[Bug kprobes/5759] probes in sched.c on preempt kernel cause reboot

jkenisto at us dot ibm dot com sourceware-bugzilla@sourceware.org
Tue Feb 12 21:49:00 GMT 2008


------- Additional Comments From jkenisto at us dot ibm dot com  2008-02-12 21:48 -------
Created an attachment (id=2252)
 --> (http://sourceware.org/bugzilla/attachment.cgi?id=2252&action=view)
oops from trying to install a kprobe on init_sched_debug_procfs

Following is a summary of the oops.  For some reason, the kernel can't handle
the READ from the probepoint when we're trying to copy the probed instruction
into the kprobe_insn_page.  This is from a 4-CPU x86_64 system running vanilla
2.6.24.

Unable to handle kernel paging request at ffffffff8086c9d0
rip is in arch_copy_kprobe(), 2nd line, first insn of inline memcpy:
     movsl  %ds:(%rsi),%es:(%rdi)
rip = ffffffff80472891
rsi = ffffffff8086c9d0
rdi = ffffffff8800000f
ds = 0
es = 0
rsi contains the address of init_sched_debug_procfs (i.e., the probepoint).
rdi contains MODULES_VADDR+0xf.  This could be correct if the first
kprobe_insn_page was allocated right at MODULES_VADDR.	MAX_INSN_SIZE=0xf for
x86_64, so this would be the start of the 2nd slot.

BTW, on 2.6.24, register_kprobe() works on many functions in kernel/sched.c,
but I've also see it fail on other probepoints besides this one.

-- 


http://sourceware.org/bugzilla/show_bug.cgi?id=5759

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



More information about the Systemtap mailing list