# if RTLNX_VER_MIN(6,3,0)
1516	vm_flags_set(papVMAs[rc], VM_DONTCOPY \| VM_LOCKED);
1517	# else
1518	papVMAs[rc]->vm_flags \|= VM_DONTCOPY \| VM_LOCKED;
1519	# endif
1520	# endif
1521	}
1522
1523	LNX_MM_UP_READ(pTask->mm);
1524
1525	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1526	RTMemFree(papVMAs);
1527	# endif
1528
1529	pMemLnx->Core.u.Lock.R0Process = R0Process;
1530	pMemLnx->cPages = cPages;
1531	Assert(!pMemLnx->fMappedToRing0);
1532	*ppMem = &pMemLnx->Core;
1533
1534	IPRT_LINUX_RESTORE_EFL_AC();
1535	return VINF_SUCCESS;
1536	}
1537
1538	/*
1539	* Failed - we need to unlock any pages that we succeeded to lock.
1540	*/
1541	while (rc-- > 0)
1542	{
1543	if (!PageReserved(pMemLnx->apPages[rc]))
1544	SetPageDirty(pMemLnx->apPages[rc]);
1545	#if RTLNX_VER_MIN(4,6,0)
1546	put_page(pMemLnx->apPages[rc]);
1547	#else
1548	page_cache_release(pMemLnx->apPages[rc]);
1549	#endif
1550	}
1551
1552	LNX_MM_UP_READ(pTask->mm);
1553
1554	rc = VERR_LOCK_FAILED;
1555
1556	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1557	RTMemFree(papVMAs);
1558	}
1559	# endif
1560
1561	rtR0MemObjDelete(&pMemLnx->Core);
1562	IPRT_LINUX_RESTORE_EFL_AC();
1563	return rc;
1564	}
1565
1566
1567	DECLHIDDEN(int) rtR0MemObjNativeLockKernel(PPRTR0MEMOBJINTERNAL ppMem, void pv, size_t cb, uint32_t fAccess, const char pszTag)
1568	{
1569	IPRT_LINUX_SAVE_EFL_AC();
1570	void pvLast = (uint8_t )pv + cb - 1;
1571	size_t const cPages = cb >> PAGE_SHIFT;
1572	PRTR0MEMOBJLNX pMemLnx;
1573	bool fLinearMapping;
1574	int rc;
1575	uint8_t *pbPage;
1576	size_t iPage;
1577	NOREF(fAccess);
1578
1579	if ( !RTR0MemKernelIsValidAddr(pv)
1580	\|\| !RTR0MemKernelIsValidAddr(pv + cb))
1581	return VERR_INVALID_PARAMETER;
1582
1583	/*
1584	* The lower part of the kernel memory has a linear mapping between
1585	* physical and virtual addresses. So we take a short cut here. This is
1586	* assumed to be the cleanest way to handle those addresses (and the code
1587	* is well tested, though the test for determining it is not very nice).
1588	* If we ever decide it isn't we can still remove it.
1589	*/
1590	#if 0
1591	fLinearMapping = (unsigned long)pvLast < VMALLOC_START;
1592	#else
1593	fLinearMapping = (unsigned long)pv >= (unsigned long)__va(0)
1594	&& (unsigned long)pvLast < (unsigned long)high_memory;
1595	#endif
1596
1597	/*
1598	* Allocate the memory object.
1599	*/
1600	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_UOFFSETOF_DYN(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK,
1601	pv, cb, pszTag);
1602	if (!pMemLnx)
1603	{
1604	IPRT_LINUX_RESTORE_EFL_AC();
1605	return VERR_NO_MEMORY;
1606	}
1607
1608	/*
1609	* Gather the pages.
1610	* We ASSUME all kernel pages are non-swappable and non-movable.
1611	*/
1612	rc = VINF_SUCCESS;
1613	pbPage = (uint8_t *)pvLast;
1614	iPage = cPages;
1615	if (!fLinearMapping)
1616	{
1617	while (iPage-- > 0)
1618	{
1619	struct page *pPage = rtR0MemObjLinuxVirtToPage(pbPage);
1620	if (RT_UNLIKELY(!pPage))
1621	{
1622	rc = VERR_LOCK_FAILED;
1623	break;
1624	}
1625	pMemLnx->apPages[iPage] = pPage;
1626	pbPage -= PAGE_SIZE;
1627	}
1628	}
1629	else
1630	{
1631	while (iPage-- > 0)
1632	{
1633	pMemLnx->apPages[iPage] = virt_to_page(pbPage);
1634	pbPage -= PAGE_SIZE;
1635	}
1636	}
1637	if (RT_SUCCESS(rc))
1638	{
1639	/*
1640	* Complete the memory object and return.
