2 * arch/sparc64/mm/fault.c: Page fault handlers for the 64-bit Sparc.
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>
26 #include <asm/pgtable.h>
27 #include <asm/openprom.h>
28 #include <asm/oplib.h>
29 #include <asm/uaccess.h>
32 #include <asm/sections.h>
33 #include <asm/mmu_context.h>
35 int show_unhandled_signals = 1;
37 static inline __kprobes int notify_page_fault(struct pt_regs *regs)
41 /* kprobe_running() needs smp_processor_id() */
42 if (kprobes_built_in() && !user_mode(regs)) {
44 if (kprobe_running() && kprobe_fault_handler(regs, 0))
51 static void __kprobes unhandled_fault(unsigned long address,
52 struct task_struct *tsk,
55 if ((unsigned long) address < PAGE_SIZE) {
56 printk(KERN_ALERT "Unable to handle kernel NULL "
57 "pointer dereference\n");
59 printk(KERN_ALERT "Unable to handle kernel paging request "
60 "at virtual address %016lx\n", (unsigned long)address);
62 printk(KERN_ALERT "tsk->{mm,active_mm}->context = %016lx\n",
64 CTX_HWBITS(tsk->mm->context) :
65 CTX_HWBITS(tsk->active_mm->context)));
66 printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %016lx\n",
67 (tsk->mm ? (unsigned long) tsk->mm->pgd :
68 (unsigned long) tsk->active_mm->pgd));
69 die_if_kernel("Oops", regs);
72 static void __kprobes bad_kernel_pc(struct pt_regs *regs, unsigned long vaddr)
74 printk(KERN_CRIT "OOPS: Bogus kernel PC [%016lx] in fault handler\n",
76 printk(KERN_CRIT "OOPS: RPC [%016lx]\n", regs->u_regs[15]);
77 printk("OOPS: RPC <%pS>\n", (void *) regs->u_regs[15]);
78 printk(KERN_CRIT "OOPS: Fault was to vaddr[%lx]\n", vaddr);
80 unhandled_fault(regs->tpc, current, regs);
84 * We now make sure that mmap_sem is held in all paths that call
85 * this. Additionally, to prevent kswapd from ripping ptes from
86 * under us, raise interrupts around the time that we look at the
87 * pte, kswapd will have to wait to get his smp ipi response from
88 * us. vmtruncate likewise. This saves us having to get pte lock.
90 static unsigned int get_user_insn(unsigned long tpc)
92 pgd_t *pgdp = pgd_offset(current->mm, tpc);
102 pudp = pud_offset(pgdp, tpc);
105 pmdp = pmd_offset(pudp, tpc);
109 /* This disables preemption for us as well. */
110 __asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate));
111 __asm__ __volatile__("wrpr %0, %1, %%pstate"
112 : : "r" (pstate), "i" (PSTATE_IE));
113 ptep = pte_offset_map(pmdp, tpc);
115 if (!pte_present(pte))
118 pa = (pte_pfn(pte) << PAGE_SHIFT);
119 pa += (tpc & ~PAGE_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 __asm__ __volatile__("wrpr %0, 0x0, %%pstate" : : "r" (pstate));
134 show_signal_msg(struct pt_regs *regs, int sig, int code,
135 unsigned long address, struct task_struct *tsk)
137 if (!unhandled_signal(tsk, sig))
140 if (!printk_ratelimit())
143 printk("%s%s[%d]: segfault at %lx ip %p (rpc %p) sp %p error %x",
144 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
145 tsk->comm, task_pid_nr(tsk), address,
146 (void *)regs->tpc, (void *)regs->u_regs[UREG_I7],
147 (void *)regs->u_regs[UREG_FP], code);
149 print_vma_addr(KERN_CONT " in ", regs->tpc);
151 printk(KERN_CONT "\n");
154 extern unsigned long compute_effective_address(struct pt_regs *, unsigned int, unsigned int);
156 static void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
157 unsigned int insn, int fault_code)
165 if (fault_code & FAULT_CODE_ITLB)
168 addr = compute_effective_address(regs, insn, 0);
169 info.si_addr = (void __user *) addr;
172 if (unlikely(show_unhandled_signals))
173 show_signal_msg(regs, sig, code, addr, current);
175 force_sig_info(sig, &info, current);
178 extern int handle_ldf_stq(u32, struct pt_regs *);
179 extern int handle_ld_nf(u32, struct pt_regs *);
181 static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
184 if (!regs->tpc || (regs->tpc & 0x3))
186 if (regs->tstate & TSTATE_PRIV) {
187 insn = *(unsigned int *) regs->tpc;
189 insn = get_user_insn(regs->tpc);
195 static void __kprobes do_kernel_fault(struct pt_regs *regs, int si_code,
196 int fault_code, unsigned int insn,
197 unsigned long address)
199 unsigned char asi = ASI_P;
201 if ((!insn) && (regs->tstate & TSTATE_PRIV))
204 /* If user insn could be read (thus insn is zero), that
205 * is fine. We will just gun down the process with a signal
209 if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
210 (insn & 0xc0800000) == 0xc0800000) {
212 asi = (regs->tstate >> 24);
215 if ((asi & 0xf2) == 0x82) {
216 if (insn & 0x1000000) {
217 handle_ldf_stq(insn, regs);
219 /* This was a non-faulting load. Just clear the
220 * destination register(s) and continue with the next
223 handle_ld_nf(insn, regs);
229 /* Is this in ex_table? */
230 if (regs->tstate & TSTATE_PRIV) {
231 const struct exception_table_entry *entry;
233 entry = search_exception_tables(regs->tpc);
235 regs->tpc = entry->fixup;
236 regs->tnpc = regs->tpc + 4;
240 /* The si_code was set to make clear whether
241 * this was a SEGV_MAPERR or SEGV_ACCERR fault.
243 do_fault_siginfo(si_code, SIGSEGV, regs, insn, fault_code);
248 unhandled_fault (address, current, regs);
251 static void noinline __kprobes bogus_32bit_fault_tpc(struct pt_regs *regs)
256 printk(KERN_ERR "FAULT[%s:%d]: 32-bit process reports "
257 "64-bit TPC [%lx]\n",
258 current->comm, current->pid,
263 static void noinline __kprobes bogus_32bit_fault_address(struct pt_regs *regs,
269 printk(KERN_ERR "FAULT[%s:%d]: 32-bit process "
270 "reports 64-bit fault address [%lx]\n",
271 current->comm, current->pid, addr);
275 asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
277 struct mm_struct *mm = current->mm;
278 struct vm_area_struct *vma;
279 unsigned int insn = 0;
280 int si_code, fault_code, fault;
281 unsigned long address, mm_rss;
282 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
284 fault_code = get_thread_fault_code();
286 if (notify_page_fault(regs))
289 si_code = SEGV_MAPERR;
290 address = current_thread_info()->fault_address;
292 if ((fault_code & FAULT_CODE_ITLB) &&
293 (fault_code & FAULT_CODE_DTLB))
296 if (test_thread_flag(TIF_32BIT)) {
297 if (!(regs->tstate & TSTATE_PRIV)) {
298 if (unlikely((regs->tpc >> 32) != 0)) {
299 bogus_32bit_fault_tpc(regs);
303 if (unlikely((address >> 32) != 0)) {
304 bogus_32bit_fault_address(regs, address);
309 if (regs->tstate & TSTATE_PRIV) {
310 unsigned long tpc = regs->tpc;
312 /* Sanity check the PC. */
313 if ((tpc >= KERNBASE && tpc < (unsigned long) __init_end) ||
314 (tpc >= MODULES_VADDR && tpc < MODULES_END)) {
315 /* Valid, no problems... */
317 bad_kernel_pc(regs, address);
323 * If we're in an interrupt or have no user
324 * context, we must not take the fault..
326 if (in_atomic() || !mm)
329 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
331 if (!down_read_trylock(&mm->mmap_sem)) {
332 if ((regs->tstate & TSTATE_PRIV) &&
333 !search_exception_tables(regs->tpc)) {
334 insn = get_fault_insn(regs, insn);
335 goto handle_kernel_fault;
339 down_read(&mm->mmap_sem);
342 vma = find_vma(mm, address);
346 /* Pure DTLB misses do not tell us whether the fault causing
347 * load/store/atomic was a write or not, it only says that there
348 * was no match. So in such a case we (carefully) read the
349 * instruction to try and figure this out. It's an optimization
350 * so it's ok if we can't do this.
352 * Special hack, window spill/fill knows the exact fault type.
