1 /*
2 * VFIO BASE CONTAINER
3 *
4 * Copyright (C) 2023 Intel Corporation.
5 * Copyright Red Hat, Inc. 2023
6 *
7 * Authors: Yi Liu <yi.l.liu@intel.com>
8 * Eric Auger <eric.auger@redhat.com>
9 *
10 * SPDX-License-Identifier: GPL-2.0-or-later
11 */
12
13 #ifndef HW_VFIO_VFIO_CONTAINER_BASE_H
14 #define HW_VFIO_VFIO_CONTAINER_BASE_H
15
16 #include "system/memory.h"
17
18 typedef struct VFIODevice VFIODevice;
19 typedef struct VFIOIOMMUClass VFIOIOMMUClass;
20
21 typedef struct {
22 unsigned long *bitmap;
23 hwaddr size;
24 hwaddr pages;
25 } VFIOBitmap;
26
27 typedef struct VFIOAddressSpace {
28 AddressSpace *as;
29 QLIST_HEAD(, VFIOContainerBase) containers;
30 QLIST_ENTRY(VFIOAddressSpace) list;
31 } VFIOAddressSpace;
32
33 /*
34 * This is the base object for vfio container backends
35 */
36 typedef struct VFIOContainerBase {
37 Object parent;
38 VFIOAddressSpace *space;
39 MemoryListener listener;
40 Error *error;
41 bool initialized;
42 uint64_t dirty_pgsizes;
43 uint64_t max_dirty_bitmap_size;
44 unsigned long pgsizes;
45 unsigned int dma_max_mappings;
46 bool dirty_pages_supported;
47 bool dirty_pages_started; /* Protected by BQL */
48 QLIST_HEAD(, VFIOGuestIOMMU) giommu_list;
49 QLIST_HEAD(, VFIORamDiscardListener) vrdl_list;
50 QLIST_ENTRY(VFIOContainerBase) next;
51 QLIST_HEAD(, VFIODevice) device_list;
52 GList *iova_ranges;
53 NotifierWithReturn cpr_reboot_notifier;
54 } VFIOContainerBase;
55
56 typedef struct VFIOGuestIOMMU {
57 VFIOContainerBase *bcontainer;
58 IOMMUMemoryRegion *iommu_mr;
59 hwaddr iommu_offset;
60 IOMMUNotifier n;
61 QLIST_ENTRY(VFIOGuestIOMMU) giommu_next;
62 } VFIOGuestIOMMU;
63
64 typedef struct VFIORamDiscardListener {
65 VFIOContainerBase *bcontainer;
66 MemoryRegion *mr;
67 hwaddr offset_within_address_space;
68 hwaddr size;
69 uint64_t granularity;
70 RamDiscardListener listener;
71 QLIST_ENTRY(VFIORamDiscardListener) next;
72 } VFIORamDiscardListener;
73
74 VFIOAddressSpace *vfio_address_space_get(AddressSpace *as);
75 void vfio_address_space_put(VFIOAddressSpace *space);
76 void vfio_address_space_insert(VFIOAddressSpace *space,
77 VFIOContainerBase *bcontainer);
78
79 int vfio_container_dma_map(VFIOContainerBase *bcontainer,
80 hwaddr iova, ram_addr_t size,
81 void *vaddr, bool readonly, MemoryRegion *mr);
82 int vfio_container_dma_unmap(VFIOContainerBase *bcontainer,
83 hwaddr iova, ram_addr_t size,
84 IOMMUTLBEntry *iotlb, bool unmap_all);
85 bool vfio_container_add_section_window(VFIOContainerBase *bcontainer,
86 MemoryRegionSection *section,
87 Error **errp);
88 void vfio_container_del_section_window(VFIOContainerBase *bcontainer,
89 MemoryRegionSection *section);
90 int vfio_container_set_dirty_page_tracking(VFIOContainerBase *bcontainer,
91 bool start, Error **errp);
92 bool vfio_container_dirty_tracking_is_started(
93 const VFIOContainerBase *bcontainer);
94 bool vfio_container_devices_dirty_tracking_is_supported(
95 const VFIOContainerBase *bcontainer);
96 int vfio_container_query_dirty_bitmap(const VFIOContainerBase *bcontainer,
97 uint64_t iova, uint64_t size, ram_addr_t ram_addr, Error **errp);
98
99 GList *vfio_container_get_iova_ranges(const VFIOContainerBase *bcontainer);
100
101 static inline uint64_t
vfio_container_get_page_size_mask(const VFIOContainerBase * bcontainer)102 vfio_container_get_page_size_mask(const VFIOContainerBase *bcontainer)
103 {
104 assert(bcontainer);
105 return bcontainer->pgsizes;
106 }
107
108 #define TYPE_VFIO_IOMMU "vfio-iommu"
109 #define TYPE_VFIO_IOMMU_LEGACY TYPE_VFIO_IOMMU "-legacy"
110 #define TYPE_VFIO_IOMMU_SPAPR TYPE_VFIO_IOMMU "-spapr"
111 #define TYPE_VFIO_IOMMU_IOMMUFD TYPE_VFIO_IOMMU "-iommufd"
112 #define TYPE_VFIO_IOMMU_USER TYPE_VFIO_IOMMU "-user"
113
114 OBJECT_DECLARE_TYPE(VFIOContainerBase, VFIOIOMMUClass, VFIO_IOMMU)
115
116 struct VFIOIOMMUClass {
117 ObjectClass parent_class;
118
119 /**
120 * @setup
121 *
122 * Perform basic setup of the container, including configuring IOMMU
123 * capabilities, IOVA ranges, supported page sizes, etc.
124 *
125 * @bcontainer: #VFIOContainerBase
126 * @errp: pointer to Error*, to store an error if it happens.
127 *
128 * Returns true to indicate success and false for error.
129 */
130 bool (*setup)(VFIOContainerBase *bcontainer, Error **errp);
131
132 /**
133 * @listener_begin
134 *
135 * Called at the beginning of an address space update transaction.
136 * See #MemoryListener.
137 *
138 * @bcontainer: #VFIOContainerBase
139 */
140 void (*listener_begin)(VFIOContainerBase *bcontainer);
141
142 /**
143 * @listener_commit
144 *
145 * Called at the end of an address space update transaction,
146 * See #MemoryListener.
147 *
148 * @bcontainer: #VFIOContainerBase
149 */
150 void (*listener_commit)(VFIOContainerBase *bcontainer);
151
152 /**
153 * @dma_map
154 *
155 * Map an address range into the container. Note that the memory region is
156 * referenced within an RCU read lock region across this call.
157 *
158 * @bcontainer: #VFIOContainerBase to use
159 * @iova: start address to map
160 * @size: size of the range to map
161 * @vaddr: process virtual address of mapping
162 * @readonly: true if mapping should be readonly
163 * @mr: the memory region for this mapping
164 *
165 * Returns 0 to indicate success and -errno otherwise.
166 */
167 int (*dma_map)(const VFIOContainerBase *bcontainer,
168 hwaddr iova, ram_addr_t size,
169 void *vaddr, bool readonly, MemoryRegion *mr);
170 /**
171 * @dma_map_file
172 *
173 * Map a file range for the container.
174 *
175 * @bcontainer: #VFIOContainerBase to use for map
176 * @iova: start address to map
177 * @size: size of the range to map
178 * @fd: descriptor of the file to map
179 * @start: starting file offset of the range to map
180 * @readonly: map read only if true
181 */
182 int (*dma_map_file)(const VFIOContainerBase *bcontainer,
183 hwaddr iova, ram_addr_t size,
184 int fd, unsigned long start, bool readonly);
185 /**
186 * @dma_unmap
187 *
188 * Unmap an address range from the container.
189 *
190 * @bcontainer: #VFIOContainerBase to use for unmap
191 * @iova: start address to unmap
192 * @size: size of the range to unmap
193 * @iotlb: The IOMMU TLB mapping entry (or NULL)
194 * @unmap_all: if set, unmap the entire address space
195 *
196 * Returns 0 to indicate success and -errno otherwise.
197 */
198 int (*dma_unmap)(const VFIOContainerBase *bcontainer,
199 hwaddr iova, ram_addr_t size,
200 IOMMUTLBEntry *iotlb, bool unmap_all);
201
202
203 /**
204 * @attach_device
205 *
206 * Associate the given device with a container and do some related
207 * initialization of the device context.
208 *
209 * @name: name of the device
210 * @vbasedev: the device
211 * @as: address space to use
212 * @errp: pointer to Error*, to store an error if it happens.
213 *
214 * Returns true to indicate success and false for error.
215 */
216 bool (*attach_device)(const char *name, VFIODevice *vbasedev,
217 AddressSpace *as, Error **errp);
218
219 /*
220 * @detach_device
221 *
222 * Detach the given device from its container and clean up any necessary
223 * state.
224 *
225 * @vbasedev: the device to disassociate
226 */
227 void (*detach_device)(VFIODevice *vbasedev);
228
229 /* migration feature */
230
231 /**
232 * @set_dirty_page_tracking
233 *
234 * Start or stop dirty pages tracking on VFIO container
235 *
236 * @bcontainer: #VFIOContainerBase on which to de/activate dirty
237 * page tracking
238 * @start: indicates whether to start or stop dirty pages tracking
239 * @errp: pointer to Error*, to store an error if it happens.
240 *
241 * Returns zero to indicate success and negative for error.
242 */
243 int (*set_dirty_page_tracking)(const VFIOContainerBase *bcontainer,
244 bool start, Error **errp);
245 /**
246 * @query_dirty_bitmap
247 *
248 * Get bitmap of dirty pages from container
249 *
250 * @bcontainer: #VFIOContainerBase from which to get dirty pages
251 * @vbmap: #VFIOBitmap internal bitmap structure
252 * @iova: iova base address
253 * @size: size of iova range
254 * @errp: pointer to Error*, to store an error if it happens.
255 *
256 * Returns zero to indicate success and negative for error.
257 */
258 int (*query_dirty_bitmap)(const VFIOContainerBase *bcontainer,
259 VFIOBitmap *vbmap, hwaddr iova, hwaddr size, Error **errp);
260 /* PCI specific */
261 int (*pci_hot_reset)(VFIODevice *vbasedev, bool single);
262
263 /* SPAPR specific */
264 bool (*add_window)(VFIOContainerBase *bcontainer,
265 MemoryRegionSection *section,
266 Error **errp);
267 void (*del_window)(VFIOContainerBase *bcontainer,
268 MemoryRegionSection *section);
269 void (*release)(VFIOContainerBase *bcontainer);
270 };
271
272 VFIORamDiscardListener *vfio_find_ram_discard_listener(
273 VFIOContainerBase *bcontainer, MemoryRegionSection *section);
274
275 void vfio_container_region_add(VFIOContainerBase *bcontainer,
276 MemoryRegionSection *section, bool cpr_remap);
277
278 #endif /* HW_VFIO_VFIO_CONTAINER_BASE_H */
279