Merge tag 'pull-xenfv-20250109-1' of https://gitlab.com/dwmw2/qemu into staging

Xen emulation fixes

# -----BEGIN PGP SIGNATURE-----
#
# iQJGBAABCAAwFiEEMUsIrNDeSBEzpfKGm+mA/QrAFUQFAmd/qNYSHGR3bXdAYW1h
# em9uLmNvLnVrAAoJEJvpgP0KwBVEtHsP/1qdpeVDCW1LAdGsOl9vixBXTR5/85G4
# m1KilpAPyxla8WfChRIagIdSAYGP5gN+yzbZ74AGb8HxumqJdl0bj6Gtqms2r8EQ
# 4T7IU1iNONDkncApkHdQW9BdKg4Atq3dY8dEaN1UxzCfRjHC/KS5vHPN3OzGKqJ1
# tAk8wOcDtp7cfW+utw2ssjVR14cfJLQCR7/ehBfeFkC0DSd8p/yTJ31bFnLyPpBn
# vh03MrslqV+h47D0uQxKwx5rtvNQhhIc/eRR/RymY3BSzAqRiyed/hTvsrRy4y/Z
# EXB8ACQ6U2Ikrj//VXimSTx5aQDeGIU8nD6zvNRWZ1rTmTtD3n5dOxL2U9U5DBHb
# TtlYhyochV6zO76mbINyjkSkGdj8ZZgF+5w5IIEhjazfHdWDuMdG0IjcRxl0r2Qz
# 4jaoVjxMUT/MLI4noSVYFF29/aWYxsk/nsYCPOM2X4WuzK4/ragIWbpZZqOIFn4X
# NyEc7xD2z9iL3MZe0Ygsa1eRpi/Gak0ih6W/u6ngON2EGESdF4T+CI+zTp6I4xtp
# jOrAGltp6012pRJibHrKKdpnTYuQCRj3kSFAEP+JhNSBDUhbZ5lJWTnxiW7BkBO4
# BujmX3TMFsdt4jDqNQzht84Tgf4JEAYbGCks9msFcoYdZovKcyG3kgfZyAVfEap2
# kvCgGk7JMz1A
# =5kvA
# -----END PGP SIGNATURE-----
# gpg: Signature made Thu 09 Jan 2025 05:45:42 EST
# gpg:                using RSA key 314B08ACD0DE481133A5F2869BE980FD0AC01544
# gpg:                issuer "dwmw@amazon.co.uk"
# gpg: Good signature from "David Woodhouse <dwmw@amazon.co.uk>" [unknown]
# gpg:                 aka "David Woodhouse <dwmw@amazon.com>" [unknown]
# gpg: WARNING: The key's User ID is not certified with a trusted signature!
# gpg:          There is no indication that the signature belongs to the owner.
# Primary key fingerprint: 314B 08AC D0DE 4811 33A5  F286 9BE9 80FD 0AC0 1544

* tag 'pull-xenfv-20250109-1' of https://gitlab.com/dwmw2/qemu:
  hw/xen: Check if len is 0 before memcpy()
  hw/i386/pc: Fix level interrupt sharing for Xen event channel GSI

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

4 files changed, 73 insertions, 25 deletions

diff --git a/hw/i386/kvm/xen_evtchn.c b/hw/i386/kvm/xen_evtchn.c
index bd2a3cb..58484f3 100644
--- a/hw/i386/kvm/xen_evtchn.c
+++ b/hw/i386/kvm/xen_evtchn.c
@@ -140,6 +140,8 @@ struct XenEvtchnState {
 
     uint64_t callback_param;
     bool evtchn_in_kernel;
+    bool setting_callback_gsi;
+    int extern_gsi_level;
     uint32_t callback_gsi;
 
     QEMUBH *gsi_bh;
@@ -431,9 +433,22 @@ void xen_evtchn_set_callback_level(int level)
     }
 
     if (s->callback_gsi && s->callback_gsi < s->nr_callback_gsis) {
-        qemu_set_irq(s->callback_gsis[s->callback_gsi], level);
-        if (level) {
-            /* Ensure the vCPU polls for deassertion */
+        /*
+         * Ugly, but since we hold the BQL we can set this flag so that
+         * xen_evtchn_set_gsi() can tell the difference between this code
+         * setting the GSI, and an external device (PCI INTx) doing so.
+         */
+        s->setting_callback_gsi = true;
+        /* Do not deassert the line if an external device is asserting it. */
+        qemu_set_irq(s->callback_gsis[s->callback_gsi],
+                     level || s->extern_gsi_level);
+        s->setting_callback_gsi = false;
+
+        /*
+         * If the callback GSI is the only one asserted, ensure the status
+         * is polled for deassertion in kvm_arch_post_run().
+         */
+        if (level && !s->extern_gsi_level) {
             kvm_xen_set_callback_asserted();
         }
     }
@@ -1596,7 +1611,7 @@ static int allocate_pirq(XenEvtchnState *s, int type, int gsi)
     return pirq;
 }
 
-bool xen_evtchn_set_gsi(int gsi, int level)
+bool xen_evtchn_set_gsi(int gsi, int *level)
 {
     XenEvtchnState *s = xen_evtchn_singleton;
     int pirq;
@@ -1608,16 +1623,35 @@ bool xen_evtchn_set_gsi(int gsi, int level)
     }
 
     /*
-     * Check that that it *isn't* the event channel GSI, and thus
-     * that we are not recursing and it's safe to take s->port_lock.
-     *
-     * Locking aside, it's perfectly sane to bail out early for that
-     * special case, as it would make no sense for the event channel
-     * GSI to be routed back to event channels, when the delivery
-     * method is to raise the GSI... that recursion wouldn't *just*
-     * be a locking issue.
+     * For the callback_gsi we need to implement a logical OR of the event
+     * channel GSI and the external input (e.g. from PCI INTx), because
+     * QEMU itself doesn't support shared level interrupts via demux or
+     * resamplers.
      */
     if (gsi && gsi == s->callback_gsi) {
+        /* Remember the external state of the GSI pin (e.g. from PCI INTx) */
+        if (!s->setting_callback_gsi) {
+            s->extern_gsi_level = *level;
+
+            /*
+             * Don't allow the external device to deassert the line if the
+             * eveht channel GSI should still be asserted.
+             */
+            if (!s->extern_gsi_level) {
+                struct vcpu_info *vi = kvm_xen_get_vcpu_info_hva(0);
+                if (vi && vi->evtchn_upcall_pending) {
+                    /* Need to poll for deassertion */
+                    kvm_xen_set_callback_asserted();
+                    *level = 1;
+                }
+            }
+        }
+
+        /*
+         * The event channel GSI cannot be routed to PIRQ, as that would make
+         * no sense. It could also deadlock on s->port_lock, if we proceed.
+         * So bail out now.
+         */
         return false;
     }
 
@@ -1628,7 +1662,7 @@ bool xen_evtchn_set_gsi(int gsi, int level)
         return false;
     }
 
-    if (level) {
+    if (*level) {
         int port = s->pirq[pirq].port;
 
         s->pirq_gsi_set |= (1U << gsi);
diff --git a/hw/i386/kvm/xen_evtchn.h b/hw/i386/kvm/xen_evtchn.h
index b740acf..0521ebc 100644
--- a/hw/i386/kvm/xen_evtchn.h
+++ b/hw/i386/kvm/xen_evtchn.h
@@ -23,7 +23,7 @@ void xen_evtchn_set_callback_level(int level);
 
 int xen_evtchn_set_port(uint16_t port);
 
-bool xen_evtchn_set_gsi(int gsi, int level);
+bool xen_evtchn_set_gsi(int gsi, int *level);
 void xen_evtchn_snoop_msi(PCIDevice *dev, bool is_msix, unsigned int vector,
                           uint64_t addr, uint32_t data, bool is_masked);
 void xen_evtchn_remove_pci_device(PCIDevice *dev);
diff --git a/hw/i386/kvm/xen_xenstore.c b/hw/i386/kvm/xen_xenstore.c
index 5969105..17802aa 100644
--- a/hw/i386/kvm/xen_xenstore.c
+++ b/hw/i386/kvm/xen_xenstore.c
@@ -532,6 +532,10 @@ static void xs_read(XenXenstoreState *s, unsigned int req_id,
         return;
     }
 
+    if (!len) {
+        return;
+    }
+
     memcpy(&rsp_data[rsp->len], data->data, len);
     rsp->len += len;
 }
diff --git a/hw/i386/x86-common.c b/hw/i386/x86-common.c
index a7d46c3..97b4f7d 100644
--- a/hw/i386/x86-common.c
+++ b/hw/i386/x86-common.c
@@ -450,8 +450,27 @@ static long get_file_size(FILE *f)
 void gsi_handler(void *opaque, int n, int level)
 {
     GSIState *s = opaque;
+    bool bypass_ioapic = false;
 
     trace_x86_gsi_interrupt(n, level);
+
+#ifdef CONFIG_XEN_EMU
+    /*
+     * Xen delivers the GSI to the Legacy PIC (not that Legacy PIC
+     * routing actually works properly under Xen). And then to
+     * *either* the PIRQ handling or the I/OAPIC depending on whether
+     * the former wants it.
+     *
+     * Additionally, this hook allows the Xen event channel GSI to
+     * work around QEMU's lack of support for shared level interrupts,
+     * by keeping track of the externally driven state of the pin and
+     * implementing a logical OR with the state of the evtchn GSI.
+     */
+    if (xen_mode == XEN_EMULATE) {
+        bypass_ioapic = xen_evtchn_set_gsi(n, &level);
+    }
+#endif
+
     switch (n) {
     case 0 ... ISA_NUM_IRQS - 1:
         if (s->i8259_irq[n]) {
@@ -460,18 +479,9 @@ void gsi_handler(void *opaque, int n, int level)
         }
         /* fall through */
     case ISA_NUM_IRQS ... IOAPIC_NUM_PINS - 1:
-#ifdef CONFIG_XEN_EMU
-        /*
-         * Xen delivers the GSI to the Legacy PIC (not that Legacy PIC
-         * routing actually works properly under Xen). And then to
-         * *either* the PIRQ handling or the I/OAPIC depending on
-         * whether the former wants it.
-         */
-        if (xen_mode == XEN_EMULATE && xen_evtchn_set_gsi(n, level)) {
-            break;
+        if (!bypass_ioapic) {
+            qemu_set_irq(s->ioapic_irq[n], level);
         }
-#endif
-        qemu_set_irq(s->ioapic_irq[n], level);
         break;
     case IO_APIC_SECONDARY_IRQBASE
         ... IO_APIC_SECONDARY_IRQBASE + IOAPIC_NUM_PINS - 1: