block,xd: Delay allocation of DMA buffers until device is known

block,xd: Delay allocation of DMA buffers until device is known

Loading the XD module triggers a warning like

 WARNING: at mm/page_alloc.c:1805
 __alloc_pages_nodemask+0x127/0x48f()
 Hardware name: System Product Name
 Modules linked in:
 Pid: 1, comm: swapper Not tainted 2.6.32-rc8-git5 #1
 Call Trace:
  [<c103d94b>] warn_slowpath_common+0x65/0x95
  [<c103d98d>] warn_slowpath_null+0x12/0x15
  [<c109550c>] __alloc_pages_nodemask+0x127/0x48f
  [<c10be964>] ? get_slab+0x8/0x50
  [<c10b8979>] alloc_page_interleave+0x2e/0x6e
  [<c10b8a10>] alloc_pages_current+0x57/0x99
  [<c2083a4a>] ? xd_init+0x0/0x482
  [<c1094c38>] __get_free_pages+0xd/0x1e
  [<c2083a94>] xd_init+0x4a/0x482
  [<c2082df0>] ? loop_init+0x104/0x16a
  [<c169162d>] ? loop_probe+0x0/0xaf
  [<c2083a4a>] ? xd_init+0x0/0x482
  [<c1001143>] do_one_initcall+0x51/0x13f
  [<c204a307>] kernel_init+0x10b/0x15f
  [<c204a1fc>] ? kernel_init+0x0/0x15f
  [<c1004347>] kernel_thread_helper+0x7/0x10
 ---[ end trace 686db6333ade6e7a ]---
 xd: Out of memory.

The warning is because the alloc_pages is called with an
order >= MAX_ORDER. The simplistic reason is that get_order(0) returns garbage
values when given 0 as a size. The more complex reason is that the XD driver
initialisation is broken.

It's not clear why this ever worked. XD allocates a buffer for DMA based
on the value of xd_maxsectors. This value is determined by the exact
type of controller in use but the value is determined *after* an attempt
has been made to allocate the buffer. i.e. the requested size of the DMA
buffer will always be 0.

This patch alters how XD is initialised slightly by allocating the
buffer when and if a device has actually been detected. The error paths
are updated to suit the new logic.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>