memobj-r0drv-linux.c@ 109071

Last change on this file since 109071 was 109071, checked in by vboxsync, 4 weeks ago
Runtime/r0drv/linux/memobj-r0drv-linux.c: GFP_REPEAT was renamed to GFP_RETRY_MAYFAIL in 2017 with commit dcda9b04713c3f6ff0875652924844fae28286ea, so we weren't instructing the allocator to try hard to allocate memory with such kernels. Hopefully fixes VERR_NO_LOW_MEMORY errors when starting a VM
Property svn:eol-style set to `native` Property svn:keywords set to `Id Rev Revision`
File size: 76.3 KB

Line
1	/* $Id: memobj-r0drv-linux.c 109071 2025-04-24 12:34:13Z 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	# elif defined (__GFP_RETRY_MAYFAIL) /* Renamed in commit dcda9b04713c3f6ff0875652924844fae28286ea . */
418	fFlagsLnx \|= __GFP_RETRY_MAYFAIL;
419	# else /* This is to notice when the flags are renamed/moved around again. */
420	# error "Was this flag renamed again?"
421	# endif
422	# ifdef __GFP_NOMEMALLOC
423	/* Introduced with Linux 2.6.12: Don't use emergency reserves */
424	fFlagsLnx \|= __GFP_NOMEMALLOC;
425	# endif
426	}
427
428	/*
429	* Allocate the pages.
430	* For small allocations we'll try contiguous first and then fall back on page by page.
431	*/
432	#if RTLNX_VER_MIN(2,4,22)
433	if ( fContiguous
434	\|\| cb <= PAGE_SIZE * 2)
435	{
436	# ifdef VBOX_USE_INSERT_PAGE
437	paPages = alloc_pages(fFlagsLnx \| __GFP_COMP \| __GFP_NOWARN, rtR0MemObjLinuxOrder(cPages));
438	# else
439	paPages = alloc_pages(fFlagsLnx \| __GFP_NOWARN, rtR0MemObjLinuxOrder(cPages));
440	# endif
441	if (paPages)
442	{
443	fContiguous = true;
444	for (iPage = 0; iPage < cPages; iPage++)
445	pMemLnx->apPages[iPage] = &paPages[iPage];
446	}
447	else if (fContiguous)
448	{
449	rtR0MemObjDelete(&pMemLnx->Core);
450	return rcNoMem;
451	}
452	}
453
454	if (!fContiguous)
455	{
456	/** @todo Try use alloc_pages_bulk_array when available, it should be faster
457	* than a alloc_page loop. Put it in #ifdefs similar to
458	* IPRT_USE_APPLY_TO_PAGE_RANGE_FOR_EXEC. */
459	for (iPage = 0; iPage < cPages; iPage++)
460	{
461	pMemLnx->apPages[iPage] = alloc_page(fFlagsLnx \| __GFP_NOWARN);
462	if (RT_UNLIKELY(!pMemLnx->apPages[iPage]))
463	{
464	while (iPage-- > 0)
465	__free_page(pMemLnx->apPages[iPage]);
466	rtR0MemObjDelete(&pMemLnx->Core);
467	return rcNoMem;
468	}
469	}
470	}
471
472	#else /* < 2.4.22 */
473	/** @todo figure out why we didn't allocate page-by-page on 2.4.21 and older... */
474	paPages = alloc_pages(fFlagsLnx, rtR0MemObjLinuxOrder(cPages));
475	if (!paPages)
476	{
477	rtR0MemObjDelete(&pMemLnx->Core);
478	return rcNoMem;
479	}
480	for (iPage = 0; iPage < cPages; iPage++)
481	{
482	pMemLnx->apPages[iPage] = &paPages[iPage];
483	if (fExecutable)
484	MY_SET_PAGES_EXEC(pMemLnx->apPages[iPage], 1);
485	if (PageHighMem(pMemLnx->apPages[iPage]))
486	BUG();
487	}
488
489	fContiguous = true;
490	#endif /* < 2.4.22 */
491	pMemLnx->fContiguous = fContiguous;
492	pMemLnx->fExecutable = fExecutable;
493
494	#if RTLNX_VER_MAX(4,5,0)
495	/*
496	* Reserve the pages.
497	*
498	* Linux >= 4.5 with CONFIG_DEBUG_VM panics when setting PG_reserved on compound
499	* pages. According to Michal Hocko this shouldn't be necessary anyway because
500	* as pages which are not on the LRU list are never evictable.
501	*/
502	for (iPage = 0; iPage < cPages; iPage++)
503	SetPageReserved(pMemLnx->apPages[iPage]);
504	#endif
505
506	/*
507	* Note that the physical address of memory allocated with alloc_pages(flags, order)
508	* is always 2^(PAGE_SHIFT+order)-aligned.
509	*/
510	if ( fContiguous
511	&& uAlignment > PAGE_SIZE)
512	{
513	/*
514	* Check for alignment constraints. The physical address of memory allocated with
515	* alloc_pages(flags, order) is always 2^(PAGE_SHIFT+order)-aligned.
516	*/
517	if (RT_UNLIKELY(page_to_phys(pMemLnx->apPages[0]) & (uAlignment - 1)))
518	{
519	/*
520	* This should never happen!
521	*/
522	printk("rtR0MemObjLinuxAllocPages(cb=0x%lx, uAlignment=0x%lx): alloc_pages(..., %d) returned physical memory at 0x%lx!\n",
523	(unsigned long)cb, (unsigned long)uAlignment, rtR0MemObjLinuxOrder(cPages), (unsigned long)page_to_phys(pMemLnx->apPages[0]));
524	rtR0MemObjLinuxFreePages(pMemLnx);
525	return rcNoMem;
526	}
527	}
528
529	*ppMemLnx = pMemLnx;
530	return VINF_SUCCESS;
531	}
532
533
534	/**
535	* Frees the physical pages allocated by the rtR0MemObjLinuxAllocPages() call.
536	*
537	* This method does NOT free the object.
538	*
539	* @param pMemLnx The object which physical pages should be freed.
540	*/
541	static void rtR0MemObjLinuxFreePages(PRTR0MEMOBJLNX pMemLnx)
542	{
543	size_t iPage = pMemLnx->cPages;
544	if (iPage > 0)
545	{
546	/*
547	* Restore the page flags.
548	*/
549	while (iPage-- > 0)
550	{
551	#if RTLNX_VER_MAX(4,5,0)
552	/* See SetPageReserved() in rtR0MemObjLinuxAllocPages() */
553	ClearPageReserved(pMemLnx->apPages[iPage]);
554	#endif
555	#if RTLNX_VER_MAX(2,4,22)
556	if (pMemLnx->fExecutable)
557	MY_SET_PAGES_NOEXEC(pMemLnx->apPages[iPage], 1);
558	#endif
559	}
560
561	/*
562	* Free the pages.
563	*/
564	#if RTLNX_VER_MIN(2,4,22)
565	if (!pMemLnx->fContiguous)
566	{
567	iPage = pMemLnx->cPages;
568	while (iPage-- > 0)
569	__free_page(pMemLnx->apPages[iPage]);
570	}
571	else
572	#endif
573	__free_pages(pMemLnx->apPages[0], rtR0MemObjLinuxOrder(pMemLnx->cPages));
574
575	pMemLnx->cPages = 0;
576	}
577	}
578
579
580	#ifdef IPRT_USE_APPLY_TO_PAGE_RANGE_FOR_EXEC
581	/**
582	* User data passed to the apply_to_page_range() callback.
583	*/
584	typedef struct LNXAPPLYPGRANGE
585	{
586	/** Pointer to the memory object. */
587	PRTR0MEMOBJLNX pMemLnx;
588	/** The page protection flags to apply. */
589	pgprot_t fPg;
590	} LNXAPPLYPGRANGE;
591	/** Pointer to the user data. */
592	typedef LNXAPPLYPGRANGE *PLNXAPPLYPGRANGE;
593	/** Pointer to the const user data. */
594	typedef const LNXAPPLYPGRANGE *PCLNXAPPLYPGRANGE;
595
596	/**
597	* Callback called in apply_to_page_range().
598	*
599	* @returns Linux status code.
600	* @param pPte Pointer to the page table entry for the given address.
601	* @param uAddr The address to apply the new protection to.
602	* @param pvUser The opaque user data.
603	*/
604	static int rtR0MemObjLinuxApplyPageRange(pte_t pPte, unsigned long uAddr, void pvUser)
605	{
606	PCLNXAPPLYPGRANGE pArgs = (PCLNXAPPLYPGRANGE)pvUser;
607	PRTR0MEMOBJLNX pMemLnx = pArgs->pMemLnx;
608	size_t idxPg = (uAddr - (unsigned long)pMemLnx->Core.pv) >> PAGE_SHIFT;
609
610	set_pte(pPte, mk_pte(pMemLnx->apPages[idxPg], pArgs->fPg));
611	return 0;
612	}
613	#endif
614
615
616	/**
617	* Maps the allocation into ring-0.
618	*
619	* This will update the RTR0MEMOBJLNX::Core.pv and RTR0MEMOBJ::fMappedToRing0 members.
620	*
621	* Contiguous mappings that isn't in 'high' memory will already be mapped into kernel
622	* space, so we'll use that mapping if possible. If execute access is required, we'll
623	* play safe and do our own mapping.
624	*
625	* @returns IPRT status code.
626	* @param pMemLnx The linux memory object to map.
627	* @param fExecutable Whether execute access is required.
628	*/
629	static int rtR0MemObjLinuxVMap(PRTR0MEMOBJLNX pMemLnx, bool fExecutable)
630	{
631	int rc = VINF_SUCCESS;
632
633	/*
634	* Choose mapping strategy.
635	*/
636	bool fMustMap = fExecutable
637	\|\| !pMemLnx->fContiguous;
638	if (!fMustMap)
639	{
640	size_t iPage = pMemLnx->cPages;
641	while (iPage-- > 0)
642	if (PageHighMem(pMemLnx->apPages[iPage]))
643	{
644	fMustMap = true;
645	break;
646	}
647	}
648
649	Assert(!pMemLnx->Core.pv);
650	Assert(!pMemLnx->fMappedToRing0);
651
652	if (fMustMap)
653	{
654	/*
655	* Use vmap - 2.4.22 and later.
656	*/
657	#if RTLNX_VER_MIN(2,4,22) && (defined(RT_ARCH_AMD64) \|\| defined(RT_ARCH_X86) \|\| defined(RT_ARCH_ARM64))
658	pgprot_t fPg;
659	# if defined(RT_ARCH_ARM64)
660	/* ARM64 architecture has no _PAGE_NX, _PAGE_PRESENT and _PAGE_RW flags.
661	* Closest alternatives would be PTE_PXN, PTE_UXN, PROT_DEFAULT and PTE_WRITE. */
662	# if RTLNX_VER_MIN(6,5,0)
663	pgprot_val(fPg) = _PAGE_KERNEL; /* (PROT_DEFAULT \| PTE_PXN \| PTE_UXN \| PTE_WRITE \| PTE_ATTRINDX(MT_NORMAL). */
664	# else /* < 6.5.0 */
665	pgprot_val(fPg) = PROT_NORMAL; /* (PROT_DEFAULT \| PTE_PXN \| PTE_UXN \| PTE_WRITE \| PTE_ATTRINDX(MT_NORMAL). */
666	# endif /* 6.5.0 */
667	# else /* !RT_ARCH_ARM64 */
668	pgprot_val(fPg) = _PAGE_PRESENT \| _PAGE_RW;
669	# ifdef _PAGE_NX
670	if (!fExecutable)
671	pgprot_val(fPg) \|= _PAGE_NX;
672	# endif
673	# endif /* RT_ARCH_ARM64 */
674
675	# ifdef IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
676	if (fExecutable)
677	{
678	# if RTLNX_VER_MIN(3,2,51)
679	pte_t papPtes = (pte_t )kmalloc_array(pMemLnx->cPages, sizeof(papPtes[0]), GFP_KERNEL);
680	# else
681	pte_t papPtes = (pte_t )kmalloc(pMemLnx->cPages * sizeof(papPtes[0]), GFP_KERNEL);
682	# endif
683	if (papPtes)
684	{
685	pMemLnx->pArea = alloc_vm_area(pMemLnx->Core.cb, papPtes); /* Note! pArea->nr_pages is not set. */
686	if (pMemLnx->pArea)
687	{
688	size_t i;
689	Assert(pMemLnx->pArea->size >= pMemLnx->Core.cb); /* Note! includes guard page. */
690	Assert(pMemLnx->pArea->addr);
691	# if !defined(RT_ARCH_ARM64) && defined(_PAGE_NX)
692	pgprot_val(fPg) \|= _PAGE_NX; /* Uses RTR0MemObjProtect to clear NX when memory ready, W^X fashion. */
693	# endif
694	pMemLnx->papPtesForArea = papPtes;
695	for (i = 0; i < pMemLnx->cPages; i++)
696	*papPtes[i] = mk_pte(pMemLnx->apPages[i], fPg);
697	pMemLnx->Core.pv = pMemLnx->pArea->addr;
698	pMemLnx->fMappedToRing0 = true;
699	}
700	else
701	{
702	kfree(papPtes);
703	rc = VERR_MAP_FAILED;
704	}
705	}
706	else
707	rc = VERR_MAP_FAILED;
708	}
709	else
710	# endif
711	{
712	# if !defined(RT_ARCH_ARM64) && defined(IPRT_USE_APPLY_TO_PAGE_RANGE_FOR_EXEC)
713	if (fExecutable)
714	pgprot_val(fPg) \|= _PAGE_NX; /* Uses RTR0MemObjProtect to clear NX when memory ready, W^X fashion. */
715	# endif
716
717	# ifdef VM_MAP
718	pMemLnx->Core.pv = vmap(&pMemLnx->apPages[0], pMemLnx->cPages, VM_MAP, fPg);
719	# else
720	pMemLnx->Core.pv = vmap(&pMemLnx->apPages[0], pMemLnx->cPages, VM_ALLOC, fPg);
721	# endif
722	if (pMemLnx->Core.pv)
723	pMemLnx->fMappedToRing0 = true;
724	else
725	rc = VERR_MAP_FAILED;
726	}
727	#else /* < 2.4.22 */
728	rc = VERR_NOT_SUPPORTED;
729	#endif
730	}
731	else
732	{
733	/*
734	* Use the kernel RAM mapping.
735	*/
736	pMemLnx->Core.pv = phys_to_virt(page_to_phys(pMemLnx->apPages[0]));
737	Assert(pMemLnx->Core.pv);
738	}
739
740	return rc;
741	}
742
743
744	/**
745	* Undoes what rtR0MemObjLinuxVMap() did.
746	*
747	* @param pMemLnx The linux memory object.
748	*/
749	static void rtR0MemObjLinuxVUnmap(PRTR0MEMOBJLNX pMemLnx)
750	{
751	#if RTLNX_VER_MIN(2,4,22)
752	# ifdef IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
753	if (pMemLnx->pArea)
754	{
755	# if 0
756	pte_t **papPtes = pMemLnx->papPtesForArea;
757	size_t i;
758	for (i = 0; i < pMemLnx->cPages; i++)
759	*papPtes[i] = 0;
760	# endif
761	free_vm_area(pMemLnx->pArea);
762	kfree(pMemLnx->papPtesForArea);
763	pMemLnx->pArea = NULL;
764	pMemLnx->papPtesForArea = NULL;
765	}
766	else
767	# endif
768	if (pMemLnx->fMappedToRing0)
769	{
770	Assert(pMemLnx->Core.pv);
771	vunmap(pMemLnx->Core.pv);
772	pMemLnx->fMappedToRing0 = false;
773	}
774	#else /* < 2.4.22 */
775	Assert(!pMemLnx->fMappedToRing0);
776	#endif
777	pMemLnx->Core.pv = NULL;
778	}
779
780
781	DECLHIDDEN(int) rtR0MemObjNativeFree(RTR0MEMOBJ pMem)
782	{
783	IPRT_LINUX_SAVE_EFL_AC();
784	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
785
786	/*
787	* Release any memory that we've allocated or locked.
788	*/
789	switch (pMemLnx->Core.enmType)
790	{
791	case RTR0MEMOBJTYPE_PAGE:
792	case RTR0MEMOBJTYPE_LOW:
793	case RTR0MEMOBJTYPE_CONT:
794	case RTR0MEMOBJTYPE_PHYS:
795	case RTR0MEMOBJTYPE_PHYS_NC:
796	rtR0MemObjLinuxVUnmap(pMemLnx);
797	rtR0MemObjLinuxFreePages(pMemLnx);
798	break;
799
800	case RTR0MEMOBJTYPE_LARGE_PAGE:
801	{
802	uint32_t const cLargePages = pMemLnx->Core.cb >> (pMemLnx->cLargePageOrder + PAGE_SHIFT);
803	uint32_t iLargePage;
804	for (iLargePage = 0; iLargePage < cLargePages; iLargePage++)
805	__free_pages(pMemLnx->apPages[iLargePage << pMemLnx->cLargePageOrder], pMemLnx->cLargePageOrder);
806	pMemLnx->cPages = 0;
807
808	#ifdef IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
809	Assert(!pMemLnx->pArea);
810	Assert(!pMemLnx->papPtesForArea);
811	#endif
812	break;
813	}
814
815	case RTR0MEMOBJTYPE_LOCK:
816	if (pMemLnx->Core.u.Lock.R0Process != NIL_RTR0PROCESS)
817	{
818	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
819	size_t iPage;
820	Assert(pTask);
821	if (pTask && pTask->mm)
822	LNX_MM_DOWN_READ(pTask->mm);
823
824	iPage = pMemLnx->cPages;
825	while (iPage-- > 0)
826	{
827	if (!PageReserved(pMemLnx->apPages[iPage]))
828	SetPageDirty(pMemLnx->apPages[iPage]);
829	#if RTLNX_VER_MIN(4,6,0)
830	put_page(pMemLnx->apPages[iPage]);
831	#else
832	page_cache_release(pMemLnx->apPages[iPage]);
833	#endif
834	}
835
836	if (pTask && pTask->mm)
837	LNX_MM_UP_READ(pTask->mm);
838	}
839	/* else: kernel memory - nothing to do here. */
840	break;
841
842	case RTR0MEMOBJTYPE_RES_VIRT:
843	Assert(pMemLnx->Core.pv);
844	if (pMemLnx->Core.u.ResVirt.R0Process != NIL_RTR0PROCESS)
845	{
846	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
847	Assert(pTask);
848	if (pTask && pTask->mm)
849	rtR0MemObjLinuxDoMunmap(pMemLnx->Core.pv, pMemLnx->Core.cb, pTask);
850	}
851	else
852	{
853	vunmap(pMemLnx->Core.pv);
854
855	Assert(pMemLnx->cPages == 1 && pMemLnx->apPages[0] != NULL);
856	__free_page(pMemLnx->apPages[0]);
857	pMemLnx->apPages[0] = NULL;
858	pMemLnx->cPages = 0;
859	}
860	pMemLnx->Core.pv = NULL;
861	break;
862
863	case RTR0MEMOBJTYPE_MAPPING:
864	Assert(pMemLnx->cPages == 0); Assert(pMemLnx->Core.pv);
865	if (pMemLnx->Core.u.ResVirt.R0Process != NIL_RTR0PROCESS)
866	{
867	struct task_struct *pTask = rtR0ProcessToLinuxTask(pMemLnx->Core.u.Lock.R0Process);
868	Assert(pTask);
869	if (pTask && pTask->mm)
870	rtR0MemObjLinuxDoMunmap(pMemLnx->Core.pv, pMemLnx->Core.cb, pTask);
871	}
872	else
873	vunmap(pMemLnx->Core.pv);
874	pMemLnx->Core.pv = NULL;
875	break;
876
877	default:
878	AssertMsgFailed(("enmType=%d\n", pMemLnx->Core.enmType));
879	return VERR_INTERNAL_ERROR;
880	}
881	IPRT_LINUX_RESTORE_EFL_ONLY_AC();
882	return VINF_SUCCESS;
883	}
884
885
886	DECLHIDDEN(int) rtR0MemObjNativeAllocPage(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable, const char *pszTag)
887	{
888	IPRT_LINUX_SAVE_EFL_AC();
889	PRTR0MEMOBJLNX pMemLnx;
890	int rc;
891
892	#if RTLNX_VER_MIN(2,4,22)
893	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_PAGE, cb, PAGE_SIZE, GFP_HIGHUSER,
894	false /* non-contiguous */, fExecutable, VERR_NO_MEMORY, pszTag);
895	#else
896	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_PAGE, cb, PAGE_SIZE, GFP_USER,
897	false /* non-contiguous */, fExecutable, VERR_NO_MEMORY, pszTag);
898	#endif
899	if (RT_SUCCESS(rc))
900	{
901	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
902	if (RT_SUCCESS(rc))
903	{
904	*ppMem = &pMemLnx->Core;
905	IPRT_LINUX_RESTORE_EFL_AC();
906	return rc;
907	}
908
909	rtR0MemObjLinuxFreePages(pMemLnx);
910	rtR0MemObjDelete(&pMemLnx->Core);
911	}
912
913	IPRT_LINUX_RESTORE_EFL_AC();
914	return rc;
915	}
916
917
918	DECLHIDDEN(int) rtR0MemObjNativeAllocLarge(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, size_t cbLargePage, uint32_t fFlags,
919	const char *pszTag)
920	{
921	#ifdef GFP_TRANSHUGE
922	/*
923	* Allocate a memory object structure that's large enough to contain
924	* the page pointer array.
925	*/
926	# ifdef __GFP_MOVABLE
927	unsigned const fGfp = (GFP_TRANSHUGE \| __GFP_ZERO) & ~__GFP_MOVABLE;
928	# else
929	unsigned const fGfp = (GFP_TRANSHUGE \| __GFP_ZERO);
930	# endif
931	size_t const cPagesPerLarge = cbLargePage >> PAGE_SHIFT;
932	unsigned const cLargePageOrder = rtR0MemObjLinuxOrder(cPagesPerLarge);
933	size_t const cLargePages = cb >> (cLargePageOrder + PAGE_SHIFT);
934	size_t const cPages = cb >> PAGE_SHIFT;
935	PRTR0MEMOBJLNX pMemLnx;
936
937	Assert(RT_BIT_64(cLargePageOrder + PAGE_SHIFT) == cbLargePage);
938	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_UOFFSETOF_DYN(RTR0MEMOBJLNX, apPages[cPages]),
939	RTR0MEMOBJTYPE_LARGE_PAGE, NULL, cb, pszTag);
940	if (pMemLnx)
941	{
942	size_t iLargePage;
943
944	pMemLnx->Core.fFlags \|= RTR0MEMOBJ_FLAGS_ZERO_AT_ALLOC;
945	pMemLnx->cLargePageOrder = cLargePageOrder;
946	pMemLnx->cPages = cPages;
947
948	/*
949	* Allocate the requested number of large pages.
950	*/
951	for (iLargePage = 0; iLargePage < cLargePages; iLargePage++)
952	{
953	struct page *paPages = alloc_pages(fGfp, cLargePageOrder);
954	if (paPages)
955	{
956	size_t const iPageBase = iLargePage << cLargePageOrder;
957	size_t iPage = cPagesPerLarge;
958	while (iPage-- > 0)
959	pMemLnx->apPages[iPageBase + iPage] = &paPages[iPage];
960	}
961	else
962	{
963	/*Log(("rtR0MemObjNativeAllocLarge: cb=%#zx cPages=%#zx cLargePages=%#zx cLargePageOrder=%u cPagesPerLarge=%#zx iLargePage=%#zx -> failed!\n",
964	cb, cPages, cLargePages, cLargePageOrder, cPagesPerLarge, iLargePage, paPages));*/
965	while (iLargePage-- > 0)
966	__free_pages(pMemLnx->apPages[iLargePage << (cLargePageOrder - PAGE_SHIFT)], cLargePageOrder);
967	rtR0MemObjDelete(&pMemLnx->Core);
968	return VERR_NO_MEMORY;
969	}
970	}
971	*ppMem = &pMemLnx->Core;
972	return VINF_SUCCESS;
973	}
974	return VERR_NO_MEMORY;
975
976	#else
977	/*
978	* We don't call rtR0MemObjFallbackAllocLarge here as it can be a really
979	* bad idea to trigger the swap daemon and whatnot. So, just fail.
980	*/
981	RT_NOREF(ppMem, cb, cbLargePage, fFlags, pszTag);
982	return VERR_NOT_SUPPORTED;
983	#endif
984	}
985
986
987	DECLHIDDEN(int) rtR0MemObjNativeAllocLow(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, bool fExecutable, const char *pszTag)
988	{
989	IPRT_LINUX_SAVE_EFL_AC();
990	PRTR0MEMOBJLNX pMemLnx;
991	int rc;
992
993	/* Try to avoid GFP_DMA. GFM_DMA32 was introduced with Linux 2.6.15. */
994	#if (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
995	/* ZONE_DMA32: 0-4GB */
996	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_DMA32,
997	false /* non-contiguous */, fExecutable, VERR_NO_LOW_MEMORY, pszTag);
998	if (RT_FAILURE(rc))
999	#endif
1000	#ifdef RT_ARCH_AMD64
1001	/* ZONE_DMA: 0-16MB */
1002	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_DMA,
1003	false /* non-contiguous */, fExecutable, VERR_NO_LOW_MEMORY, pszTag);
1004	#else
1005	# ifdef CONFIG_X86_PAE
1006	# endif
1007	/* ZONE_NORMAL: 0-896MB */
1008	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_LOW, cb, PAGE_SIZE, GFP_USER,
1009	false /* non-contiguous */, fExecutable, VERR_NO_LOW_MEMORY, pszTag);
1010	#endif
1011	if (RT_SUCCESS(rc))
1012	{
1013	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
1014	if (RT_SUCCESS(rc))
1015	{
1016	*ppMem = &pMemLnx->Core;
1017	IPRT_LINUX_RESTORE_EFL_AC();
1018	return rc;
1019	}
1020
1021	rtR0MemObjLinuxFreePages(pMemLnx);
1022	rtR0MemObjDelete(&pMemLnx->Core);
1023	}
1024
1025	IPRT_LINUX_RESTORE_EFL_AC();
1026	return rc;
1027	}
1028
1029
1030	DECLHIDDEN(int) rtR0MemObjNativeAllocCont(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest,
1031	bool fExecutable, const char *pszTag)
1032	{
1033	IPRT_LINUX_SAVE_EFL_AC();
1034	PRTR0MEMOBJLNX pMemLnx;
1035	int rc;
1036	uint32_t idxZone;
1037
1038	/*
1039	* The last zone must be able to satisfy the PhysHighest requirement or there
1040	* will be no zone at all.
1041	*/
1042	if (g_aZones[RT_ELEMENTS(g_aZones) - 1].PhysHighest > PhysHighest)
1043	{
1044	IPRT_LINUX_RESTORE_EFL_AC();
1045	AssertMsgFailedReturn(("No zone can satisfy PhysHighest=%RHp!\n", PhysHighest),
1046	VERR_NO_CONT_MEMORY);
1047	}
1048
1049	/* Find the first zone matching our PhysHighest requirement. */
1050	idxZone = 0;
1051	for (;;)
1052	{
1053	if (g_aZones[idxZone].PhysHighest <= PhysHighest)
1054	break; /* We found a zone satisfying the requirement. */
1055	idxZone++;
1056	}
1057
1058	/* Now try to allocate pages from all the left zones until one succeeds. */
1059	for (;;)
1060	{
1061	rc = rtR0MemObjLinuxAllocPages(&pMemLnx, RTR0MEMOBJTYPE_CONT, cb, PAGE_SIZE, g_aZones[idxZone].fGfp,
1062	true /* contiguous */, fExecutable, VERR_NO_CONT_MEMORY, pszTag);
1063	idxZone++;
1064	if (RT_SUCCESS(rc) \|\| idxZone == RT_ELEMENTS(g_aZones))
1065	break;
1066	}
1067	if (RT_SUCCESS(rc))
1068	{
1069	rc = rtR0MemObjLinuxVMap(pMemLnx, fExecutable);
1070	if (RT_SUCCESS(rc))
1071	{
1072	#if defined(RT_STRICT)
1073	size_t iPage = pMemLnx->cPages;
1074	while (iPage-- > 0)
1075	Assert(page_to_phys(pMemLnx->apPages[iPage]) < PhysHighest);
1076	#endif
1077	pMemLnx->Core.u.Cont.Phys = page_to_phys(pMemLnx->apPages[0]);
1078	*ppMem = &pMemLnx->Core;
1079	IPRT_LINUX_RESTORE_EFL_AC();
1080	return rc;
1081	}
1082
1083	rtR0MemObjLinuxFreePages(pMemLnx);
1084	rtR0MemObjDelete(&pMemLnx->Core);
1085	}
1086
1087	IPRT_LINUX_RESTORE_EFL_AC();
1088	return rc;
1089	}
1090
1091
1092	/**
1093	* Worker for rtR0MemObjLinuxAllocPhysSub that tries one allocation strategy.
1094	*
1095	* @returns IPRT status code.
1096	* @param ppMemLnx Where to
1097	* @param enmType The object type.
1098	* @param cb The size of the allocation.
1099	* @param uAlignment The alignment of the physical memory.
1100	* Only valid for fContiguous == true, ignored otherwise.
1101	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
1102	* @param pszTag Allocation tag used for statistics and such.
1103	* @param fGfp The Linux GFP flags to use for the allocation.
1104	*/
1105	static int rtR0MemObjLinuxAllocPhysSub2(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType,
1106	size_t cb, size_t uAlignment, RTHCPHYS PhysHighest, const char *pszTag, gfp_t fGfp)
1107	{
1108	PRTR0MEMOBJLNX pMemLnx;
1109	int rc = rtR0MemObjLinuxAllocPages(&pMemLnx, enmType, cb, uAlignment, fGfp,
1110	enmType == RTR0MEMOBJTYPE_PHYS /* contiguous / non-contiguous */,
1111	false /fExecutable/, VERR_NO_PHYS_MEMORY, pszTag);
1112	if (RT_FAILURE(rc))
1113	return rc;
1114
1115	/*
1116	* Check the addresses if necessary. (Can be optimized a bit for PHYS.)
1117	*/
1118	if (PhysHighest != NIL_RTHCPHYS)
1119	{
1120	size_t iPage = pMemLnx->cPages;
1121	while (iPage-- > 0)
1122	if (page_to_phys(pMemLnx->apPages[iPage]) > PhysHighest)
1123	{
1124	rtR0MemObjLinuxFreePages(pMemLnx);
1125	rtR0MemObjDelete(&pMemLnx->Core);
1126	return VERR_NO_MEMORY;
1127	}
1128	}
1129
1130	/*
1131	* Complete the object.
1132	*/
1133	if (enmType == RTR0MEMOBJTYPE_PHYS)
1134	{
1135	pMemLnx->Core.u.Phys.PhysBase = page_to_phys(pMemLnx->apPages[0]);
1136	pMemLnx->Core.u.Phys.fAllocated = true;
1137	}
1138	*ppMem = &pMemLnx->Core;
1139	return rc;
1140	}
1141
1142
1143	/**
1144	* Worker for rtR0MemObjNativeAllocPhys and rtR0MemObjNativeAllocPhysNC.
1145	*
1146	* @returns IPRT status code.
1147	* @param ppMem Where to store the memory object pointer on success.
1148	* @param enmType The object type.
1149	* @param cb The size of the allocation.
1150	* @param uAlignment The alignment of the physical memory.
1151	* Only valid for enmType == RTR0MEMOBJTYPE_PHYS, ignored otherwise.
1152	* @param PhysHighest See rtR0MemObjNativeAllocPhys.
1153	* @param pszTag Allocation tag used for statistics and such.
1154	*/
1155	static int rtR0MemObjLinuxAllocPhysSub(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJTYPE enmType,
1156	size_t cb, size_t uAlignment, RTHCPHYS PhysHighest, const char *pszTag)
1157	{
1158	int rc;
1159	IPRT_LINUX_SAVE_EFL_AC();
1160
1161	/*
1162	* There are two clear cases and that's the <=16MB and anything-goes ones.
1163	* When the physical address limit is somewhere in-between those two we'll
1164	* just have to try, starting with HIGHUSER and working our way thru the
1165	* different types, hoping we'll get lucky.
1166	*
1167	* We should probably move this physical address restriction logic up to
1168	* the page alloc function as it would be more efficient there. But since
1169	* we don't expect this to be a performance issue just yet it can wait.
1170	*/
1171	if (PhysHighest == NIL_RTHCPHYS)
1172	/* ZONE_HIGHMEM: the whole physical memory */
1173	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_HIGHUSER);
1174	else if (PhysHighest <= _1M * 16)
1175	/* ZONE_DMA: 0-16MB */
1176	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_DMA);
1177	else
1178	{
1179	rc = VERR_NO_MEMORY;
1180	if (RT_FAILURE(rc))
1181	/* ZONE_HIGHMEM: the whole physical memory */
1182	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_HIGHUSER);
1183	if (RT_FAILURE(rc))
1184	/* ZONE_NORMAL: 0-896MB */
1185	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_USER);
1186	#ifdef GFP_DMA32
1187	if (RT_FAILURE(rc))
1188	/* ZONE_DMA32: 0-4GB */
1189	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_DMA32);
1190	#endif
1191	if (RT_FAILURE(rc))
1192	/* ZONE_DMA: 0-16MB */
1193	rc = rtR0MemObjLinuxAllocPhysSub2(ppMem, enmType, cb, uAlignment, PhysHighest, pszTag, GFP_DMA);
1194	}
1195	IPRT_LINUX_RESTORE_EFL_AC();
1196	return rc;
1197	}
1198
1199
1200	/**
1201	* Translates a kernel virtual address to a linux page structure by walking the
1202	* page tables.
1203	*
1204	* @note We do assume that the page tables will not change as we are walking
1205	* them. This assumption is rather forced by the fact that I could not
1206	* immediately see any way of preventing this from happening. So, we
1207	* take some extra care when accessing them.
1208	*
1209	* Because of this, we don't want to use this function on memory where
1210	* attribute changes to nearby pages is likely to cause large pages to
1211	* be used or split up. So, don't use this for the linear mapping of
1212	* physical memory.
1213	*
1214	* @returns Pointer to the page structur or NULL if it could not be found.
1215	* @param pv The kernel virtual address.
1216	*/
1217	RTDECL(struct page ) rtR0MemObjLinuxVirtToPage(void pv)
1218	{
1219	#if defined(RT_ARCH_AMD64) \|\| defined(RT_ARCH_X86)
1220	unsigned long ulAddr = (unsigned long)pv;
1221	unsigned long pfn;
1222	struct page *pPage;
1223	pte_t *pEntry;
1224	union
1225	{
1226	pgd_t Global;
1227	# if RTLNX_VER_MIN(4,12,0)
1228	p4d_t Four;
1229	# endif
1230	# if RTLNX_VER_MIN(2,6,11)
1231	pud_t Upper;
1232	# endif
1233	pmd_t Middle;
1234	pte_t Entry;
1235	} u;
1236
1237	/* Should this happen in a situation this code will be called in? And if
1238	* so, can it change under our feet? See also
1239	* "Documentation/vm/active_mm.txt" in the kernel sources. */
1240	if (RT_UNLIKELY(!current->active_mm))
1241	return NULL;
1242	u.Global = *pgd_offset(current->active_mm, ulAddr);
1243	if (RT_UNLIKELY(pgd_none(u.Global)))
1244	return NULL;
1245	# if RTLNX_VER_MIN(2,6,11)
1246	# if RTLNX_VER_MIN(4,12,0)
1247	u.Four = *p4d_offset(&u.Global, ulAddr);
1248	if (RT_UNLIKELY(p4d_none(u.Four)))
1249	return NULL;
1250	# if RTLNX_VER_MIN(5,6,0)
1251	if (p4d_leaf(u.Four))
1252	# else
1253	if (p4d_large(u.Four))
1254	# endif
1255	{
1256	pPage = p4d_page(u.Four);
1257	AssertReturn(pPage, NULL);
1258	pfn = page_to_pfn(pPage); /* doing the safe way... */
1259	AssertCompile(P4D_SHIFT - PAGE_SHIFT < 31);
1260	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (P4D_SHIFT - PAGE_SHIFT)) - 1);
1261	return pfn_to_page(pfn);
1262	}
1263	u.Upper = *pud_offset(&u.Four, ulAddr);
1264	# else /* < 4.12 */
1265	u.Upper = *pud_offset(&u.Global, ulAddr);
1266	# endif /* < 4.12 */
1267	if (RT_UNLIKELY(pud_none(u.Upper)))
1268	return NULL;
1269	# if RTLNX_VER_MIN(2,6,25)
1270	# if RTLNX_VER_MIN(5,6,0)
1271	if (pud_leaf(u.Upper))
1272	# else
1273	if (pud_large(u.Upper))
1274	# endif
1275	{
1276	pPage = pud_page(u.Upper);
1277	AssertReturn(pPage, NULL);
1278	pfn = page_to_pfn(pPage); /* doing the safe way... */
1279	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (PUD_SHIFT - PAGE_SHIFT)) - 1);
1280	return pfn_to_page(pfn);
1281	}
1282	# endif
1283	u.Middle = *pmd_offset(&u.Upper, ulAddr);
1284	# else /* < 2.6.11 */
1285	u.Middle = *pmd_offset(&u.Global, ulAddr);
1286	# endif /* < 2.6.11 */
1287	if (RT_UNLIKELY(pmd_none(u.Middle)))
1288	return NULL;
1289	# if RTLNX_VER_MIN(2,6,0)
1290	# if RTLNX_VER_MIN(5,6,0)
1291	if (pmd_leaf(u.Middle))
1292	# else
1293	if (pmd_large(u.Middle))
1294	# endif
1295	{
1296	pPage = pmd_page(u.Middle);
1297	AssertReturn(pPage, NULL);
1298	pfn = page_to_pfn(pPage); /* doing the safe way... */
1299	pfn += (ulAddr >> PAGE_SHIFT) & ((UINT32_C(1) << (PMD_SHIFT - PAGE_SHIFT)) - 1);
1300	return pfn_to_page(pfn);
1301	}
1302	# endif
1303
1304	# if RTLNX_VER_MIN(6,5,0) \|\| RTLNX_RHEL_RANGE(9,4, 9,99)
1305	pEntry = __pte_map(&u.Middle, ulAddr);
1306	# elif RTLNX_VER_MIN(2,5,5) \|\| defined(pte_offset_map) /* As usual, RHEL 3 had pte_offset_map earlier. */
1307	pEntry = pte_offset_map(&u.Middle, ulAddr);
1308	# else
1309	pEntry = pte_offset(&u.Middle, ulAddr);
1310	# endif
1311	if (RT_UNLIKELY(!pEntry))
1312	return NULL;
1313	u.Entry = *pEntry;
1314	# if RTLNX_VER_MIN(2,5,5) \|\| defined(pte_offset_map)
1315	pte_unmap(pEntry);
1316	# endif
1317
1318	if (RT_UNLIKELY(!pte_present(u.Entry)))
1319	return NULL;
1320	return pte_page(u.Entry);
1321	#else /* !defined(RT_ARCH_AMD64) && !defined(RT_ARCH_X86) */
1322
1323	if (is_vmalloc_addr(pv))
1324	return vmalloc_to_page(pv);
1325
1326	return virt_to_page(pv);
1327	#endif
1328	}
1329	RT_EXPORT_SYMBOL(rtR0MemObjLinuxVirtToPage);
1330
1331
1332	DECLHIDDEN(int) rtR0MemObjNativeAllocPhys(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest, size_t uAlignment,
1333	const char *pszTag)
1334	{
1335	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS, cb, uAlignment, PhysHighest, pszTag);
1336	}
1337
1338
1339	DECLHIDDEN(int) rtR0MemObjNativeAllocPhysNC(PPRTR0MEMOBJINTERNAL ppMem, size_t cb, RTHCPHYS PhysHighest, const char *pszTag)
1340	{
1341	return rtR0MemObjLinuxAllocPhysSub(ppMem, RTR0MEMOBJTYPE_PHYS_NC, cb, PAGE_SIZE, PhysHighest, pszTag);
1342	}
1343
1344
1345	DECLHIDDEN(int) rtR0MemObjNativeEnterPhys(PPRTR0MEMOBJINTERNAL ppMem, RTHCPHYS Phys, size_t cb, uint32_t uCachePolicy,
1346	const char *pszTag)
1347	{
1348	IPRT_LINUX_SAVE_EFL_AC();
1349
1350	/*
1351	* All we need to do here is to validate that we can use
1352	* ioremap on the specified address (32/64-bit dma_addr_t).
1353	*/
1354	PRTR0MEMOBJLNX pMemLnx;
1355	dma_addr_t PhysAddr = Phys;
1356	AssertMsgReturn(PhysAddr == Phys, ("%#llx\n", (unsigned long long)Phys), VERR_ADDRESS_TOO_BIG);
1357
1358	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_PHYS, NULL, cb, pszTag);
1359	if (!pMemLnx)
1360	{
1361	IPRT_LINUX_RESTORE_EFL_AC();
1362	return VERR_NO_MEMORY;
1363	}
1364
1365	pMemLnx->Core.u.Phys.PhysBase = PhysAddr;
1366	pMemLnx->Core.u.Phys.fAllocated = false;
1367	pMemLnx->Core.u.Phys.uCachePolicy = uCachePolicy;
1368	Assert(!pMemLnx->cPages);
1369	*ppMem = &pMemLnx->Core;
1370	IPRT_LINUX_RESTORE_EFL_AC();
1371	return VINF_SUCCESS;
1372	}
1373
1374	/* openSUSE Leap 42.3 detection :-/ */
1375	#if RTLNX_VER_RANGE(4,4,0, 4,6,0) && defined(FAULT_FLAG_REMOTE)
1376	# define GET_USER_PAGES_API KERNEL_VERSION(4, 10, 0) /* no typo! */
1377	#else
1378	# define GET_USER_PAGES_API LINUX_VERSION_CODE
1379	#endif
1380
1381	DECLHIDDEN(int) rtR0MemObjNativeLockUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3Ptr, size_t cb, uint32_t fAccess,
1382	RTR0PROCESS R0Process, const char *pszTag)
1383	{
1384	IPRT_LINUX_SAVE_EFL_AC();
1385	const int cPages = cb >> PAGE_SHIFT;
1386	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1387	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1388	struct vm_area_struct **papVMAs;
1389	# endif
1390	PRTR0MEMOBJLNX pMemLnx;
1391	int rc = VERR_NO_MEMORY;
1392	int const fWrite = fAccess & RTMEM_PROT_WRITE ? 1 : 0;
1393
1394	/*
1395	* Check for valid task and size overflows.
1396	*/
1397	if (!pTask)
1398	return VERR_NOT_SUPPORTED;
1399	if (((size_t)cPages << PAGE_SHIFT) != cb)
1400	return VERR_OUT_OF_RANGE;
1401
1402	/*
1403	* Allocate the memory object and a temporary buffer for the VMAs.
1404	*/
1405	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_UOFFSETOF_DYN(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK,
1406	(void *)R3Ptr, cb, pszTag);
1407	if (!pMemLnx)
1408	{
1409	IPRT_LINUX_RESTORE_EFL_AC();
1410	return VERR_NO_MEMORY;
1411	}
1412
1413	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1414	papVMAs = (struct vm_area_struct *)RTMemAlloc(sizeof(papVMAs) * cPages);
1415	if (papVMAs)
1416	{
1417	# endif
1418	LNX_MM_DOWN_READ(pTask->mm);
1419
1420	/*
1421	* Get user pages.
1422	*/
1423	/** @todo r=bird: Should we not force read access too? */
1424	#if GET_USER_PAGES_API >= KERNEL_VERSION(4, 6, 0)
1425	if (R0Process == RTR0ProcHandleSelf())
1426	rc = get_user_pages(R3Ptr, /* Where from. */
1427	cPages, /* How many pages. */
1428	# if GET_USER_PAGES_API >= KERNEL_VERSION(4, 9, 0)
1429	fWrite ? FOLL_WRITE \| /* Write to memory. */
1430	FOLL_FORCE /* force write access. */
1431	: 0, /* Write to memory. */
1432	# else
1433	fWrite, /* Write to memory. */
1434	fWrite, /* force write access. */
1435	# endif
1436	&pMemLnx->apPages[0] /* Page array. */
1437	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_SUSE_MAJ_PREREQ(15, 6) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1438	, papVMAs /* vmas */
1439	# endif
1440	);
1441	/*
1442	* Actually this should not happen at the moment as call this function
1443	* only for our own process.
1444	*/
1445	else
1446	rc = get_user_pages_remote(
1447	# if GET_USER_PAGES_API < KERNEL_VERSION(5, 9, 0)
1448	pTask, /* Task for fault accounting. */
1449	# endif
1450	pTask->mm, /* Whose pages. */
1451	R3Ptr, /* Where from. */
1452	cPages, /* How many pages. */
1453	# if GET_USER_PAGES_API >= KERNEL_VERSION(4, 9, 0)
1454	fWrite ? FOLL_WRITE \| /* Write to memory. */
1455	FOLL_FORCE /* force write access. */
1456	: 0, /* Write to memory. */
1457	# else
1458	fWrite, /* Write to memory. */
1459	fWrite, /* force write access. */
1460	# endif
1461	&pMemLnx->apPages[0] /* Page array. */
1462	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1463	, papVMAs /* vmas */
1464	# endif
1465	# if GET_USER_PAGES_API >= KERNEL_VERSION(4, 10, 0)
1466	, NULL /* locked */
1467	# endif
1468	);
1469	#else /* GET_USER_PAGES_API < KERNEL_VERSION(4, 6, 0) */
1470	rc = get_user_pages(pTask, /* Task for fault accounting. */
1471	pTask->mm, /* Whose pages. */
1472	R3Ptr, /* Where from. */
1473	cPages, /* How many pages. */
1474	/* The get_user_pages API change was back-ported to 4.4.168. */
1475	# if RTLNX_VER_RANGE(4,4,168, 4,5,0)
1476	fWrite ? FOLL_WRITE \| /* Write to memory. */
1477	FOLL_FORCE /* force write access. */
1478	: 0, /* Write to memory. */
1479	# else
1480	fWrite, /* Write to memory. */
1481	fWrite, /* force write access. */
1482	# endif
1483	&pMemLnx->apPages[0], /* Page array. */
1484	papVMAs /* vmas */
1485	);
1486	#endif /* GET_USER_PAGES_API < KERNEL_VERSION(4, 6, 0) */
1487	if (rc == cPages)
1488	{
1489	/*
1490	* Flush dcache (required?), protect against fork and _really_ pin the page
1491	* table entries. get_user_pages() will protect against swapping out the
1492	* pages but it will NOT protect against removing page table entries. This
1493	* can be achieved with
1494	* - using mlock / mmap(..., MAP_LOCKED, ...) from userland. This requires
1495	* an appropriate limit set up with setrlimit(..., RLIMIT_MEMLOCK, ...).
1496	* Usual Linux distributions support only a limited size of locked pages
1497	* (e.g. 32KB).
1498	* - setting the PageReserved bit (as we do in rtR0MemObjLinuxAllocPages()
1499	* or by
1500	* - setting the VM_LOCKED flag. This is the same as doing mlock() without
1501	* a range check.
1502	*/
1503	/** @todo The Linux fork() protection will require more work if this API
1504	* is to be used for anything but locking VM pages. */
1505	while (rc-- > 0)
1506	{
1507	flush_dcache_page(pMemLnx->apPages[rc]);
1508	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_SUSE_MAJ_PREREQ(15, 6) && !RTLNX_RHEL_RANGE(9,5, 9,99)
1509	# if RTLNX_VER_MIN(6,3,0)
1510	vm_flags_set(papVMAs[rc], VM_DONTCOPY \| VM_LOCKED);
1511	# else
1512	papVMAs[rc]->vm_flags \|= VM_DONTCOPY \| VM_LOCKED;
1513	# endif
1514	# endif
1515	}
1516
1517	LNX_MM_UP_READ(pTask->mm);
1518
1519	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1520	RTMemFree(papVMAs);
1521	# endif
1522
1523	pMemLnx->Core.u.Lock.R0Process = R0Process;
1524	pMemLnx->cPages = cPages;
1525	Assert(!pMemLnx->fMappedToRing0);
1526	*ppMem = &pMemLnx->Core;
1527
1528	IPRT_LINUX_RESTORE_EFL_AC();
1529	return VINF_SUCCESS;
1530	}
1531
1532	/*
1533	* Failed - we need to unlock any pages that we succeeded to lock.
1534	*/
1535	while (rc-- > 0)
1536	{
1537	if (!PageReserved(pMemLnx->apPages[rc]))
1538	SetPageDirty(pMemLnx->apPages[rc]);
1539	#if RTLNX_VER_MIN(4,6,0)
1540	put_page(pMemLnx->apPages[rc]);
1541	#else
1542	page_cache_release(pMemLnx->apPages[rc]);
1543	#endif
1544	}
1545
1546	LNX_MM_UP_READ(pTask->mm);
1547
1548	rc = VERR_LOCK_FAILED;
1549
1550	# if GET_USER_PAGES_API < KERNEL_VERSION(6, 5, 0) && !RTLNX_RHEL_RANGE(9,6, 9,99)
1551	RTMemFree(papVMAs);
1552	}
1553	# endif
1554
1555	rtR0MemObjDelete(&pMemLnx->Core);
1556	IPRT_LINUX_RESTORE_EFL_AC();
1557	return rc;
1558	}
1559
1560
1561	DECLHIDDEN(int) rtR0MemObjNativeLockKernel(PPRTR0MEMOBJINTERNAL ppMem, void pv, size_t cb, uint32_t fAccess, const char pszTag)
1562	{
1563	IPRT_LINUX_SAVE_EFL_AC();
1564	void pvLast = (uint8_t )pv + cb - 1;
1565	size_t const cPages = cb >> PAGE_SHIFT;
1566	PRTR0MEMOBJLNX pMemLnx;
1567	bool fLinearMapping;
1568	int rc;
1569	uint8_t *pbPage;
1570	size_t iPage;
1571	NOREF(fAccess);
1572
1573	if ( !RTR0MemKernelIsValidAddr(pv)
1574	\|\| !RTR0MemKernelIsValidAddr(pv + cb))
1575	return VERR_INVALID_PARAMETER;
1576
1577	/*
1578	* The lower part of the kernel memory has a linear mapping between
1579	* physical and virtual addresses. So we take a short cut here. This is
1580	* assumed to be the cleanest way to handle those addresses (and the code
1581	* is well tested, though the test for determining it is not very nice).
1582	* If we ever decide it isn't we can still remove it.
1583	*/
1584	#if 0
1585	fLinearMapping = (unsigned long)pvLast < VMALLOC_START;
1586	#else
1587	fLinearMapping = (unsigned long)pv >= (unsigned long)__va(0)
1588	&& (unsigned long)pvLast < (unsigned long)high_memory;
1589	#endif
1590
1591	/*
1592	* Allocate the memory object.
1593	*/
1594	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(RT_UOFFSETOF_DYN(RTR0MEMOBJLNX, apPages[cPages]), RTR0MEMOBJTYPE_LOCK,
1595	pv, cb, pszTag);
1596	if (!pMemLnx)
1597	{
1598	IPRT_LINUX_RESTORE_EFL_AC();
1599	return VERR_NO_MEMORY;
1600	}
1601
1602	/*
1603	* Gather the pages.
1604	* We ASSUME all kernel pages are non-swappable and non-movable.
1605	*/
1606	rc = VINF_SUCCESS;
1607	pbPage = (uint8_t *)pvLast;
1608	iPage = cPages;
1609	if (!fLinearMapping)
1610	{
1611	while (iPage-- > 0)
1612	{
1613	struct page *pPage = rtR0MemObjLinuxVirtToPage(pbPage);
1614	if (RT_UNLIKELY(!pPage))
1615	{
1616	rc = VERR_LOCK_FAILED;
1617	break;
1618	}
1619	pMemLnx->apPages[iPage] = pPage;
1620	pbPage -= PAGE_SIZE;
1621	}
1622	}
1623	else
1624	{
1625	while (iPage-- > 0)
1626	{
1627	pMemLnx->apPages[iPage] = virt_to_page(pbPage);
1628	pbPage -= PAGE_SIZE;
1629	}
1630	}
1631	if (RT_SUCCESS(rc))
1632	{
1633	/*
1634	* Complete the memory object and return.
1635	*/
1636	pMemLnx->Core.u.Lock.R0Process = NIL_RTR0PROCESS;
1637	pMemLnx->cPages = cPages;
1638	Assert(!pMemLnx->fMappedToRing0);
1639	*ppMem = &pMemLnx->Core;
1640
1641	IPRT_LINUX_RESTORE_EFL_AC();
1642	return VINF_SUCCESS;
1643	}
1644
1645	rtR0MemObjDelete(&pMemLnx->Core);
1646	IPRT_LINUX_RESTORE_EFL_AC();
1647	return rc;
1648	}
1649
1650
1651	DECLHIDDEN(int) rtR0MemObjNativeReserveKernel(PPRTR0MEMOBJINTERNAL ppMem, void *pvFixed, size_t cb, size_t uAlignment,
1652	const char *pszTag)
1653	{
1654	#if RTLNX_VER_MIN(2,4,22)
1655	IPRT_LINUX_SAVE_EFL_AC();
1656	const size_t cPages = cb >> PAGE_SHIFT;
1657	struct page *pDummyPage;
1658	struct page **papPages;
1659
1660	/* check for unsupported stuff. */
1661	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
1662	if (uAlignment > PAGE_SIZE)
1663	return VERR_NOT_SUPPORTED;
1664
1665	/*
1666	* Allocate a dummy page and create a page pointer array for vmap such that
1667	* the dummy page is mapped all over the reserved area.
1668	*/
1669	pDummyPage = alloc_page(GFP_HIGHUSER \| __GFP_NOWARN);
1670	if (pDummyPage)
1671	{
1672	papPages = RTMemAlloc(sizeof(papPages) cPages);
1673	if (papPages)
1674	{
1675	void *pv;
1676	size_t iPage = cPages;
1677	while (iPage-- > 0)
1678	papPages[iPage] = pDummyPage;
1679	# ifdef VM_MAP
1680	pv = vmap(papPages, cPages, VM_MAP, PAGE_KERNEL_RO);
1681	# else
1682	pv = vmap(papPages, cPages, VM_ALLOC, PAGE_KERNEL_RO);
1683	# endif
1684	RTMemFree(papPages);
1685	if (pv)
1686	{
1687	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb, pszTag);
1688	if (pMemLnx)
1689	{
1690	pMemLnx->Core.u.ResVirt.R0Process = NIL_RTR0PROCESS;
1691	pMemLnx->cPages = 1;
1692	pMemLnx->apPages[0] = pDummyPage;
1693	*ppMem = &pMemLnx->Core;
1694	IPRT_LINUX_RESTORE_EFL_AC();
1695	return VINF_SUCCESS;
1696	}
1697	vunmap(pv);
1698	}
1699	}
1700	__free_page(pDummyPage);
1701	}
1702	IPRT_LINUX_RESTORE_EFL_AC();
1703	return VERR_NO_MEMORY;
1704
1705	#else /* < 2.4.22 */
1706	/*
1707	* Could probably use ioremap here, but the caller is in a better position than us
1708	* to select some safe physical memory.
1709	*/
1710	return VERR_NOT_SUPPORTED;
1711	#endif
1712	}
1713
1714
1715	DECLHIDDEN(int) rtR0MemObjNativeReserveUser(PPRTR0MEMOBJINTERNAL ppMem, RTR3PTR R3PtrFixed, size_t cb, size_t uAlignment,
1716	RTR0PROCESS R0Process, const char *pszTag)
1717	{
1718	IPRT_LINUX_SAVE_EFL_AC();
1719	PRTR0MEMOBJLNX pMemLnx;
1720	void *pv;
1721	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1722	if (!pTask)
1723	return VERR_NOT_SUPPORTED;
1724
1725	/*
1726	* Check that the specified alignment is supported.
1727	*/
1728	if (uAlignment > PAGE_SIZE)
1729	return VERR_NOT_SUPPORTED;
1730
1731	/*
1732	* Let rtR0MemObjLinuxDoMmap do the difficult bits.
1733	*/
1734	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, cb, uAlignment, pTask, RTMEM_PROT_NONE);
1735	if (pv == (void *)-1)
1736	{
1737	IPRT_LINUX_RESTORE_EFL_AC();
1738	return VERR_NO_MEMORY;
1739	}
1740
1741	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_RES_VIRT, pv, cb, pszTag);
1742	if (!pMemLnx)
1743	{
1744	rtR0MemObjLinuxDoMunmap(pv, cb, pTask);
1745	IPRT_LINUX_RESTORE_EFL_AC();
1746	return VERR_NO_MEMORY;
1747	}
1748
1749	pMemLnx->Core.u.ResVirt.R0Process = R0Process;
1750	*ppMem = &pMemLnx->Core;
1751	IPRT_LINUX_RESTORE_EFL_AC();
1752	return VINF_SUCCESS;
1753	}
1754
1755
1756	DECLHIDDEN(int) rtR0MemObjNativeMapKernel(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap, void *pvFixed, size_t uAlignment,
1757	unsigned fProt, size_t offSub, size_t cbSub, const char *pszTag)
1758	{
1759	int rc = VERR_NO_MEMORY;
1760	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1761	PRTR0MEMOBJLNX pMemLnx;
1762	IPRT_LINUX_SAVE_EFL_AC();
1763
1764	/* Fail if requested to do something we can't. */
1765	AssertMsgReturn(pvFixed == (void *)-1, ("%p\n", pvFixed), VERR_NOT_SUPPORTED);
1766	if (uAlignment > PAGE_SIZE)
1767	return VERR_NOT_SUPPORTED;
1768
1769	/*
1770	* Create the IPRT memory object.
1771	*/
1772	if (!cbSub)
1773	cbSub = pMemLnxToMap->Core.cb - offSub;
1774	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, cbSub, pszTag);
1775	if (pMemLnx)
1776	{
1777	if (pMemLnxToMap->cPages)
1778	{
1779	#if RTLNX_VER_MIN(2,4,22)
1780	/*
1781	* Use vmap - 2.4.22 and later.
1782	*/
1783	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, true /* kernel */);
1784	/** @todo We don't really care too much for EXEC here... 5.8 always adds NX. */
1785	Assert(((offSub + cbSub) >> PAGE_SHIFT) <= pMemLnxToMap->cPages);
1786	# ifdef VM_MAP
1787	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[offSub >> PAGE_SHIFT], cbSub >> PAGE_SHIFT, VM_MAP, fPg);
1788	# else
1789	pMemLnx->Core.pv = vmap(&pMemLnxToMap->apPages[offSub >> PAGE_SHIFT], cbSub >> PAGE_SHIFT, VM_ALLOC, fPg);
1790	# endif
1791	if (pMemLnx->Core.pv)
1792	{
1793	pMemLnx->fMappedToRing0 = true;
1794	rc = VINF_SUCCESS;
1795	}
1796	else
1797	rc = VERR_MAP_FAILED;
1798
1799	#else /* < 2.4.22 */
1800	/*
1801	* Only option here is to share mappings if possible and forget about fProt.
1802	*/
1803	if (rtR0MemObjIsRing3(pMemToMap))
1804	rc = VERR_NOT_SUPPORTED;
1805	else
1806	{
1807	rc = VINF_SUCCESS;
1808	if (!pMemLnxToMap->Core.pv)
1809	rc = rtR0MemObjLinuxVMap(pMemLnxToMap, !!(fProt & RTMEM_PROT_EXEC));
1810	if (RT_SUCCESS(rc))
1811	{
1812	Assert(pMemLnxToMap->Core.pv);
1813	pMemLnx->Core.pv = (uint8_t *)pMemLnxToMap->Core.pv + offSub;
1814	}
1815	}
1816	#endif
1817	}
1818	else
1819	{
1820	/*
1821	* MMIO / physical memory.
1822	*/
1823	Assert(pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS && !pMemLnxToMap->Core.u.Phys.fAllocated);
1824	#if RTLNX_VER_MIN(2,6,25)
1825	/*
1826	* ioremap() defaults to no caching since the 2.6 kernels.
1827	* ioremap_nocache() has been removed finally in 5.6-rc1.
1828	*/
1829	pMemLnx->Core.pv = pMemLnxToMap->Core.u.Phys.uCachePolicy == RTMEM_CACHE_POLICY_MMIO
1830	? ioremap(pMemLnxToMap->Core.u.Phys.PhysBase + offSub, cbSub)
1831	: ioremap_cache(pMemLnxToMap->Core.u.Phys.PhysBase + offSub, cbSub);
1832	#else /* KERNEL_VERSION < 2.6.25 */
1833	pMemLnx->Core.pv = pMemLnxToMap->Core.u.Phys.uCachePolicy == RTMEM_CACHE_POLICY_MMIO
1834	? ioremap_nocache(pMemLnxToMap->Core.u.Phys.PhysBase + offSub, cbSub)
1835	: ioremap(pMemLnxToMap->Core.u.Phys.PhysBase + offSub, cbSub);
1836	#endif /* KERNEL_VERSION < 2.6.25 */
1837	if (pMemLnx->Core.pv)
1838	{
1839	/** @todo fix protection. */
1840	rc = VINF_SUCCESS;
1841	}
1842	}
1843	if (RT_SUCCESS(rc))
1844	{
1845	pMemLnx->Core.u.Mapping.R0Process = NIL_RTR0PROCESS;
1846	*ppMem = &pMemLnx->Core;
1847	IPRT_LINUX_RESTORE_EFL_AC();
1848	return VINF_SUCCESS;
1849	}
1850	rtR0MemObjDelete(&pMemLnx->Core);
1851	}
1852
1853	IPRT_LINUX_RESTORE_EFL_AC();
1854	return rc;
1855	}
1856
1857
1858	#ifdef VBOX_USE_PAE_HACK
1859	/**
1860	* Replace the PFN of a PTE with the address of the actual page.
1861	*
1862	* The caller maps a reserved dummy page at the address with the desired access
1863	* and flags.
1864	*
1865	* This hack is required for older Linux kernels which don't provide
1866	* remap_pfn_range().
1867	*
1868	* @returns 0 on success, -ENOMEM on failure.
1869	* @param mm The memory context.
1870	* @param ulAddr The mapping address.
1871	* @param Phys The physical address of the page to map.
1872	*/
1873	static int rtR0MemObjLinuxFixPte(struct mm_struct *mm, unsigned long ulAddr, RTHCPHYS Phys)
1874	{
1875	int rc = -ENOMEM;
1876	pgd_t *pgd;
1877
1878	spin_lock(&mm->page_table_lock);
1879
1880	pgd = pgd_offset(mm, ulAddr);
1881	if (!pgd_none(pgd) && !pgd_bad(pgd))
1882	{
1883	pmd_t *pmd = pmd_offset(pgd, ulAddr);
1884	if (!pmd_none(*pmd))
1885	{
1886	pte_t *ptep = pte_offset_map(pmd, ulAddr);
1887	if (ptep)
1888	{
1889	pte_t pte = *ptep;
1890	pte.pte_high &= 0xfff00000;
1891	pte.pte_high \|= ((Phys >> 32) & 0x000fffff);
1892	pte.pte_low &= 0x00000fff;
1893	pte.pte_low \|= (Phys & 0xfffff000);
1894	set_pte(ptep, pte);
1895	pte_unmap(ptep);
1896	rc = 0;
1897	}
1898	}
1899	}
1900
1901	spin_unlock(&mm->page_table_lock);
1902	return rc;
1903	}
1904	#endif /* VBOX_USE_PAE_HACK */
1905
1906
1907	DECLHIDDEN(int) rtR0MemObjNativeMapUser(PPRTR0MEMOBJINTERNAL ppMem, RTR0MEMOBJ pMemToMap, RTR3PTR R3PtrFixed, size_t uAlignment,
1908	unsigned fProt, RTR0PROCESS R0Process, size_t offSub, size_t cbSub, const char *pszTag)
1909	{
1910	struct task_struct *pTask = rtR0ProcessToLinuxTask(R0Process);
1911	PRTR0MEMOBJLNX pMemLnxToMap = (PRTR0MEMOBJLNX)pMemToMap;
1912	int rc = VERR_NO_MEMORY;
1913	PRTR0MEMOBJLNX pMemLnx;
1914	#ifdef VBOX_USE_PAE_HACK
1915	struct page *pDummyPage;
1916	RTHCPHYS DummyPhys;
1917	#endif
1918	IPRT_LINUX_SAVE_EFL_AC();
1919
1920	/*
1921	* Check for restrictions.
1922	*/
1923	if (!pTask)
1924	return VERR_NOT_SUPPORTED;
1925	if (uAlignment > PAGE_SIZE)
1926	return VERR_NOT_SUPPORTED;
1927
1928	#ifdef VBOX_USE_PAE_HACK
1929	/*
1930	* Allocate a dummy page for use when mapping the memory.
1931	*/
1932	pDummyPage = alloc_page(GFP_USER \| __GFP_NOWARN);
1933	if (!pDummyPage)
1934	{
1935	IPRT_LINUX_RESTORE_EFL_AC();
1936	return VERR_NO_MEMORY;
1937	}
1938	SetPageReserved(pDummyPage);
1939	DummyPhys = page_to_phys(pDummyPage);
1940	#endif
1941
1942	/*
1943	* Create the IPRT memory object.
1944	*/
1945	Assert(!offSub \|\| cbSub);
1946	if (cbSub == 0)
1947	cbSub = pMemLnxToMap->Core.cb;
1948	pMemLnx = (PRTR0MEMOBJLNX)rtR0MemObjNew(sizeof(*pMemLnx), RTR0MEMOBJTYPE_MAPPING, NULL, cbSub, pszTag);
1949	if (pMemLnx)
1950	{
1951	/*
1952	* Allocate user space mapping.
1953	*/
1954	void *pv;
1955	pv = rtR0MemObjLinuxDoMmap(R3PtrFixed, cbSub, uAlignment, pTask, fProt);
1956	if (pv != (void *)-1)
1957	{
1958	/*
1959	* Map page by page into the mmap area.
1960	* This is generic, paranoid and not very efficient.
1961	*/
1962	pgprot_t fPg = rtR0MemObjLinuxConvertProt(fProt, false /* user */);
1963	unsigned long ulAddrCur = (unsigned long)pv;
1964	const size_t cPages = (offSub + cbSub) >> PAGE_SHIFT;
1965	size_t iPage;
1966
1967	LNX_MM_DOWN_WRITE(pTask->mm);
1968
1969	rc = VINF_SUCCESS;
1970	if (pMemLnxToMap->cPages)
1971	{
1972	for (iPage = offSub >> PAGE_SHIFT; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE)
1973	{
1974	#if RTLNX_VER_MAX(2,6,11)
1975	RTHCPHYS Phys = page_to_phys(pMemLnxToMap->apPages[iPage]);
1976	#endif
1977	#if RTLNX_VER_MIN(2,6,0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
1978	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
1979	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
1980	#endif
1981	#if RTLNX_VER_MAX(2,6,0) && defined(RT_ARCH_X86)
1982	/* remap_page_range() limitation on x86 */
1983	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
1984	#endif
1985
1986	#if defined(VBOX_USE_INSERT_PAGE) && RTLNX_VER_MIN(2,6,22)
1987	rc = vm_insert_page(vma, ulAddrCur, pMemLnxToMap->apPages[iPage]);
1988	/* Thes flags help making 100% sure some bad stuff wont happen (swap, core, ++).
1989	* See remap_pfn_range() in mm/memory.c */
1990
1991	#if RTLNX_VER_MIN(6,3,0) \|\| RTLNX_RHEL_RANGE(9,5, 9,99)
1992	vm_flags_set(vma, VM_DONTEXPAND \| VM_DONTDUMP);
1993	#elif RTLNX_VER_MIN(3,7,0)
1994	vma->vm_flags \|= VM_DONTEXPAND \| VM_DONTDUMP;
1995	#else
1996	vma->vm_flags \|= VM_RESERVED;
1997	#endif
1998	#elif RTLNX_VER_MIN(2,6,11)
1999	rc = remap_pfn_range(vma, ulAddrCur, page_to_pfn(pMemLnxToMap->apPages[iPage]), PAGE_SIZE, fPg);
2000	#elif defined(VBOX_USE_PAE_HACK)
2001	rc = remap_page_range(vma, ulAddrCur, DummyPhys, PAGE_SIZE, fPg);
2002	if (!rc)
2003	rc = rtR0MemObjLinuxFixPte(pTask->mm, ulAddrCur, Phys);
2004	#elif RTLNX_VER_MIN(2,6,0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
2005	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
2006	#else /* 2.4 */
2007	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
2008	#endif
2009	if (rc)
2010	{
2011	rc = VERR_NO_MEMORY;
2012	break;
2013	}
2014	}
2015	}
2016	else
2017	{
2018	RTHCPHYS Phys;
2019	if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_PHYS)
2020	Phys = pMemLnxToMap->Core.u.Phys.PhysBase;
2021	else if (pMemLnxToMap->Core.enmType == RTR0MEMOBJTYPE_CONT)
2022	Phys = pMemLnxToMap->Core.u.Cont.Phys;
2023	else
2024	{
2025	AssertMsgFailed(("%d\n", pMemLnxToMap->Core.enmType));
2026	Phys = NIL_RTHCPHYS;
2027	}
2028	if (Phys != NIL_RTHCPHYS)
2029	{
2030	for (iPage = offSub >> PAGE_SHIFT; iPage < cPages; iPage++, ulAddrCur += PAGE_SIZE, Phys += PAGE_SIZE)
2031	{
2032	#if RTLNX_VER_MIN(2,6,0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
2033	struct vm_area_struct vma = find_vma(pTask->mm, ulAddrCur); / this is probably the same for all the pages... */
2034	AssertBreakStmt(vma, rc = VERR_INTERNAL_ERROR);
2035	#endif
2036	#if RTLNX_VER_MAX(2,6,0) && defined(RT_ARCH_X86)
2037	/* remap_page_range() limitation on x86 */
2038	AssertBreakStmt(Phys < _4G, rc = VERR_NO_MEMORY);
2039	#endif
2040
2041	#if RTLNX_VER_MIN(2,6,11)
2042	rc = remap_pfn_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
2043	#elif defined(VBOX_USE_PAE_HACK)
2044	rc = remap_page_range(vma, ulAddrCur, DummyPhys, PAGE_SIZE, fPg);
2045	if (!rc)
2046	rc = rtR0MemObjLinuxFixPte(pTask->mm, ulAddrCur, Phys);
2047	#elif RTLNX_VER_MIN(2,6,0) \|\| defined(HAVE_26_STYLE_REMAP_PAGE_RANGE)
2048	rc = remap_page_range(vma, ulAddrCur, Phys, PAGE_SIZE, fPg);
2049	#else /* 2.4 */
2050	rc = remap_page_range(ulAddrCur, Phys, PAGE_SIZE, fPg);
2051	#endif
2052	if (rc)
2053	{
2054	rc = VERR_NO_MEMORY;
2055	break;
2056	}
2057	}
2058	}
2059	}
2060
2061	#ifdef CONFIG_NUMA_BALANCING
2062	# if RTLNX_VER_MAX(3,13,0) && RTLNX_RHEL_MAX(7,0)
2063	# define VBOX_NUMA_HACK_OLD
2064	# endif
2065	if (RT_SUCCESS(rc))
2066	{
2067	/** @todo Ugly hack! But right now we have no other means to
2068	* disable automatic NUMA page balancing. */
2069	# ifdef RT_OS_X86
2070	# ifdef VBOX_NUMA_HACK_OLD
2071	pTask->mm->numa_next_reset = jiffies + 0x7fffffffUL;
2072	# endif
2073	pTask->mm->numa_next_scan = jiffies + 0x7fffffffUL;
2074	# else
2075	# ifdef VBOX_NUMA_HACK_OLD
2076	pTask->mm->numa_next_reset = jiffies + 0x7fffffffffffffffUL;
2077	# endif
2078	pTask->mm->numa_next_scan = jiffies + 0x7fffffffffffffffUL;
2079	# endif
2080	}
2081	#endif /* CONFIG_NUMA_BALANCING */
2082
2083	LNX_MM_UP_WRITE(pTask->mm);
2084
2085	if (RT_SUCCESS(rc))
2086	{
2087	#ifdef VBOX_USE_PAE_HACK
2088	__free_page(pDummyPage);
2089	#endif
2090	pMemLnx->Core.pv = pv;
2091	pMemLnx->Core.u.Mapping.R0Process = R0Process;
2092	*ppMem = &pMemLnx->Core;
2093	IPRT_LINUX_RESTORE_EFL_AC();
2094	return VINF_SUCCESS;
2095	}
2096
2097	/*
2098	* Bail out.
2099	*/
2100	rtR0MemObjLinuxDoMunmap(pv, cbSub, pTask);
2101	}
2102	rtR0MemObjDelete(&pMemLnx->Core);
2103	}
2104	#ifdef VBOX_USE_PAE_HACK
2105	__free_page(pDummyPage);
2106	#endif
2107
2108	IPRT_LINUX_RESTORE_EFL_AC();
2109	return rc;
2110	}
2111
2112
2113	DECLHIDDEN(int) rtR0MemObjNativeProtect(PRTR0MEMOBJINTERNAL pMem, size_t offSub, size_t cbSub, uint32_t fProt)
2114	{
2115	# ifdef IPRT_USE_ALLOC_VM_AREA_FOR_EXEC
2116	/*
2117	* Currently only supported when we've got addresses PTEs from the kernel.
2118	*/
2119	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
2120	if (pMemLnx->pArea && pMemLnx->papPtesForArea)
2121	{
2122	pgprot_t const fPg = rtR0MemObjLinuxConvertProt(fProt, true /fKernel/);
2123	size_t const cPages = (offSub + cbSub) >> PAGE_SHIFT;
2124	pte_t **papPtes = pMemLnx->papPtesForArea;
2125	size_t i;
2126
2127	for (i = offSub >> PAGE_SHIFT; i < cPages; i++)
2128	{
2129	set_pte(papPtes[i], mk_pte(pMemLnx->apPages[i], fPg));
2130	}
2131	preempt_disable();
2132	__flush_tlb_all();
2133	preempt_enable();
2134	return VINF_SUCCESS;
2135	}
2136	# elif defined(IPRT_USE_APPLY_TO_PAGE_RANGE_FOR_EXEC)
2137	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
2138	if ( pMemLnx->fExecutable
2139	&& pMemLnx->fMappedToRing0)
2140	{
2141	LNXAPPLYPGRANGE Args;
2142	Args.pMemLnx = pMemLnx;
2143	Args.fPg = rtR0MemObjLinuxConvertProt(fProt, true /fKernel/);
2144	int rcLnx = apply_to_page_range(current->active_mm, (unsigned long)pMemLnx->Core.pv + offSub, cbSub,
2145	rtR0MemObjLinuxApplyPageRange, (void *)&Args);
2146	if (rcLnx)
2147	return VERR_NOT_SUPPORTED;
2148
2149	return VINF_SUCCESS;
2150	}
2151	# endif
2152
2153	NOREF(pMem);
2154	NOREF(offSub);
2155	NOREF(cbSub);
2156	NOREF(fProt);
2157	return VERR_NOT_SUPPORTED;
2158	}
2159
2160
2161	DECLHIDDEN(RTHCPHYS) rtR0MemObjNativeGetPagePhysAddr(PRTR0MEMOBJINTERNAL pMem, size_t iPage)
2162	{
2163	PRTR0MEMOBJLNX pMemLnx = (PRTR0MEMOBJLNX)pMem;
2164
2165	if (pMemLnx->cPages)
2166	return page_to_phys(pMemLnx->apPages[iPage]);
2167
2168	switch (pMemLnx->Core.enmType)
2169	{
2170	case RTR0MEMOBJTYPE_CONT:
2171	return pMemLnx->Core.u.Cont.Phys + (iPage << PAGE_SHIFT);
2172
2173	case RTR0MEMOBJTYPE_PHYS:
2174	return pMemLnx->Core.u.Phys.PhysBase + (iPage << PAGE_SHIFT);
2175
2176	/* the parent knows */
2177	case RTR0MEMOBJTYPE_MAPPING:
2178	return rtR0MemObjNativeGetPagePhysAddr(pMemLnx->Core.uRel.Child.pParent, iPage);
2179
2180	/* cPages > 0 */
2181	case RTR0MEMOBJTYPE_LOW:
2182	case RTR0MEMOBJTYPE_LOCK:
2183	case RTR0MEMOBJTYPE_PHYS_NC:
2184	case RTR0MEMOBJTYPE_PAGE:
2185	case RTR0MEMOBJTYPE_LARGE_PAGE:
2186	default:
2187	AssertMsgFailed(("%d\n", pMemLnx->Core.enmType));
2188	RT_FALL_THROUGH();
2189
2190	case RTR0MEMOBJTYPE_RES_VIRT:
2191	return NIL_RTHCPHYS;
2192	}
2193	}
2194
2195
2196	DECLHIDDEN(int) rtR0MemObjNativeZeroInitWithoutMapping(PRTR0MEMOBJINTERNAL pMem)
2197	{
2198	PRTR0MEMOBJLNX const pMemLnx = (PRTR0MEMOBJLNX)pMem;
2199	size_t const cPages = pMemLnx->Core.cb >> PAGE_SHIFT;
2200	size_t iPage;
2201	/** @todo optimize this. */
2202	for (iPage = 0; iPage < cPages; iPage++)
2203	{
2204	void *pvPage;
2205
2206	/* Get the physical address of the page. */
2207	RTHCPHYS const HCPhys = rtR0MemObjNativeGetPagePhysAddr(&pMemLnx->Core, iPage);
2208	AssertReturn(HCPhys != NIL_RTHCPHYS, VERR_INTERNAL_ERROR_3);
2209	Assert(!(HCPhys & PAGE_OFFSET_MASK));
2210
2211	/* Would've like to use valid_phys_addr_range for this test, but it isn't exported. */
2212	AssertReturn((HCPhys \| PAGE_OFFSET_MASK) < __pa(high_memory), VERR_INTERNAL_ERROR_3);
2213
2214	/* Map it. */
2215	pvPage = phys_to_virt(HCPhys);
2216	AssertPtrReturn(pvPage, VERR_INTERNAL_ERROR_3);
2217
2218	/* Zero it. */
2219	RT_BZERO(pvPage, PAGE_SIZE);
2220	}
2221	return VINF_SUCCESS;
2222	}
2223

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

source: vbox/trunk/src/VBox/Runtime/r0drv/linux/memobj-r0drv-linux.c@ 109071

Download in other formats: