We saw these issues in recent ChangeSet: 1. Reboot fail in Guest OS: Reboot in guest will close current qemu windows but does not create a new qemu window to restart 32bit HV/xen0: 2. 4 LTP math cases(float_bessel, float_exp_log, float_power and float_trigo) sometime fail. (775) 3. Kernel build in VMX may meet kernel panic. It complains like: "<0> Kernel panic - not syncing: kernel/sched.c:460: spin_lock(kernel/sched.c:c03ea120) already locked by kernel/sched.c/2689" 32bit PAE HV/xen0: 4. Cannot bootup pae SMP VMX domain. guest crashed with attached output. 5. Create 32 guest with balloon driver will hang xen0: If I do not set a fixed size memory using "dom0_mem=" option in grub.conf for dom0, dom0 will hang after a vmx guest is created. 6. 32 SMP Guest only see about 900M memory when assign 1.5G memory for the guest: If I assign 1.5G memory in vmx config file and boot up the guest, I still can only see 903256KB memory using "cat /proc/meminfo" in the guest, though I can see 1.5G memory allocated to the guest. IA32E HV/xen0 7. CPU2K on 64bit guest has a performance regression. INT down from ~95% to 88% and FP down from ~90% to 78% (VMX/Native) 8. 64bit SMP VMX can not be up, if acpi=0 : If try with acpi=0 apic=1 vcpus=2 pae=1 to create FC3 IA32e guest, Guest will kernel panic when booting. 9. FC5 IA32 VMX Guest crash when try to boot SMP Guest: set vcpus=2 acpi=1 apic=1 pae=1 in configure file and create SMP FC5 IA32 VMX guest, guest wil crash with attached output. 10. 32pae SMP guest with 1500MB memory will cause xen0 reboot 11. Bootup 64bit guest with pae=0 in configure file, guest will complain "Your CPU does not support long mode" Bests Jeff _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Puthiyaparambil, Aravindh
2006-Aug-23 15:51 UTC
RE: [Xen-devel] Some issues in recent ChangeSet.
Jeff, Which changeset are you seeing these problems on?> IA32E HV/xen0 > 7. CPU2K on 64bit guest has a performance regression. INT down from > ~95% to 88% and FP down from ~90% to 78% (VMX/Native) > 8. 64bit SMP VMX can not be up, if acpi=0 : If try with acpi=0 apic=1 > vcpus=2 pae=1 to create FC3 IA32e guest, Guest will kernel panicwhen> booting. > 9. FC5 IA32 VMX Guest crash when try to boot SMP Guest: set vcpus=2 > acpi=1 apic=1 pae=1 in configure file and create SMP FC5 IA32 VMX > guest, guest wil crash with attached output. > 10. 32pae SMP guest with 1500MB memory will cause xen0 reboot > 11. Bootup 64bit guest with pae=0 in configure file, guest will > complain "Your CPU does not support long mode"Are you able to bring up an uni-processor 32-bit VMX guest on IA32E HV/dom0? _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
We believe we''ve tracked down the root cause of at least some of
these issues - a proper fix is in the staging tree and should get
pushed tonight (-unstable cset 11243:51a98a6c2c05).
If you want to check it out earlier, or if the staging tree fails
to push for any reason, you can just apply the below patch to xen.
In our testing this fixes the LTP math cases as well as the kernel
panic - not synching: kernel/sched.c issue.
cheers,
S.
--- a/xen/arch/x86/x86_emulate.c Wed Aug 23 17:14:38 2006 +0100
+++ b/xen/arch/x86/x86_emulate.c Wed Aug 23 17:25:11 2006 +0100
@@ -846,6 +846,9 @@ x86_emulate_memop(
emulate_1op("dec", dst, _regs.eflags);
break;
case 6: /* push */
+ /* Don''t emulate if fault was on stack */
+ if ( _regs.error_code & 2 )
+ goto cannot_emulate;
/* 64-bit mode: PUSH always pushes a 64-bit operand. */
if ( mode == X86EMUL_MODE_PROT64 )
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel
These problems happen on ChangeSets in this week but not a specific ChangeSet. If you need, I can double check and provide you some. Bests Jeff>-----Original Message----- >From: Puthiyaparambil, Aravindh >[mailto:aravindh.puthiyaparambil@unisys.com] >Sent: 2006年8月23日 23:51 >To: Zheng, Jeff; xen-devel@lists.xensource.com >Subject: RE: [Xen-devel] Some issues in recent ChangeSet. > >Jeff, > >Which changeset are you seeing these problems on? > >> IA32E HV/xen0 >> 7. CPU2K on 64bit guest has a performance regression. INT down from >> ~95% to 88% and FP down from ~90% to 78% (VMX/Native) >> 8. 64bit SMP VMX can not be up, if acpi=0 : If try with acpi=0 apic=1 >> vcpus=2 pae=1 to create FC3 IA32e guest, Guest will kernel panic >when >> booting. >> 9. FC5 IA32 VMX Guest crash when try to boot SMP Guest: set vcpus=2 >> acpi=1 apic=1 pae=1 in configure file and create SMP FC5 IA32 VMX >> guest, guest wil crash with attached output. >> 10. 32pae SMP guest with 1500MB memory will cause xen0 reboot >> 11. Bootup 64bit guest with pae=0 in configure file, guest will >> complain "Your CPU does not support long mode" > >Are you able to bring up an uni-processor 32-bit VMX guest on IA32E >HV/dom0? >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Puthiyaparambil, Aravindh
2006-Aug-24 01:01 UTC
RE: [Xen-devel] Some issues in recent ChangeSet.
Jeff, Don’t worry about it. I think it has been fixed in unstable cset 11243:51a98a6c2c05. Thanks, Aravindh> -----Original Message----- > From: Zheng, Jeff [mailto:jeff.zheng@intel.com] > Sent: Wednesday, August 23, 2006 8:52 PM > To: Puthiyaparambil, Aravindh; xen-devel@lists.xensource.com > Subject: RE: [Xen-devel] Some issues in recent ChangeSet. > > These problems happen on ChangeSets in this week but not a specific > ChangeSet. If you need, I can double check and provide you some. > > Bests > Jeff > > >-----Original Message----- > >From: Puthiyaparambil, Aravindh > >[mailto:aravindh.puthiyaparambil@unisys.com] > >Sent: 2006年8月23日 23:51 > >To: Zheng, Jeff; xen-devel@lists.xensource.com > >Subject: RE: [Xen-devel] Some issues in recent ChangeSet. > > > >Jeff, > > > >Which changeset are you seeing these problems on? > > > >> IA32E HV/xen0 > >> 7. CPU2K on 64bit guest has a performance regression. INT down from > >> ~95% to 88% and FP down from ~90% to 78% (VMX/Native) > >> 8. 64bit SMP VMX can not be up, if acpi=0 : If try with acpi=0 apic=1 > >> vcpus=2 pae=1 to create FC3 IA32e guest, Guest will kernel panic > >when > >> booting. > >> 9. FC5 IA32 VMX Guest crash when try to boot SMP Guest: set vcpus=2 > >> acpi=1 apic=1 pae=1 in configure file and create SMP FC5 IA32 VMX > >> guest, guest wil crash with attached output. > >> 10. 32pae SMP guest with 1500MB memory will cause xen0 reboot > >> 11. Bootup 64bit guest with pae=0 in configure file, guest will > >> complain "Your CPU does not support long mode" > > > >Are you able to bring up an uni-processor 32-bit VMX guest on IA32E > >HV/dom0? > >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Yes. Your patch should at least fix ltp math issue. And we see your patch is checked in to CS11242. But unfortunately your patch as 11242 was removed in CS 11243. Bests Jeff>-----Original Message----- >From: Steven Hand [mailto:Steven.Hand@cl.cam.ac.uk] >Sent: 2006年8月24日 2:09 >To: Zheng, Jeff >Cc: xen-devel@lists.xensource.com; Steven.Hand@cl.cam.ac.uk >Subject: Re: [Xen-devel] Some issues in recent ChangeSet. > > >We believe we''ve tracked down the root cause of at least some of >these issues - a proper fix is in the staging tree and should get >pushed tonight (-unstable cset 11243:51a98a6c2c05). > >If you want to check it out earlier, or if the staging tree fails >to push for any reason, you can just apply the below patch to xen. >In our testing this fixes the LTP math cases as well as the kernel >panic - not synching: kernel/sched.c issue. > >cheers, > >S. > > > > >--- a/xen/arch/x86/x86_emulate.c Wed Aug 23 17:14:38 2006 +0100 >+++ b/xen/arch/x86/x86_emulate.c Wed Aug 23 17:25:11 2006 +0100 >@@ -846,6 +846,9 @@ x86_emulate_memop( > emulate_1op("dec", dst, _regs.eflags); > break; > case 6: /* push */ >+ /* Don''t emulate if fault was on stack */ >+ if ( _regs.error_code & 2 ) >+ goto cannot_emulate; > /* 64-bit mode: PUSH always pushes a 64-bit operand. */ > if ( mode == X86EMUL_MODE_PROT64 ) > >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
>Yes. Your patch should at least fix ltp math issue. And we see your >patch is checked in to CS11242.Ok, great!>But unfortunately your patch as 11242 was removed in CS 11243.Yes - CS 11243 is a more comprehensive patch which addresses the same general issue but in more potential cases. Can you confirm that CS 11243 also resolves the ltp math issue for you? (As mentioned in the original post, this stack corruption bug is the likely culprit of a number of other instabilities people have been seeing recently with HVM guests. Hence we''re very keen to hear of any recurrance of those issues after the fix in CS 11243). cheers, S. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
No. The issue still exists. We also found that "kernel panic - not sync: kernel/sched.c" issue in kernel build at CS11246. Bests Jeff>-----Original Message----- >From: Steven Hand [mailto:Steven.Hand@cl.cam.ac.uk] >Sent: 2006年8月24日 14:45 >To: Zheng, Jeff >Cc: Steven Hand; xen-devel@lists.xensource.com; >Steven.Hand@cl.cam.ac.uk >Subject: Re: [Xen-devel] Some issues in recent ChangeSet. > > >>Yes. Your patch should at least fix ltp math issue. And we see your >>patch is checked in to CS11242. > >Ok, great! > >>But unfortunately your patch as 11242 was removed in CS 11243. > >Yes - CS 11243 is a more comprehensive patch which addresses the same >general issue but in more potential cases. Can you confirm >that CS 11243 >also resolves the ltp math issue for you? > >(As mentioned in the original post, this stack corruption bug is the >likely culprit of a number of other instabilities people have been >seeing recently with HVM guests. Hence we''re very keen to hear of any >recurrance of those issues after the fix in CS 11243). > > >cheers, > >S. >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On 24/8/06 7:48 am, "Zheng, Jeff" <jeff.zheng@intel.com> wrote:> No. The issue still exists. We also found that "kernel panic - not sync: > kernel/sched.c" issue in kernel build at CS11246.Ok, I see why the ''better'' patch 11243 doesn''t in fact work. It''s good to hear that 11242 worked okay for you though. We''ll sort out a real, final, proper fix today! -- Keir _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On 24/8/06 8:20 am, "Keir Fraser" <Keir.Fraser@cl.cam.ac.uk> wrote:>> No. The issue still exists. We also found that "kernel panic - not sync: >> kernel/sched.c" issue in kernel build at CS11246. > > Ok, I see why the ''better'' patch 11243 doesn''t in fact work. It''s good to > hear that 11242 worked okay for you though. We''ll sort out a real, final, > proper fix today!Hopefully 11249:1ff4cc298bec is a final solution to this issue. -- Keir _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Yes. LTP math issue & "kerne panic - not sync" issue resolved.
Resolved issues contain:
32bit HV/xen0:
2. 4 LTP math cases(float_bessel, float_exp_log, float_power and
float_trigo) sometime fail. (775)
3. Kernel build in VMX may meet kernel panic. It complains like: "<0>
Kernel panic - not syncing: kernel/sched.c:460:
spin_lock(kernel/sched.c:c03ea120) already locked by
kernel/sched.c/2689"
IA32E HV/xen0
8. 64bit SMP VMX can not be up, if acpi=0 : If try with acpi=0 apic=1
vcpus=2 pae=1 to create FC3 IA32e guest, Guest will kernel panic when
booting.
But following issues still exist:
1. Reboot fail in Guest OS: Reboot in guest will close current qemu
windows but does not create a new qemu window to restart
32bit PAE HV/xen0:
4. Cannot bootup pae SMP VMX domain. guest crashed with attached
output.
5. Create 32 guest with balloon driver will hang xen0: If I do not set
a fixed size memory using "dom0_mem=" option in grub.conf for dom0,
dom0 will hang after a vmx guest is created.
6. 32 SMP Guest only see about 900M memory when assign 1.5G memory for
the guest: If I assign 1.5G memory in vmx config file and boot up the
guest, I still can only see 903256KB memory using "cat
/proc/meminfo"
in the guest, though I can see 1.5G memory allocated to the guest.
IA32E HV/xen0
7. 64bit guest shows io-specific performance improvement but cpu-specific
performance downgrad
9. FC5 IA32 VMX Guest crash when try to boot SMP Guest: set vcpus=2
acpi=1 apic=1 pae=1 in configure file and create SMP FC5 IA32 VMX
guest, guest wil crash with attached output.
10. 32pae SMP guest with 1500MB memory will cause xen0 reboot
11. Bootup 64bit guest with pae=0 in configure file, guest will
complain "Your CPU does not support long mode"
Bests
Jeff
>-----Original Message-----
>From: Keir Fraser [mailto:Keir.Fraser@cl.cam.ac.uk]
>Sent: 2006年8月24日 16:53
>To: Keir Fraser; Zheng, Jeff; Steven Hand
>Cc: xen-devel@lists.xensource.com
>Subject: Re: [Xen-devel] Some issues in recent ChangeSet.
>
>
>
>
>On 24/8/06 8:20 am, "Keir Fraser" <Keir.Fraser@cl.cam.ac.uk>
wrote:
>
>>> No. The issue still exists. We also found that "kernel
>panic - not sync:
>>> kernel/sched.c" issue in kernel build at CS11246.
>>
>> Ok, I see why the ''better'' patch 11243
doesn''t in fact work.
>It''s good to
>> hear that 11242 worked okay for you though. We''ll sort out a
>real, final,
>> proper fix today!
>
>Hopefully 11249:1ff4cc298bec is a final solution to this issue.
