[dpdk-stable] [PATCH 1/3] bus/vmbus: fix leak on device scan

Long Li longli at microsoft.com
Fri Oct 1 18:28:40 CEST 2021


> Subject: Re: [PATCH 1/3] bus/vmbus: fix leak on device scan
> 
> On Thu, Sep 30, 2021 at 9:14 PM Long Li <longli at microsoft.com> wrote:
> > > rte_device name only points at some location where the name is stored.
> > > In general this storage is in the bus object or (in some buses) the
> > > devarg that resulted in the rte_device object creation.
> > >
> > > If we won't store the name in the bus object, then we lose the
> > > ability to release the name later.
> > > This is probably fine as long as we never release rte_vmbus_device
> > > objects which is the case atm.
> > > But I don't understand why vmbus should be an exception when
> > > comparing to other buses.
> >
> > I don’t understand why you want to put a name there if it's never used outside
> vmbus_scan_one().
> 
> Since you don't care about releasing rte_vmbus_device objects, I'll rework as
> you suggest.

I don't have any objection defining new data fields in the device struct, as long as it's used during the life cycle of the device.

The rest of the patch looks good, thank you.

Long


More information about the stable mailing list