Hi William and Jan, sorry for the late answer. I'll look into the problem and see what I'm able to do because I can reproduce the problem and also the workaround with the I/O-APIC works here. We already have a bug ticket in our internal database where all available information is collected. Thank you very much for the findings.
Kind regards, Alexander Eichner Am Freitag, den 21.11.2008, 17:26 +0100 schrieb William Schumann: > Alexander, all, > > In my testing, I was unable to reproduce this case with multiple virtual > SATA drives - the symptoms did not appear with 2 and 4 drives. > > Also, the delay seems to clear by itself after just 2+ minutes. > > I used TD test driver tdmgtst, just doing disk and partition discovery. > > Running tdmgtst with truss just after boot, then a second time, I see > calls by lxstat() about hard disk device links that are made the second > run only (/dev/rdsk/c4t0d0s0...15,p0...4). > > I would (roughly) speculate that some basic Solaris device discovery and > device node building is being delayed - possibly due to timeouts in the > SATA device driver under certain conditions. > > William > jan damborsky wrote: > > Hi Alexander, > > > > > > Alexander Eichner wrote: > >> Hi Jan, > >> > >> I found the following bug report for OpenSolaris Express Community > >> Edition, Build 59: > >> http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6533668 > > > > This bug was actually filed against legacy SXCE installer and the > > problem was identified in the part which is designed and implemented > > in different way in Caiman installer. > > > >> > >> and I found a similar report in the OpenSolaris forum: > >> http://opensolaris.org/jive/thread.jspa?threadID=60203 > >> The latter case is a bit different because the one reporting the bug > >> already has some data on the disk. > > > > Agreed. > > > >> > >> I never checked it myself on real hardware because I don't have a blank > >> hard disk here at the moment. > > > > I have tried on Ultra 20 with 'zeroed' SATA disk and I didn't see the > > hang in that case. > > > > I tried to run format(1M) on VB guest in question and it hangs as well. > > After IO-APIC was enabled, that hang disappeared. Based on this, I guess > > that the problem is somewhere between device driver and emulated HW layer > > and the behavior is affected by the fact if IO-APIC is available or not. > > > > I think that somebody who is familiar with that stuff might be able to > > proceed further with the evaluation. Would you like me to file bug for > > this and capture there the observations we have done so far ? > > > > Thank you, > > Jan > > > > > >> > >> Kind regards, > >> Alexander Eichner > >> > >> Am Donnerstag, den 20.11.2008, 19:57 +0100 schrieb jan damborsky: > >>> Hi Alexander, > >>> > >>> thanks for the update ! > >>> I have tried to reproduce that hang outside the installer > >>> and it can be also seen if I use Target Discovery test driver > >>> (it uses the same mechanism as installer for collecting > >>> the information about disk configuration). > >>> truss(1) shows that most of the time is being spent in ioctl > >>> system calls (I have attached complete truss output). > >>> > >>> At this point it doesn't seem that the problem is related to > >>> the installer, but it is likely the hang is happening in device > >>> driver itself and can be triggered when empty SATA drive is > >>> attached to the VB guest and IO-APIC is left disabled. > >>> > >>> You originally mentioned that similar hang could be also seen > >>> on real hardware - do you happen to know more details about that > >>> particular configuration or where it was being observed ? > >>> > >>> Thank you, > >>> Jan > >>> > >>> > >>> Alexander Eichner wrote: > >>>> Hi Jan, > >>>> > >>>> tried with IO-APIC enabled and the problem is not reproducible > >>>> anymore. > >>>> > >>>> Kind regards, > >>>> Alexander Eichner > >>>> > >>>> Am Mittwoch, den 19.11.2008, 16:42 +0100 schrieb jan damborsky: > >>>>> Hi Alexander, > >>>>> > >>>>> > >>>>> Alexander Eichner wrote: > >>>>>> Hi Jan, > >>>>>> > >>>>>> this issue doesn't happen always here. > >>>>>> But sometimes the installer hangs about 1 minute while detecting the > >>>>>> hard disk freezing the GUI completely. > >>>>>> It seems to happen here only if the hard disk is completely > >>>>>> uninitialized. > >>>>> Yes, I can reproduce this if new disk is created and attached as > >>>>> SATA. > >>>>> > >>>>>> I tested it also with 10 disks attached but they all had a partition > >>>>>> with ZFS and the installer detected them immediately (the spinning > >>>>>> circle didn't showed up). > >>>>>> > >>>>>> The guest configuration is: > >>>>>> RAM: 1024MB > >>>>>> ACPI: enabled > >>>>>> IO-APIC: disabled > >>>>> Could you please try if that hang can be seen with 'IO-APIC' > >>>>> enabled ? > >>>>> In my case, if I enable IO-APIC that hang disappears. > >>>>> > >>>>> > >>>>>> AMD-V/VT-x: disabled > >>>>>> PAE/NX: disabled > >>>>>> > >>>>>> My host is a Dell Inspiron 1520 > >>>>>> RAM: 4GB > >>>>>> OS: 64bit Linux (Ubuntu Intrepid Ibex) > >>>>>> > >>>>>> I created a log of the installer with LS_DBG_LVL=4. It is > >>>>>> attached to > >>>>>> this mail. > >>>>> Thanks ! William is looking at this - we can see the delays there. > >>>>> However, they don't seem to be happening in the installer itself, > >>>>> but rather in underlying libdiskmgt library or device driver. Once > >>>>> it is more investigated, we would file bug for this. > >>>>> > >>>>> Thank you, > >>>>> Jan > >>>>> > >>>>>> Kind regards, > >>>>>> Alexander Eichner > >>>>>> > >>>>>> Am Dienstag, den 18.11.2008, 18:28 +0100 schrieb jan damborsky: > >>>>>>> Hi Achim, > >>>>>>> > >>>>>>> > >>>>>>> Achim Hasenmueller wrote: > >>>>>>>> I can reproduce the iostat -En issue (SATA is reported as ATA). > >>>>>>>> I am not sure that this is what's causing the installer to take > >>>>>>>> several minutes discovering the SATA disk. > >>>>>>> It is unlikely this is the reason why you see > >>>>>>> the installer hanging in disk discovery. > >>>>>>> > >>>>>>> I have upgraded from 2.0.2 to 2.0.4 and I am able to > >>>>>>> boot LiveCD in VB guest with SATA disk attached - now > >>>>>>> with VT-x/AMD-V enabled > >>>>>>> (fix for http://www.virtualbox.org/ticket/2292 works fine) > >>>>>>> and install on SATA drive. I didn't encounter delay during > >>>>>>> the disk discovery phase as installer Disk screen was brought > >>>>>>> up almost immediately. > >>>>>>> > >>>>>>> Then I tried several combinations of VT-x/AMD-V, IO APIC, > >>>>>>> ... options and could see the delay when following configuration > >>>>>>> was used: > >>>>>>> > >>>>>>> ACPI: Enabled > >>>>>>> IO APIC: Disabled > >>>>>>> VT-x/AMD-v: Disabled > >>>>>>> PAE/NX: Enabled > >>>>>>> > >>>>>>> So it seems that the problem with hang might be specific > >>>>>>> to particular host/guest configuration or/and VB version. > >>>>>>> > >>>>>>> Do you happen to know in which configuration you are seeing > >>>>>>> the hang ? > >>>>>>> Is there any difference if you enable 'IO APIC' or play > >>>>>>> with other options ? > >>>>>>> > >>>>>>> Thank you, > >>>>>>> Jan > >>>>>>> > >>>>>>> > >>>>>>>> If it is, I'd strongly suggest to get the fix into 2008.11. > >>>>>>>> > >>>>>>>> Achim. > >>>>>>>> > >>>>>>>> On Nov 18, 2008, at 3:15 PM, jan damborsky wrote: > >>>>>>>> > >>>>>>>>> Hi Achim, Alexander, > >>>>>>>>> > >>>>>>>>> as far as problem seen in VirtualBox is concerned, could you > >>>>>>>>> please > >>>>>>>>> provide more details about configuration you used when > >>>>>>>>> encountering > >>>>>>>>> that issue (HW, host operating system, how much memory you > >>>>>>>>> dedicated to the guest, other settings - VT-x/AMD-V,ACPI, ...) ? > >>>>>>>>> > >>>>>>>>> There were couple of SATA issues reported in 2.0.2, which are > >>>>>>>>> fixed > >>>>>>>>> in 2.0.4: > >>>>>>>>> > >>>>>>>>> http://www.virtualbox.org/ticket/2292 > >>>>>>>>> http://www.virtualbox.org/ticket/2182 > >>>>>>>>> > >>>>>>>>> I have tried to launch the installer on VB 2.0.2 guest (1GB > >>>>>>>>> memory) > >>>>>>>>> with one SATA disk attached and I haven't seen that hang in > >>>>>>>>> disk discovery - > >>>>>>>>> Disk screen was displayed immediately. > >>>>>>>>> I haven't tried the installation. > >>>>>>>>> > >>>>>>>>> I have used following configuration: > >>>>>>>>> > >>>>>>>>> host: > >>>>>>>>> ----- > >>>>>>>>> HW: Lenovo Thinkpad X61 (4GB RWM) > >>>>>>>>> SW: OpenSolaris build 101a > >>>>>>>>> VB: 2.0.2 > >>>>>>>>> > >>>>>>>>> guest: > >>>>>>>>> ------ > >>>>>>>>> base memory: 1024 MB > >>>>>>>>> OS Type: OpenSolaris > >>>>>>>>> ACPI: Enabled > >>>>>>>>> IO APIC: Enabled > >>>>>>>>> VT-x/AMD-v: disabled (workaround for bug 2292) > >>>>>>>>> PAE/NX: Enabled > >>>>>>>>> > >>>>>>>>> > >>>>>>>>> With respect to problems with SATA disk discovery hanging on > >>>>>>>>> bare metal, > >>>>>>>>> do you happen to know, which HW configuration might be suffering > >>>>>>>>> from this issue ? > >>>>>>>>> > >>>>>>>>> Thank you, > >>>>>>>>> Jan > >>>>>>>>> > >>>>>>>>> > >>>>>>>>> William Schumann wrote: > >>>>>>>>>> There are two separate bugs that are addressing problems with > >>>>>>>>>> library libdiskmgt correctly reporting type on SATA drives > >>>>>>>>>> for real hardware: > >>>>>>>>>> > >>>>>>>>>> 6558646 DM_CTYPE attribute is set to "unknown" for SATA > >>>>>>>>>> controllers handled by "sata" driver > >>>>>>>>>> http://monaco.sfbay/detail.jsf?cr=6558646 > >>>>>>>>>> > >>>>>>>>>> auto-installer can not identify SATA disk type (most likely a > >>>>>>>>>> duplicate of the former) > >>>>>>>>>> http://defect.opensolaris.org/bz/show_bug.cgi?id=4387 > >>>>>>>>>> > >>>>>>>>>> The former has been in the 'accept' state since June. If > >>>>>>>>>> libdiskmgt bugs are now affecting VirtualBox, they should > >>>>>>>>>> perhaps be escalated (currently P3). > >>>>>>>>>> > >>>>>>>>>> I will inquire about status of fixes for libdiskmgt - they > >>>>>>>>>> are not presently being handled by the install group. > >>>>>>>>>> William Schumann > >>>>>>>>>> > >>>>>>>>>> Dave Miner wrote: > >>>>>>>>>>> Vincent Murphy wrote: > >>>>>>>>>>>> Achim, > >>>>>>>>>>>> > >>>>>>>>>>>> Best person to contact is Dave Miner - however, I would > >>>>>>>>>>>> also send mail to indiana-eng as this may be a known issue. > >>>>>>>>>>>> If not then raise a bug :-) > >>>>>>>>>>>> > >>>>>>>>>>> Achim, Alexander: William Schumann on the install team is > >>>>>>>>>>> the best person to work with on disk discovery issues; he's > >>>>>>>>>>> in Prague and should be back in the office tomorrow since > >>>>>>>>>>> today's a holiday there. I've copied him. > >>>>>>>>>>> > >>>>>>>>>>> I believe I've seen this issue, though I rarely have used > >>>>>>>>>>> SATA in my vbox instances, and I'm not aware of any existing > >>>>>>>>>>> bug reports related to it. > >>>>>>>>>>> > >>>>>>>>>>> Dave > >>>>>>>>>>> > >>>>>>>>>>>> Thanks > >>>>>>>>>>>> Vincent > >>>>>>>>>>>> > >>>>>>>>>>>> On 11/17/08 12:46, Achim Hasenmueller wrote: > >>>>>>>>>>>>> Hi Vincent, > >>>>>>>>>>>>> > >>>>>>>>>>>>> when installing OpenSolaris (2008.5 or .11) in a > >>>>>>>>>>>>> VirtualBox VM with a SATA disk configured for the VM, the > >>>>>>>>>>>>> installer takes several minutes to discover the disk. > >>>>>>>>>>>>> > >>>>>>>>>>>>> We have been trying to debug this from our end but without > >>>>>>>>>>>>> success. Also we have found reports on the web about > >>>>>>>>>>>>> people having the same issue with physical hardware. > >>>>>>>>>>>>> Therefore, we now need help from the OS team. Who in your > >>>>>>>>>>>>> team could Alexander work with? > >>>>>>>>>>>>> > >>>>>>>>>>>>> Thanks! > >>>>>>>>>>>>> -- > >>>>>>>>>>>>> Achim Hasenmueller > >>>>>>>>>>>>> Director Engineering, VirtualBox > >>>>>>>>>>>>> Sun Microsystems GmbH > >>>>>>>>>>>>> Werkstrasse 24 > >>>>>>>>>>>>> 71384 Weinstadt, Germany > >>>>>>>>>>>>> phone: +49 7151 604050 > >>>>>>>>>>>>> > >>>>>>>>>>>>> > >>>>>>>>>>>>> ================================================ > >>>>>>>>>>>>> Sitz der Gesellschaft: Sun Microsystems GmbH, > >>>>>>>>>>>>> Sonnenallee 1, 85551 Kirchheim-Heimstetten > >>>>>>>>>>>>> Amtsgericht Muenchen: HRB 161028 > >>>>>>>>>>>>> Geschaeftsfuehrer: Thomas Schroeder, > >>>>>>>>>>>>> Wolfgang Engels, Dr. Roland Boehmer > >>>>>>>>>>>>> Vorsitzender des Aufsichtsrates: Martin Haering > >>>>>>>>>>>>> ================================================ > >>>>>>>>>>>>> > >>>>>>>>>>>>> > >> > >
