alloc-r0drv-linux.c@ 96716

Last change on this file since 96716 was 96716, checked in by vboxsync, 2 years ago
Backed out r153600: Incorrect fix, bugref:4567:65.
Property svn:eol-style set to `native` Property svn:keywords set to `Id Revision`
File size: 15.2 KB

Line
1	/* $Id: alloc-r0drv-linux.c 96716 2022-09-13 10:12:48Z vboxsync $ */
2	/** @file
3	* IPRT - Memory Allocation, Ring-0 Driver, Linux.
4	*/
5
6	/*
7	* Copyright (C) 2006-2022 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	#include "internal/iprt.h"
43	#include <iprt/mem.h>
44
45	#include <iprt/assert.h>
46	#include <iprt/errcore.h>
47	#include "r0drv/alloc-r0drv.h"
48
49
50	#if (defined(RT_ARCH_AMD64) \|\| defined(DOXYGEN_RUNNING)) && !defined(RTMEMALLOC_EXEC_HEAP)
51	# if RTLNX_VER_MIN(2,6,23) && RTLNX_VER_MAX(5,8,0) && !RTLNX_RHEL_MAJ_PREREQ(8,5)
52	/**
53	* Starting with 2.6.23 we can use __get_vm_area and map_vm_area to allocate
54	* memory in the moduel range. This is preferrable to the exec heap below.
55	*/
56	# define RTMEMALLOC_EXEC_VM_AREA
57	# else
58	/**
59	* We need memory in the module range (~2GB to ~0) this can only be obtained
60	* thru APIs that are not exported (see module_alloc()).
61	*
62	* So, we'll have to create a quick and dirty heap here using BSS memory.
63	* Very annoying and it's going to restrict us!
64	*/
65	# define RTMEMALLOC_EXEC_HEAP
66	# endif
67	#endif
68
69	#ifdef RTMEMALLOC_EXEC_HEAP
70	# include <iprt/heap.h>
71	# include <iprt/spinlock.h>
72	# include <iprt/errcore.h>
73	#endif
74
75	#include "internal/initterm.h"
76
77
78	/*********************************************************************************************************************************
79	* Structures and Typedefs *
80	*********************************************************************************************************************************/
81	#ifdef RTMEMALLOC_EXEC_VM_AREA
82	/**
83	* Extended header used for headers marked with RTMEMHDR_FLAG_EXEC_VM_AREA.
84	*
85	* This is used with allocating executable memory, for things like generated
86	* code and loaded modules.
87	*/
88	typedef struct RTMEMLNXHDREX
89	{
90	/** The VM area for this allocation. */
91	struct vm_struct *pVmArea;
92	void *pvDummy;
93	/** The header we present to the generic API. */
94	RTMEMHDR Hdr;
95	} RTMEMLNXHDREX;
96	AssertCompileSize(RTMEMLNXHDREX, 32);
97	/** Pointer to an extended memory header. */
98	typedef RTMEMLNXHDREX *PRTMEMLNXHDREX;
99	#endif
100
101
102	/*********************************************************************************************************************************
103	* Global Variables *
104	*********************************************************************************************************************************/
105	#ifdef RTMEMALLOC_EXEC_HEAP
106	/** The heap. */
107	static RTHEAPSIMPLE g_HeapExec = NIL_RTHEAPSIMPLE;
108	/** Spinlock protecting the heap. */
109	static RTSPINLOCK g_HeapExecSpinlock = NIL_RTSPINLOCK;
110	#endif
111
112
113	/**
114	* API for cleaning up the heap spinlock on IPRT termination.
115	* This is as RTMemExecDonate specific to AMD64 Linux/GNU.
116	*/
117	DECLHIDDEN(void) rtR0MemExecCleanup(void)
118	{
119	#ifdef RTMEMALLOC_EXEC_HEAP
120	RTSpinlockDestroy(g_HeapExecSpinlock);
121	g_HeapExecSpinlock = NIL_RTSPINLOCK;
122	#endif
123	}
124
125
126	/**
127	* Donate read+write+execute memory to the exec heap.
