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 HV_RING_SIZE 512 /* pages */
40 #define SEND_BUFFER_SIZE (16 * 1024 * 1024)
41 #define RECV_BUFFER_SIZE (31 * 1024 * 1024)
42
43 /*
44 * List of resources to be mapped to user space
45 * can be extended up to MAX_UIO_MAPS(5) items
46 */
47 enum hv_uio_map {
48 TXRX_RING_MAP = 0,
49 INT_PAGE_MAP,
50 MON_PAGE_MAP,
51 RECV_BUF_MAP,
52 SEND_BUF_MAP
53 };
54
55 struct hv_uio_private_data {
56 struct uio_info info;
57 struct hv_device *device;
58 atomic_t refcnt;
59
60 void *recv_buf;
61 struct vmbus_gpadl recv_gpadl;
62 char recv_name[32]; /* "recv_4294967295" */
63
64 void *send_buf;
65 struct vmbus_gpadl send_gpadl;
66 char send_name[32];
67 };
68
69 /*
70 * This is the irqcontrol callback to be registered to uio_info.
71 * It can be used to disable/enable interrupt from user space processes.
72 *
73 * @param info
74 * pointer to uio_info.
75 * @param irq_state
76 * state value. 1 to enable interrupt, 0 to disable interrupt.
77 */
78 static int
hv_uio_irqcontrol(struct uio_info * info,s32 irq_state)79 hv_uio_irqcontrol(struct uio_info *info, s32 irq_state)
80 {
81 struct hv_uio_private_data *pdata = info->priv;
82 struct hv_device *dev = pdata->device;
83
84 dev->channel->inbound.ring_buffer->interrupt_mask = !irq_state;
85 virt_mb();
86
87 return 0;
88 }
89
90 /*
91 * Callback from vmbus_event when something is in inbound ring.
92 */
hv_uio_channel_cb(void * context)93 static void hv_uio_channel_cb(void *context)
94 {
95 struct vmbus_channel *chan = context;
96 struct hv_device *hv_dev = chan->device_obj;
97 struct hv_uio_private_data *pdata = hv_get_drvdata(hv_dev);
98
99 chan->inbound.ring_buffer->interrupt_mask = 1;
100 virt_mb();
101
102 uio_event_notify(&pdata->info);
103 }
104
105 /*
106 * Callback from vmbus_event when channel is rescinded.
107 */
hv_uio_rescind(struct vmbus_channel * channel)108 static void hv_uio_rescind(struct vmbus_channel *channel)
109 {
110 struct hv_device *hv_dev = channel->primary_channel->device_obj;
111 struct hv_uio_private_data *pdata = hv_get_drvdata(hv_dev);
112
113 /*
114 * Turn off the interrupt file handle
115 * Next read for event will return -EIO
116 */
117 pdata->info.irq = 0;
118
119 /* Wake up reader */
120 uio_event_notify(&pdata->info);
121 }
122
123 /* Sysfs API to allow mmap of the ring buffers
124 * The ring buffer is allocated as contiguous memory by vmbus_open
125 */
hv_uio_ring_mmap(struct file * filp,struct kobject * kobj,struct bin_attribute * attr,struct vm_area_struct * vma)126 static int hv_uio_ring_mmap(struct file *filp, struct kobject *kobj,
127 struct bin_attribute *attr,
128 struct vm_area_struct *vma)
129 {
130 struct vmbus_channel *channel
131 = container_of(kobj, struct vmbus_channel, kobj);
132 void *ring_buffer = page_address(channel->ringbuffer_page);
133
134 if (channel->state != CHANNEL_OPENED_STATE)
135 return -ENODEV;
136
137 return vm_iomap_memory(vma, virt_to_phys(ring_buffer),
138 channel->ringbuffer_pagecount << PAGE_SHIFT);
139 }
140
141 static const struct bin_attribute ring_buffer_bin_attr = {
142 .attr = {
143 .name = "ring",
144 .mode = 0600,
145 },
146 .size = 2 * HV_RING_SIZE * PAGE_SIZE,
147 .mmap = hv_uio_ring_mmap,
148 };
149
150 /* Callback from VMBUS subsystem when new channel created. */
151 static void
hv_uio_new_channel(struct vmbus_channel * new_sc)152 hv_uio_new_channel(struct vmbus_channel *new_sc)
153 {
154 struct hv_device *hv_dev = new_sc->primary_channel->device_obj;
155 struct device *device = &hv_dev->device;
156 const size_t ring_bytes = HV_RING_SIZE * PAGE_SIZE;
157 int ret;
158
159 /* Create host communication ring */
160 ret = vmbus_open(new_sc, ring_bytes, ring_bytes, NULL, 0,
161 hv_uio_channel_cb, new_sc);
162 if (ret) {
163 dev_err(device, "vmbus_open subchannel failed: %d\n", ret);
164 return;
165 }
166
167 /* Disable interrupts on sub channel */
168 new_sc->inbound.ring_buffer->interrupt_mask = 1;
169 set_channel_read_mode(new_sc, HV_CALL_ISR);
170
171 ret = sysfs_create_bin_file(&new_sc->kobj, &ring_buffer_bin_attr);
172 if (ret) {
173 dev_err(device, "sysfs create ring bin file failed; %d\n", ret);
174 vmbus_close(new_sc);
175 }
176 }
177
178 /* free the reserved buffers for send and receive */
179 static void
hv_uio_cleanup(struct hv_device * dev,struct hv_uio_private_data * pdata)180 hv_uio_cleanup(struct hv_device *dev, struct hv_uio_private_data *pdata)
181 {
182 if (pdata->send_gpadl.gpadl_handle) {
183 vmbus_teardown_gpadl(dev->channel, &pdata->send_gpadl);
184 vfree(pdata->send_buf);
185 }
186
187 if (pdata->recv_gpadl.gpadl_handle) {
188 vmbus_teardown_gpadl(dev->channel, &pdata->recv_gpadl);
189 vfree(pdata->recv_buf);
190 }
191 }
192
193 /* VMBus primary channel is opened on first use */
194 static int
hv_uio_open(struct uio_info * info,struct inode * inode)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
hv_uio_release(struct uio_info * info,struct inode * inode)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
hv_uio_probe(struct hv_device * dev,const struct hv_vmbus_device_id * dev_id)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
242 /* Communicating with host has to be via shared memory not hypercall */
243 if (!channel->offermsg.monitor_allocated) {
244 dev_err(&dev->device, "vmbus channel requires hypercall\n");
245 return -ENOTSUPP;
246 }
247
248 pdata = devm_kzalloc(&dev->device, sizeof(*pdata), GFP_KERNEL);
249 if (!pdata)
250 return -ENOMEM;
251
252 ret = vmbus_alloc_ring(channel, HV_RING_SIZE * PAGE_SIZE,
253 HV_RING_SIZE * PAGE_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 = 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 = PAGE_SIZE;
287 pdata->info.mem[MON_PAGE_MAP].memtype = UIO_MEM_LOGICAL;
288
289 pdata->recv_buf = vzalloc(RECV_BUFFER_SIZE);
290 if (pdata->recv_buf == NULL) {
291 ret = -ENOMEM;
292 goto fail_free_ring;
293 }
294
295 ret = vmbus_establish_gpadl(channel, pdata->recv_buf,
296 RECV_BUFFER_SIZE, &pdata->recv_gpadl);
297 if (ret) {
298 vfree(pdata->recv_buf);
299 goto fail_close;
300 }
301
302 /* put Global Physical Address Label in name */
303 snprintf(pdata->recv_name, sizeof(pdata->recv_name),
304 "recv:%u", pdata->recv_gpadl.gpadl_handle);
305 pdata->info.mem[RECV_BUF_MAP].name = pdata->recv_name;
306 pdata->info.mem[RECV_BUF_MAP].addr
307 = (uintptr_t)pdata->recv_buf;
308 pdata->info.mem[RECV_BUF_MAP].size = RECV_BUFFER_SIZE;
309 pdata->info.mem[RECV_BUF_MAP].memtype = UIO_MEM_VIRTUAL;
310
311 pdata->send_buf = vzalloc(SEND_BUFFER_SIZE);
312 if (pdata->send_buf == NULL) {
313 ret = -ENOMEM;
314 goto fail_close;
315 }
316
317 ret = vmbus_establish_gpadl(channel, pdata->send_buf,
318 SEND_BUFFER_SIZE, &pdata->send_gpadl);
319 if (ret) {
320 vfree(pdata->send_buf);
321 goto fail_close;
322 }
323
324 snprintf(pdata->send_name, sizeof(pdata->send_name),
325 "send:%u", pdata->send_gpadl.gpadl_handle);
326 pdata->info.mem[SEND_BUF_MAP].name = pdata->send_name;
327 pdata->info.mem[SEND_BUF_MAP].addr
328 = (uintptr_t)pdata->send_buf;
329 pdata->info.mem[SEND_BUF_MAP].size = SEND_BUFFER_SIZE;
330 pdata->info.mem[SEND_BUF_MAP].memtype = UIO_MEM_VIRTUAL;
331
332 pdata->info.priv = pdata;
333 pdata->device = dev;
334
335 ret = uio_register_device(&dev->device, &pdata->info);
336 if (ret) {
337 dev_err(&dev->device, "hv_uio register failed\n");
338 goto fail_close;
339 }
340
341 ret = sysfs_create_bin_file(&channel->kobj, &ring_buffer_bin_attr);
342 if (ret)
343 dev_notice(&dev->device,
344 "sysfs create ring bin file failed; %d\n", ret);
345
346 hv_set_drvdata(dev, pdata);
347
348 return 0;
349
350 fail_close:
351 hv_uio_cleanup(dev, pdata);
352 fail_free_ring:
353 vmbus_free_ring(dev->channel);
354
355 return ret;
356 }
357
358 static int
hv_uio_remove(struct hv_device * dev)359 hv_uio_remove(struct hv_device *dev)
360 {
361 struct hv_uio_private_data *pdata = hv_get_drvdata(dev);
362
363 if (!pdata)
364 return 0;
365
366 sysfs_remove_bin_file(&dev->channel->kobj, &ring_buffer_bin_attr);
367 uio_unregister_device(&pdata->info);
368 hv_uio_cleanup(dev, pdata);
369
370 vmbus_free_ring(dev->channel);
371 return 0;
372 }
373
374 static struct hv_driver hv_uio_drv = {
375 .name = "uio_hv_generic",
376 .id_table = NULL, /* only dynamic id's */
377 .probe = hv_uio_probe,
378 .remove = hv_uio_remove,
379 };
380
381 static int __init
hyperv_module_init(void)382 hyperv_module_init(void)
383 {
384 return vmbus_driver_register(&hv_uio_drv);
385 }
386
387 static void __exit
hyperv_module_exit(void)388 hyperv_module_exit(void)
389 {
390 vmbus_driver_unregister(&hv_uio_drv);
391 }
392
393 module_init(hyperv_module_init);
394 module_exit(hyperv_module_exit);
395
396 MODULE_VERSION(DRIVER_VERSION);
397 MODULE_LICENSE("GPL v2");
398 MODULE_AUTHOR(DRIVER_AUTHOR);
399 MODULE_DESCRIPTION(DRIVER_DESC);
400