I compiled the kernel without deadline, and the dasd code exits the old
scheduler (CFQ), fails to load the new one (deadline), and then things just
hang - with one of these (sorry about the weird chars - I copy & pasted it
from a 3270 console):
dasd(eckd): 0.0.0151: 3390/0A(CU:3990/01) Cyl:3338 Head:15 Sec:224
------------ cut here ------------
Badness at kernel/mutex.c:134
Modules linked in: dasd_eckd_mod dasd_mod
CPU: 0 Not tainted 2.6.25-rc3 #9
Process exe (pid: 538, task: 
000000000d172000, ksp: 
000000000d21ef88)
Krnl PSW : 
0404000180000000 000000000022fb5c (mutex_lock_nested+0x2a4/0x2cc)
           R:0 T:1 IO:0 EX:0 Key:0 M:1 W:0 P:0 AS:0 CC:0 PM:0 EA:3
Krnl GPRS: 
0000000000024218 000000000076fc78 0000000000000000 000000000000000f
           000000000022f92e 0000000000449898 000000000f921c00 000003e000162590
           00000000001539c4 000000000d172000 070000007fffffff 000000000d21f400
           000000000f8f2560 00000000002413f8 000000000022fb44 000000000d21f400
Krnl Code: 
000000000022fb50: 
bf2f1000           icm     %r2,15,0(%r1)
           
000000000022fb54: 
a774fef6           brc     7,22f940
           
000000000022fb58: 
a7f40001           brc     15,22fb5a
          >
000000000022fb5c: 
a7f4fef2           brc     15,22f940
           
000000000022fb60: 
c0e5fffa112a       brasl   %r14,171db4
           
000000000022fb66: 1222               ltr     %r2,%r2
           
000000000022fb68: 
a784fedb           brc     8,22f91e
           
000000000022fb6c: 
c010002a0086       larl    %r1,76fc78
Call Trace:
(<
000000000022f92e> mutex_lock_nested+0x76/0x2cc)
 <
00000000001539c4> elevator_exit+0x38/0x80
 <
0000000000156ffe> blk_cleanup_queue+0x62/0x7c
 <
000003e0001d5414> dasd_change_state+0xe0/0x8ec
 <
000003e0001d5cae> dasd_set_target_state+0x8e/0x9c
 <
000003e0001d5f74> dasd_generic_set_online+0x160/0x284
 <
000003e00011e83a> dasd_eckd_set_online+0x2e/0x40
 <
0000000000199bf4> ccw_device_set_online+0x170/0x2c0
 <
0000000000199d9e> online_store_recog_and_online+0x5a/0x14c
 <
000000000019a08a> online_store+0xbe/0x2ec
 <
000000000018456c> dev_attr_store+0x38/0x58
 <
000000000010efbc> sysfs_write_file+0x130/0x190
 <
00000000000af582> vfs_write+0xb2/0x160
 <
00000000000afc7c> sys_write+0x54/0x9c
 <
0000000000025e16> sys32_write+0x2e/0x50
 <
0000000000024218> sysc_noemu+0x10/0x16
 <
0000000077e82bd2> 0x77e82bd2
Set elevator pointer to NULL in order to avoid double elevator_exit
calls when elevator_init call for deadline iosched fails.
Also make sure the dasd device driver depends on IOSCHED_DEADLINE so
the default IO scheduler of the dasd driver is present.
Signed-off-by: Josef 'Jeff' Sipek <jeffpc@josefsipek.net>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
 config DASD
        tristate "Support for DASD devices"
        depends on CCW && BLOCK
+       select IOSCHED_DEADLINE
        help
          Enable this option if you want to access DASDs directly utilizing
          S/390s channel subsystem commands. This is necessary for running
 
        block->request_queue->queuedata = block;
 
        elevator_exit(block->request_queue->elevator);
+       block->request_queue->elevator = NULL;
        rc = elevator_init(block->request_queue, "deadline");
        if (rc) {
                blk_cleanup_queue(block->request_queue);