xref: /openbmc/linux/arch/x86/kernel/cpu/mtrr/cleanup.c (revision 2359ccdd)
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/init.h>
21 #include <linux/pci.h>
22 #include <linux/smp.h>
23 #include <linux/cpu.h>
24 #include <linux/mutex.h>
25 #include <linux/uaccess.h>
26 #include <linux/kvm_para.h>
27 #include <linux/range.h>
28 
29 #include <asm/processor.h>
30 #include <asm/e820/api.h>
31 #include <asm/mtrr.h>
32 #include <asm/msr.h>
33 
34 #include "mtrr.h"
35 
36 struct var_mtrr_range_state {
37 	unsigned long	base_pfn;
38 	unsigned long	size_pfn;
39 	mtrr_type	type;
40 };
41 
42 struct var_mtrr_state {
43 	unsigned long	range_startk;
44 	unsigned long	range_sizek;
45 	unsigned long	chunk_sizek;
46 	unsigned long	gran_sizek;
47 	unsigned int	reg;
48 };
49 
50 /* Should be related to MTRR_VAR_RANGES nums */
51 #define RANGE_NUM				256
52 
53 static struct range __initdata		range[RANGE_NUM];
54 static int __initdata				nr_range;
55 
56 static struct var_mtrr_range_state __initdata	range_state[RANGE_NUM];
57 
58 static int __initdata debug_print;
59 #define Dprintk(x...) do { if (debug_print) pr_debug(x); } while (0)
60 
61 #define BIOS_BUG_MSG \
62 	"WARNING: BIOS bug: VAR MTRR %d contains strange UC entry under 1M, check with your system vendor!\n"
63 
64 static int __init
65 x86_get_mtrr_mem_range(struct range *range, int nr_range,
66 		       unsigned long extra_remove_base,
67 		       unsigned long extra_remove_size)
68 {
69 	unsigned long base, size;
70 	mtrr_type type;
71 	int i;
72 
73 	for (i = 0; i < num_var_ranges; i++) {
74 		type = range_state[i].type;
75 		if (type != MTRR_TYPE_WRBACK)
76 			continue;
77 		base = range_state[i].base_pfn;
78 		size = range_state[i].size_pfn;
79 		nr_range = add_range_with_merge(range, RANGE_NUM, nr_range,
80 						base, base + size);
81 	}
82 	if (debug_print) {
83 		pr_debug("After WB checking\n");
84 		for (i = 0; i < nr_range; i++)
85 			pr_debug("MTRR MAP PFN: %016llx - %016llx\n",
86 				 range[i].start, range[i].end);
87 	}
88 
89 	/* Take out UC ranges: */
90 	for (i = 0; i < num_var_ranges; i++) {
91 		type = range_state[i].type;
92 		if (type != MTRR_TYPE_UNCACHABLE &&
93 		    type != MTRR_TYPE_WRPROT)
94 			continue;
95 		size = range_state[i].size_pfn;
96 		if (!size)
97 			continue;
98 		base = range_state[i].base_pfn;
99 		if (base < (1<<(20-PAGE_SHIFT)) && mtrr_state.have_fixed &&
100 		    (mtrr_state.enabled & MTRR_STATE_MTRR_ENABLED) &&
101 		    (mtrr_state.enabled & MTRR_STATE_MTRR_FIXED_ENABLED)) {
102 			/* Var MTRR contains UC entry below 1M? Skip it: */
103 			pr_warn(BIOS_BUG_MSG, i);
104 			if (base + size <= (1<<(20-PAGE_SHIFT)))
105 				continue;
106 			size -= (1<<(20-PAGE_SHIFT)) - base;
107 			base = 1<<(20-PAGE_SHIFT);
108 		}
109 		subtract_range(range, RANGE_NUM, base, base + size);
110 	}
111 	if (extra_remove_size)
112 		subtract_range(range, RANGE_NUM, extra_remove_base,
113 				 extra_remove_base + extra_remove_size);
114 
115 	if  (debug_print) {
116 		pr_debug("After UC checking\n");
117 		for (i = 0; i < RANGE_NUM; i++) {
118 			if (!range[i].end)
119 				continue;
120 			pr_debug("MTRR MAP PFN: %016llx - %016llx\n",
121 				 range[i].start, range[i].end);
122 		}
123 	}
124 
125 	/* sort the ranges */
126 	nr_range = clean_sort_range(range, RANGE_NUM);
127 	if  (debug_print) {
128 		pr_debug("After sorting\n");
