cache-sh4.c 19.8 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4
/*
 * arch/sh/mm/cache-sh4.c
 *
 * Copyright (C) 1999, 2000, 2002  Niibe Yutaka
5
 * Copyright (C) 2001 - 2006  Paul Mundt
L
Linus Torvalds 已提交
6 7 8 9 10 11 12 13
 * Copyright (C) 2003  Richard Curnow
 *
 * This file is subject to the terms and conditions of the GNU General Public
 * License.  See the file "COPYING" in the main directory of this archive
 * for more details.
 */
#include <linux/init.h>
#include <linux/mm.h>
P
Paul Mundt 已提交
14 15
#include <linux/io.h>
#include <linux/mutex.h>
L
Linus Torvalds 已提交
16 17 18
#include <asm/mmu_context.h>
#include <asm/cacheflush.h>

19 20 21 22 23 24 25
/*
 * The maximum number of pages we support up to when doing ranged dcache
 * flushing. Anything exceeding this will simply flush the dcache in its
 * entirety.
 */
#define MAX_DCACHE_PAGES	64	/* XXX: Tune for ways */

26 27 28 29 30 31 32 33
static void __flush_dcache_segment_1way(unsigned long start,
					unsigned long extent);
static void __flush_dcache_segment_2way(unsigned long start,
					unsigned long extent);
static void __flush_dcache_segment_4way(unsigned long start,
					unsigned long extent);

static void __flush_cache_4096(unsigned long addr, unsigned long phys,
34
			       unsigned long exec_offset);
35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56

/*
 * This is initialised here to ensure that it is not placed in the BSS.  If
 * that were to happen, note that cache_init gets called before the BSS is
 * cleared, so this would get nulled out which would be hopeless.
 */
static void (*__flush_dcache_segment_fn)(unsigned long, unsigned long) =
	(void (*)(unsigned long, unsigned long))0xdeadbeef;

static void compute_alias(struct cache_info *c)
{
	c->alias_mask = ((c->sets - 1) << c->entry_shift) & ~(PAGE_SIZE - 1);
	c->n_aliases = (c->alias_mask >> PAGE_SHIFT) + 1;
}

static void __init emit_cache_params(void)
{
	printk("PVR=%08x CVR=%08x PRR=%08x\n",
		ctrl_inl(CCN_PVR),
		ctrl_inl(CCN_CVR),
		ctrl_inl(CCN_PRR));
	printk("I-cache : n_ways=%d n_sets=%d way_incr=%d\n",
57 58 59
		current_cpu_data.icache.ways,
		current_cpu_data.icache.sets,
		current_cpu_data.icache.way_incr);
60
	printk("I-cache : entry_mask=0x%08x alias_mask=0x%08x n_aliases=%d\n",
61 62 63
		current_cpu_data.icache.entry_mask,
		current_cpu_data.icache.alias_mask,
		current_cpu_data.icache.n_aliases);
64
	printk("D-cache : n_ways=%d n_sets=%d way_incr=%d\n",
65 66 67
		current_cpu_data.dcache.ways,
		current_cpu_data.dcache.sets,
		current_cpu_data.dcache.way_incr);
68
	printk("D-cache : entry_mask=0x%08x alias_mask=0x%08x n_aliases=%d\n",
69 70 71
		current_cpu_data.dcache.entry_mask,
		current_cpu_data.dcache.alias_mask,
		current_cpu_data.dcache.n_aliases);
72 73 74 75

	if (!__flush_dcache_segment_fn)
		panic("unknown number of cache ways\n");
}
L
Linus Torvalds 已提交
76 77 78 79 80

/*
 * SH-4 has virtually indexed and physically tagged cache.
 */

81
/* Worst case assumed to be 64k cache, direct-mapped i.e. 4 synonym bits. */
P
Paul Mundt 已提交
82
#define MAX_P3_MUTEXES 16
83

P
Paul Mundt 已提交
84
struct mutex p3map_mutex[MAX_P3_MUTEXES];
L
Linus Torvalds 已提交
85 86 87

void __init p3_cache_init(void)
{
88 89
	int i;

90 91
	compute_alias(&current_cpu_data.icache);
	compute_alias(&current_cpu_data.dcache);
92

93
	switch (current_cpu_data.dcache.ways) {
94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109
	case 1:
		__flush_dcache_segment_fn = __flush_dcache_segment_1way;
		break;
	case 2:
		__flush_dcache_segment_fn = __flush_dcache_segment_2way;
		break;
	case 4:
		__flush_dcache_segment_fn = __flush_dcache_segment_4way;
		break;
	default:
		__flush_dcache_segment_fn = NULL;
		break;
	}

	emit_cache_params();

110
	if (ioremap_page_range(P3SEG, P3SEG + (PAGE_SIZE * 4), 0, PAGE_KERNEL))
L
Linus Torvalds 已提交
111 112
		panic("%s failed.", __FUNCTION__);

113
	for (i = 0; i < current_cpu_data.dcache.n_aliases; i++)
P
Paul Mundt 已提交
114
		mutex_init(&p3map_mutex[i]);
L
Linus Torvalds 已提交
115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187
}

/*
 * Write back the dirty D-caches, but not invalidate them.
 *
 * START: Virtual Address (U0, P1, or P3)
 * SIZE: Size of the region.
 */
void __flush_wback_region(void *start, int size)
{
	unsigned long v;
	unsigned long begin, end;

	begin = (unsigned long)start & ~(L1_CACHE_BYTES-1);
	end = ((unsigned long)start + size + L1_CACHE_BYTES-1)
		& ~(L1_CACHE_BYTES-1);
	for (v = begin; v < end; v+=L1_CACHE_BYTES) {
		asm volatile("ocbwb	%0"
			     : /* no output */
			     : "m" (__m(v)));
	}
}

/*
 * Write back the dirty D-caches and invalidate them.
 *
 * START: Virtual Address (U0, P1, or P3)
 * SIZE: Size of the region.
 */
void __flush_purge_region(void *start, int size)
{
	unsigned long v;
	unsigned long begin, end;

	begin = (unsigned long)start & ~(L1_CACHE_BYTES-1);
	end = ((unsigned long)start + size + L1_CACHE_BYTES-1)
		& ~(L1_CACHE_BYTES-1);
	for (v = begin; v < end; v+=L1_CACHE_BYTES) {
		asm volatile("ocbp	%0"
			     : /* no output */
			     : "m" (__m(v)));
	}
}

/*
 * No write back please
 */
void __flush_invalidate_region(void *start, int size)
{
	unsigned long v;
	unsigned long begin, end;

	begin = (unsigned long)start & ~(L1_CACHE_BYTES-1);
	end = ((unsigned long)start + size + L1_CACHE_BYTES-1)
		& ~(L1_CACHE_BYTES-1);
	for (v = begin; v < end; v+=L1_CACHE_BYTES) {
		asm volatile("ocbi	%0"
			     : /* no output */
			     : "m" (__m(v)));
	}
}

/*
 * Write back the range of D-cache, and purge the I-cache.
 *
 * Called from kernel/module.c:sys_init_module and routine for a.out format.
 */
void flush_icache_range(unsigned long start, unsigned long end)
{
	flush_cache_all();
}

/*
188
 * Write back the D-cache and purge the I-cache for signal trampoline.
L
Linus Torvalds 已提交
189 190 191 192 193 194
 * .. which happens to be the same behavior as flush_icache_range().
 * So, we simply flush out a line.
 */
void flush_cache_sigtramp(unsigned long addr)
{
	unsigned long v, index;
195
	unsigned long flags;
L
Linus Torvalds 已提交
196 197 198 199 200 201 202
	int i;

	v = addr & ~(L1_CACHE_BYTES-1);
	asm volatile("ocbwb	%0"
		     : /* no output */
		     : "m" (__m(v)));

203 204
	index = CACHE_IC_ADDRESS_ARRAY |
			(v & current_cpu_data.icache.entry_mask);
L
Linus Torvalds 已提交
205 206 207

	local_irq_save(flags);
	jump_to_P2();
208

209 210
	for (i = 0; i < current_cpu_data.icache.ways;
	     i++, index += current_cpu_data.icache.way_incr)
L
Linus Torvalds 已提交
211
		ctrl_outl(0, index);	/* Clear out Valid-bit */
212

