x86/efi: Enforce CONFIG_RELOCATABLE for EFI boot stub
[linux/fpc-iii.git] / arch / sparc / mm / fault_64.c
blob3841a081beb3967d9f2f08d0c6d1b4abf5390ff0
1 /*
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)
6 */
8 #include <asm/head.h>
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>
16 #include <linux/mm.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>
25 #include <asm/page.h>
26 #include <asm/pgtable.h>
27 #include <asm/openprom.h>
28 #include <asm/oplib.h>
29 #include <asm/uaccess.h>
30 #include <asm/asi.h>
31 #include <asm/lsu.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)
39 int ret = 0;
41 /* kprobe_running() needs smp_processor_id() */
42 if (kprobes_built_in() && !user_mode(regs)) {
43 preempt_disable();
44 if (kprobe_running() && kprobe_fault_handler(regs, 0))
45 ret = 1;
46 preempt_enable();
48 return ret;
51 static void __kprobes unhandled_fault(unsigned long address,
52 struct task_struct *tsk,
53 struct pt_regs *regs)
55 if ((unsigned long) address < PAGE_SIZE) {
56 printk(KERN_ALERT "Unable to handle kernel NULL "
57 "pointer dereference\n");
58 } else {
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",
63 (tsk->mm ?
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",
75 regs->tpc);
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);
79 dump_stack();
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);
93 pud_t *pudp;
94 pmd_t *pmdp;
95 pte_t *ptep, pte;
96 unsigned long pa;
97 u32 insn = 0;
99 if (pgd_none(*pgdp) || unlikely(pgd_bad(*pgdp)))
100 goto out;
101 pudp = pud_offset(pgdp, tpc);
102 if (pud_none(*pudp) || unlikely(pud_bad(*pudp)))
103 goto out;
105 /* This disables preemption for us as well. */
106 local_irq_disable();
108 pmdp = pmd_offset(pudp, tpc);
109 if (pmd_none(*pmdp) || unlikely(pmd_bad(*pmdp)))
110 goto out_irq_enable;
112 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
113 if (pmd_trans_huge(*pmdp)) {
114 if (pmd_trans_splitting(*pmdp))
115 goto out_irq_enable;
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"
122 : "=r" (insn)
123 : "r" (pa), "i" (ASI_PHYS_USE_EC));
124 } else
125 #endif
127 ptep = pte_offset_map(pmdp, tpc);
128 pte = *ptep;
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"
135 : "=r" (insn)
136 : "r" (pa), "i" (ASI_PHYS_USE_EC));
138 pte_unmap(ptep);
140 out_irq_enable:
141 local_irq_enable();
142 out:
143 return insn;
146 static inline void
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))
151 return;
153 if (!printk_ratelimit())
154 return;
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,
169 int fault_code)
171 unsigned long addr;
172 siginfo_t info;
174 info.si_code = code;
175 info.si_signo = sig;
176 info.si_errno = 0;
177 if (fault_code & FAULT_CODE_ITLB) {
178 addr = regs->tpc;
179 } else {
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.
184 if (insn)
185 addr = compute_effective_address(regs, insn, 0);
186 else
187 addr = fault_addr;
189 info.si_addr = (void __user *) addr;
190 info.si_trapno = 0;
192 if (unlikely(show_unhandled_signals))
193 show_signal_msg(regs, sig, code, addr, current);
195 force_sig_info(sig, &info, current);
198 extern int handle_ldf_stq(u32, struct pt_regs *);
199 extern int handle_ld_nf(u32, struct pt_regs *);
201 static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
203 if (!insn) {
204 if (!regs->tpc || (regs->tpc & 0x3))
205 return 0;
206 if (regs->tstate & TSTATE_PRIV) {
207 insn = *(unsigned int *) regs->tpc;
208 } else {
209 insn = get_user_insn(regs->tpc);
212 return insn;
215 static void __kprobes do_kernel_fault(struct pt_regs *regs, int si_code,
216 int fault_code, unsigned int insn,
217 unsigned long address)
219 unsigned char asi = ASI_P;
221 if ((!insn) && (regs->tstate & TSTATE_PRIV))
222 goto cannot_handle;
224 /* If user insn could be read (thus insn is zero), that
225 * is fine. We will just gun down the process with a signal
226 * in that case.
