[RFC,13/34] x86-64: Disable prefer_map_32bit_exec tunable on non-Linux
Checks
Commit Message
While we could/should implement MAP_32BIT for the Hurd port by setting
all the high bits of mask in a vm_map () call, neither MAP_32BIT nor
glibc.cpu.prefer_map_32bit_exec exist on the Hurd as of now. Compile
this code out to fix build failures.
Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
---
sysdeps/x86/cpu-features.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
Comments
Hello,
Sergey Bugaev, le dim. 19 mars 2023 18:09:56 +0300, a ecrit:
> While we could/should implement MAP_32BIT for the Hurd port by setting
> all the high bits of mask in a vm_map () call, neither MAP_32BIT nor
> glibc.cpu.prefer_map_32bit_exec exist on the Hurd as of now. Compile
> this code out to fix build failures.
Rather use defined(MAP_32BIT)? That way it will nicely work for BSD
ports as well, and when hurd eventually supports MAP_32BIT.
> Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
> ---
> sysdeps/x86/cpu-features.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/sysdeps/x86/cpu-features.c b/sysdeps/x86/cpu-features.c
> index 822688e2..78631d7a 100644
> --- a/sysdeps/x86/cpu-features.c
> +++ b/sysdeps/x86/cpu-features.c
> @@ -27,7 +27,7 @@
> extern void TUNABLE_CALLBACK (set_hwcaps) (tunable_val_t *)
> attribute_hidden;
>
> -# ifdef __LP64__
> +# if defined (__LP64__) && defined (__linux__)
> static void
> TUNABLE_CALLBACK (set_prefer_map_32bit_exec) (tunable_val_t *valp)
> {
> @@ -715,7 +715,7 @@ no_cpuid:
> #if HAVE_TUNABLES
> TUNABLE_GET (hwcaps, tunable_val_t *, TUNABLE_CALLBACK (set_hwcaps));
>
> -# ifdef __LP64__
> +# if defined (__LP64__) && defined (__linux__)
> TUNABLE_GET (prefer_map_32bit_exec, tunable_val_t *,
> TUNABLE_CALLBACK (set_prefer_map_32bit_exec));
> # endif
> --
> 2.39.2
>
On Mon, Apr 3, 2023 at 2:09 AM Samuel Thibault <samuel.thibault@gnu.org> wrote:
> Sergey Bugaev, le dim. 19 mars 2023 18:09:56 +0300, a ecrit:
> > While we could/should implement MAP_32BIT for the Hurd port by setting
> > all the high bits of mask in a vm_map () call, neither MAP_32BIT nor
> > glibc.cpu.prefer_map_32bit_exec exist on the Hurd as of now. Compile
> > this code out to fix build failures.
>
> Rather use defined(MAP_32BIT)? That way it will nicely work for BSD
> ports as well, and when hurd eventually supports MAP_32BIT.
Not really. What breaks compilation here is not missing MAP_32BIT, but
rather the undefined set_prefer_map_32bit_exec tunable. It is defined
declaratively in sysdeps/unix/sysv/linux/x86_64/64/dl-tunables.list
(see commit 317f1c0a8a71a862b1e600ff5386b08e02cf4b95); so as-is, it
really is x86_64 Linux specific.
Is there some way to check if a specific tunable is defined, like #if
HAVE_TUNABLE (prefer_map_32bit_exec)? AFAICS the generated
dl-tunable-list.h defines the C enum members, but not any checkable
preprocessor macros.
Alternatively: maybe this part of cpu-features.c should be moved into
sysdeps/unix/sysv/linux/x86_64/64/cpu-features.c and #include_next the
common cpu-features.c?
Sergey
On Mon, Apr 3, 2023 at 3:10 AM Sergey Bugaev <bugaevc@gmail.com> wrote:
>
> On Mon, Apr 3, 2023 at 2:09 AM Samuel Thibault <samuel.thibault@gnu.org> wrote:
> > Sergey Bugaev, le dim. 19 mars 2023 18:09:56 +0300, a ecrit:
> > > While we could/should implement MAP_32BIT for the Hurd port by setting
> > > all the high bits of mask in a vm_map () call, neither MAP_32BIT nor
> > > glibc.cpu.prefer_map_32bit_exec exist on the Hurd as of now. Compile
> > > this code out to fix build failures.
> >
> > Rather use defined(MAP_32BIT)? That way it will nicely work for BSD
> > ports as well, and when hurd eventually supports MAP_32BIT.
>
> Not really. What breaks compilation here is not missing MAP_32BIT, but
> rather the undefined set_prefer_map_32bit_exec tunable. It is defined
> declaratively in sysdeps/unix/sysv/linux/x86_64/64/dl-tunables.list
> (see commit 317f1c0a8a71a862b1e600ff5386b08e02cf4b95); so as-is, it
> really is x86_64 Linux specific.
>
> Is there some way to check if a specific tunable is defined, like #if
> HAVE_TUNABLE (prefer_map_32bit_exec)? AFAICS the generated
> dl-tunable-list.h defines the C enum members, but not any checkable
> preprocessor macros.
>
> Alternatively: maybe this part of cpu-features.c should be moved into
> sysdeps/unix/sysv/linux/x86_64/64/cpu-features.c and #include_next the
> common cpu-features.c?
>
Or something like this.
On Mon, Apr 3, 2023 at 10:02 PM H.J. Lu <hjl.tools@gmail.com> wrote:
> Or something like this.
Or like that, yeah!
On a second thought, I actually like Samuel's idea (ifdef MAP_32BIT):
instead of only building this piece of code when the tunable *is*
defined, we'd build it when the tunable *should be* defined (that is,
when your port has MAP_32BIT). And if you get build errors, that is
your not-so-gentle reminder to go and define the tunable for your
port.
What do you think?
Sergey
@@ -27,7 +27,7 @@
extern void TUNABLE_CALLBACK (set_hwcaps) (tunable_val_t *)
attribute_hidden;
-# ifdef __LP64__
+# if defined (__LP64__) && defined (__linux__)
static void
TUNABLE_CALLBACK (set_prefer_map_32bit_exec) (tunable_val_t *valp)
{
@@ -715,7 +715,7 @@ no_cpuid:
#if HAVE_TUNABLES
TUNABLE_GET (hwcaps, tunable_val_t *, TUNABLE_CALLBACK (set_hwcaps));
-# ifdef __LP64__
+# if defined (__LP64__) && defined (__linux__)
TUNABLE_GET (prefer_map_32bit_exec, tunable_val_t *,
TUNABLE_CALLBACK (set_prefer_map_32bit_exec));
# endif