xref: /openbmc/linux/arch/mips/mm/c-r4k.c (revision 7fb6f7b0)

/*
 * 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.
 *
 * Copyright (C) 1996 David S. Miller (davem@davemloft.net)
 * Copyright (C) 1997, 1998, 1999, 2000, 2001, 2002 Ralf Baechle (ralf@gnu.org)
 * Copyright (C) 1999, 2000 Silicon Graphics, Inc.
 */
#include <linux/cpu_pm.h>
#include <linux/hardirq.h>
#include <linux/init.h>
#include <linux/highmem.h>
#include <linux/kernel.h>
#include <linux/linkage.h>
#include <linux/preempt.h>
#include <linux/sched.h>
#include <linux/smp.h>
#include <linux/mm.h>
#include <linux/export.h>
#include <linux/bitops.h>
#include <linux/dma-map-ops.h> /* for dma_default_coherent */

#include <asm/bcache.h>
#include <asm/bootinfo.h>
#include <asm/cache.h>
#include <asm/cacheops.h>
#include <asm/cpu.h>
#include <asm/cpu-features.h>
#include <asm/cpu-type.h>
#include <asm/io.h>
#include <asm/page.h>
#include <asm/r4kcache.h>
#include <asm/sections.h>
#include <asm/mmu_context.h>
#include <asm/cacheflush.h> /* for run_uncached() */
#include <asm/traps.h>
#include <asm/mips-cps.h>

/*
 * Bits describing what cache ops an SMP callback function may perform.
 *
 * R4K_HIT   -	Virtual user or kernel address based cache operations. The
 *		active_mm must be checked before using user addresses, falling
 *		back to kmap.
 * R4K_INDEX -	Index based cache operations.
 */

#define R4K_HIT		BIT(0)
#define R4K_INDEX	BIT(1)

/**
 * r4k_op_needs_ipi() - Decide if a cache op needs to be done on every core.
 * @type:	Type of cache operations (R4K_HIT or R4K_INDEX).
 *
 * Decides whether a cache op needs to be performed on every core in the system.
 * This may change depending on the @type of cache operation, as well as the set
 * of online CPUs, so preemption should be disabled by the caller to prevent CPU
 * hotplug from changing the result.
 *
 * Returns:	1 if the cache operation @type should be done on every core in
 *		the system.
 *		0 if the cache operation @type is globalized and only needs to
 *		be performed on a simple CPU.
 */
static inline bool r4k_op_needs_ipi(unsigned int type)
{
	/* The MIPS Coherence Manager (CM) globalizes address-based cache ops */
	if (type == R4K_HIT && mips_cm_present())
		return false;

	/*
	 * Hardware doesn't globalize the required cache ops, so SMP calls may
	 * be needed, but only if there are foreign CPUs (non-siblings with
	 * separate caches).
	 */
	/* cpu_foreign_map[] undeclared when !CONFIG_SMP */
#ifdef CONFIG_SMP
	return !cpumask_empty(&cpu_foreign_map[0]);
#else
	return false;
#endif
}

/*
 * Special Variant of smp_call_function for use by cache functions:
 *
 *  o No return value
 *  o collapses to normal function call on UP kernels
 *  o collapses to normal function call on systems with a single shared
 *    primary cache.
 *  o doesn't disable interrupts on the local CPU
 */
static inline void r4k_on_each_cpu(unsigned int type,
				   void (*func)(void *info), void *info)
{
	preempt_disable();
	if (r4k_op_needs_ipi(type))
		smp_call_function_many(&cpu_foreign_map[smp_processor_id()],
				       func, info, 1);
	func(info);
	preempt_enable();
}

/*
 * Must die.
 */
static unsigned long icache_size __read_mostly;
static unsigned long dcache_size __read_mostly;
static unsigned long vcache_size __read_mostly;
static unsigned long scache_size __read_mostly;

#define cpu_is_r4600_v1_x()	((read_c0_prid() & 0xfffffff0) == 0x00002010)
#define cpu_is_r4600_v2_x()	((read_c0_prid() & 0xfffffff0) == 0x00002020)

#define R4600_HIT_CACHEOP_WAR_IMPL					\
do {									\
	if (IS_ENABLED(CONFIG_WAR_R4600_V2_HIT_CACHEOP) &&		\
	    cpu_is_r4600_v2_x())					\
		*(volatile unsigned long *)CKSEG1;			\
	if (IS_ENABLED(CONFIG_WAR_R4600_V1_HIT_CACHEOP))					\
		__asm__ __volatile__("nop;nop;nop;nop");		\
} while (0)

static void (*r4k_blast_dcache_page)(unsigned long addr);

static inline void r4k_blast_dcache_page_dc32(unsigned long addr)
{
	R4600_HIT_CACHEOP_WAR_IMPL;
	blast_dcache32_page(addr);
}

static inline void r4k_blast_dcache_page_dc64(unsigned long addr)
{
	blast_dcache64_page(addr);
}

static inline void r4k_blast_dcache_page_dc128(unsigned long addr)
{
	blast_dcache128_page(addr);
}

static void r4k_blast_dcache_page_setup(void)
{
	unsigned long  dc_lsize = cpu_dcache_line_size();

	switch (dc_lsize) {
	case 0:
		r4k_blast_dcache_page = (void *)cache_noop;
		break;
	case 16:
		r4k_blast_dcache_page = blast_dcache16_page;
		break;
	case 32:
		r4k_blast_dcache_page = r4k_blast_dcache_page_dc32;
		break;
	case 64:
		r4k_blast_dcache_page = r4k_blast_dcache_page_dc64;
		break;
	case 128:
		r4k_blast_dcache_page = r4k_blast_dcache_page_dc128;
		break;
	default:
		break;
	}
}

#ifndef CONFIG_EVA
#define r4k_blast_dcache_user_page  r4k_blast_dcache_page
#else

static void (*r4k_blast_dcache_user_page)(unsigned long addr);

static void r4k_blast_dcache_user_page_setup(void)
{
	unsigned long  dc_lsize = cpu_dcache_line_size();

	if (dc_lsize == 0)
		r4k_blast_dcache_user_page = (void *)cache_noop;
	else if (dc_lsize == 16)
		r4k_blast_dcache_user_page = blast_dcache16_user_page;
	else if (dc_lsize == 32)
		r4k_blast_dcache_user_page = blast_dcache32_user_page;
	else if (dc_lsize == 64)
		r4k_blast_dcache_user_page = blast_dcache64_user_page;
}

#endif

void (* r4k_blast_dcache)(void);
EXPORT_SYMBOL(r4k_blast_dcache);

static void r4k_blast_dcache_setup(void)
{
	unsigned long dc_lsize = cpu_dcache_line_size();

	if (dc_lsize == 0)
		r4k_blast_dcache = (void *)cache_noop;
	else if (dc_lsize == 16)
		r4k_blast_dcache = blast_dcache16;
	else if (dc_lsize == 32)
		r4k_blast_dcache = blast_dcache32;
	else if (dc_lsize == 64)
		r4k_blast_dcache = blast_dcache64;
	else if (dc_lsize == 128)
		r4k_blast_dcache = blast_dcache128;
}

/* force code alignment (used for CONFIG_WAR_TX49XX_ICACHE_INDEX_INV) */
#define JUMP_TO_ALIGN(order) \
	__asm__ __volatile__( \
		"b\t1f\n\t" \
		".align\t" #order "\n\t" \
		"1:\n\t" \
		)
#define CACHE32_UNROLL32_ALIGN	JUMP_TO_ALIGN(10) /* 32 * 32 = 1024 */
#define CACHE32_UNROLL32_ALIGN2 JUMP_TO_ALIGN(11)

static inline void blast_r4600_v1_icache32(void)
{
	unsigned long flags;

	local_irq_save(flags);
	blast_icache32();
	local_irq_restore(flags);
}

static inline void tx49_blast_icache32(void)
{
	unsigned long start = INDEX_BASE;
	unsigned long end = start + current_cpu_data.icache.waysize;
	unsigned long ws_inc = 1UL << current_cpu_data.icache.waybit;
	unsigned long ws_end = current_cpu_data.icache.ways <<
			       current_cpu_data.icache.waybit;
	unsigned long ws, addr;

	CACHE32_UNROLL32_ALIGN2;
	/* I'm in even chunk.  blast odd chunks */
	for (ws = 0; ws < ws_end; ws += ws_inc)
		for (addr = start + 0x400; addr < end; addr += 0x400 * 2)
			cache_unroll(32, kernel_cache, Index_Invalidate_I,
				     addr | ws, 32);
	CACHE32_UNROLL32_ALIGN;
	/* I'm in odd chunk.  blast even chunks */
	for (ws = 0; ws < ws_end; ws += ws_inc)
		for (addr = start; addr < end; addr += 0x400 * 2)
			cache_unroll(32, kernel_cache, Index_Invalidate_I,
				     addr | ws, 32);
}

static void (* r4k_blast_icache_page)(unsigned long addr);

static void r4k_blast_icache_page_setup(void)
{
	unsigned long ic_lsize = cpu_icache_line_size();

	if (ic_lsize == 0)
		r4k_blast_icache_page = (void *)cache_noop;
	else if (ic_lsize == 16)
		r4k_blast_icache_page = blast_icache16_page;
	else if (ic_lsize == 32 && current_cpu_type() == CPU_LOONGSON2EF)
		r4k_blast_icache_page = loongson2_blast_icache32_page;
	else if (ic_lsize == 32)
		r4k_blast_icache_page = blast_icache32_page;
	else if (ic_lsize == 64)
		r4k_blast_icache_page = blast_icache64_page;
	else if (ic_lsize == 128)
		r4k_blast_icache_page = blast_icache128_page;
}

#ifndef CONFIG_EVA
#define r4k_blast_icache_user_page  r4k_blast_icache_page
#else

static void (*r4k_blast_icache_user_page)(unsigned long addr);

static void r4k_blast_icache_user_page_setup(void)
{
	unsigned long ic_lsize = cpu_icache_line_size();

	if (ic_lsize == 0)
		r4k_blast_icache_user_page = (void *)cache_noop;
	else if (ic_lsize == 16)
		r4k_blast_icache_user_page = blast_icache16_user_page;
	else if (ic_lsize == 32)
		r4k_blast_icache_user_page = blast_icache32_user_page;
	else if (ic_lsize == 64)
		r4k_blast_icache_user_page = blast_icache64_user_page;
}

#endif

void (* r4k_blast_icache)(void);
EXPORT_SYMBOL(r4k_blast_icache);

static void r4k_blast_icache_setup(void)
{
	unsigned long ic_lsize = cpu_icache_line_size();

	if (ic_lsize == 0)
		r4k_blast_icache = (void *)cache_noop;
	else if (ic_lsize == 16)
		r4k_blast_icache = blast_icache16;
	else if (ic_lsize == 32) {
		if (IS_ENABLED(CONFIG_WAR_R4600_V1_INDEX_ICACHEOP) &&
		    cpu_is_r4600_v1_x())
			r4k_blast_icache = blast_r4600_v1_icache32;
		else if (IS_ENABLED(CONFIG_WAR_TX49XX_ICACHE_INDEX_INV))
			r4k_blast_icache = tx49_blast_icache32;
		else if (current_cpu_type() == CPU_LOONGSON2EF)
			r4k_blast_icache = loongson2_blast_icache32;
		else
			r4k_blast_icache = blast_icache32;
	} else if (ic_lsize == 64)
		r4k_blast_icache = blast_icache64;
	else if (ic_lsize == 128)
		r4k_blast_icache = blast_icache128;
}

static void (* r4k_blast_scache_page)(unsigned long addr);

static void r4k_blast_scache_page_setup(void)
{
	unsigned long sc_lsize = cpu_scache_line_size();

	if (scache_size == 0)
		r4k_blast_scache_page = (void *)cache_noop;
	else if (sc_lsize == 16)
		r4k_blast_scache_page = blast_scache16_page;
	else if (sc_lsize == 32)
		r4k_blast_scache_page = blast_scache32_page;
	else if (sc_lsize == 64)
		r4k_blast_scache_page = blast_scache64_page;
	else if (sc_lsize == 128)
		r4k_blast_scache_page = blast_scache128_page;
}

static void (* r4k_blast_scache)(void);

static void r4k_blast_scache_setup(void)
{
	unsigned long sc_lsize = cpu_scache_line_size();

	if (scache_size == 0)
		r4k_blast_scache = (void *)cache_noop;
	else if (sc_lsize == 16)
		r4k_blast_scache = blast_scache16;
	else if (sc_lsize == 32)
		r4k_blast_scache = blast_scache32;
	else if (sc_lsize == 64)
		r4k_blast_scache = blast_scache64;
	else if (sc_lsize == 128)
		r4k_blast_scache = blast_scache128;
}

static void (*r4k_blast_scache_node)(long node);

static void r4k_blast_scache_node_setup(void)
{
	unsigned long sc_lsize = cpu_scache_line_size();

	if (current_cpu_type() != CPU_LOONGSON64)
		r4k_blast_scache_node = (void *)cache_noop;
	else if (sc_lsize == 16)
		r4k_blast_scache_node = blast_scache16_node;
	else if (sc_lsize == 32)
		r4k_blast_scache_node = blast_scache32_node;
	else if (sc_lsize == 64)
		r4k_blast_scache_node = blast_scache64_node;
	else if (sc_lsize == 128)
		r4k_blast_scache_node = blast_scache128_node;
}

static inline void local_r4k___flush_cache_all(void * args)
{
	switch (current_cpu_type()) {
	case CPU_LOONGSON2EF:
	case CPU_R4000SC:
	case CPU_R4000MC:
	case CPU_R4400SC:
	case CPU_R4400MC:
	case CPU_R10000:
	case CPU_R12000:
	case CPU_R14000:
	case CPU_R16000:
		/*
		 * These caches are inclusive caches, that is, if something
		 * is not cached in the S-cache, we know it also won't be
		 * in one of the primary caches.
		 */
		r4k_blast_scache();
		break;

	case CPU_LOONGSON64:
		/* Use get_ebase_cpunum() for both NUMA=y/n */
		r4k_blast_scache_node(get_ebase_cpunum() >> 2);
		break;

	case CPU_BMIPS5000:
		r4k_blast_scache();
		__sync();
		break;

	default:
		r4k_blast_dcache();
		r4k_blast_icache();
		break;
	}
}

static void r4k___flush_cache_all(void)
{
	r4k_on_each_cpu(R4K_INDEX, local_r4k___flush_cache_all, NULL);
}

/**
 * has_valid_asid() - Determine if an mm already has an ASID.
 * @mm:		Memory map.
 * @type:	R4K_HIT or R4K_INDEX, type of cache op.
 *
 * Determines whether @mm already has an ASID on any of the CPUs which cache ops
 * of type @type within an r4k_on_each_cpu() call will affect. If
 * r4k_on_each_cpu() does an SMP call to a single VPE in each core, then the
 * scope of the operation is confined to sibling CPUs, otherwise all online CPUs
 * will need to be checked.
 *
 * Must be called in non-preemptive context.
 *
 * Returns:	1 if the CPUs affected by @type cache ops have an ASID for @mm.
 *		0 otherwise.
 */
static inline int has_valid_asid(const struct mm_struct *mm, unsigned int type)
{
	unsigned int i;
	const cpumask_t *mask = cpu_present_mask;

	if (cpu_has_mmid)
		return cpu_context(0, mm) != 0;

	/* cpu_sibling_map[] undeclared when !CONFIG_SMP */
#ifdef CONFIG_SMP
	/*
	 * If r4k_on_each_cpu does SMP calls, it does them to a single VPE in
	 * each foreign core, so we only need to worry about siblings.
	 * Otherwise we need to worry about all present CPUs.
	 */
	if (r4k_op_needs_ipi(type))
		mask = &cpu_sibling_map[smp_processor_id()];
#endif
	for_each_cpu(i, mask)
		if (cpu_context(i, mm))
			return 1;
	return 0;
}

static void r4k__flush_cache_vmap(void)
{
	r4k_blast_dcache();
}

static void r4k__flush_cache_vunmap(void)
{
	r4k_blast_dcache();
}

/*
 * Note: flush_tlb_range() assumes flush_cache_range() sufficiently flushes
 * whole caches when vma is executable.
 */
static inline void local_r4k_flush_cache_range(void * args)
{
	struct vm_area_struct *vma = args;
	int exec = vma->vm_flags & VM_EXEC;

	if (!has_valid_asid(vma->vm_mm, R4K_INDEX))
		return;

	/*
	 * If dcache can alias, we must blast it since mapping is changing.
	 * If executable, we must ensure any dirty lines are written back far
	 * enough to be visible to icache.
	 */
	if (cpu_has_dc_aliases || (exec && !cpu_has_ic_fills_f_dc))
		r4k_blast_dcache();
	/* If executable, blast stale lines from icache */
	if (exec)
		r4k_blast_icache();
}

static void r4k_flush_cache_range(struct vm_area_struct *vma,
	unsigned long start, unsigned long end)
{
	int exec = vma->vm_flags & VM_EXEC;

	if (cpu_has_dc_aliases || exec)
		r4k_on_each_cpu(R4K_INDEX, local_r4k_flush_cache_range, vma);
}

static inline void local_r4k_flush_cache_mm(void * args)
{
	struct mm_struct *mm = args;

	if (!has_valid_asid(mm, R4K_INDEX))
		return;

	/*
	 * Kludge alert.  For obscure reasons R4000SC and R4400SC go nuts if we
	 * only flush the primary caches but R1x000 behave sane ...
	 * R4000SC and R4400SC indexed S-cache ops also invalidate primary
	 * caches, so we can bail out early.
	 */
	if (current_cpu_type() == CPU_R4000SC ||
	    current_cpu_type() == CPU_R4000MC ||
	    current_cpu_type() == CPU_R4400SC ||
	    current_cpu_type() == CPU_R4400MC) {
		r4k_blast_scache();
		return;
	}

	r4k_blast_dcache();
}

static void r4k_flush_cache_mm(struct mm_struct *mm)
{
	if (!cpu_has_dc_aliases)
		return;

	r4k_on_each_cpu(R4K_INDEX, local_r4k_flush_cache_mm, mm);
}

struct flush_cache_page_args {
	struct vm_area_struct *vma;
	unsigned long addr;
	unsigned long pfn;
};

static inline void local_r4k_flush_cache_page(void *args)
{
	struct flush_cache_page_args *fcp_args = args;
	struct vm_area_struct *vma = fcp_args->vma;
	unsigned long addr = fcp_args->addr;
	struct page *page = pfn_to_page(fcp_args->pfn);
	int exec = vma->vm_flags & VM_EXEC;
	struct mm_struct *mm = vma->vm_mm;
	int map_coherent = 0;
	pmd_t *pmdp;
	pte_t *ptep;
	void *vaddr;

	/*
	 * If owns no valid ASID yet, cannot possibly have gotten
	 * this page into the cache.
	 */
	if (!has_valid_asid(mm, R4K_HIT))
		return;

	addr &= PAGE_MASK;
	pmdp = pmd_off(mm, addr);
	ptep = pte_offset_kernel(pmdp, addr);

	/*
	 * If the page isn't marked valid, the page cannot possibly be
	 * in the cache.
	 */
	if (!(pte_present(*ptep)))
		return;

	if ((mm == current->active_mm) && (pte_val(*ptep) & _PAGE_VALID))
		vaddr = NULL;
	else {
		/*
		 * Use kmap_coherent or kmap_atomic to do flushes for
		 * another ASID than the current one.
		 */
		map_coherent = (cpu_has_dc_aliases &&
				page_mapcount(page) &&
				!Page_dcache_dirty(page));
		if (map_coherent)
			vaddr = kmap_coherent(page, addr);
		else
			vaddr = kmap_atomic(page);
		addr = (unsigned long)vaddr;
	}

	if (cpu_has_dc_aliases || (exec && !cpu_has_ic_fills_f_dc)) {
		vaddr ? r4k_blast_dcache_page(addr) :
			r4k_blast_dcache_user_page(addr);
		if (exec && !cpu_icache_snoops_remote_store)
			r4k_blast_scache_page(addr);
	}
	if (exec) {
		if (vaddr && cpu_has_vtag_icache && mm == current->active_mm) {
			drop_mmu_context(mm);
		} else
			vaddr ? r4k_blast_icache_page(addr) :
				r4k_blast_icache_user_page(addr);
	}

	if (vaddr) {
		if (map_coherent)
			kunmap_coherent();
		else
			kunmap_atomic(vaddr);
	}
}

static void r4k_flush_cache_page(struct vm_area_struct *vma,
	unsigned long addr, unsigned long pfn)
{
	struct flush_cache_page_args args;

	args.vma = vma;
	args.addr = addr;
	args.pfn = pfn;

	r4k_on_each_cpu(R4K_HIT, local_r4k_flush_cache_page, &args);
}

static inline void local_r4k_flush_data_cache_page(void * addr)
{
	r4k_blast_dcache_page((unsigned long) addr);
}

static void r4k_flush_data_cache_page(unsigned long addr)
{
	if (in_atomic())
		local_r4k_flush_data_cache_page((void *)addr);
	else
		r4k_on_each_cpu(R4K_HIT, local_r4k_flush_data_cache_page,
				(void *) addr);
}

struct flush_icache_range_args {
	unsigned long start;
	unsigned long end;
	unsigned int type;
	bool user;
};

static inline void __local_r4k_flush_icache_range(unsigned long start,
						  unsigned long end,
						  unsigned int type,
						  bool user)
{
	if (!cpu_has_ic_fills_f_dc) {
		if (type == R4K_INDEX ||
		    (type & R4K_INDEX && end - start >= dcache_size)) {
			r4k_blast_dcache();
		} else {
			R4600_HIT_CACHEOP_WAR_IMPL;
			if (user)
				protected_blast_dcache_range(start, end);
			else
				blast_dcache_range(start, end);
		}
	}

	if (type == R4K_INDEX ||
	    (type & R4K_INDEX && end - start > icache_size))
		r4k_blast_icache();
	else {
		switch (boot_cpu_type()) {
		case CPU_LOONGSON2EF:
			protected_loongson2_blast_icache_range(start, end);
			break;

		default:
			if (user)
				protected_blast_icache_range(start, end);
			else
				blast_icache_range(start, end);
			break;
		}
	}
}

static inline void local_r4k_flush_icache_range(unsigned long start,
						unsigned long end)
{
	__local_r4k_flush_icache_range(start, end, R4K_HIT | R4K_INDEX, false);
}

static inline void local_r4k_flush_icache_user_range(unsigned long start,
						     unsigned long end)
{
	__local_r4k_flush_icache_range(start, end, R4K_HIT | R4K_INDEX, true);
}

static inline void local_r4k_flush_icache_range_ipi(void *args)
{
	struct flush_icache_range_args *fir_args = args;
	unsigned long start = fir_args->start;
	unsigned long end = fir_args->end;
	unsigned int type = fir_args->type;
	bool user = fir_args->user;

	__local_r4k_flush_icache_range(start, end, type, user);
}

static void __r4k_flush_icache_range(unsigned long start, unsigned long end,
				     bool user)
{
	struct flush_icache_range_args args;
	unsigned long size, cache_size;

	args.start = start;
	args.end = end;
	args.type = R4K_HIT | R4K_INDEX;
	args.user = user;

	/*
	 * Indexed cache ops require an SMP call.
	 * Consider if that can or should be avoided.
	 */
	preempt_disable();
	if (r4k_op_needs_ipi(R4K_INDEX) && !r4k_op_needs_ipi(R4K_HIT)) {
		/*
		 * If address-based cache ops don't require an SMP call, then
		 * use them exclusively for small flushes.
		 */
		size = end - start;
		cache_size = icache_size;
		if (!cpu_has_ic_fills_f_dc) {
			size *= 2;
			cache_size += dcache_size;
		}
		if (size <= cache_size)
			args.type &= ~R4K_INDEX;
	}
	r4k_on_each_cpu(args.type, local_r4k_flush_icache_range_ipi, &args);
	preempt_enable();
	instruction_hazard();
}

static void r4k_flush_icache_range(unsigned long start, unsigned long end)
{
	return __r4k_flush_icache_range(start, end, false);
}

static void r4k_flush_icache_user_range(unsigned long start, unsigned long end)
{
	return __r4k_flush_icache_range(start, end, true);
}

#ifdef CONFIG_DMA_NONCOHERENT

static void r4k_dma_cache_wback_inv(unsigned long addr, unsigned long size)
{
	/* Catch bad driver code */
	if (WARN_ON(size == 0))
		return;

	preempt_disable();
	if (cpu_has_inclusive_pcaches) {
		if (size >= scache_size) {
			if (current_cpu_type() != CPU_LOONGSON64)
				r4k_blast_scache();
			else
				r4k_blast_scache_node(pa_to_nid(addr));
		} else {
			blast_scache_range(addr, addr + size);
		}
		preempt_enable();
		__sync();
		return;
	}

	/*
	 * Either no secondary cache or the available caches don't have the
	 * subset property so we have to flush the primary caches
	 * explicitly.
	 * If we would need IPI to perform an INDEX-type operation, then
	 * we have to use the HIT-type alternative as IPI cannot be used
	 * here due to interrupts possibly being disabled.
	 */
	if (!r4k_op_needs_ipi(R4K_INDEX) && size >= dcache_size) {
		r4k_blast_dcache();
	} else {
		R4600_HIT_CACHEOP_WAR_IMPL;
		blast_dcache_range(addr, addr + size);
	}
	preempt_enable();

	bc_wback_inv(addr, size);
	__sync();
}

static void prefetch_cache_inv(unsigned long addr, unsigned long size)
{
	unsigned int linesz = cpu_scache_line_size();
	unsigned long addr0 = addr, addr1;

	addr0 &= ~(linesz - 1);
	addr1 = (addr0 + size - 1) & ~(linesz - 1);

	protected_writeback_scache_line(addr0);
	if (likely(addr1 != addr0))
		protected_writeback_scache_line(addr1);
	else
		return;

	addr0 += linesz;
	if (likely(addr1 != addr0))
		protected_writeback_scache_line(addr0);
	else
		return;

	addr1 -= linesz;
	if (likely(addr1 > addr0))
		protected_writeback_scache_line(addr0);
}

static void r4k_dma_cache_inv(unsigned long addr, unsigned long size)
{
	/* Catch bad driver code */
	if (WARN_ON(size == 0))
		return;

	preempt_disable();

	if (current_cpu_type() == CPU_BMIPS5000)
		prefetch_cache_inv(addr, size);

	if (cpu_has_inclusive_pcaches) {
		if (size >= scache_size) {
			if (current_cpu_type() != CPU_LOONGSON64)
				r4k_blast_scache();
			else
				r4k_blast_scache_node(pa_to_nid(addr));
		} else {
			/*
			 * There is no clearly documented alignment requirement
			 * for the cache instruction on MIPS processors and
			 * some processors, among them the RM5200 and RM7000
			 * QED processors will throw an address error for cache
			 * hit ops with insufficient alignment.	 Solved by
			 * aligning the address to cache line size.
			 */
			blast_inv_scache_range(addr, addr + size);
		}
		preempt_enable();
		__sync();
		return;
	}

	if (!r4k_op_needs_ipi(R4K_INDEX) && size >= dcache_size) {
		r4k_blast_dcache();
	} else {
		R4600_HIT_CACHEOP_WAR_IMPL;
		blast_inv_dcache_range(addr, addr + size);
	}
	preempt_enable();

	bc_inv(addr, size);
	__sync();
}
#endif /* CONFIG_DMA_NONCOHERENT */

static void r4k_flush_icache_all(void)
{
	if (cpu_has_vtag_icache)
		r4k_blast_icache();
}

struct flush_kernel_vmap_range_args {
	unsigned long	vaddr;
	int		size;
};

static inline void local_r4k_flush_kernel_vmap_range_index(void *args)
{
	/*
	 * Aliases only affect the primary caches so don't bother with
	 * S-caches or T-caches.
	 */
	r4k_blast_dcache();
}

static inline void local_r4k_flush_kernel_vmap_range(void *args)
{
	struct flush_kernel_vmap_range_args *vmra = args;
	unsigned long vaddr = vmra->vaddr;
	int size = vmra->size;

	/*
	 * Aliases only affect the primary caches so don't bother with
	 * S-caches or T-caches.
	 */
	R4600_HIT_CACHEOP_WAR_IMPL;
	blast_dcache_range(vaddr, vaddr + size);
}

static void r4k_flush_kernel_vmap_range(unsigned long vaddr, int size)
{
	struct flush_kernel_vmap_range_args args;

	args.vaddr = (unsigned long) vaddr;
	args.size = size;

	if (size >= dcache_size)
		r4k_on_each_cpu(R4K_INDEX,
				local_r4k_flush_kernel_vmap_range_index, NULL);
	else
		r4k_on_each_cpu(R4K_HIT, local_r4k_flush_kernel_vmap_range,
				&args);
}

static inline void rm7k_erratum31(void)
{
	const unsigned long ic_lsize = 32;
	unsigned long addr;

	/* RM7000 erratum #31. The icache is screwed at startup. */
	write_c0_taglo(0);
	write_c0_taghi(0);

	for (addr = INDEX_BASE; addr <= INDEX_BASE + 4096; addr += ic_lsize) {
		__asm__ __volatile__ (
			".set push\n\t"
			".set noreorder\n\t"
			".set mips3\n\t"
			"cache\t%1, 0(%0)\n\t"
			"cache\t%1, 0x1000(%0)\n\t"
			"cache\t%1, 0x2000(%0)\n\t"
			"cache\t%1, 0x3000(%0)\n\t"
			"cache\t%2, 0(%0)\n\t"
			"cache\t%2, 0x1000(%0)\n\t"
			"cache\t%2, 0x2000(%0)\n\t"
			"cache\t%2, 0x3000(%0)\n\t"
			"cache\t%1, 0(%0)\n\t"
			"cache\t%1, 0x1000(%0)\n\t"
			"cache\t%1, 0x2000(%0)\n\t"
			"cache\t%1, 0x3000(%0)\n\t"
			".set pop\n"
			:
			: "r" (addr), "i" (Index_Store_Tag_I), "i" (Fill_I));
	}
}

static inline int alias_74k_erratum(struct cpuinfo_mips *c)
{
	unsigned int imp = c->processor_id & PRID_IMP_MASK;
	unsigned int rev = c->processor_id & PRID_REV_MASK;
	int present = 0;

	/*
	 * Early versions of the 74K do not update the cache tags on a
	 * vtag miss/ptag hit which can occur in the case of KSEG0/KUSEG
	 * aliases.  In this case it is better to treat the cache as always
	 * having aliases.  Also disable the synonym tag update feature
	 * where available.  In this case no opportunistic tag update will
	 * happen where a load causes a virtual address miss but a physical
	 * address hit during a D-cache look-up.
	 */
	switch (imp) {
	case PRID_IMP_74K:
		if (rev <= PRID_REV_ENCODE_332(2, 4, 0))
			present = 1;
		if (rev == PRID_REV_ENCODE_332(2, 4, 0))
			write_c0_config6(read_c0_config6() | MTI_CONF6_SYND);
		break;
	case PRID_IMP_1074K:
		if (rev <= PRID_REV_ENCODE_332(1, 1, 0)) {
			present = 1;
			write_c0_config6(read_c0_config6() | MTI_CONF6_SYND);
		}
		break;
	default:
		BUG();
	}

	return present;
}

static void b5k_instruction_hazard(void)
{
	__sync();
	__sync();
	__asm__ __volatile__(
	"       nop; nop; nop; nop; nop; nop; nop; nop\n"
	"       nop; nop; nop; nop; nop; nop; nop; nop\n"
	"       nop; nop; nop; nop; nop; nop; nop; nop\n"
	"       nop; nop; nop; nop; nop; nop; nop; nop\n"
	: : : "memory");
}

static char *way_string[] = { NULL, "direct mapped", "2-way",
	"3-way", "4-way", "5-way", "6-way", "7-way", "8-way",
	"9-way", "10-way", "11-way", "12-way",
	"13-way", "14-way", "15-way", "16-way",
};

static void probe_pcache(void)
{
	struct cpuinfo_mips *c = &current_cpu_data;
	unsigned int config = read_c0_config();
	unsigned int prid = read_c0_prid();
	int has_74k_erratum = 0;
	unsigned long config1;
	unsigned int lsize;

	switch (current_cpu_type()) {
	case CPU_R4600:			/* QED style two way caches? */
	case CPU_R4700:
	case CPU_R5000:
	case CPU_NEVADA:
		icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
		c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
		c->icache.ways = 2;
		c->icache.waybit = __ffs(icache_size/2);

		dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
		c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
		c->dcache.ways = 2;
		c->dcache.waybit= __ffs(dcache_size/2);

		c->options |= MIPS_CPU_CACHE_CDEX_P;
		break;

	case CPU_R5500:
		icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
		c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
		c->icache.ways = 2;
		c->icache.waybit= 0;

		dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
		c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
		c->dcache.ways = 2;
		c->dcache.waybit = 0;

		c->options |= MIPS_CPU_CACHE_CDEX_P | MIPS_CPU_PREFETCH;
		break;

	case CPU_TX49XX:
		icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
		c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
		c->icache.ways = 4;
		c->icache.waybit= 0;

		dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
		c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
		c->dcache.ways = 4;
		c->dcache.waybit = 0;

		c->options |= MIPS_CPU_CACHE_CDEX_P;
		c->options |= MIPS_CPU_PREFETCH;
		break;

	case CPU_R4000PC:
	case CPU_R4000SC:
	case CPU_R4000MC:
	case CPU_R4400PC:
	case CPU_R4400SC:
	case CPU_R4400MC:
	case CPU_R4300:
		icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
		c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
		c->icache.ways = 1;
		c->icache.waybit = 0;	/* doesn't matter */

		dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
		c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
		c->dcache.ways = 1;
		c->dcache.waybit = 0;	/* does not matter */

		c->options |= MIPS_CPU_CACHE_CDEX_P;
		break;

	case CPU_R10000:
	case CPU_R12000:
	case CPU_R14000:
	case CPU_R16000:
		icache_size = 1 << (12 + ((config & R10K_CONF_IC) >> 29));
		c->icache.linesz = 64;
		c->icache.ways = 2;
		c->icache.waybit = 0;

		dcache_size = 1 << (12 + ((config & R10K_CONF_DC) >> 26));
		c->dcache.linesz = 32;
		c->dcache.ways = 2;
		c->dcache.waybit = 0;

		c->options |= MIPS_CPU_PREFETCH;
		break;

	case CPU_RM7000:
		rm7k_erratum31();

		icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
		c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
		c->icache.ways = 4;
		c->icache.waybit = __ffs(icache_size / c->icache.ways);

		dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
		c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
		c->dcache.ways = 4;
		c->dcache.waybit = __ffs(dcache_size / c->dcache.ways);

		c->options |= MIPS_CPU_CACHE_CDEX_P;
		c->options |= MIPS_CPU_PREFETCH;
		break;

	case CPU_LOONGSON2EF:
		icache_size = 1 << (12 + ((config & CONF_IC) >> 9));
		c->icache.linesz = 16 << ((config & CONF_IB) >> 5);
		if (prid & 0x3)
			c->icache.ways = 4;
		else
			c->icache.ways = 2;
		c->icache.waybit = 0;

		dcache_size = 1 << (12 + ((config & CONF_DC) >> 6));
		c->dcache.linesz = 16 << ((config & CONF_DB) >> 4);
		if (prid & 0x3)
			c->dcache.ways = 4;
		else
			c->dcache.ways = 2;
		c->dcache.waybit = 0;
		break;

	case CPU_LOONGSON64:
		config1 = read_c0_config1();
		lsize = (config1 >> 19) & 7;
		if (lsize)
			c->icache.linesz = 2 << lsize;
		else
			c->icache.linesz = 0;
		c->icache.sets = 64 << ((config1 >> 22) & 7);
		c->icache.ways = 1 + ((config1 >> 16) & 7);
		icache_size = c->icache.sets *
					  c->icache.ways *
					  c->icache.linesz;
		c->icache.waybit = 0;

		lsize = (config1 >> 10) & 7;
		if (lsize)
			c->dcache.linesz = 2 << lsize;
		else
			c->dcache.linesz = 0;
		c->dcache.sets = 64 << ((config1 >> 13) & 7);
		c->dcache.ways = 1 + ((config1 >> 7) & 7);
		dcache_size = c->dcache.sets *
					  c->dcache.ways *
					  c->dcache.linesz;
		c->dcache.waybit = 0;
		if ((c->processor_id & (PRID_IMP_MASK | PRID_REV_MASK)) >=
				(PRID_IMP_LOONGSON_64C | PRID_REV_LOONGSON3A_R2_0) ||
				(c->processor_id & PRID_IMP_MASK) == PRID_IMP_LOONGSON_64R)
			c->options |= MIPS_CPU_PREFETCH;
		break;

	case CPU_CAVIUM_OCTEON3:
		/* For now lie about the number of ways. */
		c->icache.linesz = 128;
		c->icache.sets = 16;
		c->icache.ways = 8;
		c->icache.flags |= MIPS_CACHE_VTAG;
		icache_size = c->icache.sets * c->icache.ways * c->icache.linesz;

		c->dcache.linesz = 128;
		c->dcache.ways = 8;
		c->dcache.sets = 8;
		dcache_size = c->dcache.sets * c->dcache.ways * c->dcache.linesz;
		c->options |= MIPS_CPU_PREFETCH;
		break;

	default:
		if (!(config & MIPS_CONF_M))
			panic("Don't know how to probe P-caches on this cpu.");

		/*
		 * So we seem to be a MIPS32 or MIPS64 CPU
		 * So let's probe the I-cache ...
		 */
		config1 = read_c0_config1();

		lsize = (config1 >> 19) & 7;

		/* IL == 7 is reserved */
		if (lsize == 7)
			panic("Invalid icache line size");

		c->icache.linesz = lsize ? 2 << lsize : 0;

		c->icache.sets = 32 << (((config1 >> 22) + 1) & 7);
		c->icache.ways = 1 + ((config1 >> 16) & 7);

		icache_size = c->icache.sets *
			      c->icache.ways *
			      c->icache.linesz;
		c->icache.waybit = __ffs(icache_size/c->icache.ways);

		if (config & MIPS_CONF_VI)
			c->icache.flags |= MIPS_CACHE_VTAG;

		/*
		 * Now probe the MIPS32 / MIPS64 data cache.
		 */
		c->dcache.flags = 0;

		lsize = (config1 >> 10) & 7;

		/* DL == 7 is reserved */
		if (lsize == 7)
			panic("Invalid dcache line size");

		c->dcache.linesz = lsize ? 2 << lsize : 0;

		c->dcache.sets = 32 << (((config1 >> 13) + 1) & 7);
		c->dcache.ways = 1 + ((config1 >> 7) & 7);

		dcache_size = c->dcache.sets *
			      c->dcache.ways *
			      c->dcache.linesz;
		c->dcache.waybit = __ffs(dcache_size/c->dcache.ways);

		c->options |= MIPS_CPU_PREFETCH;
		break;
	}

	/*
	 * Processor configuration sanity check for the R4000SC erratum
	 * #5.	With page sizes larger than 32kB there is no possibility
	 * to get a VCE exception anymore so we don't care about this
	 * misconfiguration.  The case is rather theoretical anyway;
	 * presumably no vendor is shipping his hardware in the "bad"
	 * configuration.
	 */
	if ((prid & PRID_IMP_MASK) == PRID_IMP_R4000 &&
	    (prid & PRID_REV_MASK) < PRID_REV_R4400 &&
	    !(config & CONF_SC) && c->icache.linesz != 16 &&
	    PAGE_SIZE <= 0x8000)
		panic("Improper R4000SC processor configuration detected");

	/* compute a couple of other cache variables */
	c->icache.waysize = icache_size / c->icache.ways;
	c->dcache.waysize = dcache_size / c->dcache.ways;

	c->icache.sets = c->icache.linesz ?
		icache_size / (c->icache.linesz * c->icache.ways) : 0;
	c->dcache.sets = c->dcache.linesz ?
		dcache_size / (c->dcache.linesz * c->dcache.ways) : 0;

	/*
	 * R1x000 P-caches are odd in a positive way.  They're 32kB 2-way
	 * virtually indexed so normally would suffer from aliases.  So
	 * normally they'd suffer from aliases but magic in the hardware deals
	 * with that for us so we don't need to take care ourselves.
	 */
	switch (current_cpu_type()) {
	case CPU_20KC:
	case CPU_25KF:
	case CPU_I6400:
	case CPU_I6500:
	case CPU_SB1:
	case CPU_SB1A:
		c->dcache.flags |= MIPS_CACHE_PINDEX;
		break;

	case CPU_R10000:
	case CPU_R12000:
	case CPU_R14000:
	case CPU_R16000:
		break;

	case CPU_74K:
	case CPU_1074K:
		has_74k_erratum = alias_74k_erratum(c);
		fallthrough;
	case CPU_M14KC:
	case CPU_M14KEC:
	case CPU_24K:
	case CPU_34K:
	case CPU_1004K:
	case CPU_INTERAPTIV:
	case CPU_P5600:
	case CPU_PROAPTIV:
	case CPU_M5150:
	case CPU_QEMU_GENERIC:
	case CPU_P6600:
	case CPU_M6250:
		if (!(read_c0_config7() & MIPS_CONF7_IAR) &&
		    (c->icache.waysize > PAGE_SIZE))
			c->icache.flags |= MIPS_CACHE_ALIASES;
		if (!has_74k_erratum && (read_c0_config7() & MIPS_CONF7_AR)) {
			/*
			 * Effectively physically indexed dcache,
			 * thus no virtual aliases.
			*/
			c->dcache.flags |= MIPS_CACHE_PINDEX;
			break;
		}
		fallthrough;
	default:
		if (has_74k_erratum || c->dcache.waysize > PAGE_SIZE)
			c->dcache.flags |= MIPS_CACHE_ALIASES;
	}

	/* Physically indexed caches don't suffer from virtual aliasing */
	if (c->dcache.flags & MIPS_CACHE_PINDEX)
		c->dcache.flags &= ~MIPS_CACHE_ALIASES;

	/*
	 * In systems with CM the icache fills from L2 or closer caches, and
	 * thus sees remote stores without needing to write them back any
	 * further than that.
	 */
	if (mips_cm_present())
		c->icache.flags |= MIPS_IC_SNOOPS_REMOTE;

	switch (current_cpu_type()) {
	case CPU_20KC:
		/*
		 * Some older 20Kc chips doesn't have the 'VI' bit in
		 * the config register.
		 */
		c->icache.flags |= MIPS_CACHE_VTAG;
		break;

	case CPU_ALCHEMY:
	case CPU_I6400:
	case CPU_I6500:
		c->icache.flags |= MIPS_CACHE_IC_F_DC;
		break;

	case CPU_BMIPS5000:
		c->icache.flags |= MIPS_CACHE_IC_F_DC;
		/* Cache aliases are handled in hardware; allow HIGHMEM */
		c->dcache.flags &= ~MIPS_CACHE_ALIASES;
		break;

	case CPU_LOONGSON2EF:
		/*
		 * LOONGSON2 has 4 way icache, but when using indexed cache op,
		 * one op will act on all 4 ways
		 */
		c->icache.ways = 1;
	}

	pr_info("Primary instruction cache %ldkB, %s, %s, linesize %d bytes.\n",
		icache_size >> 10,
		c->icache.flags & MIPS_CACHE_VTAG ? "VIVT" : "VIPT",
		way_string[c->icache.ways], c->icache.linesz);

	pr_info("Primary data cache %ldkB, %s, %s, %s, linesize %d bytes\n",
		dcache_size >> 10, way_string[c->dcache.ways],
		(c->dcache.flags & MIPS_CACHE_PINDEX) ? "PIPT" : "VIPT",
		(c->dcache.flags & MIPS_CACHE_ALIASES) ?
			"cache aliases" : "no aliases",
		c->dcache.linesz);
}

static void probe_vcache(void)
{
	struct cpuinfo_mips *c = &current_cpu_data;
	unsigned int config2, lsize;

	if (current_cpu_type() != CPU_LOONGSON64)
		return;

	config2 = read_c0_config2();
	if ((lsize = ((config2 >> 20) & 15)))
		c->vcache.linesz = 2 << lsize;
	else
		c->vcache.linesz = lsize;

	c->vcache.sets = 64 << ((config2 >> 24) & 15);
	c->vcache.ways = 1 + ((config2 >> 16) & 15);

	vcache_size = c->vcache.sets * c->vcache.ways * c->vcache.linesz;

	c->vcache.waybit = 0;
	c->vcache.waysize = vcache_size / c->vcache.ways;

	pr_info("Unified victim cache %ldkB %s, linesize %d bytes.\n",
		vcache_size >> 10, way_string[c->vcache.ways], c->vcache.linesz);
}

/*
 * If you even _breathe_ on this function, look at the gcc output and make sure
 * it does not pop things on and off the stack for the cache sizing loop that
 * executes in KSEG1 space or else you will crash and burn badly.  You have
 * been warned.
 */
static int probe_scache(void)
{
	unsigned long flags, addr, begin, end, pow2;
	unsigned int config = read_c0_config();
	struct cpuinfo_mips *c = &current_cpu_data;

	if (config & CONF_SC)
		return 0;

	begin = (unsigned long) &_stext;
	begin &= ~((4 * 1024 * 1024) - 1);
	end = begin + (4 * 1024 * 1024);

	/*
	 * This is such a bitch, you'd think they would make it easy to do
	 * this.  Away you daemons of stupidity!
	 */
	local_irq_save(flags);

	/* Fill each size-multiple cache line with a valid tag. */
	pow2 = (64 * 1024);
	for (addr = begin; addr < end; addr = (begin + pow2)) {
		unsigned long *p = (unsigned long *) addr;
		__asm__ __volatile__("nop" : : "r" (*p)); /* whee... */
		pow2 <<= 1;
	}

	/* Load first line with zero (therefore invalid) tag. */
	write_c0_taglo(0);
	write_c0_taghi(0);
	__asm__ __volatile__("nop; nop; nop; nop;"); /* avoid the hazard */
	cache_op(Index_Store_Tag_I, begin);
	cache_op(Index_Store_Tag_D, begin);
	cache_op(Index_Store_Tag_SD, begin);

	/* Now search for the wrap around point. */
	pow2 = (128 * 1024);
	for (addr = begin + (128 * 1024); addr < end; addr = begin + pow2) {
		cache_op(Index_Load_Tag_SD, addr);
		__asm__ __volatile__("nop; nop; nop; nop;"); /* hazard... */
		if (!read_c0_taglo())
			break;
		pow2 <<= 1;
	}
	local_irq_restore(flags);
	addr -= begin;

	scache_size = addr;
	c->scache.linesz = 16 << ((config & R4K_CONF_SB) >> 22);
	c->scache.ways = 1;
	c->scache.waybit = 0;		/* does not matter */

	return 1;
}

static void loongson2_sc_init(void)
{
	struct cpuinfo_mips *c = &current_cpu_data;

	scache_size = 512*1024;
	c->scache.linesz = 32;
	c->scache.ways = 4;
	c->scache.waybit = 0;
	c->scache.waysize = scache_size / (c->scache.ways);
	c->scache.sets = scache_size / (c->scache.linesz * c->scache.ways);
	pr_info("Unified secondary cache %ldkB %s, linesize %d bytes.\n",
	       scache_size >> 10, way_string[c->scache.ways], c->scache.linesz);

	c->options |= MIPS_CPU_INCLUSIVE_CACHES;
}

static void loongson3_sc_init(void)
{
	struct cpuinfo_mips *c = &current_cpu_data;
	unsigned int config2, lsize;

	config2 = read_c0_config2();
	lsize = (config2 >> 4) & 15;
	if (lsize)
		c->scache.linesz = 2 << lsize;
	else
		c->scache.linesz = 0;
	c->scache.sets = 64 << ((config2 >> 8) & 15);
	c->scache.ways = 1 + (config2 & 15);

	/* Loongson-3 has 4-Scache banks, while Loongson-2K have only 2 banks */
	if ((c->processor_id & PRID_IMP_MASK) == PRID_IMP_LOONGSON_64R)
		c->scache.sets *= 2;
	else
		c->scache.sets *= 4;

	scache_size = c->scache.sets * c->scache.ways * c->scache.linesz;

	c->scache.waybit = 0;
	c->scache.waysize = scache_size / c->scache.ways;
	pr_info("Unified secondary cache %ldkB %s, linesize %d bytes.\n",
	       scache_size >> 10, way_string[c->scache.ways], c->scache.linesz);
	if (scache_size)
		c->options |= MIPS_CPU_INCLUSIVE_CACHES;
	return;
}

extern int r5k_sc_init(void);
extern int rm7k_sc_init(void);
extern int mips_sc_init(void);

static void setup_scache(void)
{
	struct cpuinfo_mips *c = &current_cpu_data;
	unsigned int config = read_c0_config();
	int sc_present = 0;

	/*
	 * Do the probing thing on R4000SC and R4400SC processors.  Other
	 * processors don't have a S-cache that would be relevant to the
	 * Linux memory management.
	 */
	switch (current_cpu_type()) {
	case CPU_R4000SC:
	case CPU_R4000MC:
	case CPU_R4400SC:
	case CPU_R4400MC:
		sc_present = run_uncached(probe_scache);
		if (sc_present)
			c->options |= MIPS_CPU_CACHE_CDEX_S;
		break;

	case CPU_R10000:
	case CPU_R12000:
	case CPU_R14000:
	case CPU_R16000:
		scache_size = 0x80000 << ((config & R10K_CONF_SS) >> 16);
		c->scache.linesz = 64 << ((config >> 13) & 1);
		c->scache.ways = 2;
		c->scache.waybit= 0;
		sc_present = 1;
		break;

	case CPU_R5000:
	case CPU_NEVADA:
#ifdef CONFIG_R5000_CPU_SCACHE
		r5k_sc_init();
#endif
		return;

	case CPU_RM7000:
#ifdef CONFIG_RM7000_CPU_SCACHE
		rm7k_sc_init();
#endif
		return;

	case CPU_LOONGSON2EF:
		loongson2_sc_init();
		return;

	case CPU_LOONGSON64:
		loongson3_sc_init();
		return;

	case CPU_CAVIUM_OCTEON3:
		/* don't need to worry about L2, fully coherent */
		return;

	default:
		if (c->isa_level & (MIPS_CPU_ISA_M32R1 | MIPS_CPU_ISA_M64R1 |
				    MIPS_CPU_ISA_M32R2 | MIPS_CPU_ISA_M64R2 |
				    MIPS_CPU_ISA_M32R5 | MIPS_CPU_ISA_M64R5 |
				    MIPS_CPU_ISA_M32R6 | MIPS_CPU_ISA_M64R6)) {
#ifdef CONFIG_MIPS_CPU_SCACHE
			if (mips_sc_init ()) {
				scache_size = c->scache.ways * c->scache.sets * c->scache.linesz;
				printk("MIPS secondary cache %ldkB, %s, linesize %d bytes.\n",
				       scache_size >> 10,
				       way_string[c->scache.ways], c->scache.linesz);

				if (current_cpu_type() == CPU_BMIPS5000)
					c->options |= MIPS_CPU_INCLUSIVE_CACHES;
			}

#else
			if (!(c->scache.flags & MIPS_CACHE_NOT_PRESENT))
				panic("Dunno how to handle MIPS32 / MIPS64 second level cache");
#endif
			return;
		}
		sc_present = 0;
	}

	if (!sc_present)
		return;

	/* compute a couple of other cache variables */
	c->scache.waysize = scache_size / c->scache.ways;

	c->scache.sets = scache_size / (c->scache.linesz * c->scache.ways);

	printk("Unified secondary cache %ldkB %s, linesize %d bytes.\n",
	       scache_size >> 10, way_string[c->scache.ways], c->scache.linesz);

	c->options |= MIPS_CPU_INCLUSIVE_CACHES;
}

void au1x00_fixup_config_od(void)
{
	/*
	 * c0_config.od (bit 19) was write only (and read as 0)
	 * on the early revisions of Alchemy SOCs.  It disables the bus
	 * transaction overlapping and needs to be set to fix various errata.
	 */
	switch (read_c0_prid()) {
	case 0x00030100: /* Au1000 DA */
	case 0x00030201: /* Au1000 HA */
	case 0x00030202: /* Au1000 HB */
	case 0x01030200: /* Au1500 AB */
	/*
	 * Au1100 errata actually keeps silence about this bit, so we set it
	 * just in case for those revisions that require it to be set according
	 * to the (now gone) cpu table.
	 */
	case 0x02030200: /* Au1100 AB */
	case 0x02030201: /* Au1100 BA */
	case 0x02030202: /* Au1100 BC */
		set_c0_config(1 << 19);
		break;
	}
}

/* CP0 hazard avoidance. */
#define NXP_BARRIER()							\
	 __asm__ __volatile__(						\
	".set noreorder\n\t"						\
	"nop; nop; nop; nop; nop; nop;\n\t"				\
	".set reorder\n\t")

static void nxp_pr4450_fixup_config(void)
{
	unsigned long config0;

	config0 = read_c0_config();

	/* clear all three cache coherency fields */
	config0 &= ~(0x7 | (7 << 25) | (7 << 28));
	config0 |= (((_page_cachable_default >> _CACHE_SHIFT) <<  0) |
		    ((_page_cachable_default >> _CACHE_SHIFT) << 25) |
		    ((_page_cachable_default >> _CACHE_SHIFT) << 28));
	write_c0_config(config0);
	NXP_BARRIER();
}

static int cca = -1;

static int __init cca_setup(char *str)
{
	get_option(&str, &cca);

	return 0;
}

early_param("cca", cca_setup);

static void coherency_setup(void)
{
	if (cca < 0 || cca > 7)
		cca = read_c0_config() & CONF_CM_CMASK;
	_page_cachable_default = cca << _CACHE_SHIFT;

	pr_debug("Using cache attribute %d\n", cca);
	change_c0_config(CONF_CM_CMASK, cca);

	/*
	 * c0_status.cu=0 specifies that updates by the sc instruction use
	 * the coherency mode specified by the TLB; 1 means cachable
	 * coherent update on write will be used.  Not all processors have
	 * this bit and; some wire it to zero, others like Toshiba had the
	 * silly idea of putting something else there ...
	 */
	switch (current_cpu_type()) {
	case CPU_R4000PC:
	case CPU_R4000SC:
	case CPU_R4000MC:
	case CPU_R4400PC:
	case CPU_R4400SC:
	case CPU_R4400MC:
		clear_c0_config(CONF_CU);
		break;
	/*
	 * We need to catch the early Alchemy SOCs with
	 * the write-only co_config.od bit and set it back to one on:
	 * Au1000 rev DA, HA, HB;  Au1100 AB, BA, BC, Au1500 AB
	 */
	case CPU_ALCHEMY:
		au1x00_fixup_config_od();
		break;

	case PRID_IMP_PR4450:
		nxp_pr4450_fixup_config();
		break;
	}
}

static void r4k_cache_error_setup(void)
{
	extern char __weak except_vec2_generic;
	extern char __weak except_vec2_sb1;

	switch (current_cpu_type()) {
	case CPU_SB1:
	case CPU_SB1A:
		set_uncached_handler(0x100, &except_vec2_sb1, 0x80);
		break;

	default:
		set_uncached_handler(0x100, &except_vec2_generic, 0x80);
		break;
	}
}

void r4k_cache_init(void)
{
	extern void build_clear_page(void);
	extern void build_copy_page(void);
	struct cpuinfo_mips *c = &current_cpu_data;

	probe_pcache();
	probe_vcache();
	setup_scache();

	r4k_blast_dcache_page_setup();
	r4k_blast_dcache_setup();
	r4k_blast_icache_page_setup();
	r4k_blast_icache_setup();
	r4k_blast_scache_page_setup();
	r4k_blast_scache_setup();
	r4k_blast_scache_node_setup();
#ifdef CONFIG_EVA
	r4k_blast_dcache_user_page_setup();
	r4k_blast_icache_user_page_setup();
#endif

	/*
	 * Some MIPS32 and MIPS64 processors have physically indexed caches.
	 * This code supports virtually indexed processors and will be
	 * unnecessarily inefficient on physically indexed processors.
	 */
	if (c->dcache.linesz && cpu_has_dc_aliases)
		shm_align_mask = max_t( unsigned long,
					c->dcache.sets * c->dcache.linesz - 1,
					PAGE_SIZE - 1);
	else
		shm_align_mask = PAGE_SIZE-1;

	__flush_cache_vmap	= r4k__flush_cache_vmap;
	__flush_cache_vunmap	= r4k__flush_cache_vunmap;

	flush_cache_all		= cache_noop;
	__flush_cache_all	= r4k___flush_cache_all;
	flush_cache_mm		= r4k_flush_cache_mm;
	flush_cache_page	= r4k_flush_cache_page;
	flush_cache_range	= r4k_flush_cache_range;

	__flush_kernel_vmap_range = r4k_flush_kernel_vmap_range;

	flush_icache_all	= r4k_flush_icache_all;
	flush_data_cache_page	= r4k_flush_data_cache_page;
	flush_icache_range	= r4k_flush_icache_range;
	local_flush_icache_range	= local_r4k_flush_icache_range;
	__flush_icache_user_range	= r4k_flush_icache_user_range;
	__local_flush_icache_user_range	= local_r4k_flush_icache_user_range;

#ifdef CONFIG_DMA_NONCOHERENT
	_dma_cache_wback_inv	= r4k_dma_cache_wback_inv;
	_dma_cache_wback	= r4k_dma_cache_wback_inv;
	_dma_cache_inv		= r4k_dma_cache_inv;
#endif /* CONFIG_DMA_NONCOHERENT */

	build_clear_page();
	build_copy_page();

	/*
	 * We want to run CMP kernels on core with and without coherent
	 * caches. Therefore, do not use CONFIG_MIPS_CMP to decide whether
	 * or not to flush caches.
	 */
	local_r4k___flush_cache_all(NULL);

	coherency_setup();
	board_cache_error_setup = r4k_cache_error_setup;

	/*
	 * Per-CPU overrides
	 */
	switch (current_cpu_type()) {
	case CPU_BMIPS4350:
	case CPU_BMIPS4380:
		/* No IPI is needed because all CPUs share the same D$ */
		flush_data_cache_page = r4k_blast_dcache_page;
		break;
	case CPU_BMIPS5000:
		/* We lose our superpowers if L2 is disabled */
		if (c->scache.flags & MIPS_CACHE_NOT_PRESENT)
			break;

		/* I$ fills from D$ just by emptying the write buffers */
		flush_cache_page = (void *)b5k_instruction_hazard;
		flush_cache_range = (void *)b5k_instruction_hazard;
		flush_data_cache_page = (void *)b5k_instruction_hazard;
		flush_icache_range = (void *)b5k_instruction_hazard;
		local_flush_icache_range = (void *)b5k_instruction_hazard;


		/* Optimization: an L2 flush implicitly flushes the L1 */
		current_cpu_data.options |= MIPS_CPU_INCLUSIVE_CACHES;
		break;
	case CPU_LOONGSON64:
		/* Loongson-3 maintains cache coherency by hardware */
		__flush_cache_all	= cache_noop;
		__flush_cache_vmap	= cache_noop;
		__flush_cache_vunmap	= cache_noop;
		__flush_kernel_vmap_range = (void *)cache_noop;
		flush_cache_mm		= (void *)cache_noop;
		flush_cache_page	= (void *)cache_noop;
		flush_cache_range	= (void *)cache_noop;
		flush_icache_all	= (void *)cache_noop;
		flush_data_cache_page	= (void *)cache_noop;
		break;
	}
}

static int r4k_cache_pm_notifier(struct notifier_block *self, unsigned long cmd,
			       void *v)
{
	switch (cmd) {
	case CPU_PM_ENTER_FAILED:
	case CPU_PM_EXIT:
		coherency_setup();
		break;
	}

	return NOTIFY_OK;
}

static struct notifier_block r4k_cache_pm_notifier_block = {
	.notifier_call = r4k_cache_pm_notifier,
};

int __init r4k_cache_init_pm(void)
{
	return cpu_pm_register_notifier(&r4k_cache_pm_notifier_block);
}
arch_initcall(r4k_cache_init_pm);