Back to Home / #uml / 2007 / 05 / Prev Day | Next Day
#uml IRC Logs for 2007-05-31

---Logopened Thu May 31 00:00:21 2007
04:42|-|motp [~motp@] has joined #uml
04:43|-|aroscha [] has quit [Quit: aroscha]
05:12|-|motp [~motp@] has quit [Quit: Leaving]
05:25|-|tchan [] has quit [Read error: Connection reset by peer]
06:13|-|aroscha [~aroscha@] has joined #uml
07:08|-|krau [~cktakahas@] has quit [Read error: Operation timed out]
07:12|-|krau [~cktakahas@] has joined #uml
07:47|-|aroscha [~aroscha@] has quit [Quit: aroscha]
07:48|-|tchan [] has joined #uml
08:03|-|jdike [] has joined #uml
08:03<jdike>Hi guys
08:18<jdike>So, I was wondering why UML performance sucked so badly all of a sudden
08:18<jdike>a 6.5 minute kernel build was now taking 30
08:18<jdike>I checked both my main development trees - they both did the same thing
08:19<jdike>starting to wonder about the host kernel, since I've been playing with it
08:20<jdike>then I remembered booting it with 32M because this guy yesterday was complaining about that not working
08:21<jdike>sure enough, that's what was in my command history, and I was just finding the last UML boot command and running it without looking at it
08:28|-|aroscha [~aroscha@] has joined #uml
09:29|-|krau [~cktakahas@] has quit [Read error: Operation timed out]
09:40|-|aroscha [~aroscha@] has quit [Ping timeout: 480 seconds]
09:49|-|krau [~cktakahas@] has joined #uml
10:24|-|hfb [] has joined #uml
10:39|-|LemyDanger [~icechat5@] has joined #uml
10:45|-|aroscha [] has joined #uml
10:45<LemyDanger>is it better to compile the host with high mem supprot ? i have have 1 GB on the machine
10:47<jdike>32-bit host?
10:47<jdike>I guess, but 1G is pretty marginal
10:49<LemyDanger>unfortunately i cant upgrade
10:53|-|kokoko1 [~Slacker@] has joined #uml
10:53<kokoko1>Hello guys
10:54|-|hfb [] has left #uml [Leaving]
10:59[~]jdike puts printks in userspace code
10:59<jdike>that won't work so well
11:35<peterz>heh, just put a printk wrapper in your userspace libs :-)
11:59<kokoko1>Firstly don't panic! :-)
11:59|-|ram [] has joined #uml
12:02|-|aroscha [] has quit [Quit: aroscha]
12:09|-|LemyDanger [~icechat5@] has quit [Quit: Never put off till tomorrow, what you can do the day after tomorrow]
12:10|-|linux__alien [~linux__al@] has joined #uml
12:10<linux__alien>hello jdike
12:13<linux__alien>couldnt log on to my system for the past 2 days. I am not able to copy a file from my host kernel to UML using scp . Its unable to locate my file
12:13<linux__alien>but the file exists in the same path
12:13<jdike>what's the scp command?
12:16<linux__alien>scp .
12:16<linux__alien>this command i am giving it in UML
12:16<linux__alien> is the IP address of tap0 interface in host kernel
12:16<peterz>linux__alien: you seem confused
12:16<peterz>linux__alien: you gave 2 hosts
12:16<linux__alien>2 hosts how ?
12:17<peterz>the syntax is user@host:path
12:17<jdike>do you understand scp?
12:17<linux__alien>ve not used it at all
12:17<linux__alien>didnt get a chance to use it
12:21|-|linux__alien [~linux__al@] has quit [Remote host closed the connection]
12:23|-|linux__alien [~linux__al@] has joined #uml
12:24<linux__alien>instead of shutting down UML did that in my host kernel and my machine halted :-)
12:24<jdike>why are you on the host as root at all?
12:25<linux__alien>no actually in that shell alone i had logged in as root and this happened
12:26<linux__alien>from UML can i connect to the tap0 interface in the host?
12:26<linux__alien>scp balaji@ .
12:26<linux__alien>sh: connect to host port 22: Connection refused
12:26<linux__alien>i get this error
12:26<jdike>what normally causes that?
12:27<linux__alien>sorry SSH was not running started that
12:29<linux__alien>have a doubt wont i be in a position to execute telnet from inside UML?
12:30<jdike>is telnetd running on the host?
12:31<linux__alien>i dont have telnetd in my host
12:32<linux__alien>ve sshd though
12:32<jdike>well, telnetd running wherever you want to telnet to
12:33<linux__alien>you mean to say that i should have telnetd running wherever i want to telnet to is it
12:35<linux__alien>jdike, ok so now have couple of questions. I ve extracted my tar package but when i did ./configure it said no gcc and exited so now for me to install gcc is there any easy method here ? and where do i get that telnetd from
12:35<jdike>you know about yum?
12:36<linux__alien>ve used it also
12:36<linux__alien>but thats in host right ?
12:38<jdike>what do you know about yum?
12:39<linux__alien>with yum you can install rpm packages . The advantage of yum is that it checks for dependencies and installs the packages too from the internet
12:40<jdike>so, what do you think about using yum to install gcc?
12:40<linux__alien>yes ve done that too in my host but how do i do it in UML
12:40<linux__alien>infact i upgraded my Gcc in Fedora 6 using yum . yum update gcc
12:41<linux__alien>so you want me to compile the package in Host and move the binaries alone to UML is it?
12:41<linux__alien>is that what you mean by any chance?
12:42<jdike>you have a fedora filesystem booted in UML, right?
12:42<linux__alien>ok thats got yum just checked
12:42<linux__alien>it but somehow internet does not work in UML :-)
12:42<jdike>well, you should fix that
12:42|-|aroscha [] has joined #uml
12:45<linux__alien>Ok thanks jdike Will play with it. Thanks for your inputs.
12:50<linux__alien>have a nice day
12:50|-|linux__alien [~linux__al@] has quit [Quit: Leaving]
12:50<peterz>jdike: you're so patient
12:50<jdike>amazing cluelessness
12:51<peterz>yes, looks like a microsoftie's first encounter with a unix cli
12:51<jdike>and no ability to recognize sarcasm
12:52<peterz>he's from india
12:53<peterz>it might be a language/cultural barrier for sarcasm
12:54<jdike>the Brits say the same thing about us
12:54<peterz>us being usaians I take it :-)
12:55[~]peterz notes he lives on the same side of the big pond as those quaint brits
13:00[~]jdike sends off a batch of syslet demo patches
13:00<peterz>you've been playing with syslets?
13:01<peterz>to make uml go faster?
13:01<peterz>and did it?
13:01<jdike>patience, grasshopper
13:02<jdike>I wrote a couple small demos to understand the interface
13:02<peterz>I take it its mostly the batching of syscalls thats interresting
13:02<jdike>but that's interesting for later
13:03<jdike>the interesting part is that syslets can be used to collect status on all async events in which UML is interested
13:03<jdike>disk I/O
13:03<jdike>timer expirations
13:03<jdike>child process status
13:03<jdike>all in one loop, with no interrupts
13:04<jdike>the syscall batching is interesting as a skas4 mechanism
13:04<peterz>the extended epoll interface can do some of those too
13:04<peterz>they have signalfd and timerfds 'n stuff
13:05<jdike>yeah, but it can't automatically background something which is going to block
13:05<jdike>while immediately returning for something which is available now
13:05<peterz>right, but each uml process has its own thread, right?
13:05<jdike>right now, yes, but they won't use this
13:06<jdike>the kernel threads use it, and they need to worry about what's synchronous and what's not
13:06<peterz>right, and with these syslets thats fixed
13:06<peterz>is it bounded?
13:07<jdike>you mean the asynchronousness?
13:07<peterz>at some point spawning more syslets will not make things better
13:07<jdike>there's a fixed number of async threads
13:08[~]peterz obviously hasn't read all those syslet emails
13:08<peterz>perhaps I should
13:08|-|da-x [] has quit [Ping timeout: 480 seconds]
13:12<jdike>if I add new_mm() and switch_mm() system calls, I can use the batching mechanism to fiddle process address spaces with
13:13<jdike>batch( switch_mm(proc_mm), mmap, munmap, mprotect, etc, etc, switch_mm(my_mm) )
13:13<peterz>yeah, that'd save a bunch of ctx
13:14<jdike>and possibly a bunch of tlb flushes
13:14<peterz>right, those might be even more interesting
13:14<peterz>those suckers are expensive
13:14<jdike>but that would need some more surgery
13:16<jdike>but the attraction is that new_mm/switch_mm looks to me like a reasonable interface
13:16<peterz>yeah, you tried that before, right?
13:17<jdike>the batching allows the whole sequence to be automated in the kernel
13:17[~]peterz lost on
13:17<jdike>which is nice since doing anything immediately after switching address spaces is problematic
13:19|-|da-x [] has joined #uml
13:20<jdike>and off goes one bug fix in the async stuff itself
13:25<peterz>darn, I wrote myself an non obvious infinite loop
13:55<dgraves>peterz: awesome!
13:55<peterz>dgraves: yeah, some days.. :-)
13:56<peterz>basically: while (!(foo >> shift)) { shift += c; }
13:57<dgraves>how is that infinite? shouldn't foo left shifted by shift eventually = 0?
13:58<dgraves>right shifted
13:58<peterz>compiler translates that into sar %cl, %edx
13:58<peterz>and cl gets masked by 31
13:59<peterz>if 32 mod c != 0
13:59<peterz>you'll never get a full shift
13:59<dgraves>what's sar?
13:59<peterz>and hence never terminate
13:59<peterz>shift right
13:59<dgraves>peterz: wow. that is unobvious.
13:59<jdike>shift arithmetic right
14:01|-|weasel_ [] has joined #uml
14:01|-|weasel_ changed nick to Guest763
14:01|-|Netsplit <-> quits: weasel
14:01|-|Guest763 changed nick to weasel
14:12|-|newb_prog [] has joined #uml
14:15|-|flatronf700B [~flatronf7@] has quit [Ping timeout: 480 seconds]
15:09|-|hfb [] has joined #uml
15:12|-|hfb [] has left #uml []
15:14|-|Web-sidux473 [] has joined #uml
15:14|-|Web-sidux473 [] has quit [Remote host closed the connection]
15:20|-|aroscha [] has quit [Quit: aroscha]
15:27|-|newb_prog [] has quit [Quit: Leaving]
15:36|-|kokoko1 [~Slacker@] has quit [Ping timeout: 480 seconds]
16:05|-|aroscha [] has joined #uml
16:30|-|mjf [] has joined #uml
16:41|-|mjf [] has quit [Quit: leaving]
17:49|-|aroscha [] has quit [Quit: aroscha]
18:32|-|aroscha [] has joined #uml
18:35|-|aroscha_ [] has joined #uml
18:40|-|aroscha [] has quit [Ping timeout: 480 seconds]
19:04|-|jdike [] has quit [Quit: Leaving]
20:20|-|ram [] has quit [Ping timeout: 480 seconds]
22:25|-|newb_prog [] has joined #uml
22:49|-|newb_prog [] has quit [Quit: Leaving]
22:59|-|VS_ChanLog [] has left #uml [Rotating Logs]
22:59|-|VS_ChanLog [] has joined #uml
23:16|-|ElectricElf [] has quit [Ping timeout: 480 seconds]
23:20|-|ElectricElf [] has joined #uml
---Logclosed Fri Jun 01 00:00:56 2007