[PATCH] Enlarge default NR_CPUS to 64 for x86_64

[PATCH] Enlarge default NR_CPUS to 64 for x86_64

[Wei, Gang]

[-- Attachment #1: Type: text/plain, Size: 720 bytes --]

By far a lot of systems are with more sockets (>=4), more cores (>=8), meanwhile with HyperThread, so enlarge the default NR_CPUS would easier the build for more threads systems.

Jimmy

-------------------------------------------------------------
Enlarge default NR_CPUS to 64 for x86_64.

Signed-off-by: Wei Gang <gang.wei@intel.com>

diff -r b9cdcf502aa3 xen/include/asm-x86/config.h
--- a/xen/include/asm-x86/config.h	Thu Aug 06 11:14:48 2009 +0100
+++ b/xen/include/asm-x86/config.h	Fri Aug 07 14:54:06 2009 +0800
@@ -49,7 +49,11 @@
 #ifdef MAX_PHYS_CPUS
 #define NR_CPUS MAX_PHYS_CPUS
 #else
+#ifdef __i386__
 #define NR_CPUS 32
+#else
+#define NR_CPUS 64
+#endif
 #endif
 
 #ifdef __i386__

[-- Attachment #2: nr_cpus-64.patch --]
[-- Type: application/octet-stream, Size: 449 bytes --]

Enlarge default NR_CPUS to 64 for x86_64.

Signed-off-by: Wei Gang <gang.wei@intel.com>

diff -r b9cdcf502aa3 xen/include/asm-x86/config.h
--- a/xen/include/asm-x86/config.h	Thu Aug 06 11:14:48 2009 +0100
+++ b/xen/include/asm-x86/config.h	Fri Aug 07 14:54:06 2009 +0800
@@ -49,7 +49,11 @@
 #ifdef MAX_PHYS_CPUS
 #define NR_CPUS MAX_PHYS_CPUS
 #else
+#ifdef __i386__
 #define NR_CPUS 32
+#else
+#define NR_CPUS 64
+#endif
 #endif
 
 #ifdef __i386__

[-- Attachment #3: Type: text/plain, Size: 138 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Re: [PATCH] Enlarge default NR_CPUS to 64 for x86_64

[Keir Fraser]

Shall we just bump it for i386 too?

 -- Keir

On 07/08/2009 10:37, "Wei, Gang" <gang.wei@intel.com> wrote:

> By far a lot of systems are with more sockets (>=4), more cores (>=8),
> meanwhile with HyperThread, so enlarge the default NR_CPUS would easier the
> build for more threads systems.
> 
> Jimmy
> 
> -------------------------------------------------------------
> Enlarge default NR_CPUS to 64 for x86_64.
> 
> Signed-off-by: Wei Gang <gang.wei@intel.com>
> 
> diff -r b9cdcf502aa3 xen/include/asm-x86/config.h
> --- a/xen/include/asm-x86/config.h Thu Aug 06 11:14:48 2009 +0100
> +++ b/xen/include/asm-x86/config.h Fri Aug 07 14:54:06 2009 +0800
> @@ -49,7 +49,11 @@
>  #ifdef MAX_PHYS_CPUS
>  #define NR_CPUS MAX_PHYS_CPUS
>  #else
> +#ifdef __i386__
>  #define NR_CPUS 32
> +#else
> +#define NR_CPUS 64
> +#endif
>  #endif
>  
>  #ifdef __i386__

Re: [PATCH] Enlarge default NR_CPUS to 64 for x86_64

[Jan Beulich]

I think there's little point in having both have the same limit. I'd rather
question whether the limit shouldn't be pushed up higher for x86-64.

Jan

>>> Keir Fraser <keir.fraser@eu.citrix.com> 07.08.09 11:52 >>>
Shall we just bump it for i386 too?

 -- Keir

On 07/08/2009 10:37, "Wei, Gang" <gang.wei@intel.com> wrote:

> By far a lot of systems are with more sockets (>=4), more cores (>=8),
> meanwhile with HyperThread, so enlarge the default NR_CPUS would easier the
> build for more threads systems.
> 
> Jimmy
> 
> -------------------------------------------------------------
> Enlarge default NR_CPUS to 64 for x86_64.
> 
> Signed-off-by: Wei Gang <gang.wei@intel.com>
> 
> diff -r b9cdcf502aa3 xen/include/asm-x86/config.h
> --- a/xen/include/asm-x86/config.h Thu Aug 06 11:14:48 2009 +0100
> +++ b/xen/include/asm-x86/config.h Fri Aug 07 14:54:06 2009 +0800
> @@ -49,7 +49,11 @@
>  #ifdef MAX_PHYS_CPUS
>  #define NR_CPUS MAX_PHYS_CPUS
>  #else
> +#ifdef __i386__
>  #define NR_CPUS 32
> +#else
> +#define NR_CPUS 64
> +#endif
>  #endif
>  
>  #ifdef __i386__



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com 
http://lists.xensource.com/xen-devel

Re: [PATCH] Enlarge default NR_CPUS to 64 for x86_64

[Keir Fraser]

Yes indeed. We could push to 128 no problem. I think Intel's only concern is
that 64 is as much as they test currently.

 -- Keir

On 07/08/2009 11:02, "Jan Beulich" <JBeulich@novell.com> wrote:

> I think there's little point in having both have the same limit. I'd rather
> question whether the limit shouldn't be pushed up higher for x86-64.
> 
> Jan
> 
>>>> Keir Fraser <keir.fraser@eu.citrix.com> 07.08.09 11:52 >>>
> Shall we just bump it for i386 too?
> 
>  -- Keir
> 
> On 07/08/2009 10:37, "Wei, Gang" <gang.wei@intel.com> wrote:
> 
>> By far a lot of systems are with more sockets (>=4), more cores (>=8),
>> meanwhile with HyperThread, so enlarge the default NR_CPUS would easier the
>> build for more threads systems.
>> 
>> Jimmy
>> 
>> -------------------------------------------------------------
>> Enlarge default NR_CPUS to 64 for x86_64.
>> 
>> Signed-off-by: Wei Gang <gang.wei@intel.com>
>> 
>> diff -r b9cdcf502aa3 xen/include/asm-x86/config.h
>> --- a/xen/include/asm-x86/config.h Thu Aug 06 11:14:48 2009 +0100
>> +++ b/xen/include/asm-x86/config.h Fri Aug 07 14:54:06 2009 +0800
>> @@ -49,7 +49,11 @@
>>  #ifdef MAX_PHYS_CPUS
>>  #define NR_CPUS MAX_PHYS_CPUS
>>  #else
>> +#ifdef __i386__
>>  #define NR_CPUS 32
>> +#else
>> +#define NR_CPUS 64
>> +#endif
>>  #endif
>>  
>>  #ifdef __i386__
> 
> 
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
> 

RE: [PATCH] Enlarge default NR_CPUS to 64 for x86_64

[Wei, Gang]

Not sure about whether we can raise it for i386 too. I am a little confused by below lines in config.h:

    #ifdef __i386__
    #if NR_CPUS > 32
    #error "Maximum of 32 physical processors supported by Xen on x86_32"
    #endif
    /* Maximum number of virtual CPUs in multi-processor guests. */
    #define MAX_VIRT_CPUS XEN_LEGACY_MAX_VCPUS
    #endif

I don't know why this compiling time error was put here for i386 yet. Any body know it?

Jimmy

Keir Fraser wrote:
> Shall we just bump it for i386 too?
> 
> -- Keir
> 
> On 07/08/2009 10:37, "Wei, Gang" <gang.wei@intel.com> wrote:
> 
>> By far a lot of systems are with more sockets (>=4), more cores
>> (>=8), meanwhile with HyperThread, so enlarge the default NR_CPUS
>> would easier the build for more threads systems.
>> 
>> Jimmy
>> 
>> -------------------------------------------------------------
>> Enlarge default NR_CPUS to 64 for x86_64.
>> 
>> Signed-off-by: Wei Gang <gang.wei@intel.com>
>> 
>> diff -r b9cdcf502aa3 xen/include/asm-x86/config.h
>> --- a/xen/include/asm-x86/config.h Thu Aug 06 11:14:48 2009 +0100
>> +++ b/xen/include/asm-x86/config.h Fri Aug 07 14:54:06 2009 +0800 @@
>>  -49,7 +49,11 @@ #ifdef MAX_PHYS_CPUS
>>  #define NR_CPUS MAX_PHYS_CPUS
>>  #else
>> +#ifdef __i386__
>>  #define NR_CPUS 32
>> +#else
>> +#define NR_CPUS 64
>> +#endif
>>  #endif
>> 
>>  #ifdef __i386__

RE: [PATCH] Enlarge default NR_CPUS to 64 for x86_64

[Wei, Gang]

Yes, it is. We can push to 128 if there really are a lot of needs. The slightly side-effect is a little larger default memory footprint for hypervisor if not really need to support that much of pcpus. We can do it step by step.

Jimmy

Keir Fraser wrote:
> Yes indeed. We could push to 128 no problem. I think Intel's only
> concern is that 64 is as much as they test currently.
> 
> -- Keir
> 
> On 07/08/2009 11:02, "Jan Beulich" <JBeulich@novell.com> wrote:
> 
>> I think there's little point in having both have the same limit. I'd
>> rather question whether the limit shouldn't be pushed up higher for
>> x86-64. 
>> 
>> Jan
>> 
>>>>> Keir Fraser <keir.fraser@eu.citrix.com> 07.08.09 11:52 >>>
>> Shall we just bump it for i386 too?
>> 
>>  -- Keir
>> 
>> On 07/08/2009 10:37, "Wei, Gang" <gang.wei@intel.com> wrote:
>> 
>>> By far a lot of systems are with more sockets (>=4), more cores
>>> (>=8), meanwhile with HyperThread, so enlarge the default NR_CPUS
>>> would easier the build for more threads systems.
>>> 
>>> Jimmy
>>> 
>>> -------------------------------------------------------------
>>> Enlarge default NR_CPUS to 64 for x86_64.
>>> 
>>> Signed-off-by: Wei Gang <gang.wei@intel.com>
>>> 
>>> diff -r b9cdcf502aa3 xen/include/asm-x86/config.h
>>> --- a/xen/include/asm-x86/config.h Thu Aug 06 11:14:48 2009 +0100
>>> +++ b/xen/include/asm-x86/config.h Fri Aug 07 14:54:06 2009 +0800
>>>  @@ -49,7 +49,11 @@ #ifdef MAX_PHYS_CPUS
>>>  #define NR_CPUS MAX_PHYS_CPUS
>>>  #else
>>> +#ifdef __i386__
>>>  #define NR_CPUS 32
>>> +#else
>>> +#define NR_CPUS 64
>>> +#endif
>>>  #endif
>>> 
>>>  #ifdef __i386__
>> 
>> 
>> 
>> _______________________________________________
>> Xen-devel mailing list
>> Xen-devel@lists.xensource.com
>> http://lists.xensource.com/xen-devel

Re: [PATCH] Enlarge default NR_CPUS to 64 for x86_64

[Keir Fraser]

On 07/08/2009 15:59, "Wei, Gang" <gang.wei@intel.com> wrote:

> Not sure about whether we can raise it for i386 too. I am a little confused by
> below lines in config.h:
> 
>     #ifdef __i386__
>     #if NR_CPUS > 32
>     #error "Maximum of 32 physical processors supported by Xen on x86_32"
>     #endif
>     /* Maximum number of virtual CPUs in multi-processor guests. */
>     #define MAX_VIRT_CPUS XEN_LEGACY_MAX_VCPUS
>     #endif
> 
> I don't know why this compiling time error was put here for i386 yet. Any body
> know it?

I think it's a hangover from when there was a cpumask_t inside struct
page_info. That structure had to be fixed size for i386 to fit in the
hypervisor's virtual memory footprint.

 -- Keir

RE: [PATCH] Enlarge default NR_CPUS to 64 for x86_64

[Dan Magenheimer]

> I think it's a hangover from when there was a cpumask_t inside struct
> page_info. That structure had to be fixed size for i386 to fit in the
> hypervisor's virtual memory footprint.

FYI, xen/include/asm-x86/mm.h still #includes <xen/cpumask.h>
and that include can be removed now (compile tested only).

> -----Original Message-----
> From: Keir Fraser [mailto:keir.fraser@eu.citrix.com]
> Sent: Friday, August 07, 2009 10:18 AM
> To: Wei, Gang; xen-devel@lists.xensource.com
> Subject: Re: [Xen-devel] [PATCH] Enlarge default NR_CPUS to 64 for
> x86_64
> 
> 
> On 07/08/2009 15:59, "Wei, Gang" <gang.wei@intel.com> wrote:
> 
> > Not sure about whether we can raise it for i386 too. I am a 
> little confused by
> > below lines in config.h:
> > 
> >     #ifdef __i386__
> >     #if NR_CPUS > 32
> >     #error "Maximum of 32 physical processors supported by 
> Xen on x86_32"
> >     #endif
> >     /* Maximum number of virtual CPUs in multi-processor guests. */
> >     #define MAX_VIRT_CPUS XEN_LEGACY_MAX_VCPUS
> >     #endif
> > 
> > I don't know why this compiling time error was put here for 
> i386 yet. Any body
> > know it?
> 
> I think it's a hangover from when there was a cpumask_t inside struct
> page_info. That structure had to be fixed size for i386 to fit in the
> hypervisor's virtual memory footprint.
> 
>  -- Keir
> 
> 
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
>

Re: [PATCH] Enlarge default NR_CPUS to 64 for x86_64

[Keir Fraser]

On 08/08/2009 22:46, "Dan Magenheimer" <dan.magenheimer@oracle.com> wrote:

>> I think it's a hangover from when there was a cpumask_t inside struct
>> page_info. That structure had to be fixed size for i386 to fit in the
>> hypervisor's virtual memory footprint.
> 
> FYI, xen/include/asm-x86/mm.h still #includes <xen/cpumask.h>
> and that include can be removed now (compile tested only).

Okay, I'll remove it.

 Thanks,
 Keir