lkml.org 
[lkml]   [2019]   [Jul]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH net-next v6 04/15] ethtool: introduce ethtool netlink interface
On Mon, Jul 08, 2019 at 09:26:29PM +0200, Jiri Pirko wrote:
> Mon, Jul 08, 2019 at 07:27:29PM CEST, mkubecek@suse.cz wrote:
> >
> >There are two reasons for this design. First is to reduce the number of
> >requests needed to get the information. This is not so much a problem of
> >ethtool itself; the only existing commands that would result in multiple
> >request messages would be "ethtool <dev>" and "ethtool -s <dev>". Maybe
> >also "ethtool -x/-X <dev>" but even if the indirection table and hash
> >key have different bits assigned now, they don't have to be split even
> >if we split other commands. It may be bigger problem for daemons wanting
> >to keep track of system configuration which would have to issue many
> >requests whenever a new device appears.
> >
> >Second reason is that with 8-bit genetlink command/message id, the space
> >is not as infinite as it might seem. I counted quickly, right now the
> >full series uses 14 ids for kernel messages, with split you propose it
> >would most likely grow to 44. For full implementation of all ethtool
> >functionality, we could get to ~60 ids. It's still only 1/4 of the
> >available space but it's not clear what the future development will look
> >like. We would certainly need to be careful not to start allocating new
> >commands for single parameters and try to be foreseeing about what can
> >be grouped together. But we will need to do that in any case.
> >
> >On kernel side, splitting existing messages would make some things a bit
> >easier. It would also reduce the number of scenarios where only part of
> >requested information is available or only part of a SET request fails.
>
> Okay, I got your point. So why don't we look at if from the other angle.
> Why don't we have only single get/set command that would be in general
> used to get/set ALL info from/to the kernel. Where we can have these
> bits (perhaps rather varlen bitfield) to for user to indicate which data
> is he interested in? This scales. The other commands would be
> just for action.
>
> Something like RTM_GETLINK/RTM_SETLINK. Makes sense?

It's certainly an option but at the first glance it seems as just moving
what I tried to avoid one level lower. It would work around the u8 issue
(but as Johannes pointed out, we can handle it with genetlink when/if
the time comes). We would almost certainly have to split the replies
into multiple messages to keep the packet size reasonable. I'll have to
think more about the consequences for both kernel and userspace.

My gut feeling is that out of the two extreme options (one universal
message type and message types corresponding to current infomask bits),
the latter is more appealing. After all, ethtool has been gathering
features that would need those ~60 message types for 20 years.

Michal

\
 
 \ /
  Last update: 2019-07-08 22:22    [W:0.198 / U:0.128 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site