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
 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 */
 77static int
 78hv_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 */
 95static 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 */
111static 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/* Sysfs API to allow mmap of the ring buffers
135 * The ring buffer is allocated as contiguous memory by vmbus_open
136 */
137static int hv_uio_ring_mmap(struct file *filp, struct kobject *kobj,
138			    const struct bin_attribute *attr,
139			    struct vm_area_struct *vma)
140{
141	struct vmbus_channel *channel
142		= container_of(kobj, struct vmbus_channel, kobj);
143	void *ring_buffer = page_address(channel->ringbuffer_page);
144
145	if (channel->state != CHANNEL_OPENED_STATE)
146		return -ENODEV;
147
148	return vm_iomap_memory(vma, virt_to_phys(ring_buffer),
149			       channel->ringbuffer_pagecount << PAGE_SHIFT);
150}
151
152static const struct bin_attribute ring_buffer_bin_attr = {
153	.attr = {
154		.name = "ring",
155		.mode = 0600,
156	},
157	.size = 2 * SZ_2M,
158	.mmap = hv_uio_ring_mmap,
159};
160
161/* Callback from VMBUS subsystem when new channel created. */
162static void
163hv_uio_new_channel(struct vmbus_channel *new_sc)
164{
165	struct hv_device *hv_dev = new_sc->primary_channel->device_obj;
166	struct device *device = &hv_dev->device;
167	const size_t ring_bytes = SZ_2M;
168	int ret;
169
170	/* Create host communication ring */
171	ret = vmbus_open(new_sc, ring_bytes, ring_bytes, NULL, 0,
172			 hv_uio_channel_cb, new_sc);
173	if (ret) {
174		dev_err(device, "vmbus_open subchannel failed: %d\n", ret);
175		return;
176	}
177
178	/* Disable interrupts on sub channel */
179	new_sc->inbound.ring_buffer->interrupt_mask = 1;
180	set_channel_read_mode(new_sc, HV_CALL_ISR);
181
182	ret = sysfs_create_bin_file(&new_sc->kobj, &ring_buffer_bin_attr);
183	if (ret) {
184		dev_err(device, "sysfs create ring bin file failed; %d\n", ret);
185		vmbus_close(new_sc);
186	}
187}
188
189/* free the reserved buffers for send and receive */
190static void
191hv_uio_cleanup(struct hv_device *dev, struct hv_uio_private_data *pdata)
192{
193	if (pdata->send_gpadl.gpadl_handle) {
194		vmbus_teardown_gpadl(dev->channel, &pdata->send_gpadl);
195		if (!pdata->send_gpadl.decrypted)
196			vfree(pdata->send_buf);
197	}
198
199	if (pdata->recv_gpadl.gpadl_handle) {
200		vmbus_teardown_gpadl(dev->channel, &pdata->recv_gpadl);
201		if (!pdata->recv_gpadl.decrypted)
202			vfree(pdata->recv_buf);
203	}
204}
205
206/* VMBus primary channel is opened on first use */
207static int
208hv_uio_open(struct uio_info *info, struct inode *inode)
209{
210	struct hv_uio_private_data *pdata
211		= container_of(info, struct hv_uio_private_data, info);
212	struct hv_device *dev = pdata->device;
213	int ret;
214
215	if (atomic_inc_return(&pdata->refcnt) != 1)
216		return 0;
217
218	vmbus_set_chn_rescind_callback(dev->channel, hv_uio_rescind);
219	vmbus_set_sc_create_callback(dev->channel, hv_uio_new_channel);
220
221	ret = vmbus_connect_ring(dev->channel,
222				 hv_uio_channel_cb, dev->channel);
223	if (ret == 0)
224		dev->channel->inbound.ring_buffer->interrupt_mask = 1;
225	else
226		atomic_dec(&pdata->refcnt);
227
228	return ret;
229}
230
231/* VMBus primary channel is closed on last close */
232static int
233hv_uio_release(struct uio_info *info, struct inode *inode)
234{
235	struct hv_uio_private_data *pdata
236		= container_of(info, struct hv_uio_private_data, info);
237	struct hv_device *dev = pdata->device;
238	int ret = 0;
239
240	if (atomic_dec_and_test(&pdata->refcnt))
241		ret = vmbus_disconnect_ring(dev->channel);
242
243	return ret;
244}
245
246static int
247hv_uio_probe(struct hv_device *dev,
248	     const struct hv_vmbus_device_id *dev_id)
249{
250	struct vmbus_channel *channel = dev->channel;
251	struct hv_uio_private_data *pdata;
252	void *ring_buffer;
253	int ret;
254	size_t ring_size = hv_dev_ring_size(channel);
255
256	if (!ring_size)
257		ring_size = SZ_2M;
258
259	pdata = devm_kzalloc(&dev->device, sizeof(*pdata), GFP_KERNEL);
260	if (!pdata)
261		return -ENOMEM;
262
263	ret = vmbus_alloc_ring(channel, ring_size, ring_size);
264	if (ret)
265		return ret;
266
267	set_channel_read_mode(channel, HV_CALL_ISR);
268
269	/* Fill general uio info */
270	pdata->info.name = "uio_hv_generic";
271	pdata->info.version = DRIVER_VERSION;
272	pdata->info.irqcontrol = hv_uio_irqcontrol;
273	pdata->info.open = hv_uio_open;
274	pdata->info.release = hv_uio_release;
275	pdata->info.irq = UIO_IRQ_CUSTOM;
276	atomic_set(&pdata->refcnt, 0);
277
278	/* mem resources */
279	pdata->info.mem[TXRX_RING_MAP].name = "txrx_rings";
280	ring_buffer = page_address(channel->ringbuffer_page);
281	pdata->info.mem[TXRX_RING_MAP].addr
282		= (uintptr_t)virt_to_phys(ring_buffer);
283	pdata->info.mem[TXRX_RING_MAP].size
284		= channel->ringbuffer_pagecount << PAGE_SHIFT;
285	pdata->info.mem[TXRX_RING_MAP].memtype = UIO_MEM_IOVA;
286
287	pdata->info.mem[INT_PAGE_MAP].name = "int_page";
288	pdata->info.mem[INT_PAGE_MAP].addr
289		= (uintptr_t)vmbus_connection.int_page;
290	pdata->info.mem[INT_PAGE_MAP].size = PAGE_SIZE;
291	pdata->info.mem[INT_PAGE_MAP].memtype = UIO_MEM_LOGICAL;
292
293	pdata->info.mem[MON_PAGE_MAP].name = "monitor_page";
294	pdata->info.mem[MON_PAGE_MAP].addr
295		= (uintptr_t)vmbus_connection.monitor_pages[1];
296	pdata->info.mem[MON_PAGE_MAP].size = PAGE_SIZE;
297	pdata->info.mem[MON_PAGE_MAP].memtype = UIO_MEM_LOGICAL;
298
299	pdata->recv_buf = vzalloc(RECV_BUFFER_SIZE);
300	if (pdata->recv_buf == NULL) {
301		ret = -ENOMEM;
302		goto fail_free_ring;
303	}
304
305	ret = vmbus_establish_gpadl(channel, pdata->recv_buf,
306				    RECV_BUFFER_SIZE, &pdata->recv_gpadl);
307	if (ret) {
308		if (!pdata->recv_gpadl.decrypted)
309			vfree(pdata->recv_buf);
310		goto fail_close;
311	}
312
313	/* put Global Physical Address Label in name */
314	snprintf(pdata->recv_name, sizeof(pdata->recv_name),
315		 "recv:%u", pdata->recv_gpadl.gpadl_handle);
316	pdata->info.mem[RECV_BUF_MAP].name = pdata->recv_name;
317	pdata->info.mem[RECV_BUF_MAP].addr
318		= (uintptr_t)pdata->recv_buf;
319	pdata->info.mem[RECV_BUF_MAP].size = RECV_BUFFER_SIZE;
320	pdata->info.mem[RECV_BUF_MAP].memtype = UIO_MEM_VIRTUAL;
321
322	pdata->send_buf = vzalloc(SEND_BUFFER_SIZE);
323	if (pdata->send_buf == NULL) {
324		ret = -ENOMEM;
325		goto fail_close;
326	}
327
328	ret = vmbus_establish_gpadl(channel, pdata->send_buf,
329				    SEND_BUFFER_SIZE, &pdata->send_gpadl);
330	if (ret) {
331		if (!pdata->send_gpadl.decrypted)
332			vfree(pdata->send_buf);
333		goto fail_close;
334	}
335
336	snprintf(pdata->send_name, sizeof(pdata->send_name),
337		 "send:%u", pdata->send_gpadl.gpadl_handle);
338	pdata->info.mem[SEND_BUF_MAP].name = pdata->send_name;
339	pdata->info.mem[SEND_BUF_MAP].addr
340		= (uintptr_t)pdata->send_buf;
341	pdata->info.mem[SEND_BUF_MAP].size = SEND_BUFFER_SIZE;
342	pdata->info.mem[SEND_BUF_MAP].memtype = UIO_MEM_VIRTUAL;
343
344	pdata->info.priv = pdata;
345	pdata->device = dev;
346
347	ret = uio_register_device(&dev->device, &pdata->info);
348	if (ret) {
349		dev_err(&dev->device, "hv_uio register failed\n");
350		goto fail_close;
351	}
352
353	ret = sysfs_create_bin_file(&channel->kobj, &ring_buffer_bin_attr);
354	if (ret)
355		dev_notice(&dev->device,
356			   "sysfs create ring bin file failed; %d\n", ret);
357
358	hv_set_drvdata(dev, pdata);
359
360	return 0;
361
362fail_close:
363	hv_uio_cleanup(dev, pdata);
364fail_free_ring:
365	vmbus_free_ring(dev->channel);
366
367	return ret;
368}
369
370static void
371hv_uio_remove(struct hv_device *dev)
372{
373	struct hv_uio_private_data *pdata = hv_get_drvdata(dev);
374
375	if (!pdata)
376		return;
377
378	sysfs_remove_bin_file(&dev->channel->kobj, &ring_buffer_bin_attr);
379	uio_unregister_device(&pdata->info);
380	hv_uio_cleanup(dev, pdata);
381
382	vmbus_free_ring(dev->channel);
383}
384
385static struct hv_driver hv_uio_drv = {
386	.name = "uio_hv_generic",
387	.id_table = NULL, /* only dynamic id's */
388	.probe = hv_uio_probe,
389	.remove = hv_uio_remove,
390};
391
392static int __init
393hyperv_module_init(void)
394{
395	return vmbus_driver_register(&hv_uio_drv);
396}
397
398static void __exit
399hyperv_module_exit(void)
400{
401	vmbus_driver_unregister(&hv_uio_drv);
402}
403
404module_init(hyperv_module_init);
405module_exit(hyperv_module_exit);
406
407MODULE_VERSION(DRIVER_VERSION);
408MODULE_LICENSE("GPL v2");
409MODULE_AUTHOR(DRIVER_AUTHOR);
410MODULE_DESCRIPTION(DRIVER_DESC);