229 if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
230 (insn & 0xc0800000) == 0xc0800000) {
231 if (insn & 0x2000)
232 asi = (regs->tstate >> 24);
233 else
234 asi = (insn >> 5);
235 if ((asi & 0xf2) == 0x82) {
236 if (insn & 0x1000000) {
237 handle_ldf_stq(insn, regs);
238 } else {
239 /* This was a non-faulting load. Just clear the
240 * destination register(s) and continue with the next
241 * instruction. -jj
243 handle_ld_nf(insn, regs);
245 return;
249 /* Is this in ex_table? */
250 if (regs->tstate & TSTATE_PRIV) {
251 const struct exception_table_entry *entry;
253 entry = search_exception_tables(regs->tpc);
254 if (entry) {
255 regs->tpc = entry->fixup;
256 regs->tnpc = regs->tpc + 4;
257 return;
259 } else {
260 /* The si_code was set to make clear whether
261 * this was a SEGV_MAPERR or SEGV_ACCERR fault.
263 do_fault_siginfo(si_code, SIGSEGV, regs, address, insn, fault_code);
264 return;
267 cannot_handle:
268 unhandled_fault (address, current, regs);
271 static void noinline __kprobes bogus_32bit_fault_tpc(struct pt_regs *regs)
273 static int times;
275 if (times++ < 10)
276 printk(KERN_ERR "FAULT[%s:%d]: 32-bit process reports "
277 "64-bit TPC [%lx]\n",
278 current->comm, current->pid,
279 regs->tpc);
280 show_regs(regs);
283 asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
285 struct mm_struct *mm = current->mm;
286 struct vm_area_struct *vma;
287 unsigned int insn = 0;
288 int si_code, fault_code, fault;
289 unsigned long address, mm_rss;
290 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
292 fault_code = get_thread_fault_code();
294 if (notify_page_fault(regs))
295 return;
297 si_code = SEGV_MAPERR;
298 address = current_thread_info()->fault_address;
300 if ((fault_code & FAULT_CODE_ITLB) &&
301 (fault_code & FAULT_CODE_DTLB))
302 BUG();
304 if (test_thread_flag(TIF_32BIT)) {
305 if (!(regs->tstate & TSTATE_PRIV)) {
306 if (unlikely((regs->tpc >> 32) != 0)) {
307 bogus_32bit_fault_tpc(regs);
308 goto intr_or_no_mm;
311 if (unlikely((address >> 32) != 0))
312 goto intr_or_no_mm;
315 if (regs->tstate & TSTATE_PRIV) {
316 unsigned long tpc = regs->tpc;
318 /* Sanity check the PC. */
319 if ((tpc >= KERNBASE && tpc < (unsigned long) __init_end) ||
320 (tpc >= MODULES_VADDR && tpc < MODULES_END)) {
321 /* Valid, no problems... */
322 } else {
323 bad_kernel_pc(regs, address);
324 return;
326 } else
327 flags |= FAULT_FLAG_USER;
330 * If we're in an interrupt or have no user
331 * context, we must not take the fault..
333 if (in_atomic() || !mm)
334 goto intr_or_no_mm;
336 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
338 if (!down_read_trylock(&mm->mmap_sem)) {
339 if ((regs->tstate & TSTATE_PRIV) &&
340 !search_exception_tables(regs->tpc)) {
341 insn = get_fault_insn(regs, insn);
342 goto handle_kernel_fault;
345 retry:
346 down_read(&mm->mmap_sem);
349 vma = find_vma(mm, address);
350 if (!vma)
351 goto bad_area;
353 /* Pure DTLB misses do not tell us whether the fault causing
354 * load/store/atomic was a write or not, it only says that there
355 * was no match. So in such a case we (carefully) read the
356 * instruction to try and figure this out. It's an optimization
357 * so it's ok if we can't do this.
359 * Special hack, window spill/fill knows the exact fault type.
361 if (((fault_code &
362 (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) &&
363 (vma->vm_flags & VM_WRITE) != 0) {
364 insn = get_fault_insn(regs, 0);
365 if (!insn)
366 goto continue_fault;
367 /* All loads, stores and atomics have bits 30 and 31 both set
368 * in the instruction. Bit 21 is set in all stores, but we
369 * have to avoid prefetches which also have bit 21 set.
371 if ((insn & 0xc0200000) == 0xc0200000 &&
372 (insn & 0x01780000) != 0x01680000) {
373 /* Don't bother updating thread struct value,
374 * because update_mmu_cache only cares which tlb
375 * the access came from.
377 fault_code |= FAULT_CODE_WRITE;
380 continue_fault:
382 if (vma->vm_start <= address)
383 goto good_area;
384 if (!(vma->vm_flags & VM_GROWSDOWN))
385 goto bad_area;
386 if (!(fault_code & FAULT_CODE_WRITE)) {
387 /* Non-faulting loads shouldn't expand stack. */
388 insn = get_fault_insn(regs, insn);
389 if ((insn & 0xc0800000) == 0xc0800000) {
390 unsigned char asi;
392 if (insn & 0x2000)
393 asi = (regs->tstate >> 24);
394 else
395 asi = (insn >> 5);
396 if ((asi & 0xf2) == 0x82)
397 goto bad_area;
400 if (expand_stack(vma, address))
401 goto bad_area;
403 * Ok, we have a good vm_area for this memory access, so
404 * we can handle it..
406 good_area:
407 si_code = SEGV_ACCERR;
409 /* If we took a ITLB miss on a non-executable page, catch
410 * that here.
412 if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) {
413 BUG_ON(address != regs->tpc);
414 BUG_ON(regs->tstate & TSTATE_PRIV);
415 goto bad_area;
418 if (fault_code & FAULT_CODE_WRITE) {
419 if (!(vma->vm_flags & VM_WRITE))
420 goto bad_area;
422 /* Spitfire has an icache which does not snoop
423 * processor stores. Later processors do...
425 if (tlb_type == spitfire &&
426 (vma->vm_flags & VM_EXEC) != 0 &&
427 vma->vm_file != NULL)
428 set_thread_fault_code(fault_code |
429 FAULT_CODE_BLKCOMMIT);
431 flags |= FAULT_FLAG_WRITE;
432 } else {
433 /* Allow reads even for write-only mappings */
434 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
435 goto bad_area;
438 fault = handle_mm_fault(mm, vma, address, flags);
440 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
441 return;
443 if (unlikely(fault & VM_FAULT_ERROR)) {
444 if (fault & VM_FAULT_OOM)
445 goto out_of_memory;
446 else if (fault & VM_FAULT_SIGBUS)
447 goto do_sigbus;
448 BUG();
451 if (flags & FAULT_FLAG_ALLOW_RETRY) {
452 if (fault & VM_FAULT_MAJOR) {
453 current->maj_flt++;
454 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ,
455 1, regs, address);
456 } else {
457 current->min_flt++;
458 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN,
459 1, regs, address);
461 if (fault & VM_FAULT_RETRY) {
462 flags &= ~FAULT_FLAG_ALLOW_RETRY;
463 flags |= FAULT_FLAG_TRIED;
465 /* No need to up_read(&mm->mmap_sem) as we would
466 * have already released it in __lock_page_or_retry
467 * in mm/filemap.c.
470 goto retry;
473 up_read(&mm->mmap_sem);
475 mm_rss = get_mm_rss(mm);
476 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
477 mm_rss -= (mm->context.huge_pte_count * (HPAGE_SIZE / PAGE_SIZE));
478 #endif
479 if (unlikely(mm_rss >
480 mm->context.tsb_block[MM_TSB_BASE].tsb_rss_limit))
481 tsb_grow(mm, MM_TSB_BASE, mm_rss);
482 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
483 mm_rss = mm->context.huge_pte_count;
484 if (unlikely(mm_rss >
485 mm->context.tsb_block[MM_TSB_HUGE].tsb_rss_limit)) {
486 if (mm->context.tsb_block[MM_TSB_HUGE].tsb)
487 tsb_grow(mm, MM_TSB_HUGE, mm_rss);
488 else
489 hugetlb_setup(regs);
492 #endif
493 return;
496 * Something tried to access memory that isn't in our memory map..
497 * Fix it, but check if it's kernel or user first..
499 bad_area:
500 insn = get_fault_insn(regs, insn);
501 up_read(&mm->mmap_sem);
503 handle_kernel_fault:
504 do_kernel_fault(regs, si_code, fault_code, insn, address);
505 return;
508 * We ran out of memory, or some other thing happened to us that made
509 * us unable to handle the page fault gracefully.
511 out_of_memory:
512 insn = get_fault_insn(regs, insn);
513 up_read(&mm->mmap_sem);
514 if (!(regs->tstate & TSTATE_PRIV)) {
515 pagefault_out_of_memory();
516 return;
518 goto handle_kernel_fault;
520 intr_or_no_mm:
521 insn = get_fault_insn(regs, 0);
522 goto handle_kernel_fault;
524 do_sigbus:
525 insn = get_fault_insn(regs, insn);
526 up_read(&mm->mmap_sem);
529 * Send a sigbus, regardless of whether we were in kernel
530 * or user mode.
532 do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, address, insn, fault_code);
534 /* Kernel mode? Handle exceptions or die */
535 if (regs->tstate & TSTATE_PRIV)
536 goto handle_kernel_fault;