In general, and without trying to break it, HVM save/restore works quite well... There is one obvious and immediate problem: The SDL display is (correctly) removed as part of the first save, but restore never restores the display. The more complex but rarer problem: However, my attempts to break it does indeed break it after some random amount of time. The breaking is in the form of relatively heavy disk activity whilst looping over save/restore/sleep. The application in this case is a simple C-program run on sles 9.3 (32bit, HVM, 128MB on 64-bit hypervisor) generates a file of 25MB with a incrementing pattern, then starts reading the data and writing the inverse back, in blocks of 512 bytes, checking that the read data is what''s expected. The current behaviour indicates that the guest gets into a state where it''s idle, but no longer communicating on the ssh-connection, and from what I can tell, nor on the disk-io. Trying to bring up a VNC display doesn''t work either. Since I loose the SDL console display, I can''t see if there''s any error/warning messages from the guest. It is entirely possible that the disk-io is continuing, but the network connection to the guest is DEFINITELY lost. I''m at a loss on even where to start looking at the problem - one thing I do know: the guest is not killed and restarted, because I''ve set "on_crash=destroy" in the configuration file. I believe I can also reproduce the same problem with a "simple guest" (single executable file loaded instead of hvmloader that executes disk-testing to a disk-image that is filled with a running 32-bit count [writen back inverted). This guest also fails to properly operate after some number of save/restore cycles, but I''m not 100% sure if this is a problem with the guest itself not handling aborted IDE operations in some way - this is why I created the disk-io in Linux application above. -- Mats _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
At 15:53 +0200 on 20 Apr (1177084394), Petersson, Mats wrote:> There is one obvious and immediate problem: > The SDL display is (correctly) removed as part of the first save, but > restore never restores the display.It looks at the moment as if the guest is restores with a VNC display, so you should at least be able to see the console when the other hang occurs, by connecting a VNC client to localhost:5900. I''ll have a look at the xend runes that are meant to track the display options. Cheers, Tim. -- Tim Deegan <Tim.Deegan@xensource.com>, XenSource UK Limited Registered office c/o EC2Y 5EB, UK; company number 05334508 _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
> -----Original Message----- > From: Tim Deegan [mailto:Tim.Deegan@xensource.com] > Sent: 20 April 2007 15:06 > To: Petersson, Mats > Cc: xen-devel@lists.xensource.com > Subject: Re: Problems with HVM Save/Restore. > > At 15:53 +0200 on 20 Apr (1177084394), Petersson, Mats wrote: > > There is one obvious and immediate problem: > > The SDL display is (correctly) removed as part of the first > save, but > > restore never restores the display. > > It looks at the moment as if the guest is restores with a VNC display, > so you should at least be able to see the console when the other hang > occurs, by connecting a VNC client to localhost:5900.You are right, using VNC works fine [although it feels slower in some cases than SDL, and I''ve previously had problems with VNC which is why I was using SDL in the first place]. -- Mats> > I''ll have a look at the xend runes that are meant to track the display > options. > > Cheers, > > Tim. > > -- > Tim Deegan <Tim.Deegan@xensource.com>, XenSource UK Limited > Registered office c/o EC2Y 5EB, UK; company number 05334508 > > >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
> I believe I can also reproduce the same problem with a "simple guest" > (single executable file loaded instead of hvmloader that executes > disk-testing to a disk-image that is filled with a running 32-bitcount> [writen back inverted). This guest also fails to properly operateafter> some number of save/restore cycles, but I''m not 100% sure if this is a > problem with the guest itself not handling aborted IDE operations in > some way - this is why I created the disk-io in Linux applicationabove. What happens if you build and use the PV Linux drivers for SLES9.3 ? Best, Ian _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
> -----Original Message----- > From: xen-devel-bounces@lists.xensource.com > [mailto:xen-devel-bounces@lists.xensource.com] On Behalf Of Ian Pratt > Sent: 20 April 2007 15:31 > To: Petersson, Mats; xen-devel@lists.xensource.com > Cc: Tim Deegan; Woller, Thomas > Subject: RE: [Xen-devel] Problems with HVM Save/Restore. > > > I believe I can also reproduce the same problem with a > "simple guest" > > (single executable file loaded instead of hvmloader that executes > > disk-testing to a disk-image that is filled with a running 32-bit > count > > [writen back inverted). This guest also fails to properly operate > after > > some number of save/restore cycles, but I''m not 100% sure > if this is a > > problem with the guest itself not handling aborted IDE operations in > > some way - this is why I created the disk-io in Linux application > above. > > What happens if you build and use the PV Linux drivers for SLES9.3 ?Haven''t tried that. I''ll see if I can cobble that together on Monday. -- Mats> > Best, > Ian > > _______________________________________________ > 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
> -----Original Message----- > From: xen-devel-bounces@lists.xensource.com > [mailto:xen-devel-bounces@lists.xensource.com] On Behalf Of > Petersson, Mats > Sent: 20 April 2007 16:04 > To: Ian Pratt; xen-devel@lists.xensource.com > Cc: Tim Deegan; Woller, Thomas > Subject: RE: [Xen-devel] Problems with HVM Save/Restore. > > > > > -----Original Message----- > > From: xen-devel-bounces@lists.xensource.com > > [mailto:xen-devel-bounces@lists.xensource.com] On Behalf Of > Ian Pratt > > Sent: 20 April 2007 15:31 > > To: Petersson, Mats; xen-devel@lists.xensource.com > > Cc: Tim Deegan; Woller, Thomas > > Subject: RE: [Xen-devel] Problems with HVM Save/Restore. > > > > > I believe I can also reproduce the same problem with a > > "simple guest" > > > (single executable file loaded instead of hvmloader that executes > > > disk-testing to a disk-image that is filled with a running 32-bit > > count > > > [writen back inverted). This guest also fails to properly operate > > after > > > some number of save/restore cycles, but I''m not 100% sure > > if this is a > > > problem with the guest itself not handling aborted IDE > operations in > > > some way - this is why I created the disk-io in Linux application > > above. > > > > What happens if you build and use the PV Linux drivers for SLES9.3 ? > > Haven''t tried that. I''ll see if I can cobble that together on Monday.By the way, I presume you''re not really implying that the IDE controller shouldn''t survive multiple save/restore cycles? I''m pretty sure there will be customers wanting to run less common OS''s that don''t have PV-on-HVM drivers readily available (and since the drivers aren''t publicly available for non-Linux OS''s, they would require some pretty qualified staff to produce PV-drivers). -- Mats> > -- > Mats > > > > Best, > > Ian > > > > _______________________________________________ > > 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 > > >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On 20/4/07 16:22, "Petersson, Mats" <Mats.Petersson@amd.com> wrote:>>> What happens if you build and use the PV Linux drivers for SLES9.3 ? >> >> Haven''t tried that. I''ll see if I can cobble that together on Monday. > > By the way, I presume you''re not really implying that the IDE controller > shouldn''t survive multiple save/restore cycles? I''m pretty sure there > will be customers wanting to run less common OS''s that don''t have > PV-on-HVM drivers readily available (and since the drivers aren''t > publicly available for non-Linux OS''s, they would require some pretty > qualified staff to produce PV-drivers).Well yes, of course it should survive. If it doesn''t, that''s a bug. :-) -- Keir _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
At 16:28 +0200 on 20 Apr (1177086531), Petersson, Mats wrote:> You are right, using VNC works fine [although it feels slower in some > cases than SDL, and I''ve previously had problems with VNC which is why I > was using SDL in the first place].Fair enough; I find SDL''s habit of grabbing the X pointer and crashing far more annoying than anything VNC ever did to me. :) Anyway, cset 14884:23cf7d6c30f4 of -unstable should fix SDL across save/restore. Cheers, Tim. -- Tim Deegan <Tim.Deegan@xensource.com>, XenSource UK Limited Registered office c/o EC2Y 5EB, UK; company number 05334508 _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
> -----Original Message----- > From: xen-devel-bounces@lists.xensource.com > [mailto:xen-devel-bounces@lists.xensource.com] On Behalf Of Ian Pratt > Sent: 20 April 2007 15:31 > To: Petersson, Mats; xen-devel@lists.xensource.com > Cc: Tim Deegan; Woller, Thomas > Subject: RE: [Xen-devel] Problems with HVM Save/Restore. > > > I believe I can also reproduce the same problem with a > "simple guest" > > (single executable file loaded instead of hvmloader that executes > > disk-testing to a disk-image that is filled with a running 32-bit > count > > [writen back inverted). This guest also fails to properly operate > after > > some number of save/restore cycles, but I''m not 100% sure > if this is a > > problem with the guest itself not handling aborted IDE operations in > > some way - this is why I created the disk-io in Linux application > above. > > What happens if you build and use the PV Linux drivers for SLES9.3 ?An initial attempt on this fails due to the xen-platform-pci.ko requiring a "system_state" reference, which isn''t exported by the kernel version that my SLES9.3 is built from (2.6.5-7). I''ll have a look to see if I can patch around this problem. -- Mats> > Best, > Ian > > _______________________________________________ > 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