Skip to contentSkip to conversation

Of hanging menuconfig [cause found]

DISCUSSION Dr. David Alan Gilbert 4 participants 5 messages Aug 24, 2002 – Aug 25, 2002

lore
Share & tools
lore
#0 lore raw

Hi,
make menuconfig will hang just after the parsing in the
activate_menu loop in the case where the file scripts/lxdialog/lxdialog won't execute. Some error codes in this case are caught; but the case where the file scripts/lxdialog/lxdialog is a binary for the wrong architecture (case 126) is not caught. This is quite easy to trip if you are swapping between native and cross building - you get a couple of errors when you try and build make menuconfig for the first time about wrong binaries; in my case I just deleted those binaries and did the make again; however this failure is silent - it just hangs.

A make mrproper is probably the best thing to do when switching - but
the error case needs catching, and I'm sure there are other similar cases.

Dave

P.S. This was on 2.4.18-rmk7 but I believe it is general. P.P.S. Is it a good idea to keep binaries in the scripts subdirectory?

---------------- Have a happy GNU millennium! ----------------------

/ Dr. David Alan Gilbert    | Running GNU/Linux on Alpha,68K| Happy  \ 
\ gro.gilbert @ treblig.org | MIPS,x86,ARM, SPARC and HP-PA | In Hex /
 \ _________________________|_____ http://www.treblig.org   |_______/
#1 lore raw
Quoted 10 lines · Dr. David Alan Gilbert

Hi,
make menuconfig will hang just after the parsing in the
activate_menu loop in the case where the file scripts/lxdialog/lxdialog won't execute. Some error codes in this case are caught; but the case where the file scripts/lxdialog/lxdialog is a binary for the wrong architecture (case 126) is not caught. This is quite easy to trip if you are swapping between native and cross building - you get a couple of errors when you try and build make menuconfig for the first time about wrong binaries; in my case I just deleted those binaries and did the make again; however this failure is silent - it just hangs.

This does not make sense...
lxdialog are compiled utilising HOSTCC, and HOSTCC always points to gcc. So unless you fail to keep gcc for native in PATH and use:
$> make CROSS_COMPILE=arm all
to do cross-compile the above scenario should not be possible. In other words
$> which gcc
shall always point to the gcc used for native architecture. Cross compiling are done by specifying another gcc using the above syntax. [I've only looked in 2.5 sources by the way, 2.4 may differ here].

Another point is that the current kbuild is too weak when architecture is changed. Changing architecture should require a make mrproper.

Sam

#2 lore raw
Quoted 1 line · * Sam Ravnborg ([email protected]) wrote:

On Sat, Aug 24, 2002 at 04:13:29PM +0100, I wrote:

<in short, menuconfig hangs if lxdialog is built for wrong architecture>

Quoted 5 lines

This does not make sense...
lxdialog are compiled utilising HOSTCC, and HOSTCC always points to gcc. So unless you fail to keep gcc for native in PATH and use:
$> make CROSS_COMPILE=arm all
to do cross-compile the above scenario should not be possible.

No, much simpler scenario; your kernel source is on an NFS partition. You cross compile it; many days later you compile the same code natively on the target from the same directory (for example if you suspect instability is caused by cross compilation).

Dave
---------------- Have a happy GNU millennium! ----------------------

/ Dr. David Alan Gilbert    | Running GNU/Linux on Alpha,68K| Happy  \ 
\ gro.gilbert @ treblig.org | MIPS,x86,ARM, SPARC and HP-PA | In Hex /
 \ _________________________|_____ http://www.treblig.org   |_______/
#3 lore raw
In reply to Dr. David Alan Gilbert
Quoted 11 lines · Dr. David Alan Gilbert

Hi,
make menuconfig will hang just after the parsing in the
activate_menu loop in the case where the file scripts/lxdialog/lxdialog won't execute. Some error codes in this case are caught; but the case where the file scripts/lxdialog/lxdialog is a binary for the wrong architecture (case 126) is not caught. This is quite easy to trip if you are swapping between native and cross building - you get a couple of errors when you try and build make menuconfig for the first time about wrong binaries; in my case I just deleted those binaries and did the make again; however this failure is silent - it just hangs.

Quoted 3 lines

A make mrproper is probably the best thing to do when switching - but
the error case needs catching, and I'm sure there are other similar cases.

look at dmesg and add an
alias binfmt-xxxx off
to /etc/modules.conf so similar problems get caught properly - unless you want to actually use an emulator for this architecture of course :)

This is one of the cases where I wish kmod would do something more intelligent by default than endless loop. Would it be a good idea to attempt loading of emulator modules only for formats that are previously somehow registered + a few well known like aout,misc,elf?

Looking at exec.c, why isn't the result of request_module() tested?

Richard

#4 lore raw
Quoted 4 lines · Richard Zidlicky

look at dmesg and add an
alias binfmt-xxxx off
to /etc/modules.conf so similar problems get caught properly - unless you want to actually use an emulator for this architecture of course :)

I believe Dave was saying that the scripts try to run lxdialog, which returns with an error code (to the program that called it.) This error code is not checked by the caller, who just tries again. So the execve() system call is already correctly failing.

The problem lies in activate_menu in scripts/Menuconfig; it contains a loop that calls a small script which then calls lxdialog. It only tests for a limited range of return codes:

	0 1 2 3 4 5 6 139 255

The important thing here is the missing two codes (from bash's man page):

If a command is not found, the child process created to execute it returns a status of 127. If a command is found but is not executable, the return status is 126.

We currently handle neither; we just loop and try again.

Quoted 1 line

Looking at exec.c, why isn't the result of request_module() tested?

That doesn't really tell you anything; that tells you something successful happened. It doesn't tell you that you now have (say) the a.out binary format loaded; that's why we rescan the formats list afterwards.

Anyway, here's a patch that should solve the problem. Dave - thanks for finding this odd behaviour.

scripts/Menuconfig +20 -0
@@ -905,6 +905,26 @@
 			cleanup
 			exit 139
 			;;
+		126|127)
+			stty sane
+			clear
+			cat << EOM
+
+There seems to be a problem with the lxdialog companion utility which is
+built prior to running Menuconfig.  lxdialog could not be found, or could
+not be executed.  This can be caused when lxdialog has been built for a
+different architecture.
+
+You should rebuild lxdialog.  This can be done by moving to the
+/usr/src/linux/scripts/lxdialog directory and issuing the "make clean all"
+command.
+
+If the problem persists, you may email the maintainer <[email protected]> or
+post a message to <[email protected]> for additional assistance. 
+
+EOM
+			cleanup
+			exit 1
 		esac
 	done
 }
Signature · 2 lines
Russell King ([email protected])                The developer of ARM Linux
             http://www.arm.linux.org.uk/personal/aboutme.html