ksysfs: Expose hardware name

[PATCH] ksysfs: Expose hardware name

Posted by Richard Weinberger 1 month, 3 weeks ago

Userspace tools often require a short, descriptive name of the hardware
platform. Currently, this information is fragmented: Device Tree-based
platforms typically use /proc/device-tree/model, while x86 platforms
combine multiple values from /sys/class/dmi/id/. There is no uniform,
platform-agnostic way to retrieve this string.

The kernel already maintains dump_stack_arch_desc_str[], which contains
the model name or DMI values used for the "Hardware name:" field in
stack traces.

Expose dump_stack_arch_desc_str[] at /sys/kernel/hardware_name to
provide userspace with a consistent interface for identifying the
underlying hardware.

Signed-off-by: Richard Weinberger <richard@nod.at>
---
 .../ABI/testing/sysfs-kernel-hardware_name          |  8 ++++++++
 include/linux/printk.h                              |  1 +
 kernel/ksysfs.c                                     | 13 +++++++++++++
 lib/dump_stack.c                                    |  2 +-
 4 files changed, 23 insertions(+), 1 deletion(-)
 create mode 100644 Documentation/ABI/testing/sysfs-kernel-hardware_name

diff --git a/Documentation/ABI/testing/sysfs-kernel-hardware_name b/Documentation/ABI/testing/sysfs-kernel-hardware_name
new file mode 100644
index 000000000000..21d130da1932
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-kernel-hardware_name
@@ -0,0 +1,8 @@
+What:		/sys/kernel/hardware_name
+Date:		Feb 2026
+KernelVersion:	7.1
+Contact:	Richard Weinberger <richard@nod.at>
+Description:
+		The hardware name as shown in stack traces.
+
+		Access: Read
diff --git a/include/linux/printk.h b/include/linux/printk.h
index 63d516c873b4..b8929495e35e 100644
--- a/include/linux/printk.h
+++ b/include/linux/printk.h
@@ -13,6 +13,7 @@ struct console;
 
 extern const char linux_banner[];
 extern const char linux_proc_banner[];
+extern char dump_stack_arch_desc_str[128];
 
 extern int oops_in_progress;	/* If set, an oops, panic(), BUG() or die() is in progress */
 
diff --git a/kernel/ksysfs.c b/kernel/ksysfs.c
index a9e6354d9e25..af7265754183 100644
--- a/kernel/ksysfs.c
+++ b/kernel/ksysfs.c
@@ -176,6 +176,16 @@ static ssize_t rcu_normal_store(struct kobject *kobj,
 KERNEL_ATTR_RW(rcu_normal);
 #endif /* #ifndef CONFIG_TINY_RCU */
 
+#ifdef CONFIG_PRINTK
+/* hardware name */
+static ssize_t hardware_name_show(struct kobject *kobj,
+				  struct kobj_attribute *attr, char *buf)
+{
+	return sysfs_emit(buf, "%s\n", dump_stack_arch_desc_str);
+}
+KERNEL_ATTR_RO(hardware_name);
+#endif
+
 /*
  * Make /sys/kernel/notes give the raw contents of our kernel .notes section.
  */
@@ -205,6 +215,9 @@ static struct attribute * kernel_attrs[] = {
 #ifndef CONFIG_TINY_RCU
 	&rcu_expedited_attr.attr,
 	&rcu_normal_attr.attr,
+#endif
+#ifdef CONFIG_PRINTK
+	&hardware_name_attr.attr,
 #endif
 	NULL
 };
diff --git a/lib/dump_stack.c b/lib/dump_stack.c
index f0c78b5b5324..bce41e603cff 100644
--- a/lib/dump_stack.c
+++ b/lib/dump_stack.c
@@ -15,7 +15,7 @@
 #include <linux/utsname.h>
 #include <linux/stop_machine.h>
 
-static char dump_stack_arch_desc_str[128];
+char dump_stack_arch_desc_str[128];
 
 /**
  * dump_stack_set_arch_desc - set arch-specific str to show with task dumps
-- 
2.51.0

