xref: /openbmc/linux/arch/sparc/mm/fault_64.c (revision 8d7071af)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * arch/sparc64/mm/fault.c: Page fault handlers for the 64-bit Sparc.
4  *
5  * Copyright (C) 1996, 2008 David S. Miller (davem@davemloft.net)
6  * Copyright (C) 1997, 1999 Jakub Jelinek (jj@ultra.linux.cz)
7  */
8 
9 #include <asm/head.h>
10 
11 #include <linux/string.h>
12 #include <linux/types.h>
13 #include <linux/sched.h>
14 #include <linux/sched/debug.h>
15 #include <linux/ptrace.h>
16 #include <linux/mman.h>
17 #include <linux/signal.h>
18 #include <linux/mm.h>
19 #include <linux/extable.h>
20 #include <linux/init.h>
21 #include <linux/perf_event.h>
22 #include <linux/interrupt.h>
23 #include <linux/kprobes.h>
24 #include <linux/kdebug.h>
25 #include <linux/percpu.h>
26 #include <linux/context_tracking.h>
27 #include <linux/uaccess.h>
28 
29 #include <asm/page.h>
30 #include <asm/openprom.h>
31 #include <asm/oplib.h>
32 #include <asm/asi.h>
33 #include <asm/lsu.h>
34 #include <asm/sections.h>
35 #include <asm/mmu_context.h>
36 #include <asm/setup.h>
37 
38 int show_unhandled_signals = 1;
39 
unhandled_fault(unsigned long address,struct task_struct * tsk,struct pt_regs * regs)40 static void __kprobes unhandled_fault(unsigned long address,
41 				      struct task_struct *tsk,
42 				      struct pt_regs *regs)
43 {
44 	if ((unsigned long) address < PAGE_SIZE) {
45 		printk(KERN_ALERT "Unable to handle kernel NULL "
46 		       "pointer dereference\n");
47 	} else {
48 		printk(KERN_ALERT "Unable to handle kernel paging request "
49 		       "at virtual address %016lx\n", (unsigned long)address);
50 	}
51 	printk(KERN_ALERT "tsk->{mm,active_mm}->context = %016lx\n",
52 	       (tsk->mm ?
53 		CTX_HWBITS(tsk->mm->context) :
54 		CTX_HWBITS(tsk->active_mm->context)));
55 	printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %016lx\n",
56 	       (tsk->mm ? (unsigned long) tsk->mm->pgd :
57 		          (unsigned long) tsk->active_mm->pgd));
58 	die_if_kernel("Oops", regs);
59 }
60 
bad_kernel_pc(struct pt_regs * regs,unsigned long vaddr)61 static void __kprobes bad_kernel_pc(struct pt_regs *regs, unsigned long vaddr)
62 {
63 	printk(KERN_CRIT "OOPS: Bogus kernel PC [%016lx] in fault handler\n",
64 	       regs->tpc);
65 	printk(KERN_CRIT "OOPS: RPC [%016lx]\n", regs->u_regs[15]);
66 	printk("OOPS: RPC <%pS>\n", (void *) regs->u_regs[15]);
67 	printk(KERN_CRIT "OOPS: Fault was to vaddr[%lx]\n", vaddr);
68 	dump_stack();
69 	unhandled_fault(regs->tpc, current, regs);
70 }
71 
72 /*
73  * We now make sure that mmap_lock is held in all paths that call
74  * this. Additionally, to prevent kswapd from ripping ptes from
75  * under us, raise interrupts around the time that we look at the
76  * pte, kswapd will have to wait to get his smp ipi response from
77  * us. vmtruncate likewise. This saves us having to get pte lock.
78  */
get_user_insn(unsigned long tpc)79 static unsigned int get_user_insn(unsigned long tpc)
80 {
81 	pgd_t *pgdp = pgd_offset(current->mm, tpc);
82 	p4d_t *p4dp;
83 	pud_t *pudp;
84 	pmd_t *pmdp;
85 	pte_t *ptep, pte;
86 	unsigned long pa;
87 	u32 insn = 0;
88 
89 	if (pgd_none(*pgdp) || unlikely(pgd_bad(*pgdp)))
90 		goto out;
91 	p4dp = p4d_offset(pgdp, tpc);
92 	if (p4d_none(*p4dp) || unlikely(p4d_bad(*p4dp)))
93 		goto out;
94 	pudp = pud_offset(p4dp, tpc);
95 	if (pud_none(*pudp) || unlikely(pud_bad(*pudp)))
96 		goto out;
97 
98 	/* This disables preemption for us as well. */
99 	local_irq_disable();
100 
101 	pmdp = pmd_offset(pudp, tpc);
102 again:
103 	if (pmd_none(*pmdp) || unlikely(pmd_bad(*pmdp)))
104 		goto out_irq_enable;
105 
106 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
107 	if (is_hugetlb_pmd(*pmdp)) {
108 		pa  = pmd_pfn(*pmdp) << PAGE_SHIFT;
109 		pa += tpc & ~HPAGE_MASK;
110 
111 		/* Use phys bypass so we don't pollute dtlb/dcache. */
112 		__asm__ __volatile__("lduwa [%1] %2, %0"
113 				     : "=r" (insn)
114 				     : "r" (pa), "i" (ASI_PHYS_USE_EC));
115 	} else
116 #endif
117 	{
118 		ptep = pte_offset_map(pmdp, tpc);
119 		if (!ptep)
120 			goto again;
121 		pte = *ptep;
122 		if (pte_present(pte)) {
123 			pa  = (pte_pfn(pte) << PAGE_SHIFT);
124 			pa += (tpc & ~PAGE_MASK);
125 
126 			/* Use phys bypass so we don't pollute dtlb/dcache. */
127 			__asm__ __volatile__("lduwa [%1] %2, %0"
128 					     : "=r" (insn)
129 					     : "r" (pa), "i" (ASI_PHYS_USE_EC));
130 		}
131 		pte_unmap(ptep);
132 	}
133 out_irq_enable:
134 	local_irq_enable();
135 out:
136 	return insn;
137 }
138 
139 static inline void
show_signal_msg(struct pt_regs * regs,int sig,int code,unsigned long address,struct task_struct * tsk)140 show_signal_msg(struct pt_regs *regs, int sig, int code,
141 		unsigned long address, struct task_struct *tsk)
142 {
143 	if (!unhandled_signal(tsk, sig))
144 		return;
145 
146 	if (!printk_ratelimit())
147 		return;
148 
149 	printk("%s%s[%d]: segfault at %lx ip %px (rpc %px) sp %px error %x",
150 	       task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
151 	       tsk->comm, task_pid_nr(tsk), address,
152 	       (void *)regs->tpc, (void *)regs->u_regs[UREG_I7],
153 	       (void *)regs->u_regs[UREG_FP], code);
154 
155 	print_vma_addr(KERN_CONT " in ", regs->tpc);
156 
157 	printk(KERN_CONT "\n");
158 }
159 
do_fault_siginfo(int code,int sig,struct pt_regs * regs,unsigned long fault_addr,unsigned int insn,int fault_code)160 static void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
161 			     unsigned long fault_addr, unsigned int insn,
162 			     int fault_code)
163 {
164 	unsigned long addr;
165 
166 	if (fault_code & FAULT_CODE_ITLB) {
167 		addr = regs->tpc;
168 	} else {
169 		/* If we were able to probe the faulting instruction, use it
170 		 * to compute a precise fault address.  Otherwise use the fault
171 		 * time provided address which may only have page granularity.
172 		 */
173 		if (insn)
174 			addr = compute_effective_address(regs, insn, 0);
175 		else
176 			addr = fault_addr;
177 	}
178 
179 	if (unlikely(show_unhandled_signals))
180 		show_signal_msg(regs, sig, code, addr, current);
181 
182 	force_sig_fault(sig, code, (void __user *) addr);
183 }
184 
get_fault_insn(struct pt_regs * regs,unsigned int insn)185 static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
186 {
187 	if (!insn) {
188 		if (!regs->tpc || (regs->tpc & 0x3))
189 			return 0;
190 		if (regs->tstate & TSTATE_PRIV) {
191 			insn = *(unsigned int *) regs->tpc;
192 		} else {
193 			insn = get_user_insn(regs->tpc);
194 		}
195 	}
196 	return insn;
197 }
198 
do_kernel_fault(struct pt_regs * regs,int si_code,int fault_code,unsigned int insn,unsigned long address)199 static void __kprobes do_kernel_fault(struct pt_regs *regs, int si_code,
200 				      int fault_code, unsigned int insn,
201 				      unsigned long address)
202 {
203 	unsigned char asi = ASI_P;
204 
205 	if ((!insn) && (regs->tstate & TSTATE_PRIV))
206 		goto cannot_handle;
207 
208 	/* If user insn could be read (thus insn is zero), that
209 	 * is fine.  We will just gun down the process with a signal
210 	 * in that case.
211 	 */
212 
213 	if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
214 	    (insn & 0xc0800000) == 0xc0800000) {
215 		if (insn & 0x2000)
216 			asi = (regs->tstate >> 24);
217 		else
218 			asi = (insn >> 5);
219 		if ((asi & 0xf2) == 0x82) {
220 			if (insn & 0x1000000) {
221 				handle_ldf_stq(insn, regs);
222 			} else {
223 				/* This was a non-faulting load. Just clear the
224 				 * destination register(s) and continue with the next
225 				 * instruction. -jj
226 				 */
227 				handle_ld_nf(insn, regs);
228 			}
229 			return;
230 		}
231 	}
232 
233 	/* Is this in ex_table? */
234 	if (regs->tstate & TSTATE_PRIV) {
235 		const struct exception_table_entry *entry;
236 
237 		entry = search_exception_tables(regs->tpc);
238 		if (entry) {
239 			regs->tpc = entry->fixup;
240 			regs->tnpc = regs->tpc + 4;
241 			return;
242 		}
243 	} else {
244 		/* The si_code was set to make clear whether
245 		 * this was a SEGV_MAPERR or SEGV_ACCERR fault.
246 		 */
247 		do_fault_siginfo(si_code, SIGSEGV, regs, address, insn, fault_code);
248 		return;
249 	}
250 
251 cannot_handle:
252 	unhandled_fault (address, current, regs);
253 }
254 
bogus_32bit_fault_tpc(struct pt_regs * regs)255 static void noinline __kprobes bogus_32bit_fault_tpc(struct pt_regs *regs)
256 {
257 	static int times;
258 
259 	if (times++ < 10)
260 		printk(KERN_ERR "FAULT[%s:%d]: 32-bit process reports "
261 		       "64-bit TPC [%lx]\n",
262 		       current->comm, current->pid,
263 		       regs->tpc);
264 	show_regs(regs);
265 }
266 
do_sparc64_fault(struct pt_regs * regs)267 asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
268 {
269 	enum ctx_state prev_state = exception_enter();
270 	struct mm_struct *mm = current->mm;
271 	struct vm_area_struct *vma;
272 	unsigned int insn = 0;
273 	int si_code, fault_code;
274 	vm_fault_t fault;
275 	unsigned long address, mm_rss;
276 	unsigned int flags = FAULT_FLAG_DEFAULT;
277 
278 	fault_code = get_thread_fault_code();
279 
280 	if (kprobe_page_fault(regs, 0))
281 		goto exit_exception;
282 
283 	si_code = SEGV_MAPERR;
284 	address = current_thread_info()->fault_address;
285 
286 	if ((fault_code & FAULT_CODE_ITLB) &&
287 	    (fault_code & FAULT_CODE_DTLB))
288 		BUG();
289 
290 	if (test_thread_flag(TIF_32BIT)) {
291 		if (!(regs->tstate & TSTATE_PRIV)) {
292 			if (unlikely((regs->tpc >> 32) != 0)) {
293 				bogus_32bit_fault_tpc(regs);
294 				goto intr_or_no_mm;
295 			}
296 		}
297 		if (unlikely((address >> 32) != 0))
298 			goto intr_or_no_mm;
299 	}
300 
301 	if (regs->tstate & TSTATE_PRIV) {
302 		unsigned long tpc = regs->tpc;
303 
304 		/* Sanity check the PC. */
305 		if ((tpc >= KERNBASE && tpc < (unsigned long) __init_end) ||
306 		    (tpc >= MODULES_VADDR && tpc < MODULES_END)) {
307 			/* Valid, no problems... */
308 		} else {
309 			bad_kernel_pc(regs, address);
310 			goto exit_exception;
311 		}
312 	} else
313 		flags |= FAULT_FLAG_USER;
314 
315 	/*
316 	 * If we're in an interrupt or have no user
317 	 * context, we must not take the fault..
318 	 */
319 	if (faulthandler_disabled() || !mm)
320 		goto intr_or_no_mm;
321 
322 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
323 
324 	if (!mmap_read_trylock(mm)) {
325 		if ((regs->tstate & TSTATE_PRIV) &&
326 		    !search_exception_tables(regs->tpc)) {
327 			insn = get_fault_insn(regs, insn);
328 			goto handle_kernel_fault;
329 		}
330 
331 retry:
332 		mmap_read_lock(mm);
333 	}
334 
335 	if (fault_code & FAULT_CODE_BAD_RA)
336 		goto do_sigbus;
337 
338 	vma = find_vma(mm, address);
339 	if (!vma)
340 		goto bad_area;
341 
342 	/* Pure DTLB misses do not tell us whether the fault causing
343 	 * load/store/atomic was a write or not, it only says that there
344 	 * was no match.  So in such a case we (carefully) read the
345 	 * instruction to try and figure this out.  It's an optimization
346 	 * so it's ok if we can't do this.
347 	 *
348 	 * Special hack, window spill/fill knows the exact fault type.
349 	 */
350 	if (((fault_code &
351 	      (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) &&
352 	    (vma->vm_flags & VM_WRITE) != 0) {
353 		insn = get_fault_insn(regs, 0);
354 		if (!insn)
355 			goto continue_fault;
356 		/* All loads, stores and atomics have bits 30 and 31 both set
357 		 * in the instruction.  Bit 21 is set in all stores, but we
358 		 * have to avoid prefetches which also have bit 21 set.
359 		 */
360 		if ((insn & 0xc0200000) == 0xc0200000 &&
361 		    (insn & 0x01780000) != 0x01680000) {
362 			/* Don't bother updating thread struct value,
363 			 * because update_mmu_cache only cares which tlb
364 			 * the access came from.
365 			 */
366 			fault_code |= FAULT_CODE_WRITE;
367 		}
368 	}
369 continue_fault:
370 
371 	if (vma->vm_start <= address)
372 		goto good_area;
373 	if (!(vma->vm_flags & VM_GROWSDOWN))
374 		goto bad_area;
375 	if (!(fault_code & FAULT_CODE_WRITE)) {
376 		/* Non-faulting loads shouldn't expand stack. */
377 		insn = get_fault_insn(regs, insn);
378 		if ((insn & 0xc0800000) == 0xc0800000) {
379 			unsigned char asi;
380 
381 			if (insn & 0x2000)
382 				asi = (regs->tstate >> 24);
383 			else
384 				asi = (insn >> 5);
385 			if ((asi & 0xf2) == 0x82)
386 				goto bad_area;
387 		}
388 	}
389 	vma = expand_stack(mm, address);
390 	if (!vma)
391 		goto bad_area_nosemaphore;
392 	/*
393 	 * Ok, we have a good vm_area for this memory access, so
394 	 * we can handle it..
395 	 */
396 good_area:
397 	si_code = SEGV_ACCERR;
398 
399 	/* If we took a ITLB miss on a non-executable page, catch
400 	 * that here.
401 	 */
402 	if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) {
403 		WARN(address != regs->tpc,
404 		     "address (%lx) != regs->tpc (%lx)\n", address, regs->tpc);
405 		WARN_ON(regs->tstate & TSTATE_PRIV);
406 		goto bad_area;
407 	}
408 
409 	if (fault_code & FAULT_CODE_WRITE) {
410 		if (!(vma->vm_flags & VM_WRITE))
411 			goto bad_area;
412 
413 		/* Spitfire has an icache which does not snoop
414 		 * processor stores.  Later processors do...
415 		 */
416 		if (tlb_type == spitfire &&
417 		    (vma->vm_flags & VM_EXEC) != 0 &&
418 		    vma->vm_file != NULL)
419 			set_thread_fault_code(fault_code |
420 					      FAULT_CODE_BLKCOMMIT);
421 
422 		flags |= FAULT_FLAG_WRITE;
423 	} else {
424 		/* Allow reads even for write-only mappings */
425 		if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
426 			goto bad_area;
427 	}
428 
429 	fault = handle_mm_fault(vma, address, flags, regs);
430 
431 	if (fault_signal_pending(fault, regs)) {
432 		if (regs->tstate & TSTATE_PRIV) {
433 			insn = get_fault_insn(regs, insn);
434 			goto handle_kernel_fault;
435 		}
436 		goto exit_exception;
437 	}
438 
439 	/* The fault is fully completed (including releasing mmap lock) */
440 	if (fault & VM_FAULT_COMPLETED)
441 		goto lock_released;
442 
443 	if (unlikely(fault & VM_FAULT_ERROR)) {
444 		if (fault & VM_FAULT_OOM)
445 			goto out_of_memory;
446 		else if (fault & VM_FAULT_SIGSEGV)
447 			goto bad_area;
448 		else if (fault & VM_FAULT_SIGBUS)
449 			goto do_sigbus;
450 		BUG();
451 	}
452 
453 	if (fault & VM_FAULT_RETRY) {
454 		flags |= FAULT_FLAG_TRIED;
455 
456 		/* No need to mmap_read_unlock(mm) as we would
457 		 * have already released it in __lock_page_or_retry
458 		 * in mm/filemap.c.
459 		 */
460 
461 		goto retry;
462 	}
463 	mmap_read_unlock(mm);
464 
465 lock_released:
466 	mm_rss = get_mm_rss(mm);
467 #if defined(CONFIG_TRANSPARENT_HUGEPAGE)
468 	mm_rss -= (mm->context.thp_pte_count * (HPAGE_SIZE / PAGE_SIZE));
469 #endif
470 	if (unlikely(mm_rss >
471 		     mm->context.tsb_block[MM_TSB_BASE].tsb_rss_limit))
472 		tsb_grow(mm, MM_TSB_BASE, mm_rss);
473 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
474 	mm_rss = mm->context.hugetlb_pte_count + mm->context.thp_pte_count;
475 	mm_rss *= REAL_HPAGE_PER_HPAGE;
476 	if (unlikely(mm_rss >
477 		     mm->context.tsb_block[MM_TSB_HUGE].tsb_rss_limit)) {
478 		if (mm->context.tsb_block[MM_TSB_HUGE].tsb)
479 			tsb_grow(mm, MM_TSB_HUGE, mm_rss);
480 		else
481 			hugetlb_setup(regs);
482 
483 	}
484 #endif
485 exit_exception:
486 	exception_exit(prev_state);
487 	return;
488 
489 	/*
490 	 * Something tried to access memory that isn't in our memory map..
491 	 * Fix it, but check if it's kernel or user first..
492 	 */
493 bad_area:
494 	mmap_read_unlock(mm);
495 bad_area_nosemaphore:
496 	insn = get_fault_insn(regs, insn);
497 
498 handle_kernel_fault:
499 	do_kernel_fault(regs, si_code, fault_code, insn, address);
500 	goto exit_exception;
501 
502 /*
503  * We ran out of memory, or some other thing happened to us that made
504  * us unable to handle the page fault gracefully.
505  */
506 out_of_memory:
507 	insn = get_fault_insn(regs, insn);
508 	mmap_read_unlock(mm);
509 	if (!(regs->tstate & TSTATE_PRIV)) {
510 		pagefault_out_of_memory();
511 		goto exit_exception;
512 	}
513 	goto handle_kernel_fault;
514 
515 intr_or_no_mm:
516 	insn = get_fault_insn(regs, 0);
517 	goto handle_kernel_fault;
518 
519 do_sigbus:
520 	insn = get_fault_insn(regs, insn);
521 	mmap_read_unlock(mm);
522 
523 	/*
524 	 * Send a sigbus, regardless of whether we were in kernel
525 	 * or user mode.
526 	 */
527 	do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, address, insn, fault_code);
528 
529 	/* Kernel mode? Handle exceptions or die */
530 	if (regs->tstate & TSTATE_PRIV)
531 		goto handle_kernel_fault;
532 }
533