2016-05-09 13:36+0800, Peter Xu: > On Fri, May 06, 2016 at 10:53:44PM +0200, Radim Krčmář wrote: >> This series bases on Peter's IR v6 and depends on patches that were just >> posted to kvm-list, "[RFC 0/9] KVM: x86: break the xAPIC barrier". >> >> The kernel interface could use your comments, but internal QEMU one is >> in dire need of them. Please see [1/4]. >> >> I have tested the series and seems to work as well as it can. >> Peter's IR v6 didn't boot on my setup, so I reverted to the latest >> version I know was working, v4, and rebased paches for testing. > > Radim, > > Would you please provide your test setup? So that I can try to > reproduce it on my machine and debug it.
I could reproduce with a kernel based off kvm/queue (basically 4.6-rc3), that was make olconfig with fedora rawhide config for 4.6-rc3 and qemu/master (53db932604d) after pulling your vtd-intr-v6 and doing the minimum to make ./configure happy. The bug was caused by pci-bridge, which I didn't remove from a copy-pasted qemu line ... Linux boot hangs if QEMU is ran with -device i82801b11-bridge,id=pci.1,bus=pcie.0,addr=0x1e \ -device pci-bridge,chassis_nr=2,id=pci.2,bus=pci.1,addr=0x0 \ -drive file=/home/rhel7.qcow2,format=qcow2,if=none,id=drive-virtio-disk0 \ -device virtio-blk-pci,scsi=off,bus=pci.2,addr=0x3,drive=drive-virtio-disk0,id=virtio-disk0,bootindex=1 \ but the following works -drive file=/home/rhel7.qcow2,format=qcow2,if=none,id=drive-virtio-disk0 \ -device virtio-blk-pci,scsi=off,bus=pcie.0,drive=drive-virtio-disk0,id=virtio-disk0,bootindex=1 \ I can give you ssh access to the machine too. >> The setup from Igor's latest x2APIC QEMU patches creates two VCPUs, >> first has id 0 and second has 280. Edge IO-APIC and MSI interrupts >> were being delivered to both of them, but level didn't work -- only >> one interrupt was ever delivered, I blame EOI. > > Not sure whether this is related to the v5 fix on kernel EOI hack > (the two patches that you have reviewed)? That two patches should > solve the level-triggered interrupt issue that I have encountered, > like e1000 cards. I didn't try the R/O preserving patch, but the other one on top of v4 changed nothing. (There might be some kernel bugs too, because I expected that EOI would start working with it ...)