1641	*/
1642	pMemLnx->Core.u.Lock.R0Process = NIL_RTR0PROCESS;
1643	pMemLnx->cPages = cPages;
1644	Assert(!pMemLnx->fMappedToRing0);
1645	*ppMem = &pMemLnx->Core;
1646
1647	IPRT_LINUX_RESTORE_EFL_AC();
1648	return VINF_SUCCESS;
1649	}
1650
1651	rtR0MemObjDelete(&pMemLnx->Core);
1652	IPRT_LINUX_RESTORE_EFL_AC();
1653	return rc;
1654	}
1655
1656
1657	DECLHIDDEN(int) rtR0MemObjNativeReserveKernel(PPRTR0MEMOBJINTERNAL ppMem, void *pvFixed, size_t cb, size_t uAlignment,
1658	const char *pszTag)
1659	{
1660	#if RTLNX_VER_MIN(2,4,22)
1661	IPRT_LINUX_SAVE_EFL_AC();
1662	const size_t cPages = cb >> PAGE_SHIFT;
1663	struct page *pDummyPage;
1664	struct page **papPages;
1665
1666	/* check for unsupported stuff. */
1667	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
1668	if (uAlignment > PAGE_SIZE)
1669	return VERR_NOT_SUPPORTED;
1670
1671	/*
1672	* Allocate a dummy page and create a page pointer array for vmap such that
1673	* the dummy page is mapped all over the reserved area.
1674	*/
1675	pDummyPage = alloc_page(GFP_HIGHUSER \| __GFP_NOWARN);
1676	if (pDummyPage)
1677	{
1678	papPages = RTMemAlloc(sizeof(papPages) cPages);
1679	if (papPages)
1680	{
1681	void *pv;
1682	size_t iPage = cPages;
1683	while (iPage-- > 0)
1684	papPages[iPage] = pDummyPage;
1685	# ifdef VM_MAP
1686	pv = vmap(papPages, cPages, VM_MAP, PAGE_KERNEL_RO);
1687	# else
1688	pv = vmap(papPages, cPages, VM_ALLOC, PAGE_KERNEL_RO);
1689	# endif
1690	RTMemFree(papPages);
1691	if (pv)
1692	{
1693	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb, pszTag);
1694	if (pMemLnx)
1695	{
1696	pMemLnx->Core.u.ResVirt.R0Process = NIL_RTR0PROCESS;
1697	pMemLnx->cPages = 1;
1698	pMemLnx->apPages[0] = pDummyPage;
1699	*ppMem = &pMemLnx->Core;
1700	IPRT_LINUX_RESTORE_EFL_AC();
1701	return VINF_SUCCESS;
1702	}
1703	vunmap(pv);
1704	}
1705	}
1706	__free_page(pDummyPage);
1707	}
1708	IPRT_LINUX_RESTORE_EFL_AC();
1709	return VERR_NO_MEMORY;
1710
1711	#else /* < 2.4.22 */
1712	/*
1713	* Could probably use ioremap here, but the caller is in a better position than us
1714	* to select some safe physical memory.
1715	*/
1716	return VERR_NOT_SUPPORTED;
1717	#endif
1718	}
1719
1720
1721	DECLHIDDEN(int) rtR0MemObjNativeReserveUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment,
1722	RTR0PROCESS R0Process, const char *pszTag)
1723	{
1724	IPRT_LINUX_SAVE_EFL_AC();
1725	PRTR0MEMOBJLNX pMemLnx;
1726	void *pv;
1727	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1728	if (!pTask)
1729	return VERR_NOT_SUPPORTED;
1730
1731	/*
1732	* Check that the specified alignment is supported.
1733	*/
1734	if (uAlignment > PAGE_SIZE)
1735	return VERR_NOT_SUPPORTED;
1736
1737	/*
1738	* Let rtR0MemObjLinuxDoMmap do the difficult bits.
1739	*/
1740	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, cb, uAlignment, pTask, RTMEM_PROT_NONE);
1741	if (pv == (void *)-1)
1742	{
1743	IPRT_LINUX_RESTORE_EFL_AC();
1744	return VERR_NO_MEMORY;
1745	}
1746
1747	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb, pszTag);
1748	if (!pMemLnx)
1749	{
1750	rtR0MemObjLinuxDoMunmap(pv, cb, pTask);
1751	IPRT_LINUX_RESTORE_EFL_AC();
1752	return VERR_NO_MEMORY;
1753	}
1754
1755	pMemLnx->Core.u.ResVirt.R0Process = R0Process;
1756	*ppMem = &pMemLnx->Core;
1757	IPRT_LINUX_RESTORE_EFL_AC();
1758	return VINF_SUCCESS;
1759	}
1760
1761
1762	DECLHIDDEN(int) rtR0MemObjNativeMapKernel(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap, void *pvFixed, size_t uAlignment,
1763	unsigned fProt, size_t offSub, size_t cbSub, const char *pszTag)
1764	{
1765	int rc = VERR_NO_MEMORY;
1766	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1767	PRTR0MEMOBJLNX pMemLnx;
1768	IPRT_LINUX_SAVE_EFL_AC();
1769
1770	/* Fail if requested to do something we can't. */
1771	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
1772	if (uAlignment > PAGE_SIZE)
1773	return VERR_NOT_SUPPORTED;
1774
1775	/*
1776	* Create the IPRT memory object.
1777	*/
1778	if (!cbSub)
1779	cbSub = pMemLnxToMap->Core.cb - offSub;
1780	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, cbSub, pszTag);
1781	if (pMemLnx)
1782	{
1783	if (pMemLnxToMap->cPages)
1784	{
1785	#if RTLNX_VER_MIN(2,4,22)
1786	/*
1787	* Use vmap - 2.4.22 and later.
1788	*/
1789	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, true /* kernel */);
1790	/** @todo We don't really care too much for EXEC here... 5.8 always adds NX. */
1791	Assert(((offSub + cbSub) >> PAGE_SHIFT) <= pMemLnxToMap->cPages);
1792	# ifdef VM_MAP
1793	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[offSub >> PAGE_SHIFT], cbSub >> PAGE_SHIFT, VM_MAP, fPg);
1794	# else
1795	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[offSub >> PAGE_SHIFT], cbSub >> PAGE_SHIFT, VM_ALLOC, fPg);
1796	# endif
1797	if (pMemLnx->Core.pv)
1798	{
1799	pMemLnx->fMappedToRing0 = true;
1800	rc = VINF_SUCCESS;
1801	}
1802	else
1803	rc = VERR_MAP_FAILED;
1804
1805	#else /* < 2.4.22 */
1806	/*
1807	* Only option here is to share mappings if possible and forget about fProt.
1808	*/
1809	if (rtR0MemObjIsRing3(pMemToMap))
1810	rc = VERR_NOT_SUPPORTED;
1811	else
1812	{
1813	rc = VINF_SUCCESS;
1814	if (!pMemLnxToMap->Core.pv)
1815	rc = rtR0MemObjLinuxVMap(pMemLnxToMap, !!(fProt & RTMEM_PROT_EXEC));
1816	if (RT_SUCCESS(rc))
1817	{
1818	Assert(pMemLnxToMap->Core.pv);
1819	pMemLnx->Core.pv = (uint8_t *)pMemLnxToMap->Core.pv + offSub;
1820	}
1821	}
1822	#endif
1823	}
1824	else
1825	{
1826	/*
1827	* MMIO / physical memory.
1828	*/
1829	Assert(pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS && !pMemLnxToMap->Core.u.Phys.fAllocated);
1830	#if RTLNX_VER_MIN(2,6,25)
1831	/*
1832	* ioremap() defaults to no caching since the 2.6 kernels.
1833	* ioremap_nocache() has been removed finally in 5.6-rc1.
