Re: [PATCH] net: ethernet: mediatek: ppe: add support for flow accounting

[Daniel Golle <daniel@makrotopia.org>]

Hi Felix,

On Tue, Nov 01, 2022 at 08:38:59PM +0100, Felix Fietkau wrote:
> On 01.11.22 20:12, Daniel Golle wrote:
> > The PPE units found in MT7622 and newer support packet and byte
> > accounting of hw-offloaded flows. Add support for reading those
> > counters as found in MediaTek's SDK[1].
> > 
> > [1]: https://git01.mediatek.com/plugins/gitiles/openwrt/feeds/mtk-openwrt-feeds/+/bc6a6a375c800dc2b80e1a325a2c732d1737df92
> > Signed-off-by: Daniel Golle <daniel@makrotopia.org>
> > ---
> >   drivers/net/ethernet/mediatek/mtk_eth_soc.c   |  11 +-
> >   drivers/net/ethernet/mediatek/mtk_eth_soc.h   |   1 +
> >   drivers/net/ethernet/mediatek/mtk_ppe.c       | 122 +++++++++++++++++-
> >   drivers/net/ethernet/mediatek/mtk_ppe.h       |  23 +++-
> >   .../net/ethernet/mediatek/mtk_ppe_debugfs.c   |   9 +-
> >   .../net/ethernet/mediatek/mtk_ppe_offload.c   |   7 +
> >   drivers/net/ethernet/mediatek/mtk_ppe_regs.h  |  14 ++
> >   7 files changed, 178 insertions(+), 9 deletions(-)
> > 
> > diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.c b/drivers/net/ethernet/mediatek/mtk_eth_soc.c
> > index 789268b15106ec..5fcd66fca7a089 100644
> > --- a/drivers/net/ethernet/mediatek/mtk_eth_soc.c
> > +++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.c
> > @@ -4221,6 +4223,7 @@ static const struct mtk_soc_data mt2701_data = {
> >   	.hw_features = MTK_HW_FEATURES,
> >   	.required_clks = MT7623_CLKS_BITMAP,
> >   	.required_pctl = true,
> > +	.has_accounting = false,
> >   	.txrx = {
> >   		.txd_size = sizeof(struct mtk_tx_dma),
> >   		.rxd_size = sizeof(struct mtk_rx_dma),
> > @@ -4239,6 +4242,7 @@ static const struct mtk_soc_data mt7621_data = {
> >   	.required_pctl = false,
> >   	.offload_version = 2,
> >   	.hash_offset = 2,
> > +	.has_accounting = false,
> >   	.foe_entry_size = sizeof(struct mtk_foe_entry) - 16,
> >   	.txrx = {
> >   		.txd_size = sizeof(struct mtk_tx_dma),
> > @@ -4259,6 +4263,7 @@ static const struct mtk_soc_data mt7622_data = {
> >   	.required_pctl = false,
> >   	.offload_version = 2,
> >   	.hash_offset = 2,
> > +	.has_accounting = true,
> >   	.foe_entry_size = sizeof(struct mtk_foe_entry) - 16,
> >   	.txrx = {
> >   		.txd_size = sizeof(struct mtk_tx_dma),
> > @@ -4278,6 +4283,7 @@ static const struct mtk_soc_data mt7623_data = {
> >   	.required_pctl = true,
> >   	.offload_version = 2,
> >   	.hash_offset = 2,
> > +	.has_accounting = false,
> >   	.foe_entry_size = sizeof(struct mtk_foe_entry) - 16,
> >   	.txrx = {
> >   		.txd_size = sizeof(struct mtk_tx_dma),
> > @@ -4296,6 +4302,7 @@ static const struct mtk_soc_data mt7629_data = {
> >   	.hw_features = MTK_HW_FEATURES,
> >   	.required_clks = MT7629_CLKS_BITMAP,
> >   	.required_pctl = false,
> > +	.has_accounting = true,
> >   	.txrx = {
> >   		.txd_size = sizeof(struct mtk_tx_dma),
> >   		.rxd_size = sizeof(struct mtk_rx_dma),
> > @@ -4315,6 +4322,7 @@ static const struct mtk_soc_data mt7986_data = {
> >   	.required_pctl = false,
> >   	.hash_offset = 4,
> >   	.foe_entry_size = sizeof(struct mtk_foe_entry),
> > +	.has_accounting = true,
> >   	.txrx = {
> >   		.txd_size = sizeof(struct mtk_tx_dma_v2),
> >   		.rxd_size = sizeof(struct mtk_rx_dma_v2),
> > @@ -4331,6 +4339,7 @@ static const struct mtk_soc_data rt5350_data = {
> >   	.hw_features = MTK_HW_FEATURES_MT7628,
> >   	.required_clks = MT7628_CLKS_BITMAP,
> >   	.required_pctl = false,
> > +	.has_accounting = false,
> >   	.txrx = {
> >   		.txd_size = sizeof(struct mtk_tx_dma),
> >   		.rxd_size = sizeof(struct mtk_rx_dma),
> You can leave out the .has_accounting = false assignments.
Ack.

> 
> > diff --git a/drivers/net/ethernet/mediatek/mtk_ppe.c b/drivers/net/ethernet/mediatek/mtk_ppe.c
> > index 2d8ca99f2467ff..bc4660da28451b 100644
> > --- a/drivers/net/ethernet/mediatek/mtk_ppe.c
> > +++ b/drivers/net/ethernet/mediatek/mtk_ppe.c
> > @@ -545,6 +592,16 @@ __mtk_foe_entry_commit(struct mtk_ppe *ppe, struct mtk_foe_entry *entry,
> >   	wmb();
> >   	hwe->ib1 = entry->ib1;
> > +	if (ppe->accounting) {
> > +		int type;
> > +
> > +		type = FIELD_GET(MTK_FOE_IB1_PACKET_TYPE, entry->ib1);
> > +		if (type >= MTK_PPE_PKT_TYPE_IPV4_DSLITE)
> > +			hwe->ipv6.ib2 |= MTK_FOE_IB2_MIB_CNT;
> > +		else
> > +			hwe->ipv4.ib2 |= MTK_FOE_IB2_MIB_CNT;
> > +	}
> Use mtk_foe_entry_ib2() here.

Ack, turns it into a one-liner ;)

> 
> > @@ -654,11 +711,8 @@ void __mtk_ppe_check_skb(struct mtk_ppe *ppe, struct sk_buff *skb, u16 hash)
> >   			continue;
> >   		}
> > -		if (found || !mtk_flow_entry_match(ppe->eth, entry, hwe)) {
> > -			if (entry->hash != 0xffff)
> > -				entry->hash = 0xffff;
> > +		if (found || !mtk_flow_entry_match(ppe->eth, entry, hwe))
> >   			continue;
> > -		}
> What is the reason for this change?

This has slipped in accidentally as it was part of the patch in
mtk-openwrt-feeds. I suppose it may be related to the rather mystical
2nd part of the commit message there:
Patches-4 fix __schedule_bug issue in the __mtk_foe_entry_clear().

However, as things seem to work just fine also without that I will send
v3 skipping this part.


Thanks for the review!


Daniel