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/ptrace.h>
14 #include <linux/mman.h>
15 #include <linux/signal.h>
17 #include <linux/module.h>
18 #include <linux/init.h>
19 #include <linux/perf_event.h>
20 #include <linux/interrupt.h>
21 #include <linux/kprobes.h>
22 #include <linux/kdebug.h>
23 #include <linux/percpu.h>
24 #include <linux/context_tracking.h>
27 #include <asm/pgtable.h>
28 #include <asm/openprom.h>
29 #include <asm/oplib.h>
30 #include <asm/uaccess.h>
33 #include <asm/sections.h>
34 #include <asm/mmu_context.h>
36 int show_unhandled_signals = 1;
38 static inline __kprobes int notify_page_fault(struct pt_regs *regs)
42 /* kprobe_running() needs smp_processor_id() */
43 if (kprobes_built_in() && !user_mode(regs)) {
45 if (kprobe_running() && kprobe_fault_handler(regs, 0))
52 static void __kprobes unhandled_fault(unsigned long address,
53 struct task_struct *tsk,
56 if ((unsigned long) address < PAGE_SIZE) {
57 printk(KERN_ALERT "Unable to handle kernel NULL "
58 "pointer dereference\n");
60 printk(KERN_ALERT "Unable to handle kernel paging request "
61 "at virtual address %016lx\n", (unsigned long)address);
63 printk(KERN_ALERT "tsk->{mm,active_mm}->context = %016lx\n",
65 CTX_HWBITS(tsk->mm->context) :
66 CTX_HWBITS(tsk->active_mm->context)));
67 printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %016lx\n",
68 (tsk->mm ? (unsigned long) tsk->mm->pgd :
69 (unsigned long) tsk->active_mm->pgd));
70 die_if_kernel("Oops", regs);
73 static void __kprobes bad_kernel_pc(struct pt_regs *regs, unsigned long vaddr)
75 printk(KERN_CRIT "OOPS: Bogus kernel PC [%016lx] in fault handler\n",
77 printk(KERN_CRIT "OOPS: RPC [%016lx]\n", regs->u_regs[15]);
78 printk("OOPS: RPC <%pS>\n", (void *) regs->u_regs[15]);
79 printk(KERN_CRIT "OOPS: Fault was to vaddr[%lx]\n", vaddr);
81 unhandled_fault(regs->tpc, current, regs);
85 * We now make sure that mmap_sem is held in all paths that call
86 * this. Additionally, to prevent kswapd from ripping ptes from
87 * under us, raise interrupts around the time that we look at the
88 * pte, kswapd will have to wait to get his smp ipi response from
89 * us. vmtruncate likewise. This saves us having to get pte lock.
91 static unsigned int get_user_insn(unsigned long tpc)
93 pgd_t *pgdp = pgd_offset(current->mm, tpc);
100 if (pgd_none(*pgdp) || unlikely(pgd_bad(*pgdp)))
102 pudp = pud_offset(pgdp, tpc);
103 if (pud_none(*pudp) || unlikely(pud_bad(*pudp)))
106 /* This disables preemption for us as well. */
109 pmdp = pmd_offset(pudp, tpc);
110 if (pmd_none(*pmdp) || unlikely(pmd_bad(*pmdp)))
113 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
114 if (pmd_trans_huge(*pmdp)) {
115 if (pmd_trans_splitting(*pmdp))
118 pa = pmd_pfn(*pmdp) << PAGE_SHIFT;
119 pa += tpc & ~HPAGE_MASK;
121 /* Use phys bypass so we don't pollute dtlb/dcache. */
122 __asm__ __volatile__("lduwa [%1] %2, %0"
124 : "r" (pa), "i" (ASI_PHYS_USE_EC));
128 ptep = pte_offset_map(pmdp, tpc);
130 if (pte_present(pte)) {
131 pa = (pte_pfn(pte) << PAGE_SHIFT);
132 pa += (tpc & ~PAGE_MASK);
134 /* Use phys bypass so we don't pollute dtlb/dcache. */
135 __asm__ __volatile__("lduwa [%1] %2, %0"
137 : "r" (pa), "i" (ASI_PHYS_USE_EC));
148 show_signal_msg(struct pt_regs *regs, int sig, int code,
149 unsigned long address, struct task_struct *tsk)
151 if (!unhandled_signal(tsk, sig))
154 if (!printk_ratelimit())
157 printk("%s%s[%d]: segfault at %lx ip %p (rpc %p) sp %p error %x",
158 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
159 tsk->comm, task_pid_nr(tsk), address,
160 (void *)regs->tpc, (void *)regs->u_regs[UREG_I7],
161 (void *)regs->u_regs[UREG_FP], code);
163 print_vma_addr(KERN_CONT " in ", regs->tpc);
165 printk(KERN_CONT "\n");
168 static void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
169 unsigned long fault_addr, unsigned int insn,
178 if (fault_code & FAULT_CODE_ITLB) {
181 /* If we were able to probe the faulting instruction, use it
182 * to compute a precise fault address. Otherwise use the fault
183 * time provided address which may only have page granularity.
186 addr = compute_effective_address(regs, insn, 0);
190 info.si_addr = (void __user *) addr;
193 if (unlikely(show_unhandled_signals))
194 show_signal_msg(regs, sig, code, addr, current);
196 force_sig_info(sig, &info, current);
199 extern int handle_ldf_stq(u32, struct pt_regs *);
200 extern int handle_ld_nf(u32, struct pt_regs *);
202 static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
205 if (!regs->tpc || (regs->tpc & 0x3))
207 if (regs->tstate & TSTATE_PRIV) {
208 insn = *(unsigned int *) regs->tpc;
210 insn = get_user_insn(regs->tpc);
216 static void __kprobes do_kernel_fault(struct pt_regs *regs, int si_code,
217 int fault_code, unsigned int insn,
218 unsigned long address)
220 unsigned char asi = ASI_P;
222 if ((!insn) && (regs->tstate & TSTATE_PRIV))
225 /* If user insn could be read (thus insn is zero), that
226 * is fine. We will just gun down the process with a signal
230 if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
231 (insn & 0xc0800000) == 0xc0800000) {
233 asi = (regs->tstate >> 24);
236 if ((asi & 0xf2) == 0x82) {
237 if (insn & 0x1000000) {
238 handle_ldf_stq(insn, regs);
240 /* This was a non-faulting load. Just clear the
241 * destination register(s) and continue with the next
244 handle_ld_nf(insn, regs);
250 /* Is this in ex_table? */
251 if (regs->tstate & TSTATE_PRIV) {
252 const struct exception_table_entry *entry;
254 entry = search_exception_tables(regs->tpc);
256 regs->tpc = entry->fixup;
257 regs->tnpc = regs->tpc + 4;
261 /* The si_code was set to make clear whether
262 * this was a SEGV_MAPERR or SEGV_ACCERR fault.
264 do_fault_siginfo(si_code, SIGSEGV, regs, address, insn, fault_code);
269 unhandled_fault (address, current, regs);
272 static void noinline __kprobes bogus_32bit_fault_tpc(struct pt_regs *regs)
277 printk(KERN_ERR "FAULT[%s:%d]: 32-bit process reports "
278 "64-bit TPC [%lx]\n",
279 current->comm, current->pid,
284 asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
286 enum ctx_state prev_state = exception_enter();
287 struct mm_struct *mm = current->mm;
288 struct vm_area_struct *vma;
289 unsigned int insn = 0;
290 int si_code, fault_code, fault;
291 unsigned long address, mm_rss;
292 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
294 fault_code = get_thread_fault_code();
296 if (notify_page_fault(regs))
299 si_code = SEGV_MAPERR;
300 address = current_thread_info()->fault_address;
302 if ((fault_code & FAULT_CODE_ITLB) &&
303 (fault_code & FAULT_CODE_DTLB))
306 if (test_thread_flag(TIF_32BIT)) {
307 if (!(regs->tstate & TSTATE_PRIV)) {
308 if (unlikely((regs->tpc >> 32) != 0)) {
309 bogus_32bit_fault_tpc(regs);
313 if (unlikely((address >> 32) != 0))
317 if (regs->tstate & TSTATE_PRIV) {
318 unsigned long tpc = regs->tpc;
320 /* Sanity check the PC. */
321 if ((tpc >= KERNBASE && tpc < (unsigned long) __init_end) ||
322 (tpc >= MODULES_VADDR && tpc < MODULES_END)) {
323 /* Valid, no problems... */
325 bad_kernel_pc(regs, address);
329 flags |= FAULT_FLAG_USER;
332 * If we're in an interrupt or have no user
333 * context, we must not take the fault..
