[PATCH] arm/arm64: KVM: Fix set_clear_sgi_pend_reg offset

[christoffer.dall@linaro.org (Christoffer Dall)]

On Fri, Sep 26, 2014 at 12:16:35PM +0200, Christoffer Dall wrote:
> On Fri, Sep 26, 2014 at 05:26:00PM +0800, Shannon Zhao wrote:
> > 
> > 
> > On 2014/9/26 16:44, Christoffer Dall wrote:
> > > Hi Shannon,
> > > 
> > > On Fri, Sep 26, 2014 at 01:57:46PM +0800, Shannon Zhao wrote:
> > >>
> > >> On 2014/9/26 1:49, Christoffer Dall wrote:
> > >>> The sgi values calculated in read_set_clear_sgi_pend_reg() and
> > >>> write_set_clear_sgi_pend_reg() were horribly incorrectly multiplied by 4
> > >>> with catastrophic results in that subfunctions ended up overwriting
> > >>> memory not allocated for the expected purpose.
> > >>>
> > >>> This showed up as bugs in kfree() and the kernel complaining a lot of
> > >>> you turn on memory debugging.
> > >>>
> > >>> This addresses: http://marc.info/?l=kvm&m=141164910007868&w=2
> > >>>
> > >>> Reported-by: Shannon Zhao <zhaoshenglong@huawei.com>
> > >>> Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
> > >>> ---
> > >>>  virt/kvm/arm/vgic.c | 4 ++--
> > >>>  1 file changed, 2 insertions(+), 2 deletions(-)
> > >>>
> > >>> diff --git a/virt/kvm/arm/vgic.c b/virt/kvm/arm/vgic.c
> > >>> index b6fab0f..8629678 100644
> > >>> --- a/virt/kvm/arm/vgic.c
> > >>> +++ b/virt/kvm/arm/vgic.c
> > >>> @@ -816,7 +816,7 @@ static bool read_set_clear_sgi_pend_reg(struct kvm_vcpu *vcpu,
> > >>>  {
> > >>>  	struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
> > >>>  	int sgi;
> > >>> -	int min_sgi = (offset & ~0x3) * 4;
> > >>> +	int min_sgi = (offset & ~0x3);
> > >>>  	int max_sgi = min_sgi + 3;
> > >>>  	int vcpu_id = vcpu->vcpu_id;
> > >>>  	u32 reg = 0;
> > >>> @@ -837,7 +837,7 @@ static bool write_set_clear_sgi_pend_reg(struct kvm_vcpu *vcpu,
> > >>>  {
> > >>>  	struct vgic_dist *dist = &vcpu->kvm->arch.vgic;
> > >>>  	int sgi;
> > >>> -	int min_sgi = (offset & ~0x3) * 4;
> > >>> +	int min_sgi = (offset & ~0x3);
> > >>>  	int max_sgi = min_sgi + 3;
> > >>>  	int vcpu_id = vcpu->vcpu_id;
> > >>>  	u32 reg;
> > >>>
> > >> Hi Christoffer,
> > >>
> > >> I have test this patch for a few hours. The kfree() bug doesn't appear again.
> > >> But I come to another problem as followed.
> > >> The test is that start 2 VMs, sleep 10 and do pkill qemu.
> > >>
> > >> qemu-system-aar[1207]: unhandled level 1 permission fault (11) at 0xffffc01ed6c200, esr 0x9200000d
> > >> pgd = ffffffc012986000
> > >> [ffffc01ed6c200] *pgd=0000000000000000, *pud=0000000000000000
> > >>
> > >> CPU: 1 PID: 1207 Comm: qemu-system-aar Not tainted 3.17.0-rc4+ #1
> > >> task: ffffffc87b072900 ti: ffffffc0129e0000 task.ti: ffffffc0129e0000
> > >> PC is at 0x4181a0
> > >> LR is at 0x41826c
> > >> pc : [<00000000004181a0>] lr : [<000000000041826c>] pstate: 80000000
> > >> sp : 0000007fcd38ace0
> > >> x29: 0000007fcd38ace0 x28: 0000000000000000
> > >> x27: 0000000000000000 x26: 0000000000000000
> > >> x25: 0000000000000000 x24: 0000000000000000
> > >> x23: 0000000000000000 x22: 0000000000000000
> > >> x21: 0000000000000000 x20: 0000000000000000
> > >> x19: 0000007fcd38b070 x18: 0000007fcd38ab10
> > >> x17: 0000007f9bb14480 x16: 00000000009f2370
> > >> x15: ffffffffffffffff x14: 0000000000000000
> > >> x13: 0000000000000000 x12: 0000000000000268
> > >> x11: 00000000115e5520 x10: 0101010101010101
> > >> x9 : 0000000000000004 x8 : 0000000000ac7a78
> > >> x7 : 0000000000000000 x6 : 000000000000003f
> > >> x5 : 0000000000000040 x4 : 0000000000000000
> > >> x3 : 0000000000000030 x2 : 0000000000000001
> > >> x1 : ffffffc01ed6c200 x0 : ffffffc01ed6c200
> > >>
> > > Hmmm, I just ran a similar loop with a number of tests in the VM for a
> > > few hours and I didn't see this error.
> > Yeah, it really need to run longer.
> > After running about one hour this problem first appears and after running
> > about 4 hours it second appears.
> > > 
> > > In any case, this patch should still be merged, but we should try to
> > > reproduce your setup.
> > Your patch really solves the kfree() bug. I'll add tested-by line.
> > > 
> > > What is your command line, exact QEMU version, the file system you use,
> > > and the guest kernel you are running?
> > My test script is as followed. QEMU version is v2.1.0 release.
> > The fs is linaro-image-lamp-genericarmv8-20140727-701.rootfs.tar.gz.
> > Host kernel is based on marc's branch "kvmtool-vgic-dyn" with your patch
> > "Fix set_clear_sgi_pend_reg offset".
> > Guest kernel is 3.16 release.
> > 
> > while true
> > do
> > qemu-system-aarch64 \
> >     -enable-kvm -smp 4 \
> >     -kernel Image \
> >     -m 512 -machine virt,kernel_irqchip=on \
> >     -initrd guestfs.cpio.gz \
> >     -cpu host \
> >     -chardev pty,id=pty0,mux=on -monitor chardev:pty0 \
> >     -serial chardev:pty0 -daemonize \
> >     -vnc 0.0.0.0:0 \
> >     -append "rdinit=/sbin/init console=ttyAMA0 mem=512M root=/dev/ram earlyprintk=pl011,0x9000000 rw" &
> > 
> > qemu-system-aarch64 \
> >     -enable-kvm -smp 4 \
> >     -kernel Image \
> >     -m 512 -machine virt,kernel_irqchip=on \
> >     -initrd guestfs.cpio.gz \
> >     -cpu host \
> >     -chardev pty,id=pty0,mux=on -monitor chardev:pty0 \
> >     -serial chardev:pty0 -daemonize \
> >     -vnc 0.0.0.0:1 \
> >     -append "rdinit=/sbin/init console=ttyAMA0 mem=512M root=/dev/ram earlyprintk=pl011,0x9000000 rw" &
> >         sleep 5
> >         pkill qemu
> 
> ok, I'll try to reproduce.
> 
With kvmarm/queue as both host and guest and otherwise not using vnc but
nographic and a serial output, I've now been running this for 5 hours
straight without any issues. That's 1131 runs (2x number of guests
booted) and counting without seeing this...

-Christoffer