ieee1394 sbp2 driver is broken for kernel >= 2.4.21-rc2

... at least for PPC targets.
There was a set of patches applied for the 2.4.21-rc2 kernel (some seem _very_ strange, maybe endianness issues...) that makes sbp2 fail to recognize my external firewire drives. The only way I have for now to make things work is to come back to the 2.4.21-rc1 version, which work correctly for my Mac and my PC (VIA Chipset & Firewire chip).
On the Mac (Ibook), I can see my disks in /proc/bus/ieee1394/devices:

Node[01:1023] GUID[000393fffeab0e76]:
Vendor ID: `Linux OHCI-1394' [0x000000]
Capabilities: 0x0083c0
Bus Options:

    IRMC(1) CMC(1) ISC(1) BMC(0) PMC(0) GEN(8)
    LSPD(2) MAX_REC(2048) CYC_CLK_ACC(0)

Host Node Status:
Host Driver : ohci1394
Nodes connected : 2
Nodes active : 2
SelfIDs received: 2
Irm ID : [01:1023]
BusMgr ID : [01:1023]
In Bus Reset : no
Root : yes
Cycle Master : yes
IRM : yes
Bus Manager : yes
Node[00:1023] GUID[00d04b0ce0900195]:
Vendor ID: `Oxford ' [0x00d04b]
Capabilities: 0x0083c0
Bus Options:

    IRMC(0) CMC(0) ISC(0) BMC(0) PMC(0) GEN(0)
    LSPD(0) MAX_REC(64) CYC_CLK_ACC(255)

Unit Directory 0:
Vendor/Model ID: Oxford [00d04b] / 911G [000001]
Software Specifier ID: 00609e
Software Version: 010483
Driver: SBP2 Driver
Length (in quads): 8

but not in /proc/scsi/scsi
Here's the output of /proc/scsi/sbp2/1 (with the working driver):
IEEE-1394 SBP-2 protocol driver (host: ohci1394)
$Rev: 878 $ James Goodwin <[email protected]>
SBP-2 module load options:
- Max speed supported: S400
- Max sectors per I/O supported: 255
- Max outstanding commands supported: 64
- Max outstanding commands per lun supported: 1
- Serialized I/O (debug): no
- Exclusive login: yes

I may do more tests to localize the bug, if I only knew which ones I should do...

/proc/cpuinfo:
cpu : 750FX
temperature : 2 C (uncalibrated)
clock : 700MHz
revision : 1.18 (pvr 7000 0112)
bogomips : 1389.36
machine : PowerBook4,3
motherboard : PowerBook4,3 MacRISC2 MacRISC Power Macintosh
detected as : 257 (iBook 2 rev. 2)
pmac flags : 0000000b
L2 cache : 512K unified
memory : 384MB
pmac-generation : NewWorld

/proc/pci:
PCI devices found:
Bus 0, device 11, function 0:
Host bridge: Apple Computer Inc. UniNorth/Pangea AGP (rev 0).
Master Capable. Latency=16.
Bus 16, device 11, function 0:
Host bridge: Apple Computer Inc. UniNorth/Pangea PCI (rev 0).
Master Capable. Latency=16.
Bus 16, device 23, function 0:
Class ff00: Apple Computer Inc. KeyLargo/Pangea Mac I/O (rev 0).
Master Capable. Latency=16.
Non-prefetchable 32 bit memory at 0x80000000 [0x8007ffff].

Jocelyn Mayer <[email protected]>

As a datapoint, works fine for me with my x86 laptop:

(keller) 497$ cat /proc/bus/ieee1394/devices
Node[00:1023] GUID[0001d2000003a4ec]:
Vendor ID: `Oxford Semiconductor Ltd. ' [0x0001d2]
Capabilities: 0x0083c0
Bus Options:

    IRMC(0) CMC(0) ISC(0) BMC(0) PMC(0) GEN(0)
    LSPD(0) MAX_REC(64) CYC_CLK_ACC(255)

Unit Directory 0:
Vendor/Model ID: Oxford Semiconductor Ltd. [0001d2] / OXFORD IDE Device LUN 0 [42a258]
Software Specifier ID: 00609e
Software Version: 010483
Driver: SBP2 Driver
Length (in quads): 8
Node[01:1023] GUID[474fc000075f5001]:
Vendor ID: `Linux OHCI-1394' [0x000000]
Capabilities: 0x0083c0
Bus Options:

    IRMC(1) CMC(1) ISC(1) BMC(0) PMC(0) GEN(0)
    LSPD(2) MAX_REC(2048) CYC_CLK_ACC(0)

Host Node Status:
Host Driver : ohci1394
Nodes connected : 2
Nodes active : 2
SelfIDs received: 2
Irm ID : [01:1023]
BusMgr ID : [63:1023]
In Bus Reset : no
Root : yes
Cycle Master : yes
IRM : yes
Bus Manager : no
You have new mail in /var/mail/georgn
(keller) 498$ cat /proc/scsi/scsi
Attached devices:
Host: scsi0 Channel: 00 Id: 00 Lun: 00
Vendor: IC35L120 Model: AVVA07-0 Rev:
Type: Direct-Access ANSI SCSI revision: 06
(keller) 499$ cat /proc/scsi/sbp2_0/0
Host scsi0 : SBP-2 IEEE-1394 (ohci1394)
Driver version : $Rev: 906 $ James Goodwin <[email protected]>

Module options         :
  max_speed            : S800
  max_sectors          : 255
  serialize_io         : no
  exclusive_login      : yes

Attached devices :
[Channel: 00, Id: 00, Lun: 00] Direct-Access IC35L120 AVVA07-0

(keller) 500$ uname -a
Linux keller 2.4.21-rc6-rmap15j #1 Sun Jun 1 18:43:25 EDT 2003 i686 GNU/Linux

Hi,

OK, so it should be an endianness related problem... I didn't test this on a PC because I need (want ?) to always use the same kernel on my Mac & my PC so I can test my patches always in the same conditions. It gives me a start point to investigate...

Regards.

Jocelyn Mayer <[email protected]>

No, it's a rescan-scsi-bus.sh issue. Get the script, and execute it. Hotplug for 2.4.x scsi is a fantasy. Just so happens it used to work, but that "work" used to cause oopses.

Debian     - http://www.debian.org/
Linux 1394 - http://www.linux1394.org/
Subversion - http://subversion.tigris.org/
Deqo       - http://www.deqo.com/

Hi,

Thanks for your help, but I think you're wrong:

First, I never trust hotplug or other tools like this:
I do all insmod by hand, so I know all drivers have been loaded. What is hotplug supposed to do (but wasn't in previous driver version...) ?

The second thing I see is that it used to work, before 2.4.21-rc2. The only difference is in the kernel driver, so it should work with no user-space tool, as it used to. If not, the driver is now buggy...

Regards.

Jocelyn Mayer <[email protected]>

I doubt it. Ben's one of the authorities on this stuff.

I compile this stuff directly into my kernel. Doesn't make a difference.

Well, I've _always_ needed either rescan-scsi-bus.sh (or scsiadd -s since I switched to Debian). If there's some magic that you've been doing that obviates this requirement, we're all ears.

-g

Right, I can believe this.

No magics.
I must precise that I don't use any standard distribution:
I bootstrapped my own distrib from scratch,
using sources from the web.
Also note that I don't have neither rescan-scsi-bus nor scsiadd anywhere on this machine.

I just did a new test a few minutes ago.
To be sure that there are nothing done that I can't control, I removed /sbin/hotplug (rm -f /sbin/hotplug).

I did exactly:

modprobe ieee1394
modprobe cmp
modprobe ohci1394
modprobe amdtp
modprobe sd_mod
modprobe sbp2

(I use modprobe so I don't have to write the whole path for insmod,
but do all step by step...)

Before this I had:
/proc/scsi/scsi:

root:~$ cat /proc/scsi/scsi 
Attached devices: 
Host: scsi0 Channel: 00 Id: 00 Lun: 00
  Vendor: SONY     Model: CD-RW  CRX800E   Rev: 1.3p
  Type:   CD-ROM                           ANSI SCSI revision: 02

After I inserted sbp2, I get:

root:~$ cat /proc/scsi/scsi 
Attached devices: 
Host: scsi0 Channel: 00 Id: 00 Lun: 00
  Vendor: SONY     Model: CD-RW  CRX800E   Rev: 1.3p
  Type:   CD-ROM                           ANSI SCSI revision: 02
Host: scsi1 Channel: 00 Id: 00 Lun: 00
  Vendor: Maxtor 6 Model: Y080L0           Rev: YAR4
  Type:   Direct-Access                    ANSI SCSI revision: 06

The Maxtor disk is my Firewire one !

