xref: /linux/arch/um/kernel/tlb.c (revision 4eb7ae7a301d3586c3351e81d5c3cfe2304a1a6a) 
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
4  */
5 
6 #include <linux/mm.h>
7 #include <linux/module.h>
8 #include <linux/sched/signal.h>
9 
10 #include <asm/pgtable.h>
11 #include <asm/tlbflush.h>
12 #include <as-layout.h>
13 #include <mem_user.h>
14 #include <os.h>
15 #include <skas.h>
16 #include <kern_util.h>
17 
18 struct host_vm_change {
19 	struct host_vm_op {
20 		enum { NONE, MMAP, MUNMAP, MPROTECT } type;
21 		union {
22 			struct {
23 				unsigned long addr;
24 				unsigned long len;
25 				unsigned int prot;
26 				int fd;
27 				__u64 offset;
28 			} mmap;
29 			struct {
30 				unsigned long addr;
31 				unsigned long len;
32 			} munmap;
33 			struct {
34 				unsigned long addr;
35 				unsigned long len;
36 				unsigned int prot;
37 			} mprotect;
38 		} u;
39 	} ops[1];
40 	int userspace;
41 	int index;
42 	struct mm_struct *mm;
43 	void *data;
44 	int force;
45 };
46 
47 #define INIT_HVC(mm, force, userspace) \
48 	((struct host_vm_change) \
49 	 { .ops		= { { .type = NONE } },	\
50 	   .mm		= mm, \
51        	   .data	= NULL, \
52 	   .userspace	= userspace, \
53 	   .index	= 0, \
54 	   .force	= force })
55 
56 static void report_enomem(void)
57 {
58 	printk(KERN_ERR "UML ran out of memory on the host side! "
59 			"This can happen due to a memory limitation or "
60 			"vm.max_map_count has been reached.\n");
61 }
62 
63 static int do_ops(struct host_vm_change *hvc, int end,
64 		  int finished)
65 {
66 	struct host_vm_op *op;
67 	int i, ret = 0;
68 
69 	for (i = 0; i < end && !ret; i++) {
70 		op = &hvc->ops[i];
71 		switch (op->type) {
72 		case MMAP:
73 			if (hvc->userspace)
74 				ret = map(&hvc->mm->context.id, op->u.mmap.addr,
75 					  op->u.mmap.len, op->u.mmap.prot,
76 					  op->u.mmap.fd,
77 					  op->u.mmap.offset, finished,
78 					  &hvc->data);
79 			else
80 				map_memory(op->u.mmap.addr, op->u.mmap.offset,
81 					   op->u.mmap.len, 1, 1, 1);
82 			break;
83 		case MUNMAP:
84 			if (hvc->userspace)
85 				ret = unmap(&hvc->mm->context.id,
86 					    op->u.munmap.addr,
87 					    op->u.munmap.len, finished,
88 					    &hvc->data);
89 			else
90 				ret = os_unmap_memory(
91 					(void *) op->u.munmap.addr,
92 						      op->u.munmap.len);
93 
94 			break;
95 		case MPROTECT:
96 			if (hvc->userspace)
97 				ret = protect(&hvc->mm->context.id,
98 					      op->u.mprotect.addr,
99 					      op->u.mprotect.len,
100 					      op->u.mprotect.prot,
101 					      finished, &hvc->data);
102 			else
103 				ret = os_protect_memory(
104 					(void *) op->u.mprotect.addr,
105 							op->u.mprotect.len,
106 							1, 1, 1);
107 			break;
108 		default:
109 			printk(KERN_ERR "Unknown op type %d in do_ops\n",
110 			       op->type);
111 			BUG();
112 			break;
113 		}
114 	}
115 
116 	if (ret == -ENOMEM)
117 		report_enomem();
118 
119 	return ret;
120 }
121 
122 static int add_mmap(unsigned long virt, unsigned long phys, unsigned long len,
123 		    unsigned int prot, struct host_vm_change *hvc)
124 {
125 	__u64 offset;
126 	struct host_vm_op *last;
127 	int fd = -1, ret = 0;
128 
129 	if (hvc->userspace)
130 		fd = phys_mapping(phys, &offset);
131 	else
132 		offset = phys;
133 	if (hvc->index != 0) {
134 		last = &hvc->ops[hvc->index - 1];
135 		if ((last->type == MMAP) &&
136 		   (last->u.mmap.addr + last->u.mmap.len == virt) &&
137 		   (last->u.mmap.prot == prot) && (last->u.mmap.fd == fd) &&
138 		   (last->u.mmap.offset + last->u.mmap.len == offset)) {
139 			last->u.mmap.len += len;
140 			return 0;
141 		}
142 	}
143 
144 	if (hvc->index == ARRAY_SIZE(hvc->ops)) {
145 		ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
146 		hvc->index = 0;
147 	}
148 
149 	hvc->ops[hvc->index++] = ((struct host_vm_op)
150 				  { .type	= MMAP,
151 				    .u = { .mmap = { .addr	= virt,
152 						     .len	= len,
153 						     .prot	= prot,
154 						     .fd	= fd,
155 						     .offset	= offset }
156 			   } });
157 	return ret;
158 }
159 
160 static int add_munmap(unsigned long addr, unsigned long len,
161 		      struct host_vm_change *hvc)
162 {
163 	struct host_vm_op *last;
164 	int ret = 0;
165 
166 	if ((addr >= STUB_START) && (addr < STUB_END))
167 		return -EINVAL;
168 
169 	if (hvc->index != 0) {
170 		last = &hvc->ops[hvc->index - 1];
171 		if ((last->type == MUNMAP) &&
172 		   (last->u.munmap.addr + last->u.mmap.len == addr)) {
173 			last->u.munmap.len += len;
174 			return 0;
175 		}
176 	}
177 
178 	if (hvc->index == ARRAY_SIZE(hvc->ops)) {
179 		ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
180 		hvc->index = 0;
181 	}
182 
183 	hvc->ops[hvc->index++] = ((struct host_vm_op)
184 				  { .type	= MUNMAP,
185 			     	    .u = { .munmap = { .addr	= addr,
186 						       .len	= len } } });
187 	return ret;
188 }
189 
190 static int add_mprotect(unsigned long addr, unsigned long len,
191 			unsigned int prot, struct host_vm_change *hvc)
192 {
193 	struct host_vm_op *last;
194 	int ret = 0;
195 
196 	if (hvc->index != 0) {
197 		last = &hvc->ops[hvc->index - 1];
198 		if ((last->type == MPROTECT) &&
199 		   (last->u.mprotect.addr + last->u.mprotect.len == addr) &&
200 		   (last->u.mprotect.prot == prot)) {
201 			last->u.mprotect.len += len;
202 			return 0;
203 		}
204 	}
205 
206 	if (hvc->index == ARRAY_SIZE(hvc->ops)) {
207 		ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
208 		hvc->index = 0;
209 	}
210 
211 	hvc->ops[hvc->index++] = ((struct host_vm_op)
212 				  { .type	= MPROTECT,
213 			     	    .u = { .mprotect = { .addr	= addr,
214 							 .len	= len,
215 							 .prot	= prot } } });
216 	return ret;
217 }
218 
219 #define ADD_ROUND(n, inc) (((n) + (inc)) & ~((inc) - 1))
220 
221 static inline int update_pte_range(pmd_t *pmd, unsigned long addr,
222 				   unsigned long end,
223 				   struct host_vm_change *hvc)
224 {
225 	pte_t *pte;
226 	int r, w, x, prot, ret = 0;
227 
228 	pte = pte_offset_kernel(pmd, addr);
229 	do {
230 		if ((addr >= STUB_START) && (addr < STUB_END))
231 			continue;
232 
233 		r = pte_read(*pte);
234 		w = pte_write(*pte);
235 		x = pte_exec(*pte);
236 		if (!pte_young(*pte)) {
237 			r = 0;
238 			w = 0;
239 		} else if (!pte_dirty(*pte))
240 			w = 0;
241 
242 		prot = ((r ? UM_PROT_READ : 0) | (w ? UM_PROT_WRITE : 0) |
243 			(x ? UM_PROT_EXEC : 0));
244 		if (hvc->force || pte_newpage(*pte)) {
245 			if (pte_present(*pte)) {
246 				if (pte_newpage(*pte))
247 					ret = add_mmap(addr, pte_val(*pte) & PAGE_MASK,
248 						       PAGE_SIZE, prot, hvc);
249 			} else
250 				ret = add_munmap(addr, PAGE_SIZE, hvc);
251 		} else if (pte_newprot(*pte))
252 			ret = add_mprotect(addr, PAGE_SIZE, prot, hvc);
253 		*pte = pte_mkuptodate(*pte);
254 	} while (pte++, addr += PAGE_SIZE, ((addr < end) && !ret));
255 	return ret;
256 }
257 
258 static inline int update_pmd_range(pud_t *pud, unsigned long addr,
259 				   unsigned long end,
260 				   struct host_vm_change *hvc)
261 {
262 	pmd_t *pmd;
263 	unsigned long next;
264 	int ret = 0;
265 
266 	pmd = pmd_offset(pud, addr);
267 	do {
268 		next = pmd_addr_end(addr, end);
269 		if (!pmd_present(*pmd)) {
270 			if (hvc->force || pmd_newpage(*pmd)) {
271 				ret = add_munmap(addr, next - addr, hvc);
272 				pmd_mkuptodate(*pmd);
273 			}
274 		}
275 		else ret = update_pte_range(pmd, addr, next, hvc);
276 	} while (pmd++, addr = next, ((addr < end) && !ret));
277 	return ret;
278 }
279 
280 static inline int update_pud_range(p4d_t *p4d, unsigned long addr,
281 				   unsigned long end,
282 				   struct host_vm_change *hvc)
283 {
284 	pud_t *pud;
285 	unsigned long next;
286 	int ret = 0;
287 
288 	pud = pud_offset(p4d, addr);
289 	do {
290 		next = pud_addr_end(addr, end);
291 		if (!pud_present(*pud)) {
292 			if (hvc->force || pud_newpage(*pud)) {
293 				ret = add_munmap(addr, next - addr, hvc);
294 				pud_mkuptodate(*pud);
295 			}
296 		}
297 		else ret = update_pmd_range(pud, addr, next, hvc);
298 	} while (pud++, addr = next, ((addr < end) && !ret));
299 	return ret;
300 }
301 
302 static inline int update_p4d_range(pgd_t *pgd, unsigned long addr,
303 				   unsigned long end,
304 				   struct host_vm_change *hvc)
305 {
306 	p4d_t *p4d;
307 	unsigned long next;
308 	int ret = 0;
309 
310 	p4d = p4d_offset(pgd, addr);
311 	do {
312 		next = p4d_addr_end(addr, end);
313 		if (!p4d_present(*p4d)) {
314 			if (hvc->force || p4d_newpage(*p4d)) {
315 				ret = add_munmap(addr, next - addr, hvc);
316 				p4d_mkuptodate(*p4d);
317 			}
318 		} else
319 			ret = update_pud_range(p4d, addr, next, hvc);
320 	} while (p4d++, addr = next, ((addr < end) && !ret));
321 	return ret;
322 }
323 
324 void fix_range_common(struct mm_struct *mm, unsigned long start_addr,
325 		      unsigned long end_addr, int force)
326 {
327 	pgd_t *pgd;
328 	struct host_vm_change hvc;
329 	unsigned long addr = start_addr, next;
330 	int ret = 0, userspace = 1;
331 
332 	hvc = INIT_HVC(mm, force, userspace);
333 	pgd = pgd_offset(mm, addr);
334 	do {
335 		next = pgd_addr_end(addr, end_addr);
336 		if (!pgd_present(*pgd)) {
337 			if (force || pgd_newpage(*pgd)) {
338 				ret = add_munmap(addr, next - addr, &hvc);
339 				pgd_mkuptodate(*pgd);
340 			}
341 		} else
342 			ret = update_p4d_range(pgd, addr, next, &hvc);
343 	} while (pgd++, addr = next, ((addr < end_addr) && !ret));
344 
345 	if (!ret)
346 		ret = do_ops(&hvc, hvc.index, 1);
347 
348 	/* This is not an else because ret is modified above */
349 	if (ret) {
350 		printk(KERN_ERR "fix_range_common: failed, killing current "
351 		       "process: %d\n", task_tgid_vnr(current));
352 		/* We are under mmap_sem, release it such that current can terminate */
353 		up_write(&current->mm->mmap_sem);
354 		force_sig(SIGKILL);
355 		do_signal(&current->thread.regs);
356 	}
357 }
358 
359 static int flush_tlb_kernel_range_common(unsigned long start, unsigned long end)
360 {
361 	struct mm_struct *mm;
362 	pgd_t *pgd;
363 	p4d_t *p4d;
364 	pud_t *pud;
365 	pmd_t *pmd;
366 	pte_t *pte;
367 	unsigned long addr, last;
368 	int updated = 0, err = 0, force = 0, userspace = 0;
369 	struct host_vm_change hvc;
370 
371 	mm = &init_mm;
372 	hvc = INIT_HVC(mm, force, userspace);
373 	for (addr = start; addr < end;) {
374 		pgd = pgd_offset(mm, addr);
375 		if (!pgd_present(*pgd)) {
376 			last = ADD_ROUND(addr, PGDIR_SIZE);
377 			if (last > end)
378 				last = end;
379 			if (pgd_newpage(*pgd)) {
380 				updated = 1;
381 				err = add_munmap(addr, last - addr, &hvc);
382 				if (err < 0)
383 					panic("munmap failed, errno = %d\n",
384 					      -err);
385 			}
386 			addr = last;
387 			continue;
388 		}
389 
390 		p4d = p4d_offset(pgd, addr);
391 		if (!p4d_present(*p4d)) {
392 			last = ADD_ROUND(addr, P4D_SIZE);
393 			if (last > end)
394 				last = end;
395 			if (p4d_newpage(*p4d)) {
396 				updated = 1;
397 				err = add_munmap(addr, last - addr, &hvc);
398 				if (err < 0)
399 					panic("munmap failed, errno = %d\n",
400 					      -err);
401 			}
402 			addr = last;
403 			continue;
404 		}
405 
406 		pud = pud_offset(p4d, addr);
407 		if (!pud_present(*pud)) {
408 			last = ADD_ROUND(addr, PUD_SIZE);
409 			if (last > end)
410 				last = end;
411 			if (pud_newpage(*pud)) {
412 				updated = 1;
413 				err = add_munmap(addr, last - addr, &hvc);
414 				if (err < 0)
415 					panic("munmap failed, errno = %d\n",
416 					      -err);
417 			}
418 			addr = last;
419 			continue;
420 		}
421 
422 		pmd = pmd_offset(pud, addr);
423 		if (!pmd_present(*pmd)) {
424 			last = ADD_ROUND(addr, PMD_SIZE);
425 			if (last > end)
426 				last = end;
427 			if (pmd_newpage(*pmd)) {
428 				updated = 1;
429 				err = add_munmap(addr, last - addr, &hvc);
430 				if (err < 0)
431 					panic("munmap failed, errno = %d\n",
432 					      -err);
433 			}
434 			addr = last;
435 			continue;
436 		}
437 
438 		pte = pte_offset_kernel(pmd, addr);
439 		if (!pte_present(*pte) || pte_newpage(*pte)) {
440 			updated = 1;
441 			err = add_munmap(addr, PAGE_SIZE, &hvc);
442 			if (err < 0)
443 				panic("munmap failed, errno = %d\n",
444 				      -err);
445 			if (pte_present(*pte))
446 				err = add_mmap(addr, pte_val(*pte) & PAGE_MASK,
447 					       PAGE_SIZE, 0, &hvc);
448 		}
449 		else if (pte_newprot(*pte)) {
450 			updated = 1;
451 			err = add_mprotect(addr, PAGE_SIZE, 0, &hvc);
452 		}
453 		addr += PAGE_SIZE;
454 	}
455 	if (!err)
456 		err = do_ops(&hvc, hvc.index, 1);
457 
458 	if (err < 0)
459 		panic("flush_tlb_kernel failed, errno = %d\n", err);
460 	return updated;
461 }
462 
463 void flush_tlb_page(struct vm_area_struct *vma, unsigned long address)
464 {
465 	pgd_t *pgd;
466 	p4d_t *p4d;
467 	pud_t *pud;
468 	pmd_t *pmd;
469 	pte_t *pte;
470 	struct mm_struct *mm = vma->vm_mm;
471 	void *flush = NULL;
472 	int r, w, x, prot, err = 0;
473 	struct mm_id *mm_id;
474 
475 	address &= PAGE_MASK;
476 	pgd = pgd_offset(mm, address);
477 	if (!pgd_present(*pgd))
478 		goto kill;
479 
480 	p4d = p4d_offset(pgd, address);
481 	if (!p4d_present(*p4d))
482 		goto kill;
483 
484 	pud = pud_offset(p4d, address);
485 	if (!pud_present(*pud))
486 		goto kill;
487 
488 	pmd = pmd_offset(pud, address);
489 	if (!pmd_present(*pmd))
490 		goto kill;
491 
492 	pte = pte_offset_kernel(pmd, address);
493 
494 	r = pte_read(*pte);
495 	w = pte_write(*pte);
496 	x = pte_exec(*pte);
497 	if (!pte_young(*pte)) {
498 		r = 0;
499 		w = 0;
500 	} else if (!pte_dirty(*pte)) {
501 		w = 0;
502 	}
503 
504 	mm_id = &mm->context.id;
505 	prot = ((r ? UM_PROT_READ : 0) | (w ? UM_PROT_WRITE : 0) |
506 		(x ? UM_PROT_EXEC : 0));
507 	if (pte_newpage(*pte)) {
508 		if (pte_present(*pte)) {
509 			unsigned long long offset;
510 			int fd;
511 
512 			fd = phys_mapping(pte_val(*pte) & PAGE_MASK, &offset);
513 			err = map(mm_id, address, PAGE_SIZE, prot, fd, offset,
514 				  1, &flush);
515 		}
516 		else err = unmap(mm_id, address, PAGE_SIZE, 1, &flush);
517 	}
518 	else if (pte_newprot(*pte))
519 		err = protect(mm_id, address, PAGE_SIZE, prot, 1, &flush);
520 
521 	if (err) {
522 		if (err == -ENOMEM)
523 			report_enomem();
524 
525 		goto kill;
526 	}
527 
528 	*pte = pte_mkuptodate(*pte);
529 
530 	return;
531 
532 kill:
533 	printk(KERN_ERR "Failed to flush page for address 0x%lx\n", address);
534 	force_sig(SIGKILL);
535 }
536 
537 void flush_tlb_all(void)
538 {
539 	/*
540 	 * Don't bother flushing if this address space is about to be
541 	 * destroyed.
542 	 */
543 	if (atomic_read(&current->mm->mm_users) == 0)
544 		return;
545 
546 	flush_tlb_mm(current->mm);
547 }
548 
549 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
550 {
551 	flush_tlb_kernel_range_common(start, end);
552 }
553 
554 void flush_tlb_kernel_vm(void)
555 {
556 	flush_tlb_kernel_range_common(start_vm, end_vm);
557 }
558 
559 void __flush_tlb_one(unsigned long addr)
560 {
561 	flush_tlb_kernel_range_common(addr, addr + PAGE_SIZE);
562 }
563 
564 static void fix_range(struct mm_struct *mm, unsigned long start_addr,
565 		      unsigned long end_addr, int force)
566 {
567 	/*
568 	 * Don't bother flushing if this address space is about to be
569 	 * destroyed.
570 	 */
571 	if (atomic_read(&mm->mm_users) == 0)
572 		return;
573 
574 	fix_range_common(mm, start_addr, end_addr, force);
575 }
576 
577 void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
578 		     unsigned long end)
579 {
580 	if (vma->vm_mm == NULL)
581 		flush_tlb_kernel_range_common(start, end);
582 	else fix_range(vma->vm_mm, start, end, 0);
583 }
584 EXPORT_SYMBOL(flush_tlb_range);
585 
586 void flush_tlb_mm_range(struct mm_struct *mm, unsigned long start,
587 			unsigned long end)
588 {
589 	fix_range(mm, start, end, 0);
590 }
591 
592 void flush_tlb_mm(struct mm_struct *mm)
593 {
594 	struct vm_area_struct *vma = mm->mmap;
595 
596 	while (vma != NULL) {
597 		fix_range(mm, vma->vm_start, vma->vm_end, 0);
598 		vma = vma->vm_next;
599 	}
600 }
601 
602 void force_flush_all(void)
603 {
604 	struct mm_struct *mm = current->mm;
605 	struct vm_area_struct *vma = mm->mmap;
606 
607 	while (vma != NULL) {
608 		fix_range(mm, vma->vm_start, vma->vm_end, 1);
609 		vma = vma->vm_next;
610 	}
611 }
612