1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * uio_hv_generic - generic UIO driver for VMBus 4 * 5 * Copyright (c) 2013-2016 Brocade Communications Systems, Inc. 6 * Copyright (c) 2016, Microsoft Corporation. 7 * 8 * Since the driver does not declare any device ids, you must allocate 9 * id and bind the device to the driver yourself. For example: 10 * 11 * Associate Network GUID with UIO device 12 * # echo "f8615163-df3e-46c5-913f-f2d2f965ed0e" \ 13 * > /sys/bus/vmbus/drivers/uio_hv_generic/new_id 14 * Then rebind 15 * # echo -n "ed963694-e847-4b2a-85af-bc9cfc11d6f3" \ 16 * > /sys/bus/vmbus/drivers/hv_netvsc/unbind 17 * # echo -n "ed963694-e847-4b2a-85af-bc9cfc11d6f3" \ 18 * > /sys/bus/vmbus/drivers/uio_hv_generic/bind 19 */ 20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 21 22 #include <linux/device.h> 23 #include <linux/kernel.h> 24 #include <linux/module.h> 25 #include <linux/uio_driver.h> 26 #include <linux/netdevice.h> 27 #include <linux/if_ether.h> 28 #include <linux/skbuff.h> 29 #include <linux/hyperv.h> 30 #include <linux/vmalloc.h> 31 #include <linux/slab.h> 32 33 #include "../hv/hyperv_vmbus.h" 34 35 #define DRIVER_VERSION "0.02.1" 36 #define DRIVER_AUTHOR "Stephen Hemminger <sthemmin at microsoft.com>" 37 #define DRIVER_DESC "Generic UIO driver for VMBus devices" 38 39 #define SEND_BUFFER_SIZE (16 * 1024 * 1024) 40 #define RECV_BUFFER_SIZE (31 * 1024 * 1024) 41 42 /* 43 * List of resources to be mapped to user space 44 * can be extended up to MAX_UIO_MAPS(5) items 45 */ 46 enum hv_uio_map { 47 TXRX_RING_MAP = 0, 48 INT_PAGE_MAP, 49 MON_PAGE_MAP, 50 RECV_BUF_MAP, 51 SEND_BUF_MAP 52 }; 53 54 struct hv_uio_private_data { 55 struct uio_info info; 56 struct hv_device *device; 57 atomic_t refcnt; 58 59 void *recv_buf; 60 struct vmbus_gpadl recv_gpadl; 61 char recv_name[32]; /* "recv_4294967295" */ 62 63 void *send_buf; 64 struct vmbus_gpadl send_gpadl; 65 char send_name[32]; 66 }; 67 68 /* 69 * This is the irqcontrol callback to be registered to uio_info. 70 * It can be used to disable/enable interrupt from user space processes. 71 * 72 * @param info 73 * pointer to uio_info. 74 * @param irq_state 75 * state value. 1 to enable interrupt, 0 to disable interrupt. 76 */ 77 static int 78 hv_uio_irqcontrol(struct uio_info *info, s32 irq_state) 79 { 80 struct hv_uio_private_data *pdata = info->priv; 81 struct hv_device *dev = pdata->device; 82 83 dev->channel->inbound.ring_buffer->interrupt_mask = !irq_state; 84 virt_mb(); 85 86 if (!dev->channel->offermsg.monitor_allocated && irq_state) 87 vmbus_setevent(dev->channel); 88 89 return 0; 90 } 91 92 /* 93 * Callback from vmbus_event when something is in inbound ring. 94 */ 95 static void hv_uio_channel_cb(void *context) 96 { 97 struct vmbus_channel *chan = context; 98 struct hv_device *hv_dev = chan->device_obj; 99 struct hv_uio_private_data *pdata = hv_get_drvdata(hv_dev); 100 101 chan->inbound.ring_buffer->interrupt_mask = 1; 102 virt_mb(); 103 104 uio_event_notify(&pdata->info); 105 } 106 107 /* 108 * Callback from vmbus_event when channel is rescinded. 109 * It is meant for rescind of primary channels only. 110 */ 111 static void hv_uio_rescind(struct vmbus_channel *channel) 112 { 113 struct hv_device *hv_dev = channel->device_obj; 114 struct hv_uio_private_data *pdata = hv_get_drvdata(hv_dev); 115 116 /* 117 * Turn off the interrupt file handle 118 * Next read for event will return -EIO 119 */ 120 pdata->info.irq = 0; 121 122 /* Wake up reader */ 123 uio_event_notify(&pdata->info); 124 125 /* 126 * With rescind callback registered, rescind path will not unregister the device 127 * from vmbus when the primary channel is rescinded. 128 * Without it, rescind handling is incomplete and next onoffer msg does not come. 129 * Unregister the device from vmbus here. 130 */ 131 vmbus_device_unregister(channel->device_obj); 132 } 133 134 /* Function used for mmap of ring buffer sysfs interface. 135 * The ring buffer is allocated as contiguous memory by vmbus_open 136 */ 137 static int 138 hv_uio_ring_mmap(struct vmbus_channel *channel, struct vm_area_struct *vma) 139 { 140 void *ring_buffer = page_address(channel->ringbuffer_page); 141 142 if (channel->state != CHANNEL_OPENED_STATE) 143 return -ENODEV; 144 145 return vm_iomap_memory(vma, virt_to_phys(ring_buffer), 146 channel->ringbuffer_pagecount << PAGE_SHIFT); 147 } 148 149 /* Callback from VMBUS subsystem when new channel created. */ 150 static void 151 hv_uio_new_channel(struct vmbus_channel *new_sc) 152 { 153 struct hv_device *hv_dev = new_sc->primary_channel->device_obj; 154 struct device *device = &hv_dev->device; 155 const size_t ring_bytes = SZ_2M; 156 int ret; 157 158 /* Create host communication ring */ 159 ret = vmbus_open(new_sc, ring_bytes, ring_bytes, NULL, 0, 160 hv_uio_channel_cb, new_sc); 161 if (ret) { 162 dev_err(device, "vmbus_open subchannel failed: %d\n", ret); 163 return; 164 } 165 166 /* Disable interrupts on sub channel */ 167 new_sc->inbound.ring_buffer->interrupt_mask = 1; 168 set_channel_read_mode(new_sc, HV_CALL_ISR); 169 ret = hv_create_ring_sysfs(new_sc, hv_uio_ring_mmap); 170 if (ret) { 171 dev_err(device, "sysfs create ring bin file failed; %d\n", ret); 172 vmbus_close(new_sc); 173 } 174 } 175 176 /* free the reserved buffers for send and receive */ 177 static void 178 hv_uio_cleanup(struct hv_device *dev, struct hv_uio_private_data *pdata) 179 { 180 if (pdata->send_gpadl.gpadl_handle) { 181 vmbus_teardown_gpadl(dev->channel, &pdata->send_gpadl); 182 if (!pdata->send_gpadl.decrypted) 183 vfree(pdata->send_buf); 184 } 185 186 if (pdata->recv_gpadl.gpadl_handle) { 187 vmbus_teardown_gpadl(dev->channel, &pdata->recv_gpadl); 188 if (!pdata->recv_gpadl.decrypted) 189 vfree(pdata->recv_buf); 190 } 191 } 192 193 /* VMBus primary channel is opened on first use */ 194 static int 195 hv_uio_open(struct uio_info *info, struct inode *inode) 196 { 197 struct hv_uio_private_data *pdata 198 = container_of(info, struct hv_uio_private_data, info); 199 struct hv_device *dev = pdata->device; 200 int ret; 201 202 if (atomic_inc_return(&pdata->refcnt) != 1) 203 return 0; 204 205 vmbus_set_chn_rescind_callback(dev->channel, hv_uio_rescind); 206 vmbus_set_sc_create_callback(dev->channel, hv_uio_new_channel); 207 208 ret = vmbus_connect_ring(dev->channel, 209 hv_uio_channel_cb, dev->channel); 210 if (ret == 0) 211 dev->channel->inbound.ring_buffer->interrupt_mask = 1; 212 else 213 atomic_dec(&pdata->refcnt); 214 215 return ret; 216 } 217 218 /* VMBus primary channel is closed on last close */ 219 static int 220 hv_uio_release(struct uio_info *info, struct inode *inode) 221 { 222 struct hv_uio_private_data *pdata 223 = container_of(info, struct hv_uio_private_data, info); 224 struct hv_device *dev = pdata->device; 225 int ret = 0; 226 227 if (atomic_dec_and_test(&pdata->refcnt)) 228 ret = vmbus_disconnect_ring(dev->channel); 229 230 return ret; 231 } 232 233 static int 234 hv_uio_probe(struct hv_device *dev, 235 const struct hv_vmbus_device_id *dev_id) 236 { 237 struct vmbus_channel *channel = dev->channel; 238 struct hv_uio_private_data *pdata; 239 void *ring_buffer; 240 int ret; 241 size_t ring_size = hv_dev_ring_size(channel); 242 243 if (!ring_size) 244 ring_size = SZ_2M; 245 246 /* Adjust ring size if necessary to have it page aligned */ 247 ring_size = VMBUS_RING_SIZE(ring_size); 248 249 pdata = devm_kzalloc(&dev->device, sizeof(*pdata), GFP_KERNEL); 250 if (!pdata) 251 return -ENOMEM; 252 253 ret = vmbus_alloc_ring(channel, ring_size, ring_size); 254 if (ret) 255 return ret; 256 257 set_channel_read_mode(channel, HV_CALL_ISR); 258 259 /* Fill general uio info */ 260 pdata->info.name = "uio_hv_generic"; 261 pdata->info.version = DRIVER_VERSION; 262 pdata->info.irqcontrol = hv_uio_irqcontrol; 263 pdata->info.open = hv_uio_open; 264 pdata->info.release = hv_uio_release; 265 pdata->info.irq = UIO_IRQ_CUSTOM; 266 atomic_set(&pdata->refcnt, 0); 267 268 /* mem resources */ 269 pdata->info.mem[TXRX_RING_MAP].name = "txrx_rings"; 270 ring_buffer = page_address(channel->ringbuffer_page); 271 pdata->info.mem[TXRX_RING_MAP].addr 272 = (uintptr_t)virt_to_phys(ring_buffer); 273 pdata->info.mem[TXRX_RING_MAP].size 274 = channel->ringbuffer_pagecount << PAGE_SHIFT; 275 pdata->info.mem[TXRX_RING_MAP].memtype = UIO_MEM_IOVA; 276 277 pdata->info.mem[INT_PAGE_MAP].name = "int_page"; 278 pdata->info.mem[INT_PAGE_MAP].addr 279 = (uintptr_t)vmbus_connection.int_page; 280 pdata->info.mem[INT_PAGE_MAP].size = HV_HYP_PAGE_SIZE; 281 pdata->info.mem[INT_PAGE_MAP].memtype = UIO_MEM_LOGICAL; 282 283 pdata->info.mem[MON_PAGE_MAP].name = "monitor_page"; 284 pdata->info.mem[MON_PAGE_MAP].addr 285 = (uintptr_t)vmbus_connection.monitor_pages[1]; 286 pdata->info.mem[MON_PAGE_MAP].size = HV_HYP_PAGE_SIZE; 287 pdata->info.mem[MON_PAGE_MAP].memtype = UIO_MEM_LOGICAL; 288 289 if (channel->device_id == HV_NIC) { 290 pdata->recv_buf = vzalloc(RECV_BUFFER_SIZE); 291 if (!pdata->recv_buf) { 292 ret = -ENOMEM; 293 goto fail_free_ring; 294 } 295 296 ret = vmbus_establish_gpadl(channel, pdata->recv_buf, 297 RECV_BUFFER_SIZE, &pdata->recv_gpadl); 298 if (ret) { 299 if (!pdata->recv_gpadl.decrypted) 300 vfree(pdata->recv_buf); 301 goto fail_close; 302 } 303 304 /* put Global Physical Address Label in name */ 305 snprintf(pdata->recv_name, sizeof(pdata->recv_name), 306 "recv:%u", pdata->recv_gpadl.gpadl_handle); 307 pdata->info.mem[RECV_BUF_MAP].name = pdata->recv_name; 308 pdata->info.mem[RECV_BUF_MAP].addr = (uintptr_t)pdata->recv_buf; 309 pdata->info.mem[RECV_BUF_MAP].size = RECV_BUFFER_SIZE; 310 pdata->info.mem[RECV_BUF_MAP].memtype = UIO_MEM_VIRTUAL; 311 312 pdata->send_buf = vzalloc(SEND_BUFFER_SIZE); 313 if (!pdata->send_buf) { 314 ret = -ENOMEM; 315 goto fail_close; 316 } 317 318 ret = vmbus_establish_gpadl(channel, pdata->send_buf, 319 SEND_BUFFER_SIZE, &pdata->send_gpadl); 320 if (ret) { 321 if (!pdata->send_gpadl.decrypted) 322 vfree(pdata->send_buf); 323 goto fail_close; 324 } 325 326 snprintf(pdata->send_name, sizeof(pdata->send_name), 327 "send:%u", pdata->send_gpadl.gpadl_handle); 328 pdata->info.mem[SEND_BUF_MAP].name = pdata->send_name; 329 pdata->info.mem[SEND_BUF_MAP].addr = (uintptr_t)pdata->send_buf; 330 pdata->info.mem[SEND_BUF_MAP].size = SEND_BUFFER_SIZE; 331 pdata->info.mem[SEND_BUF_MAP].memtype = UIO_MEM_VIRTUAL; 332 } 333 334 pdata->info.priv = pdata; 335 pdata->device = dev; 336 337 ret = uio_register_device(&dev->device, &pdata->info); 338 if (ret) { 339 dev_err(&dev->device, "hv_uio register failed\n"); 340 goto fail_close; 341 } 342 343 /* 344 * This internally calls sysfs_update_group, which returns a non-zero value if it executes 345 * before sysfs_create_group. This is expected as the 'ring' will be created later in 346 * vmbus_device_register() -> vmbus_add_channel_kobj(). Thus, no need to check the return 347 * value and print warning. 348 * 349 * Creating/exposing sysfs in driver probe is not encouraged as it can lead to race 350 * conditions with userspace. For backward compatibility, "ring" sysfs could not be removed 351 * or decoupled from uio_hv_generic probe. Userspace programs can make use of inotify 352 * APIs to make sure that ring is created. 353 */ 354 hv_create_ring_sysfs(channel, hv_uio_ring_mmap); 355 356 hv_set_drvdata(dev, pdata); 357 358 return 0; 359 360 fail_close: 361 hv_uio_cleanup(dev, pdata); 362 fail_free_ring: 363 vmbus_free_ring(dev->channel); 364 365 return ret; 366 } 367 368 static void 369 hv_uio_remove(struct hv_device *dev) 370 { 371 struct hv_uio_private_data *pdata = hv_get_drvdata(dev); 372 373 if (!pdata) 374 return; 375 376 hv_remove_ring_sysfs(dev->channel); 377 uio_unregister_device(&pdata->info); 378 hv_uio_cleanup(dev, pdata); 379 380 vmbus_free_ring(dev->channel); 381 } 382 383 static struct hv_driver hv_uio_drv = { 384 .name = "uio_hv_generic", 385 .id_table = NULL, /* only dynamic id's */ 386 .probe = hv_uio_probe, 387 .remove = hv_uio_remove, 388 }; 389 390 static int __init 391 hyperv_module_init(void) 392 { 393 return vmbus_driver_register(&hv_uio_drv); 394 } 395 396 static void __exit 397 hyperv_module_exit(void) 398 { 399 vmbus_driver_unregister(&hv_uio_drv); 400 } 401 402 module_init(hyperv_module_init); 403 module_exit(hyperv_module_exit); 404 405 MODULE_VERSION(DRIVER_VERSION); 406 MODULE_LICENSE("GPL v2"); 407 MODULE_AUTHOR(DRIVER_AUTHOR); 408 MODULE_DESCRIPTION(DRIVER_DESC); 409