xref: /openbmc/linux/arch/powerpc/kvm/book3s_64_vio.c (revision a28b2ed9)
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 unsigned long kvmppc_tce_pages(unsigned long iommu_pages)
36 {
37 	return ALIGN(iommu_pages * sizeof(u64), PAGE_SIZE) / PAGE_SIZE;
38 }
39 
40 static unsigned long kvmppc_stt_pages(unsigned long tce_pages)
41 {
42 	unsigned long stt_bytes = sizeof(struct kvmppc_spapr_tce_table) +
43 			(tce_pages * sizeof(struct page *));
44 
45 	return tce_pages + ALIGN(stt_bytes, PAGE_SIZE) / PAGE_SIZE;
46 }
47 
48 static void kvm_spapr_tce_iommu_table_free(struct rcu_head *head)
49 {
50 	struct kvmppc_spapr_tce_iommu_table *stit = container_of(head,
51 			struct kvmppc_spapr_tce_iommu_table, rcu);
52 
53 	iommu_tce_table_put(stit->tbl);
54 
55 	kfree(stit);
56 }
57 
58 static void kvm_spapr_tce_liobn_put(struct kref *kref)
59 {
60 	struct kvmppc_spapr_tce_iommu_table *stit = container_of(kref,
61 			struct kvmppc_spapr_tce_iommu_table, kref);
62 
63 	list_del_rcu(&stit->next);
64 
65 	call_rcu(&stit->rcu, kvm_spapr_tce_iommu_table_free);
66 }
67 
68 extern void kvm_spapr_tce_release_iommu_group(struct kvm *kvm,
69 		struct iommu_group *grp)
70 {
71 	int i;
72 	struct kvmppc_spapr_tce_table *stt;
73 	struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
74 	struct iommu_table_group *table_group = NULL;
75 
76 	rcu_read_lock();
77 	list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
78 
79 		table_group = iommu_group_get_iommudata(grp);
80 		if (WARN_ON(!table_group))
81 			continue;
82 
83 		list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
84 			for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
85 				if (table_group->tables[i] != stit->tbl)
86 					continue;
87 
88 				kref_put(&stit->kref, kvm_spapr_tce_liobn_put);
89 			}
90 		}
91 		cond_resched_rcu();
92 	}
93 	rcu_read_unlock();
94 }
95 
96 extern long kvm_spapr_tce_attach_iommu_group(struct kvm *kvm, int tablefd,
97 		struct iommu_group *grp)
98 {
99 	struct kvmppc_spapr_tce_table *stt = NULL;
100 	bool found = false;
101 	struct iommu_table *tbl = NULL;
102 	struct iommu_table_group *table_group;
103 	long i;
104 	struct kvmppc_spapr_tce_iommu_table *stit;
105 	struct fd f;
106 
107 	f = fdget(tablefd);
108 	if (!f.file)
109 		return -EBADF;
110 
111 	rcu_read_lock();
112 	list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
113 		if (stt == f.file->private_data) {
114 			found = true;
115 			break;
116 		}
117 	}
118 	rcu_read_unlock();
119 
120 	fdput(f);
121 
122 	if (!found)
123 		return -EINVAL;
124 
125 	table_group = iommu_group_get_iommudata(grp);
126 	if (WARN_ON(!table_group))
127 		return -EFAULT;
128 
129 	for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
130 		struct iommu_table *tbltmp = table_group->tables[i];
131 
132 		if (!tbltmp)
133 			continue;
134 		/* Make sure hardware table parameters are compatible */
135 		if ((tbltmp->it_page_shift <= stt->page_shift) &&
136 				(tbltmp->it_offset << tbltmp->it_page_shift ==
137 				 stt->offset << stt->page_shift) &&
138 				(tbltmp->it_size << tbltmp->it_page_shift >=
139 				 stt->size << stt->page_shift)) {
140 			/*
141 			 * Reference the table to avoid races with
142 			 * add/remove DMA windows.
143 			 */
144 			tbl = iommu_tce_table_get(tbltmp);
145 			break;
146 		}
147 	}
148 	if (!tbl)
149 		return -EINVAL;
150 
151 	rcu_read_lock();
152 	list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
153 		if (tbl != stit->tbl)
154 			continue;
155 
156 		if (!kref_get_unless_zero(&stit->kref)) {
157 			/* stit is being destroyed */
158 			iommu_tce_table_put(tbl);
159 			rcu_read_unlock();
160 			return -ENOTTY;
161 		}
162 		/*
163 		 * The table is already known to this KVM, we just increased
164 		 * its KVM reference counter and can return.
