2 * arch/sparc64/mm/fault.c: Page fault handlers for the 64-bit Sparc.
4 * Copyright (C) 1996, 2008 David S. Miller (davem@davemloft.net)
5 * Copyright (C) 1997, 1999 Jakub Jelinek (jj@ultra.linux.cz)
10 #include <linux/string.h>
11 #include <linux/types.h>
12 #include <linux/sched.h>
13 #include <linux/sched/debug.h>
14 #include <linux/ptrace.h>
15 #include <linux/mman.h>
16 #include <linux/signal.h>
18 #include <linux/extable.h>
19 #include <linux/init.h>
20 #include <linux/perf_event.h>
21 #include <linux/interrupt.h>
22 #include <linux/kprobes.h>
23 #include <linux/kdebug.h>
24 #include <linux/percpu.h>
25 #include <linux/context_tracking.h>
26 #include <linux/uaccess.h>
29 #include <asm/pgtable.h>
30 #include <asm/openprom.h>
31 #include <asm/oplib.h>
34 #include <asm/sections.h>
35 #include <asm/mmu_context.h>
36 #include <asm/setup.h>
38 int show_unhandled_signals
= 1;
40 static inline __kprobes
int notify_page_fault(struct pt_regs
*regs
)
44 /* kprobe_running() needs smp_processor_id() */
45 if (kprobes_built_in() && !user_mode(regs
)) {
47 if (kprobe_running() && kprobe_fault_handler(regs
, 0))
54 static void __kprobes
unhandled_fault(unsigned long address
,
55 struct task_struct
*tsk
,
58 if ((unsigned long) address
< PAGE_SIZE
) {
59 printk(KERN_ALERT
"Unable to handle kernel NULL "
60 "pointer dereference\n");
62 printk(KERN_ALERT
"Unable to handle kernel paging request "
63 "at virtual address %016lx\n", (unsigned long)address
);
65 printk(KERN_ALERT
"tsk->{mm,active_mm}->context = %016lx\n",
67 CTX_HWBITS(tsk
->mm
->context
) :
68 CTX_HWBITS(tsk
->active_mm
->context
)));
69 printk(KERN_ALERT
"tsk->{mm,active_mm}->pgd = %016lx\n",
70 (tsk
->mm
? (unsigned long) tsk
->mm
->pgd
:
71 (unsigned long) tsk
->active_mm
->pgd
));
72 die_if_kernel("Oops", regs
);
75 static void __kprobes
bad_kernel_pc(struct pt_regs
*regs
, unsigned long vaddr
)
77 printk(KERN_CRIT
"OOPS: Bogus kernel PC [%016lx] in fault handler\n",
79 printk(KERN_CRIT
"OOPS: RPC [%016lx]\n", regs
->u_regs
[15]);
80 printk("OOPS: RPC <%pS>\n", (void *) regs
->u_regs
[15]);
81 printk(KERN_CRIT
"OOPS: Fault was to vaddr[%lx]\n", vaddr
);
83 unhandled_fault(regs
->tpc
, current
, regs
);
87 * We now make sure that mmap_sem is held in all paths that call
88 * this. Additionally, to prevent kswapd from ripping ptes from
89 * under us, raise interrupts around the time that we look at the
90 * pte, kswapd will have to wait to get his smp ipi response from
91 * us. vmtruncate likewise. This saves us having to get pte lock.
