Re: [PATCH] x86: fix pte_flags() to only return flags, fix lguest (updated)

[Rusty Russell <rusty@rustcorp.com.au>]

On Tuesday 22 July 2008 19:04:32 Ingo Molnar wrote:
> i'm wondering. My randconfig tests boot up an lguest enabled kernel
> every 30 minutes or so:
>
>  config-Mon_Jul_21_19_05_54_CEST_2008.good:CONFIG_LGUEST=y
>  config-Mon_Jul_21_19_43_13_CEST_2008.good:CONFIG_LGUEST=y
>  config-Mon_Jul_21_19_47_40_CEST_2008.good:CONFIG_LGUEST=y
>  config-Mon_Jul_21_20_37_41_CEST_2008.good:CONFIG_LGUEST=y
>  config-Mon_Jul_21_22_11_42_CEST_2008.good:CONFIG_LGUEST=y
>  config-Mon_Jul_21_22_16_59_CEST_2008.good:CONFIG_LGUEST=y
>  config-Mon_Jul_21_22_32_22_CEST_2008.good:CONFIG_LGUEST=y
>  config-Mon_Jul_21_23_25_55_CEST_2008.good:CONFIG_LGUEST=y
>  config-Mon_Jul_21_23_51_29_CEST_2008.good:CONFIG_LGUEST=y
>
> Would it be possible to have some really stupid lguest self-test which
> would complain spectacularly in the host kernel if it fails to reach
> some minimal user-space?
>
> Something that could be self-contained within a single bzImage. (i.e. it
> would contain a minimalistic image of some sort with a very minimalistic
> userspace component as well - or something like that)

Well, adding "make -C Documentation/lguest" to the build is a good start (this 
finds those "e820.h not longer includable from userspace" bugs).

Secondly, if you put the resulting Documentation/lguest/lguest somewhere on 
your booting test machine, it can just do something like

	./lguest 64 /boot/vmlinuz-`uname -r` | grep 'VFS: Unable to mount root'

Won't quite test userspace, but it's easy and will get the worst breakage.
If we want to be more ambitious, I'd suggest a tiny initrd with a 
statically-linked /hello_world to test userspace:

	./lguest --initrd=/boot/hello_world.initrd 64 /boot/vmlinuz-`uname -r` \
		rdinit=/hello_world | grep 'Hello world'

I can create one (and test the example) for you if you're interested?

Thanks,
Rusty.