1/*
  2 * Copyright (C) 2009 Wind River Systems Inc
  3 *   Implemented by fredrik.markstrom@gmail.com and ivarholmqvist@gmail.com
  4 *
  5 * based on arch/mips/mm/fault.c which is:
  6 *
  7 * Copyright (C) 1995-2000 Ralf Baechle
  8 *
  9 * This file is subject to the terms and conditions of the GNU General Public
 10 * License.  See the file "COPYING" in the main directory of this archive
 11 * for more details.
 12 */
 13
 14#include <linux/signal.h>
 15#include <linux/sched.h>
 16#include <linux/sched/debug.h>
 17#include <linux/interrupt.h>
 18#include <linux/kernel.h>
 19#include <linux/errno.h>
 20#include <linux/string.h>
 21#include <linux/types.h>
 22#include <linux/ptrace.h>
 23#include <linux/mman.h>
 24#include <linux/mm.h>
 25#include <linux/extable.h>
 26#include <linux/uaccess.h>
 27#include <linux/ptrace.h>
 28
 29#include <asm/mmu_context.h>
 30#include <asm/traps.h>
 31
 32#define EXC_SUPERV_INSN_ACCESS	9  /* Supervisor only instruction address */
 33#define EXC_SUPERV_DATA_ACCESS	11 /* Supervisor only data address */
 34#define EXC_X_PROTECTION_FAULT	13 /* TLB permission violation (x) */
 35#define EXC_R_PROTECTION_FAULT	14 /* TLB permission violation (r) */
 36#define EXC_W_PROTECTION_FAULT	15 /* TLB permission violation (w) */
 37
 38/*
 39 * This routine handles page faults.  It determines the address,
 40 * and the problem, and then passes it off to one of the appropriate
 41 * routines.
 42 */
 43asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long cause,
 44				unsigned long address)
 45{
 46	struct vm_area_struct *vma = NULL;
 47	struct task_struct *tsk = current;
 48	struct mm_struct *mm = tsk->mm;
 49	int code = SEGV_MAPERR;
 50	int fault;
 51	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 52
 53	cause >>= 2;
 54
 55	/* Restart the instruction */
 56	regs->ea -= 4;
 57
 58	/*
 59	 * We fault-in kernel-space virtual memory on-demand. The
 60	 * 'reference' page table is init_mm.pgd.
 61	 *
 62	 * NOTE! We MUST NOT take any locks for this case. We may
 63	 * be in an interrupt or a critical region, and should
 64	 * only copy the information from the master page table,
 65	 * nothing more.
 66	 */
 67	if (unlikely(address >= VMALLOC_START && address <= VMALLOC_END)) {
 68		if (user_mode(regs))
 69			goto bad_area_nosemaphore;
 70		else
 71			goto vmalloc_fault;
 72	}
 73
 74	if (unlikely(address >= TASK_SIZE))
 75		goto bad_area_nosemaphore;
 76
 77	/*
 78	 * If we're in an interrupt or have no user
 79	 * context, we must not take the fault..
 80	 */
 81	if (faulthandler_disabled() || !mm)
 82		goto bad_area_nosemaphore;
 83
 84	if (user_mode(regs))
 85		flags |= FAULT_FLAG_USER;
 86
 87	if (!down_read_trylock(&mm->mmap_sem)) {
 88		if (!user_mode(regs) && !search_exception_tables(regs->ea))
 89			goto bad_area_nosemaphore;
 90retry:
 91		down_read(&mm->mmap_sem);
 92	}
 93
 94	vma = find_vma(mm, address);
 95	if (!vma)
 96		goto bad_area;
 97	if (vma->vm_start <= address)
 98		goto good_area;
 99	if (!(vma->vm_flags & VM_GROWSDOWN))
100		goto bad_area;
101	if (expand_stack(vma, address))
102		goto bad_area;
103/*
104 * Ok, we have a good vm_area for this memory access, so
105 * we can handle it..
106 */
107good_area:
108	code = SEGV_ACCERR;
109
110	switch (cause) {
111	case EXC_SUPERV_INSN_ACCESS:
112		goto bad_area;
113	case EXC_SUPERV_DATA_ACCESS:
114		goto bad_area;
115	case EXC_X_PROTECTION_FAULT:
116		if (!(vma->vm_flags & VM_EXEC))
117			goto bad_area;
118		break;
119	case EXC_R_PROTECTION_FAULT:
120		if (!(vma->vm_flags & VM_READ))
121			goto bad_area;
122		break;
123	case EXC_W_PROTECTION_FAULT:
124		if (!(vma->vm_flags & VM_WRITE))
125			goto bad_area;
126		flags = FAULT_FLAG_WRITE;
127		break;
128	}
129
130	/*
131	 * If for any reason at all we couldn't handle the fault,
132	 * make sure we exit gracefully rather than endlessly redo
133	 * the fault.
134	 */
135	fault = handle_mm_fault(vma, address, flags);
136
137	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
138		return;
139
140	if (unlikely(fault & VM_FAULT_ERROR)) {
141		if (fault & VM_FAULT_OOM)
142			goto out_of_memory;
143		else if (fault & VM_FAULT_SIGSEGV)
144			goto bad_area;
145		else if (fault & VM_FAULT_SIGBUS)
146			goto do_sigbus;
147		BUG();
148	}
149
150	/*
151	 * Major/minor page fault accounting is only done on the
152	 * initial attempt. If we go through a retry, it is extremely
153	 * likely that the page will be found in page cache at that point.
154	 */
155	if (flags & FAULT_FLAG_ALLOW_RETRY) {
156		if (fault & VM_FAULT_MAJOR)
157			current->maj_flt++;
158		else
159			current->min_flt++;
160		if (fault & VM_FAULT_RETRY) {
161			/* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
162			 * of starvation. */
163			flags &= ~FAULT_FLAG_ALLOW_RETRY;
164			flags |= FAULT_FLAG_TRIED;
165
166			/*
167			 * No need to up_read(&mm->mmap_sem) as we would
168			 * have already released it in __lock_page_or_retry
169			 * in mm/filemap.c.
170			 */
171
172			goto retry;
173		}
174	}
175
176	up_read(&mm->mmap_sem);
177	return;
178
179/*
180 * Something tried to access memory that isn't in our memory map..
181 * Fix it, but check if it's kernel or user first..
182 */
183bad_area:
184	up_read(&mm->mmap_sem);
185
186bad_area_nosemaphore:
187	/* User mode accesses just cause a SIGSEGV */
188	if (user_mode(regs)) {
189		if (unhandled_signal(current, SIGSEGV) && printk_ratelimit()) {
190			pr_info("%s: unhandled page fault (%d) at 0x%08lx, "
191				"cause %ld\n", current->comm, SIGSEGV, address, cause);
192			show_regs(regs);
193		}
194		_exception(SIGSEGV, regs, code, address);
195		return;
196	}
197
198no_context:
199	/* Are we prepared to handle this kernel fault? */
200	if (fixup_exception(regs))
201		return;
202
203	/*
204	 * Oops. The kernel tried to access some bad page. We'll have to
205	 * terminate things with extreme prejudice.
206	 */
207	bust_spinlocks(1);
208
209	pr_alert("Unable to handle kernel %s at virtual address %08lx",
210		address < PAGE_SIZE ? "NULL pointer dereference" :
211		"paging request", address);
212	pr_alert("ea = %08lx, ra = %08lx, cause = %ld\n", regs->ea, regs->ra,
213		cause);
214	panic("Oops");
215	return;
216
217/*
218 * We ran out of memory, or some other thing happened to us that made
219 * us unable to handle the page fault gracefully.
220 */
221out_of_memory:
222	up_read(&mm->mmap_sem);
223	if (!user_mode(regs))
224		goto no_context;
225	pagefault_out_of_memory();
226	return;
227
228do_sigbus:
229	up_read(&mm->mmap_sem);
230
231	/* Kernel mode? Handle exceptions or die */
232	if (!user_mode(regs))
233		goto no_context;
234
235	_exception(SIGBUS, regs, BUS_ADRERR, address);
236	return;
237
238vmalloc_fault:
239	{
240		/*
241		 * Synchronize this task's top level page-table
242		 * with the 'reference' page table.
243		 *
244		 * Do _not_ use "tsk" here. We might be inside
245		 * an interrupt in the middle of a task switch..
246		 */
247		int offset = pgd_index(address);
248		pgd_t *pgd, *pgd_k;
249		pud_t *pud, *pud_k;
250		pmd_t *pmd, *pmd_k;
251		pte_t *pte_k;
252
253		pgd = pgd_current + offset;
254		pgd_k = init_mm.pgd + offset;
255
256		if (!pgd_present(*pgd_k))
257			goto no_context;
258		set_pgd(pgd, *pgd_k);
259
260		pud = pud_offset(pgd, address);
261		pud_k = pud_offset(pgd_k, address);
262		if (!pud_present(*pud_k))
263			goto no_context;
264		pmd = pmd_offset(pud, address);
265		pmd_k = pmd_offset(pud_k, address);
266		if (!pmd_present(*pmd_k))
267			goto no_context;
268		set_pmd(pmd, *pmd_k);
269
270		pte_k = pte_offset_kernel(pmd_k, address);
271		if (!pte_present(*pte_k))
272			goto no_context;
273
274		flush_tlb_one(address);
275		return;
276	}
277}