[deliverable/linux.git] / arch / sh / mm / cache-sh4.c

/*
 * arch/sh/mm/cache-sh4.c
 *
 * Copyright (C) 1999, 2000, 2002  Niibe Yutaka
 * Copyright (C) 2001, 2002, 2003, 2004, 2005  Paul Mundt
 * 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/mman.h>
#include <linux/mm.h>
#include <linux/threads.h>
#include <asm/addrspace.h>
#include <asm/page.h>
#include <asm/pgtable.h>
#include <asm/processor.h>
#include <asm/cache.h>
#include <asm/io.h>
#include <asm/uaccess.h>
#include <asm/pgalloc.h>
#include <asm/mmu_context.h>
#include <asm/cacheflush.h>

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,
			       unsigned long exec_offset);

/*
 * 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",
		cpu_data->icache.ways,
		cpu_data->icache.sets,
		cpu_data->icache.way_incr);
	printk("I-cache : entry_mask=0x%08x alias_mask=0x%08x n_aliases=%d\n",
		cpu_data->icache.entry_mask,
		cpu_data->icache.alias_mask,
		cpu_data->icache.n_aliases);
	printk("D-cache : n_ways=%d n_sets=%d way_incr=%d\n",
		cpu_data->dcache.ways,
		cpu_data->dcache.sets,
		cpu_data->dcache.way_incr);
	printk("D-cache : entry_mask=0x%08x alias_mask=0x%08x n_aliases=%d\n",
		cpu_data->dcache.entry_mask,
		cpu_data->dcache.alias_mask,
		cpu_data->dcache.n_aliases);

	if (!__flush_dcache_segment_fn)
		panic("unknown number of cache ways\n");
}

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

/* Worst case assumed to be 64k cache, direct-mapped i.e. 4 synonym bits. */
#define MAX_P3_SEMAPHORES 16

struct semaphore p3map_sem[MAX_P3_SEMAPHORES];

void __init p3_cache_init(void)
{
	int i;

	compute_alias(&cpu_data->icache);
	compute_alias(&cpu_data->dcache);

	switch (cpu_data->dcache.ways) {
	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();

	if (remap_area_pages(P3SEG, 0, PAGE_SIZE * 4, _PAGE_CACHABLE))
		panic("%s failed.", __FUNCTION__);

	for (i = 0; i < cpu_data->dcache.n_aliases; i++)
		sema_init(&p3map_sem[i], 1);
}

/*
 * 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();
}

/*
 * Write back the D-cache and purge the I-cache for signal trampoline.
 * .. 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;
	unsigned long flags;
	int i;

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

	index = CACHE_IC_ADDRESS_ARRAY | (v & cpu_data->icache.entry_mask);

	local_irq_save(flags);
	jump_to_P2();

	for (i = 0; i < cpu_data->icache.ways;
	     i++, index += cpu_data->icache.way_incr)
		ctrl_outl(0, index);	/* Clear out Valid-bit */

	back_to_P1();
	wmb();
	local_irq_restore(flags);
}

static inline void flush_cache_4096(unsigned long start,
				    unsigned long phys)
{
	unsigned long flags;

	/*
	 * 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.
	 */
	if ((cpu_data->flags & CPU_HAS_P2_FLUSH_BUG)
	   || start < CACHE_OC_ADDRESS_ARRAY) {
		local_irq_save(flags);
		__flush_cache_4096(start | SH_CACHE_ASSOC,
				   P1SEGADDR(phys), 0x20000000);
		local_irq_restore(flags);
	} else {
		__flush_cache_4096(start | SH_CACHE_ASSOC,
				   P1SEGADDR(phys), 0);
	}
}

/*
 * Write back & invalidate the D-cache of the page.
 * (To avoid "alias" issues)
 */
void flush_dcache_page(struct page *page)
{
	if (test_bit(PG_mapped, &page->flags)) {
		unsigned long phys = PHYSADDR(page_address(page));
		unsigned long addr = CACHE_OC_ADDRESS_ARRAY;
		int i, n;

		/* Loop all the D-cache */
		n = cpu_data->dcache.n_aliases;
		for (i = 0; i < n; i++, addr += PAGE_SIZE)
			flush_cache_4096(addr, phys);
	}

	wmb();
}

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);

	/*
	 * back_to_P1() will take care of the barrier for us, don't add
	 * another one!
	 */

	back_to_P1();
	local_irq_restore(flags);
}

void flush_dcache_all(void)
{
	(*__flush_dcache_segment_fn)(0UL, cpu_data->dcache.way_size);
	wmb();
}

void flush_cache_all(void)
{
	flush_dcache_all();
	flush_icache_all();
}

void flush_cache_mm(struct mm_struct *mm)
{
	/*
	 * 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.
	 */
	flush_cache_all();
}

/*
 * Write back and invalidate I/D-caches for the page.
 *
 * ADDR: Virtual Address (U0 address)
 * PFN: Physical page number
 */
void flush_cache_page(struct vm_area_struct *vma, unsigned long address, unsigned long pfn)
{
	unsigned long phys = pfn << PAGE_SHIFT;
	unsigned int alias_mask;

	alias_mask = cpu_data->dcache.alias_mask;

	/* We only need to flush D-cache when we have alias */
	if ((address^phys) & alias_mask) {
		/* Loop 4K of the D-cache */
		flush_cache_4096(
			CACHE_OC_ADDRESS_ARRAY | (address & alias_mask),
			phys);
		/* Loop another 4K of the D-cache */
		flush_cache_4096(
			CACHE_OC_ADDRESS_ARRAY | (phys & alias_mask),
			phys);
	}

	alias_mask = cpu_data->icache.alias_mask;
	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.
		 */
		flush_cache_4096(
			CACHE_IC_ADDRESS_ARRAY | (address & alias_mask),
			phys);
	}
}

/*
 * 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)
{
	unsigned long d = 0, p = start & PAGE_MASK;
	unsigned long alias_mask = cpu_data->dcache.alias_mask;
	unsigned long n_aliases = cpu_data->dcache.n_aliases;
	unsigned long select_bit;
	unsigned long all_aliases_mask;
	unsigned long addr_offset;
	unsigned long phys;
	pgd_t *dir;
	pmd_t *pmd;
	pud_t *pud;
	pte_t *pte;
	pte_t entry;
	int i;

	/*
	 * 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.
	 */
	if (n_aliases == 0)
		return;

	all_aliases_mask = (1 << n_aliases) - 1;

	/*
	 * 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.
	 */
	if (((end - start) >> PAGE_SHIFT) >= 64) {
		flush_dcache_all();

		if (vma->vm_flags & VM_EXEC)
			flush_icache_all();

		return;
	}

	dir = pgd_offset(vma->vm_mm, p);
	pud = pud_offset(dir, p);
	pmd = pmd_offset(pud, p);
	end = PAGE_ALIGN(end);

	do {
		if (pmd_none(*pmd) || pmd_bad(*pmd)) {
			p &= ~((1 << PMD_SHIFT) - 1);
			p += (1 << PMD_SHIFT);
			pmd++;

			continue;
		}

		pte = pte_offset_kernel(pmd, p);

		do {
			entry = *pte;

			if ((pte_val(entry) & _PAGE_PRESENT)) {
				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:
	for (i = 0, select_bit = 0x1, addr_offset = 0x0; i < n_aliases;
	     i++, select_bit <<= 1, addr_offset += PAGE_SIZE)
		if (d & select_bit) {
			(*__flush_dcache_segment_fn)(addr_offset, PAGE_SIZE);
			wmb();
		}

	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.
		 */
		flush_icache_all();
	}
}

/*
 * 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));
	mb();
}

/**
 * __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;

	dcache = &cpu_data->dcache;
	/* 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;

	dcache = &cpu_data->dcache;
	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;

	dcache = &cpu_data->dcache;
	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;

	dcache = &cpu_data->dcache;
	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);
}