Dan Hollis wrote:
On Sun, 24 Oct 2004, Olaf Hering wrote:
On Sun, Oct 24, James Bottomley wrote:
The trace doesn't show any error handler activity at all. Are there no
messages in the log about offlining the device? If not, it sounds like
there's a problem somewhere in the firewire system.
How should sbp2_remove_device get rid of the device? It calls
scsi_remove_host, which calls scsi_remove_device, which sets the mode to
SDEV_CANCEL, then calls to device_del. sr_do_ioctl expects SDEV_OFFLINE.
This has been subtly changed in the latest sbp2 source from linux1394.org.
scsi_remove_device is now called before sbp2_logout_device and
sbp2_remove_device. The latter is still the caller of scsi_remove_host.
However, this little tweak only helps when sbp2 is unloaded or detached
from a device while the device is still physically present and working.
It does not improve the situation when a device was physically removed
while the drivers were still attached.
Has there been any progress on this issue? I just got my _entire scsi
subsystem_ sent out to lunch because a firewire device lost power.
The situation has got even worse. Since RBC conversions were moved from
sbp2 to sd_mod in linux-2.6.13-rc2, the same or a similar problem is now
exhibited by sd_mod too with most of the FireWire harddisks. The culprit
in sd_mod is obviously sd_shutdown/ sd_sync_cache, which blocks at
scsi_wait_req if the device is physically gone, thus effectively freezing
ieee1394's nodemgr thread. I have not yet looked closer into what might
be the equivalent in sr_mod.
However, there is a positive side of this: 1. With harddisks triggering
the same or a similar problem now, the pressure got really high to
finally fix this extremely annoying bug. I am trying to understand the
nature of the problem right now, but I am new in linux-scsi land. 2. At
least unplugging should be less of a problem if you always remember to
detach or unload sbp2 first. (Or alternatively sr_mod/ sd_mod?) Of
course this does not help if power on the SBP-2 device failed or
somebody tripped over the wire. And detaching the driver manually is of
course quite awkward.
--
Stefan Richter
-=====-=-=-= -=== =====
http://arcgraph.de/sr/
-
: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
