[PATCH] tree-optimization/111736 - avoid address sanitizing of __seg_gs

[PATCH] tree-optimization/111736 - avoid address sanitizing of __seg_gs

[Richard Biener]

The following more thoroughly avoids address sanitizing accesses
to non-generic address-spaces.

Bootstrapped and tested on x86_64-unknown-linux-gnu.

OK?

Thanks,
Richard.

	PR tree-optimization/111736
	* asan.cc (instrument_derefs): Do not instrument accesses
	to non-generic address-spaces.

	* gcc.target/i386/pr111736.c: New testcase.
---
 gcc/asan.cc                              |  4 ++++
 gcc/testsuite/gcc.target/i386/pr111736.c | 23 +++++++++++++++++++++++
 2 files changed, 27 insertions(+)
 create mode 100644 gcc/testsuite/gcc.target/i386/pr111736.c

diff --git a/gcc/asan.cc b/gcc/asan.cc
index cfe83106460..04caf8802e2 100644
--- a/gcc/asan.cc
+++ b/gcc/asan.cc
@@ -2755,6 +2755,10 @@ instrument_derefs (gimple_stmt_iterator *iter, tree t,
   if (VAR_P (inner) && DECL_HARD_REGISTER (inner))
     return;
 
+  /* Accesses to non-generic address-spaces are not handled.  */
+  if (!ADDR_SPACE_GENERIC_P (TYPE_ADDR_SPACE (TREE_TYPE (inner))))
+    return;
+
   poly_int64 decl_size;
   if ((VAR_P (inner)
        || (TREE_CODE (inner) == RESULT_DECL
diff --git a/gcc/testsuite/gcc.target/i386/pr111736.c b/gcc/testsuite/gcc.target/i386/pr111736.c
new file mode 100644
index 00000000000..231fdd07e80
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr111736.c
@@ -0,0 +1,23 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fsanitize=address" } */
+
+int __seg_gs m;
+
+int foo (void)
+{
+  return m;
+}
+
+extern int  __seg_gs n;
+
+int bar (void)
+{
+  return n;
+}
+
+int baz (int __seg_gs *o)
+{
+  return *o;
+}
+
+/* { dg-final { scan-assembler-not "asan_report_load" } } */
-- 
2.35.3

Re: [PATCH] tree-optimization/111736 - avoid address sanitizing of __seg_gs

[Richard Biener]

On Thu, 21 Mar 2024, Jakub Jelinek wrote:

> On Thu, Mar 21, 2024 at 10:50:04AM +0100, Richard Biener wrote:
> > Fixed and pushed.  I suppose for address-spaces nested within the
> > generic address space we could instrument the address converted to
> > the generic address space value.
> 
> Unlike TLS, we don't know if address-spaces are nested within the generic
> address space and how to map the as address to a generic as address.

In theory there's hooks for this (TARGET_ADDR_SPACE_SUBSET_P) and
ADDR_SPACE_CONVERT_EXPR to do the conversion.  But hardly worth it
I guess.

Richard.

Re: [PATCH] tree-optimization/111736 - avoid address sanitizing of __seg_gs

[Jakub Jelinek]

On Thu, Mar 21, 2024 at 10:50:04AM +0100, Richard Biener wrote:
> Fixed and pushed.  I suppose for address-spaces nested within the
> generic address space we could instrument the address converted to
> the generic address space value.

Unlike TLS, we don't know if address-spaces are nested within the generic
address space and how to map the as address to a generic as address.

	Jakub

Re: [PATCH] tree-optimization/111736 - avoid address sanitizing of __seg_gs

[Richard Biener]

On Thu, 21 Mar 2024, Jakub Jelinek wrote:

> On Thu, Mar 21, 2024 at 10:25:24AM +0100, Richard Biener wrote:
> > The following more thoroughly avoids address sanitizing accesses
> > to non-generic address-spaces.
> > 
> > Bootstrapped and tested on x86_64-unknown-linux-gnu.
> > 
> > OK?
> > 
> > Thanks,
> > Richard.
> > 
> > 	PR tree-optimization/111736
> > 	* asan.cc (instrument_derefs): Do not instrument accesses
> > 	to non-generic address-spaces.
> > 
> > 	* gcc.target/i386/pr111736.c: New testcase.
> > ---
> >  gcc/asan.cc                              |  4 ++++
> >  gcc/testsuite/gcc.target/i386/pr111736.c | 23 +++++++++++++++++++++++
> >  2 files changed, 27 insertions(+)
> >  create mode 100644 gcc/testsuite/gcc.target/i386/pr111736.c
> > 
> > diff --git a/gcc/asan.cc b/gcc/asan.cc
> > index cfe83106460..04caf8802e2 100644
> > --- a/gcc/asan.cc
> > +++ b/gcc/asan.cc
> > @@ -2755,6 +2755,10 @@ instrument_derefs (gimple_stmt_iterator *iter, tree t,
> >    if (VAR_P (inner) && DECL_HARD_REGISTER (inner))
> >      return;
> >  
> > +  /* Accesses to non-generic address-spaces are not handled.  */
> 
> I'd say s/are not handled/should not be instrumented/

Fixed and pushed.  I suppose for address-spaces nested within the
generic address space we could instrument the address converted to
the generic address space value.

Richard.

> > +  if (!ADDR_SPACE_GENERIC_P (TYPE_ADDR_SPACE (TREE_TYPE (inner))))
> > +    return;
> > +
> 
> Otherwise LGTM.
> 
> >    poly_int64 decl_size;
> >    if ((VAR_P (inner)
> >         || (TREE_CODE (inner) == RESULT_DECL
> > diff --git a/gcc/testsuite/gcc.target/i386/pr111736.c b/gcc/testsuite/gcc.target/i386/pr111736.c
> > new file mode 100644
> > index 00000000000..231fdd07e80
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr111736.c
> > @@ -0,0 +1,23 @@
> > +/* { dg-do compile } */
> > +/* { dg-options "-O2 -fsanitize=address" } */
> > +
> > +int __seg_gs m;
> > +
> > +int foo (void)
> > +{
> > +  return m;
> > +}
> > +
> > +extern int  __seg_gs n;
> > +
> > +int bar (void)
> > +{
> > +  return n;
> > +}
> > +
> > +int baz (int __seg_gs *o)
> > +{
> > +  return *o;
> > +}
> > +
> > +/* { dg-final { scan-assembler-not "asan_report_load" } } */
> > -- 
> > 2.35.3
> 
> 	Jakub
> 
> 

-- 
Richard Biener <rguenther@suse.de>
SUSE Software Solutions Germany GmbH,
Frankenstrasse 146, 90461 Nuernberg, Germany;
GF: Ivo Totev, Andrew McDonald, Werner Knoblich; (HRB 36809, AG Nuernberg)

Re: [PATCH] tree-optimization/111736 - avoid address sanitizing of __seg_gs

[Jakub Jelinek]

On Thu, Mar 21, 2024 at 10:25:24AM +0100, Richard Biener wrote:
> The following more thoroughly avoids address sanitizing accesses
> to non-generic address-spaces.
> 
> Bootstrapped and tested on x86_64-unknown-linux-gnu.
> 
> OK?
> 
> Thanks,
> Richard.
> 
> 	PR tree-optimization/111736
> 	* asan.cc (instrument_derefs): Do not instrument accesses
> 	to non-generic address-spaces.
> 
> 	* gcc.target/i386/pr111736.c: New testcase.
> ---
>  gcc/asan.cc                              |  4 ++++
>  gcc/testsuite/gcc.target/i386/pr111736.c | 23 +++++++++++++++++++++++
>  2 files changed, 27 insertions(+)
>  create mode 100644 gcc/testsuite/gcc.target/i386/pr111736.c
> 
> diff --git a/gcc/asan.cc b/gcc/asan.cc
> index cfe83106460..04caf8802e2 100644
> --- a/gcc/asan.cc
> +++ b/gcc/asan.cc
> @@ -2755,6 +2755,10 @@ instrument_derefs (gimple_stmt_iterator *iter, tree t,
>    if (VAR_P (inner) && DECL_HARD_REGISTER (inner))
>      return;
>  
> +  /* Accesses to non-generic address-spaces are not handled.  */

I'd say s/are not handled/should not be instrumented/

> +  if (!ADDR_SPACE_GENERIC_P (TYPE_ADDR_SPACE (TREE_TYPE (inner))))
> +    return;
> +

Otherwise LGTM.

>    poly_int64 decl_size;
>    if ((VAR_P (inner)
>         || (TREE_CODE (inner) == RESULT_DECL
> diff --git a/gcc/testsuite/gcc.target/i386/pr111736.c b/gcc/testsuite/gcc.target/i386/pr111736.c
> new file mode 100644
> index 00000000000..231fdd07e80
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr111736.c
> @@ -0,0 +1,23 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -fsanitize=address" } */
> +
> +int __seg_gs m;
> +
> +int foo (void)
> +{
> +  return m;
> +}
> +
> +extern int  __seg_gs n;
> +
> +int bar (void)
> +{
> +  return n;
> +}
> +
> +int baz (int __seg_gs *o)
> +{
> +  return *o;
> +}
> +
> +/* { dg-final { scan-assembler-not "asan_report_load" } } */
> -- 
> 2.35.3

	Jakub