[net v2] net: ethernet: mtk_eth_soc: handle dma buffer size soc specific

[Daniel Golle]

On Mon, May 27, 2024 at 03:55:55PM GMT, Sunil Kovvuri Goutham wrote:
> 
> 
> > -----Original Message-----
> > From: Frank Wunderlich <linux at fw-web.de>
> > Sent: Monday, May 27, 2024 7:52 PM
> > To: Felix Fietkau <nbd at nbd.name>; Sean Wang <sean.wang at mediatek.com>;
> > Mark Lee <Mark-MC.Lee at mediatek.com>; Lorenzo Bianconi
> > <lorenzo at kernel.org>; David S. Miller <davem at davemloft.net>; Eric Dumazet
> > <edumazet at google.com>; Jakub Kicinski <kuba at kernel.org>; Paolo Abeni
> > <pabeni at redhat.com>; Matthias Brugger <matthias.bgg at gmail.com>;
> > AngeloGioacchino Del Regno <angelogioacchino.delregno at collabora.com>
> > Cc: Frank Wunderlich <frank-w at public-files.de>; John Crispin
> > <john at phrozen.org>; netdev at vger.kernel.org; linux-kernel at vger.kernel.org;
> > linux-arm-kernel at lists.infradead.org; linux-mediatek at lists.infradead.org;
> > Daniel Golle <daniel at makrotopia.org>
> > Subject: [net v2] net: ethernet: mtk_eth_soc: handle dma buffer size soc specific
> > 
> > From: Frank Wunderlich <frank-w at public-files.de>
> > 
> > The mainline MTK ethernet driver suffers long time from rarly but annoying tx
> > queue timeouts. We think that this is caused by fixed dma sizes hardcoded for
> > all SoCs.
> > 
> > Use the dma-size implementation from SDK in a per SoC manner.
> > 
> > Fixes: 656e705243fd ("net-next: mediatek: add support for MT7623
> > ethernet")
> > Suggested-by: Daniel Golle <daniel at makrotopia.org>
> > Signed-off-by: Frank Wunderlich <frank-w at public-files.de>
> 
> ..............
> > 
> > diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.c
> > b/drivers/net/ethernet/mediatek/mtk_eth_soc.c
> > index cae46290a7ae..f1ff1be73926 100644
> > --- a/drivers/net/ethernet/mediatek/mtk_eth_soc.c
> > +++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.c
> 
> .............
> > @@ -1142,40 +1142,46 @@ static int mtk_init_fq_dma(struct mtk_eth *eth)
> >  						       cnt * soc->tx.desc_size,
> >  						       &eth->phy_scratch_ring,
> >  						       GFP_KERNEL);
> 
> ..............
> > -	for (i = 0; i < cnt; i++) {
> > -		dma_addr_t addr = dma_addr + i * MTK_QDMA_PAGE_SIZE;
> > -		struct mtk_tx_dma_v2 *txd;
> > +		dma_addr = dma_map_single(eth->dma_dev,
> > +					  eth->scratch_head[j], len *
> > MTK_QDMA_PAGE_SIZE,
> > +					  DMA_FROM_DEVICE);
> > 
> 
> As per commit msg, the fix is for transmit queue timeouts.
> But the DMA buffer changes seems for receive pkts.
> Can you please elaborate the connection here.

*I guess* the memory window used for both, TX and RX DMA descriptors
needs to be wisely split to not risk TX queue overruns, depending on the
SoC speed and without hurting RX performance...

Maybe someone inside MediaTek (I've added to Cc now) and more familiar
with the design can elaborate in more detail.