* Michael Tokarev <[email protected]> [11-05-2013 10:15 EEST]: > Control: tag -1 + upstream patch pending > > 10.05.2013 00:52, Edward J. Shornock wrote: [..]
> Oh. So it is guest crash, not qemu process on host crash. Please indicate > as such in the subject. I'll be sure to do so in the future. > This is a known issue, I spent half a day pre-yesterday diagnosing it. > Current upstream qemu git should be fine, after the following commit: > > commit 66e61b55f158ef5628e4c056dd2f233c9351a3f5 > Author: Aurelien Jarno <[email protected]> > Date: Wed May 8 22:36:39 2013 +0200 > > tcg/optimize: fix setcond2 optimization > > When setcond2 is rewritten into setcond, the state of the destination > temp should be reset, so that a copy of the previous value is not > used instead of the result. > > Reported-by: Michael Tokarev <[email protected]> > Reviewed-by: Richard Henderson <[email protected]> > Signed-off-by: Aurelien Jarno <[email protected]> > > You can revert just that commit and retry the test - it should fail as > before. To test this I did the following: git clean -xdf git reset --hard 47ec15cdd44877e553ed0bd0a16aea8a295dad62 ./configure make -j6 [WAIT] x86_64-softmmu/qemu-system-x86_64 -m 256 -cdrom /path/to/kfreebsd/mini.iso --enable-kvm and the guest didn't crash. I pulled the source to get back to revision 38ebb396c955ceb2ef7e246248ceb7f8bfe1b774. Then I merged 5e773833688b49d8a3cce6c75dde8cdc1e0c7807 from git://anonscm.debian.org/pkg-qemu/qemu.git to make packages with "fakeroot debian/rules binary". It crashes with the packages. I see "qemu-system-x86_64: pci_add_option_rom: failed to find romfile "efi-e1000.rom" but don't know if that "matters". (I'm guessing it doesn't) So at this point - packaged => crash - run from git source => works It's definitely not a build-dependency issue since I didn't build in a clean chroot as I normally would--I built these on my live system. > Note that this happens only when you run something 32bit on a 64bit host, > 32bit on 32bit is unaffected. > > > I also had this problem with 1.4.0. Booting the iso will fail both with > > and without KVM acceleration. > > 1.4.0 had no this bug, so it must be something different. It indeed looks like it's a different problem. Whatever I'm experiencing (with this issue booting a kfreebsd iso) I could reproduce with the 1.4.0 packages. I didn't start testing from git until after 1.5.0 made it to experimental. > But at this point. 1.4 is mostly history, we should concentrate on getting > 1.5 out > as soon as possible, with all the needed fixes, and, more important, on > wheezy version which is in stable which will be used by most people. Sounds like a good plan. I'm not sure that I remember correctly what works and what doesn't with 1.1.2 so I'll set up a testbed for my reports.
signature.asc
Description: Digital signature