>
> -- Keir
>
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel
OK. Following two issues are invalid/trivial: + 32 SMP Guest only see about 900M memory when assign 1.5G memory This is because 32bit guest was built without enable HIGH_MEM. + Bootup 64bit guest with pae=0 in configure file, guest will complain "Your CPU does not support long mode" 64bit guest assumes pae=1 And following issues still exist: + (up/SMP) pae guest cannot bootup on 32pae/64bit platform. Vista guest cannot bootup either. + Boot 32 guest will hang 32pae xen0 with balloon driver: If I do not set a fixed size memory using "dom0_mem=" option in grub.conf for dom0, dom0 will hang after a vmx guest is created. + 64bit guest shows cpu-specific performance downgrad + FC5 IA32 VMX Guest crash when try to boot SMP Guest: set vcpus=2 acpi=1 apic=1 pae=1 in configure file and create SMP FC5 IA32 VMX guest, guest wil crash + 32pae SMP guest with 1500MB memory will cause 64bit xen0 reboot + Creating PAE SMP base kernel in VMX causes 32pae xen0 reboot(xen0_crash.1) + VNIF does not work Bests Jeff>-----Original Message----- >From: xen-devel-bounces@lists.xensource.com >[mailto:xen-devel-bounces@lists.xensource.com] On Behalf Of Zheng, Jeff >Sent: 2006年8月25日 23:15 >To: Keir Fraser; Steven Hand >Cc: xen-devel@lists.xensource.com >Subject: RE: [Xen-devel] Some issues in recent ChangeSet. > >Yes. LTP math issue & "kerne panic - not sync" issue resolved. > >Resolved issues contain: >32bit HV/xen0: >2. 4 LTP math cases(float_bessel, float_exp_log, float_power and > float_trigo) sometime fail. (775) >3. Kernel build in VMX may meet kernel panic. It complains like: "<0> > Kernel panic - not syncing: kernel/sched.c:460: > spin_lock(kernel/sched.c:c03ea120) already locked by > kernel/sched.c/2689" >IA32E HV/xen0 >8. 64bit SMP VMX can not be up, if acpi=0 : If try with acpi=0 apic=1 > vcpus=2 pae=1 to create FC3 IA32e guest, Guest will kernel >panic when > booting. > > >But following issues still exist: >1. Reboot fail in Guest OS: Reboot in guest will close current qemu > windows but does not create a new qemu window to restart > >32bit PAE HV/xen0: >4. Cannot bootup pae SMP VMX domain. guest crashed with attached > output. >5. Create 32 guest with balloon driver will hang xen0: If I do not set > a fixed size memory using "dom0_mem=" option in grub.conf for dom0, > dom0 will hang after a vmx guest is created. >6. 32 SMP Guest only see about 900M memory when assign 1.5G memory for > the guest: If I assign 1.5G memory in vmx config file and >boot up the > guest, I still can only see 903256KB memory using "cat >/proc/meminfo" > in the guest, though I can see 1.5G memory allocated to the guest. > >IA32E HV/xen0 >7. 64bit guest shows io-specific performance improvement but >cpu-specific > performance downgrad >9. FC5 IA32 VMX Guest crash when try to boot SMP Guest: set vcpus=2 > acpi=1 apic=1 pae=1 in configure file and create SMP FC5 IA32 VMX > guest, guest wil crash with attached output. >10. 32pae SMP guest with 1500MB memory will cause xen0 reboot >11. Bootup 64bit guest with pae=0 in configure file, guest will > complain "Your CPU does not support long mode" > > >Bests >Jeff > >>-----Original Message----- >>From: Keir Fraser [mailto:Keir.Fraser@cl.cam.ac.uk] >>Sent: 2006年8月24日 16:53 >>To: Keir Fraser; Zheng, Jeff; Steven Hand >>Cc: xen-devel@lists.xensource.com >>Subject: Re: [Xen-devel] Some issues in recent ChangeSet. >> >> >> >> >>On 24/8/06 8:20 am, "Keir Fraser" <Keir.Fraser@cl.cam.ac.uk> wrote: >> >>>> No. The issue still exists. We also found that "kernel >>panic - not sync: >>>> kernel/sched.c" issue in kernel build at CS11246. >>> >>> Ok, I see why the ''better'' patch 11243 doesn''t in fact work. >>It''s good to >>> hear that 11242 worked okay for you though. We''ll sort out a >>real, final, >>> proper fix today! >> >>Hopefully 11249:1ff4cc298bec is a final solution to this issue. >> >> -- Keir >> > >_______________________________________________ >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
>And following issues still exist: >+ (up/SMP) pae guest cannot bootup on 32pae/64bit platform. Vista guest > cannot bootup either.This is believed fixed by -unstable cset 11270:28824bd2701c which is still in the staging tree. If you''d like to test it out now, the relevant part of the patch is below. --- a/xen/arch/x86/mm/shadow/multi.c Mon Aug 28 20:22:56 2006 +0100 +++ b/xen/arch/x86/mm/shadow/multi.c Mon Aug 28 21:25:13 2006 +0100 @@ -2543,6 +2543,16 @@ static int validate_gl3e(struct vcpu *v, perfc_incrc(shadow_validate_gl3e_calls); +#if (SHADOW_PAGING_LEVELS == 3) && (GUEST_PAGING_LEVELS == 3) + { + /* If we''ve updated a subshadow which is unreferenced then + we don''t care what value is being written - bail. */ + struct pae_l3_bookkeeping *info = sl3p_to_info(se); + if(!info->refcount) + return result; + } +#endif + if ( guest_l3e_get_flags(*new_gl3e) & _PAGE_PRESENT ) { gfn_t gl2gfn = guest_l3e_get_gfn(*new_gl3e); cheers, S. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
We can see the pae guest works. And previous open bug number reduces to 3: + Boot 32 guest will hang 32pae xen0 with balloon driver: If I do not set a fixed size memory using "dom0_mem=" option in grub.conf for dom0, dom0 will hang after a vmx guest is created. + 64bit guest shows cpu-specific performance downgrad + FC5 IA32 VMX Guest crash when try to boot SMP Guest: set vcpus=2 acpi=1 apic=1 pae=1 in configure file and create SMP FC5 IA32 VMX guest, guest wil crash We also see these issues: + VNIF does not work + Build kernel in IA32 SMP VMX domain may fail + scp big file from Xen0 to VMX may fail + With ACPI=1, booting Windows guest works, after destroy and recreate Windows guest the qemu window become black + Using vi in VMX may hang guest Bests Jeff _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
I also see HVM domain reboot failed again in recent cset. A quick look showes qemu-dm and HV use diferrent MFN to map shared_page after rebooting, this will cause DM event channel binding failure. In my box, DM use mfn 0x2d406 while HV use mfn 0x2d426 to map shared page, they differ 0x20 pages. Looks like xc_get_pfn_list is incorrect after rebooting. I will look into this issue. Best Regards Ke Zheng, Jeff wrote:> We can see the pae guest works. And previous open bug number reduces > to 3: > + Boot 32 guest will hang 32pae xen0 with balloon driver: If I do not > set > a fixed size memory using "dom0_mem=" option in grub.conf for dom0, > dom0 will hang after a vmx guest is created. > + 64bit guest shows cpu-specific performance downgrad > + FC5 IA32 VMX Guest crash when try to boot SMP Guest: set vcpus=2 > acpi=1 apic=1 pae=1 in configure file and create SMP FC5 IA32 VMX > guest, guest wil crash > > We also see these issues: > + VNIF does not work > + Build kernel in IA32 SMP VMX domain may fail > + scp big file from Xen0 to VMX may fail > + With ACPI=1, booting Windows guest works, after destroy and recreate > Windows guest the qemu window become black > + Using vi in VMX may hang guest > > > Bests > Jeff > > _______________________________________________ > 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 4/9/06 8:19 am, "Yu, Ke" <ke.yu@intel.com> wrote:> I also see HVM domain reboot failed again in recent cset. > > A quick look showes qemu-dm and HV use diferrent MFN to map shared_page after > rebooting, this will cause DM event channel binding failure. In my box, DM use > mfn 0x2d406 while HV use mfn 0x2d426 to map shared page, they differ 0x20 > pages. Looks like xc_get_pfn_list is incorrect after rebooting. I will look > into this issue.Short-term fix: For HVM guests, return frame list by scanning the P2M table so you definitely get the frames in the right order (currently it always scans the ownership linked list, which actually has no guaranteed order). Long-term fix: Always map HVM guest pages by PFN rather than MFN. This will avoid domain0 needing to manually do P->M translations -- it''ll instead happen automatically inside Xen. Much neater. Overall we should be aiming to get read of get_pfn_list entirely, especially for HVM guests. -- Keir _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Keir Fraser wrote:> On 4/9/06 8:19 am, "Yu, Ke" <ke.yu@intel.com> wrote: > >> I also see HVM domain reboot failed again in recent cset. >> >> A quick look showes qemu-dm and HV use diferrent MFN to map >> shared_page after rebooting, this will cause DM event channel >> binding failure. In my box, DM use mfn 0x2d406 while HV use mfn >> 0x2d426 to map shared page, they differ 0x20 pages. Looks like >> xc_get_pfn_list is incorrect after rebooting. I will look into this >> issue. > > Short-term fix: For HVM guests, return frame list by scanning the P2M > table so you definitely get the frames in the right order (currently > it always scans the ownership linked list, which actually has no > guaranteed order). > > Long-term fix: Always map HVM guest pages by PFN rather than MFN. > This will avoid domain0 needing to manually do P->M translations -- > it''ll instead happen automatically inside Xen. Much neater. > > Overall we should be aiming to get read of get_pfn_list entirely, > especially for HVM guests. > > -- KeirHi Keir, Please see the attached short-term fix. Hopefully, it can be applied before 3.0.3. I have tested it with several memory combinations, and all pass: ia32 guest, ia32 host, 1G guest mem ia32 pae guest, ia32pae host, 1.5G guest mem, ia32 guest, ia32e host, 1G guest mem ia32e guest, ia32e host, 4G guest mem ia32e guest, ia32e host, 8G guest mem Best Regards Ke ---------------- Intel Opensource Technology Center _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On Thu, Sep 14, 2006 at 01:47:11PM +0800, Yu, Ke wrote:> Keir Fraser wrote: > > On 4/9/06 8:19 am, "Yu, Ke" <ke.yu@intel.com> wrote: > > > >> I also see HVM domain reboot failed again in recent cset. > >> > >> A quick look showes qemu-dm and HV use diferrent MFN to map > >> shared_page after rebooting, this will cause DM event channel > >> binding failure. In my box, DM use mfn 0x2d406 while HV use mfn > >> 0x2d426 to map shared page, they differ 0x20 pages. Looks like > >> xc_get_pfn_list is incorrect after rebooting. I will look into this > >> issue.keir, same issue happened when vmx save/restore. ke''s short-term patch can fix it. any comments? thanks,> > > > Short-term fix: For HVM guests, return frame list by scanning the P2M > > table so you definitely get the frames in the right order (currently > > it always scans the ownership linked list, which actually has no > > guaranteed order). > > > > Long-term fix: Always map HVM guest pages by PFN rather than MFN. > > This will avoid domain0 needing to manually do P->M translations -- > > it''ll instead happen automatically inside Xen. Much neater. > > > > Overall we should be aiming to get read of get_pfn_list entirely, > > especially for HVM guests. > > > > -- Keir > > Hi Keir, > > Please see the attached short-term fix. Hopefully, it can be applied before 3.0.3. > > I have tested it with several memory combinations, and all pass: > ia32 guest, ia32 host, 1G guest mem > ia32 pae guest, ia32pae host, 1.5G guest mem, > ia32 guest, ia32e host, 1G guest mem > ia32e guest, ia32e host, 4G guest mem > ia32e guest, ia32e host, 8G guest mem > > Best Regards > Ke > ---------------- > Intel Opensource Technology Center> _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.xensource.com > http://lists.xensource.com/xen-devel-- best rgds, edwin _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On 15/9/06 04:11, "Zhai, Edwin" <edwin.zhai@intel.com> wrote:> same issue happened when vmx save/restore. > ke''s short-term patch can fix it. > any comments?I need to look at it some more but I''ll probably apply it. It seems a reasonable approach. -- Keir _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel