Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
[linux-btrfs-devel.git] / arch / sparc / kernel / unaligned_64.c
blob76e4ac1a13e1b9978be67102ed45ef5f0f19e274
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/system.h>
20 #include <asm/uaccess.h>
21 #include <linux/smp.h>
22 #include <linux/bitops.h>
23 #include <linux/perf_event.h>
24 #include <linux/ratelimit.h>
25 #include <linux/bitops.h>
26 #include <asm/fpumacro.h>
28 enum direction {
29 load, /* ld, ldd, ldh, ldsh */
30 store, /* st, std, sth, stsh */
31 both, /* Swap, ldstub, cas, ... */
32 fpld,
33 fpst,
34 invalid,
37 static inline enum direction decode_direction(unsigned int insn)
39 unsigned long tmp = (insn >> 21) & 1;
41 if (!tmp)
42 return load;
43 else {
44 switch ((insn>>19)&0xf) {
45 case 15: /* swap* */
46 return both;
47 default:
48 return store;
53 /* 16 = double-word, 8 = extra-word, 4 = word, 2 = half-word */
54 static inline int decode_access_size(struct pt_regs *regs, unsigned int insn)
56 unsigned int tmp;
58 tmp = ((insn >> 19) & 0xf);
59 if (tmp == 11 || tmp == 14) /* ldx/stx */
60 return 8;
61 tmp &= 3;
62 if (!tmp)
63 return 4;
64 else if (tmp == 3)
65 return 16; /* ldd/std - Although it is actually 8 */
66 else if (tmp == 2)
67 return 2;
68 else {
69 printk("Impossible unaligned trap. insn=%08x\n", insn);
70 die_if_kernel("Byte sized unaligned access?!?!", regs);
72 /* GCC should never warn that control reaches the end
73 * of this function without returning a value because
74 * die_if_kernel() is marked with attribute 'noreturn'.
75 * Alas, some versions do...
78 return 0;
82 static inline int decode_asi(unsigned int insn, struct pt_regs *regs)
84 if (insn & 0x800000) {
85 if (insn & 0x2000)
86 return (unsigned char)(regs->tstate >> 24); /* %asi */
87 else
88 return (unsigned char)(insn >> 5); /* imm_asi */
89 } else
90 return ASI_P;
93 /* 0x400000 = signed, 0 = unsigned */
94 static inline int decode_signedness(unsigned int insn)
96 return (insn & 0x400000);
99 static inline void maybe_flush_windows(unsigned int rs1, unsigned int rs2,
100 unsigned int rd, int from_kernel)
102 if (rs2 >= 16 || rs1 >= 16 || rd >= 16) {
103 if (from_kernel != 0)
104 __asm__ __volatile__("flushw");
105 else
106 flushw_user();
110 static inline long sign_extend_imm13(long imm)
112 return imm << 51 >> 51;
115 static unsigned long fetch_reg(unsigned int reg, struct pt_regs *regs)
117 unsigned long value;
119 if (reg < 16)
120 return (!reg ? 0 : regs->u_regs[reg]);
121 if (regs->tstate & TSTATE_PRIV) {
122 struct reg_window *win;
123 win = (struct reg_window *)(regs->u_regs[UREG_FP] + STACK_BIAS);
124 value = win->locals[reg - 16];
125 } else if (test_thread_flag(TIF_32BIT)) {
126 struct reg_window32 __user *win32;
127 win32 = (struct reg_window32 __user *)((unsigned long)((u32)regs->u_regs[UREG_FP]));
128 get_user(value, &win32->locals[reg - 16]);
129 } else {
130 struct reg_window __user *win;
131 win = (struct reg_window __user *)(regs->u_regs[UREG_FP] + STACK_BIAS);
132 get_user(value, &win->locals[reg - 16]);
134 return value;
137 static unsigned long *fetch_reg_addr(unsigned int reg, struct pt_regs *regs)
139 if (reg < 16)
140 return &regs->u_regs[reg];
141 if (regs->tstate & TSTATE_PRIV) {
142 struct reg_window *win;
143 win = (struct reg_window *)(regs->u_regs[UREG_FP] + STACK_BIAS);
144 return &win->locals[reg - 16];
145 } else if (test_thread_flag(TIF_32BIT)) {
146 struct reg_window32 *win32;
147 win32 = (struct reg_window32 *)((unsigned long)((u32)regs->u_regs[UREG_FP]));
148 return (unsigned long *)&win32->locals[reg - 16];
149 } else {
150 struct reg_window *win;
151 win = (struct reg_window *)(regs->u_regs[UREG_FP] + STACK_BIAS);
152 return &win->locals[reg - 16];
156 unsigned long compute_effective_address(struct pt_regs *regs,
157 unsigned int insn, unsigned int rd)
159 unsigned int rs1 = (insn >> 14) & 0x1f;
160 unsigned int rs2 = insn & 0x1f;
161 int from_kernel = (regs->tstate & TSTATE_PRIV) != 0;
163 if (insn & 0x2000) {
164 maybe_flush_windows(rs1, 0, rd, from_kernel);
165 return (fetch_reg(rs1, regs) + sign_extend_imm13(insn));
166 } else {
167 maybe_flush_windows(rs1, rs2, rd, from_kernel);
168 return (fetch_reg(rs1, regs) + fetch_reg(rs2, regs));
172 /* This is just to make gcc think die_if_kernel does return... */
173 static void __used unaligned_panic(char *str, struct pt_regs *regs)
175 die_if_kernel(str, regs);
178 extern int do_int_load(unsigned long *dest_reg, int size,
179 unsigned long *saddr, int is_signed, int asi);
181 extern int __do_int_store(unsigned long *dst_addr, int size,
182 unsigned long src_val, int asi);
184 static inline int do_int_store(int reg_num, int size, unsigned long *dst_addr,
185 struct pt_regs *regs, int asi, int orig_asi)
187 unsigned long zero = 0;
188 unsigned long *src_val_p = &zero;
189 unsigned long src_val;
191 if (size == 16) {
192 size = 8;
193 zero = (((long)(reg_num ?
194 (unsigned)fetch_reg(reg_num, regs) : 0)) << 32) |
195 (unsigned)fetch_reg(reg_num + 1, regs);
196 } else if (reg_num) {
197 src_val_p = fetch_reg_addr(reg_num, regs);
199 src_val = *src_val_p;
200 if (unlikely(asi != orig_asi)) {
201 switch (size) {
202 case 2:
203 src_val = swab16(src_val);
204 break;
205 case 4:
206 src_val = swab32(src_val);
207 break;
208 case 8:
209 src_val = swab64(src_val);
210 break;
211 case 16:
212 default:
213 BUG();
214 break;
217 return __do_int_store(dst_addr, size, src_val, asi);
220 static inline void advance(struct pt_regs *regs)
222 regs->tpc = regs->tnpc;
223 regs->tnpc += 4;
224 if (test_thread_flag(TIF_32BIT)) {
225 regs->tpc &= 0xffffffff;
226 regs->tnpc &= 0xffffffff;
230 static inline int floating_point_load_or_store_p(unsigned int insn)
232 return (insn >> 24) & 1;
235 static inline int ok_for_kernel(unsigned int insn)
237 return !floating_point_load_or_store_p(insn);
240 static void kernel_mna_trap_fault(int fixup_tstate_asi)
242 struct pt_regs *regs = current_thread_info()->kern_una_regs;
243 unsigned int insn = current_thread_info()->kern_una_insn;
244 const struct exception_table_entry *entry;
246 entry = search_exception_tables(regs->tpc);
247 if (!entry) {
248 unsigned long address;
250 address = compute_effective_address(regs, insn,
251 ((insn >> 25) & 0x1f));
252 if (address < PAGE_SIZE) {
253 printk(KERN_ALERT "Unable to handle kernel NULL "
254 "pointer dereference in mna handler");
255 } else
256 printk(KERN_ALERT "Unable to handle kernel paging "
257 "request in mna handler");
258 printk(KERN_ALERT " at virtual address %016lx\n",address);
259 printk(KERN_ALERT "current->{active_,}mm->context = %016lx\n",
260 (current->mm ? CTX_HWBITS(current->mm->context) :
261 CTX_HWBITS(current->active_mm->context)));
262 printk(KERN_ALERT "current->{active_,}mm->pgd = %016lx\n",
263 (current->mm ? (unsigned long) current->mm->pgd :
264 (unsigned long) current->active_mm->pgd));
265 die_if_kernel("Oops", regs);
266 /* Not reached */
268 regs->tpc = entry->fixup;
269 regs->tnpc = regs->tpc + 4;
271 if (fixup_tstate_asi) {
272 regs->tstate &= ~TSTATE_ASI;
273 regs->tstate |= (ASI_AIUS << 24UL);
277 static void log_unaligned(struct pt_regs *regs)
279 static DEFINE_RATELIMIT_STATE(ratelimit, 5 * HZ, 5);
281 if (__ratelimit(&ratelimit)) {
282 printk("Kernel unaligned access at TPC[%lx] %pS\n",
283 regs->tpc, (void *) regs->tpc);
287 asmlinkage void kernel_unaligned_trap(struct pt_regs *regs, unsigned int insn)
289 enum direction dir = decode_direction(insn);
290 int size = decode_access_size(regs, insn);
291 int orig_asi, asi;
293 current_thread_info()->kern_una_regs = regs;
294 current_thread_info()->kern_una_insn = insn;
296 orig_asi = asi = decode_asi(insn, regs);
298 /* If this is a {get,put}_user() on an unaligned userspace pointer,
299 * just signal a fault and do not log the event.
301 if (asi == ASI_AIUS) {
302 kernel_mna_trap_fault(0);
303 return;
306 log_unaligned(regs);
308 if (!ok_for_kernel(insn) || dir == both) {
309 printk("Unsupported unaligned load/store trap for kernel "
310 "at <%016lx>.\n", regs->tpc);
311 unaligned_panic("Kernel does fpu/atomic "
312 "unaligned load/store.", regs);
314 kernel_mna_trap_fault(0);
315 } else {
316 unsigned long addr, *reg_addr;
317 int err;
319 addr = compute_effective_address(regs, insn,
320 ((insn >> 25) & 0x1f));
321 perf_sw_event(PERF_COUNT_SW_ALIGNMENT_FAULTS, 1, regs, addr);
322 switch (asi) {
323 case ASI_NL:
324 case ASI_AIUPL:
325 case ASI_AIUSL:
326 case ASI_PL:
327 case ASI_SL:
328 case ASI_PNFL:
329 case ASI_SNFL:
330 asi &= ~0x08;
331 break;
333 switch (dir) {
334 case load:
335 reg_addr = fetch_reg_addr(((insn>>25)&0x1f), regs);
336 err = do_int_load(reg_addr, size,
337 (unsigned long *) addr,
338 decode_signedness(insn), asi);
339 if (likely(!err) && unlikely(asi != orig_asi)) {
340 unsigned long val_in = *reg_addr;
341 switch (size) {
342 case 2:
343 val_in = swab16(val_in);
344 break;
345 case 4:
346 val_in = swab32(val_in);
347 break;
348 case 8:
349 val_in = swab64(val_in);
350 break;
351 case 16:
352 default:
353 BUG();
354 break;
356 *reg_addr = val_in;
358 break;
360 case store:
361 err = do_int_store(((insn>>25)&0x1f), size,
362 (unsigned long *) addr, regs,
363 asi, orig_asi);
364 break;
366 default:
367 panic("Impossible kernel unaligned trap.");
368 /* Not reached... */
370 if (unlikely(err))
371 kernel_mna_trap_fault(1);
372 else
373 advance(regs);
377 int handle_popc(u32 insn, struct pt_regs *regs)
379 int from_kernel = (regs->tstate & TSTATE_PRIV) != 0;
380 int ret, rd = ((insn >> 25) & 0x1f);
381 u64 value;
383 perf_sw_event(PERF_COUNT_SW_EMULATION_FAULTS, 1, regs, 0);
384 if (insn & 0x2000) {
385 maybe_flush_windows(0, 0, rd, from_kernel);
386 value = sign_extend_imm13(insn);
387 } else {
388 maybe_flush_windows(0, insn & 0x1f, rd, from_kernel);
389 value = fetch_reg(insn & 0x1f, regs);
391 ret = hweight64(value);
392 if (rd < 16) {
393 if (rd)
394 regs->u_regs[rd] = ret;
395 } else {
396 if (test_thread_flag(TIF_32BIT)) {
397 struct reg_window32 __user *win32;
398 win32 = (struct reg_window32 __user *)((unsigned long)((u32)regs->u_regs[UREG_FP]));
399 put_user(ret, &win32->locals[rd - 16]);
400 } else {
401 struct reg_window __user *win;
402 win = (struct reg_window __user *)(regs->u_regs[UREG_FP] + STACK_BIAS);
403 put_user(ret, &win->locals[rd - 16]);
406 advance(regs);
407 return 1;
410 extern void do_fpother(struct pt_regs *regs);
411 extern void do_privact(struct pt_regs *regs);
412 extern void spitfire_data_access_exception(struct pt_regs *regs,
413 unsigned long sfsr,
414 unsigned long sfar);
415 extern void sun4v_data_access_exception(struct pt_regs *regs,
416 unsigned long addr,
417 unsigned long type_ctx);
419 int handle_ldf_stq(u32 insn, struct pt_regs *regs)
421 unsigned long addr = compute_effective_address(regs, insn, 0);
422 int freg = ((insn >> 25) & 0x1e) | ((insn >> 20) & 0x20);
423 struct fpustate *f = FPUSTATE;
424 int asi = decode_asi(insn, regs);
425 int flag = (freg < 32) ? FPRS_DL : FPRS_DU;
427 perf_sw_event(PERF_COUNT_SW_EMULATION_FAULTS, 1, regs, 0);
429 save_and_clear_fpu();
430 current_thread_info()->xfsr[0] &= ~0x1c000;
431 if (freg & 3) {
432 current_thread_info()->xfsr[0] |= (6 << 14) /* invalid_fp_register */;
433 do_fpother(regs);
434 return 0;
436 if (insn & 0x200000) {
437 /* STQ */
438 u64 first = 0, second = 0;
440 if (current_thread_info()->fpsaved[0] & flag) {
441 first = *(u64 *)&f->regs[freg];
442 second = *(u64 *)&f->regs[freg+2];
444 if (asi < 0x80) {
445 do_privact(regs);
446 return 1;
448 switch (asi) {
449 case ASI_P:
450 case ASI_S: break;
451 case ASI_PL:
452 case ASI_SL:
454 /* Need to convert endians */
455 u64 tmp = __swab64p(&first);
457 first = __swab64p(&second);
458 second = tmp;
459 break;
461 default:
462 if (tlb_type == hypervisor)
463 sun4v_data_access_exception(regs, addr, 0);
464 else
465 spitfire_data_access_exception(regs, 0, addr);
466 return 1;
468 if (put_user (first >> 32, (u32 __user *)addr) ||
469 __put_user ((u32)first, (u32 __user *)(addr + 4)) ||
470 __put_user (second >> 32, (u32 __user *)(addr + 8)) ||
471 __put_user ((u32)second, (u32 __user *)(addr + 12))) {
472 if (tlb_type == hypervisor)
473 sun4v_data_access_exception(regs, addr, 0);
474 else
475 spitfire_data_access_exception(regs, 0, addr);
476 return 1;
478 } else {
479 /* LDF, LDDF, LDQF */
480 u32 data[4] __attribute__ ((aligned(8)));
481 int size, i;
482 int err;
484 if (asi < 0x80) {
485 do_privact(regs);
486 return 1;
487 } else if (asi > ASI_SNFL) {
488 if (tlb_type == hypervisor)
489 sun4v_data_access_exception(regs, addr, 0);
490 else
491 spitfire_data_access_exception(regs, 0, addr);
492 return 1;
494 switch (insn & 0x180000) {
495 case 0x000000: size = 1; break;
496 case 0x100000: size = 4; break;
497 default: size = 2; break;
499 for (i = 0; i < size; i++)
500 data[i] = 0;
502 err = get_user (data[0], (u32 __user *) addr);
503 if (!err) {
504 for (i = 1; i < size; i++)
505 err |= __get_user (data[i], (u32 __user *)(addr + 4*i));
507 if (err && !(asi & 0x2 /* NF */)) {
508 if (tlb_type == hypervisor)
509 sun4v_data_access_exception(regs, addr, 0);
510 else
511 spitfire_data_access_exception(regs, 0, addr);
512 return 1;
514 if (asi & 0x8) /* Little */ {
515 u64 tmp;
517 switch (size) {
518 case 1: data[0] = le32_to_cpup(data + 0); break;
519 default:*(u64 *)(data + 0) = le64_to_cpup((u64 *)(data + 0));
520 break;
521 case 4: tmp = le64_to_cpup((u64 *)(data + 0));
522 *(u64 *)(data + 0) = le64_to_cpup((u64 *)(data + 2));
523 *(u64 *)(data + 2) = tmp;
524 break;
527 if (!(current_thread_info()->fpsaved[0] & FPRS_FEF)) {
528 current_thread_info()->fpsaved[0] = FPRS_FEF;
529 current_thread_info()->gsr[0] = 0;
531 if (!(current_thread_info()->fpsaved[0] & flag)) {
532 if (freg < 32)
533 memset(f->regs, 0, 32*sizeof(u32));
534 else
535 memset(f->regs+32, 0, 32*sizeof(u32));
537 memcpy(f->regs + freg, data, size * 4);
538 current_thread_info()->fpsaved[0] |= flag;
540 advance(regs);
541 return 1;
544 void handle_ld_nf(u32 insn, struct pt_regs *regs)
546 int rd = ((insn >> 25) & 0x1f);
547 int from_kernel = (regs->tstate & TSTATE_PRIV) != 0;
548 unsigned long *reg;
550 perf_sw_event(PERF_COUNT_SW_EMULATION_FAULTS, 1, regs, 0);
552 maybe_flush_windows(0, 0, rd, from_kernel);
553 reg = fetch_reg_addr(rd, regs);
554 if (from_kernel || rd < 16) {
555 reg[0] = 0;
556 if ((insn & 0x780000) == 0x180000)
557 reg[1] = 0;
558 } else if (test_thread_flag(TIF_32BIT)) {
559 put_user(0, (int __user *) reg);
560 if ((insn & 0x780000) == 0x180000)
561 put_user(0, ((int __user *) reg) + 1);
562 } else {
563 put_user(0, (unsigned long __user *) reg);
564 if ((insn & 0x780000) == 0x180000)
565 put_user(0, (unsigned long __user *) reg + 1);
567 advance(regs);
570 void handle_lddfmna(struct pt_regs *regs, unsigned long sfar, unsigned long sfsr)
