IRC log started Thu Aug 19 00:00:00 1999 For multitasking... how do I make the OS premtive? Just have an interupt every time slice that switches the currnet task? yep... when the thread's quantum runs out just save it's register state and restore the register state of some other task Ok [msg(TUNES)] permlog 1999.0819 quantum = time slice? yep it's easier to type ok yes :) were either of you the person with whom i discussed whether or not hw-based tasking vs sw-based tasking was more efficient a few weeks ago? twasn't me nope... I think I talked to you about that a long time ago though it's just that i found an os developers' web site with a definite answer to all those kinds of questions i wish i'd had that a month ago :) i had to go around harassing people to get the answer Oh! Cool! What is the answer? sw but it requires more coding care to get it done that way _thank you_ I was about to write my kernel using hardware... those TSSes sure looked funky... but you can eliminate managing lots of TSS's also which s a bonus the TSSes look bloated since you need a GDT entry for every TSS abi: osweb? water: i haven't a clue abi: os web? os web is the best place to find operating system information at http://osweb.8m.com/ you still have to multiplex one TSS a little, though So what needs to be saved between tasks? just register and stack state? segment registers, general purpose registers, cr3 - the page directory base register optionally, the floating point registers as well lots of stuff Do I need to save _every_ register though? you have to save any registers you expect tasks to be using locally So those exotic cache registers like tr4 can be left out? 12:10am yah Is it best to always run in V86 mode or sould I switch in and out of that with each task if the taks needs it? no experience with v86 Doesn't linux run in v86? i know linux has a system call specifically for switching into vm86 mode Ohh er v86 So linux runs in protected mode until a program needs v86 linux itself is always in protected mode -:- overfien [overfiend@m243.telcomplus.net] has joined #tunes y0 y0 y0h only a process can go into v86 mode the fiend! -d werd ;) Ok So a message is what exactly... if thread A wants to tell thread B to do something how does that happen? usually just some piece of data how that data is transferred is up to your implementation the only useful characteristic of messages is that they can cross address spaces or any other nasty boundaries If I am not using a uK are messages still nesscary? if you want process to be able to communicate the message idea also allows processes and threads to synchronize that's not necessarily inherent to messages themselves which means that switching tasks can be optimized for the kernel and app co-operatively or otherwise So apps have there own data for messages, the message handler just carries what ever data the thread says? it's just the fact that they can use communication to notify other processes of their state well, you didn't offer analternative with the same semantics pretty much So thread A tells the kernel, hey tell thread B over there xyzpdq, and thread B will know to check its zipper. The kernel dosnt care? if you mean the kernel shouldn't care about the contents of the message, then yes 12:20am but most of the time it's a two way transaction thread B has to agree to receive messages before it can get any from thread A -:- SignOff water: #TUNES (Ping timeout for water[ppp-tnt-85.tscnet.net]) unless you have things like signals, where you still have to give the kernel a signal handler (in unix) So whats the best way to get kernel services to execute? I mean for apps to talk to the kernel. not quite sure there i'm only aware of the various ways you can do it, not necessarily which is better What ways can you do it? you can use call gates (inter segment calls), interrupts or at a more extreme case, maybe even dynamically linking applications into the kernel Ahh... why on earth would you dynamicly link to the kernel?? Isnt that asking for instablity? i think that's what Win9x might do :) Yes right now i'm thinking of using interrupts Interrupts don't have that much overhead right? they have a considerable amount compared to a normal call but i think it's unavoidable Thats what I was thinking either way return address has to get put on the stack and such and plus in a system call you usually have to change the privilege level Very true So any way you go you have overhead unless you do an exokernel but you still have the system call overhead there, yep Whats an exokernel? an exokernel is a system with no run-time kernel or see pdos.lcs.mit.edu exokernels simply multiplex hardware I don't like that idea they're more or less extreme microkernels that try to provide as little functionality as possible, only what's necessary so that you can write more kernelly services on top Were the people that came up with uKs on something? I don't understand why this uK is a _good_ thing there's very little reason actually :) 12:30am only reason i'm making a microkernel is it's simple :) if i was really involved and trying to make something useable i'd go with a modular kernel Ok but i don't claim to be an expert, i'm just a neophyte myself But you are quite knoladgeable but have yet to actually utilize a lot of this knowledge :) still only about half way done with my kernel Thats futher then I am been spending the last two months just reading kernel code and then soon as i find an answer, i find 30 more questions That sounds like whats happning over here Thats why I have to bug all you guys for answers ;) i'm still doing that :) yes I appreciate all the help I get here very much. If I hadn't of talked to you guys Raven would be a real mode os... ala dos? It would be very similar to dos 12:40am -:- water [water@ppp-tnt-31.tscnet.net] has joined #tunes How is Netscape different from Mozilla? mozilla crashes more Heh the Raven OS page yours? I didn't make the page, but its for my os i find this mildly funny, "utilizes paging to access memory quickly and efficiently" :) 12:50am -:- SignOff lar1: #TUNES (Ping timeout for lar1[dialup-209.245.138.19.SanJose1.Level3.net]) -:- lar1 [LARMAN@dialup-209.245.128.26.SanJose1.Level3.net] has joined #tunes Stupid windows! bsod? well, bsod is a feature invented by Microsoft abi: pioneered Not quite... Invalid page fault in Kernel32.dll | Unable to load Kernel32.dll Reinstall windows | black screen mmm... dangling pointers 01:00am in the kernel no less Whats so funny about that paging comment? paging is the antithesis of efficient you use it for many things, but surely not efficiency Its better then segmentation but neither segmentation nor paging is more efficient than either So then how is paging the antithesis of efficient? because... every memory address you use within the address space is converted to a physical address or rather, needs to be the initial conversion takes some time after that you have a small cache of commonly used conversions True and context switching empties that cache then you have the memory resources used by managing page tables... but then you have the cool parts: shared memory (mapping memory), memory protection, paging to disk, etc... But both their isn't a really efficent memory access method on the x86 Yes yes there is use an unprotected flat model rather than a protected flat model you can support effectively one process with nothing preventing the process from screwing with the kernel but memory accesses would be fast note my sarcasm :) Yes Most of us like to refer to that as windows no... that's more dosish it's hard to do multitasking without paging at all Dos didn't multitask it could have 01:10am very simply... there were a few multitaskers for it but you had nothing protecting the tasks from eachother... which is a downside I just invision this pice of paper 'a task' that has a mean look on its face chacing another peice of paper with a crayon :D nah... more like one paper doing some useful work while the other task is drawing a kick me sign on it's back :) Ohh I like that and maybe even poking it with the crayon But still with a crayon... a _green_ crayon snot green! * water/#tunes wonders how far the metaphor can be stretched Yes! Muahaha pretty far I rather like it. ack... that pentium optimization manual is giving me headaches after-the-fact hehe heh * lar1/#tunes thinks he could be coding now err, should be coding should be sleeping both very good options -:- lar2 [LARMAN@dialup-209.245.128.26.SanJose1.Level3.net] has joined #tunes 01:20am Stupid ghost lar1, meet lar2 :) -:- NetSplit: bear.openprojects.net split from varley.openprojects.net [01:22am] -:- BitchX+Deb1an: Press Ctrl-F to see who left Ctrl-E to change to [bear.openprojects.net] I whish IRC had some fast way of killing ghosts -:- lar2 is now known as lar1 I got this really cool idea today Apps should beable to request a larger quantum or even no quantum like if app a needed to do something big, it should beable to ask the kernel to make the next cycle of apps 1 second quantums instead of 500 ms the better solution is usually to give tasks priorities Yes... like I have nothing to do leave me alone! That just moves higher priority apps up higher on the que, right? but both can me complementary :) er be yep but sometimes you want tasks to run until they're ready to give up i.e. they have no quantum, and only by say, doing a system call and entering into the kernel will they be preempted/switched out like in some driver process, if you were doing a microkernel But if that program crashed... the system would go with it no it wouldn't not necessarily higher priority tasks get run before that program and crashing is not a big issue I see what you mean as if it truly crashed, you could terminate the task if it say... gave itself an indefinite quantum, and just sat there not making any sort of system call and there were no higher priority tasks to preempt it, you might have a problem 01:30am so you'd probably want to make sure only privileged tasks can completely remove their quantum But if the kernel checked every once and a while, it could preempt the stupid task sure... but what if you have a driver that absolutely CANNOT except a preemption or else it will screw up the work it's doing and perhaps compromise the system But that kind of thing should be stable ;) not checking at all is probably the better solution Yes since you will usually always have drivers that run at higher priority than any of the user level applications Kinda like standing by a bathroom door... 'done yet? 'no' and as soon as a hardware event triggers that driver to preempt it, the offender gets switched out yep except you have a bouncer in the bathroom it'll only let people in if they're cool enough :) heh and if their is simply someone cooler than the person taking a dump 'but I _really_ have to go!!' they get kicked out good deal and sent to the end of the line so other people equally cool as the one kicked out can get in afterwards So dynamic quantums and prioritys together? Or just prioritys... both Cool, so my idea was right I am so proud of myself! basically, you just alternate between tasks in the same priority level and the quantum's ensure that alternation within the same priority level happens unless the process explicitly doesn't want it to Yes My stupid boot block is broken i don't even have one... so consider yourself lucky :) It gets to protected mode and then I am pretty sure it hangs It won't put the letter a in the corner of the screen like I so politely ask it why not get rest of the os written first? :) 01:40am Because its good to wirte the boot block (or at least part of it) done first. If not, how can you test your code? but what is there to test? it's like having a corkscrew without a bottle of wine Kernel functions Its like having a corkscrew with an empty bottle of wine but again... if the kernel itself isn't written really then testing is pointless empty bottle of wine without a cork :) and maybe split down the seems because it hasn't been glued :) But you do test the kernel as you go along, right? nope i'm writing a lot of untested code How do you know it works? and then when it's all written i'll have a whole untested kernel and all that's left is to write the bootsector and debug it will, once it's debugged Oh is that all? well umm, writing the kernel is the big part which is the part i'm doing right now true What are you writing your kernel in? assembly... i'm probably going to rewrite it in C afterwards, i only grudgling went to assembly since C was awkward in some respects I am using asm for speed And I like to know exactly whats going on that's the wrong reason uh oh your compiler will usually generate code as fast as you :) Not always sometimes it screws up i'm only using assembly because it's easier to represent the concepts of interfacing with the hardware as interfacing with the hardware even so... it's the wrong reason compilers eventually get better even though the code itself stays the same whereas assembly code can never proactively get better it always sucker, for better or worse unless you change it er it always sucks But saying your OS is in pure asm is alot more impressive to the unsuspecting layman... umm... bs i'm not targetting my OS at any layman i'm not even building it for the purpose of releasing it to other people Just for you? i'm doing it for myself, and therefor the only reasons i have are personal ones i'm not necessarily trying to impress anyone, just build something sound and working 01:50am and elegant, if possible No I'm not saying that I am only to impress people. That totally misses the mark of Raven's purpose then what are your real reasons for using assembly? :) At the time I thought it would be faster I decided this a year or so agao but what about now? That and I had no real C compiler Now? I like having low level control... if something goes wrong I want to know who what where when and why. Not like gosh, function blah failed. I guess I am beging to think I have no real reason other then I like it ;) that's another bad reason as you can know where something went wrong in any language and in assembly, you don't even have notions of grouping whereas in C, you atleast have the function I know... I guess I have no valid reason Whats wrong with address dw 0 | mov address, bx -:- SignOff Plundis: #TUNES (Ping timeout for Plundis[chaosdev.org]) -:- SignOff Crimson: #TUNES (Ping timeout for Crimson[chaosdev.org]) well... wouldn't "address: dw 0" get executed? seeing as it's right in the instruction streams er wait... mov [address], bx it what you be wanting address is just a little temp place that the physical address of a function goes bottom line is I need to put 0x8 in CS and the physical address of a label named hang in IP address itself is just an immediate address Ah, very true it's not like in C :) 02:00am Well, I should be off to bed _Thanks_ _much_ -:- SignOff lar1: #TUNES (Bye Bye) 02:10am -:- Plundis [plundis@chaosdev.org] has joined #tunes * Plundis/#tunes is gone. Gone since Tue Aug 3 18:38:00 1999 -:- Crimson [crimson@194.17.41.8] has joined #tunes * Crimson/#tunes is gone. Gone since Sat Aug 14 22:23:00 1999 -:- SignOff eihrul: #TUNES (Leaving) -:- SignOff Plundis: #TUNES (Ping timeout for Plundis[chaosdev.org]) -:- SignOff Crimson: #TUNES (Ping timeout for Crimson[194.17.41.8]) -:- Crimson [crimson@chaosdev.org] has joined #tunes * Crimson/#tunes is gone. Gone since Sat Aug 14 22:23:00 1999 -:- Plundis [plundis@chaosdev.org] has joined #tunes * Plundis/#tunes is gone. Gone since Tue Aug 3 18:38:00 1999 -:- SignOff overfien: #TUNES (Ping timeout for overfien[m243.telcomplus.net]) -:- SignOff fare: #TUNES (Ping timeout for fare[quatramaran.ens.fr]) -:- hcf [nef@me-portland-us1045.javanet.com] has joined #tunes oh hmm 04:50am -:- AlonzoTG [Alonzo@216-164-241-195.s195.tnt10.lnh.md.dialup.rcn.com] has joined #tunes hoy AlonzoTG 06:40am -:- hcf_ [nef@me-portland-us941.javanet.com] has joined #tunes -:- SignOff hcf: #TUNES (Ping timeout for hcf[me-portland-us1045.javanet.com]) -:- hcf_ is now known as hcf -:- smkl [sami@MCCCLVIII.rdyn.saunalahti.fi] has joined #tunes hoy smkl hello hcf 08:30am -:- SignOff smkl: #TUNES (Ping timeout for smkl[MCCCLVIII.rdyn.saunalahti.fi]) -:- smkl [sami@CMX.rdyn.saunalahti.fi] has joined #tunes -:- NetSplit: sterling.openprojects.net split from irc.linux.com [09:56am] -:- BitchX+Deb1an: Press Ctrl-F to see who left Ctrl-E to change to [sterling.openprojects.net] -:- Netjoined: sterling.openprojects.net irc.linux.com -:- Crimson [crimson@chaosdev.org] has joined #Tunes -:- Plundis [plundis@chaosdev.org] has joined #Tunes -:- AlonzoTG [Alonzo@216-164-241-195.s195.tnt10.lnh.md.dialup.rcn.com] has joined #Tunes !sterling.openprojects.net!! Received :irc.linux.com SERVER hogan.openprojects.net from clarke.openprojects.net !?! -:- fare [fare@quatramaran.ens.fr] has joined #Tunes -:- ElGato [dave@bespin.dhs.org] has joined #tunes * ElGato/#tunes is away: (Auto-Away after 10 mins) [BX-MsgLog On] -:- TB [chayne@p20-max20.chc.ihug.co.nz] has joined #tunes Howdy hi Wooo, a reply :) yep 10:50am I thought i'd come and see what Tunes is about seeing as my site got linked on the page :) what's your site? Amiga.org (which is semi down for a few days) wowsers Under the listing of AmigaOS is simply my/our site :) So I though well, time to spread the word a little better :) wowsers? uh...well im not the best person to represent the project but im the most active alot of people like the amiga here And so they should :) :D The new AmigaOS is to be released this weekend cooly Bout damn time and all ;) freeware? Nope, but its cheap. :) :/ After that the OS dev is slowed down/stoped and the OE dev is put to the front OE? Operating Enviroment oh well I like to think we're an os project but the tunes project puts language and os together or under the philosophy, they already are together just not enough right -- see the FAQ so you end up with funny looking os's :) hi fare Interesting fare is the unadmitately head honcho of the project Hmmm (FAQ), some interesting ideas. pardon my spelling spelling pardoned :) :D Ive been wokring on OS ideas myself for a long time, but do not have much time for it job? Amiga.org <-- Job :) Being an AMigan is a fulltime thing :) oh -:- SignOff hcf: #TUNES (Ping timeout for hcf[me-portland-us941.javanet.com]) Big words in the FAQ :) yay, i didn't know tril and i share the same first name :) 11:00am TB: yes, very indimidating to people including me I understand long words, I just cant spell them :) heh try reading brian's arrow paper The Tunes thing has many of the things the new AmigaOE has. your brain will emlt melt rather I think I might have to read up on the definition of some of the tunes things, they seem interesting not to mention some are very familar to me I just love this kinda crap :) :) you're welcome to help out if you ever have any time Okay so let me gets this stright. The idea is to make a new os with all the good stuff from the others to make a mega cool OS? under the tunes philosophy everyone's resources are welcome and hinder the project in no way TB: no sorta the idea of tunes is to create an os the fixes all the bad things with computing systems already around and adding even more stuff not unify them all Ahhh, I see. or better yet just get rid of bad ideas although for that you can check out unios Im on the track now :) Hmmm, Amiga would love some of the ideas :) TB: when we're done we can port it :D hehe :) Port? Naaa, the new OE runs on any platform. amiga is an architecure right? Yes and No. Amiga is fed up with the way hardware and especialy OS's are done, so they've started making a new one but not for PCs, for Information Devices. It has advanced objects, code that works on anything even different cpus, highspeed multimedia, transparent distributed clustering, wireless networking, a better interface, and an OE (not OS) that spans across machines not limited to just one. And more of course.... but my fingers are tired ;) 11:10am hum fare: how is IRCnet ? dunno