?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'x86-mm-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6/kvm.git] / arch / x86 / kernel / cpu / mtrr / cleanup.c
blob09b1698e0466be4c7c0b041d80be57992000667e
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 static int __init clean_sort_range(struct res_range *range, int az)
174 {
175 int i, j, k = az - 1, nr_range = 0;
176
177 for (i = 0; i < k; i++) {
178 if (range[i].end)
179 continue;
180 for (j = k; j > i; j--) {
181 if (range[j].end) {
182 k = j;
183 break;
184 }
185 }
186 if (j == i)
187 break;
188 range[i].start = range[k].start;
189 range[i].end = range[k].end;
190 range[k].start = 0;
191 range[k].end = 0;
192 k--;
193 }
194 /* count it */
195 for (i = 0; i < az; i++) {
196 if (!range[i].end) {
197 nr_range = i;
198 break;
199 }
200 }
201
202 /* sort them */
203 sort(range, nr_range, sizeof(struct res_range), cmp_range, NULL);
204
205 return nr_range;
206 }
207
208 #define BIOS_BUG_MSG KERN_WARNING \
209 "WARNING: BIOS bug: VAR MTRR %d contains strange UC entry under 1M, check with your system vendor!\n"
210
211 static int __init
212 x86_get_mtrr_mem_range(struct res_range *range, int nr_range,
213 unsigned long extra_remove_base,
214 unsigned long extra_remove_size)
215 {
216 unsigned long base, size;
217 mtrr_type type;
218 int i;
219
220 for (i = 0; i < num_var_ranges; i++) {
221 type = range_state[i].type;
222 if (type != MTRR_TYPE_WRBACK)
223 continue;
224 base = range_state[i].base_pfn;
225 size = range_state[i].size_pfn;
226 nr_range = add_range_with_merge(range, nr_range, base,
227 base + size - 1);
228 }
229 if (debug_print) {
230 printk(KERN_DEBUG "After WB checking\n");
231 for (i = 0; i < nr_range; i++)
232 printk(KERN_DEBUG "MTRR MAP PFN: %016lx - %016lx\n",
233 range[i].start, range[i].end + 1);
234 }
235
236 /* Take out UC ranges: */
237 for (i = 0; i < num_var_ranges; i++) {
238 type = range_state[i].type;
239 if (type != MTRR_TYPE_UNCACHABLE &&
240 type != MTRR_TYPE_WRPROT)
241 continue;
242 size = range_state[i].size_pfn;
243 if (!size)
244 continue;
245 base = range_state[i].base_pfn;
246 if (base < (1<<(20-PAGE_SHIFT)) && mtrr_state.have_fixed &&
247 (mtrr_state.enabled & 1)) {
248 /* Var MTRR contains UC entry below 1M? Skip it: */
249 printk(BIOS_BUG_MSG, i);
250 if (base + size <= (1<<(20-PAGE_SHIFT)))
251 continue;
252 size -= (1<<(20-PAGE_SHIFT)) - base;
253 base = 1<<(20-PAGE_SHIFT);
254 }
255 subtract_range(range, base, base + size - 1);
256 }
257 if (extra_remove_size)
258 subtract_range(range, extra_remove_base,
259 extra_remove_base + extra_remove_size - 1);
260
261 if (debug_print) {
262 printk(KERN_DEBUG "After UC checking\n");
263 for (i = 0; i < RANGE_NUM; i++) {
264 if (!range[i].end)
265 continue;
266 printk(KERN_DEBUG "MTRR MAP PFN: %016lx - %016lx\n",
267 range[i].start, range[i].end + 1);
268 }
269 }
270
271 /* sort the ranges */
272 nr_range = clean_sort_range(range, RANGE_NUM);
273 if (debug_print) {
274 printk(KERN_DEBUG "After sorting\n");
275 for (i = 0; i < nr_range; i++)
276 printk(KERN_DEBUG "MTRR MAP PFN: %016lx - %016lx\n",
277 range[i].start, range[i].end + 1);
278 }
279
280 /* clear those is not used */
281 for (i = nr_range; i < RANGE_NUM; i++)
282 memset(&range[i], 0, sizeof(range[i]));
283
284 return nr_range;
285 }
286
287 #ifdef CONFIG_MTRR_SANITIZER
288
289 static unsigned long __init sum_ranges(struct res_range *range, int nr_range)
290 {
