From: <pie...@mp...> - 2003-10-25 13:26:55
|
Hello all, Has anyone tried to implement user-mode-linux (skas patch and um architecture) on an Itanium IA-64 machine ? //// (o o) ===============oOO==(_)==OOo============= Pierre Gentile |
From: Jeff D. <jd...@ad...> - 2003-10-25 17:03:00
|
pie...@mp... said: > Has anyone tried to implement user-mode-linux (skas patch and um > architecture) on an Itanium IA-64 machine ? Not really. I have an IA64 box sitting next to me, and starting a port to it is on my list once I acquire some free time. Jeff |
From: Matt Z. <md...@de...> - 2003-10-26 04:41:00
|
On Sat, Oct 25, 2003 at 01:03:22PM -0400, Jeff Dike wrote: > pie...@mp... said: > > Has anyone tried to implement user-mode-linux (skas patch and um > > architecture) on an Itanium IA-64 machine ? > > Not really. I have an IA64 box sitting next to me, and starting a port to > it is on my list once I acquire some free time. I did some preliminary investigation into making ia64 UML-friendly as a host (the changing syscall numbers bit), and after discovering why it won't work, I sent pleas for help to some HP folks to find out if it would be easy to fix. I did get one interested response, but no reply to my followup as yet. -- - mdz |
From: Michael R. <ra...@st...> - 2003-10-26 05:26:38
|
I've come across a problem with running a user-mode-linux instance with RR-Scheduling. When I attempt to start a server with chrt the host system panics. I can start a uml in normal (other) scheduling, and then change it to RR-Scheduling and it works fine, until I try to stop the uml... then I get the same kernel panic. I'm using the skas-3 patch on a 2.4.21 kernel, and 2.4.21-6um. If you need the specific output of the kernel panic, I can get it... Just as a related question... how would different scheduling the different 'threads' of the UML effect performance? As far as I know.. the first 'linux' process actually does all the work and the extra processes are for the different IO the UML does, eg network, block devices etc? Would increasing the priority of the IO over the priority of the main process have an effect on network performance for example? Michael Ralston |
From: Jeff D. <jd...@ad...> - 2003-10-26 19:07:37
|
ra...@st... said: > I've come across a problem with running a user-mode-linux instance > with RR-Scheduling. When I attempt to start a server with chrt the > host system panics. I can start a uml in normal (other) scheduling, > and then change it to RR-Scheduling and it works fine, until I try to > stop the uml... then I get the same kernel panic. Any time the host panics, it's time to send the details to LKML. > how would different scheduling the different 'threads' of the UML > effect performance? As far as I know.. the first 'linux' process > actually does all the work and the extra processes are for the > different IO the UML does, eg network, block devices etc? Would > increasing the priority of the IO over the priority of the main > process have an effect on network performance for example? In skas mode, the first process runs the kernel, the second runs processes, and the other two handle SIGIO emulation and SIGWINCHes. So, those two don't matter, and first two are never running at the same time, so their relative priorities don't matter. Jeff |
From: Nuno S. <nun...@vg...> - 2003-10-26 21:23:43
|
Hi! Jeff Dike wrote: [..] > In skas mode, the first process runs the kernel, the second runs processes, > and the other two handle SIGIO emulation and SIGWINCHes. So, those two don't > matter, and first two are never running at the same time, so their relative > priorities don't matter. > > Jeff > Wouldn't it be nice if the threads updated their argv[0] to something like "argv[0]-kernel" "argv[0]-userspace" "argv[0]-sigio" "argv[0]-sigwinch"? I would send a patch but my first attempt didn't work... Can this be done? Maybe renicing the threads could help in environments where you have lots of UMLs running? Like having the kernel runnig at -10 and userspace at 19 and sigio at 19? Maybe this could make the "time" inside UML more consistent in high loads? Comments? Regards, Nuno Silva |
From: BlaisorBlade <bla...@ya...> - 2003-10-29 19:10:41
|
Alle 22:23, domenica 26 ottobre 2003, Nuno Silva ha scritto: > Hi! > > Wouldn't it be nice if the threads updated their argv[0] to something like > "argv[0]-kernel" > "argv[0]-userspace" > "argv[0]-sigio" > "argv[0]-sigwinch"? > > I would send a patch but my first attempt didn't work... Can this be > done? > Maybe renicing the threads could help in environments where you > have lots of UMLs running? Like having the kernel runnig at -10 and > userspace at 19 and sigio at 19? Avoid nice=19 since there have been some notable patches in 2.5(even present in 2.4-ck tree) to make such task run only in place of the idle task. And having the kernel running at -10 means having high permissions. Adding an option to make UML switch to some UID after startup is not difficult. But first we need to know until which point it needs root priviledges. And probably, capabilities are just enough. And at some point, in 2.6 there will be setcapability extended attributes(it's not hard now that extended attrs work). > Maybe this could make the "time" inside > UML more consistent in high loads? > > Comments? > > Regards, > Nuno Silva -- cat <<EOSIGN Paolo Giarrusso, aka Blaisorblade Linux Kernel 2.4.21/2.6.0-test on an i686; Linux registered user n. 292729 EOSIGN |
From: Nuno S. <nun...@vg...> - 2003-10-30 00:39:50
|
Hello! BlaisorBlade wrote: > Alle 22:23, domenica 26 ottobre 2003, Nuno Silva ha scritto: > >>Hi! >> >>Wouldn't it be nice if the threads updated their argv[0] to something like >>"argv[0]-kernel" >>"argv[0]-userspace" >>"argv[0]-sigio" >>"argv[0]-sigwinch"? >> >>I would send a patch but my first attempt didn't work... Can this be >>done? >>Maybe renicing the threads could help in environments where you >>have lots of UMLs running? Like having the kernel runnig at -10 and >>userspace at 19 and sigio at 19? > > Avoid nice=19 since there have been some notable patches in 2.5(even present > in 2.4-ck tree) to make such task run only in place of the idle task. And > having the kernel running at -10 means having high permissions. > Adding an option to make UML switch to some UID after startup is not > difficult. But first we need to know until which point it needs root I wasn't talking about running UML as root. Just an easy way to identify each thread. After a script (maybe run by root, maybe not) can easily spot which thread does what we could renice the process, remove linux capabilities, killall -STOP linux-sigio, etc... ;) > priviledges. And probably, capabilities are just enough. And at some point, > in 2.6 there will be setcapability extended attributes(it's not hard now that > extended attrs work). > >>Maybe this could make the "time" inside >>UML more consistent in high loads? >> This would be really good! :) Regards, Nuno Silva >>Comments? >> >>Regards, >>Nuno Silva |