Re: [usb-storage] Re: Make UAS work on HS for devices with and without command tagging support

[James Bottomley <James.Bottomley@HansenPartnership.com>]

On Fri, 2011-12-16 at 13:36 -0800, Sarah Sharp wrote:
> On Fri, Dec 16, 2011 at 09:42:41PM +0100, Sebastian Andrzej Siewior wrote:
> > * Matthew Wilcox | 2011-12-16 15:31:46 [-0500]:
> > 
> > >On Fri, Dec 16, 2011 at 09:12:36PM +0100, Sebastian Andrzej Siewior wrote:
> > >> * Sebastian Andrzej Siewior | 2011-12-16 15:47:24 [+0100]:
> > >> 
> > >> >>If you want to take a stab at redoing your patch #2 to use only one
> > >> >>status URB for USB 2.0 devices, I would appreciate it.  Then I can build
> > >> >>the abort/reset synchronization on top of it.
> > >> >Okay.
> > >> 
> > >> Just once things started to become easy.... So while I tried to have
> > >> only one status urb which I always re-submit (as Matthew/ You suggested)
> > >> I run into the problem that I don't have struct scsi_device yet. So I
> > >> just created a device with two luns to see if this struct happens always
> > >> to be same. Ofcourse it is not. 
> > >
> > >Can you not send one status URB per LUN (instead of one per command)?
> > 
> > The thing is by the time a status URB completes I have only the *TAG*
> > number from the device which tells to which command it belongs. Sending
> > one status per LUN does not help because once a status URB with TAG 1
> > arrived I have no idea to which device/LUN it does belong. ->context
> > does not help here at all.
> 
> Oh, right, now I understand.  With USB 2.0, we might have one status URB
> per LUN, but they're being queued to the same endpoint ring.  Since the
> device can re-order the commands any way it likes, we can't rely on the
> scsi_device that's stored on the urb->context to be correct.
> 
> Ok, I think we just need to divide the tag address space equally between
> devices.  The devinfo can keep track of the pointers to the scsi_devices
> for each LUN, and what tag range each LUN has.  So if you have 255 tags
> available, and two LUNs, you can give one of them tag 1 to 122, and the
> other tags 123 to 254.

Well, no, what you want to do is use a shared tag map at the block
level.  That will manage a joint tag space for N queues without your
having to partition it arbitrarily.  see scsi_host_find_tag() and
scsi_init_shared_tag_map().

James