[dpdk-dev] [PATCH v6 12/17] pci: add a helper for device name

Jan Viktorin viktorin at rehivetech.com
Fri Jul 15 13:32:20 CEST 2016


On Fri, 15 Jul 2016 11:56:38 +0200
Thomas Monjalon <thomas.monjalon at 6wind.com> wrote:

> 2016-07-15 15:09, Shreyansh jain:
> > On Thursday 14 July 2016 10:25 PM, Jan Viktorin wrote:  
> > > What is meant by "resources" here?  
> > 
> > This has historic context (from earlier version of this patch). 
> > But I could relate the word 'resources' to EAL representation of devices - whether PCI or Crypto.
> > Or, Resource == Device.  
> 
> We need to define a precise wording, especially for the words
> 	- interface
> 	- resource
> 	- device
> 	- driver
> 	- module
> Following are my thoughts:
> 
> An interface is a view of a resource through an API (e.g. an ethdev port).
> A resource is a well identified hardware (e.g. a PCI device id).
> A device is a hardware function (e.g. a networking port).
> Note that some PCI NICs have only one PCI device id for several ports
> (Chelsio and Mellanox cases). That's why we need to distinguish device
> and resource.
> 
> A driver is the code handling a device.
> I have read the word module in some patch proposals but I don't really know
> what it refers to. Is it a group of drivers in the same file/directory?

Well, yes, initially there was proposed a kind of module. However, after David
as sent the patchset prepare for rte_device/driver, you can see that there
is no need for such object (probably). A module might be helpful when thinking
about the metadata (as done by Neil Horman). But as far as I know there was no
need for this... So a module has no data abstraction now and I am not going to
make any.

Jan


More information about the dev mailing list