mail archive of the barebox mailing list
 help / color / mirror / Atom feed
From: Antony Pavlov <antonynpavlov@gmail.com>
To: Lucas Stach <dev@lynxeye.de>
Cc: barebox@lists.infradead.org
Subject: Re: [PATCH] net: rtl8169: remove unnecessary cache maintenance
Date: Tue, 6 Jan 2015 13:56:11 +0300	[thread overview]
Message-ID: <20150106135611.e7d20dffcf2f94ee0ccfbdaa@gmail.com> (raw)
In-Reply-To: <1420479449.2221.3.camel@lynxeye.de>

On Mon, 05 Jan 2015 18:37:29 +0100
Lucas Stach <dev@lynxeye.de> wrote:

> Am Montag, den 05.01.2015, 20:19 +0300 schrieb Antony Pavlov:
> > On Sun,  4 Jan 2015 22:09:05 +0100
> > Lucas Stach <dev@lynxeye.de> wrote:
> > 
> > I suppose that this patch can make problems on MIPS with explicid cache handling.
> > 
> How would it?
> This driver is ARM only for now (as we are using ARM MMU functions to
> flush buffers) until someone gets around to properly implement generic
> dma sync ops.

Anyway it is reasonable to add 'depends on ARM' to 'config DRIVER_NET_RTL8169'
record in drivers/net/Kconfig.

> But even then MIPS knows about uncached memory types which is exactly
> what should be returned for a dma coherent memory allocation.

Alas! We can't just add 'volatile' modifier to make pointer uncached on MIPS in general case.

> 
> Regards,
> Lucas 
> > > The buffer descriptors are allocated from coherent memory, so there
> > > is no cache maintenance needed. Only tell the compiler that the descriptors
> > > can be modified by the hardware.
> > > 
> > > Signed-off-by: Lucas Stach <dev@lynxeye.de>
> > > ---
> > >  drivers/net/rtl8169.c | 24 +++++-------------------
> > >  1 file changed, 5 insertions(+), 19 deletions(-)
> > > 
> > > diff --git a/drivers/net/rtl8169.c b/drivers/net/rtl8169.c
> > > index 5702900..19f5763 100644
> > > --- a/drivers/net/rtl8169.c
> > > +++ b/drivers/net/rtl8169.c
> > > @@ -50,11 +50,11 @@ struct rtl8169_priv {
> > >  	struct pci_dev		*pci_dev;
> > >  	int			chipset;
> > >  
> > > -	struct bufdesc		*tx_desc;
> > > +	volatile struct bufdesc	*tx_desc;
> > >  	void			*tx_buf;
> > >  	unsigned int		cur_tx;
> > >  
> > > -	struct bufdesc		*rx_desc;
> > > +	volatile struct bufdesc	*rx_desc;
> > >  	void			*rx_buf;
> > >  	unsigned int		cur_rx;
> > >  
> > > @@ -250,10 +250,6 @@ static void rtl8169_init_ring(struct rtl8169_priv *priv)
> > >  		priv->rx_desc[i].buf_addr =
> > >  				virt_to_phys(priv->rx_buf + i * PKT_BUF_SIZE);
> > >  	}
> > > -
> > > -	dma_flush_range((unsigned long)priv->rx_desc,
> > > -			(unsigned long)priv->rx_desc +
> > > -			NUM_RX_DESC * sizeof(struct bufdesc));
> > >  }
> > >  
> > >  static void rtl8169_hw_start(struct rtl8169_priv *priv)
> > > @@ -386,14 +382,10 @@ static int rtl8169_eth_send(struct eth_device *edev, void *packet,
> > >  			((packet_length > ETH_ZLEN) ? packet_length : ETH_ZLEN);
> > >  	}
> > >  
> > > -	dma_flush_range((unsigned long)&priv->tx_desc[entry],
> > > -			(unsigned long)&priv->tx_desc[entry + 1]);
> > > -
> > >  	RTL_W8(priv, TxPoll, 0x40);
> > > -	do {
> > > -		dma_inv_range((unsigned long)&priv->tx_desc[entry],
> > > -		              (unsigned long)&priv->tx_desc[entry + 1]);
> > > -	} while (priv->tx_desc[entry].status & BD_STAT_OWN);
> > > +
> > > +	while (priv->tx_desc[entry].status & BD_STAT_OWN)
> > > +		;
> > >  
> > >  	priv->cur_tx++;
> > >  
> > > @@ -408,9 +400,6 @@ static int rtl8169_eth_rx(struct eth_device *edev)
> > >  
> > >  	entry = priv->cur_rx % NUM_RX_DESC;
> > >  
> > > -	dma_inv_range((unsigned long)&priv->rx_desc[entry],
> > > -	              (unsigned long)&priv->rx_desc[entry + 1]);
> > > -
> > >  	if ((priv->rx_desc[entry].status & BD_STAT_OWN) == 0) {
> > >  		if (!(priv->rx_desc[entry].status & BD_STAT_RX_RES)) {
> > >  			pkt_size = (priv->rx_desc[entry].status & 0x1fff) - 4;
> > > @@ -441,9 +430,6 @@ static int rtl8169_eth_rx(struct eth_device *edev)
> > >  			priv->rx_desc[entry].buf_addr =
> > >  				virt_to_phys(priv->rx_buf +
> > >  				             entry * PKT_BUF_SIZE);
> > > -
> > > -			dma_flush_range((unsigned long)&priv->rx_desc[entry],
> > > -			                (unsigned long)&priv->rx_desc[entry + 1]);
> > >  		} else {
> > >  			dev_err(&edev->dev, "rx error\n");
> > >  		}
> > > -- 
> > > 2.1.0
> > > 
> > > 
> > > _______________________________________________
> > > barebox mailing list
> > > barebox@lists.infradead.org
> > > http://lists.infradead.org/mailman/listinfo/barebox
> > 
> > 
> 
> 


-- 
-- 
Best regards,
  Antony Pavlov

_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox

  reply	other threads:[~2015-01-06 10:54 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2015-01-04 21:09 Lucas Stach
2015-01-05 11:52 ` Sascha Hauer
2015-01-05 17:19 ` Antony Pavlov
2015-01-05 17:37   ` Lucas Stach
2015-01-06 10:56     ` Antony Pavlov [this message]
2015-01-06 11:11       ` Lucas Stach

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20150106135611.e7d20dffcf2f94ee0ccfbdaa@gmail.com \
    --to=antonynpavlov@gmail.com \
    --cc=barebox@lists.infradead.org \
    --cc=dev@lynxeye.de \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox