parisc: fixes and cleanups in page cache flushing (2/4)
[deliverable/linux.git] / arch / parisc / kernel / pacache.S
1 /*
2 * PARISC TLB and cache flushing support
3 * Copyright (C) 2000-2001 Hewlett-Packard (John Marvin)
4 * Copyright (C) 2001 Matthew Wilcox (willy at parisc-linux.org)
5 * Copyright (C) 2002 Richard Hirst (rhirst with parisc-linux.org)
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2, or (at your option)
10 * any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 */
21
22 /*
23 * NOTE: fdc,fic, and pdc instructions that use base register modification
24 * should only use index and base registers that are not shadowed,
25 * so that the fast path emulation in the non access miss handler
26 * can be used.
27 */
28
29 #ifdef CONFIG_64BIT
30 .level 2.0w
31 #else
32 .level 2.0
33 #endif
34
35 #include <asm/psw.h>
36 #include <asm/assembly.h>
37 #include <asm/pgtable.h>
38 #include <asm/cache.h>
39 #include <linux/linkage.h>
40
41 .text
42 .align 128
43
44 ENTRY(flush_tlb_all_local)
45 .proc
46 .callinfo NO_CALLS
47 .entry
48
49 /*
50 * The pitlbe and pdtlbe instructions should only be used to
51 * flush the entire tlb. Also, there needs to be no intervening
52 * tlb operations, e.g. tlb misses, so the operation needs
53 * to happen in real mode with all interruptions disabled.
54 */
55
56 /* pcxt_ssm_bug - relied upon translation! PA 2.0 Arch. F-4 and F-5 */
57 rsm PSW_SM_I, %r19 /* save I-bit state */
58 load32 PA(1f), %r1
59 nop
60 nop
61 nop
62 nop
63 nop
64
65 rsm PSW_SM_Q, %r0 /* prep to load iia queue */
66 mtctl %r0, %cr17 /* Clear IIASQ tail */
67 mtctl %r0, %cr17 /* Clear IIASQ head */
68 mtctl %r1, %cr18 /* IIAOQ head */
69 ldo 4(%r1), %r1
70 mtctl %r1, %cr18 /* IIAOQ tail */
71 load32 REAL_MODE_PSW, %r1
72 mtctl %r1, %ipsw
73 rfi
74 nop
75
76 1: load32 PA(cache_info), %r1
77
78 /* Flush Instruction Tlb */
79
80 LDREG ITLB_SID_BASE(%r1), %r20
81 LDREG ITLB_SID_STRIDE(%r1), %r21
82 LDREG ITLB_SID_COUNT(%r1), %r22
83 LDREG ITLB_OFF_BASE(%r1), %arg0
84 LDREG ITLB_OFF_STRIDE(%r1), %arg1
85 LDREG ITLB_OFF_COUNT(%r1), %arg2
86 LDREG ITLB_LOOP(%r1), %arg3
87
88 addib,COND(=) -1, %arg3, fitoneloop /* Preadjust and test */
89 movb,<,n %arg3, %r31, fitdone /* If loop < 0, skip */
90 copy %arg0, %r28 /* Init base addr */
91
92 fitmanyloop: /* Loop if LOOP >= 2 */
93 mtsp %r20, %sr1
94 add %r21, %r20, %r20 /* increment space */
95 copy %arg2, %r29 /* Init middle loop count */
96
97 fitmanymiddle: /* Loop if LOOP >= 2 */
98 addib,COND(>) -1, %r31, fitmanymiddle /* Adjusted inner loop decr */
99 pitlbe 0(%sr1, %r28)
100 pitlbe,m %arg1(%sr1, %r28) /* Last pitlbe and addr adjust */
101 addib,COND(>) -1, %r29, fitmanymiddle /* Middle loop decr */
102 copy %arg3, %r31 /* Re-init inner loop count */
103
104 movb,tr %arg0, %r28, fitmanyloop /* Re-init base addr */
105 addib,COND(<=),n -1, %r22, fitdone /* Outer loop count decr */
106
107 fitoneloop: /* Loop if LOOP = 1 */
108 mtsp %r20, %sr1
109 copy %arg0, %r28 /* init base addr */
110 copy %arg2, %r29 /* init middle loop count */
111
112 fitonemiddle: /* Loop if LOOP = 1 */
113 addib,COND(>) -1, %r29, fitonemiddle /* Middle loop count decr */
114 pitlbe,m %arg1(%sr1, %r28) /* pitlbe for one loop */
115
116 addib,COND(>) -1, %r22, fitoneloop /* Outer loop count decr */
117 add %r21, %r20, %r20 /* increment space */
118
119 fitdone:
120
121 /* Flush Data Tlb */
122
123 LDREG DTLB_SID_BASE(%r1), %r20
124 LDREG DTLB_SID_STRIDE(%r1), %r21
125 LDREG DTLB_SID_COUNT(%r1), %r22
126 LDREG DTLB_OFF_BASE(%r1), %arg0
127 LDREG DTLB_OFF_STRIDE(%r1), %arg1
128 LDREG DTLB_OFF_COUNT(%r1), %arg2
129 LDREG DTLB_LOOP(%r1), %arg3
130
131 addib,COND(=) -1, %arg3, fdtoneloop /* Preadjust and test */
132 movb,<,n %arg3, %r31, fdtdone /* If loop < 0, skip */
133 copy %arg0, %r28 /* Init base addr */
134
135 fdtmanyloop: /* Loop if LOOP >= 2 */
136 mtsp %r20, %sr1
137 add %r21, %r20, %r20 /* increment space */
138 copy %arg2, %r29 /* Init middle loop count */
139
140 fdtmanymiddle: /* Loop if LOOP >= 2 */
141 addib,COND(>) -1, %r31, fdtmanymiddle /* Adjusted inner loop decr */
142 pdtlbe 0(%sr1, %r28)
143 pdtlbe,m %arg1(%sr1, %r28) /* Last pdtlbe and addr adjust */
144 addib,COND(>) -1, %r29, fdtmanymiddle /* Middle loop decr */
145 copy %arg3, %r31 /* Re-init inner loop count */
146
147 movb,tr %arg0, %r28, fdtmanyloop /* Re-init base addr */
148 addib,COND(<=),n -1, %r22,fdtdone /* Outer loop count decr */
149
150 fdtoneloop: /* Loop if LOOP = 1 */
151 mtsp %r20, %sr1
152 copy %arg0, %r28 /* init base addr */
153 copy %arg2, %r29 /* init middle loop count */
154
155 fdtonemiddle: /* Loop if LOOP = 1 */
156 addib,COND(>) -1, %r29, fdtonemiddle /* Middle loop count decr */
157 pdtlbe,m %arg1(%sr1, %r28) /* pdtlbe for one loop */
158
159 addib,COND(>) -1, %r22, fdtoneloop /* Outer loop count decr */
160 add %r21, %r20, %r20 /* increment space */
161
162
163 fdtdone:
164 /*
165 * Switch back to virtual mode
166 */
167 /* pcxt_ssm_bug */
168 rsm PSW_SM_I, %r0
169 load32 2f, %r1
170 nop
171 nop
172 nop
173 nop
174 nop
175
176 rsm PSW_SM_Q, %r0 /* prep to load iia queue */
177 mtctl %r0, %cr17 /* Clear IIASQ tail */
178 mtctl %r0, %cr17 /* Clear IIASQ head */
179 mtctl %r1, %cr18 /* IIAOQ head */
180 ldo 4(%r1), %r1
181 mtctl %r1, %cr18 /* IIAOQ tail */
182 load32 KERNEL_PSW, %r1
183 or %r1, %r19, %r1 /* I-bit to state on entry */
184 mtctl %r1, %ipsw /* restore I-bit (entire PSW) */
185 rfi
186 nop
187
188 2: bv %r0(%r2)
189 nop
190
191 .exit
192 .procend
193 ENDPROC(flush_tlb_all_local)
194
195 .import cache_info,data
196
197 ENTRY(flush_instruction_cache_local)
198 .proc
199 .callinfo NO_CALLS
200 .entry
201
202 load32 cache_info, %r1
203
204 /* Flush Instruction Cache */
205
206 LDREG ICACHE_BASE(%r1), %arg0
207 LDREG ICACHE_STRIDE(%r1), %arg1
208 LDREG ICACHE_COUNT(%r1), %arg2
209 LDREG ICACHE_LOOP(%r1), %arg3
210 rsm PSW_SM_I, %r22 /* No mmgt ops during loop*/
211 mtsp %r0, %sr1
212 addib,COND(=) -1, %arg3, fioneloop /* Preadjust and test */
213 movb,<,n %arg3, %r31, fisync /* If loop < 0, do sync */
214
215 fimanyloop: /* Loop if LOOP >= 2 */
216 addib,COND(>) -1, %r31, fimanyloop /* Adjusted inner loop decr */
217 fice %r0(%sr1, %arg0)
218 fice,m %arg1(%sr1, %arg0) /* Last fice and addr adjust */
219 movb,tr %arg3, %r31, fimanyloop /* Re-init inner loop count */
220 addib,COND(<=),n -1, %arg2, fisync /* Outer loop decr */
221
222 fioneloop: /* Loop if LOOP = 1 */
223 /* Some implementations may flush with a single fice instruction */
224 cmpib,COND(>>=),n 15, %arg2, fioneloop2
225
226 fioneloop1:
227 fice,m %arg1(%sr1, %arg0)
228 fice,m %arg1(%sr1, %arg0)
229 fice,m %arg1(%sr1, %arg0)
230 fice,m %arg1(%sr1, %arg0)
231 fice,m %arg1(%sr1, %arg0)
232 fice,m %arg1(%sr1, %arg0)
233 fice,m %arg1(%sr1, %arg0)
234 fice,m %arg1(%sr1, %arg0)
235 fice,m %arg1(%sr1, %arg0)
236 fice,m %arg1(%sr1, %arg0)
237 fice,m %arg1(%sr1, %arg0)
238 fice,m %arg1(%sr1, %arg0)
239 fice,m %arg1(%sr1, %arg0)
240 fice,m %arg1(%sr1, %arg0)
241 fice,m %arg1(%sr1, %arg0)
242 addib,COND(>) -16, %arg2, fioneloop1
243 fice,m %arg1(%sr1, %arg0)
244
245 /* Check if done */
246 cmpb,COND(=),n %arg2, %r0, fisync /* Predict branch taken */
247
248 fioneloop2:
249 addib,COND(>) -1, %arg2, fioneloop2 /* Outer loop count decr */
250 fice,m %arg1(%sr1, %arg0) /* Fice for one loop */
251
252 fisync:
253 sync
254 mtsm %r22 /* restore I-bit */
255 bv %r0(%r2)
256 nop
257 .exit
258
259 .procend
260 ENDPROC(flush_instruction_cache_local)
261
262
263 .import cache_info, data
264 ENTRY(flush_data_cache_local)
265 .proc
266 .callinfo NO_CALLS
267 .entry
268
269 load32 cache_info, %r1
270
271 /* Flush Data Cache */
272
273 LDREG DCACHE_BASE(%r1), %arg0
274 LDREG DCACHE_STRIDE(%r1), %arg1
275 LDREG DCACHE_COUNT(%r1), %arg2
276 LDREG DCACHE_LOOP(%r1), %arg3
277 rsm PSW_SM_I, %r22 /* No mmgt ops during loop*/
278 mtsp %r0, %sr1
279 addib,COND(=) -1, %arg3, fdoneloop /* Preadjust and test */
280 movb,<,n %arg3, %r31, fdsync /* If loop < 0, do sync */
281
282 fdmanyloop: /* Loop if LOOP >= 2 */
283 addib,COND(>) -1, %r31, fdmanyloop /* Adjusted inner loop decr */
284 fdce %r0(%sr1, %arg0)
285 fdce,m %arg1(%sr1, %arg0) /* Last fdce and addr adjust */
286 movb,tr %arg3, %r31, fdmanyloop /* Re-init inner loop count */
287 addib,COND(<=),n -1, %arg2, fdsync /* Outer loop decr */
288
289 fdoneloop: /* Loop if LOOP = 1 */
290 /* Some implementations may flush with a single fdce instruction */
291 cmpib,COND(>>=),n 15, %arg2, fdoneloop2
292
293 fdoneloop1:
294 fdce,m %arg1(%sr1, %arg0)
295 fdce,m %arg1(%sr1, %arg0)
296 fdce,m %arg1(%sr1, %arg0)
297 fdce,m %arg1(%sr1, %arg0)
298 fdce,m %arg1(%sr1, %arg0)
299 fdce,m %arg1(%sr1, %arg0)
300 fdce,m %arg1(%sr1, %arg0)
301 fdce,m %arg1(%sr1, %arg0)
302 fdce,m %arg1(%sr1, %arg0)
303 fdce,m %arg1(%sr1, %arg0)
304 fdce,m %arg1(%sr1, %arg0)
305 fdce,m %arg1(%sr1, %arg0)
306 fdce,m %arg1(%sr1, %arg0)
307 fdce,m %arg1(%sr1, %arg0)
308 fdce,m %arg1(%sr1, %arg0)
309 addib,COND(>) -16, %arg2, fdoneloop1
310 fdce,m %arg1(%sr1, %arg0)
311
312 /* Check if done */
313 cmpb,COND(=),n %arg2, %r0, fdsync /* Predict branch taken */
314
315 fdoneloop2:
316 addib,COND(>) -1, %arg2, fdoneloop2 /* Outer loop count decr */
317 fdce,m %arg1(%sr1, %arg0) /* Fdce for one loop */
318
319 fdsync:
320 syncdma
321 sync
322 mtsm %r22 /* restore I-bit */
323 bv %r0(%r2)
324 nop
325 .exit
326
327 .procend
328 ENDPROC(flush_data_cache_local)
329
330 .align 16
331
332 /* Macros to serialize TLB purge operations on SMP. */
333
334 .macro tlb_lock la,flags,tmp
335 #ifdef CONFIG_SMP
336 ldil L%pa_tlb_lock,%r1
337 ldo R%pa_tlb_lock(%r1),\la
338 rsm PSW_SM_I,\flags
339 1: LDCW 0(\la),\tmp
340 cmpib,<>,n 0,\tmp,3f
341 2: ldw 0(\la),\tmp
342 cmpb,<> %r0,\tmp,1b
343 nop
344 b,n 2b
345 3:
346 #endif
347 .endm
348
349 .macro tlb_unlock la,flags,tmp
350 #ifdef CONFIG_SMP
351 ldi 1,\tmp
352 stw \tmp,0(\la)
353 mtsm \flags
354 #endif
355 .endm
356
357 /* Clear page using kernel mapping. */
358
359 ENTRY(clear_page_asm)
360 .proc
361 .callinfo NO_CALLS
362 .entry
363
364 #ifdef CONFIG_64BIT
365
366 /* Unroll the loop. */
367 ldi (PAGE_SIZE / 128), %r1
368
369 1:
370 std %r0, 0(%r26)
371 std %r0, 8(%r26)
372 std %r0, 16(%r26)
373 std %r0, 24(%r26)
374 std %r0, 32(%r26)
375 std %r0, 40(%r26)
376 std %r0, 48(%r26)
377 std %r0, 56(%r26)
378 std %r0, 64(%r26)
379 std %r0, 72(%r26)
380 std %r0, 80(%r26)
381 std %r0, 88(%r26)
382 std %r0, 96(%r26)
383 std %r0, 104(%r26)
384 std %r0, 112(%r26)
385 std %r0, 120(%r26)
386
387 /* Note reverse branch hint for addib is taken. */
388 addib,COND(>),n -1, %r1, 1b
389 ldo 128(%r26), %r26
390
391 #else
392
393 /*
394 * Note that until (if) we start saving the full 64-bit register
395 * values on interrupt, we can't use std on a 32 bit kernel.
396 */
397 ldi (PAGE_SIZE / 64), %r1
398
399 1:
400 stw %r0, 0(%r26)
401 stw %r0, 4(%r26)
402 stw %r0, 8(%r26)
403 stw %r0, 12(%r26)
404 stw %r0, 16(%r26)
405 stw %r0, 20(%r26)
406 stw %r0, 24(%r26)
407 stw %r0, 28(%r26)
408 stw %r0, 32(%r26)
409 stw %r0, 36(%r26)
410 stw %r0, 40(%r26)
411 stw %r0, 44(%r26)
412 stw %r0, 48(%r26)
413 stw %r0, 52(%r26)
414 stw %r0, 56(%r26)
415 stw %r0, 60(%r26)
416
417 addib,COND(>),n -1, %r1, 1b
418 ldo 64(%r26), %r26
419 #endif
420 bv %r0(%r2)
421 nop
422 .exit
423
424 .procend
425 ENDPROC(clear_page_asm)
426
427 /* Copy page using kernel mapping. */
428
429 ENTRY(copy_page_asm)
430 .proc
431 .callinfo NO_CALLS
432 .entry
433
434 #ifdef CONFIG_64BIT
435 /* PA8x00 CPUs can consume 2 loads or 1 store per cycle.
436 * Unroll the loop by hand and arrange insn appropriately.
437 * Prefetch doesn't improve performance on rp3440.
438 * GCC probably can do this just as well...
