Re: Patch [2/2] DaVinci CPPI TX DMA tasklet

[David Brownell <david-b-yBeKhBN/0LDR7s880joybQ@public.gmane.org>]

On Thursday 21 August 2008, Subbrathnam, Swaminathan wrote:
> Without these changes DMA on DaVinci has never worked traditionally.
> We fixed this issue pretty early in DaVinci program ~2006 on the 2.6.10
> treee. 

I remember a recommendation not to use DMA, but that was primarily
because it seemed to hurt performance (unless it was configured to
provide incorrect semantics).  If there were also correctness issues,
2006 is too long ago for me to remember.  :)


> We are now in the process of forward porting the musb patches 
> from that tree to the git tree.
> 
> We did some performance enhancements/bug fixes on that tree and hope to make
> it available to the general community w.r.t MUSB.  This (specific to DaVinci)
> and the following patches (musb generic) are in that direction.

Great.  Seems to me those could be ready for the 2.6.28 merge window,
given some updates to the DaVinci platform support.

They'll need to be in nicely reviewable chunks ... I remember seeing
some DaVinci patches in late 2006 which broke on TUSB6010 silicon,
for example.

DMA in particular could really stand some cleanup.  Having four
different chunks of DMA code -- RX/TX vs Host/Peripheral -- with
ifdeffery for multiple DMA engines is ... chaotic.


> I discussed with Kevin on the DaVinci Git tree and Filipe here and they recommended that
> I take the changes to linux-usb, linux-omap tree (w.r.t musb changes) to get a wider 
> audience for review and acceptance.
> 
> Kevin would then pull in the changes as part of his regular synch ups.

I think the plan should be to have various SOC-specific trees (DaVinci,
OMAP, Blackfin, etc) stop hosting MUSB-specific patches.  They should
be pushed up to mainline ASAP ... the linux-omap tree should stop being
the place where the latest MUSB code sits.


> This patch (+ CPPI DMA changes) in a different form has already been 
> tested in the DaVinci git community

It has???   It's not git://source.mvista.com/git/linux-davinci-2.6 ...


>	 and it resolved the USB DMA issues and 
> improved performance considerably as communicated by the community
> members like Neuros.
> 
> I will be submitting the CPPI DMA changes as a seperate patch as it is big
> and based on preliminary review needs to be broken down significantly to
> enable better review and acceptance.

I'm quite willing to believe the CPPI stuff can be improved.  :)

- Dave


> Thanks for your comments.
> 
> regards
> swami
> 
> ________________________________________
> From: David Brownell [david-b-yBeKhBN/0LDR7s880joybQ@public.gmane.org]
> Sent: Friday, August 22, 2008 12:45 AM
> To: Dmitry Krivoschekov; Subbrathnam, Swaminathan
> Cc: linux-omap-u79uwXL29TY76Z2rM5mHXA@public.gmane.org; linux-usb-u79uwXL29TY76Z2rM5mHXA@public.gmane.org; felipe.balbi-xNZwKgViW5gAvxtiuMwx3w@public.gmane.org; Kevin Hilman
> Subject: Re: Patch [2/2] DaVinci CPPI TX DMA tasklet
> 
> On Thursday 21 August 2008, Dmitry Krivoschekov wrote:
> > why don't rely on an endpoint interrupt which is generated
> > after a USB transaction is complete?
> 
> That's what I thought *was* working a couple years ago...
> 
> I take it that something broke since then?  Or is this
> a corner case which didn't always work right?
> 
> I'm not sure this would be appropriate for mainline
> until the DaVinci code can run there.  There are a
> bunch of arch/arm/mach-davinci updates needed first.
> I know Kevin was looking at some of this, but just
> now I think enabling MUSB on a 2.6.27-rc4 DaVinci
> config means a broken build.
> 
> - Dave


--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html