[kpatch] can't set ftrace filter

[Josh Poimboeuf]

On Thu, Jun 11, 2015 at 07:44:59PM +0530, Somdeep Das wrote:
> Hi,
> 
> I have been trying out kpatch for some time now. I was able to create,
> install and load a patch successfully for my simple device driver. So far
> so good. Next, I tried to patch a bit more complex kernel module having
> multiple object files. I then created and installed the patch using the
> kpatch tools but while loading the patch I am hitting an issue. Kpatch is
> reporting the following error:
> kpatch: can't set ftrace filter at address <some address>
> 
> I did some investigation and reached to a conclusion that
> ftrace_match_addr() is returning -22 (-EINVAL) as ftrace_location() is
> failing to find the target function’s address in ftrace table.
> 
> I would be really grateful if I can get any help on resolving this issue.

Hi Somdeep,

Most kernel and module functions have a "call mcount" instruction as the
beginning, which is the ftrace hook which kpatch needs in order to patch
a function.

"objdump -d" should show something like the following:

0000000000000000 <drm_getmagic>:
       0:       e8 00 00 00 00          callq  5 <drm_getmagic+0x5>

(Ignore the <drm_getmagic+0x5> target.  objdump isn't smart enough to
understand ELF relocations.  But it should always be a "e8 00 00 00 00"
instruction.)

It sounds like the function you're trying to patch doesn't have an
ftrace hook at the beginning for some reason.  Did you do anything
"different" when building the more complex module?  Or any chance the
function was annotated with "notrace"?

-- 
Josh