xref: /openbmc/linux/arch/sparc/mm/fault_64.c (revision b830f94f) 
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/pgtable.h>
31 #include <asm/openprom.h>
32 #include <asm/oplib.h>
33 #include <asm/asi.h>
34 #include <asm/lsu.h>
35 #include <asm/sections.h>
36 #include <asm/mmu_context.h>
37 #include <asm/setup.h>
38 
39 int show_unhandled_signals = 1;
40 
41 static void __kprobes unhandled_fault(unsigned long address,
42 				      struct task_struct *tsk,
43 				      struct pt_regs *regs)
44 {
45 	if ((unsigned long) address < PAGE_SIZE) {
46 		printk(KERN_ALERT "Unable to handle kernel NULL "
47 		       "pointer dereference\n");
48 	} else {
49 		printk(KERN_ALERT "Unable to handle kernel paging request "
50 		       "at virtual address %016lx\n", (unsigned long)address);
51 	}
52 	printk(KERN_ALERT "tsk->{mm,active_mm}->context = %016lx\n",
53 	       (tsk->mm ?
54 		CTX_HWBITS(tsk->mm->context) :
55 		CTX_HWBITS(tsk->active_mm->context)));
56 	printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %016lx\n",
57 	       (tsk->mm ? (unsigned long) tsk->mm->pgd :
58 		          (unsigned long) tsk->active_mm->pgd));
59 	die_if_kernel("Oops", regs);
60 }
61 
62 static void __kprobes bad_kernel_pc(struct pt_regs *regs, unsigned long vaddr)
63 {
64 	printk(KERN_CRIT "OOPS: Bogus kernel PC [%016lx] in fault handler\n",
65 	       regs->tpc);
66 	printk(KERN_CRIT "OOPS: RPC [%016lx]\n", regs->u_regs[15]);
67 	printk("OOPS: RPC <%pS>\n", (void *) regs->u_regs[15]);
68 	printk(KERN_CRIT "OOPS: Fault was to vaddr[%lx]\n", vaddr);
69 	dump_stack();
70 	unhandled_fault(regs->tpc, current, regs);
71 }
72 
73 /*
74  * We now make sure that mmap_sem is held in all paths that call
75  * this. Additionally, to prevent kswapd from ripping ptes from
76  * under us, raise interrupts around the time that we look at the
77  * pte, kswapd will have to wait to get his smp ipi response from
78  * us. vmtruncate likewise. This saves us having to get pte lock.
79  */
80 static unsigned int get_user_insn(unsigned long tpc)
81 {
82 	pgd_t *pgdp = pgd_offset(current->mm, tpc);
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 	pudp = pud_offset(pgdp, tpc);
92 	if (pud_none(*pudp) || unlikely(pud_bad(*pudp)))
93 		goto out;
94 
95 	/* This disables preemption for us as well. */
96 	local_irq_disable();
97 
98 	pmdp = pmd_offset(pudp, tpc);
99 	if (pmd_none(*pmdp) || unlikely(pmd_bad(*pmdp)))
100 		goto out_irq_enable;
101 
102 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
103 	if (is_hugetlb_pmd(*pmdp)) {
104 		pa  = pmd_pfn(*pmdp) << PAGE_SHIFT;
105 		pa += tpc & ~HPAGE_MASK;
106 
107 		/* Use phys bypass so we don't pollute dtlb/dcache. */
108 		__asm__ __volatile__("lduwa [%1] %2, %0"
109 				     : "=r" (insn)
110 				     : "r" (pa), "i" (ASI_PHYS_USE_EC));
111 	} else
112 #endif
113 	{
114 		ptep = pte_offset_map(pmdp, tpc);
115 		pte = *ptep;
116 		if (pte_present(pte)) {
117 			pa  = (pte_pfn(pte) << PAGE_SHIFT);
118 			pa += (tpc & ~PAGE_MASK);
119 
120 			/* Use phys bypass so we don't pollute dtlb/dcache. */
121 			__asm__ __volatile__("lduwa [%1] %2, %0"
122 					     : "=r" (insn)
123 					     : "r" (pa), "i" (ASI_PHYS_USE_EC));
124 		}
125 		pte_unmap(ptep);
126 	}
127 out_irq_enable:
128 	local_irq_enable();
129 out:
130 	return insn;
131 }
132 
133 static inline void
134 show_signal_msg(struct pt_regs *regs, int sig, int code,
135 		unsigned long address, struct task_struct *tsk)
136 {
137 	if (!unhandled_signal(tsk, sig))
138 		return;
139 
140 	if (!printk_ratelimit())
141 		return;
142 
143 	printk("%s%s[%d]: segfault at %lx ip %px (rpc %px) sp %px error %x",
144 	       task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
145 	       tsk->comm, task_pid_nr(tsk), address,
146 	       (void *)regs->tpc, (void *)regs->u_regs[UREG_I7],
147 	       (void *)regs->u_regs[UREG_FP], code);
148 
149 	print_vma_addr(KERN_CONT " in ", regs->tpc);
150 
151 	printk(KERN_CONT "\n");
152 }
153 
154 static void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
155 			     unsigned long fault_addr, unsigned int insn,
156 			     int fault_code)
157 {
158 	unsigned long addr;
159 
160 	if (fault_code & FAULT_CODE_ITLB) {
161 		addr = regs->tpc;
162 	} else {
163 		/* If we were able to probe the faulting instruction, use it
164 		 * to compute a precise fault address.  Otherwise use the fault
165 		 * time provided address which may only have page granularity.
166 		 */
167 		if (insn)
168 			addr = compute_effective_address(regs, insn, 0);
169 		else
170 			addr = fault_addr;
171 	}
172 
173 	if (unlikely(show_unhandled_signals))
174 		show_signal_msg(regs, sig, code, addr, current);
175 
176 	force_sig_fault(sig, code, (void __user *) addr, 0);
177 }
178 
179 static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
180 {
181 	if (!insn) {
182 		if (!regs->tpc || (regs->tpc & 0x3))
183 			return 0;
184 		if (regs->tstate & TSTATE_PRIV) {
185 			insn = *(unsigned int *) regs->tpc;
186 		} else {
187 			insn = get_user_insn(regs->tpc);
188 		}
189 	}
190 	return insn;
191 }
192 
193 static void __kprobes do_kernel_fault(struct pt_regs *regs, int si_code,
194 				      int fault_code, unsigned int insn,
195 				      unsigned long address)
196 {
197 	unsigned char asi = ASI_P;
198 
199 	if ((!insn) && (regs->tstate & TSTATE_PRIV))
200 		goto cannot_handle;
201 
202 	/* If user insn could be read (thus insn is zero), that
203 	 * is fine.  We will just gun down the process with a signal
204 	 * in that case.
205 	 */
206 
207 	if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
208 	    (insn & 0xc0800000) == 0xc0800000) {
209 		if (insn & 0x2000)
210 			asi = (regs->tstate >> 24);
211 		else
212 			asi = (insn >> 5);
213 		if ((asi & 0xf2) == 0x82) {
214 			if (insn & 0x1000000) {
215 				handle_ldf_stq(insn, regs);
216 			} else {
217 				/* This was a non-faulting load. Just clear the
218 				 * destination register(s) and continue with the next
219 				 * instruction. -jj
220 				 */
221 				handle_ld_nf(insn, regs);
222 			}
223 			return;
224 		}
225 	}
226 
227 	/* Is this in ex_table? */
228 	if (regs->tstate & TSTATE_PRIV) {
229 		const struct exception_table_entry *entry;
230 
231 		entry = search_exception_tables(regs->tpc);
232 		if (entry) {
233 			regs->tpc = entry->fixup;
234 			regs->tnpc = regs->tpc + 4;
235 			return;
236 		}
237 	} else {
238 		/* The si_code was set to make clear whether
239 		 * this was a SEGV_MAPERR or SEGV_ACCERR fault.
240 		 */
241 		do_fault_siginfo(si_code, SIGSEGV, regs, address, insn, fault_code);
242 		return;
243 	}
244 
245 cannot_handle:
246 	unhandled_fault (address, current, regs);
247 }
248 
249 static void noinline __kprobes bogus_32bit_fault_tpc(struct pt_regs *regs)
250 {
251 	static int times;
252 
253 	if (times++ < 10)
254 		printk(KERN_ERR "FAULT[%s:%d]: 32-bit process reports "
255 		       "64-bit TPC [%lx]\n",
256 		       current->comm, current->pid,
257 		       regs->tpc);
258 	show_regs(regs);
259 }
260 
261 asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
262 {
263 	enum ctx_state prev_state = exception_enter();
264 	struct mm_struct *mm = current->mm;
265 	struct vm_area_struct *vma;
266 	unsigned int insn = 0;
267 	int si_code, fault_code;
268 	vm_fault_t fault;
269 	unsigned long address, mm_rss;
270 	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
271 
272 	fault_code = get_thread_fault_code();
273 
274 	if (kprobe_page_fault(regs, 0))
275 		goto exit_exception;
276 
277 	si_code = SEGV_MAPERR;
278 	address = current_thread_info()->fault_address;
279 
280 	if ((fault_code & FAULT_CODE_ITLB) &&
281 	    (fault_code & FAULT_CODE_DTLB))
282 		BUG();
283 
284 	if (test_thread_flag(TIF_32BIT)) {
285 		if (!(regs->tstate & TSTATE_PRIV)) {
286 			if (unlikely((regs->tpc >> 32) != 0)) {
287 				bogus_32bit_fault_tpc(regs);
288 				goto intr_or_no_mm;
289 			}
290 		}
291 		if (unlikely((address >> 32) != 0))
292 			goto intr_or_no_mm;
293 	}
294 
295 	if (regs->tstate & TSTATE_PRIV) {
296 		unsigned long tpc = regs->tpc;
297 
298 		/* Sanity check the PC. */
299 		if ((tpc >= KERNBASE && tpc < (unsigned long) __init_end) ||
300 		    (tpc >= MODULES_VADDR && tpc < MODULES_END)) {
301 			/* Valid, no problems... */
302 		} else {
303 			bad_kernel_pc(regs, address);
304 			goto exit_exception;
305 		}
306 	} else
307 		flags |= FAULT_FLAG_USER;
308 
309 	/*
310 	 * If we're in an interrupt or have no user
311 	 * context, we must not take the fault..
312 	 */
313 	if (faulthandler_disabled() || !mm)
314 		goto intr_or_no_mm;
315 
316 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
317 
318 	if (!down_read_trylock(&mm->mmap_sem)) {
319 		if ((regs->tstate & TSTATE_PRIV) &&
320 		    !search_exception_tables(regs->tpc)) {
321 			insn = get_fault_insn(regs, insn);
322 			goto handle_kernel_fault;
323 		}
324 
325 retry:
326 		down_read(&mm->mmap_sem);
327 	}
328 
329 	if (fault_code & FAULT_CODE_BAD_RA)
330 		goto do_sigbus;
331 
332 	vma = find_vma(mm, address);
333 	if (!vma)
334 		goto bad_area;
335 
336 	/* Pure DTLB misses do not tell us whether the fault causing
337 	 * load/store/atomic was a write or not, it only says that there
338 	 * was no match.  So in such a case we (carefully) read the
339 	 * instruction to try and figure this out.  It's an optimization
340 	 * so it's ok if we can't do this.
341 	 *
342 	 * Special hack, window spill/fill knows the exact fault type.
343 	 */
344 	if (((fault_code &
345 	      (FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) &&
346 	    (vma->vm_flags & VM_WRITE) != 0) {
347 		insn = get_fault_insn(regs, 0);
348 		if (!insn)
349 			goto continue_fault;
350 		/* All loads, stores and atomics have bits 30 and 31 both set
351 		 * in the instruction.  Bit 21 is set in all stores, but we
352 		 * have to avoid prefetches which also have bit 21 set.
353 		 */
354 		if ((insn & 0xc0200000) == 0xc0200000 &&
355 		    (insn & 0x01780000) != 0x01680000) {
356 			/* Don't bother updating thread struct value,
357 			 * because update_mmu_cache only cares which tlb
358 			 * the access came from.
359 			 */
360 			fault_code |= FAULT_CODE_WRITE;
361 		}
362 	}
363 continue_fault:
364 
365 	if (vma->vm_start <= address)
366 		goto good_area;
367 	if (!(vma->vm_flags & VM_GROWSDOWN))
368 		goto bad_area;
369 	if (!(fault_code & FAULT_CODE_WRITE)) {
370 		/* Non-faulting loads shouldn't expand stack. */
371 		insn = get_fault_insn(regs, insn);
372 		if ((insn & 0xc0800000) == 0xc0800000) {
373 			unsigned char asi;
374 
375 			if (insn & 0x2000)
376 				asi = (regs->tstate >> 24);
377 			else
378 				asi = (insn >> 5);
379 			if ((asi & 0xf2) == 0x82)
380 				goto bad_area;
381 		}
382 	}
383 	if (expand_stack(vma, address))
384 		goto bad_area;
385 	/*
386 	 * Ok, we have a good vm_area for this memory access, so
387 	 * we can handle it..
388 	 */
389 good_area:
390 	si_code = SEGV_ACCERR;
391 
392 	/* If we took a ITLB miss on a non-executable page, catch
393 	 * that here.
394 	 */
395 	if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) {
396 		WARN(address != regs->tpc,
397 		     "address (%lx) != regs->tpc (%lx)\n", address, regs->tpc);
398 		WARN_ON(regs->tstate & TSTATE_PRIV);
399 		goto bad_area;
400 	}
401 
402 	if (fault_code & FAULT_CODE_WRITE) {
403 		if (!(vma->vm_flags & VM_WRITE))
404 			goto bad_area;
405 
406 		/* Spitfire has an icache which does not snoop
407 		 * processor stores.  Later processors do...
408 		 */
409 		if (tlb_type == spitfire &&
410 		    (vma->vm_flags & VM_EXEC) != 0 &&
411 		    vma->vm_file != NULL)
412 			set_thread_fault_code(fault_code |
413 					      FAULT_CODE_BLKCOMMIT);
414 
415 		flags |= FAULT_FLAG_WRITE;
416 	} else {
417 		/* Allow reads even for write-only mappings */
418 		if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
419 			goto bad_area;
420 	}
421 
422 	fault = handle_mm_fault(vma, address, flags);
423 
424 	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
425 		goto exit_exception;
426 
427 	if (unlikely(fault & VM_FAULT_ERROR)) {
428 		if (fault & VM_FAULT_OOM)
429 			goto out_of_memory;
430 		else if (fault & VM_FAULT_SIGSEGV)
431 			goto bad_area;
432 		else if (fault & VM_FAULT_SIGBUS)
433 			goto do_sigbus;
434 		BUG();
435 	}
436 
437 	if (flags & FAULT_FLAG_ALLOW_RETRY) {
438 		if (fault & VM_FAULT_MAJOR) {
439 			current->maj_flt++;
440 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ,
441 				      1, regs, address);
442 		} else {
443 			current->min_flt++;
444 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN,
445 				      1, regs, address);
446 		}
447 		if (fault & VM_FAULT_RETRY) {
448 			flags &= ~FAULT_FLAG_ALLOW_RETRY;
449 			flags |= FAULT_FLAG_TRIED;
450 
451 			/* No need to up_read(&mm->mmap_sem) as we would
452 			 * have already released it in __lock_page_or_retry
453 			 * in mm/filemap.c.
454 			 */
455 
456 			goto retry;
457 		}
458 	}
459 	up_read(&mm->mmap_sem);
460 
461 	mm_rss = get_mm_rss(mm);
462 #if defined(CONFIG_TRANSPARENT_HUGEPAGE)
463 	mm_rss -= (mm->context.thp_pte_count * (HPAGE_SIZE / PAGE_SIZE));
464 #endif
465 	if (unlikely(mm_rss >
466 		     mm->context.tsb_block[MM_TSB_BASE].tsb_rss_limit))
467 		tsb_grow(mm, MM_TSB_BASE, mm_rss);
468 #if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
469 	mm_rss = mm->context.hugetlb_pte_count + mm->context.thp_pte_count;
470 	mm_rss *= REAL_HPAGE_PER_HPAGE;
471 	if (unlikely(mm_rss >
472 		     mm->context.tsb_block[MM_TSB_HUGE].tsb_rss_limit)) {
473 		if (mm->context.tsb_block[MM_TSB_HUGE].tsb)
474 			tsb_grow(mm, MM_TSB_HUGE, mm_rss);
475 		else
476 			hugetlb_setup(regs);
477 
478 	}
479 #endif
480 exit_exception:
481 	exception_exit(prev_state);
482 	return;
483 
484 	/*
485 	 * Something tried to access memory that isn't in our memory map..
486 	 * Fix it, but check if it's kernel or user first..
487 	 */
488 bad_area:
489 	insn = get_fault_insn(regs, insn);
490 	up_read(&mm->mmap_sem);
491 
492 handle_kernel_fault:
493 	do_kernel_fault(regs, si_code, fault_code, insn, address);
494 	goto exit_exception;
495 
496 /*
497  * We ran out of memory, or some other thing happened to us that made
498  * us unable to handle the page fault gracefully.
499  */
500 out_of_memory:
501 	insn = get_fault_insn(regs, insn);
502 	up_read(&mm->mmap_sem);
503 	if (!(regs->tstate & TSTATE_PRIV)) {
504 		pagefault_out_of_memory();
505 		goto exit_exception;
506 	}
507 	goto handle_kernel_fault;
508 
509 intr_or_no_mm:
510 	insn = get_fault_insn(regs, 0);
511 	goto handle_kernel_fault;
512 
513 do_sigbus:
514 	insn = get_fault_insn(regs, insn);
515 	up_read(&mm->mmap_sem);
516 
517 	/*
518 	 * Send a sigbus, regardless of whether we were in kernel
519 	 * or user mode.
520 	 */
521 	do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, address, insn, fault_code);
522 
523 	/* Kernel mode? Handle exceptions or die */
524 	if (regs->tstate & TSTATE_PRIV)
525 		goto handle_kernel_fault;
526 }
527