Merge master.kernel.org:/home/rmk/linux-2.6-arm
[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 int do_int_load(unsigned long *dest_reg, int size,
184                        unsigned long *saddr, int is_signed, int asi);
185         
186 extern int __do_int_store(unsigned long *dst_addr, int size,
187                           unsigned long src_val, int asi);
188
189 static inline int 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         return __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 static 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         const struct exception_table_entry *entry;
250
251         entry = search_exception_tables(regs->tpc);
252         if (!entry) {
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 = entry->fixup;
274         regs->tnpc = regs->tpc + 4;
275
276         regs->tstate &= ~TSTATE_ASI;
277         regs->tstate |= (ASI_AIUS << 24UL);
278 }
279
280 asmlinkage void kernel_unaligned_trap(struct pt_regs *regs, unsigned int insn, unsigned long sfar, unsigned long sfsr)
281 {
282         enum direction dir = decode_direction(insn);
283         int size = decode_access_size(insn);
284
285         current_thread_info()->kern_una_regs = regs;
286         current_thread_info()->kern_una_insn = insn;
287
288         if (!ok_for_kernel(insn) || dir == both) {
289                 printk("Unsupported unaligned load/store trap for kernel "
290                        "at <%016lx>.\n", regs->tpc);
291                 unaligned_panic("Kernel does fpu/atomic "
292                                 "unaligned load/store.", regs);
293
294                 kernel_mna_trap_fault();
295         } else {
296                 unsigned long addr, *reg_addr;
297                 int orig_asi, asi, err;
298
299                 addr = compute_effective_address(regs, insn,
300                                                  ((insn >> 25) & 0x1f));
301 #ifdef DEBUG_MNA
302                 printk("KMNA: pc=%016lx [dir=%s addr=%016lx size=%d] "
303                        "retpc[%016lx]\n",
304                        regs->tpc, dirstrings[dir], addr, size,
305                        regs->u_regs[UREG_RETPC]);
306 #endif
307                 orig_asi = asi = decode_asi(insn, regs);
308                 switch (asi) {
309                 case ASI_NL:
310                 case ASI_AIUPL:
311                 case ASI_AIUSL:
312                 case ASI_PL:
313                 case ASI_SL:
314                 case ASI_PNFL:
315                 case ASI_SNFL:
316                         asi &= ~0x08;
317                         break;
318                 };
319                 switch (dir) {
320                 case load:
321                         reg_addr = fetch_reg_addr(((insn>>25)&0x1f), regs);
322                         err = do_int_load(reg_addr, size,
323                                           (unsigned long *) addr,
324                                           decode_signedness(insn), asi);
325                         if (likely(!err) && unlikely(asi != orig_asi)) {
326                                 unsigned long val_in = *reg_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                                 *reg_addr = val_in;
343                         }
344                         break;
345
346                 case store:
347                         err = 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                 if (unlikely(err))
357                         kernel_mna_trap_fault();
358                 else
359                         advance(regs);
360         }
361 }
362
363 static char popc_helper[] = {
364 0, 1, 1, 2, 1, 2, 2, 3,
365 1, 2, 2, 3, 2, 3, 3, 4, 
366 };
367
368 int handle_popc(u32 insn, struct pt_regs *regs)
369 {
370         u64 value;
371         int ret, i, rd = ((insn >> 25) & 0x1f);
372         int from_kernel = (regs->tstate & TSTATE_PRIV) != 0;
373                                 
374         if (insn & 0x2000) {
375                 maybe_flush_windows(0, 0, rd, from_kernel);
376                 value = sign_extend_imm13(insn);
377         } else {
378                 maybe_flush_windows(0, insn & 0x1f, rd, from_kernel);
379                 value = fetch_reg(insn & 0x1f, regs);
380         }
381         for (ret = 0, i = 0; i < 16; i++) {
382                 ret += popc_helper[value & 0xf];
383                 value >>= 4;
384         }
385         if (rd < 16) {
386                 if (rd)
387                         regs->u_regs[rd] = ret;
388         } else {
389                 if (test_thread_flag(TIF_32BIT)) {
390                         struct reg_window32 __user *win32;
391                         win32 = (struct reg_window32 __user *)((unsigned long)((u32)regs->u_regs[UREG_FP]));
392                         put_user(ret, &win32->locals[rd - 16]);
393                 } else {
394                         struct reg_window __user *win;
395                         win = (struct reg_window __user *)(regs->u_regs[UREG_FP] + STACK_BIAS);
396                         put_user(ret, &win->locals[rd - 16]);
397                 }
398         }
399         advance(regs);
400         return 1;
401 }
402
403 extern void do_fpother(struct pt_regs *regs);
404 extern void do_privact(struct pt_regs *regs);
405 extern void spitfire_data_access_exception(struct pt_regs *regs,
406                                            unsigned long sfsr,
407                                            unsigned long sfar);
408
409 int handle_ldf_stq(u32 insn, struct pt_regs *regs)
410 {
411         unsigned long addr = compute_effective_address(regs, insn, 0);
412         int freg = ((insn >> 25) & 0x1e) | ((insn >> 20) & 0x20);
413         struct fpustate *f = FPUSTATE;
414         int asi = decode_asi(insn, regs);
415         int flag = (freg < 32) ? FPRS_DL : FPRS_DU;
416
417         save_and_clear_fpu();
418         current_thread_info()->xfsr[0] &= ~0x1c000;
419         if (freg & 3) {
420                 current_thread_info()->xfsr[0] |= (6 << 14) /* invalid_fp_register */;
421                 do_fpother(regs);
422                 return 0;
423         }
424         if (insn & 0x200000) {
425                 /* STQ */
426                 u64 first = 0, second = 0;
427                 
428                 if (current_thread_info()->fpsaved[0] & flag) {
429                         first = *(u64 *)&f->regs[freg];
430                         second = *(u64 *)&f->regs[freg+2];
431                 }
432                 if (asi < 0x80) {
433                         do_privact(regs);
434                         return 1;
435                 }
436                 switch (asi) {
437                 case ASI_P:
438                 case ASI_S: break;
439                 case ASI_PL:
440                 case ASI_SL: 
441                         {
442                                 /* Need to convert endians */
443                                 u64 tmp = __swab64p(&first);
444                                 
445                                 first = __swab64p(&second);
446                                 second = tmp;
447                                 break;
448                         }
449                 default:
450                         spitfire_data_access_exception(regs, 0, addr);
451                         return 1;
452                 }
453                 if (put_user (first >> 32, (u32 __user *)addr) ||
454                     __put_user ((u32)first, (u32 __user *)(addr + 4)) ||
455                     __put_user (second >> 32, (u32 __user *)(addr + 8)) ||
456                     __put_user ((u32)second, (u32 __user *)(addr + 12))) {
457                         spitfire_data_access_exception(regs, 0, addr);
458                         return 1;
459                 }
460         } else {
461                 /* LDF, LDDF, LDQF */
462                 u32 data[4] __attribute__ ((aligned(8)));
463                 int size, i;
464                 int err;
465
466                 if (asi < 0x80) {
467                         do_privact(regs);
468                         return 1;
469                 } else if (asi > ASI_SNFL) {
470                         spitfire_data_access_exception(regs, 0, addr);
471                         return 1;
472                 }
473                 switch (insn & 0x180000) {
474                 case 0x000000: size = 1; break;
475                 case 0x100000: size = 4; break;
476                 default: size = 2; break;
477                 }
478                 for (i = 0; i < size; i++)
479                         data[i] = 0;
480                 
481                 err = get_user (data[0], (u32 __user *) addr);
482                 if (!err) {
483                         for (i = 1; i < size; i++)
484                                 err |= __get_user (data[i], (u32 __user *)(addr + 4*i));
485                 }
486                 if (err && !(asi & 0x2 /* NF */)) {
487                         spitfire_data_access_exception(regs, 0, addr);
488                         return 1;
489                 }
490                 if (asi & 0x8) /* Little */ {
491                         u64 tmp;
492
493                         switch (size) {
494                         case 1: data[0] = le32_to_cpup(data + 0); break;
495                         default:*(u64 *)(data + 0) = le64_to_cpup((u64 *)(data + 0));
496                                 break;
497                         case 4: tmp = le64_to_cpup((u64 *)(data + 0));
498                                 *(u64 *)(data + 0) = le64_to_cpup((u64 *)(data + 2));
499                                 *(u64 *)(data + 2) = tmp;
500                                 break;
501                         }
502                 }
503                 if (!(current_thread_info()->fpsaved[0] & FPRS_FEF)) {
504                         current_thread_info()->fpsaved[0] = FPRS_FEF;
505                         current_thread_info()->gsr[0] = 0;
506                 }
507                 if (!(current_thread_info()->fpsaved[0] & flag)) {
508                         if (freg < 32)
509                                 memset(f->regs, 0, 32*sizeof(u32));
510                         else
511                                 memset(f->regs+32, 0, 32*sizeof(u32));
512                 }
513                 memcpy(f->regs + freg, data, size * 4);
514                 current_thread_info()->fpsaved[0] |= flag;
515         }
516         advance(regs);
517         return 1;
518 }
519
520 void handle_ld_nf(u32 insn, struct pt_regs *regs)
521 {
522         int rd = ((insn >> 25) & 0x1f);
523         int from_kernel = (regs->tstate & TSTATE_PRIV) != 0;
524         unsigned long *reg;
525                                 
526         maybe_flush_windows(0, 0, rd, from_kernel);
527         reg = fetch_reg_addr(rd, regs);
528         if (from_kernel || rd < 16) {
529                 reg[0] = 0;
530                 if ((insn & 0x780000) == 0x180000)
531                         reg[1] = 0;
532         } else if (test_thread_flag(TIF_32BIT)) {
533                 put_user(0, (int __user *) reg);
534                 if ((insn & 0x780000) == 0x180000)
535                         put_user(0, ((int __user *) reg) + 1);
536         } else {
537                 put_user(0, (unsigned long __user *) reg);
538                 if ((insn & 0x780000) == 0x180000)
539                         put_user(0, (unsigned long __user *) reg + 1);
540         }
541         advance(regs);
542 }
543
544 void handle_lddfmna(struct pt_regs *regs, unsigned long sfar, unsigned long sfsr)
545 {
546         unsigned long pc = regs->tpc;
547         unsigned long tstate = regs->tstate;
548         u32 insn;
549         u32 first, second;
550         u64 value;
551         u8 asi, freg;
552         int flag;
553         struct fpustate *f = FPUSTATE;
554
555         if (tstate & TSTATE_PRIV)
556                 die_if_kernel("lddfmna from kernel", regs);
557         if (test_thread_flag(TIF_32BIT))
558                 pc = (u32)pc;
559         if (get_user(insn, (u32 __user *) pc) != -EFAULT) {
560                 asi = sfsr >> 16;
561                 if ((asi > ASI_SNFL) ||
562                     (asi < ASI_P))
563                         goto daex;
564                 if (get_user(first, (u32 __user *)sfar) ||
565                      get_user(second, (u32 __user *)(sfar + 4))) {
566                         if (asi & 0x2) /* NF */ {
567                                 first = 0; second = 0;
568                         } else
569                                 goto daex;
570                 }
571                 save_and_clear_fpu();
572                 freg = ((insn >> 25) & 0x1e) | ((insn >> 20) & 0x20);
573                 value = (((u64)first) << 32) | second;
574                 if (asi & 0x8) /* Little */
575                         value = __swab64p(&value);
576                 flag = (freg < 32) ? FPRS_DL : FPRS_DU;
577                 if (!(current_thread_info()->fpsaved[0] & FPRS_FEF)) {
578                         current_thread_info()->fpsaved[0] = FPRS_FEF;
579                         current_thread_info()->gsr[0] = 0;
580                 }
581                 if (!(current_thread_info()->fpsaved[0] & flag)) {
582                         if (freg < 32)
583                                 memset(f->regs, 0, 32*sizeof(u32));
584                         else
585                                 memset(f->regs+32, 0, 32*sizeof(u32));
586                 }
587                 *(u64 *)(f->regs + freg) = value;
588                 current_thread_info()->fpsaved[0] |= flag;
589         } else {
590 daex:           spitfire_data_access_exception(regs, sfsr, sfar);
591                 return;
592         }
593         advance(regs);
594         return;
595 }
596
597 void handle_stdfmna(struct pt_regs *regs, unsigned long sfar, unsigned long sfsr)
598 {
599         unsigned long pc = regs->tpc;
600         unsigned long tstate = regs->tstate;
601         u32 insn;
602         u64 value;
603         u8 asi, freg;
604         int flag;
605         struct fpustate *f = FPUSTATE;
606
607         if (tstate & TSTATE_PRIV)
608                 die_if_kernel("stdfmna from kernel", regs);
609         if (test_thread_flag(TIF_32BIT))
610                 pc = (u32)pc;
611         if (get_user(insn, (u32 __user *) pc) != -EFAULT) {
612                 freg = ((insn >> 25) & 0x1e) | ((insn >> 20) & 0x20);
613                 asi = sfsr >> 16;
614                 value = 0;
615                 flag = (freg < 32) ? FPRS_DL : FPRS_DU;
616                 if ((asi > ASI_SNFL) ||
617                     (asi < ASI_P))
618                         goto daex;
619                 save_and_clear_fpu();
620                 if (current_thread_info()->fpsaved[0] & flag)
621                         value = *(u64 *)&f->regs[freg];
622                 switch (asi) {
623                 case ASI_P:
624                 case ASI_S: break;
625                 case ASI_PL:
626                 case ASI_SL: 
627                         value = __swab64p(&value); break;
628                 default: goto daex;
629                 }
630                 if (put_user (value >> 32, (u32 __user *) sfar) ||
631                     __put_user ((u32)value, (u32 __user *)(sfar + 4)))
632                         goto daex;
633         } else {
634 daex:           spitfire_data_access_exception(regs, sfsr, sfar);
635                 return;
636         }
637         advance(regs);
638         return;
639 }