1/*
  2 *  arch/cris/mm/fault.c
  3 *
  4 *  Copyright (C) 2000-2010  Axis Communications AB
  5 */
  6
  7#include <linux/mm.h>
  8#include <linux/interrupt.h>
  9#include <linux/module.h>
 10#include <linux/wait.h>
 11#include <asm/uaccess.h>
 12#include <arch/system.h>
 13
 14extern int find_fixup_code(struct pt_regs *);
 15extern void die_if_kernel(const char *, struct pt_regs *, long);
 16extern void show_registers(struct pt_regs *regs);
 17
 18/* debug of low-level TLB reload */
 19#undef DEBUG
 20
 21#ifdef DEBUG
 22#define D(x) x
 23#else
 24#define D(x)
 25#endif
 26
 27/* debug of higher-level faults */
 28#define DPG(x)
 29
 30/* current active page directory */
 31
 32DEFINE_PER_CPU(pgd_t *, current_pgd);
 33unsigned long cris_signal_return_page;
 34
 35/*
 36 * This routine handles page faults.  It determines the address,
 37 * and the problem, and then passes it off to one of the appropriate
 38 * routines.
 39 *
 40 * Notice that the address we're given is aligned to the page the fault
 41 * occurred in, since we only get the PFN in R_MMU_CAUSE not the complete
 42 * address.
 43 *
 44 * error_code:
 45 *      bit 0 == 0 means no page found, 1 means protection fault
 46 *      bit 1 == 0 means read, 1 means write
 47 *
 48 * If this routine detects a bad access, it returns 1, otherwise it
 49 * returns 0.
 50 */
 51
 52asmlinkage void
 53do_page_fault(unsigned long address, struct pt_regs *regs,
 54	      int protection, int writeaccess)
 55{
 56	struct task_struct *tsk;
 57	struct mm_struct *mm;
 58	struct vm_area_struct * vma;
 59	siginfo_t info;
 60	int fault;
 61	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 62
 63	D(printk(KERN_DEBUG
 64		 "Page fault for %lX on %X at %lX, prot %d write %d\n",
 65		 address, smp_processor_id(), instruction_pointer(regs),
 66		 protection, writeaccess));
 67
 68	tsk = current;
 69
 70	/*
 71	 * We fault-in kernel-space virtual memory on-demand. The
 72	 * 'reference' page table is init_mm.pgd.
 73	 *
 74	 * NOTE! We MUST NOT take any locks for this case. We may
 75	 * be in an interrupt or a critical region, and should
 76	 * only copy the information from the master page table,
 77	 * nothing more.
 78	 *
 79	 * NOTE2: This is done so that, when updating the vmalloc
 80	 * mappings we don't have to walk all processes pgdirs and
 81	 * add the high mappings all at once. Instead we do it as they
 82	 * are used. However vmalloc'ed page entries have the PAGE_GLOBAL
 83	 * bit set so sometimes the TLB can use a lingering entry.
 84	 *
 85	 * This verifies that the fault happens in kernel space
 86	 * and that the fault was not a protection error (error_code & 1).
 87	 */
 88
 89	if (address >= VMALLOC_START &&
 90	    !protection &&
 91	    !user_mode(regs))
 92		goto vmalloc_fault;
 93
 94	/* When stack execution is not allowed we store the signal
 95	 * trampolines in the reserved cris_signal_return_page.
 96	 * Handle this in the exact same way as vmalloc (we know
 97	 * that the mapping is there and is valid so no need to
 98	 * call handle_mm_fault).
 99	 */
100	if (cris_signal_return_page &&
101	    address == cris_signal_return_page &&
102	    !protection && user_mode(regs))
103		goto vmalloc_fault;
104
105	/* we can and should enable interrupts at this point */
106	local_irq_enable();
107
108	mm = tsk->mm;
109	info.si_code = SEGV_MAPERR;
110
111	/*
112	 * If we're in an interrupt or "atomic" operation or have no
113	 * user context, we must not take the fault.
