RE: [PATCH 07/10] spi: rzv2m-csi: Switch to using {read,write}s{b,w}

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi Geert,

Thanks for your reply!

> From: Geert Uytterhoeven <geert@xxxxxxxxxxxxxx>
> Subject: Re: [PATCH 07/10] spi: rzv2m-csi: Switch to using
> {read,write}s{b,w}
> 
> Hi Fabrizio,
> 
> On Sat, Jul 15, 2023 at 3:04 AM Fabrizio Castro
> <fabrizio.castro.jz@xxxxxxxxxxx> wrote:
> > The RX/TX FIFOs implemented by the CSI IP are accessed by
> > repeatedly reading/writing the same memory address, and
> > therefore they are the ideal candidate for {read,write}s{b,w}.
> > The RZ/V2M CSI driver currently implements loops to fill up
> > the TX FIFO and empty the RX FIFO, differentiating between
> > 8-bit and 16-bit word size.
> > Switch to using {read,write}s{b,w} to get rid of the bespoke
> > loops.
> >
> > Signed-off-by: Fabrizio Castro <fabrizio.castro.jz@xxxxxxxxxxx>
> 
> Thanks for your patch!
> 
> > --- a/drivers/spi/spi-rzv2m-csi.c
> > +++ b/drivers/spi/spi-rzv2m-csi.c
> 
> > @@ -157,22 +157,15 @@ static int
> rzv2m_csi_start_stop_operation(const struct rzv2m_csi_priv *csi,
> >
> >  static int rzv2m_csi_fill_txfifo(struct rzv2m_csi_priv *csi)
> >  {
> > -       int i;
> > -
> >         if (readl(csi->base + CSI_OFIFOL))
> >                 return -EIO;
> >
> > -       if (csi->bytes_per_word == 2) {
> > -               u16 *buf = (u16 *)csi->txbuf;
> > -
> > -               for (i = 0; i < csi->words_to_transfer; i++)
> > -                       writel(buf[i], csi->base + CSI_OFIFO);
> > -       } else {
> > -               u8 *buf = (u8 *)csi->txbuf;
> > -
> > -               for (i = 0; i < csi->words_to_transfer; i++)
> > -                       writel(buf[i], csi->base + CSI_OFIFO);
> > -       }
> > +       if (csi->bytes_per_word == 2)
> > +               writesw(csi->base + CSI_OFIFO, csi->txbuf,
> > +                       csi->words_to_transfer);
> > +       else
> > +               writesb(csi->base + CSI_OFIFO, csi->txbuf,
> > +                       csi->words_to_transfer);
> 
> According to the hardware documentation[1], the access size for both
> the
> CSI_OFIFO and CSI_IFIFO registers is 32 bits, so you must use writel()
> resp. readl().  So please check with the hardware people first.
> 
> [1] RZ/V2M User's Manual Hardware, Rev. 1.30.

You are right, access is 32 bits (and although this patch works fine,
we should avoid accessing those regs any other way). Now I remember
why I decided to go for the bespoke loops in the first place, writesl
and readsl provide the right register access, but the wrong pointer
arithmetic for this use case.
For this patch I ended up selecting writesw/writesb/readsw/readsb to
get the right pointer arithmetic, but the register access is not as
per manual.

I can either fallback to using the bespoke loops (I can still
remove the unnecessary u8* and u16* casting ;-) ), or I can add
new APIs for this sort of access to io.h (e.g. writesbl, writeswl,
readsbl, readswl, in order to get the pointer arithmetic right for
the type of array handled, while keeping memory access as expected).

What are your thoughts on that?

Thanks,
Fab

> 
> Gr{oetje,eeting}s,
> 
>                         Geert
> 
> --
> Geert Uytterhoeven -- There's lots of Linux beyond ia32 --
> geert@xxxxxxxxxxxxxx
> 
> In personal conversations with technical people, I call myself a
> hacker. But
> when I'm talking to journalists I just say "programmer" or something
> like that.
>                                 -- Linus Torvalds




[Index of Archives]     [Linux Kernel]     [Linux ARM (vger)]     [Linux ARM MSM]     [Linux Omap]     [Linux Arm]     [Linux Tegra]     [Fedora ARM]     [Linux for Samsung SOC]     [eCos]     [Linux Fastboot]     [Gcc Help]     [Git]     [DCCP]     [IETF Announce]     [Security]     [Linux MIPS]     [Yosemite Campsites]

  Powered by Linux