Fix ARCH=um segfault on x86-64.

Apparently, nobody other than me has ever attempted to use User Mode Linux built from 2.6.28 on x86-64, because it doesn't work. It still doesn't work in current git. I complained about it not working back in January:

http://sourceforge.net/mailarchive/forum.php?thread_name=200901130159.04389.rob%40landley.net&forum_name=user-
mode-linux-devel
http://lkml.indiana.edu/hypermail/linux/kernel/0901.2/00669.html

And today, I bothered to track down why.

This is the commit that broke it, when Peter Anvin merged x86 and x86-64 for ARCH=um: http://kernel.org/hg/linux-2.6/rev/117978

Here's a patch that fixes it for me:

diff -r 178a096e9e38 arch/um/Kconfig.x86
@@ -26,9 +26,8 @@
 	def_bool !X86_XADD
 
 config 3_LEVEL_PGTABLES
-	bool "Three-level pagetables (EXPERIMENTAL)" if !64BIT
+	bool
 	default 64BIT
-	depends on EXPERIMENTAL
 	help
 	Three-level pagetables will let UML have more than 4G of physical
 	memory.  All the memory that can't be mapped directly will be treated

What changed is that the resulting .config no longer contains the line "CONFIG_3_LEVEL_PGTABLES=y" (it's not visible, and thus not written out into the config file file). Without that symbol defined, x86-64 dies trying to boot. If you tweak the Kconfig so the symbol gets written out, it starts working again.

I have no idea how ANYBODY has EVER managed to use 2.6.28 User Mode Linux on an x86-64 host. My theory is that nobody ever did. I suspect that very few people use UML anymore now that KVM and the rustyvisor and such are available, and those legacy users still fiddling with it are apparently all either using old versions or 32-bit hosts. (I still like being able to stick printfs into the kernel.)

Here's the panic, in case you're wondering:

$ ./linux rw init=/bin/bash rootfstype=hostfs
Core dump limits :
	soft - 0
	hard - NONE
Checking that ptrace can change system call numbers...OK
Checking syscall emulation patch for ptrace...OK
Checking advanced syscall emulation patch for ptrace...OK
Checking for tmpfs mount on /dev/shm...OK
Checking PROT_EXEC mmap in /dev/shm/...OK
Checking for the skas3 patch in the host:
  - /proc/mm...not found: No such file or directory
  - PTRACE_FAULTINFO...not found
  - PTRACE_LDT...not found
UML running in SKAS0 mode
Adding 4390912 bytes to physical memory to account for exec-shield gap
Linux version 2.6.29-rc7 (landley@driftwood) (gcc version 4.3.2 (Ubuntu 
4.3.2-1ubuntu11) ) #1 Thu Mar 5 21:20:14 CST 2009
Built 1 zonelists in Zone order, mobility grouping on.  Total pages: 9137
Kernel command line: rw init=/bin/bash rootfstype=hostfs root=98:0
PID hash table entries: 256 (order: 8, 2048 bytes)
Dentry cache hash table entries: 8192 (order: 4, 65536 bytes)
Inode-cache hash table entries: 4096 (order: 3, 32768 bytes)
Memory: 29244k available
SLUB: Genslabs=12, HWalign=64, Order=0-3, MinObjects=0, CPUs=1, Nodes=1
Calibrating delay loop... 209.30 BogoMIPS (lpj=1046528)
Mount-cache hash table entries: 256
Checking that host ptys support output SIGIO...Yes
Checking that host ptys support SIGIO on close...No, enabling workaround
Using 2.6 host AIO
bio: create slab <bio-0> at 0
Switched to NOHz mode on CPU #0
io scheduler noop registered (default)
loop: module loaded
Initialized stdio console driver
Using a channel type which is configured out of UML
parse_chan_pair failed for device 1 : Configuration failed
Using a channel type which is configured out of UML
parse_chan_pair failed for device 2 : Configuration failed
Using a channel type which is configured out of UML
parse_chan_pair failed for device 3 : Configuration failed
Using a channel type which is configured out of UML
parse_chan_pair failed for device 4 : Configuration failed
Using a channel type which is configured out of UML
parse_chan_pair failed for device 5 : Configuration failed
Using a channel type which is configured out of UML
parse_chan_pair failed for device 6 : Configuration failed
Using a channel type which is configured out of UML
parse_chan_pair failed for device 7 : Configuration failed
Using a channel type which is configured out of UML
parse_chan_pair failed for device 8 : Configuration failed
Using a channel type which is configured out of UML
parse_chan_pair failed for device 9 : Configuration failed
Using a channel type which is configured out of UML
parse_chan_pair failed for device 10 : Configuration failed
Using a channel type which is configured out of UML
parse_chan_pair failed for device 11 : Configuration failed
Using a channel type which is configured out of UML
parse_chan_pair failed for device 12 : Configuration failed
Using a channel type which is configured out of UML
parse_chan_pair failed for device 13 : Configuration failed
Using a channel type which is configured out of UML
parse_chan_pair failed for device 14 : Configuration failed
Using a channel type which is configured out of UML
parse_chan_pair failed for device 15 : Configuration failed
Console initialized on /dev/tty0
console [tty0] enabled
VFS: Mounted root (hostfs filesystem) on device 0:8.
IRQ 3/console-write: IRQF_DISABLED is not guaranteed on shared IRQs
IRQ 2/console: IRQF_DISABLED is not guaranteed on shared IRQs
IRQ 10/winch: IRQF_DISABLED is not guaranteed on shared IRQs

Pid: 1, comm: swapper Not tainted 2.6.29-rc7

RIP: 0033:[<000000006001b342>]
RSP: 0000000062029dd0  EFLAGS: 00010216
RAX: 00000000622af800 RBX: 00000000621b0000 RCX: 0000000003ffc09f
RDX: fffffffffff02800 RSI: 0000000060313900 RDI: 00000000622af800
RBP: 0000000060d10048 R08: 0000000000000000 R09: 0000000000100000
R10: 0000000000000000 R11: 0000000060197a00 R12: 000000006211f300
R13: 000000006211f300 R14: 0000000060206440 R15: 0000000062020300
Call Trace: 
602058f8:  [<600160c5>] timer_one_shot+0x55/0x80
60205908:  [<6000e4b9>] segv+0x2a9/0x2d0
60205918:  [<6001b342>] __memcpy+0xe/0xac
60205928:  [<6003f376>] tick_dev_program_event+0x36/0xb0
60205958:  [<6003f5c4>] tick_check_oneshot_change+0xf4/0x100
60205968:  [<6002bc6d>] run_timer_softirq+0x1cd/0x210
602059e8:  [<6000e530>] segv_handler+0x50/0xe0
60205a08:  [<6003f250>] tick_handle_periodic+0x10/0x60
60205a48:  [<60026a7d>] do_softirq+0x4d/0x70
60205a68:  [<60026bf2>] irq_exit+0x42/0xa0
60205a88:  [<6000aecf>] do_IRQ+0x2f/0x50
60205aa8:  [<600154e4>] sig_handler_common+0x64/0xe0
60205b30:  [<6001b342>] __memcpy+0xe/0xac
60205b50:  [<600ae3de>] sysfs_new_dirent+0xfe/0x120
60205bd8:  [<600156aa>] sig_handler+0x1a/0x40
60205be8:  [<60015983>] handle_signal+0x73/0xb0
60205c28:  [<60100140>] __restore_rt+0x0/0x10
60205cd8:  [<6001b342>] __memcpy+0xe/0xac

Kernel panic - not syncing: Segfault with no mm

Pid: 1, comm: swapper Not tainted 2.6.29-rc7

RIP: 0033:[<00000000601003a7>]
RSP: 00007fff8026e2a8  EFLAGS: 00000246
RAX: 0000000000000000 RBX: 0000000000001cc4 RCX: ffffffffffffffff
RDX: 0000000000000000 RSI: 0000000000000013 RDI: 0000000000001cc4
RBP: 0000000000001cc0 R08: 00007fff8026e1f0 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000246 R12: 00007fff8026e3b8
R13: 0000000000000004 R14: 00007fff8026e580 R15: 00007fff8026e414
Call Trace: 
602057b8:  [<6003aacd>] up+0x1d/0x50
602057c8:  [<6000e77d>] panic_exit+0x2d/0x50
602057d8:  [<600214ac>] release_console_sem+0x19c/0x1e0
602057e8:  [<6003ac87>] notifier_call_chain+0x37/0x70
60205818:  [<60167cd7>] panic+0xd0/0x164
60205858:  [<60100370>] __sigprocmask+0x10/0x40
60205878:  [<60167df6>] printk+0x8b/0x95
60205898:  [<6001604e>] os_nsecs+0xe/0x30
602058b8:  [<6001b342>] __memcpy+0xe/0xac
602058c8:  [<6000d080>] show_trace+0x60/0xc0
602058e8:  [<6001b148>] show_regs+0x28/0x30
60205908:  [<6000e4c5>] segv+0x2b5/0x2d0
60205918:  [<6001b342>] __memcpy+0xe/0xac
60205928:  [<6003f376>] tick_dev_program_event+0x36/0xb0
60205958:  [<6003f5c4>] tick_check_oneshot_change+0xf4/0x100
60205968:  [<6002bc6d>] run_timer_softirq+0x1cd/0x210
602059e8:  [<6000e530>] segv_handler+0x50/0xe0
60205a08:  [<6003f250>] tick_handle_periodic+0x10/0x60
60205a48:  [<60026a7d>] do_softirq+0x4d/0x70
60205a68:  [<60026bf2>] irq_exit+0x42/0xa0
60205a88:  [<6000aecf>] do_IRQ+0x2f/0x50
60205aa8:  [<600154e4>] sig_handler_common+0x64/0xe0
60205b30:  [<6001b342>] __memcpy+0xe/0xac
60205b50:  [<600ae3de>] sysfs_new_dirent+0xfe/0x120
60205bd8:  [<600156aa>] sig_handler+0x1a/0x40
60205be8:  [<60015983>] handle_signal+0x73/0xb0
60205c28:  [<60100140>] __restore_rt+0x0/0x10
60205cd8:  [<6001b342>] __memcpy+0xe/0xac

Segmentation fault

Rob

Thanks, Bob!

    depends on 64BIT || EXPERIMENTAL

I am sorry that I never have an x86_64 machine to use. :(

Do what you love, f**k the rest! F**k the regulations!

I've just did a build of plain v2.6.28 on amd64 aka x86-64. The resulting image ran fine.

I attached my .config.

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds

So I'm wondering: why does it work for me and not for you?

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds

It does work well x86_64, but my question is that whether this will break i386 or not, since before, CONFIG_3_LEVEL_PGTABLES depends on EXPERIMENTAL on i386, this patch removes it.

Do what you love, f**k the rest! F**k the regulations!

As long as CONFIG_3_LEVEL_PGTABLES is off on 32-bit, it should be fine.

It did work, last I checked, but 3-level page tables on 32-bit is a very rarely used combination, and not useful.

Jeff

Work email - jdike at linux dot intel dot com

I have no idea if my patch is the _right_ fix, I just know I couldn't use UML for 2 months and now I can.

If you enable CONFIG_EXPERIMENTAL, then it writes it out to the config file. If you don't, it hides it and doesn't write it out even though the value would be y. (The visibility predicates affect the resulting data.)

Rob

So this patch should be fine, right? :)

Thanks for your comments!

Do what you love, f**k the rest! F**k the regulations!

Which contains:
CONFIG_3_LEVEL_PGTABLES=y

So the question is, why is your config saving that value, and mine isn't?

Ah, I found it. You enabled CONFIG_EXPERIMENTAL, and I didn't. That's the difference.

Ok, CONFIG_EXPERIMENTAL is required in order for UML to initialize its memory management. That makes a bit more sense why other people haven't seen this...

Rob