Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* $Id: fault.c,v 1.59 2002/02/09 19:49:31 davem Exp $ |
2 | * arch/sparc64/mm/fault.c: Page fault handlers for the 64-bit Sparc. | |
3 | * | |
4 | * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu) | |
5 | * Copyright (C) 1997, 1999 Jakub Jelinek (jj@ultra.linux.cz) | |
6 | */ | |
7 | ||
8 | #include <asm/head.h> | |
9 | ||
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/smp_lock.h> | |
19 | #include <linux/init.h> | |
20 | #include <linux/interrupt.h> | |
05e14cb3 | 21 | #include <linux/kprobes.h> |
1da177e4 LT |
22 | |
23 | #include <asm/page.h> | |
24 | #include <asm/pgtable.h> | |
25 | #include <asm/openprom.h> | |
26 | #include <asm/oplib.h> | |
27 | #include <asm/uaccess.h> | |
28 | #include <asm/asi.h> | |
29 | #include <asm/lsu.h> | |
30 | #include <asm/sections.h> | |
31 | #include <asm/kdebug.h> | |
32 | ||
33 | #define ELEMENTS(arr) (sizeof (arr)/sizeof (arr[0])) | |
34 | ||
1da177e4 LT |
35 | /* |
36 | * To debug kernel to catch accesses to certain virtual/physical addresses. | |
37 | * Mode = 0 selects physical watchpoints, mode = 1 selects virtual watchpoints. | |
38 | * flags = VM_READ watches memread accesses, flags = VM_WRITE watches memwrite accesses. | |
39 | * Caller passes in a 64bit aligned addr, with mask set to the bytes that need to be | |
40 | * watched. This is only useful on a single cpu machine for now. After the watchpoint | |
41 | * is detected, the process causing it will be killed, thus preventing an infinite loop. | |
42 | */ | |
43 | void set_brkpt(unsigned long addr, unsigned char mask, int flags, int mode) | |
44 | { | |
45 | unsigned long lsubits; | |
46 | ||
47 | __asm__ __volatile__("ldxa [%%g0] %1, %0" | |
48 | : "=r" (lsubits) | |
49 | : "i" (ASI_LSU_CONTROL)); | |
50 | lsubits &= ~(LSU_CONTROL_PM | LSU_CONTROL_VM | | |
51 | LSU_CONTROL_PR | LSU_CONTROL_VR | | |
52 | LSU_CONTROL_PW | LSU_CONTROL_VW); | |
53 | ||
54 | __asm__ __volatile__("stxa %0, [%1] %2\n\t" | |
55 | "membar #Sync" | |
56 | : /* no outputs */ | |
57 | : "r" (addr), "r" (mode ? VIRT_WATCHPOINT : PHYS_WATCHPOINT), | |
58 | "i" (ASI_DMMU)); | |
59 | ||
60 | lsubits |= ((unsigned long)mask << (mode ? 25 : 33)); | |
61 | if (flags & VM_READ) | |
62 | lsubits |= (mode ? LSU_CONTROL_VR : LSU_CONTROL_PR); | |
63 | if (flags & VM_WRITE) | |
64 | lsubits |= (mode ? LSU_CONTROL_VW : LSU_CONTROL_PW); | |
65 | __asm__ __volatile__("stxa %0, [%%g0] %1\n\t" | |
66 | "membar #Sync" | |
67 | : /* no outputs */ | |
68 | : "r" (lsubits), "i" (ASI_LSU_CONTROL) | |
69 | : "memory"); | |
70 | } | |
71 | ||
05e14cb3 PP |
72 | static void __kprobes unhandled_fault(unsigned long address, |
73 | struct task_struct *tsk, | |
74 | struct pt_regs *regs) | |
1da177e4 LT |
75 | { |
76 | if ((unsigned long) address < PAGE_SIZE) { | |
77 | printk(KERN_ALERT "Unable to handle kernel NULL " | |
78 | "pointer dereference\n"); | |
79 | } else { | |
80 | printk(KERN_ALERT "Unable to handle kernel paging request " | |
81 | "at virtual address %016lx\n", (unsigned long)address); | |
82 | } | |
83 | printk(KERN_ALERT "tsk->{mm,active_mm}->context = %016lx\n", | |
84 | (tsk->mm ? | |
85 | CTX_HWBITS(tsk->mm->context) : | |
86 | CTX_HWBITS(tsk->active_mm->context))); | |
87 | printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %016lx\n", | |
88 | (tsk->mm ? (unsigned long) tsk->mm->pgd : | |
89 | (unsigned long) tsk->active_mm->pgd)); | |
90 | if (notify_die(DIE_GPF, "general protection fault", regs, | |
91 | 0, 0, SIGSEGV) == NOTIFY_STOP) | |
92 | return; | |
93 | die_if_kernel("Oops", regs); | |
94 | } | |
95 | ||
96 | static void bad_kernel_pc(struct pt_regs *regs) | |
97 | { | |
98 | unsigned long *ksp; | |
99 | ||
100 | printk(KERN_CRIT "OOPS: Bogus kernel PC [%016lx] in fault handler\n", | |
101 | regs->tpc); | |
102 | __asm__("mov %%sp, %0" : "=r" (ksp)); | |
103 | show_stack(current, ksp); | |
104 | unhandled_fault(regs->tpc, current, regs); | |
105 | } | |
106 | ||
107 | /* | |
108 | * We now make sure that mmap_sem is held in all paths that call | |
109 | * this. Additionally, to prevent kswapd from ripping ptes from | |
110 | * under us, raise interrupts around the time that we look at the | |
111 | * pte, kswapd will have to wait to get his smp ipi response from | |
112 | * us. This saves us having to get page_table_lock. | |
113 | */ | |
114 | static unsigned int get_user_insn(unsigned long tpc) | |
115 | { | |
116 | pgd_t *pgdp = pgd_offset(current->mm, tpc); | |
117 | pud_t *pudp; | |
118 | pmd_t *pmdp; | |
119 | pte_t *ptep, pte; | |
120 | unsigned long pa; | |
121 | u32 insn = 0; | |
122 | unsigned long pstate; | |
123 | ||
124 | if (pgd_none(*pgdp)) | |
125 | goto outret; | |
126 | pudp = pud_offset(pgdp, tpc); | |
127 | if (pud_none(*pudp)) | |
128 | goto outret; | |
129 | pmdp = pmd_offset(pudp, tpc); | |
130 | if (pmd_none(*pmdp)) | |
131 | goto outret; | |
132 | ||
133 | /* This disables preemption for us as well. */ | |
134 | __asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate)); | |
135 | __asm__ __volatile__("wrpr %0, %1, %%pstate" | |
136 | : : "r" (pstate), "i" (PSTATE_IE)); | |
137 | ptep = pte_offset_map(pmdp, tpc); | |
138 | pte = *ptep; | |
139 | if (!pte_present(pte)) | |
140 | goto out; | |
141 | ||
142 | pa = (pte_val(pte) & _PAGE_PADDR); | |
143 | pa += (tpc & ~PAGE_MASK); | |
144 | ||
145 | /* Use phys bypass so we don't pollute dtlb/dcache. */ | |
146 | __asm__ __volatile__("lduwa [%1] %2, %0" | |
147 | : "=r" (insn) | |
148 | : "r" (pa), "i" (ASI_PHYS_USE_EC)); | |
149 | ||
150 | out: | |
151 | pte_unmap(ptep); | |
152 | __asm__ __volatile__("wrpr %0, 0x0, %%pstate" : : "r" (pstate)); | |
153 | outret: | |
154 | return insn; | |
155 | } | |
156 | ||
157 | extern unsigned long compute_effective_address(struct pt_regs *, unsigned int, unsigned int); | |
158 | ||
159 | static void do_fault_siginfo(int code, int sig, struct pt_regs *regs, | |
160 | unsigned int insn, int fault_code) | |
161 | { | |
162 | siginfo_t info; | |
163 | ||
164 | info.si_code = code; | |
165 | info.si_signo = sig; | |
166 | info.si_errno = 0; | |
167 | if (fault_code & FAULT_CODE_ITLB) | |
168 | info.si_addr = (void __user *) regs->tpc; | |
169 | else | |
170 | info.si_addr = (void __user *) | |
171 | compute_effective_address(regs, insn, 0); | |
172 | info.si_trapno = 0; | |
173 | force_sig_info(sig, &info, current); | |
174 | } | |
175 | ||
176 | extern int handle_ldf_stq(u32, struct pt_regs *); | |
177 | extern int handle_ld_nf(u32, struct pt_regs *); | |
178 | ||
179 | static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn) | |
180 | { | |
181 | if (!insn) { | |
182 | if (!regs->tpc || (regs->tpc & 0x3)) | |
183 | return 0; | |
184 | if (regs->tstate & TSTATE_PRIV) { | |
185 | insn = *(unsigned int *) regs->tpc; | |
186 | } else { | |
187 | insn = get_user_insn(regs->tpc); | |
188 | } | |
189 | } | |
190 | return insn; | |
191 | } | |
192 | ||
193 | static void do_kernel_fault(struct pt_regs *regs, int si_code, int fault_code, | |
194 | unsigned int insn, unsigned long address) | |
195 | { | |
1da177e4 LT |
196 | unsigned char asi = ASI_P; |
197 | ||
198 | if ((!insn) && (regs->tstate & TSTATE_PRIV)) | |
199 | goto cannot_handle; | |
200 | ||
201 | /* If user insn could be read (thus insn is zero), that | |
202 | * is fine. We will just gun down the process with a signal | |
203 | * in that case. | |
204 | */ | |
205 | ||
206 | if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) && | |
207 | (insn & 0xc0800000) == 0xc0800000) { | |
208 | if (insn & 0x2000) | |
209 | asi = (regs->tstate >> 24); | |
210 | else | |
211 | asi = (insn >> 5); | |
212 | if ((asi & 0xf2) == 0x82) { | |
213 | if (insn & 0x1000000) { | |
214 | handle_ldf_stq(insn, regs); | |
215 | } else { | |
216 | /* This was a non-faulting load. Just clear the | |
217 | * destination register(s) and continue with the next | |
218 | * instruction. -jj | |
219 | */ | |
220 | handle_ld_nf(insn, regs); | |
221 | } | |
222 | return; | |
223 | } | |
224 | } | |
225 | ||
1da177e4 LT |
226 | /* Is this in ex_table? */ |
227 | if (regs->tstate & TSTATE_PRIV) { | |
8cf14af0 | 228 | const struct exception_table_entry *entry; |
1da177e4 LT |
229 | |
230 | if (asi == ASI_P && (insn & 0xc0800000) == 0xc0800000) { | |
231 | if (insn & 0x2000) | |
232 | asi = (regs->tstate >> 24); | |
233 | else | |
234 | asi = (insn >> 5); | |
235 | } | |
236 | ||
237 | /* Look in asi.h: All _S asis have LS bit set */ | |
238 | if ((asi & 0x1) && | |
8cf14af0 DM |
239 | (entry = search_exception_tables(regs->tpc))) { |
240 | regs->tpc = entry->fixup; | |
1da177e4 | 241 | regs->tnpc = regs->tpc + 4; |
1da177e4 LT |
242 | return; |
243 | } | |
244 | } else { | |
245 | /* The si_code was set to make clear whether | |
246 | * this was a SEGV_MAPERR or SEGV_ACCERR fault. | |
247 | */ | |
248 | do_fault_siginfo(si_code, SIGSEGV, regs, insn, fault_code); | |
249 | return; | |
250 | } | |
251 | ||
252 | cannot_handle: | |
253 | unhandled_fault (address, current, regs); | |
254 | } | |
255 | ||
05e14cb3 | 256 | asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs) |
1da177e4 LT |
257 | { |
258 | struct mm_struct *mm = current->mm; | |
259 | struct vm_area_struct *vma; | |
260 | unsigned int insn = 0; | |
261 | int si_code, fault_code; | |
262 | unsigned long address; | |
263 | ||
264 | fault_code = get_thread_fault_code(); | |
