Filter: add FILTER_OP_RETURN_S64 instruction

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Perform bitwise ops on unsigned types

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Filter: catch shift undefined behavior

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Filter: add lshift, rshift, bit not ops

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Filter: index array, sequences, implement bitwise binary operators

Implement indexing of array and sequence of integers, as well as bitwise
binary operators &, |, ^.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: update ext4 instrumentation for kernel 4.13

See this upstream commit :

  commit a627b0a7c15ee4d2c87a86d5be5c8167382e8d0d
  Author: Eric Whitney <enwlinux@gmail.com>
  Date:   Sun Jul 30 22:30:11 2017 -0400

      ext4: remove unused metadata accounting variables

      Two variables in ext4_inode_info, i_reserved_meta_blocks and
      i_allocated_meta_blocks, are unused.  Removing them saves a little
      memory per in-memory inode and cleans up clutter in several tracepoints.
      Adjust tracepoint output from ext4_alloc_da_blocks() for consistency
      and fix a typo and whitespace near these changes.

Signed-off-by: Eric Whitney <enwlinux@gmail.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: Sleeping function called from invalid context

It affects system call instrumentation for accept, accept4 and connect,
only on the x86-64 architecture.

We need to use the LTTng accessing functions to touch user-space memory,
which take care of disabling the page fault handler, so we don't preempt
while in preempt-off context (tracepoints disable preemption).

Fixes #1111

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: sched for v4.11.5-rt1

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Make vim users life easier

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Rename Makefile.ABI.workarounds to Kbuild.common

This file is now used for code which is common to all Kbuild files.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: handle missing ftrace header on v4.12

Properly handle the case where we build against the distro headers of a
kernel >= 4.12 and ftrace is enabled but the private header is
unavailable.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: pid tracker should track "pgid"

The "pid" notion exposed by LTTng translates to the "pgid" notion in the
Linux kernel. Therefore using "current->pid" as argument to the PID
tracker actually ends up behaving as a "tid" tracker, which does not
match the intent nor the user-space tracer behavior.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: typo in lttng pid tracker

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: Build ftrace probe on kernels prior to 4.12

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: update ftrace probe for kernel 4.12

Follow changes introduced by Linux upstream commits:
  ec19b85913486993d7d6f747beed1a711afd47d8
  bca6c8d0480a8aa5c86f8f416db96c71f6b79e29
  b5f081b563a6cdcb85a543df8c851951a8978275
  6e4443199e5354255e8a4c1e8e5cfc8ef064c3ce

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: update block instrumentation for kernel 4.12

Follow changes introduced by Linux upstream commits:
  48b77ad6084481ef9330a5d2bee289966da0975b
  cee4b7ce3f9161c88f7255a3d73c1c4d5bbabea7
  caf7df12272118e0274c8353bcfeaf60c7743a47

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Calculate context length outside of retry loop

Allow context length calculation to have side-effects (e.g. page faults)
which trigger event tracing by moving the calculation outside of the
buffer space reservation retry loop.

This also paves the way to have dynamically sized contexts, which
would expect to put their size of the internal stack. Note that the
context length calculation is performed *after* the event payload field
length calculation, so the stack needs to be used accordingly.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: Add support for 4.9.27-rt18 kernel

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: update btrfs instrumentation for kernel 4.12

See upstream commit 490b54d6fb75f6ffd0471ec58bb38a992e2b40cd

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: update ringbuffer for kernel 4.12

flags removed from splice_pipe_desc in 4.12.

See upstream commit f81dc7d7d5a2528f98f26a0b9406e822d0b35011

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: update sched instrumentation for kernel 4.12

See upstream commit b91473ff6e979c0028f02f90e40c844959c736d8

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: ext3 was completely removed from the kernel in v4.3

Don't display the warning about missing ext3 headers on kernels >= 4.3

See upstream commit e31fb9e00543e5d3c5b686747d3c862bc09b59f3

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: NULL pointer dereference of THIS_MODULE with built-in modules

THIS MODULE is defined to 0 when a module is built-in the kernel [1].
This caused NULL pointer dereference when booting a kernel with the
lttng-modules built-in.
To fix this issue, add #if guard around the wrapper_lttng_fixup_sig
function checking if the MODULE macro is defined to confirm that this
piece of code will end up in a module and not in the kernel itself.

[1]: linux/include/linux/export.h:32
Fixes: #1107
Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: add "flush empty" ioctl for stream intersection

Changing the behavior of the "snapshot" lttng command to implicitly do a
buffer "flush" (even when current packet is empty) had unwanted
side-effects: for instance, the snapshot ABI is used by the live timer
to grab the buffer positions, and we don't want to generate useless
empty packets in that scenario.