335 if (in_atomic() || !mm)
338 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
340 if (!down_read_trylock(&mm->mmap_sem)) {
341 if ((regs->tstate & TSTATE_PRIV) &&
342 !search_exception_tables(regs->tpc)) {
343 insn = get_fault_insn(regs, insn);
344 goto handle_kernel_fault;
348 down_read(&mm->mmap_sem);
351 vma = find_vma(mm, address);
355 /* Pure DTLB misses do not tell us whether the fault causing
356 * load/store/atomic was a write or not, it only says that there
357 * was no match. So in such a case we (carefully) read the
358 * instruction to try and figure this out. It's an optimization
359 * so it's ok if we can't do this.
361 * Special hack, window spill/fill knows the exact fault type.
364 (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) &&
365 (vma->vm_flags & VM_WRITE) != 0) {
366 insn = get_fault_insn(regs, 0);
369 /* All loads, stores and atomics have bits 30 and 31 both set
370 * in the instruction. Bit 21 is set in all stores, but we
371 * have to avoid prefetches which also have bit 21 set.
373 if ((insn & 0xc0200000) == 0xc0200000 &&
374 (insn & 0x01780000) != 0x01680000) {
375 /* Don't bother updating thread struct value,
376 * because update_mmu_cache only cares which tlb
377 * the access came from.
379 fault_code |= FAULT_CODE_WRITE;
384 if (vma->vm_start <= address)
386 if (!(vma->vm_flags & VM_GROWSDOWN))
388 if (!(fault_code & FAULT_CODE_WRITE)) {
389 /* Non-faulting loads shouldn't expand stack. */
390 insn = get_fault_insn(regs, insn);
391 if ((insn & 0xc0800000) == 0xc0800000) {
395 asi = (regs->tstate >> 24);
398 if ((asi & 0xf2) == 0x82)
402 if (expand_stack(vma, address))
405 * Ok, we have a good vm_area for this memory access, so
409 si_code = SEGV_ACCERR;
411 /* If we took a ITLB miss on a non-executable page, catch
414 if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) {
415 BUG_ON(address != regs->tpc);
416 BUG_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(mm, 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_SIGBUS)
453 if (flags & FAULT_FLAG_ALLOW_RETRY) {
454 if (fault & VM_FAULT_MAJOR) {
456 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ,
460 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN,
463 if (fault & VM_FAULT_RETRY) {
464 flags &= ~FAULT_FLAG_ALLOW_RETRY;
465 flags |= FAULT_FLAG_TRIED;
467 /* No need to up_read(&mm->mmap_sem) as we would
468 * have already released it in __lock_page_or_retry
475 up_read(&mm->mmap_sem);
477 mm_rss = get_mm_rss(mm);
478 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
479 mm_rss -= (mm->context.huge_pte_count * (HPAGE_SIZE / PAGE_SIZE));
481 if (unlikely(mm_rss >
482 mm->context.tsb_block[MM_TSB_BASE].tsb_rss_limit))
483 tsb_grow(mm, MM_TSB_BASE, mm_rss);
484 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
485 mm_rss = mm->context.huge_pte_count;
486 if (unlikely(mm_rss >
487 mm->context.tsb_block[MM_TSB_HUGE].tsb_rss_limit)) {
488 if (mm->context.tsb_block[MM_TSB_HUGE].tsb)
489 tsb_grow(mm, MM_TSB_HUGE, mm_rss);
496 exception_exit(prev_state);
500 * Something tried to access memory that isn't in our memory map..
501 * Fix it, but check if it's kernel or user first..
504 insn = get_fault_insn(regs, insn);
505 up_read(&mm->mmap_sem);
508 do_kernel_fault(regs, si_code, fault_code, insn, address);
512 * We ran out of memory, or some other thing happened to us that made
513 * us unable to handle the page fault gracefully.
516 insn = get_fault_insn(regs, insn);
517 up_read(&mm->mmap_sem);
518 if (!(regs->tstate & TSTATE_PRIV)) {
519 pagefault_out_of_memory();
522 goto handle_kernel_fault;
525 insn = get_fault_insn(regs, 0);
526 goto handle_kernel_fault;
529 insn = get_fault_insn(regs, insn);
530 up_read(&mm->mmap_sem);
533 * Send a sigbus, regardless of whether we were in kernel
536 do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, address, insn, fault_code);
538 /* Kernel mode? Handle exceptions or die */
539 if (regs->tstate & TSTATE_PRIV)
540 goto handle_kernel_fault;