Announcement

Collapse
No announcement yet.

Many Power Management Updates For The Linux 4.10 Kernel

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts
    gilboa
    Senior Member

  • gilboa
    replied
    Originally posted by indepe View Post

    Sounds good, does it happen to be open source?
    Hopefully, down the road.

    Specifically atomic counters work well in std C, though, or are they in some way special? Just curious...
    Back when I started working on it, it wasn't in std library.
    Plus, I rather not be limited to modern compilers (I still support older GCC and MSVC).

    [Hope I'm not derailing the thread, but it's a quiet older one anyway.]
    *We* are, but as you pointed out, its a quiet thread

    - Gilboa

    Leave a comment:

  • indepe
    Premium Member

  • indepe
    replied
    Originally posted by gilboa View Post

    Actually, I have such library for simple stuff, rdtsc, atomic counters, atomic bitmaps, etc.
    Sounds good, does it happen to be open source?

    Specifically atomic counters work well in std C, though, or are they in some way special? Just curious...

    [Hope I'm not derailing the thread, but it's a quiet older one anyway.]

    Leave a comment:

  • gilboa
    Senior Member

  • gilboa
    replied
    Originally posted by indepe View Post

    Too good to be true!
    Actually, I have such library for simple stuff, rdtsc, atomic counters, atomic bitmaps, etc.

    - Gilboa

    Leave a comment:

  • indepe
    Premium Member

  • indepe
    replied
    Originally posted by gilboa View Post

    Yeah, but nothing beats having a cross OS / platform / user-mode/kernel-mode ASM library
    Too good to be true!

    Leave a comment:

  • gilboa
    Senior Member

  • gilboa
    replied
    Originally posted by indepe View Post

    By the way, would be nice to have a portable userspace lib function around RDTSC as well, if such a thing is possible.
    Yeah, but nothing beats having a cross OS / platform / user-mode/kernel-mode ASM library

    - Gilboa

    Leave a comment:

  • indepe
    Premium Member

  • indepe
    replied
    Regarding power management:

    It would seem to make sense if threads were able to switch between performance / power-efficient / default modes.

    In default mode, the scaling governor would determine the frequency as it does now.

    In power-efficient mode, a long-running task can run at the frequency optimal for power-consumption, even if its CPU-usage will be 100%.

    In performance mode, it will run at the best available frequency, even if it runs only for short time intervals and averages a low CPU-usage.

    Leave a comment:

  • indepe
    Premium Member

  • indepe
    replied
    Originally posted by gilboa View Post

    I'm not sure I understand your request: You want the kernel to enforce zero-PM on machines that doesn't support constant TSC? Why? Kernel / user-land developers on such platforms should be aware of this limitation and use other clock sources instead of TSC..
    By the way, would be nice to have a portable userspace lib function around RDTSC as well, if such a thing is possible.
    indepe
    Premium Member
    Last edited by indepe; 20 December 2016, 09:22 AM.

    Leave a comment:

  • gilboa
    Senior Member

  • gilboa
    replied
    Originally posted by Azrael5 View Post
    what about TSC which is the default source clock used by linux operating systems on the different machins!? When the cpus implements constant TSC (not invariant) the C1E and or EIST should be disabled!?
    I'm not sure I understand your request: You want the kernel to enforce zero-PM on machines that doesn't support constant TSC?
    Why? Kernel / user-land developers on such platforms should be aware of this limitation and use other clock sources instead of TSC...

    - Gilboa

    Leave a comment:

  • indepe
    Premium Member

  • indepe
    replied
    Originally posted by davidbepo View Post
    pstate + schedutil sounds nice
    Especially if that helps schedutil to provide the same performance as "performance", when needed.

    I hope the final version of this won't require another flavor of boot parameter ("intel_pstate=passive") as the experimental version apparently did, thereby making all existing options inaccessible. Who would even know about this feature?

    Leave a comment:

  • davidbepo
    Senior Member

  • davidbepo
    replied
    pstate + schedutil sounds nice

    Leave a comment:

Working...
X