I've been experiencing hangs with kernel 2.6.10-ac11 and also previous
ac-series. The affected box is a quite loaded Dual-Xeon HT system.
The kernel was built with gcc-2.95 (Debian woody).
Sysrq-b on ac11 brings the following and the completely hangs, i.e. no
sysrq responses anymore:
SysRq : Emergency Sync
SysRq : Emergency Remount R/O
SysRq : Resetting
Badness in smp_call_function at arch/i386/kernel/smp.c:523
[<c010c718>] smp_call_function+0x4c/0xf0
[<c0116dc7>] release_console_sem+0x1f/0xa8
[<c010c7fc>] smp_send_stop+0x10/0x1c
[<c010c7bc>] stop_this_cpu+0x0/0x30
[<c010c238>] machine_restart+0x7c/0xf8
[<c02518fb>] sysrq_handle_reboot+0x7/0xc
[<c0251a87>] __handle_sysrq+0x6b/0x104
[<c0251b3d>] handle_sysrq+0x1d/0x24
[<c0258030>] receive_chars+0x138/0x204
[<c02582f2>] serial8250_interrupt+0x66/0xe4
[<c012d750>] handle_IRQ_event+0x28/0x58
[<c012d87b>] __do_IRQ+0xfb/0x150
[<c010415b>] do_IRQ+0x1b/0x28
[<c0102bd2>] common_interrupt+0x1a/0x20
[<c03266a6>] _spin_lock+0xa/0x10
[<c010c752>] smp_call_function+0x86/0xf0
[<c01362e8>] do_drain+0x0/0x44
[<c01362e8>] do_drain+0x0/0x44
[<c01362d6>] smp_call_function_all_cpus+0x1a/0x2c
[<c01362e8>] do_drain+0x0/0x44
[<c013633d>] drain_cpu_caches+0x11/0x40
[<c01362e8>] do_drain+0x0/0x44
[<c0136379>] __cache_shrink+0xd/0x8c
[<c013641e>] kmem_cache_shrink+0x26/0x2c
[<c022636c>] xfs_inode_shake+0xc/0x24
[<c0138676>] shrink_slab+0x86/0x1a0
[<c013990e>] try_to_free_pages+0xd2/0x188
[<c0132a95>] __alloc_pages+0x1e5/0x308
[<c0135596>] do_page_cache_readahead+0x10a/0x194
[<c01357b1>] page_cache_readahead+0x191/0x1c8
[<c012f036>] do_generic_mapping_read+0xe6/0x464
[<c012f839>] generic_file_sendfile+0x51/0x64
[<c012f798>] file_send_actor+0x0/0x50
[<c0224eaa>] xfs_sendfile+0x152/0x1a4
[<c012f798>] file_send_actor+0x0/0x50
[<c012f798>] file_send_actor+0x0/0x50
[<c0221d2a>] linvfs_sendfile+0x36/0x40
[<c012f798>] file_send_actor+0x0/0x50
[<c014a43a>] do_sendfile+0x246/0x294
[<c012f798>] file_send_actor+0x0/0x50
[<c014a55c>] sys_sendfile64+0x3c/0xa0
[<c0102263>] syscall_call+0x7/0xb
Btw, would it be possible to directly boot the box in the sysrq case
instead of going through the smp functions as it looks they do not
always have the desired effect?
Adam
--
Adam [email protected]
Lackorzynski http://os.inf.tu-dresden.de/~adam/