hyperthreading disabled, one NVMe device

Run fio on a single NVMe device. We see that with AES-NI it saturates the device
without consuming too much CPU time.

QAT with 512-byte sector size is totally unuseable (200 - 300 MiB/s)
QAT with 4kB sector size has read throughput is less than AES-NI and QAT
consumes more CPU time.
QAT with 4kB sector size saturates the device when writing to it - but there was
a deadlock that can be fixed by the patch.

fio --ioengine=psync --iodepth=1 --rw=randread --direct=1 --end_fsync=1 --bs=64k --numjobs=56 --time_based --runtime=10 --group_reporting --name=job --filename=/dev/mapper/cr
fio --ioengine=psync --iodepth=1 --rw=randwrite --direct=1 --end_fsync=1 --bs=64k --numjobs=56 --time_based --runtime=10 --group_reporting --name=job --filename=/dev/mapper/cr

raw device:
READ: bw=2893MiB/s
busy: 0.591141%
idle: 99.214848%
irq:  0.112351%

WRITE: bw=1094MiB/s
busy: 0.436918%
idle: 99.473930%
irq:  0.044901%

dm-crypt, aes-ni, 4096:
READ: bw=2985MiB/s
busy: 2.033783%
idle: 97.618126%
irq:  0.158701%

WRITE: bw=1092MiB/s
busy: 2.877080%
idle: 96.933613%
irq:  0.098376%

dm-crypt, qat, 4096:
READ: bw=1779MiB/s
busy: 3.345561%
idle: 93.220190%
irq:  1.828929%

WRITE:	-- !!!! QAT OCCASIONALY DEADLOCKS !!!!; fix with qat-fix.patch
WRITE: bw=1093MiB/s
busy: 3.831038%
idle: 94.548982%
irq:  0.919794%

dm-crypt, aes-ni, 512:
READ: bw=2863MiB/s
busy: 2.653832%
idle: 96.996943%
irq:  0.150217

WRITE: bw=1097MiB/s
busy: 3.420346%
idle: 96.392972%
irq:  0.103595%

dm-crypt, qat, 512:
READ: bw=335MiB/s
busy: 3.235961%
idle: 93.666922%
irq:  1.726651%

WRITE: bw=243MiB/s
busy: 3.033301%
idle: 93.827130%
irq:  1.769282%