Re: USB transfer_buffer allocations on 64bit systems

[Takashi Iwai <tiwai@suse.de>]

At Wed, 7 Apr 2010 11:55:19 -0400 (EDT),
Alan Stern wrote:
> 
> On Wed, 7 Apr 2010, Greg KH wrote:
> 
> > Yeah, I really don't want to have to change every driver in different
> > ways just depending on if someone thinks it is going to need to run on
> > this wierd hardware.
> 
> It's not weird hardware, as far as I know.  It's just a 64-bit system
> with a 32-bit USB host controller.
> 
> (And remember, while there are 64-bit EHCI controllers, there are not 
> any 64-bit OHCI or UHCI controllers.  So whenever somebody plugs a 
> full-speed or low-speed device into a 64-bit machine, they will face 
> this problem.  It's like the old problem of ISA devices that could 
> only do DMA to addresses in the first 16 MB of memory -- what the 
> original GFP_DMA flag was intended for.)
> 
> > Alan, any objection to just using usb_buffer_alloc() for every driver?
> > Or is that too much overhead?
> 
> I don't know what the overhead is.  But usb_buffer_alloc() requires the 
> caller to keep track of the buffer's DMA address, so it's not a simple 
> plug-in replacement.  In addition, the consistent memory that 
> usb_buffer_alloc() provides is a scarce resource on some platforms.

Yeah, also the area is aligned to kernel pages, and it may be much
bigger than the requested (power-of-two).  If not needed, we should
avoid it.


thanks,

Takashi