xref: /openbmc/qemu/hw/vfio/pci.h (revision 29b62a10)
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_device.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 #include "sysemu/kvm.h"
23 
24 #define PCI_ANY_ID (~0)
25 
26 struct VFIOPCIDevice;
27 
28 typedef struct VFIOIOEventFD {
29     QLIST_ENTRY(VFIOIOEventFD) next;
30     MemoryRegion *mr;
31     hwaddr addr;
32     unsigned size;
33     uint64_t data;
34     EventNotifier e;
35     VFIORegion *region;
36     hwaddr region_addr;
37     bool dynamic; /* Added runtime, removed on device reset */
38     bool vfio;
39 } VFIOIOEventFD;
40 
41 typedef struct VFIOQuirk {
42     QLIST_ENTRY(VFIOQuirk) next;
43     void *data;
44     QLIST_HEAD(, VFIOIOEventFD) ioeventfds;
45     int nr_mem;
46     MemoryRegion *mem;
47     void (*reset)(struct VFIOPCIDevice *vdev, struct VFIOQuirk *quirk);
48 } VFIOQuirk;
49 
50 typedef struct VFIOBAR {
51     VFIORegion region;
52     MemoryRegion *mr;
53     size_t size;
54     uint8_t type;
55     bool ioport;
56     bool mem64;
57     QLIST_HEAD(, VFIOQuirk) quirks;
58 } VFIOBAR;
59 
60 typedef struct VFIOVGARegion {
61     MemoryRegion mem;
62     off_t offset;
63     int nr;
64     QLIST_HEAD(, VFIOQuirk) quirks;
65 } VFIOVGARegion;
66 
67 typedef struct VFIOVGA {
68     off_t fd_offset;
69     int fd;
70     VFIOVGARegion region[QEMU_PCI_VGA_NUM_REGIONS];
71 } VFIOVGA;
72 
73 typedef struct VFIOINTx {
74     bool pending; /* interrupt pending */
75     bool kvm_accel; /* set when QEMU bypass through KVM enabled */
76     uint8_t pin; /* which pin to pull for qemu_set_irq */
77     EventNotifier interrupt; /* eventfd triggered on interrupt */
78     EventNotifier unmask; /* eventfd for unmask on QEMU bypass */
79     PCIINTxRoute route; /* routing info for QEMU bypass */
80     uint32_t mmap_timeout; /* delay to re-enable mmaps after interrupt */
81     QEMUTimer *mmap_timer; /* enable mmaps after periods w/o interrupts */
82 } VFIOINTx;
83 
84 typedef struct VFIOMSIVector {
85     /*
86      * Two interrupt paths are configured per vector.  The first, is only used
87      * for interrupts injected via QEMU.  This is typically the non-accel path,
88      * but may also be used when we want QEMU to handle masking and pending
89      * bits.  The KVM path bypasses QEMU and is therefore higher performance,
90      * but requires masking at the device.  virq is used to track the MSI route
91      * through KVM, thus kvm_interrupt is only available when virq is set to a
92      * valid (>= 0) value.
93      */
94     EventNotifier interrupt;
95     EventNotifier kvm_interrupt;
96     struct VFIOPCIDevice *vdev; /* back pointer to device */
97     int virq;
98     bool use;
99 } VFIOMSIVector;
100 
101 enum {
102     VFIO_INT_NONE = 0,
103     VFIO_INT_INTx = 1,
104     VFIO_INT_MSI  = 2,
105     VFIO_INT_MSIX = 3,
106 };
107 
108 /* Cache of MSI-X setup */
109 typedef struct VFIOMSIXInfo {
110     uint8_t table_bar;
111     uint8_t pba_bar;
112     uint16_t entries;
113     uint32_t table_offset;
114     uint32_t pba_offset;
115     unsigned long *pending;
116 } VFIOMSIXInfo;
117 
118 #define TYPE_VFIO_PCI "vfio-pci"
119 OBJECT_DECLARE_SIMPLE_TYPE(VFIOPCIDevice, VFIO_PCI)
120 
121 struct VFIOPCIDevice {
122     PCIDevice pdev;
123     VFIODevice vbasedev;
124     VFIOINTx intx;
125     unsigned int config_size;
126     uint8_t *emulated_config_bits; /* QEMU emulated bits, little-endian */
127     off_t config_offset; /* Offset of config space region within device fd */
128     unsigned int rom_size;
129     off_t rom_offset; /* Offset of ROM region within device fd */
130     void *rom;
131     int msi_cap_size;
132     VFIOMSIVector *msi_vectors;
133     VFIOMSIXInfo *msix;
134     int nr_vectors; /* Number of MSI/MSIX vectors currently in use */
135     int interrupt; /* Current interrupt type */
136     VFIOBAR bars[PCI_NUM_REGIONS - 1]; /* No ROM */
137     VFIOVGA *vga; /* 0xa0000, 0x3b0, 0x3c0 */
138     void *igd_opregion;
139     PCIHostDeviceAddress host;
140     EventNotifier err_notifier;
141     EventNotifier req_notifier;
142     int (*resetfn)(struct VFIOPCIDevice *);
143     uint32_t vendor_id;
144     uint32_t device_id;
145     uint32_t sub_vendor_id;
146     uint32_t sub_device_id;
147     uint32_t features;
148 #define VFIO_FEATURE_ENABLE_VGA_BIT 0
149 #define VFIO_FEATURE_ENABLE_VGA (1 << VFIO_FEATURE_ENABLE_VGA_BIT)
150 #define VFIO_FEATURE_ENABLE_REQ_BIT 1
151 #define VFIO_FEATURE_ENABLE_REQ (1 << VFIO_FEATURE_ENABLE_REQ_BIT)
152 #define VFIO_FEATURE_ENABLE_IGD_OPREGION_BIT 2
153 #define VFIO_FEATURE_ENABLE_IGD_OPREGION \
154                                 (1 << VFIO_FEATURE_ENABLE_IGD_OPREGION_BIT)
155     OnOffAuto display;
156     uint32_t display_xres;
157     uint32_t display_yres;
158     int32_t bootindex;
159     uint32_t igd_gms;
160     OffAutoPCIBAR msix_relo;
161     uint8_t pm_cap;
162     uint8_t nv_gpudirect_clique;
163     bool pci_aer;
164     bool req_enabled;
165     bool has_flr;
166     bool has_pm_reset;
167     bool rom_read_failed;
168     bool no_kvm_intx;
169     bool no_kvm_msi;
170     bool no_kvm_msix;
171     bool no_geforce_quirks;
172     bool no_kvm_ioeventfd;
173     bool no_vfio_ioeventfd;
174     bool enable_ramfb;
175     bool defer_kvm_irq_routing;
176     VFIODisplay *dpy;
177     Notifier irqchip_change_notifier;
178 };
179 
180 /* Use uin32_t for vendor & device so PCI_ANY_ID expands and cannot match hw */
181 static inline bool vfio_pci_is(VFIOPCIDevice *vdev, uint32_t vendor, uint32_t device)
182 {
183     return (vendor == PCI_ANY_ID || vendor == vdev->vendor_id) &&
184            (device == PCI_ANY_ID || device == vdev->device_id);
185 }
186 
187 static inline bool vfio_is_vga(VFIOPCIDevice *vdev)
188 {
189     PCIDevice *pdev = &vdev->pdev;
190     uint16_t class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
191 
192     return class == PCI_CLASS_DISPLAY_VGA;
193 }
194 
195 uint32_t vfio_pci_read_config(PCIDevice *pdev, uint32_t addr, int len);
196 void vfio_pci_write_config(PCIDevice *pdev,
197                            uint32_t addr, uint32_t val, int len);
198 
199 uint64_t vfio_vga_read(void *opaque, hwaddr addr, unsigned size);
200 void vfio_vga_write(void *opaque, hwaddr addr, uint64_t data, unsigned size);
201 
202 bool vfio_opt_rom_in_denylist(VFIOPCIDevice *vdev);
203 void vfio_vga_quirk_setup(VFIOPCIDevice *vdev);
204 void vfio_vga_quirk_exit(VFIOPCIDevice *vdev);
205 void vfio_vga_quirk_finalize(VFIOPCIDevice *vdev);
206 void vfio_bar_quirk_setup(VFIOPCIDevice *vdev, int nr);
207 void vfio_bar_quirk_exit(VFIOPCIDevice *vdev, int nr);
208 void vfio_bar_quirk_finalize(VFIOPCIDevice *vdev, int nr);
209 void vfio_setup_resetfn_quirk(VFIOPCIDevice *vdev);
210 int vfio_add_virt_caps(VFIOPCIDevice *vdev, Error **errp);
211 void vfio_quirk_reset(VFIOPCIDevice *vdev);
212 VFIOQuirk *vfio_quirk_alloc(int nr_mem);
213 void vfio_probe_igd_bar4_quirk(VFIOPCIDevice *vdev, int nr);
214 
215 extern const PropertyInfo qdev_prop_nv_gpudirect_clique;
216 
217 int vfio_populate_vga(VFIOPCIDevice *vdev, Error **errp);
218 
219 int vfio_pci_igd_opregion_init(VFIOPCIDevice *vdev,
220                                struct vfio_region_info *info,
221                                Error **errp);
222 int vfio_pci_nvidia_v100_ram_init(VFIOPCIDevice *vdev, Error **errp);
223 int vfio_pci_nvlink2_init(VFIOPCIDevice *vdev, Error **errp);
224 
225 void vfio_display_reset(VFIOPCIDevice *vdev);
226 int vfio_display_probe(VFIOPCIDevice *vdev, Error **errp);
227 void vfio_display_finalize(VFIOPCIDevice *vdev);
228 
229 #endif /* HW_VFIO_VFIO_PCI_H */
230