xref: /openbmc/qemu/hw/vfio/pci.h (revision 284d697c)
1 /*
2  * vfio based device assignment support - PCI devices
3  *
4  * Copyright Red Hat, Inc. 2012-2015
5  *
6  * Authors:
7  *  Alex Williamson <alex.williamson@redhat.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  */
12 #ifndef HW_VFIO_VFIO_PCI_H
13 #define HW_VFIO_VFIO_PCI_H
14 
15 #include "exec/memory.h"
16 #include "hw/pci/pci.h"
17 #include "hw/vfio/vfio-common.h"
18 #include "qemu/event_notifier.h"
19 #include "qemu/queue.h"
20 #include "qemu/timer.h"
21 #include "qom/object.h"
22 
23 #define PCI_ANY_ID (~0)
24 
25 struct VFIOPCIDevice;
26 
27 typedef struct VFIOIOEventFD {
28     QLIST_ENTRY(VFIOIOEventFD) next;
29     MemoryRegion *mr;
30     hwaddr addr;
31     unsigned size;
32     uint64_t data;
33     EventNotifier e;
34     VFIORegion *region;
35     hwaddr region_addr;
36     bool dynamic; /* Added runtime, removed on device reset */
37     bool vfio;
38 } VFIOIOEventFD;
39 
40 typedef struct VFIOQuirk {
41     QLIST_ENTRY(VFIOQuirk) next;
42     void *data;
43     QLIST_HEAD(, VFIOIOEventFD) ioeventfds;
44     int nr_mem;
45     MemoryRegion *mem;
46     void (*reset)(struct VFIOPCIDevice *vdev, struct VFIOQuirk *quirk);
47 } VFIOQuirk;
48 
49 typedef struct VFIOBAR {
50     VFIORegion region;
51     MemoryRegion *mr;
52     size_t size;
53     uint8_t type;
54     bool ioport;
55     bool mem64;
56     QLIST_HEAD(, VFIOQuirk) quirks;
57 } VFIOBAR;
58 
59 typedef struct VFIOVGARegion {
60     MemoryRegion mem;
61     off_t offset;
62     int nr;
63     QLIST_HEAD(, VFIOQuirk) quirks;
64 } VFIOVGARegion;
65 
66 typedef struct VFIOVGA {
67     off_t fd_offset;
68     int fd;
69     VFIOVGARegion region[QEMU_PCI_VGA_NUM_REGIONS];
70 } VFIOVGA;
71 
72 typedef struct VFIOINTx {
73     bool pending; /* interrupt pending */
74     bool kvm_accel; /* set when QEMU bypass through KVM enabled */
75     uint8_t pin; /* which pin to pull for qemu_set_irq */
76     EventNotifier interrupt; /* eventfd triggered on interrupt */
77     EventNotifier unmask; /* eventfd for unmask on QEMU bypass */
78     PCIINTxRoute route; /* routing info for QEMU bypass */
79     uint32_t mmap_timeout; /* delay to re-enable mmaps after interrupt */
80     QEMUTimer *mmap_timer; /* enable mmaps after periods w/o interrupts */
81 } VFIOINTx;
82 
83 typedef struct VFIOMSIVector {
84     /*
85      * Two interrupt paths are configured per vector.  The first, is only used
86      * for interrupts injected via QEMU.  This is typically the non-accel path,
87      * but may also be used when we want QEMU to handle masking and pending
88      * bits.  The KVM path bypasses QEMU and is therefore higher performance,
89      * but requires masking at the device.  virq is used to track the MSI route
90      * through KVM, thus kvm_interrupt is only available when virq is set to a
91      * valid (>= 0) value.
92      */
93     EventNotifier interrupt;
94     EventNotifier kvm_interrupt;
95     struct VFIOPCIDevice *vdev; /* back pointer to device */
96     int virq;
97     bool use;
98 } VFIOMSIVector;
99 
100 enum {
101     VFIO_INT_NONE = 0,
102     VFIO_INT_INTx = 1,
103     VFIO_INT_MSI  = 2,
104     VFIO_INT_MSIX = 3,
105 };
106 
107 /* Cache of MSI-X setup */
108 typedef struct VFIOMSIXInfo {
109     uint8_t table_bar;
110     uint8_t pba_bar;
111     uint16_t entries;
112     uint32_t table_offset;
113     uint32_t pba_offset;
114     unsigned long *pending;
115 } VFIOMSIXInfo;
116 
117 #define TYPE_VFIO_PCI "vfio-pci"
118 OBJECT_DECLARE_SIMPLE_TYPE(VFIOPCIDevice, VFIO_PCI)
119 
120 struct VFIOPCIDevice {
121     PCIDevice pdev;
122     VFIODevice vbasedev;
123     VFIOINTx intx;
124     unsigned int config_size;
125     uint8_t *emulated_config_bits; /* QEMU emulated bits, little-endian */
126     off_t config_offset; /* Offset of config space region within device fd */
127     unsigned int rom_size;
128     off_t rom_offset; /* Offset of ROM region within device fd */
129     void *rom;
130     int msi_cap_size;
131     VFIOMSIVector *msi_vectors;
132     VFIOMSIXInfo *msix;
133     int nr_vectors; /* Number of MSI/MSIX vectors currently in use */
134     int interrupt; /* Current interrupt type */
135     VFIOBAR bars[PCI_NUM_REGIONS - 1]; /* No ROM */
136     VFIOVGA *vga; /* 0xa0000, 0x3b0, 0x3c0 */
137     void *igd_opregion;
138     PCIHostDeviceAddress host;
139     EventNotifier err_notifier;
140     EventNotifier req_notifier;
141     int (*resetfn)(struct VFIOPCIDevice *);
142     uint32_t vendor_id;
143     uint32_t device_id;
144     uint32_t sub_vendor_id;
145     uint32_t sub_device_id;
146     uint32_t features;
147 #define VFIO_FEATURE_ENABLE_VGA_BIT 0
148 #define VFIO_FEATURE_ENABLE_VGA (1 << VFIO_FEATURE_ENABLE_VGA_BIT)
149 #define VFIO_FEATURE_ENABLE_REQ_BIT 1
150 #define VFIO_FEATURE_ENABLE_REQ (1 << VFIO_FEATURE_ENABLE_REQ_BIT)
151 #define VFIO_FEATURE_ENABLE_IGD_OPREGION_BIT 2
152 #define VFIO_FEATURE_ENABLE_IGD_OPREGION \
153                                 (1 << VFIO_FEATURE_ENABLE_IGD_OPREGION_BIT)
154     OnOffAuto display;
155     uint32_t display_xres;
156     uint32_t display_yres;
157     int32_t bootindex;
158     uint32_t igd_gms;
159     OffAutoPCIBAR msix_relo;
160     uint8_t pm_cap;
161     uint8_t nv_gpudirect_clique;
162     bool pci_aer;
163     bool req_enabled;
164     bool has_flr;
165     bool has_pm_reset;
166     bool rom_read_failed;
167     bool no_kvm_intx;
168     bool no_kvm_msi;
169     bool no_kvm_msix;
170     bool no_geforce_quirks;
171     bool no_kvm_ioeventfd;
172     bool no_vfio_ioeventfd;
173     bool enable_ramfb;
174     VFIODisplay *dpy;
175     Notifier irqchip_change_notifier;
176 };
177 
178 /* Use uin32_t for vendor & device so PCI_ANY_ID expands and cannot match hw */
179 static inline bool vfio_pci_is(VFIOPCIDevice *vdev, uint32_t vendor, uint32_t device)
180 {
181     return (vendor == PCI_ANY_ID || vendor == vdev->vendor_id) &&
182            (device == PCI_ANY_ID || device == vdev->device_id);
183 }
184 
185 static inline bool vfio_is_vga(VFIOPCIDevice *vdev)
186 {
187     PCIDevice *pdev = &vdev->pdev;
188     uint16_t class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
189 
190     return class == PCI_CLASS_DISPLAY_VGA;
191 }
192 
193 uint32_t vfio_pci_read_config(PCIDevice *pdev, uint32_t addr, int len);
194 void vfio_pci_write_config(PCIDevice *pdev,
195                            uint32_t addr, uint32_t val, int len);
196 
197 uint64_t vfio_vga_read(void *opaque, hwaddr addr, unsigned size);
198 void vfio_vga_write(void *opaque, hwaddr addr, uint64_t data, unsigned size);
199 
200 bool vfio_blacklist_opt_rom(VFIOPCIDevice *vdev);
201 void vfio_vga_quirk_setup(VFIOPCIDevice *vdev);
202 void vfio_vga_quirk_exit(VFIOPCIDevice *vdev);
203 void vfio_vga_quirk_finalize(VFIOPCIDevice *vdev);
204 void vfio_bar_quirk_setup(VFIOPCIDevice *vdev, int nr);
205 void vfio_bar_quirk_exit(VFIOPCIDevice *vdev, int nr);
206 void vfio_bar_quirk_finalize(VFIOPCIDevice *vdev, int nr);
207 void vfio_setup_resetfn_quirk(VFIOPCIDevice *vdev);
208 int vfio_add_virt_caps(VFIOPCIDevice *vdev, Error **errp);
209 void vfio_quirk_reset(VFIOPCIDevice *vdev);
210 VFIOQuirk *vfio_quirk_alloc(int nr_mem);
211 void vfio_probe_igd_bar4_quirk(VFIOPCIDevice *vdev, int nr);
212 
213 extern const PropertyInfo qdev_prop_nv_gpudirect_clique;
214 
215 int vfio_populate_vga(VFIOPCIDevice *vdev, Error **errp);
216 
217 int vfio_pci_igd_opregion_init(VFIOPCIDevice *vdev,
218                                struct vfio_region_info *info,
219                                Error **errp);
220 int vfio_pci_nvidia_v100_ram_init(VFIOPCIDevice *vdev, Error **errp);
221 int vfio_pci_nvlink2_init(VFIOPCIDevice *vdev, Error **errp);
222 
223 void vfio_display_reset(VFIOPCIDevice *vdev);
224 int vfio_display_probe(VFIOPCIDevice *vdev, Error **errp);
225 void vfio_display_finalize(VFIOPCIDevice *vdev);
226 
227 #endif /* HW_VFIO_VFIO_PCI_H */
228