Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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 */
46enum 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
54struct 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
68static void set_event(struct vmbus_channel *channel, s32 irq_state)
69{
70 channel->inbound.ring_buffer->interrupt_mask = !irq_state;
71 if (!channel->offermsg.monitor_allocated && irq_state) {
72 /* MB is needed for host to see the interrupt mask first */
73 virt_mb();
74 vmbus_set_event(channel);
75 }
76}
77
78/*
79 * This is the irqcontrol callback to be registered to uio_info.
80 * It can be used to disable/enable interrupt from user space processes.
81 *
82 * @param info
83 * pointer to uio_info.
84 * @param irq_state
85 * state value. 1 to enable interrupt, 0 to disable interrupt.
86 */
87static int
88hv_uio_irqcontrol(struct uio_info *info, s32 irq_state)
89{
90 struct hv_uio_private_data *pdata = info->priv;
91 struct hv_device *dev = pdata->device;
92 struct vmbus_channel *primary, *sc;
93
94 primary = dev->channel;
95 set_event(primary, irq_state);
96
97 mutex_lock(&vmbus_connection.channel_mutex);
98 list_for_each_entry(sc, &primary->sc_list, sc_list)
99 set_event(sc, irq_state);
100 mutex_unlock(&vmbus_connection.channel_mutex);
101
102 return 0;
103}
104
105/*
106 * Callback from vmbus_event when something is in inbound ring.
107 */
108static void hv_uio_channel_cb(void *context)
109{
110 struct vmbus_channel *chan = context;
111 struct hv_device *hv_dev;
112 struct hv_uio_private_data *pdata;
113
114 virt_mb();
115
116 /*
117 * The callback may come from a subchannel, in which case look
118 * for the hv device in the primary channel
119 */
120 hv_dev = chan->primary_channel ?
121 chan->primary_channel->device_obj : chan->device_obj;
122 pdata = hv_get_drvdata(hv_dev);
123 uio_event_notify(&pdata->info);
124}
125
126/*
127 * Callback from vmbus_event when channel is rescinded.
128 * It is meant for rescind of primary channels only.
129 */
130static void hv_uio_rescind(struct vmbus_channel *channel)
131{
132 struct hv_device *hv_dev = channel->device_obj;
133 struct hv_uio_private_data *pdata = hv_get_drvdata(hv_dev);
134
135 /*
136 * Turn off the interrupt file handle
137 * Next read for event will return -EIO
138 */
139 pdata->info.irq = 0;
140
141 /* Wake up reader */
142 uio_event_notify(&pdata->info);
143
144 /*
145 * With rescind callback registered, rescind path will not unregister the device
146 * from vmbus when the primary channel is rescinded.
147 * Without it, rescind handling is incomplete and next onoffer msg does not come.
148 * Unregister the device from vmbus here.
149 */
150 vmbus_device_unregister(channel->device_obj);
151}
152
153/* Function used for mmap of ring buffer sysfs interface.
154 * The ring buffer is allocated as contiguous memory by vmbus_open
155 */
156static int
157hv_uio_ring_mmap_prepare(struct vmbus_channel *channel, struct vm_area_desc *desc)
158{
159 void *ring_buffer = page_address(channel->ringbuffer_page);
160
161 if (channel->state != CHANNEL_OPENED_STATE)
162 return -ENODEV;
163
164 mmap_action_simple_ioremap(desc, virt_to_phys(ring_buffer),
165 channel->ringbuffer_pagecount << PAGE_SHIFT);
166 return 0;
167}
168
169/* Callback from VMBUS subsystem when new channel created. */
170static void
171hv_uio_new_channel(struct vmbus_channel *new_sc)
172{
173 struct hv_device *hv_dev = new_sc->primary_channel->device_obj;
174 struct device *device = &hv_dev->device;
175 const size_t ring_bytes = SZ_2M;
176 int ret;
177
178 /* Create host communication ring */
179 ret = vmbus_open(new_sc, ring_bytes, ring_bytes, NULL, 0,
180 hv_uio_channel_cb, new_sc);
181 if (ret) {
182 dev_err(device, "vmbus_open subchannel failed: %d\n", ret);
183 return;
184 }
185
186 set_channel_read_mode(new_sc, HV_CALL_ISR);
187 ret = hv_create_ring_sysfs(new_sc, hv_uio_ring_mmap_prepare);
188 if (ret) {
189 dev_err(device, "sysfs create ring bin file failed; %d\n", ret);
190 vmbus_close(new_sc);
191 }
192}
193
194/* free the reserved buffers for send and receive */
195static void
196hv_uio_cleanup(struct hv_device *dev, struct hv_uio_private_data *pdata)
197{
198 if (pdata->send_gpadl.gpadl_handle) {
199 vmbus_teardown_gpadl(dev->channel, &pdata->send_gpadl);
200 if (!pdata->send_gpadl.decrypted)
201 vfree(pdata->send_buf);
202 }
203
204 if (pdata->recv_gpadl.gpadl_handle) {
205 vmbus_teardown_gpadl(dev->channel, &pdata->recv_gpadl);
206 if (!pdata->recv_gpadl.decrypted)
207 vfree(pdata->recv_buf);
208 }
209}
210
211/* VMBus primary channel is opened on first use */
212static int
213hv_uio_open(struct uio_info *info, struct inode *inode)
214{
215 struct hv_uio_private_data *pdata
216 = container_of(info, struct hv_uio_private_data, info);
217 struct hv_device *dev = pdata->device;
218 int ret;
219
220 if (atomic_inc_return(&pdata->refcnt) != 1)
221 return 0;
222
223 vmbus_set_chn_rescind_callback(dev->channel, hv_uio_rescind);
224 vmbus_set_sc_create_callback(dev->channel, hv_uio_new_channel);
225
226 ret = vmbus_connect_ring(dev->channel,
227 hv_uio_channel_cb, dev->channel);
228 if (ret)
229 atomic_dec(&pdata->refcnt);
230
231 return ret;
232}
233
234/* VMBus primary channel is closed on last close */
235static int
236hv_uio_release(struct uio_info *info, struct inode *inode)
237{
238 struct hv_uio_private_data *pdata
239 = container_of(info, struct hv_uio_private_data, info);
240 struct hv_device *dev = pdata->device;
241 int ret = 0;
242
243 if (atomic_dec_and_test(&pdata->refcnt))
244 ret = vmbus_disconnect_ring(dev->channel);
245
246 return ret;
247}
248
249static int
250hv_uio_probe(struct hv_device *dev,
251 const struct hv_vmbus_device_id *dev_id)
252{
253 struct vmbus_channel *channel = dev->channel;
254 struct hv_uio_private_data *pdata;
255 void *ring_buffer;
256 int ret;
257 size_t ring_size = hv_dev_ring_size(channel);
258
259 if (!ring_size)
260 ring_size = SZ_2M;
261
262 /* Adjust ring size if necessary to have it page aligned */
263 ring_size = VMBUS_RING_SIZE(ring_size);
264
265 pdata = devm_kzalloc(&dev->device, sizeof(*pdata), GFP_KERNEL);
266 if (!pdata)
267 return -ENOMEM;
268
269 ret = vmbus_alloc_ring(channel, ring_size, ring_size);
270 if (ret)
271 return ret;
272
273 set_channel_read_mode(channel, HV_CALL_ISR);
274
275 /* Fill general uio info */
276 pdata->info.name = "uio_hv_generic";
277 pdata->info.version = DRIVER_VERSION;
278 pdata->info.irqcontrol = hv_uio_irqcontrol;
279 pdata->info.open = hv_uio_open;
280 pdata->info.release = hv_uio_release;
281 pdata->info.irq = UIO_IRQ_CUSTOM;
282 atomic_set(&pdata->refcnt, 0);
283
284 /* mem resources */
285 pdata->info.mem[TXRX_RING_MAP].name = "txrx_rings";
286 ring_buffer = page_address(channel->ringbuffer_page);
287 pdata->info.mem[TXRX_RING_MAP].addr
288 = (uintptr_t)virt_to_phys(ring_buffer);
289 pdata->info.mem[TXRX_RING_MAP].size
290 = channel->ringbuffer_pagecount << PAGE_SHIFT;
291 pdata->info.mem[TXRX_RING_MAP].memtype = UIO_MEM_IOVA;
292
293 pdata->info.mem[INT_PAGE_MAP].name = "int_page";
294 pdata->info.mem[INT_PAGE_MAP].addr
295 = (uintptr_t)vmbus_connection.int_page;
296 pdata->info.mem[INT_PAGE_MAP].size = HV_HYP_PAGE_SIZE;
297 pdata->info.mem[INT_PAGE_MAP].memtype = UIO_MEM_LOGICAL;
298
299 pdata->info.mem[MON_PAGE_MAP].name = "monitor_page";
300 pdata->info.mem[MON_PAGE_MAP].addr
301 = (uintptr_t)vmbus_connection.monitor_pages[1];
302 pdata->info.mem[MON_PAGE_MAP].size = HV_HYP_PAGE_SIZE;
303 pdata->info.mem[MON_PAGE_MAP].memtype = UIO_MEM_LOGICAL;
304
305 if (channel->device_id == HV_NIC) {
306 pdata->recv_buf = vzalloc(RECV_BUFFER_SIZE);
307 if (!pdata->recv_buf) {
308 ret = -ENOMEM;
309 goto fail_free_ring;
310 }
311
312 ret = vmbus_establish_gpadl(channel, pdata->recv_buf,
313 RECV_BUFFER_SIZE, &pdata->recv_gpadl);
314 if (ret) {
315 if (!pdata->recv_gpadl.decrypted)
316 vfree(pdata->recv_buf);
317 goto fail_close;
318 }
319
320 /* put Global Physical Address Label in name */
321 snprintf(pdata->recv_name, sizeof(pdata->recv_name),
322 "recv:%u", pdata->recv_gpadl.gpadl_handle);
323 pdata->info.mem[RECV_BUF_MAP].name = pdata->recv_name;
324 pdata->info.mem[RECV_BUF_MAP].addr = (uintptr_t)pdata->recv_buf;
325 pdata->info.mem[RECV_BUF_MAP].size = RECV_BUFFER_SIZE;
326 pdata->info.mem[RECV_BUF_MAP].memtype = UIO_MEM_VIRTUAL;
327
328 pdata->send_buf = vzalloc(SEND_BUFFER_SIZE);
329 if (!pdata->send_buf) {
330 ret = -ENOMEM;
331 goto fail_close;
332 }
333
334 ret = vmbus_establish_gpadl(channel, pdata->send_buf,
335 SEND_BUFFER_SIZE, &pdata->send_gpadl);
336 if (ret) {
337 if (!pdata->send_gpadl.decrypted)
338 vfree(pdata->send_buf);
339 goto fail_close;
340 }
341
342 snprintf(pdata->send_name, sizeof(pdata->send_name),
343 "send:%u", pdata->send_gpadl.gpadl_handle);
344 pdata->info.mem[SEND_BUF_MAP].name = pdata->send_name;
345 pdata->info.mem[SEND_BUF_MAP].addr = (uintptr_t)pdata->send_buf;
346 pdata->info.mem[SEND_BUF_MAP].size = SEND_BUFFER_SIZE;
347 pdata->info.mem[SEND_BUF_MAP].memtype = UIO_MEM_VIRTUAL;
348 }
349
350 pdata->info.priv = pdata;
351 pdata->device = dev;
352
353 ret = uio_register_device(&dev->device, &pdata->info);
354 if (ret) {
355 dev_err(&dev->device, "hv_uio register failed\n");
356 goto fail_close;
357 }
358
359 /*
360 * This internally calls sysfs_update_group, which returns a non-zero value if it executes
361 * before sysfs_create_group. This is expected as the 'ring' will be created later in
362 * vmbus_device_register() -> vmbus_add_channel_kobj(). Thus, no need to check the return
363 * value and print warning.
364 *
365 * Creating/exposing sysfs in driver probe is not encouraged as it can lead to race
366 * conditions with userspace. For backward compatibility, "ring" sysfs could not be removed
367 * or decoupled from uio_hv_generic probe. Userspace programs can make use of inotify
368 * APIs to make sure that ring is created.
369 */
370 hv_create_ring_sysfs(channel, hv_uio_ring_mmap_prepare);
371
372 hv_set_drvdata(dev, pdata);
373
374 return 0;
375
376fail_close:
377 hv_uio_cleanup(dev, pdata);
378fail_free_ring:
379 vmbus_free_ring(dev->channel);
380
381 return ret;
382}
383
384static void
385hv_uio_remove(struct hv_device *dev)
386{
387 struct hv_uio_private_data *pdata = hv_get_drvdata(dev);
388
389 if (!pdata)
390 return;
391
392 hv_remove_ring_sysfs(dev->channel);
393 uio_unregister_device(&pdata->info);
394 hv_uio_cleanup(dev, pdata);
395
396 vmbus_free_ring(dev->channel);
397}
398
399static struct hv_driver hv_uio_drv = {
400 .name = "uio_hv_generic",
401 .id_table = NULL, /* only dynamic id's */
402 .probe = hv_uio_probe,
403 .remove = hv_uio_remove,
404};
405
406static int __init
407hyperv_module_init(void)
408{
409 return vmbus_driver_register(&hv_uio_drv);
410}
411
412static void __exit
413hyperv_module_exit(void)
414{
415 vmbus_driver_unregister(&hv_uio_drv);
416}
417
418module_init(hyperv_module_init);
419module_exit(hyperv_module_exit);
420
421MODULE_VERSION(DRIVER_VERSION);
422MODULE_LICENSE("GPL v2");
423MODULE_AUTHOR(DRIVER_AUTHOR);
424MODULE_DESCRIPTION(DRIVER_DESC);