291 unsigned long sum = 0;
292 int i;
293
294 for (i = 0; i < nr_range; i++)
295 sum += range[i].end + 1 - range[i].start;
296
297 return sum;
298 }
299
300 static int enable_mtrr_cleanup __initdata =
301 CONFIG_MTRR_SANITIZER_ENABLE_DEFAULT;
302
303 static int __init disable_mtrr_cleanup_setup(char *str)
304 {
305 enable_mtrr_cleanup = 0;
306 return 0;
307 }
308 early_param("disable_mtrr_cleanup", disable_mtrr_cleanup_setup);
309
310 static int __init enable_mtrr_cleanup_setup(char *str)
311 {
312 enable_mtrr_cleanup = 1;
313 return 0;
314 }
315 early_param("enable_mtrr_cleanup", enable_mtrr_cleanup_setup);
316
317 static int __init mtrr_cleanup_debug_setup(char *str)
318 {
319 debug_print = 1;
320 return 0;
321 }
322 early_param("mtrr_cleanup_debug", mtrr_cleanup_debug_setup);
323
324 static void __init
325 set_var_mtrr(unsigned int reg, unsigned long basek, unsigned long sizek,
326 unsigned char type, unsigned int address_bits)
327 {
328 u32 base_lo, base_hi, mask_lo, mask_hi;
329 u64 base, mask;
330
331 if (!sizek) {
332 fill_mtrr_var_range(reg, 0, 0, 0, 0);
333 return;
334 }
335
336 mask = (1ULL << address_bits) - 1;
337 mask &= ~((((u64)sizek) << 10) - 1);
338
339 base = ((u64)basek) << 10;
340
341 base |= type;
342 mask |= 0x800;
343
344 base_lo = base & ((1ULL<<32) - 1);
345 base_hi = base >> 32;
346
347 mask_lo = mask & ((1ULL<<32) - 1);
348 mask_hi = mask >> 32;
349
350 fill_mtrr_var_range(reg, base_lo, base_hi, mask_lo, mask_hi);
351 }
352
353 static void __init
354 save_var_mtrr(unsigned int reg, unsigned long basek, unsigned long sizek,
355 unsigned char type)
356 {
357 range_state[reg].base_pfn = basek >> (PAGE_SHIFT - 10);
358 range_state[reg].size_pfn = sizek >> (PAGE_SHIFT - 10);
359 range_state[reg].type = type;
360 }
361
362 static void __init set_var_mtrr_all(unsigned int address_bits)
363 {
364 unsigned long basek, sizek;
365 unsigned char type;
366 unsigned int reg;
367
368 for (reg = 0; reg < num_var_ranges; reg++) {
369 basek = range_state[reg].base_pfn << (PAGE_SHIFT - 10);
370 sizek = range_state[reg].size_pfn << (PAGE_SHIFT - 10);
371 type = range_state[reg].type;
372
373 set_var_mtrr(reg, basek, sizek, type, address_bits);
374 }
375 }
376
377 static unsigned long to_size_factor(unsigned long sizek, char *factorp)
378 {
379 unsigned long base = sizek;
380 char factor;
381
382 if (base & ((1<<10) - 1)) {
383 /* Not MB-aligned: */
384 factor = 'K';
385 } else if (base & ((1<<20) - 1)) {
386 factor = 'M';
387 base >>= 10;
388 } else {
389 factor = 'G';
390 base >>= 20;
391 }
392
393 *factorp = factor;
394
395 return base;
396 }
397
398 static unsigned int __init
399 range_to_mtrr(unsigned int reg, unsigned long range_startk,
400 unsigned long range_sizek, unsigned char type)
401 {
402 if (!range_sizek || (reg >= num_var_ranges))
403 return reg;
404
405 while (range_sizek) {
406 unsigned long max_align, align;
407 unsigned long sizek;
408
409 /* Compute the maximum size with which we can make a range: */
410 if (range_startk)
411 max_align = ffs(range_startk) - 1;
412 else
413 max_align = 32;
414
415 align = fls(range_sizek) - 1;
416 if (align > max_align)
417 align = max_align;
418
419 sizek = 1 << align;
420 if (debug_print) {
421 char start_factor = 'K', size_factor = 'K';
422 unsigned long start_base, size_base;
423
424 start_base = to_size_factor(range_startk, &start_factor);
425 size_base = to_size_factor(sizek, &size_factor);
426
427 Dprintk("Setting variable MTRR %d, "
428 "base: %ld%cB, range: %ld%cB, type %s\n",
429 reg, start_base, start_factor,
430 size_base, size_factor,
431 (type == MTRR_TYPE_UNCACHABLE) ? "UC" :
432 ((type == MTRR_TYPE_WRBACK) ? "WB" : "Other")
433 );
434 }
435 save_var_mtrr(reg++, range_startk, sizek, type);
436 range_startk += sizek;
437 range_sizek -= sizek;
438 if (reg >= num_var_ranges)
439 break;
440 }
441 return reg;
442 }
443
444 static unsigned __init
445 range_to_mtrr_with_hole(struct var_mtrr_state *state, unsigned long basek,
446 unsigned long sizek)
447 {
448 unsigned long hole_basek, hole_sizek;
449 unsigned long second_basek, second_sizek;
450 unsigned long range0_basek, range0_sizek;
451 unsigned long range_basek, range_sizek;
452 unsigned long chunk_sizek;
453 unsigned long gran_sizek;
454
455 hole_basek = 0;
456 hole_sizek = 0;
457 second_basek = 0;
458 second_sizek = 0;
459 chunk_sizek = state->chunk_sizek;
460 gran_sizek = state->gran_sizek;
461
462 /* Align with gran size, prevent small block used up MTRRs: */
463 range_basek = ALIGN(state->range_startk, gran_sizek);
464 if ((range_basek > basek) && basek)
465 return second_sizek;
466
467 state->range_sizek -= (range_basek - state->range_startk);
468 range_sizek = ALIGN(state->range_sizek, gran_sizek);
469
470 while (range_sizek > state->range_sizek) {
471 range_sizek -= gran_sizek;
472 if (!range_sizek)
473 return 0;
474 }
475 state->range_sizek = range_sizek;
476
477 /* Try to append some small hole: */
478 range0_basek = state->range_startk;
479 range0_sizek = ALIGN(state->range_sizek, chunk_sizek);
480
481 /* No increase: */
482 if (range0_sizek == state->range_sizek) {
483 Dprintk("rangeX: %016lx - %016lx\n",
484 range0_basek<<10,
485 (range0_basek + state->range_sizek)<<10);
486 state->reg = range_to_mtrr(state->reg, range0_basek,
487 state->range_sizek, MTRR_TYPE_WRBACK);
488 return 0;
489 }
490
491 /* Only cut back when it is not the last: */
492 if (sizek) {
493 while (range0_basek + range0_sizek > (basek + sizek)) {
494 if (range0_sizek >= chunk_sizek)
495 range0_sizek -= chunk_sizek;
496 else
497 range0_sizek = 0;
498
499 if (!range0_sizek)
500 break;
501 }
502 }
503
504 second_try:
505 range_basek = range0_basek + range0_sizek;
506
507 /* One hole in the middle: */
508 if (range_basek > basek && range_basek <= (basek + sizek))
509 second_sizek = range_basek - basek;
510
511 if (range0_sizek > state->range_sizek) {
512
513 /* One hole in middle or at the end: */
514 hole_sizek = range0_sizek - state->range_sizek - second_sizek;
515
516 /* Hole size should be less than half of range0 size: */
517 if (hole_sizek >= (range0_sizek >> 1) &&
518 range0_sizek >= chunk_sizek) {
519 range0_sizek -= chunk_sizek;
520 second_sizek = 0;
521 hole_sizek = 0;
522
523 goto second_try;
524 }
525 }
526
527 if (range0_sizek) {
528 Dprintk("range0: %016lx - %016lx\n",
529 range0_basek<<10,
530 (range0_basek + range0_sizek)<<10);
531 state->reg = range_to_mtrr(state->reg, range0_basek,
532 range0_sizek, MTRR_TYPE_WRBACK);
533 }
534
535 if (range0_sizek < state->range_sizek) {
536 /* Need to handle left over range: */
537 range_sizek = state->range_sizek - range0_sizek;
538
539 Dprintk("range: %016lx - %016lx\n",
540 range_basek<<10,
541 (range_basek + range_sizek)<<10);
542
543 state->reg = range_to_mtrr(state->reg, range_basek,
544 range_sizek, MTRR_TYPE_WRBACK);
545 }
546
547 if (hole_sizek) {
