Re: [dpdk-dev] [PATCH 1/2] net/tap: fix setting speed by argument

[Pascal Mazon <pascal.mazon@6wind.com>]

On 25/10/2017 03:24, Ferruh Yigit wrote:
> On 9/19/2017 5:45 AM, Pascal Mazon wrote:
>> Hi,
>>
>> The patch looks mainly ok to me.
>>
>> I'll put some comments inline.
>>
>> On 18/09/2017 20:47, Ferruh Yigit wrote:
>>> From: Vipin Varghese <vipin.varghese@intel.com>
>>>
>>> tap speed argument is not working without generating any error.
>> Can you describe the error, paste the log?
>>> This patch sets the configured speed during device start.
>>>
>>> Fixes: 02f96a0a82d1 ("net/tap: add TUN/TAP device PMD")
>>> Cc: stable@dpdk.org
>>>
>>> Signed-off-by: Vipin Varghese <vipin.varghese@intel.com>
>>> ---
>>>  drivers/net/tap/rte_eth_tap.c | 27 +++++++++++++++++++++++++++
>>>  drivers/net/tap/rte_eth_tap.h |  2 ++
>>>  2 files changed, 29 insertions(+)
>>>
>>> diff --git a/drivers/net/tap/rte_eth_tap.c b/drivers/net/tap/rte_eth_tap.c
>>> index 01cba0fa1..00dad167f 100644
>>> --- a/drivers/net/tap/rte_eth_tap.c
>>> +++ b/drivers/net/tap/rte_eth_tap.c
>>> @@ -545,6 +545,7 @@ tap_ioctl(struct pmd_internals *pmd, unsigned long request,
>>>  	case SIOCGIFHWADDR:
>>>  	case SIOCSIFHWADDR:
>>>  	case SIOCSIFMTU:
>>> +	case SIOCETHTOOL:
>>>  		break;
>>>  	default:
>>>  		RTE_ASSERT(!"unsupported request type: must not happen");
>>> @@ -585,6 +586,32 @@ static int
>>>  tap_dev_start(struct rte_eth_dev *dev)
>>>  {
>>>  	int err;
>>> +	struct ifreq ifr;
>>> +	struct ethtool_cmd edata = {0};
>>> +	struct pmd_internals *pmd = dev->data->dev_private;
>>> +
>>> +	/*set & get speed to device*/
>>> +	edata.speed = pmd_link.link_speed;
>>> +	edata.cmd = ETHTOOL_SSET;
>>> +	ifr.ifr_data = (caddr_t)&edata;
>>> +	if (tap_ioctl(pmd, SIOCETHTOOL, &ifr, 0, LOCAL_ONLY) < 0) {
>> I think it would be best to also try setting the speed on the remote
>> (tap_ioctl(REMOTE_ONLY)), but continue execution even if that fails.
> Hi Pascal,
>
> It looks like this ioctl fails to set tap speed. And as far as I can see speed
> is hardcoded in kernel [1], do you know if it is possible to set speed of tap
> interface?
>
> Thanks,
> ferruh
>
> [1]
> http://elixir.free-electrons.com/linux/v4.13.9/source/drivers/net/tun.c#L2460
Hi Ferruh,

I concur with you, it definitely looks like speed is hardcoded in the
kernel.
And there is no set_settings() callback defined either, so it won't be
possible to set anything using ethtool.

I haven't seen the error and I don't think Vipin Varghese sent a log for
that.
I'm still not sure what the patch was fixing; it would be good to have
that feedback to improve the patch.

Best regards,
Pascal
>
>>> +		RTE_LOG(WARNING, PMD,
>>> +			"Could not set param speed %d for ifindex for %s.\n",
>>> +				pmd_link.link_speed,
>>> +				pmd->name);
>>> +
>>> +		/* fetch current speed of created device */
>>> +		edata.cmd = ETHTOOL_GSET;
>>> +		ifr.ifr_data = (caddr_t)&edata;
>>> +		if (tap_ioctl(pmd, SIOCETHTOOL, &ifr, 0, LOCAL_ONLY) < 0)
>>> +			return 0;
>>> +
>>> +		pmd_link.link_speed = edata.speed;
>>> +		RTE_LOG(INFO, PMD, "get speed %d for ifindex for %s.\n",
>> I would say "got".
>>
>>> +			pmd_link.link_speed,
>>> +			pmd->name);
>>> +	}
>>> +	dev->data->dev_link = pmd_link;
>>>  
>>>  	err = tap_intr_handle_set(dev, 1);
>>>  	if (err)
>>> diff --git a/drivers/net/tap/rte_eth_tap.h b/drivers/net/tap/rte_eth_tap.h
>>> index 928a0454e..3e91db4ae 100644
>>> --- a/drivers/net/tap/rte_eth_tap.h
>>> +++ b/drivers/net/tap/rte_eth_tap.h
>>> @@ -40,6 +40,8 @@
>>>  #include <net/if.h>
>>>  
>>>  #include <linux/if_tun.h>
>>> +#include <linux/ethtool.h>
>>> +#include <linux/sockios.h>
>>>  
>>>  #include <rte_ethdev.h>
>>>  #include <rte_ether.h>
>> These includes are only useful inside rte_eth_tap.c, I would put them
>> there only.
>>