Re: PCI device initialisation and RAID assembly

[Mark J Hewitt <mjh@elsabio.demon.co.uk>]

I know, bad form to reply to one's own postings, but I now have this 
working and I thought I'd just post the answer that works for me.
All I need is to preload the appropriate driver in initrd.
I wrote a tiny script so I can use this when I do kernel upgrades:  it's 
yours if it is of any help!


#!/bin/bash
# 
###########################################################################
# Preload the SATA controller driver so that the RAID array starts after any
# drives connected there are available.
#
# Mark J Hewitt.  July 2009.
#
# 
###########################################################################
#

module=sata_sil

version=$(uname -r)

if ! grep -qE 
"^[[:space:]]*alias[[:space:]]+scsi_hostadapter[[:space:]]+\<$module\>" 
/etc/modprobe.conf
then
  echo "alias scsi_hostadapter $module" >> /etc/modprobe.conf
fi

mkinitrd --preload=$module -v -f "/boot/initrd-$version.img" "$version"


You may prefer to supply the version on the command line so the next 
reboot works immediately.
The driver you need may well also differ.

Mark.

Mark J Hewitt wrote:
> I think I have an issue here with initialization timing of a PCI SATA 
> card and a RAID 5 array assembly.
>
> The history ...
>
> I'm just putting together a small storage appliance using old 
> components.  The Motherboard came from an old HP Compaq and has two 
> SATA and two PATA connectors.  I build the first RAID 5 array from one 
> PATA and two SATA drives, and this was working fine but very slow.  I 
> decided to play with a PCI SATA card (StarTech 4 port) and migrate the 
> PATA element onto another SATA drive.  Doing this is the most obvious 
> way (add the new drive, fault the old one and remove it, wait for 
> resync) seems to work fine.  However, when I reboot, the array is 
> built from the original 2 SATA drives only in degraded mode.  I can 
> re-add the SATA drive attached to the PCI card, but of course, it 
> takes the best part of a day to resync.
>
> So my guess is that the kernel (2.6.29.6 running in Fedora 11 compiled 
> with PAE support) is not bringing up the PCI card before assembling 
> the RAID array.
>
> I say this because the log looks like this:
> ...
> ...
> Jul 29 19:56:53 janus kernel: raid6: int32x2    761 MB/s
> Jul 29 19:56:53 janus kernel: raid6: int32x4    699 MB/s
> Jul 29 19:56:53 janus kernel: raid6: int32x8    457 MB/s
> Jul 29 19:56:53 janus kernel: raid6: mmxx1     2140 MB/s
> Jul 29 19:56:53 janus kernel: raid6: mmxx2     2417 MB/s
> Jul 29 19:56:53 janus kernel: raid6: sse1x1    1265 MB/s
> Jul 29 19:56:53 janus kernel: raid6: sse1x2    2121 MB/s
> Jul 29 19:56:53 janus kernel: raid6: sse2x1    2539 MB/s
> Jul 29 19:56:53 janus kernel: raid6: sse2x2    2652 MB/s
> Jul 29 19:56:53 janus kernel: raid6: using algorithm sse2x2 (2652 MB/s)
> Jul 29 19:56:53 janus kernel: md: raid6 personality registered for 
> level 6
> Jul 29 19:56:53 janus kernel: md: raid5 personality registered for 
> level 5
> Jul 29 19:56:53 janus kernel: md: raid4 personality registered for 
> level 4
> Jul 29 19:56:53 janus kernel: md: md0 stopped.
> Jul 29 19:56:53 janus kernel: md: bind<sda3>                           
> *<-- This is the removed PATA drive*
> Jul 29 19:56:53 janus kernel: md: bind<sdc3>                           
> *<-- Original SATA drive*
> Jul 29 19:56:53 janus kernel: md: bind<sdb3>                           
> *<-- Original SATA drive*
> Jul 29 19:56:53 janus kernel: md: kicking non-fresh sda3 from array!
> Jul 29 19:56:53 janus kernel: md: unbind<sda3>
> Jul 29 19:56:53 janus kernel: md: export_rdev(sda3)
> Jul 29 19:56:53 janus kernel: raid5: device sdb3 operational as raid 
> disk 1
> Jul 29 19:56:53 janus kernel: raid5: device sdc3 operational as raid 
> disk 2
> Jul 29 19:56:53 janus kernel: raid5: allocated 3176kB for md0
> Jul 29 19:56:53 janus kernel: raid5: raid level 5 set md0 active with 
> 2 out of 3 devices, algorithm 2
> Jul 29 19:56:53 janus kernel: RAID5 conf printout:
> Jul 29 19:56:53 janus kernel: --- rd:3 wd:2
> Jul 29 19:56:53 janus kernel: disk 1, o:1, dev:sdb3
> Jul 29 19:56:53 janus kernel: disk 2, o:1, dev:sdc3
> Jul 29 19:56:53 janus kernel: md0: unknown partition table
>
> ...
> ...
> ... later!
> ...
> ...
> Jul 29 19:56:53 janus kernel: sata_sil 0000:05:09.0: PCI INT A -> GSI 
> 18 (level, low) -> IRQ 18
> Jul 29 19:56:53 janus kernel: scsi4 : sata_sil
> Jul 29 19:56:53 janus kernel: scsi5 : sata_sil
> Jul 29 19:56:53 janus kernel: scsi6 : sata_sil
> Jul 29 19:56:53 janus kernel: scsi7 : sata_sil
> Jul 29 19:56:53 janus kernel: ata5: SATA max UDMA/100 mmio 
> m1024@0xfc510000 tf 0xfc510080 irq 18
> Jul 29 19:56:53 janus kernel: ata6: SATA max UDMA/100 mmio 
> m1024@0xfc510000 tf 0xfc5100c0 irq 18
> Jul 29 19:56:53 janus kernel: ata7: SATA max UDMA/100 mmio 
> m1024@0xfc510000 tf 0xfc510280 irq 18
> Jul 29 19:56:53 janus kernel: ata8: SATA max UDMA/100 mmio 
> m1024@0xfc510000 tf 0xfc5102c0 irq 18
>
> Jul 29 19:56:53 janus kernel: ata5: SATA link up 1.5 Gbps (SStatus 113 
> SControl 310)
> Jul 29 19:56:53 janus kernel: ata5.00: ATA-8: Hitachi HDT721050SLA360, 
> ST3OA3AA, max UDMA/133  *<-- The third SATA drive*
> Jul 29 19:56:53 janus kernel: ata5.00: 976773168 sectors, multi 16: 
> LBA48 NCQ (depth 0/32)
>
> ...
> ... Finally, (and manually :-( )
> ...
>
> Jul 29 20:00:12 janus kernel: md: bind<sdd3>
> Jul 29 20:00:12 janus kernel: RAID5 conf printout:
> Jul 29 20:00:12 janus kernel: --- rd:3 wd:2
> Jul 29 20:00:12 janus kernel: disk 0, o:1, dev:sdd3
> Jul 29 20:00:12 janus kernel: disk 1, o:1, dev:sdb3
> Jul 29 20:00:12 janus kernel: disk 2, o:1, dev:sdc3
> Jul 29 20:00:12 janus kernel: md: recovery of RAID array md0
> Jul 29 20:00:12 janus kernel: md: minimum _guaranteed_  speed: 1000 
> KB/sec/disk.
> Jul 29 20:00:12 janus kernel: md: using maximum available idle IO 
> bandwidth (but not more than 200000 KB/sec) for recovery.
> Jul 29 20:00:12 janus kernel: md: using 128k window, over a total of 
> 487098624 blocks.
>
> Is there anything I can do about this?
> Mark.
>
>