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.
4 * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu)
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/smp_lock.h>
19 #include <linux/init.h>
20 #include <linux/interrupt.h>
23 #include <asm/pgtable.h>
24 #include <asm/openprom.h>
25 #include <asm/oplib.h>
26 #include <asm/uaccess.h>
29 #include <asm/sections.h>
30 #include <asm/kdebug.h>
32 #define ELEMENTS(arr) (sizeof (arr)/sizeof (arr[0]))
34 extern struct sparc_phys_banks sp_banks[SPARC_PHYS_BANKS];
37 * To debug kernel during syscall entry.
39 void syscall_trace_entry(struct pt_regs *regs)
41 printk("scall entry: %s[%d]/cpu%d: %d\n", current->comm, current->pid, smp_processor_id(), (int) regs->u_regs[UREG_G1]);
45 * To debug kernel during syscall exit.
47 void syscall_trace_exit(struct pt_regs *regs)
49 printk("scall exit: %s[%d]/cpu%d: %d\n", current->comm, current->pid, smp_processor_id(), (int) regs->u_regs[UREG_G1]);
53 * To debug kernel to catch accesses to certain virtual/physical addresses.
54 * Mode = 0 selects physical watchpoints, mode = 1 selects virtual watchpoints.
55 * flags = VM_READ watches memread accesses, flags = VM_WRITE watches memwrite accesses.
56 * Caller passes in a 64bit aligned addr, with mask set to the bytes that need to be
57 * watched. This is only useful on a single cpu machine for now. After the watchpoint
58 * is detected, the process causing it will be killed, thus preventing an infinite loop.
60 void set_brkpt(unsigned long addr, unsigned char mask, int flags, int mode)
62 unsigned long lsubits;
64 __asm__ __volatile__("ldxa [%%g0] %1, %0"
66 : "i" (ASI_LSU_CONTROL));
67 lsubits &= ~(LSU_CONTROL_PM | LSU_CONTROL_VM |
68 LSU_CONTROL_PR | LSU_CONTROL_VR |
69 LSU_CONTROL_PW | LSU_CONTROL_VW);
71 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
74 : "r" (addr), "r" (mode ? VIRT_WATCHPOINT : PHYS_WATCHPOINT),
77 lsubits |= ((unsigned long)mask << (mode ? 25 : 33));
79 lsubits |= (mode ? LSU_CONTROL_VR : LSU_CONTROL_PR);
81 lsubits |= (mode ? LSU_CONTROL_VW : LSU_CONTROL_PW);
82 __asm__ __volatile__("stxa %0, [%%g0] %1\n\t"
85 : "r" (lsubits), "i" (ASI_LSU_CONTROL)
89 /* Nice, simple, prom library does all the sweating for us. ;) */
90 unsigned long __init prom_probe_memory (void)
92 register struct linux_mlist_p1275 *mlist;
93 register unsigned long bytes, base_paddr, tally;
97 mlist = *prom_meminfo()->p1275_available;
98 bytes = tally = mlist->num_bytes;
99 base_paddr = mlist->start_adr;
101 sp_banks[0].base_addr = base_paddr;
102 sp_banks[0].num_bytes = bytes;
104 while (mlist->theres_more != (void *) 0) {
106 mlist = mlist->theres_more;
107 bytes = mlist->num_bytes;
109 if (i >= SPARC_PHYS_BANKS-1) {
110 printk ("The machine has more banks than "
111 "this kernel can support\n"
112 "Increase the SPARC_PHYS_BANKS "
113 "setting (currently %d)\n",
115 i = SPARC_PHYS_BANKS-1;
119 sp_banks[i].base_addr = mlist->start_adr;
120 sp_banks[i].num_bytes = mlist->num_bytes;
124 sp_banks[i].base_addr = 0xdeadbeefbeefdeadUL;
125 sp_banks[i].num_bytes = 0;
127 /* Now mask all bank sizes on a page boundary, it is all we can
130 for (i = 0; sp_banks[i].num_bytes != 0; i++)
131 sp_banks[i].num_bytes &= PAGE_MASK;
136 static void unhandled_fault(unsigned long address, struct task_struct *tsk,
137 struct pt_regs *regs)
139 if ((unsigned long) address < PAGE_SIZE) {
140 printk(KERN_ALERT "Unable to handle kernel NULL "
141 "pointer dereference\n");
143 printk(KERN_ALERT "Unable to handle kernel paging request "
144 "at virtual address %016lx\n", (unsigned long)address);
146 printk(KERN_ALERT "tsk->{mm,active_mm}->context = %016lx\n",
148 CTX_HWBITS(tsk->mm->context) :
149 CTX_HWBITS(tsk->active_mm->context)));
150 printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %016lx\n",
151 (tsk->mm ? (unsigned long) tsk->mm->pgd :
152 (unsigned long) tsk->active_mm->pgd));
153 if (notify_die(DIE_GPF, "general protection fault", regs,
154 0, 0, SIGSEGV) == NOTIFY_STOP)
156 die_if_kernel("Oops", regs);
159 static void bad_kernel_pc(struct pt_regs *regs)
163 printk(KERN_CRIT "OOPS: Bogus kernel PC [%016lx] in fault handler\n",
165 __asm__("mov %%sp, %0" : "=r" (ksp));
166 show_stack(current, ksp);
167 unhandled_fault(regs->tpc, current, regs);
171 * We now make sure that mmap_sem is held in all paths that call
172 * this. Additionally, to prevent kswapd from ripping ptes from
173 * under us, raise interrupts around the time that we look at the
174 * pte, kswapd will have to wait to get his smp ipi response from
175 * us. This saves us having to get page_table_lock.
177 static unsigned int get_user_insn(unsigned long tpc)
179 pgd_t *pgdp = pgd_offset(current->mm, tpc);
185 unsigned long pstate;
189 pudp = pud_offset(pgdp, tpc);
192 pmdp = pmd_offset(pudp, tpc);
196 /* This disables preemption for us as well. */
197 __asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate));
198 __asm__ __volatile__("wrpr %0, %1, %%pstate"
199 : : "r" (pstate), "i" (PSTATE_IE));
200 ptep = pte_offset_map(pmdp, tpc);
202 if (!pte_present(pte))
205 pa = (pte_val(pte) & _PAGE_PADDR);
206 pa += (tpc & ~PAGE_MASK);
208 /* Use phys bypass so we don't pollute dtlb/dcache. */
209 __asm__ __volatile__("lduwa [%1] %2, %0"
211 : "r" (pa), "i" (ASI_PHYS_USE_EC));
215 __asm__ __volatile__("wrpr %0, 0x0, %%pstate" : : "r" (pstate));
220 extern unsigned long compute_effective_address(struct pt_regs *, unsigned int, unsigned int);
222 static void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
223 unsigned int insn, int fault_code)
230 if (fault_code & FAULT_CODE_ITLB)
231 info.si_addr = (void __user *) regs->tpc;
233 info.si_addr = (void __user *)
234 compute_effective_address(regs, insn, 0);
236 force_sig_info(sig, &info, current);
239 extern int handle_ldf_stq(u32, struct pt_regs *);
240 extern int handle_ld_nf(u32, struct pt_regs *);
242 static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
245 if (!regs->tpc || (regs->tpc & 0x3))
247 if (regs->tstate & TSTATE_PRIV) {
248 insn = *(unsigned int *) regs->tpc;
250 insn = get_user_insn(regs->tpc);
256 static void do_kernel_fault(struct pt_regs *regs, int si_code, int fault_code,
257 unsigned int insn, unsigned long address)
260 unsigned char asi = ASI_P;
262 if ((!insn) && (regs->tstate & TSTATE_PRIV))
265 /* If user insn could be read (thus insn is zero), that
266 * is fine. We will just gun down the process with a signal
270 if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
271 (insn & 0xc0800000) == 0xc0800000) {
273 asi = (regs->tstate >> 24);
276 if ((asi & 0xf2) == 0x82) {
277 if (insn & 0x1000000) {
278 handle_ldf_stq(insn, regs);
280 /* This was a non-faulting load. Just clear the
281 * destination register(s) and continue with the next
284 handle_ld_nf(insn, regs);
290 g2 = regs->u_regs[UREG_G2];
292 /* Is this in ex_table? */
293 if (regs->tstate & TSTATE_PRIV) {
296 if (asi == ASI_P && (insn & 0xc0800000) == 0xc0800000) {
298 asi = (regs->tstate >> 24);
303 /* Look in asi.h: All _S asis have LS bit set */
305 (fixup = search_extables_range(regs->tpc, &g2))) {
307 regs->tnpc = regs->tpc + 4;
308 regs->u_regs[UREG_G2] = g2;
312 /* The si_code was set to make clear whether
313 * this was a SEGV_MAPERR or SEGV_ACCERR fault.
315 do_fault_siginfo(si_code, SIGSEGV, regs, insn, fault_code);
320 unhandled_fault (address, current, regs);
323 asmlinkage void do_sparc64_fault(struct pt_regs *regs)
325 struct mm_struct *mm = current->mm;
326 struct vm_area_struct *vma;
327 unsigned int insn = 0;
328 int si_code, fault_code;
329 unsigned long address;
331 fault_code = get_thread_fault_code();
333 if (notify_die(DIE_PAGE_FAULT, "page_fault", regs,
334 fault_code, 0, SIGSEGV) == NOTIFY_STOP)
337 si_code = SEGV_MAPERR;
338 address = current_thread_info()->fault_address;
340 if ((fault_code & FAULT_CODE_ITLB) &&
341 (fault_code & FAULT_CODE_DTLB))
344 if (regs->tstate & TSTATE_PRIV) {
345 unsigned long tpc = regs->tpc;
347 /* Sanity check the PC. */
348 if ((tpc >= KERNBASE && tpc < (unsigned long) _etext) ||
349 (tpc >= MODULES_VADDR && tpc < MODULES_END)) {
350 /* Valid, no problems... */
358 * If we're in an interrupt or have no user
359 * context, we must not take the fault..
361 if (in_atomic() || !mm)
364 if (test_thread_flag(TIF_32BIT)) {
365 if (!(regs->tstate & TSTATE_PRIV))
366 regs->tpc &= 0xffffffff;
367 address &= 0xffffffff;
370 if (!down_read_trylock(&mm->mmap_sem)) {
371 if ((regs->tstate & TSTATE_PRIV) &&
372 !search_exception_tables(regs->tpc)) {
373 insn = get_fault_insn(regs, insn);
374 goto handle_kernel_fault;
376 down_read(&mm->mmap_sem);
379 vma = find_vma(mm, address);
383 /* Pure DTLB misses do not tell us whether the fault causing
384 * load/store/atomic was a write or not, it only says that there
385 * was no match. So in such a case we (carefully) read the
386 * instruction to try and figure this out. It's an optimization
387 * so it's ok if we can't do this.
389 * Special hack, window spill/fill knows the exact fault type.
392 (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) &&
393 (vma->vm_flags & VM_WRITE) != 0) {
394 insn = get_fault_insn(regs, 0);
397 if ((insn & 0xc0200000) == 0xc0200000 &&
398 (insn & 0x1780000) != 0x1680000) {
399 /* Don't bother updating thread struct value,
400 * because update_mmu_cache only cares which tlb
401 * the access came from.
403 fault_code |= FAULT_CODE_WRITE;
408 if (vma->vm_start <= address)
410 if (!(vma->vm_flags & VM_GROWSDOWN))
412 if (!(fault_code & FAULT_CODE_WRITE)) {
413 /* Non-faulting loads shouldn't expand stack. */
414 insn = get_fault_insn(regs, insn);
415 if ((insn & 0xc0800000) == 0xc0800000) {
419 asi = (regs->tstate >> 24);
422 if ((asi & 0xf2) == 0x82)
426 if (expand_stack(vma, address))
429 * Ok, we have a good vm_area for this memory access, so
433 si_code = SEGV_ACCERR;
435 /* If we took a ITLB miss on a non-executable page, catch
438 if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) {
439 BUG_ON(address != regs->tpc);
440 BUG_ON(regs->tstate & TSTATE_PRIV);
444 if (fault_code & FAULT_CODE_WRITE) {
445 if (!(vma->vm_flags & VM_WRITE))
448 /* Spitfire has an icache which does not snoop
449 * processor stores. Later processors do...
451 if (tlb_type == spitfire &&
452 (vma->vm_flags & VM_EXEC) != 0 &&
453 vma->vm_file != NULL)
454 set_thread_fault_code(fault_code |
455 FAULT_CODE_BLKCOMMIT);
457 /* Allow reads even for write-only mappings */
458 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
462 switch (handle_mm_fault(mm, vma, address, (fault_code & FAULT_CODE_WRITE))) {
469 case VM_FAULT_SIGBUS:
477 up_read(&mm->mmap_sem);
481 * Something tried to access memory that isn't in our memory map..
482 * Fix it, but check if it's kernel or user first..
485 insn = get_fault_insn(regs, insn);
486 up_read(&mm->mmap_sem);
489 do_kernel_fault(regs, si_code, fault_code, insn, address);
494 * We ran out of memory, or some other thing happened to us that made
495 * us unable to handle the page fault gracefully.
498 insn = get_fault_insn(regs, insn);
499 up_read(&mm->mmap_sem);
500 printk("VM: killing process %s\n", current->comm);
501 if (!(regs->tstate & TSTATE_PRIV))
503 goto handle_kernel_fault;
506 insn = get_fault_insn(regs, 0);
507 goto handle_kernel_fault;
510 insn = get_fault_insn(regs, insn);
511 up_read(&mm->mmap_sem);
514 * Send a sigbus, regardless of whether we were in kernel
517 do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, insn, fault_code);
519 /* Kernel mode? Handle exceptions or die */
520 if (regs->tstate & TSTATE_PRIV)
521 goto handle_kernel_fault;
524 /* These values are no longer needed, clear them. */
525 set_thread_fault_code(0);
526 current_thread_info()->fault_address = 0;