proc-xscale.s

linux 内核源代码

/*
 *  linux/arch/arm/mm/proc-xscale.S
 *
 *  Author:	Nicolas Pitre
 *  Created:	November 2000
 *  Copyright:	(C) 2000, 2001 MontaVista Software Inc.
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation.
 *
 * MMU functions for the Intel XScale CPUs
 *
 * 2001 Aug 21:
 *	some contributions by Brett Gaines <[e-mail]>
 *	Copyright 2001 by Intel Corp.
 *
 * 2001 Sep 08:
 *	Completely revisited, many important fixes
 *	Nicolas Pitre <[e-mail]>
 */

#include <linux/linkage.h>
#include <linux/init.h>
#include <asm/assembler.h>
#include <asm/elf.h>
#include <asm/pgtable.h>
#include <asm/pgtable-hwdef.h>
#include <asm/page.h>
#include <asm/ptrace.h>
#include "proc-macros.S"

/*
 * This is the maximum size of an area which will be flushed.  If the area
 * is larger than this, then we flush the whole cache
 */
#define MAX_AREA_SIZE	32768

/*
 * the cache line size of the I and D cache
 */
#define CACHELINESIZE	32

/*
 * the size of the data cache
 */
#define CACHESIZE	32768

/*
 * Virtual address used to allocate the cache when flushed
 *
 * This must be an address range which is _never_ used.  It should
 * apparently have a mapping in the corresponding page table for
 * compatibility with future CPUs that _could_ require it.  For instance we
 * don't care.
 *
 * This must be aligned on a 2*CACHESIZE boundary.  The code selects one of
 * the 2 areas in alternance each time the clean_d_cache macro is used.
 * Without this the XScale core exhibits cache eviction problems and no one
 * knows why.
 *
 * Reminder: the vector table is located at 0xffff0000-0xffff0fff.
 */
#define CLEAN_ADDR	0xfffe0000

/*
 * This macro is used to wait for a CP15 write and is needed
 * when we have to ensure that the last operation to the co-pro
 * was completed before continuing with operation.
 */
	.macro	cpwait, rd
	mrc	p15, 0, \rd, c2, c0, 0		@ arbitrary read of cp15
	mov	\rd, \rd			@ wait for completion
	sub 	pc, pc, #4			@ flush instruction pipeline
	.endm

	.macro	cpwait_ret, lr, rd
	mrc	p15, 0, \rd, c2, c0, 0		@ arbitrary read of cp15
	sub	pc, \lr, \rd, LSR #32		@ wait for completion and
						@ flush instruction pipeline
	.endm

/*
 * This macro cleans the entire dcache using line allocate.
 * The main loop has been unrolled to reduce loop overhead.
 * rd and rs are two scratch registers.
 */
	.macro  clean_d_cache, rd, rs
	ldr	\rs, =clean_addr
	ldr	\rd, [\rs]
	eor	\rd, \rd, #CACHESIZE
	str	\rd, [\rs]
	add	\rs, \rd, #CACHESIZE
1:	mcr	p15, 0, \rd, c7, c2, 5		@ allocate D cache line
	add	\rd, \rd, #CACHELINESIZE
	mcr	p15, 0, \rd, c7, c2, 5		@ allocate D cache line
	add	\rd, \rd, #CACHELINESIZE
	mcr	p15, 0, \rd, c7, c2, 5		@ allocate D cache line
	add	\rd, \rd, #CACHELINESIZE
	mcr	p15, 0, \rd, c7, c2, 5		@ allocate D cache line
	add	\rd, \rd, #CACHELINESIZE
	teq	\rd, \rs
	bne	1b
	.endm

	.data
clean_addr:	.word	CLEAN_ADDR

	.text

/*
 * cpu_xscale_proc_init()
 *
 * Nothing too exciting at the moment
 */
ENTRY(cpu_xscale_proc_init)
	mov	pc, lr

/*
 * cpu_xscale_proc_fin()
 */
ENTRY(cpu_xscale_proc_fin)
	str	lr, [sp, #-4]!
	mov	r0, #PSR_F_BIT|PSR_I_BIT|SVC_MODE
	msr	cpsr_c, r0
	bl	xscale_flush_kern_cache_all	@ clean caches
	mrc	p15, 0, r0, c1, c0, 0		@ ctrl register
	bic	r0, r0, #0x1800			@ ...IZ...........
	bic	r0, r0, #0x0006			@ .............CA.
	mcr	p15, 0, r0, c1, c0, 0		@ disable caches
	ldr	pc, [sp], #4

/*
 * cpu_xscale_reset(loc)
 *
 * Perform a soft reset of the system.  Put the CPU into the
 * same state as it would be if it had been reset, and branch
 * to what would be the reset vector.
 *
 * loc: location to jump to for soft reset
 *
 * Beware PXA270 erratum E7.
 */
	.align	5
ENTRY(cpu_xscale_reset)
	mov	r1, #PSR_F_BIT|PSR_I_BIT|SVC_MODE
	msr	cpsr_c, r1			@ reset CPSR
	mcr	p15, 0, r1, c10, c4, 1		@ unlock I-TLB
	mcr	p15, 0, r1, c8, c5, 0		@ invalidate I-TLB
	mrc	p15, 0, r1, c1, c0, 0		@ ctrl register
	bic	r1, r1, #0x0086			@ ........B....CA.
	bic	r1, r1, #0x3900			@ ..VIZ..S........
	sub	pc, pc, #4			@ flush pipeline
	@ *** cache line aligned ***
	mcr	p15, 0, r1, c1, c0, 0		@ ctrl register
	bic	r1, r1, #0x0001			@ ...............M
	mcr	p15, 0, ip, c7, c7, 0		@ invalidate I,D caches & BTB
	mcr	p15, 0, r1, c1, c0, 0		@ ctrl register
	@ CAUTION: MMU turned off from this point. We count on the pipeline
	@ already containing those two last instructions to survive.
	mcr	p15, 0, ip, c8, c7, 0		@ invalidate I & D TLBs
	mov	pc, r0

/*
 * cpu_xscale_do_idle()
 *
 * Cause the processor to idle
 *
 * For now we do nothing but go to idle mode for every case
 *
 * XScale supports clock switching, but using idle mode support
 * allows external hardware to react to system state changes.
 */
	.align	5

ENTRY(cpu_xscale_do_idle)
	mov	r0, #1
	mcr	p14, 0, r0, c7, c0, 0		@ Go to IDLE
	mov	pc, lr

/* ================================= CACHE ================================ */

/*
 *	flush_user_cache_all()
 *
 *	Invalidate all cache entries in a particular address
 *	space.
 */
ENTRY(xscale_flush_user_cache_all)
	/* FALLTHROUGH */

/*
 *	flush_kern_cache_all()
 *
 *	Clean and invalidate the entire cache.
 */
ENTRY(xscale_flush_kern_cache_all)
	mov	r2, #VM_EXEC
	mov	ip, #0
__flush_whole_cache:
	clean_d_cache r0, r1
	tst	r2, #VM_EXEC
	mcrne	p15, 0, ip, c7, c5, 0		@ Invalidate I cache & BTB
	mcrne	p15, 0, ip, c7, c10, 4		@ Drain Write (& Fill) Buffer
	mov	pc, lr

/*
 *	flush_user_cache_range(start, end, vm_flags)
 *
 *	Invalidate a range of cache entries in the specified
 *	address space.
 *
 *	- start - start address (may not be aligned)
 *	- end	- end address (exclusive, may not be aligned)
 *	- vma	- vma_area_struct describing address space
 */
	.align	5
ENTRY(xscale_flush_user_cache_range)
	mov	ip, #0
	sub	r3, r1, r0			@ calculate total size
	cmp	r3, #MAX_AREA_SIZE
	bhs	__flush_whole_cache

1:	tst	r2, #VM_EXEC
	mcrne	p15, 0, r0, c7, c5, 1		@ Invalidate I cache line
	mcr	p15, 0, r0, c7, c10, 1		@ Clean D cache line
	mcr	p15, 0, r0, c7, c6, 1		@ Invalidate D cache line
	add	r0, r0, #CACHELINESIZE
	cmp	r0, r1
	blo	1b
	tst	r2, #VM_EXEC
	mcrne	p15, 0, ip, c7, c5, 6		@ Invalidate BTB
	mcrne	p15, 0, ip, c7, c10, 4		@ Drain Write (& Fill) Buffer
	mov	pc, lr

/*
 *	coherent_kern_range(start, end)
 *
 *	Ensure coherency between the Icache and the Dcache in the
 *	region described by start.  If you have non-snooping
 *	Harvard caches, you need to implement this function.
 *
 *	- start  - virtual start address
 *	- end	 - virtual end address
 *
 *	Note: single I-cache line invalidation isn't used here since
 *	it also trashes the mini I-cache used by JTAG debuggers.
 */
ENTRY(xscale_coherent_kern_range)
	bic	r0, r0, #CACHELINESIZE - 1
1:	mcr	p15, 0, r0, c7, c10, 1		@ clean D entry
	add	r0, r0, #CACHELINESIZE
	cmp	r0, r1
	blo	1b
	mov	r0, #0
	mcr	p15, 0, r0, c7, c5, 0		@ Invalidate I cache & BTB
	mcr	p15, 0, r0, c7, c10, 4		@ Drain Write (& Fill) Buffer
	mov	pc, lr

/*
 *	coherent_user_range(start, end)
 *
 *	Ensure coherency between the Icache and the Dcache in the
 *	region described by start.  If you have non-snooping
 *	Harvard caches, you need to implement this function.
 *
 *	- start  - virtual start address
 *	- end	 - virtual end address
 */
ENTRY(xscale_coherent_user_range)
	bic	r0, r0, #CACHELINESIZE - 1
1:	mcr	p15, 0, r0, c7, c10, 1		@ clean D entry
	mcr	p15, 0, r0, c7, c5, 1		@ Invalidate I cache entry
	add	r0, r0, #CACHELINESIZE
	cmp	r0, r1
	blo	1b
	mov	r0, #0
	mcr	p15, 0, r0, c7, c5, 6		@ Invalidate BTB
	mcr	p15, 0, r0, c7, c10, 4		@ Drain Write (& Fill) Buffer
	mov	pc, lr

/*
 *	flush_kern_dcache_page(void *page)
 *
 *	Ensure no D cache aliasing occurs, either with itself or
 *	the I cache
 *
 *	- addr	- page aligned address
 */
ENTRY(xscale_flush_kern_dcache_page)
	add	r1, r0, #PAGE_SZ
1:	mcr	p15, 0, r0, c7, c10, 1		@ clean D entry
	mcr	p15, 0, r0, c7, c6, 1		@ invalidate D entry
	add	r0, r0, #CACHELINESIZE
	cmp	r0, r1
	blo	1b
	mov	r0, #0
	mcr	p15, 0, r0, c7, c5, 0		@ Invalidate I cache & BTB
	mcr	p15, 0, r0, c7, c10, 4		@ Drain Write (& Fill) Buffer
	mov	pc, lr

/*
 *	dma_inv_range(start, end)
 *
 *	Invalidate (discard) the specified virtual address range.
 *	May not write back any entries.  If 'start' or 'end'
 *	are not cache line aligned, those lines must be written
 *	back.
 *
 *	- start  - virtual start address
 *	- end	 - virtual end address
 */
ENTRY(xscale_dma_inv_range)
	tst	r0, #CACHELINESIZE - 1
	bic	r0, r0, #CACHELINESIZE - 1
	mcrne	p15, 0, r0, c7, c10, 1		@ clean D entry
	tst	r1, #CACHELINESIZE - 1
	mcrne	p15, 0, r1, c7, c10, 1		@ clean D entry
1:	mcr	p15, 0, r0, c7, c6, 1		@ invalidate D entry
	add	r0, r0, #CACHELINESIZE
	cmp	r0, r1
	blo	1b
	mcr	p15, 0, r0, c7, c10, 4		@ Drain Write (& Fill) Buffer
	mov	pc, lr

/*
 *	dma_clean_range(start, end)
 *
 *	Clean the specified virtual address range.
 *
 *	- start  - virtual start address
 *	- end	 - virtual end address
 */
ENTRY(xscale_dma_clean_range)
	bic	r0, r0, #CACHELINESIZE - 1
1:	mcr	p15, 0, r0, c7, c10, 1		@ clean D entry
	add	r0, r0, #CACHELINESIZE
	cmp	r0, r1
	blo	1b
	mcr	p15, 0, r0, c7, c10, 4		@ Drain Write (& Fill) Buffer
	mov	pc, lr

/*
 *	dma_flush_range(start, end)
 *
 *	Clean and invalidate the specified virtual address range.
 *
 *	- start  - virtual start address
 *	- end	 - virtual end address
 */
ENTRY(xscale_dma_flush_range)
	bic	r0, r0, #CACHELINESIZE - 1
1:	mcr	p15, 0, r0, c7, c10, 1		@ clean D entry
	mcr	p15, 0, r0, c7, c6, 1		@ invalidate D entry
	add	r0, r0, #CACHELINESIZE
	cmp	r0, r1
	blo	1b
	mcr	p15, 0, r0, c7, c10, 4		@ Drain Write (& Fill) Buffer
	mov	pc, lr

ENTRY(xscale_cache_fns)
	.long	xscale_flush_kern_cache_all
	.long	xscale_flush_user_cache_all
	.long	xscale_flush_user_cache_range
	.long	xscale_coherent_kern_range
	.long	xscale_coherent_user_range
	.long	xscale_flush_kern_dcache_page
	.long	xscale_dma_inv_range
	.long	xscale_dma_clean_range
	.long	xscale_dma_flush_range

/*
 * On stepping A0/A1 of the 80200, invalidating D-cache by line doesn't
 * clear the dirty bits, which means that if we invalidate a dirty line,
 * the dirty data can still be written back to external memory later on.
 *
 * The recommended workaround is to always do a clean D-cache line before
 * doing an invalidate D-cache line, so on the affected processors,
 * dma_inv_range() is implemented as dma_flush_range().
 *
 * See erratum #25 of "Intel 80200 Processor Specification Update",
 * revision January 22, 2003, available at:
 *     http://www.intel.com/design/iio/specupdt/273415.htm
 */
ENTRY(xscale_80200_A0_A1_cache_fns)
	.long	xscale_flush_kern_cache_all
	.long	xscale_flush_user_cache_all
	.long	xscale_flush_user_cache_range
	.long	xscale_coherent_kern_range
	.long	xscale_coherent_user_range
	.long	xscale_flush_kern_dcache_page
	.long	xscale_dma_flush_range
	.long	xscale_dma_clean_range
	.long	xscale_dma_flush_range

ENTRY(cpu_xscale_dcache_clean_area)
1:	mcr	p15, 0, r0, c7, c10, 1		@ clean D entry
	add	r0, r0, #CACHELINESIZE
	subs	r1, r1, #CACHELINESIZE
	bhi	1b
	mov	pc, lr

/* =============================== PageTable ============================== */

#define PTE_CACHE_WRITE_ALLOCATE 0

/*
 * cpu_xscale_switch_mm(pgd)
 *
 * Set the translation base pointer to be as described by pgd.
 *
 * pgd: new page tables
 */
	.align	5
ENTRY(cpu_xscale_switch_mm)
	clean_d_cache r1, r2
	mcr	p15, 0, ip, c7, c5, 0		@ Invalidate I cache & BTB
	mcr	p15, 0, ip, c7, c10, 4		@ Drain Write (& Fill) Buffer
	mcr	p15, 0, r0, c2, c0, 0		@ load page table pointer
	mcr	p15, 0, ip, c8, c7, 0		@ invalidate I & D TLBs
	cpwait_ret lr, ip

/*
 * cpu_xscale_set_pte_ext(ptep, pte, ext)
 *
 * Set a PTE and flush it out
 *
 * Errata 40: must set memory to write-through for user read-only pages.
 */
	.align	5
ENTRY(cpu_xscale_set_pte_ext)
	str	r1, [r0], #-2048		@ linux version

	bic	r2, r1, #0xff0
	orr	r2, r2, #PTE_TYPE_EXT		@ extended page

	eor	r3, r1, #L_PTE_PRESENT | L_PTE_YOUNG | L_PTE_WRITE | L_PTE_DIRTY

	tst	r3, #L_PTE_USER			@ User?
	orrne	r2, r2, #PTE_EXT_AP_URO_SRW	@ yes -> user r/o, system r/w

	tst	r3, #L_PTE_WRITE | L_PTE_DIRTY	@ Write and Dirty?
	orreq	r2, r2, #PTE_EXT_AP_UNO_SRW	@ yes -> user n/a, system r/w
						@ combined with user -> user r/w

	@
	@ Handle the X bit.  We want to set this bit for the minicache
	@ (U = E = B = W = 0, C = 1) or when write allocate is enabled,
	@ and we have a writeable, cacheable region.  If we ignore the
	@ U and E bits, we can allow user space to use the minicache as
	@ well.
	@
	@  X = (C & ~W & ~B) | (C & W & B & write_allocate)
	@
	eor	ip, r1, #L_PTE_CACHEABLE
	tst	ip, #L_PTE_CACHEABLE | L_PTE_WRITE | L_PTE_BUFFERABLE
#if PTE_CACHE_WRITE_ALLOCATE
	eorne	ip, r1, #L_PTE_CACHEABLE | L_PTE_WRITE | L_PTE_BUFFERABLE
	tstne	ip, #L_PTE_CACHEABLE | L_PTE_WRITE | L_PTE_BUFFERABLE
#endif
	orreq	r2, r2, #PTE_EXT_TEX(1)

	@
	@ Erratum 40: The B bit must be cleared for a user read-only
	@ cacheable page.
	@
	@  B = B & ~(U & C & ~W)
	@
	and	ip, r1, #L_PTE_USER | L_PTE_WRITE | L_PTE_CACHEABLE
	teq	ip, #L_PTE_USER | L_PTE_CACHEABLE
	biceq	r2, r2, #PTE_BUFFERABLE