Skip to content

Conversation

@bges
Copy link

@bges bges commented Oct 29, 2025

Add a new channel to support SBTSIx50 and SBTSIx54 for reporting maximum HBM temperature.

Test result on MI300C platform
CPU temperature:
$ cat /sys/bus/i3c/drivers/sbtsi_i3c/0-22400000001/hwmon/hwmon16/temp1_input
41750

Max HBM Temperature:
$ cat /sys/bus/i3c/drivers/sbtsi_i3c/0-22400000001/hwmon/hwmon16/temp2_input
32000

Copy link
Collaborator

@akky16 akky16 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thank you for raising the PR, can you please help to address few review comments?

Add a new channel to support SBTSIx50 and SBTSIx54 for reporting maximum HBM temperature.
@bges bges force-pushed the add_hbm_temp_sensor branch from 229e87a to 6617133 Compare November 6, 2025 07:42
@bges
Copy link
Author

bges commented Nov 6, 2025

Add the Label, Max and Min:

# cat temp1_label 
CpuTemp
# cat temp1_input 
42750
# cat temp1_max   
98000
# cat temp1_min 
0

# cat temp2_label 
MaxMemTemp
# cat temp2_input 
33000
# cat temp2_max   
98000
# cat temp2_min 
1000

@bges bges requested a review from akky16 November 13, 2025 03:19
@nchatrad
Copy link
Collaborator

Akshay, TSI registers 0x50 and 0x54 do not report any value on EPYC processors, SB-TSI driver has no way to know the platform to conditionally support the hwmon entries.

HBM temp information is available via APML, DIMM temp info is also available for EPYC processors, as SBRMI driver is aware of the platform via CPUID or simliar. can we expose hwmon entries for HBM/DIMM temp from SBRMI instead.

@akky16
Copy link
Collaborator

akky16 commented Dec 2, 2025

SBRMI through mailbox provides way to get the max HBM temp (highest of all stacks) using mailbox command 0xA1,
will implement the required changes in SBRMI to support this.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants