Fix: ext4 instrumentation for 4.2 kernels

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

Fix: kvm instrumentation for 4.2 kernels

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

Fix: mm_page_alloc_extfrag instrumentation for kernel 3.18.10

The following commit has been backported from kernel 3.19.2 into
3.18.10, introducing a change to mm_page_alloc_extfrag and therefore
provoking a build failure on a 3.18.10 kernel:

commit a4f3f96f850db69890f47984c60606a2362bee89
Author: Vlastimil Babka <vbabka@suse.cz>
Date:   Wed Feb 11 15:28:15 2015 -0800

    mm: when stealing freepages, also take pages created by splitting buddy page

    commit 99592d598eca62bdbbf62b59941c189176dfc614 upstream.

This patch adds the proper version check to allow building on 3.18.y.

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

Fix: allow LTTng to be built within kernel tree

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

Version 2.7.0-rc1

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

Fix: Building the event list fails on fragmented memory

On a small arm imx6 solo with 256MB RAM it often happens that memory
becomes fragmented rather quickly, so that kmalloc will not be able to
get enough consecutive pages (enocuntered for example if you enable all
kernel events: lttng enable-event -k syscall --all).

This patch switches the allocation to vmalloc. Tested for x86 on Ubuntu
12.04 Lts and on imx6 solo 256MB RAM

If this patch is not applied, you can identify low and/or fragmented
memory failures by looking at the kernel ring buffer (please ignore DMA,
it is due to some memory setup misconfiguration, should read Normal):

...
[  321.993820] lttng-sessiond: page allocation failure: order:4, mode:0x1040d0
...
[  321.994711] lowmem_reserve[]: 0 0 0
[  321.994727] DMA: 801*4kB (UEMC) 424*8kB (EMC) 355*16kB (UEMC) 344*32kB (MC) 340*64kB (C) 8*128kB (C) 0*256kB 0*512kB 0*1024kB 0*2048kB 0*4096kB 0*8192kB 0*16384kB 0*32768kB = 46068kB

[ Edit by Mathieu: use vzalloc() rather than vmalloc() + memset 0. ]

Signed-off-by: Martin Leisener <martin@leisener.de>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: use after free on metadata cache reallocation

When the metadata cache is expanded (reallocated) by
lttng_metadata_printf(), the metadata cache reader
(lttng_metadata_output_channel()) may use freed memory, because the
metadata cache is not protected from concurrent read accesses. The
metadata cache updates are protected from each other by the sessions
mutex, but metadata cache reads do not hold the sessions mutex.
Actually, the comment on top of lttng_metadata_output_channel() stating
"We have exclusive access to our metadata buffer (protected by the
sessions_mutex)" is simply wrong, because this mutex is never held when
calling lttng_metadata_output_channel().

Promote the per-stream lock to the metadata cache used by each of those
metadata streams, thus ensuring mutual exclusion between metadata cache
reallocation and readers.

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

Fix: version checks for kernel 4.0

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

Fix: Incorrect Ubuntu kernel patchlevel

The kernel version used should be the one in Makefile and generated
version.h, not the one tweaked by Ubuntu as "0" in the built package.

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

Fix: cpu-id context should use int type

cycles_t makes no sense as type for cpu-id context.

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

Implement cpu-id context

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

Implement cpu_id context for filtering

cpu_id context can now be used for filtering, e.g.:

lttng enable-event -k sched_switch --filter '$ctx.cpu_id==1'

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

Fix: context printk mismatch on error

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

Cleanup: misleading comment about deferrable timer

Timer in lttng-modules are not deferrable, because we cannot hook into
the Linux kernel CPU idle events, because no notifiers are exposed for
those.

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

Fix: arm-32 build

The current master branch of lttng-modules doesn't seem to compile for
arm-32 builds.

The patch below repairs the build for arm (and still builds on a x86)
(tested on Ubuntu 12.04 LTS x86 + imx6 solo)

Signed-off-by: Martin Leisener <martin@leisener.de>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: Update kmem event for Ubuntu's 3.16 kernel

The upstream patch in question in commit 72906e1 was cherry-picked into
Ubuntu's 3.16 kernel backport branch (called lts-utopic). The version
check should also include this branch.

The other backport branch, lts-vivid, and all following ones will be
covered by the standard

  LINUX_VERSION_CODE >= KERNEL_VERSION(3,19,2)

check, so should not require other workarounds.

References #889.

Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: RHEL 7.1 block instrumentation

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

Fix: add RHEL version macros

RHEL 7.1 has backported upstream patches that change the kvm.h
tracepoint prototype. Add version check specific to RHEL to handle this
case.

RHEL 7.x has the required fixes backported that allow removing the 3.10
blacklist check for them, so take this into account.

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

Fix: add missing parenthesis in ubuntu version check

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

Fix: remove regmap instrumentation for kernels < 4.1

Modifications to regmap.h instrumentation in mainline Linux kernel has
been no less than erratic. First, the public instrumentation header
regmap.h include a private driver header in Linux 4.0. Then, regmap.h
is moved to the private driver directory in Linux 4.1. To make things
worse, the 4.0 commit has been picked into stable branches of the Linux
kernel.

Since this does not appear to be an instrumentation of utmost
importance, only build this instrumentation probe if the private header
is found in the driver directory (need full kernel sources).

This removes regmap instrumentation for older kernels.

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

Fix: Update kmem event for recent Ubuntu kernel

Ubuntu backported a patch into their 3.13 (Trusty) branch which
changes kmem event definition.

Upstream commit
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=99592d598eca62bdbbf62b59941c189176dfc614
first appears in 4.0 (and was later backported to 3.19.2).
We already have the proper check for this.

This corresponds to Ubuntu commit
http://kernel.ubuntu.com/git/ubuntu/ubuntu-trusty.git/commit/?id=d938e4bff1089a1f4757ed94a5cace803c6a363b
which was first released with 3.13.0-50.82.
This required adding a new check specifically for the
Ubuntu version.

Signed-off-by: Alexandre Montplaisir <alexmonthy@voxpopuli.im>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: missing statedump end event when block device listing available

Incorrect handling of lttng_enumerate_block_devices() return value of 0
(which means "ok"), causing do_lttng_statedump() to return early.

To causes missing "lttng_statedump_end" in traces where block device
enumeration is available (CONFIG_KALLSYMS=y).

Reported-by: Antoine Busque <abusque@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: don't generate incorrect macros for debian/ubuntu version check

When the debian/ubuntu macros are unset, we end up generating macros
that have no right hand operand to "+", which seems to break compilation
on some compilers, even though it will always be optimized away.

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

Revert "Fix: build against Ubuntu kernels"

This reverts commit a748963b76f353d459a3faf85eecc96f52811eee.

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

Fix: build against Ubuntu kernels

Recent Ubuntu kernels appear to define DEBIAN_API_VERSION, but they are
not really a Debian kernel per-se: the abi-debian-version.sh cannot find
the expected Debian kernel version string layout.

Therefore, change the conditional check to use directly the
LTTNG_DEBIAN_VERSION_CODE define, which is defined only when the
abi-debian-version.sh script figures out it builds against a Debian
kernel.

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

Add namespace info in sched_fork and statedump

Modifications to sched_fork : Add a field for parent and child pid
namespace inode (as in /proc/$PID/ns/pid). Also add a field for the
child VTID.

Modifications to lttng_statedump_process_state : Add a field for the pid
namespace inode of the task.

Signed-off-by: Sebastien Lorrain <sebastien.lorrain01@gmail.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: ext4 instrumentation: flags parameter did not exist prior to 3.11

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

Fix: printk instrumentation: remove "condition" tracepoints

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

Fix: regmap build against kernel 4.0 kernel headers

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

Add mmu_valid_gen field to mmutrace events

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

Fix: add missing x86 mmutrace kernel header include

Needed to ensure the compiler checks prototypes of lttng-modules probe
against the kernel prototypes.

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

Fix: x86 mmutrace event extra field for kernels >= 3.11

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

Fix: remove execute flag from kvm.h file

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

Fix: build x86 kvm probes on kernels >= 4.1-rc1

iodev.h has moved to include/kvm/.

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

Fix: compilation on 4.1-rc1 kernel, remove unused WB_WORK_REASON

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

Fix: 4.1-rc1 Linux build and regmap instrumentation

regmap header has been privatized.

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

Fix: don't clash with older Debian kernel versions

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

Fix: update get_pfnblock_flags_mask wrapper for debian kernels

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

Fix: discover Debian API

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

Fix: Update README.md to match reality: kernels >= 2.6.36 are supported

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

Fix: atomic.h wrapper for kernels < 2.6.37

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

Fix: disable x86 kvm instrumentation for kernels < 2.6.38

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

Fix: add missing types.h include for older kernels

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

Fix: add hlist lttng wrappers

For older kernels (e.g. 2.6.38).

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

Fix: add struct file_handle forward declaration

For older kernels (e.g. 2.6.38).

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

Fix: string length calculation reversed user vs kernel args

Only affects feature recently merged into master.

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

Fix: use lttng_rcu_dereference within lttng-syscalls.c

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

Fix: add lttng rcu wrapper to allow tracing RCU

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

Fix: double-unlock of session mutex

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

Fix: check that class and instance prototype match

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

Fix: lttng-events.h check function takes void

The kernel nerver really emit this, but our function prototype should
not take the void *data parameter, so we are similar to the the "with
argument" prototype, which does not have the private data pointer.

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

Fix: regmap instrumentation for Linux 3.19.4 and 4.0

This is necessary after commit c6b570d97c0e77f570bb6b2ed30d372b2b1e9aae
in the kernel.

[ Edit by Mathieu Desnoyers: fix DECLARE_EVENT_CLASS regmap_reg too. ]

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

Fix: kernels >= 2.6.33 put utsrelease.h in generated/

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

Fix: Include utsrelease.h in lttng-kernel-version.h

Including module.h includes other headers, which is an issue for the
get_pfnblock_flags_mask wrapper.

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

Fix: ext4 ext4_da_write_pages instrumentation for kernels < 3.11.0

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

Fix: old kernels need module.h included before vermagic.h

This is at least true for 3.2.

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

Fix: net instrumentation for kernels 2.6.39 and below

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

Fix: version condition in makefile

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

Implement wrapper around get_pfnblock_flags_mask

This is used by kmem instrumentation, and changed in 4.0.0. This change
has been backported to 3.19.2 stable kernel.

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

Be more restrictive in provider name check

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

Fix: net instrumentation namespacing

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

Fix: compaction instrumentation namespacing

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

Fix: Support 4.x kernel versions in Makefile

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

Fix: remove stale TP_STRUCT_entry from kvm instrumentation

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

Fix: add missing module.h include to lttng-probe-user.c

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

Fix: use lttng hlist wrapper in wildcard and filter

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

Fix: namespace asoc and kmem instrumentation

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

Fix: update instrumentation after wildcard rebase

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

Fix: lttng_abi_syscall_list does not need to be exported

Fix rebase of wildcard feature.

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

Migrate ARM, powerpc, MIPS syscall instrumentation to TP_FIELDS

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

Remove now unused syscalls.h from lttng-modules

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

Filter: start with enabler "disabled"

Since we can create the enabler, then attach the filter, and then enable
it, ensure it is in a "disabled" state initially.

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

Fix coverity warning about sizeof(void **) != sizeof(void *)

The size of the copied data is actually void *, not void **.

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

Fix length type and text sequence base

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

Fix: TP_FIELDS: add missing headers

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

Migrate syscall instrumentation to TP_FIELDS

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

Migrate tracepoint instrumentation to TP_FIELDS

At this stage syscall instrumentation is disabled. Will come in a
following commit.

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

Implement filtering infrastructure

Implement filtering infrastructure in the enablers. It is not yet active
in the tracepoint probes, since we need to modify the
LTTNG_TRACEPOINT_EVENT() to follow lttng-ust more closely beforehand.

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

syscall wildcards: apply syscall filtering

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

Implement syscall wildcard support

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

Implement tracepoint wildcard support

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

Namespace all tracepoints

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

Implement list.h wrapper for hlist in older kernels

The hlist API has changed between kernels 3.2 and recent (e.g. 3.19.2)
kernels. Implement a wrapper header to implement the macros missing from
older kernels that we support.

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

Implement listing of pid tracker content

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

Implement PID tracking

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

Fix: mm_page_alloc_extfrag instrumentation for kernel 3.14.36

The following commit has been backported from kernel 3.19.2 into
3.14.36, introducing a change to mm_page_alloc_extfrag and therefore
provoking a build failure on a 3.14.36 kernel:

commit a4f3f96f850db69890f47984c60606a2362bee89
Author: Vlastimil Babka <vbabka@suse.cz>
Date:   Wed Feb 11 15:28:15 2015 -0800

    mm: when stealing freepages, also take pages created by splitting buddy page

    commit 99592d598eca62bdbbf62b59941c189176dfc614 upstream.

This patch adds the proper version check to allow building on 3.14.y.

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

Fix: update writeback instrumentation for kernel 4.0+

Introduced by Linux kernel commit:

commit 0ae45f63d4ef8d8eeec49c7d8b44a1775fff13e8
Author: Theodore Ts'o <tytso@mit.edu>
Date:   Mon Feb 2 00:37:00 2015 -0500

    vfs: add support for a lazytime mount option

And:

commit de1414a654e66b81b5348dbc5259ecf2fb61655e
Author: Christoph Hellwig <hch@lst.de>
Date:   Wed Jan 14 10:42:36 2015 +0100

    fs: export inode_to_bdi and use it in favor of mapping->backing_dev_info

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

Fix: compaction mm_compaction_isolate_template update for Linux 4.0+

Updated by Linux commit:

commit b7aba6984dc048503b69c2a885098cdd430832bf
Author: Mel Gorman <mel@csn.ul.ie>
Date:   Thu Jan 13 15:45:54 2011 -0800

    mm: compaction: add trace events for memory compaction activity

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

Fix: kmem instrumentation update for mm_page_alloc_extfrag kernel 3.19.2+

This commit introduces an API change for mm_page_alloc_extfrag
instrumentation:

commit cdf476685b5c9dcd20f0305791bd60212cace25d
Author: Vlastimil Babka <vbabka@suse.cz>
Date:   Wed Feb 11 15:28:15 2015 -0800

    mm: when stealing freepages, also take pages created by splitting buddy page

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

Fix: kmem instrumentation: mm_page_alloc extfrag change_ownership field

Linux kernel 3.12.30 and onward changes the parameter received by this
callback. It's the same type, but the meaning differs. This is why it
did compile fine, but the data gathered did not really represent
change_ownership.

This change has been introduced in the Linux kernel by this commit:

commit 52c8f6a5aeb0bdd396849ecaa72d96f8175528f5
Author: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Date:   Tue Nov 12 15:08:19 2013 -0800

    mm: get rid of unnecessary overhead of trace_mm_page_alloc_extfrag()

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

Fix: missing parenthesis in offset_align_floor

Is currently unused. Triggers a compile error when used.

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

Fix: implement time.h wrapper for FD_ISSET

Kernels v3.4.0 to v3.4.7 still define FD_ISSET, but __FD_ISSET has been
removed. Therefore, we cannot use the FD_ISSET define to find out of we
need to use the new close_on_exec(). Implement a wrapper based on kernel
version detection instead.

Fixes #872

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

Fix: update sched instrumentation for 3.19+ kernels

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

Fix: update rcu instrumentation for 3.19+ kernels

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

Fix: update scsi instrumentation for 3.19+ kernels

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

Fix: update module instrumentation for 3.19+ kernels

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

Fix: Add f_dentry wrapper for 3.19+ kernels

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

Fix: introduce lttng_get_unused_fd() wrapper for 3.19 kernels

get_unused_fd disappears in 3.19 kernel, where get_unused_fd_flags(0)
should be used instead.

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

Fix: lttng_this_cpu_ptr wrapper for kernel 3.19+

Starting from kernel 3.19-rc1, __get_cpu_var() disappears, replaced by
this_cpu_ptr(). Introduce a wrapper using either depending on the kernel
version. Older kernels did not expose a this_cpu_ptr(), so keep on using
__get_cpu_var() on old kernels.

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

Fix: Update btrfs instrumentation for 3.18 Linux kernel

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