fix MSI injection on Xen

On Xen MSIs can be remapped into pirqs, which are a type of event
channels. It's mostly for the benefit of PCI passthrough devices, to
avoid the overhead of interacting with the emulated lapic.

However remapping interrupts and MSIs is also supported for emulated
devices, such as the e1000 and virtio-net.

When an interrupt or an MSI is remapped into a pirq, masking and
unmasking is done by masking and unmasking the event channel. The
masking bit on the PCI config space or MSI-X table should be ignored,
but it isn't at the moment.

As a consequence emulated devices which use MSI or MSI-X, such as
virtio-net, don't work properly (the guest doesn't receive any
notifications). The mechanism was working properly when xen_apic was
introduced, but I haven't narrowed down which commit in particular is
causing the regression.

Fix the issue by ignoring the masking bit for MSI and MSI-X which have
been remapped into pirqs.

Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

6 files changed, 34 insertions, 6 deletions

diff --git a/hw/pci/msi.c b/hw/pci/msi.c
index 8efa23d376..85f21b8c4b 100644
--- a/hw/pci/msi.c
+++ b/hw/pci/msi.c
@@ -20,6 +20,7 @@
 
 #include "qemu/osdep.h"
 #include "hw/pci/msi.h"
+#include "hw/xen/xen.h"
 #include "qemu/range.h"
 
 /* PCI_MSI_ADDRESS_LO */
@@ -254,13 +255,19 @@ void msi_reset(PCIDevice *dev)
 static bool msi_is_masked(const PCIDevice *dev, unsigned int vector)
 {
     uint16_t flags = pci_get_word(dev->config + msi_flags_off(dev));
-    uint32_t mask;
+    uint32_t mask, data;
+    bool msi64bit = flags & PCI_MSI_FLAGS_64BIT;
     assert(vector < PCI_MSI_VECTORS_MAX);
 
     if (!(flags & PCI_MSI_FLAGS_MASKBIT)) {
         return false;
     }
 
+    data = pci_get_word(dev->config + msi_data_off(dev, msi64bit));
+    if (xen_is_pirq_msi(data)) {
+        return false;
+    }
+
     mask = pci_get_long(dev->config +
                         msi_mask_off(dev, flags & PCI_MSI_FLAGS_64BIT));
     return mask & (1U << vector);
diff --git a/hw/pci/msix.c b/hw/pci/msix.c
index 4fea7edc89..eb4ef113d1 100644
--- a/hw/pci/msix.c
+++ b/hw/pci/msix.c
@@ -19,6 +19,7 @@
 #include "hw/pci/msi.h"
 #include "hw/pci/msix.h"
 #include "hw/pci/pci.h"
+#include "hw/xen/xen.h"
 #include "qemu/range.h"
 
 #define MSIX_CAP_LENGTH 12
@@ -78,8 +79,15 @@ static void msix_clr_pending(PCIDevice *dev, int vector)
 
 static bool msix_vector_masked(PCIDevice *dev, unsigned int vector, bool fmask)
 {
-    unsigned offset = vector * PCI_MSIX_ENTRY_SIZE + PCI_MSIX_ENTRY_VECTOR_CTRL;
-    return fmask || dev->msix_table[offset] & PCI_MSIX_ENTRY_CTRL_MASKBIT;
+    unsigned offset = vector * PCI_MSIX_ENTRY_SIZE;
+    uint32_t *data = (uint32_t *)&dev->msix_table[offset + PCI_MSIX_ENTRY_DATA];
+    /* MSIs on Xen can be remapped into pirqs. In those cases, masking
+     * and unmasking go through the PV evtchn path. */
+    if (xen_is_pirq_msi(*data)) {
+        return false;
+    }
+    return fmask || dev->msix_table[offset + PCI_MSIX_ENTRY_VECTOR_CTRL] &
+        PCI_MSIX_ENTRY_CTRL_MASKBIT;
 }
 
 bool msix_is_masked(PCIDevice *dev, unsigned int vector)
diff --git a/hw/xen/xen_pt_msi.c b/hw/xen/xen_pt_msi.c
index 5624685b20..9a16f2bff1 100644
--- a/hw/xen/xen_pt_msi.c
+++ b/hw/xen/xen_pt_msi.c
@@ -115,9 +115,7 @@ static int msi_msix_setup(XenPCIPassthroughState *s,
 
     assert((!is_msix && msix_entry == 0) || is_msix);
 
-    if (gvec == 0) {
-        /* if gvec is 0, the guest is asking for a particular pirq that
-         * is passed as dest_id */
+    if (xen_is_pirq_msi(data)) {
         *ppirq = msi_ext_dest_id(addr >> 32) | msi_dest_id(addr);
         if (!*ppirq) {
             /* this probably identifies an misconfiguration of the guest,
diff --git a/include/hw/xen/xen.h b/include/hw/xen/xen.h
index 1b81b4be9a..c57735419c 100644
--- a/include/hw/xen/xen.h
+++ b/include/hw/xen/xen.h
@@ -33,6 +33,7 @@ int xen_pci_slot_get_pirq(PCIDevice *pci_dev, int irq_num);
 void xen_piix3_set_irq(void *opaque, int irq_num, int level);
 void xen_piix_pci_write_config_client(uint32_t address, uint32_t val, int len);
 void xen_hvm_inject_msi(uint64_t addr, uint32_t data);
+int xen_is_pirq_msi(uint32_t msi_data);
 
 qemu_irq *xen_interrupt_controller_init(void);
 
diff --git a/xen-hvm-stub.c b/xen-hvm-stub.c
index a6cb5d358f..c5003251cb 100644
--- a/xen-hvm-stub.c
+++ b/xen-hvm-stub.c
@@ -31,6 +31,11 @@ void xen_hvm_inject_msi(uint64_t addr, uint32_t data)
 {
 }
 
+int xen_is_pirq_msi(uint32_t msi_data)
+{
+    return 0;
+}
+
 void xen_ram_alloc(ram_addr_t ram_addr, ram_addr_t size, MemoryRegion *mr,
                    Error **errp)
 {
diff --git a/xen-hvm.c b/xen-hvm.c
index 1c9fb12955..6861c51ef9 100644
--- a/xen-hvm.c
+++ b/xen-hvm.c
@@ -13,6 +13,7 @@
 
 #include "hw/pci/pci.h"
 #include "hw/i386/pc.h"
+#include "hw/i386/apic-msidef.h"
 #include "hw/xen/xen_common.h"
 #include "hw/xen/xen_backend.h"
 #include "qmp-commands.h"
@@ -158,6 +159,14 @@ void xen_piix_pci_write_config_client(uint32_t address, uint32_t val, int len)
     }
 }
 
+int xen_is_pirq_msi(uint32_t msi_data)
+{
+    /* If vector is 0, the msi is remapped into a pirq, passed as
+     * dest_id.
+     */
+    return ((msi_data & MSI_DATA_VECTOR_MASK) >> MSI_DATA_VECTOR_SHIFT) == 0;
+}
+
 void xen_hvm_inject_msi(uint64_t addr, uint32_t data)
 {
     xen_xc_hvm_inject_msi(xen_xc, xen_domid, addr, data);