From: Blaisorblade <bla...@ya...> - 2007-04-28 12:27:49
|
On gioved=EC 26 aprile 2007, Herbert Volkmann wrote: > Hi Blaisorblade, > > thanks for answering that fast: Yes, and sorry, but I'm short on time and I do not want private emails -=20 please reply on the list (on uml-user actually). So anybody else can contin= ue=20 answering. I'll just say to try a more recent kernel first. Second suggestion, which is more probably the correct one, where is the UML= =20 RAM file? It goes in /tmp or in /dev/shm usually (lsof, or ls /proc/<host U= ML=20 pid>/fd can show what happens - or even df on the host). If the host has 1G= =20 of RAM, by default a tmpfs mount is 512MB wide. If that fills, UML crashes= =20 with a characteristic error message (IIRC 'Kernel mode signal N', where N i= s=20 the numeric code of SIGBUS). But maybe, since you append con=3Dnull, you get no kernel messages at all..= =2E and=20 that's bad from you. > - Filesystem is ext3 on host and uml > - the things you guessed were right > - it's a 32bit system > - uml kernel release 2.6.18, exactly (apt-cache show user-mode-linux): > Architecture: i386 > Version: 2.6.18-1um-2 > > I do not write to tmpfs, i execute the following command in /root > dd if=3D/dev/zero of=3Dtest.img bs=3D1024k count=3D512 > right before ending dd, uml crashes, all uml processes end working which > i can see on host system by executing ps axuf. > > what is also interesting is, that i have 256M swap which is never used > by uml (fstab shows: > /dev/ubd1 none swap defaults 0 0 > ) > > and i have the same effect when i copy with scp FROM the UML Machine to > my Host System. cache is increasing, memory goes down and shortly before > memory is gone, uml stops working. > > Any other suggestions from your side? > > Thanks for supporting me with this strange behavior, > Best Herbert > > Am Donnerstag, den 26.04.2007, 16:17 +0200 schrieb Blaisorblade: > > On gioved=EC 26 aprile 2007, Herbert Volkmann wrote: > > > Hi List, > > > > > > i ran in the following problem: > > > > Very strange... > > > > > when i produce files (with dd) or copy files to or from the uml > > > machine, > > > > Which filesystems are involved? Can you give a sample command? > > > > > i can see (with top), > > > > Run on the guest (i.e. UML) I guess, right? > > > > > that the used memory and the cached things > > > increases, while the unused memory will go down. > > > > That is reasonable, but the cache should not make the system stop. > > > > > shortly before memory > > > will come to zero, the uml stops working completely without any > > > messages in the /var/log/* files on uml or host system. > > > > > > if i do in between a "echo 3 > /proc/sys/vm/drop_caches" > > > > On the guest (i.e. UML) I assume, right? > > > > > it cleans caches, frees memory and works fine! but i have to do this > > > all 10 seconds when i use dd for producing large files ;-( > > > > > > these are the components i use: > > > - debian etch on host with debian kernel patched with skas3 from debi= an > > > > 32bit system, right? > > > > > - 1024 MB RAM on host > > > - RAID5 System with 3ware Controller (5x2G) almost 1 TB > > > - on uml i use debian etch, with bridged networking, 512MB RAM for uml > > > > Which UML kernel release? That's the most important thing. > > > > > - 6G root_fs for uml and 256M swap > > > > > > here is the command with which i start the uml: > > > sudo -u uml /usr/bin/linux.uml umid=3Duml mem=3D512M eth0=3Dtuntap,ta= p1 > > > ubd0=3D/usr/media/uml-linux/root_fs_2007_04_23 > > > ubd1=3D/usr/media/uml-linux/swap_fs con=3Dnull root=3D/dev/ubda uml_d= ir=3D/tmp/ > > > > > > I would be really happy if someone could tell me what i can test more > > > to find out why this system is instable with i/o operations and cache= s. > > > > I have only a doubt: if you are filling a shmfs/tmpfs filesystem (which > > is backed by ram), that's an almost normal behaviour, and can happen ev= en > > on host systems. > > > > > thanx for any reply, > > > Herbert =2D-=20 Inform me of my mistakes, so I can add them to my list! Paolo Giarrusso, aka Blaisorblade http://www.user-mode-linux.org/~blaisorblade |
From: Herbert V. <sys...@au...> - 2007-04-29 08:21:32
|
Hi all, refering to the actual mail correspondence, I can give the following update: on host i have mounted (not by fstab, i think thats automatically done by system) two tmpfs systems (why two systems?) tmpfs 507M 0 507M 0% /lib/init/rw tmpfs 507M 437M 70M 87% /dev/shm it's very interesting, that /dev/shm is filled to 87%. lsof /dev/shm shows, that it is used by uml system. but in uml if i do free, i can see Mem: 512908 (total) 198704 (used) 314204 (free) =20 Actually my question is, why do i have two tmpfs mounts and why is one filled up to 87%, when uml is using only less than 50%? I hope someone can help me with this stuff... Best, Herbert Am Samstag, den 28.04.2007, 14:27 +0200 schrieb Blaisorblade: > On gioved=EC 26 aprile 2007, Herbert Volkmann wrote: > > Hi Blaisorblade, > > > > thanks for answering that fast: >=20 > Yes, and sorry, but I'm short on time and I do not want private emails = -=20 > please reply on the list (on uml-user actually). So anybody else can co= ntinue=20 > answering. >=20 > I'll just say to try a more recent kernel first. >=20 > Second suggestion, which is more probably the correct one, where is the= UML=20 > RAM file? It goes in /tmp or in /dev/shm usually (lsof, or ls /proc/<ho= st UML=20 > pid>/fd can show what happens - or even df on the host). If the host ha= s 1G=20 > of RAM, by default a tmpfs mount is 512MB wide. If that fills, UML cras= hes=20 > with a characteristic error message (IIRC 'Kernel mode signal N', where= N is=20 > the numeric code of SIGBUS). >=20 > But maybe, since you append con=3Dnull, you get no kernel messages at a= ll... and=20 > that's bad from you. >=20 > > - Filesystem is ext3 on host and uml > > - the things you guessed were right > > - it's a 32bit system > > - uml kernel release 2.6.18, exactly (apt-cache show user-mode-linux)= : > > Architecture: i386 > > Version: 2.6.18-1um-2 > > > > I do not write to tmpfs, i execute the following command in /root > > dd if=3D/dev/zero of=3Dtest.img bs=3D1024k count=3D512 > > right before ending dd, uml crashes, all uml processes end working wh= ich > > i can see on host system by executing ps axuf. > > > > what is also interesting is, that i have 256M swap which is never use= d > > by uml (fstab shows: > > /dev/ubd1 none swap defaults 0 0 > > ) > > > > and i have the same effect when i copy with scp FROM the UML Machine = to > > my Host System. cache is increasing, memory goes down and shortly bef= ore > > memory is gone, uml stops working. > > > > Any other suggestions from your side? > > > > Thanks for supporting me with this strange behavior, > > Best Herbert > > > > Am Donnerstag, den 26.04.2007, 16:17 +0200 schrieb Blaisorblade: > > > On gioved=EC 26 aprile 2007, Herbert Volkmann wrote: > > > > Hi List, > > > > > > > > i ran in the following problem: > > > > > > Very strange... > > > > > > > when i produce files (with dd) or copy files to or from the uml > > > > machine, > > > > > > Which filesystems are involved? Can you give a sample command? > > > > > > > i can see (with top), > > > > > > Run on the guest (i.e. UML) I guess, right? > > > > > > > that the used memory and the cached things > > > > increases, while the unused memory will go down. > > > > > > That is reasonable, but the cache should not make the system stop. > > > > > > > shortly before memory > > > > will come to zero, the uml stops working completely without any > > > > messages in the /var/log/* files on uml or host system. > > > > > > > > if i do in between a "echo 3 > /proc/sys/vm/drop_caches" > > > > > > On the guest (i.e. UML) I assume, right? > > > > > > > it cleans caches, frees memory and works fine! but i have to do t= his > > > > all 10 seconds when i use dd for producing large files ;-( > > > > > > > > these are the components i use: > > > > - debian etch on host with debian kernel patched with skas3 from = debian > > > > > > 32bit system, right? > > > > > > > - 1024 MB RAM on host > > > > - RAID5 System with 3ware Controller (5x2G) almost 1 TB > > > > - on uml i use debian etch, with bridged networking, 512MB RAM fo= r uml > > > > > > Which UML kernel release? That's the most important thing. > > > > > > > - 6G root_fs for uml and 256M swap > > > > > > > > here is the command with which i start the uml: > > > > sudo -u uml /usr/bin/linux.uml umid=3Duml mem=3D512M eth0=3Dtunta= p,tap1 > > > > ubd0=3D/usr/media/uml-linux/root_fs_2007_04_23 > > > > ubd1=3D/usr/media/uml-linux/swap_fs con=3Dnull root=3D/dev/ubda u= ml_dir=3D/tmp/ > > > > > > > > I would be really happy if someone could tell me what i can test = more > > > > to find out why this system is instable with i/o operations and c= aches. > > > > > > I have only a doubt: if you are filling a shmfs/tmpfs filesystem (w= hich > > > is backed by ram), that's an almost normal behaviour, and can happe= n even > > > on host systems. > > > > > > > thanx for any reply, > > > > Herbert >=20 >=20 >=20 > --=20 > Inform me of my mistakes, so I can add them to my list! > Paolo Giarrusso, aka Blaisorblade > http://www.user-mode-linux.org/~blaisorblade |
From: Jeff D. <jd...@ad...> - 2007-04-29 15:00:22
|
On Sun, Apr 29, 2007 at 10:21:08AM +0200, Herbert Volkmann wrote: > on host i have mounted (not by fstab, i think thats automatically done > by system) two tmpfs systems (why two systems?) > tmpfs 507M 0 507M 0% /lib/init/rw > tmpfs 507M 437M 70M 87% /dev/shm /dev/shm is pretty universal now, I think. I've never seen /lib/init/rw before. > it's very interesting, that /dev/shm is filled to 87%. lsof /dev/shm > shows, that it is used by uml system. but in uml if i do free, i can see > Mem: 512908 (total) 198704 (used) 314204 (free) Used tmpfs pages will be anything that the UML has ever touched. Those pages were once in use, but may be free as far as the UML is concerned. So there's no particular correlation between free tmpfs space and free memory inside the UML. Jeff -- Work email - jdike at linux dot intel dot com |
From: Blaisorblade <bla...@ya...> - 2007-04-29 21:09:26
|
On domenica 29 aprile 2007, Herbert Volkmann wrote: > Hi all, > > refering to the actual mail correspondence, I can give the following > update: > > on host i have mounted (not by fstab, i think thats automatically done > by system) two tmpfs systems (why two systems?) > tmpfs 507M 0 507M 0% /lib/init/rw Probably used by some of the scripts of your distro. I have no idea about what exactly (I can just guess), but you can leave that alone. > tmpfs 507M 437M 70M 87% /dev/shm > > it's very interesting, that /dev/shm is filled to 87%. lsof /dev/shm > shows, that it is used by uml system. but in uml if i do free, i can see > Mem: 512908 (total) 198704 (used) 314204 (free) > > Actually my question is, why do i have two tmpfs mounts and why is one > filled up to 87%, when uml is using only less than 50%? On this one Jeff already answered. However, that tmpfs is less than 512Mb, so I confirm my explaination. If there is a line in /etc/fstab about /dev/shm, you can add an option 'size=XXXM', say 'size=520M'. Or reduce UML amount of used memory. Recent uml kernels (if the console can be seen, I complained about con=null) display a specific error message saying that host memory was exhausted (thanks Jeff). -- Inform me of my mistakes, so I can add them to my list! Paolo Giarrusso, aka Blaisorblade http://www.user-mode-linux.org/~blaisorblade |
From: Jeff D. <jd...@ad...> - 2007-04-29 23:42:51
|
On Sun, Apr 29, 2007 at 11:08:59PM +0200, Blaisorblade wrote: > Recent uml kernels (if the console can be seen, I complained about con=null) > display a specific error message saying that host memory was exhausted Not that recently - it's been there for quite a while. But even more recently, UML will dump core on a panic (rlimits permitting), so even if there's something crazy like con=null, you can gdb the core file and look at the contents of log_buf (the symbol table permitting). Jeff -- Work email - jdike at linux dot intel dot com |
From: Herbert V. <sys...@au...> - 2007-04-30 07:39:32
|
Thanks for all your help, I can now confirm, that if i use a MEM=256 instead of MEM=512 it works fine without crashes when i'm producing files larger than 256M with dd! So I think, I have to give an entry in fstab for /dev/shm and have to explicitly define the size of /dev/shm to an equal value as i did to start uml linux with mem parameter. con=pty is my new change, when i run uml linux to get messages of my system. hope this can help other guys, which have same problems expirienced. thanx & bye, Herbert Am Sonntag, den 29.04.2007, 19:38 -0400 schrieb Jeff Dike: > On Sun, Apr 29, 2007 at 11:08:59PM +0200, Blaisorblade wrote: > > Recent uml kernels (if the console can be seen, I complained about con=null) > > display a specific error message saying that host memory was exhausted > > Not that recently - it's been there for quite a while. > > But even more recently, UML will dump core on a panic (rlimits > permitting), so even if there's something crazy like con=null, you can > gdb the core file and look at the contents of log_buf (the symbol > table permitting). > > Jeff > |
From: Antoine M. <an...@na...> - 2007-04-30 17:17:52
|
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA512 Herbert Volkmann wrote: > Thanks for all your help, > > I can now confirm, that if i use a MEM=256 instead of MEM=512 it works > fine without crashes when i'm producing files larger than 256M with dd! > > So I think, I have to give an entry in fstab for /dev/shm and have to > explicitly define the size of /dev/shm to an equal value as i did to > start uml linux with mem parameter. An easier way of doing this is to tell the guest which mount point to use for tmpfs storage and mount it before starting the guest: TMP_DIR=/somewhere/else linux ubda=./root_fs That way you leave your host's /dev/shm alone for what it is meant to be used. Also saves you (security) issues with mounting it exec/noexec. Antoine > > con=pty is my new change, when i run uml linux to get messages of my > system. hope this can help other guys, which have same problems > expirienced. > > thanx & bye, > Herbert > > Am Sonntag, den 29.04.2007, 19:38 -0400 schrieb Jeff Dike: >> On Sun, Apr 29, 2007 at 11:08:59PM +0200, Blaisorblade wrote: >>> Recent uml kernels (if the console can be seen, I complained about con=null) >>> display a specific error message saying that host memory was exhausted >> Not that recently - it's been there for quite a while. >> >> But even more recently, UML will dump core on a panic (rlimits >> permitting), so even if there's something crazy like con=null, you can >> gdb the core file and look at the contents of log_buf (the symbol >> table permitting). >> >> Jeff >> > > > ------------------------------------------------------------------------- > This SF.net email is sponsored by DB2 Express > Download DB2 Express C - the FREE version of DB2 express and take > control of your XML. No limits. Just data. Click to get it now. > http://sourceforge.net/powerbar/db2/ > _______________________________________________ > User-mode-linux-user mailing list > Use...@li... > https://lists.sourceforge.net/lists/listinfo/user-mode-linux-user > -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.3 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFGNiSxGK2zHPGK1rsRCpgDAJ9iAOZEtRpkTH1KqQHuMrsJzErwgwCfZYGE /dK2thS1GlIlHW8G/ZgFC5Y= =XBzg -----END PGP SIGNATURE----- |