439 */
440
441 ldi (PAGE_SIZE / 128), %r1
442
443 1: ldd 0(%r25), %r19
444 ldd 8(%r25), %r20
445
446 ldd 16(%r25), %r21
447 ldd 24(%r25), %r22
448 std %r19, 0(%r26)
449 std %r20, 8(%r26)
450
451 ldd 32(%r25), %r19
452 ldd 40(%r25), %r20
453 std %r21, 16(%r26)
454 std %r22, 24(%r26)
455
456 ldd 48(%r25), %r21
457 ldd 56(%r25), %r22
458 std %r19, 32(%r26)
459 std %r20, 40(%r26)
460
461 ldd 64(%r25), %r19
462 ldd 72(%r25), %r20
463 std %r21, 48(%r26)
464 std %r22, 56(%r26)
465
466 ldd 80(%r25), %r21
467 ldd 88(%r25), %r22
468 std %r19, 64(%r26)
469 std %r20, 72(%r26)
470
471 ldd 96(%r25), %r19
472 ldd 104(%r25), %r20
473 std %r21, 80(%r26)
474 std %r22, 88(%r26)
475
476 ldd 112(%r25), %r21
477 ldd 120(%r25), %r22
478 ldo 128(%r25), %r25
479 std %r19, 96(%r26)
480 std %r20, 104(%r26)
481
482 std %r21, 112(%r26)
483 std %r22, 120(%r26)
484
485 /* Note reverse branch hint for addib is taken. */
486 addib,COND(>),n -1, %r1, 1b
487 ldo 128(%r26), %r26
488
489 #else
490
491 /*
492 * This loop is optimized for PCXL/PCXL2 ldw/ldw and stw/stw
493 * bundles (very restricted rules for bundling).
494 * Note that until (if) we start saving
495 * the full 64 bit register values on interrupt, we can't
496 * use ldd/std on a 32 bit kernel.
497 */
498 ldw 0(%r25), %r19
499 ldi (PAGE_SIZE / 64), %r1
500
501 1:
502 ldw 4(%r25), %r20
503 ldw 8(%r25), %r21
504 ldw 12(%r25), %r22
505 stw %r19, 0(%r26)
506 stw %r20, 4(%r26)
507 stw %r21, 8(%r26)
508 stw %r22, 12(%r26)
509 ldw 16(%r25), %r19
510 ldw 20(%r25), %r20
511 ldw 24(%r25), %r21
512 ldw 28(%r25), %r22
513 stw %r19, 16(%r26)
514 stw %r20, 20(%r26)
515 stw %r21, 24(%r26)
516 stw %r22, 28(%r26)
517 ldw 32(%r25), %r19
518 ldw 36(%r25), %r20
519 ldw 40(%r25), %r21
520 ldw 44(%r25), %r22
521 stw %r19, 32(%r26)
522 stw %r20, 36(%r26)
523 stw %r21, 40(%r26)
524 stw %r22, 44(%r26)
525 ldw 48(%r25), %r19
526 ldw 52(%r25), %r20
527 ldw 56(%r25), %r21
528 ldw 60(%r25), %r22
529 stw %r19, 48(%r26)
530 stw %r20, 52(%r26)
531 ldo 64(%r25), %r25
532 stw %r21, 56(%r26)
533 stw %r22, 60(%r26)
534 ldo 64(%r26), %r26
535 addib,COND(>),n -1, %r1, 1b
536 ldw 0(%r25), %r19
537 #endif
538 bv %r0(%r2)
539 nop
540 .exit
541
542 .procend
543 ENDPROC(copy_page_asm)
544
545 /*
546 * NOTE: Code in clear_user_page has a hard coded dependency on the
547 * maximum alias boundary being 4 Mb. We've been assured by the
548 * parisc chip designers that there will not ever be a parisc
549 * chip with a larger alias boundary (Never say never :-) ).
550 *
551 * Subtle: the dtlb miss handlers support the temp alias region by
552 * "knowing" that if a dtlb miss happens within the temp alias
553 * region it must have occurred while in clear_user_page. Since
554 * this routine makes use of processor local translations, we
555 * don't want to insert them into the kernel page table. Instead,
556 * we load up some general registers (they need to be registers
557 * which aren't shadowed) with the physical page numbers (preshifted
558 * for tlb insertion) needed to insert the translations. When we
559 * miss on the translation, the dtlb miss handler inserts the
560 * translation into the tlb using these values:
561 *
562 * %r26 physical page (shifted for tlb insert) of "to" translation
563 * %r23 physical page (shifted for tlb insert) of "from" translation
564 */
565
566 /*
567 * We can't do this since copy_user_page is used to bring in
568 * file data that might have instructions. Since the data would
569 * then need to be flushed out so the i-fetch can see it, it
570 * makes more sense to just copy through the kernel translation
571 * and flush it.
572 *
573 * I'm still keeping this around because it may be possible to
574 * use it if more information is passed into copy_user_page().
575 * Have to do some measurements to see if it is worthwhile to
576 * lobby for such a change.
577 *
578 */
579
580 ENTRY(copy_user_page_asm)
581 .proc
582 .callinfo NO_CALLS
583 .entry
584
585 /* Convert virtual `to' and `from' addresses to physical addresses.
586 Move `from' physical address to non shadowed register. */
587 ldil L%(__PAGE_OFFSET), %r1
588 sub %r26, %r1, %r26
589 sub %r25, %r1, %r23
590
591 ldil L%(TMPALIAS_MAP_START), %r28
592 /* FIXME for different page sizes != 4k */
593 #ifdef CONFIG_64BIT
594 #if (TMPALIAS_MAP_START >= 0x80000000)
595 depdi 0, 31,32, %r28 /* clear any sign extension */
596 #endif
597 extrd,u %r26,56,32, %r26 /* convert phys addr to tlb insert format */
598 extrd,u %r23,56,32, %r23 /* convert phys addr to tlb insert format */
599 depd %r24,63,22, %r28 /* Form aliased virtual address 'to' */
600 depdi 0, 63,12, %r28 /* Clear any offset bits */
601 copy %r28, %r29
602 depdi 1, 41,1, %r29 /* Form aliased virtual address 'from' */
603 #else
604 extrw,u %r26, 24,25, %r26 /* convert phys addr to tlb insert format */
605 extrw,u %r23, 24,25, %r23 /* convert phys addr to tlb insert format */
606 depw %r24, 31,22, %r28 /* Form aliased virtual address 'to' */
607 depwi 0, 31,12, %r28 /* Clear any offset bits */
608 copy %r28, %r29
609 depwi 1, 9,1, %r29 /* Form aliased virtual address 'from' */
610 #endif
611
612 /* Purge any old translations */
613
614 #ifdef CONFIG_PA20
615 pdtlb,l 0(%r28)
616 pdtlb,l 0(%r29)
617 #else
618 tlb_lock %r20,%r21,%r22
619 pdtlb 0(%r28)
620 pdtlb 0(%r29)
621 tlb_unlock %r20,%r21,%r22
622 #endif
623
624 #ifdef CONFIG_64BIT
625 /* PA8x00 CPUs can consume 2 loads or 1 store per cycle.
