2
* Cache flushing routines.
4
* Copyright (C) 1999-2001, 2005 Hewlett-Packard Co
5
* David Mosberger-Tang <davidm@hpl.hp.com>
7
* 05/28/05 Zoltan Menyhart Dynamic stride size
10
#include <asm/asmmacro.h>
14
* flush_icache_range(start,end)
16
* Make i-cache(s) coherent with d-caches.
18
* Must deal with range from start to end-1 but nothing else (need to
19
* be careful not to touch addresses that may be unmapped).
21
* Note: "in0" and "in1" are preserved for debugging purposes.
23
.section .kprobes.text,"ax"
24
GLOBAL_ENTRY(flush_icache_range)
27
alloc r2=ar.pfs,2,0,0,0
28
movl r3=ia64_i_cache_stride_shift
31
ld8 r20=[r3] // r20: stride shift
32
sub r22=in1,r0,1 // last byte address
34
shr.u r23=in0,r20 // start / (stride size)
35
shr.u r22=r22,r20 // (last byte address) / (stride size)
36
shl r21=r21,r20 // r21: stride size of the i-cache(s)
38
sub r8=r22,r23 // number of strides - 1
39
shl r24=r23,r20 // r24: addresses for "fc.i" =
40
// "start" rounded down to stride boundary
42
mov r3=ar.lc // save ar.lc
49
* 32 byte aligned loop, even number of (actually 2) bundles
51
.Loop: fc.i r24 // issuable on M0 only
52
add r24=r21,r24 // we flush "stride size" bytes per iteration
54
br.cloop.sptk.few .Loop
60
mov ar.lc=r3 // restore ar.lc
62
END(flush_icache_range)
65
* clflush_cache_range(start,size)
67
* Flush cache lines from start to start+size-1.
69
* Must deal with range from start to start+size-1 but nothing else
70
* (need to be careful not to touch addresses that may be
73
* Note: "in0" and "in1" are preserved for debugging purposes.
75
.section .kprobes.text,"ax"
76
GLOBAL_ENTRY(clflush_cache_range)
79
alloc r2=ar.pfs,2,0,0,0
80
movl r3=ia64_cache_stride_shift
84
ld8 r20=[r3] // r20: stride shift
85
sub r22=r22,r0,1 // last byte address
87
shr.u r23=in0,r20 // start / (stride size)
88
shr.u r22=r22,r20 // (last byte address) / (stride size)
89
shl r21=r21,r20 // r21: stride size of the i-cache(s)
91
sub r8=r22,r23 // number of strides - 1
92
shl r24=r23,r20 // r24: addresses for "fc" =
93
// "start" rounded down to stride
96
mov r3=ar.lc // save ar.lc
103
* 32 byte aligned loop, even number of (actually 2) bundles
106
fc r24 // issuable on M0 only
107
add r24=r21,r24 // we flush "stride size" bytes per iteration
109
br.cloop.sptk.few .Loop_fc
115
mov ar.lc=r3 // restore ar.lc
117
END(clflush_cache_range)