Clean-up: consumer: fix -Wshadow error in lttng_consumer_rotate_channel

We can safely re-use the chunk_status declared in the broader scope.

Change-Id: I0dc501168b1c35522a2ce4805d4514854162cefc
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>

Clean-up: config: fix -Wshadow error in config_load_session

Fix:

      CC       libconfig_la-session-config.lo
    /home/simark/src/lttng-tools/src/common/config/session-config.c: In function ‘config_load_session’:
    /home/simark/src/lttng-tools/src/common/config/session-config.c:4020:9: error: declaration of ‘path’ shadows a parameter [-Werror=shadow]
     4020 |    char path[PATH_MAX];
          |         ^~~~
    /home/simark/src/lttng-tools/src/common/config/session-config.c:3999:37: note: shadowed declaration is here
     3999 | int config_load_session(const char *path, const char *session_name,
          |                         ~~~~~~~~~~~~^~~~

Change-Id: Id7d5f6fcf0e88f43dee420dfb174d9bf01aa8fbd
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>

Clean-up: ust-consumer: fix -Wshadow issues in lttng_ustconsumer_recv_cmd

Same rationale and idea as previous patch, but for UST.

To differentiante the top-level `channel` variable, which owns a
channel, and the non-owning `channel` variables used when looking up
existing channels, the latter are renamed `found_channel`.

Change-Id: I6bf94be8d4368b5716756f4e12bae17c3e0d7cee
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>

Clean-up: kernel-consumer: fix -Wshadow issues in lttng_kconsumer_recv_cmd

There is a "ret" variable declared at the top and used throughout, but
there are also some other "ret" variables declared in narrower scopes.

Most of the time, the return value is not communicated from the "case"s
to the return value of the function. On failure, they goto an error
label which overrides ret to -1. There is one exception in "case
LTTNG_CONSUMER_ADD_STREAM", which goes directly to label end, not sure
if that's on purpose or not.

So what I did is: declare variables in narrower scopes when possible,
name them according to their intended use (e.g. ret_send, ret_recv). In
the end, there is no variable simply named "ret". The variable holding
the function's return value is "ret_func".

Change-Id: Ic02067c23757bd98d4b2d3d0f2b77384b4b0d743
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>

Clean-up: kernel-ctl: rename local variables in LTTNG_IOCTL_{,NO_}CHECK

Same rationale as previous patch.

Change-Id: Ie1782adad99616b3705c015360fa31d7e9273718
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>

Clean-up: common: rename local variables in PERROR

If PERROR is used in a function where a `buf` or `tmp` variable already
exists, we get this when enabling -Wshadow:

      CC       poll.lo
    In file included from /home/simark/src/lttng-tools/src/common/compat/poll.c:15:
    /home/simark/src/lttng-tools/src/common/compat/poll.c: In function ‘compat_epoll_set_max_size’:
    /home/simark/src/lttng-tools/src/common/error.h:247:9: error: declaration of ‘buf’ shadows a previous local [-Werror=shadow]
      247 |   char *buf; \
          |         ^~~
    /home/simark/src/lttng-tools/src/common/compat/poll.c:335:3: note: in expansion of macro ‘PERROR’
      335 |   PERROR("read set max size");
          |   ^~~~~~
    /home/simark/src/lttng-tools/src/common/compat/poll.c:313:7: note: shadowed declaration is here
      313 |  char buf[64];
          |       ^~~

Avoid that by giving PERROR's variables names that are unlikely to be
used elsewhere.

Change-Id: I167491fd3b232e6cfd86d13c0003ad4facb40c58
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>

Clean-up: filter: fix variable shadowing in visit_node_load_expression

It seems safe to just re-use the `ret` variable at the function-level
scope.

Change-Id: I7826edf2b8bb41f9c202af75a8b5e7dff9cbbcc8
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>

Clean-up: filter: rename variable in filter-grammar-test.c

... to avoid variable shadowing.

Change-Id: If5781b7e2f41ab5b1df361b9b247714b17751a32
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>

Clean-up: sessiond: remove some declarations from lttng-sessiond.h

As far as I can see, they aren't used anymore, there is no corresponding
definition.

Change-Id: Ie72e643e04da033bd590525de617201303f6dd23
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>

Clean-up: sessiond: prepend `the_` to global variable names

This avoids name clashes between global variables and local variables or
function parameters (notification_thread_handle, for example). This is a
step towards enabling -Wshadow.

This also helps readability, in my opinion, as it helps quickly spot
that some code is using a global variable.

Change-Id: Ib0e35ad7efcc54fa88e1900cab3388b98a06b8d9
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>

Fix: utils: avoid strncpy overlap in utils_partial_realpath

When running the test_utils_expand_path test with ASan enabled, I get:

➜  lttng-tools ./tests/unit/test_utils_expand_path
1..29
INPUT: /a/b/c/d/e
=================================================================
==1485873==ERROR: AddressSanitizer: strncpy-param-overlap: memory ranges [0x621000021d00,0x621000021d0b) and [0x621000021d00, 0x621000021d0b) overlap
    #0 0x7ffff761fd97 in __interceptor_strncpy /build/gcc/src/gcc/libsanitizer/asan/asan_interceptors.cpp:481
    #1 0x555555573834 in utils_partial_realpath /home/simark/src/lttng-tools/src/common/utils.c:195
    #2 0x55555557410b in _utils_expand_path /home/simark/src/lttng-tools/src/common/utils.c:374
    #3 0x555555574340 in utils_expand_path /home/simark/src/lttng-tools/src/common/utils.c:420
    #4 0x555555570b28 in test_utils_expand_path /home/simark/src/lttng-tools/tests/unit/test_utils_expand_path.c:274
    #5 0x55555557119e in main /home/simark/src/lttng-tools/tests/unit/test_utils_expand_path.c:345
    #6 0x7ffff725fb24 in __libc_start_main (/usr/lib/libc.so.6+0x27b24)
    #7 0x55555556fa3d in _start (/home/simark/build/lttng-tools/tests/unit/test_utils_expand_path+0x1ba3d)

0x621000021d00 is located 0 bytes inside of 4096-byte region [0x621000021d00,0x621000022d00)
allocated by thread T0 here:
    #0 0x7ffff7677639 in __interceptor_calloc /build/gcc/src/gcc/libsanitizer/asan/asan_malloc_linux.cpp:154
    #1 0x55555557269d in zmalloc /home/simark/src/lttng-tools/src/common/macros.h:45
    #2 0x555555573d34 in _utils_expand_path /home/simark/src/lttng-tools/src/common/utils.c:335
    #3 0x555555574340 in utils_expand_path /home/simark/src/lttng-tools/src/common/utils.c:420
    #4 0x555555570b28 in test_utils_expand_path /home/simark/src/lttng-tools/tests/unit/test_utils_expand_path.c:274
    #5 0x55555557119e in main /home/simark/src/lttng-tools/tests/unit/test_utils_expand_path.c:345
    #6 0x7ffff725fb24 in __libc_start_main (/usr/lib/libc.so.6+0x27b24)

0x621000021d00 is located 0 bytes inside of 4096-byte region [0x621000021d00,0x621000022d00)
allocated by thread T0 here:
    #0 0x7ffff7677639 in __interceptor_calloc /build/gcc/src/gcc/libsanitizer/asan/asan_malloc_linux.cpp:154
    #1 0x55555557269d in zmalloc /home/simark/src/lttng-tools/src/common/macros.h:45
    #2 0x555555573d34 in _utils_expand_path /home/simark/src/lttng-tools/src/common/utils.c:335
    #3 0x555555574340 in utils_expand_path /home/simark/src/lttng-tools/src/common/utils.c:420
    #4 0x555555570b28 in test_utils_expand_path /home/simark/src/lttng-tools/tests/unit/test_utils_expand_path.c:274
    #5 0x55555557119e in main /home/simark/src/lttng-tools/tests/unit/test_utils_expand_path.c:345
    #6 0x7ffff725fb24 in __libc_start_main (/usr/lib/libc.so.6+0x27b24)

The sole caller of utils_partial_realpath, _utils_expand_path, passes
the same buffer (resolved_path) for the input and output.  This causes
utils_partial_realpath to call strncpy with overlapping strings.

Fix it by making utils_partial_realpath allocate new memory for the
returned string itself.  This causes one more allocation than the
current code, because we don't re-use the existing buffer, but this
should be fine since this isn't exactly performance-critical code.
I think the code is easier to follow as a result.

Change-Id: I98a9aafc08d3bef45e3a83cbeef049f249b86f59
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>

Clean-up: utils: make utils_partial_realpath static

It is only used within its file.

Change-Id: I3e9b01e0cb765c74d3564a7c57598dace8104a29
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>

Fix: lttng: list_triggers: use proper format specifier for uint64_t

This fixes a warning on 32 bit systems.

Change-Id: Ie00e340cbe5a2b4e82bb544275bea914708d8e38
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>

Fix: sessiond: error accounting: wrong sizeof argument cpu_counters

1452205 Wrong sizeof argument

The wrong sizeof value is used in an expression or as argument to a function. The result is an incorrect value that may cause unexpected program behaviors.

In create_uid_accounting_entry: The sizeof operator is invoked on the wrong argument (CWE-569)

Reported-by: Coverity Scan
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I1912964bd1579f3b1ff013000848c28cea6a9d52

Fix: sessiond: error accounting: wrong sizeof argument cpu_counters

`cpu_counters` is an array of `struct lttng_ust_abi_object_data *`, not
of `struct lttng_ust_abi_object_data`. This didn't cause any problems
since the structure is larger than a pointer.

1452203 Wrong sizeof argument

The wrong sizeof value is used in an expression or as argument to a function. The result is an incorrect value that may cause unexpected program behaviors.

In event_notifier_error_accounting_register_app: The sizeof operator is invoked on the wrong argument (CWE-569)

Reported-by: Coverity Scan
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I9278949aab0a98f050ec232d7b8ce024240020d0

Fix: sessiond: notification: missing unlock on client list on error

1452204 Missing unlock

May result in deadlock if there is another attempt to acquire the lock.

In notification_thread_client_subscribe: Missing a release of a lock on a path (CWE-667)

Reported-by: Coverity Scan
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Iabfe3eecfbea89c96c7f16fbb25a0beef48f78b7

Docs: sessiond: document the rationale for the iteration on triggers list

Document the reason why an iteration is performed on the list of
triggers of a "client list" when a client subscribes to a condition.

Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I7d1c15016064ff2f6498c925f4933450fc7da30b

Fix: added missing test_channel and removed double defined test_syscall

Signed-off-by: Anders Wallin <wallinux@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I3f4963e3e10d81f18e4327e8189f9cb8316fe5ed

Clean-up: sessiond: blasphemous declaration of variable in code

Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I5200a2a3e077e3e33f6773a2b0ada1db04261a9f

Fix: sessiond: error accounting: leak of cpu_counter array on error path

1452202 Resource leak

The system resource will not be reclaimed and reused, reducing the future availability of the resource.

In event_notifier_error_accounting_register_app: Leak of memory or pointers to system resources (CWE-404)

Reported-by: Coverity Scan
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I07cc8808b05c497b9bc3099deca58e14f721c5a1

Fix: sessiond: error accounting: dereference of null pointer on error path

1452201 Dereference after null check

Either the check against null is unnecessary, or there may be a null pointer dereference.

In event_notifier_error_accounting_register_app: Pointer is checked against null but then dereferenced anyway (CWE-476)

Reported-by: Coverity Scan
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I169fd6434606b2a35e076ae9258de6597565effe

Fix: sessiond: error accounting: unchecked lttng_ht_del return value

There is no reason for this deletion to fail; assert that it succeeds
to detect any internal error.

1451556 Unchecked return value

If the function returns an error value, the error value may be mistaken for a normal value.

In event_notifier_error_accounting_unregister_event_notifier: Value returned from a function is not checked for errors before being used (CWE-252)

Reported-by: Coverity Scan
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I3d0f8995608ac96cd3c94b5e92c4459fff592702

Fix: sessiond: kernel: invalid error code check

The `EEXIST` error code does not make sense for a kernctl_disable()
operation. In fact, not existing would be an error.

Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Iae193c9c8d8697af3154a9862bf14d4398c60b60

Fix: sessiond: kernel: missing check for unregistration error

1451463 Unused value

An assigned value that is never used may represent unnecessary computation, an incorrect algorithm, or possibly the need for cleanup or refactoring.

In synchronize_tracer_notifier_unregister: A value assigned to a variable is never used. (CWE-563)

Reported-by: Coverity Scan
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Iaea550a2d70730cc4c90f8d1bdcd8cc0a4af6bd8

Tests: fix: unchecked sscanf return value

1407934 Unchecked return value

If the function returns an error value, the error value may be mistaken for a normal value.

In parse_arguments: Value returned from a function is not checked for errors before being used (CWE-252)

Reported-by: Coverity Scan
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I7733e88ea6af6313379bd8d4ca131cd243cda401

on-event evaluation: remove trigger name accessor

The trigger name accessor on an on-event evaluation is no longer needed
as the originating trigger is now accessible on a notification.

The tests are adapted to use the name of the trigger as returned by
the lttng_notification received by the client.

Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I57bd64d7f5ec35f015dfb115ecfadd75806b9a75

notification: add trigger accessor to lttng_notification

Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: If154de643f0d3ec0978b96551b3c3873f4b3380b

notification: transmit originating trigger as part of evaluation

Transmit the full trigger object instead of its condition when sending a
notification to a client. This allows clients to reliably filter on
triggers as, currently, only the name of a trigger is made available to
clients as part of the "on_event" evaluation object (the owner uid is
needed to perform a correct comparison for filtering).

This will also make the trigger accessible for all condition types.

Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I61d587436177eb34744487a1d3c486d87fe2e1f1

Fix: .gitignore: use full paths for test binaries

Using only the binary name in the .gitignore paths yields confusion
for git and search tools relying on this file's content (e.g. ag
silversearcher), because the binary file name may also be part of a
path, e.g. the filename "gen-ust-events" is also part of the path
leading to the .c file implementing this test:
"./tests/utils/testapp/gen-ust-events/gen-ust-events.c", which means
this .c file is ignored, even though it should not be.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I309b268fed8e66b693c1e0670044b40bf90593c4

Tests: use TRACEPOINT_INCLUDE

The macro TRACEPOINT_INCLUDE_FILE is deprecated since LTTng-UST 2.1. In
preparation for LTTng-UST 2.13, this compatibility macro is finally
removed.

The tests in lttng-tools need to keep up with times and start using its
replacement: TRACEPOINT_INCLUDE.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Id2bb15062f37878de03656dc5a47d9dd35687b41

Fix: sessiond: ust-app: assertion failure on registration of dead app

While setting up the event notifier group of a newly registered
application (see `ust_app_setup_event_notifier_group()`), communications
with the application can fail (generally because the application exited)
while sending the error counter configuration. This causes the UST
handle to the event notifier group to be released.

However, a second release is performed which results in undefined
behaviour (I have seen it succeed, assert, and crash under different
circumstances) while disposing of the application (see
`delete_ust_app()`).

Set the event notifier group ust handle to NULL to mark it as invalid
and prevent a second release during the clean-up.

Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ia5623a8d3f57174effa51c278e9654b6adc43289

common: credentials: hide symbols

Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I089502d46bdd0cf34b0721703a08fdd28f61b8f7

trigger: consider name in trigger _is_equal()`

For the moment, this is a breaking change since, before 2.13, users
could not register two identical triggers.

Triggers were previously unnamed, but a name is now generated
automatically. This means that a pre-2.13 user that could expect
a TRIGGER_ALREADY_EXISTS reply when registering a trigger that already
existed will now get "OK".

However, since it doesn't know about trigger names, it could now fail
to unregister a trigger (if the instance used to register the trigger
was not kept around).

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I9549980677e2387a7cc0f275233a3b5717fb915e

Fix: sessiond: notification: use after free of trigger object

Background
==========
* Clients can subscribe to certain specific conditions (e.g. buffer
  usage) using the `lttng_notification_channel_subscribe()` function.

* This subscription is only useful once a trigger with that condition
  AND at least one notify action is registered.

* The sessiond keeps a list for client subscribed to each registered
  condition.

* More than one trigger with the same condition may be registered
  the sessiond at the same time if they have different actions.

Issue
=====
Currently, when registering a trigger (T1) the sessiond looks if there
is already a client list for the condition of this trigger. If not, the
sessiond links the newly created client list object to that trigger T1
by keeping a pointer to it.

This means that if another trigger (T2) is registered with the same
condition (but a different name, or different actions) it will reuse the
same client list object and use the pointer to the T1.

This causes problems if T1 is unregistered before T2. In that case, the
pointer to T1 in the client list object is pointing to a deallocated
trigger object.

This issue is not encountered with the current test suite, namely the
`test_notification_multi_app` test case, because triggers with the same
condition also had the same action, so they are considered identical and
are not registered.

This issue was first witnessed when adding a trigger name comparison in
the `lttng_trigger_is_equal()` function.

Fix
===
Change the client list object so that it has its own copy of the
condition and a list of dependent triggers. Each trigger with that
condition has a reference on this client list object.

When unregistering a trigger, the notification thread removes it for the
client list's triggers list and put its reference on the client list
object.

Tests
=====
This commit adds a parameter to the base_client that dictates if the
notify action should be in a group. This is a trick to create triggers
that are not equal but have the same behaviour.

The `test_notification_multi_app` test case is modified to turn on this
option every other trigger registration.

Semi-related cleanups
=====================
* Merge `notification_client_list_create()` and
  `publish_notification_client_list()` functions since they are used
  together anyway. This removes the need to call
  `notification_client_list_put()` at the end of the
  `_register_trigger()` function

* Rename `trigger_applies_to_client()` to `condition_applies_to_client()`.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I3ebb90a1a64236a440a085e6fc1b82726a0e5af9

sessiond: Extract `{setup,teardown}_tracer_notifier()` functions

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ic394b1a50f166afc911b58607aece6ae69422841

Clean-up: sessiond: Extract trigger_ht_element removal function

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I4994d1dce7a8fc188b6df6ad72e5cd1081cb0cc7

Tests: event notifier error counters

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I9e2a4bab833d9204717029c1d5f594455c2262ce

sessiond: Implement UST event notifier error counter

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I254b2cec39b8c61ad945376043b5b9053de98f6f

common: document why spawn_viewer() is public

Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I09574ebd82bb583db5a7acec2b8f129a7dcd36fc

Docs: document trigger condition and action ownership in 2.13+

Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I8a52f166565efc87c6073713ce0f6569ba6d3661

Fix: sessiond: kernel: close on badfd on initialization error

The various kernel tracer file descriptors are closed during an
initialization error, but are not set to `-1` which causes them to
be closed again when the session daemon performs its clean-up.

Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I11d3ce7accea51394c34b9ca7ea0445e4ee422fc

Fix: sessiond: kernel error accounting fd still open when unloading modules

Issue
=====
I noticed that some kernel modules were not unloading properly when
shutting down the sessiond. Namely those modules:
  lttng_counter_client_percpu_64_modular
  lttng_ring_buffer_event_notifier_client

The call to `modprobe -r` for those modules is failing because the
kernel error accounting fd has not been closed.

Fix
===
Only call `modprobe -r` after all kernel resources (file descriptors)
have been released.

Side note
=========
The `modprobe_remove_lttng()` is currently printing that the removal of a
_OPTIONAL module was a success even if it failed. I rework this function
to print a debug message in such situation. I used the
`modprobe_lttng()` as a template for this function.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ibb51425ef3deedf6e982084ef879bc83ac823ba3

sessiond: Implement kernel event notifier error counter

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I390113873567a98d7e53db605d48b59a60a1074e

Fix: on-event condition: don't abort() on invalid event expression type

Protocol errors should not result in crashes. Simply report the
failure to deserialize the event expression.

Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ic5a0be6b248c430fcf1fe487cd734d9bdf5dd36a

Fix: common: index_allocator_get_index_count() returns size

The lttng_index_allocator_get_index_count() should return the number of
allocated indexes. Right now it returns the maximum number of indexes.

I witnessed the following message when running testing the event
notifier error counter feature:
  Warning: Destroying index allocator with 4096 slot indexes still in use

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ibbcb3f128553df0a4118901ef85ae6fe17ebcecf

on-event: add error counter fields to condition

Modified from proposed change: don't include error counter index in serdes

The error counter index is an internal implementation detail that
doesn't need to be exposed to the client (the accessor is not reacheble
to them anyway).

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ibc84d4cc3bcc7e304f4b52bd45f1de010dd8b87b

compiler warning cleanup: is_signed_type: compare -1 to 1

Comparing -1 to 0 triggers compiler warnings (gcc -Wtype-limits and
-Wbool-compare) and Coverity warning "Macro compares unsigned to 0".

Comparing -1 to 1 instead takes care of silencing those warnings while
keeping the same behavior.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Id4aebce6eed214f0ccba2c0289fb3b034808cb64

common: shm.c: improve logging on various error paths

Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I1d68f952b8a01b79aeab7c8d05d98085071a45a7

Cleanup: Move `create_posix_shm()` to common/shm.c

* This function will be used for trigger error accounting.

* This function is renamed to `shm_create_anonymous()` and now takes one
  parameter for the owner of the shared memory area. Again this is code
  reuse.

* Remove erroneous comment about the `name` parameter for the
  `shm_open()` function. It's the absence (and not its presence) of the
  '/' at the beginning of the name that makes the behaviour implementation
  defined.

* Move the bin/lttng-sessiond/shm.c to common/shm.c so that all shm
  related functions are in one location.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I690b38801ffb4772e7c5069c4cccf470b0671f63

common: Add index allocator for error counters

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I677a8f5b774315eadb8e3b2c2f2a5b1aae02133f

Fix: sessiond: agent: lazy creation of agent on removal

When unregistering a trigger that was previously registered, the
agent corresponding to the trigger's domain will already exist. It is
superfluous to handle the case where it doesn't exist and it would only
hide an internal error.

Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ifc6d14b74586af8423a70b3466e0a2c00491d3c3

Clean-up: sessiond: make some accesses to conditions const

Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I88edd184c06d2f2912331da8e83b624f8ae8bcc2

Extract synchronize tracer notifier functions

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I2d92227d93ede0b04bf7314ff819c2b777247262

Cleanup: Extract `ust_app_synchronize_all_channels()` function

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Id24eff10cc07c5e061fbb45e871087c17eac9b1d

ust-abi: add missing lttng_ust_abi prefixes

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I03f465659cd7b044d1ec10458fa77332d5e9db41

clock override: use getter API from LTTng-UST

Currently, the lttng session daemon reimplements its own copy of
struct lttng_trace_clock, which is pretty bad as ABI go.

Use getter functions for each of the clock callback instead. This
refactoring needs to be introduced with a matching commit to implement
the new ABI in LTTng-UST.