265 | ||
266 | if (notify_die(DIE_PAGE_FAULT, "page_fault", regs, | |
267 | fault_code, 0, SIGSEGV) == NOTIFY_STOP) | |
268 | return; | |
269 | ||
270 | si_code = SEGV_MAPERR; | |
271 | address = current_thread_info()->fault_address; | |
272 | ||
273 | if ((fault_code & FAULT_CODE_ITLB) && | |
274 | (fault_code & FAULT_CODE_DTLB)) | |
275 | BUG(); | |
276 | ||
277 | if (regs->tstate & TSTATE_PRIV) { | |
278 | unsigned long tpc = regs->tpc; | |
279 | ||
280 | /* Sanity check the PC. */ | |
281 | if ((tpc >= KERNBASE && tpc < (unsigned long) _etext) || | |
282 | (tpc >= MODULES_VADDR && tpc < MODULES_END)) { | |
283 | /* Valid, no problems... */ | |
284 | } else { | |
285 | bad_kernel_pc(regs); | |
286 | return; | |
287 | } | |
288 | } | |
289 | ||
290 | /* | |
291 | * If we're in an interrupt or have no user | |
292 | * context, we must not take the fault.. | |
293 | */ | |
294 | if (in_atomic() || !mm) | |
295 | goto intr_or_no_mm; | |
296 | ||
297 | if (test_thread_flag(TIF_32BIT)) { | |
298 | if (!(regs->tstate & TSTATE_PRIV)) | |
299 | regs->tpc &= 0xffffffff; | |
300 | address &= 0xffffffff; | |
301 | } | |
302 | ||
303 | if (!down_read_trylock(&mm->mmap_sem)) { | |
304 | if ((regs->tstate & TSTATE_PRIV) && | |
305 | !search_exception_tables(regs->tpc)) { | |
306 | insn = get_fault_insn(regs, insn); | |
307 | goto handle_kernel_fault; | |
308 | } | |
309 | down_read(&mm->mmap_sem); | |
310 | } | |
311 | ||
312 | vma = find_vma(mm, address); | |
313 | if (!vma) | |
314 | goto bad_area; | |
315 | ||
316 | /* Pure DTLB misses do not tell us whether the fault causing | |
317 | * load/store/atomic was a write or not, it only says that there | |
318 | * was no match. So in such a case we (carefully) read the | |
319 | * instruction to try and figure this out. It's an optimization | |
320 | * so it's ok if we can't do this. | |
321 | * | |
322 | * Special hack, window spill/fill knows the exact fault type. | |
323 | */ | |
324 | if (((fault_code & | |
325 | (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) && | |
326 | (vma->vm_flags & VM_WRITE) != 0) { | |
327 | insn = get_fault_insn(regs, 0); | |
328 | if (!insn) | |
329 | goto continue_fault; | |
330 | if ((insn & 0xc0200000) == 0xc0200000 && | |
331 | (insn & 0x1780000) != 0x1680000) { | |
332 | /* Don't bother updating thread struct value, | |
333 | * because update_mmu_cache only cares which tlb | |
334 | * the access came from. | |
335 | */ | |
336 | fault_code |= FAULT_CODE_WRITE; | |
337 | } | |
338 | } | |
339 | continue_fault: | |
340 | ||
341 | if (vma->vm_start <= address) | |
342 | goto good_area; | |
343 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
344 | goto bad_area; | |
345 | if (!(fault_code & FAULT_CODE_WRITE)) { | |
346 | /* Non-faulting loads shouldn't expand stack. */ | |
347 | insn = get_fault_insn(regs, insn); | |
348 | if ((insn & 0xc0800000) == 0xc0800000) { | |
349 | unsigned char asi; | |
350 | ||
351 | if (insn & 0x2000) | |
352 | asi = (regs->tstate >> 24); | |
353 | else | |
354 | asi = (insn >> 5); | |
355 | if ((asi & 0xf2) == 0x82) | |
356 | goto bad_area; | |
357 | } | |
358 | } | |
359 | if (expand_stack(vma, address)) | |
360 | goto bad_area; | |
361 | /* | |
362 | * Ok, we have a good vm_area for this memory access, so | |
363 | * we can handle it.. | |
364 | */ | |
365 | good_area: | |
366 | si_code = SEGV_ACCERR; | |
367 | ||
368 | /* If we took a ITLB miss on a non-executable page, catch | |
369 | * that here. | |
370 | */ | |
371 | if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) { | |
372 | BUG_ON(address != regs->tpc); | |
373 | BUG_ON(regs->tstate & TSTATE_PRIV); | |
374 | goto bad_area; | |
375 | } | |
376 | ||
377 | if (fault_code & FAULT_CODE_WRITE) { | |
378 | if (!(vma->vm_flags & VM_WRITE)) | |
379 | goto bad_area; | |
380 | ||
381 | /* Spitfire has an icache which does not snoop | |
382 | * processor stores. Later processors do... | |
383 | */ | |
384 | if (tlb_type == spitfire && | |
385 | (vma->vm_flags & VM_EXEC) != 0 && | |
386 | vma->vm_file != NULL) | |
387 | set_thread_fault_code(fault_code | | |
388 | FAULT_CODE_BLKCOMMIT); | |
389 | } else { | |
390 | /* Allow reads even for write-only mappings */ | |
391 | if (!(vma->vm_flags & (VM_READ | VM_EXEC))) | |
392 | goto bad_area; | |
393 | } | |
394 | ||
395 | switch (handle_mm_fault(mm, vma, address, (fault_code & FAULT_CODE_WRITE))) { | |
396 | case VM_FAULT_MINOR: | |
397 | current->min_flt++; | |
398 | break; | |
399 | case VM_FAULT_MAJOR: | |
400 | current->maj_flt++; | |
401 | break; | |
402 | case VM_FAULT_SIGBUS: | |
403 | goto do_sigbus; | |
404 | case VM_FAULT_OOM: | |
405 | goto out_of_memory; | |
406 | default: | |
407 | BUG(); | |
408 | } | |
409 | ||
410 | up_read(&mm->mmap_sem); | |
efdc1e20 | 411 | return; |
1da177e4 LT |
412 | |
413 | /* | |
414 | * Something tried to access memory that isn't in our memory map.. | |
415 | * Fix it, but check if it's kernel or user first.. | |
416 | */ | |
417 | bad_area: | |
418 | insn = get_fault_insn(regs, insn); | |
419 | up_read(&mm->mmap_sem); | |
420 | ||
421 | handle_kernel_fault: | |
422 | do_kernel_fault(regs, si_code, fault_code, insn, address); | |
efdc1e20 | 423 | return; |
1da177e4 LT |
424 | |
425 | /* | |
426 | * We ran out of memory, or some other thing happened to us that made | |
427 | * us unable to handle the page fault gracefully. | |
428 | */ | |
429 | out_of_memory: | |
430 | insn = get_fault_insn(regs, insn); | |
431 | up_read(&mm->mmap_sem); | |
432 | printk("VM: killing process %s\n", current->comm); | |
433 | if (!(regs->tstate & TSTATE_PRIV)) | |
434 | do_exit(SIGKILL); | |
435 | goto handle_kernel_fault; | |
436 | ||
437 | intr_or_no_mm: | |
438 | insn = get_fault_insn(regs, 0); | |
439 | goto handle_kernel_fault; | |
440 | ||
441 | do_sigbus: | |
442 | insn = get_fault_insn(regs, insn); | |
443 | up_read(&mm->mmap_sem); | |
444 | ||
445 | /* | |
446 | * Send a sigbus, regardless of whether we were in kernel | |
447 | * or user mode. | |
448 | */ | |
449 | do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, insn, fault_code); | |
450 | ||
451 | /* Kernel mode? Handle exceptions or die */ | |
452 | if (regs->tstate & TSTATE_PRIV) | |
453 | goto handle_kernel_fault; | |
1da177e4 | 454 | } |