See near line 570 in loc2c.c. At least one of those FAIL's can be made to trigger on both the PR 1868 kernel versions, using this script: % stap -p2 -e 'probe kernel.syscall.* {}' There are lots of other errors, but you'll see a dwarf_addrloclists or two in there. Could you diagnose whether this is a compiler, elfutils, or other problem?
I reproduced this using: ./stap -v -p2 -e 'probe kernel.syscall.exit_group {}' on 2.6.12-1.1381_FC3smp A separate problem this displays is that systemtap is choosing a bogus probe PC here--the PC off the end of the function. That is an issue that was reported for other cases on the mailing list, and needs a separate bug filed. Correcting for that to select the same scopes systemtap gets, I reproduce it with: ./loc2c-test -k 0xc0124411 error_code Here we see the trouble in the debug info: [28a2cd] subprogram sibling [28a2f3] external name "sys_exit_group" decl_file 1 decl_line 917 prototyped low_pc 0xc0124411 high_pc 0xc012441e [28a2e3] formal_parameter name "error_code" decl_file 1 decl_line 916 type [27e6d4] location 2 byte block [ 0] fbreg 4 Though fbreg is used, there is no frame_base attribute. The code in question is a tail call: c0124411 <sys_exit_group>: c0124411: 0f b6 44 24 04 movzbl 0x4(%esp),%eax c0124416: c1 e0 08 shl $0x8,%eax c0124419: e9 63 ff ff ff jmp c0124381 <do_group_exit> There are definitely some gcc problems with the tail-call case.
The frame_base attribute does appear in the 2.6.9-22.24.EL rebuild done by gcc-3.4.4-3. It appears this was fixed along with the other gcc problems. This case should be checked on other kernel builds (FC4's gcc 4.0.x and rawhide's gcc, for example) to be sure.
See bug #1934 about the prologue detection issues in choosing the probe PC. That is a separate problem that shows up in testing the cases here, and confuses folks trying to test. Will did some testing on a kernel built by the new compiler on i686 and x86-64, and the results look good though the bug #1934 problem makes it harder to puzzle out.
AFAIK this is about old GCC bugs that are gone now. Please verify you no longer think there is an outstanding problem.
ok now