4 * It is designed to allow debugging traces of kvm to be generated
5 * on UP / SMP machines. Each trace entry can be timestamped so that
6 * it's possible to reconstruct a chronological record of trace events.
7 * The implementation refers to blktrace kernel support.
9 * Copyright (c) 2008 Intel Corporation
10 * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
12 * Authors: Feng(Eric) Liu, eric.e.liu@intel.com
17 #include <linux/module.h>
18 #include <linux/relay.h>
19 #include <linux/debugfs.h>
21 #include <linux/kvm_host.h>
23 #define KVM_TRACE_STATE_RUNNING (1 << 0)
24 #define KVM_TRACE_STATE_PAUSE (1 << 1)
25 #define KVM_TRACE_STATE_CLEARUP (1 << 2)
30 struct dentry *lost_file;
31 atomic_t lost_records;
33 static struct kvm_trace *kvm_trace;
35 struct kvm_trace_probe {
39 marker_probe_func *probe_func;
42 static inline int calc_rec_size(int cycle, int extra)
44 int rec_size = KVM_TRC_HEAD_SIZE;
47 return cycle ? rec_size += KVM_TRC_CYCLE_SIZE : rec_size;
50 static void kvm_add_trace(void *probe_private, void *call_data,
51 const char *format, va_list *args)
53 struct kvm_trace_probe *p = probe_private;
54 struct kvm_trace *kt = kvm_trace;
55 struct kvm_trace_rec rec;
56 struct kvm_vcpu *vcpu;
60 if (unlikely(kt->trace_state != KVM_TRACE_STATE_RUNNING))
63 rec.rec_val = TRACE_REC_EVENT_ID(va_arg(*args, u32));
64 vcpu = va_arg(*args, struct kvm_vcpu *);
65 rec.pid = current->tgid;
66 rec.vcpu_id = vcpu->vcpu_id;
68 extra = va_arg(*args, u32);
69 WARN_ON(!(extra <= KVM_TRC_EXTRA_MAX));
70 extra = min_t(u32, extra, KVM_TRC_EXTRA_MAX);
72 rec.rec_val |= TRACE_REC_TCS(p->cycle_in)
73 | TRACE_REC_NUM_DATA_ARGS(extra);
76 rec.u.cycle.cycle_u64 = get_cycles();
78 for (i = 0; i < extra; i++)
79 rec.u.cycle.extra_u32[i] = va_arg(*args, u32);
81 for (i = 0; i < extra; i++)
82 rec.u.nocycle.extra_u32[i] = va_arg(*args, u32);
85 size = calc_rec_size(p->cycle_in, extra * sizeof(u32));
86 relay_write(kt->rchan, &rec, size);
89 static struct kvm_trace_probe kvm_trace_probes[] = {
90 { "kvm_trace_entryexit", "%u %p %u %u %u %u %u %u", 1, kvm_add_trace },
91 { "kvm_trace_handler", "%u %p %u %u %u %u %u %u", 0, kvm_add_trace },
94 static int lost_records_get(void *data, u64 *val)
96 struct kvm_trace *kt = data;
98 *val = atomic_read(&kt->lost_records);
102 DEFINE_SIMPLE_ATTRIBUTE(kvm_trace_lost_ops, lost_records_get, NULL, "%llu\n");
105 * The relay channel is used in "no-overwrite" mode, it keeps trace of how
106 * many times we encountered a full subbuffer, to tell user space app the
107 * lost records there were.
109 static int kvm_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
110 void *prev_subbuf, size_t prev_padding)
112 struct kvm_trace *kt;
114 if (!relay_buf_full(buf)) {
117 * executed only once when the channel is opened
118 * save metadata as first record
120 subbuf_start_reserve(buf, sizeof(u32));
121 *(u32 *)subbuf = 0x12345678;
127 kt = buf->chan->private_data;
128 atomic_inc(&kt->lost_records);
133 static struct dentry *kvm_create_buf_file_callack(const char *filename,
134 struct dentry *parent,
136 struct rchan_buf *buf,
139 return debugfs_create_file(filename, mode, parent, buf,
140 &relay_file_operations);
143 static int kvm_remove_buf_file_callback(struct dentry *dentry)
145 debugfs_remove(dentry);
149 static struct rchan_callbacks kvm_relay_callbacks = {
150 .subbuf_start = kvm_subbuf_start_callback,
151 .create_buf_file = kvm_create_buf_file_callack,
152 .remove_buf_file = kvm_remove_buf_file_callback,
155 static int do_kvm_trace_enable(struct kvm_user_trace_setup *kuts)
157 struct kvm_trace *kt;
160 if (!kuts->buf_size || !kuts->buf_nr)
163 kt = kzalloc(sizeof(*kt), GFP_KERNEL);
168 atomic_set(&kt->lost_records, 0);
169 kt->lost_file = debugfs_create_file("lost_records", 0444, kvm_debugfs_dir,
170 kt, &kvm_trace_lost_ops);
174 kt->rchan = relay_open("trace", kvm_debugfs_dir, kuts->buf_size,
175 kuts->buf_nr, &kvm_relay_callbacks, kt);
181 for (i = 0; i < ARRAY_SIZE(kvm_trace_probes); i++) {
182 struct kvm_trace_probe *p = &kvm_trace_probes[i];
184 r = marker_probe_register(p->name, p->format, p->probe_func, p);
186 printk(KERN_INFO "Unable to register probe %s\n",
190 kvm_trace->trace_state = KVM_TRACE_STATE_RUNNING;
196 debugfs_remove(kt->lost_file);
198 relay_close(kt->rchan);
204 static int kvm_trace_enable(char __user *arg)
206 struct kvm_user_trace_setup kuts;
209 ret = copy_from_user(&kuts, arg, sizeof(kuts));
213 ret = do_kvm_trace_enable(&kuts);
220 static int kvm_trace_pause(void)
222 struct kvm_trace *kt = kvm_trace;
228 if (kt->trace_state == KVM_TRACE_STATE_RUNNING) {
229 kt->trace_state = KVM_TRACE_STATE_PAUSE;
230 relay_flush(kt->rchan);
237 void kvm_trace_cleanup(void)
239 struct kvm_trace *kt = kvm_trace;
245 if (kt->trace_state == KVM_TRACE_STATE_RUNNING ||
246 kt->trace_state == KVM_TRACE_STATE_PAUSE) {
248 kt->trace_state = KVM_TRACE_STATE_CLEARUP;
250 for (i = 0; i < ARRAY_SIZE(kvm_trace_probes); i++) {
251 struct kvm_trace_probe *p = &kvm_trace_probes[i];
252 marker_probe_unregister(p->name, p->probe_func, p);
255 relay_close(kt->rchan);
256 debugfs_remove(kt->lost_file);
261 int kvm_trace_ioctl(unsigned int ioctl, unsigned long arg)
263 void __user *argp = (void __user *)arg;
266 if (!capable(CAP_SYS_ADMIN))
270 case KVM_TRACE_ENABLE:
271 r = kvm_trace_enable(argp);
273 case KVM_TRACE_PAUSE:
274 r = kvm_trace_pause();
276 case KVM_TRACE_DISABLE: