[uml-devel] skas3 causes kernel panic before UML finishes booting

[Dan F. <ds...@gb...>]

Howdy -

	I'm running Gentoo Linux 1.4rc4 with the latest version of all
Gentoo tools and kernel version 2.4.20-gentoo-r5 (uname -r output) with the
skas3 patch cleanly applied on a x86 box.

	In the host and UML kernel config file, CONFIG_PROC_MM=y is set.
	In the UML kernel config file, CONFIG_MODE_SKAS=y is set.
	/proc/mm does exist on the host system with (octal) 222 perms set.

	The UML kernel source tree (built in a separate directory, per
instructions), is based on the plain vanilla 2.4.20 kernel along with
uml-patch-2.4.20-6 applied, 'make config' (with ARCH=um) and the other
normal steps done to build UML properly including making sure the built
modules were put in the root_fs image via a loopback mount.

	UML utilities version is 20030605.

	I start a single UML instance (with no others running) with:

	./linux eth0=mcast ubd0=../cow/root_fs_cow_r1,../root_fs/root_fs_slack81 umid=r1

	It boots the UML kernel, and first line says it detects skas3
successfully and uses it. Normal looking messages - no errors seen. At the
end is this:

[...snipped for brevity - ask if you'd like to see the whole thing...]
0VFS: Mounted root (ext2 filesystem) readonly.
Mounted devfs on /dev
Kernel panic: switch_mm_skas - PTRACE_SWITCH_MM failed, errno = 1

In idle task - not syncing

	And it returns to the host shell prompt, not having started UML.

	What am I doing wrong? It works fine in tt mode, not at all with
skas3 applied or enabled...so I know my setup is good. errno=1 is EPERM,
isn't it? I get this error regardless of whether I run it as my unprivved
user or as root. (Although, obviously, I prefer to run things non-root!)

	In the host kernel source file arch/i386/kernel/ptrace.c is this:

        case PTRACE_SWITCH_MM: {
                struct mm_struct *old = child->mm;
                struct mm_struct *new = proc_mm_get_mm(data);
 
                if(IS_ERR(new)){
                        ret = PTR_ERR(new);
                        break;
                }
 
                atomic_inc(&new->mm_users);
                child->mm = new;
                child->active_mm = new;
                mmput(old);
                ret = 0;
                break;
        }

	....so I know the skas3 patch applied there fine, and there really
isn't much that could cause it to fail there? Since it doesn't complete the
boot, can't fire up a debugger on it easily :(

	More than happy to apply any custom patches to the kernel for
further debugging if desired - it's my laptop so I can reboot at will. It's
also a reasonably 'fast' box so it's no sweat to recompile and reboot.

	Well, I've got a binary for no-skas-enabled, and a binary for
skas-enabled, so I can easily switch between the two for testing. Anything
I can do to help out with testing/debugging of this problem? If it isn't
some 'stupid operator error', of course. ;)

	Final comment... I took all the defaults (because they looked
reasonable and appropriate) when doing 'make config ARCH=um' in the UML
kernel build.  So, nothing exotic there.

	Also scourged all the web pages at the sourceforge web site for
this project and didn't find any mention of this as a known bug or issue,
nor a common problem with workaround/fix. Ditto with a Google search.

	UML's a most *impressive* piece of work with so much raw potential,
so I am quite enthused and eagerly seek assistance in ironing out this one.
I will be rather disappointed (and quite sheepish) if this turns out to be
a 'stupid user error' ;) Hopefully not, though!

-Dan