165 		 */
166 		rcu_read_unlock();
167 		return 0;
168 	}
169 	rcu_read_unlock();
170 
171 	stit = kzalloc(sizeof(*stit), GFP_KERNEL);
172 	if (!stit) {
173 		iommu_tce_table_put(tbl);
174 		return -ENOMEM;
175 	}
176 
177 	stit->tbl = tbl;
178 	kref_init(&stit->kref);
179 
180 	list_add_rcu(&stit->next, &stt->iommu_tables);
181 
182 	return 0;
183 }
184 
185 static void release_spapr_tce_table(struct rcu_head *head)
186 {
187 	struct kvmppc_spapr_tce_table *stt = container_of(head,
188 			struct kvmppc_spapr_tce_table, rcu);
189 	unsigned long i, npages = kvmppc_tce_pages(stt->size);
190 
191 	for (i = 0; i < npages; i++)
192 		if (stt->pages[i])
193 			__free_page(stt->pages[i]);
194 
195 	kfree(stt);
196 }
197 
198 static struct page *kvm_spapr_get_tce_page(struct kvmppc_spapr_tce_table *stt,
199 		unsigned long sttpage)
200 {
201 	struct page *page = stt->pages[sttpage];
202 
203 	if (page)
204 		return page;
205 
206 	mutex_lock(&stt->alloc_lock);
207 	page = stt->pages[sttpage];
208 	if (!page) {
209 		page = alloc_page(GFP_KERNEL | __GFP_ZERO);
210 		WARN_ON_ONCE(!page);
211 		if (page)
212 			stt->pages[sttpage] = page;
213 	}
214 	mutex_unlock(&stt->alloc_lock);
215 
216 	return page;
217 }
218 
219 static vm_fault_t kvm_spapr_tce_fault(struct vm_fault *vmf)
220 {
221 	struct kvmppc_spapr_tce_table *stt = vmf->vma->vm_file->private_data;
222 	struct page *page;
223 
224 	if (vmf->pgoff >= kvmppc_tce_pages(stt->size))
225 		return VM_FAULT_SIGBUS;
226 
227 	page = kvm_spapr_get_tce_page(stt, vmf->pgoff);
228 	if (!page)
229 		return VM_FAULT_OOM;
230 
231 	get_page(page);
232 	vmf->page = page;
233 	return 0;
234 }
235 
236 static const struct vm_operations_struct kvm_spapr_tce_vm_ops = {
237 	.fault = kvm_spapr_tce_fault,
238 };
239 
240 static int kvm_spapr_tce_mmap(struct file *file, struct vm_area_struct *vma)
241 {
242 	vma->vm_ops = &kvm_spapr_tce_vm_ops;
243 	return 0;
244 }
245 
246 static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
247 {
248 	struct kvmppc_spapr_tce_table *stt = filp->private_data;
249 	struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
250 	struct kvm *kvm = stt->kvm;
251 
252 	mutex_lock(&kvm->lock);
253 	list_del_rcu(&stt->list);
254 	mutex_unlock(&kvm->lock);
255 
256 	list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
257 		WARN_ON(!kref_read(&stit->kref));
258 		while (1) {
259 			if (kref_put(&stit->kref, kvm_spapr_tce_liobn_put))
260 				break;
261 		}
262 	}
263 
264 	account_locked_vm(kvm->mm,
265 		kvmppc_stt_pages(kvmppc_tce_pages(stt->size)), false);
266 
267 	kvm_put_kvm(stt->kvm);
268 
269 	call_rcu(&stt->rcu, release_spapr_tce_table);
270 
271 	return 0;
272 }
273 
274 static const struct file_operations kvm_spapr_tce_fops = {
275 	.mmap           = kvm_spapr_tce_mmap,
276 	.release	= kvm_spapr_tce_release,
277 };
278 
279 long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
280 				   struct kvm_create_spapr_tce_64 *args)
281 {
282 	struct kvmppc_spapr_tce_table *stt = NULL;
283 	struct kvmppc_spapr_tce_table *siter;
284 	struct mm_struct *mm = kvm->mm;
285 	unsigned long npages, size = args->size;
286 	int ret = -ENOMEM;
287 
288 	if (!args->size || args->page_shift < 12 || args->page_shift > 34 ||
289 		(args->offset + args->size > (ULLONG_MAX >> args->page_shift)))
290 		return -EINVAL;
291 
292 	npages = kvmppc_tce_pages(size);
293 	ret = account_locked_vm(mm, kvmppc_stt_pages(npages), true);
294 	if (ret)
295 		return ret;
296 
297 	ret = -ENOMEM;
298 	stt = kzalloc(sizeof(*stt) + npages * sizeof(struct page *),
299 		      GFP_KERNEL);
300 	if (!stt)
301 		goto fail_acct;
302 
303 	stt->liobn = args->liobn;
304 	stt->page_shift = args->page_shift;
305 	stt->offset = args->offset;
306 	stt->size = size;
307 	stt->kvm = kvm;
308 	mutex_init(&stt->alloc_lock);
309 	INIT_LIST_HEAD_RCU(&stt->iommu_tables);
310 
311 	mutex_lock(&kvm->lock);
312 
313 	/* Check this LIOBN hasn't been previously allocated */
314 	ret = 0;
315 	list_for_each_entry(siter, &kvm->arch.spapr_tce_tables, list) {
316 		if (siter->liobn == args->liobn) {
317 			ret = -EBUSY;
318 			break;
319 		}
320 	}
321 
322 	kvm_get_kvm(kvm);
323 	if (!ret)
324 		ret = anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops,
325 				       stt, O_RDWR | O_CLOEXEC);
326 
327 	if (ret >= 0)
328 		list_add_rcu(&stt->list, &kvm->arch.spapr_tce_tables);
329 	else
330 		kvm_put_kvm_no_destroy(kvm);
331 
332 	mutex_unlock(&kvm->lock);
333 
334 	if (ret >= 0)
335 		return ret;
336 
337 	kfree(stt);
338  fail_acct:
339 	account_locked_vm(mm, kvmppc_stt_pages(npages), false);
340 	return ret;
341 }
342 
343 static long kvmppc_tce_to_ua(struct kvm *kvm, unsigned long tce,
344 		unsigned long *ua)
345 {
346 	unsigned long gfn = tce >> PAGE_SHIFT;
347 	struct kvm_memory_slot *memslot;
348 
349 	memslot = search_memslots(kvm_memslots(kvm), gfn);
350 	if (!memslot)
351 		return -EINVAL;
352 
353 	*ua = __gfn_to_hva_memslot(memslot, gfn) |
354 		(tce & ~(PAGE_MASK | TCE_PCI_READ | TCE_PCI_WRITE));
355 
356 	return 0;
357 }
358 
359 static long kvmppc_tce_validate(struct kvmppc_spapr_tce_table *stt,
360 		unsigned long tce)
361 {
362 	unsigned long gpa = tce & ~(TCE_PCI_READ | TCE_PCI_WRITE);
363 	enum dma_data_direction dir = iommu_tce_direction(tce);
364 	struct kvmppc_spapr_tce_iommu_table *stit;
365 	unsigned long ua = 0;
366 
367 	/* Allow userspace to poison TCE table */
368 	if (dir == DMA_NONE)
369 		return H_SUCCESS;
370 
371 	if (iommu_tce_check_gpa(stt->page_shift, gpa))
372 		return H_TOO_HARD;
373 
374 	if (kvmppc_tce_to_ua(stt->kvm, tce, &ua))
375 		return H_TOO_HARD;
376 
377 	rcu_read_lock();
378 	list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
379 		unsigned long hpa = 0;
380 		struct mm_iommu_table_group_mem_t *mem;
381 		long shift = stit->tbl->it_page_shift;
382 
383 		mem = mm_iommu_lookup(stt->kvm->mm, ua, 1ULL << shift);
384 		if (!mem || mm_iommu_ua_to_hpa(mem, ua, shift, &hpa)) {
385 			rcu_read_unlock();
386 			return H_TOO_HARD;
387 		}
388 	}
389 	rcu_read_unlock();
390 
391 	return H_SUCCESS;
392 }
393 
394 /*
395  * Handles TCE requests for emulated devices.
396  * Puts guest TCE values to the table and expects user space to convert them.
397  * Cannot fail so kvmppc_tce_validate must be called before it.
