Linux To Begin Tightening Up Ability To Write To CPU MSRs From User-Space

[phoronix]

Phoronix: Linux To Begin Tightening Up Ability To Write To CPU MSRs From User-Space

The Linux 5.9 kernel is slated to begin introducing new restrictions on allowing writes to CPU model specific registers (MSRs) from user-space...

http://www.phoronix.com/scan.php?page=news_item&px=Linux-Filter-Tightening-MSRs

[birdie]

CPU undervolting is already not available under Linux - yeah, let's tighten it up even further. Meanwhile Windows even in secure UEFI mode has no qualms allowing unfettered access to all CPU MSR registers.

[linuxgeex]

I use to use MSR's from userspace to throttle the CPU on IVB to extend my battery life. With Skylake that stopped working... well it still worked but it didn't improve battery live. Instead, lowering the max frequency scaling was more effective. Ie limiting the CPU to 1 core, 2 threads, and 1.6GHz makes my Zenbook UX310ua with 6200U last over 9.5 hours for surfing and editing code. Part of that long life comes from having 24G RAM so the drive is very idle and there's not much IO spinlock activity.

[Danny3]

I think that a miner that I'm using when the computer is idle is using some MSR.
I don't want any other jump through hoops crap.
Sudo password is enough and if the want a kernel parameter they should do the reverse, allow by default setting the MSR with sudo and block it only if some kernel parameter is passed at boot.

[dragorth]

So, are these only very specific registers, specifically for things related to configuration and updating microcode? Or will this require changes to the generated code of GCC and/or LLVM. If the latter, how will this affect performance for such things as games, 3D software such as Maya and Blender?

[dragorth]

Also, Michael, could we get a link to the quote or the list? Thanks.

[coder]

Maybe this would violate some conventions, but it seems like they could at least prevent "poking at MSRs under the kernel's feet" by routing these accesses through the kernel, instead of the driver changing them directly.

[coder]

So, are these only very specific registers, specifically for things related to configuration and updating microcode? Or will this require changes to the generated code of GCC and/or LLVM. If the latter, how will this affect performance for such things as games, 3D software such as Maya and Blender?

No, it said these already require root access, so it's not going to be any registers in the ABI for apps.

[tildearrow]

Typo:,

MSR writes from user-space do not happen. This behavior right now can be toggled via the msr.allow_writes= kernel module paramrter with on/off/default.

[NotMine999]

CPU undervolting is already not available under Linux - yeah, let's tighten it up even further. Meanwhile Windows even in secure UEFI mode has no qualms allowing unfettered access to all CPU MSR registers.

When people in Linux forums attempt to equate Linux features with Windows features, it only goes to prove that some folks should have stayed on Windows and never learned to use Linux.