1 // SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause
2 /*
3 * Copyright(c) 2018 Intel Corporation.
4 */
5
6 #include <linux/debugfs.h>
7 #include <linux/seq_file.h>
8 #include <linux/kernel.h>
9 #include <linux/types.h>
10 #include <linux/bitmap.h>
11
12 #include "debugfs.h"
13 #include "fault.h"
14 #include "trace.h"
15
16 #define HFI1_FAULT_DIR_TX BIT(0)
17 #define HFI1_FAULT_DIR_RX BIT(1)
18 #define HFI1_FAULT_DIR_TXRX (HFI1_FAULT_DIR_TX | HFI1_FAULT_DIR_RX)
19
_fault_stats_seq_start(struct seq_file * s,loff_t * pos)20 static void *_fault_stats_seq_start(struct seq_file *s, loff_t *pos)
21 {
22 struct hfi1_opcode_stats_perctx *opstats;
23
24 if (*pos >= ARRAY_SIZE(opstats->stats))
25 return NULL;
26 return pos;
27 }
28
_fault_stats_seq_next(struct seq_file * s,void * v,loff_t * pos)29 static void *_fault_stats_seq_next(struct seq_file *s, void *v, loff_t *pos)
30 {
31 struct hfi1_opcode_stats_perctx *opstats;
32
33 ++*pos;
34 if (*pos >= ARRAY_SIZE(opstats->stats))
35 return NULL;
36 return pos;
37 }
38
_fault_stats_seq_stop(struct seq_file * s,void * v)39 static void _fault_stats_seq_stop(struct seq_file *s, void *v)
40 {
41 }
42
_fault_stats_seq_show(struct seq_file * s,void * v)43 static int _fault_stats_seq_show(struct seq_file *s, void *v)
44 {
45 loff_t *spos = v;
46 loff_t i = *spos, j;
47 u64 n_packets = 0, n_bytes = 0;
48 struct hfi1_ibdev *ibd = (struct hfi1_ibdev *)s->private;
49 struct hfi1_devdata *dd = dd_from_dev(ibd);
50 struct hfi1_ctxtdata *rcd;
51
52 for (j = 0; j < dd->first_dyn_alloc_ctxt; j++) {
53 rcd = hfi1_rcd_get_by_index(dd, j);
54 if (rcd) {
55 n_packets += rcd->opstats->stats[i].n_packets;
56 n_bytes += rcd->opstats->stats[i].n_bytes;
57 }
58 hfi1_rcd_put(rcd);
59 }
60 for_each_possible_cpu(j) {
61 struct hfi1_opcode_stats_perctx *sp =
62 per_cpu_ptr(dd->tx_opstats, j);
63
64 n_packets += sp->stats[i].n_packets;
65 n_bytes += sp->stats[i].n_bytes;
66 }
67 if (!n_packets && !n_bytes)
68 return SEQ_SKIP;
69 if (!ibd->fault->n_rxfaults[i] && !ibd->fault->n_txfaults[i])
70 return SEQ_SKIP;
71 seq_printf(s, "%02llx %llu/%llu (faults rx:%llu faults: tx:%llu)\n", i,
72 (unsigned long long)n_packets,
73 (unsigned long long)n_bytes,
74 (unsigned long long)ibd->fault->n_rxfaults[i],
75 (unsigned long long)ibd->fault->n_txfaults[i]);
76 return 0;
77 }
78
79 DEBUGFS_SEQ_FILE_OPS(fault_stats);
80 DEBUGFS_SEQ_FILE_OPEN(fault_stats);
81 DEBUGFS_FILE_OPS(fault_stats);
82
fault_opcodes_open(struct inode * inode,struct file * file)83 static int fault_opcodes_open(struct inode *inode, struct file *file)
84 {
85 file->private_data = inode->i_private;
86 return nonseekable_open(inode, file);
87 }
88
fault_opcodes_write(struct file * file,const char __user * buf,size_t len,loff_t * pos)89 static ssize_t fault_opcodes_write(struct file *file, const char __user *buf,
90 size_t len, loff_t *pos)
91 {
92 ssize_t ret = 0;
93 /* 1280 = 256 opcodes * 4 chars/opcode + 255 commas + NULL */
94 size_t copy, datalen = 1280;
95 char *data, *token, *ptr, *end;
96 struct fault *fault = file->private_data;
97
98 data = kcalloc(datalen, sizeof(*data), GFP_KERNEL);
99 if (!data)
100 return -ENOMEM;
101 copy = min(len, datalen - 1);
102 if (copy_from_user(data, buf, copy)) {
103 ret = -EFAULT;
104 goto free_data;
105 }
106
107 ret = debugfs_file_get(file->f_path.dentry);
108 if (unlikely(ret))
109 goto free_data;
110 ptr = data;
111 token = ptr;
112 for (ptr = data; *ptr; ptr = end + 1, token = ptr) {
113 char *dash;
114 unsigned long range_start, range_end, i;
115 bool remove = false;
116 unsigned long bound = 1U << BITS_PER_BYTE;
117
118 end = strchr(ptr, ',');
119 if (end)
120 *end = '\0';
121 if (token[0] == '-') {
122 remove = true;
123 token++;
124 }
125 dash = strchr(token, '-');
126 if (dash)
127 *dash = '\0';
128 if (kstrtoul(token, 0, &range_start))
129 break;
130 if (dash) {
131 token = dash + 1;
132 if (kstrtoul(token, 0, &range_end))
133 break;
134 } else {
135 range_end = range_start;
136 }
137 if (range_start == range_end && range_start == -1UL) {
138 bitmap_zero(fault->opcodes, sizeof(fault->opcodes) *
139 BITS_PER_BYTE);
140 break;
141 }
142 /* Check the inputs */
143 if (range_start >= bound || range_end >= bound)
144 break;
145
146 for (i = range_start; i <= range_end; i++) {
147 if (remove)
148 clear_bit(i, fault->opcodes);
149 else
150 set_bit(i, fault->opcodes);
151 }
152 if (!end)
153 break;
154 }
155 ret = len;
156
157 debugfs_file_put(file->f_path.dentry);
158 free_data:
159 kfree(data);
160 return ret;
161 }
162
fault_opcodes_read(struct file * file,char __user * buf,size_t len,loff_t * pos)163 static ssize_t fault_opcodes_read(struct file *file, char __user *buf,
164 size_t len, loff_t *pos)
165 {
166 ssize_t ret = 0;
167 char *data;
168 size_t datalen = 1280, size = 0; /* see fault_opcodes_write() */
169 unsigned long bit = 0, zero = 0;
170 struct fault *fault = file->private_data;
171 size_t bitsize = sizeof(fault->opcodes) * BITS_PER_BYTE;
172
173 data = kcalloc(datalen, sizeof(*data), GFP_KERNEL);
174 if (!data)
175 return -ENOMEM;
176 ret = debugfs_file_get(file->f_path.dentry);
177 if (unlikely(ret))
178 goto free_data;
179 bit = find_first_bit(fault->opcodes, bitsize);
180 while (bit < bitsize) {
181 zero = find_next_zero_bit(fault->opcodes, bitsize, bit);
182 if (zero - 1 != bit)
183 size += scnprintf(data + size,
184 datalen - size - 1,
185 "0x%lx-0x%lx,", bit, zero - 1);
186 else
187 size += scnprintf(data + size,
188 datalen - size - 1, "0x%lx,",
189 bit);
190 bit = find_next_bit(fault->opcodes, bitsize, zero);
191 }
192 debugfs_file_put(file->f_path.dentry);
193 data[size - 1] = '\n';
194 data[size] = '\0';
195 ret = simple_read_from_buffer(buf, len, pos, data, size);
196 free_data:
197 kfree(data);
198 return ret;
199 }
200
201 static const struct file_operations __fault_opcodes_fops = {
202 .owner = THIS_MODULE,
203 .open = fault_opcodes_open,
204 .read = fault_opcodes_read,
205 .write = fault_opcodes_write,
206 .llseek = no_llseek
207 };
208
hfi1_fault_exit_debugfs(struct hfi1_ibdev * ibd)209 void hfi1_fault_exit_debugfs(struct hfi1_ibdev *ibd)
210 {
211 if (ibd->fault)
212 debugfs_remove_recursive(ibd->fault->dir);
213 kfree(ibd->fault);
214 ibd->fault = NULL;
215 }
216
hfi1_fault_init_debugfs(struct hfi1_ibdev * ibd)217 int hfi1_fault_init_debugfs(struct hfi1_ibdev *ibd)
218 {
219 struct dentry *parent = ibd->hfi1_ibdev_dbg;
220 struct dentry *fault_dir;
221
222 ibd->fault = kzalloc(sizeof(*ibd->fault), GFP_KERNEL);
223 if (!ibd->fault)
224 return -ENOMEM;
225
226 ibd->fault->attr.interval = 1;
227 ibd->fault->attr.require_end = ULONG_MAX;
228 ibd->fault->attr.stacktrace_depth = 32;
229 ibd->fault->attr.dname = NULL;
230 ibd->fault->attr.verbose = 0;
231 ibd->fault->enable = false;
232 ibd->fault->opcode = false;
233 ibd->fault->fault_skip = 0;
234 ibd->fault->skip = 0;
235 ibd->fault->direction = HFI1_FAULT_DIR_TXRX;
236 ibd->fault->suppress_err = false;
237 bitmap_zero(ibd->fault->opcodes,
238 sizeof(ibd->fault->opcodes) * BITS_PER_BYTE);
239
240 fault_dir =
241 fault_create_debugfs_attr("fault", parent, &ibd->fault->attr);
242 if (IS_ERR(fault_dir)) {
243 kfree(ibd->fault);
244 ibd->fault = NULL;
245 return -ENOENT;
246 }
247 ibd->fault->dir = fault_dir;
248
249 debugfs_create_file("fault_stats", 0444, fault_dir, ibd,
250 &_fault_stats_file_ops);
251 debugfs_create_bool("enable", 0600, fault_dir, &ibd->fault->enable);
252 debugfs_create_bool("suppress_err", 0600, fault_dir,
253 &ibd->fault->suppress_err);
254 debugfs_create_bool("opcode_mode", 0600, fault_dir,
255 &ibd->fault->opcode);
256 debugfs_create_file("opcodes", 0600, fault_dir, ibd->fault,
257 &__fault_opcodes_fops);
258 debugfs_create_u64("skip_pkts", 0600, fault_dir,
259 &ibd->fault->fault_skip);
260 debugfs_create_u64("skip_usec", 0600, fault_dir,
261 &ibd->fault->fault_skip_usec);
262 debugfs_create_u8("direction", 0600, fault_dir, &ibd->fault->direction);
263
264 return 0;
265 }
266
hfi1_dbg_fault_suppress_err(struct hfi1_ibdev * ibd)267 bool hfi1_dbg_fault_suppress_err(struct hfi1_ibdev *ibd)
268 {
269 if (ibd->fault)
270 return ibd->fault->suppress_err;
271 return false;
272 }
273
__hfi1_should_fault(struct hfi1_ibdev * ibd,u32 opcode,u8 direction)274 static bool __hfi1_should_fault(struct hfi1_ibdev *ibd, u32 opcode,
275 u8 direction)
276 {
277 bool ret = false;
278
279 if (!ibd->fault || !ibd->fault->enable)
280 return false;
281 if (!(ibd->fault->direction & direction))
282 return false;
283 if (ibd->fault->opcode) {
284 if (bitmap_empty(ibd->fault->opcodes,
285 (sizeof(ibd->fault->opcodes) *
286 BITS_PER_BYTE)))
287 return false;
288 if (!(test_bit(opcode, ibd->fault->opcodes)))
289 return false;
290 }
291 if (ibd->fault->fault_skip_usec &&
292 time_before(jiffies, ibd->fault->skip_usec))
293 return false;
294 if (ibd->fault->fault_skip && ibd->fault->skip) {
295 ibd->fault->skip--;
296 return false;
297 }
298 ret = should_fail(&ibd->fault->attr, 1);
299 if (ret) {
300 ibd->fault->skip = ibd->fault->fault_skip;
301 ibd->fault->skip_usec = jiffies +
302 usecs_to_jiffies(ibd->fault->fault_skip_usec);
303 }
304 return ret;
305 }
306
hfi1_dbg_should_fault_tx(struct rvt_qp * qp,u32 opcode)307 bool hfi1_dbg_should_fault_tx(struct rvt_qp *qp, u32 opcode)
308 {
309 struct hfi1_ibdev *ibd = to_idev(qp->ibqp.device);
310
311 if (__hfi1_should_fault(ibd, opcode, HFI1_FAULT_DIR_TX)) {
312 trace_hfi1_fault_opcode(qp, opcode);
313 ibd->fault->n_txfaults[opcode]++;
314 return true;
315 }
316 return false;
317 }
318
hfi1_dbg_should_fault_rx(struct hfi1_packet * packet)319 bool hfi1_dbg_should_fault_rx(struct hfi1_packet *packet)
320 {
321 struct hfi1_ibdev *ibd = &packet->rcd->dd->verbs_dev;
322
323 if (__hfi1_should_fault(ibd, packet->opcode, HFI1_FAULT_DIR_RX)) {
324 trace_hfi1_fault_packet(packet);
325 ibd->fault->n_rxfaults[packet->opcode]++;
326 return true;
327 }
328 return false;
329 }
330