Merge /spare/repo/linux-2.6/
[linux-2.6] / arch / sparc64 / kernel / unaligned.c
1 /* $Id: unaligned.c,v 1.24 2002/02/09 19:49:31 davem Exp $
2  * unaligned.c: Unaligned load/store trap handling with special
3  *              cases for the kernel to do them more quickly.
4  *
5  * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu)
6  * Copyright (C) 1996,1997 Jakub Jelinek (jj@sunsite.mff.cuni.cz)
7  */
8
9
10 #include <linux/kernel.h>
11 #include <linux/sched.h>
12 #include <linux/mm.h>
13 #include <linux/module.h>
14 #include <asm/asi.h>
15 #include <asm/ptrace.h>
16 #include <asm/pstate.h>
17 #include <asm/processor.h>
18 #include <asm/system.h>
19 #include <asm/uaccess.h>
20 #include <linux/smp.h>
21 #include <linux/smp_lock.h>
22 #include <linux/bitops.h>
23 #include <asm/fpumacro.h>
24
25 /* #define DEBUG_MNA */
26
27 enum direction {
28         load,    /* ld, ldd, ldh, ldsh */
29         store,   /* st, std, sth, stsh */
30         both,    /* Swap, ldstub, cas, ... */
31         fpld,
32         fpst,
33         invalid,
34 };
35
36 #ifdef DEBUG_MNA
37 static char *dirstrings[] = {
38   "load", "store", "both", "fpload", "fpstore", "invalid"
39 };
40 #endif
41
42 static inline enum direction decode_direction(unsigned int insn)
43 {
44         unsigned long tmp = (insn >> 21) & 1;
45
46         if (!tmp)
47                 return load;
48         else {
49                 switch ((insn>>19)&0xf) {
50                 case 15: /* swap* */
51                         return both;
52                 default:
53                         return store;
54                 }
55         }
56 }
57
58 /* 16 = double-word, 8 = extra-word, 4 = word, 2 = half-word */
59 static inline int decode_access_size(unsigned int insn)
60 {
61         unsigned int tmp;
62
63         tmp = ((insn >> 19) & 0xf);
64         if (tmp == 11 || tmp == 14) /* ldx/stx */
65                 return 8;
66         tmp &= 3;
67         if (!tmp)
68                 return 4;
69         else if (tmp == 3)
70                 return 16;      /* ldd/std - Although it is actually 8 */
71         else if (tmp == 2)
72                 return 2;
73         else {
74                 printk("Impossible unaligned trap. insn=%08x\n", insn);
75                 die_if_kernel("Byte sized unaligned access?!?!", current_thread_info()->kregs);
76
77                 /* GCC should never warn that control reaches the end
78                  * of this function without returning a value because
79                  * die_if_kernel() is marked with attribute 'noreturn'.
80                  * Alas, some versions do...
81                  */
82
83                 return 0;
84         }
85 }
86
87 static inline int decode_asi(unsigned int insn, struct pt_regs *regs)
88 {
89         if (insn & 0x800000) {
90                 if (insn & 0x2000)
91                         return (unsigned char)(regs->tstate >> 24);     /* %asi */
92                 else
93                         return (unsigned char)(insn >> 5);              /* imm_asi */
94         } else
95                 return ASI_P;
96 }
97
98 /* 0x400000 = signed, 0 = unsigned */
99 static inline int decode_signedness(unsigned int insn)
100 {
101         return (insn & 0x400000);
102 }
103
104 static inline void maybe_flush_windows(unsigned int rs1, unsigned int rs2,
105                                        unsigned int rd, int from_kernel)
106 {
107         if (rs2 >= 16 || rs1 >= 16 || rd >= 16) {
108                 if (from_kernel != 0)
109                         __asm__ __volatile__("flushw");
110                 else
111                         flushw_user();
112         }
113 }
114
115 static inline long sign_extend_imm13(long imm)
116 {
117         return imm << 51 >> 51;
118 }
119
120 static unsigned long fetch_reg(unsigned int reg, struct pt_regs *regs)
121 {
122         unsigned long value;
123         
124         if (reg < 16)
125                 return (!reg ? 0 : regs->u_regs[reg]);
126         if (regs->tstate & TSTATE_PRIV) {
127                 struct reg_window *win;
128                 win = (struct reg_window *)(regs->u_regs[UREG_FP] + STACK_BIAS);
129                 value = win->locals[reg - 16];
130         } else if (test_thread_flag(TIF_32BIT)) {
131                 struct reg_window32 __user *win32;
132                 win32 = (struct reg_window32 __user *)((unsigned long)((u32)regs->u_regs[UREG_FP]));
133                 get_user(value, &win32->locals[reg - 16]);
134         } else {
135                 struct reg_window __user *win;
136                 win = (struct reg_window __user *)(regs->u_regs[UREG_FP] + STACK_BIAS);
137                 get_user(value, &win->locals[reg - 16]);
138         }
139         return value;
140 }
141
142 static unsigned long *fetch_reg_addr(unsigned int reg, struct pt_regs *regs)
143 {
144         if (reg < 16)
145                 return &regs->u_regs[reg];
146         if (regs->tstate & TSTATE_PRIV) {
147                 struct reg_window *win;
148                 win = (struct reg_window *)(regs->u_regs[UREG_FP] + STACK_BIAS);
149                 return &win->locals[reg - 16];
150         } else if (test_thread_flag(TIF_32BIT)) {
151                 struct reg_window32 *win32;
152                 win32 = (struct reg_window32 *)((unsigned long)((u32)regs->u_regs[UREG_FP]));
153                 return (unsigned long *)&win32->locals[reg - 16];
154         } else {
155                 struct reg_window *win;
156                 win = (struct reg_window *)(regs->u_regs[UREG_FP] + STACK_BIAS);
157                 return &win->locals[reg - 16];
158         }
159 }
160
161 unsigned long compute_effective_address(struct pt_regs *regs,
162                                         unsigned int insn, unsigned int rd)
163 {
164         unsigned int rs1 = (insn >> 14) & 0x1f;
165         unsigned int rs2 = insn & 0x1f;
166         int from_kernel = (regs->tstate & TSTATE_PRIV) != 0;
167
168         if (insn & 0x2000) {
169                 maybe_flush_windows(rs1, 0, rd, from_kernel);
170                 return (fetch_reg(rs1, regs) + sign_extend_imm13(insn));
171         } else {
172                 maybe_flush_windows(rs1, rs2, rd, from_kernel);
173                 return (fetch_reg(rs1, regs) + fetch_reg(rs2, regs));
174         }
175 }
176
177 /* This is just to make gcc think die_if_kernel does return... */
178 static void __attribute_used__ unaligned_panic(char *str, struct pt_regs *regs)
179 {
180         die_if_kernel(str, regs);
181 }
182
183 extern void do_int_load(unsigned long *dest_reg, int size,
184                         unsigned long *saddr, int is_signed, int asi);
185         
186 extern void __do_int_store(unsigned long *dst_addr, int size,
187                            unsigned long src_val, int asi);
188
189 static inline void do_int_store(int reg_num, int size, unsigned long *dst_addr,
190                                 struct pt_regs *regs, int asi, int orig_asi)
191 {
192         unsigned long zero = 0;
193         unsigned long *src_val_p = &zero;
194         unsigned long src_val;
195
196         if (size == 16) {
197                 size = 8;
198                 zero = (((long)(reg_num ?