128	*
129	* This API is specific to AMD64 and Linux/GNU. A kernel module that desires to
130	* use RTMemExecAlloc on AMD64 Linux/GNU will have to donate some statically
131	* allocated memory in the module if it wishes for GCC generated code to work.
132	* GCC can only generate modules that work in the address range ~2GB to ~0
133	* currently.
134	*
135	* The API only accept one single donation.
136	*
137	* @returns IPRT status code.
138	* @retval VERR_NOT_SUPPORTED if the code isn't enabled.
139	* @param pvMemory Pointer to the memory block.
140	* @param cb The size of the memory block.
141	*/
142	RTR0DECL(int) RTR0MemExecDonate(void *pvMemory, size_t cb)
143	{
144	#ifdef RTMEMALLOC_EXEC_HEAP
145	int rc;
146	AssertReturn(g_HeapExec == NIL_RTHEAPSIMPLE, VERR_WRONG_ORDER);
147
148	rc = RTSpinlockCreate(&g_HeapExecSpinlock, RTSPINLOCK_FLAGS_INTERRUPT_SAFE, "RTR0MemExecDonate");
149	if (RT_SUCCESS(rc))
150	{
151	rc = RTHeapSimpleInit(&g_HeapExec, pvMemory, cb);
152	if (RT_FAILURE(rc))
153	rtR0MemExecCleanup();
154	}
155	return rc;
156	#else
157	RT_NOREF_PV(pvMemory); RT_NOREF_PV(cb);
158	return VERR_NOT_SUPPORTED;
159	#endif
160	}
161	RT_EXPORT_SYMBOL(RTR0MemExecDonate);
162
163
164
165	#ifdef RTMEMALLOC_EXEC_VM_AREA
166	/**
167	* Allocate executable kernel memory in the module range.
168	*
169	* @returns Pointer to a allocation header success. NULL on failure.
170	*
171	* @param cb The size the user requested.
172	*/
173	static PRTMEMHDR rtR0MemAllocExecVmArea(size_t cb)
174	{
175	size_t const cbAlloc = RT_ALIGN_Z(sizeof(RTMEMLNXHDREX) + cb, PAGE_SIZE);
176	size_t const cPages = cbAlloc >> PAGE_SHIFT;
177	struct page **papPages;
178	struct vm_struct *pVmArea;
179	size_t iPage;
180
181	pVmArea = __get_vm_area(cbAlloc, VM_ALLOC, MODULES_VADDR, MODULES_END);
182	if (!pVmArea)
183	return NULL;
184	pVmArea->nr_pages = 0; /* paranoia? */
185	pVmArea->pages = NULL; /* paranoia? */
186
187	papPages = (struct page *)kmalloc(cPages sizeof(papPages[0]), GFP_KERNEL \| __GFP_NOWARN);
188	if (!papPages)
189	{
190	vunmap(pVmArea->addr);
191	return NULL;
192	}
193
194	for (iPage = 0; iPage < cPages; iPage++)
195	{
196	papPages[iPage] = alloc_page(GFP_KERNEL \| __GFP_HIGHMEM \| __GFP_NOWARN);
197	if (!papPages[iPage])
198	break;
199	}
200	if (iPage == cPages)
201	{
202	/*
203	* Map the pages.
204	*
205	* Not entirely sure we really need to set nr_pages and pages here, but
206	* they provide a very convenient place for storing something we need
207	* in the free function, if nothing else...