572 unsigned long pc = regs->tpc;
573 unsigned long tstate = regs->tstate;
574 u32 insn;
575 u64 value;
576 u8 freg;
577 int flag;
578 struct fpustate *f = FPUSTATE;
580 if (tstate & TSTATE_PRIV)
581 die_if_kernel("lddfmna from kernel", regs);
582 perf_sw_event(PERF_COUNT_SW_ALIGNMENT_FAULTS, 1, regs, sfar);
583 if (test_thread_flag(TIF_32BIT))
584 pc = (u32)pc;
585 if (get_user(insn, (u32 __user *) pc) != -EFAULT) {
586 int asi = decode_asi(insn, regs);
587 u32 first, second;
588 int err;
590 if ((asi > ASI_SNFL) ||
591 (asi < ASI_P))
592 goto daex;
593 first = second = 0;
594 err = get_user(first, (u32 __user *)sfar);
595 if (!err)
596 err = get_user(second, (u32 __user *)(sfar + 4));
597 if (err) {
598 if (!(asi & 0x2))
599 goto daex;
600 first = second = 0;
602 save_and_clear_fpu();
603 freg = ((insn >> 25) & 0x1e) | ((insn >> 20) & 0x20);
604 value = (((u64)first) << 32) | second;
605 if (asi & 0x8) /* Little */
606 value = __swab64p(&value);
607 flag = (freg < 32) ? FPRS_DL : FPRS_DU;
608 if (!(current_thread_info()->fpsaved[0] & FPRS_FEF)) {
609 current_thread_info()->fpsaved[0] = FPRS_FEF;
610 current_thread_info()->gsr[0] = 0;
612 if (!(current_thread_info()->fpsaved[0] & flag)) {
613 if (freg < 32)
614 memset(f->regs, 0, 32*sizeof(u32));
615 else
616 memset(f->regs+32, 0, 32*sizeof(u32));
618 *(u64 *)(f->regs + freg) = value;
619 current_thread_info()->fpsaved[0] |= flag;
620 } else {
621 daex:
622 if (tlb_type == hypervisor)
623 sun4v_data_access_exception(regs, sfar, sfsr);
624 else
625 spitfire_data_access_exception(regs, sfsr, sfar);
626 return;
628 advance(regs);
631 void handle_stdfmna(struct pt_regs *regs, unsigned long sfar, unsigned long sfsr)
633 unsigned long pc = regs->tpc;
634 unsigned long tstate = regs->tstate;
635 u32 insn;
636 u64 value;
637 u8 freg;
638 int flag;
639 struct fpustate *f = FPUSTATE;
641 if (tstate & TSTATE_PRIV)
642 die_if_kernel("stdfmna from kernel", regs);
643 perf_sw_event(PERF_COUNT_SW_ALIGNMENT_FAULTS, 1, regs, sfar);
644 if (test_thread_flag(TIF_32BIT))
645 pc = (u32)pc;
646 if (get_user(insn, (u32 __user *) pc) != -EFAULT) {
647 int asi = decode_asi(insn, regs);
648 freg = ((insn >> 25) & 0x1e) | ((insn >> 20) & 0x20);
649 value = 0;
650 flag = (freg < 32) ? FPRS_DL : FPRS_DU;
651 if ((asi > ASI_SNFL) ||
652 (asi < ASI_P))
653 goto daex;
654 save_and_clear_fpu();
655 if (current_thread_info()->fpsaved[0] & flag)
656 value = *(u64 *)&f->regs[freg];
657 switch (asi) {
658 case ASI_P:
659 case ASI_S: break;
660 case ASI_PL:
661 case ASI_SL:
662 value = __swab64p(&value); break;
663 default: goto daex;
665 if (put_user (value >> 32, (u32 __user *) sfar) ||
666 __put_user ((u32)value, (u32 __user *)(sfar + 4)))
667 goto daex;
668 } else {
669 daex:
670 if (tlb_type == hypervisor)
671 sun4v_data_access_exception(regs, sfar, sfsr);
672 else
673 spitfire_data_access_exception(regs, sfsr, sfar);
674 return;
676 advance(regs);