On Thu, Jun 29, 2023 at 12:24 AM Huacai Chen <chenhuacai@xxxxxxxxxx> wrote: > > Hi, Zhao, > > On Thu, Jun 29, 2023 at 8:07 AM Yu Zhao <yuzhao@xxxxxxxxxx> wrote: > > > > On Mon, Jun 26, 2023 at 6:33 AM Paolo Bonzini <pbonzini@xxxxxxxxxx> wrote: > > > > > > On Mon, Jun 26, 2023 at 9:59 AM Huacai Chen <chenhuacai@xxxxxxxxxxx> wrote: > > > > > > > > After commit 45c7e8af4a5e3f0bea4ac209 ("MIPS: Remove KVM_TE support") we > > > > get a NULL pointer dereference when creating a KVM guest: > > > > > > To be honest, a bug that needed 2 years to be reproduced is probably a > > > sign that KVM/MIPS has no users. Any objections to removing it > > > altogether? > > > > ACK: > > 1. It's still broken after this patch [1]. The most well-tested MIPS > > distros, i.e., Debian/OpenWrt, have CONFIG_KVM=n. (The latter doesn't > > even provide the QEMU package on MIPS.) > > 2. Burden on QEMU dev. There is no guarantee that QEMU would work with > > KVM even if we could fix the kernel -- it actually does not until > > v8.0 [1], which is by luck: > > > > commit a844873512400fae6bed9e87694dc96ff2f15f39 > > Author: Paolo Bonzini <pbonzini@xxxxxxxxxx> > > Date: Sun Dec 18 01:06:45 2022 +0100 > > > > mips: Remove support for trap and emulate KVM > > > > This support was limited to the Malta board, drop it. > > I do not have a machine that can run VZ KVM, so I am assuming > > that it works for -M malta as well. > > > > (The latest Debian stable only ships v7.2.) > > > > [1] https://lore.kernel.org/r/ZI0R76Fx25Q2EThZ@xxxxxxxxxx/ > My testbed is Loongson-3A4000 host + Loongson-3A4000 guest + Qemu8.0, > both TCG and KVM works. I tried TCG too, as reported in the link above, and had no luck. > Some thoughts: > 1, I think your host is malta, but you cannot use a malta host to boot > a Loongson guest, at least their kernels use different page sizes. Good to know. Thanks. > 2, commit a844873512400fae6bed9e87694dc96f remove the TE KVM, so if > you are trying VZ KVM (but it seems you are using TE KVM), it can > break nothing. I fully understand :) I was trying to point out that QEMU/KVM had been broken for over 2 years -- neither TE (removed from KVM) nor VZ works with 7.2, which is probably why you used 8.0 too -- until that commit *accidentally* fixed VZ (it was supposed to be a cleanup, not a fix).