scsi_add_device() broken? (was Re: SBP2 hotplug doesn't update /proc/partitions)

scsi_add_device() broken? (was Re: SBP2 hotplug doesn't update /proc/partitions)

[Ben Collins]

On Thu, Jun 12, 2003 at 03:52:43PM -0400, Ben Collins wrote:
> On Thu, Jun 12, 2003 at 12:28:00PM -0700, Torrey Hoffman wrote:
> > I am now running 2.5.70-bk15, and with slab debugging turned off SBP2
> > mostly works.  However, I just had an interesting glitch show up.
> > 
> > I plugged in a 120 GB drive which had two VFAT partitions, mounted them,
> > copied some data to them, unmounted them, and unplugged the drive.  
> > That worked perfectly. (This was the first use of SBP2 after booting.)
> > 
> > Then I plugged in a 250 GB drive with a single reiserfs partition.  The
> > SBP2 driver detected the drive correctly, but the kernel's idea of what
> > partitions are available was not updated.  
> > 
> > /proc/partitions still has the old, stale data from the 120 GB drive and
> > looks like this: (skipping my hda partitions)
> 
> Sounds like the scsi layer is keeping stale info. I'd say this is
> suspiciously similar to what's causing your oops in your later email.
> Track down where the stale info comes from, and I think you'll find the
> cause of both your problems.

scsi0 : SCSI emulation for IEEE-1394 SBP-2 Devices
ieee1394: sbp2: Query logins to SBP-2 device successful
ieee1394: sbp2: Maximum concurrent logins supported: 1
ieee1394: sbp2: Number of active logins: 0
ieee1394: sbp2: Logged into SBP-2 device
ieee1394: sbp2: Node[02:1023]: Max speed [S400] - Max payload [2048]
  Vendor: FireWire  Model:  1394 Disk Drive  Rev: G603
  Type:   Direct-Access                      ANSI SCSI revision: 02
SCSI device sda: 240121728 512-byte hdwr sectors (122942 MB)
sda: cache data unavailable
sda: assuming drive cache: write through
 sda: unknown partition table
devfs_mk_dir: invalid argument.<5>Attached scsi disk sda at scsi0, channel 0, id 0, lun 0


Note, the "unknown partition table" is ok for me, because I am using a
whole-disk filesystem. However, the devfs_mk_dir() error is suspicious.
Note, I use devfs. 2.5.69+bk (just before 2.6.70) things worked fine.
Now, /dev/sda is not being created. Nothing in /dev/scsi/ either, but:

hopper:~# cat /proc/scsi/sbp2/0
Host scsi0             : SBP-2 IEEE-1394 (ohci1394)
Driver version         : $Rev: 942 $ Ben Collins <bcollins@debian.org>

Module options         :
  max_speed            : S400
  max_sectors          : 255
  serialize_io         : no
  exclusive_login      : no

Attached devices       :
  [Channel: 00, Id: 00, Lun: 00]  Direct-Access     FireWire  1394 Disk Drive


Also, /sys/bus/scsi/devices/0:0:0:0 exists, with a block link to sda.

I suspect this is a problem where scsi_add_device() is not doing all the
stuff it needs to (does anything besides ieee1394 use scsi_add_device
and scsi_remove_device?).


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

Re: scsi_add_device() broken? (was Re: SBP2 hotplug doesn't update /proc/partitions)

[B. D. Elliott]

On Thu, Jun 12, 2003 at 10:40:44PM -0400, Ben Collins wrote:
> On Thu, Jun 12, 2003 at 03:52:43PM -0400, Ben Collins wrote:
> > On Thu, Jun 12, 2003 at 12:28:00PM -0700, Torrey Hoffman wrote:
> > > I am now running 2.5.70-bk15, and with slab debugging turned off SBP2
> > > mostly works.  However, I just had an interesting glitch show up.
> > > 
> > > I plugged in a 120 GB drive which had two VFAT partitions, mounted them,
> > > copied some data to them, unmounted them, and unplugged the drive.  
> > > That worked perfectly. (This was the first use of SBP2 after booting.)
> > > 
> > > Then I plugged in a 250 GB drive with a single reiserfs partition.  The
> > > SBP2 driver detected the drive correctly, but the kernel's idea of what
> > > partitions are available was not updated.  
> > > 
> > > /proc/partitions still has the old, stale data from the 120 GB drive and
> > > looks like this: (skipping my hda partitions)
> > 
> > Sounds like the scsi layer is keeping stale info. I'd say this is
> > suspiciously similar to what's causing your oops in your later email.
> > Track down where the stale info comes from, and I think you'll find the
> > cause of both your problems.
> 
> scsi0 : SCSI emulation for IEEE-1394 SBP-2 Devices
> ieee1394: sbp2: Query logins to SBP-2 device successful
> ieee1394: sbp2: Maximum concurrent logins supported: 1
> ieee1394: sbp2: Number of active logins: 0
> ieee1394: sbp2: Logged into SBP-2 device
> ieee1394: sbp2: Node[02:1023]: Max speed [S400] - Max payload [2048]
>   Vendor: FireWire  Model:  1394 Disk Drive  Rev: G603
>   Type:   Direct-Access                      ANSI SCSI revision: 02
> SCSI device sda: 240121728 512-byte hdwr sectors (122942 MB)
> sda: cache data unavailable
> sda: assuming drive cache: write through
>  sda: unknown partition table
> devfs_mk_dir: invalid argument.<5>Attached scsi disk sda at scsi0, channel 0, id 0, lun 0

This particular error occurs because the scsi/host*/bus*/target*/lun* path
hasn't yet been set up in scsi_add_lun before calling scsi_device_register.
This also occurs for usb disk registration.  (I don't know why this doesn't
happen for devices that register at boot time.)  Moving the call to
scsi_device_register (at scsi_scan.c:622) after the sprintf appears to fix
both the usb and ieee1934 problems.

> Note, the "unknown partition table" is ok for me, because I am using a
> whole-disk filesystem. However, the devfs_mk_dir() error is suspicious.
> Note, I use devfs. 2.5.69+bk (just before 2.6.70) things worked fine.
> Now, /dev/sda is not being created. Nothing in /dev/scsi/ either, but:
> 
> hopper:~# cat /proc/scsi/sbp2/0
> Host scsi0             : SBP-2 IEEE-1394 (ohci1394)
> Driver version         : $Rev: 942 $ Ben Collins <bcollins@debian.org>

A side note here:  The "$Rev: 942..." message is marked as __devinitdata
which causes an oops on sparc64 when attempting the above "cat", as the
mapping for that data is apparently deleted after module initialization.

> Module options         :
>   max_speed            : S400
>   max_sectors          : 255
>   serialize_io         : no
>   exclusive_login      : no
> 
> Attached devices       :
>   [Channel: 00, Id: 00, Lun: 00]  Direct-Access     FireWire  1394 Disk Drive
> 
> 
> Also, /sys/bus/scsi/devices/0:0:0:0 exists, with a block link to sda.
> 
> I suspect this is a problem where scsi_add_device() is not doing all the
> stuff it needs to (does anything besides ieee1394 use scsi_add_device
> and scsi_remove_device?).
> 
> 
> -- 
> Debian     - http://www.debian.org/
> Linux 1394 - http://www.linux1394.org/
> Subversion - http://subversion.tigris.org/
> Deqo       - http://www.deqo.com/
> -
> To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 

-- 
B. D. Elliott   bde@nwlink.com

[PATCH] Re: scsi_add_device() broken? (was Re: SBP2 hotplug doesn't update /proc/partitions)

[Ben Collins]

> scsi0 : SCSI emulation for IEEE-1394 SBP-2 Devices
> ieee1394: sbp2: Query logins to SBP-2 device successful
> ieee1394: sbp2: Maximum concurrent logins supported: 1
> ieee1394: sbp2: Number of active logins: 0
> ieee1394: sbp2: Logged into SBP-2 device
> ieee1394: sbp2: Node[02:1023]: Max speed [S400] - Max payload [2048]
>   Vendor: FireWire  Model:  1394 Disk Drive  Rev: G603
>   Type:   Direct-Access                      ANSI SCSI revision: 02
> SCSI device sda: 240121728 512-byte hdwr sectors (122942 MB)
> sda: cache data unavailable
> sda: assuming drive cache: write through
>  sda: unknown partition table
> devfs_mk_dir: invalid argument.<5>Attached scsi disk sda at scsi0, channel 0, id 0, lun 0

Here's the scenario. scsi_add_lun doesn't set sdp->devfs_name before
calling scsi_register_device(). Since scsi_register_device calls down to
things like sd_probe, which do try to use sdp->devfs_name, things fail.

Just an easy change, moving the sdp->devfs_name creation before calling
scsi_register_device(). Patch fixes this.

Index: linux-2.5/drivers/scsi/scsi_scan.c
===================================================================
--- linux-2.5/drivers/scsi/scsi_scan.c	(revision 10937)
+++ linux-2.5/drivers/scsi/scsi_scan.c	(working copy)
@@ -619,12 +619,12 @@
 	if (inq_result[7] & 0x10)
 		sdev->sdtr = 1;
 
-	scsi_device_register(sdev);
-
 	sprintf(sdev->devfs_name, "scsi/host%d/bus%d/target%d/lun%d",
 				sdev->host->host_no, sdev->channel,
 				sdev->id, sdev->lun);
 
+	scsi_device_register(sdev);
+
 	/*
 	 * End driverfs/devfs code.
 	 */

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

Re: [PATCH] Re: scsi_add_device() broken? (was Re: SBP2 hotplug doesn't update /proc/partitions)

[Patrick Mansfield]

On Fri, Jun 13, 2003 at 12:08:12PM -0400, Ben Collins wrote:

> Here's the scenario. scsi_add_lun doesn't set sdp->devfs_name before
> calling scsi_register_device(). Since scsi_register_device calls down to
> things like sd_probe, which do try to use sdp->devfs_name, things fail.
> 
> Just an easy change, moving the sdp->devfs_name creation before calling
> scsi_register_device(). Patch fixes this.

It really needs to move into the caller of scsi_add_lun, so we setup all
the other fields, and possibly call scsi_unlock_floptical before
registering.

And the return value should be checked - then on failure just set res (in
scsi_probe_and_add_lun) so it is cleaned up.

-- Patrick Mansfield