93 static unsigned int get_user_insn(unsigned long tpc
)
95 pgd_t
*pgdp
= pgd_offset(current
->mm
, tpc
);
102 if (pgd_none(*pgdp
) || unlikely(pgd_bad(*pgdp
)))
104 pudp
= pud_offset(pgdp
, tpc
);
105 if (pud_none(*pudp
) || unlikely(pud_bad(*pudp
)))
108 /* This disables preemption for us as well. */
111 pmdp
= pmd_offset(pudp
, tpc
);
112 if (pmd_none(*pmdp
) || unlikely(pmd_bad(*pmdp
)))
115 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
116 if (is_hugetlb_pmd(*pmdp
)) {
117 pa
= pmd_pfn(*pmdp
) << PAGE_SHIFT
;
118 pa
+= tpc
& ~HPAGE_MASK
;
120 /* Use phys bypass so we don't pollute dtlb/dcache. */
121 __asm__
__volatile__("lduwa [%1] %2, %0"
123 : "r" (pa
), "i" (ASI_PHYS_USE_EC
));
127 ptep
= pte_offset_map(pmdp
, tpc
);
129 if (pte_present(pte
)) {
130 pa
= (pte_pfn(pte
) << PAGE_SHIFT
);
131 pa
+= (tpc
& ~PAGE_MASK
);
133 /* Use phys bypass so we don't pollute dtlb/dcache. */
134 __asm__
__volatile__("lduwa [%1] %2, %0"
136 : "r" (pa
), "i" (ASI_PHYS_USE_EC
));
147 show_signal_msg(struct pt_regs
*regs
, int sig
, int code
,
148 unsigned long address
, struct task_struct
*tsk
)
150 if (!unhandled_signal(tsk
, sig
))
153 if (!printk_ratelimit())
156 printk("%s%s[%d]: segfault at %lx ip %p (rpc %p) sp %p error %x",
157 task_pid_nr(tsk
) > 1 ? KERN_INFO
: KERN_EMERG
,
158 tsk
->comm
, task_pid_nr(tsk
), address
,
159 (void *)regs
->tpc
, (void *)regs
->u_regs
[UREG_I7
],
160 (void *)regs
->u_regs
[UREG_FP
], code
);
162 print_vma_addr(KERN_CONT
" in ", regs
->tpc
);
164 printk(KERN_CONT
"\n");
167 static void do_fault_siginfo(int code
, int sig
, struct pt_regs
*regs
,
168 unsigned long fault_addr
, unsigned int insn
,
177 if (fault_code
& FAULT_CODE_ITLB
) {
180 /* If we were able to probe the faulting instruction, use it
181 * to compute a precise fault address. Otherwise use the fault
182 * time provided address which may only have page granularity.
185 addr
= compute_effective_address(regs
, insn
, 0);
189 info
.si_addr
= (void __user
*) addr
;
192 if (unlikely(show_unhandled_signals
))
193 show_signal_msg(regs
, sig
, code
, addr
, current
);
195 force_sig_info(sig
, &info
, current
);
198 static unsigned int get_fault_insn(struct pt_regs
*regs
, unsigned int insn
)
201 if (!regs
->tpc
|| (regs
->tpc
& 0x3))
203 if (regs
->tstate
& TSTATE_PRIV
) {
204 insn
= *(unsigned int *) regs
->tpc
;
206 insn
= get_user_insn(regs
->tpc
);
212 static void __kprobes
do_kernel_fault(struct pt_regs
*regs
, int si_code
,
213 int fault_code
, unsigned int insn
,
214 unsigned long address
)
216 unsigned char asi
= ASI_P
;
218 if ((!insn
) && (regs
->tstate
& TSTATE_PRIV
))
221 /* If user insn could be read (thus insn is zero), that
222 * is fine. We will just gun down the process with a signal
226 if (!(fault_code
& (FAULT_CODE_WRITE
|FAULT_CODE_ITLB
)) &&
227 (insn
& 0xc0800000) == 0xc0800000) {
229 asi
= (regs
->tstate
>> 24);
232 if ((asi
& 0xf2) == 0x82) {
233 if (insn
& 0x1000000) {
234 handle_ldf_stq(insn
, regs
);
236 /* This was a non-faulting load. Just clear the
237 * destination register(s) and continue with the next
240 handle_ld_nf(insn
, regs
);
246 /* Is this in ex_table? */
247 if (regs
->tstate
& TSTATE_PRIV
) {
248 const struct exception_table_entry
*entry
;
250 entry
= search_exception_tables(regs
->tpc
);
252 regs
->tpc
= entry
->fixup
;
253 regs
->tnpc
= regs
->tpc
+ 4;
257 /* The si_code was set to make clear whether
258 * this was a SEGV_MAPERR or SEGV_ACCERR fault.
