[dpdk-dev] [PATCH v2 4/4] net/failsafe: fix removed device handling

Gaëtan Rivet gaetan.rivet at 6wind.com
Thu Dec 14 14:27:01 CET 2017


On Thu, Dec 14, 2017 at 01:07:31PM +0000, Matan Azrad wrote:
> Hi Gaetan
> 
> > -----Original Message-----
> > From: Gaëtan Rivet [mailto:gaetan.rivet at 6wind.com]
> > Sent: Thursday, December 14, 2017 12:49 PM
> > To: Matan Azrad <matan at mellanox.com>
> > Cc: Adrien Mazarguil <adrien.mazarguil at 6wind.com>; Thomas Monjalon
> > <thomas at monjalon.net>; dev at dpdk.org; stable at dpdk.org
> > Subject: Re: [PATCH v2 4/4] net/failsafe: fix removed device handling
> > 
> > On Thu, Dec 14, 2017 at 10:40:22AM +0000, Matan Azrad wrote:
> > > Hi Gaetan
> > >
> > 

<snip>

> > > > Ok, actually you were right here to do it this way. The "is_removed"
> > > > check needs to happen after the operation attempt to effectively
> > > > mitigate the possible race. Checking before attempting the call will
> > > > be much less effective.
> > > >
> > > > That being said, would it be cleaner to have eth_dev ops return
> > > > -ENODEV directly, and check against it within fail-safe?
> > > >
> > >
> > > I think that according to "is_removed" semantic we must return a Boolean
> > value (Each value different from '0' means that the device is removed) like
> > other functions in c library (for example isspace()).
> > >
> > 
> > Sure, I wasn't discussing the interface proposed by
> > rte_eth_dev_is_removed().
> > 
> > What I meant was to ask whether checking rte_eth_dev_is_removed()
> > would be more interesting in the ethdev layer, making the eth_dev_ops
> > return -ENODEV regardless of the previous error if this check is supported by
> > the driver and signal that the port is removed.
> > 
> > I think this information could be interesting to other systems, not just fail-
> > safe.
> > 
> 
> Ok. Got you now.
> Interesting approach - plan:
> 	1. update fs_link_update to use rte_eth* functions.

I'm surprised it doesn't already.
Either the rte_eth* function was introduced after the failsafe, or be
wary of potential issues. I don't see a problem right now though.

> 	2. maybe -EIO is preferred because -ENODEV is used for no port error?

Good point, didn't think about it.
Prepare yourself maybe to some arguments about the most relevant error
code. -EIO seems fine to me, but maybe use a wrapper for all this.

Something like:

---8<---

static int
eth_error(pid, int original_ret)
{
    int ret;

    if (original_ret == 0)
        return original_ret;
    ret = rte_eth_is_removed(pid);
    if (ret == 0 || ret == -ENOTSUP)
        return original_ret;
    return -EIO;
}

int
rte_eth_ops_xyz(pid)
{
        int ret;
        ret = eth_dev(pid).ops_xyz();
        return eth_error(pid, ret);
}

--->8---

This way you would be able to change it easily and the logic would be
insulated.

> 	3. update all relevant rte_eth* to use "is_removed" in error flows(1 patch for flow APIs and 1 for the others).
> 	4. Change fs checks in error flows to check rte_eth* return values.
> 	5. Remove CC stable from commit massage.
> 
> What do you think?
> 

Agreed otherwise.

Thanks,

> > --
> > Gaëtan Rivet
> > 6WIND

-- 
Gaëtan Rivet
6WIND


More information about the dev mailing list