Re: [PATCH] perf probe: Fix add event failed when 32-bit perf running in 64-bit kernel

[Masami Hiramatsu]

On Thu, 15 Jul 2021 09:20:01 +0800
Yang Jihong <yangjihong1@xxxxxxxxxx> wrote:

> Hello Hiramatsu,
> 
> On 2021/7/14 16:35, Masami Hiramatsu wrote:
> > Hi Yang,
> > 
> > On Wed, 14 Jul 2021 14:54:32 +0800
> > Yang Jihong <yangjihong1@xxxxxxxxxx> wrote:
> > 
> >> The "address" member  of "struct probe_trace_point" uses long data type.
> >> If kernel is 64-bit and perf program is 32-bit, size of "address" variable is
> >> 32 bits. As a result, upper 32 bits of address read from kernel are truncated,
> >> An error occurs during address comparison in kprobe_warn_out_range function.
> > 
> > Good catch!
> > I didn't imagine that such a use case. But that is important because perf
> > probe can be used for cross-arch probe definition too.
> > 
> >>
> >> Before:
> >>
> >>    # perf probe -a schedule
> >>    schedule is out of .text, skip it.
> >>      Error: Failed to add events.
> >>
> >> Solution:
> >>    Change data type of "address" variable to u64 and change corresponding
> >> address printing and value assignment.
> > 
> > OK, as far as I can see, the other parts of the perf also uses u64 for
> > "address" storing variables. (e.g. symbols, maps etc.)
> > 
> >>
> >> After:
> >>
> >>    # perf.new.new probe -a schedule
> >>    Added new event:
> >>      probe:schedule       (on schedule)
> >>
> >>    You can now use it in all perf tools, such as:
> >>
> >>            perf record -e probe:schedule -aR sleep 1
> >>
> >>    # perf probe -l
> >>      probe:schedule       (on schedule)
> > 
> > I think you missed one thing here.
> > Usually, this shows the filename and line number of schedule().
> Yes,  I tried the following diff and now it can show the filename (as is 
> function entry, relative line number is 0), thank you very much :)
> 
> The test result in my environment is as follows:
> 
> # perf probe -a schedule
> Added new event:
>    probe:schedule       (on schedule)
> 
> You can now use it in all perf tools, such as:
> 
>          perf record -e probe:schedule -aR sleep 1
> 
> # perf probe -l schedule -k vmlinux.debug
>    probe:schedule       (on schedule@kernel/sched/core.c)
> 

OK, good!

> 
> Can I put the following diff together and submit a v2 patch?

Yes, please include this fix in your patch, because it is what your patch
has to do :)

Thank you!

> > 
> > Could you try below diff?
> > 
> > diff --git a/tools/perf/util/dwarf-aux.c b/tools/perf/util/dwarf-aux.c
> > index 7d2ba8419b0c..609ca1671501 100644
> > --- a/tools/perf/util/dwarf-aux.c
> > +++ b/tools/perf/util/dwarf-aux.c
> > @@ -113,14 +113,14 @@ static Dwarf_Line *cu_getsrc_die(Dwarf_Die *cu_die, Dwarf_Addr addr)
> >    *
> >    * Find a line number and file name for @addr in @cu_die.
> >    */
> > -int cu_find_lineinfo(Dwarf_Die *cu_die, unsigned long addr,
> > -		    const char **fname, int *lineno)
> > +int cu_find_lineinfo(Dwarf_Die *cu_die, Dwarf_Addr addr,
> > +		     const char **fname, int *lineno)
> >   {
> >   	Dwarf_Line *line;
> >   	Dwarf_Die die_mem;
> >   	Dwarf_Addr faddr;
> >   
> > -	if (die_find_realfunc(cu_die, (Dwarf_Addr)addr, &die_mem)
> > +	if (die_find_realfunc(cu_die, addr, &die_mem)
> >   	    && die_entrypc(&die_mem, &faddr) == 0 &&
> >   	    faddr == addr) {
> >   		*fname = dwarf_decl_file(&die_mem);
> > @@ -128,7 +128,7 @@ int cu_find_lineinfo(Dwarf_Die *cu_die, unsigned long addr,
> >   		goto out;
> >   	}
> >   
> > -	line = cu_getsrc_die(cu_die, (Dwarf_Addr)addr);
> > +	line = cu_getsrc_die(cu_die, addr);
> >   	if (line && dwarf_lineno(line, lineno) == 0) {
> >   		*fname = dwarf_linesrc(line, NULL, NULL);
> >   		if (!*fname)
> > diff --git a/tools/perf/util/dwarf-aux.h b/tools/perf/util/dwarf-aux.h
> > index cb99646843a9..7ee0fa19b5c4 100644
> > --- a/tools/perf/util/dwarf-aux.h
> > +++ b/tools/perf/util/dwarf-aux.h
> > @@ -19,7 +19,7 @@ const char *cu_find_realpath(Dwarf_Die *cu_die, const char *fname);
> >   const char *cu_get_comp_dir(Dwarf_Die *cu_die);
> >   
> >   /* Get a line number and file name for given address */
> > -int cu_find_lineinfo(Dwarf_Die *cudie, unsigned long addr,
> > +int cu_find_lineinfo(Dwarf_Die *cudie, Dwarf_Addr addr,
> >   		     const char **fname, int *lineno);
> >   
> >   /* Walk on functions at given address */
> > diff --git a/tools/perf/util/probe-finder.c b/tools/perf/util/probe-finder.c
> > index 8ba01bbdf05c..50d861a80f57 100644
> > --- a/tools/perf/util/probe-finder.c
> > +++ b/tools/perf/util/probe-finder.c
> > @@ -1727,7 +1727,7 @@ int debuginfo__find_probe_point(struct debuginfo *dbg, u64 addr,
> >   	}
> >   
> >   	/* Find a corresponding line (filename and lineno) */
> > -	cu_find_lineinfo(&cudie, addr, &fname, &lineno);
> > +	cu_find_lineinfo(&cudie, (Dwarf_Addr)addr, &fname, &lineno);
> >   	/* Don't care whether it failed or not */
> >   
> >   	/* Find a corresponding function (name, baseline and baseaddr) */
> > @@ -1828,8 +1828,7 @@ static int line_range_add_line(const char *src, unsigned int lineno,
> >   }
> >   
> >   static int line_range_walk_cb(const char *fname, int lineno,
> > -			      Dwarf_Addr addr __maybe_unused,
> > -			      void *data)
> > +			      Dwarf_Addr addr, void *data)
> >   {
> >   	struct line_finder *lf = data;
> >   	const char *__fname;
> > 
> > 


-- 
Masami Hiramatsu <mhiramat@xxxxxxxxxx>