Re: [PATCH] ksysfs: Expose hardware name

Posted by John Ogness 1 month, 3 weeks ago

Hi Richard,

On 2026-02-23, Richard Weinberger <richard@nod.at> wrote:
> Userspace tools often require a short, descriptive name of the hardware
> platform. Currently, this information is fragmented: Device Tree-based
> platforms typically use /proc/device-tree/model, while x86 platforms
> combine multiple values from /sys/class/dmi/id/. There is no uniform,
> platform-agnostic way to retrieve this string.
>
> The kernel already maintains dump_stack_arch_desc_str[], which contains
> the model name or DMI values used for the "Hardware name:" field in
> stack traces.
>
> Expose dump_stack_arch_desc_str[] at /sys/kernel/hardware_name to
> provide userspace with a consistent interface for identifying the
> underlying hardware.
>
> Signed-off-by: Richard Weinberger <richard@nod.at>
> ---
>  .../ABI/testing/sysfs-kernel-hardware_name          |  8 ++++++++
>  include/linux/printk.h                              |  1 +
>  kernel/ksysfs.c                                     | 13 +++++++++++++
>  lib/dump_stack.c                                    |  2 +-
>  4 files changed, 23 insertions(+), 1 deletion(-)
>  create mode 100644 Documentation/ABI/testing/sysfs-kernel-hardware_name
>
> diff --git a/kernel/ksysfs.c b/kernel/ksysfs.c
> index a9e6354d9e25..af7265754183 100644
> --- a/kernel/ksysfs.c
> +++ b/kernel/ksysfs.c
> @@ -176,6 +176,16 @@ static ssize_t rcu_normal_store(struct kobject *kobj,
>  KERNEL_ATTR_RW(rcu_normal);
>  #endif /* #ifndef CONFIG_TINY_RCU */
>  
> +#ifdef CONFIG_PRINTK
> +/* hardware name */
> +static ssize_t hardware_name_show(struct kobject *kobj,
> +				  struct kobj_attribute *attr, char *buf)
> +{
> +	return sysfs_emit(buf, "%s\n", dump_stack_arch_desc_str);
> +}
> +KERNEL_ATTR_RO(hardware_name);
> +#endif

It seems a bit odd to make this a dependency of CONFIG_PRINTK. Perhaps
@dump_stack_arch_desc_str and dump_stack_set_arch_desc() should move out
of lib/dump_stack.c if this new sysfs attribute is accepted?

John Ogness

Re: [PATCH] ksysfs: Expose hardware name

Posted by Richard Weinberger 1 month, 3 weeks ago

----- Ursprüngliche Mail -----
> Von: "John Ogness" <john.ogness@linutronix.de>
> An: "richard" <richard@nod.at>, "linux-kernel" <linux-kernel@vger.kernel.org>
> 
> It seems a bit odd to make this a dependency of CONFIG_PRINTK. Perhaps
> @dump_stack_arch_desc_str and dump_stack_set_arch_desc() should move out
> of lib/dump_stack.c if this new sysfs attribute is accepted?

Sure. I was surprised too that lib/dump_stack.c depends on CONFIG_PRINTK.

Thanks,
//richard

Re: [PATCH] ksysfs: Expose hardware name

Posted by Steven Rostedt 1 month, 2 weeks ago

On Mon, 23 Feb 2026 15:20:22 +0100 (CET)
Richard Weinberger <richard@nod.at> wrote:

> ----- Ursprüngliche Mail -----
> > Von: "John Ogness" <john.ogness@linutronix.de>
> > An: "richard" <richard@nod.at>, "linux-kernel" <linux-kernel@vger.kernel.org>
> > 
> > It seems a bit odd to make this a dependency of CONFIG_PRINTK. Perhaps
> > @dump_stack_arch_desc_str and dump_stack_set_arch_desc() should move out
> > of lib/dump_stack.c if this new sysfs attribute is accepted?  
> 
> Sure. I was surprised too that lib/dump_stack.c depends on CONFIG_PRINTK.
>

Or possibly we should make dump_stack.c not depend on CONFIG_PRINTK.
I'm guessing that it was done that way because it expects to dump the
stack to the console?

(haven't looked at the code yet)

-- Steve

Re: [PATCH] ksysfs: Expose hardware name

Posted by Petr Mladek 1 month ago

On Wed 2026-02-25 04:00:59, Steven Rostedt wrote:
> On Mon, 23 Feb 2026 15:20:22 +0100 (CET)
> Richard Weinberger <richard@nod.at> wrote:
> 
> > ----- Ursprüngliche Mail -----
> > > Von: "John Ogness" <john.ogness@linutronix.de>
> > > An: "richard" <richard@nod.at>, "linux-kernel" <linux-kernel@vger.kernel.org>
> > > 
> > > It seems a bit odd to make this a dependency of CONFIG_PRINTK. Perhaps
> > > @dump_stack_arch_desc_str and dump_stack_set_arch_desc() should move out
> > > of lib/dump_stack.c if this new sysfs attribute is accepted?  
> > 
> > Sure. I was surprised too that lib/dump_stack.c depends on CONFIG_PRINTK.
> >
> 
> Or possibly we should make dump_stack.c not depend on CONFIG_PRINTK.
> I'm guessing that it was done that way because it expects to dump the
> stack to the console?

My guess is that dump_stack.c depends on CONFIG_PRINTK because
the stack is stored using printk(). Nothing is stored when
printk() is a NOP ;-)

A clean solution would be to move dump_stack_arch_desc_str
and dump_stack_set_arch_desc() somewhere else (ksysfs.c ?)
and rename them to hardware_name and set_hardware_name().

Best Regards,
Petr

Re: [PATCH] ksysfs: Expose hardware name

Posted by Greg KH 1 month, 3 weeks ago

On Mon, Feb 23, 2026 at 02:34:48PM +0100, Richard Weinberger wrote:
> Userspace tools often require a short, descriptive name of the hardware
> platform. Currently, this information is fragmented: Device Tree-based
> platforms typically use /proc/device-tree/model, while x86 platforms
> combine multiple values from /sys/class/dmi/id/. There is no uniform,
> platform-agnostic way to retrieve this string.
> 
> The kernel already maintains dump_stack_arch_desc_str[], which contains
> the model name or DMI values used for the "Hardware name:" field in
> stack traces.
> 
> Expose dump_stack_arch_desc_str[] at /sys/kernel/hardware_name to
> provide userspace with a consistent interface for identifying the
> underlying hardware.

No objection from me, but now that this is a global symbol in the
kernel, "dump_" seems like an odd prefix for this.  How about
"linux_hardware_name" to match up with the sysfs file name?

</bikeshed>

thanks,

greg k-h

Re: [PATCH] ksysfs: Expose hardware name

Posted by Richard Weinberger 1 month, 3 weeks ago

----- Ursprüngliche Mail -----
> Von: "Greg Kroah-Hartman" <gregkh@linuxfoundation.org>
> An: "richard" <richard@nod.at>
> CC: "linux-kernel" <linux-kernel@vger.kernel.org>, "bhe" <bhe@redhat.com>, sourabhjain@linux.ibm.com, "Andrew Morton"
> <akpm@linux-foundation.org>, "senozhatsky" <senozhatsky@chromium.org>, "John Ogness" <john.ogness@linutronix.de>,
> "Steven Rostedt" <rostedt@goodmis.org>, "Petr Mladek" <pmladek@suse.com>, upstream+linux@sigma-star.at
> Gesendet: Montag, 23. Februar 2026 14:49:31
> Betreff: Re: [PATCH] ksysfs: Expose hardware name

> On Mon, Feb 23, 2026 at 02:34:48PM +0100, Richard Weinberger wrote:
>> Userspace tools often require a short, descriptive name of the hardware
>> platform. Currently, this information is fragmented: Device Tree-based
>> platforms typically use /proc/device-tree/model, while x86 platforms
>> combine multiple values from /sys/class/dmi/id/. There is no uniform,
>> platform-agnostic way to retrieve this string.
>> 
>> The kernel already maintains dump_stack_arch_desc_str[], which contains
>> the model name or DMI values used for the "Hardware name:" field in
>> stack traces.
>> 
>> Expose dump_stack_arch_desc_str[] at /sys/kernel/hardware_name to
>> provide userspace with a consistent interface for identifying the
>> underlying hardware.
> 
> No objection from me, but now that this is a global symbol in the
> kernel, "dump_" seems like an odd prefix for this.  How about
> "linux_hardware_name" to match up with the sysfs file name?

I don't care much. The reason why I kept the variable as-is is because
the function to set it is dump_stack_set_arch_desc().
So the prefix makes kinda sense.

Thanks,
//richard