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)
10 #include <linux/jiffies.h>
11 #include <linux/kernel.h>
12 #include <linux/sched.h>
14 #include <linux/extable.h>
16 #include <asm/ptrace.h>
17 #include <asm/pstate.h>
18 #include <asm/processor.h>
19 #include <linux/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 <linux/context_tracking.h>
25 #include <asm/fpumacro.h>
26 #include <asm/cacheflush.h>
27 #include <asm/setup.h>
33 load
, /* ld, ldd, ldh, ldsh */
34 store
, /* st, std, sth, stsh */
35 both
, /* Swap, ldstub, cas, ... */
41 static inline enum direction
decode_direction(unsigned int insn
)
43 unsigned long tmp
= (insn
>> 21) & 1;
48 switch ((insn
>>19)&0xf) {
57 /* 16 = double-word, 8 = extra-word, 4 = word, 2 = half-word */
58 static inline int decode_access_size(struct pt_regs
*regs
, unsigned int insn
)
62 tmp
= ((insn
>> 19) & 0xf);
63 if (tmp
== 11 || tmp
== 14) /* ldx/stx */
69 return 16; /* ldd/std - Although it is actually 8 */
73 printk("Impossible unaligned trap. insn=%08x\n", insn
);
74 die_if_kernel("Byte sized unaligned access?!?!", regs
);
76 /* GCC should never warn that control reaches the end
77 * of this function without returning a value because
78 * die_if_kernel() is marked with attribute 'noreturn'.
79 * Alas, some versions do...
86 static inline int decode_asi(unsigned int insn
, struct pt_regs
*regs
)
88 if (insn
& 0x800000) {
90 return (unsigned char)(regs
->tstate
>> 24); /* %asi */
92 return (unsigned char)(insn
>> 5); /* imm_asi */
97 /* 0x400000 = signed, 0 = unsigned */
98 static inline int decode_signedness(unsigned int insn
)
100 return (insn
& 0x400000);
103 static inline void maybe_flush_windows(unsigned int rs1
, unsigned int rs2
,
104 unsigned int rd
, int from_kernel
)
106 if (rs2
>= 16 || rs1
>= 16 || rd
>= 16) {
107 if (from_kernel
!= 0)
108 __asm__
__volatile__("flushw");
114 static inline long sign_extend_imm13(long imm
)
116 return imm
<< 51 >> 51;
119 static unsigned long fetch_reg(unsigned int reg
, struct pt_regs
*regs
)
121 unsigned long value
, fp
;
124 return (!reg
? 0 : regs
->u_regs
[reg
]);
126 fp
= regs
->u_regs
[UREG_FP
];
128 if (regs
->tstate
& TSTATE_PRIV
) {
129 struct reg_window
*win
;
130 win
= (struct reg_window
*)(fp
+ STACK_BIAS
);
131 value
= win
->locals
[reg
- 16];
132 } else if (!test_thread_64bit_stack(fp
)) {
133 struct reg_window32 __user
*win32
;
134 win32
= (struct reg_window32 __user
*)((unsigned long)((u32
)fp
));
135 get_user(value
, &win32
->locals
[reg
- 16]);
137 struct reg_window __user
*win
;
138 win
= (struct reg_window __user
*)(fp
+ STACK_BIAS
);
139 get_user(value
, &win
->locals
[reg
- 16]);
144 static unsigned long *fetch_reg_addr(unsigned int reg
, struct pt_regs
*regs
)
149 return ®s
->u_regs
[reg
];
151 fp
= regs
->u_regs
[UREG_FP
];
153 if (regs
->tstate
& TSTATE_PRIV
) {
154 struct reg_window
*win
;
155 win
= (struct reg_window
*)(fp
+ STACK_BIAS
);
156 return &win
->locals
[reg
- 16];
157 } else if (!test_thread_64bit_stack(fp
)) {
158 struct reg_window32
*win32
;
159 win32
= (struct reg_window32
*)((unsigned long)((u32
)fp
));
160 return (unsigned long *)&win32
->locals
[reg
- 16];
162 struct reg_window
*win
;
163 win
= (struct reg_window
*)(fp
+ STACK_BIAS
);
164 return &win
->locals
[reg
- 16];
168 unsigned long compute_effective_address(struct pt_regs
*regs
,
169 unsigned int insn
, unsigned int rd
)
171 int from_kernel
= (regs
->tstate
& TSTATE_PRIV
) != 0;
172 unsigned int rs1
= (insn
>> 14) & 0x1f;
173 unsigned int rs2
= insn
& 0x1f;
177 maybe_flush_windows(rs1
, 0, rd
, from_kernel
);
178 addr
= (fetch_reg(rs1
, regs
) + sign_extend_imm13(insn
));
180 maybe_flush_windows(rs1
, rs2
, rd
, from_kernel
);
181 addr
= (fetch_reg(rs1
, regs
) + fetch_reg(rs2
, regs
));
184 if (!from_kernel
&& test_thread_flag(TIF_32BIT
))
190 /* This is just to make gcc think die_if_kernel does return... */
191 static void __used
unaligned_panic(char *str
, struct pt_regs
*regs
)
193 die_if_kernel(str
, regs
);
196 extern int do_int_load(unsigned long *dest_reg
, int size
,
197 unsigned long *saddr
, int is_signed
, int asi
);
199 extern int __do_int_store(unsigned long *dst_addr
, int size
,
200 unsigned long src_val
, int asi
);
202 static inline int do_int_store(int reg_num
, int size
, unsigned long *dst_addr
,
203 struct pt_regs
*regs
, int asi
, int orig_asi
)
205 unsigned long zero
= 0;
206 unsigned long *src_val_p
= &zero
;
207 unsigned long src_val
;
211 zero
= (((long)(reg_num
?
212 (unsigned int)fetch_reg(reg_num
, regs
) : 0)) << 32) |
213 (unsigned int)fetch_reg(reg_num
+ 1, regs
);
214 } else if (reg_num
) {
215 src_val_p
= fetch_reg_addr(reg_num
, regs
);
217 src_val
= *src_val_p
;
218 if (unlikely(asi
!= orig_asi
)) {
221 src_val
= swab16(src_val
);
224 src_val
= swab32(src_val
);
227 src_val
= swab64(src_val
);
235 return __do_int_store(dst_addr
, size
, src_val
, asi
);
238 static inline void advance(struct pt_regs
*regs
)
240 regs
->tpc
= regs
->tnpc
;
242 if (test_thread_flag(TIF_32BIT
)) {
243 regs
->tpc
&= 0xffffffff;
244 regs
->tnpc
&= 0xffffffff;
248 static inline int floating_point_load_or_store_p(unsigned int insn
)
250 return (insn
>> 24) & 1;
253 static inline int ok_for_kernel(unsigned int insn
)
255 return !floating_point_load_or_store_p(insn
);
258 static void kernel_mna_trap_fault(int fixup_tstate_asi
)
260 struct pt_regs
*regs
= current_thread_info()->kern_una_regs
;
261 unsigned int insn
= current_thread_info()->kern_una_insn
;
262 const struct exception_table_entry
*entry
;
264 entry
= search_exception_tables(regs
->tpc
);
266 unsigned long address
;
268 address
= compute_effective_address(regs
, insn
,
269 ((insn
>> 25) & 0x1f));
270 if (address
< PAGE_SIZE
) {
271 printk(KERN_ALERT
"Unable to handle kernel NULL "
272 "pointer dereference in mna handler");
274 printk(KERN_ALERT
"Unable to handle kernel paging "
275 "request in mna handler");
276 printk(KERN_ALERT
" at virtual address %016lx\n",address
);
277 printk(KERN_ALERT
"current->{active_,}mm->context = %016lx\n",
278 (current
->mm
? CTX_HWBITS(current
->mm
->context
) :
279 CTX_HWBITS(current
->active_mm
->context
)));
280 printk(KERN_ALERT
"current->{active_,}mm->pgd = %016lx\n",
281 (current
->mm
? (unsigned long) current
->mm
->pgd
:
282 (unsigned long) current
->active_mm
->pgd
));
283 die_if_kernel("Oops", regs
);
286 regs
->tpc
= entry
->fixup
;
287 regs
->tnpc
= regs
->tpc
+ 4;
289 if (fixup_tstate_asi
) {
290 regs
->tstate
&= ~TSTATE_ASI
;
291 regs
->tstate
|= (ASI_AIUS
<< 24UL);
295 static void log_unaligned(struct pt_regs
*regs
)
297 static DEFINE_RATELIMIT_STATE(ratelimit
, 5 * HZ
, 5);
299 if (__ratelimit(&ratelimit
)) {
300 printk("Kernel unaligned access at TPC[%lx] %pS\n",
301 regs
->tpc
, (void *) regs
->tpc
);
305 asmlinkage
void kernel_unaligned_trap(struct pt_regs
*regs
, unsigned int insn
)
307 enum direction dir
= decode_direction(insn
);
308 int size
= decode_access_size(regs
, insn
);
311 current_thread_info()->kern_una_regs
= regs
;
312 current_thread_info()->kern_una_insn
= insn
;
314 orig_asi
= asi
= decode_asi(insn
, regs
);
316 /* If this is a {get,put}_user() on an unaligned userspace pointer,
317 * just signal a fault and do not log the event.
319 if (asi
== ASI_AIUS
) {
320 kernel_mna_trap_fault(0);
326 if (!ok_for_kernel(insn
) || dir
== both
) {
327 printk("Unsupported unaligned load/store trap for kernel "
328 "at <%016lx>.\n", regs
->tpc
);
329 unaligned_panic("Kernel does fpu/atomic "
330 "unaligned load/store.", regs
);
332 kernel_mna_trap_fault(0);
334 unsigned long addr
, *reg_addr
;
337 addr
= compute_effective_address(regs
, insn
,
338 ((insn
>> 25) & 0x1f));
339 perf_sw_event(PERF_COUNT_SW_ALIGNMENT_FAULTS
, 1, regs
, addr
);
353 reg_addr
= fetch_reg_addr(((insn
>>25)&0x1f), regs
);
354 err
= do_int_load(reg_addr
, size
,
355 (unsigned long *) addr
,
356 decode_signedness(insn
), asi
);
357 if (likely(!err
) && unlikely(asi
!= orig_asi
)) {
358 unsigned long val_in
= *reg_addr
;
361 val_in
= swab16(val_in
);
364 val_in
= swab32(val_in
);
367 val_in
= swab64(val_in
);
379 err
= do_int_store(((insn
>>25)&0x1f), size
,
380 (unsigned long *) addr
, regs
,
385 panic("Impossible kernel unaligned trap.");
389 kernel_mna_trap_fault(1);
395 int handle_popc(u32 insn
, struct pt_regs
*regs
)
397 int from_kernel
= (regs
->tstate
& TSTATE_PRIV
) != 0;
398 int ret
, rd
= ((insn
>> 25) & 0x1f);
401 perf_sw_event(PERF_COUNT_SW_EMULATION_FAULTS
, 1, regs
, 0);
403 maybe_flush_windows(0, 0, rd
, from_kernel
);
404 value
= sign_extend_imm13(insn
);
406 maybe_flush_windows(0, insn
& 0x1f, rd
, from_kernel
);
407 value
= fetch_reg(insn
& 0x1f, regs
);
409 ret
= hweight64(value
);
412 regs
->u_regs
[rd
] = ret
;
414 unsigned long fp
= regs
->u_regs
[UREG_FP
];
416 if (!test_thread_64bit_stack(fp
)) {
417 struct reg_window32 __user
*win32
;
418 win32
= (struct reg_window32 __user
*)((unsigned long)((u32
)fp
));
419 put_user(ret
, &win32
->locals
[rd
- 16]);
421 struct reg_window __user
*win
;
422 win
= (struct reg_window __user
*)(fp
+ STACK_BIAS
);
423 put_user(ret
, &win
->locals
[rd
- 16]);
430 extern void do_fpother(struct pt_regs
*regs
);
431 extern void do_privact(struct pt_regs
*regs
);
432 extern void sun4v_data_access_exception(struct pt_regs
*regs
,
434 unsigned long type_ctx
);
436 int handle_ldf_stq(u32 insn
, struct pt_regs
*regs
)
438 unsigned long addr
= compute_effective_address(regs
, insn
, 0);
440 struct fpustate
*f
= FPUSTATE
;
441 int asi
= decode_asi(insn
, regs
);
444 perf_sw_event(PERF_COUNT_SW_EMULATION_FAULTS
, 1, regs
, 0);
446 save_and_clear_fpu();
447 current_thread_info()->xfsr
[0] &= ~0x1c000;
448 if (insn
& 0x200000) {
450 u64 first
= 0, second
= 0;
452 freg
= ((insn
>> 25) & 0x1e) | ((insn
>> 20) & 0x20);
453 flag
= (freg
< 32) ? FPRS_DL
: FPRS_DU
;
455 current_thread_info()->xfsr
[0] |= (6 << 14) /* invalid_fp_register */;
459 if (current_thread_info()->fpsaved
[0] & flag
) {
460 first
= *(u64
*)&f
->regs
[freg
];
461 second
= *(u64
*)&f
->regs
[freg
+2];
473 /* Need to convert endians */
474 u64 tmp
= __swab64p(&first
);
476 first
= __swab64p(&second
);
481 if (tlb_type
== hypervisor
)
482 sun4v_data_access_exception(regs
, addr
, 0);
484 spitfire_data_access_exception(regs
, 0, addr
);
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);
494 spitfire_data_access_exception(regs
, 0, addr
);
498 /* LDF, LDDF, LDQF */
499 u32 data
[4] __attribute__ ((aligned(8)));
506 } else if (asi
> ASI_SNFL
) {
507 if (tlb_type
== hypervisor
)
508 sun4v_data_access_exception(regs
, addr
, 0);
510 spitfire_data_access_exception(regs
, 0, addr
);
513 switch (insn
& 0x180000) {
514 case 0x000000: size
= 1; break;
515 case 0x100000: size
= 4; break;
516 default: size
= 2; break;
519 freg
= (insn
>> 25) & 0x1f;
521 freg
= ((insn
>> 25) & 0x1e) | ((insn
>> 20) & 0x20);
522 flag
= (freg
< 32) ? FPRS_DL
: FPRS_DU
;
524 for (i
= 0; i
< size
; i
++)
527 err
= get_user (data
[0], (u32 __user
*) addr
);
529 for (i
= 1; i
< size
; i
++)
530 err
|= __get_user (data
[i
], (u32 __user
*)(addr
+ 4*i
));
532 if (err
&& !(asi
& 0x2 /* NF */)) {
533 if (tlb_type
== hypervisor
)
534 sun4v_data_access_exception(regs
, addr
, 0);
536 spitfire_data_access_exception(regs
, 0, addr
);
539 if (asi
& 0x8) /* Little */ {
543 case 1: data
[0] = le32_to_cpup(data
+ 0); break;
544 default:*(u64
*)(data
+ 0) = le64_to_cpup((u64
*)(data
+ 0));
546 case 4: tmp
= le64_to_cpup((u64
*)(data
+ 0));
547 *(u64
*)(data
+ 0) = le64_to_cpup((u64
*)(data
+ 2));
548 *(u64
*)(data
+ 2) = tmp
;
552 if (!(current_thread_info()->fpsaved
[0] & FPRS_FEF
)) {
553 current_thread_info()->fpsaved
[0] = FPRS_FEF
;
554 current_thread_info()->gsr
[0] = 0;
556 if (!(current_thread_info()->fpsaved
[0] & flag
)) {
558 memset(f
->regs
, 0, 32*sizeof(u32
));
560 memset(f
->regs
+32, 0, 32*sizeof(u32
));
562 memcpy(f
->regs
+ freg
, data
, size
* 4);
563 current_thread_info()->fpsaved
[0] |= flag
;
569 void handle_ld_nf(u32 insn
, struct pt_regs
*regs
)
571 int rd
= ((insn
>> 25) & 0x1f);
572 int from_kernel
= (regs
->tstate
& TSTATE_PRIV
) != 0;
575 perf_sw_event(PERF_COUNT_SW_EMULATION_FAULTS
, 1, regs
, 0);
577 maybe_flush_windows(0, 0, rd
, from_kernel
);
578 reg
= fetch_reg_addr(rd
, regs
);
579 if (from_kernel
|| rd
< 16) {
581 if ((insn
& 0x780000) == 0x180000)
583 } else if (!test_thread_64bit_stack(regs
->u_regs
[UREG_FP
])) {
584 put_user(0, (int __user
*) reg
);
585 if ((insn
& 0x780000) == 0x180000)
586 put_user(0, ((int __user
*) reg
) + 1);
588 put_user(0, (unsigned long __user
*) reg
);
589 if ((insn
& 0x780000) == 0x180000)
590 put_user(0, (unsigned long __user
*) reg
+ 1);
595 void handle_lddfmna(struct pt_regs
*regs
, unsigned long sfar
, unsigned long sfsr
)
597 enum ctx_state prev_state
= exception_enter();
598 unsigned long pc
= regs
->tpc
;
599 unsigned long tstate
= regs
->tstate
;
604 struct fpustate
*f
= FPUSTATE
;
606 if (tstate
& TSTATE_PRIV
)
607 die_if_kernel("lddfmna from kernel", regs
);
608 perf_sw_event(PERF_COUNT_SW_ALIGNMENT_FAULTS
, 1, regs
, sfar
);
609 if (test_thread_flag(TIF_32BIT
))
611 if (get_user(insn
, (u32 __user
*) pc
) != -EFAULT
) {
612 int asi
= decode_asi(insn
, regs
);
616 if ((asi
> ASI_SNFL
) ||
620 err
= get_user(first
, (u32 __user
*)sfar
);
622 err
= get_user(second
, (u32 __user
*)(sfar
+ 4));
628 save_and_clear_fpu();
629 freg
= ((insn
>> 25) & 0x1e) | ((insn
>> 20) & 0x20);
630 value
= (((u64
)first
) << 32) | second
;
631 if (asi
& 0x8) /* Little */
632 value
= __swab64p(&value
);
633 flag
= (freg
< 32) ? FPRS_DL
: FPRS_DU
;
634 if (!(current_thread_info()->fpsaved
[0] & FPRS_FEF
)) {
635 current_thread_info()->fpsaved
[0] = FPRS_FEF
;
636 current_thread_info()->gsr
[0] = 0;
638 if (!(current_thread_info()->fpsaved
[0] & flag
)) {
640 memset(f
->regs
, 0, 32*sizeof(u32
));
642 memset(f
->regs
+32, 0, 32*sizeof(u32
));
644 *(u64
*)(f
->regs
+ freg
) = value
;
645 current_thread_info()->fpsaved
[0] |= flag
;
648 if (tlb_type
== hypervisor
)
649 sun4v_data_access_exception(regs
, sfar
, sfsr
);
651 spitfire_data_access_exception(regs
, sfsr
, sfar
);
656 exception_exit(prev_state
);
659 void handle_stdfmna(struct pt_regs
*regs
, unsigned long sfar
, unsigned long sfsr
)
661 enum ctx_state prev_state
= exception_enter();
662 unsigned long pc
= regs
->tpc
;
663 unsigned long tstate
= regs
->tstate
;
668 struct fpustate
*f
= FPUSTATE
;
670 if (tstate
& TSTATE_PRIV
)
671 die_if_kernel("stdfmna from kernel", regs
);
672 perf_sw_event(PERF_COUNT_SW_ALIGNMENT_FAULTS
, 1, regs
, sfar
);
673 if (test_thread_flag(TIF_32BIT
))
675 if (get_user(insn
, (u32 __user
*) pc
) != -EFAULT
) {
676 int asi
= decode_asi(insn
, regs
);
677 freg
= ((insn
>> 25) & 0x1e) | ((insn
>> 20) & 0x20);
679 flag
= (freg
< 32) ? FPRS_DL
: FPRS_DU
;
680 if ((asi
> ASI_SNFL
) ||
683 save_and_clear_fpu();
684 if (current_thread_info()->fpsaved
[0] & flag
)
685 value
= *(u64
*)&f
->regs
[freg
];
691 value
= __swab64p(&value
); break;
694 if (put_user (value
>> 32, (u32 __user
*) sfar
) ||
695 __put_user ((u32
)value
, (u32 __user
*)(sfar
+ 4)))
699 if (tlb_type
== hypervisor
)
700 sun4v_data_access_exception(regs
, sfar
, sfsr
);
702 spitfire_data_access_exception(regs
, sfsr
, sfar
);
707 exception_exit(prev_state
);