diff options
Diffstat (limited to 'arch/x86/kernel/cpu/mtrr/main.c')
-rw-r--r-- | arch/x86/kernel/cpu/mtrr/main.c | 768 |
1 files changed, 768 insertions, 0 deletions
diff --git a/arch/x86/kernel/cpu/mtrr/main.c b/arch/x86/kernel/cpu/mtrr/main.c new file mode 100644 index 00000000000..c48b6fea5ab --- /dev/null +++ b/arch/x86/kernel/cpu/mtrr/main.c | |||
@@ -0,0 +1,768 @@ | |||
1 | /* Generic MTRR (Memory Type Range Register) driver. | ||
2 | |||
3 | Copyright (C) 1997-2000 Richard Gooch | ||
4 | Copyright (c) 2002 Patrick Mochel | ||
5 | |||
6 | This library is free software; you can redistribute it and/or | ||
7 | modify it under the terms of the GNU Library General Public | ||
8 | License as published by the Free Software Foundation; either | ||
9 | version 2 of the License, or (at your option) any later version. | ||
10 | |||
11 | This library is distributed in the hope that it will be useful, | ||
12 | but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
14 | Library General Public License for more details. | ||
15 | |||
16 | You should have received a copy of the GNU Library General Public | ||
17 | License along with this library; if not, write to the Free | ||
18 | Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | ||
19 | |||
20 | Richard Gooch may be reached by email at rgooch@atnf.csiro.au | ||
21 | The postal address is: | ||
22 | Richard Gooch, c/o ATNF, P. O. Box 76, Epping, N.S.W., 2121, Australia. | ||
23 | |||
24 | Source: "Pentium Pro Family Developer's Manual, Volume 3: | ||
25 | Operating System Writer's Guide" (Intel document number 242692), | ||
26 | section 11.11.7 | ||
27 | |||
28 | This was cleaned and made readable by Patrick Mochel <mochel@osdl.org> | ||
29 | on 6-7 March 2002. | ||
30 | Source: Intel Architecture Software Developers Manual, Volume 3: | ||
31 | System Programming Guide; Section 9.11. (1997 edition - PPro). | ||
32 | */ | ||
33 | |||
34 | #include <linux/module.h> | ||
35 | #include <linux/init.h> | ||
36 | #include <linux/pci.h> | ||
37 | #include <linux/smp.h> | ||
38 | #include <linux/cpu.h> | ||
39 | #include <linux/mutex.h> | ||
40 | |||
41 | #include <asm/mtrr.h> | ||
42 | |||
43 | #include <asm/uaccess.h> | ||
44 | #include <asm/processor.h> | ||
45 | #include <asm/msr.h> | ||
46 | #include "mtrr.h" | ||
47 | |||
48 | u32 num_var_ranges = 0; | ||
49 | |||
50 | unsigned int *usage_table; | ||
51 | static DEFINE_MUTEX(mtrr_mutex); | ||
52 | |||
53 | u64 size_or_mask, size_and_mask; | ||
54 | |||
55 | static struct mtrr_ops * mtrr_ops[X86_VENDOR_NUM] = {}; | ||
56 | |||
57 | struct mtrr_ops * mtrr_if = NULL; | ||
58 | |||
59 | static void set_mtrr(unsigned int reg, unsigned long base, | ||
60 | unsigned long size, mtrr_type type); | ||
61 | |||
62 | #ifndef CONFIG_X86_64 | ||
63 | extern int arr3_protected; | ||
64 | #else | ||
65 | #define arr3_protected 0 | ||
66 | #endif | ||
67 | |||
68 | void set_mtrr_ops(struct mtrr_ops * ops) | ||
69 | { | ||
70 | if (ops->vendor && ops->vendor < X86_VENDOR_NUM) | ||
71 | mtrr_ops[ops->vendor] = ops; | ||
72 | } | ||
73 | |||
74 | /* Returns non-zero if we have the write-combining memory type */ | ||
75 | static int have_wrcomb(void) | ||
76 | { | ||
77 | struct pci_dev *dev; | ||
78 | u8 rev; | ||
79 | |||
80 | if ((dev = pci_get_class(PCI_CLASS_BRIDGE_HOST << 8, NULL)) != NULL) { | ||
81 | /* ServerWorks LE chipsets < rev 6 have problems with write-combining | ||
82 | Don't allow it and leave room for other chipsets to be tagged */ | ||
83 | if (dev->vendor == PCI_VENDOR_ID_SERVERWORKS && | ||
84 | dev->device == PCI_DEVICE_ID_SERVERWORKS_LE) { | ||
85 | pci_read_config_byte(dev, PCI_CLASS_REVISION, &rev); | ||
86 | if (rev <= 5) { | ||
87 | printk(KERN_INFO "mtrr: Serverworks LE rev < 6 detected. Write-combining disabled.\n"); | ||
88 | pci_dev_put(dev); | ||
89 | return 0; | ||
90 | } | ||
91 | } | ||
92 | /* Intel 450NX errata # 23. Non ascending cacheline evictions to | ||
93 | write combining memory may resulting in data corruption */ | ||
94 | if (dev->vendor == PCI_VENDOR_ID_INTEL && | ||
95 | dev->device == PCI_DEVICE_ID_INTEL_82451NX) { | ||
96 | printk(KERN_INFO "mtrr: Intel 450NX MMC detected. Write-combining disabled.\n"); | ||
97 | pci_dev_put(dev); | ||
98 | return 0; | ||
99 | } | ||
100 | pci_dev_put(dev); | ||
101 | } | ||
102 | return (mtrr_if->have_wrcomb ? mtrr_if->have_wrcomb() : 0); | ||
103 | } | ||
104 | |||
105 | /* This function returns the number of variable MTRRs */ | ||
106 | static void __init set_num_var_ranges(void) | ||
107 | { | ||
108 | unsigned long config = 0, dummy; | ||
109 | |||
110 | if (use_intel()) { | ||
111 | rdmsr(MTRRcap_MSR, config, dummy); | ||
112 | } else if (is_cpu(AMD)) | ||
113 | config = 2; | ||
114 | else if (is_cpu(CYRIX) || is_cpu(CENTAUR)) | ||
115 | config = 8; | ||
116 | num_var_ranges = config & 0xff; | ||
117 | } | ||
118 | |||
119 | static void __init init_table(void) | ||
120 | { | ||
121 | int i, max; | ||
122 | |||
123 | max = num_var_ranges; | ||
124 | if ((usage_table = kmalloc(max * sizeof *usage_table, GFP_KERNEL)) | ||
125 | == NULL) { | ||
126 | printk(KERN_ERR "mtrr: could not allocate\n"); | ||
127 | return; | ||
128 | } | ||
129 | for (i = 0; i < max; i++) | ||
130 | usage_table[i] = 1; | ||
131 | } | ||
132 | |||
133 | struct set_mtrr_data { | ||
134 | atomic_t count; | ||
135 | atomic_t gate; | ||
136 | unsigned long smp_base; | ||
137 | unsigned long smp_size; | ||
138 | unsigned int smp_reg; | ||
139 | mtrr_type smp_type; | ||
140 | }; | ||
141 | |||
142 | #ifdef CONFIG_SMP | ||
143 | |||
144 | static void ipi_handler(void *info) | ||
145 | /* [SUMMARY] Synchronisation handler. Executed by "other" CPUs. | ||
146 | [RETURNS] Nothing. | ||
147 | */ | ||
148 | { | ||
149 | struct set_mtrr_data *data = info; | ||
150 | unsigned long flags; | ||
151 | |||
152 | local_irq_save(flags); | ||
153 | |||
154 | atomic_dec(&data->count); | ||
155 | while(!atomic_read(&data->gate)) | ||
156 | cpu_relax(); | ||
157 | |||
158 | /* The master has cleared me to execute */ | ||
159 | if (data->smp_reg != ~0U) | ||
160 | mtrr_if->set(data->smp_reg, data->smp_base, | ||
161 | data->smp_size, data->smp_type); | ||
162 | else | ||
163 | mtrr_if->set_all(); | ||
164 | |||
165 | atomic_dec(&data->count); | ||
166 | while(atomic_read(&data->gate)) | ||
167 | cpu_relax(); | ||
168 | |||
169 | atomic_dec(&data->count); | ||
170 | local_irq_restore(flags); | ||
171 | } | ||
172 | |||
173 | #endif | ||
174 | |||
175 | static inline int types_compatible(mtrr_type type1, mtrr_type type2) { | ||
176 | return type1 == MTRR_TYPE_UNCACHABLE || | ||
177 | type2 == MTRR_TYPE_UNCACHABLE || | ||
178 | (type1 == MTRR_TYPE_WRTHROUGH && type2 == MTRR_TYPE_WRBACK) || | ||
179 | (type1 == MTRR_TYPE_WRBACK && type2 == MTRR_TYPE_WRTHROUGH); | ||
180 | } | ||
181 | |||
182 | /** | ||
183 | * set_mtrr - update mtrrs on all processors | ||
184 | * @reg: mtrr in question | ||
185 | * @base: mtrr base | ||
186 | * @size: mtrr size | ||
187 | * @type: mtrr type | ||
188 | * | ||
189 | * This is kinda tricky, but fortunately, Intel spelled it out for us cleanly: | ||
190 | * | ||
191 | * 1. Send IPI to do the following: | ||
192 | * 2. Disable Interrupts | ||
193 | * 3. Wait for all procs to do so | ||
194 | * 4. Enter no-fill cache mode | ||
195 | * 5. Flush caches | ||
196 | * 6. Clear PGE bit | ||
197 | * 7. Flush all TLBs | ||
198 | * 8. Disable all range registers | ||
199 | * 9. Update the MTRRs | ||
200 | * 10. Enable all range registers | ||
201 | * 11. Flush all TLBs and caches again | ||
202 | * 12. Enter normal cache mode and reenable caching | ||
203 | * 13. Set PGE | ||
204 | * 14. Wait for buddies to catch up | ||
205 | * 15. Enable interrupts. | ||
206 | * | ||
207 | * What does that mean for us? Well, first we set data.count to the number | ||
208 | * of CPUs. As each CPU disables interrupts, it'll decrement it once. We wait | ||
209 | * until it hits 0 and proceed. We set the data.gate flag and reset data.count. | ||
210 | * Meanwhile, they are waiting for that flag to be set. Once it's set, each | ||
211 | * CPU goes through the transition of updating MTRRs. The CPU vendors may each do it | ||
212 | * differently, so we call mtrr_if->set() callback and let them take care of it. | ||
213 | * When they're done, they again decrement data->count and wait for data.gate to | ||
214 | * be reset. | ||
215 | * When we finish, we wait for data.count to hit 0 and toggle the data.gate flag. | ||
216 | * Everyone then enables interrupts and we all continue on. | ||
217 | * | ||
218 | * Note that the mechanism is the same for UP systems, too; all the SMP stuff | ||
219 | * becomes nops. | ||
220 | */ | ||
221 | static void set_mtrr(unsigned int reg, unsigned long base, | ||
222 | unsigned long size, mtrr_type type) | ||
223 | { | ||
224 | struct set_mtrr_data data; | ||
225 | unsigned long flags; | ||
226 | |||
227 | data.smp_reg = reg; | ||
228 | data.smp_base = base; | ||
229 | data.smp_size = size; | ||
230 | data.smp_type = type; | ||
231 | atomic_set(&data.count, num_booting_cpus() - 1); | ||
232 | /* make sure data.count is visible before unleashing other CPUs */ | ||
233 | smp_wmb(); | ||
234 | atomic_set(&data.gate,0); | ||
235 | |||
236 | /* Start the ball rolling on other CPUs */ | ||
237 | if (smp_call_function(ipi_handler, &data, 1, 0) != 0) | ||
238 | panic("mtrr: timed out waiting for other CPUs\n"); | ||
239 | |||
240 | local_irq_save(flags); | ||
241 | |||
242 | while(atomic_read(&data.count)) | ||
243 | cpu_relax(); | ||
244 | |||
245 | /* ok, reset count and toggle gate */ | ||
246 | atomic_set(&data.count, num_booting_cpus() - 1); | ||
247 | smp_wmb(); | ||
248 | atomic_set(&data.gate,1); | ||
249 | |||
250 | /* do our MTRR business */ | ||
251 | |||
252 | /* HACK! | ||
253 | * We use this same function to initialize the mtrrs on boot. | ||
254 | * The state of the boot cpu's mtrrs has been saved, and we want | ||
255 | * to replicate across all the APs. | ||
256 | * If we're doing that @reg is set to something special... | ||
257 | */ | ||
258 | if (reg != ~0U) | ||
259 | mtrr_if->set(reg,base,size,type); | ||
260 | |||
261 | /* wait for the others */ | ||
262 | while(atomic_read(&data.count)) | ||
263 | cpu_relax(); | ||
264 | |||
265 | atomic_set(&data.count, num_booting_cpus() - 1); | ||
266 | smp_wmb(); | ||
267 | atomic_set(&data.gate,0); | ||
268 | |||
269 | /* | ||
270 | * Wait here for everyone to have seen the gate change | ||
271 | * So we're the last ones to touch 'data' | ||
272 | */ | ||
273 | while(atomic_read(&data.count)) | ||
274 | cpu_relax(); | ||
275 | |||
276 | local_irq_restore(flags); | ||
277 | } | ||
278 | |||
279 | /** | ||
280 | * mtrr_add_page - Add a memory type region | ||
281 | * @base: Physical base address of region in pages (in units of 4 kB!) | ||
282 | * @size: Physical size of region in pages (4 kB) | ||
283 | * @type: Type of MTRR desired | ||
284 | * @increment: If this is true do usage counting on the region | ||
285 | * | ||
286 | * Memory type region registers control the caching on newer Intel and | ||
287 | * non Intel processors. This function allows drivers to request an | ||
288 | * MTRR is added. The details and hardware specifics of each processor's | ||
289 | * implementation are hidden from the caller, but nevertheless the | ||
290 | * caller should expect to need to provide a power of two size on an | ||
291 | * equivalent power of two boundary. | ||
292 | * | ||
293 | * If the region cannot be added either because all regions are in use | ||
294 | * or the CPU cannot support it a negative value is returned. On success | ||
295 | * the register number for this entry is returned, but should be treated | ||
296 | * as a cookie only. | ||
297 | * | ||
298 | * On a multiprocessor machine the changes are made to all processors. | ||
299 | * This is required on x86 by the Intel processors. | ||
300 | * | ||
301 | * The available types are | ||
302 | * | ||
303 | * %MTRR_TYPE_UNCACHABLE - No caching | ||
304 | * | ||
305 | * %MTRR_TYPE_WRBACK - Write data back in bursts whenever | ||
306 | * | ||
307 | * %MTRR_TYPE_WRCOMB - Write data back soon but allow bursts | ||
308 | * | ||
309 | * %MTRR_TYPE_WRTHROUGH - Cache reads but not writes | ||
310 | * | ||
311 | * BUGS: Needs a quiet flag for the cases where drivers do not mind | ||
312 | * failures and do not wish system log messages to be sent. | ||
313 | */ | ||
314 | |||
315 | int mtrr_add_page(unsigned long base, unsigned long size, | ||
316 | unsigned int type, char increment) | ||
317 | { | ||
318 | int i, replace, error; | ||
319 | mtrr_type ltype; | ||
320 | unsigned long lbase, lsize; | ||
321 | |||
322 | if (!mtrr_if) | ||
323 | return -ENXIO; | ||
324 | |||
325 | if ((error = mtrr_if->validate_add_page(base,size,type))) | ||
326 | return error; | ||
327 | |||
328 | if (type >= MTRR_NUM_TYPES) { | ||
329 | printk(KERN_WARNING "mtrr: type: %u invalid\n", type); | ||
330 | return -EINVAL; | ||
331 | } | ||
332 | |||
333 | /* If the type is WC, check that this processor supports it */ | ||
334 | if ((type == MTRR_TYPE_WRCOMB) && !have_wrcomb()) { | ||
335 | printk(KERN_WARNING | ||
336 | "mtrr: your processor doesn't support write-combining\n"); | ||
337 | return -ENOSYS; | ||
338 | } | ||
339 | |||
340 | if (!size) { | ||
341 | printk(KERN_WARNING "mtrr: zero sized request\n"); | ||
342 | return -EINVAL; | ||
343 | } | ||
344 | |||
345 | if (base & size_or_mask || size & size_or_mask) { | ||
346 | printk(KERN_WARNING "mtrr: base or size exceeds the MTRR width\n"); | ||
347 | return -EINVAL; | ||
348 | } | ||
349 | |||
350 | error = -EINVAL; | ||
351 | replace = -1; | ||
352 | |||
353 | /* No CPU hotplug when we change MTRR entries */ | ||
354 | lock_cpu_hotplug(); | ||
355 | /* Search for existing MTRR */ | ||
356 | mutex_lock(&mtrr_mutex); | ||
357 | for (i = 0; i < num_var_ranges; ++i) { | ||
358 | mtrr_if->get(i, &lbase, &lsize, <ype); | ||
359 | if (!lsize || base > lbase + lsize - 1 || base + size - 1 < lbase) | ||
360 | continue; | ||
361 | /* At this point we know there is some kind of overlap/enclosure */ | ||
362 | if (base < lbase || base + size - 1 > lbase + lsize - 1) { | ||
363 | if (base <= lbase && base + size - 1 >= lbase + lsize - 1) { | ||
364 | /* New region encloses an existing region */ | ||
365 | if (type == ltype) { | ||
366 | replace = replace == -1 ? i : -2; | ||
367 | continue; | ||
368 | } | ||
369 | else if (types_compatible(type, ltype)) | ||
370 | continue; | ||
371 | } | ||
372 | printk(KERN_WARNING | ||
373 | "mtrr: 0x%lx000,0x%lx000 overlaps existing" | ||
374 | " 0x%lx000,0x%lx000\n", base, size, lbase, | ||
375 | lsize); | ||
376 | goto out; | ||
377 | } | ||
378 | /* New region is enclosed by an existing region */ | ||
379 | if (ltype != type) { | ||
380 | if (types_compatible(type, ltype)) | ||
381 | continue; | ||
382 | printk (KERN_WARNING "mtrr: type mismatch for %lx000,%lx000 old: %s new: %s\n", | ||
383 | base, size, mtrr_attrib_to_str(ltype), | ||
384 | mtrr_attrib_to_str(type)); | ||
385 | goto out; | ||
386 | } | ||
387 | if (increment) | ||
388 | ++usage_table[i]; | ||
389 | error = i; | ||
390 | goto out; | ||
391 | } | ||
392 | /* Search for an empty MTRR */ | ||
393 | i = mtrr_if->get_free_region(base, size, replace); | ||
394 | if (i >= 0) { | ||
395 | set_mtrr(i, base, size, type); | ||
396 | if (likely(replace < 0)) | ||
397 | usage_table[i] = 1; | ||
398 | else { | ||
399 | usage_table[i] = usage_table[replace] + !!increment; | ||
400 | if (unlikely(replace != i)) { | ||
401 | set_mtrr(replace, 0, 0, 0); | ||
402 | usage_table[replace] = 0; | ||
403 | } | ||
404 | } | ||
405 | } else | ||
406 | printk(KERN_INFO "mtrr: no more MTRRs available\n"); | ||
407 | error = i; | ||
408 | out: | ||
409 | mutex_unlock(&mtrr_mutex); | ||
410 | unlock_cpu_hotplug(); | ||
411 | return error; | ||
412 | } | ||
413 | |||
414 | static int mtrr_check(unsigned long base, unsigned long size) | ||
415 | { | ||
416 | if ((base & (PAGE_SIZE - 1)) || (size & (PAGE_SIZE - 1))) { | ||
417 | printk(KERN_WARNING | ||
418 | "mtrr: size and base must be multiples of 4 kiB\n"); | ||
419 | printk(KERN_DEBUG | ||
420 | "mtrr: size: 0x%lx base: 0x%lx\n", size, base); | ||
421 | dump_stack(); | ||
422 | return -1; | ||
423 | } | ||
424 | return 0; | ||
425 | } | ||
426 | |||
427 | /** | ||
428 | * mtrr_add - Add a memory type region | ||
429 | * @base: Physical base address of region | ||
430 | * @size: Physical size of region | ||
431 | * @type: Type of MTRR desired | ||
432 | * @increment: If this is true do usage counting on the region | ||
433 | * | ||
434 | * Memory type region registers control the caching on newer Intel and | ||
435 | * non Intel processors. This function allows drivers to request an | ||
436 | * MTRR is added. The details and hardware specifics of each processor's | ||
437 | * implementation are hidden from the caller, but nevertheless the | ||
438 | * caller should expect to need to provide a power of two size on an | ||
439 | * equivalent power of two boundary. | ||
440 | * | ||
441 | * If the region cannot be added either because all regions are in use | ||
442 | * or the CPU cannot support it a negative value is returned. On success | ||
443 | * the register number for this entry is returned, but should be treated | ||
444 | * as a cookie only. | ||
445 | * | ||
446 | * On a multiprocessor machine the changes are made to all processors. | ||
447 | * This is required on x86 by the Intel processors. | ||
448 | * | ||
449 | * The available types are | ||
450 | * | ||
451 | * %MTRR_TYPE_UNCACHABLE - No caching | ||
452 | * | ||
453 | * %MTRR_TYPE_WRBACK - Write data back in bursts whenever | ||
454 | * | ||
455 | * %MTRR_TYPE_WRCOMB - Write data back soon but allow bursts | ||
456 | * | ||
457 | * %MTRR_TYPE_WRTHROUGH - Cache reads but not writes | ||
458 | * | ||
459 | * BUGS: Needs a quiet flag for the cases where drivers do not mind | ||
460 | * failures and do not wish system log messages to be sent. | ||
461 | */ | ||
462 | |||
463 | int | ||
464 | mtrr_add(unsigned long base, unsigned long size, unsigned int type, | ||
465 | char increment) | ||
466 | { | ||
467 | if (mtrr_check(base, size)) | ||
468 | return -EINVAL; | ||
469 | return mtrr_add_page(base >> PAGE_SHIFT, size >> PAGE_SHIFT, type, | ||
470 | increment); | ||
471 | } | ||
472 | |||
473 | /** | ||
474 | * mtrr_del_page - delete a memory type region | ||
475 | * @reg: Register returned by mtrr_add | ||
476 | * @base: Physical base address | ||
477 | * @size: Size of region | ||
478 | * | ||
479 | * If register is supplied then base and size are ignored. This is | ||
480 | * how drivers should call it. | ||
481 | * | ||
482 | * Releases an MTRR region. If the usage count drops to zero the | ||
483 | * register is freed and the region returns to default state. | ||
484 | * On success the register is returned, on failure a negative error | ||
485 | * code. | ||
486 | */ | ||
487 | |||
488 | int mtrr_del_page(int reg, unsigned long base, unsigned long size) | ||
489 | { | ||
490 | int i, max; | ||
491 | mtrr_type ltype; | ||
492 | unsigned long lbase, lsize; | ||
493 | int error = -EINVAL; | ||
494 | |||
495 | if (!mtrr_if) | ||
496 | return -ENXIO; | ||
497 | |||
498 | max = num_var_ranges; | ||
499 | /* No CPU hotplug when we change MTRR entries */ | ||
500 | lock_cpu_hotplug(); | ||
501 | mutex_lock(&mtrr_mutex); | ||
502 | if (reg < 0) { | ||
503 | /* Search for existing MTRR */ | ||
504 | for (i = 0; i < max; ++i) { | ||
505 | mtrr_if->get(i, &lbase, &lsize, <ype); | ||
506 | if (lbase == base && lsize == size) { | ||
507 | reg = i; | ||
508 | break; | ||
509 | } | ||
510 | } | ||
511 | if (reg < 0) { | ||
512 | printk(KERN_DEBUG "mtrr: no MTRR for %lx000,%lx000 found\n", base, | ||
513 | size); | ||
514 | goto out; | ||
515 | } | ||
516 | } | ||
517 | if (reg >= max) { | ||
518 | printk(KERN_WARNING "mtrr: register: %d too big\n", reg); | ||
519 | goto out; | ||
520 | } | ||
521 | if (is_cpu(CYRIX) && !use_intel()) { | ||
522 | if ((reg == 3) && arr3_protected) { | ||
523 | printk(KERN_WARNING "mtrr: ARR3 cannot be changed\n"); | ||
524 | goto out; | ||
525 | } | ||
526 | } | ||
527 | mtrr_if->get(reg, &lbase, &lsize, <ype); | ||
528 | if (lsize < 1) { | ||
529 | printk(KERN_WARNING "mtrr: MTRR %d not used\n", reg); | ||
530 | goto out; | ||
531 | } | ||
532 | if (usage_table[reg] < 1) { | ||
533 | printk(KERN_WARNING "mtrr: reg: %d has count=0\n", reg); | ||
534 | goto out; | ||
535 | } | ||
536 | if (--usage_table[reg] < 1) | ||
537 | set_mtrr(reg, 0, 0, 0); | ||
538 | error = reg; | ||
539 | out: | ||
540 | mutex_unlock(&mtrr_mutex); | ||
541 | unlock_cpu_hotplug(); | ||
542 | return error; | ||
543 | } | ||
544 | /** | ||
545 | * mtrr_del - delete a memory type region | ||
546 | * @reg: Register returned by mtrr_add | ||
547 | * @base: Physical base address | ||
548 | * @size: Size of region | ||
549 | * | ||
550 | * If register is supplied then base and size are ignored. This is | ||
551 | * how drivers should call it. | ||
552 | * | ||
553 | * Releases an MTRR region. If the usage count drops to zero the | ||
