On Tue, 2011-05-24 at 22:33 +0800, Jiang, Yunhong wrote: > Hi, Leann, when I run KVM guest in ubuntu 10.04 (lucid amd64) with > 32bit qemu, I noticed the 32bit qemu always seg fault because the %gs > register is clobbered after KVM_RUN ioctl. > > After google and check the code, I noticed > http://bugs.debian.org/cgi-bin/bugreport.cgi?msg=47;bug=604956 and > commit 6cbdc7ba48dff850c233a1ea1abe87a6d0cde2dd in > git://kernel.ubuntu.com/ubuntu/ubuntu-lucid.git. But I'm a bit > confused of this commit and hope get some help from you. > > According to comments, the commit > 6cbdc7ba48dff850c233a1ea1abe87a6d0cde2dd is from upstream linux > c8770e7ba63bb5dd8fe5f9d251275a8fa717fb78. > However, comparing the two patchset, I noticed the ubuntu patch is a > bit different with the upstream patch, especially, the ubuntu patch > includes followed hunk, which removes the load_gs_index(). Will this > change cause the gs index not be loaded, and thus the gs will possibly > be 0? Are there any special reason to add this changes? I think I may > missed some important reason, hope to get some hints from you. > > #ifdef CONFIG_X86_64 > - load_gs_index(vmx->host_state.gs_sel); > wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gs); > #else
Hi Jiang, Bah, good catch. I've obviously bodged the backport by accident. I've opened up a new bug [1] to submit for Ubuntu SRU to get this properly fixed up in Lucid. Care to subscribe to the bug and help test? > Also I noticed the 2.6.35-28-50 tag in this git repo does not > contained the related changeset, will they be ported also? The fix in Maverick will actually be in the 2.6.35-29.51 release which is currently in maverick-proposed. It will then propagate down in the Maverick LTS backport kernel for Lucid. Thanks, Leann [1] https://bugs.launchpad.net/ubuntu/lucid/+source/linux/+bug/787675 -- To UNSUBSCRIBE, email to debian-kernel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/1306259951.1548.45.camel@adamo