Re: [PATCH net-next v3 2/2] net: ethernet: ti: am65-cpsw: enable DSCP to priority map for RX

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

 




On 14/11/2024 02:16, Guillaume Nault wrote:
> On Sat, Nov 09, 2024 at 01:00:08PM +0200, Roger Quadros wrote:
>> AM65 CPSW hardware can map the 6-bit DSCP/TOS field to
>> appropriate priority queue via DSCP to Priority mapping registers
>> (CPSW_PN_RX_PRI_MAP_REG).
>>
>> We use the upper 3 bits of the DSCP field that indicate IP Precedence
>> to map traffic to 8 priority queues.
>>
>> Signed-off-by: Roger Quadros <rogerq@xxxxxxxxxx>
>> ---
>>  drivers/net/ethernet/ti/am65-cpsw-nuss.c | 54 ++++++++++++++++++++++++++++++++
>>  1 file changed, 54 insertions(+)
>>
>> diff --git a/drivers/net/ethernet/ti/am65-cpsw-nuss.c b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
>> index 0520e9f4bea7..fab35e6aac7f 100644
>> --- a/drivers/net/ethernet/ti/am65-cpsw-nuss.c
>> +++ b/drivers/net/ethernet/ti/am65-cpsw-nuss.c
>> @@ -71,6 +71,8 @@
>>  #define AM65_CPSW_PORT_REG_RX_PRI_MAP		0x020
>>  #define AM65_CPSW_PORT_REG_RX_MAXLEN		0x024
>>  
>> +#define AM65_CPSW_PORTN_REG_CTL			0x004
>> +#define AM65_CPSW_PORTN_REG_DSCP_MAP		0x120
>>  #define AM65_CPSW_PORTN_REG_SA_L		0x308
>>  #define AM65_CPSW_PORTN_REG_SA_H		0x30c
>>  #define AM65_CPSW_PORTN_REG_TS_CTL              0x310
>> @@ -94,6 +96,10 @@
>>  /* AM65_CPSW_PORT_REG_PRI_CTL */
>>  #define AM65_CPSW_PORT_REG_PRI_CTL_RX_PTYPE_RROBIN	BIT(8)
>>  
>> +/* AM65_CPSW_PN_REG_CTL */
>> +#define AM65_CPSW_PN_REG_CTL_DSCP_IPV4_EN	BIT(1)
>> +#define AM65_CPSW_PN_REG_CTL_DSCP_IPV6_EN	BIT(2)
>> +
>>  /* AM65_CPSW_PN_TS_CTL register fields */
>>  #define AM65_CPSW_PN_TS_CTL_TX_ANX_F_EN		BIT(4)
>>  #define AM65_CPSW_PN_TS_CTL_TX_VLAN_LT1_EN	BIT(5)
>> @@ -176,6 +182,53 @@ static void am65_cpsw_port_set_sl_mac(struct am65_cpsw_port *slave,
>>  	writel(mac_lo, slave->port_base + AM65_CPSW_PORTN_REG_SA_L);
>>  }
>>  
>> +#define AM65_CPSW_DSCP_MAX	GENMASK(5, 0)
>> +#define AM65_CPSW_PRI_MAX	GENMASK(2, 0)
>> +#define AM65_CPSW_DSCP_PRI_PER_REG	8
>> +#define AM65_CPSW_DSCP_PRI_SIZE		4	/* in bits */
>> +static int am65_cpsw_port_set_dscp_map(struct am65_cpsw_port *slave, u8 dscp, u8 pri)
>> +{
>> +	int reg_ofs;
>> +	int bit_ofs;
>> +	u32 val;
>> +
>> +	if (dscp > AM65_CPSW_DSCP_MAX)
>> +		return -EINVAL;
>> +
>> +	if (pri > AM65_CPSW_PRI_MAX)
>> +		return -EINVAL;
>> +
>> +	/* 32-bit register offset to this dscp */
>> +	reg_ofs = (dscp / AM65_CPSW_DSCP_PRI_PER_REG) * 4;
>> +	/* bit field offset to this dscp */
>> +	bit_ofs = AM65_CPSW_DSCP_PRI_SIZE * (dscp % AM65_CPSW_DSCP_PRI_PER_REG);
>> +
>> +	val = readl(slave->port_base + AM65_CPSW_PORTN_REG_DSCP_MAP + reg_ofs);
>> +	val &= ~(AM65_CPSW_PRI_MAX << bit_ofs);	/* clear */
>> +	val |= pri << bit_ofs;			/* set */
>> +	writel(val, slave->port_base + AM65_CPSW_PORTN_REG_DSCP_MAP + reg_ofs);
>> +
>> +	return 0;
>> +}
>> +
>> +static void am65_cpsw_port_enable_dscp_map(struct am65_cpsw_port *slave)
>> +{
>> +	int dscp, pri;
>> +	u32 val;
>> +
>> +	/* Map IP Precedence field to Priority */
>> +	for (dscp = 0; dscp <= AM65_CPSW_DSCP_MAX; dscp++) {
>> +		pri = dscp >> 3; /* Extract IP Precedence */
>> +		am65_cpsw_port_set_dscp_map(slave, dscp, pri);
>> +	}
>> +
>> +	/* enable port IPV4 and IPV6 DSCP for this port */
>> +	val = readl(slave->port_base + AM65_CPSW_PORTN_REG_CTL);
>> +	val |= AM65_CPSW_PN_REG_CTL_DSCP_IPV4_EN |
>> +		AM65_CPSW_PN_REG_CTL_DSCP_IPV6_EN;
>> +	writel(val, slave->port_base + AM65_CPSW_PORTN_REG_CTL);
>> +}
> 
> It seems that this hardware is capable of mapping all possible DSCP
yes.

> values. Then why restricting the mapping to the 3 high order bits only?

Currently, the 64 DSCP values are mapped to 8 User Priorities (UP) based
on just the Class Selector Codepoint field (first 3 bits of DSCP).

But now looking at rfc8325#section-4.3.
"Note: All unused codepoints are RECOMMENDED to be mapped to UP 0"

So what this patch does doesn't look like a good idea.

> According to RFC 8325 section 2.3, this seem to be a common practice,
> which this RFC considers a problem:
> https://datatracker.ietf.org/doc/html/rfc8325#section-2.3

Good to know about this.

> 
> I know this RFC is about 802.11, not 802.1p, but as far as I know, the
> user priority (UP) are the same for both, so that shouldn't make a
> difference.
> 
> So what about following the IETF mapping found in section 4.3?
> https://datatracker.ietf.org/doc/html/rfc8325#section-4.3

Thanks for this tip.
I will update this patch to have the default DSCP to UP mapping as per
above link and map all unused DSCP to UP 0.

Is there any mechanism/API for network administrator to change this
default mapping in the network drivers?

> 
>>  static void am65_cpsw_sl_ctl_reset(struct am65_cpsw_port *port)
>>  {
>>  	cpsw_sl_reset(port->slave.mac_sl, 100);
>> @@ -921,6 +974,7 @@ static int am65_cpsw_nuss_ndo_slave_open(struct net_device *ndev)
>>  	common->usage_count++;
>>  
>>  	am65_cpsw_port_set_sl_mac(port, ndev->dev_addr);
>> +	am65_cpsw_port_enable_dscp_map(port);
>>  
>>  	if (common->is_emac_mode)
>>  		am65_cpsw_init_port_emac_ale(port);
>>
>> -- 
>> 2.34.1
>>
>>
> 

-- 
cheers,
-roger





[Index of Archives]     [Linux Arm (vger)]     [ARM Kernel]     [ARM MSM]     [Linux Tegra]     [Linux WPAN Networking]     [Linux Wireless Networking]     [Maemo Users]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite Trails]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux