xref: /linux/arch/um/kernel/tlb.c (revision 76ed9158e1d474e963fc59da7a461b27a2212c5a) 
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/tlbflush.h>
11 #include <asm/mmu_context.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 				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);
78 			else
79 				map_memory(op->u.mmap.addr, op->u.mmap.offset,
80 					   op->u.mmap.len, 1, 1, 1);
81 			break;
82 		case MUNMAP:
83 			if (hvc->userspace)
84 				unmap(&hvc->mm->context.id,
85 				      op->u.munmap.addr,
86 				      op->u.munmap.len);
87 			else
88 				ret = os_unmap_memory(
89 					(void *) op->u.munmap.addr,
90 						      op->u.munmap.len);
91 
92 			break;
93 		case MPROTECT:
94 			if (hvc->userspace)
95 				protect(&hvc->mm->context.id,
96 					op->u.mprotect.addr,
97 					op->u.mprotect.len,
98 					op->u.mprotect.prot);
99 			else
100 				ret = os_protect_memory(
101 					(void *) op->u.mprotect.addr,
102 							op->u.mprotect.len,
103 							1, 1, 1);
104 			break;
105 		default:
106 			printk(KERN_ERR "Unknown op type %d in do_ops\n",
107 			       op->type);
108 			BUG();
109 			break;
110 		}
111 	}
112 
113 	if (hvc->userspace && finished)
114 		ret = syscall_stub_flush(&hvc->mm->context.id);
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 (hvc->index != 0) {
167 		last = &hvc->ops[hvc->index - 1];
168 		if ((last->type == MUNMAP) &&
169 		   (last->u.munmap.addr + last->u.mmap.len == addr)) {
170 			last->u.munmap.len += len;
171 			return 0;
172 		}
173 	}
174 
175 	if (hvc->index == ARRAY_SIZE(hvc->ops)) {
176 		ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
177 		hvc->index = 0;
178 	}
179 
180 	hvc->ops[hvc->index++] = ((struct host_vm_op)
181 				  { .type	= MUNMAP,
182 			     	    .u = { .munmap = { .addr	= addr,
183 						       .len	= len } } });
184 	return ret;
185 }
186 
187 static int add_mprotect(unsigned long addr, unsigned long len,
188 			unsigned int prot, struct host_vm_change *hvc)
189 {
190 	struct host_vm_op *last;
191 	int ret = 0;
192 
193 	if (hvc->index != 0) {
194 		last = &hvc->ops[hvc->index - 1];
195 		if ((last->type == MPROTECT) &&
196 		   (last->u.mprotect.addr + last->u.mprotect.len == addr) &&
197 		   (last->u.mprotect.prot == prot)) {
198 			last->u.mprotect.len += len;
199 			return 0;
200 		}
201 	}
202 
203 	if (hvc->index == ARRAY_SIZE(hvc->ops)) {
204 		ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
205 		hvc->index = 0;
206 	}
207 
208 	hvc->ops[hvc->index++] = ((struct host_vm_op)
209 				  { .type	= MPROTECT,
210 			     	    .u = { .mprotect = { .addr	= addr,
211 							 .len	= len,
212 							 .prot	= prot } } });
213 	return ret;
214 }
215 
216 #define ADD_ROUND(n, inc) (((n) + (inc)) & ~((inc) - 1))
217 
218 static inline int update_pte_range(pmd_t *pmd, unsigned long addr,
219 				   unsigned long end,
220 				   struct host_vm_change *hvc)
221 {
222 	pte_t *pte;
223 	int r, w, x, prot, ret = 0;
224 
225 	pte = pte_offset_kernel(pmd, addr);
226 	do {
227 		r = pte_read(*pte);
228 		w = pte_write(*pte);
229 		x = pte_exec(*pte);
230 		if (!pte_young(*pte)) {
231 			r = 0;
232 			w = 0;
233 		} else if (!pte_dirty(*pte))
234 			w = 0;
235 
236 		prot = ((r ? UM_PROT_READ : 0) | (w ? UM_PROT_WRITE : 0) |
237 			(x ? UM_PROT_EXEC : 0));
238 		if (hvc->force || pte_newpage(*pte)) {
239 			if (pte_present(*pte)) {
240 				if (pte_newpage(*pte))
241 					ret = add_mmap(addr, pte_val(*pte) & PAGE_MASK,
242 						       PAGE_SIZE, prot, hvc);
243 			} else
244 				ret = add_munmap(addr, PAGE_SIZE, hvc);
245 		} else if (pte_newprot(*pte))
246 			ret = add_mprotect(addr, PAGE_SIZE, prot, hvc);
247 		*pte = pte_mkuptodate(*pte);
248 	} while (pte++, addr += PAGE_SIZE, ((addr < end) && !ret));
249 	return ret;
250 }
251 
252 static inline int update_pmd_range(pud_t *pud, unsigned long addr,
253 				   unsigned long end,
254 				   struct host_vm_change *hvc)
255 {
256 	pmd_t *pmd;
257 	unsigned long next;
258 	int ret = 0;
259 
260 	pmd = pmd_offset(pud, addr);
261 	do {
262 		next = pmd_addr_end(addr, end);
263 		if (!pmd_present(*pmd)) {
264 			if (hvc->force || pmd_newpage(*pmd)) {
265 				ret = add_munmap(addr, next - addr, hvc);
266 				pmd_mkuptodate(*pmd);
267 			}
268 		}
269 		else ret = update_pte_range(pmd, addr, next, hvc);
270 	} while (pmd++, addr = next, ((addr < end) && !ret));
271 	return ret;
272 }
273 
274 static inline int update_pud_range(p4d_t *p4d, unsigned long addr,
275 				   unsigned long end,
276 				   struct host_vm_change *hvc)
277 {
278 	pud_t *pud;
279 	unsigned long next;
280 	int ret = 0;
281 
282 	pud = pud_offset(p4d, addr);
283 	do {
284 		next = pud_addr_end(addr, end);
285 		if (!pud_present(*pud)) {
286 			if (hvc->force || pud_newpage(*pud)) {
287 				ret = add_munmap(addr, next - addr, hvc);
288 				pud_mkuptodate(*pud);
289 			}
290 		}
291 		else ret = update_pmd_range(pud, addr, next, hvc);
292 	} while (pud++, addr = next, ((addr < end) && !ret));
293 	return ret;
294 }
295 
296 static inline int update_p4d_range(pgd_t *pgd, unsigned long addr,
297 				   unsigned long end,
298 				   struct host_vm_change *hvc)
299 {
300 	p4d_t *p4d;
301 	unsigned long next;
302 	int ret = 0;
303 
304 	p4d = p4d_offset(pgd, addr);
305 	do {
306 		next = p4d_addr_end(addr, end);
307 		if (!p4d_present(*p4d)) {
308 			if (hvc->force || p4d_newpage(*p4d)) {
309 				ret = add_munmap(addr, next - addr, hvc);
310 				p4d_mkuptodate(*p4d);
311 			}
312 		} else
313 			ret = update_pud_range(p4d, addr, next, hvc);
314 	} while (p4d++, addr = next, ((addr < end) && !ret));
315 	return ret;
316 }
317 
318 static void fix_range_common(struct mm_struct *mm, unsigned long start_addr,
319 			     unsigned long end_addr, int force)
320 {
321 	pgd_t *pgd;
322 	struct host_vm_change hvc;
323 	unsigned long addr = start_addr, next;
324 	int ret = 0, userspace = 1;
325 
326 	hvc = INIT_HVC(mm, force, userspace);
327 	pgd = pgd_offset(mm, addr);
328 	do {
329 		next = pgd_addr_end(addr, end_addr);
330 		if (!pgd_present(*pgd)) {
331 			if (force || pgd_newpage(*pgd)) {
332 				ret = add_munmap(addr, next - addr, &hvc);
333 				pgd_mkuptodate(*pgd);
334 			}
335 		} else
336 			ret = update_p4d_range(pgd, addr, next, &hvc);
337 	} while (pgd++, addr = next, ((addr < end_addr) && !ret));
338 
339 	if (!ret)
340 		ret = do_ops(&hvc, hvc.index, 1);
341 
342 	/* This is not an else because ret is modified above */
343 	if (ret) {
344 		struct mm_id *mm_idp = &current->mm->context.id;
345 
346 		printk(KERN_ERR "fix_range_common: failed, killing current "
347 		       "process: %d\n", task_tgid_vnr(current));
348 		mm_idp->kill = 1;
349 	}
350 }
351 
352 static int flush_tlb_kernel_range_common(unsigned long start, unsigned long end)
353 {
354 	struct mm_struct *mm;
355 	pgd_t *pgd;
356 	p4d_t *p4d;
357 	pud_t *pud;
358 	pmd_t *pmd;
359 	pte_t *pte;
360 	unsigned long addr, last;
361 	int updated = 0, err = 0, force = 0, userspace = 0;
362 	struct host_vm_change hvc;
363 
364 	mm = &init_mm;
365 	hvc = INIT_HVC(mm, force, userspace);
366 	for (addr = start; addr < end;) {
367 		pgd = pgd_offset(mm, addr);
368 		if (!pgd_present(*pgd)) {
369 			last = ADD_ROUND(addr, PGDIR_SIZE);
370 			if (last > end)
371 				last = end;
372 			if (pgd_newpage(*pgd)) {
373 				updated = 1;
374 				err = add_munmap(addr, last - addr, &hvc);
375 				if (err < 0)
376 					panic("munmap failed, errno = %d\n",
377 					      -err);
378 			}
379 			addr = last;
380 			continue;
381 		}
382 
383 		p4d = p4d_offset(pgd, addr);
384 		if (!p4d_present(*p4d)) {
385 			last = ADD_ROUND(addr, P4D_SIZE);
386 			if (last > end)
387 				last = end;
388 			if (p4d_newpage(*p4d)) {
389 				updated = 1;
390 				err = add_munmap(addr, last - addr, &hvc);
391 				if (err < 0)
392 					panic("munmap failed, errno = %d\n",
393 					      -err);
394 			}
395 			addr = last;
396 			continue;
397 		}
398 
399 		pud = pud_offset(p4d, addr);
400 		if (!pud_present(*pud)) {
401 			last = ADD_ROUND(addr, PUD_SIZE);
402 			if (last > end)
403 				last = end;
404 			if (pud_newpage(*pud)) {
405 				updated = 1;
406 				err = add_munmap(addr, last - addr, &hvc);
407 				if (err < 0)
408 					panic("munmap failed, errno = %d\n",
409 					      -err);
410 			}
411 			addr = last;
412 			continue;
413 		}
414 
415 		pmd = pmd_offset(pud, addr);
416 		if (!pmd_present(*pmd)) {
417 			last = ADD_ROUND(addr, PMD_SIZE);
418 			if (last > end)
419 				last = end;
420 			if (pmd_newpage(*pmd)) {
421 				updated = 1;
422 				err = add_munmap(addr, last - addr, &hvc);
423 				if (err < 0)
424 					panic("munmap failed, errno = %d\n",
425 					      -err);
426 			}
427 			addr = last;
428 			continue;
429 		}
430 
431 		pte = pte_offset_kernel(pmd, addr);
432 		if (!pte_present(*pte) || pte_newpage(*pte)) {
433 			updated = 1;
434 			err = add_munmap(addr, PAGE_SIZE, &hvc);
435 			if (err < 0)
436 				panic("munmap failed, errno = %d\n",
437 				      -err);
438 			if (pte_present(*pte))
439 				err = add_mmap(addr, pte_val(*pte) & PAGE_MASK,
440 					       PAGE_SIZE, 0, &hvc);
441 		}
442 		else if (pte_newprot(*pte)) {
443 			updated = 1;
444 			err = add_mprotect(addr, PAGE_SIZE, 0, &hvc);
445 		}
446 		addr += PAGE_SIZE;
447 	}
448 	if (!err)
449 		err = do_ops(&hvc, hvc.index, 1);
450 
451 	if (err < 0)
452 		panic("flush_tlb_kernel failed, errno = %d\n", err);
453 	return updated;
454 }
455 
456 void flush_tlb_page(struct vm_area_struct *vma, unsigned long address)
457 {
458 	pgd_t *pgd;
459 	p4d_t *p4d;
460 	pud_t *pud;
461 	pmd_t *pmd;
462 	pte_t *pte;
463 	struct mm_struct *mm = vma->vm_mm;
464 	int r, w, x, prot, err = 0;
465 	struct mm_id *mm_id;
466 
467 	address &= PAGE_MASK;
468 
469 	pgd = pgd_offset(mm, address);
470 	if (!pgd_present(*pgd))
471 		goto kill;
472 
473 	p4d = p4d_offset(pgd, address);
474 	if (!p4d_present(*p4d))
475 		goto kill;
476 
477 	pud = pud_offset(p4d, address);
478 	if (!pud_present(*pud))
479 		goto kill;
480 
481 	pmd = pmd_offset(pud, address);
482 	if (!pmd_present(*pmd))
483 		goto kill;
484 
485 	pte = pte_offset_kernel(pmd, address);
486 
487 	r = pte_read(*pte);
488 	w = pte_write(*pte);
489 	x = pte_exec(*pte);
490 	if (!pte_young(*pte)) {
491 		r = 0;
492 		w = 0;
493 	} else if (!pte_dirty(*pte)) {
494 		w = 0;
495 	}
496 
497 	mm_id = &mm->context.id;
498 	prot = ((r ? UM_PROT_READ : 0) | (w ? UM_PROT_WRITE : 0) |
499 		(x ? UM_PROT_EXEC : 0));
500 	if (pte_newpage(*pte)) {
501 		if (pte_present(*pte)) {
502 			unsigned long long offset;
503 			int fd;
504 
505 			fd = phys_mapping(pte_val(*pte) & PAGE_MASK, &offset);
506 			map(mm_id, address, PAGE_SIZE, prot, fd, offset);
507 		} else
508 			unmap(mm_id, address, PAGE_SIZE);
509 	} else if (pte_newprot(*pte))
510 		protect(mm_id, address, PAGE_SIZE, prot);
511 
512 	err = syscall_stub_flush(mm_id);
513 	if (err) {
514 		if (err == -ENOMEM)
515 			report_enomem();
516 
517 		goto kill;
518 	}
519 
520 	*pte = pte_mkuptodate(*pte);
521 
522 	return;
523 
524 kill:
525 	printk(KERN_ERR "Failed to flush page for address 0x%lx\n", address);
526 	force_sig(SIGKILL);
527 }
528 
529 void flush_tlb_all(void)
530 {
531 	/*
532 	 * Don't bother flushing if this address space is about to be
533 	 * destroyed.
534 	 */
535 	if (atomic_read(&current->mm->mm_users) == 0)
536 		return;
537 
538 	flush_tlb_mm(current->mm);
539 }
540 
541 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
542 {
543 	flush_tlb_kernel_range_common(start, end);
544 }
545 
546 void flush_tlb_kernel_vm(void)
547 {
548 	flush_tlb_kernel_range_common(start_vm, end_vm);
549 }
550 
551 void __flush_tlb_one(unsigned long addr)
552 {
553 	flush_tlb_kernel_range_common(addr, addr + PAGE_SIZE);
554 }
555 
556 static void fix_range(struct mm_struct *mm, unsigned long start_addr,
557 		      unsigned long end_addr, int force)
558 {
559 	/*
560 	 * Don't bother flushing if this address space is about to be
561 	 * destroyed.
562 	 */
563 	if (atomic_read(&mm->mm_users) == 0)
564 		return;
565 
566 	fix_range_common(mm, start_addr, end_addr, force);
567 }
568 
569 void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
570 		     unsigned long end)
571 {
572 	if (vma->vm_mm == NULL)
573 		flush_tlb_kernel_range_common(start, end);
574 	else fix_range(vma->vm_mm, start, end, 0);
575 }
576 EXPORT_SYMBOL(flush_tlb_range);
577 
578 void flush_tlb_mm(struct mm_struct *mm)
579 {
580 	struct vm_area_struct *vma;
581 	VMA_ITERATOR(vmi, mm, 0);
582 
583 	for_each_vma(vmi, vma)
584 		fix_range(mm, vma->vm_start, vma->vm_end, 0);
585 }
586 
587 void force_flush_all(void)
588 {
589 	struct mm_struct *mm = current->mm;
590 	struct vm_area_struct *vma;
591 	VMA_ITERATOR(vmi, mm, 0);
592 
593 	mmap_read_lock(mm);
594 	for_each_vma(vmi, vma)
595 		fix_range(mm, vma->vm_start, vma->vm_end, 1);
596 	mmap_read_unlock(mm);
597 }
598