Re: [PATCH 1/3]USB: add Freescale high-speed USB SOC device controller driver

> -----Original Message-----
> From: David Brownell [mailto:[email protected]]
> Sent: Friday, April 20, 2007 12:53 PM
> To: Li Yang-r58472
> Cc: [email protected]; [email protected]; Schmid Bruce-R62923
> Subject: Re: [linux-usb-devel] [PATCH 1/3]USB: add Freescale high-speed USB SOC
> device controller driver
> 
> On Thursday 19 April 2007 8:56 pm, Li Yang-r58472 wrote:
> >
> > > > > On Thursday 19 April 2007 2:53 am, Li Yang wrote:
> > > > > > +static void *fsl_alloc_buffer(struct usb_ep *_ep, unsigned bytes,
> > > > > > +               dma_addr_t *dma, gfp_t gfp_flags)
> > > > > > +{...
> > > > > > +}
> > > > >
> > > > > You still need to fix this to use dma_alloc_coherent(), and its
> > > > > sibling function to use dma_free_coherent().
> > > > >
> > > > > This particular code assumes !CONFIG_NOT_COHERENT_CACHE ... which
> > > > > ISTR you've said is not true on all the relevant PPC platforms, and
> > > > > is certainly not true on the versions found (already?) in Freescale
> > > > > ARM products.
> > > >
> > > > This buffer is synced using dma_sync_single_*/dma_(un)map_single when
> > > > the request is queued and when the transmission is complete.  To
> > >
> > > No.  See <linux/usb_gadget.h> for the definition of usb_ep_alloc_buffer();
> > > it says explicitly that such mapping calls are not required.
> > >
> > > The only portable way to guarantee that is to allocate coherent memory.
> >
> > Then I can remove the dma_sync_single calls from ep_queue() and done()?
> 
> Not a good idea.  There are two ways that a DMA address gets
> provided to the lower level code:  (a) coherent buffer allocation,
> as above, and (b) mapping done by driver, and re-used.  The former
> doesn't need dma_sync_single().  The latter does.

Do you mean that the buf of a usb_request could be allocated other than calling usb_ep_alloc_buffer()?  If not, the buf would be guaranteed to be coherent buffer.
And will usb_ep_alloc_buffer() be used to allocate buffer other than buffer to be queued with ep_queue()?

I mean if usb_ep_alloc_buffer() is required and only used in allocating buffer to be processed by ep_queue() and done().  We can choose to use one of the ways you mentioned above, not both.  In this case, I preferred to choose the second way for better performance.

- Leo

-------------------------------------------------------------------------
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