On 6/12/2017 3:19 PM, Bruce Richardson wrote: > On Mon, Jun 12, 2017 at 03:08:31PM +0100, Ferruh Yigit wrote: >> On 6/12/2017 2:25 PM, Bruce Richardson wrote: >>> On Fri, Jun 09, 2017 at 06:51:19PM +0100, Ferruh Yigit wrote: >>>> When ring PMD created via PMD specific API instead of EAL abstraction >>>> it is missing the virtual device creation done by EAL vdev. >>>> >>>> And this makes eth_dev unusable exact same as other PMDs used, because >>>> of some missing fields, like rte_device->name. >>>> >>>> Now API creates a virtual device and sets proper fields, not all, and it >>>> still won't be linked in the virtual device list eal keeps track. But >>>> makes PMD usable in usual manner. >>>> >>>> Signed-off-by: Ferruh Yigit <ferruh.yi...@intel.com> >>>> --- >>> >>> Is a better fix not to have this API call into the EAL to create the >>> vdev and add it to the lists as with other vdevs? [If it makes it easier, >>> the extra parameters passed in to the library-local function can be >>> saved in a context that can be accessed when the EAL calls back into the >>> driver, rather than having to flatten them out into devargs and re-parsed >>> again.] >> >> Let me send the patch as suggested. >> >> Using EAL API is better idea I think, but overall this ring PMD looked >> like hack after changes. >> >> Please check the latest patch, if we want to keep ring PMD API, perhaps >> we should postpone removing drv_name patch. >> > The new patch looks ok to me. I actually don't think it looks that > hacked together. :-)
That is good, if you are happy with the patch, I guess we can get the patchset. Thanks, ferruh > > /Bruce >
