Re: [PATCH v6 21/21] perf-probe: Add array argument support

[Masami Hiramatsu <mhiramat@kernel.org>]

On Mon, 19 Mar 2018 13:29:59 +0530
Ravi Bangoria <ravi.bangoria@linux.vnet.ibm.com> wrote:

> Hi Masami,
> 
> On 03/17/2018 06:23 PM, Masami Hiramatsu wrote:
> > Since kprobes events support an array argument, perf-probe
> > can also support dumping array now.
> > The syntax are
> >
> >  <array-var>[<range>]
> > or
> >  <pointer-var>[<range>]
> >
> > where the <range> is <start>..<end>. e.g. array[0..5].
> > This can also be available with string type. In this
> > case, the string array should be "char *array[]" or
> > "char **array_ptr".
> >
> > Note that this feature is only available on the kernel which
> > supports array type.
> 
> User can still do,
> 
> # perf probe -x ~/hello main:3 'a=a:x32[3]'
> 
> which will successfully be installed in uprobe_events. But for a
> perf tool, x32[3] is not a valid type. And it seems perf tool don't
> validate the 'type' field:
> 
>     static int parse_perf_probe_arg(char *str, struct perf_probe_arg *arg)
>     {
>         ....
>         tmp = strchr(str, ':');
>         if (tmp) {      /* Type setting */
>                 *tmp = '\0';
>                 arg->type = strdup(tmp + 1);
>                 if (arg->type == NULL)
>                         return -ENOMEM;
>                 pr_debug("type:%s ", arg->type);
>         }
> 
> Is it okay to allow user to specify array size with type field?

Fro this patch, yes. The availability of type is checked only when
it is automatically generated.
IMO, it should be done in another patch, something like
"Validate user specified type casting" patch. Would you need it?

Thank you,

-- 
Masami Hiramatsu <mhiramat@kernel.org>