Re: [PATCH v6 net-next 6/7] net: ethernet: ti: am65-cpsw-qos: Add Frame Preemption MAC Merge support

[Roger Quadros]

On 21/11/2023 01:26, Vladimir Oltean wrote:
> On Mon, Nov 20, 2023 at 04:01:46PM +0200, Roger Quadros wrote:
>> Changelog:
>>
>> v6:
>> - get mutex around am65_cpsw_iet_commit_preemptible_tcs() in
>>   am65_cpsw_iet_change_preemptible_tcs()
>> - use "preemption" instead of "pre-emption"
>> - call am65_cpsw_setup_mqprio() from within am65_cpsw_setup_taprio()
>> - Now works with kselftest except the last test which fails
>>

<snip>

>> diff --git a/drivers/net/ethernet/ti/am65-cpsw-qos.c b/drivers/net/ethernet/ti/am65-cpsw-qos.c
>> index 60f625e1350b..8443c13c9ff6 100644
>> --- a/drivers/net/ethernet/ti/am65-cpsw-qos.c
>> +++ b/drivers/net/ethernet/ti/am65-cpsw-qos.c
>> @@ -4,9 +4,11 @@
>>   *
>>   * quality of service module includes:
>>   * Enhanced Scheduler Traffic (EST - P802.1Qbv/D2.2)
>> + * Interspersed Express Traffic (IET - P802.3br/D2.0)
>>   */
>>  
>>  #include <linux/pm_runtime.h>
>> +#include <linux/units.h>
> 
> If this was ordered alphabetically, u comes after t.
> 
>>  #include <linux/time.h>
>>  #include <net/pkt_cls.h>
>>  
>> @@ -259,6 +266,196 @@ static int am65_cpsw_setup_mqprio(struct net_device *ndev, void *type_data)
>>  	return ret;
>>  }
>>  
>> +static int am65_cpsw_iet_set_verify_timeout_count(struct am65_cpsw_port *port)
>> +{
>> +	int verify_time_ms = port->qos.iet.verify_time_ms;
>> +	int link_speed = port->qos.link_speed;
>> +	u32 val;
>> +
>> +	if (WARN_ON(link_speed == SPEED_UNKNOWN))
>> +		return -ENODEV;
>> +
>> +	/* The number of wireside clocks contained in the verify
>> +	 * timeout counter. The default is 0x1312d0
>> +	 * (10ms at 125Mhz in 1G mode).
>> +	 */
>> +	val = 125 * HZ_PER_MHZ;	/* assuming 125MHz wireside clock */
>> +
>> +	val /= MILLIHZ_PER_HZ;		/* count per ms timeout */
>> +	val *= verify_time_ms;		/* count for timeout ms */
>> +	if (link_speed < SPEED_1000)
>> +		val <<= 1;	/* FIXME: Is this correct? */
> 
> Is there any way to find out and remove the FIXME? Like running the
> selftest at SPEED_100?

I checked at SPEED_100 without the above if condition and it works so I'll
drop it in next revision.

> 
>> +
>> +	if (val > AM65_CPSW_PN_MAC_VERIFY_CNT_MASK)
>> +		return -EINVAL;
>> +
>> +	writel(val, port->port_base + AM65_CPSW_PN_REG_IET_VERIFY);
>> +
>> +	return 0;
>> +}
>> +

-- 
cheers,
-roger