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/moduleparam.h>
 29#include <linux/pci.h>
 30#include <linux/stat.h>
 31#include <linux/iommu.h>
 32#include "assigned-dev.h"
 33
 34static bool allow_unsafe_assigned_interrupts;
 35module_param_named(allow_unsafe_assigned_interrupts,
 36		   allow_unsafe_assigned_interrupts, bool, S_IRUGO | S_IWUSR);
 37MODULE_PARM_DESC(allow_unsafe_assigned_interrupts,
 38 "Enable device assignment on platforms without interrupt remapping support.");
 39
 40static int kvm_iommu_unmap_memslots(struct kvm *kvm);
 41static void kvm_iommu_put_pages(struct kvm *kvm,
 42				gfn_t base_gfn, unsigned long npages);
 43
 44static kvm_pfn_t kvm_pin_pages(struct kvm_memory_slot *slot, gfn_t gfn,
 45			   unsigned long npages)
 46{
 47	gfn_t end_gfn;
 48	kvm_pfn_t pfn;
 49
 50	pfn     = gfn_to_pfn_memslot(slot, gfn);
 51	end_gfn = gfn + npages;
 52	gfn    += 1;
 53
 54	if (is_error_noslot_pfn(pfn))
 55		return pfn;
 56
 57	while (gfn < end_gfn)
 58		gfn_to_pfn_memslot(slot, gfn++);
 59
 60	return pfn;
 61}
 62
 63static void kvm_unpin_pages(struct kvm *kvm, kvm_pfn_t pfn,
 64		unsigned long npages)
 65{
 66	unsigned long i;
 67
 68	for (i = 0; i < npages; ++i)
 69		kvm_release_pfn_clean(pfn + i);
 70}
 71
 72int kvm_iommu_map_pages(struct kvm *kvm, struct kvm_memory_slot *slot)
 73{
 74	gfn_t gfn, end_gfn;
 75	kvm_pfn_t pfn;
 76	int r = 0;
 77	struct iommu_domain *domain = kvm->arch.iommu_domain;
 78	int flags;
 79
 80	/* check if iommu exists and in use */
 81	if (!domain)
 82		return 0;
 83
 84	gfn     = slot->base_gfn;
 85	end_gfn = gfn + slot->npages;
 86
 87	flags = IOMMU_READ;
 88	if (!(slot->flags & KVM_MEM_READONLY))
 89		flags |= IOMMU_WRITE;
 90	if (!kvm->arch.iommu_noncoherent)
 91		flags |= IOMMU_CACHE;
 92
 93
 94	while (gfn < end_gfn) {
 95		unsigned long page_size;
 96
 97		/* Check if already mapped */
 98		if (iommu_iova_to_phys(domain, gfn_to_gpa(gfn))) {
 99			gfn += 1;
100			continue;
101		}
102
103		/* Get the page size we could use to map */
104		page_size = kvm_host_page_size(kvm, gfn);
105
106		/* Make sure the page_size does not exceed the memslot */
107		while ((gfn + (page_size >> PAGE_SHIFT)) > end_gfn)
108			page_size >>= 1;
109
110		/* Make sure gfn is aligned to the page size we want to map */
111		while ((gfn << PAGE_SHIFT) & (page_size - 1))
112			page_size >>= 1;
113
114		/* Make sure hva is aligned to the page size we want to map */
115		while (__gfn_to_hva_memslot(slot, gfn) & (page_size - 1))
116			page_size >>= 1;
117
118		/*
119		 * Pin all pages we are about to map in memory. This is
120		 * important because we unmap and unpin in 4kb steps later.
121		 */
122		pfn = kvm_pin_pages(slot, gfn, page_size >> PAGE_SHIFT);
123		if (is_error_noslot_pfn(pfn)) {
124			gfn += 1;
125			continue;
126		}
127
128		/* Map into IO address space */
129		r = iommu_map(domain, gfn_to_gpa(gfn), pfn_to_hpa(pfn),
130			      page_size, flags);
131		if (r) {
132			printk(KERN_ERR "kvm_iommu_map_address:"
133			       "iommu failed to map pfn=%llx\n", pfn);
134			kvm_unpin_pages(kvm, pfn, page_size >> PAGE_SHIFT);
135			goto unmap_pages;
136		}
137
138		gfn += page_size >> PAGE_SHIFT;
139
140		cond_resched();
141	}
142
143	return 0;
144
145unmap_pages:
146	kvm_iommu_put_pages(kvm, slot->base_gfn, gfn - slot->base_gfn);
147	return r;
148}
149
150static int kvm_iommu_map_memslots(struct kvm *kvm)
151{
152	int idx, r = 0;
153	struct kvm_memslots *slots;
154	struct kvm_memory_slot *memslot;
155
156	if (kvm->arch.iommu_noncoherent)
157		kvm_arch_register_noncoherent_dma(kvm);
158
159	idx = srcu_read_lock(&kvm->srcu);
160	slots = kvm_memslots(kvm);
161
162	kvm_for_each_memslot(memslot, slots) {
163		r = kvm_iommu_map_pages(kvm, memslot);
164		if (r)
165			break;
166	}
167	srcu_read_unlock(&kvm->srcu, idx);
168
169	return r;
170}
171
172int kvm_assign_device(struct kvm *kvm, struct pci_dev *pdev)
173{
174	struct iommu_domain *domain = kvm->arch.iommu_domain;
175	int r;
176	bool noncoherent;
177
178	/* check if iommu exists and in use */
179	if (!domain)
180		return 0;
181
182	if (pdev == NULL)
183		return -ENODEV;
184
185	r = iommu_attach_device(domain, &pdev->dev);
186	if (r) {
187		dev_err(&pdev->dev, "kvm assign device failed ret %d", r);
188		return r;
189	}
190
191	noncoherent = !iommu_capable(&pci_bus_type, IOMMU_CAP_CACHE_COHERENCY);
192
193	/* Check if need to update IOMMU page table for guest memory */
194	if (noncoherent != kvm->arch.iommu_noncoherent) {
195		kvm_iommu_unmap_memslots(kvm);
196		kvm->arch.iommu_noncoherent = noncoherent;
197		r = kvm_iommu_map_memslots(kvm);
198		if (r)
199			goto out_unmap;
200	}
201
202	kvm_arch_start_assignment(kvm);
203	pci_set_dev_assigned(pdev);
204
205	dev_info(&pdev->dev, "kvm assign device\n");
206
207	return 0;
208out_unmap:
209	kvm_iommu_unmap_memslots(kvm);
210	return r;
211}
212
213int kvm_deassign_device(struct kvm *kvm, struct pci_dev *pdev)
214{
215	struct iommu_domain *domain = kvm->arch.iommu_domain;
216
217	/* check if iommu exists and in use */
218	if (!domain)
219		return 0;
220
221	if (pdev == NULL)
222		return -ENODEV;
223
224	iommu_detach_device(domain, &pdev->dev);
225
226	pci_clear_dev_assigned(pdev);
227	kvm_arch_end_assignment(kvm);
228
229	dev_info(&pdev->dev, "kvm deassign device\n");
230
231	return 0;
232}
233
234int kvm_iommu_map_guest(struct kvm *kvm)
235{
236	int r;
237
238	if (!iommu_present(&pci_bus_type)) {
239		printk(KERN_ERR "%s: iommu not found\n", __func__);
240		return -ENODEV;
241	}
242
243	mutex_lock(&kvm->slots_lock);
244
245	kvm->arch.iommu_domain = iommu_domain_alloc(&pci_bus_type);
246	if (!kvm->arch.iommu_domain) {
247		r = -ENOMEM;
248		goto out_unlock;
249	}
250
251	if (!allow_unsafe_assigned_interrupts &&
252	    !iommu_capable(&pci_bus_type, IOMMU_CAP_INTR_REMAP)) {
253		printk(KERN_WARNING "%s: No interrupt remapping support,"
254		       " disallowing device assignment."
255		       " Re-enable with \"allow_unsafe_assigned_interrupts=1\""
256		       " module option.\n", __func__);
257		iommu_domain_free(kvm->arch.iommu_domain);
258		kvm->arch.iommu_domain = NULL;
259		r = -EPERM;
260		goto out_unlock;
261	}
262
263	r = kvm_iommu_map_memslots(kvm);
264	if (r)
265		kvm_iommu_unmap_memslots(kvm);
266
267out_unlock:
268	mutex_unlock(&kvm->slots_lock);
269	return r;
270}
271
272static void kvm_iommu_put_pages(struct kvm *kvm,
273				gfn_t base_gfn, unsigned long npages)
274{
275	struct iommu_domain *domain;
276	gfn_t end_gfn, gfn;
277	kvm_pfn_t pfn;
278	u64 phys;
279
280	domain  = kvm->arch.iommu_domain;
281	end_gfn = base_gfn + npages;
282	gfn     = base_gfn;
283
284	/* check if iommu exists and in use */
285	if (!domain)
286		return;
287
288	while (gfn < end_gfn) {
289		unsigned long unmap_pages;
290		size_t size;
291
292		/* Get physical address */
293		phys = iommu_iova_to_phys(domain, gfn_to_gpa(gfn));
294
295		if (!phys) {
296			gfn++;
297			continue;
298		}
299
300		pfn  = phys >> PAGE_SHIFT;
301
302		/* Unmap address from IO address space */
303		size       = iommu_unmap(domain, gfn_to_gpa(gfn), PAGE_SIZE);
304		unmap_pages = 1ULL << get_order(size);
305
306		/* Unpin all pages we just unmapped to not leak any memory */
307		kvm_unpin_pages(kvm, pfn, unmap_pages);
308
309		gfn += unmap_pages;
310
311		cond_resched();
312	}
313}
314
315void kvm_iommu_unmap_pages(struct kvm *kvm, struct kvm_memory_slot *slot)
316{
317	kvm_iommu_put_pages(kvm, slot->base_gfn, slot->npages);
318}
319
320static int kvm_iommu_unmap_memslots(struct kvm *kvm)
321{
322	int idx;
323	struct kvm_memslots *slots;
324	struct kvm_memory_slot *memslot;
325
326	idx = srcu_read_lock(&kvm->srcu);
327	slots = kvm_memslots(kvm);
328
329	kvm_for_each_memslot(memslot, slots)
330		kvm_iommu_unmap_pages(kvm, memslot);
331
332	srcu_read_unlock(&kvm->srcu, idx);
333
334	if (kvm->arch.iommu_noncoherent)
335		kvm_arch_unregister_noncoherent_dma(kvm);
336
337	return 0;
338}
339
340int kvm_iommu_unmap_guest(struct kvm *kvm)
341{
342	struct iommu_domain *domain = kvm->arch.iommu_domain;
343
344	/* check if iommu exists and in use */
345	if (!domain)
346		return 0;
347
348	mutex_lock(&kvm->slots_lock);
349	kvm_iommu_unmap_memslots(kvm);
350	kvm->arch.iommu_domain = NULL;
351	kvm->arch.iommu_noncoherent = false;
352	mutex_unlock(&kvm->slots_lock);
353
354	iommu_domain_free(domain);
355	return 0;
356}