L
Linus Torvalds 已提交
213
	back_to_P1();
214
	wmb();
L
Linus Torvalds 已提交
215 216 217 218 219 220
	local_irq_restore(flags);
}

static inline void flush_cache_4096(unsigned long start,
				    unsigned long phys)
{
221 222
	unsigned long flags, exec_offset = 0;

L
Linus Torvalds 已提交
223
	/*
224 225
	 * All types of SH-4 require PC to be in P2 to operate on the I-cache.
	 * Some types of SH-4 require PC to be in P2 to operate on the D-cache.
L
Linus Torvalds 已提交
226
	 */
227
	if ((current_cpu_data.flags & CPU_HAS_P2_FLUSH_BUG) ||
228
	    (start < CACHE_OC_ADDRESS_ARRAY))
229
		exec_offset = 0x20000000;
230 231 232 233 234

	local_irq_save(flags);
	__flush_cache_4096(start | SH_CACHE_ASSOC,
			   P1SEGADDR(phys), exec_offset);
	local_irq_restore(flags);
L
Linus Torvalds 已提交
235 236 237 238 239
}

/*
 * Write back & invalidate the D-cache of the page.
 * (To avoid "alias" issues)
240 241 242
 *
 * This uses a lazy write-back on UP, which is explicitly
 * disabled on SMP.
L
Linus Torvalds 已提交
243 244 245
 */
void flush_dcache_page(struct page *page)
{
246 247 248 249 250 251 252 253
#ifndef CONFIG_SMP
	struct address_space *mapping = page_mapping(page);

	if (mapping && !mapping_mapped(mapping))
		set_bit(PG_dcache_dirty, &page->flags);
	else
#endif
	{
L
Linus Torvalds 已提交
254
		unsigned long phys = PHYSADDR(page_address(page));
255 256
		unsigned long addr = CACHE_OC_ADDRESS_ARRAY;
		int i, n;
L
Linus Torvalds 已提交
257 258

		/* Loop all the D-cache */
259
		n = current_cpu_data.dcache.n_aliases;
260
		for (i = 0; i < n; i++, addr += 4096)
261
			flush_cache_4096(addr, phys);
L
Linus Torvalds 已提交
262
	}
263 264

	wmb();
L
Linus Torvalds 已提交
265 266
}

267
/* TODO: Selective icache invalidation through IC address array.. */
L
Linus Torvalds 已提交
268 269 270 271 272 273 274 275 276 277 278 279
static inline void flush_icache_all(void)
{
	unsigned long flags, ccr;

	local_irq_save(flags);
	jump_to_P2();

	/* Flush I-cache */
	ccr = ctrl_inl(CCR);
	ccr |= CCR_CACHE_ICI;
	ctrl_outl(ccr, CCR);

P
Paul Mundt 已提交
280 281 282 283 284
	/*
	 * back_to_P1() will take care of the barrier for us, don't add
	 * another one!
	 */

L
Linus Torvalds 已提交
285 286 287 288
	back_to_P1();
	local_irq_restore(flags);
}

289
void flush_dcache_all(void)
L
Linus Torvalds 已提交
290
{
291
	(*__flush_dcache_segment_fn)(0UL, current_cpu_data.dcache.way_size);
292
	wmb();
293 294 295 296 297
}

void flush_cache_all(void)
{
	flush_dcache_all();
L
Linus Torvalds 已提交
298 299 300
	flush_icache_all();
}

301 302 303 304
static void __flush_cache_mm(struct mm_struct *mm, unsigned long start,
			     unsigned long end)
{
	unsigned long d = 0, p = start & PAGE_MASK;
305 306
	unsigned long alias_mask = current_cpu_data.dcache.alias_mask;
	unsigned long n_aliases = current_cpu_data.dcache.n_aliases;
307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386
	unsigned long select_bit;
	unsigned long all_aliases_mask;
	unsigned long addr_offset;
	pgd_t *dir;
	pmd_t *pmd;
	pud_t *pud;
	pte_t *pte;
	int i;

	dir = pgd_offset(mm, p);
	pud = pud_offset(dir, p);
	pmd = pmd_offset(pud, p);
	end = PAGE_ALIGN(end);

	all_aliases_mask = (1 << n_aliases) - 1;

