Re: [microblaze-uclinux] [patch] memory management initialisationcleanup

[John Williams <jwilliams@xxxxxxxxxxxxxx>]

Hey Yashi,

Yasushi SHOJI wrote:
> At Fri, 30 Jul 2004 15:29:46 +1000,
> John Williams wrote:
> 
>>Basically I've just lifted the mechanism used in arm/armnommu kernels. 
>>They take a clean approach to the whole thing.
> 
> yup, it is clean, but it has some unneeded code for mmu-less cpu, ie
> most of free_memmap(), {O,V}_PFN_{UP,DOWN}.  but the patch looks good
> for the start.

Yeah.. it was a quick hack, so I just concentrated on getting it 
working.  It could certainly be trimmed down a little.

> IMHO, we need to provide __pa() __va() family macro for seemless
> integration for arch independent code.  but we should keep
> arch/microblaze/* as clean as we can.

Not quite sure what you mean here - __pa() and __va() macros are already 
defined for microblaze in asm-microblaze/page.h  They just reduce to 
identity functions of course since physical == virtual for nommu system.

> or, we can keep arch/microblaze/mm/* code in sync with arm so that we
> can play with other fun stuff ;p
> 
> any plan in your mind, John?

Well, our research agenda here basically requires discontigmem, and from 
a quick look armnommu seemed the only no-mmu target with clean support 
(indeed /any/ support).  So, I'm not committed to tracking arm as such, 
but rather just providing the functionality.

This arch-specific mem mgmt stuff changes so rarely, so I don't think we 
need to achieve patch-level compatibility with armnommu or any other 
target.  Common architectural roots should be enough to do what we need.

>>This patch reverts the link scripts (mbvanilla.ld and suzaku.ld) to page 
>>align the __init_end, as discussed in my previous email.  Yashi (and 
>>anyone else who's interested), can you please give this a try and let me 
>>know if it works for you?
> 
> 
> it worked on SUZAKU without discon mem.  I haven't tested with discon
> mem setup, yet.  

That's good news.  I'd be v. surprised if discontigmem works yet, I've 
certainly not tested it.  I'll try to look at it this week.

> we need to modify a few line to tell the kernel how
> many banks we have on each platform.

I'm picturing a function that goes in {mbvanilla|suzaku}.c, which builds 
a meminfo struct that describes that platforms memory architecture. 
Then the common code in kernel/setup.c passes this off to bootmem_init() 
and paging_init().

Anyway I'll apply the first patch now - it should be ready for cvs 
updates shortly.

John
___________________________
microblaze-uclinux mailing list
microblaze-uclinux@itee.uq.edu.au
Project Home Page : http://www.itee.uq.edu.au/~jwilliams/mblaze-uclinux
Mailing List Archive : http://www.itee.uq.edu.au/~listarch/microblaze-uclinux/