Re: [Qemu-devel] [PATCH v12 2/5] apic: use per CPU AS to map APIC MMIO for TCG

[Eduardo Habkost]

On Fri, Oct 02, 2015 at 03:24:24PM -0300, Eduardo Habkost wrote:
> On Tue, Sep 22, 2015 at 02:29:01PM +0200, Igor Mammedov wrote:
> > On Wed, 16 Sep 2015 17:19:12 +0800
> > Zhu Guihua <address@hidden> wrote:
> > 
> > > TCG supports per CPU address space, and the emulation quality is
> > > a bit better with it.
> > > So use per CPU address space to map APIC MMIO area. This allows the
> > > APIC base address of each cpu to be moved independent of others.
> > > 
> > > Signed-off-by: Zhu Guihua <address@hidden>
> > 
> > Reviewed-by: Igor Mammedov <address@hidden>
> 
> The series seems to be OK when using KVM, but something is wrong with TCG.
> 
> After applying this patch, my Fedora 20 image hangs on boot at:
> 
> [  OK  ] Started udev Coldplug all Devices.
>          Starting dracut initqueue hook...
> [  OK  ] Reached target System Initialization.
> [  OK  ] Reached target Basic System.
> [    8.988420] ACPI: PCI Interrupt Link [LNKD] enabled at IRQ 11
> 
> 
> Before this patch, boot continues with:
> 
> [  OK  ] Reached target Basic System.
> [    8.523293] ACPI: PCI Interrupt Link [LNKD] enabled at IRQ 11
> [    9.230955]  vda: vda1
> [  OK  ] Started dracut initqueue hook.
>          Mounting /sysroot...
> 
> I am running QEMU as:
>   $ ./x86_64-softmmu/qemu-system-x86_64 -drive 
> if=virtio,file=/home/ehabkost/system/vmachines/Fedora-x86_64-20-20131211.1-sda.qcow2,format=qcow2
>  -machine pc,accel=tcg  -smp 8,sockets=2,threads=2,cores=2,maxcpus=16 -serial 
> stdio -nographic -monitor none

Debugging revealed that this broke MSI delivery of virtio notifications.
Writes to PCIDevice::bus_master_as (at msi_send_message()) don't get
delivered to the APIC memory region anymore.

This patch doesn't seem to be required for the removal of ICC bus, so I
am applying the rest of the series anyway. If you see any problem with
that, please let me know.

-- 
Eduardo