1834	*/
1835	pMemLnx->Core.pv = pMemLnxToMap->Core.u.Phys.uCachePolicy == RTMEM_CACHE_POLICY_MMIO
1836	? ioremap(pMemLnxToMap->Core.u.Phys.PhysBase + offSub, cbSub)
1837	: ioremap_cache(pMemLnxToMap->Core.u.Phys.PhysBase + offSub, cbSub);
1838	#else /* KERNEL_VERSION < 2.6.25 */
1839	pMemLnx->Core.pv = pMemLnxToMap->Core.u.Phys.uCachePolicy == RTMEM_CACHE_POLICY_MMIO
1840	? ioremap_nocache(pMemLnxToMap->Core.u.Phys.PhysBase + offSub, cbSub)
1841	: ioremap(pMemLnxToMap->Core.u.Phys.PhysBase + offSub, cbSub);
1842	#endif /* KERNEL_VERSION < 2.6.25 */
1843	if (pMemLnx->Core.pv)
1844	{
1845	/** @todo fix protection. */
1846	rc = VINF_SUCCESS;
1847	}
1848	}
1849	if (RT_SUCCESS(rc))
1850	{
1851	pMemLnx->Core.u.Mapping.R0Process = NIL_RTR0PROCESS;
1852	*ppMem = &pMemLnx->Core;
1853	IPRT_LINUX_RESTORE_EFL_AC();
1854	return VINF_SUCCESS;
1855	}
1856	rtR0MemObjDelete(&pMemLnx->Core);
1857	}
1858
1859	IPRT_LINUX_RESTORE_EFL_AC();
1860	return rc;
1861	}
1862
1863
1864	#ifdef VBOX_USE_PAE_HACK
1865	/**
1866	* Replace the PFN of a PTE with the address of the actual page.
1867	*
1868	* The caller maps a reserved dummy page at the address with the desired access
1869	* and flags.
1870	*
1871	* This hack is required for older Linux kernels which don't provide
1872	* remap_pfn_range().
1873	*
1874	* @returns 0 on success, -ENOMEM on failure.
1875	* @param mm The memory context.
1876	* @param ulAddr The mapping address.
1877	* @param Phys The physical address of the page to map.
1878	*/
1879	static int rtR0MemObjLinuxFixPte(struct mm_struct *mm, unsigned long ulAddr, RTHCPHYS Phys)
1880	{
1881	int rc = -ENOMEM;
1882	pgd_t *pgd;
1883
1884	spin_lock(&mm->page_table_lock);
1885
1886	pgd = pgd_offset(mm, ulAddr);
1887	if (!pgd_none(pgd) && !pgd_bad(pgd))
1888	{
1889	pmd_t *pmd = pmd_offset(pgd, ulAddr);
1890	if (!pmd_none(*pmd))
1891	{
1892	pte_t *ptep = pte_offset_map(pmd, ulAddr);
1893	if (ptep)
1894	{
1895	pte_t pte = *ptep;
1896	pte.pte_high &= 0xfff00000;
1897	pte.pte_high \|= ((Phys >> 32) & 0x000fffff);
1898	pte.pte_low &= 0x00000fff;
1899	pte.pte_low \|= (Phys & 0xfffff000);
1900	set_pte(ptep, pte);
1901	pte_unmap(ptep);
1902	rc = 0;
1903	}
1904	}
1905	}
1906
1907	spin_unlock(&mm->page_table_lock);
1908	return rc;
1909	}
1910	#endif /* VBOX_USE_PAE_HACK */
1911
1912
1913	DECLHIDDEN(int) rtR0MemObjNativeMapUser(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap, RTR3PTR R3PtrFixed, size_t uAlignment,
1914	unsigned fProt, RTR0PROCESS R0Process, size_t offSub, size_t cbSub, const char *pszTag)
1915	{
1916	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1917	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1918	int rc = VERR_NO_MEMORY;
1919	PRTR0MEMOBJLNX pMemLnx;
1920	#ifdef VBOX_USE_PAE_HACK
1921	struct page *pDummyPage;
1922	RTHCPHYS DummyPhys;
1923	#endif
1924	IPRT_LINUX_SAVE_EFL_AC();
1925
1926	/*
1927	* Check for restrictions.
1928	*/
1929	if (!pTask)
1930	return VERR_NOT_SUPPORTED;
1931	if (uAlignment > PAGE_SIZE)
1932	return VERR_NOT_SUPPORTED;
1933
1934	#ifdef VBOX_USE_PAE_HACK
1935	/*
1936	* Allocate a dummy page for use when mapping the memory.
