xref: /openbmc/qemu/hw/remote/proxy.c (revision 1bec145cd91c82b87ae715770f8961960b1d93b0)
1 /*
2  * Copyright © 2018, 2021 Oracle and/or its affiliates.
3  *
4  * This work is licensed under the terms of the GNU GPL, version 2 or later.
5  * See the COPYING file in the top-level directory.
6  *
7  */
8 
9 #include "qemu/osdep.h"
10 #include "qemu-common.h"
11 
12 #include "hw/remote/proxy.h"
13 #include "hw/pci/pci.h"
14 #include "qapi/error.h"
15 #include "io/channel-util.h"
16 #include "hw/qdev-properties.h"
17 #include "monitor/monitor.h"
18 #include "migration/blocker.h"
19 #include "qemu/sockets.h"
20 #include "hw/remote/mpqemu-link.h"
21 #include "qemu/error-report.h"
22 #include "hw/remote/proxy-memory-listener.h"
23 #include "qom/object.h"
24 #include "qemu/event_notifier.h"
25 #include "sysemu/kvm.h"
26 #include "util/event_notifier-posix.c"
27 
28 static void probe_pci_info(PCIDevice *dev, Error **errp);
29 
30 static void proxy_intx_update(PCIDevice *pci_dev)
31 {
32     PCIProxyDev *dev = PCI_PROXY_DEV(pci_dev);
33     PCIINTxRoute route;
34     int pin = pci_get_byte(pci_dev->config + PCI_INTERRUPT_PIN) - 1;
35 
36     if (dev->virq != -1) {
37         kvm_irqchip_remove_irqfd_notifier_gsi(kvm_state, &dev->intr, dev->virq);
38         dev->virq = -1;
39     }
40 
41     route = pci_device_route_intx_to_irq(pci_dev, pin);
42 
43     dev->virq = route.irq;
44 
45     if (dev->virq != -1) {
46         kvm_irqchip_add_irqfd_notifier_gsi(kvm_state, &dev->intr,
47                                            &dev->resample, dev->virq);
48     }
49 }
50 
51 static void setup_irqfd(PCIProxyDev *dev)
52 {
53     PCIDevice *pci_dev = PCI_DEVICE(dev);
54     MPQemuMsg msg;
55     Error *local_err = NULL;
56 
57     event_notifier_init(&dev->intr, 0);
58     event_notifier_init(&dev->resample, 0);
59 
60     memset(&msg, 0, sizeof(MPQemuMsg));
61     msg.cmd = MPQEMU_CMD_SET_IRQFD;
62     msg.num_fds = 2;
63     msg.fds[0] = event_notifier_get_fd(&dev->intr);
64     msg.fds[1] = event_notifier_get_fd(&dev->resample);
65     msg.size = 0;
66 
67     if (!mpqemu_msg_send(&msg, dev->ioc, &local_err)) {
68         error_report_err(local_err);
69     }
70 
71     dev->virq = -1;
72 
73     proxy_intx_update(pci_dev);
74 
75     pci_device_set_intx_routing_notifier(pci_dev, proxy_intx_update);
76 }
77 
78 static void pci_proxy_dev_realize(PCIDevice *device, Error **errp)
79 {
80     ERRP_GUARD();
81     PCIProxyDev *dev = PCI_PROXY_DEV(device);
82     uint8_t *pci_conf = device->config;
83     int fd;
84 
85     if (!dev->fd) {
86         error_setg(errp, "fd parameter not specified for %s",
87                    DEVICE(device)->id);
88         return;
89     }
90 
91     fd = monitor_fd_param(monitor_cur(), dev->fd, errp);
92     if (fd == -1) {
93         error_prepend(errp, "proxy: unable to parse fd %s: ", dev->fd);
94         return;
95     }
96 
97     if (!fd_is_socket(fd)) {
98         error_setg(errp, "proxy: fd %d is not a socket", fd);
99         close(fd);
100         return;
101     }
102 
103     dev->ioc = qio_channel_new_fd(fd, errp);
104 
105     error_setg(&dev->migration_blocker, "%s does not support migration",
106                TYPE_PCI_PROXY_DEV);
107     migrate_add_blocker(dev->migration_blocker, errp);
108 
109     qemu_mutex_init(&dev->io_mutex);
110     qio_channel_set_blocking(dev->ioc, true, NULL);
111 
112     pci_conf[PCI_LATENCY_TIMER] = 0xff;
113     pci_conf[PCI_INTERRUPT_PIN] = 0x01;
114 
115     proxy_memory_listener_configure(&dev->proxy_listener, dev->ioc);
116 
117     setup_irqfd(dev);
118 
119     probe_pci_info(PCI_DEVICE(dev), errp);
120 }
121 
122 static void pci_proxy_dev_exit(PCIDevice *pdev)
123 {
124     PCIProxyDev *dev = PCI_PROXY_DEV(pdev);
125 
126     if (dev->ioc) {
127         qio_channel_close(dev->ioc, NULL);
128     }
129 
130     migrate_del_blocker(dev->migration_blocker);
131 
132     error_free(dev->migration_blocker);
133 
134     proxy_memory_listener_deconfigure(&dev->proxy_listener);
135 
136     event_notifier_cleanup(&dev->intr);
137     event_notifier_cleanup(&dev->resample);
138 }
139 
140 static void config_op_send(PCIProxyDev *pdev, uint32_t addr, uint32_t *val,
141                            int len, unsigned int op)
142 {
143     MPQemuMsg msg = { 0 };
144     uint64_t ret = -EINVAL;
145     Error *local_err = NULL;
146 
147     msg.cmd = op;
148     msg.data.pci_conf_data.addr = addr;
149     msg.data.pci_conf_data.val = (op == MPQEMU_CMD_PCI_CFGWRITE) ? *val : 0;
150     msg.data.pci_conf_data.len = len;
151     msg.size = sizeof(PciConfDataMsg);
152 
153     ret = mpqemu_msg_send_and_await_reply(&msg, pdev, &local_err);
154     if (local_err) {
155         error_report_err(local_err);
156     }
157 
158     if (ret == UINT64_MAX) {
159         error_report("Failed to perform PCI config %s operation",
160                      (op == MPQEMU_CMD_PCI_CFGREAD) ? "READ" : "WRITE");
161     }
162 
163     if (op == MPQEMU_CMD_PCI_CFGREAD) {
164         *val = (uint32_t)ret;
165     }
166 }
167 
168 static uint32_t pci_proxy_read_config(PCIDevice *d, uint32_t addr, int len)
169 {
170     uint32_t val;
171 
172     config_op_send(PCI_PROXY_DEV(d), addr, &val, len, MPQEMU_CMD_PCI_CFGREAD);
173 
174     return val;
175 }
176 
177 static void pci_proxy_write_config(PCIDevice *d, uint32_t addr, uint32_t val,
178                                    int len)
179 {
180     /*
181      * Some of the functions access the copy of remote device's PCI config
182      * space which is cached in the proxy device. Therefore, maintain
183      * it updated.
184      */
185     pci_default_write_config(d, addr, val, len);
186 
187     config_op_send(PCI_PROXY_DEV(d), addr, &val, len, MPQEMU_CMD_PCI_CFGWRITE);
188 }
189 
190 static Property proxy_properties[] = {
191     DEFINE_PROP_STRING("fd", PCIProxyDev, fd),
192     DEFINE_PROP_END_OF_LIST(),
193 };
194 
195 static void pci_proxy_dev_class_init(ObjectClass *klass, void *data)
196 {
197     DeviceClass *dc = DEVICE_CLASS(klass);
198     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
199 
200     k->realize = pci_proxy_dev_realize;
201     k->exit = pci_proxy_dev_exit;
202     k->config_read = pci_proxy_read_config;
203     k->config_write = pci_proxy_write_config;
204 
205     device_class_set_props(dc, proxy_properties);
206 }
207 
208 static const TypeInfo pci_proxy_dev_type_info = {
209     .name          = TYPE_PCI_PROXY_DEV,
210     .parent        = TYPE_PCI_DEVICE,
211     .instance_size = sizeof(PCIProxyDev),
212     .class_init    = pci_proxy_dev_class_init,
213     .interfaces = (InterfaceInfo[]) {
214         { INTERFACE_CONVENTIONAL_PCI_DEVICE },
215         { },
216     },
217 };
218 
219 static void pci_proxy_dev_register_types(void)
220 {
221     type_register_static(&pci_proxy_dev_type_info);
222 }
223 
224 type_init(pci_proxy_dev_register_types)
225 
226 static void send_bar_access_msg(PCIProxyDev *pdev, MemoryRegion *mr,
227                                 bool write, hwaddr addr, uint64_t *val,
228                                 unsigned size, bool memory)
229 {
230     MPQemuMsg msg = { 0 };
231     long ret = -EINVAL;
232     Error *local_err = NULL;
233 
234     msg.size = sizeof(BarAccessMsg);
235     msg.data.bar_access.addr = mr->addr + addr;
236     msg.data.bar_access.size = size;
237     msg.data.bar_access.memory = memory;
238 
239     if (write) {
240         msg.cmd = MPQEMU_CMD_BAR_WRITE;
241         msg.data.bar_access.val = *val;
242     } else {
243         msg.cmd = MPQEMU_CMD_BAR_READ;
244     }
245 
246     ret = mpqemu_msg_send_and_await_reply(&msg, pdev, &local_err);
247     if (local_err) {
248         error_report_err(local_err);
249     }
250 
251     if (!write) {
252         *val = ret;
253     }
254 }
255 
256 static void proxy_bar_write(void *opaque, hwaddr addr, uint64_t val,
257                             unsigned size)
258 {
259     ProxyMemoryRegion *pmr = opaque;
260 
261     send_bar_access_msg(pmr->dev, &pmr->mr, true, addr, &val, size,
262                         pmr->memory);
263 }
264 
265 static uint64_t proxy_bar_read(void *opaque, hwaddr addr, unsigned size)
266 {
267     ProxyMemoryRegion *pmr = opaque;
268     uint64_t val;
269 
270     send_bar_access_msg(pmr->dev, &pmr->mr, false, addr, &val, size,
271                         pmr->memory);
272 
273     return val;
274 }
275 
276 const MemoryRegionOps proxy_mr_ops = {
277     .read = proxy_bar_read,
278     .write = proxy_bar_write,
279     .endianness = DEVICE_NATIVE_ENDIAN,
280     .impl = {
281         .min_access_size = 1,
282         .max_access_size = 8,
283     },
284 };
285 
286 static void probe_pci_info(PCIDevice *dev, Error **errp)
287 {
288     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
289     uint32_t orig_val, new_val, base_class, val;
290     PCIProxyDev *pdev = PCI_PROXY_DEV(dev);
291     DeviceClass *dc = DEVICE_CLASS(pc);
292     uint8_t type;
293     int i, size;
294 
295     config_op_send(pdev, PCI_VENDOR_ID, &val, 2, MPQEMU_CMD_PCI_CFGREAD);
296     pc->vendor_id = (uint16_t)val;
297 
298     config_op_send(pdev, PCI_DEVICE_ID, &val, 2, MPQEMU_CMD_PCI_CFGREAD);
299     pc->device_id = (uint16_t)val;
300 
301     config_op_send(pdev, PCI_CLASS_DEVICE, &val, 2, MPQEMU_CMD_PCI_CFGREAD);
302     pc->class_id = (uint16_t)val;
303 
304     config_op_send(pdev, PCI_SUBSYSTEM_ID, &val, 2, MPQEMU_CMD_PCI_CFGREAD);
305     pc->subsystem_id = (uint16_t)val;
306 
307     base_class = pc->class_id >> 4;
308     switch (base_class) {
309     case PCI_BASE_CLASS_BRIDGE:
310         set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
311         break;
312     case PCI_BASE_CLASS_STORAGE:
313         set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
314         break;
315     case PCI_BASE_CLASS_NETWORK:
316         set_bit(DEVICE_CATEGORY_NETWORK, dc->categories);
317         break;
318     case PCI_BASE_CLASS_INPUT:
319         set_bit(DEVICE_CATEGORY_INPUT, dc->categories);
320         break;
321     case PCI_BASE_CLASS_DISPLAY:
322         set_bit(DEVICE_CATEGORY_DISPLAY, dc->categories);
323         break;
324     case PCI_BASE_CLASS_PROCESSOR:
325         set_bit(DEVICE_CATEGORY_CPU, dc->categories);
326         break;
327     default:
328         set_bit(DEVICE_CATEGORY_MISC, dc->categories);
329         break;
330     }
331 
332     for (i = 0; i < PCI_NUM_REGIONS; i++) {
333         config_op_send(pdev, PCI_BASE_ADDRESS_0 + (4 * i), &orig_val, 4,
334                        MPQEMU_CMD_PCI_CFGREAD);
335         new_val = 0xffffffff;
336         config_op_send(pdev, PCI_BASE_ADDRESS_0 + (4 * i), &new_val, 4,
337                        MPQEMU_CMD_PCI_CFGWRITE);
338         config_op_send(pdev, PCI_BASE_ADDRESS_0 + (4 * i), &new_val, 4,
339                        MPQEMU_CMD_PCI_CFGREAD);
340         size = (~(new_val & 0xFFFFFFF0)) + 1;
341         config_op_send(pdev, PCI_BASE_ADDRESS_0 + (4 * i), &orig_val, 4,
342                        MPQEMU_CMD_PCI_CFGWRITE);
343         type = (new_val & 0x1) ?
344                    PCI_BASE_ADDRESS_SPACE_IO : PCI_BASE_ADDRESS_SPACE_MEMORY;
345 
346         if (size) {
347             g_autofree char *name;
348             pdev->region[i].dev = pdev;
349             pdev->region[i].present = true;
350             if (type == PCI_BASE_ADDRESS_SPACE_MEMORY) {
351                 pdev->region[i].memory = true;
352             }
353             name = g_strdup_printf("bar-region-%d", i);
354             memory_region_init_io(&pdev->region[i].mr, OBJECT(pdev),
355                                   &proxy_mr_ops, &pdev->region[i],
356                                   name, size);
357             pci_register_bar(dev, i, type, &pdev->region[i].mr);
358         }
359     }
360 }
361