0000125: Compile Busybox without MMU failed. - BusyBox Bug and Patch Tracking

(0000741)
landley
12-07-05 22:13

What version of uClibc are you building against, and could you post your .config to the mailing list?

Rob

(0000758)
frankteoth
12-10-05 10:02

Uses:
buildroot-20051209.tar.bz2
busybox-1.01.tar.bz2
uClibc-0.9.28.tar.bz2

binutils-2.16.1.tar.bz2
gcc-3.4.2.tar.bz2
linux-libc-headers-2.4.31.tar.bz2

.config for buildroot, busybox & uClibc are attached.
.config.BUGS & .config.GOOD for busybox are also attached.
Please refer to attachment: BUG-0000125.tar.bz2

(0000909)
landley
01-09-06 23:28

Your libc isn't providing daemonize(), and those things (quite understandably) want it. This is a toolchain issue, not a busybox issue.

man 3 daemonize

(0000912)
vapier
01-10-06 06:01

not really

would also say it's a toolchain/kernel problem that fork() isnt available on uClinux ?

(0001201)
jamie
03-28-06 13:23

It's fixed with the attached patch "patch_busybox_daemon",
for i386 and ARM uClinux.

Other architectures are easy to add, and I'll do so if someone asks
(but you can probably do it yourself).

Because uClinux kernel doesn't implement fork(), uClibc doesn't
implement daemon() on those targets. However daemon() can be
implemented using clone() and an assembly stub.

Only use this patch with supported MMUless uClinux targets.
It has the wrong behaviour otherwise.

It would be great if someone includes this patch in the Busybox
source tree and adds an Autoconf test to decide when to use it,
which is when there isn't a working fork().

Enjoy, --Jamie

(0001202)
vapier
03-28-06 16:25

that patch isnt acceptable, sorry

(0001412)
bernhardf
06-11-06 10:34

We will take care of this in bb_xdaemon for the BB_NOMMU case.

E.g.:
#if !BB_NOMMU /* normal case, we have fork */
#define bb_xdaemon(chd,clo,str) common_xdaemon(chd,clo)
#else
#define bb_xdaemon(chd,clo,str) vfork_daemon_rexec(chd,clo,str)
#endif

If by then we still have not exported argv and argc, we'll have to also pass these in -- for the case we end up calling vfork_daemon_rexec().

(0004064)
vda
02-11-08 10:19

Appears to be fixed in svn. start_stop_daemon works for me on NOMMU.

Issue History
Date Modified	Username	Field	Change
02-25-05 22:31	frankteoth	New Issue
03-16-05 12:27	andersen	Assigned To	andersen => BusyBox
09-17-05 03:56	Dmitry Bazhenov	Issue Monitored: Dmitry Bazhenov
12-07-05 22:13	landley	Note Added: 0000741
12-07-05 22:13	landley	Status	assigned => feedback
12-10-05 08:40	frankteoth	Issue Monitored: frankteoth
12-10-05 09:53	frankteoth	Issue End Monitor: frankteoth
12-10-05 09:58	frankteoth	File Added: BUG-0000125.tar.bz2
12-10-05 10:02	frankteoth	Note Added: 0000758
01-09-06 23:28	landley	Status	feedback => closed
01-09-06 23:28	landley	Note Added: 0000909
01-09-06 23:28	landley	Resolution	open => no change required
01-10-06 06:01	vapier	Status	closed => feedback
01-10-06 06:01	vapier	Resolution	no change required => reopened
01-10-06 06:01	vapier	Note Added: 0000912
03-28-06 12:26	jamie	Issue Monitored: jamie
03-28-06 13:11	jamie	File Added: patch_busybox_daemon
03-28-06 13:23	jamie	Note Added: 0001201
03-28-06 16:25	vapier	Note Added: 0001202
06-11-06 10:34	bernhardf	Note Added: 0001412
02-11-08 10:19	vda	Status	feedback => closed
02-11-08 10:19	vda	Note Added: 0004064
02-11-08 10:19	vda	Resolution	reopened => fixed

Notes
(0000741) landley 12-07-05 22:13	What version of uClibc are you building against, and could you post your .config to the mailing list? Rob

(0000758) frankteoth 12-10-05 10:02	Uses: buildroot-20051209.tar.bz2 busybox-1.01.tar.bz2 uClibc-0.9.28.tar.bz2 binutils-2.16.1.tar.bz2 gcc-3.4.2.tar.bz2 linux-libc-headers-2.4.31.tar.bz2 .config for buildroot, busybox & uClibc are attached. .config.BUGS & .config.GOOD for busybox are also attached. Please refer to attachment: BUG-0000125.tar.bz2

(0000909) landley 01-09-06 23:28	Your libc isn't providing daemonize(), and those things (quite understandably) want it. This is a toolchain issue, not a busybox issue. man 3 daemonize

(0000912) vapier 01-10-06 06:01	not really would also say it's a toolchain/kernel problem that fork() isnt available on uClinux ?

(0001201) jamie 03-28-06 13:23	It's fixed with the attached patch "patch_busybox_daemon", for i386 and ARM uClinux. Other architectures are easy to add, and I'll do so if someone asks (but you can probably do it yourself). Because uClinux kernel doesn't implement fork(), uClibc doesn't implement daemon() on those targets. However daemon() can be implemented using clone() and an assembly stub. Only use this patch with supported MMUless uClinux targets. It has the wrong behaviour otherwise. It would be great if someone includes this patch in the Busybox source tree and adds an Autoconf test to decide when to use it, which is when there isn't a working fork(). Enjoy, --Jamie

(0001202) vapier 03-28-06 16:25	that patch isnt acceptable, sorry

(0001412) bernhardf 06-11-06 10:34	We will take care of this in bb_xdaemon for the BB_NOMMU case. E.g.: #if !BB_NOMMU /* normal case, we have fork */ #define bb_xdaemon(chd,clo,str) common_xdaemon(chd,clo) #else #define bb_xdaemon(chd,clo,str) vfork_daemon_rexec(chd,clo,str) #endif If by then we still have not exported argv and argc, we'll have to also pass these in -- for the case we end up calling vfork_daemon_rexec().

(0004064) vda 02-11-08 10:19	Appears to be fixed in svn. start_stop_daemon works for me on NOMMU.

Relationships