XOR on Helios 4

Mangix · December 6, 2020

Is it me or it the mv_xor driver being loaded late?

```

[    0.007976] xor: measuring software checksum speed
[    0.164062] xor: using function: arm4regs (2534.000 MB/sec)
[    0.316068] raid6: neonx8   xor() 1096 MB/s
[    0.452063] raid6: neonx4   xor() 1378 MB/s
[    0.588069] raid6: neonx2   xor() 1610 MB/s
[    0.724066] raid6: neonx1   xor() 1346 MB/s
[    0.860075] raid6: int32x8 xor()   328 MB/s
[    0.996072] raid6: int32x4 xor()   369 MB/s
[    1.132082] raid6: int32x2 xor()   332 MB/s
[    1.268063] raid6: int32x1 xor()   285 MB/s
[    1.268067] raid6: .... xor() 1610 MB/s, rmw enabled
[    1.665163] mv_xor f1060800.xor: Marvell shared XOR driver
[    1.694139] mv_xor f1060800.xor: Marvell XOR (Descriptor Mode): ( xor cpy intr )
[    1.694257] mv_xor f1060900.xor: Marvell shared XOR driver
[    1.722115] mv_xor f1060900.xor: Marvell XOR (Descriptor Mode): ( xor cpy intr )

```

I see that /proc/interrupts shows no action despite me using RAID5.

edit: from looking at kernel source, it seems btrfs uses xor_blocks instead of async_tx. The former seems software only.

aprayoga · December 8, 2020

yes, mv_xor loaded late. i think after async_tx loaded.

mdadm RAID5 use the xor engine

gprovost · December 14, 2020

@Mangix You can check that you system effectively use mv_xor engine by looking at the interrupt counters.

cat /proc/interrupts 

[...]
 47:      27799          0     GIC-0  54 Level     f1060800.xor
 48:      29949          0     GIC-0  97 Level     f1060900.xor

You will see the counter increase when doing operation on your RAID array (FYI XOR only used for RAID5 and 6).