On Tue, 9 Jan 2007 15:40:03 +0200 (EET)
Pekka J Enberg <penberg@cs.helsinki.fi> wrote:

quoted text
> The current implementation of cache_grow() has to either (1) use pre-allocated
> memory for the slab or (2) allocate the memory itself which makes the error
> paths messy. Move __GFP_NO_GROW and __GFP_WAIT processing to kmem_getpages()
> and introduce a new __cache_grow() variant that expects the memory for a new
> slab to always be handed over in the 'objp' parameter.

This gets its local interrupt state mucked up.

BUG: sleeping function called from invalid context at mm/slab.c:3038
in_atomic():0, irqs_disabled():1
no locks held by init/1.
irq event stamp: 656902
hardirqs last  enabled at (656901): [<c015f48b>] mod_zone_page_state+0x4b/0x50
hardirqs last disabled at (656902): [<c0171f54>] cache_alloc_refill+0x384/0x6a0
softirqs last  enabled at (650628): [<c03a6317>] unix_release_sock+0x67/0x220
softirqs last disabled at (650626): [<c03cfe8e>] _write_lock_bh+0xe/0x40
 [<c0103f7a>] show_trace_log_lvl+0x1a/0x30
 [<c0104682>] show_trace+0x12/0x20
 [<c0104736>] dump_stack+0x16/0x20
 [<c01175ea>] __might_sleep+0xba/0xd0
 [<c017262f>] kmem_cache_alloc+0xaf/0xd0
 [<c0172131>] cache_alloc_refill+0x561/0x6a0
 [<c0172574>] kmem_cache_zalloc+0xe4/0xf0
 [<c011cb49>] copy_process+0x89/0x1280
 [<c011dfcb>] do_fork+0x6b/0x1c0
 [<c0100e13>] sys_clone+0x33/0x40
 [<c0102edc>] sysenter_past_esp+0x5d/0x99
 =======================
-
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/