Could you please merge the x86_64 EFI boot support patchset?

Could you please merge the x86_64 EFI boot support patchset?

[Huang, Ying]

Hi, Linus,

Could you please merge the following patchset:

[PATCH 0/2 -v3] x86_64 EFI boot support
[PATCH 1/2 -v3] x86_64 EFI boot support: EFI frame buffer driver
[PATCH 2/2 -v3] x86_64 EFI boot support: EFI boot document

The patchset has been in -mm tree from 2.6.23-rc2-mm2 on. Andrew Moton
had suggested it to be merged into 2.6.24 during early merge window of
2.6.24. It was not merged because the 32-bit boot protocol had not been
done at that time.

Now, the 32-bit boot protocol has been merged into 2.6.24. And this
patch has been in x86 patch queue.

I know that it is a little late for this patchset to be merged into
2.6.24. But this patchset is very simple, just adds a framebuffer
driver, so it is impossible for this patchset to break anything. And
this patchset will be helpful for people have machine with UEFI 64
firmware instead of legacy BIOS.

Best Regards,
Huang Ying

Re: Could you please merge the x86_64 EFI boot support patchset?

[Paul Jackson]

In Nov 2007, Huang Ying wrote:
> Could you please merge the following patchset:
> 
> [PATCH 0/2 -v3] x86_64 EFI boot support
> [PATCH 1/2 -v3] x86_64 EFI boot support: EFI frame buffer driver
> [PATCH 2/2 -v3] x86_64 EFI boot support: EFI boot document

Huang - what has become of this patchset?

We (SGI) have designs on a big honkin NUMA box using x86_64 arch, and
we are runnning up against the arbitrary limits on the memory map size
due to the ancient 4k zero page size limit (hence H. Peter Anvin added
to the CC list, since he knows a bazillion times more about any such
limits than I do.).  The limits on 128 or so local chunks of memory
imposed by the kernel code that invokes Int 15 E820h are too small for
us.

Since we are already accustomed to dealing with EFI on our IA64
Itanium boxes, I'm figuring that it will be easier in the short run,
and better in the long run, to just use EFI on these upcoming big
x86_64 NUMA boxes.

But we'd need to get this patch, or something equivalent, into the
Linux kernel in the next 2.6.* cycle or so, for this to work for us.

Also, is there a current version of this patch set available, against
either 2.6.24-rc8-mm1 or Ingo's recent x86 git tree?  I should try it
out.

-- 
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <pj@sgi.com> 1.940.382.4214

Re: Could you please merge the x86_64 EFI boot support patchset?

[Huang, Ying]

On Wed, 2008-01-23 at 01:00 -0600, Paul Jackson wrote:
> In Nov 2007, Huang Ying wrote:
> > Could you please merge the following patchset:
> > 
> > [PATCH 0/2 -v3] x86_64 EFI boot support
> > [PATCH 1/2 -v3] x86_64 EFI boot support: EFI frame buffer driver
> > [PATCH 2/2 -v3] x86_64 EFI boot support: EFI boot document
> 
> Huang - what has become of this patchset?

This patchset has been merged into Linux 2.6.24.

> We (SGI) have designs on a big honkin NUMA box using x86_64 arch, and
> we are runnning up against the arbitrary limits on the memory map size
> due to the ancient 4k zero page size limit (hence H. Peter Anvin added
> to the CC list, since he knows a bazillion times more about any such
> limits than I do.).  The limits on 128 or so local chunks of memory
> imposed by the kernel code that invokes Int 15 E820h are too small for
> us.
> 
> Since we are already accustomed to dealing with EFI on our IA64
> Itanium boxes, I'm figuring that it will be easier in the short run,
> and better in the long run, to just use EFI on these upcoming big
> x86_64 NUMA boxes.

Unfortunately, the new EFI support patches do not use EFI memory map for
system boot up (just for runtime service support). The EFI memory map is
converted into E820 memory map in bootloader. The main reason for this
change is to remove the duplication between E820 memory map and EFI
memory map handling code.

So, I think the resolution for your problem is the "struct setup_data"
mechanism proposed by H. Peter Anvin. That is a linked list data
structure for boot parameter without size limitation. I have ever writen
a patch for it, but there are some issues for implementation scheme.
Most people think it should be based on the "early reservation/early
allocation" mechanism from Andi Kleen. So I am waiting that is merged by
-mm or git-x86.

Best Regards,
Huang Ying

Re: Could you please merge the x86_64 EFI boot support patchset?

[Paul Jackson]

Huang wrote:
> This patchset has been merged into Linux 2.6.24.

Excellent.

> Unfortunately, the new EFI support patches do not use EFI memory map for
> system boot up ...  So, I think the resolution for your problem is the
> "struct setup_data" mechanism proposed by H. Peter Anvin.

So you're saying that the EFI in the kernel now still won't support more
than 128 or so chunks of memory in the boottime memory map, because it
still goes via the legacy E820h memory map code?

I'll have to study the code more and give it a try.

Are you optimistic that some variation of H. Peter Anvin's "struct
setup_data" mechanism will make it into 2.6.25 or thereabouts?

-- 
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <pj@sgi.com> 1.940.382.4214

Re: Could you please merge the x86_64 EFI boot support patchset?

[Huang, Ying]

On Wed, 2008-01-23 at 01:46 -0600, Paul Jackson wrote:
> Huang wrote:
> > This patchset has been merged into Linux 2.6.24.
> 
> Excellent.
> 
> > Unfortunately, the new EFI support patches do not use EFI memory map for
> > system boot up ...  So, I think the resolution for your problem is the
> > "struct setup_data" mechanism proposed by H. Peter Anvin.
> 
> So you're saying that the EFI in the kernel now still won't support more
> than 128 or so chunks of memory in the boottime memory map, because it
> still goes via the legacy E820h memory map code?

The kernel with EFI support can not support more than 128 or so chunks
of memory. Because the EFI memory map is converted to E820 memory map in
boot-loader (such as elilo).

> I'll have to study the code more and give it a try.
> 
> Are you optimistic that some variation of H. Peter Anvin's "struct
> setup_data" mechanism will make it into 2.6.25 or thereabouts?

I think it may be merged into 2.6.26.

Best Regards,
Huang Ying

Re: Could you please merge the x86_64 EFI boot support patchset?

[Paul Jackson]

Ying Huang wrote:
> > Are you optimistic that some variation of H. Peter Anvin's "struct
> > setup_data" mechanism will make it into 2.6.25 or thereabouts?
> 
> I think it may be merged into 2.6.26.

That (2.6.26) is just barely within the schedule SGI needs.
I hope the chances of making it by then are good.  I'm no
expert in this area, but I'll do what little I can to help
review and test.

Good luck ;).

-- 
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <pj@sgi.com> 1.940.382.4214