208	*/
209	# if RTLNX_VER_MAX(3,17,0)
210	struct page **papPagesIterator = papPages;
211	# endif
212	pVmArea->nr_pages = cPages;
213	pVmArea->pages = papPages;
214	if (!map_vm_area(pVmArea, PAGE_KERNEL_EXEC,
215	# if RTLNX_VER_MAX(3,17,0)
216	&papPagesIterator
217	# else
218	papPages
219	# endif
220	))
221	{
222	PRTMEMLNXHDREX pHdrEx = (PRTMEMLNXHDREX)pVmArea->addr;
223	pHdrEx->pVmArea = pVmArea;
224	pHdrEx->pvDummy = NULL;
225	return &pHdrEx->Hdr;
226	}
227	/* bail out */
228	# if RTLNX_VER_MAX(3,17,0)
229	pVmArea->nr_pages = papPagesIterator - papPages;
230	# endif
231	}
232
233	vunmap(pVmArea->addr);
234
235	while (iPage-- > 0)
236	__free_page(papPages[iPage]);
237	kfree(papPages);
238
239	return NULL;
240	}
241	#endif /* RTMEMALLOC_EXEC_VM_AREA */
242
243
244	/**
245	* OS specific allocation function.
246	*/
247	DECLHIDDEN(int) rtR0MemAllocEx(size_t cb, uint32_t fFlags, PRTMEMHDR *ppHdr)
248	{
249	PRTMEMHDR pHdr;
250	IPRT_LINUX_SAVE_EFL_AC();
251
252	/*
253	* Allocate.
254	*/
255	if (fFlags & RTMEMHDR_FLAG_EXEC)
256	{
257	if (fFlags & RTMEMHDR_FLAG_ANY_CTX)
258	return VERR_NOT_SUPPORTED;
259
260	#if defined(RT_ARCH_AMD64)
261	# ifdef RTMEMALLOC_EXEC_HEAP
262	if (g_HeapExec != NIL_RTHEAPSIMPLE)
263	{
264	RTSpinlockAcquire(g_HeapExecSpinlock);
265	pHdr = (PRTMEMHDR)RTHeapSimpleAlloc(g_HeapExec, cb + sizeof(*pHdr), 0);
266	RTSpinlockRelease(g_HeapExecSpinlock);
267	fFlags \|= RTMEMHDR_FLAG_EXEC_HEAP;
268	}
269	else
270	pHdr = NULL;
271
272	# elif defined(RTMEMALLOC_EXEC_VM_AREA)
273	pHdr = rtR0MemAllocExecVmArea(cb);
274	fFlags \|= RTMEMHDR_FLAG_EXEC_VM_AREA;
275
276	# else /* !RTMEMALLOC_EXEC_HEAP */
277	# error "you don not want to go here..."
278	pHdr = (PRTMEMHDR)__vmalloc(cb + sizeof(*pHdr), GFP_KERNEL \| __GFP_HIGHMEM \| __GFP_NOWARN, MY_PAGE_KERNEL_EXEC);
279	# endif /* !RTMEMALLOC_EXEC_HEAP */
280
281	#elif defined(PAGE_KERNEL_EXEC) && defined(CONFIG_X86_PAE)
282	pHdr = (PRTMEMHDR)__vmalloc(cb + sizeof(*pHdr), GFP_KERNEL \| __GFP_HIGHMEM \| __GFP_NOWARN, MY_PAGE_KERNEL_EXEC);
283	#else
284	pHdr = (PRTMEMHDR)vmalloc(cb + sizeof(*pHdr));
285	#endif
286	}
287	else
288	{
289	if (
290	#if 1 /* vmalloc has serious performance issues, avoid it. */
291	cb <= PAGE_SIZE16 - sizeof(pHdr)
292	#else
293	cb <= PAGE_SIZE
294	#endif
295	\|\| (fFlags & RTMEMHDR_FLAG_ANY_CTX)
296	)
297	{
298	fFlags \|= RTMEMHDR_FLAG_KMALLOC;
299	pHdr = kmalloc(cb + sizeof(*pHdr),
300	(fFlags & RTMEMHDR_FLAG_ANY_CTX_ALLOC) ? (GFP_ATOMIC \| __GFP_NOWARN)
301	: (GFP_KERNEL \| __GFP_NOWARN));
302	if (RT_UNLIKELY( !pHdr
303	&& cb > PAGE_SIZE
304	&& !(fFlags & RTMEMHDR_FLAG_ANY_CTX) ))
305	{
306	fFlags &= ~RTMEMHDR_FLAG_KMALLOC;
307	pHdr = vmalloc(cb + sizeof(*pHdr));
308	}
309	}
310	else
311	pHdr = vmalloc(cb + sizeof(*pHdr));
312	}
313	if (RT_UNLIKELY(!pHdr))
314	{
315	IPRT_LINUX_RESTORE_EFL_AC();
316	return VERR_NO_MEMORY;
317	}
318
319	/*
320	* Initialize.