398  */
399 static void kvmppc_tce_put(struct kvmppc_spapr_tce_table *stt,
400 		unsigned long idx, unsigned long tce)
401 {
402 	struct page *page;
403 	u64 *tbl;
404 	unsigned long sttpage;
405 
406 	idx -= stt->offset;
407 	sttpage = idx / TCES_PER_PAGE;
408 	page = stt->pages[sttpage];
409 
410 	if (!page) {
411 		/* We allow any TCE, not just with read|write permissions */
412 		if (!tce)
413 			return;
414 
415 		page = kvm_spapr_get_tce_page(stt, sttpage);
416 		if (!page)
417 			return;
418 	}
419 	tbl = page_to_virt(page);
420 
421 	tbl[idx % TCES_PER_PAGE] = tce;
422 }
423 
424 static void kvmppc_clear_tce(struct mm_struct *mm, struct iommu_table *tbl,
425 		unsigned long entry)
426 {
427 	unsigned long hpa = 0;
428 	enum dma_data_direction dir = DMA_NONE;
429 
430 	iommu_tce_xchg_no_kill(mm, tbl, entry, &hpa, &dir);
431 }
432 
433 static long kvmppc_tce_iommu_mapped_dec(struct kvm *kvm,
434 		struct iommu_table *tbl, unsigned long entry)
435 {
436 	struct mm_iommu_table_group_mem_t *mem = NULL;
437 	const unsigned long pgsize = 1ULL << tbl->it_page_shift;
438 	__be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY_RO(tbl, entry);
439 
440 	if (!pua)
441 		return H_SUCCESS;
442 
443 	mem = mm_iommu_lookup(kvm->mm, be64_to_cpu(*pua), pgsize);
444 	if (!mem)
445 		return H_TOO_HARD;
446 
447 	mm_iommu_mapped_dec(mem);
448 
449 	*pua = cpu_to_be64(0);
450 
451 	return H_SUCCESS;
452 }
453 
454 static long kvmppc_tce_iommu_do_unmap(struct kvm *kvm,
455 		struct iommu_table *tbl, unsigned long entry)
456 {
457 	enum dma_data_direction dir = DMA_NONE;
458 	unsigned long hpa = 0;
459 	long ret;
460 
461 	if (WARN_ON_ONCE(iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa,
462 					&dir)))
463 		return H_TOO_HARD;
464 
465 	if (dir == DMA_NONE)
466 		return H_SUCCESS;
467 
468 	ret = kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
469 	if (ret != H_SUCCESS)
470 		iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa, &dir);
471 
472 	return ret;
473 }
474 
475 static long kvmppc_tce_iommu_unmap(struct kvm *kvm,
476 		struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
477 		unsigned long entry)
478 {
479 	unsigned long i, ret = H_SUCCESS;
480 	unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
481 	unsigned long io_entry = entry * subpages;
482 
483 	for (i = 0; i < subpages; ++i) {
484 		ret = kvmppc_tce_iommu_do_unmap(kvm, tbl, io_entry + i);
485 		if (ret != H_SUCCESS)
486 			break;
487 	}
488 
489 	return ret;
490 }
491 
492 long kvmppc_tce_iommu_do_map(struct kvm *kvm, struct iommu_table *tbl,
493 		unsigned long entry, unsigned long ua,
494 		enum dma_data_direction dir)
495 {
496 	long ret;
497 	unsigned long hpa;
498 	__be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
499 	struct mm_iommu_table_group_mem_t *mem;
500 
501 	if (!pua)
502 		/* it_userspace allocation might be delayed */
503 		return H_TOO_HARD;
504 
505 	mem = mm_iommu_lookup(kvm->mm, ua, 1ULL << tbl->it_page_shift);
506 	if (!mem)
507 		/* This only handles v2 IOMMU type, v1 is handled via ioctl() */
508 		return H_TOO_HARD;
509 
510 	if (WARN_ON_ONCE(mm_iommu_ua_to_hpa(mem, ua, tbl->it_page_shift, &hpa)))
511 		return H_TOO_HARD;
512 
513 	if (mm_iommu_mapped_inc(mem))
514 		return H_TOO_HARD;
515 
516 	ret = iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa, &dir);
517 	if (WARN_ON_ONCE(ret)) {
518 		mm_iommu_mapped_dec(mem);
519 		return H_TOO_HARD;
520 	}
521 
522 	if (dir != DMA_NONE)
523 		kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
524 
525 	*pua = cpu_to_be64(ua);
526 
527 	return 0;
528 }
529 
530 static long kvmppc_tce_iommu_map(struct kvm *kvm,
531 		struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
532 		unsigned long entry, unsigned long ua,
533 		enum dma_data_direction dir)
534 {
535 	unsigned long i, pgoff, ret = H_SUCCESS;
536 	unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
537 	unsigned long io_entry = entry * subpages;
538 
539 	for (i = 0, pgoff = 0; i < subpages;
540 			++i, pgoff += IOMMU_PAGE_SIZE(tbl)) {
541 
542 		ret = kvmppc_tce_iommu_do_map(kvm, tbl,
543 				io_entry + i, ua + pgoff, dir);
544 		if (ret != H_SUCCESS)
545 			break;
546 	}
547 
548 	return ret;
549 }
550 
551 long kvmppc_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
552 		      unsigned long ioba, unsigned long tce)
553 {
554 	struct kvmppc_spapr_tce_table *stt;
555 	long ret, idx;
556 	struct kvmppc_spapr_tce_iommu_table *stit;
557 	unsigned long entry, ua = 0;
558 	enum dma_data_direction dir;
559 
560 	/* udbg_printf("H_PUT_TCE(): liobn=0x%lx ioba=0x%lx, tce=0x%lx\n", */
561 	/* 	    liobn, ioba, tce); */
562 
563 	stt = kvmppc_find_table(vcpu->kvm, liobn);
564 	if (!stt)
565 		return H_TOO_HARD;
566 
567 	ret = kvmppc_ioba_validate(stt, ioba, 1);
568 	if (ret != H_SUCCESS)
569 		return ret;
570 
571 	idx = srcu_read_lock(&vcpu->kvm->srcu);
572 
573 	ret = kvmppc_tce_validate(stt, tce);
574 	if (ret != H_SUCCESS)
575 		goto unlock_exit;
576 
577 	dir = iommu_tce_direction(tce);
578 
579 	if ((dir != DMA_NONE) && kvmppc_tce_to_ua(vcpu->kvm, tce, &ua)) {
580 		ret = H_PARAMETER;
581 		goto unlock_exit;
582 	}
583 
584 	entry = ioba >> stt->page_shift;
585 
586 	list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
587 		if (dir == DMA_NONE)
588 			ret = kvmppc_tce_iommu_unmap(vcpu->kvm, stt,
589 					stit->tbl, entry);
590 		else
591 			ret = kvmppc_tce_iommu_map(vcpu->kvm, stt, stit->tbl,
592 					entry, ua, dir);
593 
594 		iommu_tce_kill(stit->tbl, entry, 1);
595 
596 		if (ret != H_SUCCESS) {
597 			kvmppc_clear_tce(vcpu->kvm->mm, stit->tbl, entry);
598 			goto unlock_exit;
599 		}
600 	}
601 
602 	kvmppc_tce_put(stt, entry, tce);
603 
604 unlock_exit:
605 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
606 
607 	return ret;
608 }
609 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce);
610 
611 long kvmppc_h_put_tce_indirect(struct kvm_vcpu *vcpu,
612 		unsigned long liobn, unsigned long ioba,
613 		unsigned long tce_list, unsigned long npages)
614 {
615 	struct kvmppc_spapr_tce_table *stt;
616 	long i, ret = H_SUCCESS, idx;
617 	unsigned long entry, ua = 0;
618 	u64 __user *tces;
619 	u64 tce;
620 	struct kvmppc_spapr_tce_iommu_table *stit;
621 
622 	stt = kvmppc_find_table(vcpu->kvm, liobn);
623 	if (!stt)
624 		return H_TOO_HARD;
625 
626 	entry = ioba >> stt->page_shift;
627 	/*
628 	 * SPAPR spec says that the maximum size of the list is 512 TCEs
629 	 * so the whole table fits in 4K page
630 	 */
631 	if (npages > 512)
632 		return H_PARAMETER;
633 
634 	if (tce_list & (SZ_4K - 1))
635 		return H_PARAMETER;
636 
637 	ret = kvmppc_ioba_validate(stt, ioba, npages);
638 	if (ret != H_SUCCESS)
639 		return ret;
640 
641 	idx = srcu_read_lock(&vcpu->kvm->srcu);
642 	if (kvmppc_tce_to_ua(vcpu->kvm, tce_list, &ua)) {
643 		ret = H_TOO_HARD;
644 		goto unlock_exit;
645 	}
646 	tces = (u64 __user *) ua;
647 
648 	for (i = 0; i < npages; ++i) {
649 		if (get_user(tce, tces + i)) {
650 			ret = H_TOO_HARD;
651 			goto unlock_exit;
652 		}
653 		tce = be64_to_cpu(tce);
654 
655 		ret = kvmppc_tce_validate(stt, tce);
656 		if (ret != H_SUCCESS)
657 			goto unlock_exit;
658 	}
659 
660 	for (i = 0; i < npages; ++i) {
661 		/*
662 		 * This looks unsafe, because we validate, then regrab
663 		 * the TCE from userspace which could have been changed by
664 		 * another thread.
665 		 *
666 		 * But it actually is safe, because the relevant checks will be
667 		 * re-executed in the following code.  If userspace tries to
668 		 * change this dodgily it will result in a messier failure mode
669 		 * but won't threaten the host.
670 		 */
671 		if (get_user(tce, tces + i)) {
672 			ret = H_TOO_HARD;
673 			goto invalidate_exit;
674 		}
675 		tce = be64_to_cpu(tce);
676 
677 		if (kvmppc_tce_to_ua(vcpu->kvm, tce, &ua)) {
678 			ret = H_PARAMETER;
679 			goto invalidate_exit;
680 		}
681 
682 		list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
683 			ret = kvmppc_tce_iommu_map(vcpu->kvm, stt,
684 					stit->tbl, entry + i, ua,
685 					iommu_tce_direction(tce));
686 
687 			if (ret != H_SUCCESS) {
688 				kvmppc_clear_tce(vcpu->kvm->mm, stit->tbl,
689 						entry);
690 				goto invalidate_exit;
691 			}
692 		}
693 
694 		kvmppc_tce_put(stt, entry + i, tce);
695 	}
696 
697 invalidate_exit:
698 	list_for_each_entry_lockless(stit, &stt->iommu_tables, next)
699 		iommu_tce_kill(stit->tbl, entry, npages);
700 
701 unlock_exit:
702 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
703 
704 	return ret;
705 }
706 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce_indirect);
707 
708 long kvmppc_h_stuff_tce(struct kvm_vcpu *vcpu,
709 		unsigned long liobn, unsigned long ioba,
710 		unsigned long tce_value, unsigned long npages)
711 {
712 	struct kvmppc_spapr_tce_table *stt;
713 	long i, ret;
714 	struct kvmppc_spapr_tce_iommu_table *stit;
715 
716 	stt = kvmppc_find_table(vcpu->kvm, liobn);
717 	if (!stt)
718 		return H_TOO_HARD;
719 
720 	ret = kvmppc_ioba_validate(stt, ioba, npages);
721 	if (ret != H_SUCCESS)
722 		return ret;
723 
724 	/* Check permission bits only to allow userspace poison TCE for debug */
725 	if (tce_value & (TCE_PCI_WRITE | TCE_PCI_READ))
726 		return H_PARAMETER;
727 
728 	list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
729 		unsigned long entry = ioba >> stt->page_shift;
730 
731 		for (i = 0; i < npages; ++i) {
732 			ret = kvmppc_tce_iommu_unmap(vcpu->kvm, stt,
733 					stit->tbl, entry + i);
734 
735 			if (ret == H_SUCCESS)
736 				continue;
737 
738 			if (ret == H_TOO_HARD)
739 				goto invalidate_exit;
740 
741 			WARN_ON_ONCE(1);
742 			kvmppc_clear_tce(vcpu->kvm->mm, stit->tbl, entry);
743 		}
744 	}
745 
746 	for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift))
747 		kvmppc_tce_put(stt, ioba >> stt->page_shift, tce_value);
748 
749 invalidate_exit:
750 	list_for_each_entry_lockless(stit, &stt->iommu_tables, next)
751 		iommu_tce_kill(stit->tbl, ioba >> stt->page_shift, npages);
752 
753 	return ret;
754 }
755 EXPORT_SYMBOL_GPL(kvmppc_h_stuff_tce);
756