260 do_fault_siginfo(si_code
, SIGSEGV
, regs
, address
, insn
, fault_code
);
265 unhandled_fault (address
, current
, regs
);
268 static void noinline __kprobes
bogus_32bit_fault_tpc(struct pt_regs
*regs
)
273 printk(KERN_ERR
"FAULT[%s:%d]: 32-bit process reports "
274 "64-bit TPC [%lx]\n",
275 current
->comm
, current
->pid
,
280 asmlinkage
void __kprobes
do_sparc64_fault(struct pt_regs
*regs
)
282 enum ctx_state prev_state
= exception_enter();
283 struct mm_struct
*mm
= current
->mm
;
284 struct vm_area_struct
*vma
;
285 unsigned int insn
= 0;
286 int si_code
, fault_code
, fault
;
287 unsigned long address
, mm_rss
;
288 unsigned int flags
= FAULT_FLAG_ALLOW_RETRY
| FAULT_FLAG_KILLABLE
;
290 fault_code
= get_thread_fault_code();
292 if (notify_page_fault(regs
))
295 si_code
= SEGV_MAPERR
;
296 address
= current_thread_info()->fault_address
;
298 if ((fault_code
& FAULT_CODE_ITLB
) &&
299 (fault_code
& FAULT_CODE_DTLB
))
302 if (test_thread_flag(TIF_32BIT
)) {
303 if (!(regs
->tstate
& TSTATE_PRIV
)) {
304 if (unlikely((regs
->tpc
>> 32) != 0)) {
305 bogus_32bit_fault_tpc(regs
);
309 if (unlikely((address
>> 32) != 0))
313 if (regs
->tstate
& TSTATE_PRIV
) {
314 unsigned long tpc
= regs
->tpc
;
316 /* Sanity check the PC. */
317 if ((tpc
>= KERNBASE
&& tpc
< (unsigned long) __init_end
) ||
318 (tpc
>= MODULES_VADDR
&& tpc
< MODULES_END
)) {
319 /* Valid, no problems... */
321 bad_kernel_pc(regs
, address
);
325 flags
|= FAULT_FLAG_USER
;
328 * If we're in an interrupt or have no user
329 * context, we must not take the fault..
331 if (faulthandler_disabled() || !mm
)
334 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS
, 1, regs
, address
);
336 if (!down_read_trylock(&mm
->mmap_sem
)) {
337 if ((regs
->tstate
& TSTATE_PRIV
) &&
338 !search_exception_tables(regs
->tpc
)) {
339 insn
= get_fault_insn(regs
, insn
);
340 goto handle_kernel_fault
;
344 down_read(&mm
->mmap_sem
);
347 if (fault_code
& FAULT_CODE_BAD_RA
)
350 vma
= find_vma(mm
, address
);
354 /* Pure DTLB misses do not tell us whether the fault causing
355 * load/store/atomic was a write or not, it only says that there
356 * was no match. So in such a case we (carefully) read the
357 * instruction to try and figure this out. It's an optimization
358 * so it's ok if we can't do this.
360 * Special hack, window spill/fill knows the exact fault type.
363 (FAULT_CODE_DTLB
| FAULT_CODE_WRITE
| FAULT_CODE_WINFIXUP
)) == FAULT_CODE_DTLB
) &&
364 (vma
->vm_flags
& VM_WRITE
) != 0) {
365 insn
= get_fault_insn(regs
, 0);
368 /* All loads, stores and atomics have bits 30 and 31 both set
369 * in the instruction. Bit 21 is set in all stores, but we
370 * have to avoid prefetches which also have bit 21 set.
372 if ((insn
& 0xc0200000) == 0xc0200000 &&
373 (insn
& 0x01780000) != 0x01680000) {
374 /* Don't bother updating thread struct value,
375 * because update_mmu_cache only cares which tlb
376 * the access came from.
378 fault_code
|= FAULT_CODE_WRITE
;
383 if (vma
->vm_start
<= address
)
385 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
387 if (!(fault_code
& FAULT_CODE_WRITE
)) {
388 /* Non-faulting loads shouldn't expand stack. */
389 insn
= get_fault_insn(regs
, insn
);
390 if ((insn
& 0xc0800000) == 0xc0800000) {
394 asi
= (regs
->tstate
>> 24);
397 if ((asi
& 0xf2) == 0x82)
401 if (expand_stack(vma
, address
))
404 * Ok, we have a good vm_area for this memory access, so
408 si_code
= SEGV_ACCERR
;
410 /* If we took a ITLB miss on a non-executable page, catch
413 if ((fault_code
& FAULT_CODE_ITLB
) && !(vma
->vm_flags
& VM_EXEC
)) {
414 WARN(address
!= regs
->tpc
,
415 "address (%lx) != regs->tpc (%lx)\n", address
, regs
->tpc
);
416 WARN_ON(regs
->tstate
& TSTATE_PRIV
);
420 if (fault_code
& FAULT_CODE_WRITE
) {
421 if (!(vma
->vm_flags
& VM_WRITE
))
424 /* Spitfire has an icache which does not snoop
425 * processor stores. Later processors do...
427 if (tlb_type
== spitfire
&&
428 (vma
->vm_flags
& VM_EXEC
) != 0 &&
429 vma
->vm_file
!= NULL
)
430 set_thread_fault_code(fault_code
|
431 FAULT_CODE_BLKCOMMIT
);
433 flags
|= FAULT_FLAG_WRITE
;
435 /* Allow reads even for write-only mappings */
436 if (!(vma
->vm_flags
& (VM_READ
| VM_EXEC
)))
440 fault
= handle_mm_fault(vma
, address
, flags
);
442 if ((fault
& VM_FAULT_RETRY
) && fatal_signal_pending(current
))
445 if (unlikely(fault
& VM_FAULT_ERROR
)) {
446 if (fault
& VM_FAULT_OOM
)
448 else if (fault
& VM_FAULT_SIGSEGV
)
450 else if (fault
& VM_FAULT_SIGBUS
)
455 if (flags
& FAULT_FLAG_ALLOW_RETRY
) {
456 if (fault
& VM_FAULT_MAJOR
) {
458 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ
,
462 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN
,
465 if (fault
& VM_FAULT_RETRY
) {
466 flags
&= ~FAULT_FLAG_ALLOW_RETRY
;
467 flags
|= FAULT_FLAG_TRIED
;
469 /* No need to up_read(&mm->mmap_sem) as we would
470 * have already released it in __lock_page_or_retry
477 up_read(&mm
->mmap_sem
);
479 mm_rss
= get_mm_rss(mm
);
480 #if defined(CONFIG_TRANSPARENT_HUGEPAGE)
481 mm_rss
-= (mm
->context
.thp_pte_count
* (HPAGE_SIZE
/ PAGE_SIZE
));
483 if (unlikely(mm_rss
>
484 mm
->context
.tsb_block
[MM_TSB_BASE
].tsb_rss_limit
))
485 tsb_grow(mm
, MM_TSB_BASE
, mm_rss
);
486 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
487 mm_rss
= mm
->context
.hugetlb_pte_count
+ mm
->context
.thp_pte_count
;
488 mm_rss
*= REAL_HPAGE_PER_HPAGE
;
489 if (unlikely(mm_rss
>
490 mm
->context
.tsb_block
[MM_TSB_HUGE
].tsb_rss_limit
)) {
491 if (mm
->context
.tsb_block
[MM_TSB_HUGE
].tsb
)
492 tsb_grow(mm
, MM_TSB_HUGE
, mm_rss
);
499 exception_exit(prev_state
);
503 * Something tried to access memory that isn't in our memory map..
504 * Fix it, but check if it's kernel or user first..
507 insn
= get_fault_insn(regs
, insn
);
508 up_read(&mm
->mmap_sem
);
511 do_kernel_fault(regs
, si_code
, fault_code
, insn
, address
);
515 * We ran out of memory, or some other thing happened to us that made
516 * us unable to handle the page fault gracefully.
519 insn
= get_fault_insn(regs
, insn
);
520 up_read(&mm
->mmap_sem
);
521 if (!(regs
->tstate
& TSTATE_PRIV
)) {
522 pagefault_out_of_memory();
525 goto handle_kernel_fault
;
528 insn
= get_fault_insn(regs
, 0);
529 goto handle_kernel_fault
;
532 insn
= get_fault_insn(regs
, insn
);
533 up_read(&mm
->mmap_sem
);
536 * Send a sigbus, regardless of whether we were in kernel
539 do_fault_siginfo(BUS_ADRERR
, SIGBUS
, regs
, address
, insn
, fault_code
);
541 /* Kernel mode? Handle exceptions or die */
542 if (regs
->tstate
& TSTATE_PRIV
)
543 goto handle_kernel_fault
;