Re: [RFC updated PATCHES] Bug Translator 3016 : Error accessing members of anonymous structs / unions

[Masami Hiramatsu <mhiramat at redhat dot com>]

Hi Prerana,

Thank you for fixing!

I checked this patch worked well.
$ stap -e 'probe module("libsas").function("sas_ex_revalidate_domain"){print($port_dev->ex_dev->children)}' -vp2
Pass 1: parsed user script and 45 library script(s) in 370usr/30sys/401real ms.
semantic error: struct/union 'children' is being accessed instead of a member of the struct/union: identifier 
'$port_dev' at <input>:1:67
Pass 2: analyzed script: 2 probe(s), 0 function(s), 0 embed(s), 0 global(s) in 160usr/1930sys/2092real ms.
Pass 2: analysis failed.  Try again with more '-v' (verbose) options.

Ack!

Prerna Saxena wrote:
> Hi Masami,
> Thanks for letting me know..
> It pointed me to an incorrect pointer reference that I had 
> overlooked...Its funny why this was never uncovered in all test-runs I 
> had on x86 & x86_64 :-)
> Here's the updated patch...with this, your test script correctly throws 
> a semantic error instead of causing SEGV.
>
> Masami Hiramatsu wrote:
> > Hi Prerna,
> >
> > Prerna Saxena wrote:
> > > Agree, but making "die_mem" as the 6th arg will disturb the call flow.
> > > This is because in each recursive call, the 6th arg is overwritten by 
> > > a die which is pointed to by DW_AT_type attribute of the original 
> > > attr_mem. If die_mem is re-used as 6th arg instead of a new variable, 
> > > the old contents of die (in the parent recursive call) will also be 
> > > lost as both die & die_mem point to the same location, (whose 
> > > contents would be overwritten). This is not a problem if search in a 
> > > branch  has succeeded-- but in case a search path fails and a new 
> > > branch needs to be tried, this will deem it impossible.
> > >
> > > I agree with your concern about "temp_die_2" being local memory on 
> > > the stack which may be reused, so I've fixed it by copying the 
> > > contents of "temp_die_2" to "die_mem" in case of a successful match. 
> > > This should take care of memory errors.
> >
> > Hmm, you seems right, but the below command still cause SEGV.
> >
> > $ stap -e 'probe 
> > module("libsas").function("sas_ex_revalidate_domain"){print($port_dev->ex_dev->children)}' 
> > -vp2
> > Pass 1: parsed user script and 45 library script(s) in 
> > 370usr/30sys/404real ms.
> > Segmentation fault
> >
> > Would you run above command for testing & debugging?
> >
> > Thank you,
> Thanks & regards,

--
Masami Hiramatsu

Software Engineer
Hitachi Computer Products (America) Inc.
Software Solutions Division

e-mail: mhiramat@redhat.com