1937	*/
1938	pDummyPage = alloc_page(GFP_USER \| __GFP_NOWARN);
1939	if (!pDummyPage)
1940	{
1941	IPRT_LINUX_RESTORE_EFL_AC();
1942	return VERR_NO_MEMORY;
1943	}
1944	SetPageReserved(pDummyPage);
1945	DummyPhys = page_to_phys(pDummyPage);
1946	#endif
1947
1948	/*
1949	* Create the IPRT memory object.
1950	*/
1951	Assert(!offSub \|\| cbSub);
1952	if (cbSub == 0)
1953	cbSub = pMemLnxToMap->Core.cb;
1954	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, cbSub, pszTag);
1955	if (pMemLnx)
1956	{
1957	/*
1958	* Allocate user space mapping.
1959	*/
1960	void *pv;
1961	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, cbSub, uAlignment, pTask, fProt);
1962	if (pv != (void *)-1)
1963	{
1964	/*
1965	* Map page by page into the mmap area.
1966	* This is generic, paranoid and not very efficient.
1967	*/
1968	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, false /* user */);
1969	unsigned long ulAddrCur = (unsigned long)pv;
1970	const size_t cPages = (offSub + cbSub) >> PAGE_SHIFT;
1971	size_t iPage;
1972
1973	LNX_MM_DOWN_WRITE(pTask->mm);
1974
1975	rc = VINF_SUCCESS;
1976	if (pMemLnxToMap->cPages)
1977	{
1978	for (iPage = offSub >> PAGE_SHIFT; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE)
1979	{
1980	#if RTLNX_VER_MAX(2,6,11)
1981	RTHCPHYS Phys = page_to_phys(pMemLnxToMap->apPages[iPage]);
1982	#endif
1983	#if RTLNX_VER_MIN(2,6,0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1984	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
1985	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
1986	#endif
1987	#if RTLNX_VER_MAX(2,6,0) && defined(RT_ARCH_X86)
1988	/* remap_page_range() limitation on x86 */
1989	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
1990	#endif
1991
1992	#if defined(VBOX_USE_INSERT_PAGE) && RTLNX_VER_MIN(2,6,22)
1993	rc = vm_insert_page(vma, ulAddrCur, pMemLnxToMap->apPages[iPage]);
1994	/* Thes flags help making 100% sure some bad stuff wont happen (swap, core, ++).
1995	* See remap_pfn_range() in mm/memory.c */
1996
1997	#if RTLNX_VER_MIN(6,3,0) \|\| RTLNX_RHEL_RANGE(9,5, 9,99)
1998	vm_flags_set(vma, VM_DONTEXPAND \| VM_DONTDUMP);
1999	#elif RTLNX_VER_MIN(3,7,0)
2000	vma->vm_flags \|= VM_DONTEXPAND \| VM_DONTDUMP;
2001	#else
2002	vma->vm_flags \|= VM_RESERVED;
2003	#endif
2004	#elif RTLNX_VER_MIN(2,6,11)
2005	rc = remap_pfn_range(vma, ulAddrCur, page_to_pfn(pMemLnxToMap->apPages[iPage]), PAGE_SIZE, fPg);
2006	#elif defined(VBOX_USE_PAE_HACK)
2007	rc = remap_page_range(vma, ulAddrCur, DummyPhys, PAGE_SIZE, fPg);
2008	if (!rc)
2009	rc = rtR0MemObjLinuxFixPte(pTask->mm, ulAddrCur, Phys);
2010	#elif RTLNX_VER_MIN(2,6,0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
2011	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
2012	#else /* 2.4 */
2013	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
2014	#endif
2015	if (rc)
2016	{
2017	rc = VERR_NO_MEMORY;
2018	break;
2019	}
2020	}
2021	}
2022	else
2023	{
2024	RTHCPHYS Phys;
2025	if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS)
2026	Phys = pMemLnxToMap->Core.u.Phys.PhysBase;
2027	else if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_CONT)
2028	Phys = pMemLnxToMap->Core.u.Cont.Phys;
2029	else
2030	{
2031	AssertMsgFailed(("%d\n", pMemLnxToMap->Core.enmType));
2032	Phys = NIL_RTHCPHYS;
2033	}
2034	if (Phys != NIL_RTHCPHYS)
2035	{
2036	for (iPage = offSub >> PAGE_SHIFT; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE, Phys += PAGE_SIZE)
2037	{
2038	#if RTLNX_VER_MIN(2,6,0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
2039	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
2040	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
2041	#endif
2042	#if RTLNX_VER_MAX(2,6,0) && defined(RT_ARCH_X86)
2043	/* remap_page_range() limitation on x86 */
2044	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
2045	#endif
2046
2047	#if RTLNX_VER_MIN(2,6,11)
2048	rc = remap_pfn_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
2049	#elif defined(VBOX_USE_PAE_HACK)
2050	rc = remap_page_range(vma, ulAddrCur, DummyPhys, PAGE_SIZE, fPg);
2051	if (!rc)
2052	rc = rtR0MemObjLinuxFixPte(pTask->mm, ulAddrCur, Phys);
2053	#elif RTLNX_VER_MIN(2,6,0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
2054	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
2055	#else /* 2.4 */
2056	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
2057	#endif
2058	if (rc)
2059	{
2060	rc = VERR_NO_MEMORY;
2061	break;
2062	}
2063	}
2064	}
2065	}
2066
2067	#ifdef CONFIG_NUMA_BALANCING
2068	# if RTLNX_VER_MAX(3,13,0) && RTLNX_RHEL_MAX(7,0)
2069	# define VBOX_NUMA_HACK_OLD
2070	# endif
2071	if (RT_SUCCESS(rc))
2072	{
2073	/** @todo Ugly hack! But right now we have no other means to
2074	* disable automatic NUMA page balancing. */
2075	# ifdef RT_OS_X86
2076	# ifdef VBOX_NUMA_HACK_OLD
2077	pTask->mm->numa_next_reset = jiffies + 0x7fffffffUL;
2078	# endif
2079	pTask->mm->numa_next_scan = jiffies + 0x7fffffffUL;
2080	# else
2081	# ifdef VBOX_NUMA_HACK_OLD
2082	pTask->mm->numa_next_reset = jiffies + 0x7fffffffffffffffUL;
2083	# endif
2084	pTask->mm->numa_next_scan = jiffies + 0x7fffffffffffffffUL;
2085	# endif
2086	}
2087	#endif /* CONFIG_NUMA_BALANCING */
2088
2089	LNX_MM_UP_WRITE(pTask->mm);
2090
2091	if (RT_SUCCESS(rc))
2092	{
2093	#ifdef VBOX_USE_PAE_HACK
2094	__free_page(pDummyPage);
2095	#endif
2096	pMemLnx->Core.pv = pv;
2097	pMemLnx->Core.u.Mapping.R0Process = R0Process;
2098	*ppMem = &pMemLnx->Core;
2099	IPRT_LINUX_RESTORE_EFL_AC();
2100	return VINF_SUCCESS;
2101	}
2102
2103	/*
2104	* Bail out.
2105	*/
2106	rtR0MemObjLinuxDoMunmap(pv, cbSub, pTask);
2107	}
2108	rtR0MemObjDelete(&pMemLnx->Core);
2109	}
2110	#ifdef VBOX_USE_PAE_HACK
2111	__free_page(pDummyPage);
2112	#endif
2113
2114	IPRT_LINUX_RESTORE_EFL_AC();
2115	return rc;
2116	}
2117
2118
2119	DECLHIDDEN(int) rtR0MemObjNativeProtect(PRTR0MEMOBJINTERNAL pMem, size_t offSub, size_t cbSub, uint32_t fProt)
2120	{
2121	# ifdef IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
2122	/*
2123	* Currently only supported when we've got addresses PTEs from the kernel.
2124	*/
2125	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
2126	if (pMemLnx->pArea && pMemLnx->papPtesForArea)
2127	{
2128	pgprot_t const fPg = rtR0MemObjLinuxConvertProt(fProt, true /fKernel/);
2129	size_t const cPages = (offSub + cbSub) >> PAGE_SHIFT;
2130	pte_t **papPtes = pMemLnx->papPtesForArea;
2131	size_t i;
2132
2133	for (i = offSub >> PAGE_SHIFT; i < cPages; i++)
2134	{
2135	set_pte(papPtes[i], mk_pte(pMemLnx->apPages[i], fPg));
2136	}
2137	preempt_disable();
2138	__flush_tlb_all();
2139	preempt_enable();
2140	return VINF_SUCCESS;
2141	}
2142	# elif defined(IPRT_USE_APPLY_TO_PAGE_RANGE_FOR_EXEC)
2143	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
2144	if ( pMemLnx->fExecutable
2145	&& pMemLnx->fMappedToRing0)
2146	{
2147	LNXAPPLYPGRANGE Args;
2148	Args.pMemLnx = pMemLnx;
2149	Args.fPg = rtR0MemObjLinuxConvertProt(fProt, true /fKernel/);
2150	int rcLnx = apply_to_page_range(current->active_mm, (unsigned long)pMemLnx->Core.pv + offSub, cbSub,
2151	rtR0MemObjLinuxApplyPageRange, (void *)&Args);
2152	if (rcLnx)
2153	return VERR_NOT_SUPPORTED;
2154
2155	return VINF_SUCCESS;
2156	}
2157	# endif
2158
2159	NOREF(pMem);
2160	NOREF(offSub);
2161	NOREF(cbSub);
2162	NOREF(fProt);
2163	return VERR_NOT_SUPPORTED;
2164	}
2165
2166
2167	DECLHIDDEN(RTHCPHYS) rtR0MemObjNativeGetPagePhysAddr(PRTR0MEMOBJINTERNAL pMem, size_t iPage)
2168	{
2169	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
2170
2171	if (pMemLnx->cPages)
2172	return page_to_phys(pMemLnx->apPages[iPage]);
2173
2174	switch (pMemLnx->Core.enmType)
2175	{
2176	case RTR0MEMOBJTYPE_CONT:
2177	return pMemLnx->Core.u.Cont.Phys + (iPage << PAGE_SHIFT);
2178
2179	case RTR0MEMOBJTYPE_PHYS:
2180	return pMemLnx->Core.u.Phys.PhysBase + (iPage << PAGE_SHIFT);
2181
2182	/* the parent knows */
2183	case RTR0MEMOBJTYPE_MAPPING:
2184	return rtR0MemObjNativeGetPagePhysAddr(pMemLnx->Core.uRel.Child.pParent, iPage);
2185
2186	/* cPages > 0 */
2187	case RTR0MEMOBJTYPE_LOW:
2188	case RTR0MEMOBJTYPE_LOCK:
2189	case RTR0MEMOBJTYPE_PHYS_NC:
2190	case RTR0MEMOBJTYPE_PAGE:
2191	case RTR0MEMOBJTYPE_LARGE_PAGE:
2192	default:
2193	AssertMsgFailed(("%d\n", pMemLnx->Core.enmType));
2194	RT_FALL_THROUGH();
2195
2196	case RTR0MEMOBJTYPE_RES_VIRT:
2197	return NIL_RTHCPHYS;
2198	}
2199	}
2200
2201
2202	DECLHIDDEN(int) rtR0MemObjNativeZeroInitWithoutMapping(PRTR0MEMOBJINTERNAL pMem)
2203	{
2204	PRTR0MEMOBJLNX const pMemLnx = (PRTR0MEMOBJLNX)pMem;
2205	size_t const cPages = pMemLnx->Core.cb >> PAGE_SHIFT;
2206	size_t iPage;
2207	/** @todo optimize this. */
2208	for (iPage = 0; iPage < cPages; iPage++)
2209	{
2210	void *pvPage;
2211
2212	/* Get the physical address of the page. */
2213	RTHCPHYS const HCPhys = rtR0MemObjNativeGetPagePhysAddr(&pMemLnx->Core, iPage);
2214	AssertReturn(HCPhys != NIL_RTHCPHYS, VERR_INTERNAL_ERROR_3);
2215	Assert(!(HCPhys & PAGE_OFFSET_MASK));
2216
2217	/* Would've like to use valid_phys_addr_range for this test, but it isn't exported. */
2218	AssertReturn((HCPhys \| PAGE_OFFSET_MASK) < __pa(high_memory), VERR_INTERNAL_ERROR_3);
2219
2220	/* Map it. */
2221	pvPage = phys_to_virt(HCPhys);
2222	AssertPtrReturn(pvPage, VERR_INTERNAL_ERROR_3);
2223
2224	/* Zero it. */
2225	RT_BZERO(pvPage, PAGE_SIZE);
2226	}
2227	return VINF_SUCCESS;
2228	}
2229

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: