Doug / smarc-fsl-linux-kernel

Download as
Browse Code ยป

Commit bcf6edcd6fdb8965290f0b635a530fa3c6c212e1

Authored by Xiao Guangrong
Committed by Arnaldo Carvalho de Melo
1 parent 26bf264e87
Exists in smarc-l5.0.0_1.0.0-ga and in 5 other branches imx_3.10.17_1.0.1_ga, imx_3.10.53_1.1.0_ga, imx_3.14.28_1.0.0_ga, smarc-imx_3.10.53_1.1.0_ga, smarc-imx_3.14.28_1.0.0_ga

perf kvm: Events analysis tool 

Add 'perf kvm stat' support to analyze kvm vmexit/mmio/ioport smartly

Usage:
- kvm stat
  run a command and gather performance counter statistics, it is the alias of
  perf stat

- trace kvm events:
  perf kvm stat record, or, if other tracepoints are interesting as well, we
  can append the events like this:
  perf kvm stat record -e timer:* -a

  If many guests are running, we can track the specified guest by using -p or
  --pid, -a is used to track events generated by all guests.

- show the result:
  perf kvm stat report

The output example is following:
13005
13059

total 2 guests are running on the host

Then, track the guest whose pid is 13059:
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.253 MB perf.data.guest (~11065 samples) ]

See the vmexit events:

Analyze events for all VCPUs:

             VM-EXIT    Samples  Samples%     Time%         Avg time

         APIC_ACCESS        460    70.55%     0.01%     22.44us ( +-   1.75% )
                 HLT         93    14.26%    99.98% 832077.26us ( +-  10.42% )
  EXTERNAL_INTERRUPT         64     9.82%     0.00%     35.35us ( +-  14.21% )
   PENDING_INTERRUPT         24     3.68%     0.00%      9.29us ( +-  31.39% )
           CR_ACCESS          7     1.07%     0.00%      8.12us ( +-   5.76% )
      IO_INSTRUCTION          3     0.46%     0.00%     18.00us ( +-  11.79% )
       EXCEPTION_NMI          1     0.15%     0.00%      5.83us ( +-   -nan% )

Total Samples:652, Total events handled time:77396109.80us.

See the mmio events:

Analyze events for all VCPUs:

         MMIO Access    Samples  Samples%     Time%         Avg time

        0xfee00380:W        387    84.31%    79.28%      8.29us ( +-   3.32% )
        0xfee00300:W         24     5.23%     9.96%     16.79us ( +-   1.97% )
        0xfee00300:R         24     5.23%     7.83%     13.20us ( +-   3.00% )
        0xfee00310:W         24     5.23%     2.93%      4.94us ( +-   3.84% )

Total Samples:459, Total events handled time:4044.59us.

See the ioport event:

Analyze events for all VCPUs:

      IO Port Access    Samples  Samples%     Time%         Avg time

         0xc050:POUT          3   100.00%   100.00%     13.75us ( +-  10.83% )

Total Samples:3, Total events handled time:41.26us.

And, --vcpu is used to track the specified vcpu and --key is used to sort the
result:

Analyze events for VCPU 0:

             VM-EXIT    Samples  Samples%     Time%         Avg time

                 HLT         27    13.85%    99.97% 405790.24us ( +-  12.70% )
  EXTERNAL_INTERRUPT         13     6.67%     0.00%     27.94us ( +-  22.26% )
         APIC_ACCESS        146    74.87%     0.03%     21.69us ( +-   2.91% )
      IO_INSTRUCTION          2     1.03%     0.00%     17.77us ( +-  20.56% )
           CR_ACCESS          2     1.03%     0.00%      8.55us ( +-   6.47% )
   PENDING_INTERRUPT          5     2.56%     0.00%      6.27us ( +-   3.94% )

Total Samples:195, Total events handled time:10959950.90us.

Signed-off-by: Dong Hao <haodong@linux.vnet.ibm.com>
Signed-off-by: Runzhen Wang <runzhen@linux.vnet.ibm.com>
[ Dong Hao <haodong@linux.vnet.ibm.com>
  Runzhen Wang <runzhen@linux.vnet.ibm.com>:
     - rebase it on current acme's tree
     - fix the compiling-error on i386 ]
Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Acked-by: David Ahern <dsahern@gmail.com>
Cc: Avi Kivity <avi@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: kvm@vger.kernel.org
Cc: Runzhen Wang <runzhen@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/1347870675-31495-4-git-send-email-haodong@linux.vnet.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

Showing 6 changed files with 929 additions and 6 deletions Side-by-side Diff

tools/perf/Documentation/perf-kvm.txt
Diff comments   View file @ bcf6edc
... ... @@ -12,7 +12,7 @@
12 12 [--guestkallsyms=<path> --guestmodules=<path> | --guestvmlinux=<path>]]
13 13 {top|record|report|diff|buildid-list}
14 14 'perf kvm' [--host] [--guest] [--guestkallsyms=<path> --guestmodules=<path>
15   - | --guestvmlinux=<path>] {top|record|report|diff|buildid-list}
  15 + | --guestvmlinux=<path>] {top|record|report|diff|buildid-list|stat}
16 16  
17 17 DESCRIPTION
18 18 -----------
... ... @@ -38,6 +38,18 @@
38 38 so that other tools can be used to fetch packages with matching symbol tables
39 39 for use by perf report.
40 40  
  41 + 'perf kvm stat <command>' to run a command and gather performance counter
  42 + statistics.
  43 + Especially, perf 'kvm stat record/report' generates a statistical analysis
  44 + of KVM events. Currently, vmexit, mmio and ioport events are supported.
  45 + 'perf kvm stat record <command>' records kvm events and the events between
  46 + start and end <command>.
  47 + And this command produces a file which contains tracing results of kvm
  48 + events.
  49 +
  50 + 'perf kvm stat report' reports statistical data which includes events
  51 + handled time, samples, and so on.
  52 +
41 53 OPTIONS
42 54 -------
43 55 -i::
44 56  
... ... @@ -68,8 +80,22 @@
68 80 --guestvmlinux=<path>::
69 81 Guest os kernel vmlinux.
70 82  
  83 +STAT REPORT OPTIONS
  84 +-------------------
  85 +--vcpu=<value>::
  86 + analyze events which occures on this vcpu. (default: all vcpus)
  87 +
  88 +--events=<value>::
  89 + events to be analyzed. Possible values: vmexit, mmio, ioport.
  90 + (default: vmexit)
  91 +-k::
  92 +--key=<value>::
  93 + Sorting key. Possible values: sample (default, sort by samples
  94 + number), time (sort by average time).
  95 +
71 96 SEE ALSO
72 97 --------
73 98 linkperf:perf-top[1], linkperf:perf-record[1], linkperf:perf-report[1],
74   -linkperf:perf-diff[1], linkperf:perf-buildid-list[1]
  99 +linkperf:perf-diff[1], linkperf:perf-buildid-list[1],
  100 +linkperf:perf-stat[1]
... ... @@ -16,4 +16,7 @@
16 16 include/linux/poison.h
17 17 include/linux/magic.h
18 18 include/linux/hw_breakpoint.h
  19 +arch/x86/include/asm/svm.h
  20 +arch/x86/include/asm/vmx.h
  21 +arch/x86/include/asm/kvm_host.h
tools/perf/builtin-kvm.c
Diff comments   View file @ bcf6edc
1 1 #include "builtin.h"
2 2 #include "perf.h"
3 3  
  4 +#include "util/evsel.h"
4 5 #include "util/util.h"
5 6 #include "util/cache.h"
6 7 #include "util/symbol.h"
7 8  
... ... @@ -10,8 +11,10 @@
10 11  
11 12 #include "util/parse-options.h"
12 13 #include "util/trace-event.h"
13   -
14 14 #include "util/debug.h"
  15 +#include "util/debugfs.h"
  16 +#include "util/tool.h"
  17 +#include "util/stat.h"
15 18  
16 19 #include <sys/prctl.h>
17 20  
18 21  
... ... @@ -19,11 +22,840 @@
19 22 #include <pthread.h>
20 23 #include <math.h>
21 24  
22   -static const char *file_name;
  25 +#include "../../arch/x86/include/asm/svm.h"
  26 +#include "../../arch/x86/include/asm/vmx.h"
  27 +#include "../../arch/x86/include/asm/kvm.h"
  28 +
  29 +struct event_key {
  30 + #define INVALID_KEY (~0ULL)
  31 + u64 key;
  32 + int info;
  33 +};
  34 +
  35 +struct kvm_events_ops {
  36 + bool (*is_begin_event)(struct event_format *event, void *data,
  37 + struct event_key *key);
  38 + bool (*is_end_event)(struct event_format *event, void *data,
  39 + struct event_key *key);
  40 + void (*decode_key)(struct event_key *key, char decode[20]);
  41 + const char *name;
  42 +};
  43 +
  44 +static void exit_event_get_key(struct event_format *event, void *data,
  45 + struct event_key *key)
  46 +{
  47 + key->info = 0;
  48 + key->key = raw_field_value(event, "exit_reason", data);
  49 +}
  50 +
  51 +static bool kvm_exit_event(struct event_format *event)
  52 +{
  53 + return !strcmp(event->name, "kvm_exit");
  54 +}
  55 +
  56 +static bool exit_event_begin(struct event_format *event, void *data,
  57 + struct event_key *key)
  58 +{
  59 + if (kvm_exit_event(event)) {
  60 + exit_event_get_key(event, data, key);
  61 + return true;
  62 + }
  63 +
  64 + return false;
  65 +}
  66 +
  67 +static bool kvm_entry_event(struct event_format *event)
  68 +{
  69 + return !strcmp(event->name, "kvm_entry");
  70 +}
  71 +
  72 +static bool exit_event_end(struct event_format *event, void *data __maybe_unused,
  73 + struct event_key *key __maybe_unused)
  74 +{
