lmb.c 11.7 KB
Newer Older
1 2 3 4 5
/*
 * Procedures for maintaining information about logical memory blocks.
 *
 * Peter Bergner, IBM Corp.	June 2001.
 * Copyright (C) 2001 Peter Bergner.
6
 *
7 8 9 10 11 12 13 14 15
 *      This program is free software; you can redistribute it and/or
 *      modify it under the terms of the GNU General Public License
 *      as published by the Free Software Foundation; either version
 *      2 of the License, or (at your option) any later version.
 */

#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/bitops.h>
16
#include <linux/lmb.h>
17

18 19
#define LMB_ALLOC_ANYWHERE	0

20 21
struct lmb lmb;

22 23 24 25 26 27 28 29 30 31
static int lmb_debug;

static int __init early_lmb(char *p)
{
	if (p && strstr(p, "debug"))
		lmb_debug = 1;
	return 0;
}
early_param("lmb", early_lmb);

32
static void lmb_dump(struct lmb_region *region, char *name)
33
{
34 35 36 37 38 39 40 41 42 43 44 45 46
	unsigned long long base, size;
	int i;

	pr_info(" %s.cnt  = 0x%lx\n", name, region->cnt);

	for (i = 0; i < region->cnt; i++) {
		base = region->region[i].base;
		size = region->region[i].size;

		pr_info(" %s[0x%x]\t0x%016llx - 0x%016llx, 0x%llx bytes\n",
		    name, i, base, base + size - 1, size);
	}
}
47

48 49
void lmb_dump_all(void)
{
50 51 52
	if (!lmb_debug)
		return;

53 54 55
	pr_info("LMB configuration:\n");
	pr_info(" rmo_size    = 0x%llx\n", (unsigned long long)lmb.rmo_size);
	pr_info(" memory.size = 0x%llx\n", (unsigned long long)lmb.memory.size);
56

57 58
	lmb_dump(&lmb.memory, "memory");
	lmb_dump(&lmb.reserved, "reserved");
59 60
}

61 62
static unsigned long lmb_addrs_overlap(u64 base1, u64 size1, u64 base2,
					u64 size2)
63
{
64
	return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
65 66
}

67
static long lmb_addrs_adjacent(u64 base1, u64 size1, u64 base2, u64 size2)
68 69 70 71 72 73 74 75 76
{
	if (base2 == base1 + size1)
		return 1;
	else if (base1 == base2 + size2)
		return -1;

	return 0;
}

77
static long lmb_regions_adjacent(struct lmb_region *rgn,
78 79
		unsigned long r1, unsigned long r2)
{
80 81 82 83
	u64 base1 = rgn->region[r1].base;
	u64 size1 = rgn->region[r1].size;
	u64 base2 = rgn->region[r2].base;
	u64 size2 = rgn->region[r2].size;
84 85 86 87

	return lmb_addrs_adjacent(base1, size1, base2, size2);
}

88
static void lmb_remove_region(struct lmb_region *rgn, unsigned long r)
89 90 91
{
	unsigned long i;

92 93 94
	for (i = r; i < rgn->cnt - 1; i++) {
		rgn->region[i].base = rgn->region[i + 1].base;
		rgn->region[i].size = rgn->region[i + 1].size;
95 96 97 98
	}
	rgn->cnt--;
}

99
/* Assumption: base addr of region 1 < base addr of region 2 */
100
static void lmb_coalesce_regions(struct lmb_region *rgn,
101 102 103 104 105 106
		unsigned long r1, unsigned long r2)
{
	rgn->region[r1].size += rgn->region[r2].size;
	lmb_remove_region(rgn, r2);
}

107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131
void __init lmb_init(void)
{
	/* Create a dummy zero size LMB which will get coalesced away later.
	 * This simplifies the lmb_add() code below...
	 */
	lmb.memory.region[0].base = 0;
	lmb.memory.region[0].size = 0;
	lmb.memory.cnt = 1;

	/* Ditto. */
	lmb.reserved.region[0].base = 0;
	lmb.reserved.region[0].size = 0;
	lmb.reserved.cnt = 1;
}

void __init lmb_analyze(void)
{
	int i;

	lmb.memory.size = 0;

	for (i = 0; i < lmb.memory.cnt; i++)
		lmb.memory.size += lmb.memory.region[i].size;
}

