1 // SPDX-License-Identifier: GPL-2.0 OR MIT
2 /*
3 * Copyright 2020-2022 Advanced Micro Devices, Inc.
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be included in
13 * all copies or substantial portions of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
19 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
20 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
21 * OTHER DEALINGS IN THE SOFTWARE.
22 */
23
24 #include <linux/poll.h>
25 #include <linux/wait.h>
26 #include <linux/anon_inodes.h>
27 #include <uapi/linux/kfd_ioctl.h>
28 #include "amdgpu.h"
29 #include "amdgpu_vm.h"
30 #include "kfd_priv.h"
31 #include "kfd_smi_events.h"
32
33 struct kfd_smi_client {
34 struct list_head list;
35 struct kfifo fifo;
36 wait_queue_head_t wait_queue;
37 /* events enabled */
38 uint64_t events;
39 struct kfd_dev *dev;
40 spinlock_t lock;
41 };
42
43 #define MAX_KFIFO_SIZE 1024
44
45 static __poll_t kfd_smi_ev_poll(struct file *, struct poll_table_struct *);
46 static ssize_t kfd_smi_ev_read(struct file *, char __user *, size_t, loff_t *);
47 static ssize_t kfd_smi_ev_write(struct file *, const char __user *, size_t,
48 loff_t *);
49 static int kfd_smi_ev_release(struct inode *, struct file *);
50
51 static const char kfd_smi_name[] = "kfd_smi_ev";
52
53 static const struct file_operations kfd_smi_ev_fops = {
54 .owner = THIS_MODULE,
55 .poll = kfd_smi_ev_poll,
56 .read = kfd_smi_ev_read,
57 .write = kfd_smi_ev_write,
58 .release = kfd_smi_ev_release
59 };
60
kfd_smi_ev_poll(struct file * filep,struct poll_table_struct * wait)61 static __poll_t kfd_smi_ev_poll(struct file *filep,
62 struct poll_table_struct *wait)
63 {
64 struct kfd_smi_client *client = filep->private_data;
65 __poll_t mask = 0;
66
67 poll_wait(filep, &client->wait_queue, wait);
68
69 spin_lock(&client->lock);
70 if (!kfifo_is_empty(&client->fifo))
71 mask = EPOLLIN | EPOLLRDNORM;
72 spin_unlock(&client->lock);
73
74 return mask;
75 }
76
kfd_smi_ev_read(struct file * filep,char __user * user,size_t size,loff_t * offset)77 static ssize_t kfd_smi_ev_read(struct file *filep, char __user *user,
78 size_t size, loff_t *offset)
79 {
80 int ret;
81 size_t to_copy;
82 struct kfd_smi_client *client = filep->private_data;
83 unsigned char *buf;
84
85 size = min_t(size_t, size, MAX_KFIFO_SIZE);
86 buf = kmalloc(size, GFP_KERNEL);
87 if (!buf)
88 return -ENOMEM;
89
90 /* kfifo_to_user can sleep so we can't use spinlock protection around
91 * it. Instead, we kfifo out as spinlocked then copy them to the user.
92 */
93 spin_lock(&client->lock);
94 to_copy = kfifo_len(&client->fifo);
95 if (!to_copy) {
96 spin_unlock(&client->lock);
97 ret = -EAGAIN;
98 goto ret_err;
99 }
100 to_copy = min(size, to_copy);
101 ret = kfifo_out(&client->fifo, buf, to_copy);
102 spin_unlock(&client->lock);
103 if (ret <= 0) {
104 ret = -EAGAIN;
105 goto ret_err;
106 }
107
108 ret = copy_to_user(user, buf, to_copy);
109 if (ret) {
110 ret = -EFAULT;
111 goto ret_err;
112 }
113
114 kfree(buf);
115 return to_copy;
116
117 ret_err:
118 kfree(buf);
119 return ret;
120 }
121
kfd_smi_ev_write(struct file * filep,const char __user * user,size_t size,loff_t * offset)122 static ssize_t kfd_smi_ev_write(struct file *filep, const char __user *user,
123 size_t size, loff_t *offset)
124 {
125 struct kfd_smi_client *client = filep->private_data;
126 uint64_t events;
127
128 if (!access_ok(user, size) || size < sizeof(events))
129 return -EFAULT;
130 if (copy_from_user(&events, user, sizeof(events)))
131 return -EFAULT;
132
133 WRITE_ONCE(client->events, events);
134
135 return sizeof(events);
136 }
137
kfd_smi_ev_release(struct inode * inode,struct file * filep)138 static int kfd_smi_ev_release(struct inode *inode, struct file *filep)
139 {
140 struct kfd_smi_client *client = filep->private_data;
141 struct kfd_dev *dev = client->dev;
142
143 spin_lock(&dev->smi_lock);
144 list_del_rcu(&client->list);
145 spin_unlock(&dev->smi_lock);
146
147 synchronize_rcu();
148 kfifo_free(&client->fifo);
149 kfree(client);
150
151 return 0;
152 }
153
add_event_to_kfifo(struct kfd_dev * dev,unsigned int smi_event,char * event_msg,int len)154 static void add_event_to_kfifo(struct kfd_dev *dev, unsigned int smi_event,
155 char *event_msg, int len)
156 {
157 struct kfd_smi_client *client;
158
159 rcu_read_lock();
160
161 list_for_each_entry_rcu(client, &dev->smi_clients, list) {
162 if (!(READ_ONCE(client->events) &
163 KFD_SMI_EVENT_MASK_FROM_INDEX(smi_event)))
164 continue;
165 spin_lock(&client->lock);
166 if (kfifo_avail(&client->fifo) >= len) {
167 kfifo_in(&client->fifo, event_msg, len);
168 wake_up_all(&client->wait_queue);
169 } else {
170 pr_debug("smi_event(EventID: %u): no space left\n",
171 smi_event);
172 }
173 spin_unlock(&client->lock);
174 }
175
176 rcu_read_unlock();
177 }
178
179 __printf(3, 4)
kfd_smi_event_add(struct kfd_dev * dev,unsigned int event,char * fmt,...)180 static void kfd_smi_event_add(struct kfd_dev *dev, unsigned int event,
181 char *fmt, ...)
