Whamcloud - gitweb
LU-12352 libcfs: crashes with certain cpu part numbers
[fs/lustre-release.git] / libcfs / libcfs / linux / linux-tracefile.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.gnu.org/licenses/gpl-2.0.html
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2012, Intel Corporation.
27  */
28 /*
29  * This file is part of Lustre, http://www.lustre.org/
30  * Lustre is a trademark of Sun Microsystems, Inc.
31  */
32
33 #define DEBUG_SUBSYSTEM S_LNET
34 #define LUSTRE_TRACEFILE_PRIVATE
35
36 #include <linux/slab.h>
37 #include <libcfs/libcfs.h>
38 #include "tracefile.h"
39
40 /* percents to share the total debug memory for each type */
41 static unsigned int pages_factor[CFS_TCD_TYPE_MAX] = {
42         80,  /* 80% pages for CFS_TCD_TYPE_PROC */
43         10,  /* 10% pages for CFS_TCD_TYPE_SOFTIRQ */
44         10   /* 10% pages for CFS_TCD_TYPE_IRQ */
45 };
46
47 char *cfs_trace_console_buffers[NR_CPUS][CFS_TCD_TYPE_MAX];
48
49 static DECLARE_RWSEM(cfs_tracefile_sem);
50
51 int cfs_tracefile_init_arch()
52 {
53         int    i;
54         int    j;
55         struct cfs_trace_cpu_data *tcd;
56
57         /* initialize trace_data */
58         memset(cfs_trace_data, 0, sizeof(cfs_trace_data));
59         for (i = 0; i < CFS_TCD_TYPE_MAX; i++) {
60                 cfs_trace_data[i] =
61                         kmalloc(sizeof(union cfs_trace_data_union) *
62                                 num_possible_cpus(), GFP_KERNEL);
63                 if (cfs_trace_data[i] == NULL)
64                         goto out;
65
66         }
67
68         /* arch related info initialized */
69         cfs_tcd_for_each(tcd, i, j) {
70                 spin_lock_init(&tcd->tcd_lock);
71                 tcd->tcd_pages_factor = pages_factor[i];
72                 tcd->tcd_type = i;
73                 tcd->tcd_cpu = j;
74         }
75
76         for (i = 0; i < num_possible_cpus(); i++)
77                 for (j = 0; j < 3; j++) {
78                         cfs_trace_console_buffers[i][j] =
79                                 kmalloc(CFS_TRACE_CONSOLE_BUFFER_SIZE,
80                                         GFP_KERNEL);
81
82                         if (cfs_trace_console_buffers[i][j] == NULL)
83                                 goto out;
84                 }
85
86         return 0;
87
88 out:
89         cfs_tracefile_fini_arch();
90         printk(KERN_ERR "lnet: Not enough memory\n");
91         return -ENOMEM;
92 }
93
94 void cfs_tracefile_fini_arch()
95 {
96         int    i;
97         int    j;
98
99         for (i = 0; i < num_possible_cpus(); i++)
100                 for (j = 0; j < 3; j++)
101                         if (cfs_trace_console_buffers[i][j] != NULL) {
102                                 kfree(cfs_trace_console_buffers[i][j]);
103                                 cfs_trace_console_buffers[i][j] = NULL;
104                         }
105
106         for (i = 0; cfs_trace_data[i] != NULL; i++) {
107                 kfree(cfs_trace_data[i]);
108                 cfs_trace_data[i] = NULL;
109         }
110 }
111
112 void cfs_tracefile_read_lock()
113 {
114         down_read(&cfs_tracefile_sem);
115 }
116
117 void cfs_tracefile_read_unlock()
118 {
119         up_read(&cfs_tracefile_sem);
120 }
121
122 void cfs_tracefile_write_lock()
123 {
124         down_write(&cfs_tracefile_sem);
125 }
126
127 void cfs_tracefile_write_unlock()
128 {
129         up_write(&cfs_tracefile_sem);
130 }
131
132 enum cfs_trace_buf_type cfs_trace_buf_idx_get()
133 {
134         if (in_irq())
135                 return CFS_TCD_TYPE_IRQ;
136         else if (in_softirq())
137                 return CFS_TCD_TYPE_SOFTIRQ;
138         else
139                 return CFS_TCD_TYPE_PROC;
140 }
141
142 /*
143  * The walking argument indicates the locking comes from all tcd types
144  * iterator and we must lock it and dissable local irqs to avoid deadlocks
145  * with other interrupt locks that might be happening. See LU-1311
146  * for details.
147  */
148 int cfs_trace_lock_tcd(struct cfs_trace_cpu_data *tcd, int walking)
149 __acquires(&tcd->tcd_lock)
150 {
151         __LASSERT(tcd->tcd_type < CFS_TCD_TYPE_MAX);
152         if (tcd->tcd_type == CFS_TCD_TYPE_IRQ)
153                 spin_lock_irqsave(&tcd->tcd_lock, tcd->tcd_lock_flags);
154         else if (tcd->tcd_type == CFS_TCD_TYPE_SOFTIRQ)
155                 spin_lock_bh(&tcd->tcd_lock);
156         else if (unlikely(walking))
157                 spin_lock_irq(&tcd->tcd_lock);
158         else
159                 spin_lock(&tcd->tcd_lock);
160         return 1;
161 }
162
163 void cfs_trace_unlock_tcd(struct cfs_trace_cpu_data *tcd, int walking)
164 __releases(&tcd->tcd_lock)
165 {
166         __LASSERT(tcd->tcd_type < CFS_TCD_TYPE_MAX);
167         if (tcd->tcd_type == CFS_TCD_TYPE_IRQ)
168                 spin_unlock_irqrestore(&tcd->tcd_lock, tcd->tcd_lock_flags);
169         else if (tcd->tcd_type == CFS_TCD_TYPE_SOFTIRQ)
170                 spin_unlock_bh(&tcd->tcd_lock);
171         else if (unlikely(walking))
172                 spin_unlock_irq(&tcd->tcd_lock);
173         else
174                 spin_unlock(&tcd->tcd_lock);
175 }
176
177 int cfs_tcd_owns_tage(struct cfs_trace_cpu_data *tcd,
178                       struct cfs_trace_page *tage)
179 {
180         /*
181          * XXX nikita: do NOT call portals_debug_msg() (CDEBUG/ENTRY/EXIT)
182          * from here: this will lead to infinite recursion.
183          */
184         return tcd->tcd_cpu == tage->cpu;
185 }
186
187 void
188 cfs_set_ptldebug_header(struct ptldebug_header *header,
189                         struct libcfs_debug_msg_data *msgdata,
190                         unsigned long stack)
191 {
192         struct timespec64 ts;
193
194         ktime_get_real_ts64(&ts);
195
196         header->ph_subsys = msgdata->msg_subsys;
197         header->ph_mask = msgdata->msg_mask;
198         header->ph_cpu_id = smp_processor_id();
199         header->ph_type = cfs_trace_buf_idx_get();
200         /* y2038 safe since all user space treats this as unsigned, but
201          * will overflow in 2106
202          */
203         header->ph_sec = (u32)ts.tv_sec;
204         header->ph_usec = ts.tv_nsec / NSEC_PER_USEC;
205         header->ph_stack = stack;
206         header->ph_pid = current->pid;
207         header->ph_line_num = msgdata->msg_line;
208         header->ph_extern_pid = 0;
209         return;
210 }
211
212 static char *
213 dbghdr_to_err_string(struct ptldebug_header *hdr)
214 {
215         switch (hdr->ph_subsys) {
216
217                 case S_LND:
218                 case S_LNET:
219                         return "LNetError";
220                 default:
221                         return "LustreError";
222         }
223 }
224
225 static char *
226 dbghdr_to_info_string(struct ptldebug_header *hdr)
227 {
228         switch (hdr->ph_subsys) {
229
230                 case S_LND:
231                 case S_LNET:
232                         return "LNet";
233                 default:
234                         return "Lustre";
235         }
236 }
237
238 void cfs_print_to_console(struct ptldebug_header *hdr, int mask,
239                           const char *buf, int len, const char *file,
240                           const char *fn)
241 {
242         char *prefix = "Lustre", *ptype = NULL;
243
244         if ((mask & D_EMERG) != 0) {
245                 prefix = dbghdr_to_err_string(hdr);
246                 ptype = KERN_EMERG;
247         } else if ((mask & D_ERROR) != 0) {
248                 prefix = dbghdr_to_err_string(hdr);
249                 ptype = KERN_ERR;
250         } else if ((mask & D_WARNING) != 0) {
251                 prefix = dbghdr_to_info_string(hdr);
252                 ptype = KERN_WARNING;
253         } else if ((mask & (D_CONSOLE | libcfs_printk)) != 0) {
254                 prefix = dbghdr_to_info_string(hdr);
255                 ptype = KERN_INFO;
256         }
257
258         if ((mask & D_CONSOLE) != 0) {
259                 printk("%s%s: %.*s", ptype, prefix, len, buf);
260         } else {
261                 printk("%s%s: %d:%d:(%s:%d:%s()) %.*s", ptype, prefix,
262                        hdr->ph_pid, hdr->ph_extern_pid, file, hdr->ph_line_num,
263                        fn, len, buf);
264         }
265         return;
266 }
267
268 int cfs_trace_max_debug_mb(void)
269 {
270         int  total_mb = (totalram_pages >> (20 - PAGE_SHIFT));
271
272         return MAX(512, (total_mb * 80)/100);
273 }