Linux 6.6 Will Be Able To Handle Temperature Reporting When Having More Than 32 DIMMs
The Linux kernel's "dimmtemp" driver allows for reporting memory temperatures with capable memory modules and when exposed by the Intel processor's PECI (Platform Environment Control Interface). Currently though the DIMM temperature driver is hard-coded to only allow reporting up to 32 DIMMs while a change queued for Linux 6.6 will extend that limit.
While it's rare for servers these days to have more than 32 DIMMs installed, it's possible with Intel Xeon Scalable Sapphire Rapids servers supporting up to eight CPU sockets, each potentially with 16 DIMMs per socket. The 9elements consulting firm who authored this dimmtemp change to allow for more than 32 DIMMs for temperature reporting did successfully test the patched driver on a server with 64 DIMMs between four Sapphire Rapids processors.
For those interested, more details on this change to handle 32+ DIMMs for the dimmtemp hardware monitoring driver via this commit. The change made it into the HWMON subsystem's "-next" branch so it will be part of the upcoming Linux 6.6 kernel cycle.
While it's rare for servers these days to have more than 32 DIMMs installed, it's possible with Intel Xeon Scalable Sapphire Rapids servers supporting up to eight CPU sockets, each potentially with 16 DIMMs per socket. The 9elements consulting firm who authored this dimmtemp change to allow for more than 32 DIMMs for temperature reporting did successfully test the patched driver on a server with 64 DIMMs between four Sapphire Rapids processors.
For those interested, more details on this change to handle 32+ DIMMs for the dimmtemp hardware monitoring driver via this commit. The change made it into the HWMON subsystem's "-next" branch so it will be part of the upcoming Linux 6.6 kernel cycle.
12 Comments