Re: [1/8,v3] NUMA Hotplug Emulator: add function to hide memoryregion via e820 table.

[David Rientjes]

On Fri, 19 Nov 2010, Shaohui Zheng wrote:

> > > > > Index: linux-hpe4/arch/x86/kernel/e820.c
> > > > > ===================================================================
> > > > > --- linux-hpe4.orig/arch/x86/kernel/e820.c	2010-11-15 17:13:02.483461667 +0800
> > > > > +++ linux-hpe4/arch/x86/kernel/e820.c	2010-11-15 17:13:07.083461581 +0800
> > > > > @@ -971,6 +971,7 @@
> > > > >  }
> > > > >  
> > > > >  static int userdef __initdata;
> > > > > +static u64 max_mem_size __initdata = ULLONG_MAX;
> > > > >  
> > > > >  /* "mem=nopentium" disables the 4MB page tables. */
> > > > >  static int __init parse_memopt(char *p)
> > > > > @@ -989,12 +990,28 @@
> > > > >  
> > > > >  	userdef = 1;
> > > > >  	mem_size = memparse(p, &p);
> > > > > -	e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
> > > > > +	e820_remove_range(mem_size, max_mem_size - mem_size, E820_RAM, 1);
> > > > > +	max_mem_size = mem_size;
> > > > >  
> > > > >  	return 0;
> > > > >  }
> > > > 
> > > > This needs memmap= support as well, right?
> > > we did not do the testing after combine both memmap and numa=hide paramter, 
> > > I think that the result should similar with mem=XX, they both remove a memory
> > > region from the e820 table.
> > > 
> > 
> > You've modified the parser for mem= but not memmap= so the change needs 
> > additional support for the latter.
> > 
> 
> the parser for mem= is not modified, the changed parser is numa=, I add a addtional
> option numa=hide=.
> 

The above hunk is modifying the x86 parser for the mem= parameter.

> > Your patchset doesn't do that, I'm talking specifically about the amount 
> > of memory left behind so that the kernel at least still boots.  That seems 
> > to be a function of e820_hide_mem() to do some sanity checking so we 
> > actually still get a kernel rather than the responsibility of the 
> > command-line parser.
> 
> How much memory is enough to make sure the kernel can still boot, it is very 
> hard to measure. it is almost impossible to get the exact data. I try to leave very 
> few memory to kernel(hide most memory with numa=hide), it cause a panic directly.
> 
> I have no idea about it, do you have any suggestions?
> 

Yes, I think we should use FAKE_NODE_MIN_SIZE to represent the smallest 
node that may be added and so the appropriate behavior or e820_hide_mem() 
would be to leave at least this quantity behind for the kernel to be 
loaded.
--
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/