Messages in this thread | | | Date | Tue, 26 May 2009 17:04:48 -0400 (EDT) | From | Alan Stern <> | Subject | Re: Bug in SCSI async probing |
| |
On Tue, 26 May 2009, James Bottomley wrote:
> > > > (Which reminds me... Are the calls in wait_scan_init() really enough? > > > > wait_for_device_probe() does async_synchronize_full() and then > > > > scsi_complete_async_scans() finishes the SCSI scanning. But if this > > > > scanning involves calling sd_probe(), then more async work will be > > > > queued. Maybe a second call to wait_for_device_probe() is needed.) > > > > You didn't respond to this point. > > Well this was the response: > > > > None of this really got reviewed through the SCSI list, so I'll let > > > Arjan answer.
Whoops, for some reason I had gotten the idea that you wrote wait_scan_init().
> > Well then, how does this patch look? > > Well, it's adding complexity, the best fix is to let async only take > care of the pieces which can't fail, that way we don't need complex > error handling. The piece that slows probing isn't really the sysfs > appearances, it's the SCSI probing, and the last piece that needs error > handling is the device_add() for sysfs visibility, so that should be the > dividing line between sync and async.
I had exactly the same thought, but it seemed less intrusive to keep the dividing line where Arjan put it.
> This should restore the logical flow and fix all the error leg problems > (by eliminating the error legs).
You forgot to move the dev_set_drvdata() call into the synchronous part. Apart from that it looks fine. Should I submit it officially?
Also, I do still think that wait_scan_init() needs an extra call to async_synchronize_full() at the end.
Alan Stern
| |