CMCI support in Xen - what OS has been tested with it?

CMCI support in Xen - what OS has been tested with it?

[Konrad Rzeszutek Wilk]

Hey,

I've been tracking an bug with Solaris where all of them stopped working
with Xen 4.3 and later. I narrowed it down to this commit
975b5bdf27031adef6587bb5c92b7a5800e051f1 ("x86: vMCE emulation") which
now toggles the MCE to advertise the CMCI support.

If I do:
diff --git a/xen/arch/x86/cpu/mcheck/vmce.c b/xen/arch/x86/cpu/mcheck/vmce.c
index ed00f7c..a91bcb7 100644
--- a/xen/arch/x86/cpu/mcheck/vmce.c
+++ b/xen/arch/x86/cpu/mcheck/vmce.c
@@ -46,7 +46,6 @@
  */
 #define INTEL_GUEST_MCG_CAP (MCG_SER_P |       \
                              MCG_TES_P |       \
-                             MCG_CMCI_P |      \
                              GUEST_MC_BANK_NUM)
 
 #define AMD_GUEST_MCG_CAP GUEST_MC_BANK_NUM

or:

cpuid=['1:edx=xxxxxxxxxxxxxxxxx0xxxxxx0xxxxxxx']
(mask MCE and MCA cpuid flags)

then they boot nicely, but that is of course not the long-term
solution.

The guest stops booting - and I am tracking down what it is that it
expects - but in meantime I was wondering what type of guests have been
tested for CMCI support? Aka, which ones are suppose to work with this?

Thanks!

Re: CMCI support in Xen - what OS has been tested with it?

[Zhang, Yang Z]

Konrad Rzeszutek Wilk wrote on 2014-04-19:
> Hey,
> 
> I've been tracking an bug with Solaris where all of them stopped
> working with Xen 4.3 and later. I narrowed it down to this commit
> 975b5bdf27031adef6587bb5c92b7a5800e051f1 ("x86: vMCE emulation") which
> now toggles the MCE to advertise the CMCI support.
> 
> If I do:
> diff --git a/xen/arch/x86/cpu/mcheck/vmce.c
> b/xen/arch/x86/cpu/mcheck/vmce.c index ed00f7c..a91bcb7 100644
> --- a/xen/arch/x86/cpu/mcheck/vmce.c
> +++ b/xen/arch/x86/cpu/mcheck/vmce.c
> @@ -46,7 +46,6 @@
>   */
>  #define INTEL_GUEST_MCG_CAP (MCG_SER_P |       \
>                               MCG_TES_P |       \ -                     
>                                      MCG_CMCI_P |      \
>                               GUEST_MC_BANK_NUM)
>  #define AMD_GUEST_MCG_CAP GUEST_MC_BANK_NUM
> or:
> 
> cpuid=['1:edx=xxxxxxxxxxxxxxxxx0xxxxxx0xxxxxxx']
> (mask MCE and MCA cpuid flags)
> 
> then they boot nicely, but that is of course not the long-term solution.
> 
> The guest stops booting - and I am tracking down what it is that it
> expects - but in meantime I was wondering what type of guests have
> been tested for CMCI support? Aka, which ones are suppose to work with this?

I don't think hypervisor will inject CMCI to guest and guest should never receive the CMCI. See comments:
/*
 * MCG_SER_P:  software error recovery supported
 * MCG_TES_P:  to avoid MCi_status bit56:53 model specific
 * MCG_CMCI_P: expose CMCI capability but never really inject it to guest,
 *             for sake of performance since guest not polling periodically
 */

So the possible cause is that there may be some assumptions in Solaris which depends on CMCI. 

> 
> Thanks!
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel


Best regards,
Yang

Re: CMCI support in Xen - what OS has been tested with it?

[Zhang, Yang Z]

Zhang, Yang Z wrote on 2014-04-21:
> Konrad Rzeszutek Wilk wrote on 2014-04-19:

CC jingsong who has joined alibaba. :)

>> Hey,
>> 
>> I've been tracking an bug with Solaris where all of them stopped
>> working with Xen 4.3 and later. I narrowed it down to this commit
>> 975b5bdf27031adef6587bb5c92b7a5800e051f1 ("x86: vMCE emulation") which
>> now toggles the MCE to advertise the CMCI support.
>> 
>> If I do:
>> diff --git a/xen/arch/x86/cpu/mcheck/vmce.c
>> b/xen/arch/x86/cpu/mcheck/vmce.c index ed00f7c..a91bcb7 100644
>> --- a/xen/arch/x86/cpu/mcheck/vmce.c
>> +++ b/xen/arch/x86/cpu/mcheck/vmce.c
>> @@ -46,7 +46,6 @@
>>   */
>>  #define INTEL_GUEST_MCG_CAP (MCG_SER_P |       \
>>                               MCG_TES_P |       \ -
>>                                      MCG_CMCI_P |      \
>>                               GUEST_MC_BANK_NUM)  #define
>> AMD_GUEST_MCG_CAP GUEST_MC_BANK_NUM
>> or:
>> 
>> cpuid=['1:edx=xxxxxxxxxxxxxxxxx0xxxxxx0xxxxxxx']
>> (mask MCE and MCA cpuid flags)
>> 
>> then they boot nicely, but that is of course not the long-term solution.
>> 
>> The guest stops booting - and I am tracking down what it is that it
>> expects - but in meantime I was wondering what type of guests have
>> been tested for CMCI support? Aka, which ones are suppose to work
>> with
> this?
> 
> I don't think hypervisor will inject CMCI to guest and guest should
> never receive the CMCI. See comments:
> /*
>  * MCG_SER_P:  software error recovery supported
>  * MCG_TES_P:  to avoid MCi_status bit56:53 model specific
>  * MCG_CMCI_P: expose CMCI capability but never really inject it to guest,
>  *             for sake of performance since guest not polling periodically
>  */
> So the possible cause is that there may be some assumptions in Solaris
> which depends on CMCI.
> 
>> 
>> Thanks!
>> 
>> _______________________________________________
>> Xen-devel mailing list
>> Xen-devel@lists.xen.org
>> http://lists.xen.org/xen-devel
> 
> 
> Best regards,
> Yang


Best regards,
Yang