| Age | Commit message (Collapse) | Author |
|---|
|
commit 3b3f7e785e279 [kernel: Rearrange for 1.8] began the process of
moving the kernel source and build artefacts out of sstate control and
into a shared location.
This changed triggered some workflow issues, as well as bugs related
to the kernel source containing build output, and hence being dirty and
breaking kernel rebuilds.
To solve these issues, and to make it clear that the kernel is not under
sstate control, we move the source and build outputs to:
work-shared/MACHINE/kernel-source
work-shared/MACHINE/kernel-build-artifacts
Where kernel-build-artifacts is the kernel build output and
kernel-source is kept "pristine". The build-artifacts contain everything
that is required to build external modules against the kernel source,
and includes the defconfig, the kernel-abiversion, System.map files and
output from "make scripts".
External module builds should either pass O= on the command line, or
set KBUILD_OUTPUT to point to the build-artifacts. module-base.bbclass
takes care of setting KBUILD_OUTPUT, so most existing external module
recipes are transparently adapted to the new source/build layout.
recipes that depend on the kernel source must have a depedency on the
do_shared_workdir task:
do_configure[depends] += "virtual/kernel:do_shared_workdir"
With this dependency added, the STAGING_KERNEL_DIR will be populated and
available to the rest of the build.
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
To ensure that the minimal environment has been created for out of
tree module builds, perf and lttng should depend on the do_install
task of the kernel.
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|
|
In 1.8 we want to streamline the kernel build process. Basically we
currently have multiple copies of the kernel source floating around
and the copying/compression/decompression is painful.
Lets assume we have a kernel source per machine since in most cases
this is true (and we have a sysroot per machine anyway). Basically,
instead of extracting a source into WORKDIR, then copying to a sysroot,
we now set S to point straight at STAGING_DIR_KERNEL.
Anything using kernel source can then just point at it and use:
do_configure[depends] += "virtual/kernel:do_patch"
to depend on the kernel source being present. Note this is different
behaviour to DEPENDS += "virtual/kernel" which equates to
do_configure[depends] += "virtual/kernel:do_populate_sysroot".
Once we do this, we no longer need the copy operation in
do_populate_sysroot, in fact there is nothing to do there (yay).
The remaining part of the challenge is to kill off the horrible
do_install. This patch splits it off to a different class, the idea here
is to have a separate recipe which depends on the virtual/kernel:do_patch
and just installs and packages the source needed to build modules on
target into a specific package.
Right now this code is proof of concept. It builds kernels and kernel
modules. perf blows up in do_package with issues on finding the kernel
version which can probably be fixed by adding back the right bit of do_install,
and adding a dependency of do_package[depends] += "virtual/kernel:do_install"
to perf. The whole thing needs a good write up, the corner cases testing
and probably a good dose of cleanup to the remaining code.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
When cross compiling libunwind support for ARM a missing debug include means
that pr* macros are not expanded, and hence link failures on the undefined
functions.
Since we must be compatible with many versions of the kernel and perf, we
sed the proper include into the files, while the permanent fix goes upstream
to the mainline kernel.
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Currently, perf can't split to perf-archive, perf-tests, perf-python and
perf-perl. All files are included in perf package. Change the perfexecdir
variable to make split successfull. Add python to RDEPENDS_perf-tests.
Signed-off-by: Chong Lu <Chong.Lu@windriver.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
The powerpc perf arch port has two issues in the current 3.17-rc
seriers. Undefined dwarf symbols and undefined pr_debug calls.
commit a60335ba [perf tools powerpc: Adjust callchain based on DWARF debug info]
introduces skip-callchain-idx.c, and breaks the perf build.
Until fixes get into the mainline kernel, we temporarily fix the issues by
adding a forced linke to -ldw and modifying skip-callchain-idx.c to include
the proper debug files for pr_debug.
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
|
|
Add a new feature named 'perf-libunwind'. Adding this support to perf allows it
to do stack traces on ARM - thumb2 and MIPS targets. PERF_FEATURES variable in
perf-features.inc will enable the perf-libunwind.
Signed-off-by: Chong Lu <Chong.Lu@windriver.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Fixed:
$ bitbake perf lib32-perf
ERROR: QA Issue: lib32-perf: Files/directories were installed but not shipped
/usr/lib64
/usr/lib64/traceevent
/usr/lib64/traceevent/plugins
/usr/lib64/traceevent/plugins/plugin_function.so
/usr/lib64/traceevent/plugins/plugin_scsi.so
/usr/lib64/traceevent/plugins/plugin_hrtimer.so
/usr/lib64/traceevent/plugins/plugin_kmem.so
/usr/lib64/traceevent/plugins/plugin_jbd2.so
/usr/lib64/traceevent/plugins/plugin_xen.so
/usr/lib64/traceevent/plugins/plugin_mac80211.so
/usr/lib64/traceevent/plugins/plugin_cfg80211.so
/usr/lib64/traceevent/plugins/plugin_sched_switch.so
/usr/lib64/traceevent/plugins/plugin_kvm.so [installed-vs-shipped]
The perf.do_configure edits kernel's
${STAGING_KERNEL_DIR}/tools/perf/config/Makefile, there would be
problems since kernel doesn't have multilib, and the build result is
undetermined.
Previously, the sed command changed libdir to /usr/lib64 (or 32) in the
Makefile, so the build result was different if we build perf (64) first
or lib32-perf first.
Use the weak assignment "libdir ?=" to instead of "libdir =" will fix
the problem since the multilib builds are in different processes, and
they won't affect each other any more.
The sed command will match both $(prefix)/$(lib) and $(prefix)/lib since
the Makefile may has been modified before this patch.
Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
* multiline variables should use spaces not tabs for indentation
* do_configure is also using wrong indentation, but I'm not fixing
this one (still hoping that we'll eventually fix styleguide to
use 4 spaces everywhere)
Signed-off-by: Martin Jansa <Martin.Jansa@gmail.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
* it's not included in DEPENDS and it's detected from sysroot
Signed-off-by: Martin Jansa <Martin.Jansa@gmail.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
definition of __SANE_USERSPACE_TYPES__ was added in kernel source by
commit e3541ec75219819d3235f80125a1a75d798ff6e1
perf tools, powerpc: Fix compile warnings in tests/attr.c
Overriding the CFLAGS makes it impossible for kernel build system to
append to it, thus making the build fail in various ways as:
| CC /.../perf/1.0-r9/perf-1.0/perf.o
| In file included from builtin.h:4:0,
| from perf.c:9:
| util/util.h:74:24: fatal error: lk/debugfs.h: No such file or directory
| #include <lk/debugfs.h>
| ^
| compilation terminated.
Signed-off-by: Ting Liu <ting.liu@freescale.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
* TUI/GUI support was added in 2.6.35 based on libnewt
* since 3.10 slang replaced libnewt completly
* changing TUI_DEFINES is not necessary, because NO_NEWT is
still respected with newer kernels
* add comment about the gui history to the recipe
The patch was sponsored by sysmocom
Signed-off-by: Henning Heinold <henning@itconsulting-heinold.de>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
* some fundamental perf commands can work
without the dependency on perl, python or bash
make them separate packages and RSUGGEST them
* bump PR
The patch was sponsored by sysmocom
Signed-off-by: Henning Heinold <henning@itconsulting-heinold.de>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The patch was sponsored by sysmocom
Signed-off-by: Henning Heinold <henning@itconsulting-heinold.de>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
If you build perf in tree /xxx/treea, then cleansstate perf and build it
in /xxx/treeb having deleted treea, the build will fail, unable to find libc.
The problem is that the --sysroot option passed in through CC is missing.
This works fine if the default sysroot is ok, if it isn't, things will fail.
In 1.7 we'll start poisoning the default sysroot in gcc to catch this kind of
issue however that doesn't fix the problem with perf.
The problem is that various Makefiles set CC = $(CROSS_COMPILE)gcc. The
easist fix for now is to sed out the problematic Makefile lines. Its worth
noting the tools/lib/traceevent Makefile has a much more funky way of setting
CC which works for us and may be the way we need to fix the other Makefiles
upstream.
This fixes build failures we're occasionally seen on the autobuilders.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Recent versions of perf may install files into /usr/libexec/perf-core and in
/usr/lib/traceevent. To avoid packaging QA errors, we add these two
directories to the FILES variables.
We also add: INHIBIT_PACKAGE_DEBUG_SPLIT="1" to avoid the following issue
(due to a trailing / being removed):
ERROR: debugedit failed with exit code 256
...
debugedit: canonicalization unexpectedly shrank by one character
And finally, we must ensure that the traceevent libraries are installed to
the proper multilib library path. If building some multlibs, the incorrect
library path will be selected by perf, since it triggers via: ifeq
($(ARCH),x86_64) (or similiar mechanism per arch).
In a 32 bit build, with a 64 bit multilib, the arch won't match and the
detection of a 64 bit build (and library) are not exected. To ensure that
libraries are installed to the correct location, we can make the substitution
in the config/Makefile. For non multilib builds, this has no impact.
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Instead of using oe.path.relative, use the Python Standard Library function
os.path.relpath.
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
Turns out the sed command has not been working as intended since
kernel 3.1 due to the trailing space. Adding the WERROR=0 environment
variable is the correct way to disable warnings as errors.
