[PATCH v1 3/6] perf dwarf-aux: Better type matching for stack variables

Zecheng Li posted 6 patches 2 months, 1 week ago
There is a newer version of this series
[PATCH v1 3/6] perf dwarf-aux: Better type matching for stack variables
Posted by Zecheng Li 2 months, 1 week ago
Utilizes the previous is_breg_access_indirect function to determine if
the stack location stores the variable itself or the struct it points
to.

If the DWARF expression shows DW_OP_stack_value without DW_OP_deref, it
indicates the variable value is the reg + offset itself, and the stack
location it points to is the dereferenced type.

Signed-off-by: Zecheng Li <zecheng@google.com>
---
 tools/perf/util/dwarf-aux.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/tools/perf/util/dwarf-aux.c b/tools/perf/util/dwarf-aux.c
index 814c96ea509f..4039dbd2b8c0 100644
--- a/tools/perf/util/dwarf-aux.c
+++ b/tools/perf/util/dwarf-aux.c
@@ -1635,6 +1635,14 @@ static int __die_collect_vars_cb(Dwarf_Die *die_mem, void *arg)
 	if (die_get_real_type(die_mem, &type_die) == NULL)
 		return DIE_FIND_CB_SIBLING;
 
+	if ((ops->atom == DW_OP_fbreg || ops->atom == DW_OP_breg7) &&
+	    dwarf_tag(&type_die) == DW_TAG_pointer_type &&
+	    is_breg_access_indirect(ops, nops)) {
+		/* Get the target type of the pointer */
+		if (die_get_real_type(&type_die, &type_die) == NULL)
+			return DIE_FIND_CB_SIBLING;
+	}
+
 	vt = malloc(sizeof(*vt));
 	if (vt == NULL)
 		return DIE_FIND_CB_END;
-- 
2.50.1.470.g6ba607880d-goog
Re: [PATCH v1 3/6] perf dwarf-aux: Better type matching for stack variables
Posted by Ian Rogers 2 months, 1 week ago
On Fri, Jul 25, 2025 at 1:28 PM Zecheng Li <zecheng@google.com> wrote:
>
> Utilizes the previous is_breg_access_indirect function to determine if
> the stack location stores the variable itself or the struct it points
> to.
>
> If the DWARF expression shows DW_OP_stack_value without DW_OP_deref, it
> indicates the variable value is the reg + offset itself, and the stack
> location it points to is the dereferenced type.
>
> Signed-off-by: Zecheng Li <zecheng@google.com>
> ---
>  tools/perf/util/dwarf-aux.c | 8 ++++++++
>  1 file changed, 8 insertions(+)
>
> diff --git a/tools/perf/util/dwarf-aux.c b/tools/perf/util/dwarf-aux.c
> index 814c96ea509f..4039dbd2b8c0 100644
> --- a/tools/perf/util/dwarf-aux.c
> +++ b/tools/perf/util/dwarf-aux.c
> @@ -1635,6 +1635,14 @@ static int __die_collect_vars_cb(Dwarf_Die *die_mem, void *arg)
>         if (die_get_real_type(die_mem, &type_die) == NULL)
>                 return DIE_FIND_CB_SIBLING;
>
> +       if ((ops->atom == DW_OP_fbreg || ops->atom == DW_OP_breg7) &&

A comment saying the significance of DW_OP_fbreg and DW_OP_breg7 would
be useful, for example, why not DW_OP_breg6? Isn't breg7 going to be
x86 specific?

Thanks,
Ian


> +           dwarf_tag(&type_die) == DW_TAG_pointer_type &&
> +           is_breg_access_indirect(ops, nops)) {
> +               /* Get the target type of the pointer */
> +               if (die_get_real_type(&type_die, &type_die) == NULL)
> +                       return DIE_FIND_CB_SIBLING;
> +       }
> +
>         vt = malloc(sizeof(*vt));
>         if (vt == NULL)
>                 return DIE_FIND_CB_END;
> --
> 2.50.1.470.g6ba607880d-goog
>
Re: [PATCH v1 3/6] perf dwarf-aux: Better type matching for stack variables
Posted by Namhyung Kim 2 months, 1 week ago
On Fri, Jul 25, 2025 at 06:17:33PM -0700, Ian Rogers wrote:
> On Fri, Jul 25, 2025 at 1:28 PM Zecheng Li <zecheng@google.com> wrote:
> >
> > Utilizes the previous is_breg_access_indirect function to determine if
> > the stack location stores the variable itself or the struct it points
> > to.
> >
> > If the DWARF expression shows DW_OP_stack_value without DW_OP_deref, it
> > indicates the variable value is the reg + offset itself, and the stack
> > location it points to is the dereferenced type.
> >
> > Signed-off-by: Zecheng Li <zecheng@google.com>
> > ---
> >  tools/perf/util/dwarf-aux.c | 8 ++++++++
> >  1 file changed, 8 insertions(+)
> >
> > diff --git a/tools/perf/util/dwarf-aux.c b/tools/perf/util/dwarf-aux.c
> > index 814c96ea509f..4039dbd2b8c0 100644
> > --- a/tools/perf/util/dwarf-aux.c
> > +++ b/tools/perf/util/dwarf-aux.c
> > @@ -1635,6 +1635,14 @@ static int __die_collect_vars_cb(Dwarf_Die *die_mem, void *arg)
> >         if (die_get_real_type(die_mem, &type_die) == NULL)
> >                 return DIE_FIND_CB_SIBLING;
> >
> > +       if ((ops->atom == DW_OP_fbreg || ops->atom == DW_OP_breg7) &&
> 
> A comment saying the significance of DW_OP_fbreg and DW_OP_breg7 would
> be useful, for example, why not DW_OP_breg6? Isn't breg7 going to be
> x86 specific?

Good point!  Right, it's x86 specific and we cannot hard-code.  It needs
a way to indicate this register is a stack pointer on the target arch.
Please see the instruction tracking code which checks state->stack_reg.
 
On x86_64, DWARF reg 7 is RSP.  If RBP is used as a frame pointer, it's
probably using DW_OP_fbreg expression.

> 
> > +           dwarf_tag(&type_die) == DW_TAG_pointer_type &&
> > +           is_breg_access_indirect(ops, nops)) {

Maybe it needs to check the register inside is_breg_access_indirect().

Thanks,
Namhyung


> > +               /* Get the target type of the pointer */
> > +               if (die_get_real_type(&type_die, &type_die) == NULL)
> > +                       return DIE_FIND_CB_SIBLING;
> > +       }
> > +
> >         vt = malloc(sizeof(*vt));
> >         if (vt == NULL)
> >                 return DIE_FIND_CB_END;
> > --
> > 2.50.1.470.g6ba607880d-goog
> >