114	 */
115
116	if (in_atomic() || !mm)
117		goto no_context;
118
119	if (user_mode(regs))
120		flags |= FAULT_FLAG_USER;
121retry:
122	down_read(&mm->mmap_sem);
123	vma = find_vma(mm, address);
124	if (!vma)
125		goto bad_area;
126	if (vma->vm_start <= address)
127		goto good_area;
128	if (!(vma->vm_flags & VM_GROWSDOWN))
129		goto bad_area;
130	if (user_mode(regs)) {
131		/*
132		 * accessing the stack below usp is always a bug.
133		 * we get page-aligned addresses so we can only check
134		 * if we're within a page from usp, but that might be
135		 * enough to catch brutal errors at least.
136		 */
137		if (address + PAGE_SIZE < rdusp())
138			goto bad_area;
139	}
140	if (expand_stack(vma, address))
141		goto bad_area;
142
143	/*
144	 * Ok, we have a good vm_area for this memory access, so
145	 * we can handle it..
146	 */
147
148 good_area:
149	info.si_code = SEGV_ACCERR;
150
151	/* first do some preliminary protection checks */
152
153	if (writeaccess == 2){
154		if (!(vma->vm_flags & VM_EXEC))
155			goto bad_area;
156	} else if (writeaccess == 1) {
157		if (!(vma->vm_flags & VM_WRITE))
158			goto bad_area;
159		flags |= FAULT_FLAG_WRITE;
160	} else {
161		if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
162			goto bad_area;
163	}
164
165	/*
166	 * If for any reason at all we couldn't handle the fault,
167	 * make sure we exit gracefully rather than endlessly redo
168	 * the fault.
169	 */
170
171	fault = handle_mm_fault(mm, vma, address, flags);
172
173	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
174		return;
175
176	if (unlikely(fault & VM_FAULT_ERROR)) {
177		if (fault & VM_FAULT_OOM)
178			goto out_of_memory;
179		else if (fault & VM_FAULT_SIGBUS)
180			goto do_sigbus;
181		BUG();
182	}
183
184	if (flags & FAULT_FLAG_ALLOW_RETRY) {
185		if (fault & VM_FAULT_MAJOR)
186			tsk->maj_flt++;
187		else
188			tsk->min_flt++;
189		if (fault & VM_FAULT_RETRY) {
190			flags &= ~FAULT_FLAG_ALLOW_RETRY;
191			flags |= FAULT_FLAG_TRIED;
192
193			/*
194			 * No need to up_read(&mm->mmap_sem) as we would
195			 * have already released it in __lock_page_or_retry
196			 * in mm/filemap.c.
197			 */
198
199			goto retry;
200		}
201	}
202
203	up_read(&mm->mmap_sem);
204	return;
205
206	/*
207	 * Something tried to access memory that isn't in our memory map..
208	 * Fix it, but check if it's kernel or user first..
209	 */
210
211 bad_area:
212	up_read(&mm->mmap_sem);
213
214 bad_area_nosemaphore:
215	DPG(show_registers(regs));
216
217	/* User mode accesses just cause a SIGSEGV */
218
219	if (user_mode(regs)) {
220		printk(KERN_NOTICE "%s (pid %d) segfaults for page "
221			"address %08lx at pc %08lx\n",
222			tsk->comm, tsk->pid,
223			address, instruction_pointer(regs));
224
225		/* With DPG on, we've already dumped registers above.  */
226		DPG(if (0))
227			show_registers(regs);
228
229#ifdef CONFIG_NO_SEGFAULT_TERMINATION
230		DECLARE_WAIT_QUEUE_HEAD(wq);
231		wait_event_interruptible(wq, 0 == 1);
232#else
233		info.si_signo = SIGSEGV;
234		info.si_errno = 0;
235		/* info.si_code has been set above */
236		info.si_addr = (void *)address;
237		force_sig_info(SIGSEGV, &info, tsk);
238#endif
239		return;
240	}
241
242 no_context:
243
244	/* Are we prepared to handle this kernel fault?
245	 *
246	 * (The kernel has valid exception-points in the source
247	 *  when it accesses user-memory. When it fails in one
248	 *  of those points, we find it in a table and do a jump
249	 *  to some fixup code that loads an appropriate error
250	 *  code)
251	 */
252
253	if (find_fixup_code(regs))
254		return;
255
256	/*
257	 * Oops. The kernel tried to access some bad page. We'll have to
258	 * terminate things with extreme prejudice.
259	 */
260
261	if (!oops_in_progress) {
262		oops_in_progress = 1;
263		if ((unsigned long) (address) < PAGE_SIZE)
264			printk(KERN_ALERT "Unable to handle kernel NULL "
265				"pointer dereference");
266		else
267			printk(KERN_ALERT "Unable to handle kernel access"
268				" at virtual address %08lx\n", address);
269
270		die_if_kernel("Oops", regs, (writeaccess << 1) | protection);
271		oops_in_progress = 0;
272	}
273
274	do_exit(SIGKILL);
275
276	/*
277	 * We ran out of memory, or some other thing happened to us that made
278	 * us unable to handle the page fault gracefully.
279	 */
280
281 out_of_memory:
282	up_read(&mm->mmap_sem);
283	if (!user_mode(regs))
284		goto no_context;
285	pagefault_out_of_memory();
286	return;
287
288 do_sigbus:
289	up_read(&mm->mmap_sem);
290
291	/*
292	 * Send a sigbus, regardless of whether we were in kernel
293	 * or user mode.
294	 */
295	info.si_signo = SIGBUS;
296	info.si_errno = 0;
297	info.si_code = BUS_ADRERR;
298	info.si_addr = (void *)address;
299	force_sig_info(SIGBUS, &info, tsk);
300
301	/* Kernel mode? Handle exceptions or die */
302	if (!user_mode(regs))
303		goto no_context;
304	return;
305
306vmalloc_fault:
307	{
308		/*
309		 * Synchronize this task's top level page-table
310		 * with the 'reference' page table.
311		 *
312		 * Use current_pgd instead of tsk->active_mm->pgd
313		 * since the latter might be unavailable if this
314		 * code is executed in a misfortunately run irq
315		 * (like inside schedule() between switch_mm and
316		 *  switch_to...).
317		 */
318
319		int offset = pgd_index(address);
320		pgd_t *pgd, *pgd_k;
321		pud_t *pud, *pud_k;
322		pmd_t *pmd, *pmd_k;
323		pte_t *pte_k;
324
325		pgd = (pgd_t *)per_cpu(current_pgd, smp_processor_id()) + offset;
326		pgd_k = init_mm.pgd + offset;
327
328		/* Since we're two-level, we don't need to do both
329		 * set_pgd and set_pmd (they do the same thing). If
330		 * we go three-level at some point, do the right thing
331		 * with pgd_present and set_pgd here.
332		 *
333		 * Also, since the vmalloc area is global, we don't
334		 * need to copy individual PTE's, it is enough to
335		 * copy the pgd pointer into the pte page of the
336		 * root task. If that is there, we'll find our pte if
337		 * it exists.
338		 */
339
340		pud = pud_offset(pgd, address);
341		pud_k = pud_offset(pgd_k, address);
342		if (!pud_present(*pud_k))
343			goto no_context;
344
345		pmd = pmd_offset(pud, address);
346		pmd_k = pmd_offset(pud_k, address);
347
348		if (!pmd_present(*pmd_k))
349			goto bad_area_nosemaphore;
350
351		set_pmd(pmd, *pmd_k);
352
353		/* Make sure the actual PTE exists as well to
354		 * catch kernel vmalloc-area accesses to non-mapped
355		 * addresses. If we don't do this, this will just
356		 * silently loop forever.
357		 */
358
359		pte_k = pte_offset_kernel(pmd_k, address);
360		if (!pte_present(*pte_k))
361			goto no_context;
362
363		return;
364	}
365}
366
367/* Find fixup code. */
368int
369find_fixup_code(struct pt_regs *regs)
370{
371	const struct exception_table_entry *fixup;
372	/* in case of delay slot fault (v32) */
373	unsigned long ip = (instruction_pointer(regs) & ~0x1);
374
375	fixup = search_exception_tables(ip);
376	if (fixup != 0) {
377		/* Adjust the instruction pointer in the stackframe. */
378		instruction_pointer(regs) = fixup->fixup;
379		arch_fixup(regs);
380		return 1;
381	}
382
383	return 0;
384}