1/*
  2 * Copyright (c) 2006, Intel Corporation.
  3 *
  4 * This program is free software; you can redistribute it and/or modify it
  5 * under the terms and conditions of the GNU General Public License,
  6 * version 2, as published by the Free Software Foundation.
  7 *
  8 * This program is distributed in the hope it will be useful, but WITHOUT
  9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 10 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
 11 * more details.
 12 *
 13 * You should have received a copy of the GNU General Public License along with
 14 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
 15 * Place - Suite 330, Boston, MA 02111-1307 USA.
 16 *
 17 * Copyright (C) 2006-2008 Intel Corporation
 18 * Copyright IBM Corporation, 2008
 19 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
 20 *
 21 * Author: Allen M. Kay <allen.m.kay@intel.com>
 22 * Author: Weidong Han <weidong.han@intel.com>
 23 * Author: Ben-Ami Yassour <benami@il.ibm.com>
 24 */
 25
 26#include <linux/list.h>
 27#include <linux/kvm_host.h>
 28#include <linux/module.h>
 29#include <linux/pci.h>
 30#include <linux/stat.h>
 31#include <linux/dmar.h>
 32#include <linux/iommu.h>
 33#include <linux/intel-iommu.h>
 34
 35static bool allow_unsafe_assigned_interrupts;
 36module_param_named(allow_unsafe_assigned_interrupts,
 37		   allow_unsafe_assigned_interrupts, bool, S_IRUGO | S_IWUSR);
 38MODULE_PARM_DESC(allow_unsafe_assigned_interrupts,
 39 "Enable device assignment on platforms without interrupt remapping support.");
 40
 41static int kvm_iommu_unmap_memslots(struct kvm *kvm);
 42static void kvm_iommu_put_pages(struct kvm *kvm,
 43				gfn_t base_gfn, unsigned long npages);
 44
 45static pfn_t kvm_pin_pages(struct kvm_memory_slot *slot, gfn_t gfn,
 46			   unsigned long size)
 47{
 48	gfn_t end_gfn;
 49	pfn_t pfn;
 50
 51	pfn     = gfn_to_pfn_memslot(slot, gfn);
 52	end_gfn = gfn + (size >> PAGE_SHIFT);
 53	gfn    += 1;
 54
 55	if (is_error_noslot_pfn(pfn))
 56		return pfn;
 57
 58	while (gfn < end_gfn)
 59		gfn_to_pfn_memslot(slot, gfn++);
 60
 61	return pfn;
 62}
 63
 64int kvm_iommu_map_pages(struct kvm *kvm, struct kvm_memory_slot *slot)
 65{
 66	gfn_t gfn, end_gfn;
 67	pfn_t pfn;
 68	int r = 0;
 69	struct iommu_domain *domain = kvm->arch.iommu_domain;
 70	int flags;
 71
 72	/* check if iommu exists and in use */
 73	if (!domain)
 74		return 0;
 75
 76	gfn     = slot->base_gfn;
 77	end_gfn = gfn + slot->npages;
 78
 79	flags = IOMMU_READ;
 80	if (!(slot->flags & KVM_MEM_READONLY))
 81		flags |= IOMMU_WRITE;
 82	if (!kvm->arch.iommu_noncoherent)
 83		flags |= IOMMU_CACHE;
 84
 85
 86	while (gfn < end_gfn) {
 87		unsigned long page_size;
 88
 89		/* Check if already mapped */
 90		if (iommu_iova_to_phys(domain, gfn_to_gpa(gfn))) {
 91			gfn += 1;
 92			continue;
 93		}
 94
 95		/* Get the page size we could use to map */
 96		page_size = kvm_host_page_size(kvm, gfn);
 97
 98		/* Make sure the page_size does not exceed the memslot */
 99		while ((gfn + (page_size >> PAGE_SHIFT)) > end_gfn)
100			page_size >>= 1;
101
102		/* Make sure gfn is aligned to the page size we want to map */
103		while ((gfn << PAGE_SHIFT) & (page_size - 1))
104			page_size >>= 1;
105
106		/* Make sure hva is aligned to the page size we want to map */
107		while (__gfn_to_hva_memslot(slot, gfn) & (page_size - 1))
108			page_size >>= 1;
109
110		/*
111		 * Pin all pages we are about to map in memory. This is
112		 * important because we unmap and unpin in 4kb steps later.
113		 */
114		pfn = kvm_pin_pages(slot, gfn, page_size);
115		if (is_error_noslot_pfn(pfn)) {
116			gfn += 1;
117			continue;
118		}
119
120		/* Map into IO address space */
121		r = iommu_map(domain, gfn_to_gpa(gfn), pfn_to_hpa(pfn),
122			      page_size, flags);
123		if (r) {
124			printk(KERN_ERR "kvm_iommu_map_address:"
125			       "iommu failed to map pfn=%llx\n", pfn);
126			goto unmap_pages;
127		}
128
129		gfn += page_size >> PAGE_SHIFT;
130
131
132	}
133
134	return 0;
135
136unmap_pages:
137	kvm_iommu_put_pages(kvm, slot->base_gfn, gfn);
138	return r;
139}
140
141static int kvm_iommu_map_memslots(struct kvm *kvm)
142{
143	int idx, r = 0;
144	struct kvm_memslots *slots;
145	struct kvm_memory_slot *memslot;
146
147	if (kvm->arch.iommu_noncoherent)
148		kvm_arch_register_noncoherent_dma(kvm);
149
150	idx = srcu_read_lock(&kvm->srcu);
151	slots = kvm_memslots(kvm);
152
153	kvm_for_each_memslot(memslot, slots) {
154		r = kvm_iommu_map_pages(kvm, memslot);
155		if (r)
156			break;
157	}
158	srcu_read_unlock(&kvm->srcu, idx);
159
160	return r;
161}
162
163int kvm_assign_device(struct kvm *kvm,
164		      struct kvm_assigned_dev_kernel *assigned_dev)
165{
166	struct pci_dev *pdev = NULL;
167	struct iommu_domain *domain = kvm->arch.iommu_domain;
168	int r;
169	bool noncoherent;
170
171	/* check if iommu exists and in use */
172	if (!domain)
173		return 0;
174
175	pdev = assigned_dev->dev;
176	if (pdev == NULL)
177		return -ENODEV;
178
179	r = iommu_attach_device(domain, &pdev->dev);
180	if (r) {
181		dev_err(&pdev->dev, "kvm assign device failed ret %d", r);
182		return r;
183	}
184
185	noncoherent = !iommu_domain_has_cap(kvm->arch.iommu_domain,
186					    IOMMU_CAP_CACHE_COHERENCY);
187
188	/* Check if need to update IOMMU page table for guest memory */
189	if (noncoherent != kvm->arch.iommu_noncoherent) {
190		kvm_iommu_unmap_memslots(kvm);
191		kvm->arch.iommu_noncoherent = noncoherent;
192		r = kvm_iommu_map_memslots(kvm);
193		if (r)
194			goto out_unmap;
195	}
196
197	pdev->dev_flags |= PCI_DEV_FLAGS_ASSIGNED;
198
199	dev_info(&pdev->dev, "kvm assign device\n");
200
201	return 0;
202out_unmap:
203	kvm_iommu_unmap_memslots(kvm);
204	return r;
205}
206
207int kvm_deassign_device(struct kvm *kvm,
208			struct kvm_assigned_dev_kernel *assigned_dev)
209{
210	struct iommu_domain *domain = kvm->arch.iommu_domain;
211	struct pci_dev *pdev = NULL;
212
213	/* check if iommu exists and in use */
214	if (!domain)
215		return 0;
216
217	pdev = assigned_dev->dev;
218	if (pdev == NULL)
219		return -ENODEV;
220
221	iommu_detach_device(domain, &pdev->dev);
222
223	pdev->dev_flags &= ~PCI_DEV_FLAGS_ASSIGNED;
224
225	dev_info(&pdev->dev, "kvm deassign device\n");
226
227	return 0;
228}
229
230int kvm_iommu_map_guest(struct kvm *kvm)
231{
232	int r;
233
234	if (!iommu_present(&pci_bus_type)) {
235		printk(KERN_ERR "%s: iommu not found\n", __func__);
236		return -ENODEV;
237	}
238
239	mutex_lock(&kvm->slots_lock);
240
241	kvm->arch.iommu_domain = iommu_domain_alloc(&pci_bus_type);
242	if (!kvm->arch.iommu_domain) {
243		r = -ENOMEM;
244		goto out_unlock;
245	}
246
247	if (!allow_unsafe_assigned_interrupts &&
248	    !iommu_domain_has_cap(kvm->arch.iommu_domain,
249				  IOMMU_CAP_INTR_REMAP)) {
250		printk(KERN_WARNING "%s: No interrupt remapping support,"
251		       " disallowing device assignment."
252		       " Re-enble with \"allow_unsafe_assigned_interrupts=1\""
253		       " module option.\n", __func__);
254		iommu_domain_free(kvm->arch.iommu_domain);
255		kvm->arch.iommu_domain = NULL;
256		r = -EPERM;
257		goto out_unlock;
258	}
259
260	r = kvm_iommu_map_memslots(kvm);
261	if (r)
262		kvm_iommu_unmap_memslots(kvm);
263
264out_unlock:
265	mutex_unlock(&kvm->slots_lock);
266	return r;
267}
268
269static void kvm_unpin_pages(struct kvm *kvm, pfn_t pfn, unsigned long npages)
270{
271	unsigned long i;
272
273	for (i = 0; i < npages; ++i)
274		kvm_release_pfn_clean(pfn + i);
275}
276
277static void kvm_iommu_put_pages(struct kvm *kvm,
278				gfn_t base_gfn, unsigned long npages)
279{
280	struct iommu_domain *domain;
281	gfn_t end_gfn, gfn;
282	pfn_t pfn;
283	u64 phys;
284
285	domain  = kvm->arch.iommu_domain;
286	end_gfn = base_gfn + npages;
287	gfn     = base_gfn;
288
289	/* check if iommu exists and in use */
290	if (!domain)
291		return;
292
293	while (gfn < end_gfn) {
294		unsigned long unmap_pages;
295		size_t size;
296
297		/* Get physical address */
298		phys = iommu_iova_to_phys(domain, gfn_to_gpa(gfn));
299
300		if (!phys) {
301			gfn++;
302			continue;
303		}
304
305		pfn  = phys >> PAGE_SHIFT;
306
307		/* Unmap address from IO address space */
308		size       = iommu_unmap(domain, gfn_to_gpa(gfn), PAGE_SIZE);
309		unmap_pages = 1ULL << get_order(size);
310
311		/* Unpin all pages we just unmapped to not leak any memory */
312		kvm_unpin_pages(kvm, pfn, unmap_pages);
313
314		gfn += unmap_pages;
315	}
316}
317
318void kvm_iommu_unmap_pages(struct kvm *kvm, struct kvm_memory_slot *slot)
319{
320	kvm_iommu_put_pages(kvm, slot->base_gfn, slot->npages);
321}
322
323static int kvm_iommu_unmap_memslots(struct kvm *kvm)
324{
325	int idx;
326	struct kvm_memslots *slots;
327	struct kvm_memory_slot *memslot;
328
329	idx = srcu_read_lock(&kvm->srcu);
330	slots = kvm_memslots(kvm);
331
332	kvm_for_each_memslot(memslot, slots)
333		kvm_iommu_unmap_pages(kvm, memslot);
334
335	srcu_read_unlock(&kvm->srcu, idx);
336
337	if (kvm->arch.iommu_noncoherent)
338		kvm_arch_unregister_noncoherent_dma(kvm);
339
340	return 0;
341}
342
343int kvm_iommu_unmap_guest(struct kvm *kvm)
344{
345	struct iommu_domain *domain = kvm->arch.iommu_domain;
346
347	/* check if iommu exists and in use */
348	if (!domain)
349		return 0;
350
351	mutex_lock(&kvm->slots_lock);
352	kvm_iommu_unmap_memslots(kvm);
353	kvm->arch.iommu_domain = NULL;
354	kvm->arch.iommu_noncoherent = false;
355	mutex_unlock(&kvm->slots_lock);
356
357	iommu_domain_free(domain);
358	return 0;
359}