Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * arch/alpha/lib/strncpy_from_user.S |
| 3 | * Contributed by Richard Henderson (rth@tamu.edu) |
| 4 | * |
| 5 | * Just like strncpy except in the return value: |
| 6 | * |
| 7 | * -EFAULT if an exception occurs before the terminator is copied. |
| 8 | * N if the buffer filled. |
| 9 | * |
| 10 | * Otherwise the length of the string is returned. |
| 11 | */ |
| 12 | |
| 13 | |
| 14 | #include <asm/errno.h> |
| 15 | #include <asm/regdef.h> |
| 16 | |
| 17 | |
| 18 | /* Allow an exception for an insn; exit if we get one. */ |
| 19 | #define EX(x,y...) \ |
| 20 | 99: x,##y; \ |
| 21 | .section __ex_table,"a"; \ |
| 22 | .long 99b - .; \ |
| 23 | lda $31, $exception-99b($0); \ |
| 24 | .previous |
| 25 | |
| 26 | |
| 27 | .set noat |
| 28 | .set noreorder |
| 29 | .text |
| 30 | |
| 31 | .globl __strncpy_from_user |
| 32 | .ent __strncpy_from_user |
| 33 | .frame $30, 0, $26 |
| 34 | .prologue 0 |
| 35 | |
| 36 | .align 3 |
| 37 | $aligned: |
| 38 | /* On entry to this basic block: |
| 39 | t0 == the first destination word for masking back in |
| 40 | t1 == the first source word. */ |
| 41 | |
| 42 | /* Create the 1st output word and detect 0's in the 1st input word. */ |
| 43 | lda t2, -1 # e1 : build a mask against false zero |
| 44 | mskqh t2, a1, t2 # e0 : detection in the src word |
| 45 | mskqh t1, a1, t3 # e0 : |
| 46 | ornot t1, t2, t2 # .. e1 : |
| 47 | mskql t0, a1, t0 # e0 : assemble the first output word |
| 48 | cmpbge zero, t2, t8 # .. e1 : bits set iff null found |
| 49 | or t0, t3, t0 # e0 : |
| 50 | beq a2, $a_eoc # .. e1 : |
| 51 | bne t8, $a_eos # .. e1 : |
| 52 | |
| 53 | /* On entry to this basic block: |
| 54 | t0 == a source word not containing a null. */ |
| 55 | |
| 56 | $a_loop: |
| 57 | stq_u t0, 0(a0) # e0 : |
| 58 | addq a0, 8, a0 # .. e1 : |
| 59 | EX( ldq_u t0, 0(a1) ) # e0 : |
| 60 | addq a1, 8, a1 # .. e1 : |
| 61 | subq a2, 1, a2 # e0 : |
| 62 | cmpbge zero, t0, t8 # .. e1 (stall) |
| 63 | beq a2, $a_eoc # e1 : |
| 64 | beq t8, $a_loop # e1 : |
| 65 | |
| 66 | /* Take care of the final (partial) word store. At this point |
| 67 | the end-of-count bit is set in t8 iff it applies. |
| 68 | |
| 69 | On entry to this basic block we have: |
| 70 | t0 == the source word containing the null |
| 71 | t8 == the cmpbge mask that found it. */ |
| 72 | |
| 73 | $a_eos: |
| 74 | negq t8, t12 # e0 : find low bit set |
| 75 | and t8, t12, t12 # e1 (stall) |
| 76 | |
| 77 | /* For the sake of the cache, don't read a destination word |
| 78 | if we're not going to need it. */ |
| 79 | and t12, 0x80, t6 # e0 : |
| 80 | bne t6, 1f # .. e1 (zdb) |
| 81 | |
| 82 | /* We're doing a partial word store and so need to combine |
| 83 | our source and original destination words. */ |
| 84 | ldq_u t1, 0(a0) # e0 : |
| 85 | subq t12, 1, t6 # .. e1 : |
| 86 | or t12, t6, t8 # e0 : |
| 87 | unop # |
| 88 | zapnot t0, t8, t0 # e0 : clear src bytes > null |
| 89 | zap t1, t8, t1 # .. e1 : clear dst bytes <= null |
| 90 | or t0, t1, t0 # e1 : |
| 91 | |
| 92 | 1: stq_u t0, 0(a0) |
| 93 | br $finish_up |
| 94 | |
| 95 | /* Add the end-of-count bit to the eos detection bitmask. */ |
| 96 | $a_eoc: |
| 97 | or t10, t8, t8 |
| 98 | br $a_eos |
| 99 | |
| 100 | /*** The Function Entry Point ***/ |
| 101 | .align 3 |
| 102 | __strncpy_from_user: |
| 103 | mov a0, v0 # save the string start |
| 104 | beq a2, $zerolength |
| 105 | |
| 106 | /* Are source and destination co-aligned? */ |
| 107 | xor a0, a1, t1 # e0 : |
| 108 | and a0, 7, t0 # .. e1 : find dest misalignment |
| 109 | and t1, 7, t1 # e0 : |
| 110 | addq a2, t0, a2 # .. e1 : bias count by dest misalignment |
| 111 | subq a2, 1, a2 # e0 : |
| 112 | and a2, 7, t2 # e1 : |
| 113 | srl a2, 3, a2 # e0 : a2 = loop counter = (count - 1)/8 |
| 114 | addq zero, 1, t10 # .. e1 : |
| 115 | sll t10, t2, t10 # e0 : t10 = bitmask of last count byte |
| 116 | bne t1, $unaligned # .. e1 : |
| 117 | |
| 118 | /* We are co-aligned; take care of a partial first word. */ |
| 119 | |
| 120 | EX( ldq_u t1, 0(a1) ) # e0 : load first src word |
| 121 | addq a1, 8, a1 # .. e1 : |
| 122 | |
| 123 | beq t0, $aligned # avoid loading dest word if not needed |
| 124 | ldq_u t0, 0(a0) # e0 : |
| 125 | br $aligned # .. e1 : |
| 126 | |
| 127 | |
| 128 | /* The source and destination are not co-aligned. Align the destination |
| 129 | and cope. We have to be very careful about not reading too much and |
| 130 | causing a SEGV. */ |
| 131 | |
| 132 | .align 3 |
| 133 | $u_head: |
| 134 | /* We know just enough now to be able to assemble the first |
| 135 | full source word. We can still find a zero at the end of it |
| 136 | that prevents us from outputting the whole thing. |
| 137 | |
| 138 | On entry to this basic block: |
| 139 | t0 == the first dest word, unmasked |
| 140 | t1 == the shifted low bits of the first source word |
| 141 | t6 == bytemask that is -1 in dest word bytes */ |
| 142 | |
| 143 | EX( ldq_u t2, 8(a1) ) # e0 : load second src word |
| 144 | addq a1, 8, a1 # .. e1 : |
| 145 | mskql t0, a0, t0 # e0 : mask trailing garbage in dst |
| 146 | extqh t2, a1, t4 # e0 : |
| 147 | or t1, t4, t1 # e1 : first aligned src word complete |
| 148 | mskqh t1, a0, t1 # e0 : mask leading garbage in src |
| 149 | or t0, t1, t0 # e0 : first output word complete |
| 150 | or t0, t6, t6 # e1 : mask original data for zero test |
| 151 | cmpbge zero, t6, t8 # e0 : |
| 152 | beq a2, $u_eocfin # .. e1 : |
| 153 | bne t8, $u_final # e1 : |
| 154 | |
| 155 | lda t6, -1 # e1 : mask out the bits we have |
| 156 | mskql t6, a1, t6 # e0 : already seen |
| 157 | stq_u t0, 0(a0) # e0 : store first output word |
| 158 | or t6, t2, t2 # .. e1 : |
| 159 | cmpbge zero, t2, t8 # e0 : find nulls in second partial |
| 160 | addq a0, 8, a0 # .. e1 : |
| 161 | subq a2, 1, a2 # e0 : |
| 162 | bne t8, $u_late_head_exit # .. e1 : |
| 163 | |
| 164 | /* Finally, we've got all the stupid leading edge cases taken care |
| 165 | of and we can set up to enter the main loop. */ |
| 166 | |
| 167 | extql t2, a1, t1 # e0 : position hi-bits of lo word |
| 168 | EX( ldq_u t2, 8(a1) ) # .. e1 : read next high-order source word |
| 169 | addq a1, 8, a1 # e0 : |
| 170 | cmpbge zero, t2, t8 # e1 (stall) |
| 171 | beq a2, $u_eoc # e1 : |
| 172 | bne t8, $u_eos # e1 : |
| 173 | |
| 174 | /* Unaligned copy main loop. In order to avoid reading too much, |
| 175 | the loop is structured to detect zeros in aligned source words. |
| 176 | This has, unfortunately, effectively pulled half of a loop |
| 177 | iteration out into the head and half into the tail, but it does |
| 178 | prevent nastiness from accumulating in the very thing we want |
| 179 | to run as fast as possible. |
| 180 | |
| 181 | On entry to this basic block: |
| 182 | t1 == the shifted high-order bits from the previous source word |
| 183 | t2 == the unshifted current source word |
| 184 | |
| 185 | We further know that t2 does not contain a null terminator. */ |
| 186 | |
| 187 | .align 3 |
| 188 | $u_loop: |
| 189 | extqh t2, a1, t0 # e0 : extract high bits for current word |
| 190 | addq a1, 8, a1 # .. e1 : |
| 191 | extql t2, a1, t3 # e0 : extract low bits for next time |
| 192 | addq a0, 8, a0 # .. e1 : |
| 193 | or t0, t1, t0 # e0 : current dst word now complete |
| 194 | EX( ldq_u t2, 0(a1) ) # .. e1 : load high word for next time |
| 195 | stq_u t0, -8(a0) # e0 : save the current word |
| 196 | mov t3, t1 # .. e1 : |
| 197 | subq a2, 1, a2 # e0 : |
| 198 | cmpbge zero, t2, t8 # .. e1 : test new word for eos |
| 199 | beq a2, $u_eoc # e1 : |
| 200 | beq t8, $u_loop # e1 : |
| 201 | |
| 202 | /* We've found a zero somewhere in the source word we just read. |
| 203 | If it resides in the lower half, we have one (probably partial) |
| 204 | word to write out, and if it resides in the upper half, we |
| 205 | have one full and one partial word left to write out. |
| 206 | |
| 207 | On entry to this basic block: |
| 208 | t1 == the shifted high-order bits from the previous source word |
| 209 | t2 == the unshifted current source word. */ |
| 210 | $u_eos: |
| 211 | extqh t2, a1, t0 # e0 : |
| 212 | or t0, t1, t0 # e1 : first (partial) source word complete |
| 213 | |
| 214 | cmpbge zero, t0, t8 # e0 : is the null in this first bit? |
| 215 | bne t8, $u_final # .. e1 (zdb) |
| 216 | |
| 217 | stq_u t0, 0(a0) # e0 : the null was in the high-order bits |
| 218 | addq a0, 8, a0 # .. e1 : |
| 219 | subq a2, 1, a2 # e1 : |
| 220 | |
| 221 | $u_late_head_exit: |
| 222 | extql t2, a1, t0 # .. e0 : |
| 223 | cmpbge zero, t0, t8 # e0 : |
| 224 | or t8, t10, t6 # e1 : |
| 225 | cmoveq a2, t6, t8 # e0 : |
| 226 | nop # .. e1 : |
| 227 | |
| 228 | /* Take care of a final (probably partial) result word. |
| 229 | On entry to this basic block: |
| 230 | t0 == assembled source word |
| 231 | t8 == cmpbge mask that found the null. */ |
| 232 | $u_final: |
| 233 | negq t8, t6 # e0 : isolate low bit set |
| 234 | and t6, t8, t12 # e1 : |
| 235 | |
| 236 | and t12, 0x80, t6 # e0 : avoid dest word load if we can |
| 237 | bne t6, 1f # .. e1 (zdb) |
| 238 | |
| 239 | ldq_u t1, 0(a0) # e0 : |
| 240 | subq t12, 1, t6 # .. e1 : |
| 241 | or t6, t12, t8 # e0 : |
| 242 | zapnot t0, t8, t0 # .. e1 : kill source bytes > null |
| 243 | zap t1, t8, t1 # e0 : kill dest bytes <= null |
| 244 | or t0, t1, t0 # e1 : |
| 245 | |
| 246 | 1: stq_u t0, 0(a0) # e0 : |
| 247 | br $finish_up |
| 248 | |
| 249 | $u_eoc: # end-of-count |
| 250 | extqh t2, a1, t0 |
| 251 | or t0, t1, t0 |
| 252 | cmpbge zero, t0, t8 |
| 253 | |
| 254 | $u_eocfin: # end-of-count, final word |
| 255 | or t10, t8, t8 |
| 256 | br $u_final |
| 257 | |
| 258 | /* Unaligned copy entry point. */ |
| 259 | .align 3 |
| 260 | $unaligned: |
| 261 | |
| 262 | EX( ldq_u t1, 0(a1) ) # e0 : load first source word |
| 263 | |
| 264 | and a0, 7, t4 # .. e1 : find dest misalignment |
| 265 | and a1, 7, t5 # e0 : find src misalignment |
| 266 | |
| 267 | /* Conditionally load the first destination word and a bytemask |
| 268 | with 0xff indicating that the destination byte is sacrosanct. */ |
| 269 | |
| 270 | mov zero, t0 # .. e1 : |
| 271 | mov zero, t6 # e0 : |
| 272 | beq t4, 1f # .. e1 : |
| 273 | ldq_u t0, 0(a0) # e0 : |
| 274 | lda t6, -1 # .. e1 : |
| 275 | mskql t6, a0, t6 # e0 : |
| 276 | 1: |
| 277 | subq a1, t4, a1 # .. e1 : sub dest misalignment from src addr |
| 278 | |
| 279 | /* If source misalignment is larger than dest misalignment, we need |
| 280 | extra startup checks to avoid SEGV. */ |
| 281 | |
| 282 | cmplt t4, t5, t12 # e1 : |
| 283 | extql t1, a1, t1 # .. e0 : shift src into place |
| 284 | lda t2, -1 # e0 : for creating masks later |
| 285 | beq t12, $u_head # e1 : |
| 286 | |
| 287 | mskqh t2, t5, t2 # e0 : begin src byte validity mask |
| 288 | cmpbge zero, t1, t8 # .. e1 : is there a zero? |
| 289 | extql t2, a1, t2 # e0 : |
| 290 | or t8, t10, t5 # .. e1 : test for end-of-count too |
| 291 | cmpbge zero, t2, t3 # e0 : |
| 292 | cmoveq a2, t5, t8 # .. e1 : |
| 293 | andnot t8, t3, t8 # e0 : |
| 294 | beq t8, $u_head # .. e1 (zdb) |
| 295 | |
| 296 | /* At this point we've found a zero in the first partial word of |
| 297 | the source. We need to isolate the valid source data and mask |
| 298 | it into the original destination data. (Incidentally, we know |
| 299 | that we'll need at least one byte of that original dest word.) */ |
| 300 | |
| 301 | ldq_u t0, 0(a0) # e0 : |
| 302 | negq t8, t6 # .. e1 : build bitmask of bytes <= zero |
| 303 | mskqh t1, t4, t1 # e0 : |
| 304 | and t6, t8, t12 # .. e1 : |
| 305 | subq t12, 1, t6 # e0 : |
| 306 | or t6, t12, t8 # e1 : |
| 307 | |
| 308 | zapnot t2, t8, t2 # e0 : prepare source word; mirror changes |
| 309 | zapnot t1, t8, t1 # .. e1 : to source validity mask |
| 310 | |
| 311 | andnot t0, t2, t0 # e0 : zero place for source to reside |
| 312 | or t0, t1, t0 # e1 : and put it there |
| 313 | stq_u t0, 0(a0) # e0 : |
| 314 | |
| 315 | $finish_up: |
| 316 | zapnot t0, t12, t4 # was last byte written null? |
| 317 | cmovne t4, 1, t4 |
| 318 | |
| 319 | and t12, 0xf0, t3 # binary search for the address of the |
| 320 | and t12, 0xcc, t2 # last byte written |
| 321 | and t12, 0xaa, t1 |
| 322 | bic a0, 7, t0 |
| 323 | cmovne t3, 4, t3 |
| 324 | cmovne t2, 2, t2 |
| 325 | cmovne t1, 1, t1 |
| 326 | addq t0, t3, t0 |
| 327 | addq t1, t2, t1 |
| 328 | addq t0, t1, t0 |
| 329 | addq t0, t4, t0 # add one if we filled the buffer |
| 330 | |
| 331 | subq t0, v0, v0 # find string length |
| 332 | ret |
| 333 | |
| 334 | $zerolength: |
| 335 | clr v0 |
| 336 | $exception: |
| 337 | ret |
| 338 | |
| 339 | .end __strncpy_from_user |