xref: /openbmc/linux/arch/microblaze/mm/fault.c (revision 5d7800d9)
1 /*
2  *  arch/microblaze/mm/fault.c
3  *
4  *    Copyright (C) 2007 Xilinx, Inc.  All rights reserved.
5  *
6  *  Derived from "arch/ppc/mm/fault.c"
7  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
8  *
9  *  Derived from "arch/i386/mm/fault.c"
10  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
11  *
12  *  Modified by Cort Dougan and Paul Mackerras.
13  *
14  * This file is subject to the terms and conditions of the GNU General
15  * Public License.  See the file COPYING in the main directory of this
16  * archive for more details.
17  *
18  */
19 
20 #include <linux/extable.h>
21 #include <linux/signal.h>
22 #include <linux/sched.h>
23 #include <linux/kernel.h>
24 #include <linux/errno.h>
25 #include <linux/string.h>
26 #include <linux/types.h>
27 #include <linux/ptrace.h>
28 #include <linux/mman.h>
29 #include <linux/mm.h>
30 #include <linux/interrupt.h>
31 #include <linux/perf_event.h>
32 
33 #include <asm/page.h>
34 #include <asm/mmu.h>
35 #include <linux/mmu_context.h>
36 #include <linux/uaccess.h>
37 #include <asm/exceptions.h>
38 
39 static unsigned long pte_misses;	/* updated by do_page_fault() */
40 static unsigned long pte_errors;	/* updated by do_page_fault() */
41 
42 /*
43  * Check whether the instruction at regs->pc is a store using
44  * an update addressing form which will update r1.
45  */
46 static int store_updates_sp(struct pt_regs *regs)
47 {
48 	unsigned int inst;
49 
50 	if (get_user(inst, (unsigned int __user *)regs->pc))
51 		return 0;
52 	/* check for 1 in the rD field */
53 	if (((inst >> 21) & 0x1f) != 1)
54 		return 0;
55 	/* check for store opcodes */
56 	if ((inst & 0xd0000000) == 0xd0000000)
57 		return 1;
58 	return 0;
59 }
60 
61 
62 /*
63  * bad_page_fault is called when we have a bad access from the kernel.
64  * It is called from do_page_fault above and from some of the procedures
65  * in traps.c.
66  */
67 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
68 {
69 	const struct exception_table_entry *fixup;
70 /* MS: no context */
71 	/* Are we prepared to handle this fault?  */
72 	fixup = search_exception_tables(regs->pc);
73 	if (fixup) {
74 		regs->pc = fixup->fixup;
75 		return;
76 	}
77 
78 	/* kernel has accessed a bad area */
79 	die("kernel access of bad area", regs, sig);
80 }
81 
82 /*
83  * The error_code parameter is ESR for a data fault,
84  * 0 for an instruction fault.
85  */
86 void do_page_fault(struct pt_regs *regs, unsigned long address,
87 		   unsigned long error_code)
88 {
89 	struct vm_area_struct *vma;
90 	struct mm_struct *mm = current->mm;
91 	int code = SEGV_MAPERR;
92 	int is_write = error_code & ESR_S;
93 	vm_fault_t fault;
94 	unsigned int flags = FAULT_FLAG_DEFAULT;
95 
96 	regs->ear = address;
97 	regs->esr = error_code;
98 
99 	/* On a kernel SLB miss we can only check for a valid exception entry */
100 	if (unlikely(kernel_mode(regs) && (address >= TASK_SIZE))) {
101 		pr_warn("kernel task_size exceed");
102 		_exception(SIGSEGV, regs, code, address);
103 	}
104 
105 	/* for instr TLB miss and instr storage exception ESR_S is undefined */
106 	if ((error_code & 0x13) == 0x13 || (error_code & 0x11) == 0x11)
107 		is_write = 0;
108 
109 	if (unlikely(faulthandler_disabled() || !mm)) {
110 		if (kernel_mode(regs))
111 			goto bad_area_nosemaphore;
112 
113 		/* faulthandler_disabled() in user mode is really bad,
114 		   as is current->mm == NULL. */
115 		pr_emerg("Page fault in user mode with faulthandler_disabled(), mm = %p\n",
116 			 mm);
117 		pr_emerg("r15 = %lx  MSR = %lx\n",
118 		       regs->r15, regs->msr);
119 		die("Weird page fault", regs, SIGSEGV);
120 	}
121 
122 	if (user_mode(regs))
123 		flags |= FAULT_FLAG_USER;
124 
125 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
126 
127 	/* When running in the kernel we expect faults to occur only to
128 	 * addresses in user space.  All other faults represent errors in the
129 	 * kernel and should generate an OOPS.  Unfortunately, in the case of an
130 	 * erroneous fault occurring in a code path which already holds mmap_lock
131 	 * we will deadlock attempting to validate the fault against the
132 	 * address space.  Luckily the kernel only validly references user
133 	 * space from well defined areas of code, which are listed in the
134 	 * exceptions table.
135 	 *
136 	 * As the vast majority of faults will be valid we will only perform
137 	 * the source reference check when there is a possibility of a deadlock.
138 	 * Attempt to lock the address space, if we cannot we then validate the
139 	 * source.  If this is invalid we can skip the address space check,
140 	 * thus avoiding the deadlock.
141 	 */
142 	if (unlikely(!mmap_read_trylock(mm))) {
143 		if (kernel_mode(regs) && !search_exception_tables(regs->pc))
144 			goto bad_area_nosemaphore;
145 
146 retry:
147 		mmap_read_lock(mm);
148 	}
149 
150 	vma = find_vma(mm, address);
151 	if (unlikely(!vma))
152 		goto bad_area;
153 
154 	if (vma->vm_start <= address)
155 		goto good_area;
156 
157 	if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
158 		goto bad_area;
159 
160 	if (unlikely(!is_write))
161 		goto bad_area;
162 
163 	/*
164 	 * N.B. The ABI allows programs to access up to
165 	 * a few hundred bytes below the stack pointer (TBD).
166 	 * The kernel signal delivery code writes up to about 1.5kB
167 	 * below the stack pointer (r1) before decrementing it.
168 	 * The exec code can write slightly over 640kB to the stack
169 	 * before setting the user r1.  Thus we allow the stack to
170 	 * expand to 1MB without further checks.
171 	 */
172 	if (unlikely(address + 0x100000 < vma->vm_end)) {
173 
174 		/* get user regs even if this fault is in kernel mode */
175 		struct pt_regs *uregs = current->thread.regs;
176 		if (uregs == NULL)
177 			goto bad_area;
178 
179 		/*
180 		 * A user-mode access to an address a long way below
181 		 * the stack pointer is only valid if the instruction
182 		 * is one which would update the stack pointer to the
183 		 * address accessed if the instruction completed,
184 		 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
185 		 * (or the byte, halfword, float or double forms).
186 		 *
187 		 * If we don't check this then any write to the area
188 		 * between the last mapped region and the stack will
189 		 * expand the stack rather than segfaulting.
190 		 */
191 		if (address + 2048 < uregs->r1
192 			&& (kernel_mode(regs) || !store_updates_sp(regs)))
193 				goto bad_area;
194 	}
195 	vma = expand_stack(mm, address);
196 	if (!vma)
197 		goto bad_area_nosemaphore;
198 
199 good_area:
200 	code = SEGV_ACCERR;
201 
202 	/* a write */
203 	if (unlikely(is_write)) {
204 		if (unlikely(!(vma->vm_flags & VM_WRITE)))
205 			goto bad_area;
206 		flags |= FAULT_FLAG_WRITE;
207 	/* a read */
208 	} else {
209 		/* protection fault */
210 		if (unlikely(error_code & 0x08000000))
211 			goto bad_area;
212 		if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC))))
213 			goto bad_area;
214 	}
215 
216 	/*
217 	 * If for any reason at all we couldn't handle the fault,
218 	 * make sure we exit gracefully rather than endlessly redo
219 	 * the fault.
220 	 */
221 	fault = handle_mm_fault(vma, address, flags, regs);
222 
223 	if (fault_signal_pending(fault, regs)) {
224 		if (!user_mode(regs))
225 			bad_page_fault(regs, address, SIGBUS);
226 		return;
227 	}
228 
229 	/* The fault is fully completed (including releasing mmap lock) */
230 	if (fault & VM_FAULT_COMPLETED)
231 		return;
232 
233 	if (unlikely(fault & VM_FAULT_ERROR)) {
234 		if (fault & VM_FAULT_OOM)
235 			goto out_of_memory;
236 		else if (fault & VM_FAULT_SIGSEGV)
237 			goto bad_area;
238 		else if (fault & VM_FAULT_SIGBUS)
239 			goto do_sigbus;
240 		BUG();
241 	}
242 
243 	if (fault & VM_FAULT_RETRY) {
244 		flags |= FAULT_FLAG_TRIED;
245 
246 		/*
247 		 * No need to mmap_read_unlock(mm) as we would
248 		 * have already released it in __lock_page_or_retry
249 		 * in mm/filemap.c.
250 		 */
251 
252 		goto retry;
253 	}
254 
255 	mmap_read_unlock(mm);
256 
257 	/*
258 	 * keep track of tlb+htab misses that are good addrs but
259 	 * just need pte's created via handle_mm_fault()
260 	 * -- Cort
261 	 */
262 	pte_misses++;
263 	return;
264 
265 bad_area:
266 	mmap_read_unlock(mm);
267 
268 bad_area_nosemaphore:
269 	pte_errors++;
270 
271 	/* User mode accesses cause a SIGSEGV */
272 	if (user_mode(regs)) {
273 		_exception(SIGSEGV, regs, code, address);
274 		return;
275 	}
276 
277 	bad_page_fault(regs, address, SIGSEGV);
278 	return;
279 
280 /*
281  * We ran out of memory, or some other thing happened to us that made
282  * us unable to handle the page fault gracefully.
283  */
284 out_of_memory:
285 	mmap_read_unlock(mm);
286 	if (!user_mode(regs))
287 		bad_page_fault(regs, address, SIGKILL);
288 	else
289 		pagefault_out_of_memory();
290 	return;
291 
292 do_sigbus:
293 	mmap_read_unlock(mm);
294 	if (user_mode(regs)) {
295 		force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
296 		return;
297 	}
298 	bad_page_fault(regs, address, SIGBUS);
299 }
300