xref: /openbmc/linux/arch/parisc/mm/fault.c (revision 8dd3cdea)
1 /*
2  * This file is subject to the terms and conditions of the GNU General Public
3  * License.  See the file "COPYING" in the main directory of this archive
4  * for more details.
5  *
6  *
7  * Copyright (C) 1995, 1996, 1997, 1998 by Ralf Baechle
8  * Copyright 1999 SuSE GmbH (Philipp Rumpf, prumpf@tux.org)
9  * Copyright 1999 Hewlett Packard Co.
10  *
11  */
12 
13 #include <linux/mm.h>
14 #include <linux/ptrace.h>
15 #include <linux/sched.h>
16 #include <linux/sched/debug.h>
17 #include <linux/interrupt.h>
18 #include <linux/extable.h>
19 #include <linux/uaccess.h>
20 #include <linux/hugetlb.h>
21 #include <linux/perf_event.h>
22 
23 #include <asm/traps.h>
24 
25 /* Various important other fields */
26 #define bit22set(x)		(x & 0x00000200)
27 #define bits23_25set(x)		(x & 0x000001c0)
28 #define isGraphicsFlushRead(x)	((x & 0xfc003fdf) == 0x04001a80)
29 				/* extended opcode is 0x6a */
30 
31 #define BITSSET		0x1c0	/* for identifying LDCW */
32 
33 
34 int show_unhandled_signals = 1;
35 
36 /*
37  * parisc_acctyp(unsigned int inst) --
38  *    Given a PA-RISC memory access instruction, determine if the
39  *    the instruction would perform a memory read or memory write
40  *    operation.
41  *
42  *    This function assumes that the given instruction is a memory access
43  *    instruction (i.e. you should really only call it if you know that
44  *    the instruction has generated some sort of a memory access fault).
45  *
46  * Returns:
47  *   VM_READ  if read operation
48  *   VM_WRITE if write operation
49  *   VM_EXEC  if execute operation
50  */
51 unsigned long
52 parisc_acctyp(unsigned long code, unsigned int inst)
53 {
54 	if (code == 6 || code == 16)
55 	    return VM_EXEC;
56 
57 	switch (inst & 0xf0000000) {
58 	case 0x40000000: /* load */
59 	case 0x50000000: /* new load */
60 		return VM_READ;
61 
62 	case 0x60000000: /* store */
63 	case 0x70000000: /* new store */
64 		return VM_WRITE;
65 
66 	case 0x20000000: /* coproc */
67 	case 0x30000000: /* coproc2 */
68 		if (bit22set(inst))
69 			return VM_WRITE;
70 		fallthrough;
71 
72 	case 0x0: /* indexed/memory management */
73 		if (bit22set(inst)) {
74 			/*
75 			 * Check for the 'Graphics Flush Read' instruction.
76 			 * It resembles an FDC instruction, except for bits
77 			 * 20 and 21. Any combination other than zero will
78 			 * utilize the block mover functionality on some
79 			 * older PA-RISC platforms.  The case where a block
80 			 * move is performed from VM to graphics IO space
81 			 * should be treated as a READ.
82 			 *
83 			 * The significance of bits 20,21 in the FDC
84 			 * instruction is:
85 			 *
86 			 *   00  Flush data cache (normal instruction behavior)
87 			 *   01  Graphics flush write  (IO space -> VM)
88 			 *   10  Graphics flush read   (VM -> IO space)
89 			 *   11  Graphics flush read/write (VM <-> IO space)
90 			 */
91 			if (isGraphicsFlushRead(inst))
92 				return VM_READ;
93 			return VM_WRITE;
94 		} else {
95 			/*
96 			 * Check for LDCWX and LDCWS (semaphore instructions).
97 			 * If bits 23 through 25 are all 1's it is one of
98 			 * the above two instructions and is a write.
99 			 *
100 			 * Note: With the limited bits we are looking at,
101 			 * this will also catch PROBEW and PROBEWI. However,
102 			 * these should never get in here because they don't
103 			 * generate exceptions of the type:
104 			 *   Data TLB miss fault/data page fault
105 			 *   Data memory protection trap
106 			 */
107 			if (bits23_25set(inst) == BITSSET)
108 				return VM_WRITE;
109 		}
110 		return VM_READ; /* Default */
111 	}
112 	return VM_READ; /* Default */
113 }
114 
115 #undef bit22set
116 #undef bits23_25set
117 #undef isGraphicsFlushRead
118 #undef BITSSET
119 
120 
121 #if 0
122 /* This is the treewalk to find a vma which is the highest that has
123  * a start < addr.  We're using find_vma_prev instead right now, but
124  * we might want to use this at some point in the future.  Probably
125  * not, but I want it committed to CVS so I don't lose it :-)
126  */
127 			while (tree != vm_avl_empty) {
128 				if (tree->vm_start > addr) {
129 					tree = tree->vm_avl_left;
130 				} else {
131 					prev = tree;
132 					if (prev->vm_next == NULL)
133 						break;
134 					if (prev->vm_next->vm_start > addr)
135 						break;
136 					tree = tree->vm_avl_right;
137 				}
138 			}
139 #endif
140 
141 int fixup_exception(struct pt_regs *regs)
142 {
143 	const struct exception_table_entry *fix;
144 
145 	fix = search_exception_tables(regs->iaoq[0]);
146 	if (fix) {
147 		/*
148 		 * Fix up get_user() and put_user().
