First Support on Ginger and OMAP TI
[linux-ginger.git] / arch / sparc / kernel / unaligned_64.c
blob379209982a07f41ca7454332e41b9ee303263b8c
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 <asm/fpumacro.h>
25 /* #define DEBUG_MNA */
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 #ifdef DEBUG_MNA
37 static char *dirstrings[] = {
38 "load", "store", "both", "fpload", "fpstore", "invalid"
40 #endif
42 static inline enum direction decode_direction(unsigned int insn)
44 unsigned long tmp = (insn >> 21) & 1;
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;
58 /* 16 = double-word, 8 = extra-word, 4 = word, 2 = half-word */
59 static inline int decode_access_size(unsigned int insn)
61 unsigned int tmp;
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);
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...
83 return 0;
87 static inline int decode_asi(unsigned int insn, struct pt_regs *regs)
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;
98 /* 0x400000 = signed, 0 = unsigned */
99 static inline int decode_signedness(unsigned int insn)
101 return (insn & 0x400000);
104 static inline void maybe_flush_windows(unsigned int rs1, unsigned int rs2,
105 unsigned int rd, int from_kernel)
107 if (rs2 >= 16 || rs1 >= 16 || rd >= 16) {
108 if (from_kernel != 0)
109 __asm__ __volatile__("flushw");
110 else
111 flushw_user();
115 static inline long sign_extend_imm13(long imm)
117 return imm << 51 >> 51;
120 static unsigned long fetch_reg(unsigned int reg, struct pt_regs *regs)
122 unsigned long value;
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]);
139 return value;
142 static unsigned long *fetch_reg_addr(unsigned int reg, struct pt_regs *regs)
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];
161 unsigned long compute_effective_address(struct pt_regs *regs,
162 unsigned int insn, unsigned int rd)
164 unsigned int rs1 = (insn >> 14) & 0x1f;
165 unsigned int rs2 = insn & 0x1f;
166 int from_kernel = (regs->tstate & TSTATE_PRIV) != 0;
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));
177 /* This is just to make gcc think die_if_kernel does return... */
178 static void __used unaligned_panic(char *str, struct pt_regs *regs)
180 die_if_kernel(str, regs);
183 extern int do_int_load(unsigned long *dest_reg, int size,
184 unsigned long *saddr, int is_signed, int asi);
186 extern int __do_int_store(unsigned long *dst_addr, int size,
187 unsigned long src_val, int asi);
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)
192 unsigned long zero = 0;
193 unsigned long *src_val_p = &zero;
194 unsigned long src_val;
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);
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;
222 return __do_int_store(dst_addr, size, src_val, asi);
225 static inline void advance(struct pt_regs *regs)
227 regs->tpc = regs->tnpc;
228 regs->tnpc += 4;
229 if (test_thread_flag(TIF_32BIT)) {
230 regs->tpc &= 0xffffffff;
231 regs->tnpc &= 0xffffffff;
235 static inline int floating_point_load_or_store_p(unsigned int insn)
237 return (insn >> 24) & 1;
240 static inline int ok_for_kernel(unsigned int insn)
242 return !floating_point_load_or_store_p(insn);
245 static void kernel_mna_trap_fault(int fixup_tstate_asi)
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;
251 entry = search_exception_tables(regs->tpc);
252 if (!entry) {
253 unsigned long address;
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 */
273 regs->tpc = entry->fixup;
274 regs->tnpc = regs->tpc + 4;
276 if (fixup_tstate_asi) {
277 regs->tstate &= ~TSTATE_ASI;
278 regs->tstate |= (ASI_AIUS << 24UL);
282 static void log_unaligned(struct pt_regs *regs)
284 static unsigned long count, last_time;
286 if (time_after(jiffies, last_time + 5 * HZ))
287 count = 0;
288 if (count < 5) {
289 last_time = jiffies;
290 count++;
291 printk("Kernel unaligned access at TPC[%lx] %pS\n",
292 regs->tpc, (void *) regs->tpc);
296 asmlinkage void kernel_unaligned_trap(struct pt_regs *regs, unsigned int insn)
298 enum direction dir = decode_direction(insn);
299 int size = decode_access_size(insn);
300 int orig_asi, asi;
302 current_thread_info()->kern_una_regs = regs;
303 current_thread_info()->kern_una_insn = insn;
305 orig_asi = asi = decode_asi(insn, regs);
307 /* If this is a {get,put}_user() on an unaligned userspace pointer,
308 * just signal a fault and do not log the event.
310 if (asi == ASI_AIUS) {
311 kernel_mna_trap_fault(0);
312 return;
315 log_unaligned(regs);
317 if (!ok_for_kernel(insn) || dir == both) {
318 printk("Unsupported unaligned load/store trap for kernel "
319 "at <%016lx>.\n", regs->tpc);
320 unaligned_panic("Kernel does fpu/atomic "
321 "unaligned load/store.", regs);
323 kernel_mna_trap_fault(0);
324 } else {
325 unsigned long addr, *reg_addr;
326 int err;
328 addr = compute_effective_address(regs, insn,
329 ((insn >> 25) & 0x1f));
330 #ifdef DEBUG_MNA
331 printk("KMNA: pc=%016lx [dir=%s addr=%016lx size=%d] "
332 "retpc[%016lx]\n",
333 regs->tpc, dirstrings[dir], addr, size,
334 regs->u_regs[UREG_RETPC]);
335 #endif
336 switch (asi) {
337 case ASI_NL:
338 case ASI_AIUPL:
339 case ASI_AIUSL:
340 case ASI_PL:
341 case ASI_SL:
342 case ASI_PNFL:
343 case ASI_SNFL:
344 asi &= ~0x08;
345 break;
347 switch (dir) {
348 case load:
349 reg_addr = fetch_reg_addr(((insn>>25)&0x1f), regs);
350 err = do_int_load(reg_addr, size,
351 (unsigned long *) addr,
352 decode_signedness(insn), asi);
353 if (likely(!err) && unlikely(asi != orig_asi)) {
354 unsigned long val_in = *reg_addr;
355 switch (size) {
356 case 2:
357 val_in = swab16(val_in);
358 break;
359 case 4:
360 val_in = swab32(val_in);
361 break;
362 case 8:
363 val_in = swab64(val_in);
364 break;
365 case 16:
366 default:
367 BUG();
368 break;
370 *reg_addr = val_in;
372 break;
374 case store:
375 err = do_int_store(((insn>>25)&0x1f), size,
376 (unsigned long *) addr, regs,
377 asi, orig_asi);
378 break;
380 default:
381 panic("Impossible kernel unaligned trap.");
382 /* Not reached... */
384 if (unlikely(err))
385 kernel_mna_trap_fault(1);
386 else
387 advance(regs);
391 static char popc_helper[] = {
392 0, 1, 1, 2, 1, 2, 2, 3,
393 1, 2, 2, 3, 2, 3, 3, 4,
396 int handle_popc(u32 insn, struct pt_regs *regs)
398 u64 value;
399 int ret, i, rd = ((insn >> 25) & 0x1f);
400 int from_kernel = (regs->tstate & TSTATE_PRIV) != 0;
402 if (insn & 0x2000) {
403 maybe_flush_windows(0, 0, rd, from_kernel);
404 value = sign_extend_imm13(insn);
405 } else {
406 maybe_flush_windows(0, insn & 0x1f, rd, from_kernel);
407 value = fetch_reg(insn & 0x1f, regs);
409 for (ret = 0, i = 0; i < 16; i++) {
410 ret += popc_helper[value & 0xf];
411 value >>= 4;
413 if (rd < 16) {
414 if (rd)
415 regs->u_regs[rd] = ret;
416 } else {
417 if (test_thread_flag(TIF_32BIT)) {
418 struct reg_window32 __user *win32;
419 win32 = (struct reg_window32 __user *)((unsigned long)((u32)regs->u_regs[UREG_FP]));
420 put_user(ret, &win32->locals[rd - 16]);
421 } else {
422 struct reg_window __user *win;
423 win = (struct reg_window __user *)(regs->u_regs[UREG_FP] + STACK_BIAS);
424 put_user(ret, &win->locals[rd - 16]);
427 advance(regs);
428 return 1;
431 extern void do_fpother(struct pt_regs *regs);
432 extern void do_privact(struct pt_regs *regs);
433 extern void spitfire_data_access_exception(struct pt_regs *regs,
434 unsigned long sfsr,
435 unsigned long sfar);
436 extern void sun4v_data_access_exception(struct pt_regs *regs,
437 unsigned long addr,
438 unsigned long type_ctx);
440 int handle_ldf_stq(u32 insn, struct pt_regs *regs)
442 unsigned long addr = compute_effective_address(regs, insn, 0);
443 int freg = ((insn >> 25) & 0x1e) | ((insn >> 20) & 0x20);
444 struct fpustate *f = FPUSTATE;
445 int asi = decode_asi(insn, regs);
446 int flag = (freg < 32) ? FPRS_DL : FPRS_DU;
448 save_and_clear_fpu();
449 current_thread_info()->xfsr[0] &= ~0x1c000;
450 if (freg & 3) {
451 current_thread_info()->xfsr[0] |= (6 << 14) /* invalid_fp_register */;
452 do_fpother(regs);
453 return 0;
455 if (insn & 0x200000) {
456 /* STQ */
457 u64 first = 0, second = 0;
459 if (current_thread_info()->fpsaved[0] & flag) {
460 first = *(u64 *)&f->regs[freg];
461 second = *(u64 *)&f->regs[freg+2];
463 if (asi < 0x80) {
464 do_privact(regs);
465 return 1;
467 switch (asi) {
468 case ASI_P:
469 case ASI_S: break;
470 case ASI_PL:
471 case ASI_SL:
473 /* Need to convert endians */
474 u64 tmp = __swab64p(&first);
476 first = __swab64p(&second);
477 second = tmp;
478 break;
480 default:
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 if (put_user (first >> 32, (u32 __user *)addr) ||
488 __put_user ((u32)first, (u32 __user *)(addr + 4)) ||
489 __put_user (second >> 32, (u32 __user *)(addr + 8)) ||
490 __put_user ((u32)second, (u32 __user *)(addr + 12))) {
491 if (tlb_type == hypervisor)
492 sun4v_data_access_exception(regs, addr, 0);
493 else
494 spitfire_data_access_exception(regs, 0, addr);
495 return 1;
497 } else {
498 /* LDF, LDDF, LDQF */
499 u32 data[4] __attribute__ ((aligned(8)));
500 int size, i;
501 int err;
503 if (asi < 0x80) {
504 do_privact(regs);
505 return 1;
506 } else if (asi > ASI_SNFL) {
507 if (tlb_type == hypervisor)
508 sun4v_data_access_exception(regs, addr, 0);
509 else
510 spitfire_data_access_exception(regs, 0, addr);
511 return 1;
513 switch (insn & 0x180000) {
514 case 0x000000: size = 1; break;
515 case 0x100000: size = 4; break;
516 default: size = 2; break;
518 for (i = 0; i < size; i++)
519 data[i] = 0;
521 err = get_user (data[0], (u32 __user *) addr);
522 if (!err) {
523 for (i = 1; i < size; i++)
524 err |= __get_user (data[i], (u32 __user *)(addr + 4*i));
526 if (err && !(asi & 0x2 /* NF */)) {
527 if (tlb_type == hypervisor)
528 sun4v_data_access_exception(regs, addr, 0);
529 else
530 spitfire_data_access_exception(regs, 0, addr);
531 return 1;
533 if (asi & 0x8) /* Little */ {
534 u64 tmp;
536 switch (size) {
537 case 1: data[0] = le32_to_cpup(data + 0); break;
538 default:*(u64 *)(data + 0) = le64_to_cpup((u64 *)(data + 0));
539 break;
540 case 4: tmp = le64_to_cpup((u64 *)(data + 0));
541 *(u64 *)(data + 0) = le64_to_cpup((u64 *)(data + 2));
542 *(u64 *)(data + 2) = tmp;
543 break;
546 if (!(current_thread_info()->fpsaved[0] & FPRS_FEF)) {
547 current_thread_info()->fpsaved[0] = FPRS_FEF;
548 current_thread_info()->gsr[0] = 0;
550 if (!(current_thread_info()->fpsaved[0] & flag)) {
551 if (freg < 32)
552 memset(f->regs, 0, 32*sizeof(u32));
553 else
554 memset(f->regs+32, 0, 32*sizeof(u32));
556 memcpy(f->regs + freg, data, size * 4);
557 current_thread_info()->fpsaved[0] |= flag;
559 advance(regs);
560 return 1;
563 void handle_ld_nf(u32 insn, struct pt_regs *regs)
565 int rd = ((insn >> 25) & 0x1f);
566 int from_kernel = (regs->tstate & TSTATE_PRIV) != 0;
567 unsigned long *reg;
569 maybe_flush_windows(0, 0, rd, from_kernel);
570 reg = fetch_reg_addr(rd, regs);
571 if (from_kernel || rd < 16) {
572 reg[0] = 0;
573 if ((insn & 0x780000) == 0x180000)
574 reg[1] = 0;
575 } else if (test_thread_flag(TIF_32BIT)) {
576 put_user(0, (int __user *) reg);
577 if ((insn & 0x780000) == 0x180000)
578 put_user(0, ((int __user *) reg) + 1);
579 } else {
580 put_user(0, (unsigned long __user *) reg);
581 if ((insn & 0x780000) == 0x180000)
582 put_user(0, (unsigned long __user *) reg + 1);
584 advance(regs);
587 void handle_lddfmna(struct pt_regs *regs, unsigned long sfar, unsigned long sfsr)
589 unsigned long pc = regs->tpc;
590 unsigned long tstate = regs->tstate;
591 u32 insn;
592 u64 value;
593 u8 freg;
594 int flag;
595 struct fpustate *f = FPUSTATE;
597 if (tstate & TSTATE_PRIV)
598 die_if_kernel("lddfmna from kernel", regs);
599 if (test_thread_flag(TIF_32BIT))
600 pc = (u32)pc;
601 if (get_user(insn, (u32 __user *) pc) != -EFAULT) {
602 int asi = decode_asi(insn, regs);
603 u32 first, second;
604 int err;
606 if ((asi > ASI_SNFL) ||
607 (asi < ASI_P))
608 goto daex;
609 first = second = 0;
610 err = get_user(first, (u32 __user *)sfar);
611 if (!err)
612 err = get_user(second, (u32 __user *)(sfar + 4));
613 if (err) {
614 if (!(asi & 0x2))
615 goto daex;
616 first = second = 0;
618 save_and_clear_fpu();
619 freg = ((insn >> 25) & 0x1e) | ((insn >> 20) & 0x20);
620 value = (((u64)first) << 32) | second;
621 if (asi & 0x8) /* Little */
622 value = __swab64p(&value);
623 flag = (freg < 32) ? FPRS_DL : FPRS_DU;
624 if (!(current_thread_info()->fpsaved[0] & FPRS_FEF)) {
625 current_thread_info()->fpsaved[0] = FPRS_FEF;
626 current_thread_info()->gsr[0] = 0;
628 if (!(current_thread_info()->fpsaved[0] & flag)) {
629 if (freg < 32)
630 memset(f->regs, 0, 32*sizeof(u32));
631 else
632 memset(f->regs+32, 0, 32*sizeof(u32));
634 *(u64 *)(f->regs + freg) = value;
635 current_thread_info()->fpsaved[0] |= flag;
636 } else {
637 daex:
638 if (tlb_type == hypervisor)
639 sun4v_data_access_exception(regs, sfar, sfsr);
640 else
641 spitfire_data_access_exception(regs, sfsr, sfar);
642 return;
644 advance(regs);
645 return;
648 void handle_stdfmna(struct pt_regs *regs, unsigned long sfar, unsigned long sfsr)
650 unsigned long pc = regs->tpc;
651 unsigned long tstate = regs->tstate;
652 u32 insn;
653 u64 value;
654 u8 freg;
655 int flag;
656 struct fpustate *f = FPUSTATE;
658 if (tstate & TSTATE_PRIV)
659 die_if_kernel("stdfmna from kernel", regs);
660 if (test_thread_flag(TIF_32BIT))
661 pc = (u32)pc;
662 if (get_user(insn, (u32 __user *) pc) != -EFAULT) {
663 int asi = decode_asi(insn, regs);
664 freg = ((insn >> 25) & 0x1e) | ((insn >> 20) & 0x20);
665 value = 0;
666 flag = (freg < 32) ? FPRS_DL : FPRS_DU;
667 if ((asi > ASI_SNFL) ||
668 (asi < ASI_P))
669 goto daex;
670 save_and_clear_fpu();
671 if (current_thread_info()->fpsaved[0] & flag)
672 value = *(u64 *)&f->regs[freg];
673 switch (asi) {
674 case ASI_P:
675 case ASI_S: break;
676 case ASI_PL:
677 case ASI_SL:
678 value = __swab64p(&value); break;
679 default: goto daex;
681 if (put_user (value >> 32, (u32 __user *) sfar) ||
682 __put_user ((u32)value, (u32 __user *)(sfar + 4)))
683 goto daex;
684 } else {
685 daex:
686 if (tlb_type == hypervisor)
687 sun4v_data_access_exception(regs, sfar, sfsr);
688 else
689 spitfire_data_access_exception(regs, sfsr, sfar);
690 return;
692 advance(regs);
693 return;