  75 + return kvm_entry_event(event);
  76 +}
  77 +
  78 +struct exit_reasons_table {
  79 + unsigned long exit_code;
  80 + const char *reason;
  81 +};
  82 +
  83 +struct exit_reasons_table vmx_exit_reasons[] = {
  84 + VMX_EXIT_REASONS
  85 +};
  86 +
  87 +struct exit_reasons_table svm_exit_reasons[] = {
  88 + SVM_EXIT_REASONS
  89 +};
  90 +
  91 +static int cpu_isa;
  92 +
  93 +static const char *get_exit_reason(u64 exit_code)
  94 +{
  95 + int table_size = ARRAY_SIZE(svm_exit_reasons);
  96 + struct exit_reasons_table *table = svm_exit_reasons;
  97 +
  98 + if (cpu_isa == 1) {
  99 + table = vmx_exit_reasons;
  100 + table_size = ARRAY_SIZE(vmx_exit_reasons);
  101 + }
  102 +
  103 + while (table_size--) {
  104 + if (table->exit_code == exit_code)
  105 + return table->reason;
  106 + table++;
  107 + }
  108 +
  109 + pr_err("unknown kvm exit code:%lld on %s\n",
  110 + (unsigned long long)exit_code, cpu_isa ? "VMX" : "SVM");
  111 + return "UNKNOWN";
  112 +}
  113 +
  114 +static void exit_event_decode_key(struct event_key *key, char decode[20])
  115 +{
  116 + const char *exit_reason = get_exit_reason(key->key);
  117 +
  118 + scnprintf(decode, 20, "%s", exit_reason);
  119 +}
  120 +
  121 +static struct kvm_events_ops exit_events = {
  122 + .is_begin_event = exit_event_begin,
  123 + .is_end_event = exit_event_end,
  124 + .decode_key = exit_event_decode_key,
  125 + .name = "VM-EXIT"
  126 +};
  127 +
  128 + /*
  129 + * For the mmio events, we treat:
  130 + * the time of MMIO write: kvm_mmio(KVM_TRACE_MMIO_WRITE...) -> kvm_entry
  131 + * the time of MMIO read: kvm_exit -> kvm_mmio(KVM_TRACE_MMIO_READ...).
  132 + */
  133 +static void mmio_event_get_key(struct event_format *event, void *data,
  134 + struct event_key *key)
  135 +{
  136 + key->key = raw_field_value(event, "gpa", data);
  137 + key->info = raw_field_value(event, "type", data);
  138 +}
  139 +
  140 +#define KVM_TRACE_MMIO_READ_UNSATISFIED 0
  141 +#define KVM_TRACE_MMIO_READ 1
  142 +#define KVM_TRACE_MMIO_WRITE 2
  143 +
  144 +static bool mmio_event_begin(struct event_format *event, void *data,
  145 + struct event_key *key)
  146 +{
  147 + /* MMIO read begin event in kernel. */
  148 + if (kvm_exit_event(event))
  149 + return true;
  150 +
  151 + /* MMIO write begin event in kernel. */
  152 + if (!strcmp(event->name, "kvm_mmio") &&
  153 + raw_field_value(event, "type", data) == KVM_TRACE_MMIO_WRITE) {
  154 + mmio_event_get_key(event, data, key);
  155 + return true;
  156 + }
  157 +
  158 + return false;
  159 +}
  160 +
  161 +static bool mmio_event_end(struct event_format *event, void *data,
  162 + struct event_key *key)
  163 +{
  164 + /* MMIO write end event in kernel. */
  165 + if (kvm_entry_event(event))
  166 + return true;
  167 +
  168 + /* MMIO read end event in kernel.*/
  169 + if (!strcmp(event->name, "kvm_mmio") &&
  170 + raw_field_value(event, "type", data) == KVM_TRACE_MMIO_READ) {
  171 + mmio_event_get_key(event, data, key);
  172 + return true;
  173 + }
  174 +
  175 + return false;
  176 +}
  177 +
  178 +static void mmio_event_decode_key(struct event_key *key, char decode[20])
  179 +{
  180 + scnprintf(decode, 20, "%#lx:%s", (unsigned long)key->key,
  181 + key->info == KVM_TRACE_MMIO_WRITE ? "W" : "R");
  182 +}
  183 +
  184 +static struct kvm_events_ops mmio_events = {
  185 + .is_begin_event = mmio_event_begin,
  186 + .is_end_event = mmio_event_end,
  187 + .decode_key = mmio_event_decode_key,
  188 + .name = "MMIO Access"
  189 +};
  190 +
  191 + /* The time of emulation pio access is from kvm_pio to kvm_entry. */
  192 +static void ioport_event_get_key(struct event_format *event, void *data,
  193 + struct event_key *key)
  194 +{
  195 + key->key = raw_field_value(event, "port", data);
  196 + key->info = raw_field_value(event, "rw", data);
  197 +}
  198 +
  199 +static bool ioport_event_begin(struct event_format *event, void *data,
  200 + struct event_key *key)
  201 +{
  202 + if (!strcmp(event->name, "kvm_pio")) {
  203 + ioport_event_get_key(event, data, key);
  204 + return true;
  205 + }
  206 +
  207 + return false;
  208 +}
  209 +
  210 +static bool ioport_event_end(struct event_format *event, void *data __maybe_unused,
  211 + struct event_key *key __maybe_unused)
  212 +{
  213 + if (kvm_entry_event(event))
  214 + return true;
  215 +
  216 + return false;
  217 +}
  218 +
  219 +static void ioport_event_decode_key(struct event_key *key, char decode[20])
  220 +{
