Re: [Qemu-devel] Bug with qxl vga memory settings
On 01/25/13 15:37, Fabio Fantoni wrote: Debugging qxl problem with xen I tested first qemu without xen and I found that qxl memory mapping problem is not only xen related. After some tests with qemu-only I found that -global qxl-vga.vram_size=N and -global qxl-vga.ram_size=N (where N is for example 16384 or 65536) are bugged. ram_size + vram_size want bytes not kilobytes. Qxl ram regions with these parameters on all test are always: qxl-0: ram/pri: 32 MB [region 0] qxl-0: vram/32: 0 MB [region 1] Yep. You'll get the minimum supported sizes due to the small numbers. Qemu crashes when start using qxl. Oops. That shouldn't happen of course, I'll have a look. With -global qxl-vga.vram_size_mb=N and -global qxl-vga.ram_size_mb=N it seems correctly working. Ah, you've found the more user-friendly versions which accept megabytes instead of bytes. For example I saw that on ALL tests with right qxl memory region 0 and 1 there is also another unmapped region 0 mb. The rom bar maybe? seabios maps that only temporary to load the vgabios. Is it correct or is it wrong and can be the cause of crash with xen if qemu search for videoram region0+region1? qxl-0: ram/pri: 32 MB [region 0] qxl-0: vram/32: 32 MB [region 1] qxl-0: vram/64: 32 MB [unmapped] vram/64 is a mirror of vram/32, but in a 64bit capable pci bar. It will show up if you make vram64_size_mb larger than vram_size_mb and allows pretty big vram bars for 64bit guests. I saw also that qemu 1.3 debian package use vgabios from vgabios 0.7 package instead of seabios 1.7.2 package. Can be a problem with qxl? Both should work (assuming vgabios is taken from the repo @ qemu.org). cheers, Gerd
Re: [Qemu-devel] Bug with qxl vga memory settings
After some tests with qemu-only I found that -global qxl-vga.vram_size=N and -global qxl-vga.ram_size=N (where N is for example 16384 or 65536) are bugged. qxl-0: ram/pri: 32 MB [region 0] qxl-0: vram/32: 0 MB [region 1] Yep. You'll get the minimum supported sizes due to the small numbers. Qemu crashes when start using qxl. Oops. That shouldn't happen of course, I'll have a look. Doesn't reproduce (git master). cheers, Gerd
Re: [Qemu-devel] Bug with qxl vga memory settings
Il 28/01/2013 11:59, Gerd Hoffmann ha scritto: After some tests with qemu-only I found that -global qxl-vga.vram_size=N and -global qxl-vga.ram_size=N (where N is for example 16384 or 65536) are bugged. qxl-0: ram/pri: 32 MB [region 0] qxl-0: vram/32: 0 MB [region 1] Yep. You'll get the minimum supported sizes due to the small numbers. Qemu crashes when start using qxl. Oops. That shouldn't happen of course, I'll have a look. Doesn't reproduce (git master). cheers, Gerd - Nessun virus nel messaggio. Controllato da AVG - www.avg.com Versione: 2013.0.2890 / Database dei virus: 2639/6061 - Data di rilascio: 27/01/2013 I tried qemu-only with qemu (commit 1356b98d3e95a85071e6bf9a99e8799e1ae1bbee) compiled from xen and qemu 1.3.0 debian experimental package, they both crash if ram and vram are not setted correctly. The command tried was this: /usr/bin/qemu-system-x86_64 -name QUANTAL -k it -spice port=6002,tls-port=0,addr=0.0.0.0,disable-ticketing,agent-mouse=on -vga qxl -global qxl-vga.vram_size=16384 -global qxl-vga.ram_size=16384 -boot order=dc -smp 2,maxcpus=2 --global qxl-vga.debug=1 -global qxl-vga.guestdebug=20 -m 2016 -drive file=/mnt/vm/disks/QUANTALHVM.disk1.xm,if=ide,index=0,media=disk,format=raw,cache=writeback \ -device virtio-serial \ -chardev spicevmc,id=vdagent,name=vdagent \ -device virtserialport,chardev=vdagent,name=com.redhat.spice.0 \ -chardev stdio,id=seabios -device isa-debugcon,iobase=0x402,chardev=seabios On xen with qxl active, when it is being used, qemu crashed in any case, there is a bug in xen but I need to know in advance if there are bugs also in qemu and/or spice. This is the reason I'm testing qemu-only. About spice without qxl, vdagent, spice audio and usbredirection are working also in xen (passing qemu parameters manually for now, I'll do the patches to have them in libxl). If someone want to test qxl with xen fast and easy, a patch for qxl support is here (based on xen-unstable): http://lists.xen.org/archives/html/xen-devel/2013-01/msg02321.html smime.p7s Description: Firma crittografica S/MIME
Re: [Qemu-devel] Bug with qxl vga memory settings
[Adding spice-devel@ back - I don't think we should drop this list, even if it will be a repost there.] I've just a small comment below, -- I haven't gotten to testing it myself yet, was too busy today to do anything useful. 25.01.2013 18:37, Fabio Fantoni wrote: Debugging qxl problem with xen I tested first qemu without xen and I found that qxl memory mapping problem is not only xen related. After some tests with qemu-only I found that -global qxl-vga.vram_size=N and -global qxl-vga.ram_size=N (where N is for example 16384 or 65536) are bugged. Qxl ram regions with these parameters on all test are always: qxl-0: ram/pri: 32 MB [region 0] qxl-0: vram/32: 0 MB [region 1] qxl-0: vram/64: 0 MB [unmapped] Qemu crashes when start using qxl. For full log of one test with this problem see attachment qemulog1 Without -global qxl-vga.vram_size=N and -global qxl-vga.ram_size=N qxl ram region it seems correct and qemu doesn't crash. For full log of such test see attachment qemulog2 With -global qxl-vga.vram_size_mb=N and -global qxl-vga.ram_size_mb=N it seems correctly working. For full log of such test see attachment qemulog3 About qxl with xen there are other problems. Is there a way to understand that is an effective xen problem and not on qemu? Should I searching something specific in the logs? See attachment qemu-dm-WHEEZY For example I saw that on ALL tests with right qxl memory region 0 and 1 there is also another unmapped region 0 mb. Is it correct or is it wrong and can be the cause of crash with xen if qemu search for videoram region0+region1? qxl-0: ram/pri: 32 MB [region 0] qxl-0: vram/32: 32 MB [region 1] qxl-0: vram/64: 32 MB [unmapped] I saw also that qemu 1.3 debian package use vgabios from vgabios 0.7 package instead of seabios 1.7.2 package. Can be a problem with qxl? I haven't noticed this -- will switch in the next release, -- seavgabios in debian is already available, the only thing needed is to change the vgabios-* symlinks in /usr/share/qemu to point to seabios binaries instead of vgabios binaries. You can do that any time. If you need more test and detail tell me and I'll do/post it. Thanks for any reply and sorry for my bad english. Thanks, /mjt
Re: [Qemu-devel] Bug with qxl vga memory settings
25.01.2013 18:51, Michael Tokarev wrote: I saw also that qemu 1.3 debian package use vgabios from vgabios 0.7 package instead of seabios 1.7.2 package. Can be a problem with qxl? This is wrong. qemu-1.3, and current qemu git tree, uses vgabios 0.7a from http://www.nongnu.org/vgabios/ . While seabios provides seavgabios too nowadays, it is not used in qemu since it does not have protected mode calls support. But we may actually ship it (seavgabios) in Debian for qemu for a while, to see how it goes. That'll be interesting. I haven't noticed this -- will switch in the next release, -- seavgabios in debian is already available, the only thing needed is to change the vgabios-* symlinks in /usr/share/qemu to point to seabios binaries instead of vgabios binaries. You can do that any time. Thanks, /mjt