Dailey, Nate wrote:
> Doug, I'm attaching a new patch (again, against 2.6.9)... the primary
> difference from my previous patch is that this one now schedules a work
> item from sg_cmd_done. This work item drops the reference on the device,
> so this gets rid of the problems of calling scsi_device_put from
> sg_cmd_done or of sleeping in sg_release. Does this look okay?

Nate,
I haven't had a close look at the logic but I have
been using this patch for several days. So far I have
been testing other things with sg and have detected
no regression from this patch.

Since your patch was against 2.6.9, it needed a few
changes to apply to lk 2.6.14-rc5 . The patch against
2.6.14-rc5 is attached.

Hopefully in the next few days I can focus on
testing this patch.

Doug Gilbert