Print the machine hardware name (UTS_MACHINE) in /proc/sys/kernel/arch.
This helps people who debug kernel with initramfs with minimal
environment (i.e. without coreutils or even busybox) or allow to open
sysfs file instead of run 'uname -m' in high level languages.
Signed-off-by: Petr Vorel <[email protected]>
---
Changes v2->v3:
* fix title in the subject
@Andrew @Eric: FYI previously the sysfs file was in
/sys/devices/system/cpu/, that's why I sent previous versions to Greg.
Documentation/admin-guide/sysctl/kernel.rst | 5 +++++
kernel/utsname_sysctl.c | 7 +++++++
2 files changed, 12 insertions(+)
diff --git a/Documentation/admin-guide/sysctl/kernel.rst b/Documentation/admin-guide/sysctl/kernel.rst
index ee6572b1edad..bbaa85194695 100644
--- a/Documentation/admin-guide/sysctl/kernel.rst
+++ b/Documentation/admin-guide/sysctl/kernel.rst
@@ -65,6 +65,11 @@ combining the following values:
4 s3_beep
= =======
+arch
+====
+
+The machine hardware name, the same output as ``uname -m``
+(e.g. ``x86_64`` or ``aarch64``).
auto_msgmni
===========
diff --git a/kernel/utsname_sysctl.c b/kernel/utsname_sysctl.c
index 4ca61d49885b..7ffdd2cd5ff9 100644
--- a/kernel/utsname_sysctl.c
+++ b/kernel/utsname_sysctl.c
@@ -73,6 +73,13 @@ static DEFINE_CTL_TABLE_POLL(hostname_poll);
static DEFINE_CTL_TABLE_POLL(domainname_poll);
static struct ctl_table uts_kern_table[] = {
+ {
+ .procname = "arch",
+ .data = init_uts_ns.name.machine,
+ .maxlen = sizeof(init_uts_ns.name.machine),
+ .mode = 0444,
+ .proc_handler = proc_do_uts_string,
+ },
{
.procname = "ostype",
.data = init_uts_ns.name.sysname,
--
2.35.3
> On Thu, 1 Sep 2022 21:44:03 +0200 Petr Vorel <[email protected]> wrote:
> > Print the machine hardware name (UTS_MACHINE) in /proc/sys/kernel/arch.
> > This helps people who debug kernel with initramfs with minimal
> > environment (i.e. without coreutils or even busybox) or allow to open
> > sysfs file instead of run 'uname -m' in high level languages.
> If they're this constrained, why not run sys_uname() directly?
Hi Andrew,
of course you're right that in C is the most straightforward way to use uname(2).
I was more concerned about userspace, but I have to admit uname is probably
everywhere, therefore this is probably useless.
The background of this was my colleague while writing software running LTP tests
with qemu (using whole systems on qcow image or minimal dracut initramfs) or over SSH
got all info about running kernel from proc / sysfs, just architecture was
missing.
/proc/cpuinfo and /sys/devices/system/cpu/ are full of details about CPU, but
surprisingly no info about the architecture. Understand, this info is in struct
utsname and there is uname binary, OTOH there are
/proc/sys/kernel/{osrelease,ostype,version}, which are also in struct utsname.
Kind regards,
Petr
On Thu, 1 Sep 2022 21:44:03 +0200 Petr Vorel <[email protected]> wrote:
> Print the machine hardware name (UTS_MACHINE) in /proc/sys/kernel/arch.
>
> This helps people who debug kernel with initramfs with minimal
> environment (i.e. without coreutils or even busybox) or allow to open
> sysfs file instead of run 'uname -m' in high level languages.
If they're this constrained, why not run sys_uname() directly?
On Thu, Sep 01, 2022 at 09:44:03PM +0200, Petr Vorel wrote:
> Print the machine hardware name (UTS_MACHINE) in /proc/sys/kernel/arch.
>
> This helps people who debug kernel with initramfs with minimal
> environment (i.e. without coreutils or even busybox) or allow to open
> sysfs file instead of run 'uname -m' in high level languages.
>
> Signed-off-by: Petr Vorel <[email protected]>
Acked-by: Greg Kroah-Hartman <[email protected]>