199                         (unsigned)fetch_reg(reg_num, regs) : 0)) << 32) |
200                         (unsigned)fetch_reg(reg_num + 1, regs);
201         } else if (reg_num) {
202                 src_val_p = fetch_reg_addr(reg_num, regs);
203         }
204         src_val = *src_val_p;
205         if (unlikely(asi != orig_asi)) {
206                 switch (size) {
207                 case 2:
208                         src_val = swab16(src_val);
209                         break;
210                 case 4:
211                         src_val = swab32(src_val);
212                         break;
213                 case 8:
214                         src_val = swab64(src_val);
215                         break;
216                 case 16:
217                 default:
218                         BUG();
219                         break;
220                 };
221         }
222         __do_int_store(dst_addr, size, src_val, asi);
223 }
224
225 static inline void advance(struct pt_regs *regs)
226 {
227         regs->tpc   = regs->tnpc;
228         regs->tnpc += 4;
229         if (test_thread_flag(TIF_32BIT)) {
230                 regs->tpc &= 0xffffffff;
231                 regs->tnpc &= 0xffffffff;
232         }
233 }
234
235 static inline int floating_point_load_or_store_p(unsigned int insn)
236 {
237         return (insn >> 24) & 1;
238 }
239
240 static inline int ok_for_kernel(unsigned int insn)
241 {
242         return !floating_point_load_or_store_p(insn);
243 }
244
245 void kernel_mna_trap_fault(void)
246 {
247         struct pt_regs *regs = current_thread_info()->kern_una_regs;
248         unsigned int insn = current_thread_info()->kern_una_insn;
249         unsigned long g2 = regs->u_regs[UREG_G2];
250         unsigned long fixup = search_extables_range(regs->tpc, &g2);
251
252         if (!fixup) {
253                 unsigned long address;
254
255                 address = compute_effective_address(regs, insn,
256                                                     ((insn >> 25) & 0x1f));
257                 if (address < PAGE_SIZE) {
258                         printk(KERN_ALERT "Unable to handle kernel NULL "
259                                "pointer dereference in mna handler");
260                 } else
261                         printk(KERN_ALERT "Unable to handle kernel paging "
262                                "request in mna handler");
263                 printk(KERN_ALERT " at virtual address %016lx\n",address);
264                 printk(KERN_ALERT "current->{active_,}mm->context = %016lx\n",
265                         (current->mm ? CTX_HWBITS(current->mm->context) :
266                         CTX_HWBITS(current->active_mm->context)));
267                 printk(KERN_ALERT "current->{active_,}mm->pgd = %016lx\n",
268                         (current->mm ? (unsigned long) current->mm->pgd :
269                         (unsigned long) current->active_mm->pgd));
270                 die_if_kernel("Oops", regs);
271                 /* Not reached */
272         }
273         regs->tpc = fixup;
274         regs->tnpc = regs->tpc + 4;
275         regs->u_regs [UREG_G2] = g2;
276
277         regs->tstate &= ~TSTATE_ASI;
278         regs->tstate |= (ASI_AIUS << 24UL);
279 }
280
281 asmlinkage void kernel_unaligned_trap(struct pt_regs *regs, unsigned int insn, unsigned long sfar, unsigned long sfsr)
282 {
283         enum direction dir = decode_direction(insn);
284         int size = decode_access_size(insn);
285
286         current_thread_info()->kern_una_regs = regs;
287         current_thread_info()->kern_una_insn = insn;
288
289         if (!ok_for_kernel(insn) || dir == both) {
290                 printk("Unsupported unaligned load/store trap for kernel "
291                        "at <%016lx>.\n", regs->tpc);
292                 unaligned_panic("Kernel does fpu/atomic "
293                                 "unaligned load/store.", regs);
294
295                 kernel_mna_trap_fault();
296         } else {
297                 unsigned long addr;
298                 int orig_asi, asi;
299
300                 addr = compute_effective_address(regs, insn,
301                                                  ((insn >> 25) & 0x1f));
302 #ifdef DEBUG_MNA
303                 printk("KMNA: pc=%016lx [dir=%s addr=%016lx size=%d] "
304                        "retpc[%016lx]\n",
305                        regs->tpc, dirstrings[dir], addr, size,
306                        regs->u_regs[UREG_RETPC]);
307 #endif
308                 orig_asi = asi = decode_asi(insn, regs);
309                 switch (asi) {
310                 case ASI_NL:
311                 case ASI_AIUPL:
312                 case ASI_AIUSL:
313                 case ASI_PL:
314                 case ASI_SL:
315                 case ASI_PNFL:
316                 case ASI_SNFL:
317                         asi &= ~0x08;
318                         break;
319                 };
320                 switch (dir) {
321                 case load:
322                         do_int_load(fetch_reg_addr(((insn>>25)&0x1f), regs),
323                                     size, (unsigned long *) addr,
324                                     decode_signedness(insn), asi);
325                         if (unlikely(asi != orig_asi)) {
326                                 unsigned long val_in = *(unsigned long *) addr;
327                                 switch (size) {
328                                 case 2:
329                                         val_in = swab16(val_in);
330                                         break;
331                                 case 4:
332                                         val_in = swab32(val_in);
333                                         break;
334                                 case 8:
335                                         val_in = swab64(val_in);
336                                         break;
337                                 case 16:
338                                 default:
339                                         BUG();
340                                         break;
341                                 };
342                                 *(unsigned long *) addr = val_in;
343                         }
344                         break;
345
346                 case store:
347                         do_int_store(((insn>>25)&0x1f), size,
348                                      (unsigned long *) addr, regs,
349                                      asi, orig_asi);
350                         break;
351
352                 default:
353                         panic("Impossible kernel unaligned trap.");
354                         /* Not reached... */
355                 }
356                 advance(regs);
357         }
358 }
359
360 static char popc_helper[] = {
361 0, 1, 1, 2, 1, 2, 2, 3,
362 1, 2, 2, 3, 2, 3, 3, 4, 
363 };
364
365 int handle_popc(u32 insn, struct pt_regs *regs)
366 {
367         u64 value;
368         int ret, i, rd = ((insn >> 25) & 0x1f);
369         int from_kernel = (regs->tstate & TSTATE_PRIV) != 0;
370                                 
371         if (insn & 0x2000) {
372                 maybe_flush_windows(0, 0, rd, from_kernel);
373                 value = sign_extend_imm13(insn);
374         } else {
375                 maybe_flush_windows(0, insn & 0x1f, rd, from_kernel);
376                 value = fetch_reg(insn & 0x1f, regs);
377         }
378         for (ret = 0, i = 0; i < 16; i++) {
379                 ret += popc_helper[value & 0xf];
380                 value >>= 4;
381         }
382         if (rd < 16) {
383                 if (rd)
384                         regs->u_regs[rd] = ret;
385         } else {
386                 if (test_thread_flag(TIF_32BIT)) {
387                         struct reg_window32 __user *win32;
388                         win32 = (struct reg_window32 __user *)((unsigned long)((u32)regs->u_regs[UREG_FP]));
389                         put_user(ret, &win32->locals[rd - 16]);
390                 } else {
391                         struct reg_window __user *win;
392                         win = (struct reg_window __user *)(regs->u_regs[UREG_FP] + STACK_BIAS);
393                         put_user(ret, &win->locals[rd - 16]);
394                 }
395         }
396         advance(regs);
397         return 1;
398 }
399
400 extern void do_fpother(struct pt_regs *regs);
401 extern void do_privact(struct pt_regs *regs);
402 extern void spitfire_data_access_exception(struct pt_regs *regs,
403                                            unsigned long sfsr,
404                                            unsigned long sfar);
405
406 int handle_ldf_stq(u32 insn, struct pt_regs *regs)
407 {
408         unsigned long addr = compute_effective_address(regs, insn, 0);
409         int freg = ((insn >> 25) & 0x1e) | ((insn >> 20) & 0x20);
410         struct fpustate *f = FPUSTATE;
411         int asi = decode_asi(insn, regs);
412         int flag = (freg < 32) ? FPRS_DL : FPRS_DU;
413
414         save_and_clear_fpu();
415         current_thread_info()->xfsr[0] &= ~0x1c000;
416         if (freg & 3) {
417                 current_thread_info()->xfsr[0] |= (6 << 14) /* invalid_fp_register */;
418                 do_fpother(regs);
419                 return 0;
420         }
421         if (insn & 0x200000) {
422                 /* STQ */
423                 u64 first = 0, second = 0;
424                 
425                 if (current_thread_info()->fpsaved[0] & flag) {
426                         first = *(u64 *)&f->regs[freg];
427                         second = *(u64 *)&f->regs[freg+2];
428                 }
429                 if (asi < 0x80) {
430                         do_privact(regs);
431                         return 1;
432                 }
433                 switch (asi) {
434                 case ASI_P:
435                 case ASI_S: break;
436                 case ASI_PL:
437                 case ASI_SL: 
438                         {
439                                 /* Need to convert endians */
440                                 u64 tmp = __swab64p(&first);
441                                 
442                                 first = __swab64p(&second);
443                                 second = tmp;
444                                 break;
445                         }
446                 default:
447                         spitfire_data_access_exception(regs, 0, addr);
448                         return 1;
449                 }
450                 if (put_user (first >> 32, (u32 __user *)addr) ||
451                     __put_user ((u32)first, (u32 __user *)(addr + 4)) ||
452                     __put_user (second >> 32, (u32 __user *)(addr + 8)) ||
453                     __put_user ((u32)second, (u32 __user *)(addr + 12))) {
454                         spitfire_data_access_exception(regs, 0, addr);
455                         return 1;
456                 }
457         } else {
458                 /* LDF, LDDF, LDQF */
459                 u32 data[4] __attribute__ ((aligned(8)));
460                 int size, i;
461                 int err;
462
463                 if (asi < 0x80) {
464                         do_privact(regs);
465                         return 1;
466                 } else if (asi > ASI_SNFL) {
467                         spitfire_data_access_exception(regs, 0, addr);
468                         return 1;
469                 }
470                 switch (insn & 0x180000) {
471                 case 0x000000: size = 1; break;
472                 case 0x100000: size = 4; break;
473                 default: size = 2; break;
474                 }
475                 for (i = 0; i < size; i++)
476                         data[i] = 0;
477                 
478                 err = get_user (data[0], (u32 __user *) addr);
479                 if (!err) {
480                         for (i = 1; i < size; i++)
481                                 err |= __get_user (data[i], (u32 __user *)(addr + 4*i));
482                 }
483                 if (err && !(asi & 0x2 /* NF */)) {
484                         spitfire_data_access_exception(regs, 0, addr);
485                         return 1;
486                 }
487                 if (asi & 0x8) /* Little */ {
488                         u64 tmp;
489
490                         switch (size) {
491                         case 1: data[0] = le32_to_cpup(data + 0); break;
492                         default:*(u64 *)(data + 0) = le64_to_cpup((u64 *)(data + 0));
493                                 break;
494                         case 4: tmp = le64_to_cpup((u64 *)(data + 0));
495                                 *(u64 *)(data + 0) = le64_to_cpup((u64 *)(data + 2));
496                                 *(u64 *)(data + 2) = tmp;
497                                 break;
498                         }
499                 }
500                 if (!(current_thread_info()->fpsaved[0] & FPRS_FEF)) {
501                         current_thread_info()->fpsaved[0] = FPRS_FEF;
502                         current_thread_info()->gsr[0] = 0;
503                 }
504                 if (!(current_thread_info()->fpsaved[0] & flag)) {
505                         if (freg < 32)
506                                 memset(f->regs, 0, 32*sizeof(u32));
507                         else
508                                 memset(f->regs+32, 0, 32*sizeof(u32));
509                 }
510                 memcpy(f->regs + freg, data, size * 4);
511                 current_thread_info()->fpsaved[0] |= flag;
512         }
513         advance(regs);
514         return 1;
515 }
516
517 void handle_ld_nf(u32 insn, struct pt_regs *regs)
518 {
519         int rd = ((insn >> 25) & 0x1f);
520         int from_kernel = (regs->tstate & TSTATE_PRIV) != 0;
521         unsigned long *reg;
522                                 
523         maybe_flush_windows(0, 0, rd, from_kernel);
524         reg = fetch_reg_addr(rd, regs);
525         if (from_kernel || rd < 16) {
526                 reg[0] = 0;
527                 if ((insn & 0x780000) == 0x180000)
528                         reg[1] = 0;
529         } else if (test_thread_flag(TIF_32BIT)) {
530                 put_user(0, (int __user *) reg);
531                 if ((insn & 0x780000) == 0x180000)
532                         put_user(0, ((int __user *) reg) + 1);
533         } else {
534                 put_user(0, (unsigned long __user *) reg);
535                 if ((insn & 0x780000) == 0x180000)
536                         put_user(0, (unsigned long __user *) reg + 1);
537         }
538         advance(regs);
539 }
540
541 void handle_lddfmna(struct pt_regs *regs, unsigned long sfar, unsigned long sfsr)
542 {
543         unsigned long pc = regs->tpc;
544         unsigned long tstate = regs->tstate;
545         u32 insn;
546         u32 first, second;
547         u64 value;
548         u8 asi, freg;
549         int flag;
550         struct fpustate *f = FPUSTATE;
551
552         if (tstate & TSTATE_PRIV)
553                 die_if_kernel("lddfmna from kernel", regs);
554         if (test_thread_flag(TIF_32BIT))
555                 pc = (u32)pc;
556         if (get_user(insn, (u32 __user *) pc) != -EFAULT) {
557                 asi = sfsr >> 16;
558                 if ((asi > ASI_SNFL) ||
559                     (asi < ASI_P))
560                         goto daex;
561                 if (get_user(first, (u32 __user *)sfar) ||
562                      get_user(second, (u32 __user *)(sfar + 4))) {
563                         if (asi & 0x2) /* NF */ {
564                                 first = 0; second = 0;
565                         } else
566                                 goto daex;
567                 }
568                 save_and_clear_fpu();
569                 freg = ((insn >> 25) & 0x1e) | ((insn >> 20) & 0x20);
570                 value = (((u64)first) << 32) | second;
571                 if (asi & 0x8) /* Little */
572                         value = __swab64p(&value);
573                 flag = (freg < 32) ? FPRS_DL : FPRS_DU;
574                 if (!(current_thread_info()->fpsaved[0] & FPRS_FEF)) {
575                         current_thread_info()->fpsaved[0] = FPRS_FEF;
576                         current_thread_info()->gsr[0] = 0;
577                 }
578                 if (!(current_thread_info()->fpsaved[0] & flag)) {
579                         if (freg < 32)
580                                 memset(f->regs, 0, 32*sizeof(u32));
581                         else
582                                 memset(f->regs+32, 0, 32*sizeof(u32));
583                 }
584                 *(u64 *)(f->regs + freg) = value;
585                 current_thread_info()->fpsaved[0] |= flag;
586         } else {
587 daex:           spitfire_data_access_exception(regs, sfsr, sfar);
588                 return;
589         }
590         advance(regs);
591         return;
592 }
593
594 void handle_stdfmna(struct pt_regs *regs, unsigned long sfar, unsigned long sfsr)
595 {
596         unsigned long pc = regs->tpc;
597         unsigned long tstate = regs->tstate;
598         u32 insn;
599         u64 value;
600         u8 asi, freg;
601         int flag;
602         struct fpustate *f = FPUSTATE;
603
604         if (tstate & TSTATE_PRIV)
605                 die_if_kernel("stdfmna from kernel", regs);
606         if (test_thread_flag(TIF_32BIT))
607                 pc = (u32)pc;
608         if (get_user(insn, (u32 __user *) pc) != -EFAULT) {
609                 freg = ((insn >> 25) & 0x1e) | ((insn >> 20) & 0x20);
610                 asi = sfsr >> 16;
611                 value = 0;
612                 flag = (freg < 32) ? FPRS_DL : FPRS_DU;
613                 if ((asi > ASI_SNFL) ||
614                     (asi < ASI_P))
615                         goto daex;
616                 save_and_clear_fpu();
617                 if (current_thread_info()->fpsaved[0] & flag)
618                         value = *(u64 *)&f->regs[freg];
619                 switch (asi) {
620                 case ASI_P:
621                 case ASI_S: break;
622                 case ASI_PL:
623                 case ASI_SL: 
624                         value = __swab64p(&value); break;
625                 default: goto daex;
626                 }
627                 if (put_user (value >> 32, (u32 __user *) sfar) ||
628                     __put_user ((u32)value, (u32 __user *)(sfar + 4)))
629                         goto daex;
630         } else {
631 daex:           spitfire_data_access_exception(regs, sfsr, sfar);
632                 return;
633         }
634         advance(regs);
635         return;
636 }