[PATCH 0/4] trace-cmd: Fixes for packaging

[PATCH 0/4] trace-cmd: Fixes for packaging

[Patrick McLean]

These are various patches to trace-cmd I needed to use to package both
it and kernelshark on Gentoo. Some of these may not be needed by other
distros, as these are at least partially to help with building them
as completely separate packages.

[PATCH 1/4] trace-cmd: Allow overriding of python installation directory

[Patrick McLean]

From: Patrick McLean <patrick.mclean@sony.com>

The current Makefile hardcodes the python module install directory, so
packages can't put them in their distro's perferred location. This
allows the directory to be overriden on the command line.

Signed-off-by: Patrick McLean <patrick.mclean@sony.com>
---
 Makefile | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/Makefile b/Makefile
index ad74a96..3579f27 100644
--- a/Makefile
+++ b/Makefile
@@ -64,7 +64,7 @@ python_dir = $(HOME)/.trace-cmd/python
 var_dir = $(HOME)/.trace-cmd/
 else
 plugin_dir = $(libdir)/trace-cmd/plugins
-python_dir = $(libdir)/trace-cmd/python
+python_dir ?= $(libdir)/trace-cmd/python
 PLUGIN_DIR = -DPLUGIN_DIR="$(plugin_dir)"
 PYTHON_DIR = -DPYTHON_DIR="$(python_dir)"
 PLUGIN_DIR_SQ = '$(subst ','\'',$(PLUGIN_DIR))'
-- 
2.22.0

[PATCH 2/4] trace-cmd: No automagic dependency on udis86

[Patrick McLean]

From: Patrick McLean <patrick.mclean@sony.com>

Currently if udis86 is detected on the system, trace-cmd automatically
uses it. This is generally a problem for packagers since if the build
machine happens to have udis86, now there is a dependency on it that may
or may not be tracked.

This adds a NO_UDIS86 variable that can be set by the packager to
disable using udis86 completely.

Signed-off-by: Patrick McLean <patrick.mclean@sony.com>
---
 Makefile | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/Makefile b/Makefile
index 3579f27..6b85e21 100644
--- a/Makefile
+++ b/Makefile
@@ -136,8 +136,13 @@ export NO_PYTHON
 test-build = $(if $(shell sh -c 'echo "$(1)" | \
 	$(CC) -o /dev/null -c -x c - > /dev/null 2>&1 && echo y'), $2)
 
+ifndef NO_UDIS86
 # have udis86 disassembler library?
-udis86-flags := $(call test-build,\#include <udis86.h>,-DHAVE_UDIS86 -ludis86)
+udis86-flags := -DHAVE_UDIS86
+udis86-ldflags := -ludis86
+else
+udis86-flags := -UHAVE_UDIS86
+endif # NO_UDIS86
 
 define BLK_TC_FLUSH_SOURCE
 #include <linux/blktrace_api.h>
@@ -237,6 +242,7 @@ endif
 # Append required CFLAGS
 override CFLAGS += $(INCLUDES) $(PLUGIN_DIR_SQ) $(VAR_DIR)
 override CFLAGS += $(udis86-flags) $(blk-flags)
+override LDFLAGS += $(udis86-ldflags)
 
 CMD_TARGETS = trace-cmd $(BUILD_PYTHON)
 
-- 
2.22.0

[PATCH 3/4] trace-cmd: Install all headers needed by kernelshark to proper paths

[Patrick McLean]

From: Patrick McLean <patrick.mclean@sony.com>

The trace-cmd build system installs headers for building kernelshark,
but it is missing trace-seq.h and trace-filter-hash.h.

It additionally installs event-parse.h to the toplevel "trace-cmd"
include directory, but the generated header includes it as
"traceevent/event-parse.h" so the kernelshark build fails.

This patch installs all the headers needed to build kernelshark to the
appropiate directories.

Signed-off-by: Patrick McLean <patrick.mclean@sony.com>
---
 Makefile | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/Makefile b/Makefile
index 6b85e21..3bab851 100644
--- a/Makefile
+++ b/Makefile
@@ -345,8 +345,10 @@ install_gui: install_cmd gui
 install_libs: libs
 	$(Q)$(call do_install,$(LIBTRACECMD_SHARED),$(libdir_SQ))
 	$(Q)$(call do_install,$(LIBTRACEEVENT_SHARED),$(libdir_SQ))
-	$(Q)$(call do_install,$(src)/include/traceevent/event-parse.h,$(includedir_SQ))
+	$(Q)$(call do_install,$(src)/include/traceevent/event-parse.h,$(includedir_SQ)/traceevent)
+	$(Q)$(call do_install,$(src)/include/traceevent/trace-seq.h,$(includedir_SQ)/traceevent)
 	$(Q)$(call do_install,$(src)/include/trace-cmd/trace-cmd.h,$(includedir_SQ))
+	$(Q)$(call do_install,$(src)/include/trace-cmd/trace-filter-hash.h,$(includedir_SQ))
 
 doc:
 	$(MAKE) -C $(src)/Documentation all
-- 
2.22.0

[PATCH 4/4] trace-cmd: Set SONAME on shared libraries

[Patrick McLean]

From: Patrick McLean <patrick.mclean@sony.com>

Currently trace-cmd installs some shared libraries, but does not set a
SONAME on these. This often violates distro policies, so it should at
least set something.

This patch sets it to the name of the file, which is somewhat naieve,
but works for many libraries. Generally it should be set to have an ABI
version as the extension, so when the developer breaks ABI, they can
change the soname.

Signed-off-by: Patrick McLean <patrick.mclean@sony.com>
---
 scripts/utils.mk | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/scripts/utils.mk b/scripts/utils.mk
index 260023a..9eb127d 100644
--- a/scripts/utils.mk
+++ b/scripts/utils.mk
@@ -54,7 +54,7 @@ do_build_static_lib =				\
 
 do_compile_shared_library =			\
 	($(print_shared_lib_compile)		\
-	$(CC) --shared $^ -o $@)
+	$(CC) --shared $^ -Wl,-soname,$@ -o $@)
 
 do_compile_plugin_obj =				\
 	($(print_plugin_obj_compile)		\
-- 
2.22.0

Re: [PATCH 2/4] trace-cmd: No automagic dependency on udis86

[Steven Rostedt]

Hi Patrick,

Thanks for sending these. Some questions below.

On Fri, 19 Jul 2019 20:03:42 -0700
Patrick McLean <chutzpah@gentoo.org> wrote:

> From: Patrick McLean <patrick.mclean@sony.com>
> 
> Currently if udis86 is detected on the system, trace-cmd automatically
> uses it. This is generally a problem for packagers since if the build
> machine happens to have udis86, now there is a dependency on it that may
> or may not be tracked.
> 
> This adds a NO_UDIS86 variable that can be set by the packager to
> disable using udis86 completely.
> 
> Signed-off-by: Patrick McLean <patrick.mclean@sony.com>
> ---
>  Makefile | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
> 
> diff --git a/Makefile b/Makefile
> index 3579f27..6b85e21 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -136,8 +136,13 @@ export NO_PYTHON
>  test-build = $(if $(shell sh -c 'echo "$(1)" | \
>  	$(CC) -o /dev/null -c -x c - > /dev/null 2>&1 && echo y'), $2)
>  
> +ifndef NO_UDIS86
>  # have udis86 disassembler library?
> -udis86-flags := $(call test-build,\#include <udis86.h>,-DHAVE_UDIS86 -ludis86)
> +udis86-flags := -DHAVE_UDIS86
> +udis86-ldflags := -ludis86

Can we keep the test here? If NO_UDIS86 is not defined. That is, if you
compile without this define, this change assumes you have it.

It's fine to force not having it, but we shouldn't force having it if
we don't.


> +else
> +udis86-flags := -UHAVE_UDIS86

I'm not sure this is needed, We shouldn't have anything defining
HAVE_UDIS86.

-- Steve


> +endif # NO_UDIS86
>  
>  define BLK_TC_FLUSH_SOURCE
>  #include <linux/blktrace_api.h>
> @@ -237,6 +242,7 @@ endif
>  # Append required CFLAGS
>  override CFLAGS += $(INCLUDES) $(PLUGIN_DIR_SQ) $(VAR_DIR)
>  override CFLAGS += $(udis86-flags) $(blk-flags)
> +override LDFLAGS += $(udis86-ldflags)
>  
>  CMD_TARGETS = trace-cmd $(BUILD_PYTHON)
>  

Re: [PATCH 2/4] trace-cmd: No automagic dependency on udis86

[Patrick McLean]

On Fri, 19 Jul 2019 23:17:01 -0400
Steven Rostedt <rostedt@goodmis.org> wrote:
> Hi Patrick,
> 
> Thanks for sending these. Some questions below.
> 
> On Fri, 19 Jul 2019 20:03:42 -0700
> Patrick McLean <chutzpah@gentoo.org> wrote:
> 
> > From: Patrick McLean <patrick.mclean@sony.com>
> > 
> > Currently if udis86 is detected on the system, trace-cmd
> > automatically uses it. This is generally a problem for packagers
> > since if the build machine happens to have udis86, now there is a
> > dependency on it that may or may not be tracked.
> > 
> > This adds a NO_UDIS86 variable that can be set by the packager to
> > disable using udis86 completely.
> > 
> > Signed-off-by: Patrick McLean <patrick.mclean@sony.com>
> > ---
> >  Makefile | 8 +++++++-
> >  1 file changed, 7 insertions(+), 1 deletion(-)
> > 
> > diff --git a/Makefile b/Makefile
> > index 3579f27..6b85e21 100644
> > --- a/Makefile
> > +++ b/Makefile
> > @@ -136,8 +136,13 @@ export NO_PYTHON
> >  test-build = $(if $(shell sh -c 'echo "$(1)" | \
> >  	$(CC) -o /dev/null -c -x c - > /dev/null 2>&1 && echo y'),
> > $2) 
> > +ifndef NO_UDIS86
> >  # have udis86 disassembler library?
> > -udis86-flags := $(call test-build,\#include
> > <udis86.h>,-DHAVE_UDIS86 -ludis86) +udis86-flags := -DHAVE_UDIS86
> > +udis86-ldflags := -ludis86  
> 
> Can we keep the test here? If NO_UDIS86 is not defined. That is, if
> you compile without this define, this change assumes you have it.
> 
> It's fine to force not having it, but we shouldn't force having it if
> we don't.

That sounds good to me, preserve the current behaviour if nothing is
defined. I will attach an updated patch.

> > +else
> > +udis86-flags := -UHAVE_UDIS86  
> 
> I'm not sure this is needed, We shouldn't have anything defining
> HAVE_UDIS86.

Sure, I will drop it.


From: Patrick McLean <patrick.mclean@sony.com>

Currently if udis86 is detected on the system, trace-cmd automatically
uses it. This is generally a problem for packagers since if the build
machine happens to have udis86, now there is a dependency on it that may
or may not be tracked.

This adds a NO_UDIS86 variable that can be set by the packager to
disable using udis86 completely.

Signed-off-by: Patrick McLean <patrick.mclean@sony.com>
---
 Makefile | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/Makefile b/Makefile
index 3579f27..ee3d8d9 100644
--- a/Makefile
+++ b/Makefile
@@ -136,8 +136,11 @@ export NO_PYTHON
 test-build = $(if $(shell sh -c 'echo "$(1)" | \
 	$(CC) -o /dev/null -c -x c - > /dev/null 2>&1 && echo y'), $2)
 
+ifndef NO_UDIS86
 # have udis86 disassembler library?
 udis86-flags := $(call test-build,\#include <udis86.h>,-DHAVE_UDIS86 -ludis86)
+udis86-ldflags := -ludis86
+endif # NO_UDIS86
 
 define BLK_TC_FLUSH_SOURCE
 #include <linux/blktrace_api.h>
@@ -237,6 +240,7 @@ endif
 # Append required CFLAGS
 override CFLAGS += $(INCLUDES) $(PLUGIN_DIR_SQ) $(VAR_DIR)
 override CFLAGS += $(udis86-flags) $(blk-flags)
+override LDFLAGS += $(udis86-ldflags)
 
 CMD_TARGETS = trace-cmd $(BUILD_PYTHON)
 
-- 
2.22.0