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/perf_event.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	vm_fault_t fault;
 51	unsigned int flags = FAULT_FLAG_DEFAULT;
 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	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
 88
 89retry:
 90	vma = lock_mm_and_find_vma(mm, address, regs);
 91	if (!vma)
 92		goto bad_area_nosemaphore;
 93/*
 94 * Ok, we have a good vm_area for this memory access, so
 95 * we can handle it..
 96 */
 97	code = SEGV_ACCERR;
 98
 99	switch (cause) {
100	case EXC_SUPERV_INSN_ACCESS:
101		goto bad_area;
102	case EXC_SUPERV_DATA_ACCESS:
103		goto bad_area;
104	case EXC_X_PROTECTION_FAULT:
105		if (!(vma->vm_flags & VM_EXEC))
106			goto bad_area;
107		break;
108	case EXC_R_PROTECTION_FAULT:
109		if (!(vma->vm_flags & VM_READ))
110			goto bad_area;
111		break;
112	case EXC_W_PROTECTION_FAULT:
113		if (!(vma->vm_flags & VM_WRITE))
114			goto bad_area;
115		flags = FAULT_FLAG_WRITE;
116		break;
117	}
118
119	/*
120	 * If for any reason at all we couldn't handle the fault,
121	 * make sure we exit gracefully rather than endlessly redo
122	 * the fault.
123	 */
124	fault = handle_mm_fault(vma, address, flags, regs);
125
126	if (fault_signal_pending(fault, regs)) {
127		if (!user_mode(regs))
128			goto no_context;
129		return;
130	}
131
132	/* The fault is fully completed (including releasing mmap lock) */
133	if (fault & VM_FAULT_COMPLETED)
134		return;
135
136	if (unlikely(fault & VM_FAULT_ERROR)) {
137		if (fault & VM_FAULT_OOM)
138			goto out_of_memory;
139		else if (fault & VM_FAULT_SIGSEGV)
140			goto bad_area;
141		else if (fault & VM_FAULT_SIGBUS)
142			goto do_sigbus;
143		BUG();
144	}
145
146	if (fault & VM_FAULT_RETRY) {
147		flags |= FAULT_FLAG_TRIED;
148
149		/*
150		 * No need to mmap_read_unlock(mm) as we would
151		 * have already released it in __lock_page_or_retry
152		 * in mm/filemap.c.
153		 */
154
155		goto retry;
156	}
157
158	mmap_read_unlock(mm);
159	return;
160
161/*
162 * Something tried to access memory that isn't in our memory map..
163 * Fix it, but check if it's kernel or user first..
164 */
165bad_area:
166	mmap_read_unlock(mm);
167
168bad_area_nosemaphore:
169	/* User mode accesses just cause a SIGSEGV */
170	if (user_mode(regs)) {
171		if (unhandled_signal(current, SIGSEGV) && printk_ratelimit()) {
172			pr_info("%s: unhandled page fault (%d) at 0x%08lx, "
173				"cause %ld\n", current->comm, SIGSEGV, address, cause);
174			show_regs(regs);
175		}
176		_exception(SIGSEGV, regs, code, address);
177		return;
178	}
179
180no_context:
181	/* Are we prepared to handle this kernel fault? */
182	if (fixup_exception(regs))
183		return;
184
185	/*
186	 * Oops. The kernel tried to access some bad page. We'll have to
187	 * terminate things with extreme prejudice.
188	 */
189	bust_spinlocks(1);
190
191	pr_alert("Unable to handle kernel %s at virtual address %08lx",
192		address < PAGE_SIZE ? "NULL pointer dereference" :
193		"paging request", address);
194	pr_alert("ea = %08lx, ra = %08lx, cause = %ld\n", regs->ea, regs->ra,
195		cause);
196	panic("Oops");
197	return;
198
199/*
200 * We ran out of memory, or some other thing happened to us that made
201 * us unable to handle the page fault gracefully.
202 */
203out_of_memory:
204	mmap_read_unlock(mm);
205	if (!user_mode(regs))
206		goto no_context;
207	pagefault_out_of_memory();
208	return;
209
210do_sigbus:
211	mmap_read_unlock(mm);
212
213	/* Kernel mode? Handle exceptions or die */
214	if (!user_mode(regs))
215		goto no_context;
216
217	_exception(SIGBUS, regs, BUS_ADRERR, address);
218	return;
219
220vmalloc_fault:
221	{
222		/*
223		 * Synchronize this task's top level page-table
224		 * with the 'reference' page table.
225		 *
226		 * Do _not_ use "tsk" here. We might be inside
227		 * an interrupt in the middle of a task switch..
228		 */
229		int offset = pgd_index(address);
230		pgd_t *pgd, *pgd_k;
231		p4d_t *p4d, *p4d_k;
232		pud_t *pud, *pud_k;
233		pmd_t *pmd, *pmd_k;
234		pte_t *pte_k;
235
236		pgd = pgd_current + offset;
237		pgd_k = init_mm.pgd + offset;
238
239		if (!pgd_present(*pgd_k))
240			goto no_context;
241		set_pgd(pgd, *pgd_k);
242
243		p4d = p4d_offset(pgd, address);
244		p4d_k = p4d_offset(pgd_k, address);
245		if (!p4d_present(*p4d_k))
246			goto no_context;
247		pud = pud_offset(p4d, address);
248		pud_k = pud_offset(p4d_k, address);
249		if (!pud_present(*pud_k))
250			goto no_context;
251		pmd = pmd_offset(pud, address);
252		pmd_k = pmd_offset(pud_k, address);
253		if (!pmd_present(*pmd_k))
254			goto no_context;
255		set_pmd(pmd, *pmd_k);
256
257		pte_k = pte_offset_kernel(pmd_k, address);
258		if (!pte_present(*pte_k))
259			goto no_context;
260
261		flush_tlb_kernel_page(address);
262		return;
263	}
264}