Comment # 5 on bug 1023755 from Jiri Bohac

Hmm, your machine has very fragmented physical memory by the time the kernel
gets to allocating the crash kernel space. I added the sizes and offsets in MB
to your dmesg output:

     reserved.cnt  = 0x12
     reserved[0x0]    [0x0000000000000000-0x000000000000ffff (    0 MB @     0
MB)], 0x0000000000010000 bytes on node 0 flags: 0x0
     reserved[0x1]    [0x0000000000096000-0x000000000009bfff (    0 MB @     0
MB)], 0x0000000000006000 bytes on node 0 flags: 0x0
     reserved[0x2]    [0x000000000009c800-0x00000000000fffff (    0 MB @     0
MB)], 0x0000000000063800 bytes on node 0 flags: 0x0
     reserved[0x3]    [0x0000000001000000-0x0000000002274fff (   18 MB @    16
MB)], 0x0000000001275000 bytes on node 0 flags: 0x0
     reserved[0x4]    [0x000000000c9ac000-0x000000000c9ebfff (    0 MB @   201
MB)], 0x0000000000040000 bytes on node 0 flags: 0x0
     reserved[0x5]    [0x000000001cad2018-0x000000001cae2057 (    0 MB @   458
MB)], 0x0000000000010040 bytes on node 0 flags: 0x0
     reserved[0x6]    [0x000000001cae3018-0x000000001cafc657 (    0 MB @   458
MB)], 0x0000000000019640 bytes on node 0 flags: 0x0
     reserved[0x7]    [0x000000001cafd000-0x000000001d5abfff (   10 MB @   458
MB)], 0x0000000000aaf000 bytes on node 0 flags: 0x0
     reserved[0x8]    [0x000000002a6d3000-0x000000002c100fff (   26 MB @   678
MB)], 0x0000000001a2e000 bytes on node 0 flags: 0x0
     reserved[0x9]    [0x000000002c14c000-0x000000002c1a2fff (    0 MB @   705
MB)], 0x0000000000057000 bytes on node 0 flags: 0x0
     reserved[0xa]    [0x000000002c994000-0x000000002c99bfff (    0 MB @   713
MB)], 0x0000000000008000 bytes on node 0 flags: 0x0
     reserved[0xb]    [0x000000002c9a4018-0x000000002c9a4827 (    0 MB @   713
MB)], 0x0000000000000810 bytes on node 0 flags: 0x0
     reserved[0xc]    [0x0000000034713000-0x0000000036f8cfff (   40 MB @   839
MB)], 0x000000000287a000 bytes on node 0 flags: 0x0
     reserved[0xd]    [0x0000000036f94000-0x00000000376a8fff (    7 MB @   879
MB)], 0x0000000000715000 bytes on node 0 flags: 0x0
     reserved[0xe]    [0x000000003a274598-0x000000003a2745cf (    0 MB @   930
MB)], 0x0000000000000038 bytes on node 0 flags: 0x0
     reserved[0xf]    [0x000000003a4ff000-0x000000003a4fffff (    0 MB @   932
MB)], 0x0000000000001000 bytes on node 0 flags: 0x0
     reserved[0x10]    [0x00000008bdfe9000-0x00000008bdffefff (    0 MB @ 35807
MB)], 0x0000000000016000 bytes flags: 0x0
     reserved[0x11]    [0x00000008bdfff000-0x00000008bdffffff (    0 MB @ 35807
MB)], 0x0000000000001000 bytes on node 0 flags: 0x0
     crashkernel reservation failed - memory is in use.

There is no single contiguous space of 256 MB under 4G.
The largest available block seems to be between 201 MB and 458 MB.
With required the alignment to 16 MB, the largest available block of 240 MB
will be between 208 and 448 MB.
So with this information, I think booting with crashkernel=xM, where x <= 240
should work.
Yet you say you tried 128 and did not work either. Could you, please, try again
with crashkernel=128M and memblock=debug?