[U-Boot] [PATCH v4] zynq-gem: Use appropriate cache flush/invalidate for RX and TX

[Bin Meng]

Hi Stefan,

On Mon, Dec 17, 2018 at 3:57 PM Stefan Theil <Stefan.Theil at mixed-mode.de> wrote:
>
> > -----Ursprüngliche Nachricht-----
> > Von: Bin Meng [mailto:bmeng.cn at gmail.com]
> > Gesendet: Montag, 17. Dezember 2018 08:52
> > An: Stefan Theil
> > Cc: U-Boot Mailing List; Michal Simek
> > Betreff: Re: [PATCH v4] zynq-gem: Use appropriate cache flush/invalidate for
> > RX and TX
> >
> > Hi Stefan,
> >
> > On Mon, Dec 17, 2018 at 3:49 PM Stefan Theil <stefan.theil at mixed-
> > mode.de> wrote:
> > >
> > > The cache was only flushed before *transmitting* packets, but not when
> > > receiving them, leading to an issue where new packets were handed to
> > > the receive handler with old contents in cache. This only happens when
> > > a lot of packets are received without sending packages every now and
> > > then. Also flushing the receive buffers in the transmit function makes
> > > no sense and can be removed.
> > >
> > > Signed-off-by: Stefan Theil <stefan.theil at mixed-mode.de>
> > >
> > > ---
> > > Changes for v2:
> > >         - Use invalidate_dcache_range instead of
> > >           flush_dcache_range
> > > Changes for v3:
> > >         - Remove unnecessary flushing of all RX
> > >           buffers in zynq_gem_send
> > > Changes for v4:
> > >         - Invalidate receive buffers after allocating
> > >           them in zynq_gem_probe
> > > ---
> > >  drivers/net/zynq_gem.c | 12 ++++++------
> > >  1 file changed, 6 insertions(+), 6 deletions(-)
> > >
> > > diff --git a/drivers/net/zynq_gem.c b/drivers/net/zynq_gem.c index
> > > 9bd79b198a..79a22fb1ed 100644
> > > --- a/drivers/net/zynq_gem.c
> > > +++ b/drivers/net/zynq_gem.c
> > > @@ -570,11 +570,6 @@ static int zynq_gem_send(struct udevice *dev,
> > void *ptr, int len)
> > >         addr &= ~(ARCH_DMA_MINALIGN - 1);
> > >         size = roundup(len, ARCH_DMA_MINALIGN);
> > >         flush_dcache_range(addr, addr + size);
> > > -
> > > -       addr = (ulong)priv->rxbuffers;
> > > -       addr &= ~(ARCH_DMA_MINALIGN - 1);
> > > -       size = roundup((RX_BUF * PKTSIZE_ALIGN), ARCH_DMA_MINALIGN);
> > > -       flush_dcache_range(addr, addr + size);
> > >         barrier();
> > >
> > >         /* Start transmit */
> > > @@ -621,6 +616,9 @@ static int zynq_gem_recv(struct udevice *dev, int
> > > flags, uchar **packetp)
> > >
> > >         *packetp = (uchar *)(uintptr_t)addr;
> > >
> > > +       invalidate_dcache_range(addr, addr + roundup(PKTSIZE_ALIGN,
> > ARCH_DMA_MINALIGN));
> > > +       barrier();
> > > +
> > >         return frame_len;
> > >  }
> > >
> > > @@ -705,7 +703,9 @@ static int zynq_gem_probe(struct udevice *dev)
> > >         if (!priv->rxbuffers)
> > >                 return -ENOMEM;
> > >
> > > -       memset(priv->rxbuffers, 0, RX_BUF * PKTSIZE_ALIGN);
> > > +       u32 addr = (ulong)priv->rxbuffers;
> > > +       invalidate_dcache_range(addr, addr + roundup(RX_BUF *
> > PKTSIZE_ALIGN, ARCH_DMA_MINALIGN));
> > > +       barrier();
> > >
> >
> > Does this fix anything? I see no need to update this.
>
> I was just following Michal's suggestion (https://lists.denx.de/pipermail/u-boot/2018-December/351969.html):
> > Also in probe there should be flush of priv->rxbuffers to make sure that
> > it is initialized properly.
> > (memset(priv->rxbuffers, 0, RX_BUF * PKTSIZE_ALIGN); - this should be
> > useless)
>

Michal was suggesting 'flush', not 'invalidate', to make sure
priv->rxbuffers is really initialized to zero by memset().

> >
> > >         /* Align bd_space to MMU_SECTION_SHIFT */
> > >         bd_space = memalign(1 << MMU_SECTION_SHIFT, BD_SPACE);
> > > --
> >

Regards,
Bin