On November 17, 2001 10:41 pm, William Lee Irwin III wrote: > On Sat, Nov 17, 2001 at 10:34:23PM +0100, Daniel Phillips wrote: > > Hey! About time I landed here.... > > I'm here too! Good to see you! I was worried for a moment that things > weren't getting through, looks like it's all clear, though. Hey, wli! Read, enjoy Of course, you've already read it, haven't you? -- Daniel
--> You are now talking on #kernelnewbies --- Topic for #kernelnewbies is #define __KERNEL__ || 2.4.15-pre5 (last before Linus hands off to Marcelo?) || 2.4.13-ac8 (Alan suggests using 7; register changes in ac8 make it rather unstable) || 2.2.20 || uml-2.4.14-2 || 2.0.40-pre2 || problems building 2.4.15-pre4? See http://marc.theaimsgroup.com/?l=linux-kernel&m=100562085807293&w=2 --- Topic for #kernelnewbies set by catfish at Fri Nov 16 03:07:34 --- You are now known as philips <philips> riel? <rsilva> I dont know how to interpret those entries. Here are they: <rsilva> inode_cache 171751 171759 512 24537 24537 1 <rsilva> dentry_cache 178041 178050 128 5935 5935 1 <riel> philips ? <riel> rsilva: ok, the 5th column of numbers is the interesting one <philips> hey --> FunkyTrix (~Fun@62.98.95.18) has joined #kernelnewbies <riel> rsilva: it shows the number of _pages_ allocated for that cache <riel> rsilva: in your case, inode+dentry cache is 30500 pages, or 122 MB <-- yusufg has quit ("Client Exiting") <-- uzi has quit (forward.openprojects.net irc.openprojects.net) --> uzi (uzi@alancoxonachip.com) has joined #kernelnewbies <rsilva> What does that really mean? What is using this amount of memory? <Gandalf_> rsilva: the name of those fields say snomething_cache... so it's a cache and that cache will be freed when something needs that memory <Gandalf_> rsilva: these caches are for caching filsystem stuff and that is an very important cache <rsilva> That is good news. But, why the "free" command doesn't know about it? <-- lxrbot has quit ("So long, and thanks for all the fish.") <Gandalf_> ask the person who wrote 'free' --> lxrbot (lxrbot@inpc60.et.tudelft.nl) has joined #kernelnewbies <HDP> i compiled a module for my aureal vortex sound card, when i boot i get unresolved sumbol errors from that module; but when i do manually modprobe soundcore and afterwards insmod au3380 things work like a breeze, i tried depmod -a but that complains also about unresolved symbols ... anyone a clue ? <-- hklygre has quit (Ping timeout: 181 seconds) <rsilva> That's a good point :-) Anyhow I was getting nervous about this memory usage, your explanation made things more confortabe. Thank you very much, riel, gandalf and diablo. <riel> rsilva: the 'free' command doesn't know about it because nobody implemented that <-- boto has quit (Read error: 104 (Connection reset by peer)) <h0h0> how do you tell make to check stuff out of bk if it needs to? <rsilva> Even better point :-) Maybe in a future release of free. It interesting that I think this newmemory usage pattern appeared on one of my last kernel upgrades. Maybe ithas something to do with the new VM? <h0h0> hehe, you know how quickly development on `free` is <h0h0> runs, <Gandalf_> rsilva: this has been in the kernel for a _long_ time now <rsilva> My last kernel was 2.4.7 (now I am using 2.4.14). --> Ernimril (~d94-rol@my.nada.kth.se) has joined #kernelnewbies <rsilva> Oh, I have just noticed a detail. Riel, am I wrong, or you are leaving in Brazil now, working for Conectiva? <-- jeb-c4 has quit (Remote closed the connection) <riel> rsilva: s/leaving/living/ but indeed ;) * fbl is away [autogone:10/l:on] <s0mbre> Hi. Could you please explain, does work with parallel port in linux means reading and writing bytes in/from 0x37* port using inb/outb ant ioperm()? If this is true, than what parport_ modules are used for? <rsilva> So you must know my cousing Joao Luis, who works there from the very beggining. Say hello to him for me (my real name is Paulo). Sorry for my English, but I type looking at the keyboard :-) <Gandalf_> s0mbre: yes it works writing and reading from 0x37* and the parport module is used for platform independent parport access IIRC <-- HDP (full@D5760111.kabel.telenet.be) has left #kernelnewbies * fbl is back [TE/gone: 3m1s] <-- Scandium has quit () * fbl is away [autogone:0/l:on] <JALH> ahh <JALH> I love this quote: <JALH> Microsoft likes to discard vulnerabilities by "no standard client <JALH> would do this." No, and no "standard visitor" would apply a crowbar <JALH> to your patio door, either. <Gandalf_> hehe --> Sebas22 (~me@m126-ras4.netizen.com.ar) has joined #kernelnewbies <-- Sebas22 has quit (Remote closed the connection) --> GeoDex (geodex@i.got.shelled.at.shellyeah.org) has joined #kernelnewbies <-- s0mbre has quit ("Segmentation Fault") <GeoDex> I have compiled 2.4.14 and had it up and running but when i tried to go on net it acts like there is no ppp <GeoDex> says no ppp0 found <GeoDex> ?? <GeoDex> do i have to re insmod the ppp? <h0h0> ifconfig <h0h0> yes, every time you reboot you have to modprobe it <fredlwm> GeoDex: Are you upgradin from 2.2 ? Read Documentation/Changes <GeoDex> yes <GeoDex> k <fredlwm> The stuff for modules.conf changed <GeoDex> hmm <Justin_> you also need /dev/ppp or something <GeoDex> well, i guess that would do that :) <-- fbl has quit ("disabling autoaway") --> Sebas22 (~me@m126-ras4.netizen.com.ar) has joined #kernelnewbies <GeoDex> fredlwm: Documentation/Changes on kernelnewbies site or in package? <fredlwm> GeoDex: pkg <philips> riel, about that #kernelnewbies fork... <riel> philips: yes ? <philips> will it be forked from alan, linus, or marcelo tree? <GeoDex> modutils i guess <riel> philips: I think it should be possible to track both 2.4 and 2.5 for a while <philips> riel, are you sure? <riel> philips: but eventually I think we'll really want 2.5 only <philips> yes <GeoDex> i have modutils-2.3.14-3 not 2.4.2 <philips> how about: track 2.4 until 2.5 opens? <riel> philips: with bitkeeper it's not too hard to implement things in 2.4 and then pull things into 2.5 <riel> philips: that was the plan ;) <fredlwm> GeoDex: Then upgrade <GeoDex> :) <philips> riel, great <philips> riel, so we are opening the tree when? --> dhask (dsada@bzq-230-24.red.bezeqint.net) has joined #kernelnewbies <-- dhask has quit (Client Quit) <philips> riel, and what is the requirement for #kernelnewbies members to load in their own hacks? <riel> philips: once I figure out bitkeeper ;) <riel> philips: no requirement for having stuff in your own tree <philips> riel, ack <riel> philips: the only thing is that something should work before it's integrated into the #kn tree <-- float has quit (Remote closed the connection) <riel> philips: that is, before I'll integrate changesets ;) <philips> riel, I mean, what is the requirment for a #kernelnewbies member to access the bitkeeper <philips> riel, and write into their own branch? <riel> philips: everybody has pull access <philips> riel, and who has push? <riel> philips: bitkeeper doesn't implement branches yet ;( <riel> philips: everybody will need their own repository <riel> philips: I'm not sure how to organise push yet <philips> riel, ok, we will go with the flow <philips> riel, so there will be a small cabal of push priests <riel> philips: and a place where everybody can put their unchecked changesets, I guess ;) <-- michael has quit ("Client Exiting") <philips> riel right <philips> riel, a patchbot <riel> philips: good idea <philips> riel, I will hack one in python <philips> and send linus a copy <-- rsilva (~rsilva@200.184.84.2) has left #kernelnewbies <h0h0> if i want to compile a kernel tree i got in bitkeeper, do i have to check it out first or can i have make automaticly checkout what it needs? <riel> philips: what will it do ? <philips> riel, not much <riel> h0h0: unfortunately make doesn't check out everything ;( <h0h0> hrm <philips> riel, append a serial number, mail a copy to linus and one to lkml <riel> h0h0: it would be nice if somebody did the work to write a script which checks out the things make/gcc don't check out ... </hint> ;) <-- mindfsck has quit (Ping timeout: 181 seconds) <philips> s/append/prepend/ <riel> philips: how about combining that with a patch _archive_ ? <philips> riel, the patchbot will be scriptable <philips> riel, and allow plugins, ok? <riel> neat <-- h0h0 has quit (Read error: 104 (Connection reset by peer)) <philips> 'apatchybot' <velco> heh <-- GeoDex has quit (" The Internet Dreadnought.") --> Shark45 (ethy@poc1.inexo.com.br) has joined #kernelnewbies <philips> riel, so what goes into the #kernelnewbies fork? <philips> riel, mm-wise? * wli listens intently. <riel> philips: a working VM ? <riel> philips: I'll start with things like write clustering and reverse mapping, will add load control later <riel> philips: and _maybe_ something of an object-based VM <philips> riel wow great <riel> philips: but I haven't decided on the last point yet <philips> riel, I will do the non-blocking kswapd <velco> what is object-based VM ? <philips> riel, optional of course <riel> velco: based on objects <philips> riel, and the semaphore based resource management in alloc_pages <riel> philips: uhh, why semaphores ? <riel> philips: one thing I want in a new VM is to have every feature justified <velco> riel: well ... I got that ... <philips> riel, because they can count <philips> riel, they can count and block <riel> philips: what are they supposed to achieve ? <philips> riel, and in particular, they can count pages <riel> philips: in what situation will you use them ? <philips> riel, they will rationalize the scanning policy <philips> riel, the scheduling of the scanning policy <philips> (only the scheduling) <riel> philips: I don't get the picture <philips> riel, ok --> GeoDex (geodex@i.got.shelled.at.shellyeah.org) has joined #kernelnewbies <philips> riel, remember, it's my experimental patch <philips> riel, let me describe <GeoDex> thx all, ppp up and running =) * philips gets ready * philips thinks * philips stops typing in the other three windows <philips> well, in 2 of the 3 other windows <Sebas22> ==> any of u know of ftp sites to download rh72 .iso images that r OK ? no ftp.redhat.com s not posible.. cause y download it 2 times & does not match md5sum. did they fix the problem whith the mirrors??. <-- GeoDex has quit (Client Quit) <philips> ok, +1 on the semaphore count means 1 page freed <philips> and, if there is a process blocked waiting for a page, it becomes runnable <philips> -1 on the semaphore count means one page alloced <philips> and if there is no page available, the process blocks <philips> and kswapd goes to work <philips> if it isn't already <riel> philips: so, will we have one semaphore per zone ? <philips> riel, yes <philips> riel, and one semphore per order per zone <riel> philips: so how about processes which can allocate from multiple zones ? <philips> riel, one per order per zone per cpu <philips> riel, I'll think first <philips> riel, we *do not* use class zone <philips> riel, I'll come up with a graph-theoretic algo <philips> riel, that works for numa <philips> riel, and simplifies to a very fast, simple algo for the simple case <riel> philips: I think I have one --> yusufg (~yusufg@cm61-10-94-231.hkcable.com.hk) has joined #kernelnewbies <philips> riel, excellent <philips> riel, description? * philips decides this chat log is a keeper <wli> You've got an algorithm, I've got a NUMA machine to run it on. <riel> philips: when a process cannot find a page to allocate, it sets zone->need_balance in each of the zones it tries to allocate from <riel> philips: kswapd will try to balance the zones and will clear the flags before going to sleep and waking up the process <riel> philips: if the process succeeds now, zone->need_balance won't be set again <philips> riel, interesting --> fbl (~fbl@1-214.ctame701-1.telepar.net.br) has joined #kernelnewbies <philips> riel, ok, I will use that as a starting point <philips> riel, and generalize it <riel> philips: the number of different fallback graphs on NUMA could be so large that having any per-zone or per-classzone things are just bound to fail <riel> philips: but a simpler scheme would work ;) <philips> riel, I will change the need_balance bit to a signed integer <philips> riel, and maybe go to fuzzy logic <riel> philips: that makes sense <philips> riel, simplified of course <philips> riel, ok, that is enough for my poor aching brain today <riel> philips: kswapd can free pages from any zone, but should try harder if userspace is more desparate <philips> riel, I'm going to draw some pictures, maybe write some code <riel> ;) <philips> riel, right <philips> riel, this is why we want ints, not bits <riel> this is much more generic and functional than what's in classzone ;) <riel> and simpler, too <riel> philips: *nod* <wli> In my mind, given a graph representation of the topology, there would be a static penalty assessed for the interconnect latencies for inter-node transfers, and a dynamic penalty associated with the load, and so a graph search could consist of inserting the things in a priority queue and attempting migrations in priority queue order. Is that what you have in mind? <philips> riel, it's more analog <Shark45> gentleman; hav you ever heard about "hidden node" wireless communicatio problem? <philips> riel, but not so analog we can't prove things about it * philips hasn't <riel> Shark45: nope <philips> riel, oh by the way, #kernelnewbies alumni are invited to play in the fork, right? <riel> philips: everybody can play <riel> philips: and when a patch is liked by people, it can be pulled into the "central" bitkeeper repository <riel> philips: ... if there is such a thing as a central repository with bitkeeper --> hklygre (~hklygre@tunnel-46-147.vpn.uib.no) has joined #kernelnewbies <Shark45> all: so you haven't. Never mind. I wrote a driver to try solving this problem. <philips> riel, marcelo is #kn alumnus <philips> ;-) <riel> philips: hehehe ;) <philips> riel, not to mention prumpf --> ajax (ajax@gh0st.net) has joined #kernelnewbies <philips> riel, freitag <philips> riel, jgarzil <philips> jgarzik <philips> jdike <philips> axboe <philips> hmm <philips> many other 'names' <philips> even.... maybe.... viro * philips thinks about that <wli> Was viro ever a newbie? =) <philips> riel, have we got a kernelnewbies@nl.linux.org? <velco> :) <-- liiwi has quit (Ping timeout: 182 seconds) <philips> wli, viro hung here <philips> wli, I have the logs <riel> philips: yes, the mailing list has around 580 members now IIRC ;) <wli> I've seen him from time to time. <deviant-> yeah, but he wasn't a newbie at the time :P <philips> wli, viro started hanging here when we started rewriting the vfs <velco> wli: some people are born gurus ;) <philips> riel, how do I join? <riel> philips: mail to majordomo or listar <wli> ouch, there's a mailing list I'm not on. <philips> riel, membership in the mailing list -> can add changesets, ok? * riel points everybody at http://mail.nl.linux.org/ <philips> riel, something like that? <wli> I get twitchy and break out in hives if I don't read at least 15,000 messages a day. <philips> riel, hmm <Shark45> all; but I think I faced a 2.2/2.4 compatability. What is this ksoftirqd_CPU0? It is "eating" all CPU. 2.2 did not show it on its process list. <philips> riel, I want to mail to listar, what do I say? <riel> philips: 'subscribe kernelnewbies <philips> riel, can add changesets -> must be on the mailing list <philips> riel, can add changesets -> must know the #kernelnewbies secret handshake <philips> riel, can add changesets -> must have found at least 2 bugs in somebody else's changeset <philips> riel, and supplied patches * philips subscribes <Diablo-D3> Heh <riel> philips: I think "can add changeset" should _never_ be an individual decision <Safari> Shark45: "it handles softirqs in a more predictable way by allowing the scheduler to take care of any softirq that can't conveniently be executed immediately. Among other benefits, this approach eliminated the need to check for and execute pending softirqs on exit from system calls." <philips> riel, right <riel> philips: but something at least 3 people agree with <philips> riel, add changeset means -> discussed on the channel, voted, no vetos <philips> riel, with a quorum present <philips> riel, right <riel> philips: these could be fairly random people, if sysadmin@foo.org says "hey, this patch fixes my system's problems" we probably want to include it <riel> philips: I don't think we need any cabal of priviledged decision makers * philips adds filters for kernelnewbies * philips confirms subscription <wli> Just did a :wq on my .procmailrc myself. <Diablo-D3> hah <Diablo-D3> kernel cabal <Shark45> Safari; I can't feel any performance difference with or without my driver installed. Does it behave like it was run under "nice -19" or so? <velco> heh, me too <philips> listar's headers are sucky <riel> Diablo-D3: which is what I'd like to prevent ;) * philips filters on anti-loop <riel> philips: what's wrong with List-ID ? ;) <-- rahul has quit (Ping timeout: 187 seconds) --> liiwi (~liiwi@ip212-226-133-178.adsl.kpnqwest.fi) has joined #kernelnewbies <Shark45> Safari: I mean ksoftirqd_CPU0 performs like it was running under "nice -19"? <Diablo-D3> heh <Diablo-D3> ksoftirqd iirc is the software irq affinity process. * riel vanishes for a few minutes ... bbl <wli> Confirmation has come back. <Diablo-D3> or... i could be totally full of it. <wli> I think it propagates softirq's or something. <Diablo-D3> heh <Diablo-D3> it does something <Diablo-D3> on my system it doesnt use any cpu time <sadie> are you guys going to create another major tree 'kernelnewbies' tree ? <Diablo-D3> so either my cpu can do whatever it needs without being told to... <Diablo-D3> or someone was running 2.4 on a 386 again <Shark45> Tha's why I think I am doing somethig wrong. <-- Cameron has quit (Remote closed the connection) <philips> riel, ok, list-id it is --> rahul (rjain@sid-1129.sid.rice.edu) has joined #kernelnewbies <philips> riel, problem: the admin traffice doesn't get picked up by the filter <philips> traffic <Shark45> Safari: my driver reschedule itself over tqimmediate task queue. <velco> what is zone balancing ? --- sadie is now known as sadie_Zz <Safari> Shark45: how nice. <Shark45> Safari: thanks! <fbl> anyone knows if struct super_block in super_block_read function have something inside * philips sends a hi to kernelnewbies@nl.linuxorg <fbl> ? <JALH> :) <wli> The kernel-doc stuff is pretty easy to deal with. <billh> wli ;-) <billh> Read your post last night. ;-) <-- Sebas22 has quit (Read error: 104 (Connection reset by peer)) <wli> billh: You couldn't see my prior one, right? <billh> when ? a couple of weeks ago ? --> jme_ (~jerome@MTL-HSE-ppp206081.qc.sympatico.ca) has joined #kernelnewbies <wli> billh: The one a couple of weeks ago, yeah. <wli> Nothing really significant changed. <billh> just bug fixes ? <wli> Exactly. * philips filters on return-path instead <philips> riel, the listar headers need work <philips> riel, a little thought wouldn't hurt <wli> Rolled up bugfix + repost with clean headers impossible to get clean within IBM. <philips> riel, where do I send the patch? <billh> sure. <philips> that was a rhetorical question <wli> Hrm, looks like I just got a -buttload- of orbz.org probes. --> bootleg (~bootleg@static.dsl.ll.net) has joined #kernelnewbies <-- Shark45 has quit ("Client Exiting") <-- todin has quit (Ping timeout: 181 seconds) <-- jme_ has quit ("ircII EPIC4-1.0.1 -- Are we there yet?") <billh> Hmm, folks poking at my sshd ports. <billh> port. --> todin (~todin@usr1729-har.blueyonder.co.uk) has joined #kernelnewbies <billh> Nov 15 13:11:32 gnuppy sshd[24011]: Did not receive identification string from 216.98.134.198. <billh> That's a bit unusual. --> Submarine (~david@d174.dhcp212-223.cybercable.fr) has joined #kernelnewbies <Submarine> Hi.