	do {
		if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) {
			p &= PMD_MASK;
			p += PMD_SIZE;
			pmd++;

			continue;
		}

		pte = pte_offset_kernel(pmd, p);

		do {
			unsigned long phys;
			pte_t entry = *pte;

			if (!(pte_val(entry) & _PAGE_PRESENT)) {
				pte++;
				p += PAGE_SIZE;
				continue;
			}

			phys = pte_val(entry) & PTE_PHYS_MASK;

			if ((p ^ phys) & alias_mask) {
				d |= 1 << ((p & alias_mask) >> PAGE_SHIFT);
				d |= 1 << ((phys & alias_mask) >> PAGE_SHIFT);

				if (d == all_aliases_mask)
					goto loop_exit;
			}

			pte++;
			p += PAGE_SIZE;
		} while (p < end && ((unsigned long)pte & ~PAGE_MASK));
		pmd++;
	} while (p < end);

loop_exit:
	addr_offset = 0;
	select_bit = 1;

	for (i = 0; i < n_aliases; i++) {
		if (d & select_bit) {
			(*__flush_dcache_segment_fn)(addr_offset, PAGE_SIZE);
			wmb();
		}

		select_bit <<= 1;
		addr_offset += PAGE_SIZE;
	}
}

/*
 * Note : (RPC) since the caches are physically tagged, the only point
 * of flush_cache_mm for SH-4 is to get rid of aliases from the
 * D-cache.  The assumption elsewhere, e.g. flush_cache_range, is that
 * lines can stay resident so long as the virtual address they were
 * accessed with (hence cache set) is in accord with the physical
 * address (i.e. tag).  It's no different here.  So I reckon we don't
 * need to flush the I-cache, since aliases don't matter for that.  We
 * should try that.
 *
 * Caller takes mm->mmap_sem.
 */
L
Linus Torvalds 已提交
387 388
void flush_cache_mm(struct mm_struct *mm)
{
389
	/*
390 391 392
	 * If cache is only 4k-per-way, there are never any 'aliases'.  Since
	 * the cache is physically tagged, the data can just be left in there.
	 */
393
	if (current_cpu_data.dcache.n_aliases == 0)
394 395 396 397 398
		return;

	/*
	 * Don't bother groveling around the dcache for the VMA ranges
	 * if there are too many PTEs to make it worthwhile.
399
	 */
400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415
	if (mm->nr_ptes >= MAX_DCACHE_PAGES)
		flush_dcache_all();
	else {
		struct vm_area_struct *vma;

		/*
		 * In this case there are reasonably sized ranges to flush,
		 * iterate through the VMA list and take care of any aliases.
		 */
		for (vma = mm->mmap; vma; vma = vma->vm_next)
			__flush_cache_mm(mm, vma->vm_start, vma->vm_end);
	}

	/* Only touch the icache if one of the VMAs has VM_EXEC set. */
	if (mm->exec_vm)
		flush_icache_all();
L
Linus Torvalds 已提交
416 417 418 419 420 421 422 423
}

/*
 * Write back and invalidate I/D-caches for the page.
 *
 * ADDR: Virtual Address (U0 address)
 * PFN: Physical page number
 */
424 425
void flush_cache_page(struct vm_area_struct *vma, unsigned long address,
		      unsigned long pfn)
L
Linus Torvalds 已提交
426 427
{
	unsigned long phys = pfn << PAGE_SHIFT;
428 429
	unsigned int alias_mask;

430
	alias_mask = current_cpu_data.dcache.alias_mask;
L
Linus Torvalds 已提交
431 432

	/* We only need to flush D-cache when we have alias */
433
	if ((address^phys) & alias_mask) {
L
Linus Torvalds 已提交
434 435
		/* Loop 4K of the D-cache */
		flush_cache_4096(
436
			CACHE_OC_ADDRESS_ARRAY | (address & alias_mask),
L
Linus Torvalds 已提交
437 438 439
			phys);
		/* Loop another 4K of the D-cache */
		flush_cache_4096(
440
			CACHE_OC_ADDRESS_ARRAY | (phys & alias_mask),
L
Linus Torvalds 已提交
441 442 443
			phys);
	}

444
	alias_mask = current_cpu_data.icache.alias_mask;
445 446 447 448 449 450 451 452 453
	if (vma->vm_flags & VM_EXEC) {
		/*
		 * Evict entries from the portion of the cache from which code
		 * may have been executed at this address (virtual).  There's
		 * no need to evict from the portion corresponding to the
		 * physical address as for the D-cache, because we know the
		 * kernel has never executed the code through its identity
		 * translation.
		 */
L
Linus Torvalds 已提交
454
		flush_cache_4096(
455
			CACHE_IC_ADDRESS_ARRAY | (address & alias_mask),
L
Linus Torvalds 已提交
456
			phys);
457
	}
L
Linus Torvalds 已提交
458 459 460 461 462 463 464 465 466 467 468 469 470 471
}

/*
 * Write back and invalidate D-caches.
 *
 * START, END: Virtual Address (U0 address)
 *
 * NOTE: We need to flush the _physical_ page entry.
 * Flushing the cache lines for U0 only isn't enough.
 * We need to flush for P1 too, which may contain aliases.
 */
void flush_cache_range(struct vm_area_struct *vma, unsigned long start,
		       unsigned long end)
{
472 473 474 475
	/*
	 * If cache is only 4k-per-way, there are never any 'aliases'.  Since
	 * the cache is physically tagged, the data can just be left in there.
	 */
476
	if (current_cpu_data.dcache.n_aliases == 0)
477 478
		return;

479 480 481 482 483
	/*
	 * Don't bother with the lookup and alias check if we have a
	 * wide range to cover, just blow away the dcache in its
	 * entirety instead. -- PFM.
	 */
484
	if (((end - start) >> PAGE_SHIFT) >= MAX_DCACHE_PAGES)
485
		flush_dcache_all();
486 487
	else
		__flush_cache_mm(vma->vm_mm, start, end);
488 489 490 491 492 493 494

	if (vma->vm_flags & VM_EXEC) {
		/*
		 * TODO: Is this required???  Need to look at how I-cache
		 * coherency is assured when new programs are loaded to see if
		 * this matters.
		 */
L
Linus Torvalds 已提交
495
		flush_icache_all();
496
	}
L
Linus Torvalds 已提交
497 498 499 500 501 502 503 504 505 506 507 508 509
}

/*
 * flush_icache_user_range
 * @vma: VMA of the process
 * @page: page
 * @addr: U0 address
 * @len: length of the range (< page size)
 */
void flush_icache_user_range(struct vm_area_struct *vma,
			     struct page *page, unsigned long addr, int len)
{
	flush_cache_page(vma, addr, page_to_pfn(page));
510
	mb();
L
Linus Torvalds 已提交
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
/**
 * __flush_cache_4096
 *
 * @addr:  address in memory mapped cache array
 * @phys:  P1 address to flush (has to match tags if addr has 'A' bit
 *         set i.e. associative write)
 * @exec_offset: set to 0x20000000 if flush has to be executed from P2
 *               region else 0x0
 *
 * The offset into the cache array implied by 'addr' selects the
 * 'colour' of the virtual address range that will be flushed.  The
 * operation (purge/write-back) is selected by the lower 2 bits of
 * 'phys'.
 */
static void __flush_cache_4096(unsigned long addr, unsigned long phys,
			       unsigned long exec_offset)
{
	int way_count;
	unsigned long base_addr = addr;
	struct cache_info *dcache;
	unsigned long way_incr;
	unsigned long a, ea, p;
	unsigned long temp_pc;

537
	dcache = &current_cpu_data.dcache;
538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611
	/* Write this way for better assembly. */
	way_count = dcache->ways;
	way_incr = dcache->way_incr;

	/*
	 * Apply exec_offset (i.e. branch to P2 if required.).
	 *
	 * FIXME:
	 *
	 *	If I write "=r" for the (temp_pc), it puts this in r6 hence
	 *	trashing exec_offset before it's been added on - why?  Hence
	 *	"=&r" as a 'workaround'
	 */
	asm volatile("mov.l 1f, %0\n\t"
		     "add   %1, %0\n\t"
		     "jmp   @%0\n\t"
		     "nop\n\t"
		     ".balign 4\n\t"
		     "1:  .long 2f\n\t"
		     "2:\n" : "=&r" (temp_pc) : "r" (exec_offset));

	/*
	 * We know there will be >=1 iteration, so write as do-while to avoid
	 * pointless nead-of-loop check for 0 iterations.
	 */
	do {
		ea = base_addr + PAGE_SIZE;
		a = base_addr;
		p = phys;

		do {
			*(volatile unsigned long *)a = p;
			/*
			 * Next line: intentionally not p+32, saves an add, p
			 * will do since only the cache tag bits need to
			 * match.
			 */
			*(volatile unsigned long *)(a+32) = p;
			a += 64;
			p += 64;
		} while (a < ea);

		base_addr += way_incr;
	} while (--way_count != 0);
}

/*
 * Break the 1, 2 and 4 way variants of this out into separate functions to
 * avoid nearly all the overhead of having the conditional stuff in the function
 * bodies (+ the 1 and 2 way cases avoid saving any registers too).
 */
static void __flush_dcache_segment_1way(unsigned long start,
					unsigned long extent_per_way)
{
	unsigned long orig_sr, sr_with_bl;
	unsigned long base_addr;
	unsigned long way_incr, linesz, way_size;
	struct cache_info *dcache;
	register unsigned long a0, a0e;

	asm volatile("stc sr, %0" : "=r" (orig_sr));
	sr_with_bl = orig_sr | (1<<28);
	base_addr = ((unsigned long)&empty_zero_page[0]);

	/*
	 * The previous code aligned base_addr to 16k, i.e. the way_size of all
	 * existing SH-4 D-caches.  Whilst I don't see a need to have this
	 * aligned to any better than the cache line size (which it will be
	 * anyway by construction), let's align it to at least the way_size of
	 * any existing or conceivable SH-4 D-cache.  -- RPC
	 */
	base_addr = ((base_addr >> 16) << 16);
	base_addr |= start;

612
	dcache = &current_cpu_data.dcache;
613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653
	linesz = dcache->linesz;
	way_incr = dcache->way_incr;
	way_size = dcache->way_size;

	a0 = base_addr;
	a0e = base_addr + extent_per_way;
	do {
		asm volatile("ldc %0, sr" : : "r" (sr_with_bl));
		asm volatile("movca.l r0, @%0\n\t"
			     "ocbi @%0" : : "r" (a0));
		a0 += linesz;
		asm volatile("movca.l r0, @%0\n\t"
			     "ocbi @%0" : : "r" (a0));
		a0 += linesz;
		asm volatile("movca.l r0, @%0\n\t"
			     "ocbi @%0" : : "r" (a0));
		a0 += linesz;
		asm volatile("movca.l r0, @%0\n\t"
			     "ocbi @%0" : : "r" (a0));
		asm volatile("ldc %0, sr" : : "r" (orig_sr));
		a0 += linesz;
	} while (a0 < a0e);
}

static void __flush_dcache_segment_2way(unsigned long start,
					unsigned long extent_per_way)
{
	unsigned long orig_sr, sr_with_bl;
	unsigned long base_addr;
	unsigned long way_incr, linesz, way_size;
	struct cache_info *dcache;
	register unsigned long a0, a1, a0e;

	asm volatile("stc sr, %0" : "=r" (orig_sr));
	sr_with_bl = orig_sr | (1<<28);
	base_addr = ((unsigned long)&empty_zero_page[0]);

	/* See comment under 1-way above */
	base_addr = ((base_addr >> 16) << 16);
	base_addr |= start;

654
	dcache = &current_cpu_data.dcache;
655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712
	linesz = dcache->linesz;
	way_incr = dcache->way_incr;
	way_size = dcache->way_size;

	a0 = base_addr;
	a1 = a0 + way_incr;
	a0e = base_addr + extent_per_way;
	do {
		asm volatile("ldc %0, sr" : : "r" (sr_with_bl));
		asm volatile("movca.l r0, @%0\n\t"
			     "movca.l r0, @%1\n\t"
			     "ocbi @%0\n\t"
			     "ocbi @%1" : :
			     "r" (a0), "r" (a1));
		a0 += linesz;
		a1 += linesz;
		asm volatile("movca.l r0, @%0\n\t"
			     "movca.l r0, @%1\n\t"
			     "ocbi @%0\n\t"
			     "ocbi @%1" : :
			     "r" (a0), "r" (a1));
		a0 += linesz;
		a1 += linesz;
		asm volatile("movca.l r0, @%0\n\t"
			     "movca.l r0, @%1\n\t"
			     "ocbi @%0\n\t"
			     "ocbi @%1" : :
			     "r" (a0), "r" (a1));
		a0 += linesz;
		a1 += linesz;
		asm volatile("movca.l r0, @%0\n\t"
			     "movca.l r0, @%1\n\t"
			     "ocbi @%0\n\t"
			     "ocbi @%1" : :
			     "r" (a0), "r" (a1));
		asm volatile("ldc %0, sr" : : "r" (orig_sr));
		a0 += linesz;
		a1 += linesz;
	} while (a0 < a0e);
}

static void __flush_dcache_segment_4way(unsigned long start,
					unsigned long extent_per_way)
{
	unsigned long orig_sr, sr_with_bl;
	unsigned long base_addr;
	unsigned long way_incr, linesz, way_size;
	struct cache_info *dcache;
	register unsigned long a0, a1, a2, a3, a0e;

	asm volatile("stc sr, %0" : "=r" (orig_sr));
	sr_with_bl = orig_sr | (1<<28);
	base_addr = ((unsigned long)&empty_zero_page[0]);

	/* See comment under 1-way above */
	base_addr = ((base_addr >> 16) << 16);
	base_addr |= start;

713
	dcache = &current_cpu_data.dcache;
714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779
	linesz = dcache->linesz;
	way_incr = dcache->way_incr;
	way_size = dcache->way_size;

	a0 = base_addr;
	a1 = a0 + way_incr;
	a2 = a1 + way_incr;
	a3 = a2 + way_incr;
	a0e = base_addr + extent_per_way;
	do {
		asm volatile("ldc %0, sr" : : "r" (sr_with_bl));
		asm volatile("movca.l r0, @%0\n\t"
			     "movca.l r0, @%1\n\t"
			     "movca.l r0, @%2\n\t"
			     "movca.l r0, @%3\n\t"
			     "ocbi @%0\n\t"
			     "ocbi @%1\n\t"
			     "ocbi @%2\n\t"
			     "ocbi @%3\n\t" : :
			     "r" (a0), "r" (a1), "r" (a2), "r" (a3));
		a0 += linesz;
		a1 += linesz;
		a2 += linesz;
		a3 += linesz;
		asm volatile("movca.l r0, @%0\n\t"
			     "movca.l r0, @%1\n\t"
			     "movca.l r0, @%2\n\t"
			     "movca.l r0, @%3\n\t"
			     "ocbi @%0\n\t"
			     "ocbi @%1\n\t"
			     "ocbi @%2\n\t"
			     "ocbi @%3\n\t" : :
			     "r" (a0), "r" (a1), "r" (a2), "r" (a3));
		a0 += linesz;
		a1 += linesz;
		a2 += linesz;
		a3 += linesz;
		asm volatile("movca.l r0, @%0\n\t"
			     "movca.l r0, @%1\n\t"
			     "movca.l r0, @%2\n\t"
			     "movca.l r0, @%3\n\t"
			     "ocbi @%0\n\t"
			     "ocbi @%1\n\t"
			     "ocbi @%2\n\t"
			     "ocbi @%3\n\t" : :
			     "r" (a0), "r" (a1), "r" (a2), "r" (a3));
		a0 += linesz;
		a1 += linesz;
		a2 += linesz;
		a3 += linesz;
		asm volatile("movca.l r0, @%0\n\t"
			     "movca.l r0, @%1\n\t"
			     "movca.l r0, @%2\n\t"
			     "movca.l r0, @%3\n\t"
			     "ocbi @%0\n\t"
			     "ocbi @%1\n\t"
			     "ocbi @%2\n\t"
			     "ocbi @%3\n\t" : :
			     "r" (a0), "r" (a1), "r" (a2), "r" (a3));
		asm volatile("ldc %0, sr" : : "r" (orig_sr));
		a0 += linesz;
		a1 += linesz;
		a2 += linesz;
		a3 += linesz;
	} while (a0 < a0e);
}