VirtualBox

source: vbox/trunk/src/VBox/Runtime/r0drv/linux/alloc-r0drv-linux.c@ 50008

Last change on this file since 50008 was 50008, checked in by vboxsync, 11 years ago

supdrv,iprt: Added VBOX_WITH_TEXT_MODMEM_HACK for getting VMMR0.r0 address in the linux kernel stack crawl.

  • Property svn:eol-style set to native
  • Property svn:keywords set to Id Revision
File size: 14.2 KB
Line 
1/* $Id: alloc-r0drv-linux.c 50008 2013-12-27 14:20:34Z 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 */
76typedef 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;
84AssertCompileSize(RTMEMLNXHDREX, 32);
85/** Pointer to an extended memory header. */
86typedef RTMEMLNXHDREX *PRTMEMLNXHDREX;
87#endif
88
89
90/*******************************************************************************
91* Global Variables *
92*******************************************************************************/
93#ifdef RTMEMALLOC_EXEC_HEAP
94/** The heap. */
95static RTHEAPSIMPLE g_HeapExec = NIL_RTHEAPSIMPLE;
96/** Spinlock protecting the heap. */
97static 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 */
105DECLHIDDEN(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 */
130RTR0DECL(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}
148RT_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 */
160static 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);
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);
184 if (!papPages[iPage])
185 break;
186 }
187 if (iPage == cPages)
188 {
189 /*
190 * Map the pages. The API requires an iterator argument, which can be
191 * used, in case of failure, to figure out how much was actually
192 * mapped. Not sure how useful this really is, but whatever.
193 *
194 * Not entirely sure we really need to set nr_pages and pages here, but
195 * they provide a very convenient place for storing something we need
196 * in the free function, if nothing else...
197 */
198 struct page **papPagesIterator = papPages;
199 pVmArea->nr_pages = cPages;
200 pVmArea->pages = papPages;
201 if (!map_vm_area(pVmArea, PAGE_KERNEL_EXEC, &papPagesIterator))
202 {
203 PRTMEMLNXHDREX pHdrEx = (PRTMEMLNXHDREX)pVmArea->addr;
204 pHdrEx->pVmArea = pVmArea;
205 pHdrEx->pvDummy = NULL;
206 return &pHdrEx->Hdr;
207 }
208
209 /* bail out */
210 pVmArea->nr_pages = papPagesIterator - papPages;
211 }
212
213 vunmap(pVmArea->addr);
214
215 while (iPage-- > 0)
216 __free_page(papPages[iPage]);
217 kfree(papPages);
218
219 return NULL;
220}
221#endif /* RTMEMALLOC_EXEC_VM_AREA */
222
223
224/**
225 * OS specific allocation function.
226 */
227DECLHIDDEN(int) rtR0MemAllocEx(size_t cb, uint32_t fFlags, PRTMEMHDR *ppHdr)
228{
229 PRTMEMHDR pHdr;
230
231 /*
232 * Allocate.
233 */
234 if (fFlags & RTMEMHDR_FLAG_EXEC)
235 {
236 if (fFlags & RTMEMHDR_FLAG_ANY_CTX)
237 return VERR_NOT_SUPPORTED;
238
239#if defined(RT_ARCH_AMD64)
240# ifdef RTMEMALLOC_EXEC_HEAP
241 if (g_HeapExec != NIL_RTHEAPSIMPLE)
242 {
243 RTSpinlockAcquire(g_HeapExecSpinlock);
244 pHdr = (PRTMEMHDR)RTHeapSimpleAlloc(g_HeapExec, cb + sizeof(*pHdr), 0);
245 RTSpinlockRelease(g_HeapExecSpinlock);
246 fFlags |= RTMEMHDR_FLAG_EXEC_HEAP;
247 }
248 else
249 pHdr = NULL;
250
251# elif defined(RTMEMALLOC_EXEC_VM_AREA)
252 pHdr = rtR0MemAllocExecVmArea(cb);
253 fFlags |= RTMEMHDR_FLAG_EXEC_VM_AREA;
254
255# else /* !RTMEMALLOC_EXEC_HEAP */
256# error "you don not want to go here..."
257 pHdr = (PRTMEMHDR)__vmalloc(cb + sizeof(*pHdr), GFP_KERNEL | __GFP_HIGHMEM, MY_PAGE_KERNEL_EXEC);
258# endif /* !RTMEMALLOC_EXEC_HEAP */
259
260#elif defined(PAGE_KERNEL_EXEC) && defined(CONFIG_X86_PAE)
261 pHdr = (PRTMEMHDR)__vmalloc(cb + sizeof(*pHdr), GFP_KERNEL | __GFP_HIGHMEM, MY_PAGE_KERNEL_EXEC);
262#else
263 pHdr = (PRTMEMHDR)vmalloc(cb + sizeof(*pHdr));
264#endif
265 }
266 else
267 {
268 if (
269#if 1 /* vmalloc has serious performance issues, avoid it. */
270 cb <= PAGE_SIZE*16 - sizeof(*pHdr)
271#else
272 cb <= PAGE_SIZE
273#endif
274 || (fFlags & RTMEMHDR_FLAG_ANY_CTX)
275 )
276 {
277 fFlags |= RTMEMHDR_FLAG_KMALLOC;
278 pHdr = kmalloc(cb + sizeof(*pHdr),
279 (fFlags & RTMEMHDR_FLAG_ANY_CTX_ALLOC) ? GFP_ATOMIC : GFP_KERNEL);
280 if (RT_UNLIKELY( !pHdr
281 && cb > PAGE_SIZE
282 && !(fFlags & RTMEMHDR_FLAG_ANY_CTX) ))
283 {
284 fFlags &= ~RTMEMHDR_FLAG_KMALLOC;
285 pHdr = vmalloc(cb + sizeof(*pHdr));
286 }
287 }
288 else
289 pHdr = vmalloc(cb + sizeof(*pHdr));
290 }
291 if (RT_UNLIKELY(!pHdr))
292 return VERR_NO_MEMORY;
293
294 /*
295 * Initialize.
296 */
297 pHdr->u32Magic = RTMEMHDR_MAGIC;
298 pHdr->fFlags = fFlags;
299 pHdr->cb = cb;
300 pHdr->cbReq = cb;
301
302 *ppHdr = pHdr;
303 return VINF_SUCCESS;
304}
305
306
307/**
308 * OS specific free function.
309 */
310DECLHIDDEN(void) rtR0MemFree(PRTMEMHDR pHdr)
311{
312 pHdr->u32Magic += 1;
313 if (pHdr->fFlags & RTMEMHDR_FLAG_KMALLOC)
314 kfree(pHdr);
315#ifdef RTMEMALLOC_EXEC_HEAP
316 else if (pHdr->fFlags & RTMEMHDR_FLAG_EXEC_HEAP)
317 {
318 RTSpinlockAcquire(g_HeapExecSpinlock);
319 RTHeapSimpleFree(g_HeapExec, pHdr);
320 RTSpinlockRelease(g_HeapExecSpinlock);
321 }
322#endif
323#ifdef RTMEMALLOC_EXEC_VM_AREA
324 else if (pHdr->fFlags & RTMEMHDR_FLAG_EXEC_VM_AREA)
325 {
326 PRTMEMLNXHDREX pHdrEx = RT_FROM_MEMBER(pHdr, RTMEMLNXHDREX, Hdr);
327 size_t iPage = pHdrEx->pVmArea->nr_pages;
328 struct page **papPages = pHdrEx->pVmArea->pages;
329 void *pvMapping = pHdrEx->pVmArea->addr;
330
331 vunmap(pvMapping);
332
333 while (iPage-- > 0)
334 __free_page(papPages[iPage]);
335 kfree(papPages);
336 }
337#endif
338 else
339 vfree(pHdr);
340}
341
342
343
344/**
345 * Compute order. Some functions allocate 2^order pages.
346 *
347 * @returns order.
348 * @param cPages Number of pages.
349 */
350static int CalcPowerOf2Order(unsigned long cPages)
351{
352 int iOrder;
353 unsigned long cTmp;
354
355 for (iOrder = 0, cTmp = cPages; cTmp >>= 1; ++iOrder)
356 ;
357 if (cPages & ~(1 << iOrder))
358 ++iOrder;
359
360 return iOrder;
361}
362
363
364/**
365 * Allocates physical contiguous memory (below 4GB).
366 * The allocation is page aligned and the content is undefined.
367 *
368 * @returns Pointer to the memory block. This is page aligned.
369 * @param pPhys Where to store the physical address.
370 * @param cb The allocation size in bytes. This is always
371 * rounded up to PAGE_SIZE.
372 */
373RTR0DECL(void *) RTMemContAlloc(PRTCCPHYS pPhys, size_t cb)
374{
375 int cOrder;
376 unsigned cPages;
377 struct page *paPages;
378
379 /*
380 * validate input.
381 */
382 Assert(VALID_PTR(pPhys));
383 Assert(cb > 0);
384
385 /*
386 * Allocate page pointer array.
387 */
388 cb = RT_ALIGN_Z(cb, PAGE_SIZE);
389 cPages = cb >> PAGE_SHIFT;
390 cOrder = CalcPowerOf2Order(cPages);
391#if (defined(RT_ARCH_AMD64) || defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
392 /* ZONE_DMA32: 0-4GB */
393 paPages = alloc_pages(GFP_DMA32, cOrder);
394 if (!paPages)
395#endif
396#ifdef RT_ARCH_AMD64
397 /* ZONE_DMA; 0-16MB */
398 paPages = alloc_pages(GFP_DMA, cOrder);
399#else
400 /* ZONE_NORMAL: 0-896MB */
401 paPages = alloc_pages(GFP_USER, cOrder);
402#endif
403 if (paPages)
404 {
405 /*
406 * Reserve the pages and mark them executable.
407 */
408 unsigned iPage;
409 for (iPage = 0; iPage < cPages; iPage++)
410 {
411 Assert(!PageHighMem(&paPages[iPage]));
412 if (iPage + 1 < cPages)
413 {
414 AssertMsg( (uintptr_t)phys_to_virt(page_to_phys(&paPages[iPage])) + PAGE_SIZE
415 == (uintptr_t)phys_to_virt(page_to_phys(&paPages[iPage + 1]))
416 && page_to_phys(&paPages[iPage]) + PAGE_SIZE
417 == page_to_phys(&paPages[iPage + 1]),
418 ("iPage=%i cPages=%u [0]=%#llx,%p [1]=%#llx,%p\n", iPage, cPages,
419 (long long)page_to_phys(&paPages[iPage]), phys_to_virt(page_to_phys(&paPages[iPage])),
420 (long long)page_to_phys(&paPages[iPage + 1]), phys_to_virt(page_to_phys(&paPages[iPage + 1])) ));
421 }
422
423 SetPageReserved(&paPages[iPage]);
424#if LINUX_VERSION_CODE > KERNEL_VERSION(2, 4, 20) /** @todo find the exact kernel where change_page_attr was introduced. */
425 MY_SET_PAGES_EXEC(&paPages[iPage], 1);
426#endif
427 }
428 *pPhys = page_to_phys(paPages);
429 return phys_to_virt(page_to_phys(paPages));
430 }
431
432 return NULL;
433}
434RT_EXPORT_SYMBOL(RTMemContAlloc);
435
436
437/**
438 * Frees memory allocated using RTMemContAlloc().
439 *
440 * @param pv Pointer to return from RTMemContAlloc().
441 * @param cb The cb parameter passed to RTMemContAlloc().
442 */
443RTR0DECL(void) RTMemContFree(void *pv, size_t cb)
444{
445 if (pv)
446 {
447 int cOrder;
448 unsigned cPages;
449 unsigned iPage;
450 struct page *paPages;
451
452 /* validate */
453 AssertMsg(!((uintptr_t)pv & PAGE_OFFSET_MASK), ("pv=%p\n", pv));
454 Assert(cb > 0);
455
456 /* calc order and get pages */
457 cb = RT_ALIGN_Z(cb, PAGE_SIZE);
458 cPages = cb >> PAGE_SHIFT;
459 cOrder = CalcPowerOf2Order(cPages);
460 paPages = virt_to_page(pv);
461
462 /*
463 * Restore page attributes freeing the pages.
464 */
465 for (iPage = 0; iPage < cPages; iPage++)
466 {
467 ClearPageReserved(&paPages[iPage]);
468#if LINUX_VERSION_CODE > KERNEL_VERSION(2, 4, 20) /** @todo find the exact kernel where change_page_attr was introduced. */
469 MY_SET_PAGES_NOEXEC(&paPages[iPage], 1);
470#endif
471 }
472 __free_pages(paPages, cOrder);
473 }
474}
475RT_EXPORT_SYMBOL(RTMemContFree);
476
Note: See TracBrowser for help on using the repository browser.

© 2025 Oracle Support Privacy / Do Not Sell My Info Terms of Use Trademark Policy Automated Access Etiquette