Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/arch/arm/lib/copypage-xscale.S |
| 3 | * |
| 4 | * Copyright (C) 2001 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 | #include <linux/linkage.h> |
| 11 | #include <linux/init.h> |
| 12 | #include <asm/constants.h> |
| 13 | |
| 14 | /* |
| 15 | * General note: |
| 16 | * We don't really want write-allocate cache behaviour for these functions |
| 17 | * since that will just eat through 8K of the cache. |
| 18 | */ |
| 19 | |
| 20 | .text |
| 21 | .align 5 |
| 22 | /* |
| 23 | * XScale optimised copy_user_page |
| 24 | * r0 = destination |
| 25 | * r1 = source |
| 26 | * r2 = virtual user address of ultimate destination page |
| 27 | * |
| 28 | * The source page may have some clean entries in the cache already, but we |
| 29 | * can safely ignore them - break_cow() will flush them out of the cache |
| 30 | * if we eventually end up using our copied page. |
| 31 | * |
| 32 | * What we could do is use the mini-cache to buffer reads from the source |
| 33 | * page. We rely on the mini-cache being smaller than one page, so we'll |
| 34 | * cycle through the complete cache anyway. |
| 35 | */ |
| 36 | ENTRY(xscale_mc_copy_user_page) |
| 37 | stmfd sp!, {r4, r5, lr} |
| 38 | mov r5, r0 |
| 39 | mov r0, r1 |
| 40 | bl map_page_minicache |
| 41 | mov r1, r5 |
| 42 | mov lr, #PAGE_SZ/64-1 |
| 43 | |
| 44 | /* |
| 45 | * Strangely enough, best performance is achieved |
| 46 | * when prefetching destination as well. (NP) |
| 47 | */ |
| 48 | pld [r0, #0] |
| 49 | pld [r0, #32] |
| 50 | pld [r1, #0] |
| 51 | pld [r1, #32] |
| 52 | |
| 53 | 1: pld [r0, #64] |
| 54 | pld [r0, #96] |
| 55 | pld [r1, #64] |
| 56 | pld [r1, #96] |
| 57 | |
| 58 | 2: ldrd r2, [r0], #8 |
| 59 | ldrd r4, [r0], #8 |
| 60 | mov ip, r1 |
| 61 | strd r2, [r1], #8 |
| 62 | ldrd r2, [r0], #8 |
| 63 | strd r4, [r1], #8 |
| 64 | ldrd r4, [r0], #8 |
| 65 | strd r2, [r1], #8 |
| 66 | strd r4, [r1], #8 |
| 67 | mcr p15, 0, ip, c7, c10, 1 @ clean D line |
| 68 | ldrd r2, [r0], #8 |
| 69 | mcr p15, 0, ip, c7, c6, 1 @ invalidate D line |
| 70 | ldrd r4, [r0], #8 |
| 71 | mov ip, r1 |
| 72 | strd r2, [r1], #8 |
| 73 | ldrd r2, [r0], #8 |
| 74 | strd r4, [r1], #8 |
| 75 | ldrd r4, [r0], #8 |
| 76 | strd r2, [r1], #8 |
| 77 | strd r4, [r1], #8 |
| 78 | mcr p15, 0, ip, c7, c10, 1 @ clean D line |
| 79 | subs lr, lr, #1 |
| 80 | mcr p15, 0, ip, c7, c6, 1 @ invalidate D line |
| 81 | bgt 1b |
| 82 | beq 2b |
| 83 | |
| 84 | ldmfd sp!, {r4, r5, pc} |
| 85 | |
| 86 | .align 5 |
| 87 | /* |
| 88 | * XScale optimised clear_user_page |
| 89 | * r0 = destination |
| 90 | * r1 = virtual user address of ultimate destination page |
| 91 | */ |
| 92 | ENTRY(xscale_mc_clear_user_page) |
| 93 | mov r1, #PAGE_SZ/32 |
| 94 | mov r2, #0 |
| 95 | mov r3, #0 |
| 96 | 1: mov ip, r0 |
| 97 | strd r2, [r0], #8 |
| 98 | strd r2, [r0], #8 |
| 99 | strd r2, [r0], #8 |
| 100 | strd r2, [r0], #8 |
| 101 | mcr p15, 0, ip, c7, c10, 1 @ clean D line |
| 102 | subs r1, r1, #1 |
| 103 | mcr p15, 0, ip, c7, c6, 1 @ invalidate D line |
| 104 | bne 1b |
| 105 | mov pc, lr |
| 106 | |
| 107 | __INITDATA |
| 108 | |
| 109 | .type xscale_mc_user_fns, #object |
| 110 | ENTRY(xscale_mc_user_fns) |
| 111 | .long xscale_mc_clear_user_page |
| 112 | .long xscale_mc_copy_user_page |
| 113 | .size xscale_mc_user_fns, . - xscale_mc_user_fns |