  221 + scnprintf(decode, 20, "%#llx:%s", (unsigned long long)key->key,
  222 + key->info ? "POUT" : "PIN");
  223 +}
  224 +
  225 +static struct kvm_events_ops ioport_events = {
  226 + .is_begin_event = ioport_event_begin,
  227 + .is_end_event = ioport_event_end,
  228 + .decode_key = ioport_event_decode_key,
  229 + .name = "IO Port Access"
  230 +};
  231 +
  232 +static const char *report_event = "vmexit";
  233 +struct kvm_events_ops *events_ops;
  234 +
  235 +static bool register_kvm_events_ops(void)
  236 +{
  237 + bool ret = true;
  238 +
  239 + if (!strcmp(report_event, "vmexit"))
  240 + events_ops = &exit_events;
  241 + else if (!strcmp(report_event, "mmio"))
  242 + events_ops = &mmio_events;
  243 + else if (!strcmp(report_event, "ioport"))
  244 + events_ops = &ioport_events;
  245 + else {
  246 + pr_err("Unknown report event:%s\n", report_event);
  247 + ret = false;
  248 + }
  249 +
  250 + return ret;
  251 +}
  252 +
  253 +struct kvm_event_stats {
  254 + u64 time;
  255 + struct stats stats;
  256 +};
  257 +
  258 +struct kvm_event {
  259 + struct list_head hash_entry;
  260 + struct rb_node rb;
  261 +
  262 + struct event_key key;
  263 +
  264 + struct kvm_event_stats total;
  265 +
  266 + #define DEFAULT_VCPU_NUM 8
  267 + int max_vcpu;
  268 + struct kvm_event_stats *vcpu;
  269 +};
  270 +
  271 +struct vcpu_event_record {
  272 + int vcpu_id;
  273 + u64 start_time;
  274 + struct kvm_event *last_event;
  275 +};
  276 +
  277 +#define EVENTS_BITS 12
  278 +#define EVENTS_CACHE_SIZE (1UL << EVENTS_BITS)
  279 +
  280 +static u64 total_time;
  281 +static u64 total_count;
  282 +static struct list_head kvm_events_cache[EVENTS_CACHE_SIZE];
  283 +
  284 +static void init_kvm_event_record(void)
  285 +{
  286 + int i;
  287 +
  288 + for (i = 0; i < (int)EVENTS_CACHE_SIZE; i++)
  289 + INIT_LIST_HEAD(&kvm_events_cache[i]);
  290 +}
  291 +
  292 +static int kvm_events_hash_fn(u64 key)
  293 +{
  294 + return key & (EVENTS_CACHE_SIZE - 1);
  295 +}
  296 +
  297 +static bool kvm_event_expand(struct kvm_event *event, int vcpu_id)
  298 +{
  299 + int old_max_vcpu = event->max_vcpu;
  300 +
  301 + if (vcpu_id < event->max_vcpu)
  302 + return true;
  303 +
  304 + while (event->max_vcpu <= vcpu_id)
  305 + event->max_vcpu += DEFAULT_VCPU_NUM;
  306 +
  307 + event->vcpu = realloc(event->vcpu,
  308 + event->max_vcpu * sizeof(*event->vcpu));
  309 + if (!event->vcpu) {
  310 + pr_err("Not enough memory\n");
  311 + return false;
  312 + }
  313 +
  314 + memset(event->vcpu + old_max_vcpu, 0,
  315 + (event->max_vcpu - old_max_vcpu) * sizeof(*event->vcpu));
  316 + return true;
  317 +}
  318 +
  319 +static struct kvm_event *kvm_alloc_init_event(struct event_key *key)
  320 +{
  321 + struct kvm_event *event;
  322 +
  323 + event = zalloc(sizeof(*event));
  324 + if (!event) {
  325 + pr_err("Not enough memory\n");
  326 + return NULL;
  327 + }
  328 +
  329 + event->key = *key;
  330 + return event;
  331 +}
  332 +
  333 +static struct kvm_event *find_create_kvm_event(struct event_key *key)
  334 +{
  335 + struct kvm_event *event;
  336 + struct list_head *head;
  337 +
  338 + BUG_ON(key->key == INVALID_KEY);
  339 +
  340 + head = &kvm_events_cache[kvm_events_hash_fn(key->key)];
  341 + list_for_each_entry(event, head, hash_entry)
  342 + if (event->key.key == key->key && event->key.info == key->info)
  343 + return event;
  344 +
  345 + event = kvm_alloc_init_event(key);
  346 + if (!event)
  347 + return NULL;
  348 +
  349 + list_add(&event->hash_entry, head);
  350 + return event;
  351 +}
  352 +
  353 +static bool handle_begin_event(struct vcpu_event_record *vcpu_record,
  354 + struct event_key *key, u64 timestamp)
  355 +{
  356 + struct kvm_event *event = NULL;
  357 +
  358 + if (key->key != INVALID_KEY)
  359 + event = find_create_kvm_event(key);
  360 +
  361 + vcpu_record->last_event = event;
  362 + vcpu_record->start_time = timestamp;
  363 + return true;
  364 +}
  365 +
  366 +static void
  367 +kvm_update_event_stats(struct kvm_event_stats *kvm_stats, u64 time_diff)
  368 +{
  369 + kvm_stats->time += time_diff;
  370 + update_stats(&kvm_stats->stats, time_diff);
  371 +}
  372 +
  373 +static double kvm_event_rel_stddev(int vcpu_id, struct kvm_event *event)
  374 +{
  375 + struct kvm_event_stats *kvm_stats = &event->total;
  376 +
  377 + if (vcpu_id != -1)
  378 + kvm_stats = &event->vcpu[vcpu_id];
  379 +
  380 + return rel_stddev_stats(stddev_stats(&kvm_stats->stats),
  381 + avg_stats(&kvm_stats->stats));
  382 +}
  383 +
  384 +static bool update_kvm_event(struct kvm_event *event, int vcpu_id,
  385 + u64 time_diff)
  386 +{
  387 + kvm_update_event_stats(&event->total, time_diff);
  388 +
  389 + if (!kvm_event_expand(event, vcpu_id))
  390 + return false;
  391 +
  392 + kvm_update_event_stats(&event->vcpu[vcpu_id], time_diff);
  393 + return true;
  394 +}
  395 +
  396 +static bool handle_end_event(struct vcpu_event_record *vcpu_record,
  397 + struct event_key *key, u64 timestamp)
  398 +{
  399 + struct kvm_event *event;
  400 + u64 time_begin, time_diff;
  401 +
  402 + event = vcpu_record->last_event;
  403 + time_begin = vcpu_record->start_time;
  404 +
  405 + /* The begin event is not caught. */
  406 + if (!time_begin)
  407 + return true;
  408 +
  409 + /*
  410 + * In some case, the 'begin event' only records the start timestamp,
  411 + * the actual event is recognized in the 'end event' (e.g. mmio-event).
  412 + */
  413 +
  414 + /* Both begin and end events did not get the key. */
  415 + if (!event && key->key == INVALID_KEY)
  416 + return true;
  417 +
  418 + if (!event)
  419 + event = find_create_kvm_event(key);
  420 +
  421 + if (!event)
  422 + return false;
  423 +
  424 + vcpu_record->last_event = NULL;
  425 + vcpu_record->start_time = 0;
  426 +
  427 + BUG_ON(timestamp < time_begin);
  428 +
  429 + time_diff = timestamp - time_begin;
  430 + return update_kvm_event(event, vcpu_record->vcpu_id, time_diff);
  431 +}
  432 +
  433 +static struct vcpu_event_record
  434 +*per_vcpu_record(struct thread *thread, struct event_format *event, void *data)
  435 +{
  436 + /* Only kvm_entry records vcpu id. */
  437 + if (!thread->priv && kvm_entry_event(event)) {
  438 + struct vcpu_event_record *vcpu_record;
  439 +
  440 + vcpu_record = zalloc(sizeof(struct vcpu_event_record));
  441 + if (!vcpu_record) {
  442 + pr_err("Not enough memory\n");
  443 + return NULL;
  444 + }
  445 +
  446 + vcpu_record->vcpu_id = raw_field_value(event, "vcpu_id", data);
  447 + thread->priv = vcpu_record;
  448 + }
  449 +
  450 + return (struct vcpu_event_record *)thread->priv;
  451 +}
  452 +
  453 +static bool handle_kvm_event(struct thread *thread, struct event_format *event,
  454 + void *data, u64 timestamp)
  455 +{
  456 + struct vcpu_event_record *vcpu_record;
  457 + struct event_key key = {.key = INVALID_KEY};
  458 +
  459 + vcpu_record = per_vcpu_record(thread, event, data);
  460 + if (!vcpu_record)
  461 + return true;
  462 +
  463 + if (events_ops->is_begin_event(event, data, &key))
  464 + return handle_begin_event(vcpu_record, &key, timestamp);
  465 +
  466 + if (events_ops->is_end_event(event, data, &key))
  467 + return handle_end_event(vcpu_record, &key, timestamp);
  468 +
  469 + return true;
  470 +}
  471 +
  472 +typedef int (*key_cmp_fun)(struct kvm_event*, struct kvm_event*, int);
  473 +struct kvm_event_key {
  474 + const char *name;
  475 + key_cmp_fun key;
  476 +};
  477 +
  478 +static int trace_vcpu = -1;
  479 +#define GET_EVENT_KEY(func, field) \
  480 +static u64 get_event_ ##func(struct kvm_event *event, int vcpu) \
  481 +{ \
  482 + if (vcpu == -1) \
  483 + return event->total.field; \
  484 + \
  485 + if (vcpu >= event->max_vcpu) \
  486 + return 0; \
  487 + \
  488 + return event->vcpu[vcpu].field; \
  489 +}
  490 +
  491 +#define COMPARE_EVENT_KEY(func, field) \
  492 +GET_EVENT_KEY(func, field) \
  493 +static int compare_kvm_event_ ## func(struct kvm_event *one, \
  494 + struct kvm_event *two, int vcpu)\
  495 +{ \
  496 + return get_event_ ##func(one, vcpu) > \
  497 + get_event_ ##func(two, vcpu); \
  498 +}
  499 +
  500 +GET_EVENT_KEY(time, time);
  501 +COMPARE_EVENT_KEY(count, stats.n);
  502 +COMPARE_EVENT_KEY(mean, stats.mean);
  503 +
  504 +#define DEF_SORT_NAME_KEY(name, compare_key) \
  505 + { #name, compare_kvm_event_ ## compare_key }
  506 +
  507 +static struct kvm_event_key keys[] = {
  508 + DEF_SORT_NAME_KEY(sample, count),
  509 + DEF_SORT_NAME_KEY(time, mean),
  510 + { NULL, NULL }
  511 +};
  512 +
  513 +static const char *sort_key = "sample";
  514 +static key_cmp_fun compare;
  515 +
  516 +static bool select_key(void)
  517 +{
  518 + int i;
  519 +
  520 + for (i = 0; keys[i].name; i++) {
  521 + if (!strcmp(keys[i].name, sort_key)) {
  522 + compare = keys[i].key;
  523 + return true;
  524 + }
  525 + }
  526 +
  527 + pr_err("Unknown compare key:%s\n", sort_key);
  528 + return false;
  529 +}
  530 +
  531 +static struct rb_root result;
  532 +static void insert_to_result(struct kvm_event *event, key_cmp_fun bigger,
  533 + int vcpu)
  534 +{
  535 + struct rb_node **rb = &result.rb_node;
  536 + struct rb_node *parent = NULL;
  537 + struct kvm_event *p;
  538 +
  539 + while (*rb) {
  540 + p = container_of(*rb, struct kvm_event, rb);
  541 + parent = *rb;
  542 +
  543 + if (bigger(event, p, vcpu))
  544 + rb = &(*rb)->rb_left;
  545 + else
  546 + rb = &(*rb)->rb_right;
  547 + }
  548 +
  549 + rb_link_node(&event->rb, parent, rb);
  550 + rb_insert_color(&event->rb, &result);
  551 +}
  552 +
  553 +static void update_total_count(struct kvm_event *event, int vcpu)
  554 +{
  555 + total_count += get_event_count(event, vcpu);
  556 + total_time += get_event_time(event, vcpu);
  557 +}
  558 +
  559 +static bool event_is_valid(struct kvm_event *event, int vcpu)
  560 +{
  561 + return !!get_event_count(event, vcpu);
  562 +}
  563 +
  564 +static void sort_result(int vcpu)
  565 +{
  566 + unsigned int i;
  567 + struct kvm_event *event;
  568 +
  569 + for (i = 0; i < EVENTS_CACHE_SIZE; i++)
  570 + list_for_each_entry(event, &kvm_events_cache[i], hash_entry)
  571 + if (event_is_valid(event, vcpu)) {
  572 + update_total_count(event, vcpu);
  573 + insert_to_result(event, compare, vcpu);
  574 + }
  575 +}
  576 +
  577 +/* returns left most element of result, and erase it */
  578 +static struct kvm_event *pop_from_result(void)
  579 +{
  580 + struct rb_node *node = rb_first(&result);
  581 +
  582 + if (!node)
  583 + return NULL;
  584 +
  585 + rb_erase(node, &result);
  586 + return container_of(node, struct kvm_event, rb);
  587 +}
  588 +
  589 +static void print_vcpu_info(int vcpu)
  590 +{
  591 + pr_info("Analyze events for ");
  592 +
  593 + if (vcpu == -1)
  594 + pr_info("all VCPUs:\n\n");
  595 + else
  596 + pr_info("VCPU %d:\n\n", vcpu);
  597 +}
  598 +
  599 +static void print_result(int vcpu)
  600 +{
  601 + char decode[20];
  602 + struct kvm_event *event;
  603 +
  604 + pr_info("\n\n");
  605 + print_vcpu_info(vcpu);
  606 + pr_info("%20s ", events_ops->name);
  607 + pr_info("%10s ", "Samples");
  608 + pr_info("%9s ", "Samples%");
  609 +
  610 + pr_info("%9s ", "Time%");
  611 + pr_info("%16s ", "Avg time");
  612 + pr_info("\n\n");
  613 +
  614 + while ((event = pop_from_result())) {
  615 + u64 ecount, etime;
  616 +
  617 + ecount = get_event_count(event, vcpu);
  618 + etime = get_event_time(event, vcpu);
  619 +
  620 + events_ops->decode_key(&event->key, decode);
  621 + pr_info("%20s ", decode);
  622 + pr_info("%10llu ", (unsigned long long)ecount);
  623 + pr_info("%8.2f%% ", (double)ecount / total_count * 100);
  624 + pr_info("%8.2f%% ", (double)etime / total_time * 100);
  625 + pr_info("%9.2fus ( +-%7.2f%% )", (double)etime / ecount/1e3,
  626 + kvm_event_rel_stddev(vcpu, event));
  627 + pr_info("\n");
  628 + }
  629 +
  630 + pr_info("\nTotal Samples:%lld, Total events handled time:%.2fus.\n\n",
  631 + (unsigned long long)total_count, total_time / 1e3);
  632 +}
  633 +
  634 +static int process_sample_event(struct perf_tool *tool __maybe_unused,
  635 + union perf_event *event,
  636 + struct perf_sample *sample,
  637 + struct perf_evsel *evsel,
  638 + struct machine *machine)
  639 +{
  640 + struct thread *thread = machine__findnew_thread(machine, sample->tid);
  641 +
  642 + if (thread == NULL) {
  643 + pr_debug("problem processing %d event, skipping it.\n",
  644 + event->header.type);
  645 + return -1;
  646 + }
  647 +
  648 + if (!handle_kvm_event(thread, evsel->tp_format, sample->raw_data,
  649 + sample->time))
  650 + return -1;
  651 +
  652 + return 0;
  653 +}
  654 +
  655 +static struct perf_tool eops = {
  656 + .sample = process_sample_event,
  657 + .comm = perf_event__process_comm,
  658 + .ordered_samples = true,
  659 +};
  660 +
  661 +static int get_cpu_isa(struct perf_session *session)
  662 +{
  663 + char *cpuid;
  664 + int isa;
  665 +
  666 + cpuid = perf_header__read_feature(session, HEADER_CPUID);
  667 +
  668 + if (!cpuid) {
  669 + pr_err("read HEADER_CPUID failed.\n");
  670 + return -ENOTSUP;
  671 + }
  672 +
  673 + if (strstr(cpuid, "Intel"))
  674 + isa = 1;
  675 + else if (strstr(cpuid, "AMD"))
  676 + isa = 0;
  677 + else {
  678 + pr_err("CPU %s is not supported.\n", cpuid);
  679 + isa = -ENOTSUP;
  680 + }
  681 +
  682 + free(cpuid);
  683 + return isa;
  684 +}
  685 +
  686 +static const char *file_name;
  687 +
  688 +static int read_events(void)
  689 +{
  690 + struct perf_session *kvm_session;
  691 + int ret;
  692 +
  693 + kvm_session = perf_session__new(file_name, O_RDONLY, 0, false, &eops);
  694 + if (!kvm_session) {
  695 + pr_err("Initializing perf session failed\n");
  696 + return -EINVAL;
  697 + }
  698 +
  699 + if (!perf_session__has_traces(kvm_session, "kvm record"))
  700 + return -EINVAL;
  701 +
  702 + /*
  703 + * Do not use 'isa' recorded in kvm_exit tracepoint since it is not
  704 + * traced in the old kernel.
  705 + */
  706 + ret = get_cpu_isa(kvm_session);
  707 +
  708 + if (ret < 0)
  709 + return ret;
  710 +
  711 + cpu_isa = ret;
  712 +
  713 + return perf_session__process_events(kvm_session, &eops);
  714 +}
  715 +
  716 +static bool verify_vcpu(int vcpu)
  717 +{
  718 + if (vcpu != -1 && vcpu < 0) {
  719 + pr_err("Invalid vcpu:%d.\n", vcpu);
  720 + return false;
  721 + }
  722 +
  723 + return true;
  724 +}
  725 +
  726 +static int kvm_events_report_vcpu(int vcpu)
  727 +{
  728 + int ret = -EINVAL;
  729 +
  730 + if (!verify_vcpu(vcpu))
  731 + goto exit;
  732 +
  733 + if (!select_key())
  734 + goto exit;
  735 +
  736 + if (!register_kvm_events_ops())
  737 + goto exit;
  738 +
  739 + init_kvm_event_record();
  740 + setup_pager();
  741 +
  742 + ret = read_events();
  743 + if (ret)
  744 + goto exit;
  745 +
  746 + sort_result(vcpu);
  747 + print_result(vcpu);
  748 +exit:
  749 + return ret;
  750 +}
  751 +
  752 +static const char * const record_args[] = {
  753 + "record",
  754 + "-R",
  755 + "-f",
  756 + "-m", "1024",
  757 + "-c", "1",
  758 + "-e", "kvm:kvm_entry",
  759 + "-e", "kvm:kvm_exit",
  760 + "-e", "kvm:kvm_mmio",
  761 + "-e", "kvm:kvm_pio",
  762 +};
  763 +
  764 +#define STRDUP_FAIL_EXIT(s) \
  765 + ({ char *_p; \
  766 + _p = strdup(s); \
  767 + if (!_p) \
  768 + return -ENOMEM; \
  769 + _p; \
  770 + })
  771 +
  772 +static int kvm_events_record(int argc, const char **argv)
  773 +{
  774 + unsigned int rec_argc, i, j;
  775 + const char **rec_argv;
  776 +
  777 + rec_argc = ARRAY_SIZE(record_args) + argc + 2;
  778 + rec_argv = calloc(rec_argc + 1, sizeof(char *));
  779 +
  780 + if (rec_argv == NULL)
  781 + return -ENOMEM;
  782 +
  783 + for (i = 0; i < ARRAY_SIZE(record_args); i++)
  784 + rec_argv[i] = STRDUP_FAIL_EXIT(record_args[i]);
  785 +
  786 + rec_argv[i++] = STRDUP_FAIL_EXIT("-o");
  787 + rec_argv[i++] = STRDUP_FAIL_EXIT(file_name);
  788 +
  789 + for (j = 1; j < (unsigned int)argc; j++, i++)
  790 + rec_argv[i] = argv[j];
  791 +
  792 + return cmd_record(i, rec_argv, NULL);
  793 +}
  794 +
  795 +static const char * const kvm_events_report_usage[] = {
  796 + "perf kvm stat report [<options>]",
  797 + NULL
  798 +};
  799 +
  800 +static const struct option kvm_events_report_options[] = {
  801 + OPT_STRING(0, "event", &report_event, "report event",
  802 + "event for reporting: vmexit, mmio, ioport"),
  803 + OPT_INTEGER(0, "vcpu", &trace_vcpu,
  804 + "vcpu id to report"),
  805 + OPT_STRING('k', "key", &sort_key, "sort-key",
  806 + "key for sorting: sample(sort by samples number)"
  807 + " time (sort by avg time)"),
  808 + OPT_END()
  809 +};
  810 +
  811 +static int kvm_events_report(int argc, const char **argv)
  812 +{
  813 + symbol__init();
  814 +
  815 + if (argc) {
  816 + argc = parse_options(argc, argv,
  817 + kvm_events_report_options,
  818 + kvm_events_report_usage, 0);
  819 + if (argc)
  820 + usage_with_options(kvm_events_report_usage,
  821 + kvm_events_report_options);
  822 + }
  823 +
  824 + return kvm_events_report_vcpu(trace_vcpu);
  825 +}
  826 +
  827 +static void print_kvm_stat_usage(void)
  828 +{
  829 + printf("Usage: perf kvm stat <command>\n\n");
  830 +
  831 + printf("# Available commands:\n");
  832 + printf("\trecord: record kvm events\n");
  833 + printf("\treport: report statistical data of kvm events\n");
  834 +
  835 + printf("\nOtherwise, it is the alias of 'perf stat':\n");
  836 +}
  837 +
  838 +static int kvm_cmd_stat(int argc, const char **argv)
  839 +{
  840 + if (argc == 1) {
  841 + print_kvm_stat_usage();
  842 + goto perf_stat;
  843 + }
  844 +
  845 + if (!strncmp(argv[1], "rec", 3))
  846 + return kvm_events_record(argc - 1, argv + 1);
  847 +
  848 + if (!strncmp(argv[1], "rep", 3))
  849 + return kvm_events_report(argc - 1 , argv + 1);
  850 +
  851 +perf_stat:
  852 + return cmd_stat(argc, argv, NULL);
  853 +}
  854 +
23 855 static char name_buffer[256];
24 856  
25 857 static const char * const kvm_usage[] = {
26   - "perf kvm [<options>] {top|record|report|diff|buildid-list}",
  858 + "perf kvm [<options>] {top|record|report|diff|buildid-list|stat}",
27 859 NULL
28 860 };
29 861  
... ... @@ -135,6 +967,8 @@
135 967 return cmd_top(argc, argv, NULL);
136 968 else if (!strncmp(argv[0], "buildid-list", 12))
137 969 return __cmd_buildid_list(argc, argv);
  970 + else if (!strncmp(argv[0], "stat", 4))
  971 + return kvm_cmd_stat(argc, argv);
138 972 else
139 973 usage_with_options(kvm_usage, kvm_options);
140 974  
tools/perf/util/header.c
Diff comments   View file @ bcf6edc
... ... @@ -1673,6 +1673,11 @@
1673 1673 return 0;
1674 1674 }
1675 1675  
  1676 +static char *read_cpuid(struct perf_header *ph, int fd)
  1677 +{
  1678 + return do_read_string(fd, ph);
  1679 +}
  1680 +
1676 1681 static struct perf_evsel *
1677 1682 perf_evlist__find_by_index(struct perf_evlist *evlist, int idx)
1678 1683 {
... ... @@ -1726,6 +1731,7 @@
1726 1731 struct feature_ops {
1727 1732 int (*write)(int fd, struct perf_header *h, struct perf_evlist *evlist);
1728 1733 void (*print)(struct perf_header *h, int fd, FILE *fp);
  1734 + char *(*read)(struct perf_header *h, int fd);
1729 1735 int (*process)(struct perf_file_section *section,
1730 1736 struct perf_header *h, int feat, int fd, void *data);
1731 1737 const char *name;
... ... @@ -1740,6 +1746,9 @@
1740 1746 #define FEAT_OPF(n, func) \
1741 1747 [n] = { .name = #n, .write = write_##func, .print = print_##func, \
1742 1748 .full_only = true }
  1749 +#define FEAT_OPA_R(n, func) \
  1750 + [n] = { .name = #n, .write = write_##func, .print = print_##func, \
  1751 + .read = read_##func }
1743 1752  
1744 1753 /* feature_ops not implemented: */
1745 1754 #define print_tracing_data NULL
... ... @@ -1754,7 +1763,7 @@
1754 1763 FEAT_OPA(HEADER_ARCH, arch),
1755 1764 FEAT_OPA(HEADER_NRCPUS, nrcpus),
1756 1765 FEAT_OPA(HEADER_CPUDESC, cpudesc),
1757   - FEAT_OPA(HEADER_CPUID, cpuid),
  1766 + FEAT_OPA_R(HEADER_CPUID, cpuid),
1758 1767 FEAT_OPA(HEADER_TOTAL_MEM, total_mem),
1759 1768 FEAT_OPP(HEADER_EVENT_DESC, event_desc),
1760 1769 FEAT_OPA(HEADER_CMDLINE, cmdline),
... ... @@ -1807,6 +1816,54 @@
1807 1816 perf_header__process_sections(header, fd, &hd,
1808 1817 perf_file_section__fprintf_info);
1809 1818 return 0;
  1819 +}
  1820 +
  1821 +struct header_read_data {
  1822 + int feat;
  1823 + char *result;
  1824 +};
  1825 +
  1826 +static int perf_file_section__read_feature(struct perf_file_section *section,
  1827 + struct perf_header *ph,
  1828 + int feat, int fd, void *data)
  1829 +{
  1830 + struct header_read_data *hd = data;
  1831 +
  1832 + if (feat != hd->feat)
  1833 + return 0;
  1834 +
  1835 + if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) {
  1836 + pr_debug("Failed to lseek to %" PRIu64 " offset for feature "
  1837 + "%d, continuing...\n", section->offset, feat);
  1838 + return 0;
  1839 + }
  1840 +
  1841 + if (feat >= HEADER_LAST_FEATURE) {
  1842 + pr_warning("unknown feature %d\n", feat);
  1843 + return 0;
  1844 + }
  1845 +
  1846 + if (!feat_ops[feat].read) {
  1847 + pr_warning("read is not supported for feature %d\n", feat);
  1848 + return 0;
  1849 + }
  1850 +
  1851 + hd->result = feat_ops[feat].read(ph, fd);
  1852 + return 0;
  1853 +}
  1854 +
  1855 +char *perf_header__read_feature(struct perf_session *session, int feat)
  1856 +{
  1857 + struct perf_header *header = &session->header;
  1858 + struct header_read_data hd;
  1859 + int fd = session->fd;
  1860 +
  1861 + hd.feat = feat;
  1862 + hd.result = NULL;
  1863 +
  1864 + perf_header__process_sections(header, fd, &hd,
  1865 + perf_file_section__read_feature);
  1866 + return hd.result;
1810 1867 }
1811 1868  
1812 1869 static int do_write_feat(int fd, struct perf_header *h, int type,
tools/perf/util/header.h
Diff comments   View file @ bcf6edc
... ... @@ -94,6 +94,7 @@
94 94 int feat, int fd, void *data));
95 95  
96 96 int perf_header__fprintf_info(struct perf_session *s, FILE *fp, bool full);
  97 +char *perf_header__read_feature(struct perf_session *session, int feat);
97 98  
98 99 int build_id_cache__add_s(const char *sbuild_id, const char *debugdir,
99 100 const char *name, bool is_kallsyms, bool is_vdso);
tools/perf/util/thread.h
Diff comments   View file @ bcf6edc
... ... @@ -16,6 +16,8 @@
16 16 bool comm_set;
17 17 char *comm;
18 18 int comm_len;
  19 +
  20 + void *priv;
19 21 };
20 22  
21 23 struct machine;