MLK-24912-2 crypto: caam - fix RNG vs. hwrng kthread race

commit a067df836f474adbedad8cc910987aef4d4b1497 from
https://source.codeaurora.org/external/imx/linux-imx

The following stack trace is met when stress-testing suspend/resume:

[...]
PM: suspend devices took 1.972 seconds
[...]
SError Interrupt on CPU1, code 0xbf000002 -- SError
CPU: 1 PID: 213 Comm: hwrng Not tainted 5.4.70-2.3.0+g72209dedd129 #1
Hardware name: Freescale i.MX8DXL EVK (DT)
pstate: 60000005 (nZCv daif -PAN -UAO)
pc : _raw_spin_unlock_bh+0x0/0x28
lr : caam_jr_enqueue+0x24c/0x378
sp : ffff8000127dbd10
x29: ffff8000127dbd10 x28: ffff00003cac5940
x27: 00000000bcb5ef80 x26: 0000000000000010
x25: ffff800011c12000 x24: ffff8000127dbdb8
x23: ffff800010ca2298 x22: ffff00003c8aec10
x21: ffff00003cb5ef80 x20: 00000000ffffff8d
x19: 0000000000000010 x18: 000000000000000e
x17: 0000000000000001 x16: 0000000000000019
x15: 0000000000000033 x14: 000000000000004c
x13: 0000000000000068 x12: ffff800011188e90
x11: ffff00003c897210 x10: 0000000000000026
x9 : 00000000a4dcb313 x8 : 0000000000000000
x7 : 0000000000000001 x6 : ffff800011b59000
x5 : 0000000000000000 x4 : 0000000000000001
x3 : 0000000000000004 x2 : 0000000000000014
x1 : 00000000000001ec x0 : ffff00003cac5940
Kernel panic - not syncing: Asynchronous SError Interrupt
CPU: 1 PID: 213 Comm: hwrng Not tainted 5.4.70-2.3.0+g72209dedd129 #1
Hardware name: Freescale i.MX8DXL EVK (DT)
Call trace:
 dump_backtrace+0x0/0x140
 show_stack+0x14/0x20
 dump_stack+0xb4/0x114
 panic+0x158/0x324
 nmi_panic+0x84/0x88
 arm64_serror_panic+0x74/0x80
 do_serror+0x80/0x138
 el1_error+0x84/0xf8
 _raw_spin_unlock_bh+0x0/0x28
 caam_rng_read_one.isra.0+0x1c8/0x3a0
 caam_read+0x80/0xa8
 hwrng_fillfn+0x8c/0x140
 kthread+0x138/0x158
 ret_from_fork+0x10/0x1c
SMP: stopping secondary CPUs
Kernel Offset: disabled
CPU features: 0x0002,20002008
Memory Limit: none

This happens when:
-the generic "hwrng" kthread tries to draw entropy and
-the current rng is caam's rng and
-the job ring used for caam rng hasn't been resumed yet
(after a suspend)

The issue has been noticed also in upstream (for TPM device in ChromeOS)
and the fix proposed involved making the "hwrng" kthread freezable:
03a3bb7a ("hwrng: core - Freeze khwrng thread during suspend")
ff296293 ("random: Support freezable kthreads in add_hwgenerator_randomness()")
59b56948 ("random: Use wait_event_freezable() in add_hwgenerator_randomness()")

However, because these commits introduced a regression in virtio-rng
(Link: https://lore.kernel.org/lkml/4a45b3e0-ed3a-61d3-bfc6-957c7ba631bb@maciej.szmigiero.name

)
they were later reverted in commit
08e97aec ("Revert "hwrng: core - Freeze khwrng thread during suspend"")

Since there was no progress in upstream and fixing virtio-rng regression
is not trivial, the solution chosen is to unregister / re-register
caam rng driver from hwrng during suspend / resume.

Signed-off-by: Horia Geantă <horia.geanta@nxp.com>
Tested-by: Iuliana Prodan <iuliana.prodan@nxp.com>
Signed-off-by: Xiaolei Wang <xiaolei.wang@windriver.com>