BTW, the patch set bases on Rev 13899
changeset: 13899:df25547d7638
user: Ian Campbell <ian.campbell@xensource.com>
date: Tue Feb 13 15:32:25 2007 +0000
summary: [LINUX] Call ctrl_alt_del() to trigger a reboot so we benefit
Thanks,
Kevin
>From: Tian, Kevin
>Sent: 2007年2月14日 17:10
>
>Hi, all,
> Attached patch set is the first support to S3 (suspend to ram)
>on
>Xen environment. We know that 3.0.5 is feature freeze now, and thus
>this version is just for trial and review first.
>
> X86-64 is to go (failed to compile), but we have a preliminary
>support in house which is however on an old version and not very
>stable yet. It''s supposed to be out with next version.
>
> Totally 12 patches are attached. All files pulled from Linux are
>
>based on current xenlinux version (2.6.18).
>
>====[Xen specific changes]===>[PATCH 1/12] Add suspend/resume to devices
owned by Xen
>[PATCH 2/12] Pull necessary Linux PM files to Xen
>[PATCH 3/12] Provide basic Xen PM infrastructure
>[PATCH 4/12] PM interface between dom0 and Xen
>[PATCH 5/12] Add HVM support
>
>====[Linux specific changes]===>[PATCH 6/12] Make kthreads freezable
>[PATCH 7/12] Pull Linux PM files to sparse tree
>[PATCH 8/12] Enable ACPI sleep in Linux
>
>====[Xen SMP related changes]===>[PATCH 9/12] Pull Linux CPU hotplug
logic to Xen
>[PATCH 10/12] Add Xen CPU hotplug support
>[PATCH 11/12] Allow vcpu to pause self
>(May not require, if force all cpus to sync lazy state)
>[PATCH 12/12] SMP support to Xen PM
>
> For each patch mail, we also post in the body for easy review,
>except those simply pulling Linux files.
>
> Thanks a lot to guidance from Jun Nakajima.
>
>========================================> Some important notes are
attached here, and it''s better to read
>them carefully before you try it out, since Linux itself still has much
>thing to do with a stable PM support.
>
>#1. Ensure native linux (with same version as xenlinux) working on
>your environment first. Xen can never be more stable than native
>linux, thus please DO record all the information to make native Linux
>working stably, including kernel configuration changes, environmental
>setting, BIOS setting or any other tricky points you made. Then DO
>apply all these changes/settings to xen environment!!!
>
>#2. SATA driver suspend/resume was unstable in earlier linux
>version, which is said to have full support since 2.6.19. However in
>our side, SATA does work correctly if AHCI is disabled by configuring
>SATA as Enhanced/IDE mode for linux-2.6.18. If you see SATA failure
>unfortunately like "SATA: I/O request error", you have to switch
to
>legacy IDE mode, and use traditional IDE drivers to operate disk.
>
>#3. VGA resume is the real boring issue, which may even crash the
>system on some boxes. The best case is that your BIOS support
>automatic VGA post after resume. If yes, congratulations! Or else,
>please read carefully linux-2.6.18/Documentation/power/video.txt,
>which lists some known workaround on some known types. Some
>supplements here:
> a) Xen also support "acpi_sleep=" option. You can add it to the
>xenlinux cmdline if applied. We just verified
"acpi_sleep=s3_bios",
>which works on one box but failed another. "acpi_sleep=s3_mode"
>seems no effect on either box we have.
>
> b) On some boxes, VGA post have to be done in UP mode (For
>xen, it means UP-xenlinux), or else system may crash. S3_bios is a
>choice if applying to your box, since bios call is done in wakeup stub.
>Or else in case you have to use "vbetool post" from user space,
one
>possible workaround is to manual unplug other cpus before suspend
>and re-plug them after VGA is resumed by:
> #echo "0" > /sys/devices/system/cpu/cpuN/online
> #echo "mem" > /sys/power/state
> ----After resume to user space----
> #vbetool post
> ----Wait til VGA is fully resumed----
> #echo "1" > /sys/devices/system/cpu/cpuN/online
> I guess such issue may be related to vm86 emulation under
>SMP, for both Xen and Linux. But not sure for now. :-(
>
> Just reiterative NOTE here, #2 and #3 both applies to native
>Linux and Xen. So MAKE SURE you have a STABLE native result first.
>It''s better to run some workload when testing stability. For
example,
>kernel build is easy to trigger instability when VGA resume method is
>incorrect.
>
>Thanks,
>Ke/Kevin
>
>_______________________________________________
>Xen-devel mailing list
>Xen-devel@lists.xensource.com
>http://lists.xensource.com/xen-devel
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel