projects / deliverable / linux.git / blob
? search:


315738c74aad519b7d7e31d8e5a1d004bcf0a265
[deliverable/linux.git] / arch / x86 / kernel / cpu / mtrr / cleanup.c
1 /*
2 * MTRR (Memory Type Range Register) cleanup
3 *
4 * Copyright (C) 2009 Yinghai Lu
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 #include <linux/module.h>
21 #include <linux/init.h>
22 #include <linux/pci.h>
23 #include <linux/smp.h>
24 #include <linux/cpu.h>
25 #include <linux/sort.h>
26 #include <linux/mutex.h>
27 #include <linux/uaccess.h>
28 #include <linux/kvm_para.h>
29
30 #include <asm/processor.h>
31 #include <asm/e820.h>
32 #include <asm/mtrr.h>
33 #include <asm/msr.h>
34
35 #include "mtrr.h"
36
37 struct res_range {
38 unsigned long start;
39 unsigned long end;
40 };
41
42 struct var_mtrr_range_state {
43 unsigned long base_pfn;
44 unsigned long size_pfn;
45 mtrr_type type;
46 };
47
48 struct var_mtrr_state {
49 unsigned long range_startk;
50 unsigned long range_sizek;
51 unsigned long chunk_sizek;
52 unsigned long gran_sizek;
53 unsigned int reg;
54 };
55
56 /* Should be related to MTRR_VAR_RANGES nums */
57 #define RANGE_NUM 256
58
59 static struct res_range __initdata range[RANGE_NUM];
60 static int __initdata nr_range;
61
62 static struct var_mtrr_range_state __initdata range_state[RANGE_NUM];
63
64 static int __initdata debug_print;
65 #define Dprintk(x...) do { if (debug_print) printk(KERN_DEBUG x); } while (0)
66
67
68 static int __init
69 add_range(struct res_range *range, int nr_range,
70 unsigned long start, unsigned long end)
71 {
72 /* Out of slots: */
73 if (nr_range >= RANGE_NUM)
74 return nr_range;
75
76 range[nr_range].start = start;
77 range[nr_range].end = end;
78
79 nr_range++;
80
81 return nr_range;
82 }
83
84 static int __init
85 add_range_with_merge(struct res_range *range, int nr_range,
86 unsigned long start, unsigned long end)
87 {
88 int i;
89
90 /* Try to merge it with old one: */
91 for (i = 0; i < nr_range; i++) {
92 unsigned long final_start, final_end;
93 unsigned long common_start, common_end;
94
95 if (!range[i].end)
96 continue;
97
98 common_start = max(range[i].start, start);
99 common_end = min(range[i].end, end);
100 if (common_start > common_end + 1)
101 continue;
102
103 final_start = min(range[i].start, start);
104 final_end = max(range[i].end, end);
105
106 range[i].start = final_start;
107 range[i].end = final_end;
108 return nr_range;
109 }
110
111 /* Need to add it: */
112 return add_range(range, nr_range, start, end);
113 }
114
115 static void __init
116 subtract_range(struct res_range *range, unsigned long start, unsigned long end)
117 {
118 int i, j;
119
120 for (j = 0; j < RANGE_NUM; j++) {
121 if (!range[j].end)
122 continue;
123
124 if (start <= range[j].start && end >= range[j].end) {
125 range[j].start = 0;
126 range[j].end = 0;
127 continue;
128 }
129
130 if (start <= range[j].start && end < range[j].end &&
131 range[j].start < end + 1) {
132 range[j].start = end + 1;
133 continue;
134 }
135
136
137 if (start > range[j].start && end >= range[j].end &&
138 range[j].end > start - 1) {
139 range[j].end = start - 1;
140 continue;
141 }
142
143 if (start > range[j].start && end < range[j].end) {
144 /* Find the new spare: */
145 for (i = 0; i < RANGE_NUM; i++) {
146 if (range[i].end == 0)
147 break;
148 }
149 if (i < RANGE_NUM) {
150 range[i].end = range[j].end;
151 range[i].start = end + 1;
152 } else {
153 printk(KERN_ERR "run of slot in ranges\n");
154 }
155 range[j].end = start - 1;
156 continue;
157 }
158 }
159 }
160
161 static int __init cmp_range(const void *x1, const void *x2)
162 {
163 const struct res_range *r1 = x1;
164 const struct res_range *r2 = x2;
165 long start1, start2;
166
167 start1 = r1->start;
168 start2 = r2->start;
169
170 return start1 - start2;
171 }
172
173 #define BIOS_BUG_MSG KERN_WARNING \
174 "WARNING: BIOS bug: VAR MTRR %d contains strange UC entry under 1M, check with your system vendor!\n"
175
176 static int __init
177 x86_get_mtrr_mem_range(struct res_range *range, int nr_range,
178 unsigned long extra_remove_base,
179 unsigned long extra_remove_size)
180 {
181 unsigned long base, size;
182 mtrr_type type;
183 int i;
184
185 for (i = 0; i < num_var_ranges; i++) {
186 type = range_state[i].type;
187 if (type != MTRR_TYPE_WRBACK)
188 continue;
189 base = range_state[i].base_pfn;
190 size = range_state[i].size_pfn;
191 nr_range = add_range_with_merge(range, nr_range, base,
192 base + size - 1);
193 }
194 if (debug_print) {
195 printk(KERN_DEBUG "After WB checking\n");
196 for (i = 0; i < nr_range; i++)
197 printk(KERN_DEBUG "MTRR MAP PFN: %016lx - %016lx\n",
198 range[i].start, range[i].end + 1);
199 }
200
201 /* Take out UC ranges: */
202 for (i = 0; i < num_var_ranges; i++) {
203 type = range_state[i].type;
204 if (type != MTRR_TYPE_UNCACHABLE &&
205 type != MTRR_TYPE_WRPROT)
206 continue;
207 size = range_state[i].size_pfn;
208 if (!size)
209 continue;
210 base = range_state[i].base_pfn;
211 if (base < (1<<(20-PAGE_SHIFT)) && mtrr_state.have_fixed &&
212 (mtrr_state.enabled & 1)) {
213 /* Var MTRR contains UC entry below 1M? Skip it: */
214 printk(BIOS_BUG_MSG, i);
215 if (base + size <= (1<<(20-PAGE_SHIFT)))
216 continue;
217 size -= (1<<(20-PAGE_SHIFT)) - base;
218 base = 1<<(20-PAGE_SHIFT);
219 }
220 subtract_range(range, base, base + size - 1);
221 }
222 if (extra_remove_size)
223 subtract_range(range, extra_remove_base,
224 extra_remove_base + extra_remove_size - 1);
225
226 /* get new range num */
227 nr_range = 0;
228 for (i = 0; i < RANGE_NUM; i++) {
229 if (!range[i].end)
230 continue;
231 nr_range++;
232 }
233 if (debug_print) {
234 printk(KERN_DEBUG "After UC checking\n");
235 for (i = 0; i < nr_range; i++)
236 printk(KERN_DEBUG "MTRR MAP PFN: %016lx - %016lx\n",
237 range[i].start, range[i].end + 1);
238 }
239
240 /* sort the ranges */
241 sort(range, nr_range, sizeof(struct res_range), cmp_range, NULL);
242 if (debug_print) {
243 printk(KERN_DEBUG "After sorting\n");
244 for (i = 0; i < nr_range; i++)
245 printk(KERN_DEBUG "MTRR MAP PFN: %016lx - %016lx\n",
246 range[i].start, range[i].end + 1);
247 }
248
249 /* clear those is not used */
250 for (i = nr_range; i < RANGE_NUM; i++)
251 memset(&range[i], 0, sizeof(range[i]));
252
253 return nr_range;
254 }
255
256 #ifdef CONFIG_MTRR_SANITIZER
257
258 static unsigned long __init sum_ranges(struct res_range *range, int nr_range)
259 {
260 unsigned long sum = 0;
261 int i;
262
263 for (i = 0; i < nr_range; i++)
264 sum += range[i].end + 1 - range[i].start;
265
266 return sum;
267 }
268
269 static int enable_mtrr_cleanup __initdata =
270 CONFIG_MTRR_SANITIZER_ENABLE_DEFAULT;
271
272 static int __init disable_mtrr_cleanup_setup(char *str)
273 {
274 enable_mtrr_cleanup = 0;
275 return 0;
276 }
277 early_param("disable_mtrr_cleanup", disable_mtrr_cleanup_setup);
278
279 static int __init enable_mtrr_cleanup_setup(char *str)
280 {
281 enable_mtrr_cleanup = 1;
282 return 0;
283 }
284 early_param("enable_mtrr_cleanup", enable_mtrr_cleanup_setup);
285
286 static int __init mtrr_cleanup_debug_setup(char *str)
287 {
288 debug_print = 1;
289 return 0;
290 }
291 early_param("mtrr_cleanup_debug", mtrr_cleanup_debug_setup);
292
293 static void __init
294 set_var_mtrr(unsigned int reg, unsigned long basek, unsigned long sizek,
295 unsigned char type, unsigned int address_bits)
296 {
297 u32 base_lo, base_hi, mask_lo, mask_hi;
298 u64 base, mask;
299
300 if (!sizek) {
301 fill_mtrr_var_range(reg, 0, 0, 0, 0);
302 return;
303 }
304
305 mask = (1ULL << address_bits) - 1;
306 mask &= ~((((u64)sizek) << 10) - 1);
307
308 base = ((u64)basek) << 10;
309
310 base |= type;
311 mask |= 0x800;
312
313 base_lo = base & ((1ULL<<32) - 1);
314 base_hi = base >> 32;
315
316 mask_lo = mask & ((1ULL<<32) - 1);
317 mask_hi = mask >> 32;
318
319 fill_mtrr_var_range(reg, base_lo, base_hi, mask_lo, mask_hi);
320 }
321
322 static void __init
323 save_var_mtrr(unsigned int reg, unsigned long basek, unsigned long sizek,
324 unsigned char type)
325 {
326 range_state[reg].base_pfn = basek >> (PAGE_SHIFT - 10);
327 range_state[reg].size_pfn = sizek >> (PAGE_SHIFT - 10);
328 range_state[reg].type = type;
329 }
330
331 static void __init set_var_mtrr_all(unsigned int address_bits)
332 {
333 unsigned long basek, sizek;
334 unsigned char type;
335 unsigned int reg;
336
337 for (reg = 0; reg < num_var_ranges; reg++) {
338 basek = range_state[reg].base_pfn << (PAGE_SHIFT - 10);
339 sizek = range_state[reg].size_pfn << (PAGE_SHIFT - 10);
340 type = range_state[reg].type;
341
342 set_var_mtrr(reg, basek, sizek, type, address_bits);
343 }
344 }
345
346 static unsigned long to_size_factor(unsigned long sizek, char *factorp)
347 {
348 unsigned long base = sizek;
349 char factor;
350
351 if (base & ((1<<10) - 1)) {
352 /* Not MB-aligned: */
353 factor = 'K';
354 } else if (base & ((1<<20) - 1)) {
355 factor = 'M';
356 base >>= 10;
357 } else {
358 factor = 'G';
359 base >>= 20;
360 }
361
362 *factorp = factor;
363
364 return base;
365 }
366
367 static unsigned int __init
368 range_to_mtrr(unsigned int reg, unsigned long range_startk,
369 unsigned long range_sizek, unsigned char type)
370 {
371 if (!range_sizek || (reg >= num_var_ranges))
372 return reg;
373
374 while (range_sizek) {
375 unsigned long max_align, align;
376 unsigned long sizek;
377
378 /* Compute the maximum size with which we can make a range: */
379 if (range_startk)
380 max_align = ffs(range_startk) - 1;
381 else
382 max_align = 32;
383
384 align = fls(range_sizek) - 1;
385 if (align > max_align)
386 align = max_align;
387
388 sizek = 1 << align;
389 if (debug_print) {
390 char start_factor = 'K', size_factor = 'K';
391 unsigned long start_base, size_base;
392
393 start_base = to_size_factor(range_startk, &start_factor);
394 size_base = to_size_factor(sizek, &size_factor);
395
396 Dprintk("Setting variable MTRR %d, "
397 "base: %ld%cB, range: %ld%cB, type %s\n",
398 reg, start_base, start_factor,
399 size_base, size_factor,
400 (type == MTRR_TYPE_UNCACHABLE) ? "UC" :
401 ((type == MTRR_TYPE_WRBACK) ? "WB" : "Other")
402 );
403 }
404 save_var_mtrr(reg++, range_startk, sizek, type);
405 range_startk += sizek;
406 range_sizek -= sizek;
407 if (reg >= num_var_ranges)
408 break;
409 }
410 return reg;
411 }
412
413 static unsigned __init
414 range_to_mtrr_with_hole(struct var_mtrr_state *state, unsigned long basek,
415 unsigned long sizek)
416 {
417 unsigned long hole_basek, hole_sizek;
418 unsigned long second_basek, second_sizek;
419 unsigned long range0_basek, range0_sizek;
420 unsigned long range_basek, range_sizek;
421 unsigned long chunk_sizek;
422 unsigned long gran_sizek;
423
424 hole_basek = 0;
425 hole_sizek = 0;
426 second_basek = 0;
427 second_sizek = 0;
428 chunk_sizek = state->chunk_sizek;
429 gran_sizek = state->gran_sizek;
430
431 /* Align with gran size, prevent small block used up MTRRs: */
432 range_basek = ALIGN(state->range_startk, gran_sizek);
433 if ((range_basek > basek) && basek)
434 return second_sizek;
435
436 state->range_sizek -= (range_basek - state->range_startk);
437 range_sizek = ALIGN(state->range_sizek, gran_sizek);
438
439 while (range_sizek > state->range_sizek) {
440 range_sizek -= gran_sizek;
441 if (!range_sizek)
442 return 0;
443 }
444 state->range_sizek = range_sizek;
445
446 /* Try to append some small hole: */
447 range0_basek = state->range_startk;
448 range0_sizek = ALIGN(state->range_sizek, chunk_sizek);
449
450 /* No increase: */
451 if (range0_sizek == state->range_sizek) {
452 Dprintk("rangeX: %016lx - %016lx\n",
453 range0_basek<<10,
454 (range0_basek + state->range_sizek)<<10);
455 state->reg = range_to_mtrr(state->reg, range0_basek,
456 state->range_sizek, MTRR_TYPE_WRBACK);
457 return 0;
458 }
459
460 /* Only cut back when it is not the last: */
461 if (sizek) {
462 while (range0_basek + range0_sizek > (basek + sizek)) {
463 if (range0_sizek >= chunk_sizek)
464 range0_sizek -= chunk_sizek;
465 else
466 range0_sizek = 0;
467
468 if (!range0_sizek)
469 break;
470 }
471 }
472
473 second_try:
474 range_basek = range0_basek + range0_sizek;
475
476 /* One hole in the middle: */
477 if (range_basek > basek && range_basek <= (basek + sizek))
478 second_sizek = range_basek - basek;
479
480 if (range0_sizek > state->range_sizek) {
481
482 /* One hole in middle or at the end: */
483 hole_sizek = range0_sizek - state->range_sizek - second_sizek;
484
485 /* Hole size should be less than half of range0 size: */
486 if (hole_sizek >= (range0_sizek >> 1) &&
487 range0_sizek >= chunk_sizek) {
488 range0_sizek -= chunk_sizek;
489 second_sizek = 0;
490 hole_sizek = 0;
491
492 goto second_try;
493 }
494 }
495
496 if (range0_sizek) {
497 Dprintk("range0: %016lx - %016lx\n",
498 range0_basek<<10,
499 (range0_basek + range0_sizek)<<10);
500 state->reg = range_to_mtrr(state->reg, range0_basek,
501 range0_sizek, MTRR_TYPE_WRBACK);
502 }
503
504 if (range0_sizek < state->range_sizek) {
505 /* Need to handle left over range: */
506 range_sizek = state->range_sizek - range0_sizek;
507
508 Dprintk("range: %016lx - %016lx\n",
509 range_basek<<10,
510 (range_basek + range_sizek)<<10);
511
512 state->reg = range_to_mtrr(state->reg, range_basek,
513 range_sizek, MTRR_TYPE_WRBACK);
514 }
515
516 if (hole_sizek) {
517 hole_basek = range_basek - hole_sizek - second_sizek;
518 Dprintk("hole: %016lx - %016lx\n",
519 hole_basek<<10,
520 (hole_basek + hole_sizek)<<10);
521 state->reg = range_to_mtrr(state->reg, hole_basek,
522 hole_sizek, MTRR_TYPE_UNCACHABLE);
523 }
524
525 return second_sizek;
526 }
527
528 static void __init
529 set_var_mtrr_range(struct var_mtrr_state *state, unsigned long base_pfn,
530 unsigned long size_pfn)
531 {
532 unsigned long basek, sizek;
533 unsigned long second_sizek = 0;
534
535 if (state->reg >= num_var_ranges)
536 return;
537
538 basek = base_pfn << (PAGE_SHIFT - 10);
539 sizek = size_pfn << (PAGE_SHIFT - 10);
540
541 /* See if I can merge with the last range: */
542 if ((basek <= 1024) ||
543 (state->range_startk + state->range_sizek == basek)) {
544 unsigned long endk = basek + sizek;
545 state->range_sizek = endk - state->range_startk;
546 return;
547 }
548 /* Write the range mtrrs: */
549 if (state->range_sizek != 0)
550 second_sizek = range_to_mtrr_with_hole(state, basek, sizek);
551
552 /* Allocate an msr: */
553 state->range_startk = basek + second_sizek;
554 state->range_sizek = sizek - second_sizek;
555 }
556
557 /* Mininum size of mtrr block that can take hole: */
558 static u64 mtrr_chunk_size __initdata = (256ULL<<20);
559
560 static int __init parse_mtrr_chunk_size_opt(char *p)
561 {
562 if (!p)
563 return -EINVAL;
564 mtrr_chunk_size = memparse(p, &p);
565 return 0;
566 }
567 early_param("mtrr_chunk_size", parse_mtrr_chunk_size_opt);
568
569 /* Granularity of mtrr of block: */
570 static u64 mtrr_gran_size __initdata;
571
572 static int __init parse_mtrr_gran_size_opt(char *p)
573 {
574 if (!p)
575 return -EINVAL;
576 mtrr_gran_size = memparse(p, &p);
577 return 0;
578 }
579 early_param("mtrr_gran_size", parse_mtrr_gran_size_opt);
580
581 static unsigned long nr_mtrr_spare_reg __initdata =
582 CONFIG_MTRR_SANITIZER_SPARE_REG_NR_DEFAULT;
583
584 static int __init parse_mtrr_spare_reg(char *arg)
585 {
586 if (arg)
587 nr_mtrr_spare_reg = simple_strtoul(arg, NULL, 0);
588 return 0;
589 }
590 early_param("mtrr_spare_reg_nr", parse_mtrr_spare_reg);
591
592 static int __init
593 x86_setup_var_mtrrs(struct res_range *range, int nr_range,
594 u64 chunk_size, u64 gran_size)
595 {
596 struct var_mtrr_state var_state;
597 int num_reg;
598 int i;
599
600 var_state.range_startk = 0;
601 var_state.range_sizek = 0;
602 var_state.reg = 0;
603 var_state.chunk_sizek = chunk_size >> 10;
604 var_state.gran_sizek = gran_size >> 10;
605
606 memset(range_state, 0, sizeof(range_state));
607
608 /* Write the range: */
609 for (i = 0; i < nr_range; i++) {
610 set_var_mtrr_range(&var_state, range[i].start,
611 range[i].end - range[i].start + 1);
612 }
613
614 /* Write the last range: */
615 if (var_state.range_sizek != 0)
616 range_to_mtrr_with_hole(&var_state, 0, 0);
617
618 num_reg = var_state.reg;
619 /* Clear out the extra MTRR's: */
620 while (var_state.reg < num_var_ranges) {
621 save_var_mtrr(var_state.reg, 0, 0, 0);
622 var_state.reg++;
623 }
624
625 return num_reg;
626 }
627
628 struct mtrr_cleanup_result {
629 unsigned long gran_sizek;
630 unsigned long chunk_sizek;
631 unsigned long lose_cover_sizek;
632 unsigned int num_reg;
633 int bad;
634 };
635
636 /*
637 * gran_size: 64K, 128K, 256K, 512K, 1M, 2M, ..., 2G
638 * chunk size: gran_size, ..., 2G
639 * so we need (1+16)*8
640 */
641 #define NUM_RESULT 136
642 #define PSHIFT (PAGE_SHIFT - 10)
643
644 static struct mtrr_cleanup_result __initdata result[NUM_RESULT];
645 static unsigned long __initdata min_loss_pfn[RANGE_NUM];
646
647 static void __init print_out_mtrr_range_state(void)
648 {
649 char start_factor = 'K', size_factor = 'K';
650 unsigned long start_base, size_base;
651 mtrr_type type;
652 int i;
653
654 for (i = 0; i < num_var_ranges; i++) {
655
656 size_base = range_state[i].size_pfn << (PAGE_SHIFT - 10);
657 if (!size_base)
658 continue;
659
660 size_base = to_size_factor(size_base, &size_factor),
661 start_base = range_state[i].base_pfn << (PAGE_SHIFT - 10);
662 start_base = to_size_factor(start_base, &start_factor),
663 type = range_state[i].type;
664
665 printk(KERN_DEBUG "reg %d, base: %ld%cB, range: %ld%cB, type %s\n",
666 i, start_base, start_factor,
667 size_base, size_factor,
668 (type == MTRR_TYPE_UNCACHABLE) ? "UC" :
669 ((type == MTRR_TYPE_WRPROT) ? "WP" :
670 ((type == MTRR_TYPE_WRBACK) ? "WB" : "Other"))
671 );
672 }
673 }
674
675 static int __init mtrr_need_cleanup(void)
676 {
677 int i;
678 mtrr_type type;
679 unsigned long size;
680 /* Extra one for all 0: */
681 int num[MTRR_NUM_TYPES + 1];
682
683 /* Check entries number: */
684 memset(num, 0, sizeof(num));
685 for (i = 0; i < num_var_ranges; i++) {
686 type = range_state[i].type;
687 size = range_state[i].size_pfn;
688 if (type >= MTRR_NUM_TYPES)
689 continue;
690 if (!size)
691 type = MTRR_NUM_TYPES;
692 if (type == MTRR_TYPE_WRPROT)
693 type = MTRR_TYPE_UNCACHABLE;
694 num[type]++;
695 }
696
697 /* Check if we got UC entries: */
698 if (!num[MTRR_TYPE_UNCACHABLE])
699 return 0;
700
701 /* Check if we only had WB and UC */
702 if (num[MTRR_TYPE_WRBACK] + num[MTRR_TYPE_UNCACHABLE] !=
703 num_var_ranges - num[MTRR_NUM_TYPES])
704 return 0;
705
706 return 1;
707 }
708
709 static unsigned long __initdata range_sums;
710
711 static void __init
712 mtrr_calc_range_state(u64 chunk_size, u64 gran_size,
713 unsigned long x_remove_base,
714 unsigned long x_remove_size, int i)
715 {
716 static struct res_range range_new[RANGE_NUM];
717 unsigned long range_sums_new;
718 static int nr_range_new;
719 int num_reg;
720
721 /* Convert ranges to var ranges state: */
722 num_reg = x86_setup_var_mtrrs(range, nr_range, chunk_size, gran_size);
723
724 /* We got new setting in range_state, check it: */
725 memset(range_new, 0, sizeof(range_new));
726 nr_range_new = x86_get_mtrr_mem_range(range_new, 0,
727 x_remove_base, x_remove_size);
728 range_sums_new = sum_ranges(range_new, nr_range_new);
729
730 result[i].chunk_sizek = chunk_size >> 10;
731 result[i].gran_sizek = gran_size >> 10;
732 result[i].num_reg = num_reg;
733
734 if (range_sums < range_sums_new) {
735 result[i].lose_cover_sizek = (range_sums_new - range_sums) << PSHIFT;
736 result[i].bad = 1;
737 } else {
738 result[i].lose_cover_sizek = (range_sums - range_sums_new) << PSHIFT;
739 }
740
741 /* Double check it: */
742 if (!result[i].bad && !result[i].lose_cover_sizek) {
743 if (nr_range_new != nr_range || memcmp(range, range_new, sizeof(range)))
744 result[i].bad = 1;
745 }
746
747 if (!result[i].bad && (range_sums - range_sums_new < min_loss_pfn[num_reg]))
748 min_loss_pfn[num_reg] = range_sums - range_sums_new;
749 }
750
751 static void __init mtrr_print_out_one_result(int i)
752 {
753 unsigned long gran_base, chunk_base, lose_base;
754 char gran_factor, chunk_factor, lose_factor;
755
756 gran_base = to_size_factor(result[i].gran_sizek, &gran_factor),
757 chunk_base = to_size_factor(result[i].chunk_sizek, &chunk_factor),
758 lose_base = to_size_factor(result[i].lose_cover_sizek, &lose_factor),
759
760 pr_info("%sgran_size: %ld%c \tchunk_size: %ld%c \t",
761 result[i].bad ? "*BAD*" : " ",
762 gran_base, gran_factor, chunk_base, chunk_factor);
763 pr_cont("num_reg: %d \tlose cover RAM: %s%ld%c\n",
764 result[i].num_reg, result[i].bad ? "-" : "",
765 lose_base, lose_factor);
766 }
767
768 static int __init mtrr_search_optimal_index(void)
769 {
770 int num_reg_good;
771 int index_good;
772 int i;
773
774 if (nr_mtrr_spare_reg >= num_var_ranges)
775 nr_mtrr_spare_reg = num_var_ranges - 1;
776
777 num_reg_good = -1;
778 for (i = num_var_ranges - nr_mtrr_spare_reg; i > 0; i--) {
779 if (!min_loss_pfn[i])
780 num_reg_good = i;
781 }
782
783 index_good = -1;
784 if (num_reg_good != -1) {
785 for (i = 0; i < NUM_RESULT; i++) {
786 if (!result[i].bad &&
787 result[i].num_reg == num_reg_good &&
788 !result[i].lose_cover_sizek) {
789 index_good = i;
790 break;
791 }
792 }
793 }
794
795 return index_good;
796 }
797
798 int __init mtrr_cleanup(unsigned address_bits)
799 {
800 unsigned long x_remove_base, x_remove_size;
801 unsigned long base, size, def, dummy;
802 u64 chunk_size, gran_size;
803 mtrr_type type;
804 int index_good;
805 int i;
806
807 if (!is_cpu(INTEL) || enable_mtrr_cleanup < 1)
808 return 0;
809
810 rdmsr(MSR_MTRRdefType, def, dummy);
811 def &= 0xff;
812 if (def != MTRR_TYPE_UNCACHABLE)
813 return 0;
814
815 /* Get it and store it aside: */
816 memset(range_state, 0, sizeof(range_state));
817 for (i = 0; i < num_var_ranges; i++) {
818 mtrr_if->get(i, &base, &size, &type);
819 range_state[i].base_pfn = base;
820 range_state[i].size_pfn = size;
821 range_state[i].type = type;
822 }
823
824 /* Check if we need handle it and can handle it: */
825 if (!mtrr_need_cleanup())
826 return 0;
827
828 /* Print original var MTRRs at first, for debugging: */
829 printk(KERN_DEBUG "original variable MTRRs\n");
830 print_out_mtrr_range_state();
831
832 memset(range, 0, sizeof(range));
833 x_remove_size = 0;
834 x_remove_base = 1 << (32 - PAGE_SHIFT);
835 if (mtrr_tom2)
836 x_remove_size = (mtrr_tom2 >> PAGE_SHIFT) - x_remove_base;
837
838 nr_range = x86_get_mtrr_mem_range(range, 0, x_remove_base, x_remove_size);
839 /*
840 * [0, 1M) should always be covered by var mtrr with WB
841 * and fixed mtrrs should take effect before var mtrr for it:
842 */
843 nr_range = add_range_with_merge(range, nr_range, 0,
844 (1ULL<<(20 - PAGE_SHIFT)) - 1);
845 /* Sort the ranges: */
846 sort(range, nr_range, sizeof(struct res_range), cmp_range, NULL);
847
848 range_sums = sum_ranges(range, nr_range);
849 printk(KERN_INFO "total RAM coverred: %ldM\n",
850 range_sums >> (20 - PAGE_SHIFT));
851
852 if (mtrr_chunk_size && mtrr_gran_size) {
853 i = 0;
854 mtrr_calc_range_state(mtrr_chunk_size, mtrr_gran_size,
855 x_remove_base, x_remove_size, i);
856
857 mtrr_print_out_one_result(i);
858
859 if (!result[i].bad) {
860 set_var_mtrr_all(address_bits);
861 printk(KERN_DEBUG "New variable MTRRs\n");
862 print_out_mtrr_range_state();
863 return 1;
864 }
865 printk(KERN_INFO "invalid mtrr_gran_size or mtrr_chunk_size, "
866 "will find optimal one\n");
867 }
868
869 i = 0;
870 memset(min_loss_pfn, 0xff, sizeof(min_loss_pfn));
871 memset(result, 0, sizeof(result));
872 for (gran_size = (1ULL<<16); gran_size < (1ULL<<32); gran_size <<= 1) {
873
874 for (chunk_size = gran_size; chunk_size < (1ULL<<32);
875 chunk_size <<= 1) {
876
877 if (i >= NUM_RESULT)
878 continue;
879
880 mtrr_calc_range_state(chunk_size, gran_size,
881 x_remove_base, x_remove_size, i);
882 if (debug_print) {
883 mtrr_print_out_one_result(i);
884 printk(KERN_INFO "\n");
885 }
886
887 i++;
888 }
889 }
890
891 /* Try to find the optimal index: */
892 index_good = mtrr_search_optimal_index();
893
894 if (index_good != -1) {
895 printk(KERN_INFO "Found optimal setting for mtrr clean up\n");
896 i = index_good;
897 mtrr_print_out_one_result(i);
898
899 /* Convert ranges to var ranges state: */
900 chunk_size = result[i].chunk_sizek;
901 chunk_size <<= 10;
902 gran_size = result[i].gran_sizek;
903 gran_size <<= 10;
904 x86_setup_var_mtrrs(range, nr_range, chunk_size, gran_size);
905 set_var_mtrr_all(address_bits);
906 printk(KERN_DEBUG "New variable MTRRs\n");
907 print_out_mtrr_range_state();
908 return 1;
909 } else {
910 /* print out all */
911 for (i = 0; i < NUM_RESULT; i++)
912 mtrr_print_out_one_result(i);
913 }
914
915 printk(KERN_INFO "mtrr_cleanup: can not find optimal value\n");
916 printk(KERN_INFO "please specify mtrr_gran_size/mtrr_chunk_size\n");
917
918 return 0;
919 }
920 #else
921 int __init mtrr_cleanup(unsigned address_bits)
922 {
923 return 0;
924 }
925 #endif
926
927 static int disable_mtrr_trim;
928
929 static int __init disable_mtrr_trim_setup(char *str)
930 {
931 disable_mtrr_trim = 1;
932 return 0;
933 }
934 early_param("disable_mtrr_trim", disable_mtrr_trim_setup);
935
936 /*
937 * Newer AMD K8s and later CPUs have a special magic MSR way to force WB
938 * for memory >4GB. Check for that here.
939 * Note this won't check if the MTRRs < 4GB where the magic bit doesn't
940 * apply to are wrong, but so far we don't know of any such case in the wild.
941 */
942 #define Tom2Enabled (1U << 21)
943 #define Tom2ForceMemTypeWB (1U << 22)
944
945 int __init amd_special_default_mtrr(void)
946 {
947 u32 l, h;
948
949 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD)
950 return 0;
951 if (boot_cpu_data.x86 < 0xf || boot_cpu_data.x86 > 0x11)
952 return 0;
953 /* In case some hypervisor doesn't pass SYSCFG through: */
954 if (rdmsr_safe(MSR_K8_SYSCFG, &l, &h) < 0)
955 return 0;
956 /*
957 * Memory between 4GB and top of mem is forced WB by this magic bit.
958 * Reserved before K8RevF, but should be zero there.
959 */
960 if ((l & (Tom2Enabled | Tom2ForceMemTypeWB)) ==
961 (Tom2Enabled | Tom2ForceMemTypeWB))
962 return 1;
963 return 0;
964 }
965
966 static u64 __init
967 real_trim_memory(unsigned long start_pfn, unsigned long limit_pfn)
968 {
969 u64 trim_start, trim_size;
970
971 trim_start = start_pfn;
972 trim_start <<= PAGE_SHIFT;
973
974 trim_size = limit_pfn;
975 trim_size <<= PAGE_SHIFT;
976 trim_size -= trim_start;
977
978 return e820_update_range(trim_start, trim_size, E820_RAM, E820_RESERVED);
979 }
980
981 /**
982 * mtrr_trim_uncached_memory - trim RAM not covered by MTRRs
983 * @end_pfn: ending page frame number
984 *
985 * Some buggy BIOSes don't setup the MTRRs properly for systems with certain
986 * memory configurations. This routine checks that the highest MTRR matches
987 * the end of memory, to make sure the MTRRs having a write back type cover
988 * all of the memory the kernel is intending to use. If not, it'll trim any
989 * memory off the end by adjusting end_pfn, removing it from the kernel's
990 * allocation pools, warning the user with an obnoxious message.
991 */
992 int __init mtrr_trim_uncached_memory(unsigned long end_pfn)
993 {
994 unsigned long i, base, size, highest_pfn = 0, def, dummy;
995 mtrr_type type;
996 u64 total_trim_size;
997 /* extra one for all 0 */
998 int num[MTRR_NUM_TYPES + 1];
999
1000 /*
1001 * Make sure we only trim uncachable memory on machines that
1002 * support the Intel MTRR architecture:
1003 */
1004 if (!is_cpu(INTEL) || disable_mtrr_trim)
1005 return 0;
1006
1007 rdmsr(MSR_MTRRdefType, def, dummy);
1008 def &= 0xff;
1009 if (def != MTRR_TYPE_UNCACHABLE)
1010 return 0;
1011
1012 /* Get it and store it aside: */
1013 memset(range_state, 0, sizeof(range_state));
1014 for (i = 0; i < num_var_ranges; i++) {
1015 mtrr_if->get(i, &base, &size, &type);
1016 range_state[i].base_pfn = base;
1017 range_state[i].size_pfn = size;
1018 range_state[i].type = type;
1019 }
1020
1021 /* Find highest cached pfn: */
1022 for (i = 0; i < num_var_ranges; i++) {
1023 type = range_state[i].type;
1024 if (type != MTRR_TYPE_WRBACK)
1025 continue;
1026 base = range_state[i].base_pfn;
1027 size = range_state[i].size_pfn;
1028 if (highest_pfn < base + size)
1029 highest_pfn = base + size;
1030 }
1031
1032 /* kvm/qemu doesn't have mtrr set right, don't trim them all: */
1033 if (!highest_pfn) {
1034 printk(KERN_INFO "CPU MTRRs all blank - virtualized system.\n");
1035 return 0;
1036 }
1037
1038 /* Check entries number: */
1039 memset(num, 0, sizeof(num));
1040 for (i = 0; i < num_var_ranges; i++) {
1041 type = range_state[i].type;
1042 if (type >= MTRR_NUM_TYPES)
1043 continue;
1044 size = range_state[i].size_pfn;
1045 if (!size)
1046 type = MTRR_NUM_TYPES;
1047 num[type]++;
1048 }
1049
1050 /* No entry for WB? */
1051 if (!num[MTRR_TYPE_WRBACK])
1052 return 0;
1053
1054 /* Check if we only had WB and UC: */
1055 if (num[MTRR_TYPE_WRBACK] + num[MTRR_TYPE_UNCACHABLE] !=
1056 num_var_ranges - num[MTRR_NUM_TYPES])
1057 return 0;
1058
1059 memset(range, 0, sizeof(range));
1060 nr_range = 0;
1061 if (mtrr_tom2) {
1062 range[nr_range].start = (1ULL<<(32 - PAGE_SHIFT));
1063 range[nr_range].end = (mtrr_tom2 >> PAGE_SHIFT) - 1;
1064 if (highest_pfn < range[nr_range].end + 1)
1065 highest_pfn = range[nr_range].end + 1;
1066 nr_range++;
1067 }
1068 nr_range = x86_get_mtrr_mem_range(range, nr_range, 0, 0);
1069
1070 /* Check the head: */
1071 total_trim_size = 0;
1072 if (range[0].start)
1073 total_trim_size += real_trim_memory(0, range[0].start);
1074
1075 /* Check the holes: */
1076 for (i = 0; i < nr_range - 1; i++) {
1077 if (range[i].end + 1 < range[i+1].start)
1078 total_trim_size += real_trim_memory(range[i].end + 1,
1079 range[i+1].start);
1080 }
1081
1082 /* Check the top: */
1083 i = nr_range - 1;
1084 if (range[i].end + 1 < end_pfn)
1085 total_trim_size += real_trim_memory(range[i].end + 1,
1086 end_pfn);
1087
1088 if (total_trim_size) {
1089 pr_warning("WARNING: BIOS bug: CPU MTRRs don't cover all of memory, losing %lluMB of RAM.\n", total_trim_size >> 20);
1090
1091 if (!changed_by_mtrr_cleanup)
1092 WARN_ON(1);
1093
1094 pr_info("update e820 for mtrr\n");
1095 update_e820();
1096
1097 return 1;
1098 }
1099
1100 return 0;
1101 }
Linux kernel - Deliverables
This page took 0.05513 seconds and 4 git commands to generate.