Garrett D'Amore <gdam...@sun.com> wrote: > I know that all USB host controllers interrupt *really* frequently. > (1000Hz.) This is a flaw in the way those interfaces were designed. Its > true when *any* device is plugged into the port, I believe.
It's not ture on a Linux system. Plugging a USB stick onto a Linux system, we still can enter deep C-state. > > I'd expect that if you unplug all USB devices, you should be able to see > deep C-states. If you can't, then it does sound like a potential bug in the > EHCI driver. > Conditionally, some system is okay like supermicro's NHM-EP. some system is bad like Demodepot's NHM-EP. While Linux works perfect on both systems. If the device is going to be idle/not used, can we put it into a state called "Selective Suspend", instead of doing a regular wake-up to check on the device? Maybe it's already supported and this problem is caused by another case, I just don't know. Sorry I don't have USB background, what I can only do is to do the comparision. Hope USB experts can shed some lights on this issue. This makes a big energy efficiency gap between solaris and other OSes. Thanks, -Aubrey > - Garrett > > (Yes, this is one of the things that I hate about USB. It doesn't really > have a native interrupt facility, but instead requires the host to "poll" > the device very frequently to on an interrupt endpoint to simulate a true > interrupt.) > > On 03/ 7/10 07:18 PM, Aubrey Li wrote: >> >> Hi, >> >> We catched this problem in a solaris power related characterization >> report. >> In an idle system, all the logical CPUs are ~99% in deep C-state while >> the package >> can't enter deep C-state. Theoretically, if all the logical CPUs in a >> package can enter >> deep C-state, the package should be able to enter deep C-state as well. >> >> The cases here: >> 1) If we disabled USB2.0 in the BIOS, the problem is gone. >> 2) Remove EHCI driver and reboot, the problem is gone. >> 3) If we plugged a USB stick onto the box, without any read/write >> operations, the package >> C-state is throttled as well, this is not expected. >> >> If we boot the system with other OS, we didn't see any of these >> problems. So these are more >> likely the software problem. >> >> Any helps will be greatly appreciated. >> >> Thanks, >> -Aubrey >> _______________________________________________ >> driver-discuss mailing list >> driver-discuss@opensolaris.org >> http://mail.opensolaris.org/mailman/listinfo/driver-discuss >> > > _______________________________________________ > driver-discuss mailing list > driver-discuss@opensolaris.org > http://mail.opensolaris.org/mailman/listinfo/driver-discuss > _______________________________________________ driver-discuss mailing list driver-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/driver-discuss
