hack for getting suspend/resume to half work on an IBM Thinkpad x60s [SMP]

[John Baldwin]

On Tuesday 03 October 2006 17:33, Andrea Bittau wrote:
> On Tue, Oct 03, 2006 at 02:03:31PM -0700, Nate Lawson wrote:
> > I agree.  The standard switch to protected mode, paging, etc. needs to 
> > be performed and then resume from the saved register context.
> 
> I guess my point was that there are two pieces of code that do that:
> 1) mpboot.s bootMP() used by system bootstrap and what my current patch 
uses.  I
>    think this is what you guys are suggesting to use, and I'm doing it 
anyway in
>    my patch, but I just want to be the devil's advocate =D.
> 
> 2) acpi_wakecode.S wakeup_16() used by the BSP to wake itself up.  This is 
what
>    I was suggesting should be generalized and used by the other cores too.  
The
>    difference of this code as opposed to #1 is that #2 can "cheat".  That 
is, we
>    can create the code for #2 on the fly and do stuff like mov old_eax,eax 
etc
>    and don't have to be smart about figuring out where the CPU should land 
and
>    how it should initialize itself [as in the case of #1].
> 
> I'm just wondering whether we should do something about the assembly "code
> duplication" in #1 and #2.  I understand they serve a different purpose, but
> arguably, they do the same thing: real-mode -> jump in kernel.  What is
> different is what happens once in kernel mode: boot or resume?  That 
difference
> could be coded in the C part of the kernel leaving a single asm entry point 
both
> for bootstrap and wakeup code.  Am I making any sense? =D

1) is already tailored to work with starting up an extra AP based on the 
STARTUP IPI, so I think we should reuse it for starting up the cores.  I need 
to think about it more though.  First we need to get APIC with UP working 
though.

-- 
John Baldwin