Hello, got the following w/ 2.6.9rc2 w/o 00026.patch (kdbg) : LD [M] drivers/md/dm-multipath.o drivers/md/dm-roundrobin.o(.init.text+0x0): In function `init_module': : multiple definition of `init_module' drivers/md/dm-mpath.o(.init.text+0x0): first defined here ld: Warning: size of symbol `init_module' changed from 143 in drivers/md/dm-mpath.o to 53 in drivers/md/dm-roundrobin.o drivers/md/dm-roundrobin.o(.exit.text+0x0): In function `cleanup_module': : multiple definition of `cleanup_module' drivers/md/dm-mpath.o(.exit.text+0x0): first defined here ld: Warning: size of symbol `cleanup_module' changed from 64 in drivers/md/dm-mpath.o to 45 in drivers/md/dm-roundrobin.o make[2]: *** [drivers/md/dm-multipath.o] Error 1 make[1]: *** [drivers/md] Error 2 make: *** [drivers] Error 2 regards, Le vendredi 01 octobre 2004 �3:21 +0100, Alasdair G Kergon a �it : > ftp://sources.redhat.com/pub/dm/patches/2.6-unstable/2.6.9-rc2/2.6.9-rc2-udm2.tar.bz2 > > A new patchset for multipath. > This contains the final set of features I want us to work > on making stable and tidying up before the first submission > upstream. > > > Main changes: > Removed the experimental path initialisation code for now: > will reinstate in a more generic way after the current set of > features is ready for mainstream inclusion. > > Device refs by major/minor numbers so userspace tools can be more > efficient e.g. dmsetup info -j 253 -m 1 > Take device-mapper tools from CVS to take advantage of this. > [Userspace changes not thoroughly tested yet; a few > commands require kernel changes for complete support.] > > Reinstatement of failed paths directly - no need to reload the > whole table. e.g. 'dmsetup message mp1 2000 reinstate 253:4' > > Path selector 'select' function returns how many times to use > path before being asked again; also enables crude > load-balancing. > e.g. 'multipath 1 round-robin 2 1 253:4 1000 253:3 2000' > (send 1000 I/O to first; 2000 to second) > 0 means continue using a path till it fails. > It's still optional: 'round-robin 2 0 253:4 253:3' > > Auto-loading of unknown path selectors. > > > I've done some basic testing on this, but not much yet. > In particular, I've not checked any module code, not reviewed > all the uses of locks for correctness and not reviewed > what 'status' returns. > > The i386 kgdb patch (taken from -mm) is included for convenience. > > > Alasdair -- christophe varoqui <christophe.varoqui@xxxxxxx>