355 (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) &&
356 (vma->vm_flags & VM_WRITE) != 0) {
357 insn = get_fault_insn(regs, 0);
360 /* All loads, stores and atomics have bits 30 and 31 both set
361 * in the instruction. Bit 21 is set in all stores, but we
362 * have to avoid prefetches which also have bit 21 set.
364 if ((insn & 0xc0200000) == 0xc0200000 &&
365 (insn & 0x01780000) != 0x01680000) {
366 /* Don't bother updating thread struct value,
367 * because update_mmu_cache only cares which tlb
368 * the access came from.
370 fault_code |= FAULT_CODE_WRITE;
375 if (vma->vm_start <= address)
377 if (!(vma->vm_flags & VM_GROWSDOWN))
379 if (!(fault_code & FAULT_CODE_WRITE)) {
380 /* Non-faulting loads shouldn't expand stack. */
381 insn = get_fault_insn(regs, insn);
382 if ((insn & 0xc0800000) == 0xc0800000) {
386 asi = (regs->tstate >> 24);
389 if ((asi & 0xf2) == 0x82)
393 if (expand_stack(vma, address))
396 * Ok, we have a good vm_area for this memory access, so
400 si_code = SEGV_ACCERR;
402 /* If we took a ITLB miss on a non-executable page, catch
405 if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) {
406 BUG_ON(address != regs->tpc);
407 BUG_ON(regs->tstate & TSTATE_PRIV);
411 if (fault_code & FAULT_CODE_WRITE) {
412 if (!(vma->vm_flags & VM_WRITE))
415 /* Spitfire has an icache which does not snoop
416 * processor stores. Later processors do...
418 if (tlb_type == spitfire &&
419 (vma->vm_flags & VM_EXEC) != 0 &&
420 vma->vm_file != NULL)
421 set_thread_fault_code(fault_code |
422 FAULT_CODE_BLKCOMMIT);
424 /* Allow reads even for write-only mappings */
425 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
429 flags |= ((fault_code & FAULT_CODE_WRITE) ? FAULT_FLAG_WRITE : 0);
430 fault = handle_mm_fault(mm, vma, address, flags);
432 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
435 if (unlikely(fault & VM_FAULT_ERROR)) {
436 if (fault & VM_FAULT_OOM)
438 else if (fault & VM_FAULT_SIGBUS)
443 if (flags & FAULT_FLAG_ALLOW_RETRY) {
444 if (fault & VM_FAULT_MAJOR) {
446 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ,
450 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN,
453 if (fault & VM_FAULT_RETRY) {
454 flags &= ~FAULT_FLAG_ALLOW_RETRY;
455 flags |= FAULT_FLAG_TRIED;
457 /* No need to up_read(&mm->mmap_sem) as we would
458 * have already released it in __lock_page_or_retry
465 up_read(&mm->mmap_sem);
467 mm_rss = get_mm_rss(mm);
468 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
469 mm_rss -= (mm->context.huge_pte_count * (HPAGE_SIZE / PAGE_SIZE));
471 if (unlikely(mm_rss >
472 mm->context.tsb_block[MM_TSB_BASE].tsb_rss_limit))
473 tsb_grow(mm, MM_TSB_BASE, mm_rss);
474 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
475 mm_rss = mm->context.huge_pte_count;
476 if (unlikely(mm_rss >
477 mm->context.tsb_block[MM_TSB_HUGE].tsb_rss_limit))
478 tsb_grow(mm, MM_TSB_HUGE, mm_rss);
483 * Something tried to access memory that isn't in our memory map..
484 * Fix it, but check if it's kernel or user first..
487 insn = get_fault_insn(regs, insn);
488 up_read(&mm->mmap_sem);
491 do_kernel_fault(regs, si_code, fault_code, insn, address);
495 * We ran out of memory, or some other thing happened to us that made
496 * us unable to handle the page fault gracefully.
499 insn = get_fault_insn(regs, insn);
500 up_read(&mm->mmap_sem);
501 if (!(regs->tstate & TSTATE_PRIV)) {
502 pagefault_out_of_memory();
505 goto handle_kernel_fault;
508 insn = get_fault_insn(regs, 0);
509 goto handle_kernel_fault;
512 insn = get_fault_insn(regs, insn);
513 up_read(&mm->mmap_sem);
516 * Send a sigbus, regardless of whether we were in kernel
519 do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, insn, fault_code);
521 /* Kernel mode? Handle exceptions or die */
522 if (regs->tstate & TSTATE_PRIV)
523 goto handle_kernel_fault;