182 {
183 char fifo_in[KFD_SMI_EVENT_MSG_SIZE];
184 int len;
185 va_list args;
186
187 if (list_empty(&dev->smi_clients))
188 return;
189
190 len = snprintf(fifo_in, sizeof(fifo_in), "%x ", event);
191
192 va_start(args, fmt);
193 len += vsnprintf(fifo_in + len, sizeof(fifo_in) - len, fmt, args);
194 va_end(args);
195
196 add_event_to_kfifo(dev, event, fifo_in, len);
197 }
198
kfd_smi_event_update_gpu_reset(struct kfd_dev * dev,bool post_reset)199 void kfd_smi_event_update_gpu_reset(struct kfd_dev *dev, bool post_reset)
200 {
201 unsigned int event;
202
203 if (post_reset) {
204 event = KFD_SMI_EVENT_GPU_POST_RESET;
205 } else {
206 event = KFD_SMI_EVENT_GPU_PRE_RESET;
207 ++(dev->reset_seq_num);
208 }
209 kfd_smi_event_add(dev, event, "%x\n", dev->reset_seq_num);
210 }
211
kfd_smi_event_update_thermal_throttling(struct kfd_dev * dev,uint64_t throttle_bitmask)212 void kfd_smi_event_update_thermal_throttling(struct kfd_dev *dev,
213 uint64_t throttle_bitmask)
214 {
215 kfd_smi_event_add(dev, KFD_SMI_EVENT_THERMAL_THROTTLE, "%llx:%llx\n",
216 throttle_bitmask,
217 amdgpu_dpm_get_thermal_throttling_counter(dev->adev));
218 }
219
kfd_smi_event_update_vmfault(struct kfd_dev * dev,uint16_t pasid)220 void kfd_smi_event_update_vmfault(struct kfd_dev *dev, uint16_t pasid)
221 {
222 struct amdgpu_task_info task_info;
223
224 memset(&task_info, 0, sizeof(struct amdgpu_task_info));
225 amdgpu_vm_get_task_info(dev->adev, pasid, &task_info);
226 /* Report VM faults from user applications, not retry from kernel */
227 if (!task_info.pid)
228 return;
229
230 kfd_smi_event_add(dev, KFD_SMI_EVENT_VMFAULT, "%x:%s\n",
231 task_info.pid, task_info.task_name);
232 }
233
kfd_smi_event_open(struct kfd_dev * dev,uint32_t * fd)234 int kfd_smi_event_open(struct kfd_dev *dev, uint32_t *fd)
235 {
236 struct kfd_smi_client *client;
237 int ret;
238
239 client = kzalloc(sizeof(struct kfd_smi_client), GFP_KERNEL);
240 if (!client)
241 return -ENOMEM;
242 INIT_LIST_HEAD(&client->list);
243
244 ret = kfifo_alloc(&client->fifo, MAX_KFIFO_SIZE, GFP_KERNEL);
245 if (ret) {
246 kfree(client);
247 return ret;
248 }
249
250 init_waitqueue_head(&client->wait_queue);
251 spin_lock_init(&client->lock);
252 client->events = 0;
253 client->dev = dev;
254
255 spin_lock(&dev->smi_lock);
256 list_add_rcu(&client->list, &dev->smi_clients);
257 spin_unlock(&dev->smi_lock);
258
259 ret = anon_inode_getfd(kfd_smi_name, &kfd_smi_ev_fops, (void *)client,
260 O_RDWR);
261 if (ret < 0) {
262 spin_lock(&dev->smi_lock);
263 list_del_rcu(&client->list);
264 spin_unlock(&dev->smi_lock);
265
266 synchronize_rcu();
267
268 kfifo_free(&client->fifo);
269 kfree(client);
270 return ret;
271 }
272 *fd = ret;
273
274 return 0;
275 }
276