Linux 3.11-rc3
[cris-mirror.git] / arch / sparc / kernel / unaligned_64.c
blob8201c25e76697ad5f5a96de1b6921a3fb34372ac
1 /*
2 * unaligned.c: Unaligned load/store trap handling with special
3 * cases for the kernel to do them more quickly.
5 * Copyright (C) 1996,2008 David S. Miller (davem@davemloft.net)
6 * Copyright (C) 1996,1997 Jakub Jelinek (jj@sunsite.mff.cuni.cz)
7 */
10 #include <linux/jiffies.h>
11 #include <linux/kernel.h>
12 #include <linux/sched.h>
13 #include <linux/mm.h>
14 #include <linux/module.h>
15 #include <asm/asi.h>
16 #include <asm/ptrace.h>
17 #include <asm/pstate.h>
18 #include <asm/processor.h>
19 #include <asm/uaccess.h>
20 #include <linux/smp.h>
21 #include <linux/bitops.h>
22 #include <linux/perf_event.h>
23 #include <linux/ratelimit.h>
24 #include <asm/fpumacro.h>
25 #include <asm/cacheflush.h>
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,
36 static inline enum direction decode_direction(unsigned int insn)
38 unsigned long tmp = (insn >> 21) & 1;
40 if (!tmp)
41 return load;
42 else {
43 switch ((insn>>19)&0xf) {
44 case 15: /* swap* */
45 return both;
46 default:
47 return store;
52 /* 16 = double-word, 8 = extra-word, 4 = word, 2 = half-word */
53 static inline int decode_access_size(struct pt_regs *regs, unsigned int insn)
55 unsigned int tmp;
57 tmp = ((insn >> 19) & 0xf);
58 if (tmp == 11 || tmp == 14) /* ldx/stx */
59 return 8;
60 tmp &= 3;
61 if (!tmp)
62 return 4;
63 else if (tmp == 3)
64 return 16; /* ldd/std - Although it is actually 8 */
65 else if (tmp == 2)
66 return 2;
67 else {
68 printk("Impossible unaligned trap. insn=%08x\n", insn);
69 die_if_kernel("Byte sized unaligned access?!?!", regs);
71 /* GCC should never warn that control reaches the end
72 * of this function without returning a value because
73 * die_if_kernel() is marked with attribute 'noreturn'.
74 * Alas, some versions do...
77 return 0;
81 static inline int decode_asi(unsigned int insn, struct pt_regs *regs)
83 if (insn & 0x800000) {
84 if (insn & 0x2000)
85 return (unsigned char)(regs->tstate >> 24); /* %asi */
86 else
87 return (unsigned char)(insn >> 5); /* imm_asi */
88 } else
89 return ASI_P;
92 /* 0x400000 = signed, 0 = unsigned */
93 static inline int decode_signedness(unsigned int insn)
95 return (insn & 0x400000);
98 static inline void maybe_flush_windows(unsigned int rs1, unsigned int rs2,
99 unsigned int rd, int from_kernel)
101 if (rs2 >= 16 || rs1 >= 16 || rd >= 16) {
102 if (from_kernel != 0)
103 __asm__ __volatile__("flushw");
104 else
105 flushw_user();
109 static inline long sign_extend_imm13(long imm)
111 return imm << 51 >> 51;
114 static unsigned long fetch_reg(unsigned int reg, struct pt_regs *regs)
116 unsigned long value, fp;
118 if (reg < 16)
119 return (!reg ? 0 : regs->u_regs[reg]);
121 fp = regs->u_regs[UREG_FP];
123 if (regs->tstate & TSTATE_PRIV) {
124 struct reg_window *win;
125 win = (struct reg_window *)(fp + STACK_BIAS);
126 value = win->locals[reg - 16];
127 } else if (!test_thread_64bit_stack(fp)) {
128 struct reg_window32 __user *win32;
129 win32 = (struct reg_window32 __user *)((unsigned long)((u32)fp));
130 get_user(value, &win32->locals[reg - 16]);
131 } else {
132 struct reg_window __user *win;
133 win = (struct reg_window __user *)(fp + STACK_BIAS);
134 get_user(value, &win->locals[reg - 16]);
136 return value;
139 static unsigned long *fetch_reg_addr(unsigned int reg, struct pt_regs *regs)
141 unsigned long fp;
143 if (reg < 16)
144 return &regs->u_regs[reg];
146 fp = regs->u_regs[UREG_FP];
148 if (regs->tstate & TSTATE_PRIV) {
149 struct reg_window *win;
150 win = (struct reg_window *)(fp + STACK_BIAS);
151 return &win->locals[reg - 16];
152 } else if (!test_thread_64bit_stack(fp)) {
153 struct reg_window32 *win32;
154 win32 = (struct reg_window32 *)((unsigned long)((u32)fp));
155 return (unsigned long *)&win32->locals[reg - 16];
156 } else {
157 struct reg_window *win;
158 win = (struct reg_window *)(fp + STACK_BIAS);
159 return &win->locals[reg - 16];
163 unsigned long compute_effective_address(struct pt_regs *regs,
164 unsigned int insn, unsigned int rd)
166 unsigned int rs1 = (insn >> 14) & 0x1f;
167 unsigned int rs2 = insn & 0x1f;
168 int from_kernel = (regs->tstate & TSTATE_PRIV) != 0;
170 if (insn & 0x2000) {
171 maybe_flush_windows(rs1, 0, rd, from_kernel);
172 return (fetch_reg(rs1, regs) + sign_extend_imm13(insn));
173 } else {
174 maybe_flush_windows(rs1, rs2, rd, from_kernel);
175 return (fetch_reg(rs1, regs) + fetch_reg(rs2, regs));
179 /* This is just to make gcc think die_if_kernel does return... */
180 static void __used unaligned_panic(char *str, struct pt_regs *regs)
182 die_if_kernel(str, regs);
185 extern int do_int_load(unsigned long *dest_reg, int size,
186 unsigned long *saddr, int is_signed, int asi);
188 extern int __do_int_store(unsigned long *dst_addr, int size,
189 unsigned long src_val, int asi);
191 static inline int do_int_store(int reg_num, int size, unsigned long *dst_addr,
192 struct pt_regs *regs, int asi, int orig_asi)
194 unsigned long zero = 0;
195 unsigned long *src_val_p = &zero;
196 unsigned long src_val;
198 if (size == 16) {
199 size = 8;
200 zero = (((long)(reg_num ?
201 (unsigned)fetch_reg(reg_num, regs) : 0)) << 32) |
202 (unsigned)fetch_reg(reg_num + 1, regs);
203 } else if (reg_num) {
204 src_val_p = fetch_reg_addr(reg_num, regs);
206 src_val = *src_val_p;
207 if (unlikely(asi != orig_asi)) {
208 switch (size) {
209 case 2:
210 src_val = swab16(src_val);
211 break;
212 case 4:
213 src_val = swab32(src_val);
214 break;
215 case 8:
216 src_val = swab64(src_val);
217 break;
218 case 16:
219 default:
220 BUG();
221 break;
224 return __do_int_store(dst_addr, size, src_val, asi);
227 static inline void advance(struct pt_regs *regs)
229 regs->tpc = regs->tnpc;
230 regs->tnpc += 4;
231 if (test_thread_flag(TIF_32BIT)) {
232 regs->tpc &= 0xffffffff;
233 regs->tnpc &= 0xffffffff;
237 static inline int floating_point_load_or_store_p(unsigned int insn)
239 return (insn >> 24) & 1;
