Due to a firmware mismatch between a host and target (names withheld to
protect the innocent?), the LLDD was returning DID_RESET for every
i/o command.  This patch modifies the scsi layer to take into account
when the command which received DID_RESET was issued and eventually
give up on it instead of unconditionally reissuing it forever
when it receives a DID_RESET.  With this patch, on my test system,
the command receiving the constant DID_RESET times out after about
360 seconds.

The premise for this patch is that no command should have an infinite
lifetime.  The impetus for this patch was a system which would not
reach a command prompt without disconnecting the storage from the
host.

The significant change in this patch is to call scsi_retry_command()
instead of scsi_requeue_command() if the command which receives a
DID_RESET did not complete any i/o (good_bytes==0).  scsi_retry_command()
does not release the command and regenerate it like scsi_requeue_command()
does, hence jiffies_at_alloc reflects when the command was first issued.

This patch is based upon 2.6.19.  Thanks for taking the time to
look at this.

Mike