xref: /openbmc/linux/arch/powerpc/kvm/book3s_64_vio.c (revision 90848a25)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *
4  * Copyright 2010 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
5  * Copyright 2011 David Gibson, IBM Corporation <dwg@au1.ibm.com>
6  * Copyright 2016 Alexey Kardashevskiy, IBM Corporation <aik@au1.ibm.com>
7  */
8 
9 #include <linux/types.h>
10 #include <linux/string.h>
11 #include <linux/kvm.h>
12 #include <linux/kvm_host.h>
13 #include <linux/highmem.h>
14 #include <linux/gfp.h>
15 #include <linux/slab.h>
16 #include <linux/sched/signal.h>
17 #include <linux/hugetlb.h>
18 #include <linux/list.h>
19 #include <linux/anon_inodes.h>
20 #include <linux/iommu.h>
21 #include <linux/file.h>
22 #include <linux/mm.h>
23 
24 #include <asm/kvm_ppc.h>
25 #include <asm/kvm_book3s.h>
26 #include <asm/book3s/64/mmu-hash.h>
27 #include <asm/hvcall.h>
28 #include <asm/synch.h>
29 #include <asm/ppc-opcode.h>
30 #include <asm/udbg.h>
31 #include <asm/iommu.h>
32 #include <asm/tce.h>
33 #include <asm/mmu_context.h>
34 
35 static struct kvmppc_spapr_tce_table *kvmppc_find_table(struct kvm *kvm,
36 	unsigned long liobn)
37 {
38 	struct kvmppc_spapr_tce_table *stt;
39 
40 	list_for_each_entry_lockless(stt, &kvm->arch.spapr_tce_tables, list)
41 		if (stt->liobn == liobn)
42 			return stt;
43 
44 	return NULL;
45 }
46 
47 static unsigned long kvmppc_tce_pages(unsigned long iommu_pages)
48 {
49 	return ALIGN(iommu_pages * sizeof(u64), PAGE_SIZE) / PAGE_SIZE;
50 }
51 
52 static unsigned long kvmppc_stt_pages(unsigned long tce_pages)
53 {
54 	unsigned long stt_bytes = sizeof(struct kvmppc_spapr_tce_table) +
55 			(tce_pages * sizeof(struct page *));
56 
57 	return tce_pages + ALIGN(stt_bytes, PAGE_SIZE) / PAGE_SIZE;
58 }
59 
60 static void kvm_spapr_tce_iommu_table_free(struct rcu_head *head)
61 {
62 	struct kvmppc_spapr_tce_iommu_table *stit = container_of(head,
63 			struct kvmppc_spapr_tce_iommu_table, rcu);
64 
65 	iommu_tce_table_put(stit->tbl);
66 
67 	kfree(stit);
68 }
69 
70 static void kvm_spapr_tce_liobn_put(struct kref *kref)
71 {
72 	struct kvmppc_spapr_tce_iommu_table *stit = container_of(kref,
73 			struct kvmppc_spapr_tce_iommu_table, kref);
74 
75 	list_del_rcu(&stit->next);
76 
77 	call_rcu(&stit->rcu, kvm_spapr_tce_iommu_table_free);
78 }
79 
80 extern void kvm_spapr_tce_release_iommu_group(struct kvm *kvm,
81 		struct iommu_group *grp)
82 {
83 	int i;
84 	struct kvmppc_spapr_tce_table *stt;
85 	struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
86 	struct iommu_table_group *table_group = NULL;
87 
88 	rcu_read_lock();
89 	list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
90 
91 		table_group = iommu_group_get_iommudata(grp);
92 		if (WARN_ON(!table_group))
93 			continue;
94 
95 		list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
96 			for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
97 				if (table_group->tables[i] != stit->tbl)
98 					continue;
99 
100 				kref_put(&stit->kref, kvm_spapr_tce_liobn_put);
101 			}
102 		}
103 		cond_resched_rcu();
104 	}
105 	rcu_read_unlock();
106 }
107 
108 extern long kvm_spapr_tce_attach_iommu_group(struct kvm *kvm, int tablefd,
109 		struct iommu_group *grp)
110 {
111 	struct kvmppc_spapr_tce_table *stt = NULL;
112 	bool found = false;
113 	struct iommu_table *tbl = NULL;
114 	struct iommu_table_group *table_group;
115 	long i;
116 	struct kvmppc_spapr_tce_iommu_table *stit;
117 	struct fd f;
118 
119 	f = fdget(tablefd);
120 	if (!f.file)
121 		return -EBADF;
122 
123 	rcu_read_lock();
124 	list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
125 		if (stt == f.file->private_data) {
126 			found = true;
127 			break;
128 		}
129 	}
130 	rcu_read_unlock();
131 
132 	fdput(f);
133 
134 	if (!found)
135 		return -EINVAL;
136 
137 	table_group = iommu_group_get_iommudata(grp);
138 	if (WARN_ON(!table_group))
139 		return -EFAULT;
140 
141 	for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
142 		struct iommu_table *tbltmp = table_group->tables[i];
143 
144 		if (!tbltmp)
145 			continue;
146 		/* Make sure hardware table parameters are compatible */
147 		if ((tbltmp->it_page_shift <= stt->page_shift) &&
148 				(tbltmp->it_offset << tbltmp->it_page_shift ==
149 				 stt->offset << stt->page_shift) &&
150 				(tbltmp->it_size << tbltmp->it_page_shift >=
151 				 stt->size << stt->page_shift)) {
152 			/*
153 			 * Reference the table to avoid races with
154 			 * add/remove DMA windows.
155 			 */
156 			tbl = iommu_tce_table_get(tbltmp);
157 			break;
158 		}
159 	}
160 	if (!tbl)
161 		return -EINVAL;
162 
163 	rcu_read_lock();
164 	list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
165 		if (tbl != stit->tbl)
166 			continue;
167 
168 		if (!kref_get_unless_zero(&stit->kref)) {
169 			/* stit is being destroyed */
170 			iommu_tce_table_put(tbl);
171 			rcu_read_unlock();
172 			return -ENOTTY;
173 		}
174 		/*
175 		 * The table is already known to this KVM, we just increased
176 		 * its KVM reference counter and can return.
177 		 */
178 		rcu_read_unlock();
179 		return 0;
180 	}
181 	rcu_read_unlock();
182 
183 	stit = kzalloc(sizeof(*stit), GFP_KERNEL);
184 	if (!stit) {
185 		iommu_tce_table_put(tbl);
186 		return -ENOMEM;
187 	}
188 
189 	stit->tbl = tbl;
190 	kref_init(&stit->kref);
191 
192 	list_add_rcu(&stit->next, &stt->iommu_tables);
193 
194 	return 0;
195 }
196 
197 static void release_spapr_tce_table(struct rcu_head *head)
198 {
199 	struct kvmppc_spapr_tce_table *stt = container_of(head,
200 			struct kvmppc_spapr_tce_table, rcu);
201 	unsigned long i, npages = kvmppc_tce_pages(stt->size);
202 
203 	for (i = 0; i < npages; i++)
204 		if (stt->pages[i])
205 			__free_page(stt->pages[i]);
206 
207 	kfree(stt);
208 }
209 
210 static struct page *kvm_spapr_get_tce_page(struct kvmppc_spapr_tce_table *stt,
211 		unsigned long sttpage)
212 {
213 	struct page *page = stt->pages[sttpage];
214 
215 	if (page)
216 		return page;
217 
218 	mutex_lock(&stt->alloc_lock);
219 	page = stt->pages[sttpage];
220 	if (!page) {
221 		page = alloc_page(GFP_KERNEL | __GFP_ZERO);
222 		WARN_ON_ONCE(!page);
223 		if (page)
224 			stt->pages[sttpage] = page;
225 	}
226 	mutex_unlock(&stt->alloc_lock);
227 
228 	return page;
229 }
230 
231 static vm_fault_t kvm_spapr_tce_fault(struct vm_fault *vmf)
232 {
233 	struct kvmppc_spapr_tce_table *stt = vmf->vma->vm_file->private_data;
234 	struct page *page;
235 
236 	if (vmf->pgoff >= kvmppc_tce_pages(stt->size))
237 		return VM_FAULT_SIGBUS;
238 
239 	page = kvm_spapr_get_tce_page(stt, vmf->pgoff);
240 	if (!page)
241 		return VM_FAULT_OOM;
242 
243 	get_page(page);
244 	vmf->page = page;
245 	return 0;
246 }
247 
248 static const struct vm_operations_struct kvm_spapr_tce_vm_ops = {
249 	.fault = kvm_spapr_tce_fault,
250 };
251 
252 static int kvm_spapr_tce_mmap(struct file *file, struct vm_area_struct *vma)
253 {
254 	vma->vm_ops = &kvm_spapr_tce_vm_ops;
255 	return 0;
256 }
257 
258 static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
259 {
260 	struct kvmppc_spapr_tce_table *stt = filp->private_data;
261 	struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
262 	struct kvm *kvm = stt->kvm;
263 
264 	mutex_lock(&kvm->lock);
265 	list_del_rcu(&stt->list);
266 	mutex_unlock(&kvm->lock);
267 
268 	list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
269 		WARN_ON(!kref_read(&stit->kref));
270 		while (1) {
271 			if (kref_put(&stit->kref, kvm_spapr_tce_liobn_put))
272 				break;
273 		}
274 	}
275 
276 	account_locked_vm(kvm->mm,
277 		kvmppc_stt_pages(kvmppc_tce_pages(stt->size)), false);
278 
279 	kvm_put_kvm(stt->kvm);
280 
281 	call_rcu(&stt->rcu, release_spapr_tce_table);
282 
283 	return 0;
284 }
285 
286 static const struct file_operations kvm_spapr_tce_fops = {
287 	.mmap           = kvm_spapr_tce_mmap,
288 	.release	= kvm_spapr_tce_release,
289 };
290 
291 int kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
292 				  struct kvm_create_spapr_tce_64 *args)
293 {
294 	struct kvmppc_spapr_tce_table *stt = NULL;
295 	struct kvmppc_spapr_tce_table *siter;
296 	struct mm_struct *mm = kvm->mm;
297 	unsigned long npages;
298 	int ret;
299 
300 	if (!args->size || args->page_shift < 12 || args->page_shift > 34 ||
301 		(args->offset + args->size > (ULLONG_MAX >> args->page_shift)))
302 		return -EINVAL;
303 
304 	npages = kvmppc_tce_pages(args->size);
305 	ret = account_locked_vm(mm, kvmppc_stt_pages(npages), true);
306 	if (ret)
307 		return ret;
308 
309 	ret = -ENOMEM;
310 	stt = kzalloc(struct_size(stt, pages, npages), GFP_KERNEL | __GFP_NOWARN);
311 	if (!stt)
312 		goto fail_acct;
313 
314 	stt->liobn = args->liobn;
315 	stt->page_shift = args->page_shift;
316 	stt->offset = args->offset;
317 	stt->size = args->size;
318 	stt->kvm = kvm;
319 	mutex_init(&stt->alloc_lock);
320 	INIT_LIST_HEAD_RCU(&stt->iommu_tables);
321 
322 	mutex_lock(&kvm->lock);
323 
324 	/* Check this LIOBN hasn't been previously allocated */
325 	ret = 0;
326 	list_for_each_entry(siter, &kvm->arch.spapr_tce_tables, list) {
327 		if (siter->liobn == args->liobn) {
328 			ret = -EBUSY;
329 			break;
330 		}
331 	}
332 
333 	kvm_get_kvm(kvm);
334 	if (!ret)
335 		ret = anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops,
336 				       stt, O_RDWR | O_CLOEXEC);
337 
338 	if (ret >= 0)
339 		list_add_rcu(&stt->list, &kvm->arch.spapr_tce_tables);
340 	else
341 		kvm_put_kvm_no_destroy(kvm);
342 
343 	mutex_unlock(&kvm->lock);
344 
345 	if (ret >= 0)
346 		return ret;
347 
348 	kfree(stt);
349  fail_acct:
350 	account_locked_vm(mm, kvmppc_stt_pages(npages), false);
351 	return ret;
352 }
353 
354 static long kvmppc_tce_to_ua(struct kvm *kvm, unsigned long tce,
355 		unsigned long *ua)
356 {
357 	unsigned long gfn = tce >> PAGE_SHIFT;
358 	struct kvm_memory_slot *memslot;
359 
360 	memslot = __gfn_to_memslot(kvm_memslots(kvm), gfn);
361 	if (!memslot)
362 		return -EINVAL;
363 
364 	*ua = __gfn_to_hva_memslot(memslot, gfn) |
365 		(tce & ~(PAGE_MASK | TCE_PCI_READ | TCE_PCI_WRITE));
366 
367 	return 0;
368 }
369 
370 static long kvmppc_tce_validate(struct kvmppc_spapr_tce_table *stt,
371 		unsigned long tce)
372 {
373 	unsigned long gpa = tce & ~(TCE_PCI_READ | TCE_PCI_WRITE);
374 	enum dma_data_direction dir = iommu_tce_direction(tce);
375 	struct kvmppc_spapr_tce_iommu_table *stit;
376 	unsigned long ua = 0;
377 
378 	/* Allow userspace to poison TCE table */
379 	if (dir == DMA_NONE)
380 		return H_SUCCESS;
381 
382 	if (iommu_tce_check_gpa(stt->page_shift, gpa))
383 		return H_TOO_HARD;
384 
385 	if (kvmppc_tce_to_ua(stt->kvm, tce, &ua))
386 		return H_TOO_HARD;
387 
388 	rcu_read_lock();
389 	list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
390 		unsigned long hpa = 0;
391 		struct mm_iommu_table_group_mem_t *mem;
392 		long shift = stit->tbl->it_page_shift;
393 
394 		mem = mm_iommu_lookup(stt->kvm->mm, ua, 1ULL << shift);
395 		if (!mem || mm_iommu_ua_to_hpa(mem, ua, shift, &hpa)) {
396 			rcu_read_unlock();
397 			return H_TOO_HARD;
398 		}
399 	}
400 	rcu_read_unlock();
401 
402 	return H_SUCCESS;
403 }
404 
405 /*
406  * Handles TCE requests for emulated devices.
407  * Puts guest TCE values to the table and expects user space to convert them.
408  * Cannot fail so kvmppc_tce_validate must be called before it.
409  */
410 static void kvmppc_tce_put(struct kvmppc_spapr_tce_table *stt,
411 		unsigned long idx, unsigned long tce)
412 {
413 	struct page *page;
414 	u64 *tbl;
415 	unsigned long sttpage;
416 
417 	idx -= stt->offset;
418 	sttpage = idx / TCES_PER_PAGE;
419 	page = stt->pages[sttpage];
420 
421 	if (!page) {
422 		/* We allow any TCE, not just with read|write permissions */
423 		if (!tce)
424 			return;
425 
426 		page = kvm_spapr_get_tce_page(stt, sttpage);
427 		if (!page)
428 			return;
429 	}
430 	tbl = page_to_virt(page);
431 
432 	tbl[idx % TCES_PER_PAGE] = tce;
433 }
434 
435 static void kvmppc_clear_tce(struct mm_struct *mm, struct kvmppc_spapr_tce_table *stt,
436 		struct iommu_table *tbl, unsigned long entry)
437 {
438 	unsigned long i;
439 	unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
440 	unsigned long io_entry = entry << (stt->page_shift - tbl->it_page_shift);
441 
442 	for (i = 0; i < subpages; ++i) {
443 		unsigned long hpa = 0;
444 		enum dma_data_direction dir = DMA_NONE;
445 
446 		iommu_tce_xchg_no_kill(mm, tbl, io_entry + i, &hpa, &dir);
447 	}
448 }
449 
450 static long kvmppc_tce_iommu_mapped_dec(struct kvm *kvm,
451 		struct iommu_table *tbl, unsigned long entry)
452 {
453 	struct mm_iommu_table_group_mem_t *mem = NULL;
454 	const unsigned long pgsize = 1ULL << tbl->it_page_shift;
455 	__be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY_RO(tbl, entry);
456 
457 	if (!pua)
458 		return H_SUCCESS;
459 
460 	mem = mm_iommu_lookup(kvm->mm, be64_to_cpu(*pua), pgsize);
461 	if (!mem)
462 		return H_TOO_HARD;
463 
464 	mm_iommu_mapped_dec(mem);
465 
466 	*pua = cpu_to_be64(0);
467 
468 	return H_SUCCESS;
469 }
470 
471 static long kvmppc_tce_iommu_do_unmap(struct kvm *kvm,
472 		struct iommu_table *tbl, unsigned long entry)
473 {
474 	enum dma_data_direction dir = DMA_NONE;
475 	unsigned long hpa = 0;
476 	long ret;
477 
478 	if (WARN_ON_ONCE(iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa,
479 					&dir)))
480 		return H_TOO_HARD;
481 
482 	if (dir == DMA_NONE)
483 		return H_SUCCESS;
484 
485 	ret = kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
486 	if (ret != H_SUCCESS)
487 		iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa, &dir);
488 
489 	return ret;
490 }
491 
492 static long kvmppc_tce_iommu_unmap(struct kvm *kvm,
493 		struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
494 		unsigned long entry)
495 {
496 	unsigned long i, ret = H_SUCCESS;
497 	unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
498 	unsigned long io_entry = entry * subpages;
499 
500 	for (i = 0; i < subpages; ++i) {
501 		ret = kvmppc_tce_iommu_do_unmap(kvm, tbl, io_entry + i);
502 		if (ret != H_SUCCESS)
503 			break;
504 	}
505 
506 	iommu_tce_kill(tbl, io_entry, subpages);
507 
508 	return ret;
509 }
510 
511 static long kvmppc_tce_iommu_do_map(struct kvm *kvm, struct iommu_table *tbl,
512 		unsigned long entry, unsigned long ua,
513 		enum dma_data_direction dir)
514 {
515 	long ret;
516 	unsigned long hpa;
517 	__be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
518 	struct mm_iommu_table_group_mem_t *mem;
519 
520 	if (!pua)
521 		/* it_userspace allocation might be delayed */
522 		return H_TOO_HARD;
523 
524 	mem = mm_iommu_lookup(kvm->mm, ua, 1ULL << tbl->it_page_shift);
525 	if (!mem)
526 		/* This only handles v2 IOMMU type, v1 is handled via ioctl() */
527 		return H_TOO_HARD;
528 
529 	if (WARN_ON_ONCE(mm_iommu_ua_to_hpa(mem, ua, tbl->it_page_shift, &hpa)))
530 		return H_TOO_HARD;
531 
532 	if (mm_iommu_mapped_inc(mem))
533 		return H_TOO_HARD;
534 
535 	ret = iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa, &dir);
536 	if (WARN_ON_ONCE(ret)) {
537 		mm_iommu_mapped_dec(mem);
538 		return H_TOO_HARD;
539 	}
540 
541 	if (dir != DMA_NONE)
542 		kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
543 
544 	*pua = cpu_to_be64(ua);
545 
546 	return 0;
547 }
548 
549 static long kvmppc_tce_iommu_map(struct kvm *kvm,
550 		struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
551 		unsigned long entry, unsigned long ua,
552 		enum dma_data_direction dir)
553 {
554 	unsigned long i, pgoff, ret = H_SUCCESS;
555 	unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
556 	unsigned long io_entry = entry * subpages;
557 
558 	for (i = 0, pgoff = 0; i < subpages;
559 			++i, pgoff += IOMMU_PAGE_SIZE(tbl)) {
560 
561 		ret = kvmppc_tce_iommu_do_map(kvm, tbl,
562 				io_entry + i, ua + pgoff, dir);
563 		if (ret != H_SUCCESS)
564 			break;
565 	}
566 
567 	iommu_tce_kill(tbl, io_entry, subpages);
568 
569 	return ret;
570 }
571 
572 long kvmppc_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
573 		      unsigned long ioba, unsigned long tce)
574 {
575 	struct kvmppc_spapr_tce_table *stt;
576 	long ret, idx;
577 	struct kvmppc_spapr_tce_iommu_table *stit;
578 	unsigned long entry, ua = 0;
579 	enum dma_data_direction dir;
580 
581 	/* udbg_printf("H_PUT_TCE(): liobn=0x%lx ioba=0x%lx, tce=0x%lx\n", */
582 	/* 	    liobn, ioba, tce); */
583 
584 	stt = kvmppc_find_table(vcpu->kvm, liobn);
585 	if (!stt)
586 		return H_TOO_HARD;
587 
588 	ret = kvmppc_ioba_validate(stt, ioba, 1);
589 	if (ret != H_SUCCESS)
590 		return ret;
591 
592 	idx = srcu_read_lock(&vcpu->kvm->srcu);
593 
594 	ret = kvmppc_tce_validate(stt, tce);
595 	if (ret != H_SUCCESS)
596 		goto unlock_exit;
597 
598 	dir = iommu_tce_direction(tce);
599 
600 	if ((dir != DMA_NONE) && kvmppc_tce_to_ua(vcpu->kvm, tce, &ua)) {
601 		ret = H_PARAMETER;
602 		goto unlock_exit;
603 	}
604 
605 	entry = ioba >> stt->page_shift;
606 
607 	list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
608 		if (dir == DMA_NONE)
609 			ret = kvmppc_tce_iommu_unmap(vcpu->kvm, stt,
610 					stit->tbl, entry);
611 		else
612 			ret = kvmppc_tce_iommu_map(vcpu->kvm, stt, stit->tbl,
613 					entry, ua, dir);
614 
615 
616 		if (ret != H_SUCCESS) {
617 			kvmppc_clear_tce(vcpu->kvm->mm, stt, stit->tbl, entry);
618 			goto unlock_exit;
619 		}
620 	}
621 
622 	kvmppc_tce_put(stt, entry, tce);
623 
624 unlock_exit:
625 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
626 
627 	return ret;
628 }
629 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce);
630 
631 long kvmppc_h_put_tce_indirect(struct kvm_vcpu *vcpu,
632 		unsigned long liobn, unsigned long ioba,
633 		unsigned long tce_list, unsigned long npages)
634 {
635 	struct kvmppc_spapr_tce_table *stt;
636 	long i, ret = H_SUCCESS, idx;
637 	unsigned long entry, ua = 0;
638 	u64 __user *tces;
639 	u64 tce;
640 	struct kvmppc_spapr_tce_iommu_table *stit;
641 
642 	stt = kvmppc_find_table(vcpu->kvm, liobn);
643 	if (!stt)
644 		return H_TOO_HARD;
645 
646 	entry = ioba >> stt->page_shift;
647 	/*
648 	 * SPAPR spec says that the maximum size of the list is 512 TCEs
649 	 * so the whole table fits in 4K page
650 	 */
651 	if (npages > 512)
652 		return H_PARAMETER;
653 
654 	if (tce_list & (SZ_4K - 1))
655 		return H_PARAMETER;
656 
657 	ret = kvmppc_ioba_validate(stt, ioba, npages);
658 	if (ret != H_SUCCESS)
659 		return ret;
660 
661 	idx = srcu_read_lock(&vcpu->kvm->srcu);
662 	if (kvmppc_tce_to_ua(vcpu->kvm, tce_list, &ua)) {
663 		ret = H_TOO_HARD;
664 		goto unlock_exit;
665 	}
666 	tces = (u64 __user *) ua;
667 
668 	for (i = 0; i < npages; ++i) {
669 		if (get_user(tce, tces + i)) {
670 			ret = H_TOO_HARD;
671 			goto unlock_exit;
672 		}
673 		tce = be64_to_cpu(tce);
674 
675 		ret = kvmppc_tce_validate(stt, tce);
676 		if (ret != H_SUCCESS)
677 			goto unlock_exit;
678 	}
679 
680 	for (i = 0; i < npages; ++i) {
681 		/*
682 		 * This looks unsafe, because we validate, then regrab
683 		 * the TCE from userspace which could have been changed by
684 		 * another thread.
685 		 *
686 		 * But it actually is safe, because the relevant checks will be
687 		 * re-executed in the following code.  If userspace tries to
688 		 * change this dodgily it will result in a messier failure mode
689 		 * but won't threaten the host.
690 		 */
691 		if (get_user(tce, tces + i)) {
692 			ret = H_TOO_HARD;
693 			goto unlock_exit;
694 		}
695 		tce = be64_to_cpu(tce);
696 
697 		if (kvmppc_tce_to_ua(vcpu->kvm, tce, &ua)) {
698 			ret = H_PARAMETER;
699 			goto unlock_exit;
700 		}
701 
702 		list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
703 			ret = kvmppc_tce_iommu_map(vcpu->kvm, stt,
704 					stit->tbl, entry + i, ua,
705 					iommu_tce_direction(tce));
706 
707 			if (ret != H_SUCCESS) {
708 				kvmppc_clear_tce(vcpu->kvm->mm, stt, stit->tbl,
709 						 entry + i);
710 				goto unlock_exit;
711 			}
712 		}
713 
714 		kvmppc_tce_put(stt, entry + i, tce);
715 	}
716 
717 unlock_exit:
718 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
719 
720 	return ret;
721 }
722 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce_indirect);
723 
724 long kvmppc_h_stuff_tce(struct kvm_vcpu *vcpu,
725 		unsigned long liobn, unsigned long ioba,
726 		unsigned long tce_value, unsigned long npages)
727 {
728 	struct kvmppc_spapr_tce_table *stt;
729 	long i, ret;
730 	struct kvmppc_spapr_tce_iommu_table *stit;
731 
732 	stt = kvmppc_find_table(vcpu->kvm, liobn);
733 	if (!stt)
734 		return H_TOO_HARD;
735 
736 	ret = kvmppc_ioba_validate(stt, ioba, npages);
737 	if (ret != H_SUCCESS)
738 		return ret;
739 
740 	/* Check permission bits only to allow userspace poison TCE for debug */
741 	if (tce_value & (TCE_PCI_WRITE | TCE_PCI_READ))
742 		return H_PARAMETER;
743 
744 	list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
745 		unsigned long entry = ioba >> stt->page_shift;
746 
747 		for (i = 0; i < npages; ++i) {
748 			ret = kvmppc_tce_iommu_unmap(vcpu->kvm, stt,
749 					stit->tbl, entry + i);
750 
751 			if (ret == H_SUCCESS)
752 				continue;
753 
754 			if (ret == H_TOO_HARD)
755 				return ret;
756 
757 			WARN_ON_ONCE(1);
758 			kvmppc_clear_tce(vcpu->kvm->mm, stt, stit->tbl, entry + i);
759 		}
760 	}
761 
762 	for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift))
763 		kvmppc_tce_put(stt, ioba >> stt->page_shift, tce_value);
764 
765 	return ret;
766 }
767 EXPORT_SYMBOL_GPL(kvmppc_h_stuff_tce);
768 
769 long kvmppc_h_get_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
770 		      unsigned long ioba)
771 {
772 	struct kvmppc_spapr_tce_table *stt;
773 	long ret;
774 	unsigned long idx;
775 	struct page *page;
776 	u64 *tbl;
777 
778 	stt = kvmppc_find_table(vcpu->kvm, liobn);
779 	if (!stt)
780 		return H_TOO_HARD;
781 
782 	ret = kvmppc_ioba_validate(stt, ioba, 1);
783 	if (ret != H_SUCCESS)
784 		return ret;
785 
786 	idx = (ioba >> stt->page_shift) - stt->offset;
787 	page = stt->pages[idx / TCES_PER_PAGE];
788 	if (!page) {
789 		vcpu->arch.regs.gpr[4] = 0;
790 		return H_SUCCESS;
791 	}
792 	tbl = (u64 *)page_address(page);
793 
794 	vcpu->arch.regs.gpr[4] = tbl[idx % TCES_PER_PAGE];
795 
796 	return H_SUCCESS;
797 }
798 EXPORT_SYMBOL_GPL(kvmppc_h_get_tce);
799