I currently use 2.4.21-rc6 kernel with 2.4.21-rc1 ieee1394 stack:

jocelyn:~$ cat /proc/version 
Linux version 2.4.21-rc6-compat (jocelyn@(none)) (gcc version 3.2) #3
Sun Jun 1 21:02:23 CEST 2003

jocelyn:~$ cat /proc/modules 
sbp2                   19024   0 (unused)
sd_mod                 12332   0 (unused)
amdtp                   8596   0 (unused)
ohci1394               28016   0 [amdtp]
cmp                     2532   0 [amdtp]
ieee1394               45288   0 [sbp2 amdtp ohci1394 cmp]
sungem                 25840   1 (autoclean)
sungem_phy              5952   0 (autoclean) [sungem]

jocelyn:~$ cat /proc/scsi/sbp2/1 
IEEE-1394 SBP-2 protocol driver (host: ohci1394)
$Rev: 878 $ James Goodwin <[email protected]>
SBP-2 module load options:
- Max speed supported: S400
- Max sectors per I/O supported: 255
- Max outstanding commands supported: 64
- Max outstanding commands per lun supported: 1
- Serialized I/O (debug): no
- Exclusive login: yes

ote that I don't have neither rescan-scsi-bus nor scsiadd anywhere on this machine.

I just did a new test a few minutes ago.
To be sure that there are nothing done that I can't control, I removed /sbin/hotplug (rm -f /sbin/hotplug).

I did exactly:

modprobe ieee1394
modprobe cmp
modprobe ohci1394
modprobe amdtp
modprobe sd_mod
modprobe sbp2

as root.

(I use modprobe so I don't have to write the whole path for insmod,
but do all step by step...)

Before this I had:
/proc/scsi/scsi:

root:~$ cat /proc/scsi/scsi 
Attached devices: 
Host: scsi0 Channel: 00 Id: 00 Lun: 00
  Vendor: SONY     Model: CD-RW  CRX800E   Rev: 1.3p
  Type:   CD-ROM                           ANSI SCSI revision: 02

After I inserted sbp2, I get:

root:~$ cat /proc/scsi/scsi 
Attached devices: 
Host: scsi0 Channel: 00 Id: 00 Lun: 00
  Vendor: SONY     Model: CD-RW  CRX800E   Rev: 1.3p
  Type:   CD-ROM                           ANSI SCSI revision: 02
Host: scsi1 Channel: 00 Id: 00 Lun: 00
  Vendor: Maxtor 6 Model: Y080L0           Rev: YAR4
  Type:   Direct-Access                    ANSI SCSI revision: 06

The Maxtor disk is my Firewire one !

I currently use 2.4.21-rc6 kernel with 2.4.21-rc1 ieee1394 stack:

jocelyn:~$ cat /proc/version 
Linux version 2.4.21-rc6-compat (jocelyn@(none)) (gcc version 3.2) #3
Sun Jun 1 21:02:23 CEST 2003

jocelyn:~$ cat /proc/modules 
sbp2                   19024   0 (unused)
sd_mod                 12332   0 (unused)
amdtp                   8596   0 (unused)
ohci1394               28016   0 [amdtp]
cmp                     2532   0 [amdtp]
ieee1394               45288   0 [sbp2 amdtp ohci1394 cmp]
sungem                 25840   1 (autoclean)
sungem_phy              5952   0 (autoclean) [sungem]

jocelyn:~$ cat /proc/scsi/sbp2/1 
IEEE-1394 SBP-2 protocol driver (host: ohci1394)
$Rev: 878 $ James Goodwin <[email protected]>
SBP-2 module load options:
- Max speed supported: S400
- Max sectors per I/O supported: 255
- Max outstanding commands supported: 64
- Max outstanding commands per lun supported: 1
- Serialized I/O (debug): no
- Exclusive login: yes

jocelyn:~$ cat /proc/bus/ieee1394/devices 
Node[00:1023]  GUID[000393fffeab0e76]:
  Vendor ID: `Linux OHCI-1394' [0x000000]
  Capabilities: 0x0083c0
  Bus Options:
    IRMC(1) CMC(1) ISC(1) BMC(0) PMC(0) GEN(8)
    LSPD(2) MAX_REC(2048) CYC_CLK_ACC(0)
  Host Node Status:
    Host Driver     : ohci1394
    Nodes connected : 2
    Nodes active    : 2
    SelfIDs received: 2
    Irm ID          : [00:1023]
    BusMgr ID       : [00:1023]
    In Bus Reset    : no
    Root            : no
    Cycle Master    : no
    IRM             : yes
    Bus Manager     : yes
Node[01:1023]  GUID[00d04b0ce0900195]:
  Vendor ID: `Oxford  ' [0x00d04b]
  Capabilities: 0x0083c0
  Bus Options:
    IRMC(0) CMC(0) ISC(0) BMC(0) PMC(0) GEN(0)
    LSPD(0) MAX_REC(64) CYC_CLK_ACC(255)
  Unit Directory 0:
    Vendor/Model ID: Oxford   [00d04b] / 911G [000001]
    Software Specifier ID: 00609e
    Software Version: 010483
    Driver: SBP2 Driver
    Length (in quads): 8

So, I'm quite sure that it should work with no special helper ! Is there any other informations I should give,
or things I should do to understand the problem ?

Jocelyn Mayer <[email protected]>

Something else is fishy. If I put all the modules for my firewire drive, in the correct order, in an initial RAM disk (initrd), and then attempt to use the firewire drive as the root file-system, it is not available. If I install all the modules for the firewire drives from initrd, then mount a SCSI root file-system, the firewire drive is still not available!!! An attempt to access the raw device with od shows 'no such device or address'. If I remove all the firewire modules, then install them again, like magic the drive becomes available! This is not good magic. This is on 2.4.20 (RedHat 9 tools)

Cheers,
Dick Johnson
Penguin : Linux version 2.4.20 on an i686 machine (797.90 BogoMips). Why is the government concerned about the lunatic fringe? Think about it.

I didn't say CONFIG_HOTPLUG, I said hotplug. Basically SCSI in 2.4 will not let recognize devices that were not present when the scsi-host was initially registered with the SCSI stack. You have to run rescan-scsi-bus.sh (or manually send the add/remove commands via procfs).

Please read the linux-kernel and/or linux1394-devel mailing list archives. I really hate dredging this all up again.

Debian     - http://www.debian.org/
Linux 1394 - http://www.linux1394.org/
Subversion - http://subversion.tigris.org/
Deqo       - http://www.deqo.com/

Well, I did understand well hotplug...
I did read what's said about SBP2 on linux1394.org, BUT:

please read my latest mail, I did some try with the 2.4.21-rc1 ieee1394 stack, without /sbin/hotplug, and it worked.
I have neither rescan-scsi-bus.sh, nor scsiadd anywhere on this machine. So, I'm sure that no user-space tool have been called by the kernel.

- I NEVER need to do an "echo ... > /proc/scsi/scsi" to see the device (I tell again that I never use automatic tools to do this kind of low level stuffs).
- I NEVER have to do this to see an USB mass-storage device:
I'm *REALLY* sure of this, as I wrote a very simple hotplug and automounter for some embedded hardware which uses USB mass storage devices. So, I don't see why SBP2 should need this...

- Please take a look at this session:
First, I reboot the Ibook:

root:~$ reboot
root:~$ 
Broadcast message from root (pts/0) (Tue Jun  3 22:06:13 2003):

The system is going down for reboot NOW!
Connection to mac closed by remote host.
Connection to mac closed.
jma ~ > ssh jocelyn@mac
jocelyn@mac's password:

jocelyn:~$ su -
Password: 
root:~$ ls -l /bin/hotplug
ls: /bin/hotplug: No such file or directory
root:/lib/modules/2.4.21-rc6-compat/kernel/drivers/ieee1394$ cat
/proc/sys/kernel/hotplug 
/sbin/hotplug
root:~$ cd /lib/modules/2.4.21-rc6-compat/kernel/drivers/ieee1394/
root:/lib/modules/2.4.21-rc6-compat/kernel/drivers/ieee1394$ insmod
./ieee1394.o

root:/lib/modules/2.4.21-rc6-compat/kernel/drivers/ieee1394$ insmod
./cmp.o 
root:/lib/modules/2.4.21-rc6-compat/kernel/drivers/ieee1394$ cat
/proc/scsi/scsi

Attached devices:
Host: scsi0 Channel: 00 Id: 00 Lun: 00
Vendor: SONY Model: CD-RW CRX800E Rev: 1.3p
Type: CD-ROM ANSI SCSI revision: 02

root:/lib/modules/2.4.21-rc6-compat/kernel/drivers/ieee1394$ insmod
./ohci1394.o

root:/lib/modules/2.4.21-rc6-compat/kernel/drivers/ieee1394$ insmod
./amdtp.o 
root:/lib/modules/2.4.21-rc6-compat/kernel/drivers/ieee1394$ insmod
../scsi/sd_m
od.o 
root:/lib/modules/2.4.21-rc6-compat/kernel/drivers/ieee1394$ insmod
./sbp2.o 
root:/lib/modules/2.4.21-rc6-compat/kernel/drivers/ieee1394$ cat
/proc/scsi/scsi

Attached devices:
Host: scsi0 Channel: 00 Id: 00 Lun: 00
Vendor: SONY Model: CD-RW CRX800E Rev: 1.3p
Type: CD-ROM ANSI SCSI revision: 02
Host: scsi1 Channel: 00 Id: 00 Lun: 00
Vendor: Maxtor 6 Model: Y080L0 Rev: YAR4
Type: Direct-Access ANSI SCSI revision: 06

So I NEVER write anything to /proc/scsi/scsi, I got NO /sbin/hotplug, I use only insmod, so I can have no side-effect due to modprobe usage, and it works PERFECTLY on the Ibook,
using the 2.4.21-rc1 ieee1394 stack.

So, I'm sorry, but I keep thinking the new driver is buggy:
what used to work and doesn't...

Regards.

Jocelyn Mayer <[email protected]>

I know jumping in the middle of a conversation is bad, but....

In conversations with the SBP2 folks, they indicated to me that the way they do hotplugging is very different from the way usb-storage does it. The end result (I'm told) is that invoking a scan from userspace is often needed for SBP2 but never for usb-storage.

So, comparing the two is really pointless.

Matt

Matthew Dharm                              Home: [email protected] 
Maintainer, Linux USB Mass Storage Driver

Would you mind not using our Web server? We're trying to have a game of 
Quake here.
					-- Greg
User Friendly, 5/11/1998

Hi,

you're right, I just wanted to point that there's no reason that we need to register a device etheir by hand or using an "infamous script" (citation from Ben Collins
http://sourceforge.net/mailarchive/message.php?msg_id=4435485 ) due to the SCSI stack, but that it's only a SBP2 problem.

Regards.

Actually, it is/was a SCSI problem. The short version being that it is/was easier to hotplug an entire HBA rather than an individual device. That is being changed in 2.5 right now. usb-storage used an HBA-per-device, while SBP2 uses a single HBA.

Of course, all this is being changed right now in 2.5.x

Matt

Matthew Dharm                              Home: [email protected] 
Maintainer, Linux USB Mass Storage Driver

It's monday.  It must be monday.
					-- Greg
User Friendly, 5/4/1998

You didn't read a damn thing.

Look, things changed around rc2. They changed because the old way of sbp2 that allowed sbp2 devices to be detected on module load caused oopses. It was very buggy and very bad. Even the old logic didn't work when you loaded sbp2 and _then_ plugged the device in.

I HAVE NOT once said anything about fucking /sbin/hotplug, nor the kernel's idea of CONFIG_HOTPLUG. I just said "hotplug", which is a generic term for being able to insert devices after the bus has already been scanned for them. So stop bring up either of the two previous notions that you may have.

Your behavior is expected, even if it changed and is inconvient to you. This time, _really_ read the linux1394-devel mailing list archives. And just get the damn script and use it. If you had ever really used sbp2 previously and wanted to plug/unplug devices without unloading/reloading sbp2.o, you would already realize that the script was a must for 2.4. Then again, if you had been unloading/loading sbp2 you would have also hit the oopses I mentioned before.

Debian     - http://www.debian.org/
Linux 1394 - http://www.linux1394.org/
Subversion - http://subversion.tigris.org/
Deqo       - http://www.deqo.com/

Yes, I didn't read the mailing list,
only the SBP2 page.
I did, now, read the thread "initial bus scan gone" and YOU refer this as a known bug, not as a normal feature.

I may be lucky, but I never had a Ooops.
You're right on one point: I ALWAYS unload/reload the sbp2 driver. I do this on a PC about once a day and had an 30 days uptime. I don't do it on the Mac because it's a laptop and I usually keep the disk mounted until I stop it.
I know that echo to /proc/scsi/scsi is needed for device REMOVAL, but, as I said before, I NEVER needed it for device's insertion.

I use sbp2 devices every day for monthes, but never tried to know how it worked since it was working well for me, so please don't tell me I never used it...

Sorry to have talked to you, making you so angry,
won't disturb you again.

Time to close this thread, now.

Jocelyn Mayer <[email protected]>

Want to see buggy? After you load all this up on rc1, rmmod sbp2 and then do:

cat /proc/scsi/sbp2/1

Debian     - http://www.debian.org/
Linux 1394 - http://www.linux1394.org/
Subversion - http://subversion.tigris.org/
Deqo       - http://www.deqo.com/