<?xml version="1.0" encoding="UTF-8" standalone="yes" ?>
<!DOCTYPE bugzilla SYSTEM "https://bugzilla.yoctoproject.org/page.cgi?id=bugzilla.dtd">

<bugzilla version="5.0.6"
          urlbase="https://bugzilla.yoctoproject.org/"
          
          maintainer="it-coreprojects-helpdesk@linuxfoundation.org"
>

    <bug>
          <bug_id>7129</bug_id>
          
          <creation_ts>2015-01-09 13:07:28 +0000</creation_ts>
          <short_desc>perf build fails with current master</short_desc>
          <delta_ts>2015-03-26 06:35:09 +0000</delta_ts>
          <reporter_accessible>1</reporter_accessible>
          <cclist_accessible>1</cclist_accessible>
          <classification_id>7</classification_id>
          <classification>Build System, Metadata &amp; Runtime</classification>
          <product>OE-Core</product>
          <component>kernel</component>
          <version>1.8</version>
          <rep_platform>x86</rep_platform>
          <op_sys>x86_64</op_sys>
          <bug_status>VERIFIED</bug_status>
          <resolution>FIXED</resolution>
          
          
          <bug_file_loc></bug_file_loc>
          <status_whiteboard></status_whiteboard>
          <keywords></keywords>
          <priority>Medium+</priority>
          <bug_severity>normal</bug_severity>
          <target_milestone>1.8 M3</target_milestone>
          
          
          <everconfirmed>1</everconfirmed>
          <reporter name="Ross Burton">ross.burton</reporter>
          <assigned_to name="Tom Zanussi">tom.zanussi</assigned_to>
          <cc>alexandru.c.georgescu</cc>
    
    <cc>bruce.ashfield</cc>
    
    <cc>tom.zanussi</cc>
          
          <qa_contact name="Alexandru Georgescu">alexandru.c.georgescu</qa_contact>
          <cf_os>---</cf_os>
          <cf_regression_type>---</cf_regression_type>
          
          <cf_docchange>No (bug/feature does not impact docs)</cf_docchange>

      

      

      

          <comment_sort_order>oldest_to_newest</comment_sort_order>  
          <long_desc isprivate="0" >
    <commentid>47839</commentid>
    <comment_count>0</comment_count>
    <who name="Ross Burton">ross.burton</who>
    <bug_when>2015-01-09 13:07:28 +0000</bug_when>
    <thetext>I&apos;m getting this error with perl and master as of 9th January:

|   CC       /data/poky-master/tmp/work/nuc-poky-linux/perf/1.0-r9/perf-1.0/arch/x86/tests/regs_load.o
| arch/x86/tests/regs_load.S: Assembler messages:
| arch/x86/tests/regs_load.S:65: Error: operand type mismatch for `push&apos;
| arch/x86/tests/regs_load.S:72: Error: operand type mismatch for `pop&apos;
| make[1]: *** [/data/poky-master/tmp/work/nuc-poky-linux/perf/1.0-r9/perf-1.0/arch/x86/tests/regs_load.o] Error 1
| make[1]: *** Waiting for unfinished jobs....
| make[1]: *** wait: No child processes.  Stop.
| make: *** [all] Error 2
| make: Leaving directory `/data/poky-master/tmp/sysroots/nuc/usr/src/kernel/tools/perf&apos;
| ERROR: oe_runmake failed
| WARNING: exit code 1 from a shell command.
| ERROR: Function failed: do_compile (log file is located at /data/poky-master/tmp/work/nuc-poky-linux/perf/1.0-r9/temp/log.do_compile.27524)
ERROR: Task 10086 (/home/ross/Yocto/poky/meta/recipes-kernel/perf/perf.bb, do_compile) failed with exit code &apos;1&apos;</thetext>
  </long_desc><long_desc isprivate="0" >
    <commentid>47840</commentid>
    <comment_count>1</comment_count>
    <who name="Bruce Ashfield">bruce.ashfield</who>
    <bug_when>2015-01-09 13:08:31 +0000</bug_when>
    <thetext>TomZ was having a look at this today, we noticed it during testing
as well.</thetext>
  </long_desc><long_desc isprivate="0" >
    <commentid>47843</commentid>
    <comment_count>2</comment_count>
    <who name="Tom Zanussi">tom.zanussi</who>
    <bug_when>2015-01-09 17:14:08 +0000</bug_when>
    <thetext>For some reason, now libdw-dwarf-unwind: [ on  ] is being found, which means that the arch/x86 test cases are being executed, where before they weren&apos;t.

i.e. what previously triggered in the perf compile was:

config/Makefile:371: Disabling post unwind, no support found.

because NO_LIBDW_DWARF_UNWIND was set, and with dwarf-post-unwind set, NO_DWARF_UNWIND would also be set and the below would be skipped.

ifndef NO_DWARF_UNWIND
LIB_OBJS += $(OUTPUT)arch/$(ARCH)/tests/regs_load.o
LIB_OBJS += $(OUTPUT)arch/$(ARCH)/tests/dwarf-unwind.o
endif

One option would be to add NO_DWARF_UNWIND=1 as we do for NO_DWARF=1 but it seems that would just be curing a symptom whereas what we really want to know is why libdw-dwarf-unwind is no being detected.</thetext>
  </long_desc><long_desc isprivate="0" >
    <commentid>47847</commentid>
    <comment_count>3</comment_count>
    <who name="Tom Zanussi">tom.zanussi</who>
    <bug_when>2015-01-09 18:03:22 +0000</bug_when>
    <thetext>The two patches here fix it for me:

http://git.yoctoproject.org/cgit/cgit.cgi/poky-contrib/log/?h=tzanussi/possible-perf-7129-fix

This gets around the problem and we really shouldn&apos;t be leaving things to perf&apos;s detection code anyway, so it&apos;s a step forward, but it still leaves the &apos;perf-libunwind&apos;=enabled case broken, which in any case wouldn&apos;t be a regression and should probably be fixed by a separate bug.

I can post this but would rather understand why it suddenly became a problem and maybe reflects a deeper underlying issue.</thetext>
  </long_desc><long_desc isprivate="0" >
    <commentid>47848</commentid>
    <comment_count>4</comment_count>
    <who name="Bruce Ashfield">bruce.ashfield</who>
    <bug_when>2015-01-09 18:28:00 +0000</bug_when>
    <thetext>The fixes worked here as well, I&apos;m up and building a full image.</thetext>
  </long_desc><long_desc isprivate="0" >
    <commentid>47863</commentid>
    <comment_count>5</comment_count>
    <who name="Tom Zanussi">tom.zanussi</who>
    <bug_when>2015-01-12 17:39:04 +0000</bug_when>
    <thetext>There&apos;s actually a fix for this on lkml:

https://lkml.org/lkml/2015/1/11/302</thetext>
  </long_desc><long_desc isprivate="0" >
    <commentid>47864</commentid>
    <comment_count>6</comment_count>
    <who name="Bruce Ashfield">bruce.ashfield</who>
    <bug_when>2015-01-12 17:40:57 +0000</bug_when>
    <thetext>And now the perf bugfix pain begins.

I can fix all the linux-yocto kernels with the upstream change. But do we care about kernels that
wouldn&apos;t have this integrated ?

Stopping the floating/detected support is still a good idea, so we can always do both. Disable the
support AND fix it in the linux-yocto kernels.</thetext>
  </long_desc><long_desc isprivate="0" >
    <commentid>47865</commentid>
    <comment_count>7</comment_count>
    <who name="Tom Zanussi">tom.zanussi</who>
    <bug_when>2015-01-12 17:54:33 +0000</bug_when>
    <thetext>Well, the only thing we control are the linux-yocto kernels, so fixing those should be sufficient.

I can submit my patch adding libdw so users can disable it, but I won&apos;t submit the patch to disable it, which would fix build errors for users who don&apos;t have the kernel patch, but which would also disable userspace stackunwinding in the call graph profiles if frame pointers are disabled.</thetext>
  </long_desc><long_desc isprivate="0" >
    <commentid>48115</commentid>
    <comment_count>8</comment_count>
    <who name="Tom Zanussi">tom.zanussi</who>
    <bug_when>2015-01-21 00:22:15 +0000</bug_when>
    <thetext>The unwind disabling has been pulled into master in these 2 patches, to stop the immediate build problems:

http://git.yoctoproject.org/cgit/cgit.cgi/poky/commit/?id=4e13036d66b8eb61b998961243d28cdff8d855f4

http://git.yoctoproject.org/cgit/cgit.cgi/poky/commit/?id=14fa1343eccd6296120a5e4e054b21b83c57c3e0

But the libunwind enabled case still needs to tested/fixed once the upstream kernel patch mentioned above has been pulled into linux-yocto.</thetext>
  </long_desc><long_desc isprivate="0" >
    <commentid>48344</commentid>
    <comment_count>9</comment_count>
    <who name="Tom Zanussi">tom.zanussi</who>
    <bug_when>2015-01-27 23:40:19 +0000</bug_when>
    <thetext>Fixed in master:

commit 14fa1343eccd6296120a5e4e054b21b83c57c3e0
Author: Tom Zanussi &lt;tom.zanussi@linux.intel.com&gt;
Date:   Thu Jan 15 11:54:17 2015 +0000

    perf: Disable perf-libunwind

commit 4e13036d66b8eb61b998961243d28cdff8d855f4
Author: Tom Zanussi &lt;tom.zanussi@linux.intel.com&gt;
Date:   Thu Jan 15 11:54:16 2015 +0000

    perf: Add libdw unwind support to perf-libunwind feature</thetext>
  </long_desc><long_desc isprivate="0" >
    <commentid>49779</commentid>
    <comment_count>10</comment_count>
    <who name="Alexandru Georgescu">alexandru.c.georgescu</who>
    <bug_when>2015-03-26 06:35:09 +0000</bug_when>
    <thetext>could build successfully perf with master:

Build Configuration:
BB_VERSION        = &quot;1.27.0&quot;
BUILD_SYS         = &quot;x86_64-linux&quot;
NATIVELSBSTRING   = &quot;Fedora-20&quot;
TARGET_SYS        = &quot;i586-poky-linux&quot;
MACHINE           = &quot;qemux86&quot;
DISTRO            = &quot;poky&quot;
DISTRO_VERSION    = &quot;1.8+snapshot-20150326&quot;
TUNE_FEATURES     = &quot;m32 i586&quot;
TARGET_FPU        = &quot;&quot;
meta              
meta-yocto        
meta-yocto-bsp    = &quot;master:5c76cebc24675edb3dff473fce9b67cedff615db&quot;</thetext>
  </long_desc>
      
      

    </bug>

</bugzilla>