Fwd: [RFC, PATCH 1/3] ARM: PL08X: Add PL08x AMBA DMA Controller driver.

[Linus Walleij]

Oops this was onto the old list address, Peter is busy so does anybody
else know about
address boundaries in the RealViews?

Linus Walleij


---------- Forwarded message ----------
From: Linus Walleij <linus.ml.walleij at gmail.com>
Date: 2010/5/30
Subject: Re: [RFC, PATCH 1/3] ARM: PL08X: Add PL08x AMBA DMA Controller driver.
To: Peter Pearse <peter.pearse at arm.com>
Kopia: linux-arm-kernel <linux-arm-kernel at lists.arm.linux.org.uk>


Hi,

as you know I'm hacking around with this PL08x driver since the Versatile
support is in the way for getting my DMA work merged.

Now this piece I don't quite get:

2008/11/21 Peter Pearse <protected at arm.com>:


> +/*
> + * Return number of bytes to fill to boundary, or len
> + */
> +static int pl08x_pre_boundary(int addr, int len)
> +{
> +       int boundary;
> +
> +       if (len <= 0)
> +               printk(KERN_ERR "PL08X::pre_boundary() - zero length\n");
> +
> +       boundary = ((addr >> PL08X_BOUNDARY_SHIFT) + 1) << PL08X_BOUNDARY_SHIFT;
> +
> +       if (boundary < addr + len)
> +               return boundary - addr;
> +       else
> +               return len;
> +}

This PL08X_BOUNDARY_SHIFT is defined to
#define PL08X_BOUNDARY_SHIFT           (10)    /* 1KB 0x400 */
#define PL08X_BOUNDARY_SIZE            (1 << PL08X_BOUNDARY_SHIFT)

And then later:

> +/*
> + * Note that we assume we never have to change the burst sizes
> + */
> +int fill_LLIS_for_desc(struct pl08x_txd *local_txd, int pl08x_chan_num)
> +{
> (...)
> +               /*
> +                * Make largest possible LLIs until less than one bus width left
> +                */
> +               while (remainder > (mbus->buswidth - 1)) {
> +                       int lli_len, target_len;
> +                       int tsize;
> +                       int odd_bytes;
> +                       /*
> +                        * If enough left try to send max possible,
> +                        * otherwise try to send the remainder
> +                        */
> +                       target_len = remainder;
> +                       if (remainder > max_bytes_per_lli)
> +                               target_len = max_bytes_per_lli;
> +                       /*
> +                        *  Set bus lengths for incrementing busses
> +                        *  to number of bytes which fill
> +                        *  to next memory boundary
> +                        */
> +                       if (cctl_parm.bits.si)
> +                               local_txd->srcbus.fill_bytes
> +                               = pl08x_pre_boundary(local_txd->srcbus.addr
> +                                                               , remainder);
> +                       else
> +                               local_txd->srcbus.fill_bytes
> +                               = max_bytes_per_lli;

Here the transfer lengths for lli:s crawls up tp a boundary.
Now, the DMAC PL08x cannot move past a 0x400 boundary on the RealViews,
so much is clear. But: is this a thing that will affect:

- All RealView / Versatile equipment
- All implementations of PL08x on all silicon
- All ARM architectures and nothing to do with the PL08x at all, rather
 with some AMBA stuff or so?

Yours,
Linus Walleij