Patch question (Hyper-Threading support)

Patch question (Hyper-Threading support)

[Elizeu Santos-Neto]

	Hi community,

	I've been working on modifying Xen and the Xen-Linux kernel to make
vCPU better represent Hyper-Threading physical CPUs, as I mentioned in a
previous post.

http://lists.xensource.com/archives/html/xen-devel/2006-01/msg00853.html

	I produced some performance measurements and patches that might be
useful for some Hyper-Threading friendly workloads running on Xen. I'll
make the performance measurement report available soon.	

	Basically, the modified files were:

	xen/common/domain.c
	xen/common/sched_bvt.c
	xen/common/sched_sedf.c
	xen/common/schedule.c
	xen/include/public/vcpu.h
	xen/include/xen/sched.h
	xen/arch/x86/smpboot.c

	How should I proceed to send the patch?

	Thanks,
	./Eli
	 

Re: Patch question (Hyper-Threading support)

[Tim Freeman]

On Fri, 24 Mar 2006 13:30:18 -0600
Elizeu Santos-Neto <elizeu@cs.uchicago.edu> wrote:

> 
> 	Hi community,
> 
> 	I've been working on modifying Xen and the Xen-Linux kernel to make
> vCPU better represent Hyper-Threading physical CPUs, as I mentioned in a
> previous post.
> 
> http://lists.xensource.com/archives/html/xen-devel/2006-01/msg00853.html
> 
> 	I produced some performance measurements and patches that might be
> useful for some Hyper-Threading friendly workloads running on Xen. I'll
> make the performance measurement report available soon.	
> 
> 	Basically, the modified files were:
> 
> 	xen/common/domain.c
> 	xen/common/sched_bvt.c
> 	xen/common/sched_sedf.c
> 	xen/common/schedule.c
> 	xen/include/public/vcpu.h
> 	xen/include/xen/sched.h
> 	xen/arch/x86/smpboot.c
> 
> 	How should I proceed to send the patch?
 

There is a guide here: 

http://wiki.xensource.com/xenwiki/SubmittingXenPatches

For examples, see:

http://lists.xensource.com/archives/html/xen-devel/

Also see these guidelines: 

http://www.zip.com.au/~akpm/linux/patches/stuff/tpp.txt
http://linux.yyz.us/patch-format.html

I would think replying to your patch email with the performance measurement
report would help keep them together for people. 

Tim 

Re: Patch question (Hyper-Threading support)

[Elizeu Santos-Neto]

	Hi Tim,

	Thanks a lot. I'm going to prepare the patches and send the report
together.

	Regards,
	Eli

On Fri, 2006-03-24 at 19:15 -0600, Tim Freeman wrote:
> On Fri, 24 Mar 2006 13:30:18 -0600
> Elizeu Santos-Neto <elizeu@cs.uchicago.edu> wrote:
> 
> > 
> > 	Hi community,
> > 
> > 	I've been working on modifying Xen and the Xen-Linux kernel to make
> > vCPU better represent Hyper-Threading physical CPUs, as I mentioned in a
> > previous post.
> > 
> > http://lists.xensource.com/archives/html/xen-devel/2006-01/msg00853.html
> > 
> > 	I produced some performance measurements and patches that might be
> > useful for some Hyper-Threading friendly workloads running on Xen. I'll
> > make the performance measurement report available soon.	
> > 
> > 	Basically, the modified files were:
> > 
> > 	xen/common/domain.c
> > 	xen/common/sched_bvt.c
> > 	xen/common/sched_sedf.c
> > 	xen/common/schedule.c
> > 	xen/include/public/vcpu.h
> > 	xen/include/xen/sched.h
> > 	xen/arch/x86/smpboot.c
> > 
> > 	How should I proceed to send the patch?
>  
> 
> There is a guide here: 
> 
> http://wiki.xensource.com/xenwiki/SubmittingXenPatches
> 
> For examples, see:
> 
> http://lists.xensource.com/archives/html/xen-devel/
> 
> Also see these guidelines: 
> 
> http://www.zip.com.au/~akpm/linux/patches/stuff/tpp.txt
> http://linux.yyz.us/patch-format.html
> 
> I would think replying to your patch email with the performance measurement
> report would help keep them together for people. 
> 
> Tim 
> 
> 
> 

Re: Patch question (Hyper-Threading support)

[Elizeu Santos-Neto]

	Hi all,

	Well, based on some quick inspections on the that I checked out from
the mercurial repository this afternoon, it seems that the HT support
has been already implemented. :-)

	Thus, here is a suggestion, it would be nice to make more details
available about who is doing what and in which phase of development
things are, this information could avoid duplicated efforts.

	Thanks,
	Eli

On Fri, 2006-03-24 at 19:15 -0600, Tim Freeman wrote:
> On Fri, 24 Mar 2006 13:30:18 -0600
> Elizeu Santos-Neto <elizeu@cs.uchicago.edu> wrote:
> 
> > 
> > 	Hi community,
> > 
> > 	I've been working on modifying Xen and the Xen-Linux kernel to make
> > vCPU better represent Hyper-Threading physical CPUs, as I mentioned in a
> > previous post.
> > 
> > http://lists.xensource.com/archives/html/xen-devel/2006-01/msg00853.html
> > 
> > 	I produced some performance measurements and patches that might be
> > useful for some Hyper-Threading friendly workloads running on Xen. I'll
> > make the performance measurement report available soon.	
> > 
> > 	Basically, the modified files were:
> > 
> > 	xen/common/domain.c
> > 	xen/common/sched_bvt.c
> > 	xen/common/sched_sedf.c
> > 	xen/common/schedule.c
> > 	xen/include/public/vcpu.h
> > 	xen/include/xen/sched.h
> > 	xen/arch/x86/smpboot.c
> > 
> > 	How should I proceed to send the patch?
>  
> 
> There is a guide here: 
> 
> http://wiki.xensource.com/xenwiki/SubmittingXenPatches
> 
> For examples, see:
> 
> http://lists.xensource.com/archives/html/xen-devel/
> 
> Also see these guidelines: 
> 
> http://www.zip.com.au/~akpm/linux/patches/stuff/tpp.txt
> http://linux.yyz.us/patch-format.html
> 
> I would think replying to your patch email with the performance measurement
> report would help keep them together for people. 
> 
> Tim 
> 
> 
> 

Re: Patch question (Hyper-Threading support)

[Keir Fraser]

On 28 Mar 2006, at 01:52, Elizeu Santos-Neto wrote:

> 	Well, based on some quick inspections on the that I checked out from
> the mercurial repository this afternoon, it seems that the HT support
> has been already implemented. :-)
>
> 	Thus, here is a suggestion, it would be nice to make more details
> available about who is doing what and in which phase of development
> things are, this information could avoid duplicated efforts.

HT has been supported for years, so its development hasn't overlapped 
with your own work. The existing support isn't very HT aware though. 
For example, the schedulers treat HT pairs just the same as two 
separate physical processors, so there may be more performance to be 
had by making Xen a bit smarter.

  -- Keir

Re: Patch question (Hyper-Threading support)

[Elizeu Santos-Neto]

	Hi Keir,

	Thanks for the answer.

	Possibly I used an out of date Xen-3.0.testing source code.  

	What I meant by HT support was the construction of sibling maps by the
Guest OS smp-kernel using information about the core ids from the VCPU. 

	More precisely, I've found the following comment in the file:

	./linux-2.6-xen-sparse/arch/xen/kernel/smpboot.c

	 /* Currently, Xen gives no dynamic NUMA/HT info. */

	Thus, I modified this file and a couple of other to allow the guest os
kernel to build the siblings map properly.

	Regards,
	Eli
	
On Tue, 2006-03-28 at 08:32 +0100, Keir Fraser wrote:
> On 28 Mar 2006, at 01:52, Elizeu Santos-Neto wrote:
> 
> > 	Well, based on some quick inspections on the that I checked out from
> > the mercurial repository this afternoon, it seems that the HT support
> > has been already implemented. :-)
> >
> > 	Thus, here is a suggestion, it would be nice to make more details
> > available about who is doing what and in which phase of development
> > things are, this information could avoid duplicated efforts.
> 
> HT has been supported for years, so its development hasn't overlapped 
> with your own work. The existing support isn't very HT aware though. 
> For example, the schedulers treat HT pairs just the same as two 
> separate physical processors, so there may be more performance to be 
> had by making Xen a bit smarter.
> 
>   -- Keir
> 
> 

Re: Patch question (Hyper-Threading support)

[Keir Fraser]

On 28 Mar 2006, at 09:28, Elizeu Santos-Neto wrote:

> 	What I meant by HT support was the construction of sibling maps by the
> Guest OS smp-kernel using information about the core ids from the VCPU.
>
> 	More precisely, I've found the following comment in the file:
>
> 	./linux-2.6-xen-sparse/arch/xen/kernel/smpboot.c
>
> 	 /* Currently, Xen gives no dynamic NUMA/HT info. */
>
> 	Thus, I modified this file and a couple of other to allow the guest os
> kernel to build the siblings map properly.

Ah, I see. Those platform files (e.g., smpboot.c, mpparse.c. ,,,) are 
all kept in sync with linux-2.6 codebase. We generally don't modify 
them much for Xen and periodically (every couple of months) I pull in 
the latest versions from linux. That's what's happened here.

  -- Keir