Move the implementation of the clock frequency, uuid, name and
description default callbacks (for monotonic clock) to lttng-ust.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: If2e671bacacc55d3fb27b6b4d4fc8c16aaffd059

Fix: bytecode test should be in UST_FILTERS, not UST_STR_FILTERS

The UST test is in the wrong array, making it always succeed when it
should fail when run with a non-patched UST, because it uses a different
test application which does not contain the targeted fields.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I96f74e26ffdead5e03e069c944f215967b7890f2

Tests: test bytecode linker refusal of non-string arrays and sequences

The bytecode linker should refuse loading arrays and sequences of data
which are not strings for the purpose of comparison against other
strings.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Id3b14cb6674b0e2e1edaf039fbc5f929708e2451

Tests: utils.sh: pass arbitrary arguments to start_lttng_sessiond*

Allows for easier arguments passing to lttng-sessiond. This is useful
both for debugging manually and passing specific flag as needed in
testing. Passing  "--event-notifier-error-number-of-bucket=X" or "-vvv"
for examples.

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I36bd6eaf1c4fc84c3a1e8cfe70a37a46e7130c75

Cleanup: clarify `buffer_reg_channel` and `ust_registry_channel` variables

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I3eab9f1c2f53a72209ae95c05a5f9f3669e4ea0f

Fix: return value of register trigger to the notification thread is not acted upon

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I0fd6dea34854b82b6d4c1550c4f84a7c79f9076f

Tests: Cleanup: gen-syscall-events: remove duplicated help line

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Id5bc54598f8abf7498f3b81064a83c60379922d6

Tests: Cleanup: add `lttng_{add, remove}_trigger_ok()` bash functions

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I63b1f4d2a72b291f93df4548114014a116824864

Tests: fix: log level rule: leak of rule after test

1451079 Resource leak

The system resource will not be reclaimed and reused, reducing the future availability of the resource.

In test_log_level_rule_at_least_as_severe_as: Leak of memory or pointers to system resources (CWE-404)

Reported-by: Coverity Scan
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ica923f9f010e51217ca62664faab99cb34999c0b

Fix: lttng-sessiond: kernel: leak of event notifier rule

1445768 Resource leak

The system resource will not be reclaimed and reused, reducing the future availability of the resource.

In kernel_create_event_notifier_rule: Leak of memory or pointers to system resources (CWE-404)

Reported by: Coverity Scan
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I58862b3b38c7e61806e0047840bf68619273db49

Tests: add log level rule validation tests

Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Idabc5420ac2b188d13ed5537be81128976661357

Fix: event-rule: report log level rules as unsupported for kernel domain

Log levels are not supported for by tracepoints for the kernel domain.
Return `LTTNG_EVENT_RULE_STATUS_UNSUPPORTED` when setting a log level
rule on a kernel-domain tracepoint.

Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I73b6067ec0a11ae1f83098a6966dc956e1772224

Tests: Fix: list-triggers test case out of date

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I9d78c025693fce9dc0ae2177d71d24217a4b1d6a

Tests: add kernel --function basic test case

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I584be60bd7325433aa9c30338ceb652c48806ac6

Rename event rule kretprobe to kernel function

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ibbd0cfa74e7d8e5b922debd23fc5f89cdeb14621

on-event evaluation: introduce on-event evaluation specific status code

Before this change, `lttng_evaluation_on_event_get_captured_values`
returns `LTTNG_EVALUATION_STATUS_INVALID` when an evaluation's
condition has no capture descriptors. That status code is also used to
indicate a handful of "invalid parameter" conditions.

Not having captured values in an evaluation should not be considered an
error. Hence, a status enum that is specific to on-event evaluations is
added to introduce a "NONE" status that is not an error (positive value)
and wouldn't make sense for all evaluations (it could mean an error in
other circumstances).

`LTTNG_EVALUATION_ON_EVENT_STATUS_NONE` is returned when the condition
of an on-event evaluation has no capture descriptors.

Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ic2e3e90317a44c7bb3e95630956565690bffd760

event-notifier: implement `lttng_trigger_needs_tracer_notifier()` function

This function is used to tell if a trigger needs a tracer notifier. This
depends on its condition and its actions.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ie3c928ba1c705c32e4c2e4c6b6b22aa8c10837a9

lttng-ctl: Introduce lttng_log_level_rule

This object corresponds to the criteria a event-rule tracepoint must
respect for an event to be consired under the event-rule tracepoint
criteria.

The lttng tracepoint event rule object now have the following API
function in regards to log level rule:

  lttng_event_rule_tracepoint_set_log_level_rule
  lttng_event_rule_tracepoint_get_log_level_rule

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Iefd361d1b7d85c6803e891166429a2830e2404a2

Clean-up: sort noinst_PROGRAMS in test/unit/Makefile.am

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I99cb52748495cd7cc364af7ea04ae95a75bb5759

Clean-up: sort TESTS in test/unit/Makefile.am

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Id4d245eef4f0b99038c7ee76d9c783f40d8b849a

Clean-up: sort files in include/Markefile.am

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Idcdd023916084610669472a55164a3d6e0192496

event-rule: userspace probe: force location on create

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I5775184c62249c642af0ca0315103e040cb2502d

event-rule: userspace probe: rename get/set_name to get/set_event_name

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I4a7fbd26b4dfe352100a6fa78b828c1da0fa9928

event-rule: kernel probe: force location on create

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I485ab3bb32317d97ad46f7de71c72a37afcd64c2

event-rule: kernel probe: rename set/get_name to set/get_event_name

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I14950a663a4db37fe0f120d8f06c01cae241a871

event-rule: syscall: set the default pattern to '*'

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I8456f7f125b9341d31bc7bc2b8474abbc44b5467

event-rule: tracepoint: set default pattern to '*'

'*' is set on tracepoint creation.

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Id20b5e3924a11a336fb08ac6ce10f08ad7fcd80a

Rename files for condition event-rule to on-event

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I8bde2f15674be16c0a21c670affff4c4503ed846

Rename lttng_condition_event_rule to lttng_condition_on_event

Files renaming is following.

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Icf74aa7cc84b46b1f23bbe1418b7e1241d8dc081

Rename uprobe files to userspace-probe

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ide5aeca0d56b8a368fa26bf7cc082098e349d8d9

Rename uprobe to userspace-probe

Renaming of associated files follows-up.

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Iabb6f3599efbac25a4c31b892b791b98ed4f5990

Rename kprobe file to kernel-probe

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I379bd8d88b4fa4a7d5638dc5bee4d7dd86426dc9

Rename kprobe to kernel_probe

Renaming of the kprobe based file follows.

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: If9f500234bdb1a97752ffeff8c8549500a5f841a

Tests: Implement trigger capture test cases

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I661fd8f72c871c4c38b941e9c9c5f6054383354c
Depends-on: lttng-ust: I8423c510bf6af2f9bf85256e8d6f931d36f7054b

man: lttng-add-trigger: document --capture option

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Id4041c6647ee9b6ea19896d5a926f910fbf1d378
Depends-on: lttng-ust: I8423c510bf6af2f9bf85256e8d6f931d36f7054b

CLI: make list-triggers command print capture expressions

When printing an event rule condition, in the `list-triggers` command,
print the capture expressions associated with that condition.

The result looks like:

    - id: T1
      condition: event rule hit
        rule: allo (type: tracepoint, domain: ust)
        captures:
          - $ctx.foo
      actions:
        notify
    - id: T2
      condition: event rule hit
        rule: allo (type: tracepoint, domain: ust)
        captures:
          - $app.iga:active_clients
      actions:
        notify

Change-Id: Iaa902369a2df3edf2928a935f7565a7c21acb861
Signed-off-by: Simon Marchi <simon.marchi@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Depends-on: lttng-ust: I8423c510bf6af2f9bf85256e8d6f931d36f7054b

Support capture for kernel tracer

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I26167d259daa8e356a1bb03831ccf6e570b8e305
Depends-on: lttng-ust: I8423c510bf6af2f9bf85256e8d6f931d36f7054b

lttng: Capture is only supported by tracepoint and syscall event-rules

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Idfe07be4e12e0531dfdb778cdb4c011780af7a26
Depends-on: lttng-ust: I8423c510bf6af2f9bf85256e8d6f931d36f7054b

sessiond: notification-thread: Missing action executor status handling

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I8666c0728ca140c45d47ac1ecf252bdbf0526e46

sessiond: notification-thread: Missing domain internal header

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I0fbd0cd8bd688b1983430c9036351ca501f5abd7

sessiond: action-executor: Missing notify header

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: I6d00f41d1560a0fc73127aca1e37cd9e100f20ee

Make captured field values available to event rule cond. evaluation

liblttng-ctl:
    Add the public lttng_evaluation_get_captured_values() function.

    This function returns an array event field value which contains
    the captured field values as requested by the evaluation's
    condition's capture descriptors.

    This array event field value contains as many elements as returned
    by lttng_condition_event_rule_get_capture_descriptor_count() for the
    evaluation's condition. However,
    lttng_event_field_value_array_get_element_at_index() could return
    `LTTNG_EVENT_FIELD_VALUE_STATUS_UNAVAILABLE` for some or all indexes
    if there was no such field to capture at tracing time.

Common:
    Internally, an event rule evaluation now contains a captured field
    value payload (CFVP) as well as the array event field value to be
    returned by lttng_evaluation_get_captured_values().

    The CFVP is MessagePack-encoded.

    When you create an event rule evaluation with
    lttng_evaluation_event_rule_create(), you must pass the CFVP as well
    as the `decode_capture_payload` option parameter. If
    `decode_capture_payload` is `true`,
    lttng_evaluation_event_rule_create() creates the array event field
    value from the CFVP thanks to msgpack-c.

    lttng_evaluation_event_rule_create() also accepts the event rule
    condition now because it relies on the capture descriptors to create
    the event field values if `decode_capture_payload` is `true`.

    When an event rule evaluation is serialized:
        Only its CFVP is serialized (as is).

    When an event rule evaluation is deserialized:
        lttng_evaluation_event_rule_create() is called with the
        `decode_capture_payload` parameter set to `true`.

Session daemon:
    action_executor_notify_handler() calls
    lttng_evaluation_event_rule_create() with the
    `decode_capture_payload` parameter set to `false` to send the
    evaluation to the clients with the CFVP.

    This CFVP directly comes from the tracer.

Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ib2ae39dae005ad555eb8309e0bfd53f746d5e8dd
Depends-on: lttng-ust: I8423c510bf6af2f9bf85256e8d6f931d36f7054b

notification: fetch capture payload on notification reception

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Id52749b59969276b5390f4494df48940662d5d8c

Fix: set_ust_capture: missing ABI rename

struct lttng_ust_object_data has become struct
lttng_ust_abi_object_data. It gives a compiler warning.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Ia29eac5903b43a15c02df7204cf5344226d30381

tests: perf: UNHALTED_REFERENCE_CYCLES might not be actionable on a host

This patch does NOT address the root problem and only addresses the
validation of the context to be added during the test suite.

Observed issue
==============

The system_tests jobs for master hangs on the perf event test suites.

Cause
=====

The hang is caused by a cleanup problem (reference counting of the trace
chunk on session destroy/rotation) when the activation of a context
fails on a ust app channel.

This patch does NOT address the root problem and only addresses the
validation of the context to be added during the test suite. In all
cases we need to handle when a context fail, but for this test we need
to validate that the context can be added and skip the tests as
necessary based on the host.

The perf tests depend on the presence and accessibility of the
UNHALTED_REFERENCE_CYCLES PMU counter. This test suite was previously
run "manually" and since it required the presence and access to that
PMU. Since that the perf test suite is now run on `make check` when
libpfm is present, we need to automate the discovery of
UNHALTED_REFERENCE_CYCLES and validate that we can access it.

There are three major scenarios were we want to skip the tests.

1) UNHALTED_REFERENCE_CYCLES is simply not present in the PMU sets for
that hosts.

2) UNHALTED_REFERENCE_CYCLES is present in the PMU sets but not
actionable. This can happen on qemu guests.

3) UNHALTED_REFERENCE_CYCLES is present but not accessible. This can be
happen if the `/proc/sys/kernel/perf_event_paranoid` prevents the usage
of the PMU.

Solution
========

Two problems were found with `find_event.c`.

 1) It took the first event matching the passed name even if it was in a
 PMU not supported by the host. In our use case it worked since the only
 platform that does not use `r300` is not currently in our testing set.

 -> = PMU set currently choosen
 -* = The correct PMU set

 e.g:
 -> Intel Core
      r300
    Intel Atom
      r300
    Intel Nehalem
      r300
    Intel Nehalem EX
      r300
    Intel X86 architectural PMU
      r13c
    ...
 -* Intel Skylake
     r300

 On my system only the following are "detected" as per libpfm example
 found here [1].

   [18, ix86arch, "Intel X86 architectural PMU"]
   [51, perf, "perf_events generic PMU"]
   [110, rapl, "Intel RAPL"]
   [114, perf_raw, "perf_events raw PMU"]
   [200, skl, "Intel Skylake"]

 Hence the `skl` PMU set should be used.

 2) libpfm does not perform any validation as to if the event is actually
 usable or not.

To fix those two problems, we use pfm_get_os_event_encoding and
perf_event_open.

pfm_get_os_event_encoding [2] is responsible for performing the query
across valid PMU sets and encoding it to the perf struct format.

perf_event_open is used to validate that the event can be used. It tests
the availability on the running host and the accessibility of the PMU.

Based on the result of `find_event` the tests are skipped or failed as
necessary.

Known drawbacks
========

The only drawback is that the tests, albeit having libpfm as a
dependency, are not guarantee to run on all hosts. There is not much we
can do here. We can only validate that it is indeed run on our CI, most
probably using lava hardware-based workers.

References
==========

[1] https://sourceforge.net/p/perfmon2/libpfm4/ci/288483932c3eb83202b0d8762aa0ed8534982c3f/tree/examples/check_events.c
[2] https://man7.org/linux/man-pages/man3/pfm_get_os_event_encoding.3.html
[3] https://man7.org/linux/man-pages/man2/perf_event_open.2.html

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Jérémie Galarneau <jeremie.galarneau@efficios.com>
Change-Id: Iea7794dc28d019953930992a2237a1b606368d1f