xref: /linux/arch/powerpc/kvm/book3s_64_vio.c (revision f7697db8b1b3e80d8cd5af071a5af42c8b445fc4)
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 #include <linux/rcupdate_wait.h>
24 
25 #include <asm/kvm_ppc.h>
26 #include <asm/kvm_book3s.h>
27 #include <asm/book3s/64/mmu-hash.h>
28 #include <asm/hvcall.h>
29 #include <asm/synch.h>
30 #include <asm/ppc-opcode.h>
31 #include <asm/udbg.h>
32 #include <asm/iommu.h>
33 #include <asm/tce.h>
34 #include <asm/mmu_context.h>
35 
36 static struct kvmppc_spapr_tce_table *kvmppc_find_table(struct kvm *kvm,
37 	unsigned long liobn)
38 {
39 	struct kvmppc_spapr_tce_table *stt;
40 
41 	list_for_each_entry_lockless(stt, &kvm->arch.spapr_tce_tables, list)
42 		if (stt->liobn == liobn)
43 			return stt;
44 
45 	return NULL;
46 }
47 
48 static unsigned long kvmppc_tce_pages(unsigned long iommu_pages)
49 {
50 	return ALIGN(iommu_pages * sizeof(u64), PAGE_SIZE) / PAGE_SIZE;
51 }
52 
53 static unsigned long kvmppc_stt_pages(unsigned long tce_pages)
54 {
55 	unsigned long stt_bytes = sizeof(struct kvmppc_spapr_tce_table) +
56 			(tce_pages * sizeof(struct page *));
57 
58 	return tce_pages + ALIGN(stt_bytes, PAGE_SIZE) / PAGE_SIZE;
59 }
60 
61 static void kvm_spapr_tce_iommu_table_free(struct rcu_head *head)
62 {
63 	struct kvmppc_spapr_tce_iommu_table *stit = container_of(head,
64 			struct kvmppc_spapr_tce_iommu_table, rcu);
65 
66 	iommu_tce_table_put(stit->tbl);
67 
68 	kfree(stit);
69 }
70 
71 static void kvm_spapr_tce_liobn_put(struct kref *kref)
72 {
73 	struct kvmppc_spapr_tce_iommu_table *stit = container_of(kref,
74 			struct kvmppc_spapr_tce_iommu_table, kref);
75 
76 	list_del_rcu(&stit->next);
77 
78 	call_rcu(&stit->rcu, kvm_spapr_tce_iommu_table_free);
79 }
80 
81 void kvm_spapr_tce_release_iommu_group(struct kvm *kvm,
82 				       struct iommu_group *grp)
83 {
84 	int i;
85 	struct kvmppc_spapr_tce_table *stt;
86 	struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
87 	struct iommu_table_group *table_group = NULL;
88 
89 	rcu_read_lock();
90 	list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
91 
92 		table_group = iommu_group_get_iommudata(grp);
93 		if (WARN_ON(!table_group))
94 			continue;
95 
96 		list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
97 			for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
98 				if (table_group->tables[i] != stit->tbl)
99 					continue;
100 
101 				kref_put(&stit->kref, kvm_spapr_tce_liobn_put);
102 			}
103 		}
104 		cond_resched_rcu();
105 	}
106 	rcu_read_unlock();
107 }
108 
109 long kvm_spapr_tce_attach_iommu_group(struct kvm *kvm, int tablefd,
110 				      struct iommu_group *grp)
111 {
112 	struct kvmppc_spapr_tce_table *stt = NULL;
113 	bool found = false;
114 	struct iommu_table *tbl = NULL;
115 	struct iommu_table_group *table_group;
116 	long i;
117 	struct kvmppc_spapr_tce_iommu_table *stit;
118 	struct fd f;
119 
120 	f = fdget(tablefd);
121 	if (!f.file)
122 		return -EBADF;
123 
124 	rcu_read_lock();
125 	list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
126 		if (stt == f.file->private_data) {
127 			found = true;
128 			break;
129 		}
130 	}
131 	rcu_read_unlock();
132 
133 	if (!found) {
134 		fdput(f);
135 		return -EINVAL;
136 	}
137 
138 	table_group = iommu_group_get_iommudata(grp);
139 	if (WARN_ON(!table_group)) {
140 		fdput(f);
141 		return -EFAULT;
142 	}
143 
144 	for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
145 		struct iommu_table *tbltmp = table_group->tables[i];
146 
147 		if (!tbltmp)
148 			continue;
149 		/* Make sure hardware table parameters are compatible */
150 		if ((tbltmp->it_page_shift <= stt->page_shift) &&
151 				(tbltmp->it_offset << tbltmp->it_page_shift ==
152 				 stt->offset << stt->page_shift) &&
153 				(tbltmp->it_size << tbltmp->it_page_shift >=
154 				 stt->size << stt->page_shift)) {
155 			/*
156 			 * Reference the table to avoid races with
157 			 * add/remove DMA windows.
158 			 */
159 			tbl = iommu_tce_table_get(tbltmp);
160 			break;
161 		}
162 	}
163 	if (!tbl) {
164 		fdput(f);
165 		return -EINVAL;
166 	}
167 
168 	rcu_read_lock();
169 	list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
170 		if (tbl != stit->tbl)
171 			continue;
172 
173 		if (!kref_get_unless_zero(&stit->kref)) {
174 			/* stit is being destroyed */
175 			iommu_tce_table_put(tbl);
176 			rcu_read_unlock();
177 			fdput(f);
178 			return -ENOTTY;
179 		}
180 		/*
181 		 * The table is already known to this KVM, we just increased
182 		 * its KVM reference counter and can return.
183 		 */
184 		rcu_read_unlock();
185 		fdput(f);
186 		return 0;
187 	}
188 	rcu_read_unlock();
189 
190 	stit = kzalloc(sizeof(*stit), GFP_KERNEL);
191 	if (!stit) {
192 		iommu_tce_table_put(tbl);
193 		fdput(f);
194 		return -ENOMEM;
195 	}
196 
197 	stit->tbl = tbl;
198 	kref_init(&stit->kref);
199 
200 	list_add_rcu(&stit->next, &stt->iommu_tables);
201 
202 	fdput(f);
203 	return 0;
204 }
205 
206 static void release_spapr_tce_table(struct rcu_head *head)
207 {
208 	struct kvmppc_spapr_tce_table *stt = container_of(head,
209 			struct kvmppc_spapr_tce_table, rcu);
210 	unsigned long i, npages = kvmppc_tce_pages(stt->size);
211 
212 	for (i = 0; i < npages; i++)
213 		if (stt->pages[i])
214 			__free_page(stt->pages[i]);
215 
216 	kfree(stt);
217 }
218 
219 static struct page *kvm_spapr_get_tce_page(struct kvmppc_spapr_tce_table *stt,
220 		unsigned long sttpage)
221 {
222 	struct page *page = stt->pages[sttpage];
223 
224 	if (page)
225 		return page;
226 
227 	mutex_lock(&stt->alloc_lock);
228 	page = stt->pages[sttpage];
229 	if (!page) {
230 		page = alloc_page(GFP_KERNEL | __GFP_ZERO);
231 		WARN_ON_ONCE(!page);
232 		if (page)
233 			stt->pages[sttpage] = page;
234 	}
235 	mutex_unlock(&stt->alloc_lock);
236 
237 	return page;
238 }
239 
240 static vm_fault_t kvm_spapr_tce_fault(struct vm_fault *vmf)
241 {
242 	struct kvmppc_spapr_tce_table *stt = vmf->vma->vm_file->private_data;
243 	struct page *page;
244 
245 	if (vmf->pgoff >= kvmppc_tce_pages(stt->size))
246 		return VM_FAULT_SIGBUS;
247 
248 	page = kvm_spapr_get_tce_page(stt, vmf->pgoff);
249 	if (!page)
250 		return VM_FAULT_OOM;
251 
252 	get_page(page);
253 	vmf->page = page;
254 	return 0;
255 }
256 
257 static const struct vm_operations_struct kvm_spapr_tce_vm_ops = {
258 	.fault = kvm_spapr_tce_fault,
259 };
260 
261 static int kvm_spapr_tce_mmap(struct file *file, struct vm_area_struct *vma)
262 {
263 	vma->vm_ops = &kvm_spapr_tce_vm_ops;
264 	return 0;
265 }
266 
267 static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
268 {
269 	struct kvmppc_spapr_tce_table *stt = filp->private_data;
270 	struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
271 	struct kvm *kvm = stt->kvm;
272 
273 	mutex_lock(&kvm->lock);
274 	list_del_rcu(&stt->list);
275 	mutex_unlock(&kvm->lock);
276 
277 	list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
278 		WARN_ON(!kref_read(&stit->kref));
279 		while (1) {
280 			if (kref_put(&stit->kref, kvm_spapr_tce_liobn_put))
281 				break;
282 		}
283 	}
284 
285 	account_locked_vm(kvm->mm,
286 		kvmppc_stt_pages(kvmppc_tce_pages(stt->size)), false);
287 
288 	kvm_put_kvm(stt->kvm);
289 
290 	call_rcu(&stt->rcu, release_spapr_tce_table);
291 
292 	return 0;
293 }
294 
295 static const struct file_operations kvm_spapr_tce_fops = {
296 	.mmap           = kvm_spapr_tce_mmap,
297 	.release	= kvm_spapr_tce_release,
298 };
299 
300 int kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
301 				  struct kvm_create_spapr_tce_64 *args)
302 {
303 	struct kvmppc_spapr_tce_table *stt = NULL;
304 	struct kvmppc_spapr_tce_table *siter;
305 	struct mm_struct *mm = kvm->mm;
306 	unsigned long npages;
307 	int ret;
308 
309 	if (!args->size || args->page_shift < 12 || args->page_shift > 34 ||
310 		(args->offset + args->size > (ULLONG_MAX >> args->page_shift)))
311 		return -EINVAL;
312 
313 	npages = kvmppc_tce_pages(args->size);
314 	ret = account_locked_vm(mm, kvmppc_stt_pages(npages), true);
315 	if (ret)
316 		return ret;
317 
318 	ret = -ENOMEM;
319 	stt = kzalloc(struct_size(stt, pages, npages), GFP_KERNEL | __GFP_NOWARN);
320 	if (!stt)
321 		goto fail_acct;
322 
323 	stt->liobn = args->liobn;
324 	stt->page_shift = args->page_shift;
325 	stt->offset = args->offset;
326 	stt->size = args->size;
327 	stt->kvm = kvm;
328 	mutex_init(&stt->alloc_lock);
329 	INIT_LIST_HEAD_RCU(&stt->iommu_tables);
330 
331 	mutex_lock(&kvm->lock);
332 
333 	/* Check this LIOBN hasn't been previously allocated */
334 	ret = 0;
335 	list_for_each_entry(siter, &kvm->arch.spapr_tce_tables, list) {
336 		if (siter->liobn == args->liobn) {
337 			ret = -EBUSY;
338 			break;
339 		}
340 	}
341 
342 	kvm_get_kvm(kvm);
343 	if (!ret)
344 		ret = anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops,
345 				       stt, O_RDWR | O_CLOEXEC);
346 
347 	if (ret >= 0)
348 		list_add_rcu(&stt->list, &kvm->arch.spapr_tce_tables);
349 	else
350 		kvm_put_kvm_no_destroy(kvm);
351 
352 	mutex_unlock(&kvm->lock);
353 
354 	if (ret >= 0)
355 		return ret;
356 
357 	kfree(stt);
358  fail_acct:
359 	account_locked_vm(mm, kvmppc_stt_pages(npages), false);
360 	return ret;
361 }
362 
363 static long kvmppc_tce_to_ua(struct kvm *kvm, unsigned long tce,
364 		unsigned long *ua)
365 {
366 	unsigned long gfn = tce >> PAGE_SHIFT;
367 	struct kvm_memory_slot *memslot;
368 
369 	memslot = __gfn_to_memslot(kvm_memslots(kvm), gfn);
370 	if (!memslot)
371 		return -EINVAL;
372 
373 	*ua = __gfn_to_hva_memslot(memslot, gfn) |
374 		(tce & ~(PAGE_MASK | TCE_PCI_READ | TCE_PCI_WRITE));
375 
376 	return 0;
377 }
378 
379 static long kvmppc_tce_validate(struct kvmppc_spapr_tce_table *stt,
380 		unsigned long tce)
381 {
382 	unsigned long gpa = tce & ~(TCE_PCI_READ | TCE_PCI_WRITE);
383 	enum dma_data_direction dir = iommu_tce_direction(tce);
384 	struct kvmppc_spapr_tce_iommu_table *stit;
385 	unsigned long ua = 0;
386 
387 	/* Allow userspace to poison TCE table */
388 	if (dir == DMA_NONE)
389 		return H_SUCCESS;
390 
391 	if (iommu_tce_check_gpa(stt->page_shift, gpa))
392 		return H_TOO_HARD;
393 
394 	if (kvmppc_tce_to_ua(stt->kvm, tce, &ua))
395 		return H_TOO_HARD;
396 
397 	rcu_read_lock();
398 	list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
399 		unsigned long hpa = 0;
400 		struct mm_iommu_table_group_mem_t *mem;
401 		long shift = stit->tbl->it_page_shift;
402 
403 		mem = mm_iommu_lookup(stt->kvm->mm, ua, 1ULL << shift);
404 		if (!mem || mm_iommu_ua_to_hpa(mem, ua, shift, &hpa)) {
405 			rcu_read_unlock();
406 			return H_TOO_HARD;
407 		}
408 	}
409 	rcu_read_unlock();
410 
411 	return H_SUCCESS;
412 }
413 
414 /*
415  * Handles TCE requests for emulated devices.
416  * Puts guest TCE values to the table and expects user space to convert them.
417  * Cannot fail so kvmppc_tce_validate must be called before it.
418  */
419 static void kvmppc_tce_put(struct kvmppc_spapr_tce_table *stt,
420 		unsigned long idx, unsigned long tce)
421 {
422 	struct page *page;
423 	u64 *tbl;
424 	unsigned long sttpage;
425 
426 	idx -= stt->offset;
427 	sttpage = idx / TCES_PER_PAGE;
428 	page = stt->pages[sttpage];
429 
430 	if (!page) {
431 		/* We allow any TCE, not just with read|write permissions */
432 		if (!tce)
433 			return;
434 
435 		page = kvm_spapr_get_tce_page(stt, sttpage);
436 		if (!page)
437 			return;
438 	}
439 	tbl = page_to_virt(page);
440 
441 	tbl[idx % TCES_PER_PAGE] = tce;
442 }
443 
444 static void kvmppc_clear_tce(struct mm_struct *mm, struct kvmppc_spapr_tce_table *stt,
445 		struct iommu_table *tbl, unsigned long entry)
446 {
447 	unsigned long i;
448 	unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
449 	unsigned long io_entry = entry << (stt->page_shift - tbl->it_page_shift);
450 
451 	for (i = 0; i < subpages; ++i) {
452 		unsigned long hpa = 0;
453 		enum dma_data_direction dir = DMA_NONE;
454 
455 		iommu_tce_xchg_no_kill(mm, tbl, io_entry + i, &hpa, &dir);
456 	}
457 }
458 
459 static long kvmppc_tce_iommu_mapped_dec(struct kvm *kvm,
460 		struct iommu_table *tbl, unsigned long entry)
461 {
462 	struct mm_iommu_table_group_mem_t *mem = NULL;
463 	const unsigned long pgsize = 1ULL << tbl->it_page_shift;
464 	__be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY_RO(tbl, entry);
465 
466 	if (!pua)
467 		return H_SUCCESS;
468 
469 	mem = mm_iommu_lookup(kvm->mm, be64_to_cpu(*pua), pgsize);
470 	if (!mem)
471 		return H_TOO_HARD;
472 
473 	mm_iommu_mapped_dec(mem);
474 
475 	*pua = cpu_to_be64(0);
476 
477 	return H_SUCCESS;
478 }
479 
480 static long kvmppc_tce_iommu_do_unmap(struct kvm *kvm,
481 		struct iommu_table *tbl, unsigned long entry)
482 {
483 	enum dma_data_direction dir = DMA_NONE;
484 	unsigned long hpa = 0;
485 	long ret;
486 
487 	if (WARN_ON_ONCE(iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa,
488 					&dir)))
489 		return H_TOO_HARD;
490 
491 	if (dir == DMA_NONE)
492 		return H_SUCCESS;
493 
494 	ret = kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
495 	if (ret != H_SUCCESS)
496 		iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa, &dir);
497 
498 	return ret;
499 }
500 
501 static long kvmppc_tce_iommu_unmap(struct kvm *kvm,
502 		struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
503 		unsigned long entry)
504 {
505 	unsigned long i, ret = H_SUCCESS;
506 	unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
507 	unsigned long io_entry = entry * subpages;
508 
509 	for (i = 0; i < subpages; ++i) {
510 		ret = kvmppc_tce_iommu_do_unmap(kvm, tbl, io_entry + i);
511 		if (ret != H_SUCCESS)
512 			break;
513 	}
514 
515 	iommu_tce_kill(tbl, io_entry, subpages);
516 
517 	return ret;
518 }
519 
520 static long kvmppc_tce_iommu_do_map(struct kvm *kvm, struct iommu_table *tbl,
521 		unsigned long entry, unsigned long ua,
522 		enum dma_data_direction dir)
523 {
524 	long ret;
525 	unsigned long hpa;
526 	__be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
527 	struct mm_iommu_table_group_mem_t *mem;
528 
529 	if (!pua)
530 		/* it_userspace allocation might be delayed */
531 		return H_TOO_HARD;
532 
533 	mem = mm_iommu_lookup(kvm->mm, ua, 1ULL << tbl->it_page_shift);
534 	if (!mem)
535 		/* This only handles v2 IOMMU type, v1 is handled via ioctl() */
536 		return H_TOO_HARD;
537 
538 	if (WARN_ON_ONCE(mm_iommu_ua_to_hpa(mem, ua, tbl->it_page_shift, &hpa)))
539 		return H_TOO_HARD;
540 
541 	if (mm_iommu_mapped_inc(mem))
542 		return H_TOO_HARD;
543 
544 	ret = iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa, &dir);
545 	if (WARN_ON_ONCE(ret)) {
546 		mm_iommu_mapped_dec(mem);
547 		return H_TOO_HARD;
548 	}
549 
550 	if (dir != DMA_NONE)
551 		kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
552 
553 	*pua = cpu_to_be64(ua);
554 
555 	return 0;
556 }
557 
558 static long kvmppc_tce_iommu_map(struct kvm *kvm,
559 		struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
560 		unsigned long entry, unsigned long ua,
561 		enum dma_data_direction dir)
562 {
563 	unsigned long i, pgoff, ret = H_SUCCESS;
564 	unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
565 	unsigned long io_entry = entry * subpages;
566 
567 	for (i = 0, pgoff = 0; i < subpages;
568 			++i, pgoff += IOMMU_PAGE_SIZE(tbl)) {
569 
570 		ret = kvmppc_tce_iommu_do_map(kvm, tbl,
571 				io_entry + i, ua + pgoff, dir);
572 		if (ret != H_SUCCESS)
573 			break;
574 	}
575 
576 	iommu_tce_kill(tbl, io_entry, subpages);
577 
578 	return ret;
579 }
580 
581 long kvmppc_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
582 		      unsigned long ioba, unsigned long tce)
583 {
584 	struct kvmppc_spapr_tce_table *stt;
585 	long ret, idx;
586 	struct kvmppc_spapr_tce_iommu_table *stit;
587 	unsigned long entry, ua = 0;
588 	enum dma_data_direction dir;
589 
590 	/* udbg_printf("H_PUT_TCE(): liobn=0x%lx ioba=0x%lx, tce=0x%lx\n", */
591 	/* 	    liobn, ioba, tce); */
592 
593 	stt = kvmppc_find_table(vcpu->kvm, liobn);
594 	if (!stt)
595 		return H_TOO_HARD;
596 
597 	ret = kvmppc_ioba_validate(stt, ioba, 1);
598 	if (ret != H_SUCCESS)
599 		return ret;
600 
601 	idx = srcu_read_lock(&vcpu->kvm->srcu);
602 
603 	ret = kvmppc_tce_validate(stt, tce);
604 	if (ret != H_SUCCESS)
605 		goto unlock_exit;
606 
607 	dir = iommu_tce_direction(tce);
608 
609 	if ((dir != DMA_NONE) && kvmppc_tce_to_ua(vcpu->kvm, tce, &ua)) {
610 		ret = H_PARAMETER;
611 		goto unlock_exit;
612 	}
613 
614 	entry = ioba >> stt->page_shift;
615 
616 	list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
617 		if (dir == DMA_NONE)
618 			ret = kvmppc_tce_iommu_unmap(vcpu->kvm, stt,
619 					stit->tbl, entry);
620 		else
621 			ret = kvmppc_tce_iommu_map(vcpu->kvm, stt, stit->tbl,
622 					entry, ua, dir);
623 
624 
625 		if (ret != H_SUCCESS) {
626 			kvmppc_clear_tce(vcpu->kvm->mm, stt, stit->tbl, entry);
627 			goto unlock_exit;
628 		}
629 	}
630 
631 	kvmppc_tce_put(stt, entry, tce);
632 
633 unlock_exit:
634 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
635 
636 	return ret;
637 }
638 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce);
639 
640 long kvmppc_h_put_tce_indirect(struct kvm_vcpu *vcpu,
641 		unsigned long liobn, unsigned long ioba,
642 		unsigned long tce_list, unsigned long npages)
643 {
644 	struct kvmppc_spapr_tce_table *stt;
645 	long i, ret = H_SUCCESS, idx;
646 	unsigned long entry, ua = 0;
647 	u64 __user *tces;
648 	u64 tce;
649 	struct kvmppc_spapr_tce_iommu_table *stit;
650 
651 	stt = kvmppc_find_table(vcpu->kvm, liobn);
652 	if (!stt)
653 		return H_TOO_HARD;
654 
655 	entry = ioba >> stt->page_shift;
656 	/*
657 	 * SPAPR spec says that the maximum size of the list is 512 TCEs
658 	 * so the whole table fits in 4K page
659 	 */
660 	if (npages > 512)
661 		return H_PARAMETER;
662 
663 	if (tce_list & (SZ_4K - 1))
664 		return H_PARAMETER;
665 
666 	ret = kvmppc_ioba_validate(stt, ioba, npages);
667 	if (ret != H_SUCCESS)
668 		return ret;
669 
670 	idx = srcu_read_lock(&vcpu->kvm->srcu);
671 	if (kvmppc_tce_to_ua(vcpu->kvm, tce_list, &ua)) {
672 		ret = H_TOO_HARD;
673 		goto unlock_exit;
674 	}
675 	tces = (u64 __user *) ua;
676 
677 	for (i = 0; i < npages; ++i) {
678 		if (get_user(tce, tces + i)) {
679 			ret = H_TOO_HARD;
680 			goto unlock_exit;
681 		}
682 		tce = be64_to_cpu(tce);
683 
684 		ret = kvmppc_tce_validate(stt, tce);
685 		if (ret != H_SUCCESS)
686 			goto unlock_exit;
687 	}
688 
689 	for (i = 0; i < npages; ++i) {
690 		/*
691 		 * This looks unsafe, because we validate, then regrab
692 		 * the TCE from userspace which could have been changed by
693 		 * another thread.
694 		 *
695 		 * But it actually is safe, because the relevant checks will be
696 		 * re-executed in the following code.  If userspace tries to
697 		 * change this dodgily it will result in a messier failure mode
698 		 * but won't threaten the host.
699 		 */
700 		if (get_user(tce, tces + i)) {
701 			ret = H_TOO_HARD;
702 			goto unlock_exit;
703 		}
704 		tce = be64_to_cpu(tce);
705 
706 		if (kvmppc_tce_to_ua(vcpu->kvm, tce, &ua)) {
707 			ret = H_PARAMETER;
708 			goto unlock_exit;
709 		}
710 
711 		list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
712 			ret = kvmppc_tce_iommu_map(vcpu->kvm, stt,
713 					stit->tbl, entry + i, ua,
714 					iommu_tce_direction(tce));
715 
716 			if (ret != H_SUCCESS) {
717 				kvmppc_clear_tce(vcpu->kvm->mm, stt, stit->tbl,
718 						 entry + i);
719 				goto unlock_exit;
720 			}
721 		}
722 
723 		kvmppc_tce_put(stt, entry + i, tce);
724 	}
725 
726 unlock_exit:
727 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
728 
729 	return ret;
730 }
731 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce_indirect);
732 
733 long kvmppc_h_stuff_tce(struct kvm_vcpu *vcpu,
734 		unsigned long liobn, unsigned long ioba,
735 		unsigned long tce_value, unsigned long npages)
736 {
737 	struct kvmppc_spapr_tce_table *stt;
738 	long i, ret;
739 	struct kvmppc_spapr_tce_iommu_table *stit;
740 
741 	stt = kvmppc_find_table(vcpu->kvm, liobn);
742 	if (!stt)
743 		return H_TOO_HARD;
744 
745 	ret = kvmppc_ioba_validate(stt, ioba, npages);
746 	if (ret != H_SUCCESS)
747 		return ret;
748 
749 	/* Check permission bits only to allow userspace poison TCE for debug */
750 	if (tce_value & (TCE_PCI_WRITE | TCE_PCI_READ))
751 		return H_PARAMETER;
752 
753 	list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
754 		unsigned long entry = ioba >> stt->page_shift;
755 
756 		for (i = 0; i < npages; ++i) {
757 			ret = kvmppc_tce_iommu_unmap(vcpu->kvm, stt,
758 					stit->tbl, entry + i);
759 
760 			if (ret == H_SUCCESS)
761 				continue;
762 
763 			if (ret == H_TOO_HARD)
764 				return ret;
765 
766 			WARN_ON_ONCE(1);
767 			kvmppc_clear_tce(vcpu->kvm->mm, stt, stit->tbl, entry + i);
768 		}
769 	}
770 
771 	for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift))
772 		kvmppc_tce_put(stt, ioba >> stt->page_shift, tce_value);
773 
774 	return ret;
775 }
776 EXPORT_SYMBOL_GPL(kvmppc_h_stuff_tce);
777 
778 long kvmppc_h_get_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
779 		      unsigned long ioba)
780 {
781 	struct kvmppc_spapr_tce_table *stt;
782 	long ret;
783 	unsigned long idx;
784 	struct page *page;
785 	u64 *tbl;
786 
787 	stt = kvmppc_find_table(vcpu->kvm, liobn);
788 	if (!stt)
789 		return H_TOO_HARD;
790 
791 	ret = kvmppc_ioba_validate(stt, ioba, 1);
792 	if (ret != H_SUCCESS)
793 		return ret;
794 
795 	idx = (ioba >> stt->page_shift) - stt->offset;
796 	page = stt->pages[idx / TCES_PER_PAGE];
797 	if (!page) {
798 		kvmppc_set_gpr(vcpu, 4, 0);
799 		return H_SUCCESS;
800 	}
801 	tbl = (u64 *)page_address(page);
802 
803 	kvmppc_set_gpr(vcpu, 4, tbl[idx % TCES_PER_PAGE]);
804 
805 	return H_SUCCESS;
806 }
807 EXPORT_SYMBOL_GPL(kvmppc_h_get_tce);
808