129 		for (i = 0; i < nr_range; i++)
130 			pr_debug("MTRR MAP PFN: %016llx - %016llx\n",
131 				 range[i].start, range[i].end);
132 	}
133 
134 	return nr_range;
135 }
136 
137 #ifdef CONFIG_MTRR_SANITIZER
138 
139 static unsigned long __init sum_ranges(struct range *range, int nr_range)
140 {
141 	unsigned long sum = 0;
142 	int i;
143 
144 	for (i = 0; i < nr_range; i++)
145 		sum += range[i].end - range[i].start;
146 
147 	return sum;
148 }
149 
150 static int enable_mtrr_cleanup __initdata =
151 	CONFIG_MTRR_SANITIZER_ENABLE_DEFAULT;
152 
153 static int __init disable_mtrr_cleanup_setup(char *str)
154 {
155 	enable_mtrr_cleanup = 0;
156 	return 0;
157 }
158 early_param("disable_mtrr_cleanup", disable_mtrr_cleanup_setup);
159 
160 static int __init enable_mtrr_cleanup_setup(char *str)
161 {
162 	enable_mtrr_cleanup = 1;
163 	return 0;
164 }
165 early_param("enable_mtrr_cleanup", enable_mtrr_cleanup_setup);
166 
167 static int __init mtrr_cleanup_debug_setup(char *str)
168 {
169 	debug_print = 1;
170 	return 0;
171 }
172 early_param("mtrr_cleanup_debug", mtrr_cleanup_debug_setup);
173 
174 static void __init
175 set_var_mtrr(unsigned int reg, unsigned long basek, unsigned long sizek,
176 	     unsigned char type, unsigned int address_bits)
177 {
178 	u32 base_lo, base_hi, mask_lo, mask_hi;
179 	u64 base, mask;
180 
181 	if (!sizek) {
182 		fill_mtrr_var_range(reg, 0, 0, 0, 0);
183 		return;
184 	}
185 
186 	mask = (1ULL << address_bits) - 1;
187 	mask &= ~((((u64)sizek) << 10) - 1);
188 
189 	base = ((u64)basek) << 10;
190 
191 	base |= type;
192 	mask |= 0x800;
193 
194 	base_lo = base & ((1ULL<<32) - 1);
195 	base_hi = base >> 32;
196 
197 	mask_lo = mask & ((1ULL<<32) - 1);
198 	mask_hi = mask >> 32;
199 
200 	fill_mtrr_var_range(reg, base_lo, base_hi, mask_lo, mask_hi);
201 }
202 
203 static void __init
204 save_var_mtrr(unsigned int reg, unsigned long basek, unsigned long sizek,
205 	      unsigned char type)
206 {
207 	range_state[reg].base_pfn = basek >> (PAGE_SHIFT - 10);
208 	range_state[reg].size_pfn = sizek >> (PAGE_SHIFT - 10);
209 	range_state[reg].type = type;
210 }
211 
212 static void __init set_var_mtrr_all(unsigned int address_bits)
213 {
214 	unsigned long basek, sizek;
215 	unsigned char type;
216 	unsigned int reg;
217 
218 	for (reg = 0; reg < num_var_ranges; reg++) {
219 		basek = range_state[reg].base_pfn << (PAGE_SHIFT - 10);
220 		sizek = range_state[reg].size_pfn << (PAGE_SHIFT - 10);
221 		type = range_state[reg].type;
222 
223 		set_var_mtrr(reg, basek, sizek, type, address_bits);
224 	}
225 }
226 
227 static unsigned long to_size_factor(unsigned long sizek, char *factorp)
228 {
229 	unsigned long base = sizek;
230 	char factor;
231 
232 	if (base & ((1<<10) - 1)) {
233 		/* Not MB-aligned: */
234 		factor = 'K';
235 	} else if (base & ((1<<20) - 1)) {
236 		factor = 'M';
237 		base >>= 10;
238 	} else {
239 		factor = 'G';
240 		base >>= 20;
241 	}
242 
243 	*factorp = factor;
244 
245 	return base;
246 }
247 
248 static unsigned int __init
249 range_to_mtrr(unsigned int reg, unsigned long range_startk,
250 	      unsigned long range_sizek, unsigned char type)
251 {
252 	if (!range_sizek || (reg >= num_var_ranges))
253 		return reg;
254 
255 	while (range_sizek) {
256 		unsigned long max_align, align;
257 		unsigned long sizek;
258 
259 		/* Compute the maximum size with which we can make a range: */
