Fix oprofile build on powerpc
PATCH 1/1 Fix oprofile build on powerpc
Hi,
Since the upstep of oprofile to 0.9.8, the powerpc build was broken. This series fixes this problem.
As the last patch indicates, this is compile tested on x86 and powerpc, but not runtime-tested.
Best regards,
Thomas
PATCH 1/1 libpfm4: new package
--- package/Config.in | 1 + package/libpfm4/Config.in | 7 +++++++ package/libpfm4/libpfm4.mk | 27 +++++++++++++++++++++++++++ 3 files changed, 35 insertions(+), 0 deletions(-)
package/Config.in +1 -0
@@ -556,6 +556,7 @@ source "package/gsl/Config.in" source "package/libglib2/Config.in" source "package/libical/Config.in" source "package/libnspr/Config.in" +source "package/libpfm4/Config.in" source "package/libsigc/Config.in" source "package/libtasn1/Config.in" source "package/libtpl/Config.in"
package/libpfm4/Config.in new file +7 -0
@@ -0,0 +1,7 @@ +config BR2_PACKAGE_LIBPFM4 + bool "libpfm4" + help + A helper library to help encode Performance Events to use with + Operating system kernels performance monitoring interfaces. + + http://perfmon2.sourceforge.net/
package/libpfm4/libpfm4.mk new file +27 -0
@@ -0,0 +1,27 @@ +############################################################# +# +# libpfm4 +# +############################################################# +LIBPFM4_VERSION = 4.3.0 +LIBPFM4_SOURCE = libpfm-$(LIBPFM4_VERSION).tar.gz +LIBPFM4_SITE = http://downloads.sourceforge.net/project/perfmon2/libpfm4 +LIBPFM4_LICENSE = libpfm4 license +LIBPFM4_LICENSE_FILES = COPYING +LIBPFM4_INSTALL_STAGING = YES + +LIBPFM4_FLAGS = SYS=Linux ARCH=$(BR2_ARCH) CC="$(TARGET_CC)" LDCONFIG=true + +define LIBPFM4_BUILD_CMDS + $(MAKE) -C $(@D) $(LIBPFM4_FLAGS) +endef + +define LIBPFM4_INSTALL_STAGING_CMDS + make -C $(@D) $(LIBPFM4_FLAGS) PREFIX=$(STAGING_DIR)/usr install +endef + +define LIBPFM4_INSTALL_TARGET_CMDS + make -C $(@D) $(LIBPFM4_FLAGS) PREFIX=$(TARGET_DIR)/usr install +endef + +$(eval $(generic-package))
PATCH 1/1 libpfm4: new package
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
"Thomas" == Thomas De Schampheleire <[email protected]> writes:
Thomas> Signed-off-by: Thomas De Schampheleire <[email protected]>
Thomas> +++ b/package/libpfm4/libpfm4.mk
Thomas> @@ -0,0 +1,27 @@
Thomas> +############################################################# Thomas> +#
Thomas> +# libpfm4
Thomas> +#
Thomas> +############################################################# Thomas> +LIBPFM4_VERSION = 4.3.0
Thomas> +LIBPFM4_SOURCE = libpfm-$(LIBPFM4_VERSION).tar.gz
Thomas> +LIBPFM4_SITE = http://downloads.sourceforge.net/project/perfmon2/libpfm4
We normally have an empty line after the file header.
Committed with that fixed, thanks.
Signature · 1 line
Bye, Peter Korsgaard
PATCH 1/1 oprofile: upstep 0.9.7 patches to 0.9.8
In reply to Thomas De SchampheleireThe patches to oprofile 0.9.7 seem still relevant for 0.9.8, but for some reason they were not upstepped.
--- package/oprofile/oprofile-0.9.7-001-avr32-enable-lookup_dcookie.patch | 0 package/oprofile/oprofile-0.9.7-002-no-query-modules.patch | 0 2 files changed, 0 insertions(+), 0 deletions(-)
package/oprofile/oprofile-0.9.8-001-avr32-enable-lookup_dcookie.patch renamed from package/oprofile/oprofile-0.9.7-001-avr32-enable-lookup_dcookie.patch +0 -0
package/oprofile/oprofile-0.9.8-002-no-query-modules.patch renamed from package/oprofile/oprofile-0.9.7-002-no-query-modules.patch +0 -0
PATCH 1/1 oprofile: upstep 0.9.7 patches to 0.9.8
Dear Thomas De Schampheleire,
Quoted 2 lines · Thomas De Schampheleire
rename from
package/oprofile/oprofile-0.9.7-002-no-query-modules.patchrename topackage/oprofile/oprofile-0.9.8-002-no-query-modules.patch
We no longer want the version number in patch file names, for precisely the reason why you're doing this patch in the first place :-)
Thomas
Signature · 4 lines
Thomas Petazzoni, Free Electrons Kernel, drivers, real-time and embedded Linux development, consulting, training and support. http://free-electrons.com
PATCH 1/1 oprofile: upstep 0.9.7 patches to 0.9.8
Hi Thomas,
Quoted 9 lines · Thomas Petazzoni
Dear Thomas De Schampheleire,
Quoted 2 lines · Thomas De Schampheleire
rename from
package/oprofile/oprofile-0.9.7-002-no-query-modules.patchrename topackage/oprofile/oprofile-0.9.8-002-no-query-modules.patchWe no longer want the version number in patch file names, for precisely the reason why you're doing this patch in the first place :-)
Are we already making these changes, or are we waiting for someone to fix the central patching code according to the new rules first?
Best regards,
Thomas
PATCH 1/1 oprofile: upstep 0.9.7 patches to 0.9.8
Quoted 8 lines · Thomas De Schampheleire
Hi Thomas,
Quoted 4 lines · Thomas Petazzoni
Dear Thomas De Schampheleire,
On Tue, 12 Mar 2013 19:10:55 +0100, Thomas De Schampheleire wrote:
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
rename from
package/oprofile/oprofile-0.9.7-002-no-query-modules.patchrename topackage/oprofile/oprofile-0.9.8-002-no-query-modules.patchWe no longer want the version number in patch file names, for precisely the reason why you're doing this patch in the first place :-)
Are we already making these changes, or are we waiting for someone to fix the central patching code according to the new rules first?
There is nothing to be fixed in the central patching code (at least not
for this use case). Remove the version number.
Regards,
Arnout
Signature · 6 lines
Arnout Vandecappelle arnout at mind be Senior Embedded Software Architect +32-16-286500 Essensium/Mind http://www.mind.be G.Geenslaan 9, 3001 Leuven, Belgium BE 872 984 063 RPR Leuven LinkedIn profile: http://www.linkedin.com/in/arnoutvandecappelle GPG fingerprint: 7CB5 E4CC 6C2E EFD4 6E3D A754 F963 ECAB 2450 2F1F
PATCH 1/1 oprofile: upstep 0.9.7 patches to 0.9.8
Arnout, Thomas,
Quoted 5 lines · Arnout Vandecappelle
Quoted 2 lines
Are we already making these changes, or are we waiting for someone to fix the central patching code according to the new rules first?
There is nothing to be fixed in the central patching code (at least not
for this use case). Remove the version number.
Just to expand on Arnout answer: even though we haven't yet removed the support for versioned patches in the infrastructure, we have already been converting all the new packages, and many packages we were touching, to unversioned names. This will make it easier to change the package infrastructure later.
Best regards,
Thomas
Signature · 4 lines
Thomas Petazzoni, Free Electrons Kernel, drivers, real-time and embedded Linux development, consulting, training and support. http://free-electrons.com
PATCH 1/1 oprofile: upstep 0.9.7 patches to 0.9.8
In reply to Thomas De SchampheleireQuoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
"Thomas" == Thomas De Schampheleire <[email protected]> writes:
Thomas> The patches to oprofile 0.9.7 seem still relevant for 0.9.8, Thomas> but for some reason they were not upstepped.
Committed with the version number dropped from the file names, thanks.
Signature · 1 line
Bye, Peter Korsgaard
PATCH 1/1 oprofile 0.9.8: fix compilation on powerpc
In reply to Thomas De SchampheleireThe oprofile build was broken on powerpc since version 0.9.8. This was detected in several autobuilds, like
http://autobuild.buildroot.net/results/6f6c02d18495907d50fcdfc6003ac20d493c55fe/
The problem can be fixed by adding libpfm4 into the build and making a few changes in the configure script. Because the configure script is changed, we additionally need to prevent autoreconfiguring (which fails). At the same time, I fixed the removed --with-kernel-support configure option.
--- package/oprofile/Config.in | 1 + package/oprofile/oprofile-0.9.8-003-powerpc.patch | 27 +++++++++++++++++++++++++++ package/oprofile/oprofile.mk | 16 ++++++++++++++-- 3 files changed, 42 insertions(+), 2 deletions(-)
Note: this is compile-tested on x86 and powerpc, but I was not able to test it on target.
package/oprofile/Config.in +1 -0
@@ -2,6 +2,7 @@ config BR2_PACKAGE_OPROFILE bool "oprofile" select BR2_PACKAGE_POPT select BR2_PACKAGE_BINUTILS + select BR2_PACKAGE_LIBPFM4 if BR2_powerpc depends on BR2_INSTALL_LIBSTDCPP depends on !BR2_aarch64 # binutils help
package/oprofile/oprofile-0.9.8-003-powerpc.patch new file +27 -0
@@ -0,0 +1,27 @@ + +Signed-off-by: Thomas De Schampheleire <[email protected]> + +diff --git a/configure.ac b/configure.ac +--- a/configure.ac ++++ b/configure.ac +@@ -104,7 +104,7 @@ AC_DEFINE_UNQUOTED(HAVE_PERF_EVENTS, $HA + if test "$HAVE_PERF_EVENTS" = "1"; then + PFM_LIB= + arch="`uname -m`" +- if test "$arch" = "ppc64"; then ++ if test "$arch" = "ppc64" || test "$arch" = "ppc"; then + AC_CHECK_HEADER(perfmon/pfmlib.h,,[AC_MSG_ERROR([pfmlib.h not found; usually provided in papi devel package])]) + AC_CHECK_LIB(pfm,pfm_get_event_name, HAVE_LIBPFM3='1'; HAVE_LIBPFM='1', [ + AC_CHECK_LIB(pfm,pfm_get_os_event_encoding, HAVE_LIBPFM3='0'; HAVE_LIBPFM='1', +diff --git a/configure b/configure +--- a/configure ++++ b/configure +@@ -16985,7 +16985,7 @@ cat >>confdefs.h <<_ACEOF + if test "$HAVE_PERF_EVENTS" = "1"; then + PFM_LIB= + arch="`uname -m`" +- if test "$arch" = "ppc64"; then ++ if test "$arch" = "ppc64" || test "$arch" = "ppc"; then + if test "${ac_cv_header_perfmon_pfmlib_h+set}" = set; then + { $as_echo "$as_me:$LINENO: checking for perfmon/pfmlib.h" >&5 + $as_echo_n "checking for perfmon/pfmlib.h... " >&6; }
package/oprofile/oprofile.mk +14 -2
@@ -8,7 +8,7 @@ OPROFILE_VERSION = 0.9.8 OPROFILE_SITE = http://downloads.sourceforge.net/project/oprofile/oprofile/oprofile-$(OPROFILE_VERSION) OPROFILE_LICENSE = GPLv2+ OPROFILE_LICENSE_FILES = COPYING -OPROFILE_CONF_OPT = --localstatedir=/var --with-kernel-support +OPROFILE_CONF_OPT = --localstatedir=/var --with-kernel=$(SYSROOT_DIR) OPROFILE_BINARIES = utils/ophelp pp/opannotate pp/oparchive pp/opgprof OPROFILE_BINARIES += pp/opreport opjitconv/opjitconv daemon/oprofiled OPROFILE_BINARIES += utils/op-check-perfevents pe_profiling/operf libabi/opimport
@@ -29,7 +29,19 @@ ifeq ($(OPROFILE_ARCH),) OPROFILE_ARCH = $(BR2_ARCH) endif -OPROFILE_DEPENDENCIES = popt binutils +OPROFILE_DEPENDENCIES = popt binutils libpfm4 + +define OPROFILE_PREVENT_AUTOMAKE + # Prevent automake from running. + (cd $(@D); touch -c config* aclocal.m4 Makefile*); +endef + +define OPROFILE_FIX_CROSS_UNAME + $(SED) 's/`uname -m`/$(OPROFILE_ARCH)/' $(@D)/configure +endef + +OPROFILE_POST_PATCH_HOOKS += OPROFILE_PREVENT_AUTOMAKE +OPROFILE_POST_PATCH_HOOKS += OPROFILE_FIX_CROSS_UNAME define OPROFILE_INSTALL_TARGET_CMDS $(INSTALL) -d -m 755 $(TARGET_DIR)/usr/bin
PATCH 1/1 oprofile 0.9.8: fix compilation on powerpc
This series of patches fixes my build error I got when updating to latest buildroot sources on Friday. Fixing this was going to be my first task this Monday. Thanks!
I will check everything is working on my target later today. One possible minor change below:
Quoted 17 lines
package/oprofile/oprofile.mk +0
-0@@ -8,7 +8,7 @@ OPROFILE_VERSION = 0.9.8 OPROFILE_SITE =http://downloads.sourceforge.net/project/oprofile/oprofile/oprofile-
$(OPROFILE_VERSION)
OPROFILE_LICENSE= GPLv2+
OPROFILE_LICENSE_FILES= COPYINGpackage/oprofile/oprofile.mk +1
-1-OPROFILE_CONF_OPT = --localstatedir=/var --with-kernel-support +OPROFILE_CONF_OPT = --localstatedir=/var --with-kernel=$(SYSROOT_DIR) OPROFILE_BINARIES = utils/ophelp pp/opannotate pp/oparchive pp/opgprof OPROFILE_BINARIES += pp/opreport opjitconv/opjitconv daemon/oprofiled OPROFILE_BINARIES += utils/op-check-perfevents pe_profiling/operflibabi/opimport
package/oprofile/oprofile.mk ifeq ($(OPROFILE_ARCH),) · line 32
Quoted 5 lines
package/oprofile/oprofile.mk +1
-1@@ -29,7 +29,19 @@ ifeq ($(OPROFILE_ARCH),) OPROFILE_ARCH = $(BR2_ARCH) endif -OPROFILE_DEPENDENCIES = popt binutils +OPROFILE_DEPENDENCIES = popt binutils libpfm4
I think this dependency is only needed for powerpc. So something like:
OPROFILE_DEPENDENCIES = popt binutils
ifeq ($(BR2_powerpc),y)OPROFILE_DEPENDENCIES += libpfm4
Endif
Might be better?
Quoted 15 lines
package/oprofile/oprofile.mk +12
-0+ +define OPROFILE_PREVENT_AUTOMAKE + # Prevent automake from running. + (cd $(@D); touch -c config* aclocal.m4 Makefile*); +endef + +define OPROFILE_FIX_CROSS_UNAME + $(SED) 's/`uname -m`/$(OPROFILE_ARCH)/' $(@D)/configure +endef + +OPROFILE_POST_PATCH_HOOKS += OPROFILE_PREVENT_AUTOMAKE +OPROFILE_POST_PATCH_HOOKS += OPROFILE_FIX_CROSS_UNAME define OPROFILE_INSTALL_TARGET_CMDS $(INSTALL) -d -m 755 $(TARGET_DIR)/usr/bin
Quoted 4 lines
_______________________________________________
buildroot mailing list
buildroot at busybox.net
http://lists.busybox.net/mailman/listinfo/buildroot
The information in this email communication (inclusive of attachments) is confidential to 4RF Limited and the intended recipient(s). If you are not the intended recipient(s), please note that any use, disclosure, distribution or copying of this information or any part thereof is strictly prohibited and that the author accepts no liability for the consequences of any action taken on the basis of the information provided. If you have received this email in error, please notify the sender immediately by return email and then delete all instances of this email from your system. 4RF Limited will not accept responsibility for any consequences associated with the use of this email (including, but not limited to, damages sustained as a result of any viruses and/or any action or lack of action taken in reliance on it).
PATCH 1/1 oprofile 0.9.8: fix compilation on powerpc
Hi Reuben,
Quoted 3 lines · Reuben Dowle
This series of patches fixes my build error I got when updating to latest buildroot sources on Friday. Fixing this was going to be my first task this Monday. Thanks!
I will check everything is working on my target later today. One possible minor change below:
Thanks for your feedback.
I noticed there is still a problem with one of the patches, but I'm already looking forward to your runtime test.
See below...
Quoted 26 lines
Quoted 18 lines
package/oprofile/oprofile.mk +0
-0@@ -8,7 +8,7 @@ OPROFILE_VERSION = 0.9.8 OPROFILE_SITE =http://downloads.sourceforge.net/project/oprofile/oprofile/oprofile-
$(OPROFILE_VERSION)
OPROFILE_LICENSE= GPLv2+
OPROFILE_LICENSE_FILES= COPYINGpackage/oprofile/oprofile.mk +1
-1-OPROFILE_CONF_OPT = --localstatedir=/var --with-kernel-support +OPROFILE_CONF_OPT = --localstatedir=/var --with-kernel=$(SYSROOT_DIR) OPROFILE_BINARIES = utils/ophelp pp/opannotate pp/oparchive pp/opgprof OPROFILE_BINARIES += pp/opreport opjitconv/opjitconv daemon/oprofiled OPROFILE_BINARIES += utils/op-check-perfevents pe_profiling/operflibabi/opimport
package/oprofile/oprofile.mk ifeq ($(OPROFILE_ARCH),) · line 32
Quoted 5 lines
package/oprofile/oprofile.mk +1
-1@@ -29,7 +29,19 @@ ifeq ($(OPROFILE_ARCH),) OPROFILE_ARCH = $(BR2_ARCH) endif -OPROFILE_DEPENDENCIES = popt binutils +OPROFILE_DEPENDENCIES = popt binutils libpfm4I think this dependency is only needed for powerpc. So something like:
Quoted 6 lines
OPROFILE_DEPENDENCIES= popt binutils
ifeq ($(BR2_powerpc),y)OPROFILE_DEPENDENCIES+= libpfm4
EndifMight be better?
As far as I understand, these dependencies are only used for build-time ordering. This means that libfm4, *if selected in the config* will be built before oprofile. However, if libpfm4 is not selected, and you're not on powerpc, you can still build oprofile. The only benefit from the suggested code above, I think, is that on non-powerpc targets the unnecessary ordering of oprofile / libpfm4 is not enforced, but I think this will hardly give an impact on build time.
Best regards,
Thomas
PATCH 1/1 oprofile 0.9.8: fix compilation on powerpc
Quoted 3 lines · Thomas De Schampheleire
Hi Reuben,
On Tue, Mar 12, 2013 at 9:20 PM, Reuben Dowle <[email protected]> wrote:
[snip]
Quoted 10 lines
Quoted 10 lines
package/oprofile/oprofile.mk ifeq ($(OPROFILE_ARCH),) · line 32
Quoted 2 lines
package/oprofile/oprofile.mk +1
-1-OPROFILE_DEPENDENCIES = popt binutils +OPROFILE_DEPENDENCIES = popt binutils libpfm4I think this dependency is only needed for powerpc. So something like:
OPROFILE_DEPENDENCIES= popt binutils
ifeq ($(BR2_powerpc),y)OPROFILE_DEPENDENCIES+= libpfm4
Endif
Quoted 5 lines
Quoted 1 line
Might be better?
As far as I understand, these dependencies are only used for build-time ordering. This means that libfm4, *if selected in the config* will be built before oprofile.
Nope. If libpfm4 is a dependency, it will be built even if you haven't
configured it. So libpfm4 will be built unnecessarily.
Regards,
Arnout
Quoted 14 lines
However, if libpfm4 is not
selected, and you're not on powerpc, you can still build oprofile. The only benefit from the suggested code above, I think, is that on non-powerpc targets the unnecessary ordering of oprofile / libpfm4 is not enforced, but I think this will hardly give an impact on build time.Best regards,
Thomas
_______________________________________________
buildroot mailing list
buildroot at busybox.net
http://lists.busybox.net/mailman/listinfo/buildroot
Signature · 6 lines
Arnout Vandecappelle arnout at mind be Senior Embedded Software Architect +32-16-286500 Essensium/Mind http://www.mind.be G.Geenslaan 9, 3001 Leuven, Belgium BE 872 984 063 RPR Leuven LinkedIn profile: http://www.linkedin.com/in/arnoutvandecappelle GPG fingerprint: 7CB5 E4CC 6C2E EFD4 6E3D A754 F963 ECAB 2450 2F1F
PATCH 1/1 Fix oprofile build on powerpc
In reply to Thomas De SchampheleireDear Thomas De Schampheleire,
Quoted 5 lines · Thomas De Schampheleire
Since the upstep of oprofile to 0.9.8, the powerpc build was broken. This series fixes this problem.
As the last patch indicates, this is compile tested on x86 and powerpc, but not runtime-tested.
I think those patches are good, but when I started working on this issue, I also discovered that oprofile looks at /usr/include/linux by default to find out whether perf is supported or not in the kernel. Which obviously doesn't make sense in cross-compilation.
I've just pushed at
http://git.free-electrons.com/users/thomas-petazzoni/buildroot/commit/?h=oprofile-ppc-fix&id=dda400c3f798ba49b1b638f927b3089a9df24105
the work I started some time ago, which was unfinished. Note that I don't ask you to also fix those issues, I just thought it would be good mentioning these other potential problems.
Thomas
Signature · 4 lines
Thomas Petazzoni, Free Electrons Kernel, drivers, real-time and embedded Linux development, consulting, training and support. http://free-electrons.com
PATCH 1/1 Fix oprofile build on powerpc
Hi Thomas,
Quoted 15 lines · Thomas Petazzoni
Dear Thomas De Schampheleire,
Quoted 5 lines · Thomas De Schampheleire
Since the upstep of oprofile to 0.9.8, the powerpc build was broken. This series fixes this problem.
As the last patch indicates, this is compile tested on x86 and powerpc, but not runtime-tested.
I think those patches are good, but when I started working on this issue, I also discovered that oprofile looks at
/usr/include/linuxby default to find out whether perf is supported or not in the kernel. Which obviously doesn't make sense in cross-compilation.
Quoted 5 lines
I've just pushed at
http://git.free-electrons.com/users/thomas-petazzoni/buildroot/commit/?h=oprofile-ppc-fix&id=dda400c3f798ba49b1b638f927b3089a9df24105
the work I started some time ago, which was unfinished. Note that I don't ask you to also fix those issues, I just thought it would be good mentioning these other potential problems.
I haven't yet looked at your patches in detail, but I did also notice the check on perf_event.h in /usr/src/linux. However, this should have been fixed with the --with-kernel=$(SYSROOT_DIR).
However, I just now notice that SYSROOT_DIR does not exist here, and I need to find a better solution. I will investigate that aspect further.
Ideally we'd indeed merge your work together with these patches. Do you recall what was still missing or not working with your patches?
Thanks,
Thomas
PATCH 1/1 Fix oprofile build on powerpc
Dear Thomas De Schampheleire,
Quoted 9 lines · Thomas De Schampheleire
I haven't yet looked at your patches in detail, but I did also notice the check on
perf_event.hin/usr/src/linux. However, this should have been fixed with the --with-kernel=$(SYSROOT_DIR).
However, I just now notice thatSYSROOT_DIRdoes not exist here, and I need to find a better solution. I will investigate that aspect further.Ideally we'd indeed merge your work together with these patches. Do you recall what was still missing or not working with your patches?
No, I unfortunately don't remember. But maybe we should get your patches merged first, they were submitted first, and they look good, and they solve some of the problems. So just let's get your patches merged, and we'll fix/improve later on. It doesn't make sense to "block" your patches just because I did some other related work that I never submitted.
Best regards,
Thomas
Signature · 4 lines
Thomas Petazzoni, Free Electrons Kernel, drivers, real-time and embedded Linux development, consulting, training and support. http://free-electrons.com