132
static long lmb_add_region(struct lmb_region *rgn, u64 base, u64 size)
133
{
134 135
	unsigned long coalesced = 0;
	long adjacent, i;
136

137 138 139 140 141 142
	if ((rgn->cnt == 1) && (rgn->region[0].size == 0)) {
		rgn->region[0].base = base;
		rgn->region[0].size = size;
		return 0;
	}

143
	/* First try and coalesce this LMB with another. */
144
	for (i = 0; i < rgn->cnt; i++) {
145 146
		u64 rgnbase = rgn->region[i].base;
		u64 rgnsize = rgn->region[i].size;
147

148 149 150 151
		if ((rgnbase == base) && (rgnsize == size))
			/* Already have this region, so we're done */
			return 0;

152 153
		adjacent = lmb_addrs_adjacent(base, size, rgnbase, rgnsize);
		if (adjacent > 0) {
154 155 156 157
			rgn->region[i].base -= size;
			rgn->region[i].size += size;
			coalesced++;
			break;
158
		} else if (adjacent < 0) {
159 160 161 162 163 164
			rgn->region[i].size += size;
			coalesced++;
			break;
		}
	}

165
	if ((i < rgn->cnt - 1) && lmb_regions_adjacent(rgn, i, i+1)) {
166 167 168 169 170 171 172 173 174 175
		lmb_coalesce_regions(rgn, i, i+1);
		coalesced++;
	}

	if (coalesced)
		return coalesced;
	if (rgn->cnt >= MAX_LMB_REGIONS)
		return -1;

	/* Couldn't coalesce the LMB, so add it to the sorted table. */
176
	for (i = rgn->cnt - 1; i >= 0; i--) {
177 178 179 180 181 182 183 184 185
		if (base < rgn->region[i].base) {
			rgn->region[i+1].base = rgn->region[i].base;
			rgn->region[i+1].size = rgn->region[i].size;
		} else {
			rgn->region[i+1].base = base;
			rgn->region[i+1].size = size;
			break;
		}
	}
186 187 188 189 190

	if (base < rgn->region[0].base) {
		rgn->region[0].base = base;
		rgn->region[0].size = size;
	}
191 192 193 194 195
	rgn->cnt++;

	return 0;
}

196
long lmb_add(u64 base, u64 size)
197
{
198
	struct lmb_region *_rgn = &lmb.memory;
199 200 201 202 203 204 205 206 207

	/* On pSeries LPAR systems, the first LMB is our RMO region. */
	if (base == 0)
		lmb.rmo_size = size;

	return lmb_add_region(_rgn, base, size);

}

M
Michael Ellerman 已提交
208
static long __lmb_remove(struct lmb_region *rgn, u64 base, u64 size)
209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255
{
	u64 rgnbegin, rgnend;
	u64 end = base + size;
	int i;

	rgnbegin = rgnend = 0; /* supress gcc warnings */

	/* Find the region where (base, size) belongs to */
	for (i=0; i < rgn->cnt; i++) {
		rgnbegin = rgn->region[i].base;
		rgnend = rgnbegin + rgn->region[i].size;

		if ((rgnbegin <= base) && (end <= rgnend))
			break;
	}

	/* Didn't find the region */
	if (i == rgn->cnt)
		return -1;

	/* Check to see if we are removing entire region */
	if ((rgnbegin == base) && (rgnend == end)) {
		lmb_remove_region(rgn, i);
		return 0;
	}

	/* Check to see if region is matching at the front */
	if (rgnbegin == base) {
		rgn->region[i].base = end;
		rgn->region[i].size -= size;
		return 0;
	}

	/* Check to see if the region is matching at the end */
	if (rgnend == end) {
		rgn->region[i].size -= size;
		return 0;
	}

	/*
	 * We need to split the entry -  adjust the current one to the
	 * beginging of the hole and add the region after hole.
	 */
	rgn->region[i].size = base - rgn->region[i].base;
	return lmb_add_region(rgn, end, rgnend - end);
}

M
Michael Ellerman 已提交
256 257 258 259 260 261 262 263 264 265
long lmb_remove(u64 base, u64 size)
{
	return __lmb_remove(&lmb.memory, base, size);
}

long __init lmb_free(u64 base, u64 size)
{
	return __lmb_remove(&lmb.reserved, base, size);
}

266
long __init lmb_reserve(u64 base, u64 size)
267
{
268
	struct lmb_region *_rgn = &lmb.reserved;
269

270 271
	BUG_ON(0 == size);

272 273 274
	return lmb_add_region(_rgn, base, size);
}

275
long lmb_overlaps_region(struct lmb_region *rgn, u64 base, u64 size)
276 277 278
{
	unsigned long i;

279
	for (i = 0; i < rgn->cnt; i++) {
280 281
		u64 rgnbase = rgn->region[i].base;
		u64 rgnsize = rgn->region[i].size;
282
		if (lmb_addrs_overlap(base, size, rgnbase, rgnsize))
283 284 285 286 287 288
			break;
	}

	return (i < rgn->cnt) ? i : -1;
}

D
David S. Miller 已提交
289 290 291 292 293 294 295 296 297 298 299 300 301
static u64 lmb_align_down(u64 addr, u64 size)
{
	return addr & ~(size - 1);
}

static u64 lmb_align_up(u64 addr, u64 size)
{
	return (addr + (size - 1)) & ~(size - 1);
}

static u64 __init lmb_alloc_nid_unreserved(u64 start, u64 end,
					   u64 size, u64 align)
{
302
	u64 base, res_base;
D
David S. Miller 已提交
303 304 305
	long j;

	base = lmb_align_down((end - size), align);
306 307 308 309
	while (start <= base) {
		j = lmb_overlaps_region(&lmb.reserved, base, size);
		if (j < 0) {
			/* this area isn't reserved, take it */
310
			if (lmb_add_region(&lmb.reserved, base, size) < 0)
311 312 313 314 315 316 317
				base = ~(u64)0;
			return base;
		}
		res_base = lmb.reserved.region[j].base;
		if (res_base < size)
			break;
		base = lmb_align_down(res_base - size, align);
D
David S. Miller 已提交
318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355
	}

	return ~(u64)0;
}

static u64 __init lmb_alloc_nid_region(struct lmb_property *mp,
				       u64 (*nid_range)(u64, u64, int *),
				       u64 size, u64 align, int nid)
{
	u64 start, end;

	start = mp->base;
	end = start + mp->size;

	start = lmb_align_up(start, align);
	while (start < end) {
		u64 this_end;
		int this_nid;

		this_end = nid_range(start, end, &this_nid);
		if (this_nid == nid) {
			u64 ret = lmb_alloc_nid_unreserved(start, this_end,
							   size, align);
			if (ret != ~(u64)0)
				return ret;
		}
		start = this_end;
	}

	return ~(u64)0;
}

u64 __init lmb_alloc_nid(u64 size, u64 align, int nid,
			 u64 (*nid_range)(u64 start, u64 end, int *nid))
{
	struct lmb_region *mem = &lmb.memory;
	int i;

356 357 358 359
	BUG_ON(0 == size);

	size = lmb_align_up(size, align);

D
David S. Miller 已提交
360 361 362 363 364 365 366 367 368 369 370
	for (i = 0; i < mem->cnt; i++) {
		u64 ret = lmb_alloc_nid_region(&mem->region[i],
					       nid_range,
					       size, align, nid);
		if (ret != ~(u64)0)
			return ret;
	}

	return lmb_alloc(size, align);
}

371
u64 __init lmb_alloc(u64 size, u64 align)
372 373 374 375
{
	return lmb_alloc_base(size, align, LMB_ALLOC_ANYWHERE);
}

376
u64 __init lmb_alloc_base(u64 size, u64 align, u64 max_addr)
377
{
378
	u64 alloc;
379 380 381

	alloc = __lmb_alloc_base(size, align, max_addr);

382
	if (alloc == 0)
383 384
		panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
		      (unsigned long long) size, (unsigned long long) max_addr);
385 386 387 388

	return alloc;
}

389
u64 __init __lmb_alloc_base(u64 size, u64 align, u64 max_addr)
390 391
{
	long i, j;
392
	u64 base = 0;
393
	u64 res_base;
394

395 396
	BUG_ON(0 == size);

397 398
	size = lmb_align_up(size, align);

399
	/* On some platforms, make sure we allocate lowmem */
400
	/* Note that LMB_REAL_LIMIT may be LMB_ALLOC_ANYWHERE */
401
	if (max_addr == LMB_ALLOC_ANYWHERE)
402 403
		max_addr = LMB_REAL_LIMIT;

404
	for (i = lmb.memory.cnt - 1; i >= 0; i--) {
405 406
		u64 lmbbase = lmb.memory.region[i].base;
		u64 lmbsize = lmb.memory.region[i].size;
407

408 409
		if (lmbsize < size)
			continue;
410
		if (max_addr == LMB_ALLOC_ANYWHERE)
411
			base = lmb_align_down(lmbbase + lmbsize - size, align);
412 413
		else if (lmbbase < max_addr) {
			base = min(lmbbase + lmbsize, max_addr);
414
			base = lmb_align_down(base - size, align);
415 416 417
		} else
			continue;

418
		while (base && lmbbase <= base) {
419
			j = lmb_overlaps_region(&lmb.reserved, base, size);
420 421
			if (j < 0) {
				/* this area isn't reserved, take it */
422
				if (lmb_add_region(&lmb.reserved, base, size) < 0)
423 424 425 426 427
					return 0;
				return base;
			}
			res_base = lmb.reserved.region[j].base;
			if (res_base < size)
428
				break;
429
			base = lmb_align_down(res_base - size, align);
430
		}
431
	}
432
	return 0;
433 434 435
}

/* You must call lmb_analyze() before this. */
436
u64 __init lmb_phys_mem_size(void)
437 438 439 440
{
	return lmb.memory.size;
}

441
u64 lmb_end_of_DRAM(void)
442 443 444 445 446 447
{
	int idx = lmb.memory.cnt - 1;

	return (lmb.memory.region[idx].base + lmb.memory.region[idx].size);
}

448
/* You must call lmb_analyze() after this. */
449
void __init lmb_enforce_memory_limit(u64 memory_limit)
450
{
451 452
	unsigned long i;
	u64 limit;
453
	struct lmb_property *p;
454

455
	if (!memory_limit)
456 457
		return;

458
	/* Truncate the lmb regions to satisfy the memory limit. */
459 460 461 462 463 464 465 466 467 468 469
	limit = memory_limit;
	for (i = 0; i < lmb.memory.cnt; i++) {
		if (limit > lmb.memory.region[i].size) {
			limit -= lmb.memory.region[i].size;
			continue;
		}

		lmb.memory.region[i].size = limit;
		lmb.memory.cnt = i + 1;
		break;
	}
470

471 472
	if (lmb.memory.region[0].size < lmb.rmo_size)
		lmb.rmo_size = lmb.memory.region[0].size;
473

474 475
	memory_limit = lmb_end_of_DRAM();

476 477 478 479 480 481 482 483 484 485 486 487 488 489
	/* And truncate any reserves above the limit also. */
	for (i = 0; i < lmb.reserved.cnt; i++) {
		p = &lmb.reserved.region[i];

		if (p->base > memory_limit)
			p->size = 0;
		else if ((p->base + p->size) > memory_limit)
			p->size = memory_limit - p->base;

		if (p->size == 0) {
			lmb_remove_region(&lmb.reserved, i);
			i--;
		}
	}
490
}
491

492
int __init lmb_is_reserved(u64 addr)
493 494 495 496
{
	int i;

	for (i = 0; i < lmb.reserved.cnt; i++) {
497 498
		u64 upper = lmb.reserved.region[i].base +
			lmb.reserved.region[i].size - 1;
499 500 501 502 503
		if ((addr >= lmb.reserved.region[i].base) && (addr <= upper))
			return 1;
	}
	return 0;
}
504

505 506 507 508 509
int lmb_is_region_reserved(u64 base, u64 size)
{
	return lmb_overlaps_region(&lmb.reserved, base, size);
}

510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541
/*
 * Given a <base, len>, find which memory regions belong to this range.
 * Adjust the request and return a contiguous chunk.
 */
int lmb_find(struct lmb_property *res)
{
	int i;
	u64 rstart, rend;

	rstart = res->base;
	rend = rstart + res->size - 1;

	for (i = 0; i < lmb.memory.cnt; i++) {
		u64 start = lmb.memory.region[i].base;
		u64 end = start + lmb.memory.region[i].size - 1;

		if (start > rend)
			return -1;

		if ((end >= rstart) && (start < rend)) {
			/* adjust the request */
			if (rstart < start)
				rstart = start;
			if (rend > end)
				rend = end;
			res->base = rstart;
			res->size = rend - rstart + 1;
			return 0;
		}
	}
	return -1;
}