554 | * register is freed and the region returns to default state. | ||
555 | * On success the register is returned, on failure a negative error | ||
556 | * code. | ||
557 | */ | ||
558 | |||
559 | int | ||
560 | mtrr_del(int reg, unsigned long base, unsigned long size) | ||
561 | { | ||
562 | if (mtrr_check(base, size)) | ||
563 | return -EINVAL; | ||
564 | return mtrr_del_page(reg, base >> PAGE_SHIFT, size >> PAGE_SHIFT); | ||
565 | } | ||
566 | |||
567 | EXPORT_SYMBOL(mtrr_add); | ||
568 | EXPORT_SYMBOL(mtrr_del); | ||
569 | |||
570 | /* HACK ALERT! | ||
571 | * These should be called implicitly, but we can't yet until all the initcall | ||
572 | * stuff is done... | ||
573 | */ | ||
574 | extern void amd_init_mtrr(void); | ||
575 | extern void cyrix_init_mtrr(void); | ||
576 | extern void centaur_init_mtrr(void); | ||
577 | |||
578 | static void __init init_ifs(void) | ||
579 | { | ||
580 | #ifndef CONFIG_X86_64 | ||
581 | amd_init_mtrr(); | ||
582 | cyrix_init_mtrr(); | ||
583 | centaur_init_mtrr(); | ||
584 | #endif | ||
585 | } | ||
586 | |||
587 | /* The suspend/resume methods are only for CPU without MTRR. CPU using generic | ||
588 | * MTRR driver doesn't require this | ||
589 | */ | ||
590 | struct mtrr_value { | ||
591 | mtrr_type ltype; | ||
592 | unsigned long lbase; | ||
593 | unsigned long lsize; | ||
594 | }; | ||
595 | |||
596 | static struct mtrr_value * mtrr_state; | ||
597 | |||
598 | static int mtrr_save(struct sys_device * sysdev, pm_message_t state) | ||
599 | { | ||
600 | int i; | ||
601 | int size = num_var_ranges * sizeof(struct mtrr_value); | ||
602 | |||
603 | mtrr_state = kzalloc(size,GFP_ATOMIC); | ||
604 | if (!mtrr_state) | ||
605 | return -ENOMEM; | ||
606 | |||
607 | for (i = 0; i < num_var_ranges; i++) { | ||
608 | mtrr_if->get(i, | ||
609 | &mtrr_state[i].lbase, | ||
610 | &mtrr_state[i].lsize, | ||
611 | &mtrr_state[i].ltype); | ||
612 | } | ||
613 | return 0; | ||
614 | } | ||
615 | |||
616 | static int mtrr_restore(struct sys_device * sysdev) | ||
617 | { | ||
618 | int i; | ||
619 | |||
620 | for (i = 0; i < num_var_ranges; i++) { | ||
621 | if (mtrr_state[i].lsize) | ||
622 | set_mtrr(i, | ||
623 | mtrr_state[i].lbase, | ||
624 | mtrr_state[i].lsize, | ||
625 | mtrr_state[i].ltype); | ||
626 | } | ||
627 | kfree(mtrr_state); | ||
628 | return 0; | ||
629 | } | ||
630 | |||
631 | |||
632 | |||
633 | static struct sysdev_driver mtrr_sysdev_driver = { | ||
634 | .suspend = mtrr_save, | ||
635 | .resume = mtrr_restore, | ||
636 | }; | ||
637 | |||
638 | |||
639 | /** | ||
640 | * mtrr_bp_init - initialize mtrrs on the boot CPU | ||
641 | * | ||
642 | * This needs to be called early; before any of the other CPUs are | ||
643 | * initialized (i.e. before smp_init()). | ||
644 | * | ||
645 | */ | ||
646 | void __init mtrr_bp_init(void) | ||
647 | { | ||
648 | init_ifs(); | ||
649 | |||
650 | if (cpu_has_mtrr) { | ||
651 | mtrr_if = &generic_mtrr_ops; | ||
652 | size_or_mask = 0xff000000; /* 36 bits */ | ||
653 | size_and_mask = 0x00f00000; | ||
654 | |||
655 | /* This is an AMD specific MSR, but we assume(hope?) that | ||
656 | Intel will implement it to when they extend the address | ||
657 | bus of the Xeon. */ | ||
658 | if (cpuid_eax(0x80000000) >= 0x80000008) { | ||
659 | u32 phys_addr; | ||
660 | phys_addr = cpuid_eax(0x80000008) & 0xff; | ||
661 | /* CPUID workaround for Intel 0F33/0F34 CPU */ | ||
662 | if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL && | ||
663 | boot_cpu_data.x86 == 0xF && | ||
664 | boot_cpu_data.x86_model == 0x3 && | ||
665 | (boot_cpu_data.x86_mask == 0x3 || | ||
666 | boot_cpu_data.x86_mask == 0x4)) | ||
667 | phys_addr = 36; | ||
668 | |||
669 | size_or_mask = ~((1ULL << (phys_addr - PAGE_SHIFT)) - 1); | ||
670 | size_and_mask = ~size_or_mask & 0xfffff00000ULL; | ||
671 | } else if (boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR && | ||
672 | boot_cpu_data.x86 == 6) { | ||
673 | /* VIA C* family have Intel style MTRRs, but | ||
674 | don't support PAE */ | ||
675 | size_or_mask = 0xfff00000; /* 32 bits */ | ||
676 | size_and_mask = 0; | ||
677 | } | ||
678 | } else { | ||
679 | switch (boot_cpu_data.x86_vendor) { | ||
680 | case X86_VENDOR_AMD: | ||
681 | if (cpu_has_k6_mtrr) { | ||
682 | /* Pre-Athlon (K6) AMD CPU MTRRs */ | ||
683 | mtrr_if = mtrr_ops[X86_VENDOR_AMD]; | ||
684 | size_or_mask = 0xfff00000; /* 32 bits */ | ||
685 | size_and_mask = 0; | ||
686 | } | ||
687 | break; | ||
688 | case X86_VENDOR_CENTAUR: | ||
689 | if (cpu_has_centaur_mcr) { | ||
690 | mtrr_if = mtrr_ops[X86_VENDOR_CENTAUR]; | ||
691 | size_or_mask = 0xfff00000; /* 32 bits */ | ||
692 | size_and_mask = 0; | ||
693 | } | ||
694 | break; | ||
695 | case X86_VENDOR_CYRIX: | ||
696 | if (cpu_has_cyrix_arr) { | ||
697 | mtrr_if = mtrr_ops[X86_VENDOR_CYRIX]; | ||
698 | size_or_mask = 0xfff00000; /* 32 bits */ | ||
699 | size_and_mask = 0; | ||
700 | } | ||
701 | break; | ||
702 | default: | ||
703 | break; | ||
704 | } | ||
705 | } | ||
706 | |||
707 | if (mtrr_if) { | ||
708 | set_num_var_ranges(); | ||
709 | init_table(); | ||
710 | if (use_intel()) | ||
711 | get_mtrr_state(); | ||
712 | } | ||
713 | } | ||
714 | |||
715 | void mtrr_ap_init(void) | ||
716 | { | ||
717 | unsigned long flags; | ||
718 | |||
719 | if (!mtrr_if || !use_intel()) | ||
720 | return; | ||
721 | /* | ||
722 | * Ideally we should hold mtrr_mutex here to avoid mtrr entries changed, | ||
723 | * but this routine will be called in cpu boot time, holding the lock | ||
724 | * breaks it. This routine is called in two cases: 1.very earily time | ||
725 | * of software resume, when there absolutely isn't mtrr entry changes; | ||
726 | * 2.cpu hotadd time. We let mtrr_add/del_page hold cpuhotplug lock to | ||
727 | * prevent mtrr entry changes | ||
728 | */ | ||
729 | local_irq_save(flags); | ||
730 | |||
731 | mtrr_if->set_all(); | ||
732 | |||
733 | local_irq_restore(flags); | ||
734 | } | ||
735 | |||
736 | /** | ||
737 | * Save current fixed-range MTRR state of the BSP | ||
738 | */ | ||
739 | void mtrr_save_state(void) | ||
740 | { | ||
741 | int cpu = get_cpu(); | ||
742 | |||
743 | if (cpu == 0) | ||
744 | mtrr_save_fixed_ranges(NULL); | ||
745 | else | ||
746 | smp_call_function_single(0, mtrr_save_fixed_ranges, NULL, 1, 1); | ||
747 | put_cpu(); | ||
748 | } | ||
749 | |||
750 | static int __init mtrr_init_finialize(void) | ||
751 | { | ||
752 | if (!mtrr_if) | ||
753 | return 0; | ||
754 | if (use_intel()) | ||
755 | mtrr_state_warn(); | ||
756 | else { | ||
757 | /* The CPUs haven't MTRR and seemes not support SMP. They have | ||
758 | * specific drivers, we use a tricky method to support | ||
759 | * suspend/resume for them. | ||
760 | * TBD: is there any system with such CPU which supports | ||
761 | * suspend/resume? if no, we should remove the code. | ||
762 | */ | ||
763 | sysdev_driver_register(&cpu_sysdev_class, | ||
764 | &mtrr_sysdev_driver); | ||
765 | } | ||
766 | return 0; | ||
767 | } | ||
768 | subsys_initcall(mtrr_init_finialize); | ||