260 		if (range_startk)
261 			max_align = __ffs(range_startk);
262 		else
263 			max_align = BITS_PER_LONG - 1;
264 
265 		align = __fls(range_sizek);
266 		if (align > max_align)
267 			align = max_align;
268 
269 		sizek = 1UL << align;
270 		if (debug_print) {
271 			char start_factor = 'K', size_factor = 'K';
272 			unsigned long start_base, size_base;
273 
274 			start_base = to_size_factor(range_startk, &start_factor);
275 			size_base = to_size_factor(sizek, &size_factor);
276 
277 			Dprintk("Setting variable MTRR %d, "
278 				"base: %ld%cB, range: %ld%cB, type %s\n",
279 				reg, start_base, start_factor,
280 				size_base, size_factor,
281 				(type == MTRR_TYPE_UNCACHABLE) ? "UC" :
282 				   ((type == MTRR_TYPE_WRBACK) ? "WB" : "Other")
283 				);
284 		}
285 		save_var_mtrr(reg++, range_startk, sizek, type);
286 		range_startk += sizek;
287 		range_sizek -= sizek;
288 		if (reg >= num_var_ranges)
289 			break;
290 	}
291 	return reg;
292 }
293 
294 static unsigned __init
295 range_to_mtrr_with_hole(struct var_mtrr_state *state, unsigned long basek,
296 			unsigned long sizek)
297 {
298 	unsigned long hole_basek, hole_sizek;
299 	unsigned long second_basek, second_sizek;
300 	unsigned long range0_basek, range0_sizek;
301 	unsigned long range_basek, range_sizek;
302 	unsigned long chunk_sizek;
303 	unsigned long gran_sizek;
304 
305 	hole_basek = 0;
306 	hole_sizek = 0;
307 	second_basek = 0;
308 	second_sizek = 0;
309 	chunk_sizek = state->chunk_sizek;
310 	gran_sizek = state->gran_sizek;
311 
312 	/* Align with gran size, prevent small block used up MTRRs: */
313 	range_basek = ALIGN(state->range_startk, gran_sizek);
314 	if ((range_basek > basek) && basek)
315 		return second_sizek;
316 
317 	state->range_sizek -= (range_basek - state->range_startk);
318 	range_sizek = ALIGN(state->range_sizek, gran_sizek);
319 
320 	while (range_sizek > state->range_sizek) {
321 		range_sizek -= gran_sizek;
322 		if (!range_sizek)
323 			return 0;
324 	}
325 	state->range_sizek = range_sizek;
326 
327 	/* Try to append some small hole: */
328 	range0_basek = state->range_startk;
329 	range0_sizek = ALIGN(state->range_sizek, chunk_sizek);
330 
331 	/* No increase: */
332 	if (range0_sizek == state->range_sizek) {
333 		Dprintk("rangeX: %016lx - %016lx\n",
334 			range0_basek<<10,
335 			(range0_basek + state->range_sizek)<<10);
336 		state->reg = range_to_mtrr(state->reg, range0_basek,
337 				state->range_sizek, MTRR_TYPE_WRBACK);
338 		return 0;
339 	}
340 
341 	/* Only cut back when it is not the last: */
342 	if (sizek) {
343 		while (range0_basek + range0_sizek > (basek + sizek)) {
344 			if (range0_sizek >= chunk_sizek)
345 				range0_sizek -= chunk_sizek;
346 			else
347 				range0_sizek = 0;
348 
349 			if (!range0_sizek)
350 				break;
351 		}
352 	}
353 
354 second_try:
355 	range_basek = range0_basek + range0_sizek;
356 
357 	/* One hole in the middle: */
358 	if (range_basek > basek && range_basek <= (basek + sizek))
359 		second_sizek = range_basek - basek;
360 
361 	if (range0_sizek > state->range_sizek) {
362 
363 		/* One hole in middle or at the end: */
364 		hole_sizek = range0_sizek - state->range_sizek - second_sizek;
365 
366 		/* Hole size should be less than half of range0 size: */
367 		if (hole_sizek >= (range0_sizek >> 1) &&
368 		    range0_sizek >= chunk_sizek) {
369 			range0_sizek -= chunk_sizek;
370 			second_sizek = 0;
371 			hole_sizek = 0;
372 
373 			goto second_try;
374 		}
375 	}
376 
377 	if (range0_sizek) {
378 		Dprintk("range0: %016lx - %016lx\n",
379 			range0_basek<<10,
380 			(range0_basek + range0_sizek)<<10);
381 		state->reg = range_to_mtrr(state->reg, range0_basek,
382 				range0_sizek, MTRR_TYPE_WRBACK);
383 	}
384 
385 	if (range0_sizek < state->range_sizek) {
386 		/* Need to handle left over range: */
387 		range_sizek = state->range_sizek - range0_sizek;
388 
389 		Dprintk("range: %016lx - %016lx\n",
390 			 range_basek<<10,
391 			 (range_basek + range_sizek)<<10);
392 
393 		state->reg = range_to_mtrr(state->reg, range_basek,
394 				 range_sizek, MTRR_TYPE_WRBACK);
395 	}
396 
397 	if (hole_sizek) {
398 		hole_basek = range_basek - hole_sizek - second_sizek;
399 		Dprintk("hole: %016lx - %016lx\n",
400 			 hole_basek<<10,
401 			 (hole_basek + hole_sizek)<<10);
402 		state->reg = range_to_mtrr(state->reg, hole_basek,
403 				 hole_sizek, MTRR_TYPE_UNCACHABLE);
404 	}
405 
406 	return second_sizek;
407 }
408 
409 static void __init
410 set_var_mtrr_range(struct var_mtrr_state *state, unsigned long base_pfn,
411 		   unsigned long size_pfn)
412 {
413 	unsigned long basek, sizek;
414 	unsigned long second_sizek = 0;
415 
416 	if (state->reg >= num_var_ranges)
417 		return;
418 
419 	basek = base_pfn << (PAGE_SHIFT - 10);
420 	sizek = size_pfn << (PAGE_SHIFT - 10);
421 
422 	/* See if I can merge with the last range: */
423 	if ((basek <= 1024) ||
424 	    (state->range_startk + state->range_sizek == basek)) {
425 		unsigned long endk = basek + sizek;
426 		state->range_sizek = endk - state->range_startk;
427 		return;
428 	}
429 	/* Write the range mtrrs: */
430 	if (state->range_sizek != 0)
431 		second_sizek = range_to_mtrr_with_hole(state, basek, sizek);
432 
433 	/* Allocate an msr: */
434 	state->range_startk = basek + second_sizek;
435 	state->range_sizek  = sizek - second_sizek;
436 }
437 
438 /* Mininum size of mtrr block that can take hole: */
439 static u64 mtrr_chunk_size __initdata = (256ULL<<20);
440 
441 static int __init parse_mtrr_chunk_size_opt(char *p)
442 {
443 	if (!p)
444 		return -EINVAL;
445 	mtrr_chunk_size = memparse(p, &p);
446 	return 0;
447 }
448 early_param("mtrr_chunk_size", parse_mtrr_chunk_size_opt);
449 
450 /* Granularity of mtrr of block: */
451 static u64 mtrr_gran_size __initdata;
452 
453 static int __init parse_mtrr_gran_size_opt(char *p)
454 {
455 	if (!p)
456 		return -EINVAL;
457 	mtrr_gran_size = memparse(p, &p);
458 	return 0;
459 }
460 early_param("mtrr_gran_size", parse_mtrr_gran_size_opt);
461 
462 static unsigned long nr_mtrr_spare_reg __initdata =
463 				 CONFIG_MTRR_SANITIZER_SPARE_REG_NR_DEFAULT;
464 
465 static int __init parse_mtrr_spare_reg(char *arg)
466 {
467 	if (arg)
468 		nr_mtrr_spare_reg = simple_strtoul(arg, NULL, 0);
469 	return 0;
470 }
471 early_param("mtrr_spare_reg_nr", parse_mtrr_spare_reg);
472 
473 static int __init
474 x86_setup_var_mtrrs(struct range *range, int nr_range,
475 		    u64 chunk_size, u64 gran_size)
476 {
477 	struct var_mtrr_state var_state;
478 	int num_reg;
479 	int i;
480 
481 	var_state.range_startk	= 0;
482 	var_state.range_sizek	= 0;
483 	var_state.reg		= 0;
484 	var_state.chunk_sizek	= chunk_size >> 10;
485 	var_state.gran_sizek	= gran_size >> 10;
486 
487 	memset(range_state, 0, sizeof(range_state));
488 
489 	/* Write the range: */
490 	for (i = 0; i < nr_range; i++) {
491 		set_var_mtrr_range(&var_state, range[i].start,
492 				   range[i].end - range[i].start);
493 	}
494 
495 	/* Write the last range: */
496 	if (var_state.range_sizek != 0)
497 		range_to_mtrr_with_hole(&var_state, 0, 0);
498 
499 	num_reg = var_state.reg;
500 	/* Clear out the extra MTRR's: */
501 	while (var_state.reg < num_var_ranges) {
502 		save_var_mtrr(var_state.reg, 0, 0, 0);
503 		var_state.reg++;
504 	}
505 
506 	return num_reg;
507 }
508 
509 struct mtrr_cleanup_result {
510 	unsigned long	gran_sizek;
511 	unsigned long	chunk_sizek;
512 	unsigned long	lose_cover_sizek;
513 	unsigned int	num_reg;
514 	int		bad;
515 };
516 
517 /*
518  * gran_size: 64K, 128K, 256K, 512K, 1M, 2M, ..., 2G
519  * chunk size: gran_size, ..., 2G
520  * so we need (1+16)*8
521  */
522 #define NUM_RESULT	136
523 #define PSHIFT		(PAGE_SHIFT - 10)
524 
525 static struct mtrr_cleanup_result __initdata result[NUM_RESULT];
526 static unsigned long __initdata min_loss_pfn[RANGE_NUM];
527 
528 static void __init print_out_mtrr_range_state(void)
529 {
530 	char start_factor = 'K', size_factor = 'K';
531 	unsigned long start_base, size_base;
532 	mtrr_type type;
533 	int i;
534 
535 	for (i = 0; i < num_var_ranges; i++) {
536 
537 		size_base = range_state[i].size_pfn << (PAGE_SHIFT - 10);
538 		if (!size_base)
539 			continue;
540 
541 		size_base = to_size_factor(size_base, &size_factor),
542 		start_base = range_state[i].base_pfn << (PAGE_SHIFT - 10);
543 		start_base = to_size_factor(start_base, &start_factor),
544 		type = range_state[i].type;
545 
546 		pr_debug("reg %d, base: %ld%cB, range: %ld%cB, type %s\n",
547 			i, start_base, start_factor,
548 			size_base, size_factor,
549 			(type == MTRR_TYPE_UNCACHABLE) ? "UC" :
550 			    ((type == MTRR_TYPE_WRPROT) ? "WP" :
551 			     ((type == MTRR_TYPE_WRBACK) ? "WB" : "Other"))
552 			);
553 	}
554 }
555 
556 static int __init mtrr_need_cleanup(void)
557 {
558 	int i;
559 	mtrr_type type;
560 	unsigned long size;
561 	/* Extra one for all 0: */
562 	int num[MTRR_NUM_TYPES + 1];
563 
564 	/* Check entries number: */
565 	memset(num, 0, sizeof(num));
566 	for (i = 0; i < num_var_ranges; i++) {
567 		type = range_state[i].type;
568 		size = range_state[i].size_pfn;
569 		if (type >= MTRR_NUM_TYPES)
570 			continue;
571 		if (!size)
572 			type = MTRR_NUM_TYPES;
573 		num[type]++;
574 	}
575 
576 	/* Check if we got UC entries: */
577 	if (!num[MTRR_TYPE_UNCACHABLE])
578 		return 0;
579 
580 	/* Check if we only had WB and UC */
581 	if (num[MTRR_TYPE_WRBACK] + num[MTRR_TYPE_UNCACHABLE] !=
582 	    num_var_ranges - num[MTRR_NUM_TYPES])
583 		return 0;
584 
585 	return 1;
586 }
587 
588 static unsigned long __initdata range_sums;
589 
590 static void __init
591 mtrr_calc_range_state(u64 chunk_size, u64 gran_size,
592 		      unsigned long x_remove_base,
593 		      unsigned long x_remove_size, int i)
594 {
595 	/*
596 	 * range_new should really be an automatic variable, but
597 	 * putting 4096 bytes on the stack is frowned upon, to put it
598 	 * mildly. It is safe to make it a static __initdata variable,
599 	 * since mtrr_calc_range_state is only called during init and
600 	 * there's no way it will call itself recursively.
601 	 */
602 	static struct range range_new[RANGE_NUM] __initdata;
603 	unsigned long range_sums_new;
604 	int nr_range_new;
605 	int num_reg;
606 
607 	/* Convert ranges to var ranges state: */
608 	num_reg = x86_setup_var_mtrrs(range, nr_range, chunk_size, gran_size);
609 
610 	/* We got new setting in range_state, check it: */
611 	memset(range_new, 0, sizeof(range_new));
612 	nr_range_new = x86_get_mtrr_mem_range(range_new, 0,
613 				x_remove_base, x_remove_size);
614 	range_sums_new = sum_ranges(range_new, nr_range_new);
615 
616 	result[i].chunk_sizek = chunk_size >> 10;
617 	result[i].gran_sizek = gran_size >> 10;
618 	result[i].num_reg = num_reg;
619 
620 	if (range_sums < range_sums_new) {
621 		result[i].lose_cover_sizek = (range_sums_new - range_sums) << PSHIFT;
622 		result[i].bad = 1;
623 	} else {
624 		result[i].lose_cover_sizek = (range_sums - range_sums_new) << PSHIFT;
625 	}
626 
627 	/* Double check it: */
628 	if (!result[i].bad && !result[i].lose_cover_sizek) {
629 		if (nr_range_new != nr_range || memcmp(range, range_new, sizeof(range)))
630 			result[i].bad = 1;
631 	}
632 
633 	if (!result[i].bad && (range_sums - range_sums_new < min_loss_pfn[num_reg]))
634 		min_loss_pfn[num_reg] = range_sums - range_sums_new;
635 }
636 
637 static void __init mtrr_print_out_one_result(int i)
638 {
639 	unsigned long gran_base, chunk_base, lose_base;
640 	char gran_factor, chunk_factor, lose_factor;
641 
642 	gran_base = to_size_factor(result[i].gran_sizek, &gran_factor);
643 	chunk_base = to_size_factor(result[i].chunk_sizek, &chunk_factor);
644 	lose_base = to_size_factor(result[i].lose_cover_sizek, &lose_factor);
645 
646 	pr_info("%sgran_size: %ld%c \tchunk_size: %ld%c \t",
647 		result[i].bad ? "*BAD*" : " ",
648 		gran_base, gran_factor, chunk_base, chunk_factor);
649 	pr_cont("num_reg: %d  \tlose cover RAM: %s%ld%c\n",
650 		result[i].num_reg, result[i].bad ? "-" : "",
651 		lose_base, lose_factor);
652 }
653 
654 static int __init mtrr_search_optimal_index(void)
655 {
656 	int num_reg_good;
657 	int index_good;
658 	int i;
659 
660 	if (nr_mtrr_spare_reg >= num_var_ranges)
661 		nr_mtrr_spare_reg = num_var_ranges - 1;
662 
663 	num_reg_good = -1;
664 	for (i = num_var_ranges - nr_mtrr_spare_reg; i > 0; i--) {
665 		if (!min_loss_pfn[i])
666 			num_reg_good = i;
667 	}
668 
669 	index_good = -1;
670 	if (num_reg_good != -1) {
671 		for (i = 0; i < NUM_RESULT; i++) {
672 			if (!result[i].bad &&
673 			    result[i].num_reg == num_reg_good &&
674 			    !result[i].lose_cover_sizek) {
675 				index_good = i;
676 				break;
677 			}
678 		}
679 	}
680 
681 	return index_good;
682 }
683 
684 int __init mtrr_cleanup(unsigned address_bits)
685 {
686 	unsigned long x_remove_base, x_remove_size;
687 	unsigned long base, size, def, dummy;
688 	u64 chunk_size, gran_size;
689 	mtrr_type type;
690 	int index_good;
691 	int i;
692 
693 	if (!is_cpu(INTEL) || enable_mtrr_cleanup < 1)
694 		return 0;
695 
696 	rdmsr(MSR_MTRRdefType, def, dummy);
697 	def &= 0xff;
698 	if (def != MTRR_TYPE_UNCACHABLE)
699 		return 0;
700 
701 	/* Get it and store it aside: */
702 	memset(range_state, 0, sizeof(range_state));
703 	for (i = 0; i < num_var_ranges; i++) {
704 		mtrr_if->get(i, &base, &size, &type);
705 		range_state[i].base_pfn = base;
706 		range_state[i].size_pfn = size;
707 		range_state[i].type = type;
708 	}
709 
710 	/* Check if we need handle it and can handle it: */
711 	if (!mtrr_need_cleanup())
712 		return 0;
713 
714 	/* Print original var MTRRs at first, for debugging: */
715 	pr_debug("original variable MTRRs\n");
716 	print_out_mtrr_range_state();
717 
718 	memset(range, 0, sizeof(range));
719 	x_remove_size = 0;
720 	x_remove_base = 1 << (32 - PAGE_SHIFT);
721 	if (mtrr_tom2)
722 		x_remove_size = (mtrr_tom2 >> PAGE_SHIFT) - x_remove_base;
723 
724 	/*
725 	 * [0, 1M) should always be covered by var mtrr with WB
726 	 * and fixed mtrrs should take effect before var mtrr for it:
727 	 */
728 	nr_range = add_range_with_merge(range, RANGE_NUM, 0, 0,
729 					1ULL<<(20 - PAGE_SHIFT));
730 	/* add from var mtrr at last */
731 	nr_range = x86_get_mtrr_mem_range(range, nr_range,
732 					  x_remove_base, x_remove_size);
733 
734 	range_sums = sum_ranges(range, nr_range);
735 	pr_info("total RAM covered: %ldM\n",
736 	       range_sums >> (20 - PAGE_SHIFT));
737 
738 	if (mtrr_chunk_size && mtrr_gran_size) {
739 		i = 0;
740 		mtrr_calc_range_state(mtrr_chunk_size, mtrr_gran_size,
741 				      x_remove_base, x_remove_size, i);
742 
743 		mtrr_print_out_one_result(i);
744 
745 		if (!result[i].bad) {
746 			set_var_mtrr_all(address_bits);
747 			pr_debug("New variable MTRRs\n");
748 			print_out_mtrr_range_state();
749 			return 1;
750 		}
751 		pr_info("invalid mtrr_gran_size or mtrr_chunk_size, will find optimal one\n");
752 	}
753 
754 	i = 0;
755 	memset(min_loss_pfn, 0xff, sizeof(min_loss_pfn));
756 	memset(result, 0, sizeof(result));
757 	for (gran_size = (1ULL<<16); gran_size < (1ULL<<32); gran_size <<= 1) {
758 
759 		for (chunk_size = gran_size; chunk_size < (1ULL<<32);
760 		     chunk_size <<= 1) {
761 
762 			if (i >= NUM_RESULT)
763 				continue;
764 
765 			mtrr_calc_range_state(chunk_size, gran_size,
766 				      x_remove_base, x_remove_size, i);
767 			if (debug_print) {
768 				mtrr_print_out_one_result(i);
769 				pr_info("\n");
770 			}
771 
772 			i++;
773 		}
774 	}
775 
776 	/* Try to find the optimal index: */
777 	index_good = mtrr_search_optimal_index();
778 
779 	if (index_good != -1) {
780 		pr_info("Found optimal setting for mtrr clean up\n");
781 		i = index_good;
782 		mtrr_print_out_one_result(i);
783 
784 		/* Convert ranges to var ranges state: */
785 		chunk_size = result[i].chunk_sizek;
786 		chunk_size <<= 10;
787 		gran_size = result[i].gran_sizek;
788 		gran_size <<= 10;
789 		x86_setup_var_mtrrs(range, nr_range, chunk_size, gran_size);
790 		set_var_mtrr_all(address_bits);
791 		pr_debug("New variable MTRRs\n");
792 		print_out_mtrr_range_state();
793 		return 1;
794 	} else {
795 		/* print out all */
796 		for (i = 0; i < NUM_RESULT; i++)
797 			mtrr_print_out_one_result(i);
798 	}
799 
800 	pr_info("mtrr_cleanup: can not find optimal value\n");
801 	pr_info("please specify mtrr_gran_size/mtrr_chunk_size\n");
802 
803 	return 0;
804 }
805 #else
806 int __init mtrr_cleanup(unsigned address_bits)
807 {
808 	return 0;
809 }
810 #endif
811 
812 static int disable_mtrr_trim;
813 
814 static int __init disable_mtrr_trim_setup(char *str)
815 {
816 	disable_mtrr_trim = 1;
817 	return 0;
818 }
819 early_param("disable_mtrr_trim", disable_mtrr_trim_setup);
820 
821 /*
822  * Newer AMD K8s and later CPUs have a special magic MSR way to force WB
823  * for memory >4GB. Check for that here.
824  * Note this won't check if the MTRRs < 4GB where the magic bit doesn't
825  * apply to are wrong, but so far we don't know of any such case in the wild.
826  */
827 #define Tom2Enabled		(1U << 21)
828 #define Tom2ForceMemTypeWB	(1U << 22)
829 
830 int __init amd_special_default_mtrr(void)
831 {
832 	u32 l, h;
833 
834 	if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD)
835 		return 0;
836 	if (boot_cpu_data.x86 < 0xf)
837 		return 0;
838 	/* In case some hypervisor doesn't pass SYSCFG through: */
839 	if (rdmsr_safe(MSR_K8_SYSCFG, &l, &h) < 0)
840 		return 0;
841 	/*
842 	 * Memory between 4GB and top of mem is forced WB by this magic bit.
843 	 * Reserved before K8RevF, but should be zero there.
844 	 */
845 	if ((l & (Tom2Enabled | Tom2ForceMemTypeWB)) ==
846 		 (Tom2Enabled | Tom2ForceMemTypeWB))
847 		return 1;
848 	return 0;
849 }
850 
851 static u64 __init
852 real_trim_memory(unsigned long start_pfn, unsigned long limit_pfn)
853 {
854 	u64 trim_start, trim_size;
855 
856 	trim_start = start_pfn;
857 	trim_start <<= PAGE_SHIFT;
858 
859 	trim_size = limit_pfn;
860 	trim_size <<= PAGE_SHIFT;
861 	trim_size -= trim_start;
862 
863 	return e820__range_update(trim_start, trim_size, E820_TYPE_RAM, E820_TYPE_RESERVED);
864 }
865 
866 /**
867  * mtrr_trim_uncached_memory - trim RAM not covered by MTRRs
868  * @end_pfn: ending page frame number
869  *
870  * Some buggy BIOSes don't setup the MTRRs properly for systems with certain
871  * memory configurations.  This routine checks that the highest MTRR matches
872  * the end of memory, to make sure the MTRRs having a write back type cover
873  * all of the memory the kernel is intending to use.  If not, it'll trim any
874  * memory off the end by adjusting end_pfn, removing it from the kernel's
875  * allocation pools, warning the user with an obnoxious message.
876  */
877 int __init mtrr_trim_uncached_memory(unsigned long end_pfn)
878 {
879 	unsigned long i, base, size, highest_pfn = 0, def, dummy;
880 	mtrr_type type;
881 	u64 total_trim_size;
882 	/* extra one for all 0 */
883 	int num[MTRR_NUM_TYPES + 1];
884 
885 	/*
886 	 * Make sure we only trim uncachable memory on machines that
887 	 * support the Intel MTRR architecture:
888 	 */
889 	if (!is_cpu(INTEL) || disable_mtrr_trim)
890 		return 0;
891 
892 	rdmsr(MSR_MTRRdefType, def, dummy);
893 	def &= 0xff;
894 	if (def != MTRR_TYPE_UNCACHABLE)
895 		return 0;
896 
897 	/* Get it and store it aside: */
898 	memset(range_state, 0, sizeof(range_state));
899 	for (i = 0; i < num_var_ranges; i++) {
900 		mtrr_if->get(i, &base, &size, &type);
901 		range_state[i].base_pfn = base;
902 		range_state[i].size_pfn = size;
903 		range_state[i].type = type;
904 	}
905 
906 	/* Find highest cached pfn: */
907 	for (i = 0; i < num_var_ranges; i++) {
908 		type = range_state[i].type;
909 		if (type != MTRR_TYPE_WRBACK)
910 			continue;
911 		base = range_state[i].base_pfn;
912 		size = range_state[i].size_pfn;
913 		if (highest_pfn < base + size)
914 			highest_pfn = base + size;
915 	}
916 
917 	/* kvm/qemu doesn't have mtrr set right, don't trim them all: */
918 	if (!highest_pfn) {
919 		pr_info("CPU MTRRs all blank - virtualized system.\n");
920 		return 0;
921 	}
922 
923 	/* Check entries number: */
924 	memset(num, 0, sizeof(num));
925 	for (i = 0; i < num_var_ranges; i++) {
926 		type = range_state[i].type;
927 		if (type >= MTRR_NUM_TYPES)
928 			continue;
929 		size = range_state[i].size_pfn;
930 		if (!size)
931 			type = MTRR_NUM_TYPES;
932 		num[type]++;
933 	}
934 
935 	/* No entry for WB? */
936 	if (!num[MTRR_TYPE_WRBACK])
937 		return 0;
938 
939 	/* Check if we only had WB and UC: */
940 	if (num[MTRR_TYPE_WRBACK] + num[MTRR_TYPE_UNCACHABLE] !=
941 		num_var_ranges - num[MTRR_NUM_TYPES])
942 		return 0;
943 
944 	memset(range, 0, sizeof(range));
945 	nr_range = 0;
946 	if (mtrr_tom2) {
947 		range[nr_range].start = (1ULL<<(32 - PAGE_SHIFT));
948 		range[nr_range].end = mtrr_tom2 >> PAGE_SHIFT;
949 		if (highest_pfn < range[nr_range].end)
950 			highest_pfn = range[nr_range].end;
951 		nr_range++;
952 	}
953 	nr_range = x86_get_mtrr_mem_range(range, nr_range, 0, 0);
954 
955 	/* Check the head: */
956 	total_trim_size = 0;
957 	if (range[0].start)
958 		total_trim_size += real_trim_memory(0, range[0].start);
959 
960 	/* Check the holes: */
961 	for (i = 0; i < nr_range - 1; i++) {
962 		if (range[i].end < range[i+1].start)
963 			total_trim_size += real_trim_memory(range[i].end,
964 							    range[i+1].start);
965 	}
966 
967 	/* Check the top: */
968 	i = nr_range - 1;
969 	if (range[i].end < end_pfn)
970 		total_trim_size += real_trim_memory(range[i].end,
971 							 end_pfn);
972 
973 	if (total_trim_size) {
974 		pr_warn("WARNING: BIOS bug: CPU MTRRs don't cover all of memory, losing %lluMB of RAM.\n",
975 			total_trim_size >> 20);
976 
977 		if (!changed_by_mtrr_cleanup)
978 			WARN_ON(1);
979 
980 		pr_info("update e820 for mtrr\n");
981 		e820__update_table_print();
982 
983 		return 1;
984 	}
985 
986 	return 0;
987 }
988