pat.c 14.7 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13
/*
 * Handle caching attributes in page tables (PAT)
 *
 * Authors: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
 *          Suresh B Siddha <suresh.b.siddha@intel.com>
 *
 * Loosely based on earlier PAT patchset from Eric Biederman and Andi Kleen.
 */

#include <linux/mm.h>
#include <linux/kernel.h>
#include <linux/gfp.h>
#include <linux/fs.h>
14
#include <linux/bootmem.h>
15 16 17 18

#include <asm/msr.h>
#include <asm/tlbflush.h>
#include <asm/processor.h>
19
#include <asm/page.h>
20 21 22 23 24 25
#include <asm/pgtable.h>
#include <asm/pat.h>
#include <asm/e820.h>
#include <asm/cacheflush.h>
#include <asm/fcntl.h>
#include <asm/mtrr.h>
26
#include <asm/io.h>
27

28 29
#ifdef CONFIG_X86_PAT
int __read_mostly pat_wc_enabled = 1;
30

31
void __cpuinit pat_disable(char *reason)
32 33
{
	pat_wc_enabled = 0;
34
	printk(KERN_INFO "%s\n", reason);
35 36
}

37
static int nopat(char *str)
38
{
39
	pat_disable("PAT support disabled.");
40 41
	return 0;
}
42 43 44
early_param("nopat", nopat);
#endif

45 46 47 48 49 50 51 52 53 54 55 56 57

static int debug_enable;
static int __init pat_debug_setup(char *str)
{
	debug_enable = 1;
	return 0;
}
__setup("debugpat", pat_debug_setup);

#define dprintk(fmt, arg...) \
	do { if (debug_enable) printk(KERN_INFO fmt, ##arg); } while (0)


58
static u64 __read_mostly boot_pat_state;
59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74

enum {
	PAT_UC = 0,		/* uncached */
	PAT_WC = 1,		/* Write combining */
	PAT_WT = 4,		/* Write Through */
	PAT_WP = 5,		/* Write Protected */
	PAT_WB = 6,		/* Write Back (default) */
	PAT_UC_MINUS = 7,	/* UC, but can be overriden by MTRR */
};

#define PAT(x,y)	((u64)PAT_ ## y << ((x)*8))

void pat_init(void)
{
	u64 pat;

75
	if (!pat_wc_enabled)
76 77
		return;

78 79 80 81 82 83 84 85 86 87
	/* Paranoia check. */
	if (!cpu_has_pat) {
		printk(KERN_ERR "PAT enabled, but CPU feature cleared\n");
		/*
		 * Panic if this happens on the secondary CPU, and we
		 * switched to PAT on the boot CPU. We have no way to
		 * undo PAT.
		*/
		BUG_ON(boot_pat_state);
	}
88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105

	/* Set PWT to Write-Combining. All other bits stay the same */
	/*
	 * PTE encoding used in Linux:
	 *      PAT
	 *      |PCD
	 *      ||PWT
	 *      |||
	 *      000 WB		_PAGE_CACHE_WB
	 *      001 WC		_PAGE_CACHE_WC
	 *      010 UC-		_PAGE_CACHE_UC_MINUS
	 *      011 UC		_PAGE_CACHE_UC
	 * PAT bit unused
	 */
	pat = PAT(0,WB) | PAT(1,WC) | PAT(2,UC_MINUS) | PAT(3,UC) |
	      PAT(4,WB) | PAT(5,WC) | PAT(6,UC_MINUS) | PAT(7,UC);

	/* Boot CPU check */
106
	if (!boot_pat_state)
107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169
		rdmsrl(MSR_IA32_CR_PAT, boot_pat_state);

	wrmsrl(MSR_IA32_CR_PAT, pat);
	printk(KERN_INFO "x86 PAT enabled: cpu %d, old 0x%Lx, new 0x%Lx\n",
	       smp_processor_id(), boot_pat_state, pat);
}

#undef PAT

static char *cattr_name(unsigned long flags)
{
	switch (flags & _PAGE_CACHE_MASK) {
		case _PAGE_CACHE_UC:		return "uncached";
		case _PAGE_CACHE_UC_MINUS:	return "uncached-minus";
		case _PAGE_CACHE_WB:		return "write-back";
		case _PAGE_CACHE_WC:		return "write-combining";
		default:			return "broken";
	}
}

/*
 * The global memtype list keeps track of memory type for specific
 * physical memory areas. Conflicting memory types in different
 * mappings can cause CPU cache corruption. To avoid this we keep track.
 *
 * The list is sorted based on starting address and can contain multiple
 * entries for each address (this allows reference counting for overlapping
 * areas). All the aliases have the same cache attributes of course.
 * Zero attributes are represented as holes.
 *
 * Currently the data structure is a list because the number of mappings
 * are expected to be relatively small. If this should be a problem
 * it could be changed to a rbtree or similar.
 *
 * memtype_lock protects the whole list.
 */

struct memtype {
	u64 start;
	u64 end;
	unsigned long type;
	struct list_head nd;
};

static LIST_HEAD(memtype_list);
static DEFINE_SPINLOCK(memtype_lock); 	/* protects memtype list */

/*
 * Does intersection of PAT memory type and MTRR memory type and returns
 * the resulting memory type as PAT understands it.
 * (Type in pat and mtrr will not have same value)
 * The intersection is based on "Effective Memory Type" tables in IA-32
 * SDM vol 3a
 */
static int pat_x_mtrr_type(u64 start, u64 end, unsigned long prot,
				unsigned long *ret_prot)
{
	unsigned long pat_type;
	u8 mtrr_type;

	pat_type = prot & _PAGE_CACHE_MASK;
	prot &= (~_PAGE_CACHE_MASK);

170 171 172 173 174 175
	/*
	 * We return the PAT request directly for types where PAT takes
	 * precedence with respect to MTRR and for UC_MINUS.
	 * Consistency checks with other PAT requests is done later
	 * while going through memtype list.
	 */
176 177
	if (pat_type == _PAGE_CACHE_WC) {
		*ret_prot = prot | _PAGE_CACHE_WC;
178
		return 0;
179 180
	} else if (pat_type == _PAGE_CACHE_UC_MINUS) {
		*ret_prot = prot | _PAGE_CACHE_UC_MINUS;
181 182 183 184 185 186 187 188 189 190 191 192 193
		return 0;
	} else if (pat_type == _PAGE_CACHE_UC) {
		*ret_prot = prot | _PAGE_CACHE_UC;
		return 0;
	}

	/*
	 * Look for MTRR hint to get the effective type in case where PAT
	 * request is for WB.
	 */
	mtrr_type = mtrr_type_lookup(start, end);

	if (mtrr_type == MTRR_TYPE_UNCACHABLE) {
194 195 196 197 198 199 200 201 202 203
		*ret_prot = prot | _PAGE_CACHE_UC;
	} else if (mtrr_type == MTRR_TYPE_WRCOMB) {
		*ret_prot = prot | _PAGE_CACHE_WC;
	} else {
		*ret_prot = prot | _PAGE_CACHE_WB;
	}

	return 0;
}

204 205 206 207 208 209 210 211 212 213 214 215 216 217 218
/*
 * req_type typically has one of the:
 * - _PAGE_CACHE_WB
 * - _PAGE_CACHE_WC
 * - _PAGE_CACHE_UC_MINUS
 * - _PAGE_CACHE_UC
 *
 * req_type will have a special case value '-1', when requester want to inherit
 * the memory type from mtrr (if WB), existing PAT, defaulting to UC_MINUS.
 *
 * If ret_type is NULL, function will return an error if it cannot reserve the
 * region with req_type. If ret_type is non-null, function will return
 * available type in ret_type in case of no error. In case of any error
 * it will return a negative return value.
 */
219 220 221 222 223 224 225 226 227 228
int reserve_memtype(u64 start, u64 end, unsigned long req_type,
			unsigned long *ret_type)
{
	struct memtype *new_entry = NULL;
	struct memtype *parse;
	unsigned long actual_type;
	int err = 0;

	/* Only track when pat_wc_enabled */
	if (!pat_wc_enabled) {
229 230 231 232 233 234 235 236
		/* This is identical to page table setting without PAT */
		if (ret_type) {
			if (req_type == -1) {
				*ret_type = _PAGE_CACHE_WB;
			} else {
				*ret_type = req_type;
			}
		}
237 238 239 240 241 242 243 244 245 246 247
		return 0;
	}

	/* Low ISA region is always mapped WB in page table. No need to track */
	if (start >= ISA_START_ADDRESS && (end - 1) <= ISA_END_ADDRESS) {
		if (ret_type)
			*ret_type = _PAGE_CACHE_WB;

		return 0;
	}

248 249
	if (req_type == -1) {
		/*
250 251 252 253
		 * Call mtrr_lookup to get the type hint. This is an
		 * optimization for /dev/mem mmap'ers into WB memory (BIOS
		 * tools and ACPI tools). Use WB request for WB memory and use
		 * UC_MINUS otherwise.
254 255 256 257 258 259 260 261 262 263 264 265 266 267 268
		 */
		u8 mtrr_type = mtrr_type_lookup(start, end);

		if (mtrr_type == MTRR_TYPE_WRBACK) {
			req_type = _PAGE_CACHE_WB;
			actual_type = _PAGE_CACHE_WB;
		} else {
			req_type = _PAGE_CACHE_UC_MINUS;
			actual_type = _PAGE_CACHE_UC_MINUS;
		}
	} else {
		req_type &= _PAGE_CACHE_MASK;
		err = pat_x_mtrr_type(start, end, req_type, &actual_type);
	}

269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293
	if (err) {
		if (ret_type)
			*ret_type = actual_type;

		return -EINVAL;
	}

	new_entry  = kmalloc(sizeof(struct memtype), GFP_KERNEL);
	if (!new_entry)
		return -ENOMEM;

	new_entry->start = start;
	new_entry->end = end;
	new_entry->type = actual_type;

	if (ret_type)
		*ret_type = actual_type;

	spin_lock(&memtype_lock);

	/* Search for existing mapping that overlaps the current range */
	list_for_each_entry(parse, &memtype_list, nd) {
		struct memtype *saved_ptr;

		if (parse->start >= end) {
294
			dprintk("New Entry\n");
295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343
			list_add(&new_entry->nd, parse->nd.prev);
			new_entry = NULL;
			break;
		}

		if (start <= parse->start && end >= parse->start) {
			if (actual_type != parse->type && ret_type) {
				actual_type = parse->type;
				*ret_type = actual_type;
				new_entry->type = actual_type;
			}

			if (actual_type != parse->type) {
				printk(
		KERN_INFO "%s:%d conflicting memory types %Lx-%Lx %s<->%s\n",
					current->comm, current->pid,
					start, end,
					cattr_name(actual_type),
					cattr_name(parse->type));
				err = -EBUSY;
				break;
			}

			saved_ptr = parse;
			/*
			 * Check to see whether the request overlaps more
			 * than one entry in the list
			 */
			list_for_each_entry_continue(parse, &memtype_list, nd) {
				if (end <= parse->start) {
					break;
				}

				if (actual_type != parse->type) {
					printk(
		KERN_INFO "%s:%d conflicting memory types %Lx-%Lx %s<->%s\n",
						current->comm, current->pid,
						start, end,
						cattr_name(actual_type),
						cattr_name(parse->type));
					err = -EBUSY;
					break;
				}
			}

			if (err) {
				break;
			}

344
			dprintk("Overlap at 0x%Lx-0x%Lx\n",
345
			       saved_ptr->start, saved_ptr->end);
346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395
			/* No conflict. Go ahead and add this new entry */
			list_add(&new_entry->nd, saved_ptr->nd.prev);
			new_entry = NULL;
			break;
		}

		if (start < parse->end) {
			if (actual_type != parse->type && ret_type) {
				actual_type = parse->type;
				*ret_type = actual_type;
				new_entry->type = actual_type;
			}

			if (actual_type != parse->type) {
				printk(
		KERN_INFO "%s:%d conflicting memory types %Lx-%Lx %s<->%s\n",
					current->comm, current->pid,
					start, end,
					cattr_name(actual_type),
					cattr_name(parse->type));
				err = -EBUSY;
				break;
			}

			saved_ptr = parse;
			/*
			 * Check to see whether the request overlaps more
			 * than one entry in the list
			 */
			list_for_each_entry_continue(parse, &memtype_list, nd) {
				if (end <= parse->start) {
					break;
				}

				if (actual_type != parse->type) {
					printk(
		KERN_INFO "%s:%d conflicting memory types %Lx-%Lx %s<->%s\n",
						current->comm, current->pid,
						start, end,
						cattr_name(actual_type),
						cattr_name(parse->type));
					err = -EBUSY;
					break;
				}
			}

			if (err) {
				break;
			}

396
			dprintk("Overlap at 0x%Lx-0x%Lx\n",
397
				 saved_ptr->start, saved_ptr->end);
398 399 400 401 402 403 404 405
			/* No conflict. Go ahead and add this new entry */
			list_add(&new_entry->nd, &saved_ptr->nd);
			new_entry = NULL;
			break;
		}
	}

	if (err) {
I
Ingo Molnar 已提交
406
		printk(KERN_INFO
407 408 409
	"reserve_memtype failed 0x%Lx-0x%Lx, track %s, req %s\n",
			start, end, cattr_name(new_entry->type),
			cattr_name(req_type));
410 411 412 413 414 415 416 417
		kfree(new_entry);
		spin_unlock(&memtype_lock);
		return err;
	}

	if (new_entry) {
		/* No conflict. Not yet added to the list. Add to the tail */
		list_add_tail(&new_entry->nd, &memtype_list);
418
		dprintk("New Entry\n");
I
Ingo Molnar 已提交
419
	}
420 421

	if (ret_type) {
422
		dprintk(
423 424 425 426
	"reserve_memtype added 0x%Lx-0x%Lx, track %s, req %s, ret %s\n",
			start, end, cattr_name(actual_type),
			cattr_name(req_type), cattr_name(*ret_type));
	} else {
427
		dprintk(
428 429 430
	"reserve_memtype added 0x%Lx-0x%Lx, track %s, req %s\n",
			start, end, cattr_name(actual_type),
			cattr_name(req_type));
431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463
	}

	spin_unlock(&memtype_lock);
	return err;
}

int free_memtype(u64 start, u64 end)
{
	struct memtype *ml;
	int err = -EINVAL;

	/* Only track when pat_wc_enabled */
	if (!pat_wc_enabled) {
		return 0;
	}

	/* Low ISA region is always mapped WB. No need to track */
	if (start >= ISA_START_ADDRESS && end <= ISA_END_ADDRESS) {
		return 0;
	}

	spin_lock(&memtype_lock);
	list_for_each_entry(ml, &memtype_list, nd) {
		if (ml->start == start && ml->end == end) {
			list_del(&ml->nd);
			kfree(ml);
			err = 0;
			break;
		}
	}
	spin_unlock(&memtype_lock);

	if (err) {
I
Ingo Molnar 已提交
464
		printk(KERN_INFO "%s:%d freeing invalid memtype %Lx-%Lx\n",
465 466
			current->comm, current->pid, start, end);
	}
467

468
	dprintk("free_memtype request 0x%Lx-0x%Lx\n", start, end);
469 470 471
	return err;
}

472

473 474 475 476 477 478 479 480
/*
 * /dev/mem mmap interface. The memtype used for mapping varies:
 * - Use UC for mappings with O_SYNC flag
 * - Without O_SYNC flag, if there is any conflict in reserve_memtype,
 *   inherit the memtype from existing mapping.
 * - Else use UC_MINUS memtype (for backward compatibility with existing
 *   X drivers.
 */
481 482 483 484 485 486
pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
				unsigned long size, pgprot_t vma_prot)
{
	return vma_prot;
}

487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513
#ifdef CONFIG_NONPROMISC_DEVMEM
/* This check is done in drivers/char/mem.c in case of NONPROMISC_DEVMEM*/
static inline int range_is_allowed(unsigned long pfn, unsigned long size)
{
	return 1;
}
#else
static inline int range_is_allowed(unsigned long pfn, unsigned long size)
{
	u64 from = ((u64)pfn) << PAGE_SHIFT;
	u64 to = from + size;
	u64 cursor = from;

	while (cursor < to) {
		if (!devmem_is_allowed(pfn)) {
			printk(KERN_INFO
		"Program %s tried to access /dev/mem between %Lx->%Lx.\n",
				current->comm, from, to);
			return 0;
		}
		cursor += PAGE_SIZE;
		pfn++;
	}
	return 1;
}
#endif /* CONFIG_NONPROMISC_DEVMEM */

514 515 516
int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
				unsigned long size, pgprot_t *vma_prot)
{
517 518 519
	u64 offset = ((u64) pfn) << PAGE_SHIFT;
	unsigned long flags = _PAGE_CACHE_UC_MINUS;
	int retval;
520

521 522 523
	if (!range_is_allowed(pfn, size))
		return 0;

524
	if (file->f_flags & O_SYNC) {
525
		flags = _PAGE_CACHE_UC;
526 527 528 529 530 531 532 533 534 535 536 537
	}

#ifdef CONFIG_X86_32
	/*
	 * On the PPro and successors, the MTRRs are used to set
	 * memory types for physical addresses outside main memory,
	 * so blindly setting UC or PWT on those pages is wrong.
	 * For Pentiums and earlier, the surround logic should disable
	 * caching for the high addresses through the KEN pin, but
	 * we maintain the tradition of paranoia in this code.
	 */
	if (!pat_wc_enabled &&
538 539 540 541
	    ! ( boot_cpu_has(X86_FEATURE_MTRR) ||
		boot_cpu_has(X86_FEATURE_K6_MTRR) ||
		boot_cpu_has(X86_FEATURE_CYRIX_ARR) ||
		boot_cpu_has(X86_FEATURE_CENTAUR_MCR)) &&
542
	   (pfn << PAGE_SHIFT) >= __pa(high_memory)) {
543
		flags = _PAGE_CACHE_UC;
544 545 546
	}
#endif

547 548 549 550 551 552 553 554 555 556
	/*
	 * With O_SYNC, we can only take UC mapping. Fail if we cannot.
	 * Without O_SYNC, we want to get
	 * - WB for WB-able memory and no other conflicting mappings
	 * - UC_MINUS for non-WB-able memory with no other conflicting mappings
	 * - Inherit from confliting mappings otherwise
	 */
	if (flags != _PAGE_CACHE_UC_MINUS) {
		retval = reserve_memtype(offset, offset + size, flags, NULL);
	} else {
I
Ingo Molnar 已提交
557
		retval = reserve_memtype(offset, offset + size, -1, &flags);
558 559 560 561 562 563 564 565
	}

	if (retval < 0)
		return 0;

	if (pfn <= max_pfn_mapped &&
            ioremap_change_attr((unsigned long)__va(offset), size, flags) < 0) {
		free_memtype(offset, offset + size);
I
Ingo Molnar 已提交
566
		printk(KERN_INFO
567 568 569
		"%s:%d /dev/mem ioremap_change_attr failed %s for %Lx-%Lx\n",
			current->comm, current->pid,
			cattr_name(flags),
570
			offset, (unsigned long long)(offset + size));
571 572 573 574 575
		return 0;
	}

	*vma_prot = __pgprot((pgprot_val(*vma_prot) & ~_PAGE_CACHE_MASK) |
			     flags);
576 577
	return 1;
}
578 579 580 581 582 583 584 585 586

void map_devmem(unsigned long pfn, unsigned long size, pgprot_t vma_prot)
{
	u64 addr = (u64)pfn << PAGE_SHIFT;
	unsigned long flags;
	unsigned long want_flags = (pgprot_val(vma_prot) & _PAGE_CACHE_MASK);

	reserve_memtype(addr, addr + size, want_flags, &flags);
	if (flags != want_flags) {
I
Ingo Molnar 已提交
587
		printk(KERN_INFO
588 589 590
		"%s:%d /dev/mem expected mapping type %s for %Lx-%Lx, got %s\n",
			current->comm, current->pid,
			cattr_name(want_flags),
591
			addr, (unsigned long long)(addr + size),
592 593 594 595 596 597 598 599 600 601 602
			cattr_name(flags));
	}
}

void unmap_devmem(unsigned long pfn, unsigned long size, pgprot_t vma_prot)
{
	u64 addr = (u64)pfn << PAGE_SHIFT;

	free_memtype(addr, addr + size);
}