321	*/
322	pHdr->u32Magic = RTMEMHDR_MAGIC;
323	pHdr->fFlags = fFlags;
324	pHdr->cb = cb;
325	pHdr->cbReq = cb;
326
327	*ppHdr = pHdr;
328	IPRT_LINUX_RESTORE_EFL_AC();
329	return VINF_SUCCESS;
330	}
331
332
333	/**
334	* OS specific free function.
335	*/
336	DECLHIDDEN(void) rtR0MemFree(PRTMEMHDR pHdr)
337	{
338	IPRT_LINUX_SAVE_EFL_AC();
339
340	pHdr->u32Magic += 1;
341	if (pHdr->fFlags & RTMEMHDR_FLAG_KMALLOC)
342	kfree(pHdr);
343	#ifdef RTMEMALLOC_EXEC_HEAP
344	else if (pHdr->fFlags & RTMEMHDR_FLAG_EXEC_HEAP)
345	{
346	RTSpinlockAcquire(g_HeapExecSpinlock);
347	RTHeapSimpleFree(g_HeapExec, pHdr);
348	RTSpinlockRelease(g_HeapExecSpinlock);
349	}
350	#endif
351	#ifdef RTMEMALLOC_EXEC_VM_AREA
352	else if (pHdr->fFlags & RTMEMHDR_FLAG_EXEC_VM_AREA)
353	{
354	PRTMEMLNXHDREX pHdrEx = RT_FROM_MEMBER(pHdr, RTMEMLNXHDREX, Hdr);
355	size_t iPage = pHdrEx->pVmArea->nr_pages;
356	struct page **papPages = pHdrEx->pVmArea->pages;
357	void *pvMapping = pHdrEx->pVmArea->addr;
358
359	vunmap(pvMapping);
360
361	while (iPage-- > 0)
362	__free_page(papPages[iPage]);
363	kfree(papPages);
364	}
365	#endif
366	else
367	vfree(pHdr);
368
369	IPRT_LINUX_RESTORE_EFL_AC();
370	}
371
372
373
374	/**
375	* Compute order. Some functions allocate 2^order pages.
376	*
377	* @returns order.
378	* @param cPages Number of pages.
379	*/
380	static int CalcPowerOf2Order(unsigned long cPages)
381	{
382	int iOrder;
383	unsigned long cTmp;
384
385	for (iOrder = 0, cTmp = cPages; cTmp >>= 1; ++iOrder)
386	;
387	if (cPages & ~(1 << iOrder))
388	++iOrder;
389
390	return iOrder;
391	}
392
393
394	/**
395	* Allocates physical contiguous memory (below 4GB).
396	* The allocation is page aligned and the content is undefined.
397	*
398	* @returns Pointer to the memory block. This is page aligned.
399	* @param pPhys Where to store the physical address.
400	* @param cb The allocation size in bytes. This is always
401	* rounded up to PAGE_SIZE.
402	*/
403	RTR0DECL(void *) RTMemContAlloc(PRTCCPHYS pPhys, size_t cb)
404	{
405	int cOrder;
406	unsigned cPages;
407	struct page *paPages;
408	void *pvRet;
409	IPRT_LINUX_SAVE_EFL_AC();
410
411	/*
412	* validate input.
413	*/
414	AssertPtr(pPhys);
415	Assert(cb > 0);
416
417	/*
418	* Allocate page pointer array.
419	*/
420	cb = RT_ALIGN_Z(cb, PAGE_SIZE);
421	cPages = cb >> PAGE_SHIFT;
422	cOrder = CalcPowerOf2Order(cPages);
423	#if (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
424	/* ZONE_DMA32: 0-4GB */
425	paPages = alloc_pages(GFP_DMA32 \| __GFP_NOWARN, cOrder);
426	if (!paPages)
427	#endif
428	#ifdef RT_ARCH_AMD64
429	/* ZONE_DMA; 0-16MB */
430	paPages = alloc_pages(GFP_DMA \| __GFP_NOWARN, cOrder);
431	#else
432	/* ZONE_NORMAL: 0-896MB */
433	paPages = alloc_pages(GFP_USER \| __GFP_NOWARN, cOrder);
434	#endif
435	if (paPages)
436	{
437	/*
438	* Reserve the pages and mark them executable.
439	*/
440	unsigned iPage;
441	for (iPage = 0; iPage < cPages; iPage++)
442	{
443	Assert(!PageHighMem(&paPages[iPage]));
444	if (iPage + 1 < cPages)
445	{
446	AssertMsg( (uintptr_t)phys_to_virt(page_to_phys(&paPages[iPage])) + PAGE_SIZE
447	== (uintptr_t)phys_to_virt(page_to_phys(&paPages[iPage + 1]))
448	&& page_to_phys(&paPages[iPage]) + PAGE_SIZE
449	== page_to_phys(&paPages[iPage + 1]),
450	("iPage=%i cPages=%u [0]=%#llx,%p [1]=%#llx,%p\n", iPage, cPages,
451	(long long)page_to_phys(&paPages[iPage]), phys_to_virt(page_to_phys(&paPages[iPage])),
452	(long long)page_to_phys(&paPages[iPage + 1]), phys_to_virt(page_to_phys(&paPages[iPage + 1])) ));
453	}
454
455	SetPageReserved(&paPages[iPage]);
456	}
457	*pPhys = page_to_phys(paPages);
458	pvRet = phys_to_virt(page_to_phys(paPages));
459	}
460	else
461	pvRet = NULL;
462
463	IPRT_LINUX_RESTORE_EFL_AC();
464	return pvRet;
465	}
466	RT_EXPORT_SYMBOL(RTMemContAlloc);
467
468
469	/**
470	* Frees memory allocated using RTMemContAlloc().
471	*
472	* @param pv Pointer to return from RTMemContAlloc().
473	* @param cb The cb parameter passed to RTMemContAlloc().
474	*/
475	RTR0DECL(void) RTMemContFree(void *pv, size_t cb)
476	{
477	if (pv)
478	{
479	int cOrder;
480	unsigned cPages;
481	unsigned iPage;
482	struct page *paPages;
483	IPRT_LINUX_SAVE_EFL_AC();
484
485	/* validate */
486	AssertMsg(!((uintptr_t)pv & PAGE_OFFSET_MASK), ("pv=%p\n", pv));
487	Assert(cb > 0);
488
489	/* calc order and get pages */
490	cb = RT_ALIGN_Z(cb, PAGE_SIZE);
491	cPages = cb >> PAGE_SHIFT;
492	cOrder = CalcPowerOf2Order(cPages);
493	paPages = virt_to_page(pv);
494
495	/*
496	* Restore page attributes freeing the pages.
497	*/
498	for (iPage = 0; iPage < cPages; iPage++)
499	{
500	ClearPageReserved(&paPages[iPage]);
501	}
502	__free_pages(paPages, cOrder);
503	IPRT_LINUX_RESTORE_EFL_AC();
504	}
505	}
506	RT_EXPORT_SYMBOL(RTMemContFree);
507

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

source: vbox/trunk/src/VBox/Runtime/r0drv/linux/alloc-r0drv-linux.c@ 96716

Download in other formats: