On 12/22/2015 02:56 AM, Or Gerlitz wrote: > On Wed, Dec 16, 2015 at 7:53 AM, Or Gerlitz wrote: >> On 12/15/2015 9:03 PM, Doug Ledford wrote: > >>> Or, you specifically asked me to wait until this week. I made my >>> initial impressions clear (I don't necessarily like the removal of the >>> attr struct, but I like the removal of all of the query calls, and I'm >>> inclined to take the patch in spite of not liking the removal of the >>> struct). Do you have anything to add or have we beat this horse to death? > >> Hi Doug, >> Lets stop beating, both horses and people. >> I do understand that >> 1. you don't link the removal of the attr >> 2. you do like the removal of all the query calls >> >> I am proposing to take the path of a patch that >> does exactly #2 while avoiding #1. > > Doug, > > Did you look on my v1 post and the related discussion there w.r.t udata? Yes, I did. > You didn't make any comment on my response here nor on the proposed patches. I'm trying to find all of the emails, they aren't in a single thread in my mailbox (I had to do some reconstruction of my mailbox due to a problem in a mail filter late last week...missing that the rule was set to "match any" when I intended "match all" and the action of the rule was "delete" when I expected delete to be the same as "move to trash" and it wasn't, it was delete immediately, has caused me some problems). > Since we are really short in time w.r.t EOY holidays and we have the > udata matter > open (see [1]), could we move finalizing this discussion to the 4.6 time-frame? > > If you do have the time, I think it would be fair to see a response > from you on the > discussion before you pick any of the two patch sets - so?? I'm not inclined to take either patch set as they stand. Your's is closer to what I'm leaning towards though. I think I can add a single patch to yours to make it into what I want. I'm going to go work on that right now... > Or. > > [1] Christoph's patch doesn't remove the query_device callback from > mlx4 since we > report there values to libmlx4 through the udata mechanism. The > query_device callback > will need to be present in future/current drivers if they decide to > use udata as well > > >> What's wrong with that? I haven't heard any reasoning for why its >> so good to stash ~50 new fields on the IB device structure except >> for the author saying that other subsystems do that and other people >> saying they are in favor of this approach while not providing any >> reasoning, except for maybe something on bikes. >> >> Why you or anyone else has to be from now and ever the cache line police >> making sure that people don't add new attributes in random locations >> over the IB device structure? >> >> What's wrong with putting fifty attributesin a structure which is a field >> of the device struct and have people go there to see what are the d >> ifferentattrs and add news ones there? >> >> This will make the 4.5 merge window extremely complex or even totally >> threatened w.r.t to the RDMA subsystem and related drivers by 3.3K LOC >> patch. >> >> Sorry, but, I still don't get it. >> >> Or. -- Doug Ledford GPG KeyID: 0E572FDD