Re: patch to shift memory allocation s to usb_alloc_urb ()

Am Montag, 2. Juli 2007 schrieb David Brownell:
> On Saturday 30 June 2007, Oliver Neukum wrote:
> 
> > So do we have consensus for this prototype:
> > 
> > struct urb *usb_alloc_urb(struct usb_host_endpoint *ep,
> > 		unsigned int max_length, unsigned int iso_fields,
> > 		gfp_t mem_flags);  
> 
> Calling conventions that require line-wrapping are almost by
> definition unwieldy ... :(

In usbcore we have names_containing_exact_descriptions.
But yes. I intended to find a unified API that allows preallocation.
If we can live with a separate allocator for ISO URBs, we can
do better.

> That might make sense for an HCD method; size_t for length,
> and rename it so it doesn't suggest it's a "maximum" so much
> as an educated guess.

So you are suggesting that usbcore should preallocate, but
also, if need be allocate more in usb_submit_urb()? So that
the guarantee against getting ENOMEM is valid only if you
don't exceed the hint given at allocation time?
That is an idea I would have never had, but it is very good.

> But otherwise, I thought one of the goals was to get rid of
> such annoying "iso fields" (should be packets, not fields!)
> parameters, which were always being passed as zero ... ?
> 
> I'd expect drivers would just pass {ep, length, gfp} except
> for the rare ISO ones, which would pass the number of packets.
> That suggests two separate signatures at that level.

OK.

	Regards
		Oliver

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
_______________________________________________
[email protected]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel