tonyj @ suse reports a case where translator.cxx appears to emit incorrect buildid-offset values into stap-symbols.h, which leads to verification errors & aborted scripts. honeydew:/sys/module/libata/sections/:[0]# grep . /sys/module/libata/sections/.??* | sort /sys/module/libata/sections/.bss:0xd000000000c7c918 /sys/module/libata/sections/.data.rel.ro:0xd000000000c67370 /sys/module/libata/sections/.data.rel:0xd000000000c79718 /sys/module/libata/sections/.data:0xd000000000c67268 /sys/module/libata/sections/.exit.text:0xd000000000c5d2fc /sys/module/libata/sections/.gnu.linkonce.this_module:0xd000000000c7c6d8 /sys/module/libata/sections/.note.gnu.build-id:0xd000000000c67244 /sys/module/libata/sections/.opd:0xd000000000c7a788 /sys/module/libata/sections/.rodata.str1.8:0xd000000000c62e20 /sys/module/libata/sections/.rodata:0xd000000000c607e8 /sys/module/libata/sections/.strtab:0xd000000000cb9a50 /sys/module/libata/sections/.stubs:0xd000000000c5d8d8 /sys/module/libata/sections/.symtab:0xd000000000cb0000 /sys/module/libata/sections/.text:0xd000000000c40000 /sys/module/libata/sections/.toc1:0xd000000000c79a68 /sys/module/libata/sections/.toc:0xd000000000c7c6a8 the buildid-offset should be on the order of a few dozen, since runtime/sym.c properly relocates .ko buildids against the .notes section base. However, stap was found to emit ... tonyj Found build-id in libata, length 20, end at 0x87d7c tonyj build_id_offset = 0x87d7c - 0x4bd00= 0x3c07c It appears as if translate.cxx:5384, the XXX section related to build_id_offset calculation, is neglecting to perform elfutils relocation on the incoming vaddr for .ko files. There is no theory yet as to why only some .ko's are manifesting this problem.
Believed to be solved with commit 71fa1fe39faa153cf6ede620c4855e508059aa39 Author: Tony Jones <tonyj@suse.de> Date: Wed Sep 29 12:47:52 2010 -0400 PR10812: relocate module build-id *** This bug has been marked as a duplicate of bug 10812 ***