626 * Unroll the loop by hand and arrange insn appropriately.
627 * GCC probably can do this just as well.
628 */
629
630 ldd 0(%r29), %r19
631 ldi (PAGE_SIZE / 128), %r1
632
633 1: ldd 8(%r29), %r20
634
635 ldd 16(%r29), %r21
636 ldd 24(%r29), %r22
637 std %r19, 0(%r28)
638 std %r20, 8(%r28)
639
640 ldd 32(%r29), %r19
641 ldd 40(%r29), %r20
642 std %r21, 16(%r28)
643 std %r22, 24(%r28)
644
645 ldd 48(%r29), %r21
646 ldd 56(%r29), %r22
647 std %r19, 32(%r28)
648 std %r20, 40(%r28)
649
650 ldd 64(%r29), %r19
651 ldd 72(%r29), %r20
652 std %r21, 48(%r28)
653 std %r22, 56(%r28)
654
655 ldd 80(%r29), %r21
656 ldd 88(%r29), %r22
657 std %r19, 64(%r28)
658 std %r20, 72(%r28)
659
660 ldd 96(%r29), %r19
661 ldd 104(%r29), %r20
662 std %r21, 80(%r28)
663 std %r22, 88(%r28)
664
665 ldd 112(%r29), %r21
666 ldd 120(%r29), %r22
667 std %r19, 96(%r28)
668 std %r20, 104(%r28)
669
670 ldo 128(%r29), %r29
671 std %r21, 112(%r28)
672 std %r22, 120(%r28)
673 ldo 128(%r28), %r28
674
675 /* conditional branches nullify on forward taken branch, and on
676 * non-taken backward branch. Note that .+4 is a backwards branch.
677 * The ldd should only get executed if the branch is taken.
678 */
679 addib,COND(>),n -1, %r1, 1b /* bundle 10 */
680 ldd 0(%r29), %r19 /* start next loads */
681
682 #else
683 ldi (PAGE_SIZE / 64), %r1
684
685 /*
686 * This loop is optimized for PCXL/PCXL2 ldw/ldw and stw/stw
687 * bundles (very restricted rules for bundling). It probably
688 * does OK on PCXU and better, but we could do better with
689 * ldd/std instructions. Note that until (if) we start saving
690 * the full 64 bit register values on interrupt, we can't
691 * use ldd/std on a 32 bit kernel.
692 */
693
694 1: ldw 0(%r29), %r19
695 ldw 4(%r29), %r20
696 ldw 8(%r29), %r21
697 ldw 12(%r29), %r22
698 stw %r19, 0(%r28)
699 stw %r20, 4(%r28)
700 stw %r21, 8(%r28)
701 stw %r22, 12(%r28)
702 ldw 16(%r29), %r19
703 ldw 20(%r29), %r20
704 ldw 24(%r29), %r21
705 ldw 28(%r29), %r22
706 stw %r19, 16(%r28)
707 stw %r20, 20(%r28)
708 stw %r21, 24(%r28)
709 stw %r22, 28(%r28)
710 ldw 32(%r29), %r19
711 ldw 36(%r29), %r20
712 ldw 40(%r29), %r21
713 ldw 44(%r29), %r22
714 stw %r19, 32(%r28)
715 stw %r20, 36(%r28)
716 stw %r21, 40(%r28)
717 stw %r22, 44(%r28)
718 ldw 48(%r29), %r19
719 ldw 52(%r29), %r20
720 ldw 56(%r29), %r21
721 ldw 60(%r29), %r22
722 stw %r19, 48(%r28)
723 stw %r20, 52(%r28)
724 stw %r21, 56(%r28)
725 stw %r22, 60(%r28)
726 ldo 64(%r28), %r28
727
728 addib,COND(>) -1, %r1,1b
729 ldo 64(%r29), %r29
730 #endif
731
732 bv %r0(%r2)
733 nop
734 .exit
735
736 .procend
737 ENDPROC(copy_user_page_asm)
738
739 ENTRY(clear_user_page_asm)
740 .proc
741 .callinfo NO_CALLS
742 .entry
743
744 tophys_r1 %r26
745
746 ldil L%(TMPALIAS_MAP_START), %r28
747 #ifdef CONFIG_64BIT
748 #if (TMPALIAS_MAP_START >= 0x80000000)
749 depdi 0, 31,32, %r28 /* clear any sign extension */
750 /* FIXME: page size dependend */
751 #endif
752 extrd,u %r26, 56,32, %r26 /* convert phys addr to tlb insert format */
753 depd %r25, 63,22, %r28 /* Form aliased virtual address 'to' */
754 depdi 0, 63,12, %r28 /* Clear any offset bits */
755 #else
756 extrw,u %r26, 24,25, %r26 /* convert phys addr to tlb insert format */
757 depw %r25, 31,22, %r28 /* Form aliased virtual address 'to' */
758 depwi 0, 31,12, %r28 /* Clear any offset bits */
759 #endif
760
761 /* Purge any old translation */
762
763 #ifdef CONFIG_PA20
764 pdtlb,l 0(%r28)
765 #else
766 tlb_lock %r20,%r21,%r22
767 pdtlb 0(%r28)
768 tlb_unlock %r20,%r21,%r22
769 #endif
770
771 #ifdef CONFIG_64BIT
772 ldi (PAGE_SIZE / 128), %r1
773
774 /* PREFETCH (Write) has not (yet) been proven to help here */
775 /* #define PREFETCHW_OP ldd 256(%0), %r0 */
776
777 1: std %r0, 0(%r28)
778 std %r0, 8(%r28)
779 std %r0, 16(%r28)
780 std %r0, 24(%r28)
781 std %r0, 32(%r28)
782 std %r0, 40(%r28)
783 std %r0, 48(%r28)
784 std %r0, 56(%r28)
785 std %r0, 64(%r28)
786 std %r0, 72(%r28)
787 std %r0, 80(%r28)
788 std %r0, 88(%r28)
789 std %r0, 96(%r28)
790 std %r0, 104(%r28)
791 std %r0, 112(%r28)
792 std %r0, 120(%r28)
793 addib,COND(>) -1, %r1, 1b
794 ldo 128(%r28), %r28
795
796 #else /* ! CONFIG_64BIT */
797 ldi (PAGE_SIZE / 64), %r1
798
799 1: stw %r0, 0(%r28)
800 stw %r0, 4(%r28)
801 stw %r0, 8(%r28)
802 stw %r0, 12(%r28)
803 stw %r0, 16(%r28)
804 stw %r0, 20(%r28)
805 stw %r0, 24(%r28)
806 stw %r0, 28(%r28)
807 stw %r0, 32(%r28)
808 stw %r0, 36(%r28)
809 stw %r0, 40(%r28)
810 stw %r0, 44(%r28)
811 stw %r0, 48(%r28)
812 stw %r0, 52(%r28)
813 stw %r0, 56(%r28)
814 stw %r0, 60(%r28)
815 addib,COND(>) -1, %r1, 1b
816 ldo 64(%r28), %r28
817 #endif /* CONFIG_64BIT */
818
819 bv %r0(%r2)
820 nop
821 .exit
822
823 .procend
824 ENDPROC(clear_user_page_asm)
825
826 ENTRY(flush_dcache_page_asm)
827 .proc
828 .callinfo NO_CALLS
829 .entry
830
831 ldil L%(TMPALIAS_MAP_START), %r28
832 #ifdef CONFIG_64BIT
833 #if (TMPALIAS_MAP_START >= 0x80000000)
834 depdi 0, 31,32, %r28 /* clear any sign extension */
835 /* FIXME: page size dependend */
836 #endif
837 extrd,u %r26, 56,32, %r26 /* convert phys addr to tlb insert format */
838 depd %r25, 63,22, %r28 /* Form aliased virtual address 'to' */
839 depdi 0, 63,12, %r28 /* Clear any offset bits */
840 #else
841 extrw,u %r26, 24,25, %r26 /* convert phys addr to tlb insert format */
842 depw %r25, 31,22, %r28 /* Form aliased virtual address 'to' */
843 depwi 0, 31,12, %r28 /* Clear any offset bits */
844 #endif
845
846 /* Purge any old translation */
847
848 #ifdef CONFIG_PA20
849 pdtlb,l 0(%r28)
850 #else
851 tlb_lock %r20,%r21,%r22
852 pdtlb 0(%r28)
853 tlb_unlock %r20,%r21,%r22
854 #endif
855
856 ldil L%dcache_stride, %r1
857 ldw R%dcache_stride(%r1), %r1
858
859 #ifdef CONFIG_64BIT
860 depdi,z 1, 63-PAGE_SHIFT,1, %r25
861 #else
862 depwi,z 1, 31-PAGE_SHIFT,1, %r25
863 #endif
864 add %r28, %r25, %r25
865 sub %r25, %r1, %r25
866
867
868 1: fdc,m %r1(%r28)
869 fdc,m %r1(%r28)
870 fdc,m %r1(%r28)
871 fdc,m %r1(%r28)
872 fdc,m %r1(%r28)
873 fdc,m %r1(%r28)
874 fdc,m %r1(%r28)
875 fdc,m %r1(%r28)
876 fdc,m %r1(%r28)
877 fdc,m %r1(%r28)
878 fdc,m %r1(%r28)
879 fdc,m %r1(%r28)
880 fdc,m %r1(%r28)
881 fdc,m %r1(%r28)
882 fdc,m %r1(%r28)
883 cmpb,COND(<<) %r28, %r25,1b
884 fdc,m %r1(%r28)
885
886 sync
887
888 #ifdef CONFIG_PA20
889 pdtlb,l 0(%r25)
890 #else
891 tlb_lock %r20,%r21,%r22
892 pdtlb 0(%r25)
893 tlb_unlock %r20,%r21,%r22
894 #endif
895
896 bv %r0(%r2)
897 nop
898 .exit
899
900 .procend
901 ENDPROC(flush_dcache_page_asm)
902
903 ENTRY(flush_icache_page_asm)
904 .proc
905 .callinfo NO_CALLS
906 .entry
907
908 ldil L%(TMPALIAS_MAP_START), %r28
909 #ifdef CONFIG_64BIT
910 #if (TMPALIAS_MAP_START >= 0x80000000)
911 depdi 0, 31,32, %r28 /* clear any sign extension */
912 /* FIXME: page size dependend */
913 #endif
914 extrd,u %r26, 56,32, %r26 /* convert phys addr to tlb insert format */
915 depd %r25, 63,22, %r28 /* Form aliased virtual address 'to' */
916 depdi 0, 63,12, %r28 /* Clear any offset bits */
917 #else
918 extrw,u %r26, 24,25, %r26 /* convert phys addr to tlb insert format */
919 depw %r25, 31,22, %r28 /* Form aliased virtual address 'to' */
920 depwi 0, 31,12, %r28 /* Clear any offset bits */
921 #endif
922
923 /* Purge any old translation */
924
925 #ifdef CONFIG_PA20
926 pitlb,l %r0(%sr4,%r28)
927 #else
928 tlb_lock %r20,%r21,%r22
929 pitlb (%sr4,%r28)
930 tlb_unlock %r20,%r21,%r22
931 #endif
932
933 ldil L%icache_stride, %r1
934 ldw R%icache_stride(%r1), %r1
935
936 #ifdef CONFIG_64BIT
937 depdi,z 1, 63-PAGE_SHIFT,1, %r25
938 #else
939 depwi,z 1, 31-PAGE_SHIFT,1, %r25
940 #endif
941 add %r28, %r25, %r25
942 sub %r25, %r1, %r25
943
944
945 /* fic only has the type 26 form on PA1.1, requiring an
946 * explicit space specification, so use %sr4 */
947 1: fic,m %r1(%sr4,%r28)
948 fic,m %r1(%sr4,%r28)
949 fic,m %r1(%sr4,%r28)
950 fic,m %r1(%sr4,%r28)
951 fic,m %r1(%sr4,%r28)
952 fic,m %r1(%sr4,%r28)
953 fic,m %r1(%sr4,%r28)
954 fic,m %r1(%sr4,%r28)
955 fic,m %r1(%sr4,%r28)
956 fic,m %r1(%sr4,%r28)
957 fic,m %r1(%sr4,%r28)
958 fic,m %r1(%sr4,%r28)
959 fic,m %r1(%sr4,%r28)
960 fic,m %r1(%sr4,%r28)
961 fic,m %r1(%sr4,%r28)
962 cmpb,COND(<<) %r28, %r25,1b
963 fic,m %r1(%sr4,%r28)
964
965 sync
966
967 #ifdef CONFIG_PA20
968 pitlb,l %r0(%sr4,%r25)
969 #else
970 tlb_lock %r20,%r21,%r22
971 pitlb (%sr4,%r25)
972 tlb_unlock %r20,%r21,%r22
973 #endif
974
975 bv %r0(%r2)
976 nop
977 .exit
978
979 .procend
980 ENDPROC(flush_icache_page_asm)
981
982 ENTRY(flush_kernel_dcache_page_asm)
983 .proc
984 .callinfo NO_CALLS
985 .entry
986
987 ldil L%dcache_stride, %r1
988 ldw R%dcache_stride(%r1), %r23
989
990 #ifdef CONFIG_64BIT
991 depdi,z 1, 63-PAGE_SHIFT,1, %r25
992 #else
993 depwi,z 1, 31-PAGE_SHIFT,1, %r25
994 #endif
995 add %r26, %r25, %r25
996 sub %r25, %r23, %r25
997
998
999 1: fdc,m %r23(%r26)
1000 fdc,m %r23(%r26)
1001 fdc,m %r23(%r26)
1002 fdc,m %r23(%r26)
1003 fdc,m %r23(%r26)
1004 fdc,m %r23(%r26)
1005 fdc,m %r23(%r26)
1006 fdc,m %r23(%r26)
1007 fdc,m %r23(%r26)
1008 fdc,m %r23(%r26)
1009 fdc,m %r23(%r26)
1010 fdc,m %r23(%r26)
1011 fdc,m %r23(%r26)
1012 fdc,m %r23(%r26)
1013 fdc,m %r23(%r26)
1014 cmpb,COND(<<) %r26, %r25,1b
1015 fdc,m %r23(%r26)
1016
1017 sync
1018 bv %r0(%r2)
1019 nop
1020 .exit
1021
1022 .procend
1023 ENDPROC(flush_kernel_dcache_page_asm)
1024
1025 ENTRY(purge_kernel_dcache_page_asm)
1026 .proc
1027 .callinfo NO_CALLS
1028 .entry
1029
1030 ldil L%dcache_stride, %r1
1031 ldw R%dcache_stride(%r1), %r23
1032
1033 #ifdef CONFIG_64BIT
1034 depdi,z 1, 63-PAGE_SHIFT,1, %r25
1035 #else
1036 depwi,z 1, 31-PAGE_SHIFT,1, %r25
1037 #endif
1038 add %r26, %r25, %r25
1039 sub %r25, %r23, %r25
1040
1041 1: pdc,m %r23(%r26)
1042 pdc,m %r23(%r26)
1043 pdc,m %r23(%r26)
1044 pdc,m %r23(%r26)
1045 pdc,m %r23(%r26)
1046 pdc,m %r23(%r26)
1047 pdc,m %r23(%r26)
1048 pdc,m %r23(%r26)
1049 pdc,m %r23(%r26)
1050 pdc,m %r23(%r26)
1051 pdc,m %r23(%r26)
1052 pdc,m %r23(%r26)
1053 pdc,m %r23(%r26)
1054 pdc,m %r23(%r26)
1055 pdc,m %r23(%r26)
1056 cmpb,COND(<<) %r26, %r25, 1b
1057 pdc,m %r23(%r26)
1058
1059 sync
1060 bv %r0(%r2)
1061 nop
1062 .exit
1063
1064 .procend
1065 ENDPROC(purge_kernel_dcache_page_asm)
1066
1067 ENTRY(flush_user_dcache_range_asm)
1068 .proc
1069 .callinfo NO_CALLS
1070 .entry
1071
1072 ldil L%dcache_stride, %r1
1073 ldw R%dcache_stride(%r1), %r23
1074 ldo -1(%r23), %r21
1075 ANDCM %r26, %r21, %r26
1076
1077 1: cmpb,COND(<<),n %r26, %r25, 1b
1078 fdc,m %r23(%sr3, %r26)
1079
1080 sync
1081 bv %r0(%r2)
1082 nop
1083 .exit
1084
1085 .procend
1086 ENDPROC(flush_user_dcache_range_asm)
1087
1088 ENTRY(flush_kernel_dcache_range_asm)
1089 .proc
1090 .callinfo NO_CALLS
1091 .entry
1092
1093 ldil L%dcache_stride, %r1
1094 ldw R%dcache_stride(%r1), %r23
1095 ldo -1(%r23), %r21
1096 ANDCM %r26, %r21, %r26
1097
1098 1: cmpb,COND(<<),n %r26, %r25,1b
1099 fdc,m %r23(%r26)
1100
1101 sync
1102 syncdma
1103 bv %r0(%r2)
1104 nop
1105 .exit
1106
1107 .procend
1108 ENDPROC(flush_kernel_dcache_range_asm)
1109
1110 ENTRY(flush_user_icache_range_asm)
1111 .proc
1112 .callinfo NO_CALLS
1113 .entry
1114
1115 ldil L%icache_stride, %r1
1116 ldw R%icache_stride(%r1), %r23
1117 ldo -1(%r23), %r21
1118 ANDCM %r26, %r21, %r26
1119
1120 1: cmpb,COND(<<),n %r26, %r25,1b
1121 fic,m %r23(%sr3, %r26)
1122
1123 sync
1124 bv %r0(%r2)
1125 nop
1126 .exit
1127
1128 .procend
1129 ENDPROC(flush_user_icache_range_asm)
1130
1131 ENTRY(flush_kernel_icache_page)
1132 .proc
1133 .callinfo NO_CALLS
1134 .entry
1135
1136 ldil L%icache_stride, %r1
1137 ldw R%icache_stride(%r1), %r23
1138
1139 #ifdef CONFIG_64BIT
1140 depdi,z 1, 63-PAGE_SHIFT,1, %r25
1141 #else
1142 depwi,z 1, 31-PAGE_SHIFT,1, %r25
1143 #endif
1144 add %r26, %r25, %r25
1145 sub %r25, %r23, %r25
1146
1147
1148 1: fic,m %r23(%sr4, %r26)
1149 fic,m %r23(%sr4, %r26)
1150 fic,m %r23(%sr4, %r26)
1151 fic,m %r23(%sr4, %r26)
1152 fic,m %r23(%sr4, %r26)
1153 fic,m %r23(%sr4, %r26)
1154 fic,m %r23(%sr4, %r26)
1155 fic,m %r23(%sr4, %r26)
1156 fic,m %r23(%sr4, %r26)
1157 fic,m %r23(%sr4, %r26)
1158 fic,m %r23(%sr4, %r26)
1159 fic,m %r23(%sr4, %r26)
1160 fic,m %r23(%sr4, %r26)
1161 fic,m %r23(%sr4, %r26)
1162 fic,m %r23(%sr4, %r26)
1163 cmpb,COND(<<) %r26, %r25, 1b
1164 fic,m %r23(%sr4, %r26)
1165
1166 sync
1167 bv %r0(%r2)
1168 nop
1169 .exit
1170
1171 .procend
1172 ENDPROC(flush_kernel_icache_page)
1173
1174 ENTRY(flush_kernel_icache_range_asm)
1175 .proc
1176 .callinfo NO_CALLS
1177 .entry
1178
1179 ldil L%icache_stride, %r1
1180 ldw R%icache_stride(%r1), %r23
1181 ldo -1(%r23), %r21
1182 ANDCM %r26, %r21, %r26
1183
1184 1: cmpb,COND(<<),n %r26, %r25, 1b
1185 fic,m %r23(%sr4, %r26)
1186
1187 sync
1188 bv %r0(%r2)
1189 nop
1190 .exit
1191 .procend
1192 ENDPROC(flush_kernel_icache_range_asm)
1193
1194 /* align should cover use of rfi in disable_sr_hashing_asm and
1195 * srdis_done.
1196 */
1197 .align 256
1198 ENTRY(disable_sr_hashing_asm)
1199 .proc
1200 .callinfo NO_CALLS
1201 .entry
1202
1203 /*
1204 * Switch to real mode
1205 */
1206 /* pcxt_ssm_bug */
1207 rsm PSW_SM_I, %r0
1208 load32 PA(1f), %r1
1209 nop
1210 nop
1211 nop
1212 nop
1213 nop
1214
1215 rsm PSW_SM_Q, %r0 /* prep to load iia queue */
1216 mtctl %r0, %cr17 /* Clear IIASQ tail */
1217 mtctl %r0, %cr17 /* Clear IIASQ head */
1218 mtctl %r1, %cr18 /* IIAOQ head */
1219 ldo 4(%r1), %r1
1220 mtctl %r1, %cr18 /* IIAOQ tail */
1221 load32 REAL_MODE_PSW, %r1
1222 mtctl %r1, %ipsw
1223 rfi
1224 nop
1225
1226 1: cmpib,=,n SRHASH_PCXST, %r26,srdis_pcxs
1227 cmpib,=,n SRHASH_PCXL, %r26,srdis_pcxl
1228 cmpib,=,n SRHASH_PA20, %r26,srdis_pa20
1229 b,n srdis_done
1230
1231 srdis_pcxs:
1232
1233 /* Disable Space Register Hashing for PCXS,PCXT,PCXT' */
1234
1235 .word 0x141c1a00 /* mfdiag %dr0, %r28 */
1236 .word 0x141c1a00 /* must issue twice */
1237 depwi 0,18,1, %r28 /* Clear DHE (dcache hash enable) */
1238 depwi 0,20,1, %r28 /* Clear IHE (icache hash enable) */
1239 .word 0x141c1600 /* mtdiag %r28, %dr0 */
1240 .word 0x141c1600 /* must issue twice */
1241 b,n srdis_done
1242
1243 srdis_pcxl:
1244
1245 /* Disable Space Register Hashing for PCXL */
1246
1247 .word 0x141c0600 /* mfdiag %dr0, %r28 */
1248 depwi 0,28,2, %r28 /* Clear DHASH_EN & IHASH_EN */
1249 .word 0x141c0240 /* mtdiag %r28, %dr0 */
1250 b,n srdis_done
1251
1252 srdis_pa20:
1253
1254 /* Disable Space Register Hashing for PCXU,PCXU+,PCXW,PCXW+,PCXW2 */
1255
1256 .word 0x144008bc /* mfdiag %dr2, %r28 */
1257 depdi 0, 54,1, %r28 /* clear DIAG_SPHASH_ENAB (bit 54) */
1258 .word 0x145c1840 /* mtdiag %r28, %dr2 */
1259
1260
1261 srdis_done:
1262 /* Switch back to virtual mode */
1263 rsm PSW_SM_I, %r0 /* prep to load iia queue */
1264 load32 2f, %r1
1265 nop
1266 nop
1267 nop
1268 nop
1269 nop
1270
1271 rsm PSW_SM_Q, %r0 /* prep to load iia queue */
1272 mtctl %r0, %cr17 /* Clear IIASQ tail */
1273 mtctl %r0, %cr17 /* Clear IIASQ head */
1274 mtctl %r1, %cr18 /* IIAOQ head */
1275 ldo 4(%r1), %r1
1276 mtctl %r1, %cr18 /* IIAOQ tail */
1277 load32 KERNEL_PSW, %r1
1278 mtctl %r1, %ipsw
1279 rfi
1280 nop
1281
1282 2: bv %r0(%r2)
1283 nop
1284 .exit
1285
1286 .procend
1287 ENDPROC(disable_sr_hashing_asm)
1288
1289 .end
This page took 0.068326 seconds and 5 git commands to generate.