Re: [Lhms-devel] [PATCH 4/4] V4 ia64 SPARSEMEM - SPARSEMEM code

Re: [Lhms-devel] [PATCH 4/4] V4 ia64 SPARSEMEM - SPARSEMEM code

[Dave Hansen]

On Thu, 2005-09-22 at 12:14 -0400, Bob Picco wrote:
> This patch is the minimal set of changes required by ia64 to use
> SPARSEMEM.

I have a couple of little style comments.

> +#ifdef CONFIG_SPARSEMEM
> +	efi_memmap_walk(register_sparse_mem, NULL);
> +#endif
> +	sparse_init();
> +
>  	efi_memmap_walk(filter_rsvd_memory, count_node_pages);
>  
> +#ifdef CONFIG_VIRTUAL_MEM_MAP
>  	vmalloc_end -= PAGE_ALIGN(max_low_pfn * sizeof(struct page));
>  	vmem_map = (struct page *) vmalloc_end;
>  	efi_memmap_walk(create_mem_map_page_table, NULL);
>  	printk("Virtual mem_map starts at 0x%p\n", vmem_map);
> +#endif

It would be nice if these kind of .c file #ifdefs could be broken out a
little more.  For instance, you could make an arch_sparse_init() that
does all of the sparsemem bits, and make it a do{}while(0) when !
CONFIG_SPARSEMEM.

>  	for_each_online_node(node) {
>  		memset(zones_size, 0, sizeof(zones_size));
> @@ -690,7 +722,9 @@ void __init paging_init(void)
>  
>  		pfn_offset = mem_data[node].min_pfn;
>  
> +#ifdef CONFIG_VIRTUAL_MEM_MAP
>  		NODE_DATA(node)->node_mem_map = vmem_map + pfn_offset;
> +#endif

Is CONFIG_FLAT_NODE_MEM_MAP more appropriate here?  That's how
->node_mem_map is #ifdef'd to begin with in mmzone.h.

> --- linux-2.6.14-rc2.orig/arch/ia64/mm/numa.c	2005-09-21 12:28:49.000000000 -0400
> +++ linux-2.6.14-rc2/arch/ia64/mm/numa.c	2005-09-21 12:29:05.000000000 -0400
> @@ -45,5 +45,30 @@ paddr_to_nid(unsigned long paddr)
>  		    paddr < node_memblk[i].start_paddr + node_memblk[i].size)
>  			break;
>  
> -	return (i < num_node_memblks) ? node_memblk[i].nid : (num_node_memblks ? -1 : 0);
> +	return (i < num_node_memblks) ?
> +		node_memblk[i].nid : (num_node_memblks ? -1 : 0);
>  }

Looks like just a whitespace change.

-- Dave

Re: [Lhms-devel] [PATCH 4/4] V4 ia64 SPARSEMEM - SPARSEMEM code changes

[Matthew Wilcox]

On Thu, Sep 22, 2005 at 09:36:16AM -0700, Dave Hansen wrote:
> >  
> > -	return (i < num_node_memblks) ? node_memblk[i].nid : (num_node_memblks ? -1 : 0);
> > +	return (i < num_node_memblks) ?
> > +		node_memblk[i].nid : (num_node_memblks ? -1 : 0);
> >  }
> 
> Looks like just a whitespace change.

But really far less radical than it deserves ... nested conditional
expressions?  Come on.  How about:

	if (i < num_node_memblks)
		return node_memblk[i].nid;
	if (num_node_memblks)
		return -1;
	return 0;

Re: [Lhms-devel] [PATCH 4/4] V4 ia64 SPARSEMEM - SPARSEMEM code changes

[Bob Picco]

Dave Hansen wrote:	[Thu Sep 22 2005, 12:36:16PM EDT]
> On Thu, 2005-09-22 at 12:14 -0400, Bob Picco wrote:
> > This patch is the minimal set of changes required by ia64 to use
> > SPARSEMEM.
> 
> I have a couple of little style comments.
> 
> > +#ifdef CONFIG_SPARSEMEM
> > +	efi_memmap_walk(register_sparse_mem, NULL);
> > +#endif
> > +	sparse_init();
> > +
> >  	efi_memmap_walk(filter_rsvd_memory, count_node_pages);
> >  
> > +#ifdef CONFIG_VIRTUAL_MEM_MAP
> >  	vmalloc_end -= PAGE_ALIGN(max_low_pfn * sizeof(struct page));
> >  	vmem_map = (struct page *) vmalloc_end;
> >  	efi_memmap_walk(create_mem_map_page_table, NULL);
> >  	printk("Virtual mem_map starts at 0x%p\n", vmem_map);
> > +#endif
> 
> It would be nice if these kind of .c file #ifdefs could be broken out a
> little more.  For instance, you could make an arch_sparse_init() that
> does all of the sparsemem bits, and make it a do{}while(0) when !
> CONFIG_SPARSEMEM.
I could do this.  It would eliminate one conditional.  Eventually I'm hope
for only SPARSEMEM but that is a ways off.
> 
> >  	for_each_online_node(node) {
> >  		memset(zones_size, 0, sizeof(zones_size));
> > @@ -690,7 +722,9 @@ void __init paging_init(void)
> >  
> >  		pfn_offset = mem_data[node].min_pfn;
> >  
> > +#ifdef CONFIG_VIRTUAL_MEM_MAP
> >  		NODE_DATA(node)->node_mem_map = vmem_map + pfn_offset;
> > +#endif
> 
> Is CONFIG_FLAT_NODE_MEM_MAP more appropriate here?  That's how
> ->node_mem_map is #ifdef'd to begin with in mmzone.h.
Well I think this is doable but probaably shouldn't. This code is very
specific to CONFIG_VIRTUAL_MEM_MAP.
> 
> > --- linux-2.6.14-rc2.orig/arch/ia64/mm/numa.c	2005-09-21 12:28:49.000000000 -0400
> > +++ linux-2.6.14-rc2/arch/ia64/mm/numa.c	2005-09-21 12:29:05.000000000 -0400
> > @@ -45,5 +45,30 @@ paddr_to_nid(unsigned long paddr)
> >  		    paddr < node_memblk[i].start_paddr + node_memblk[i].size)
> >  			break;
> >  
> > -	return (i < num_node_memblks) ? node_memblk[i].nid : (num_node_memblks ? -1 : 0);
> > +	return (i < num_node_memblks) ?
> > +		node_memblk[i].nid : (num_node_memblks ? -1 : 0);
> >  }
> 
> Looks like just a whitespace change.
Oh dam. This shouldn't have been introduced.  It's not part of SPARSEMEM.
> 
> -- Dave
bob
> 

Re: [Lhms-devel] [PATCH 4/4] V4 ia64 SPARSEMEM - SPARSEMEM code changes

[Bob Picco]

Matthew Wilcox wrote:	[Thu Sep 22 2005, 01:13:40PM EDT]
> On Thu, Sep 22, 2005 at 09:36:16AM -0700, Dave Hansen wrote:
> > >  
> > > -	return (i < num_node_memblks) ? node_memblk[i].nid : (num_node_memblks ? -1 : 0);
> > > +	return (i < num_node_memblks) ?
> > > +		node_memblk[i].nid : (num_node_memblks ? -1 : 0);
> > >  }
> > 
> > Looks like just a whitespace change.
> 
> But really far less radical than it deserves ... nested conditional
> expressions?  Come on.  How about:
> 
> 	if (i < num_node_memblks)
> 		return node_memblk[i].nid;
> 	if (num_node_memblks)
> 		return -1;
> 	return 0;
Yup looks better. As I said to Dave, it shouldn't have been introduced. So this
is a nice patch suggestion for clean up but later.


bob

Re: [Lhms-devel] [PATCH 4/4] V4 ia64 SPARSEMEM - SPARSEMEM code

[Dave Hansen]

On Thu, 2005-09-22 at 13:59 -0400, Bob Picco wrote:
> > > +#ifdef CONFIG_VIRTUAL_MEM_MAP
> > >             NODE_DATA(node)->node_mem_map = vmem_map + pfn_offset;
> > > +#endif
> > 
> > Is CONFIG_FLAT_NODE_MEM_MAP more appropriate here?  That's how
> > ->node_mem_map is #ifdef'd to begin with in mmzone.h.
>
> Well I think this is doable but probaably shouldn't. This code is very
> specific to CONFIG_VIRTUAL_MEM_MAP.

FLAT_NODE_MEM_MAP depends on only !SPARSEMEM, as does VIRTUAL_MEM_MAP.
So, I guess they're functionally equivalent right now.  It might be most
nice to at least add a comment, saying that it actually depends on both.



-- Dave

Re: [Lhms-devel] [PATCH 4/4] V4 ia64 SPARSEMEM - SPARSEMEM code changes

[Bob Picco]

Dave Hansen wrote:	[Thu Sep 22 2005, 02:05:07PM EDT]
> On Thu, 2005-09-22 at 13:59 -0400, Bob Picco wrote:
> > > > +#ifdef CONFIG_VIRTUAL_MEM_MAP
> > > >             NODE_DATA(node)->node_mem_map = vmem_map + pfn_offset;
> > > > +#endif
> > > 
> > > Is CONFIG_FLAT_NODE_MEM_MAP more appropriate here?  That's how
> > > ->node_mem_map is #ifdef'd to begin with in mmzone.h.
> >
> > Well I think this is doable but probaably shouldn't. This code is very
> > specific to CONFIG_VIRTUAL_MEM_MAP.
> 
> FLAT_NODE_MEM_MAP depends on only !SPARSEMEM, as does VIRTUAL_MEM_MAP.
> So, I guess they're functionally equivalent right now.  It might be most
> nice to at least add a comment, saying that it actually depends on both.
> 
> 
> 
> -- Dave
> 
Okay. I'll add a comment before VIRTUAL_MEM_MAP in Kconfig file. That will
cover all uses of VIRTUAL_MEM_MAP.

thanks,

bob