Signed-off-by: Konrad Scherer <Konrad.Scherer@windriver.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
As the same reason to powerpc64, mips64 also need the flag.
Signed-off-by: Wenzong Fan <wenzong.fan@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
PPC64 uses long long for u64 in the kernel, but powerpc's asm/types.h
prevents 64-bit userland from seeing this definition, instead defaulting
to u64 == long in userspace.
Perf want LL64, flag __SANE_USERSPACE_TYPES__ to get int-ll64.h.
Fix the below issue:
| tests/attr.c:71:4: error: format '%llu' expects argument of type 'long
long unsigned int', but argument 6 has type '__u64' [-Werror=format=]
| tests/attr.c:80:7: error: format '%llu' expects argument of type 'long
long unsigned int', but argument 4 has type '__u64' [-Werror=format=]
| attr->type, attr->config, fd) < 0) {
| ^
Signed-off-by: Ting Liu <b28495@freescale.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
The LDFLAGS is required or some old kernels fails due missing
symbols and this is preferred than requiring patches to every old
supported kernel.
Fixes [YOCTO: #5221]
Signed-off-by: Otavio Salvador <otavio@ossystems.com.br>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The kernel build system does the right thing here and we should stop
overriding it. This code has been added based on a change from
'meta-metro' layer, revision 9d698004137c1a888d40d6a4808d94afa22387e7,
without any information about what problem it fixes so I am reverting
it.
Using the CFLAGS and LDFLAGS makes it impossible for kernel build
system to append to it, thus making the build fail in various ways as:
| CC /.../perf/1.0-r8/perf-1.0/perf.o
| In file included from builtin.h:4:0,
| from perf.c:9:
| util/util.h:74:24: fatal error: lk/debugfs.h: No such file or directory
| #include <lk/debugfs.h>
| ^
| compilation terminated.
The unset is done in do_compile and do_install otherwise it /rebuild/
perf as it detects the compiler options has change.
Signed-off-by: Otavio Salvador <otavio@ossystems.com.br>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
In perf.bb:
S = "${STAGING_KERNEL_DIR}"
So the source should be ready after the do_unpack, and we need this:
do_unpack[depends] += "virtual/kernel:do_populate_sysroot"
Otherwise, maybe no source after do_unpack.
[YOCTO #5168]
Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Without this we see relocation errors on mips with 3.10. This should be
safe to be included in general.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The do_populate_lic task has a race with the recipe since it relies on the
kernel being populated in the sysroot. This patch adds in the explicit missing
dependency.
[YOCTO #3534]
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Due to the use of ${@...} the code will try and expand this when
performing the initial parsing. If the sysroot doesn't exist with
an existing kernel, this will fail at parsing time.
Sinec we're already in python, just remove the ${@....} wrapping
and then we execute at do_package time which is what we want.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Have perf grab and use the kernel version it's built from for PKGV,
rather than the default perf recipe version, so the final packages get
the kernel version instead of the default 1.0, which represents a
backwards value from the previous recipe.
Signed-off-by: Tom Zanussi <tom.zanussi@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
There's nothing kernel-version-specific about the perf_3.4 recipe, so
it's actually misnamed and misleading now that it also gets used with
the 3.8 kernel.
Since the recipe isn't tied to a specific PV, and simply uses
whatever's in STAGING_KERNEL_DIR, there's no reason to add anything
else either to the bare PN, so just use that as the recipe name.
Signed-off-by: Tom Zanussi <tom.zanussi@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The contents of perf.inc are really specific to perf features and
shouldn't use the generic perf.inc name, which implies common recipe
code. It's always confusing to open up this file and find out that's
not what it is.
Signed-off-by: Tom Zanussi <tom.zanussi@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Autobuilder builds periodically and now more frequently have been
failing because of a race between the perf build and the newly
separated libtraceevent - perf tries to link libtraceevent.a, which
hasn't finished building yet in those cases.
This disables the parallel build to prevent that.
Signed-off-by: Tom Zanussi <tom.zanussi@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
There are a number of scripts in the perf installation that use bash,
so we need to add a run-time dependency on bash for them. If not, we
can generate build errors like "no package provides /bin/bash".
Fixes [YOCTO #3951].
Signed-off-by: Tom Zanussi <tom.zanussi@linux.intel.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
Signed-off-by: Christopher Larson <chris_larson@mentor.com>
the patch was imported from meta-mentor layer on yoctoproject git server
http://git.yoctoproject.org/cgit/cgit.cgi/meta-mentor as of commit id
71748b54694f4ffe2d598da71f641969df1417c0
slightly modified the patch to apply it on .bb file instead of .bbappend
Signed-off-by: Fahad Usman <fahad_usman@mentor.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
Signed-off-by: Christopher Larson <chris_larson@mentor.com>
the patch was imported from meta-mentor layer on yoctoproject git server
http://git.yoctoproject.org/cgit/cgit.cgi/meta-mentor as of commit id
9d698004137c1a888d40d6a4808d94afa22387e7
Signed-off-by: Fahad Usman <fahad_usman@mentor.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
this fails to build using a recent sourcery toolchain due to
unused-but-set-variable
Signed-off-by: Christopher Larson <chris_larson@mentor.com>
the patch was imported from meta-mentor layer on yoctoproject git server
http://git.yoctoproject.org/cgit/cgit.cgi/meta-mentor as of commit ids
82e96b3baa1c64d03412871fce56d496a338f167 and
ae325d011bd50501fe677c8b37295ae83030c526
Signed-off-by: Fahad Usman <fahad_usman@mentor.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
This allows the files installed into /usr/libexec to be
relocated to ${libexecdir}. removed unneded prefix=/usr,
which would prevent ${prefix} relocation.
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
perf depends on bison and flex for event parsing - add them as
dependencies.
Signed-off-by: Tom Zanussi <tom.zanussi@intel.com>
|
|
Removes a make install-python_ext when not present since
older versions of perf lack this install rule
This also fixes a library issue on older kernels building with
a newer toolchain where libaries that would previously be pulled
in are no longer. So we add them manually.
Signed-off-by: Matthew McClintock <msm@freescale.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
Add pythonnative to the inherits list
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
perf has perl and python compile-time dependendencies, add them.
Signed-off-by: Tom Zanussi <tom.zanussi@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Add a new feature named 'perf-tui'. Adding this into the
PERF_FEATURES variable in perf.inc will enable the perf TUI (Text-base
UI) user interface on a target, which adds libnewt and turns on the
perf text UI options in perf, if perf is included in an image.
If 'perf-tui' isn't named as a feature (the default), the perf TUI
will be disabled and unavailable.
Signed-off-by: Tom Zanussi <tom.zanussi@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Add a new feature named 'perf-scripting'. Adding this into the
PERF_FEATURES variable in perf.inc will enable perf scripting on a
target, which will turn on all the language bindings currently
available in perf (Perl and Python), if perf is included in an image.
If 'perf-scripting' isn't named as a feature (the default), all perf
language bindings will be disabled and unavailable.
Signed-off-by: Tom Zanussi <tom.zanussi@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Add a perf.inc to contain utility functions and definitions and to
avoid cluttering up the main recipe.
Signed-off-by: Tom Zanussi <tom.zanussi@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
This shouldn't be unconditional - a later patch made it so, but that's
not yet pulled in. In the meantime, to fix build failures remove the
unconditional install.
Signed-off-by: Tom Zanussi <tom.zanussi@intel.com>
|
|
libexec/perf-core contains all the pre-canned scripts and modules
needed by both the Perl and Python bindigs. Add libexec/perf-core
along with all the pre-defined perf scripts underneath it.
Signed-off-by: Tom Zanussi <tom.zanussi@intel.com>
|
|
Add support to enable the perf Perl binding.
The build depends on perl-native to retrieve the configuration
settings needed for the binding. cpan-base adds some useful functions
like is_target() and get_perl_version() that we need for
PERLCONFIGTARGET and related settings, which allow us to use the
target's Config_heavy.pl settings for ExtUtils:Embed when building for
the target.
Also adds the perl-modules dependency to give the target the perl
modules that scripts using the binding need.
Signed-off-by: Tom Zanussi <tom.zanussi@intel.com>
|
|
Add support to enable the perf python bindings. The combination of
these changes and the changes in the python-config sections in the
kernel Makefile enable all the python bindings currently available in
perf.
Signed-off-by: Tom Zanussi <tom.zanussi@intel.com>
|
|
perf has been coupled to the kernel packages via kernel.bbclass.
While maintaining the build of perf out of the kernel source tree
is desired the package coupling has proved to be awkward in
several situations such as:
- when a kernel recipe doesn't want to build/provide perf
- when licensing of dependencies would prohibit perf and hence
the kernel from being built.
To solve some of these problems, this recipe is the extraction of
the linux-tools.inc provided perf compilation into a standalone
perf recipe that builds out of the kernel source, but is otherwise
independent.
No new functionality is provided above what the linux-tools.inc
variant provided, but the separate recipe provides baseline for
adding new functionality.
Signed-off-by: Liang Li <liang.li@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
Bruce Ashfield
Richard Purdie
Chong Lu
Robert Yang
Martin Jansa
Ting Liu
Henning Heinold
Ross Burton
Konrad Scherer
Wenzong Fan
Ting Liu
Otavio Salvador
Tom Zanussi
Fahad Usman
Saul Wold
Tom Zanussi
Matthew McClintock
Morgan Little
Liang Li