alloc-r0drv-linux.c@ 32674

Last change on this file since 32674 was 32674, checked in by vboxsync, 15 years ago
IPRT: started on some internal ring-0 alloc api.
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 9.6 KB

Line
1	/* $Id: alloc-r0drv-linux.c 32674 2010-09-21 16:51:50Z vboxsync $ */
2	/** @file
3	* IPRT - Memory Allocation, Ring-0 Driver, Linux.
4	*/
5
6	/*
7	* Copyright (C) 2006-2007 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.215389.xyz. 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
34	#include <iprt/mem.h>
35	#include <iprt/assert.h>
36	#include "r0drv/alloc-r0drv.h"
37
38	#if defined(RT_ARCH_AMD64) \|\| defined(DOXYGEN_RUNNING)
39	/**
40	* We need memory in the module range (~2GB to ~0) this can only be obtained
41	* thru APIs that are not exported (see module_alloc()).
42	*
43	* So, we'll have to create a quick and dirty heap here using BSS memory.
44	* Very annoying and it's going to restrict us!
45	*/
46	# define RTMEMALLOC_EXEC_HEAP
47	#endif
48	#ifdef RTMEMALLOC_EXEC_HEAP
49	# include <iprt/heap.h>
50	# include <iprt/spinlock.h>
51	# include <iprt/err.h>
52	#endif
53
54
55	/*******************************************************************************
56	* Global Variables *
57	*******************************************************************************/
58	#ifdef RTMEMALLOC_EXEC_HEAP
59	/** The heap. */
60	static RTHEAPSIMPLE g_HeapExec = NIL_RTHEAPSIMPLE;
61	/** Spinlock protecting the heap. */
62	static RTSPINLOCK g_HeapExecSpinlock = NIL_RTSPINLOCK;
63
64
65	/**
66	* API for cleaning up the heap spinlock on IPRT termination.
67	* This is as RTMemExecDonate specific to AMD64 Linux/GNU.
68	*/
69	void rtR0MemExecCleanup(void)
70	{
71	RTSpinlockDestroy(g_HeapExecSpinlock);
72	g_HeapExecSpinlock = NIL_RTSPINLOCK;
73	}
74
75
76	/**
77	* Donate read+write+execute memory to the exec heap.
78	*
79	* This API is specific to AMD64 and Linux/GNU. A kernel module that desires to
80	* use RTMemExecAlloc on AMD64 Linux/GNU will have to donate some statically
81	* allocated memory in the module if it wishes for GCC generated code to work.
82	* GCC can only generate modules that work in the address range ~2GB to ~0
83	* currently.
84	*
85	* The API only accept one single donation.
86	*
87	* @returns IPRT status code.
88	* @param pvMemory Pointer to the memory block.
89	* @param cb The size of the memory block.
90	*/
91	RTR0DECL(int) RTR0MemExecDonate(void *pvMemory, size_t cb)
92	{
93	int rc;
94	AssertReturn(g_HeapExec == NIL_RTHEAPSIMPLE, VERR_WRONG_ORDER);
95
96	rc = RTSpinlockCreate(&g_HeapExecSpinlock);
97	if (RT_SUCCESS(rc))
98	{
99	rc = RTHeapSimpleInit(&g_HeapExec, pvMemory, cb);
100	if (RT_FAILURE(rc))
101	rtR0MemExecCleanup();
102	}
103	return rc;
104	}
105	RT_EXPORT_SYMBOL(RTR0MemExecDonate);
106	#endif /* RTMEMALLOC_EXEC_HEAP */
107
108
109
110	/**
111	* OS specific allocation function.
112	*/
113	PRTMEMHDR rtR0MemAlloc(size_t cb, uint32_t fFlags)
114	{
115	PRTMEMHDR pHdr;
116
117	/*
118	* Allocate.
119	*/
120	if (fFlags & RTMEMHDR_FLAG_EXEC)
121	{
122	AssertReturn(!(fFlags & RTMEMHDR_FLAG_ANY_CTX), NULL);
123
124	#if defined(RT_ARCH_AMD64)
125	# ifdef RTMEMALLOC_EXEC_HEAP
126	if (g_HeapExec != NIL_RTHEAPSIMPLE)
127	{
128	RTSPINLOCKTMP SpinlockTmp = RTSPINLOCKTMP_INITIALIZER;
129	RTSpinlockAcquireNoInts(g_HeapExecSpinlock, &SpinlockTmp);
130	pHdr = (PRTMEMHDR)RTHeapSimpleAlloc(g_HeapExec, cb + sizeof(*pHdr), 0);
131	RTSpinlockReleaseNoInts(g_HeapExecSpinlock, &SpinlockTmp);
132	fFlags \|= RTMEMHDR_FLAG_EXEC_HEAP;
133	}
134	else
135	pHdr = NULL;
136	# else /* !RTMEMALLOC_EXEC_HEAP */
137	pHdr = (PRTMEMHDR)__vmalloc(cb + sizeof(*pHdr), GFP_KERNEL \| __GFP_HIGHMEM, MY_PAGE_KERNEL_EXEC);
138	# endif /* !RTMEMALLOC_EXEC_HEAP */
139
140	#elif defined(PAGE_KERNEL_EXEC) && defined(CONFIG_X86_PAE)
141	pHdr = (PRTMEMHDR)__vmalloc(cb + sizeof(*pHdr), GFP_KERNEL \| __GFP_HIGHMEM, MY_PAGE_KERNEL_EXEC);
142	#else
143	pHdr = (PRTMEMHDR)vmalloc(cb + sizeof(*pHdr));
144	#endif
145	}
146	else
147	{
148	if (cb <= PAGE_SIZE \|\| (fFlags & RTMEMHDR_FLAG_ANY_CTX))
149	{
150	fFlags \|= RTMEMHDR_FLAG_KMALLOC;
151	pHdr = kmalloc(cb + sizeof(*pHdr), GFP_KERNEL);
152	}
153	else
154	pHdr = vmalloc(cb + sizeof(*pHdr));
155	}
156
157	/*
158	* Initialize.
159	*/
160	if (pHdr)
161	{
162	pHdr->u32Magic = RTMEMHDR_MAGIC;
163	pHdr->fFlags = fFlags;
164	pHdr->cb = cb;
165	pHdr->cbReq = cb;
166	}
167	return pHdr;
168	}
169
170
171	/**
172	* OS specific free function.
173	*/
174	void rtR0MemFree(PRTMEMHDR pHdr)
175	{
176	pHdr->u32Magic += 1;
177	if (pHdr->fFlags & RTMEMHDR_FLAG_KMALLOC)
178	kfree(pHdr);
179	#ifdef RTMEMALLOC_EXEC_HEAP
180	else if (pHdr->fFlags & RTMEMHDR_FLAG_EXEC_HEAP)
181	{
182	RTSPINLOCKTMP SpinlockTmp = RTSPINLOCKTMP_INITIALIZER;
183	RTSpinlockAcquireNoInts(g_HeapExecSpinlock, &SpinlockTmp);
184	RTHeapSimpleFree(g_HeapExec, pHdr);
185	RTSpinlockReleaseNoInts(g_HeapExecSpinlock, &SpinlockTmp);
186	}
187	#endif
188	else
189	vfree(pHdr);
190	}
191
192
193	/**
194	* Compute order. Some functions allocate 2^order pages.
195	*
196	* @returns order.
197	* @param cPages Number of pages.
198	*/
199	static int CalcPowerOf2Order(unsigned long cPages)
200	{
201	int iOrder;
202	unsigned long cTmp;
203
204	for (iOrder = 0, cTmp = cPages; cTmp >>= 1; ++iOrder)
205	;
206	if (cPages & ~(1 << iOrder))
207	++iOrder;
208
209	return iOrder;
210	}
211
212
213	/**
214	* Allocates physical contiguous memory (below 4GB).
215	* The allocation is page aligned and the content is undefined.
216	*
217	* @returns Pointer to the memory block. This is page aligned.
218	* @param pPhys Where to store the physical address.
219	* @param cb The allocation size in bytes. This is always
220	* rounded up to PAGE_SIZE.
221	*/
222	RTR0DECL(void *) RTMemContAlloc(PRTCCPHYS pPhys, size_t cb)
223	{
224	int cOrder;
225	unsigned cPages;
226	struct page *paPages;
227
228	/*
229	* validate input.
230	*/
231	Assert(VALID_PTR(pPhys));
232	Assert(cb > 0);
233
234	/*
235	* Allocate page pointer array.
236	*/
237	cb = RT_ALIGN_Z(cb, PAGE_SIZE);
238	cPages = cb >> PAGE_SHIFT;
239	cOrder = CalcPowerOf2Order(cPages);
240	#if (defined(RT_ARCH_AMD64) \|\| defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
241	/* ZONE_DMA32: 0-4GB */
242	paPages = alloc_pages(GFP_DMA32, cOrder);
243	if (!paPages)
244	#endif
245	#ifdef RT_ARCH_AMD64
246	/* ZONE_DMA; 0-16MB */
247	paPages = alloc_pages(GFP_DMA, cOrder);
248	#else
249	/* ZONE_NORMAL: 0-896MB */
250	paPages = alloc_pages(GFP_USER, cOrder);
251	#endif
252	if (paPages)
253	{
254	/*
255	* Reserve the pages and mark them executable.
256	*/
257	unsigned iPage;
258	for (iPage = 0; iPage < cPages; iPage++)
259	{
260	Assert(!PageHighMem(&paPages[iPage]));
261	if (iPage + 1 < cPages)
262	{
263	AssertMsg( (uintptr_t)phys_to_virt(page_to_phys(&paPages[iPage])) + PAGE_SIZE
264	== (uintptr_t)phys_to_virt(page_to_phys(&paPages[iPage + 1]))
265	&& page_to_phys(&paPages[iPage]) + PAGE_SIZE
266	== page_to_phys(&paPages[iPage + 1]),
267	("iPage=%i cPages=%u [0]=%#llx,%p [1]=%#llx,%p\n", iPage, cPages,
268	(long long)page_to_phys(&paPages[iPage]), phys_to_virt(page_to_phys(&paPages[iPage])),
269	(long long)page_to_phys(&paPages[iPage + 1]), phys_to_virt(page_to_phys(&paPages[iPage + 1])) ));
270	}
271
272	SetPageReserved(&paPages[iPage]);
273	#if LINUX_VERSION_CODE > KERNEL_VERSION(2, 4, 20) /** @todo find the exact kernel where change_page_attr was introduced. */
274	MY_SET_PAGES_EXEC(&paPages[iPage], 1);
275	#endif
276	}
277	*pPhys = page_to_phys(paPages);
278	return phys_to_virt(page_to_phys(paPages));
279	}
280
281	return NULL;
282	}
283	RT_EXPORT_SYMBOL(RTMemContAlloc);
284
285
286	/**
287	* Frees memory allocated ysing RTMemContAlloc().
288	*
289	* @param pv Pointer to return from RTMemContAlloc().
290	* @param cb The cb parameter passed to RTMemContAlloc().
291	*/
292	RTR0DECL(void) RTMemContFree(void *pv, size_t cb)
293	{
294	if (pv)
295	{
296	int cOrder;
297	unsigned cPages;
298	unsigned iPage;
299	struct page *paPages;
300
301	/* validate */
302	AssertMsg(!((uintptr_t)pv & PAGE_OFFSET_MASK), ("pv=%p\n", pv));
303	Assert(cb > 0);
304
305	/* calc order and get pages */
306	cb = RT_ALIGN_Z(cb, PAGE_SIZE);
307	cPages = cb >> PAGE_SHIFT;
308	cOrder = CalcPowerOf2Order(cPages);
309	paPages = virt_to_page(pv);
310
311	/*
312	* Restore page attributes freeing the pages.
313	*/
314	for (iPage = 0; iPage < cPages; iPage++)
315	{
316	ClearPageReserved(&paPages[iPage]);
317	#if LINUX_VERSION_CODE > KERNEL_VERSION(2, 4, 20) /** @todo find the exact kernel where change_page_attr was introduced. */
318	MY_SET_PAGES_NOEXEC(&paPages[iPage], 1);
319	#endif
320	}
321	__free_pages(paPages, cOrder);
322	}
323	}
324	RT_EXPORT_SYMBOL(RTMemContFree);
325

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

source: vbox/trunk/src/VBox/Runtime/r0drv/linux/alloc-r0drv-linux.c@ 32674

Download in other formats: