Thomas Valentino Crimi writes:
> > Whether libkvm should even exist in a perfect world (it shouldn't)
> > is an entirely different question. For now, we're stuck with it
> > until somebody changes *everything* to use sysctl instead.
>
> Just as a question, how much of a performance difference is
Excerpts from FreeBSD-Current: 24-Jan-99 Re: kvm question by Archie
co...@whistle.com
> Whether libkvm should even exist in a perfect world (it shouldn't)
> is an entirely different question. For now, we're stuck with it
> until somebody changes *everything* to use sysctl in
In message <199901251615.laa19...@khavrinen.lcs.mit.edu>, Garrett Wollman write
s:
>< said:
>
>> Strings are a whole lot more portable then integer assignments.
>
>Nonsense. Strings are not portable at all -- they only exist in
>FreeBSD. The reference implementation (4.4BSD) and its other
>de
< said:
> Strings are a whole lot more portable then integer assignments.
Nonsense. Strings are not portable at all -- they only exist in
FreeBSD. The reference implementation (4.4BSD) and its other
descendants use numbers.
-GAWollman
--
Garrett A. Wollman | O Siem / We are all family /
On Sun, 24 Jan 1999, Mike Smith wrote:
> > < > said:
> >
> > > Peter pointed out that having the sysctl's as symbols was a nice
> > > advantage of the current system. How important is this?
> >
> > I don't think it's important at all. (Then again, I liked the old
> > system.)
> >
> > > If we
In message <199901242201.raa17...@khavrinen.lcs.mit.edu>, Garrett Wollman write
s:
>< said:
>
>> Backwards compatibility is one thing, but new nodes should be named,
>> not numbered. OID_AUTO is bogus because it perpetuates the numbering
>> of nodes.
>
>Nonsense. There are plenty of contexts in
Mike Smith once stated:
=OTOH, you should consider going back to single-character directory
=names, since that's much more significant.
a) this will limit the number of directories to you-know-what
b) this will inconvinience a _user_ rather then a _programmer_,
for who
>Backwards compatibility is one thing, but new nodes should be named,
>not numbered. OID_AUTO is bogus because it perpetuates the numbering
>of nodes.
OID_AUTO is not bogus. It is just an implementation detail.
The sysctl data structures have to have a place to put a number for
old-style numbe
Julian Elischer writes:
> That is at least my opinion.. you may and do disagree. I guess you will
> say that numbers are just as dynamic, etc.etc. well I just think that in
> the REAL WORLD, as opposed to the theoretical world, names (which require
> no co-ordination between authors), are a better
:> > not numbered. OID_AUTO is bogus because it perpetuates the numbering
:> > of nodes.
:>
:> Nonsense. There are plenty of contexts in which a number makes far
:> more sense than a name -- pretty much anything in any network stack
:> other than Chaosnet, for example. If any of us ever make g
On Sun, 24 Jan 1999, Archie Cobbs wrote:
> Julian Elischer writes:
> > As soon as someone modifies sysctl to work with KLD modules
> > that would be a reasonable suggestion
> >
> > On Sun, 24 Jan 1999, Garrett Wollman wrote:
> > > > I ran into an interesting problem in the process of modifyi
>
> Pardon my intrusion, but I strongly dislike the very thought about
> my computer looking-up the same string more then once or twice. If it
> counts -- I'd take a number over a string anytime anywhere other
> then in a documentation.
Since sysctl isn't a performance interface, this isn't reall
> < said:
>
> > Backwards compatibility is one thing, but new nodes should be named,
> > not numbered. OID_AUTO is bogus because it perpetuates the numbering
> > of nodes.
>
> Nonsense. There are plenty of contexts in which a number makes far
> more sense than a name -- pretty much anything i
:Seldom. But the strings are still in the kernel, which becomes
:bigger with every build. My argument was more general, however,
:and directed against the growing tendency to use string literal
:(and copy them beck and forth). IMHO, the point of faster hardware
:is purely to have thing running fast
Julian Elischer once stated:
=> Pardon my intrusion, but I strongly dislike the very thought about
=> my computer looking-up the same string more then once or twice. If it
=> counts -- I'd take a number over a string anytime anywhere other
=> then in a documentation.
=how often do you use this?
On Sun, 24 Jan 1999, Mikhail Teterin wrote:
>
> Pardon my intrusion, but I strongly dislike the very thought about
> my computer looking-up the same string more then once or twice. If it
> counts -- I'd take a number over a string anytime anywhere other
> then in a documentation.
how often do
Matthew Dillon once stated:
=This is a silly argument. Unless the operation in question
=needs to be run a thousand times a second, a string is just
=fine as a lookup mechanism. Duh. Besides, you can always
=cache the translation.
I'll agree, that todays hardware turns this in
This is a silly argument. Unless the operation in question
needs to be run a thousand times a second, a string is just
fine as a lookup mechanism. Duh. Besides, you can always
cache the translation.
-Matt
Julian Elischer once stated:
=> Nonsense. There are plenty of contexts in which a number makes far
=> more sense than a name -- pretty much anything in any network stack
=> other than Chaosnet, for example. If any of us ever make good on the
=> threat of SNMP integration, having fixed numerical id
yeah and we should get those nice valves that used to make radios so
useful as space-heaters.
On Sun, 24 Jan 1999, Garrett Wollman wrote:
> < said:
>
> > Backwards compatibility is one thing, but new nodes should be named,
> > not numbered. OID_AUTO is bogus because it perpetuates the numberi
< said:
> Backwards compatibility is one thing, but new nodes should be named,
> not numbered. OID_AUTO is bogus because it perpetuates the numbering
> of nodes.
Nonsense. There are plenty of contexts in which a number makes far
more sense than a name -- pretty much anything in any network st
Archie Cobbs wrote:
> Julian Elischer writes:
> > As soon as someone modifies sysctl to work with KLD modules
> > that would be a reasonable suggestion
> >
> > On Sun, 24 Jan 1999, Garrett Wollman wrote:
> > > > I ran into an interesting problem in the process of modifying
> > > > "netstat" t
Julian Elischer writes:
> As soon as someone modifies sysctl to work with KLD modules
> that would be a reasonable suggestion
>
> On Sun, 24 Jan 1999, Garrett Wollman wrote:
> > > I ran into an interesting problem in the process of modifying
> > > "netstat" to understand the PF_NETGRAPH proto
> <
> said:
>
> > Peter pointed out that having the sysctl's as symbols was a nice
> > advantage of the current system. How important is this?
>
> I don't think it's important at all. (Then again, I liked the old
> system.)
>
> > If we were willing to give this up, then the SYSCTL() macro coul
<
said:
> Peter pointed out that having the sysctl's as symbols was a nice
> advantage of the current system. How important is this?
I don't think it's important at all. (Then again, I liked the old
system.)
> If we were willing to give this up, then the SYSCTL() macro could
> just expand to a
As soon as someone modifies sysctl to work with KLD modules
that would be a reasonable suggestion
On Sun, 24 Jan 1999, Garrett Wollman wrote:
> <
> said:
>
> > I ran into an interesting problem in the process of modifying
> > "netstat" to understand the PF_NETGRAPH protocol family. "netsta
Mike Smith writes:
> > My question is, should kvm_read() and friends be "enhanced" with
> > this ability to find a symbol by searching through the loaded
> > KLD modules? Seems a bit hackish, but then again so is the whole
> > kvm() idea.
>
> No; you should be using sysctl to get at the informatio
Mike Smith writes:
> Yes, there's a desire to see this fixed; it requires a significant
> rewrite of the sysctl stuff unfortunately. Mark Murray was working on
> this but probably as a 4.x feature; if you have something simpler you
> feel up to contributing in the short term you'd be very popul
> Julian Elischer writes:
> > On Fri, 22 Jan 1999, Archie Cobbs wrote:
> > > Unrelated question: SYSINIT() doesn't work from KLD modules.
> > > Is this problem being addressed?
> >
> > you mean sysctl
>
> Oops, thanks.. SYSCTL() doesn't work from KLD modules but SYSINIT() does.
What I get for be
> I ran into an interesting problem in the process of modifying
> "netstat" to understand the PF_NETGRAPH protocol family. "netstat"
> uses kvm_read(), etc. to read kernel symbols. However, this doesn't
> work when the symbols you're looking for are in an KLD module (eg,
> ng_socket.ko) -- the symb
Archie Cobbs wrote:
> Julian Elischer writes:
> > On Fri, 22 Jan 1999, Archie Cobbs wrote:
> > > Unrelated question: SYSINIT() doesn't work from KLD modules.
> > > Is this problem being addressed?
> >
> > you mean sysctl
>
> Oops, thanks.. SYSCTL() doesn't work from KLD modules but SYSINIT() does
Julian Elischer writes:
> On Fri, 22 Jan 1999, Archie Cobbs wrote:
> > Unrelated question: SYSINIT() doesn't work from KLD modules.
> > Is this problem being addressed?
>
> you mean sysctl
Oops, thanks.. SYSCTL() doesn't work from KLD modules but SYSINIT() does.
-Archie
On Fri, 22 Jan 1999, Archie Cobbs wrote:
>
> Unrelated question: SYSINIT() doesn't work from KLD modules.
> Is this problem being addressed?
you mean sysctl
>
> -Archie
>
> ___
> Archie Cobbs * Whistle Communications,
33 matches
Mail list logo
