alloc-r0drv-linux.c@ 55784

Last change on this file since 55784 was 52803, checked in by vboxsync, 10 years ago
added GFP_NOWARN to the remaining vmalloc/kmalloc operations
Property svn:eol-style set to `native` Property svn:keywords set to `Id Revision`
File size: 14.5 KB

Line
1	/* $Id: alloc-r0drv-linux.c 52803 2014-09-22 08:21:23Z vboxsync $ */
2	/** @file
3	* IPRT - Memory Allocation, Ring-0 Driver, Linux.
4	*/
5
6	/*
7	* Copyright (C) 2006-2012 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*
17	* The contents of this file may alternatively be used under the terms
18	* of the Common Development and Distribution License Version 1.0
19	* (CDDL) only, as it comes in the "COPYING.CDDL" file of the
20	* VirtualBox OSE distribution, in which case the provisions of the
21	* CDDL are applicable instead of those of the GPL.
22	*
23	* You may elect to license modified versions of this file under the
24	* terms and conditions of either the GPL or the CDDL or both.
25	*/
26
27
28	/*******************************************************************************
29	* Header Files *
30	*******************************************************************************/
31	#include "the-linux-kernel.h"
32	#include "internal/iprt.h"
33	#include <iprt/mem.h>
34
35	#include <iprt/assert.h>
36	#include <iprt/err.h>
37	#include "r0drv/alloc-r0drv.h"
38
39
40	#if (defined(RT_ARCH_AMD64) \|\| defined(DOXYGEN_RUNNING)) && !defined(RTMEMALLOC_EXEC_HEAP)
41	# if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 23)
42	/**
43	* Starting with 2.6.23 we can use __get_vm_area and map_vm_area to allocate
44	* memory in the moduel range. This is preferrable to the exec heap below.
45	*/
46	# define RTMEMALLOC_EXEC_VM_AREA
47	# else
48	/**
49	* We need memory in the module range (~2GB to ~0) this can only be obtained
50	* thru APIs that are not exported (see module_alloc()).
51	*
52	* So, we'll have to create a quick and dirty heap here using BSS memory.
53	* Very annoying and it's going to restrict us!
54	*/
55	# define RTMEMALLOC_EXEC_HEAP
56	# endif
57	#endif
58
59	#ifdef RTMEMALLOC_EXEC_HEAP
60	# include <iprt/heap.h>
61	# include <iprt/spinlock.h>
62	# include <iprt/err.h>
63	#endif
64
65
66	/*******************************************************************************
67	* Structures and Typedefs *
68	*******************************************************************************/
69	#ifdef RTMEMALLOC_EXEC_VM_AREA
70	/**
71	* Extended header used for headers marked with RTMEMHDR_FLAG_EXEC_VM_AREA.
72	*
73	* This is used with allocating executable memory, for things like generated
74	* code and loaded modules.
75	*/
76	typedef struct RTMEMLNXHDREX
77	{
78	/** The VM area for this allocation. */
79	struct vm_struct *pVmArea;
80	void *pvDummy;
81	/** The header we present to the generic API. */
82	RTMEMHDR Hdr;
83	} RTMEMLNXHDREX;
84	AssertCompileSize(RTMEMLNXHDREX, 32);
85	/** Pointer to an extended memory header. */
86	typedef RTMEMLNXHDREX *PRTMEMLNXHDREX;
87	#endif
88
89
90	/*******************************************************************************
91	* Global Variables *
92	*******************************************************************************/
93	#ifdef RTMEMALLOC_EXEC_HEAP
94	/** The heap. */
95	static RTHEAPSIMPLE g_HeapExec = NIL_RTHEAPSIMPLE;
96	/** Spinlock protecting the heap. */
97	static RTSPINLOCK g_HeapExecSpinlock = NIL_RTSPINLOCK;
98	#endif
99
100
101	/**
102	* API for cleaning up the heap spinlock on IPRT termination.
103	* This is as RTMemExecDonate specific to AMD64 Linux/GNU.
104	*/
105	DECLHIDDEN(void) rtR0MemExecCleanup(void)
106	{
107	#ifdef RTMEMALLOC_EXEC_HEAP
108	RTSpinlockDestroy(g_HeapExecSpinlock);
109	g_HeapExecSpinlock = NIL_RTSPINLOCK;
110	#endif
111	}
112
113
114	/**
115	* Donate read+write+execute memory to the exec heap.
116	*
117	* This API is specific to AMD64 and Linux/GNU. A kernel module that desires to
118	* use RTMemExecAlloc on AMD64 Linux/GNU will have to donate some statically
119	* allocated memory in the module if it wishes for GCC generated code to work.
120	* GCC can only generate modules that work in the address range ~2GB to ~0
121	* currently.
122	*
123	* The API only accept one single donation.
124	*
125	* @returns IPRT status code.
126	* @retval VERR_NOT_SUPPORTED if the code isn't enabled.
127	* @param pvMemory Pointer to the memory block.
128	* @param cb The size of the memory block.
129	*/
130	RTR0DECL(int) RTR0MemExecDonate(void *pvMemory, size_t cb)
131	{
132	#ifdef RTMEMALLOC_EXEC_HEAP
133	int rc;
134	AssertReturn(g_HeapExec == NIL_RTHEAPSIMPLE, VERR_WRONG_ORDER);
135
136	rc = RTSpinlockCreate(&g_HeapExecSpinlock, RTSPINLOCK_FLAGS_INTERRUPT_SAFE, "RTR0MemExecDonate");
137	if (RT_SUCCESS(rc))
138	{
139	rc = RTHeapSimpleInit(&g_HeapExec, pvMemory, cb);
140	if (RT_FAILURE(rc))
141	rtR0MemExecCleanup();
142	}
143	return rc;
144	#else
145	return VERR_NOT_SUPPORTED;
146	#endif
147	}
148	RT_EXPORT_SYMBOL(RTR0MemExecDonate);
149
150
151
152	#ifdef RTMEMALLOC_EXEC_VM_AREA
153	/**
154	* Allocate executable kernel memory in the module range.
155	*
156	* @returns Pointer to a allocation header success. NULL on failure.
157	*
158	* @param cb The size the user requested.
159	*/
160	static PRTMEMHDR rtR0MemAllocExecVmArea(size_t cb)
161	{
162	size_t const cbAlloc = RT_ALIGN_Z(sizeof(RTMEMLNXHDREX) + cb, PAGE_SIZE);
163	size_t const cPages = cbAlloc >> PAGE_SHIFT;
164	struct page **papPages;
165	struct vm_struct *pVmArea;
166	size_t iPage;
167
168	pVmArea = __get_vm_area(cbAlloc, VM_ALLOC, MODULES_VADDR, MODULES_END);
169	if (!pVmArea)
170	return NULL;
171	pVmArea->nr_pages = 0; /* paranoia? */
172	pVmArea->pages = NULL; /* paranoia? */
173
174	papPages = (struct page *)kmalloc(cPages sizeof(papPages[0]), GFP_KERNEL \| __GFP_NOWARN);
175	if (!papPages)
176	{
177	vunmap(pVmArea->addr);
178	return NULL;
179	}
180
181	for (iPage = 0; iPage < cPages; iPage++)
182	{
183	papPages[iPage] = alloc_page(GFP_KERNEL \| __GFP_HIGHMEM \| __GFP_NOWARN);
184	if (!papPages[iPage])
185	break;
186	}
187	if (iPage == cPages)
188	{
189	/*
190	* Map the pages.
191	*
192	* Not entirely sure we really need to set nr_pages and pages here, but
193	* they provide a very convenient place for storing something we need
194	* in the free function, if nothing else...
195	*/
196	# if LINUX_VERSION_CODE < KERNEL_VERSION(3, 17, 0)
197	struct page **papPagesIterator = papPages;
198	# endif
199	pVmArea->nr_pages = cPages;
200	pVmArea->pages = papPages;
201	if (!map_vm_area(pVmArea, PAGE_KERNEL_EXEC,
202	# if LINUX_VERSION_CODE < KERNEL_VERSION(3, 17, 0)
203	&papPagesIterator
204	# else
205	papPages
206	# endif
207	))
208	{
209	PRTMEMLNXHDREX pHdrEx = (PRTMEMLNXHDREX)pVmArea->addr;
210	pHdrEx->pVmArea = pVmArea;
211	pHdrEx->pvDummy = NULL;
212	return &pHdrEx->Hdr;
213	}
214	/* bail out */
215	# if LINUX_VERSION_CODE < KERNEL_VERSION(3, 17, 0)
216	pVmArea->nr_pages = papPagesIterator - papPages;
217	# endif
218	}
219
220	vunmap(pVmArea->addr);
221
222	while (iPage-- > 0)
223	__free_page(papPages[iPage]);
224	kfree(papPages);
225
226	return NULL;
227	}
228	#endif /* RTMEMALLOC_EXEC_VM_AREA */
229
230
231	/**
232	* OS specific allocation function.
233	*/
234	DECLHIDDEN(int) rtR0MemAllocEx(size_t cb, uint32_t fFlags, PRTMEMHDR *ppHdr)
235	{
236	PRTMEMHDR pHdr;
237
238	/*
239	* Allocate.
240	*/
241	if (fFlags & RTMEMHDR_FLAG_EXEC)
242	{
243	if (fFlags & RTMEMHDR_FLAG_ANY_CTX)
244	return VERR_NOT_SUPPORTED;
245
246	#if defined(RT_ARCH_AMD64)
247	# ifdef RTMEMALLOC_EXEC_HEAP
248	if (g_HeapExec != NIL_RTHEAPSIMPLE)
249	{
250	RTSpinlockAcquire(g_HeapExecSpinlock);
251	pHdr = (PRTMEMHDR)RTHeapSimpleAlloc(g_HeapExec, cb + sizeof(*pHdr), 0);
252	RTSpinlockRelease(g_HeapExecSpinlock);
253	fFlags \|= RTMEMHDR_FLAG_EXEC_HEAP;
254	}
255	else
256	pHdr = NULL;
257
258	# elif defined(RTMEMALLOC_EXEC_VM_AREA)
259	pHdr = rtR0MemAllocExecVmArea(cb);
260	fFlags \|= RTMEMHDR_FLAG_EXEC_VM_AREA;
261
262	# else /* !RTMEMALLOC_EXEC_HEAP */
263	# error "you don not want to go here..."
264	pHdr = (PRTMEMHDR)__vmalloc(cb + sizeof(*pHdr), GFP_KERNEL \| __GFP_HIGHMEM \| __GFP_NOWARN, MY_PAGE_KERNEL_EXEC);
265	# endif /* !RTMEMALLOC_EXEC_HEAP */
266
267	#elif defined(PAGE_KERNEL_EXEC) && defined(CONFIG_X86_PAE)
268	pHdr = (PRTMEMHDR)__vmalloc(cb + sizeof(*pHdr), GFP_KERNEL \| __GFP_HIGHMEM \| __GFP_NOWARN, MY_PAGE_KERNEL_EXEC);
269	#else
270	pHdr = (PRTMEMHDR)vmalloc(cb + sizeof(*pHdr));
271	#endif
272	}
273	else
274	{
275	if (
276	#if 1 /* vmalloc has serious performance issues, avoid it. */
277	cb <= PAGE_SIZE16 - sizeof(pHdr)
278	#else
279	cb <= PAGE_SIZE
280	#endif
281	\|\| (fFlags & RTMEMHDR_FLAG_ANY_CTX)
282	)
283	{
284	fFlags \|= RTMEMHDR_FLAG_KMALLOC;
285	pHdr = kmalloc(cb + sizeof(*pHdr),
286	(fFlags & RTMEMHDR_FLAG_ANY_CTX_ALLOC) ? (GFP_ATOMIC \| __GFP_NOWARN)
287	: (GFP_KERNEL \| __GFP_NOWARN));
288	if (RT_UNLIKELY( !pHdr
289	&& cb > PAGE_SIZE
290	&& !(fFlags & RTMEMHDR_FLAG_ANY_CTX) ))
291	{
292	fFlags &= ~RTMEMHDR_FLAG_KMALLOC;
293	pHdr = vmalloc(cb + sizeof(*pHdr));
294	}
295	}
296	else
297	pHdr = vmalloc(cb + sizeof(*pHdr));
298	}
299	if (RT_UNLIKELY(!pHdr))
300	return VERR_NO_MEMORY;
301
302	/*
303	* Initialize.
304	*/
305	pHdr->u32Magic = RTMEMHDR_MAGIC;
306	pHdr->fFlags = fFlags;
307	pHdr->cb = cb;
308	pHdr->cbReq = cb;
309
310	*ppHdr = pHdr;
311	return VINF_SUCCESS;
312	}
313
314
315	/**
316	* OS specific free function.
317	*/
318	DECLHIDDEN(void) rtR0MemFree(PRTMEMHDR pHdr)
319	{
320	pHdr->u32Magic += 1;
321	if (pHdr->fFlags & RTMEMHDR_FLAG_KMALLOC)
322	kfree(pHdr);
323	#ifdef RTMEMALLOC_EXEC_HEAP
324	else if (pHdr->fFlags & RTMEMHDR_FLAG_EXEC_HEAP)
325	{
326	RTSpinlockAcquire(g_HeapExecSpinlock);
327	RTHeapSimpleFree(g_HeapExec, pHdr);
328	RTSpinlockRelease(g_HeapExecSpinlock);
329	}
330	#endif
331	#ifdef RTMEMALLOC_EXEC_VM_AREA
332	else if (pHdr->fFlags & RTMEMHDR_FLAG_EXEC_VM_AREA)
333	{
334	PRTMEMLNXHDREX pHdrEx = RT_FROM_MEMBER(pHdr, RTMEMLNXHDREX, Hdr);
335	size_t iPage = pHdrEx->pVmArea->nr_pages;
336	struct page **papPages = pHdrEx->pVmArea->pages;
337	void *pvMapping = pHdrEx->pVmArea->addr;
338
339	vunmap(pvMapping);
340
341	while (iPage-- > 0)
342	__free_page(papPages[iPage]);
343	kfree(papPages);
344	}
345	#endif
346	else
347	vfree(pHdr);
348	}
349
350
351
352	/**
353	* Compute order. Some functions allocate 2^order pages.
354	*
355	* @returns order.
356	* @param cPages Number of pages.
357	*/
358	static int CalcPowerOf2Order(unsigned long cPages)
359	{
360	int iOrder;
361	unsigned long cTmp;
362
363	for (iOrder = 0, cTmp = cPages; cTmp >>= 1; ++iOrder)
364	;
365	if (cPages & ~(1 << iOrder))
366	++iOrder;
367
368	return iOrder;
369	}
370
371
372	/**
373	* Allocates physical contiguous memory (below 4GB).
374	* The allocation is page aligned and the content is undefined.
375	*
376	* @returns Pointer to the memory block. This is page aligned.
377	* @param pPhys Where to store the physical address.
378	* @param cb The allocation size in bytes. This is always
379	* rounded up to PAGE_SIZE.
380	*/
381	RTR0DECL(void *) RTMemContAlloc(PRTCCPHYS pPhys, size_t cb)
382	{
383	int cOrder;
384	unsigned cPages;
385	struct page *paPages;
386
387	/*
388	* validate input.
389	*/
390	Assert(VALID_PTR(pPhys));
391	Assert(cb > 0);
392
393	/*
394	* Allocate page pointer array.
395	*/
396	cb = RT_ALIGN_Z(cb, PAGE_SIZE);
397	cPages = cb >> PAGE_SHIFT;
398	cOrder = CalcPowerOf2Order(cPages);
399	#if (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
400	/* ZONE_DMA32: 0-4GB */
401	paPages = alloc_pages(GFP_DMA32 \| __GFP_NOWARN, cOrder);
402	if (!paPages)
403	#endif
404	#ifdef RT_ARCH_AMD64
405	/* ZONE_DMA; 0-16MB */
406	paPages = alloc_pages(GFP_DMA \| __GFP_NOWARN, cOrder);
407	#else
408	/* ZONE_NORMAL: 0-896MB */
409	paPages = alloc_pages(GFP_USER \| __GFP_NOWARN, cOrder);
410	#endif
411	if (paPages)
412	{
413	/*
414	* Reserve the pages and mark them executable.
415	*/
416	unsigned iPage;
417	for (iPage = 0; iPage < cPages; iPage++)
418	{
419	Assert(!PageHighMem(&paPages[iPage]));
420	if (iPage + 1 < cPages)
421	{
422	AssertMsg( (uintptr_t)phys_to_virt(page_to_phys(&paPages[iPage])) + PAGE_SIZE
423	== (uintptr_t)phys_to_virt(page_to_phys(&paPages[iPage + 1]))
424	&& page_to_phys(&paPages[iPage]) + PAGE_SIZE
425	== page_to_phys(&paPages[iPage + 1]),
426	("iPage=%i cPages=%u [0]=%#llx,%p [1]=%#llx,%p\n", iPage, cPages,
427	(long long)page_to_phys(&paPages[iPage]), phys_to_virt(page_to_phys(&paPages[iPage])),
428	(long long)page_to_phys(&paPages[iPage + 1]), phys_to_virt(page_to_phys(&paPages[iPage + 1])) ));
429	}
430
431	SetPageReserved(&paPages[iPage]);
432	#if LINUX_VERSION_CODE > KERNEL_VERSION(2, 4, 20) /** @todo find the exact kernel where change_page_attr was introduced. */
433	MY_SET_PAGES_EXEC(&paPages[iPage], 1);
434	#endif
435	}
436	*pPhys = page_to_phys(paPages);
437	return phys_to_virt(page_to_phys(paPages));
438	}
439
440	return NULL;
441	}
442	RT_EXPORT_SYMBOL(RTMemContAlloc);
443
444
445	/**
446	* Frees memory allocated using RTMemContAlloc().
447	*
448	* @param pv Pointer to return from RTMemContAlloc().
449	* @param cb The cb parameter passed to RTMemContAlloc().
450	*/
451	RTR0DECL(void) RTMemContFree(void *pv, size_t cb)
452	{
453	if (pv)
454	{
455	int cOrder;
456	unsigned cPages;
457	unsigned iPage;
458	struct page *paPages;
459
460	/* validate */
461	AssertMsg(!((uintptr_t)pv & PAGE_OFFSET_MASK), ("pv=%p\n", pv));
462	Assert(cb > 0);
463
464	/* calc order and get pages */
465	cb = RT_ALIGN_Z(cb, PAGE_SIZE);
466	cPages = cb >> PAGE_SHIFT;
467	cOrder = CalcPowerOf2Order(cPages);
468	paPages = virt_to_page(pv);
469
470	/*
471	* Restore page attributes freeing the pages.
472	*/
473	for (iPage = 0; iPage < cPages; iPage++)
474	{
475	ClearPageReserved(&paPages[iPage]);
476	#if LINUX_VERSION_CODE > KERNEL_VERSION(2, 4, 20) /** @todo find the exact kernel where change_page_attr was introduced. */
477	MY_SET_PAGES_NOEXEC(&paPages[iPage], 1);
478	#endif
479	}
480	__free_pages(paPages, cOrder);
481	}
482	}
483	RT_EXPORT_SYMBOL(RTMemContFree);
484

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

source: vbox/trunk/src/VBox/Runtime/r0drv/linux/alloc-r0drv-linux.c@ 55784

Download in other formats: