Re: [RFC PATCH 1/5] mm: intorduce __GFP_UNMAPPED and unmapped_alloc()

[Luis Chamberlain]

On Tue, Mar 28, 2023 at 10:52:08AM -0700, Luis Chamberlain wrote:
> On Tue, Mar 28, 2023 at 06:37:13PM +0100, Matthew Wilcox wrote:
> > On Tue, Mar 28, 2023 at 10:18:50AM -0700, Luis Chamberlain wrote:
> > > differences with eBPF programs is that modules *can* be rather large
> > > in size. What is the average size of modules? Well let's take a look:
> > > 
> > > mcgrof@bigtwin /mirror/code/mcgrof/linux-next (git::master)$ find ./
> > > -name \*.ko| wc -l
> > > 9173
> > 
> > ummm ... wc -c, surely?
> 
> That's the number of allmodconfig modules found.
> 
> mcgrof@fulton ~/linux (git::sysctl-next)$ find ./ -name \*.ko| head -2
> ./arch/x86/crypto/twofish-x86_64.ko
> ./arch/x86/crypto/serpent-avx2.ko
> mcgrof@fulton ~/linux (git::sysctl-next)$ find ./ -name \*.ko| head -2 |
> wc -l
> 2
> mcgrof@fulton ~/linux (git::sysctl-next)$ find ./ -name \*.ko| head -2 |
> wc -c
> 70
> 
> wc -c would give a lot more. wc -l gives me the module count.
> 
> > > mcgrof@bigtwin /mirror/code/mcgrof/linux-next (git::master)$ find ./
> > > -name \*.ko|  xargs stat -c "%s - %n" | sort -n -k 1 -r | tail
> > > -$((9173-5)) | awk 'BEGIN {sum=0} {sum+=$1} END {print sum/NR/1024}'
> > > 160.54
> > 
> > ... which invalidates all of these.
> 
> Not sure ? But regardless the *.text* lookup is what we care for though
> which was later.

Which gets me thinking it'd be super nice if kmod tools supported
querying this for us, then no fat finger could mess up the math:

For all modules available:
 * Average module size
 * Average .text module size

For only modules loaded:
 * Average module size
 * Average .text module size

  Luis