Re: linux kernel 2.6.37

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Tue, Jan 25, 2011 at 01:43:42PM -0500, Konrad Rzeszutek Wilk wrote:
> On Tue, Jan 25, 2011 at 07:33:14PM +0300, Vasiliy G Tolstov wrote:
> > On Tue, 2011-01-25 at 09:45 -0500, Konrad Rzeszutek Wilk wrote:
> > > On Mon, Jan 24, 2011 at 11:19:02PM +0300, Vasiliy G Tolstov wrote:
> > > > On Fri, 2011-01-21 at 19:16 +0300, Vasiliy G Tolstov wrote:
> > > > > On Fri, 2011-01-21 at 10:30 -0500, Konrad Rzeszutek Wilk wrote:
> > > > > 
> > > > > > Do you see any messages during bootup about memory being returned? Look for
> > > > > > xen_release_chunk .. and pages freed ?
> > > > [snip]
> > > > 
> > > > Hello. Konrad, do You have any progress about this issue?
> > > 
> > > Try booting your kernel with 'debug loglevel=10 memblock=debug'. I know that at least
> > > 64MB is used for SWIOTLB (which you don't need - you can disable that since you are
> > > not doing any PCI passthrough), 6MB for kernel + kernel data. The "memblock=debug" will
> > > give you the details to figure out who else is eating the big swaths of memory.
> > > 
> > 
> > Nice suggestion about swiotlb. Can You check my debug output and helps
> > me to find what else eats memory?
> 
> Why don't you run this through | grep memblock,
> then awk - extract those regions .. something like this:
> 
>  grep memblock | serve_range:// | sed s/\\[// | sed s/\\]// | sed s/-/\ / | awk --non-decimal-data ' { a=$2;b=$1;printf "%s %s\n", a-b, $3 }
> 
> that should give you an idea what is eating what memory. For example
> NODE_DATA looks to be eating a bunch of 20 pages (20*4096)= 81920 bytes.
> 
> thought later on it free's some of them.

Did you do what I asked you try?

Also what is the memory consumption when you boot Linux as HVM guest? You MUST have
the same amount of memory in both guests, so try the test with

mem=512
maxmem=512

for both HVM and PV. What is the difference in the log console when you use memblock=debug?
Do you see any big differences?