Linux 5.13 Will Stop Restricting CPU Power Metrics Access For AMD Energy Driver

[phoronix]

Phoronix: Linux 5.13 Will Stop Restricting CPU Power Metrics Access For AMD Energy Driver

Following the PLATYPUS discovery last year that CPU energy information could be used for possible side channel attacks, the Intel RAPL counters were not only restricted to root but the "amd_energy" driver for exposing CPU energy information on supported Zen series CPUs was also dialed back to root-only in the name of security. Linux 5.13 is introducing a new mechanism so AMD CPUs will be able to still read the energy counters even if not operating as root...

https://www.phoronix.com/scan.php?page=news_item&px=Linux-5.13-No-Root-AMD-Energy

[waxhead]

One day.... one day there will perhaps be hardware accelerated CPU's, rotating SSD's, Dot Matrix 9 needle RAM, Solid State GPU's with 9pin joystick ports, wireless keys on your keyboard and blueray powered monitors with lossless poweroff buttons... that will be the future for sure. I was actually looking for a potential AMD replacement for my old trusty Core 2 Duo, but ended up dropping the entire thing as there is so many fancy things, management engines, attack vectors, tons of sockets and insane power requirements. Perhaps it is best to wait until ARM or perhaps even RISC-V takes over everything. Hopefully things will be either on or off and maybe... just maybe a little bit simpler. X86 based stuff are starting to be messy!

[Danny3]

Oh come on!
Who the fuck comes with these imbecile ideas to restrict sensors to root ?
Why the fuck should I run conky or other monitor tools as root if I just want to see a few temperatures ?
I hate these idiotic ideas of requiring root or showing inaccurate results.
It seems the there's not a month or two without being bitten by Linux's crappy security model.

I don't know what it's so fucking hard to just create a permissions system where we allow | deny for each program to access hardware devices (webcam, mike, sensors, etc) instead of just running them as root where they have the permission to just destroy everything ?

If I were to design an Linux based OS I would do it like this:
1. Kernel give maximum permission access to only one program, the init manager
2. The init manager gives maximum permissions to only one user type (the admin)
3. The admin chooses the permissions for each program, which hardware they can access.

I find really stupid the two-level permission model (root or non-root)

[ezst036]

.............I was actually looking for a potential AMD replacement for my old trusty Core 2 Duo, but ended up dropping the entire thing.............

Careful there. You have a bunch of people running around here using "ancient hardware" as an all encompassing catch-all for pretty much "any hardware older than theirs" and would probably scoff at your more than capable Core 2. You might just find yourself cut off prematurely!

[aufkrawall]

It's really stupid that you have to run s-tui as root in order to poll power sensor. Total overkill, just like default CPU mitigation handling on Linux in general ...

[tildearrow]

One day.... one day there will perhaps be hardware accelerated CPU's,

May be actually possible. Let me introduce.... the Smart Gate Array architecture!
An FPGA inside the processor automatically programs itself to optimize heavily repeated portions of code in the CPU cache, speeding up operations by a factor of up to 10.

rotating SSD's,

The world's first rotating SSD turns out to be a commercial failure as its seek times and low random read speeds make it massively inferior to regular SSDs.

Dot Matrix 9 needle RAM,

Looks cool, but was also a commercial failure.

Solid State GPU's with 9pin joystick ports,

All GPUs are solid state by definition.
And if you meant a GPU with an SSD, check out the Radeon Pro SSG.

wireless keys on your keyboard

Laser projection keyboards. The keys are wireless.

and blueray powered monitors with lossless poweroff buttons... that will be the future for sure.

Luxury TV powered by lasers to output the image... $20 million

[ssokolow]

Perhaps it is best to wait until ARM or perhaps even RISC-V takes over everything.

AMD's analogue to the Intel Management Engine, the Platform Security Processor, is a licensed ARM TrustZone core. ARM beat AMD to running code you don't control at Ring -3.

[skeevy420]

Rotating SSDs already exist

[waxhead]

May be actually possible. Let me introduce.... the Smart Gate Array architecture!
An FPGA inside the processor automatically programs itself to optimize heavily repeated portions of code in the CPU cache, speeding up operations by a factor of up to 10.

Hehe, I actually thought of all your points when writing what I did, but in all seriousness. I have often wondered what would happen if they implemented a JIT compiler in microcode or something similar. Quite a few years ago I remember I noticed that a program ran faster under a JIT compller than on bare metal. I can't remember what it was , but I think it was WINUAE running something under VirtualBox or something similar. All I remember pretty clearly is that whatever I ran was CPU bound so that surprised me a bit.

Anyway that illustrates that the x86/amd64 architecture is not as efficient as it potentially could be and that is perhaps also the reason for many of the attacks. I still live in the dark ages, but unless I am mistaking modern processors translate the ISA to another RISC-ish load/store ISA before it even hits the microcode. That is a lot of complexity so no wonder security experts make a living

[mifritscher]

@waxshead: Do you mean Transmeta?