mm: fix exec activate_mm vs TLB shootdown and lazy tlb switching race
[linux/fpc-iii.git] / arch / xtensa / mm / fault.c
bloba14df5aa98c898f641812eef88fc6969d06ad172
1 // TODO VM_EXEC flag work-around, cache aliasing
2 /*
3 * arch/xtensa/mm/fault.c
5 * This file is subject to the terms and conditions of the GNU General Public
6 * License. See the file "COPYING" in the main directory of this archive
7 * for more details.
9 * Copyright (C) 2001 - 2010 Tensilica Inc.
11 * Chris Zankel <chris@zankel.net>
12 * Joe Taylor <joe@tensilica.com, joetylr@yahoo.com>
15 #include <linux/mm.h>
16 #include <linux/extable.h>
17 #include <linux/hardirq.h>
18 #include <linux/perf_event.h>
19 #include <linux/uaccess.h>
20 #include <asm/mmu_context.h>
21 #include <asm/cacheflush.h>
22 #include <asm/hardirq.h>
23 #include <asm/pgalloc.h>
25 DEFINE_PER_CPU(unsigned long, asid_cache) = ASID_USER_FIRST;
26 void bad_page_fault(struct pt_regs*, unsigned long, int);
28 #undef DEBUG_PAGE_FAULT
31 * This routine handles page faults. It determines the address,
32 * and the problem, and then passes it off to one of the appropriate
33 * routines.
35 * Note: does not handle Miss and MultiHit.
38 void do_page_fault(struct pt_regs *regs)
40 struct vm_area_struct * vma;
41 struct mm_struct *mm = current->mm;
42 unsigned int exccause = regs->exccause;
43 unsigned int address = regs->excvaddr;
44 siginfo_t info;
46 int is_write, is_exec;
47 int fault;
48 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
50 info.si_code = SEGV_MAPERR;
52 /* We fault-in kernel-space virtual memory on-demand. The
53 * 'reference' page table is init_mm.pgd.
55 if (address >= TASK_SIZE && !user_mode(regs))
56 goto vmalloc_fault;
58 /* If we're in an interrupt or have no user
59 * context, we must not take the fault..
61 if (faulthandler_disabled() || !mm) {
62 bad_page_fault(regs, address, SIGSEGV);
63 return;
66 is_write = (exccause == EXCCAUSE_STORE_CACHE_ATTRIBUTE) ? 1 : 0;
67 is_exec = (exccause == EXCCAUSE_ITLB_PRIVILEGE ||
68 exccause == EXCCAUSE_ITLB_MISS ||
69 exccause == EXCCAUSE_FETCH_CACHE_ATTRIBUTE) ? 1 : 0;
71 #ifdef DEBUG_PAGE_FAULT
72 printk("[%s:%d:%08x:%d:%08x:%s%s]\n", current->comm, current->pid,
73 address, exccause, regs->pc, is_write? "w":"", is_exec? "x":"");
74 #endif
76 if (user_mode(regs))
77 flags |= FAULT_FLAG_USER;
78 retry:
79 down_read(&mm->mmap_sem);
80 vma = find_vma(mm, address);
82 if (!vma)
83 goto bad_area;
84 if (vma->vm_start <= address)
85 goto good_area;
86 if (!(vma->vm_flags & VM_GROWSDOWN))
87 goto bad_area;
88 if (expand_stack(vma, address))
89 goto bad_area;
91 /* Ok, we have a good vm_area for this memory access, so
92 * we can handle it..
95 good_area:
96 info.si_code = SEGV_ACCERR;
98 if (is_write) {
99 if (!(vma->vm_flags & VM_WRITE))
100 goto bad_area;
101 flags |= FAULT_FLAG_WRITE;
102 } else if (is_exec) {
103 if (!(vma->vm_flags & VM_EXEC))
104 goto bad_area;
105 } else /* Allow read even from write-only pages. */
106 if (!(vma->vm_flags & (VM_READ | VM_WRITE)))
107 goto bad_area;
109 /* If for any reason at all we couldn't handle the fault,
110 * make sure we exit gracefully rather than endlessly redo
111 * the fault.
113 fault = handle_mm_fault(vma, address, flags);
115 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
116 return;
118 if (unlikely(fault & VM_FAULT_ERROR)) {
119 if (fault & VM_FAULT_OOM)
120 goto out_of_memory;
121 else if (fault & VM_FAULT_SIGSEGV)
122 goto bad_area;
123 else if (fault & VM_FAULT_SIGBUS)
124 goto do_sigbus;
125 BUG();
127 if (flags & FAULT_FLAG_ALLOW_RETRY) {
128 if (fault & VM_FAULT_MAJOR)
129 current->maj_flt++;
130 else
131 current->min_flt++;
132 if (fault & VM_FAULT_RETRY) {
133 flags &= ~FAULT_FLAG_ALLOW_RETRY;
134 flags |= FAULT_FLAG_TRIED;
136 /* No need to up_read(&mm->mmap_sem) as we would
137 * have already released it in __lock_page_or_retry
138 * in mm/filemap.c.
141 goto retry;
145 up_read(&mm->mmap_sem);
146 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
147 if (flags & VM_FAULT_MAJOR)
148 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
149 else
150 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
152 return;
154 /* Something tried to access memory that isn't in our memory map..
155 * Fix it, but check if it's kernel or user first..
157 bad_area:
158 up_read(&mm->mmap_sem);
159 if (user_mode(regs)) {
160 current->thread.bad_vaddr = address;
161 current->thread.error_code = is_write;
162 info.si_signo = SIGSEGV;
163 info.si_errno = 0;
164 /* info.si_code has been set above */
165 info.si_addr = (void *) address;
166 force_sig_info(SIGSEGV, &info, current);
167 return;
169 bad_page_fault(regs, address, SIGSEGV);
170 return;
173 /* We ran out of memory, or some other thing happened to us that made
174 * us unable to handle the page fault gracefully.
176 out_of_memory:
177 up_read(&mm->mmap_sem);
178 if (!user_mode(regs))
179 bad_page_fault(regs, address, SIGKILL);
180 else
181 pagefault_out_of_memory();
182 return;
184 do_sigbus:
185 up_read(&mm->mmap_sem);
187 /* Send a sigbus, regardless of whether we were in kernel
188 * or user mode.
190 current->thread.bad_vaddr = address;
191 info.si_code = SIGBUS;
192 info.si_errno = 0;
193 info.si_code = BUS_ADRERR;
194 info.si_addr = (void *) address;
195 force_sig_info(SIGBUS, &info, current);
197 /* Kernel mode? Handle exceptions or die */
198 if (!user_mode(regs))
199 bad_page_fault(regs, address, SIGBUS);
200 return;
202 vmalloc_fault:
204 /* Synchronize this task's top level page-table
205 * with the 'reference' page table.
207 struct mm_struct *act_mm = current->active_mm;
208 int index = pgd_index(address);
209 pgd_t *pgd, *pgd_k;
210 pmd_t *pmd, *pmd_k;
211 pte_t *pte_k;
213 if (act_mm == NULL)
214 goto bad_page_fault;
216 pgd = act_mm->pgd + index;
217 pgd_k = init_mm.pgd + index;
219 if (!pgd_present(*pgd_k))
220 goto bad_page_fault;
222 pgd_val(*pgd) = pgd_val(*pgd_k);
224 pmd = pmd_offset(pgd, address);
225 pmd_k = pmd_offset(pgd_k, address);
226 if (!pmd_present(*pmd) || !pmd_present(*pmd_k))
227 goto bad_page_fault;
229 pmd_val(*pmd) = pmd_val(*pmd_k);
230 pte_k = pte_offset_kernel(pmd_k, address);
232 if (!pte_present(*pte_k))
233 goto bad_page_fault;
234 return;
236 bad_page_fault:
237 bad_page_fault(regs, address, SIGKILL);
238 return;
242 void
243 bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
245 extern void die(const char*, struct pt_regs*, long);
246 const struct exception_table_entry *entry;
248 /* Are we prepared to handle this kernel fault? */
249 if ((entry = search_exception_tables(regs->pc)) != NULL) {
250 #ifdef DEBUG_PAGE_FAULT
251 printk(KERN_DEBUG "%s: Exception at pc=%#010lx (%lx)\n",
252 current->comm, regs->pc, entry->fixup);
253 #endif
254 current->thread.bad_uaddr = address;
255 regs->pc = entry->fixup;
256 return;
259 /* Oops. The kernel tried to access some bad page. We'll have to
260 * terminate things with extreme prejudice.
262 printk(KERN_ALERT "Unable to handle kernel paging request at virtual "
263 "address %08lx\n pc = %08lx, ra = %08lx\n",
264 address, regs->pc, regs->areg[0]);
265 die("Oops", regs, sig);
266 do_exit(sig);