Re: mmots: memory-hotplug: implementregister_page_bootmem_info_section of sparse-vmemmap fix

[Michal Hocko]

On Fri 11-01-13 20:38:33, Tang Chen wrote:
> On 01/11/2013 08:12 PM, Michal Hocko wrote:
> >On Fri 11-01-13 20:06:25, Tang Chen wrote:
> >>On 01/11/2013 06:47 PM, Michal Hocko wrote:
> >>>>
> >>>>Darn! And now that I am looking at the patch closer it is too x86
> >>>>centric so this cannot be in the generic code. I will try to cook
> >>>>something better. Sorry about the noise.
> >>>
> >>>It is more complicated than I thought. One would tell it's a mess.
> >>>The patch bellow fixes the compilation issue but I am not sure we want
> >>>to include memory_hotplug.h into arch/x86/mm/init_64.c. Moreover
> >>>
> >>>+void register_page_bootmem_memmap(unsigned long section_nr,
> >>>+				  struct page *start_page, unsigned long size)
> >>>+{
> >>>+	/* TODO */
> >>>+}
> >>>
> >>>for other archs would suggest that the code is not ready yet. Should
> >>>this rather be dropped for now?
> >>
> >>Hi Michal,
> >>
> >>Do you mean remove register_page_bootmem_memmap() from other
> >>architectures ?
> >
> >No I meant the patch to be dropped until it gets implementation for
> >other architectures or the users of the function would be explicit about
> >archs which are supported. What happens if the implementation is empty
> >will the generic code work properly? From my very limitted understanding
> >of the code it won't.
> 
> Hi Michal,
> 
> Hum, I see. Thank you for your remind. :)
> register_page_bootmem_info_section() will be different in other
> architectures if register_page_bootmem_memmap() is empty.

Not sure I understand what "different" means here but I suspect it would
be buggy. Is that correct?

> I think we can post a patch to make register_page_bootmem_info_section()
> the same as before, and we just implement the x86 version first. So that
> it will have no harm to other architectures.

I haven't followed the previous versions of the patch - I have noticed
this being broken only because it failed during my automatic build
testing when merging new mmots tree into mm git tree. I have no
objections for further patches of course but this one seems to be buggy
so it should be dropped until a fixed version is available.

> How do you think ?
> 
> Thanks. :)
> 
> >
> 

-- 
Michal Hocko
SUSE Labs
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/