548 hole_basek = range_basek - hole_sizek - second_sizek;
549 Dprintk("hole: %016lx - %016lx\n",
550 hole_basek<<10,
551 (hole_basek + hole_sizek)<<10);
552 state->reg = range_to_mtrr(state->reg, hole_basek,
553 hole_sizek, MTRR_TYPE_UNCACHABLE);
554 }
555
556 return second_sizek;
557 }
558
559 static void __init
560 set_var_mtrr_range(struct var_mtrr_state *state, unsigned long base_pfn,
561 unsigned long size_pfn)
562 {
563 unsigned long basek, sizek;
564 unsigned long second_sizek = 0;
565
566 if (state->reg >= num_var_ranges)
567 return;
568
569 basek = base_pfn << (PAGE_SHIFT - 10);
570 sizek = size_pfn << (PAGE_SHIFT - 10);
571
572 /* See if I can merge with the last range: */
573 if ((basek <= 1024) ||
574 (state->range_startk + state->range_sizek == basek)) {
575 unsigned long endk = basek + sizek;
576 state->range_sizek = endk - state->range_startk;
577 return;
578 }
579 /* Write the range mtrrs: */
580 if (state->range_sizek != 0)
581 second_sizek = range_to_mtrr_with_hole(state, basek, sizek);
582
583 /* Allocate an msr: */
584 state->range_startk = basek + second_sizek;
585 state->range_sizek = sizek - second_sizek;
586 }
587
588 /* Mininum size of mtrr block that can take hole: */
589 static u64 mtrr_chunk_size __initdata = (256ULL<<20);
590
591 static int __init parse_mtrr_chunk_size_opt(char *p)
592 {
593 if (!p)
594 return -EINVAL;
595 mtrr_chunk_size = memparse(p, &p);
596 return 0;
597 }
598 early_param("mtrr_chunk_size", parse_mtrr_chunk_size_opt);
599
600 /* Granularity of mtrr of block: */
601 static u64 mtrr_gran_size __initdata;
602
603 static int __init parse_mtrr_gran_size_opt(char *p)
604 {
605 if (!p)
606 return -EINVAL;
607 mtrr_gran_size = memparse(p, &p);
608 return 0;
609 }
610 early_param("mtrr_gran_size", parse_mtrr_gran_size_opt);
611
612 static unsigned long nr_mtrr_spare_reg __initdata =
613 CONFIG_MTRR_SANITIZER_SPARE_REG_NR_DEFAULT;
614
615 static int __init parse_mtrr_spare_reg(char *arg)
616 {
617 if (arg)
618 nr_mtrr_spare_reg = simple_strtoul(arg, NULL, 0);
619 return 0;
620 }
621 early_param("mtrr_spare_reg_nr", parse_mtrr_spare_reg);
622
623 static int __init
624 x86_setup_var_mtrrs(struct res_range *range, int nr_range,
625 u64 chunk_size, u64 gran_size)
626 {
627 struct var_mtrr_state var_state;
628 int num_reg;
629 int i;
630
631 var_state.range_startk = 0;
632 var_state.range_sizek = 0;
633 var_state.reg = 0;
634 var_state.chunk_sizek = chunk_size >> 10;
635 var_state.gran_sizek = gran_size >> 10;
636
637 memset(range_state, 0, sizeof(range_state));
638
639 /* Write the range: */
640 for (i = 0; i < nr_range; i++) {
641 set_var_mtrr_range(&var_state, range[i].start,
642 range[i].end - range[i].start + 1);
643 }
644
645 /* Write the last range: */
646 if (var_state.range_sizek != 0)
647 range_to_mtrr_with_hole(&var_state, 0, 0);
648
649 num_reg = var_state.reg;
650 /* Clear out the extra MTRR's: */
651 while (var_state.reg < num_var_ranges) {
652 save_var_mtrr(var_state.reg, 0, 0, 0);
653 var_state.reg++;
654 }
655
656 return num_reg;
657 }
658
659 struct mtrr_cleanup_result {
660 unsigned long gran_sizek;
661 unsigned long chunk_sizek;
662 unsigned long lose_cover_sizek;
663 unsigned int num_reg;
664 int bad;
665 };
666
667 /*
668 * gran_size: 64K, 128K, 256K, 512K, 1M, 2M, ..., 2G
669 * chunk size: gran_size, ..., 2G
670 * so we need (1+16)*8
671 */
672 #define NUM_RESULT 136
673 #define PSHIFT (PAGE_SHIFT - 10)
674
675 static struct mtrr_cleanup_result __initdata result[NUM_RESULT];
676 static unsigned long __initdata min_loss_pfn[RANGE_NUM];
677
678 static void __init print_out_mtrr_range_state(void)
679 {
680 char start_factor = 'K', size_factor = 'K';
681 unsigned long start_base, size_base;
682 mtrr_type type;
683 int i;
684
685 for (i = 0; i < num_var_ranges; i++) {
686
687 size_base = range_state[i].size_pfn << (PAGE_SHIFT - 10);
688 if (!size_base)
689 continue;
690
691 size_base = to_size_factor(size_base, &size_factor),
692 start_base = range_state[i].base_pfn << (PAGE_SHIFT - 10);
693 start_base = to_size_factor(start_base, &start_factor),
694 type = range_state[i].type;
695
696 printk(KERN_DEBUG "reg %d, base: %ld%cB, range: %ld%cB, type %s\n",
697 i, start_base, start_factor,
698 size_base, size_factor,
699 (type == MTRR_TYPE_UNCACHABLE) ? "UC" :
700 ((type == MTRR_TYPE_WRPROT) ? "WP" :
701 ((type == MTRR_TYPE_WRBACK) ? "WB" : "Other"))
702 );
703 }
704 }
705
706 static int __init mtrr_need_cleanup(void)
707 {
708 int i;
709 mtrr_type type;
710 unsigned long size;
711 /* Extra one for all 0: */
712 int num[MTRR_NUM_TYPES + 1];
713
714 /* Check entries number: */
715 memset(num, 0, sizeof(num));
716 for (i = 0; i < num_var_ranges; i++) {
717 type = range_state[i].type;
718 size = range_state[i].size_pfn;
719 if (type >= MTRR_NUM_TYPES)
720 continue;
721 if (!size)
722 type = MTRR_NUM_TYPES;
723 num[type]++;
724 }
725
726 /* Check if we got UC entries: */
727 if (!num[MTRR_TYPE_UNCACHABLE])
728 return 0;
729
730 /* Check if we only had WB and UC */
731 if (num[MTRR_TYPE_WRBACK] + num[MTRR_TYPE_UNCACHABLE] !=
732 num_var_ranges - num[MTRR_NUM_TYPES])
733 return 0;
734
735 return 1;
736 }
737
738 static unsigned long __initdata range_sums;
739
740 static void __init
741 mtrr_calc_range_state(u64 chunk_size, u64 gran_size,
742 unsigned long x_remove_base,
743 unsigned long x_remove_size, int i)
744 {
745 static struct res_range range_new[RANGE_NUM];
746 unsigned long range_sums_new;
747 static int nr_range_new;
748 int num_reg;
749
750 /* Convert ranges to var ranges state: */
751 num_reg = x86_setup_var_mtrrs(range, nr_range, chunk_size, gran_size);
752
753 /* We got new setting in range_state, check it: */
754 memset(range_new, 0, sizeof(range_new));
755 nr_range_new = x86_get_mtrr_mem_range(range_new, 0,
756 x_remove_base, x_remove_size);
757 range_sums_new = sum_ranges(range_new, nr_range_new);
758
759 result[i].chunk_sizek = chunk_size >> 10;
760 result[i].gran_sizek = gran_size >> 10;
761 result[i].num_reg = num_reg;
762
763 if (range_sums < range_sums_new) {
764 result[i].lose_cover_sizek = (range_sums_new - range_sums) << PSHIFT;
765 result[i].bad = 1;
766 } else {
767 result[i].lose_cover_sizek = (range_sums - range_sums_new) << PSHIFT;
768 }
769
770 /* Double check it: */
771 if (!result[i].bad && !result[i].lose_cover_sizek) {
772 if (nr_range_new != nr_range || memcmp(range, range_new, sizeof(range)))
773 result[i].bad = 1;
774 }
775
776 if (!result[i].bad && (range_sums - range_sums_new < min_loss_pfn[num_reg]))
777 min_loss_pfn[num_reg] = range_sums - range_sums_new;
778 }
779
780 static void __init mtrr_print_out_one_result(int i)
781 {
782 unsigned long gran_base, chunk_base, lose_base;
783 char gran_factor, chunk_factor, lose_factor;
784
785 gran_base = to_size_factor(result[i].gran_sizek, &gran_factor),
786 chunk_base = to_size_factor(result[i].chunk_sizek, &chunk_factor),
787 lose_base = to_size_factor(result[i].lose_cover_sizek, &lose_factor),
788
789 pr_info("%sgran_size: %ld%c \tchunk_size: %ld%c \t",
790 result[i].bad ? "*BAD*" : " ",
791 gran_base, gran_factor, chunk_base, chunk_factor);
792 pr_cont("num_reg: %d \tlose cover RAM: %s%ld%c\n",
793 result[i].num_reg, result[i].bad ? "-" : "",
794 lose_base, lose_factor);
795 }
796
797 static int __init mtrr_search_optimal_index(void)
798 {
799 int num_reg_good;
800 int index_good;
801 int i;
802
803 if (nr_mtrr_spare_reg >= num_var_ranges)
804 nr_mtrr_spare_reg = num_var_ranges - 1;
805
806 num_reg_good = -1;
807 for (i = num_var_ranges - nr_mtrr_spare_reg; i > 0; i--) {
808 if (!min_loss_pfn[i])
809 num_reg_good = i;
810 }
811
812 index_good = -1;
813 if (num_reg_good != -1) {
814 for (i = 0; i < NUM_RESULT; i++) {
815 if (!result[i].bad &&
816 result[i].num_reg == num_reg_good &&
817 !result[i].lose_cover_sizek) {
818 index_good = i;
819 break;
820 }
821 }
822 }
823
824 return index_good;
825 }
826
827 int __init mtrr_cleanup(unsigned address_bits)
828 {
829 unsigned long x_remove_base, x_remove_size;
830 unsigned long base, size, def, dummy;
831 u64 chunk_size, gran_size;
832 mtrr_type type;
833 int index_good;
834 int i;
835
836 if (!is_cpu(INTEL) || enable_mtrr_cleanup < 1)
837 return 0;
838
839 rdmsr(MSR_MTRRdefType, def, dummy);
840 def &= 0xff;
841 if (def != MTRR_TYPE_UNCACHABLE)
842 return 0;
843
844 /* Get it and store it aside: */
845 memset(range_state, 0, sizeof(range_state));
846 for (i = 0; i < num_var_ranges; i++) {
847 mtrr_if->get(i, &base, &size, &type);
848 range_state[i].base_pfn = base;
849 range_state[i].size_pfn = size;
850 range_state[i].type = type;
851 }
852
853 /* Check if we need handle it and can handle it: */
854 if (!mtrr_need_cleanup())
855 return 0;
856
857 /* Print original var MTRRs at first, for debugging: */
858 printk(KERN_DEBUG "original variable MTRRs\n");
859 print_out_mtrr_range_state();
860
861 memset(range, 0, sizeof(range));
862 x_remove_size = 0;
863 x_remove_base = 1 << (32 - PAGE_SHIFT);
864 if (mtrr_tom2)
865 x_remove_size = (mtrr_tom2 >> PAGE_SHIFT) - x_remove_base;
866
867 nr_range = x86_get_mtrr_mem_range(range, 0, x_remove_base, x_remove_size);
868 /*
869 * [0, 1M) should always be covered by var mtrr with WB
870 * and fixed mtrrs should take effect before var mtrr for it:
871 */
872 nr_range = add_range_with_merge(range, nr_range, 0,
873 (1ULL<<(20 - PAGE_SHIFT)) - 1);
874 /* Sort the ranges: */
875 sort(range, nr_range, sizeof(struct res_range), cmp_range, NULL);
876
877 range_sums = sum_ranges(range, nr_range);
878 printk(KERN_INFO "total RAM covered: %ldM\n",
879 range_sums >> (20 - PAGE_SHIFT));
880
881 if (mtrr_chunk_size && mtrr_gran_size) {
882 i = 0;
883 mtrr_calc_range_state(mtrr_chunk_size, mtrr_gran_size,
884 x_remove_base, x_remove_size, i);
885
886 mtrr_print_out_one_result(i);
887
888 if (!result[i].bad) {
889 set_var_mtrr_all(address_bits);
890 printk(KERN_DEBUG "New variable MTRRs\n");
891 print_out_mtrr_range_state();
892 return 1;
893 }
894 printk(KERN_INFO "invalid mtrr_gran_size or mtrr_chunk_size, "
895 "will find optimal one\n");
896 }
897
898 i = 0;
899 memset(min_loss_pfn, 0xff, sizeof(min_loss_pfn));
900 memset(result, 0, sizeof(result));
901 for (gran_size = (1ULL<<16); gran_size < (1ULL<<32); gran_size <<= 1) {
902
903 for (chunk_size = gran_size; chunk_size < (1ULL<<32);
904 chunk_size <<= 1) {
905
906 if (i >= NUM_RESULT)
907 continue;
908
909 mtrr_calc_range_state(chunk_size, gran_size,
910 x_remove_base, x_remove_size, i);
911 if (debug_print) {
912 mtrr_print_out_one_result(i);
913 printk(KERN_INFO "\n");
914 }
915
916 i++;
917 }
918 }
919
920 /* Try to find the optimal index: */
921 index_good = mtrr_search_optimal_index();
922
923 if (index_good != -1) {
924 printk(KERN_INFO "Found optimal setting for mtrr clean up\n");
925 i = index_good;
926 mtrr_print_out_one_result(i);
927
928 /* Convert ranges to var ranges state: */
929 chunk_size = result[i].chunk_sizek;
930 chunk_size <<= 10;
931 gran_size = result[i].gran_sizek;
932 gran_size <<= 10;
933 x86_setup_var_mtrrs(range, nr_range, chunk_size, gran_size);
934 set_var_mtrr_all(address_bits);
935 printk(KERN_DEBUG "New variable MTRRs\n");
936 print_out_mtrr_range_state();
937 return 1;
938 } else {
939 /* print out all */
940 for (i = 0; i < NUM_RESULT; i++)
941 mtrr_print_out_one_result(i);
942 }
943
944 printk(KERN_INFO "mtrr_cleanup: can not find optimal value\n");
945 printk(KERN_INFO "please specify mtrr_gran_size/mtrr_chunk_size\n");
946
947 return 0;
948 }
949 #else
950 int __init mtrr_cleanup(unsigned address_bits)
951 {
952 return 0;
953 }
954 #endif
955
956 static int disable_mtrr_trim;
957
958 static int __init disable_mtrr_trim_setup(char *str)
959 {
960 disable_mtrr_trim = 1;
961 return 0;
962 }
963 early_param("disable_mtrr_trim", disable_mtrr_trim_setup);
964
965 /*
966 * Newer AMD K8s and later CPUs have a special magic MSR way to force WB
967 * for memory >4GB. Check for that here.
968 * Note this won't check if the MTRRs < 4GB where the magic bit doesn't
969 * apply to are wrong, but so far we don't know of any such case in the wild.
970 */
971 #define Tom2Enabled (1U << 21)
972 #define Tom2ForceMemTypeWB (1U << 22)
973
974 int __init amd_special_default_mtrr(void)
975 {
976 u32 l, h;
977
978 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD)
979 return 0;
980 if (boot_cpu_data.x86 < 0xf || boot_cpu_data.x86 > 0x11)
981 return 0;
982 /* In case some hypervisor doesn't pass SYSCFG through: */
983 if (rdmsr_safe(MSR_K8_SYSCFG, &l, &h) < 0)
984 return 0;
985 /*
986 * Memory between 4GB and top of mem is forced WB by this magic bit.
987 * Reserved before K8RevF, but should be zero there.
988 */
989 if ((l & (Tom2Enabled | Tom2ForceMemTypeWB)) ==
990 (Tom2Enabled | Tom2ForceMemTypeWB))
991 return 1;
992 return 0;
993 }
994
995 static u64 __init
996 real_trim_memory(unsigned long start_pfn, unsigned long limit_pfn)
997 {
998 u64 trim_start, trim_size;
999
1000 trim_start = start_pfn;
1001 trim_start <<= PAGE_SHIFT;
1002
1003 trim_size = limit_pfn;
1004 trim_size <<= PAGE_SHIFT;
1005 trim_size -= trim_start;
1006
1007 return e820_update_range(trim_start, trim_size, E820_RAM, E820_RESERVED);
1008 }
1009
1010 /**
1011 * mtrr_trim_uncached_memory - trim RAM not covered by MTRRs
1012 * @end_pfn: ending page frame number
1013 *
1014 * Some buggy BIOSes don't setup the MTRRs properly for systems with certain
1015 * memory configurations. This routine checks that the highest MTRR matches
1016 * the end of memory, to make sure the MTRRs having a write back type cover
1017 * all of the memory the kernel is intending to use. If not, it'll trim any
1018 * memory off the end by adjusting end_pfn, removing it from the kernel's
1019 * allocation pools, warning the user with an obnoxious message.
1020 */
1021 int __init mtrr_trim_uncached_memory(unsigned long end_pfn)
1022 {
1023 unsigned long i, base, size, highest_pfn = 0, def, dummy;
1024 mtrr_type type;
1025 u64 total_trim_size;
1026 /* extra one for all 0 */
1027 int num[MTRR_NUM_TYPES + 1];
1028
1029 /*
1030 * Make sure we only trim uncachable memory on machines that
1031 * support the Intel MTRR architecture:
1032 */
1033 if (!is_cpu(INTEL) || disable_mtrr_trim)
1034 return 0;
1035
1036 rdmsr(MSR_MTRRdefType, def, dummy);
1037 def &= 0xff;
1038 if (def != MTRR_TYPE_UNCACHABLE)
1039 return 0;
1040
1041 /* Get it and store it aside: */
1042 memset(range_state, 0, sizeof(range_state));
1043 for (i = 0; i < num_var_ranges; i++) {
1044 mtrr_if->get(i, &base, &size, &type);
1045 range_state[i].base_pfn = base;
1046 range_state[i].size_pfn = size;
1047 range_state[i].type = type;
1048 }
1049
1050 /* Find highest cached pfn: */
1051 for (i = 0; i < num_var_ranges; i++) {
1052 type = range_state[i].type;
1053 if (type != MTRR_TYPE_WRBACK)
1054 continue;
1055 base = range_state[i].base_pfn;
1056 size = range_state[i].size_pfn;
1057 if (highest_pfn < base + size)
1058 highest_pfn = base + size;
1059 }
1060
1061 /* kvm/qemu doesn't have mtrr set right, don't trim them all: */
1062 if (!highest_pfn) {
1063 printk(KERN_INFO "CPU MTRRs all blank - virtualized system.\n");
1064 return 0;
1065 }
1066
1067 /* Check entries number: */
1068 memset(num, 0, sizeof(num));
1069 for (i = 0; i < num_var_ranges; i++) {
1070 type = range_state[i].type;
1071 if (type >= MTRR_NUM_TYPES)
1072 continue;
1073 size = range_state[i].size_pfn;
1074 if (!size)
1075 type = MTRR_NUM_TYPES;
1076 num[type]++;
1077 }
1078
1079 /* No entry for WB? */
1080 if (!num[MTRR_TYPE_WRBACK])
1081 return 0;
1082
1083 /* Check if we only had WB and UC: */
1084 if (num[MTRR_TYPE_WRBACK] + num[MTRR_TYPE_UNCACHABLE] !=
1085 num_var_ranges - num[MTRR_NUM_TYPES])
1086 return 0;
1087
1088 memset(range, 0, sizeof(range));
1089 nr_range = 0;
1090 if (mtrr_tom2) {
1091 range[nr_range].start = (1ULL<<(32 - PAGE_SHIFT));
1092 range[nr_range].end = (mtrr_tom2 >> PAGE_SHIFT) - 1;
1093 if (highest_pfn < range[nr_range].end + 1)
1094 highest_pfn = range[nr_range].end + 1;
1095 nr_range++;
1096 }
1097 nr_range = x86_get_mtrr_mem_range(range, nr_range, 0, 0);
1098
1099 /* Check the head: */
1100 total_trim_size = 0;
1101 if (range[0].start)
1102 total_trim_size += real_trim_memory(0, range[0].start);
1103
1104 /* Check the holes: */
1105 for (i = 0; i < nr_range - 1; i++) {
1106 if (range[i].end + 1 < range[i+1].start)
1107 total_trim_size += real_trim_memory(range[i].end + 1,
1108 range[i+1].start);
1109 }
1110
1111 /* Check the top: */
1112 i = nr_range - 1;
1113 if (range[i].end + 1 < end_pfn)
1114 total_trim_size += real_trim_memory(range[i].end + 1,
1115 end_pfn);
1116
1117 if (total_trim_size) {
1118 pr_warning("WARNING: BIOS bug: CPU MTRRs don't cover all of memory, losing %lluMB of RAM.\n", total_trim_size >> 20);
1119
1120 if (!changed_by_mtrr_cleanup)
1121 WARN_ON(1);
1122
1123 pr_info("update e820 for mtrr\n");
1124 update_e820();
1125
1126 return 1;
1127 }
1128
1129 return 0;
1130 }
kernel-based virtual machine