Hi,
To quickly bring you up to speed, the driver for Marvell's hardware
crypto accelerator embedded in their Armada/Kirkwood SoCs has been
rewritten as marvell_cesa and merged in kernel 4.2. The new driver
received a number of patches since and has been enabled in Debian as a
module in 4.4~rc4-1~exp1 (debian bug #807634), coexisting with the old
driver (mv_cesa).
Unfortunately, it still doesn't work on my hardware (kirkwood-ts219-6282.dts):
uname -a
Linux yukikaze 4.4.0-trunk-kirkwood #1 Debian 4.4-1~exp1 (2016-01-19)
armv5tel GNU/Linux
modprobe marvell_cesa allhwsupport=1
dmesg | tail -n 1
[ 1057.855091] marvell-cesa: probe of f1030000.crypto failed with error -12
I used to build this module on my own for kernel 4.3, where it would
fail with error -22, so that changed.
f1030000.crypto refers to, as far as I can tell, a block of memory
reserved for dma operations (?) of the hardware crypto. Notably, the
old driver (mv_cesa) did not utilize it.
dtc -I fs /proc/device-tree
[snip]
ocp@f1000000 {
crypto@30000 {
reg = <0x30000 0x10000>;
interrupts = <0x16>;
marvell,crypto-srams = <0xd>;
reg-names = "regs";
compatible = "marvell,kirkwood-crypto";
clocks = <0x3 0x11>;
marvell,crypto-sram-size = <0x800>;
status = "okay";
};
find / -name f1030000.crypto
/sys/bus/platform/devices/f1030000.crypto
/sys/devices/platform/ocp@f1000000/f1030000.crypto
/sys/bus/platform/devices/f1030000.crypto/of_node# ls
clocks compatible interrupts marvell,crypto-srams
marvell,crypto-sram-size name reg reg-names status
That's as far as I got. I suspect that the new driver might fail due
to an incorrect definition of something crypto-sram-related in the dtb
for this SoC, but I cannot offer anything concrete. Alternatively,
perhaps the driver itself is busted on this hardware (googling for
"probe of f1030000.crypto failed with error 22" yields several
results). Hopefully you have some ideas.
Best regards,
Jan
