1 | /* $Id: alloc-r0drv-linux.c 40806 2012-04-06 21:05:19Z vboxsync $ */
|
---|
2 | /** @file
|
---|
3 | * IPRT - Memory Allocation, Ring-0 Driver, Linux.
|
---|
4 | */
|
---|
5 |
|
---|
6 | /*
|
---|
7 | * Copyright (C) 2006-2010 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 | #if defined(RT_ARCH_AMD64) || defined(DOXYGEN_RUNNING)
|
---|
40 | /**
|
---|
41 | * We need memory in the module range (~2GB to ~0) this can only be obtained
|
---|
42 | * thru APIs that are not exported (see module_alloc()).
|
---|
43 | *
|
---|
44 | * So, we'll have to create a quick and dirty heap here using BSS memory.
|
---|
45 | * Very annoying and it's going to restrict us!
|
---|
46 | */
|
---|
47 | # define RTMEMALLOC_EXEC_HEAP
|
---|
48 | #endif
|
---|
49 | #ifdef RTMEMALLOC_EXEC_HEAP
|
---|
50 | # include <iprt/heap.h>
|
---|
51 | # include <iprt/spinlock.h>
|
---|
52 | # include <iprt/err.h>
|
---|
53 | #endif
|
---|
54 |
|
---|
55 |
|
---|
56 | /*******************************************************************************
|
---|
57 | * Global Variables *
|
---|
58 | *******************************************************************************/
|
---|
59 | #ifdef RTMEMALLOC_EXEC_HEAP
|
---|
60 |
|
---|
61 | # ifdef CONFIG_DEBUG_SET_MODULE_RONX
|
---|
62 | # define RTMEMALLOC_EXEC_HEAP_VM_AREA 1
|
---|
63 | # endif
|
---|
64 | /** The heap. */
|
---|
65 | static RTHEAPSIMPLE g_HeapExec = NIL_RTHEAPSIMPLE;
|
---|
66 | /** Spinlock protecting the heap. */
|
---|
67 | static RTSPINLOCK g_HeapExecSpinlock = NIL_RTSPINLOCK;
|
---|
68 | # ifdef RTMEMALLOC_EXEC_HEAP_VM_AREA
|
---|
69 | static struct page **g_apPages;
|
---|
70 | static void *g_pvHeap;
|
---|
71 | static size_t g_cPages;
|
---|
72 | # endif
|
---|
73 |
|
---|
74 |
|
---|
75 | /**
|
---|
76 | * API for cleaning up the heap spinlock on IPRT termination.
|
---|
77 | * This is as RTMemExecDonate specific to AMD64 Linux/GNU.
|
---|
78 | */
|
---|
79 | DECLHIDDEN(void) rtR0MemExecCleanup(void)
|
---|
80 | {
|
---|
81 | # ifdef RTMEMALLOC_EXEC_HEAP_VM_AREA
|
---|
82 | unsigned i;
|
---|
83 |
|
---|
84 | /* according to linux/drivers/lguest/core.c this function undoes
|
---|
85 | * map_vm_area() as well as __get_vm_area(). */
|
---|
86 | if (g_pvHeap)
|
---|
87 | vunmap(g_pvHeap);
|
---|
88 | for (i = 0; i < g_cPages; i++)
|
---|
89 | __free_page(g_apPages[i]);
|
---|
90 | kfree(g_apPages);
|
---|
91 | # endif
|
---|
92 |
|
---|
93 | RTSpinlockDestroy(g_HeapExecSpinlock);
|
---|
94 | g_HeapExecSpinlock = NIL_RTSPINLOCK;
|
---|
95 | }
|
---|
96 |
|
---|
97 |
|
---|
98 | # ifndef RTMEMALLOC_EXEC_HEAP_VM_AREA
|
---|
99 | /**
|
---|
100 | * Donate read+write+execute memory to the exec heap.
|
---|
101 | *
|
---|
102 | * This API is specific to AMD64 and Linux/GNU. A kernel module that desires to
|
---|
103 | * use RTMemExecAlloc on AMD64 Linux/GNU will have to donate some statically
|
---|
104 | * allocated memory in the module if it wishes for GCC generated code to work.
|
---|
105 | * GCC can only generate modules that work in the address range ~2GB to ~0
|
---|
106 | * currently.
|
---|
107 | *
|
---|
108 | * The API only accept one single donation.
|
---|
109 | *
|
---|
110 | * @returns IPRT status code.
|
---|
111 | * @param pvMemory Pointer to the memory block.
|
---|
112 | * @param cb The size of the memory block.
|
---|
113 | */
|
---|
114 | RTR0DECL(int) RTR0MemExecDonate(void *pvMemory, size_t cb)
|
---|
115 | {
|
---|
116 | int rc;
|
---|
117 | AssertReturn(g_HeapExec == NIL_RTHEAPSIMPLE, VERR_WRONG_ORDER);
|
---|
118 |
|
---|
119 | rc = RTSpinlockCreate(&g_HeapExecSpinlock, RTSPINLOCK_FLAGS_INTERRUPT_SAFE, "RTR0MemExecDonate");
|
---|
120 | if (RT_SUCCESS(rc))
|
---|
121 | {
|
---|
122 | rc = RTHeapSimpleInit(&g_HeapExec, pvMemory, cb);
|
---|
123 | if (RT_FAILURE(rc))
|
---|
124 | rtR0MemExecCleanup();
|
---|
125 | }
|
---|
126 | return rc;
|
---|
127 | }
|
---|
128 | RT_EXPORT_SYMBOL(RTR0MemExecDonate);
|
---|
129 |
|
---|
130 | # else /* !RTMEMALLOC_EXEC_HEAP_VM_AREA */
|
---|
131 |
|
---|
132 | /**
|
---|
133 | * RTR0MemExecDonate() does not work if CONFIG_DEBUG_SET_MODULE_RONX is enabled.
|
---|
134 | * In that case, allocate a VM area in the modules range and back it with kernel
|
---|
135 | * memory. Unfortunately __vmalloc_area() is not exported so we have to emulate
|
---|
136 | * it.
|
---|
137 | */
|
---|
138 | RTR0DECL(int) RTR0MemExecInit(size_t cb)
|
---|
139 | {
|
---|
140 | int rc;
|
---|
141 | struct vm_struct *area;
|
---|
142 | size_t cPages;
|
---|
143 | size_t cbPages;
|
---|
144 | unsigned i;
|
---|
145 | struct page **ppPages;
|
---|
146 |
|
---|
147 | AssertReturn(g_HeapExec == NIL_RTHEAPSIMPLE, VERR_WRONG_ORDER);
|
---|
148 |
|
---|
149 | rc = RTSpinlockCreate(&g_HeapExecSpinlock, RTSPINLOCK_FLAGS_INTERRUPT_SAFE, "RTR0MemExecInit");
|
---|
150 | if (RT_SUCCESS(rc))
|
---|
151 | {
|
---|
152 | cb = RT_ALIGN(cb, PAGE_SIZE);
|
---|
153 | area = __get_vm_area(cb, VM_ALLOC, MODULES_VADDR, MODULES_END);
|
---|
154 | if (!area)
|
---|
155 | {
|
---|
156 | rtR0MemExecCleanup();
|
---|
157 | return VERR_NO_MEMORY;
|
---|
158 | }
|
---|
159 | g_pvHeap = area->addr;
|
---|
160 | cPages = cb >> PAGE_SHIFT;
|
---|
161 | area->nr_pages = 0;
|
---|
162 | cbPages = cPages * sizeof(struct page *);
|
---|
163 | g_apPages = kmalloc(cbPages, GFP_KERNEL);
|
---|
164 | area->pages = g_apPages;
|
---|
165 | if (!g_apPages)
|
---|
166 | {
|
---|
167 | rtR0MemExecCleanup();
|
---|
168 | return VERR_NO_MEMORY;
|
---|
169 | }
|
---|
170 | memset(area->pages, 0, cbPages);
|
---|
171 | for (i = 0; i < cPages; i++)
|
---|
172 | {
|
---|
173 | g_apPages[i] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
|
---|
174 | if (!g_apPages[i])
|
---|
175 | {
|
---|
176 | area->nr_pages = i;
|
---|
177 | g_cPages = i;
|
---|
178 | rtR0MemExecCleanup();
|
---|
179 | return VERR_NO_MEMORY;
|
---|
180 | }
|
---|
181 | }
|
---|
182 | area->nr_pages = cPages;
|
---|
183 | g_cPages = i;
|
---|
184 | ppPages = g_apPages;
|
---|
185 | if (map_vm_area(area, PAGE_KERNEL_EXEC, &ppPages))
|
---|
186 | {
|
---|
187 | rtR0MemExecCleanup();
|
---|
188 | return VERR_NO_MEMORY;
|
---|
189 | }
|
---|
190 |
|
---|
191 | rc = RTHeapSimpleInit(&g_HeapExec, g_pvHeap, cb);
|
---|
192 | if (RT_FAILURE(rc))
|
---|
193 | rtR0MemExecCleanup();
|
---|
194 | }
|
---|
195 | return rc;
|
---|
196 | }
|
---|
197 | RT_EXPORT_SYMBOL(RTR0MemExecInit);
|
---|
198 | # endif /* RTMEMALLOC_EXEC_HEAP_VM_AREA */
|
---|
199 | #endif /* RTMEMALLOC_EXEC_HEAP */
|
---|
200 |
|
---|
201 |
|
---|
202 |
|
---|
203 | /**
|
---|
204 | * OS specific allocation function.
|
---|
205 | */
|
---|
206 | DECLHIDDEN(int) rtR0MemAllocEx(size_t cb, uint32_t fFlags, PRTMEMHDR *ppHdr)
|
---|
207 | {
|
---|
208 | PRTMEMHDR pHdr;
|
---|
209 |
|
---|
210 | /*
|
---|
211 | * Allocate.
|
---|
212 | */
|
---|
213 | if (fFlags & RTMEMHDR_FLAG_EXEC)
|
---|
214 | {
|
---|
215 | if (fFlags & RTMEMHDR_FLAG_ANY_CTX)
|
---|
216 | return VERR_NOT_SUPPORTED;
|
---|
217 |
|
---|
218 | #if defined(RT_ARCH_AMD64)
|
---|
219 | # ifdef RTMEMALLOC_EXEC_HEAP
|
---|
220 | if (g_HeapExec != NIL_RTHEAPSIMPLE)
|
---|
221 | {
|
---|
222 | RTSpinlockAcquire(g_HeapExecSpinlock);
|
---|
223 | pHdr = (PRTMEMHDR)RTHeapSimpleAlloc(g_HeapExec, cb + sizeof(*pHdr), 0);
|
---|
224 | RTSpinlockRelease(g_HeapExecSpinlock);
|
---|
225 | fFlags |= RTMEMHDR_FLAG_EXEC_HEAP;
|
---|
226 | }
|
---|
227 | else
|
---|
228 | pHdr = NULL;
|
---|
229 | # else /* !RTMEMALLOC_EXEC_HEAP */
|
---|
230 | pHdr = (PRTMEMHDR)__vmalloc(cb + sizeof(*pHdr), GFP_KERNEL | __GFP_HIGHMEM, MY_PAGE_KERNEL_EXEC);
|
---|
231 | # endif /* !RTMEMALLOC_EXEC_HEAP */
|
---|
232 |
|
---|
233 | #elif defined(PAGE_KERNEL_EXEC) && defined(CONFIG_X86_PAE)
|
---|
234 | pHdr = (PRTMEMHDR)__vmalloc(cb + sizeof(*pHdr), GFP_KERNEL | __GFP_HIGHMEM, MY_PAGE_KERNEL_EXEC);
|
---|
235 | #else
|
---|
236 | pHdr = (PRTMEMHDR)vmalloc(cb + sizeof(*pHdr));
|
---|
237 | #endif
|
---|
238 | }
|
---|
239 | else
|
---|
240 | {
|
---|
241 | if (
|
---|
242 | #if 1 /* vmalloc has serious performance issues, avoid it. */
|
---|
243 | cb <= PAGE_SIZE*16 - sizeof(*pHdr)
|
---|
244 | #else
|
---|
245 | cb <= PAGE_SIZE
|
---|
246 | #endif
|
---|
247 | || (fFlags & RTMEMHDR_FLAG_ANY_CTX)
|
---|
248 | )
|
---|
249 | {
|
---|
250 | fFlags |= RTMEMHDR_FLAG_KMALLOC;
|
---|
251 | pHdr = kmalloc(cb + sizeof(*pHdr),
|
---|
252 | (fFlags & RTMEMHDR_FLAG_ANY_CTX_ALLOC) ? GFP_ATOMIC : GFP_KERNEL);
|
---|
253 | if (RT_UNLIKELY( !pHdr
|
---|
254 | && cb > PAGE_SIZE
|
---|
255 | && !(fFlags & RTMEMHDR_FLAG_ANY_CTX) ))
|
---|
256 | {
|
---|
257 | fFlags &= ~RTMEMHDR_FLAG_KMALLOC;
|
---|
258 | pHdr = vmalloc(cb + sizeof(*pHdr));
|
---|
259 | }
|
---|
260 | }
|
---|
261 | else
|
---|
262 | pHdr = vmalloc(cb + sizeof(*pHdr));
|
---|
263 | }
|
---|
264 | if (RT_UNLIKELY(!pHdr))
|
---|
265 | return VERR_NO_MEMORY;
|
---|
266 |
|
---|
267 | /*
|
---|
268 | * Initialize.
|
---|
269 | */
|
---|
270 | pHdr->u32Magic = RTMEMHDR_MAGIC;
|
---|
271 | pHdr->fFlags = fFlags;
|
---|
272 | pHdr->cb = cb;
|
---|
273 | pHdr->cbReq = cb;
|
---|
274 |
|
---|
275 | *ppHdr = pHdr;
|
---|
276 | return VINF_SUCCESS;
|
---|
277 | }
|
---|
278 |
|
---|
279 |
|
---|
280 | /**
|
---|
281 | * OS specific free function.
|
---|
282 | */
|
---|
283 | DECLHIDDEN(void) rtR0MemFree(PRTMEMHDR pHdr)
|
---|
284 | {
|
---|
285 | pHdr->u32Magic += 1;
|
---|
286 | if (pHdr->fFlags & RTMEMHDR_FLAG_KMALLOC)
|
---|
287 | kfree(pHdr);
|
---|
288 | #ifdef RTMEMALLOC_EXEC_HEAP
|
---|
289 | else if (pHdr->fFlags & RTMEMHDR_FLAG_EXEC_HEAP)
|
---|
290 | {
|
---|
291 | RTSpinlockAcquire(g_HeapExecSpinlock);
|
---|
292 | RTHeapSimpleFree(g_HeapExec, pHdr);
|
---|
293 | RTSpinlockRelease(g_HeapExecSpinlock);
|
---|
294 | }
|
---|
295 | #endif
|
---|
296 | else
|
---|
297 | vfree(pHdr);
|
---|
298 | }
|
---|
299 |
|
---|
300 |
|
---|
301 | /**
|
---|
302 | * Compute order. Some functions allocate 2^order pages.
|
---|
303 | *
|
---|
304 | * @returns order.
|
---|
305 | * @param cPages Number of pages.
|
---|
306 | */
|
---|
307 | static int CalcPowerOf2Order(unsigned long cPages)
|
---|
308 | {
|
---|
309 | int iOrder;
|
---|
310 | unsigned long cTmp;
|
---|
311 |
|
---|
312 | for (iOrder = 0, cTmp = cPages; cTmp >>= 1; ++iOrder)
|
---|
313 | ;
|
---|
314 | if (cPages & ~(1 << iOrder))
|
---|
315 | ++iOrder;
|
---|
316 |
|
---|
317 | return iOrder;
|
---|
318 | }
|
---|
319 |
|
---|
320 |
|
---|
321 | /**
|
---|
322 | * Allocates physical contiguous memory (below 4GB).
|
---|
323 | * The allocation is page aligned and the content is undefined.
|
---|
324 | *
|
---|
325 | * @returns Pointer to the memory block. This is page aligned.
|
---|
326 | * @param pPhys Where to store the physical address.
|
---|
327 | * @param cb The allocation size in bytes. This is always
|
---|
328 | * rounded up to PAGE_SIZE.
|
---|
329 | */
|
---|
330 | RTR0DECL(void *) RTMemContAlloc(PRTCCPHYS pPhys, size_t cb)
|
---|
331 | {
|
---|
332 | int cOrder;
|
---|
333 | unsigned cPages;
|
---|
334 | struct page *paPages;
|
---|
335 |
|
---|
336 | /*
|
---|
337 | * validate input.
|
---|
338 | */
|
---|
339 | Assert(VALID_PTR(pPhys));
|
---|
340 | Assert(cb > 0);
|
---|
341 |
|
---|
342 | /*
|
---|
343 | * Allocate page pointer array.
|
---|
344 | */
|
---|
345 | cb = RT_ALIGN_Z(cb, PAGE_SIZE);
|
---|
346 | cPages = cb >> PAGE_SHIFT;
|
---|
347 | cOrder = CalcPowerOf2Order(cPages);
|
---|
348 | #if (defined(RT_ARCH_AMD64) || defined(CONFIG_X86_PAE)) && defined(GFP_DMA32)
|
---|
349 | /* ZONE_DMA32: 0-4GB */
|
---|
350 | paPages = alloc_pages(GFP_DMA32, cOrder);
|
---|
351 | if (!paPages)
|
---|
352 | #endif
|
---|
353 | #ifdef RT_ARCH_AMD64
|
---|
354 | /* ZONE_DMA; 0-16MB */
|
---|
355 | paPages = alloc_pages(GFP_DMA, cOrder);
|
---|
356 | #else
|
---|
357 | /* ZONE_NORMAL: 0-896MB */
|
---|
358 | paPages = alloc_pages(GFP_USER, cOrder);
|
---|
359 | #endif
|
---|
360 | if (paPages)
|
---|
361 | {
|
---|
362 | /*
|
---|
363 | * Reserve the pages and mark them executable.
|
---|
364 | */
|
---|
365 | unsigned iPage;
|
---|
366 | for (iPage = 0; iPage < cPages; iPage++)
|
---|
367 | {
|
---|
368 | Assert(!PageHighMem(&paPages[iPage]));
|
---|
369 | if (iPage + 1 < cPages)
|
---|
370 | {
|
---|
371 | AssertMsg( (uintptr_t)phys_to_virt(page_to_phys(&paPages[iPage])) + PAGE_SIZE
|
---|
372 | == (uintptr_t)phys_to_virt(page_to_phys(&paPages[iPage + 1]))
|
---|
373 | && page_to_phys(&paPages[iPage]) + PAGE_SIZE
|
---|
374 | == page_to_phys(&paPages[iPage + 1]),
|
---|
375 | ("iPage=%i cPages=%u [0]=%#llx,%p [1]=%#llx,%p\n", iPage, cPages,
|
---|
376 | (long long)page_to_phys(&paPages[iPage]), phys_to_virt(page_to_phys(&paPages[iPage])),
|
---|
377 | (long long)page_to_phys(&paPages[iPage + 1]), phys_to_virt(page_to_phys(&paPages[iPage + 1])) ));
|
---|
378 | }
|
---|
379 |
|
---|
380 | SetPageReserved(&paPages[iPage]);
|
---|
381 | #if LINUX_VERSION_CODE > KERNEL_VERSION(2, 4, 20) /** @todo find the exact kernel where change_page_attr was introduced. */
|
---|
382 | MY_SET_PAGES_EXEC(&paPages[iPage], 1);
|
---|
383 | #endif
|
---|
384 | }
|
---|
385 | *pPhys = page_to_phys(paPages);
|
---|
386 | return phys_to_virt(page_to_phys(paPages));
|
---|
387 | }
|
---|
388 |
|
---|
389 | return NULL;
|
---|
390 | }
|
---|
391 | RT_EXPORT_SYMBOL(RTMemContAlloc);
|
---|
392 |
|
---|
393 |
|
---|
394 | /**
|
---|
395 | * Frees memory allocated ysing RTMemContAlloc().
|
---|
396 | *
|
---|
397 | * @param pv Pointer to return from RTMemContAlloc().
|
---|
398 | * @param cb The cb parameter passed to RTMemContAlloc().
|
---|
399 | */
|
---|
400 | RTR0DECL(void) RTMemContFree(void *pv, size_t cb)
|
---|
401 | {
|
---|
402 | if (pv)
|
---|
403 | {
|
---|
404 | int cOrder;
|
---|
405 | unsigned cPages;
|
---|
406 | unsigned iPage;
|
---|
407 | struct page *paPages;
|
---|
408 |
|
---|
409 | /* validate */
|
---|
410 | AssertMsg(!((uintptr_t)pv & PAGE_OFFSET_MASK), ("pv=%p\n", pv));
|
---|
411 | Assert(cb > 0);
|
---|
412 |
|
---|
413 | /* calc order and get pages */
|
---|
414 | cb = RT_ALIGN_Z(cb, PAGE_SIZE);
|
---|
415 | cPages = cb >> PAGE_SHIFT;
|
---|
416 | cOrder = CalcPowerOf2Order(cPages);
|
---|
417 | paPages = virt_to_page(pv);
|
---|
418 |
|
---|
419 | /*
|
---|
420 | * Restore page attributes freeing the pages.
|
---|
421 | */
|
---|
422 | for (iPage = 0; iPage < cPages; iPage++)
|
---|
423 | {
|
---|
424 | ClearPageReserved(&paPages[iPage]);
|
---|
425 | #if LINUX_VERSION_CODE > KERNEL_VERSION(2, 4, 20) /** @todo find the exact kernel where change_page_attr was introduced. */
|
---|
426 | MY_SET_PAGES_NOEXEC(&paPages[iPage], 1);
|
---|
427 | #endif
|
---|
428 | }
|
---|
429 | __free_pages(paPages, cOrder);
|
---|
430 | }
|
---|
431 | }
|
---|
432 | RT_EXPORT_SYMBOL(RTMemContFree);
|
---|
433 |
|
---|