xref: /linux/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c (revision 172cdcaefea5c297fdb3d20b7d5aff60ae4fbce6)
1 /*
2  * Copyright 2020 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  */
22 
23 #include <linux/poll.h>
24 #include <linux/wait.h>
25 #include <linux/anon_inodes.h>
26 #include <uapi/linux/kfd_ioctl.h>
27 #include "amdgpu.h"
28 #include "amdgpu_vm.h"
29 #include "kfd_priv.h"
30 #include "kfd_smi_events.h"
31 
32 struct kfd_smi_client {
33 	struct list_head list;
34 	struct kfifo fifo;
35 	wait_queue_head_t wait_queue;
36 	/* events enabled */
37 	uint64_t events;
38 	struct kfd_dev *dev;
39 	spinlock_t lock;
40 };
41 
42 #define MAX_KFIFO_SIZE	1024
43 
44 static __poll_t kfd_smi_ev_poll(struct file *, struct poll_table_struct *);
45 static ssize_t kfd_smi_ev_read(struct file *, char __user *, size_t, loff_t *);
46 static ssize_t kfd_smi_ev_write(struct file *, const char __user *, size_t,
47 				loff_t *);
48 static int kfd_smi_ev_release(struct inode *, struct file *);
49 
50 static const char kfd_smi_name[] = "kfd_smi_ev";
51 
52 static const struct file_operations kfd_smi_ev_fops = {
53 	.owner = THIS_MODULE,
54 	.poll = kfd_smi_ev_poll,
55 	.read = kfd_smi_ev_read,
56 	.write = kfd_smi_ev_write,
57 	.release = kfd_smi_ev_release
58 };
59 
60 static __poll_t kfd_smi_ev_poll(struct file *filep,
61 				struct poll_table_struct *wait)
62 {
63 	struct kfd_smi_client *client = filep->private_data;
64 	__poll_t mask = 0;
65 
66 	poll_wait(filep, &client->wait_queue, wait);
67 
68 	spin_lock(&client->lock);
69 	if (!kfifo_is_empty(&client->fifo))
70 		mask = EPOLLIN | EPOLLRDNORM;
71 	spin_unlock(&client->lock);
72 
73 	return mask;
74 }
75 
76 static ssize_t kfd_smi_ev_read(struct file *filep, char __user *user,
77 			       size_t size, loff_t *offset)
78 {
79 	int ret;
80 	size_t to_copy;
81 	struct kfd_smi_client *client = filep->private_data;
82 	unsigned char *buf;
83 
84 	buf = kmalloc_array(MAX_KFIFO_SIZE, sizeof(*buf), GFP_KERNEL);
85 	if (!buf)
86 		return -ENOMEM;
87 
88 	/* kfifo_to_user can sleep so we can't use spinlock protection around
89 	 * it. Instead, we kfifo out as spinlocked then copy them to the user.
90 	 */
91 	spin_lock(&client->lock);
92 	to_copy = kfifo_len(&client->fifo);
93 	if (!to_copy) {
94 		spin_unlock(&client->lock);
95 		ret = -EAGAIN;
96 		goto ret_err;
97 	}
98 	to_copy = min3(size, sizeof(buf), to_copy);
99 	ret = kfifo_out(&client->fifo, buf, to_copy);
100 	spin_unlock(&client->lock);
101 	if (ret <= 0) {
102 		ret = -EAGAIN;
103 		goto ret_err;
104 	}
105 
106 	ret = copy_to_user(user, buf, to_copy);
107 	if (ret) {
108 		ret = -EFAULT;
109 		goto ret_err;
110 	}
111 
112 	kfree(buf);
113 	return to_copy;
114 
115 ret_err:
116 	kfree(buf);
117 	return ret;
118 }
119 
120 static ssize_t kfd_smi_ev_write(struct file *filep, const char __user *user,
121 				size_t size, loff_t *offset)
122 {
123 	struct kfd_smi_client *client = filep->private_data;
124 	uint64_t events;
125 
126 	if (!access_ok(user, size) || size < sizeof(events))
127 		return -EFAULT;
128 	if (copy_from_user(&events, user, sizeof(events)))
129 		return -EFAULT;
130 
131 	WRITE_ONCE(client->events, events);
132 
133 	return sizeof(events);
134 }
135 
136 static int kfd_smi_ev_release(struct inode *inode, struct file *filep)
137 {
138 	struct kfd_smi_client *client = filep->private_data;
139 	struct kfd_dev *dev = client->dev;
140 
141 	spin_lock(&dev->smi_lock);
142 	list_del_rcu(&client->list);
143 	spin_unlock(&dev->smi_lock);
144 
145 	synchronize_rcu();
146 	kfifo_free(&client->fifo);
147 	kfree(client);
148 
149 	return 0;
150 }
151 
152 static void add_event_to_kfifo(struct kfd_dev *dev, unsigned int smi_event,
153 			      char *event_msg, int len)
154 {
155 	struct kfd_smi_client *client;
156 
157 	rcu_read_lock();
158 
159 	list_for_each_entry_rcu(client, &dev->smi_clients, list) {
160 		if (!(READ_ONCE(client->events) &
161 				KFD_SMI_EVENT_MASK_FROM_INDEX(smi_event)))
162 			continue;
163 		spin_lock(&client->lock);
164 		if (kfifo_avail(&client->fifo) >= len) {
165 			kfifo_in(&client->fifo, event_msg, len);
166 			wake_up_all(&client->wait_queue);
167 		} else {
168 			pr_debug("smi_event(EventID: %u): no space left\n",
169 					smi_event);
170 		}
171 		spin_unlock(&client->lock);
172 	}
173 
174 	rcu_read_unlock();
175 }
176 
177 void kfd_smi_event_update_gpu_reset(struct kfd_dev *dev, bool post_reset)
178 {
179 	/*
180 	 * GpuReset msg = Reset seq number (incremented for
181 	 * every reset message sent before GPU reset).
182 	 * 1 byte event + 1 byte space + 8 bytes seq num +
183 	 * 1 byte \n + 1 byte \0 = 12
184 	 */
185 	char fifo_in[12];
186 	int len;
187 	unsigned int event;
188 
189 	if (list_empty(&dev->smi_clients))
190 		return;
191 
192 	memset(fifo_in, 0x0, sizeof(fifo_in));
193 
194 	if (post_reset) {
195 		event = KFD_SMI_EVENT_GPU_POST_RESET;
196 	} else {
197 		event = KFD_SMI_EVENT_GPU_PRE_RESET;
198 		++(dev->reset_seq_num);
199 	}
200 
201 	len = snprintf(fifo_in, sizeof(fifo_in), "%x %x\n", event,
202 						dev->reset_seq_num);
203 
204 	add_event_to_kfifo(dev, event, fifo_in, len);
205 }
206 
207 void kfd_smi_event_update_thermal_throttling(struct kfd_dev *dev,
208 					     uint32_t throttle_bitmask)
209 {
210 	struct amdgpu_device *adev = (struct amdgpu_device *)dev->kgd;
211 	/*
212 	 * ThermalThrottle msg = throttle_bitmask(8):
213 	 * 			 thermal_interrupt_count(16):
214 	 * 1 byte event + 1 byte space + 8 byte throttle_bitmask +
215 	 * 1 byte : + 16 byte thermal_interupt_counter + 1 byte \n +
216 	 * 1 byte \0 = 29
217 	 */
218 	char fifo_in[29];
219 	int len;
220 
221 	if (list_empty(&dev->smi_clients))
222 		return;
223 
224 	len = snprintf(fifo_in, sizeof(fifo_in), "%x %x:%llx\n",
225 		       KFD_SMI_EVENT_THERMAL_THROTTLE, throttle_bitmask,
226 		       atomic64_read(&adev->smu.throttle_int_counter));
227 
228 	add_event_to_kfifo(dev, KFD_SMI_EVENT_THERMAL_THROTTLE,	fifo_in, len);
229 }
230 
231 void kfd_smi_event_update_vmfault(struct kfd_dev *dev, uint16_t pasid)
232 {
233 	struct amdgpu_device *adev = (struct amdgpu_device *)dev->kgd;
234 	struct amdgpu_task_info task_info;
235 	/* VmFault msg = (hex)uint32_pid(8) + :(1) + task name(16) = 25 */
236 	/* 1 byte event + 1 byte space + 25 bytes msg + 1 byte \n +
237 	 * 1 byte \0 = 29
238 	 */
239 	char fifo_in[29];
240 	int len;
241 
242 	if (list_empty(&dev->smi_clients))
243 		return;
244 
245 	memset(&task_info, 0, sizeof(struct amdgpu_task_info));
246 	amdgpu_vm_get_task_info(adev, pasid, &task_info);
247 	/* Report VM faults from user applications, not retry from kernel */
248 	if (!task_info.pid)
249 		return;
250 
251 	len = snprintf(fifo_in, sizeof(fifo_in), "%x %x:%s\n", KFD_SMI_EVENT_VMFAULT,
252 		task_info.pid, task_info.task_name);
253 
254 	add_event_to_kfifo(dev, KFD_SMI_EVENT_VMFAULT, fifo_in, len);
255 }
256 
257 int kfd_smi_event_open(struct kfd_dev *dev, uint32_t *fd)
258 {
259 	struct kfd_smi_client *client;
260 	int ret;
261 
262 	client = kzalloc(sizeof(struct kfd_smi_client), GFP_KERNEL);
263 	if (!client)
264 		return -ENOMEM;
265 	INIT_LIST_HEAD(&client->list);
266 
267 	ret = kfifo_alloc(&client->fifo, MAX_KFIFO_SIZE, GFP_KERNEL);
268 	if (ret) {
269 		kfree(client);
270 		return ret;
271 	}
272 
273 	ret = anon_inode_getfd(kfd_smi_name, &kfd_smi_ev_fops, (void *)client,
274 			       O_RDWR);
275 	if (ret < 0) {
276 		kfifo_free(&client->fifo);
277 		kfree(client);
278 		return ret;
279 	}
280 	*fd = ret;
281 
282 	init_waitqueue_head(&client->wait_queue);
283 	spin_lock_init(&client->lock);
284 	client->events = 0;
285 	client->dev = dev;
286 
287 	spin_lock(&dev->smi_lock);
288 	list_add_rcu(&client->list, &dev->smi_clients);
289 	spin_unlock(&dev->smi_lock);
290 
291 	return 0;
292 }
293