Merge branch 'xenfv-kvm-15' of git://git.infradead.org/users/dwmw2/qemu into HEAD

This adds support for emulating Xen under Linux/KVM, based on kernel
patches which have been present since Linux v5.12. As with the kernel
support, it's derived from work started by João Martins of Oracle in
2018.

This series just adds the basic platform support — CPUID, hypercalls,
event channels, a stub of XenStore.

A full single-tenant internal implementation of XenStore, and patches
to make QEMU's Xen PV drivers work with this Xen emulation, are waiting
in the wings to be submitted in a follow-on patch series.

As noted in the documentation, it's enabled by setting the xen-version
property on the KVM accelerator, e.g.:

 qemu-system-x86_64 -serial mon:stdio -M q35 -display none -m 1G -smp 2 \
    -accel kvm,xen-version=0x4000e,kernel-irqchip=split \
    -kernel vmlinuz-6.0.7-301.fc37.x86_64 \
    -append "console=ttyS0 root=/dev/sda1" \
    -drive file=/var/lib/libvirt/images/fedora28.qcow2,if=none,id=disk \
    -device ahci,id=ahci -device ide-hd,drive=disk,bus=ahci.0

Even before this was merged, we've already been using it to find and fix
bugs in the Linux kernel Xen guest support:

https://lore.kernel.org/all/4bffa69a949bfdc92c4a18e5a1c3cbb3b94a0d32.camel@infradead.org/
https://lore.kernel.org/all/871qnunycr.ffs@tglx/

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

1 files changed, 18 insertions, 1 deletions

diff --git a/hw/pci/pci.c b/hw/pci/pci.c
index cc51f98593..10c980b9f5 100644
--- a/hw/pci/pci.c
+++ b/hw/pci/pci.c
@@ -49,6 +49,9 @@
 #include "qemu/cutils.h"
 #include "pci-internal.h"
 
+#include "hw/xen/xen.h"
+#include "hw/i386/kvm/xen_evtchn.h"
+
 //#define DEBUG_PCI
 #ifdef DEBUG_PCI
 # define PCI_DPRINTF(format, ...)       printf(format, ## __VA_ARGS__)
@@ -319,6 +322,17 @@ static void pci_msi_trigger(PCIDevice *dev, MSIMessage msg)
 {
     MemTxAttrs attrs = {};
 
+    /*
+     * Xen uses the high bits of the address to contain some of the bits
+     * of the PIRQ#. Therefore we can't just send the write cycle and
+     * trust that it's caught by the APIC at 0xfee00000 because the
+     * target of the write might be e.g. 0x0x1000fee46000 for PIRQ#4166.
+     * So we intercept the delivery here instead of in kvm_send_msi().
+     */
+    if (xen_mode == XEN_EMULATE &&
+        xen_evtchn_deliver_pirq_msi(msg.address, msg.data)) {
+        return;
+    }
     attrs.requester_id = pci_requester_id(dev);
     address_space_stl_le(&dev->bus_master_as, msg.address, msg.data,
                          attrs, NULL);
@@ -988,6 +1002,9 @@ static void do_pci_unregister_device(PCIDevice *pci_dev)
     pci_get_bus(pci_dev)->devices[pci_dev->devfn] = NULL;
     pci_config_free(pci_dev);
 
+    if (xen_mode == XEN_EMULATE) {
+        xen_evtchn_remove_pci_device(pci_dev);
+    }
     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
         memory_region_del_subregion(&pci_dev->bus_master_container_region,
                                     &pci_dev->bus_master_enable_region);
@@ -1648,7 +1665,7 @@ void pci_device_set_intx_routing_notifier(PCIDevice *dev,
  * 9.1: Interrupt routing. Table 9-1
  *
  * the PCI Express Base Specification, Revision 2.1
- * 2.2.8.1: INTx interrutp signaling - Rules
+ * 2.2.8.1: INTx interrupt signaling - Rules
  *          the Implementation Note
  *          Table 2-20
  */