arch/arm/mm/copypage-v4mc.c

   1 /*
   2  *  linux/arch/arm/lib/copypage-armv4mc.S
   3  *
   4  *  Copyright (C) 1995-2005 Russell King
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  *
  10  * This handles the mini data cache, as found on SA11x0 and XScale
  11  * processors.  When we copy a user page page, we map it in such a way
  12  * that accesses to this page will not touch the main data cache, but
  13  * will be cached in the mini data cache.  This prevents us thrashing
  14  * the main data cache on page faults.
  15  */
  16 #include <linux/init.h>
  17 #include <linux/mm.h>
  18 #include <linux/highmem.h>
  19
  20 #include <asm/pgtable.h>
  21 #include <asm/tlbflush.h>
  22 #include <asm/cacheflush.h>
  23
  24 #include "mm.h"
  25
  26 /*
  27  * 0xffff8000 to 0xffffffff is reserved for any ARM architecture
  28  * specific hacks for copying pages efficiently.
  29  */
  30 #define minicache_pgprot __pgprot(L_PTE_PRESENT | L_PTE_YOUNG | \
  31                                   L_PTE_MT_MINICACHE)
  32
  33 static DEFINE_RAW_SPINLOCK(minicache_lock);
  34
  35 /*
  36  * ARMv4 mini-dcache optimised copy_user_highpage
  37  *
  38  * We flush the destination cache lines just before we write the data into the
  39  * corresponding address.  Since the Dcache is read-allocate, this removes the
  40  * Dcache aliasing issue.  The writes will be forwarded to the write buffer,
  41  * and merged as appropriate.
  42  *
  43  * Note: We rely on all ARMv4 processors implementing the "invalidate D line"
  44  * instruction.  If your processor does not supply this, you have to write your
  45  * own copy_user_highpage that does the right thing.
  46  */
  47 static void __naked
  48 mc_copy_user_page(void *from, void *to)
  49 {
  50         asm volatile(
  51         "stmfd  sp!, {r4, lr}                   @ 2\n\
  52         mov     r4, %2                          @ 1\n\
  53         ldmia   %0!, {r2, r3, ip, lr}           @ 4\n\
  54 1:      mcr     p15, 0, %1, c7, c6, 1           @ 1   invalidate D line\n\
  55         stmia   %1!, {r2, r3, ip, lr}           @ 4\n\
  56         ldmia   %0!, {r2, r3, ip, lr}           @ 4+1\n\
  57         stmia   %1!, {r2, r3, ip, lr}           @ 4\n\
  58         ldmia   %0!, {r2, r3, ip, lr}           @ 4\n\
  59         mcr     p15, 0, %1, c7, c6, 1           @ 1   invalidate D line\n\
  60         stmia   %1!, {r2, r3, ip, lr}           @ 4\n\
  61         ldmia   %0!, {r2, r3, ip, lr}           @ 4\n\
  62         subs    r4, r4, #1                      @ 1\n\
  63         stmia   %1!, {r2, r3, ip, lr}           @ 4\n\
  64         ldmneia %0!, {r2, r3, ip, lr}           @ 4\n\
  65         bne     1b                              @ 1\n\
  66         ldmfd   sp!, {r4, pc}                   @ 3"
  67         :
  68         : "r" (from), "r" (to), "I" (PAGE_SIZE / 64));
  69 }
  70
  71 void v4_mc_copy_user_highpage(struct page *to, struct page *from,
  72         unsigned long vaddr, struct vm_area_struct *vma)
  73 {
  74         void *kto = kmap_atomic(to, KM_USER1);
  75
  76         if (!test_and_set_bit(PG_dcache_clean, &from->flags))
  77                 __flush_dcache_page(page_mapping(from), from);
  78
  79         raw_spin_lock(&minicache_lock);
  80
  81         set_pte_ext(TOP_PTE(0xffff8000), pfn_pte(page_to_pfn(from), minicache_pgprot), 0);
  82         flush_tlb_kernel_page(0xffff8000);
  83
  84         mc_copy_user_page((void *)0xffff8000, kto);
  85
  86         raw_spin_unlock(&minicache_lock);
  87
  88         kunmap_atomic(kto, KM_USER1);
  89 }
  90
  91 /*
  92  * ARMv4 optimised clear_user_page
  93  */
  94 void v4_mc_clear_user_highpage(struct page *page, unsigned long vaddr)
  95 {
  96         void *ptr, *kaddr = kmap_atomic(page, KM_USER0);
  97         asm volatile("\
  98         mov     r1, %2                          @ 1\n\
  99         mov     r2, #0                          @ 1\n\
 100         mov     r3, #0                          @ 1\n\
 101         mov     ip, #0                          @ 1\n\
 102         mov     lr, #0                          @ 1\n\
 103 1:      mcr     p15, 0, %0, c7, c6, 1           @ 1   invalidate D line\n\
 104         stmia   %0!, {r2, r3, ip, lr}           @ 4\n\
 105         stmia   %0!, {r2, r3, ip, lr}           @ 4\n\
 106         mcr     p15, 0, %0, c7, c6, 1           @ 1   invalidate D line\n\
 107         stmia   %0!, {r2, r3, ip, lr}           @ 4\n\
 108         stmia   %0!, {r2, r3, ip, lr}           @ 4\n\
 109         subs    r1, r1, #1                      @ 1\n\
 110         bne     1b                              @ 1"
 111         : "=r" (ptr)
 112         : "0" (kaddr), "I" (PAGE_SIZE / 64)
 113         : "r1", "r2", "r3", "ip", "lr");
 114         kunmap_atomic(kaddr, KM_USER0);
 115 }
 116
 117 struct cpu_user_fns v4_mc_user_fns __initdata = {
 118         .cpu_clear_user_highpage = v4_mc_clear_user_highpage,
 119         .cpu_copy_user_highpage = v4_mc_copy_user_highpage,
 120 };