Therefore, add the "flush empty" behavior as a new ioctl to the ring
buffer. This allows lttng-tools to perform buffer flush (even for empty
packets) when it needs to. Given that this new ioctl is added within
stable branches as well, lttng-tools always need to handle "-ENOSYS"
gracefully.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Revert "Fix: flush empty packets on snapshot channel"

This reverts commit dc5cd5702b74d72f0db0141c6d888a1d820aed9c.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Revert "Fix: don't perform extra flush on metadata channel"

This reverts commit 7cf44d034bdda1896f6b0c6374c90c06d45ee4fd.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Version 2.10.0-rc1

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: remove CONFIG_KALLSYMS_ALL warning on clean

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS command

There is no need to bump the LTTNG_MODULES_ABI_MINOR_VERSION
since the multiple wildcard feature introduced as part of the 2.10
release already bumps it from 2 to 3.

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

Fix: Always build vmscan probe

The mm/vmscan.c compile unit is a obj-y, even on an old 2.6.36 kernel,
always build the vmscan probe regardless of kernel configuration.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: formatting in strutils_star_glob_match explanation

Replace tabs for spaces in example scenario.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: introduce LTTNG_SIZE_MAX for older kernels

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Use SIZE_MAX instead of -1ULL for size_t parameter

strutils_star_glob_match() receives a size_t. Passing -1ULL truncates
the value implicitly on systems where size_t is 32-bit. It is cleaner to
use SIZE_T.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

filter: use SIZE_MAX for size_t

The backing type is a size_t, so use SIZE_MAX to represent infinity.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: out of bound array access in filter code

Fix ported from lttng-ust, initially found by Coverity.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add support for star globbing patterns in event names

This patch adds support for full star-only globbing patterns used in
the event names (enabler names).

strutils_star_glob_match() is always used to perform the match when
the enabler is LTTNG_ENABLER_STAR_GLOB. This enabler is set when it is
detected that its name contains at least one non-escaped star with
strutils_is_star_glob_pattern().

The match is performed by strutils_star_glob_match(), the same function
that the filter interpreter uses.

Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Filtering: add support for star-only globbing patterns

This patch adds the support for "full" star-only globbing patterns to be
used in filter literal strings. A star-only globbing pattern is a
globbing pattern with the star (`*`) being the only special character.
This means `?` and character sets (`[abc-k]`) are not supported here. We
cannot support them without a strategy to differentiate the globbing
pattern because `?` and `[` are not special characters in filter literal
strings right now. The eventual strategy to support them would probably
look like this:

    filename =* "?sys*.[ch]"

The filter bytecode generator in LTTng-tools's session daemon creates
the new FILTER_OP_LOAD_STAR_GLOB_STRING operation when the interpreter
should load a star globbing pattern literal string. Even if both
"plain", or legacy strings and star globbing pattern strings are literal
strings, they do not represent the same thing, that is, the == and !=
operators act differently.

The validation process checks that:

1. There's no binary operator between two
   FILTER_OP_LOAD_STAR_GLOB_STRING operations. It is illegal to compare
   two star globbing patterns, as this is not trivial to implement, and
   completely useless as far as I know.

2. Only the == and != binary operators are allowed between a
   star globbing pattern and a string.

For the special case of star globbing patterns with a star at the end
only, the current behaviour is not changed to preserve a maximum of
backward compatibility. This is also why the ABI version is changed from
2.2 to 2.3, not to 3.0.

== or != operations between REG_STRING and REG_STAR_GLOB_STRING
registers is specialized to FILTER_OP_EQ_STAR_GLOB_STRING and
FILTER_OP_NE_STAR_GLOB_STRING. Which side is the actual globbing pattern
(the one with the REG_STAR_GLOB_STRING type) is checked at execution
time. The strutils_star_glob_match() function is used to perform the
match operation. See the implementation for more details.

Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add string utilities

The new lttng-string-utils.c file has a few utility functions to
manipulate and check strings. See lttng-string-utils.c for more details.

Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

lttng-abi.c: cleanup whitespaces

Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: use of uninitialized ret value in lttng_abi_open_metadata_stream

Fixes the following compiler warning:

lttng-abi.c: In function ‘lttng_metadata_ioctl’:
lttng-abi.c:971:6: warning: ‘ret’ may be used uninitialized in this function [-Wmaybe-uninitialized]
  int ret;
      ^

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: kref changes for kernel 4.11

The underlying type of `struct kref` changed in kernel 4.11 from an
atomic_t to a refcount_t. This change was introduced in kernel
commit:10383ae. This commit also added a builtin overflow checks to
`kref_get()` so we use it.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: atomic_add_unless() returns true/false rather than prior value

The previous implementation assumed that `atomic_add_unless` returned
the prior value of the atomic counter when in fact it returned if the
addition was performed (true) or not performed (false).
Since `atomic_add_unless` can not return INT_MAX, the `lttng_kref_get`
always returned that the call was successful.

This issue had a low likelihood of being triggered since the two refcounts
of the counters used with this call are both bounded by the maximum
number of file descriptors on the system.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: timers cputime_t arguments replaced by ull in kernel 4.11

cputime_t was changed to ull in the kernel commit: 858cf3a

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: update scsi instrumentation for kernel 4.11

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: changes to the vm_op fault cb prototype in libringbuffer

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: update btrfs instrumentation for kernel 4.11

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: update mm_vmscan instrumentation for kernel 4.11

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: section mismatch warning caused by __exit annotation

lttng_logger_exit is used in a non-exit function so it should not be
annotated with `__exit`.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

socketpair: extend syscall socketpair tracing information

Decode the socketpair vector pointer into two file descriptors.
This exposes the connected file descriptors to analyses.

As sockerpair is a sub syscall of socketcall in x86_32,
sockerpair override must be disabled for x86_32 and x86_compatmode

Signed-off-by: Jan Willeke <jan.willeke@harman.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Remove events/mainline unused headers

We can actually diff from Linux kernel headers directly instead of
keeping stale unused copies of those headers.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

update event README

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: nmi-safe clock on 32-bit systems

On 32-bit systems, the algorithm within lttng-modules that ensures the
nmi-safe clock increases monotonically on a CPU assumes to have one
clock read per 32-bit LSB overflow period, which is not guaranteed. It
also has an issue on the first clock reads after module load, because
the initial value for the last LSB is 0. It can cause the time to stay
stuck at the same value for a few seconds at the beginning of the trace,
which is unfortunate for the first trace after module load, because this
is where the offset between realtime and trace_clock is sampled, which
prevents correlation of kernel and user-space traces for that session.

It only affects 32-bit systems with kernels >= 3.17.

Fix this by using the non-nmi-safe clock source on 32-bit systems.

While we are there, remove an implementation-defined c99 behavior
regarding casting u64 to long by using unsigned arithmetic instead:

turn:
  if (((long) now - (long) last) < 0)
into:
  if (U64_MAX / 2 < now - last)

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: only include linux/cpuhotplug.h for kernels >= 4.10

Kernels at least <= 4.4 did not have this header file.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: 4.10 hotplug adaptation backward compat

                 from /home/compudj/git/lttng-modules/lttng-context-perf-counters.c:23:
/home/compudj/git/lttng-modules/lttng-context-perf-counters.c: In function ‘lttng_add_perf_counter_to_ctx’:
/home/compudj/git/lttng-modules/lttng-context-perf-counters.c:353:22: error: ‘cpu’ undeclared (first use in this function)
  for_each_online_cpu(cpu) {
                      ^
./include/linux/cpumask.h:223:8: note: in definition of macro ‘for_each_cpu’
  for ((cpu) = -1;    \
        ^
/home/compudj/git/lttng-modules/lttng-context-perf-counters.c:353:2: note: in expansion of macro ‘for_each_online_cpu’
  for_each_online_cpu(cpu) {
  ^
/home/compudj/git/lttng-modules/lttng-context-perf-counters.c:353:22: note: each undeclared identifier is reported only once for each function it appears in
  for_each_online_cpu(cpu) {
                      ^
./include/linux/cpumask.h:223:8: note: in definition of macro ‘for_each_cpu’
  for ((cpu) = -1;    \
        ^
/home/compudj/git/lttng-modules/lttng-context-perf-counters.c:353:2: note: in expansion of macro ‘for_each_online_cpu’
  for_each_online_cpu(cpu) {
  ^
./include/linux/cpumask.h:224:38: warning: left-hand operand of comma expression has no effect [-Wunused-value]
   (cpu) = cpumask_next((cpu), (mask)), \
                                      ^
./include/linux/cpumask.h:717:36: note: in expansion of macro ‘for_each_cpu’
 #define for_each_online_cpu(cpu)   for_each_cpu((cpu), cpu_online_mask)
                                    ^
/home/compudj/git/lttng-modules/lttng-context-perf-counters.c:353:2: note: in expansion of macro ‘for_each_online_cpu’
  for_each_online_cpu(cpu) {
  ^
scripts/Makefile.build:289: recipe for target '/home/compudj/git/lttng-modules/lttng-context-perf-counters.o' failed
make[2]: *** [/home/compudj/git/lttng-modules/lttng-context-perf-counters.o] Error 1
make[2]: *** Waiting for unfinished jobs....

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: 4.10 btrfs instrumentation update backward compat

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Update btrfs instrumentation for 4.10 kernel

Based on commit 92a1bf76 "Btrfs: add 'inode' for extent map tracepoint"
in the upstream Linux kernel.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Adapt lttng-modules to Linux 4.10 cpu hotplug state machine

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

btrfs instrumentation: update to 4.10 kernel

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

timer instrumentation: adapt to ktime_t without union

Introduced in Linux upstream in 4.10.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add load/unload messages to kernel log

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Update version to 2.10.0-pre

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: asoc instrumentation for RHEL 7.3

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: SCSI instrumentation for SLES12 SP2

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add SUSE Linux Enterprise kernel version tests

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Filter code relicensing to MIT license

Relicense the filtering code to MIT license.

I am the principal author of this code. Julien Desfossez gave the
approval for his modifications.

Acked-by: Julien Desfossez <jdesfossez@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add task cpu in process statedump

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Performance: add missing unlikely in reserve

Add missing branch prediction hints within lttng_event_reserve().

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: preemptible and migratable context error handling

When built against preempt-rt and preempt kernels, the "return 0" case
means success, but lttng-modules incorrectly prints an error in the
kernel log.

Given that we handle the -ENOSYS error in lttng_context_init, there is
no need to keep the ifdefs in that function.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: bump stable kernel version ranges for clock work-around

Linux commit 27727df240c7 ("Avoid taking lock in NMI path with
CONFIG_DEBUG_TIMEKEEPING"), changed the logic to open-code
the timekeeping_get_ns() function, but forgot to include
the unit conversion from cycles to nanoseconds, breaking the
function's output, which impacts LTTng.

We expected Linux commit 58bfea9532 "timekeeping: Fix
__ktime_get_fast_ns() regression" to make its way into stable
kernels promptly, but it appears new stable kernel releases were
done before the fix was cherry-picked from the master branch.

We therefore need to bump the version ranges for the work-around
in lttng-modules.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
CC: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
CC: John Stultz <john.stultz@linaro.org>

Version 2.9.0-rc1

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: i2c: support kernels < 3.15

i2c instrumentation has only been added in kernel 3.15.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: show warning for broken clock work-around

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Bump minor ABI version

Command added: LTTNG_KERNEL_SESSION_STATEDUMP

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: work-around upstream Linux timekeeping bug

Linux commit 27727df240c7 ("Avoid taking lock in NMI path with
CONFIG_DEBUG_TIMEKEEPING"), changed the logic to open-code
the timekeeping_get_ns() function, but forgot to include
the unit conversion from cycles to nanoseconds, breaking the
function's output, which impacts LTTng.

The following kernel versions are affected: 4.8, 4.7.4+, 4.4.20+,
4.1.32+

We expect that the upstream fix will reach the master and stable
branches timely before the next releases, so we use 4.8.1, 4.7.7,
4.4.24, and 4.1.34 as upper bounds (exclusive).

Fall-back to the non-NMI-safe trace clock for those kernel versions.
We simply discard events from NMI context with a in_nmi() check,
as we did before Linux 3.17.

Link: http://lkml.kernel.org/r/1475636148-26539-1-git-send-email-john.stultz@linaro.org
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add support for i2c tracepoints

This patch teaches lttng-modules about the i2c tracepoints in the Linux
kernel.

It contains the following tracepoints:

  * i2c_write
  * i2c_read
  * i2c_reply
  * i2c_result

I translated the fields and assignments from the kernel's
include/trace/events/i2c.h as well as I could.  I also tried building
this module against a kernel without CONFIG_I2C, and it built fine (the
required types are unconditionally defined).  So I don't think any "#if
CONFIG_I2C" or similar are required.

A module parameter (extract_sensitive_payload) controls the extraction
of possibly sensitive data from events.

[ With edit by Mathieu Desnoyers. ]

Signed-off-by: Simon Marchi <simon.marchi@ericsson.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: makefile version checks with single "ge"

Version checks in makefiles should always be a disjunctive normal form
where the conjunctions consist of one or more "equals" comparisons and
at most a single greater-or-equal comparison.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Performance: special-case NULL in lttng_strlen_user_inatomic

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: lttng_inline_memcpy does not take a __user argument

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Performance: implement lttng_inline_memcpy

Because all length parameters received for serializing data coming from
applications go through a callback, they are never constant, and it
hurts performance to perform a call to memcpy each time.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Performance: cache the backend pages pointer in context

Getting the backend pages pointer requires pointer chasing through the
ring buffer backend tables. Cache the current value so it can be re-used
for all backend write operations writing fields for the same event.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: libringbuffer: remove duplicate pointer chasing in slow paths

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Performance: Only dereference commit index once

The commit fast path should not dereference the commit counter index
repeatedly for performance reasons.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Performance: Mark channel and buffer event disabled check unlikely

Channel and buffer are typically enabled when tracing, else the probes
would not be connected. Ensure the compiler optimizes for the right
usual case.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Performance: Relax atomicity constraints for crash handling

Use a store rather than a cmpxchg() for the update of the
sequential commit counter. This speeds up commit. The downside
is that short race windows between the if() check to see if the
counter is larger than the new value and the update could result
in the counter going backwards, in unlikely preemption or signal
delivery scenarios.

Accept that we may lose a few events in a crash dump for the
benefit of tracing speed.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Performance: mark ring buffer do_copy callers always inline

The underlying copy operation is more efficient if the size is a
constant, which only happens if this function is inlined in the caller.
Otherwise, we end up calling memcpy for each field.

Force inlining for performance reasons for:
  - lib_ring_buffer_do_strcpy,
  - lib_ring_buffer_do_strcpy_from_user_inatomic,
  - lib_ring_buffer_copy_from_user_inatomic.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Performance: mark lib_ring_buffer_write always inline

The underlying copy operation is more efficient if the size is a
constant, which only happens if this function is inlined in the caller.
Otherwise, we end up calling memcpy for each field.

Force inlining for performance reasons.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Kconfig: select TRACEPOINTS when built-in

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Performance: disable event counting by default

Performance improvement changelog from lttng-ust, ported back to
lttng-modules:

Disable event counting in the ring buffer, which can count the number of
events produced per ring-buffer, as well as the number of events
overwritten in overwrite mode.

This feature is currently unused anyway: it is not saved in the ring
buffer header, nor made available to lttng-tools.

This saves 70 ns/event in lttng-ust on the ARM32 Cubietruck.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: handle large number of pages or subbuffers per buffer

Do no trigger kernel console warnings when we try to allocate too many
pages, or a too large kmalloc area for page array (within a subbuffer),
or a sub-buffer array (within a buffer).

Use vmalloc/vfree for the "pages" local variable used only during
allocation, which is an array of nr_subbuf * nr_pages_per_subbuf
pointers. This ensures we do not limit the overall buffer size due to
kmalloc limitations.

Fixes #1031

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: unregister cpu hotplug notifier on buffer alloc error

The cpu hotplug notifier needs to be unregistered in the error path of
buffer allocation, else it eventually causes kernel OOPS when the kernel
accesses freed memory of the notifier block.

Fixes #1031

Signed-off-by: Jonathan Rajotte <jonathan.rajotte-julien@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: sa_family is of type unsigned short

sa_family is an unsigned short in sockaddr definitions. For instance,
the kernel's unix_getname() function sets addrlen to sizeof(short) as it
only returns the socket's family.

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

Fix: check for sizeof sa_family to save sa_family in accept and connect

The check of addrlen >= sizeof(struct sockaddr) is too restrictive
and causes sa_family to not be saved in the case of AF_UNIX sockets
as the addrlen returned by the syscall may be only sizeof(short).

Individual checks per socket family are performed anyhow in the
switch case, making this safe.

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

Fix: use printk_once() for wrapper warning messages

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: btrfs instrumentation for 4.8 kernel

Note: there are missing events.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: update mm_vmscan instrumentation for kernel 4.8

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: napi instrumentation for 4.8 kernels

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: update block instrumentation to compile on 4.8 kernels

These 2 upstream Linux commits break the block instrumentation in lttng
modules master:

https://github.com/torvalds/linux/commit/4e1b2d52a80d79296a5d899d73249748dea71a53

https://github.com/torvalds/linux/commit/288dab8a35a0bde426a09870943c8d3ee3a50dab

The errors are :

lttng-modules/probes/../instrumentation/events/lttng-module/block.h:35:13:
error: ‘REQ_DISCARD’ undeclared (first use in this function)

lttng-modules/probes/../instrumentation/events/lttng-module/block.h:41:14:
error: ‘REQ_SECURE’ undeclared (first use in this function)

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: reuse code in accept() and accept4() instrumentation

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

Add x86-64 override for accept4 syscall

This patch adds an instrumentation override for the accept4() syscall
which is almost identical to accept(), except for an additional
"flags" parameter.

A follow-up patch refactors both overrides to minimize code
duplication as is done for the select/pselect6 overrides.

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

Fix: timer wrapper: support kernels >= 4.8

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>