arch/ia64/lib/flush.S

   1 /*
   2  * Cache flushing routines.
   3  *
   4  * Copyright (C) 1999-2001, 2005 Hewlett-Packard Co
   5  *      David Mosberger-Tang <davidm@hpl.hp.com>
   6  *
   7  * 05/28/05 Zoltan Menyhart     Dynamic stride size
   8  */
   9
  10 #include <asm/asmmacro.h>
  11 #include <asm/export.h>
  12
  13
  14         /*
  15          * flush_icache_range(start,end)
  16          *
  17          *      Make i-cache(s) coherent with d-caches.
  18          *
  19          *      Must deal with range from start to end-1 but nothing else (need to
  20          *      be careful not to touch addresses that may be unmapped).
  21          *
  22          *      Note: "in0" and "in1" are preserved for debugging purposes.
  23          */
  24         .section .kprobes.text,"ax"
  25 GLOBAL_ENTRY(flush_icache_range)
  26
  27         .prologue
  28         alloc   r2=ar.pfs,2,0,0,0
  29         movl    r3=ia64_i_cache_stride_shift
  30         mov     r21=1
  31         ;;
  32         ld8     r20=[r3]                // r20: stride shift
  33         sub     r22=in1,r0,1            // last byte address
  34         ;;
  35         shr.u   r23=in0,r20             // start / (stride size)
  36         shr.u   r22=r22,r20             // (last byte address) / (stride size)
  37         shl     r21=r21,r20             // r21: stride size of the i-cache(s)
  38         ;;
  39         sub     r8=r22,r23              // number of strides - 1
  40         shl     r24=r23,r20             // r24: addresses for "fc.i" =
  41                                         //      "start" rounded down to stride boundary
  42         .save   ar.lc,r3
  43         mov     r3=ar.lc                // save ar.lc
  44         ;;
  45
  46         .body
  47         mov     ar.lc=r8
  48         ;;
  49         /*
  50          * 32 byte aligned loop, even number of (actually 2) bundles
  51          */
  52 .Loop:  fc.i    r24                     // issuable on M0 only
  53         add     r24=r21,r24             // we flush "stride size" bytes per iteration
  54         nop.i   0
  55         br.cloop.sptk.few .Loop
  56         ;;
  57         sync.i
  58         ;;
  59         srlz.i
  60         ;;
  61         mov     ar.lc=r3                // restore ar.lc
  62         br.ret.sptk.many rp
  63 END(flush_icache_range)
  64 EXPORT_SYMBOL_GPL(flush_icache_range)
  65
  66         /*
  67          * clflush_cache_range(start,size)
  68          *
  69          *      Flush cache lines from start to start+size-1.
  70          *
  71          *      Must deal with range from start to start+size-1 but nothing else
  72          *      (need to be careful not to touch addresses that may be
  73          *      unmapped).
  74          *
  75          *      Note: "in0" and "in1" are preserved for debugging purposes.
  76          */
  77         .section .kprobes.text,"ax"
  78 GLOBAL_ENTRY(clflush_cache_range)
  79
  80         .prologue
  81         alloc   r2=ar.pfs,2,0,0,0
  82         movl    r3=ia64_cache_stride_shift
  83         mov     r21=1
  84         add     r22=in1,in0
  85         ;;
  86         ld8     r20=[r3]                // r20: stride shift
  87         sub     r22=r22,r0,1            // last byte address
  88         ;;
  89         shr.u   r23=in0,r20             // start / (stride size)
  90         shr.u   r22=r22,r20             // (last byte address) / (stride size)
  91         shl     r21=r21,r20             // r21: stride size of the i-cache(s)
  92         ;;
  93         sub     r8=r22,r23              // number of strides - 1
  94         shl     r24=r23,r20             // r24: addresses for "fc" =
  95                                         //      "start" rounded down to stride
  96                                         //      boundary
  97         .save   ar.lc,r3
  98         mov     r3=ar.lc                // save ar.lc
  99         ;;
 100
 101         .body
 102         mov     ar.lc=r8
 103         ;;
 104         /*
 105          * 32 byte aligned loop, even number of (actually 2) bundles
 106          */
 107 .Loop_fc:
 108         fc      r24             // issuable on M0 only
 109         add     r24=r21,r24     // we flush "stride size" bytes per iteration
 110         nop.i   0
 111         br.cloop.sptk.few .Loop_fc
 112         ;;
 113         sync.i
 114         ;;
 115         srlz.i
 116         ;;
 117         mov     ar.lc=r3                // restore ar.lc
 118         br.ret.sptk.many rp
 119 END(clflush_cache_range)