[Bug translator/9991] New: on-the-fly kernel debuginfo construction

[Bug translator/9991] New: on-the-fly kernel debuginfo construction

[fche at redhat dot com]

mingo idea, paraphrased:

Given kernel sources & build tree, systemtap could invoke a
  % make CONFIG_DEBUG_INFO=y foo/bar.o
on the fly to generate debuginfo-laden object code.


Some complications:

- different compiler versions may generate different code;
  though maybe build-id matching would work

- we may not know which object files to rebuild for a basic
  symbolic probe like:
    probe kernel.function("foo_bar")
  since these can occur in any old compilation unit.
  Plus consider:
    probe kernel.statement("*@include/bar.h:333")

- mapping module("foo") to the right path name is not obvious;
  might require parsing of kernel **/Makefile

-- 
           Summary: on-the-fly kernel debuginfo construction
           Product: systemtap
           Version: unspecified
            Status: NEW
          Severity: normal
          Priority: P2
         Component: translator
        AssignedTo: systemtap at sources dot redhat dot com
        ReportedBy: fche at redhat dot com


http://sourceware.org/bugzilla/show_bug.cgi?id=9991

------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug translator/9991] on-the-fly kernel debuginfo construction

[wcohen at redhat dot com]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |wcohen at redhat dot com


http://sourceware.org/bugzilla/show_bug.cgi?id=9991

------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug translator/9991] on-the-fly kernel debuginfo construction

[fche at redhat dot com]

------- Additional Comments From fche at redhat dot com  2009-03-23 21:15 -------
Another complication:

- addresses derived from object files (section-based)
  would need to be mapped to run-time (segment-based)
  addresses, which are assigned by the linker slicing
  and dicing based on the linker script; solution is
  probably not as simple as "rely on / extend kallsyms API"


-- 


http://sourceware.org/bugzilla/show_bug.cgi?id=9991

------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug translator/9991] on-the-fly kernel debuginfo construction

[fche at redhat dot com]

------- Additional Comments From fche at redhat dot com  2009-03-23 21:27 -------
Another complication:

- while gcc normally observes the dictum that the presence of "-g"
  should not affect the executable code/data at all, this may not
  completely hold even now; as above, we can hope that build-id
  catches this case, or else we can find some other mechanism to
  assert byte-for-byte equality between the regen'd .o and the
  sliced & diced & linked copy in .../vmlinux or .../foo.ko

- what if vmlinuz (compressed bootable image) is the only residue,
  and the linked vmlinux is not available (as in distro packages);
  maybe simply this on-the-fly widget cannot apply there


-- 


http://sourceware.org/bugzilla/show_bug.cgi?id=9991

------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

[Bug translator/9991] on-the-fly kernel debuginfo construction

[fche at redhat dot com]

------- Additional Comments From fche at redhat dot com  2009-04-08 18:09 -------
The ksplice tools may accomplish all or most of this.

-- 


http://sourceware.org/bugzilla/show_bug.cgi?id=9991

------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.