[lkml]   [2006]   [Oct]   [17]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
    SubjectRe: [take19 1/4] kevent: Core files.
    Ulrich Drepper wrote:
    > Evgeniy Polyakov wrote:
    >> Existing design does not allow overflow.
    > And I've pointed out a number of times that this is not practical at
    > best. There are event sources which can create events which cannot be
    > coalesced into one single event as it would be required with your design.
    > Signals are one example, specifically realtime signals. If we do not
    > want the design to be limited from the start this approach has to be
    > thought over.
    >>> So zap mmap() support completely, since it is not usable at all. We
    >>> wont discuss on it.
    >> Initial implementation did not have it.
    >> But I was requested to do it, and it is ready now.
    >> No one likes it, but no one provides an alternative implementation.
    >> We are stuck.
    > We need the mapped ring buffer. The current design (before it was
    > removed) was broken but this does not mean it shouldn't be
    > implemented. We just need more time to figure out how to implement it
    > correctly.
    Considering the if at all and if then how of ring buffer implemetation
    I'd like to throw in some ideas I had when reading the discussion and
    respective code. If I understood Ulrich Drepper right, his notion of a
    generic event handling interface is, that it has to be flexible enough
    to transport additional info from origin to userspace, and to support
    queuing of events from the same origin, so that additional
    per-event-occurrence data doesn't get lost, which would happen when
    coalescing multiple events into one until delivery. From what I read he
    says ring buffer is broken because of insufficient space for additional
    data (mukevent) and the limited number of events that can be put into
    ring buffer. Another argument is missing notification of userspace about
    dropped events in case ring buffer limit is reached. (is that right?)
    I see no reason why kevent couldn't be modified to fit (all) these
    needs. While modifying the server-example and writing a client using
    kevent I came across the coalescing problem, there were more incoming
    connections than accept events, and I had to work around that. In this
    case the pure number of coalesced events would suffice, while it
    wouldn't for the example of RT-signals that Ulrich Drepper gave. So if
    coalescing can be done at all or if it is impossible depends on the type
    of event. The same goes for additional data delivered with the events.
    There might be no panacea for all possible scenarios with one fixed
    design. Either performance suffers for 'lightweight' events which don't
    need additional data and/or coalescing is not problematic and/or ring
    buffer, or kevent is not usable for other types of events. Why not treat
    different things differently, and let the (kernel-)user decide.
    I don't know if I got all this right, but if, then ring buffer is needed
    especially for cases where coalescing is not possible and additional
    data has to be delivered for each triggered notification (so the pure
    number of events is not enough; other reasons? performance? ). To me it
    doesn't make sense to have kevent fill memory and use processor-time if
    buffer is not used at all, which is the case when using kevent_getevents.
    So here are my Ideas:
    Make usage of ring buffer optional, if not required for specific
    event-type it might be chosen by userspace-code.
    Make limit of events in ring buffer optional and controllable from
    Regarding mukevent I'm thinking of a event-type specific struct, that is
    filled by the originating code, and placed into a per-event-type ring
    buffer (which requires modification of kevent_wait). To my limited
    understanding it seems that alternative or modified versions of
    kevent_storage_ready, (__)kevent_requeue and kevent_user_ring_add_event
    could return a void pointer to the position in buffer, and all kevent
    has to know about is the size of the struct.
    If coalescing doesn't hurt for a specific event-type it might just be
    modified to notify userspace about the number of coalesced events. Make
    it depend on type of event.

    I know this doesn't address all objections that have been made, and
    Evgeniy, big sorry for this being just talk again, and maybe not even
    applicable for some reasons I do not overlook, but maybe it's worth
    consideration. I'll gladly try to put that into code, and see where it
    leads. I think kevent is great, and if things can be done to increase
    it's genericity without sacrifying performance, why not.
    Sorry for the length of post and repetitions,

    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to
    More majordomo info at
    Please read the FAQ at

     \ /
      Last update: 2006-10-17 07:09    [W:0.024 / U:14.648 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site