242 static inline int ok_for_kernel(unsigned int insn)
244 return !floating_point_load_or_store_p(insn);
247 static void kernel_mna_trap_fault(int fixup_tstate_asi)
249 struct pt_regs *regs = current_thread_info()->kern_una_regs;
250 unsigned int insn = current_thread_info()->kern_una_insn;
251 const struct exception_table_entry *entry;
253 entry = search_exception_tables(regs->tpc);
254 if (!entry) {
255 unsigned long address;
257 address = compute_effective_address(regs, insn,
258 ((insn >> 25) & 0x1f));
259 if (address < PAGE_SIZE) {
260 printk(KERN_ALERT "Unable to handle kernel NULL "
261 "pointer dereference in mna handler");
262 } else
263 printk(KERN_ALERT "Unable to handle kernel paging "
264 "request in mna handler");
265 printk(KERN_ALERT " at virtual address %016lx\n",address);
266 printk(KERN_ALERT "current->{active_,}mm->context = %016lx\n",
267 (current->mm ? CTX_HWBITS(current->mm->context) :
268 CTX_HWBITS(current->active_mm->context)));
269 printk(KERN_ALERT "current->{active_,}mm->pgd = %016lx\n",
270 (current->mm ? (unsigned long) current->mm->pgd :
271 (unsigned long) current->active_mm->pgd));
272 die_if_kernel("Oops", regs);
273 /* Not reached */
275 regs->tpc = entry->fixup;
276 regs->tnpc = regs->tpc + 4;
278 if (fixup_tstate_asi) {
279 regs->tstate &= ~TSTATE_ASI;
280 regs->tstate |= (ASI_AIUS << 24UL);
284 static void log_unaligned(struct pt_regs *regs)
286 static DEFINE_RATELIMIT_STATE(ratelimit, 5 * HZ, 5);
288 if (__ratelimit(&ratelimit)) {
289 printk("Kernel unaligned access at TPC[%lx] %pS\n",
290 regs->tpc, (void *) regs->tpc);
294 asmlinkage void kernel_unaligned_trap(struct pt_regs *regs, unsigned int insn)
296 enum direction dir = decode_direction(insn);
297 int size = decode_access_size(regs, insn);
298 int orig_asi, asi;
300 current_thread_info()->kern_una_regs = regs;
301 current_thread_info()->kern_una_insn = insn;
303 orig_asi = asi = decode_asi(insn, regs);
305 /* If this is a {get,put}_user() on an unaligned userspace pointer,
306 * just signal a fault and do not log the event.
308 if (asi == ASI_AIUS) {
309 kernel_mna_trap_fault(0);
310 return;
313 log_unaligned(regs);
315 if (!ok_for_kernel(insn) || dir == both) {
316 printk("Unsupported unaligned load/store trap for kernel "
317 "at <%016lx>.\n", regs->tpc);
318 unaligned_panic("Kernel does fpu/atomic "
319 "unaligned load/store.", regs);
321 kernel_mna_trap_fault(0);
322 } else {
323 unsigned long addr, *reg_addr;
324 int err;
326 addr = compute_effective_address(regs, insn,
327 ((insn >> 25) & 0x1f));
328 perf_sw_event(PERF_COUNT_SW_ALIGNMENT_FAULTS, 1, regs, addr);
329 switch (asi) {
330 case ASI_NL:
331 case ASI_AIUPL:
332 case ASI_AIUSL:
333 case ASI_PL:
334 case ASI_SL:
335 case ASI_PNFL:
336 case ASI_SNFL:
337 asi &= ~0x08;
338 break;
340 switch (dir) {
341 case load:
342 reg_addr = fetch_reg_addr(((insn>>25)&0x1f), regs);
343 err = do_int_load(reg_addr, size,
344 (unsigned long *) addr,
345 decode_signedness(insn), asi);
346 if (likely(!err) && unlikely(asi != orig_asi)) {
347 unsigned long val_in = *reg_addr;
348 switch (size) {
349 case 2:
350 val_in = swab16(val_in);
351 break;
352 case 4:
353 val_in = swab32(val_in);
354 break;
355 case 8:
356 val_in = swab64(val_in);
357 break;
358 case 16:
359 default:
360 BUG();
361 break;
363 *reg_addr = val_in;
365 break;
367 case store:
368 err = do_int_store(((insn>>25)&0x1f), size,
369 (unsigned long *) addr, regs,
370 asi, orig_asi);
371 break;
373 default:
374 panic("Impossible kernel unaligned trap.");
375 /* Not reached... */
377 if (unlikely(err))
378 kernel_mna_trap_fault(1);
379 else
380 advance(regs);
384 int handle_popc(u32 insn, struct pt_regs *regs)
386 int from_kernel = (regs->tstate & TSTATE_PRIV) != 0;
387 int ret, rd = ((insn >> 25) & 0x1f);
388 u64 value;
390 perf_sw_event(PERF_COUNT_SW_EMULATION_FAULTS, 1, regs, 0);
391 if (insn & 0x2000) {
392 maybe_flush_windows(0, 0, rd, from_kernel);
393 value = sign_extend_imm13(insn);
394 } else {
395 maybe_flush_windows(0, insn & 0x1f, rd, from_kernel);
396 value = fetch_reg(insn & 0x1f, regs);
398 ret = hweight64(value);
399 if (rd < 16) {
400 if (rd)
401 regs->u_regs[rd] = ret;
402 } else {
403 unsigned long fp = regs->u_regs[UREG_FP];
405 if (!test_thread_64bit_stack(fp)) {
406 struct reg_window32 __user *win32;
407 win32 = (struct reg_window32 __user *)((unsigned long)((u32)fp));
408 put_user(ret, &win32->locals[rd - 16]);
409 } else {
410 struct reg_window __user *win;
411 win = (struct reg_window __user *)(fp + STACK_BIAS);
412 put_user(ret, &win->locals[rd - 16]);
415 advance(regs);
416 return 1;
419 extern void do_fpother(struct pt_regs *regs);
420 extern void do_privact(struct pt_regs *regs);
421 extern void spitfire_data_access_exception(struct pt_regs *regs,
422 unsigned long sfsr,
423 unsigned long sfar);
424 extern void sun4v_data_access_exception(struct pt_regs *regs,
425 unsigned long addr,
426 unsigned long type_ctx);
428 int handle_ldf_stq(u32 insn, struct pt_regs *regs)
430 unsigned long addr = compute_effective_address(regs, insn, 0);
431 int freg = ((insn >> 25) & 0x1e) | ((insn >> 20) & 0x20);
432 struct fpustate *f = FPUSTATE;
433 int asi = decode_asi(insn, regs);
434 int flag = (freg < 32) ? FPRS_DL : FPRS_DU;
436 perf_sw_event(PERF_COUNT_SW_EMULATION_FAULTS, 1, regs, 0);
438 save_and_clear_fpu();
439 current_thread_info()->xfsr[0] &= ~0x1c000;
440 if (freg & 3) {
441 current_thread_info()->xfsr[0] |= (6 << 14) /* invalid_fp_register */;
442 do_fpother(regs);
443 return 0;
445 if (insn & 0x200000) {
446 /* STQ */
447 u64 first = 0, second = 0;
449 if (current_thread_info()->fpsaved[0] & flag) {
450 first = *(u64 *)&f->regs[freg];
451 second = *(u64 *)&f->regs[freg+2];
453 if (asi < 0x80) {
454 do_privact(regs);
455 return 1;
457 switch (asi) {
458 case ASI_P:
459 case ASI_S: break;
460 case ASI_PL:
461 case ASI_SL:
463 /* Need to convert endians */
464 u64 tmp = __swab64p(&first);
466 first = __swab64p(&second);
467 second = tmp;
468 break;
470 default:
471 if (tlb_type == hypervisor)
472 sun4v_data_access_exception(regs, addr, 0);
473 else
474 spitfire_data_access_exception(regs, 0, addr);
475 return 1;
477 if (put_user (first >> 32, (u32 __user *)addr) ||
478 __put_user ((u32)first, (u32 __user *)(addr + 4)) ||
479 __put_user (second >> 32, (u32 __user *)(addr + 8)) ||
480 __put_user ((u32)second, (u32 __user *)(addr + 12))) {
481 if (tlb_type == hypervisor)
482 sun4v_data_access_exception(regs, addr, 0);
483 else
484 spitfire_data_access_exception(regs, 0, addr);
485 return 1;
487 } else {
488 /* LDF, LDDF, LDQF */
489 u32 data[4] __attribute__ ((aligned(8)));
490 int size, i;
491 int err;
493 if (asi < 0x80) {
494 do_privact(regs);
495 return 1;
496 } else if (asi > ASI_SNFL) {
497 if (tlb_type == hypervisor)
498 sun4v_data_access_exception(regs, addr, 0);
499 else
500 spitfire_data_access_exception(regs, 0, addr);
501 return 1;
503 switch (insn & 0x180000) {
504 case 0x000000: size = 1; break;
505 case 0x100000: size = 4; break;
506 default: size = 2; break;
508 for (i = 0; i < size; i++)
509 data[i] = 0;
511 err = get_user (data[0], (u32 __user *) addr);
512 if (!err) {
513 for (i = 1; i < size; i++)
514 err |= __get_user (data[i], (u32 __user *)(addr + 4*i));
516 if (err && !(asi & 0x2 /* NF */)) {
517 if (tlb_type == hypervisor)
518 sun4v_data_access_exception(regs, addr, 0);
519 else
520 spitfire_data_access_exception(regs, 0, addr);
521 return 1;
523 if (asi & 0x8) /* Little */ {
524 u64 tmp;
526 switch (size) {
527 case 1: data[0] = le32_to_cpup(data + 0); break;
528 default:*(u64 *)(data + 0) = le64_to_cpup((u64 *)(data + 0));
529 break;
530 case 4: tmp = le64_to_cpup((u64 *)(data + 0));
531 *(u64 *)(data + 0) = le64_to_cpup((u64 *)(data + 2));
532 *(u64 *)(data + 2) = tmp;
533 break;
536 if (!(current_thread_info()->fpsaved[0] & FPRS_FEF)) {
537 current_thread_info()->fpsaved[0] = FPRS_FEF;
538 current_thread_info()->gsr[0] = 0;
540 if (!(current_thread_info()->fpsaved[0] & flag)) {
541 if (freg < 32)
542 memset(f->regs, 0, 32*sizeof(u32));
543 else
544 memset(f->regs+32, 0, 32*sizeof(u32));
546 memcpy(f->regs + freg, data, size * 4);
547 current_thread_info()->fpsaved[0] |= flag;
549 advance(regs);
550 return 1;
553 void handle_ld_nf(u32 insn, struct pt_regs *regs)
555 int rd = ((insn >> 25) & 0x1f);
556 int from_kernel = (regs->tstate & TSTATE_PRIV) != 0;
557 unsigned long *reg;
559 perf_sw_event(PERF_COUNT_SW_EMULATION_FAULTS, 1, regs, 0);
561 maybe_flush_windows(0, 0, rd, from_kernel);
562 reg = fetch_reg_addr(rd, regs);
563 if (from_kernel || rd < 16) {
564 reg[0] = 0;
565 if ((insn & 0x780000) == 0x180000)
566 reg[1] = 0;
567 } else if (!test_thread_64bit_stack(regs->u_regs[UREG_FP])) {
568 put_user(0, (int __user *) reg);
569 if ((insn & 0x780000) == 0x180000)
570 put_user(0, ((int __user *) reg) + 1);
571 } else {
572 put_user(0, (unsigned long __user *) reg);
573 if ((insn & 0x780000) == 0x180000)
574 put_user(0, (unsigned long __user *) reg + 1);
576 advance(regs);
579 void handle_lddfmna(struct pt_regs *regs, unsigned long sfar, unsigned long sfsr)
581 unsigned long pc = regs->tpc;
582 unsigned long tstate = regs->tstate;
583 u32 insn;
584 u64 value;
585 u8 freg;
586 int flag;
587 struct fpustate *f = FPUSTATE;
589 if (tstate & TSTATE_PRIV)
590 die_if_kernel("lddfmna from kernel", regs);
591 perf_sw_event(PERF_COUNT_SW_ALIGNMENT_FAULTS, 1, regs, sfar);
592 if (test_thread_flag(TIF_32BIT))
593 pc = (u32)pc;
594 if (get_user(insn, (u32 __user *) pc) != -EFAULT) {
595 int asi = decode_asi(insn, regs);
596 u32 first, second;
597 int err;
599 if ((asi > ASI_SNFL) ||
600 (asi < ASI_P))
601 goto daex;
602 first = second = 0;
603 err = get_user(first, (u32 __user *)sfar);
604 if (!err)
605 err = get_user(second, (u32 __user *)(sfar + 4));
606 if (err) {
607 if (!(asi & 0x2))
608 goto daex;
609 first = second = 0;
611 save_and_clear_fpu();
612 freg = ((insn >> 25) & 0x1e) | ((insn >> 20) & 0x20);
613 value = (((u64)first) << 32) | second;
614 if (asi & 0x8) /* Little */
615 value = __swab64p(&value);
616 flag = (freg < 32) ? FPRS_DL : FPRS_DU;
617 if (!(current_thread_info()->fpsaved[0] & FPRS_FEF)) {
618 current_thread_info()->fpsaved[0] = FPRS_FEF;
619 current_thread_info()->gsr[0] = 0;
621 if (!(current_thread_info()->fpsaved[0] & flag)) {
622 if (freg < 32)
623 memset(f->regs, 0, 32*sizeof(u32));
624 else
625 memset(f->regs+32, 0, 32*sizeof(u32));
627 *(u64 *)(f->regs + freg) = value;
628 current_thread_info()->fpsaved[0] |= flag;
629 } else {
630 daex:
631 if (tlb_type == hypervisor)
632 sun4v_data_access_exception(regs, sfar, sfsr);
633 else
634 spitfire_data_access_exception(regs, sfsr, sfar);
635 return;
637 advance(regs);
640 void handle_stdfmna(struct pt_regs *regs, unsigned long sfar, unsigned long sfsr)
642 unsigned long pc = regs->tpc;
643 unsigned long tstate = regs->tstate;
644 u32 insn;
645 u64 value;
646 u8 freg;
647 int flag;
648 struct fpustate *f = FPUSTATE;
650 if (tstate & TSTATE_PRIV)
651 die_if_kernel("stdfmna from kernel", regs);
652 perf_sw_event(PERF_COUNT_SW_ALIGNMENT_FAULTS, 1, regs, sfar);
653 if (test_thread_flag(TIF_32BIT))
654 pc = (u32)pc;
655 if (get_user(insn, (u32 __user *) pc) != -EFAULT) {
656 int asi = decode_asi(insn, regs);
657 freg = ((insn >> 25) & 0x1e) | ((insn >> 20) & 0x20);
658 value = 0;
659 flag = (freg < 32) ? FPRS_DL : FPRS_DU;
660 if ((asi > ASI_SNFL) ||
661 (asi < ASI_P))
662 goto daex;
663 save_and_clear_fpu();
664 if (current_thread_info()->fpsaved[0] & flag)
665 value = *(u64 *)&f->regs[freg];
666 switch (asi) {
667 case ASI_P:
668 case ASI_S: break;
669 case ASI_PL:
670 case ASI_SL:
671 value = __swab64p(&value); break;
672 default: goto daex;
674 if (put_user (value >> 32, (u32 __user *) sfar) ||
675 __put_user ((u32)value, (u32 __user *)(sfar + 4)))
676 goto daex;
677 } else {
678 daex:
679 if (tlb_type == hypervisor)
680 sun4v_data_access_exception(regs, sfar, sfsr);
681 else
682 spitfire_data_access_exception(regs, sfsr, sfar);
683 return;
685 advance(regs);