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 | ||
1da177e4 LT |
33 | /* |
34 | * To debug kernel to catch accesses to certain virtual/physical addresses. | |
35 | * Mode = 0 selects physical watchpoints, mode = 1 selects virtual watchpoints. | |
36 | * flags = VM_READ watches memread accesses, flags = VM_WRITE watches memwrite accesses. | |
37 | * Caller passes in a 64bit aligned addr, with mask set to the bytes that need to be | |
38 | * watched. This is only useful on a single cpu machine for now. After the watchpoint | |
39 | * is detected, the process causing it will be killed, thus preventing an infinite loop. | |
40 | */ | |
41 | void set_brkpt(unsigned long addr, unsigned char mask, int flags, int mode) | |
42 | { | |
43 | unsigned long lsubits; | |
44 | ||
45 | __asm__ __volatile__("ldxa [%%g0] %1, %0" | |
46 | : "=r" (lsubits) | |
47 | : "i" (ASI_LSU_CONTROL)); | |
48 | lsubits &= ~(LSU_CONTROL_PM | LSU_CONTROL_VM | | |
49 | LSU_CONTROL_PR | LSU_CONTROL_VR | | |
50 | LSU_CONTROL_PW | LSU_CONTROL_VW); | |
51 | ||
52 | __asm__ __volatile__("stxa %0, [%1] %2\n\t" | |
53 | "membar #Sync" | |
54 | : /* no outputs */ | |
55 | : "r" (addr), "r" (mode ? VIRT_WATCHPOINT : PHYS_WATCHPOINT), | |
56 | "i" (ASI_DMMU)); | |
57 | ||
58 | lsubits |= ((unsigned long)mask << (mode ? 25 : 33)); | |
59 | if (flags & VM_READ) | |
60 | lsubits |= (mode ? LSU_CONTROL_VR : LSU_CONTROL_PR); | |
61 | if (flags & VM_WRITE) | |
62 | lsubits |= (mode ? LSU_CONTROL_VW : LSU_CONTROL_PW); | |
63 | __asm__ __volatile__("stxa %0, [%%g0] %1\n\t" | |
64 | "membar #Sync" | |
65 | : /* no outputs */ | |
66 | : "r" (lsubits), "i" (ASI_LSU_CONTROL) | |
67 | : "memory"); | |
68 | } | |
69 | ||
05e14cb3 PP |
70 | static void __kprobes unhandled_fault(unsigned long address, |
71 | struct task_struct *tsk, | |
72 | struct pt_regs *regs) | |
1da177e4 LT |
73 | { |
74 | if ((unsigned long) address < PAGE_SIZE) { | |
75 | printk(KERN_ALERT "Unable to handle kernel NULL " | |
76 | "pointer dereference\n"); | |
77 | } else { | |
78 | printk(KERN_ALERT "Unable to handle kernel paging request " | |
79 | "at virtual address %016lx\n", (unsigned long)address); | |
80 | } | |
81 | printk(KERN_ALERT "tsk->{mm,active_mm}->context = %016lx\n", | |
82 | (tsk->mm ? | |
83 | CTX_HWBITS(tsk->mm->context) : | |
84 | CTX_HWBITS(tsk->active_mm->context))); | |
85 | printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %016lx\n", | |
86 | (tsk->mm ? (unsigned long) tsk->mm->pgd : | |
87 | (unsigned long) tsk->active_mm->pgd)); | |
88 | if (notify_die(DIE_GPF, "general protection fault", regs, | |
89 | 0, 0, SIGSEGV) == NOTIFY_STOP) | |
90 | return; | |
91 | die_if_kernel("Oops", regs); | |
92 | } | |
93 | ||
94 | static void bad_kernel_pc(struct pt_regs *regs) | |
95 | { | |
96 | unsigned long *ksp; | |
97 | ||
98 | printk(KERN_CRIT "OOPS: Bogus kernel PC [%016lx] in fault handler\n", | |
99 | regs->tpc); | |
100 | __asm__("mov %%sp, %0" : "=r" (ksp)); | |
101 | show_stack(current, ksp); | |
102 | unhandled_fault(regs->tpc, current, regs); | |
103 | } | |
104 | ||
105 | /* | |
106 | * We now make sure that mmap_sem is held in all paths that call | |
107 | * this. Additionally, to prevent kswapd from ripping ptes from | |
108 | * under us, raise interrupts around the time that we look at the | |
109 | * pte, kswapd will have to wait to get his smp ipi response from | |
da160546 | 110 | * us. vmtruncate likewise. This saves us having to get pte lock. |
1da177e4 LT |
111 | */ |
112 | static unsigned int get_user_insn(unsigned long tpc) | |
113 | { | |
114 | pgd_t *pgdp = pgd_offset(current->mm, tpc); | |
115 | pud_t *pudp; | |
116 | pmd_t *pmdp; | |
117 | pte_t *ptep, pte; | |
118 | unsigned long pa; | |
119 | u32 insn = 0; | |
120 | unsigned long pstate; | |
121 | ||
122 | if (pgd_none(*pgdp)) | |
123 | goto outret; | |
124 | pudp = pud_offset(pgdp, tpc); | |
125 | if (pud_none(*pudp)) | |
126 | goto outret; | |
127 | pmdp = pmd_offset(pudp, tpc); | |
128 | if (pmd_none(*pmdp)) | |
129 | goto outret; | |
130 | ||
131 | /* This disables preemption for us as well. */ | |
132 | __asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate)); | |
133 | __asm__ __volatile__("wrpr %0, %1, %%pstate" | |
134 | : : "r" (pstate), "i" (PSTATE_IE)); | |
135 | ptep = pte_offset_map(pmdp, tpc); | |
136 | pte = *ptep; | |
137 | if (!pte_present(pte)) | |
138 | goto out; | |
139 | ||
c4bce90e | 140 | pa = (pte_pfn(pte) << PAGE_SHIFT); |
1da177e4 LT |
141 | pa += (tpc & ~PAGE_MASK); |
142 | ||
143 | /* Use phys bypass so we don't pollute dtlb/dcache. */ | |
144 | __asm__ __volatile__("lduwa [%1] %2, %0" | |
145 | : "=r" (insn) | |
146 | : "r" (pa), "i" (ASI_PHYS_USE_EC)); | |
147 | ||
148 | out: | |
149 | pte_unmap(ptep); | |
150 | __asm__ __volatile__("wrpr %0, 0x0, %%pstate" : : "r" (pstate)); | |
151 | outret: | |
152 | return insn; | |
153 | } | |
154 | ||
155 | extern unsigned long compute_effective_address(struct pt_regs *, unsigned int, unsigned int); | |
156 | ||
157 | static void do_fault_siginfo(int code, int sig, struct pt_regs *regs, | |
158 | unsigned int insn, int fault_code) | |
159 | { | |
160 | siginfo_t info; | |
161 | ||
162 | info.si_code = code; | |
163 | info.si_signo = sig; | |
164 | info.si_errno = 0; | |
165 | if (fault_code & FAULT_CODE_ITLB) | |
166 | info.si_addr = (void __user *) regs->tpc; | |
167 | else | |
168 | info.si_addr = (void __user *) | |
169 | compute_effective_address(regs, insn, 0); | |
170 | info.si_trapno = 0; | |
171 | force_sig_info(sig, &info, current); | |
172 | } | |
173 | ||
174 | extern int handle_ldf_stq(u32, struct pt_regs *); | |
175 | extern int handle_ld_nf(u32, struct pt_regs *); | |
176 | ||
177 | static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn) | |
178 | { | |
179 | if (!insn) { | |
180 | if (!regs->tpc || (regs->tpc & 0x3)) | |
181 | return 0; | |
182 | if (regs->tstate & TSTATE_PRIV) { | |
183 | insn = *(unsigned int *) regs->tpc; | |
184 | } else { | |
185 | insn = get_user_insn(regs->tpc); | |
186 | } | |
187 | } | |
188 | return insn; | |
189 | } | |
190 | ||
191 | static void do_kernel_fault(struct pt_regs *regs, int si_code, int fault_code, | |
192 | unsigned int insn, unsigned long address) | |
193 | { | |
1da177e4 LT |
194 | unsigned char asi = ASI_P; |
195 | ||
196 | if ((!insn) && (regs->tstate & TSTATE_PRIV)) | |
197 | goto cannot_handle; | |
198 | ||
199 | /* If user insn could be read (thus insn is zero), that | |
200 | * is fine. We will just gun down the process with a signal | |
201 | * in that case. | |
202 | */ | |
203 | ||
204 | if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) && | |
205 | (insn & 0xc0800000) == 0xc0800000) { | |
206 | if (insn & 0x2000) | |
207 | asi = (regs->tstate >> 24); | |
208 | else | |
209 | asi = (insn >> 5); | |
210 | if ((asi & 0xf2) == 0x82) { | |
211 | if (insn & 0x1000000) { | |
212 | handle_ldf_stq(insn, regs); | |
213 | } else { | |
214 | /* This was a non-faulting load. Just clear the | |
215 | * destination register(s) and continue with the next | |
216 | * instruction. -jj | |
217 | */ | |
218 | handle_ld_nf(insn, regs); | |
219 | } | |
220 | return; | |
221 | } | |
222 | } | |
223 | ||
1da177e4 LT |
224 | /* Is this in ex_table? */ |
225 | if (regs->tstate & TSTATE_PRIV) { | |
8cf14af0 | 226 | const struct exception_table_entry *entry; |
1da177e4 LT |
227 | |
228 | if (asi == ASI_P && (insn & 0xc0800000) == 0xc0800000) { | |
229 | if (insn & 0x2000) | |
230 | asi = (regs->tstate >> 24); | |
231 | else | |
232 | asi = (insn >> 5); | |
233 | } | |
234 | ||
235 | /* Look in asi.h: All _S asis have LS bit set */ | |
236 | if ((asi & 0x1) && | |
8cf14af0 DM |
237 | (entry = search_exception_tables(regs->tpc))) { |
238 | regs->tpc = entry->fixup; | |
1da177e4 | 239 | regs->tnpc = regs->tpc + 4; |
1da177e4 LT |
240 | return; |
241 | } | |
242 | } else { | |
243 | /* The si_code was set to make clear whether | |
244 | * this was a SEGV_MAPERR or SEGV_ACCERR fault. | |
245 | */ | |
246 | do_fault_siginfo(si_code, SIGSEGV, regs, insn, fault_code); | |
247 | return; | |
248 | } | |
249 | ||
250 | cannot_handle: | |
251 | unhandled_fault (address, current, regs); | |
252 | } | |
253 | ||
05e14cb3 | 254 | asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs) |
1da177e4 LT |
255 | { |
256 | struct mm_struct *mm = current->mm; | |
257 | struct vm_area_struct *vma; | |
258 | unsigned int insn = 0; | |
259 | int si_code, fault_code; | |
260 | unsigned long address; | |
261 | ||
262 | fault_code = get_thread_fault_code(); | |
263 | ||
264 | if (notify_die(DIE_PAGE_FAULT, "page_fault", regs, | |
265 | fault_code, 0, SIGSEGV) == NOTIFY_STOP) | |
266 | return; | |
267 | ||
268 | si_code = SEGV_MAPERR; | |
269 | address = current_thread_info()->fault_address; | |
270 | ||
271 | if ((fault_code & FAULT_CODE_ITLB) && | |
272 | (fault_code & FAULT_CODE_DTLB)) | |
273 | BUG(); | |
274 | ||
275 | if (regs->tstate & TSTATE_PRIV) { | |
276 | unsigned long tpc = regs->tpc; | |
277 | ||
278 | /* Sanity check the PC. */ | |
279 | if ((tpc >= KERNBASE && tpc < (unsigned long) _etext) || | |
280 | (tpc >= MODULES_VADDR && tpc < MODULES_END)) { | |
281 | /* Valid, no problems... */ | |
282 | } else { | |
283 | bad_kernel_pc(regs); | |
284 | return; | |
285 | } | |
286 | } | |
287 | ||
288 | /* | |
289 | * If we're in an interrupt or have no user | |
290 | * context, we must not take the fault.. | |
291 | */ | |
292 | if (in_atomic() || !mm) | |
293 | goto intr_or_no_mm; | |
294 | ||
295 | if (test_thread_flag(TIF_32BIT)) { | |
296 | if (!(regs->tstate & TSTATE_PRIV)) | |
297 | regs->tpc &= 0xffffffff; | |
298 | address &= 0xffffffff; | |
299 | } | |
300 | ||
301 | if (!down_read_trylock(&mm->mmap_sem)) { | |
302 | if ((regs->tstate & TSTATE_PRIV) && | |
303 | !search_exception_tables(regs->tpc)) { | |
304 | insn = get_fault_insn(regs, insn); | |
305 | goto handle_kernel_fault; | |
306 | } | |
307 | down_read(&mm->mmap_sem); | |
308 | } | |
309 | ||
310 | vma = find_vma(mm, address); | |
311 | if (!vma) | |
312 | goto bad_area; | |
313 | ||
314 | /* Pure DTLB misses do not tell us whether the fault causing | |
315 | * load/store/atomic was a write or not, it only says that there | |
316 | * was no match. So in such a case we (carefully) read the | |
317 | * instruction to try and figure this out. It's an optimization | |
318 | * so it's ok if we can't do this. | |
319 | * | |
320 | * Special hack, window spill/fill knows the exact fault type. | |
321 | */ | |
322 | if (((fault_code & | |
323 | (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) && | |
324 | (vma->vm_flags & VM_WRITE) != 0) { | |
325 | insn = get_fault_insn(regs, 0); | |
326 | if (!insn) | |
327 | goto continue_fault; | |
328 | if ((insn & 0xc0200000) == 0xc0200000 && | |
329 | (insn & 0x1780000) != 0x1680000) { | |
330 | /* Don't bother updating thread struct value, | |
331 | * because update_mmu_cache only cares which tlb | |
332 | * the access came from. | |
333 | */ | |
334 | fault_code |= FAULT_CODE_WRITE; | |
335 | } | |
336 | } | |
337 | continue_fault: | |
338 | ||
339 | if (vma->vm_start <= address) | |
340 | goto good_area; | |
341 | if (!(vma->vm_flags & VM_GROWSDOWN)) | |
342 | goto bad_area; | |
343 | if (!(fault_code & FAULT_CODE_WRITE)) { | |
344 | /* Non-faulting loads shouldn't expand stack. */ | |
345 | insn = get_fault_insn(regs, insn); | |
346 | if ((insn & 0xc0800000) == 0xc0800000) { | |
347 | unsigned char asi; | |
348 | ||
349 | if (insn & 0x2000) | |
350 | asi = (regs->tstate >> 24); | |
351 | else | |
352 | asi = (insn >> 5); | |
353 | if ((asi & 0xf2) == 0x82) | |
354 | goto bad_area; | |
355 | } | |
356 | } | |
357 | if (expand_stack(vma, address)) | |
358 | goto bad_area; | |
359 | /* | |
360 | * Ok, we have a good vm_area for this memory access, so | |
361 | * we can handle it.. | |
362 | */ | |
363 | good_area: | |
364 | si_code = SEGV_ACCERR; | |
365 | ||
366 | /* If we took a ITLB miss on a non-executable page, catch | |
367 | * that here. | |
368 | */ | |
369 | if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) { | |
370 | BUG_ON(address != regs->tpc); | |
371 | BUG_ON(regs->tstate & TSTATE_PRIV); | |
372 | goto bad_area; | |
373 | } | |
374 | ||
375 | if (fault_code & FAULT_CODE_WRITE) { | |
376 | if (!(vma->vm_flags & VM_WRITE)) | |
377 | goto bad_area; | |
378 | ||
379 | /* Spitfire has an icache which does not snoop | |
380 | * processor stores. Later processors do... | |
381 | */ | |
382 | if (tlb_type == spitfire && | |
383 | (vma->vm_flags & VM_EXEC) != 0 && | |
384 | vma->vm_file != NULL) | |
385 | set_thread_fault_code(fault_code | | |
386 | FAULT_CODE_BLKCOMMIT); | |
387 | } else { | |
388 | /* Allow reads even for write-only mappings */ | |
389 | if (!(vma->vm_flags & (VM_READ | VM_EXEC))) | |
390 | goto bad_area; | |
391 | } | |
392 | ||
393 | switch (handle_mm_fault(mm, vma, address, (fault_code & FAULT_CODE_WRITE))) { | |
394 | case VM_FAULT_MINOR: | |
395 | current->min_flt++; | |
396 | break; | |
397 | case VM_FAULT_MAJOR: | |
398 | current->maj_flt++; | |
399 | break; | |
400 | case VM_FAULT_SIGBUS: | |
401 | goto do_sigbus; | |
402 | case VM_FAULT_OOM: | |
403 | goto out_of_memory; | |
404 | default: | |
405 | BUG(); | |
406 | } | |
407 | ||
408 | up_read(&mm->mmap_sem); | |
efdc1e20 | 409 | return; |
1da177e4 LT |
410 | |
411 | /* | |
412 | * Something tried to access memory that isn't in our memory map.. | |
413 | * Fix it, but check if it's kernel or user first.. | |
414 | */ | |
415 | bad_area: | |
416 | insn = get_fault_insn(regs, insn); | |
417 | up_read(&mm->mmap_sem); | |
418 | ||
419 | handle_kernel_fault: | |
420 | do_kernel_fault(regs, si_code, fault_code, insn, address); | |
efdc1e20 | 421 | return; |
1da177e4 LT |
422 | |
423 | /* | |
424 | * We ran out of memory, or some other thing happened to us that made | |
425 | * us unable to handle the page fault gracefully. | |
426 | */ | |
427 | out_of_memory: | |
428 | insn = get_fault_insn(regs, insn); | |
429 | up_read(&mm->mmap_sem); | |
430 | printk("VM: killing process %s\n", current->comm); | |
431 | if (!(regs->tstate & TSTATE_PRIV)) | |
432 | do_exit(SIGKILL); | |
433 | goto handle_kernel_fault; | |
434 | ||
435 | intr_or_no_mm: | |
436 | insn = get_fault_insn(regs, 0); | |
437 | goto handle_kernel_fault; | |
438 | ||
439 | do_sigbus: | |
440 | insn = get_fault_insn(regs, insn); | |
441 | up_read(&mm->mmap_sem); | |
442 | ||
443 | /* | |
444 | * Send a sigbus, regardless of whether we were in kernel | |
445 | * or user mode. | |
446 | */ | |
447 | do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, insn, fault_code); | |
448 | ||
449 | /* Kernel mode? Handle exceptions or die */ | |
450 | if (regs->tstate & TSTATE_PRIV) | |
451 | goto handle_kernel_fault; | |
1da177e4 | 452 | } |