From: Dave Hansen <dave.hansen@linux.intel.com>
The existing x86_match_cpu() infrastructure can be used to match
a bunch of attributes of a CPU: vendor, family, model, steppings
and CPU features.
But, there's one more attribute that's missing and unable to be
matched against: the platform ID, enumerated on Intel CPUs in
MSR_IA32_PLATFORM_ID. It is a little more obscure and is only
queried during microcode loading. This is because Intel sometimes
has CPUs with identical family/model/stepping but which need
different microcode. These CPUs are differentiated with the
platform ID.
Add a field in 'struct x86_cpu_id' for the platform ID. Similar
to the stepping field, make the new field a mask of platform IDs.
Some examples:
0x01: matches only platform ID 0x0
0x02: matches only platform ID 0x1
0x03: matches platform IDs 0x0 or 0x1
0x80: matches only platform ID 0x7
0xff: matches all 8 possible platform IDs
Since the mask is only a byte wide, it nestles in next to another
u8 and does not even increase the size of 'struct x86_cpu_id'.
Reserve the all 0's value as the wildcard (X86_PLATFORM_ANY). This
avoids forcing changes changes to existing 'struct x86_cpu_id' users.
They can just continue to fill the field with 0's and their matching
will work exactly as before.
Note: If someone is ever looking for space in 'struct x86_cpu_id',
this new field could probably get stuck over in ->driver_data
for the one user that there is.
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Thomas Gleixner <tglx@kernel.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Pawan Gupta <pawan.kumar.gupta@linux.intel.com>
Cc: "Peter Zijlstra (Intel)" <peterz@infradead.org>
Cc: x86@kernel.org
Cc: Jon Kohler <jon@nutanix.com>
---
b/arch/x86/kernel/cpu/match.c | 3 +++
b/include/linux/mod_devicetable.h | 2 ++
2 files changed, 5 insertions(+)
diff -puN arch/x86/kernel/cpu/match.c~platform-match arch/x86/kernel/cpu/match.c
--- a/arch/x86/kernel/cpu/match.c~platform-match 2026-01-19 11:38:10.020939979 -0800
+++ b/arch/x86/kernel/cpu/match.c 2026-01-19 11:38:10.046940973 -0800
@@ -76,6 +76,9 @@ const struct x86_cpu_id *x86_match_cpu(c
if (m->steppings != X86_STEPPING_ANY &&
!(BIT(c->x86_stepping) & m->steppings))
continue;
+ if (m->platform_mask != X86_PLATFORM_ANY &&
+ !(c->x86_platform_id & m->platform_mask))
+ continue;
if (m->feature != X86_FEATURE_ANY && !cpu_has(c, m->feature))
continue;
if (!x86_match_vendor_cpu_type(c, m))
diff -puN include/linux/mod_devicetable.h~platform-match include/linux/mod_devicetable.h
--- a/include/linux/mod_devicetable.h~platform-match 2026-01-19 11:38:10.043940858 -0800
+++ b/include/linux/mod_devicetable.h 2026-01-19 11:38:10.046940973 -0800
@@ -692,6 +692,7 @@ struct x86_cpu_id {
__u16 feature; /* bit index */
/* Solely for kernel-internal use: DO NOT EXPORT to userspace! */
__u16 flags;
+ __u8 platform_mask;
__u8 type;
kernel_ulong_t driver_data;
};
@@ -703,6 +704,7 @@ struct x86_cpu_id {
#define X86_STEPPING_ANY 0
#define X86_STEP_MIN 0
#define X86_STEP_MAX 0xf
+#define X86_PLATFORM_ANY 0x0
#define X86_FEATURE_ANY 0 /* Same as FPU, you can't test for that */
#define X86_CPU_TYPE_ANY 0
_On Mon, Jan 19, 2026 at 11:50:57AM -0800, Dave Hansen wrote:
> The existing x86_match_cpu() infrastructure can be used to match
> a bunch of attributes of a CPU: vendor, family, model, steppings
> and CPU features.
>
> But, there's one more attribute that's missing and unable to be
> matched against: the platform ID, enumerated on Intel CPUs in
> MSR_IA32_PLATFORM_ID. It is a little more obscure and is only
> queried during microcode loading. This is because Intel sometimes
> has CPUs with identical family/model/stepping but which need
> different microcode. These CPUs are differentiated with the
> platform ID.
>
> Add a field in 'struct x86_cpu_id' for the platform ID. Similar
> to the stepping field, make the new field a mask of platform IDs.
> Some examples:
>
> 0x01: matches only platform ID 0x0
> 0x02: matches only platform ID 0x1
> 0x03: matches platform IDs 0x0 or 0x1
> 0x80: matches only platform ID 0x7
> 0xff: matches all 8 possible platform IDs
>
> Since the mask is only a byte wide, it nestles in next to another
> u8 and does not even increase the size of 'struct x86_cpu_id'.
>
> Reserve the all 0's value as the wildcard (X86_PLATFORM_ANY). This
> avoids forcing changes changes to existing 'struct x86_cpu_id' users.
> They can just continue to fill the field with 0's and their matching
> will work exactly as before.
>
> Note: If someone is ever looking for space in 'struct x86_cpu_id',
> this new field could probably get stuck over in ->driver_data
> for the one user that there is.
...
> struct x86_cpu_id {
> __u16 feature; /* bit index */
> /* Solely for kernel-internal use: DO NOT EXPORT to userspace! */
> __u16 flags;
> + __u8 platform_mask;
> __u8 type;
> kernel_ulong_t driver_data;
> };
...
> #define X86_STEPPING_ANY 0
> #define X86_STEP_MIN 0
> #define X86_STEP_MAX 0xf
> +#define X86_PLATFORM_ANY 0x0
> #define X86_FEATURE_ANY 0 /* Same as FPU, you can't test for that */
Wouldn't it be slightly better to locate the new one here to follow the order
in x86_cpu_id above?
> #define X86_CPU_TYPE_ANY 0
--
With Best Regards,
Andy Shevchenko
On 1/20/26 00:30, Andy Shevchenko wrote: ... >> #define X86_STEPPING_ANY 0 >> #define X86_STEP_MIN 0 >> #define X86_STEP_MAX 0xf >> +#define X86_PLATFORM_ANY 0x0 >> #define X86_FEATURE_ANY 0 /* Same as FPU, you can't test for that */ > Wouldn't it be slightly better to locate the new one here to follow the order > in x86_cpu_id above? I stuck it there because ->platform_mask acts functionally like the stepping mask does.
© 2016 - 2026 Red Hat, Inc.