xref: /openbmc/qemu/include/hw/i386/intel_iommu.h (revision 593aee5df98b4a862ff8841a57ea3dbf22131a5f)
1 /*
2  * QEMU emulation of an Intel IOMMU (VT-d)
3  *   (DMA Remapping device)
4  *
5  * Copyright (C) 2013 Knut Omang, Oracle <knut.omang@oracle.com>
6  * Copyright (C) 2014 Le Tan, <tamlokveer@gmail.com>
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License as published by
10  * the Free Software Foundation; either version 2 of the License, or
11  * (at your option) any later version.
12 
13  * This program is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  * GNU General Public License for more details.
17 
18  * You should have received a copy of the GNU General Public License along
19  * with this program; if not, see <http://www.gnu.org/licenses/>.
20  */
21 
22 #ifndef INTEL_IOMMU_H
23 #define INTEL_IOMMU_H
24 
25 #include "hw/i386/x86-iommu.h"
26 #include "qemu/iova-tree.h"
27 #include "qom/object.h"
28 
29 #define TYPE_INTEL_IOMMU_DEVICE "intel-iommu"
30 OBJECT_DECLARE_SIMPLE_TYPE(IntelIOMMUState, INTEL_IOMMU_DEVICE)
31 
32 #define TYPE_INTEL_IOMMU_MEMORY_REGION "intel-iommu-iommu-memory-region"
33 
34 /* DMAR Hardware Unit Definition address (IOMMU unit) */
35 #define Q35_HOST_BRIDGE_IOMMU_ADDR  0xfed90000ULL
36 
37 #define VTD_PCI_BUS_MAX             256
38 #define VTD_PCI_SLOT_MAX            32
39 #define VTD_PCI_FUNC_MAX            8
40 #define VTD_PCI_SLOT(devfn)         (((devfn) >> 3) & 0x1f)
41 #define VTD_PCI_FUNC(devfn)         ((devfn) & 0x07)
42 #define VTD_SID_TO_BUS(sid)         (((sid) >> 8) & 0xff)
43 #define VTD_SID_TO_DEVFN(sid)       ((sid) & 0xff)
44 
45 #define DMAR_REG_SIZE               0x230
46 #define VTD_HOST_AW_39BIT           39
47 #define VTD_HOST_AW_48BIT           48
48 #define VTD_HOST_ADDRESS_WIDTH      VTD_HOST_AW_48BIT
49 #define VTD_HAW_MASK(aw)            ((1ULL << (aw)) - 1)
50 #define VTD_MGAW_FROM_CAP(cap)      ((cap >> 16) & 0x3fULL)
51 
52 #define DMAR_REPORT_F_INTR          (1)
53 
54 #define  VTD_MSI_ADDR_HI_MASK        (0xffffffff00000000ULL)
55 #define  VTD_MSI_ADDR_HI_SHIFT       (32)
56 #define  VTD_MSI_ADDR_LO_MASK        (0x00000000ffffffffULL)
57 
58 typedef struct VTDContextEntry VTDContextEntry;
59 typedef struct VTDContextCacheEntry VTDContextCacheEntry;
60 typedef struct VTDAddressSpace VTDAddressSpace;
61 typedef struct VTDIOTLBEntry VTDIOTLBEntry;
62 typedef union VTD_IR_TableEntry VTD_IR_TableEntry;
63 typedef union VTD_IR_MSIAddress VTD_IR_MSIAddress;
64 typedef struct VTDPASIDDirEntry VTDPASIDDirEntry;
65 typedef struct VTDPASIDEntry VTDPASIDEntry;
66 
67 /* Context-Entry */
68 struct VTDContextEntry {
69     union {
70         struct {
71             uint64_t lo;
72             uint64_t hi;
73         };
74         struct {
75             uint64_t val[4];
76         };
77     };
78 };
79 
80 struct VTDContextCacheEntry {
81     /* The cache entry is obsolete if
82      * context_cache_gen!=IntelIOMMUState.context_cache_gen
83      */
84     uint32_t context_cache_gen;
85     struct VTDContextEntry context_entry;
86 };
87 
88 /* PASID Directory Entry */
89 struct VTDPASIDDirEntry {
90     uint64_t val;
91 };
92 
93 /* PASID Table Entry */
94 struct VTDPASIDEntry {
95     uint64_t val[8];
96 };
97 
98 typedef struct VTDPASIDCacheEntry {
99     struct VTDPASIDEntry pasid_entry;
100     bool valid;
101 } VTDPASIDCacheEntry;
102 
103 struct VTDAddressSpace {
104     PCIBus *bus;
105     uint8_t devfn;
106     uint32_t pasid;
107     AddressSpace as;
108     IOMMUMemoryRegion iommu;
109     MemoryRegion root;          /* The root container of the device */
110     MemoryRegion nodmar;        /* The alias of shared nodmar MR */
111     MemoryRegion iommu_ir;      /* Interrupt region: 0xfeeXXXXX */
112     MemoryRegion iommu_ir_fault; /* Interrupt region for catching fault */
113     IntelIOMMUState *iommu_state;
114     VTDContextCacheEntry context_cache_entry;
115     VTDPASIDCacheEntry pasid_cache_entry;
116     QLIST_ENTRY(VTDAddressSpace) next;
117     /* Superset of notifier flags that this address space has */
118     IOMMUNotifierFlag notifier_flags;
119     IOMMUPRINotifier *pri_notifier;
120     /*
121      * @iova_tree traces mapped IOVA ranges.
122      *
123      * The tree is not needed if no MAP notifier is registered with current
124      * VTD address space, because all guest invalidate commands can be
125      * directly passed to the IOMMU UNMAP notifiers without any further
126      * reshuffling.
127      *
128      * The tree OTOH is required for MAP typed iommu notifiers for a few
129      * reasons.
130      *
131      * Firstly, there's no way to identify whether an PSI (Page Selective
132      * Invalidations) or DSI (Domain Selective Invalidations) event is an
133      * MAP or UNMAP event within the message itself.  Without having prior
134      * knowledge of existing state vIOMMU doesn't know whether it should
135      * notify MAP or UNMAP for a PSI message it received when caching mode
136      * is enabled (for MAP notifiers).
137      *
138      * Secondly, PSI messages received from guest driver can be enlarged in
139      * range, covers but not limited to what the guest driver wanted to
140      * invalidate.  When the range to invalidates gets bigger than the
141      * limit of a PSI message, it can even become a DSI which will
142      * invalidate the whole domain.  If the vIOMMU directly notifies the
143      * registered device with the unmodified range, it may confuse the
144      * registered drivers (e.g. vfio-pci) on either:
145      *
146      *   (1) Trying to map the same region more than once (for
147      *       VFIO_IOMMU_MAP_DMA, -EEXIST will trigger), or,
148      *
149      *   (2) Trying to UNMAP a range that is still partially mapped.
150      *
151      * That accuracy is not required for UNMAP-only notifiers, but it is a
152      * must-to-have for notifiers registered with MAP events, because the
153      * vIOMMU needs to make sure the shadow page table is always in sync
154      * with the guest IOMMU pgtables for a device.
155      */
156     IOVATree *iova_tree;
157 };
158 
159 struct VTDIOTLBEntry {
160     uint64_t gfn;
161     uint16_t domain_id;
162     uint32_t pasid;
163     uint64_t pte;
164     uint64_t mask;
165     uint8_t access_flags;
166     uint8_t pgtt;
167 };
168 
169 /* VT-d Source-ID Qualifier types */
170 enum {
171     VTD_SQ_FULL = 0x00,     /* Full SID verification */
172     VTD_SQ_IGN_3 = 0x01,    /* Ignore bit 3 */
173     VTD_SQ_IGN_2_3 = 0x02,  /* Ignore bits 2 & 3 */
174     VTD_SQ_IGN_1_3 = 0x03,  /* Ignore bits 1-3 */
175     VTD_SQ_MAX,
176 };
177 
178 /* VT-d Source Validation Types */
179 enum {
180     VTD_SVT_NONE = 0x00,    /* No validation */
181     VTD_SVT_ALL = 0x01,     /* Do full validation */
182     VTD_SVT_BUS = 0x02,     /* Validate bus range */
183     VTD_SVT_MAX,
184 };
185 
186 /* Interrupt Remapping Table Entry Definition */
187 union VTD_IR_TableEntry {
188     struct {
189 #if HOST_BIG_ENDIAN
190         uint64_t dest_id:32;         /* Destination ID */
191         uint64_t __reserved_1:8;     /* Reserved 1 */
192         uint64_t vector:8;           /* Interrupt Vector */
193         uint64_t irte_mode:1;        /* IRTE Mode */
194         uint64_t __reserved_0:3;     /* Reserved 0 */
195         uint64_t __avail:4;          /* Available spaces for software */
196         uint64_t delivery_mode:3;    /* Delivery Mode */
197         uint64_t trigger_mode:1;     /* Trigger Mode */
198         uint64_t redir_hint:1;       /* Redirection Hint */
199         uint64_t dest_mode:1;        /* Destination Mode */
200         uint64_t fault_disable:1;    /* Fault Processing Disable */
201         uint64_t present:1;          /* Whether entry present/available */
202 #else
203         uint64_t present:1;          /* Whether entry present/available */
204         uint64_t fault_disable:1;    /* Fault Processing Disable */
205         uint64_t dest_mode:1;        /* Destination Mode */
206         uint64_t redir_hint:1;       /* Redirection Hint */
207         uint64_t trigger_mode:1;     /* Trigger Mode */
208         uint64_t delivery_mode:3;    /* Delivery Mode */
209         uint64_t __avail:4;          /* Available spaces for software */
210         uint64_t __reserved_0:3;     /* Reserved 0 */
211         uint64_t irte_mode:1;        /* IRTE Mode */
212         uint64_t vector:8;           /* Interrupt Vector */
213         uint64_t __reserved_1:8;     /* Reserved 1 */
214         uint64_t dest_id:32;         /* Destination ID */
215 #endif
216 #if HOST_BIG_ENDIAN
217         uint64_t __reserved_2:44;    /* Reserved 2 */
218         uint64_t sid_vtype:2;        /* Source-ID Validation Type */
219         uint64_t sid_q:2;            /* Source-ID Qualifier */
220         uint64_t source_id:16;       /* Source-ID */
221 #else
222         uint64_t source_id:16;       /* Source-ID */
223         uint64_t sid_q:2;            /* Source-ID Qualifier */
224         uint64_t sid_vtype:2;        /* Source-ID Validation Type */
225         uint64_t __reserved_2:44;    /* Reserved 2 */
226 #endif
227     } QEMU_PACKED irte;
228     uint64_t data[2];
229 };
230 
231 #define VTD_IR_INT_FORMAT_COMPAT     (0) /* Compatible Interrupt */
232 #define VTD_IR_INT_FORMAT_REMAP      (1) /* Remappable Interrupt */
233 
234 /* Programming format for MSI/MSI-X addresses */
235 union VTD_IR_MSIAddress {
236     struct {
237 #if HOST_BIG_ENDIAN
238         uint32_t __head:12;          /* Should always be: 0x0fee */
239         uint32_t index_l:15;         /* Interrupt index bit 14-0 */
240         uint32_t int_mode:1;         /* Interrupt format */
241         uint32_t sub_valid:1;        /* SHV: Sub-Handle Valid bit */
242         uint32_t index_h:1;          /* Interrupt index bit 15 */
243         uint32_t __not_care:2;
244 #else
245         uint32_t __not_care:2;
246         uint32_t index_h:1;          /* Interrupt index bit 15 */
247         uint32_t sub_valid:1;        /* SHV: Sub-Handle Valid bit */
248         uint32_t int_mode:1;         /* Interrupt format */
249         uint32_t index_l:15;         /* Interrupt index bit 14-0 */
250         uint32_t __head:12;          /* Should always be: 0x0fee */
251 #endif
252     } QEMU_PACKED addr;
253     uint32_t data;
254 };
255 
256 /* When IR is enabled, all MSI/MSI-X data bits should be zero */
257 #define VTD_IR_MSI_DATA          (0)
258 
259 /* The iommu (DMAR) device state struct */
260 struct IntelIOMMUState {
261     X86IOMMUState x86_iommu;
262     MemoryRegion csrmem;
263     MemoryRegion mr_nodmar;
264     MemoryRegion mr_ir;
265     MemoryRegion mr_sys_alias;
266     uint8_t csr[DMAR_REG_SIZE];     /* register values */
267     uint8_t wmask[DMAR_REG_SIZE];   /* R/W bytes */
268     uint8_t w1cmask[DMAR_REG_SIZE]; /* RW1C(Write 1 to Clear) bytes */
269     uint8_t womask[DMAR_REG_SIZE];  /* WO (write only - read returns 0) */
270     uint32_t version;
271 
272     bool caching_mode;              /* RO - is cap CM enabled? */
273     bool scalable_mode;             /* RO - is Scalable Mode supported? */
274     bool flts;                      /* RO - is stage-1 translation supported? */
275     bool snoop_control;             /* RO - is SNP filed supported? */
276 
277     dma_addr_t root;                /* Current root table pointer */
278     bool root_scalable;             /* Type of root table (scalable or not) */
279     bool dmar_enabled;              /* Set if DMA remapping is enabled */
280 
281     uint16_t iq_head;               /* Current invalidation queue head */
282     uint16_t iq_tail;               /* Current invalidation queue tail */
283     dma_addr_t iq;                  /* Current invalidation queue pointer */
284     uint16_t iq_size;               /* IQ Size in number of entries */
285     bool iq_dw;                     /* IQ descriptor width 256bit or not */
286     bool qi_enabled;                /* Set if the QI is enabled */
287     uint8_t iq_last_desc_type;      /* The type of last completed descriptor */
288 
289     /* The index of the Fault Recording Register to be used next.
290      * Wraps around from N-1 to 0, where N is the number of FRCD_REG.
291      */
292     uint16_t next_frcd_reg;
293 
294     uint64_t cap;                   /* The value of capability reg */
295     uint64_t ecap;                  /* The value of extended capability reg */
296 
297     uint32_t context_cache_gen;     /* Should be in [1,MAX] */
298     GHashTable *iotlb;              /* IOTLB */
299 
300     GHashTable *vtd_address_spaces;             /* VTD address spaces */
301     VTDAddressSpace *vtd_as_cache[VTD_PCI_BUS_MAX]; /* VTD address space cache */
302     /* list of registered notifiers */
303     QLIST_HEAD(, VTDAddressSpace) vtd_as_with_notifiers;
304 
305     GHashTable *vtd_host_iommu_dev;             /* HostIOMMUDevice */
306 
307     /* interrupt remapping */
308     bool intr_enabled;              /* Whether guest enabled IR */
309     dma_addr_t intr_root;           /* Interrupt remapping table pointer */
310     uint32_t intr_size;             /* Number of IR table entries */
311     bool intr_eime;                 /* Extended interrupt mode enabled */
312     OnOffAuto intr_eim;             /* Toggle for EIM cabability */
313     bool buggy_eim;                 /* Force buggy EIM unless eim=off */
314     uint8_t aw_bits;                /* Host/IOVA address width (in bits) */
315     bool dma_drain;                 /* Whether DMA r/w draining enabled */
316     bool pasid;                     /* Whether to support PASID */
317     bool fs1gp;                     /* First Stage 1-GByte Page Support */
318 
319     /* Transient Mapping, Reserved(0) since VTD spec revision 3.2 */
320     bool stale_tm;
321 
322     /*
323      * Protects IOMMU states in general.  Currently it protects the
324      * per-IOMMU IOTLB cache, and context entry cache in VTDAddressSpace.
325      */
326     QemuMutex iommu_lock;
327 };
328 
329 /* Find the VTD Address space associated with the given bus pointer,
330  * create a new one if none exists
331  */
332 VTDAddressSpace *vtd_find_add_as(IntelIOMMUState *s, PCIBus *bus,
333                                  int devfn, unsigned int pasid);
334 
335 #endif
336