Re: [PATCH 2/2] drm/xe/cri: Expose device UUID through sysfs

[Michal Wajdeczko <michal.wajdeczko@intel.com>]

On 9/7/2026 11:13 AM, Nilawar, Badal wrote:
> 
> On 04-09-2026 22:07, Michal Wajdeczko wrote:
>>
>> On 9/4/2026 6:06 PM, Badal Nilawar wrote:
>>> Expose a read-only sysfs attribute, device_uuid, that reports the
>>> GPU's unique hardware identifier.
>>>
>> Bspec: 53048,53049
> Sure
>>
>>> Assisted-by: Claude:claude-opus-4.8
>>> Signed-off-by: Badal Nilawar <badal.nilawar@intel.com>
>>> ---
>>>   .../ABI/testing/sysfs-driver-intel-xe-gpu     | 10 ++++++
>>>   drivers/gpu/drm/xe/regs/xe_regs.h             |  2 ++
>>>   drivers/gpu/drm/xe/xe_device.c                |  3 ++
>>>   drivers/gpu/drm/xe/xe_device_sysfs.c          | 36 +++++++++++++++++++
>>>   drivers/gpu/drm/xe/xe_device_types.h          |  3 ++
>>>   5 files changed, 54 insertions(+)
>>>   create mode 100644 Documentation/ABI/testing/sysfs-driver-intel-xe-gpu
>>>
>>> diff --git a/Documentation/ABI/testing/sysfs-driver-intel-xe-gpu b/Documentation/ABI/testing/sysfs-driver-intel-xe-gpu
>>> new file mode 100644
>>> index 000000000000..1a4bdfbad8d7
>>> --- /dev/null
>>> +++ b/Documentation/ABI/testing/sysfs-driver-intel-xe-gpu
>>> @@ -0,0 +1,10 @@
>>> +What:        /sys/bus/pci/drivers/xe/.../device_uuid
>>> +Date:        September 2026
>>> +KernelVersion:    7.4
>>> +Contact:    intel-xe@lists.freedesktop.org
>>> +Description:    RO. Unique 64-bit identifier of the device instance, read from
>>> +        the device's CRI_DEVICE_UUID register and printed as a
>> hmm, maybe in the user facing documentation we should refer to this register as
>>
>>     DEVICE_UUID(102008)
>>
>> instead of the meaningless macro name?
> I will drop "read from ..." statement.
>>
>>> +        zero-padded 16-digit hexadecimal string.
>> if this is UUID, shouldn't we print it in UUID format? see %pU
> I this is not UUID, i will rename the node as device_uid.
>>
>>> +
>>> +        This sysfs file is present only on Intel Xe platforms that
>>> +        provide a device UUID. It is available to all users.
>>> diff --git a/drivers/gpu/drm/xe/regs/xe_regs.h b/drivers/gpu/drm/xe/regs/xe_regs.h
>>> index ef4746b7b5d3..68226fe0b659 100644
>>> --- a/drivers/gpu/drm/xe/regs/xe_regs.h
>>> +++ b/drivers/gpu/drm/xe/regs/xe_regs.h
>>> @@ -30,6 +30,8 @@
>>>   #define XEHP_MTCFG_ADDR                XE_REG(0x101800)
>>>   #define   TILE_COUNT                REG_GENMASK(15, 8)
>>>   +#define CRI_DEVICE_UUID                XE_REG(0x102008)
>>> +
>>>   #define GGC                    XE_REG(0x108040)
>>>   #define   GMS_MASK                REG_GENMASK(15, 8)
>>>   #define   GGMS_MASK                REG_GENMASK(7, 6)
>>> diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c
>>> index 8583b2e9ecf4..e8db8871990b 100644
>>> --- a/drivers/gpu/drm/xe/xe_device.c
>>> +++ b/drivers/gpu/drm/xe/xe_device.c
>>> @@ -801,6 +801,9 @@ int xe_device_probe_early(struct xe_device *xe)
>>>       if (IS_SRIOV_VF(xe))
>>>           vf_update_device_info(xe);
>>>   +    if (xe->info.platform == XE_CRESCENTISLAND)
>>> +        xe->device_uuid = xe_mmio_read64_2x32(xe_root_tile_mmio(xe), CRI_DEVICE_UUID);
>>> +
>>>       /*
>>>        * Check for pcode uncore_init status to confirm if the SoC
>>>        * initialization is complete. Until done, any MMIO or lmem access from
>>> diff --git a/drivers/gpu/drm/xe/xe_device_sysfs.c b/drivers/gpu/drm/xe/xe_device_sysfs.c
>>> index a73e0e957cb0..a4d326159dce 100644
>>> --- a/drivers/gpu/drm/xe/xe_device_sysfs.c
>>> +++ b/drivers/gpu/drm/xe/xe_device_sysfs.c
>>> @@ -8,6 +8,7 @@
>>>   #include <linux/pci.h>
>>>   #include <linux/sysfs.h>
>>>   +#include "regs/xe_regs.h"
>>>   #include "xe_device.h"
>>>   #include "xe_device_sysfs.h"
>>>   #include "xe_mmio.h"
>>> @@ -264,6 +265,35 @@ static const struct attribute_group auto_link_downgrade_attr_group = {
>>>       .attrs = auto_link_downgrade_attrs,
>>>   };
>>>   +/**
>>> + * DOC: device_uuid
>> do we need this?
>> you already described that in Documentation/ABI
> I think I will drop the Documentation/ABI and keep DOC: section.

hmm, but all other sysfs entries are defined in Documentation/ABI
(which IMO is the correct place, not a local DOC section)


>>
>>> + *
>>> + * Each Xe device exposes a 64-bit UUID that uniquely identifies the SoC.
>>> + * The value is read from the CRI_DEVICE_UUID register and can be
>>> + * read back through the ``device_uuid`` sysfs entry.
>>> + *
>>> + *    $ cat /sys/bus/pci/devices/<bdf>/device_uuid
>>> + */
>>> +
>>> +static ssize_t
>>> +device_uuid_show(struct device *dev, struct device_attribute *attr, char *buf)
>>> +{
>>> +    struct pci_dev *pdev = to_pci_dev(dev);
>>> +    struct xe_device *xe = pdev_to_xe_device(pdev);
>>> +
>>> +    return sysfs_emit(buf, "%016llx\n", xe->device_uuid);
>> %pU ?
> Not needed.
>>
>>> +}
>>> +static DEVICE_ATTR_RO(device_uuid);
>>> +
>>> +static struct attribute *device_uuid_attrs[] = {
>>> +    &dev_attr_device_uuid.attr,
>>> +    NULL
>>> +};
>>> +
>>> +static const struct attribute_group device_uuid_attr_group = {
>>> +    .attrs = device_uuid_attrs,
>>> +};
>>> +
>>>   int xe_device_sysfs_init(struct xe_device *xe)
>>>   {
>>>       struct device *dev = xe->drm.dev;
>>> @@ -285,5 +315,11 @@ int xe_device_sysfs_init(struct xe_device *xe)
>>>               return ret;
>>>       }
>>>   +    if (xe->info.platform == XE_CRESCENTISLAND) {
>> use .is_visible instead
> 
> Will add has flag.

to be clear: my point was to move this check to the

	attribute_group.is_visible

and I don't care whether it will use .has_uid or .platform field


> 
> Thanks,
> Badal
> 
>>
>>> +        ret = devm_device_add_group(dev, &device_uuid_attr_group);
>>> +        if (ret)
>>> +            return ret;
>>> +    }
>>> +
>>>       return 0;
>>>   }
>>> diff --git a/drivers/gpu/drm/xe/xe_device_types.h b/drivers/gpu/drm/xe/xe_device_types.h
>>> index 180d450a6deb..6fdd16eac906 100644
>>> --- a/drivers/gpu/drm/xe/xe_device_types.h
>>> +++ b/drivers/gpu/drm/xe/xe_device_types.h
>>> @@ -257,6 +257,9 @@ struct xe_device {
>>>           bool oob_initialized;
>>>       } wa_active;
>>>   +    /** @device_uuid: unique 64-bit identifier read from CRI_DEVICE_UUID */
>>> +    u64 device_uuid;
>>> +
>>>       /** @survivability: survivability information for device */
>>>       struct xe_survivability survivability;
>>>