Linux Continues Crackdown On User-Space Poking CPU MSRs

[phoronix]

Phoronix: Linux Continues Crackdown On User-Space Poking CPU MSRs

The Linux kernel this year has seen new safeguards and efforts aiming to have user-space reduce their arbitrary poking of CPU machine specific registers (MSRs) in the name of security and other handling concerns. That effort has continued on with the Linux 5.11 cycle...

http://www.phoronix.com/scan.php?page=news_item&px=Linux-5.11-Stop-Poking-EPB-MSR

[sheepdestroyer]

I really love i7z to monitor my cpu.
Since 5.10 I've been getting flooded by "msr: Write to unrecognized MSR 0x38d by i7z "
It's a shame because it hasn't seen any commit for a long while and I doubt it will be fixed

[aorth]

I really love i7z to monitor my cpu.
Since 5.10 I've been getting flooded by "msr: Write to unrecognized MSR 0x38d by i7z "
It's a shame because it hasn't seen any commit for a long while and I doubt it will be fixed

Even worse, those of us who rely on throttled to fix the CPU throttling issues on ThinkPad T480 / T480s / X1C6. Now we have to explicitly allow MSR writes at boot time.

[soulsource]

Even worse, those of us who rely on throttled to fix the CPU throttling issues on ThinkPad T480 / T480s / X1C6. Now we have to explicitly allow MSR writes at boot time.

That workaround sounds awful. Shouldn't this rather be solved by a quirk in the kernel?

[birdie]

I really love i7z to monitor my cpu.
Since 5.10 I've been getting flooded by "msr: Write to unrecognized MSR 0x38d by i7z "
It's a shame because it hasn't seen any commit for a long while and I doubt it will be fixed

It's amazing how an Open OS is getting worse in terms of ... freedom vs. Windows 10 which allows to read/write any MSR registers

[kcrudup]

Shouldn't this rather be solved by a quirk in the kernel?

I just worked around it by using a systemd tmpfiles entry:

$  cat /etc/tmpfiles.d/enable_msr_writes.conf
#
# Enable writes to MSR regs (i.e., lenovo_fix/throttled)
#
w /sys/module/msr/parameters/allow_writes - - - - on

Simple enough.

[Jabberwocky]

It's amazing how an Open OS is getting worse in terms of ... freedom vs. Windows 10 which allows to read/write any MSR registers

You might be confusing convenience with freedom.

[birdie]

You might be confusing convenience with freedom.

A nice, I mean effing lame, excuse. I'm not confusing anything, the confusion is strictly in your head. Linux has always prized itself for being completely open and allowing full access to everything and anything. Nowadays Windows allows more access to your HW than once an open OS.

[Theodis]

A nice, I mean effing lame, excuse. I'm not confusing anything, the confusion is strictly in your head. Linux has always prized itself for being completely open and allowing full access to everything and anything. Nowadays Windows allows more access to your HW than once an open OS.

Windows blocks off so much kernel access to user level applications though for the sake of protecting users. If you really want to be free you should be using Temple OS...

The freedom in Linux isn't about what access applications have to the system, it's about the freedom you the user have over the system. Since it's open source, you're completely free to modify the kernel to allow whatever access you want to give to applications. You don't get that with the Windows kernel.

[soulsource]

I just worked around it by using a systemd tmpfiles entry:

$ cat /etc/tmpfiles.d/enable_msr_writes.conf
#
# Enable writes to MSR regs (i.e., lenovo_fix/throttled)
#
w /sys/module/msr/parameters/allow_writes - - - - on

Simple enough.

I meant that a userspace application is needed for proper throttling is a horrible workaround. That functionality should imho be a quirk directly in the kernel, instead of a userspace application writing to MSR.