Re: [PATCH 6/6] drivers/nfb: add support for more MAC addresses

[Ferruh Yigit <ferruh.yigit@intel.com>]

On 2/14/2022 4:53 PM, Martin Spinler wrote:
> On Mon, 2022-02-14 at 13:37 +0000, Ferruh Yigit wrote:
>> On 2/14/2022 11:25 AM, spinler@cesnet.cz wrote:
>>> From: Martin Spinler <spinler@cesnet.cz>
>>>
>>> Extend the eth_dev_ops by add/remove MAC address functions.
>>>
>>> Signed-off-by: Martin Spinler <spinler@cesnet.cz>
>>> ---
>>>    drivers/net/nfb/nfb_ethdev.c | 69 ++++++++++++++++++++++++++++++------
>>>    1 file changed, 58 insertions(+), 11 deletions(-)
>>>
>>> diff --git a/drivers/net/nfb/nfb_ethdev.c b/drivers/net/nfb/nfb_ethdev.c
>>> index 5d503e131a..7dec8e022e 100644
>>> --- a/drivers/net/nfb/nfb_ethdev.c
>>> +++ b/drivers/net/nfb/nfb_ethdev.c
>>> @@ -214,7 +214,19 @@ static int
>>>    nfb_eth_dev_info(struct rte_eth_dev *dev,
>>>    	struct rte_eth_dev_info *dev_info)
>>>    {
>>> -	dev_info->max_mac_addrs = 1;
>>> +	uint16_t i;
>>> +	uint32_t max_mac_addrs;
>>> +	struct pmd_internals *internals = dev->data->dev_private;
>>> +
>>> +	dev_info->max_mac_addrs = (uint32_t)-1;
>>> +	for (i = 0; i < internals->max_rxmac; i++) {
>>> +		max_mac_addrs = nc_rxmac_mac_address_count(internals->rxmac[i]);
>>> +		dev_info->max_mac_addrs = RTE_MIN(max_mac_addrs,
>>> +				dev_info->max_mac_addrs);
>>> +	}
>>> +	if (internals->max_rxmac == 0)
>>> +		dev_info->max_mac_addrs = 0;
>>> +
>>
>> Not sure if 'max_mac_addrs = 0' is valid value, as far as I can see
>> driver allocates memory for at least one MAC address, so there is no
>> case that NIC doesn't support any MAC address.
> 
> Oh, sorry, I missed that, will be fixed in v2.
> 
>>
>> It looks like max_mac_addrs usage is not clear, what is exactly the
>> 'max_rxmac' & 'rxmac[]' variables are?
> 
> rxmac is a firmware unit which represents the RX side of one physical
> Ethernet port (more precisely one logical Eth channel on port) and
> handles stats+CRC check+MAC filtering. So this goes through all those
> units and finds the rxmac with minimal memory space for MAC addresses
> and uses this value.
> 

Got it.

> I'll add few comments in the code to v2.
> 

Thanks.

>>
>>>    	dev_info->max_rx_pktlen = (uint32_t)-1;
>>>    	dev_info->max_rx_queues = dev->data->nb_rx_queues;
>>>    	dev_info->max_tx_queues = dev->data->nb_tx_queues;
>>> @@ -376,6 +388,18 @@ nfb_eth_dev_set_link_down(struct rte_eth_dev *dev)
>>>    	return 0;
>>>    }
>>>    
>>> +static uint64_t
>>> +nfb_eth_mac_addr_conv(struct rte_ether_addr *mac_addr)
>>> +{
>>> +	int i;
>>> +	uint64_t res = 0;
>>> +	for (i = 0; i < RTE_ETHER_ADDR_LEN; i++) {
>>> +		res <<= 8;
>>> +		res |= mac_addr->addr_bytes[i] & 0xFF;
>>> +	}
>>> +	return res;
>>> +}
>>> +
>>>    /**
>>>     * DPDK callback to set primary MAC address.
>>>     *
>>> @@ -392,26 +416,47 @@ nfb_eth_mac_addr_set(struct rte_eth_dev *dev,
>>>    	struct rte_ether_addr *mac_addr)
>>>    {
>>>    	unsigned int i;
>>> -	uint64_t mac = 0;
>>> +	uint64_t mac;
>>>    	struct rte_eth_dev_data *data = dev->data;
>>>    	struct pmd_internals *internals = (struct pmd_internals *)
>>>    		data->dev_private;
>>>    
>>> -	if (!rte_is_valid_assigned_ether_addr(mac_addr))
>>> -		return -EINVAL;
>>> +	mac = nfb_eth_mac_addr_conv(mac_addr);
>>> +	for (i = 0; i < internals->max_rxmac; ++i)
>>> +		nc_rxmac_set_mac(internals->rxmac[i], 0, mac, 1);
>>
>> This functions is to set default MAC address 'data->mac_addrs[0]',
>> why same MAC set multiple times in a loop?
> 
> Unfortunately, currently we don't have firmware support for proper
> separate port configuration, because DMA queues aren't bound to
> specific rxmac and traffic can be mixed between them. So I think the
> best way in this case is to write the same MAC address into all of the
> rxmacs inside firmware.
> 

I see the usage now, that looks OK.

>>
>>>    
>>> -	for (i = 0; i < RTE_ETHER_ADDR_LEN; i++) {
>>> -		mac <<= 8;
>>> -		mac |= mac_addr->addr_bytes[i] & 0xFF;
>>> -	}
>>> +	return 0;
>>> +}
>>>    
>>> +static int
>>> +nfb_eth_mac_addr_add(struct rte_eth_dev *dev,
>>> +	struct rte_ether_addr *mac_addr, uint32_t index, uint32_t pool __rte_unused)
>>> +{
>>> +	unsigned int i;
>>> +	uint64_t mac = 0;
>>> +	struct rte_eth_dev_data *data = dev->data;
>>> +	struct pmd_internals *internals = (struct pmd_internals *)
>>> +		data->dev_private;
>>> +
>>> +	mac = nfb_eth_mac_addr_conv(mac_addr);
>>>    	for (i = 0; i < internals->max_rxmac; ++i)
>>> -		nc_rxmac_set_mac(internals->rxmac[i], 0, mac, 1);
>>> +		nc_rxmac_set_mac(internals->rxmac[i], index, mac, 1);
>>>    
>>> -	rte_ether_addr_copy(mac_addr, data->mac_addrs);
>>>    	return 0;
>>>    }
>>>    
>>> +static void
>>> +nfb_eth_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
>>> +{
>>> +	unsigned int i;
>>> +	struct rte_eth_dev_data *data = dev->data;
>>> +	struct pmd_internals *internals = (struct pmd_internals *)
>>> +		data->dev_private;
>>> +
>>> +	for (i = 0; i < internals->max_rxmac; ++i)
>>> +		nc_rxmac_set_mac(internals->rxmac[i], index, 0, 0);
>>> +}
>>> +
>>>    static const struct eth_dev_ops ops = {
>>>    	.dev_start = nfb_eth_dev_start,
>>>    	.dev_stop = nfb_eth_dev_stop,
>>> @@ -436,6 +481,8 @@ static const struct eth_dev_ops ops = {
>>>    	.stats_get = nfb_eth_stats_get,
>>>    	.stats_reset = nfb_eth_stats_reset,
>>>    	.mac_addr_set = nfb_eth_mac_addr_set,
>>> +	.mac_addr_add = nfb_eth_mac_addr_add,
>>> +	.mac_addr_remove = nfb_eth_mac_addr_remove,
>>>    };
>>>    
>>>    /**
>>> @@ -530,7 +577,7 @@ nfb_eth_dev_init(struct rte_eth_dev *dev)
>>>    	eth_addr_init.addr_bytes[1] = eth_addr.addr_bytes[1];
>>>    	eth_addr_init.addr_bytes[2] = eth_addr.addr_bytes[2];
>>>    
>>> -	nfb_eth_mac_addr_set(dev, &eth_addr_init);
>>> +	rte_eth_dev_default_mac_addr_set(dev->data->port_id, &eth_addr_init);
>>>    
>>
>> I didn't get this change, why calling the API from the driver,
>> instead of calling the dev_ops directly as original code did?
> 
> The API does all the checks and copies the MAC value into internal data
> struct, so our code was duplicit. I didn't realize that calling the API
> could be problem. I assume it should be reverted to the prev form?
> 

It is not a problem, and APIs are used because of the reason you
mentioned in a few places, but for this case I didn't get it,
what API check is required?
Is it 'rte_is_valid_assigned_ether_addr()' check? The mac in this
function ('nfb_eth_dev_init()') is a hardcoded one, instead of
validity check, why not select a valid MAC at this stage?

I mean still drop the 'rte_is_valid_assigned_ether_addr()' check
from 'nfb_eth_mac_addr_set()', but be sure 'eth_addr_init' is
valid MAC, will it work?

>>
>>>    	data->promiscuous = nfb_eth_promiscuous_get(dev);
>>>    	data->all_multicast = nfb_eth_allmulticast_get(dev);
>>
>