149 		 * ASM_EXCEPTIONTABLE_ENTRY_EFAULT() sets the least-significant
150 		 * bit in the relative address of the fixup routine to indicate
151 		 * that gr[ASM_EXCEPTIONTABLE_REG] should be loaded with
152 		 * -EFAULT to report a userspace access error.
153 		 */
154 		if (fix->fixup & 1) {
155 			regs->gr[ASM_EXCEPTIONTABLE_REG] = -EFAULT;
156 
157 			/* zero target register for get_user() */
158 			if (parisc_acctyp(0, regs->iir) == VM_READ) {
159 				int treg = regs->iir & 0x1f;
160 				BUG_ON(treg == 0);
161 				regs->gr[treg] = 0;
162 			}
163 		}
164 
165 		regs->iaoq[0] = (unsigned long)&fix->fixup + fix->fixup;
166 		regs->iaoq[0] &= ~3;
167 		/*
168 		 * NOTE: In some cases the faulting instruction
169 		 * may be in the delay slot of a branch. We
170 		 * don't want to take the branch, so we don't
171 		 * increment iaoq[1], instead we set it to be
172 		 * iaoq[0]+4, and clear the B bit in the PSW
173 		 */
174 		regs->iaoq[1] = regs->iaoq[0] + 4;
175 		regs->gr[0] &= ~PSW_B; /* IPSW in gr[0] */
176 
177 		return 1;
178 	}
179 
180 	return 0;
181 }
182 
183 /*
184  * parisc hardware trap list
185  *
186  * Documented in section 3 "Addressing and Access Control" of the
187  * "PA-RISC 1.1 Architecture and Instruction Set Reference Manual"
188  * https://parisc.wiki.kernel.org/index.php/File:Pa11_acd.pdf
189  *
190  * For implementation see handle_interruption() in traps.c
191  */
192 static const char * const trap_description[] = {
193 	[1] "High-priority machine check (HPMC)",
194 	[2] "Power failure interrupt",
195 	[3] "Recovery counter trap",
196 	[5] "Low-priority machine check",
197 	[6] "Instruction TLB miss fault",
198 	[7] "Instruction access rights / protection trap",
199 	[8] "Illegal instruction trap",
200 	[9] "Break instruction trap",
201 	[10] "Privileged operation trap",
202 	[11] "Privileged register trap",
203 	[12] "Overflow trap",
204 	[13] "Conditional trap",
205 	[14] "FP Assist Exception trap",
206 	[15] "Data TLB miss fault",
207 	[16] "Non-access ITLB miss fault",
208 	[17] "Non-access DTLB miss fault",
209 	[18] "Data memory protection/unaligned access trap",
210 	[19] "Data memory break trap",
211 	[20] "TLB dirty bit trap",
212 	[21] "Page reference trap",
213 	[22] "Assist emulation trap",
214 	[25] "Taken branch trap",
215 	[26] "Data memory access rights trap",
216 	[27] "Data memory protection ID trap",
217 	[28] "Unaligned data reference trap",
218 };
219 
220 const char *trap_name(unsigned long code)
221 {
222 	const char *t = NULL;
223 
224 	if (code < ARRAY_SIZE(trap_description))
225 		t = trap_description[code];
226 
227 	return t ? t : "Unknown trap";
228 }
229 
230 /*
231  * Print out info about fatal segfaults, if the show_unhandled_signals
232  * sysctl is set:
233  */
234 static inline void
235 show_signal_msg(struct pt_regs *regs, unsigned long code,
236 		unsigned long address, struct task_struct *tsk,
237 		struct vm_area_struct *vma)
238 {
239 	if (!unhandled_signal(tsk, SIGSEGV))
240 		return;
241 
242 	if (!printk_ratelimit())
243 		return;
244 
245 	pr_warn("\n");
246 	pr_warn("do_page_fault() command='%s' type=%lu address=0x%08lx",
247 	    tsk->comm, code, address);
248 	print_vma_addr(KERN_CONT " in ", regs->iaoq[0]);
249 
250 	pr_cont("\ntrap #%lu: %s%c", code, trap_name(code),
251 		vma ? ',':'\n');
252 
253 	if (vma)
254 		pr_cont(" vm_start = 0x%08lx, vm_end = 0x%08lx\n",
255 			vma->vm_start, vma->vm_end);
256 
257 	show_regs(regs);
258 }
259 
260 void do_page_fault(struct pt_regs *regs, unsigned long code,
261 			      unsigned long address)
262 {
263 	struct vm_area_struct *vma, *prev_vma;
264 	struct task_struct *tsk;
265 	struct mm_struct *mm;
266 	unsigned long acc_type;
267 	vm_fault_t fault = 0;
268 	unsigned int flags;
269 	char *msg;
270 
271 	tsk = current;
272 	mm = tsk->mm;
273 	if (!mm) {
274 		msg = "Page fault: no context";
275 		goto no_context;
276 	}
277 
278 	flags = FAULT_FLAG_DEFAULT;
279 	if (user_mode(regs))
280 		flags |= FAULT_FLAG_USER;
281 
282 	acc_type = parisc_acctyp(code, regs->iir);
283 	if (acc_type & VM_WRITE)
284 		flags |= FAULT_FLAG_WRITE;
285 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
286 retry:
287 	mmap_read_lock(mm);
288 	vma = find_vma_prev(mm, address, &prev_vma);
289 	if (!vma || address < vma->vm_start)
290 		goto check_expansion;
291 /*
292  * Ok, we have a good vm_area for this memory access. We still need to
293  * check the access permissions.
294  */
295 
296 good_area:
297 
298 	if ((vma->vm_flags & acc_type) != acc_type)
299 		goto bad_area;
300 
301 	/*
302 	 * If for any reason at all we couldn't handle the fault, make
303 	 * sure we exit gracefully rather than endlessly redo the
304 	 * fault.
305 	 */
306 
307 	fault = handle_mm_fault(vma, address, flags, regs);
308 
309 	if (fault_signal_pending(fault, regs))
310 		return;
311 
312 	if (unlikely(fault & VM_FAULT_ERROR)) {
313 		/*
314 		 * We hit a shared mapping outside of the file, or some
315 		 * other thing happened to us that made us unable to
316 		 * handle the page fault gracefully.
317 		 */
318 		if (fault & VM_FAULT_OOM)
319 			goto out_of_memory;
320 		else if (fault & VM_FAULT_SIGSEGV)
321 			goto bad_area;
322 		else if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
323 				  VM_FAULT_HWPOISON_LARGE))
324 			goto bad_area;
325 		BUG();
326 	}
327 	if (fault & VM_FAULT_RETRY) {
328 		/*
329 		 * No need to mmap_read_unlock(mm) as we would
330 		 * have already released it in __lock_page_or_retry
331 		 * in mm/filemap.c.
332 		 */
333 		flags |= FAULT_FLAG_TRIED;
334 		goto retry;
335 	}
336 	mmap_read_unlock(mm);
337 	return;
338 
339 check_expansion:
340 	vma = prev_vma;
341 	if (vma && (expand_stack(vma, address) == 0))
342 		goto good_area;
343 
344 /*
345  * Something tried to access memory that isn't in our memory map..
346  */
347 bad_area:
348 	mmap_read_unlock(mm);
349 
350 	if (user_mode(regs)) {
351 		int signo, si_code;
352 
353 		switch (code) {
354 		case 15:	/* Data TLB miss fault/Data page fault */
355 			/* send SIGSEGV when outside of vma */
356 			if (!vma ||
357 			    address < vma->vm_start || address >= vma->vm_end) {
358 				signo = SIGSEGV;
359 				si_code = SEGV_MAPERR;
360 				break;
361 			}
362 
363 			/* send SIGSEGV for wrong permissions */
364 			if ((vma->vm_flags & acc_type) != acc_type) {
365 				signo = SIGSEGV;
366 				si_code = SEGV_ACCERR;
367 				break;
368 			}
369 
370 			/* probably address is outside of mapped file */
371 			fallthrough;
372 		case 17:	/* NA data TLB miss / page fault */
373 		case 18:	/* Unaligned access - PCXS only */
374 			signo = SIGBUS;
375 			si_code = (code == 18) ? BUS_ADRALN : BUS_ADRERR;
376 			break;
377 		case 16:	/* Non-access instruction TLB miss fault */
378 		case 26:	/* PCXL: Data memory access rights trap */
379 		default:
380 			signo = SIGSEGV;
381 			si_code = (code == 26) ? SEGV_ACCERR : SEGV_MAPERR;
382 			break;
383 		}
384 #ifdef CONFIG_MEMORY_FAILURE
385 		if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
386 			unsigned int lsb = 0;
387 			printk(KERN_ERR
388 	"MCE: Killing %s:%d due to hardware memory corruption fault at %08lx\n",
389 			tsk->comm, tsk->pid, address);
390 			/*
391 			 * Either small page or large page may be poisoned.
392 			 * In other words, VM_FAULT_HWPOISON_LARGE and
393 			 * VM_FAULT_HWPOISON are mutually exclusive.
394 			 */
395 			if (fault & VM_FAULT_HWPOISON_LARGE)
396 				lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
397 			else if (fault & VM_FAULT_HWPOISON)
398 				lsb = PAGE_SHIFT;
399 
400 			force_sig_mceerr(BUS_MCEERR_AR, (void __user *) address,
401 					 lsb);
402 			return;
403 		}
404 #endif
405 		show_signal_msg(regs, code, address, tsk, vma);
406 
407 		force_sig_fault(signo, si_code, (void __user *) address);
408 		return;
409 	}
410 	msg = "Page fault: bad address";
411 
412 no_context:
413 
414 	if (!user_mode(regs) && fixup_exception(regs)) {
415 		return;
416 	}
417 
418 	parisc_terminate(msg, regs, code, address);
419 
420 out_of_memory:
421 	mmap_read_unlock(mm);
422 	if (!user_mode(regs)) {
423 		msg = "Page fault: out of memory";
424 		goto no_context;
425 	}
426 	pagefault_out_of_memory();
427 }
428 
429 /* Handle non-access data TLB miss faults.
430  *
431  * For probe instructions, accesses to userspace are considered allowed
432  * if they lie in a valid VMA and the access type matches. We are not
433  * allowed to handle MM faults here so there may be situations where an
434  * actual access would fail even though a probe was successful.
435  */
436 int
437 handle_nadtlb_fault(struct pt_regs *regs)
438 {
439 	unsigned long insn = regs->iir;
440 	int breg, treg, xreg, val = 0;
441 	struct vm_area_struct *vma, *prev_vma;
442 	struct task_struct *tsk;
443 	struct mm_struct *mm;
444 	unsigned long address;
445 	unsigned long acc_type;
446 
447 	switch (insn & 0x380) {
448 	case 0x280:
449 		/* FDC instruction */
450 		fallthrough;
451 	case 0x380:
452 		/* PDC and FIC instructions */
453 		if (printk_ratelimit()) {
454 			pr_warn("BUG: nullifying cache flush/purge instruction\n");
455 			show_regs(regs);
456 		}
457 		if (insn & 0x20) {
458 			/* Base modification */
459 			breg = (insn >> 21) & 0x1f;
460 			xreg = (insn >> 16) & 0x1f;
461 			if (breg && xreg)
462 				regs->gr[breg] += regs->gr[xreg];
463 		}
464 		regs->gr[0] |= PSW_N;
465 		return 1;
466 
467 	case 0x180:
468 		/* PROBE instruction */
469 		treg = insn & 0x1f;
470 		if (regs->isr) {
471 			tsk = current;
472 			mm = tsk->mm;
473 			if (mm) {
474 				/* Search for VMA */
475 				address = regs->ior;
476 				mmap_read_lock(mm);
477 				vma = find_vma_prev(mm, address, &prev_vma);
478 				mmap_read_unlock(mm);
479 
480 				/*
481 				 * Check if access to the VMA is okay.
482 				 * We don't allow for stack expansion.
483 				 */
484 				acc_type = (insn & 0x40) ? VM_WRITE : VM_READ;
485 				if (vma
486 				    && address >= vma->vm_start
487 				    && (vma->vm_flags & acc_type) == acc_type)
488 					val = 1;
489 			}
490 		}
491 		if (treg)
492 			regs->gr[treg] = val;
493 		regs->gr[0] |= PSW_N;
494 		return 1;
495 
496 	case 0x300:
497 		/* LPA instruction */
498 		if (insn & 0x20) {
499 			/* Base modification */
500 			breg = (insn >> 21) & 0x1f;
501 			xreg = (insn >> 16) & 0x1f;
502 			if (breg && xreg)
503 				regs->gr[breg] += regs->gr[xreg];
504 		}
505 		treg = insn & 0x1f;
506 		if (treg)
507 			regs->gr[treg] = 0;
508 		regs->gr[0] |= PSW_N;
509 		return 1;
510 
511 	default:
512 		break;
513 	}
514 
515 	return 0;
516 }
517