memzero.S 2.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126
  1. /*
  2. * linux/arch/arm/lib/memzero.S
  3. *
  4. * Copyright (C) 1995-2000 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 <asm/assembler.h>
  12. .text
  13. .align 5
  14. .word 0
  15. /*
  16. * Align the pointer in r0. r3 contains the number of bytes that we are
  17. * mis-aligned by, and r1 is the number of bytes. If r1 < 4, then we
  18. * don't bother; we use byte stores instead.
  19. */
  20. 1: subs r1, r1, #4 @ 1 do we have enough
  21. blt 5f @ 1 bytes to align with?
  22. cmp r3, #2 @ 1
  23. strltb r2, [r0], #1 @ 1
  24. strleb r2, [r0], #1 @ 1
  25. strb r2, [r0], #1 @ 1
  26. add r1, r1, r3 @ 1 (r1 = r1 - (4 - r3))
  27. /*
  28. * The pointer is now aligned and the length is adjusted. Try doing the
  29. * memzero again.
  30. */
  31. ENTRY(__memzero)
  32. mov r2, #0 @ 1
  33. ands r3, r0, #3 @ 1 unaligned?
  34. bne 1b @ 1
  35. /*
  36. * r3 = 0, and we know that the pointer in r0 is aligned to a word boundary.
  37. */
  38. cmp r1, #16 @ 1 we can skip this chunk if we
  39. blt 4f @ 1 have < 16 bytes
  40. #if ! CALGN(1)+0
  41. /*
  42. * We need an extra register for this loop - save the return address and
  43. * use the LR
  44. */
  45. str lr, [sp, #-4]! @ 1
  46. mov ip, r2 @ 1
  47. mov lr, r2 @ 1
  48. 3: subs r1, r1, #64 @ 1 write 32 bytes out per loop
  49. stmgeia r0!, {r2, r3, ip, lr} @ 4
  50. stmgeia r0!, {r2, r3, ip, lr} @ 4
  51. stmgeia r0!, {r2, r3, ip, lr} @ 4
  52. stmgeia r0!, {r2, r3, ip, lr} @ 4
  53. bgt 3b @ 1
  54. ldmeqfd sp!, {pc} @ 1/2 quick exit
  55. /*
  56. * No need to correct the count; we're only testing bits from now on
  57. */
  58. tst r1, #32 @ 1
  59. stmneia r0!, {r2, r3, ip, lr} @ 4
  60. stmneia r0!, {r2, r3, ip, lr} @ 4
  61. tst r1, #16 @ 1 16 bytes or more?
  62. stmneia r0!, {r2, r3, ip, lr} @ 4
  63. ldr lr, [sp], #4 @ 1
  64. #else
  65. /*
  66. * This version aligns the destination pointer in order to write
  67. * whole cache lines at once.
  68. */
  69. stmfd sp!, {r4-r7, lr}
  70. mov r4, r2
  71. mov r5, r2
  72. mov r6, r2
  73. mov r7, r2
  74. mov ip, r2
  75. mov lr, r2
  76. cmp r1, #96
  77. andgts ip, r0, #31
  78. ble 3f
  79. rsb ip, ip, #32
  80. sub r1, r1, ip
  81. movs ip, ip, lsl #(32 - 4)
  82. stmcsia r0!, {r4, r5, r6, r7}
  83. stmmiia r0!, {r4, r5}
  84. movs ip, ip, lsl #2
  85. strcs r2, [r0], #4
  86. 3: subs r1, r1, #64
  87. stmgeia r0!, {r2-r7, ip, lr}
  88. stmgeia r0!, {r2-r7, ip, lr}
  89. bgt 3b
  90. ldmeqfd sp!, {r4-r7, pc}
  91. tst r1, #32
  92. stmneia r0!, {r2-r7, ip, lr}
  93. tst r1, #16
  94. stmneia r0!, {r4-r7}
  95. ldmfd sp!, {r4-r7, lr}
  96. #endif
  97. 4: tst r1, #8 @ 1 8 bytes or more?
  98. stmneia r0!, {r2, r3} @ 2
  99. tst r1, #4 @ 1 4 bytes or more?
  100. strne r2, [r0], #4 @ 1
  101. /*
  102. * When we get here, we've got less than 4 bytes to zero. We
  103. * may have an unaligned pointer as well.
  104. */
  105. 5: tst r1, #2 @ 1 2 bytes or more?
  106. strneb r2, [r0], #1 @ 1
  107. strneb r2, [r0], #1 @ 1
  108. tst r1, #1 @ 1 a byte left over
  109. strneb r2, [r0], #1 @ 1
  110. mov pc, lr @ 1
  111. ENDPROC(__memzero)