Re: [PATCH v10 2/2] dmaengine: sun6i: Add driver for the Allwinner A31 DMA controller

[Maxime Ripard]

Hi Mark,

On Tue, Jul 01, 2014 at 01:48:52PM +0100, Mark Rutland wrote:
> On Tue, Jul 01, 2014 at 08:21:19AM +0100, Maxime Ripard wrote:
> > On Mon, Jun 30, 2014 at 04:33:05PM +0100, Mark Rutland wrote:
> > > On Mon, Jun 30, 2014 at 04:19:06PM +0100, Maxime Ripard wrote:
> > > > On Mon, Jun 30, 2014 at 03:20:54PM +0100, Mark Rutland wrote:
> > > > > Hi Maxime,
> > > > > 
> > > > > On Mon, Jun 30, 2014 at 02:20:54PM +0100, Maxime Ripard wrote:
> > > > > > The Allwinner A31 has a 16 channels DMA controller that it shares with the
> > > > > > newer A23. Although sharing some similarities with the DMA controller of the
> > > > > > older Allwinner SoCs, it's significantly different, I don't expect it to be
> > > > > > possible to share the driver for these two.
> > > > > >
> > > > > > The A31 Controller is able to memory-to-memory or memory-to-device transfers on
> > > > > > the 16 channels in parallel.
> > > > > >
> > > > > > Signed-off-by: Maxime Ripard <maxime.ripard@xxxxxxxxxxxxxxxxxx>
> > > > > > Acked-by: Arnd Bergmann <arnd@xxxxxxxx>
> > > > > > ---
> > > > > >  drivers/dma/Kconfig     |    8 +
> > > > > >  drivers/dma/Makefile    |    1 +
> > > > > >  drivers/dma/sun6i-dma.c | 1058 +++++++++++++++++++++++++++++++++++++++++++++++
> > > > > >  3 files changed, 1067 insertions(+)
> > > > > >  create mode 100644 drivers/dma/sun6i-dma.c
> > > > > 
> > > > > [...]
> > > > > 
> > > > > > +       sdc->clk = devm_clk_get(&pdev->dev, NULL);
> > > > > > +       if (IS_ERR(sdc->clk)) {
> > > > > > +               dev_err(&pdev->dev, "No clock specified\n");
> > > > > > +               return PTR_ERR(sdc->clk);
> > > > > > +       }
> > > > > > +
> > > > > > +       mux = clk_get(NULL, "ahb1_mux");
> > > > > > +       if (IS_ERR(mux)) {
> > > > > > +               dev_err(&pdev->dev, "Couldn't get AHB1 Mux\n");
> > > > > > +               return PTR_ERR(mux);
> > > > > > +       }
> > > > > > +
> > > > > > +       pll6 = clk_get(NULL, "pll6");
> > > > > > +       if (IS_ERR(pll6)) {
> > > > > > +               dev_err(&pdev->dev, "Couldn't get PLL6\n");
> > > > > > +               clk_put(mux);
> > > > > > +               return PTR_ERR(pll6);
> > > > > > +       }
> > > > > 
> > > > > I'm slightly confused. The binding listed a single unnamed clock (the
> > > > > AHB clock). What is going on here?
> > > > 
> > > > The device itself needs only a single clock to work...
> > > > 
> > > > > 
> > > > > > +       ret = clk_set_parent(mux, pll6);
> > > > > > +       clk_put(pll6);
> > > > > > +       clk_put(mux);
> > > > > > +
> > > > > > +       if (ret) {
> > > > > > +               dev_err(&pdev->dev, "Couldn't reparent AHB1 on PLL6\n");
> > > > > > +               return ret;
> > > > > > +       }
> > > > > 
> > > > > Why do we need to reparent the mux?
> > > > 
> > > > ... but will function only if this clock is derived from PLL6.
> > > 
> > > Ok, but _why_ is that the case? Could we at least have a comment for
> > > that?
> > 
> > I have no idea, sorry.
> 
> Oh no :(
> 
> Where did this requirement turn up? Did an issue crop up in testing and
> using PLL6 just happened to work, or does some manual somewhere say that
> the device must have its clock derived from PLL6 without further
> explanation?

I first tried with the default clocks (the 24MHz oscillator) and only
transfers to devices and/or SRAM we working, but transfers from/to the
DRAM were not. Switching to PLL6 was making everything work fine. It
seems to be a bug in the A31 itself, but I don't have much details
yet.

> Does PLL6 drive some other clock that the DMA clock needs to stay in
> sync with, possibly?

PLL6 also drives some peripherals directly yes (such as the MMC or the
SPI controllers), but since I tested this using the SPI, with the main
oscillator as a source for both, I don't think it's related to
that. Or at least, not as directly related as what you were
suggesting.

> 
> > > Where does the driver get the named clocks from if they aren't provided
> > > on the device node? Is there a clock-ranges somewhere?
> > 
> > No, it just looks up the global clock name.
> 
> Ok.
> 
> > > It feels a little fragile to rely on the organisation of the clock tree
> > > and the naming thereof. If the IP block is ever reused on an SoC with a
> > > different clock tree layout then we have to handle things differently.
> > 
> > What do you suggest then?
> 
> I will admit that I don't have a better suggestion.
> 
> Without knowing which particular constraint on the mux parent clock we
> care about it's difficult to suggest anything useful.

Well, I first made it into the mach- directory, and then was told to
move it in the driver itself, so we're kind of running out of options
:)

Maxime

-- 
Maxime Ripard, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

Attachment: signature.asc
Description: Digital signature