[Qemu-devel] saving/loading PCI irq related state

[Uri Lublin]

Hello,

If one is not lucky he/she may lose PCI interrupts when saving and 
loading a VM.
It seems PCI irq related state is not being saved.
When this happens, the guest hangs/spins and the cpu usage of the 
process stays around 100%.

Attached are three patches to fix this:
  01 -- when saving/loading a pci device, save/load its irq_state
  02 -- save/load PCI-Bus irq related state
  03 -- save/load PCI-bridge (i440FX/PIIX3) irq related state

There are two alternatives, both recalculate PCI irq related state upon 
loadvm instead of saving/loading it:
(a) Make every PCI device DEV (e.g. rtl8139, ne2000, usb-uhci, etc) call 
its DEV_update_irq() from its state-load-function. or
(b) Keep patch 01 and recalculate 02 and 03 by going over all 
PCI-devices on the specific Bus/Bridge.

Comments welcome.

Thanks,
   Uri.

--- hw/pci.c    2007-11-27 22:17:33.319404340 +0200
+++ hw/pci.c.1  2007-11-27 22:11:49.889369058 +0200
@@ -83,18 +83,29 @@
 
 void pci_device_save(PCIDevice *s, QEMUFile *f)
 {
-    qemu_put_be32(f, 1); /* PCI device version */
+    int i;
+
+    qemu_put_be32(f, 2); /* PCI device version */
     qemu_put_buffer(f, s->config, 256);
+    for (i=0; i<4; i++)
+        qemu_put_be32s(f, &s->irq_state[i]);
 }
 
 int pci_device_load(PCIDevice *s, QEMUFile *f)
 {
     uint32_t version_id;
+    int i;
+
     version_id = qemu_get_be32(f);
-    if (version_id != 1)
+    if (version_id > 2)
         return -EINVAL;
     qemu_get_buffer(f, s->config, 256);
     pci_update_mappings(s);
+
+    if (version_id >= 2)
+        for (i=0; i<4; i++)
+            qemu_get_be32s(f, &s->irq_state[i]);
+
     return 0;
 }
 

--- hw/pci.c    2007-11-27 22:22:05.565379761 +0200
+++ hw/pci.c.2  2007-11-27 22:21:48.820440780 +0200
@@ -42,6 +42,7 @@
     PCIBus *next;
     /* The bus IRQ state is the logical OR of the connected devices.
        Keep a count of the number of devices with raised IRQs.  */
+    int nirq;
     int irq_count[];
 };
 
@@ -52,16 +53,51 @@
 static int pci_irq_index;
 static PCIBus *first_bus;
 
+static void pcibus_save(QEMUFile *f, void *opaque)
+{
+    PCIBus *bus = (PCIBus *)opaque;
+    int i;
+
+    qemu_put_be32s(f, &bus->nirq);
+    for (i=0; i<bus->nirq; i++)
+        qemu_put_be32s(f, &bus->irq_count[i]);
+}
+
+static int  pcibus_load(QEMUFile *f, void *opaque, int version_id)
+{
+    PCIBus *bus = (PCIBus *)opaque;
+    int i, nirq;
+
+    if (version_id != 1)
+        return -EINVAL;
+
+    qemu_get_be32s(f, &nirq);
+    if (bus->nirq != nirq) {
+        fprintf(stderr, "pcibus_load: nirq mismatch: src=%d dst=%d\n",
+                nirq, bus->nirq);
+        return -EINVAL;
+    }
+
+    for (i=0; i<nirq; i++)
+        qemu_get_be32s(f, &bus->irq_count[i]);
+
+    return 0;
+}
+
 PCIBus *pci_register_bus(pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
                          qemu_irq *pic, int devfn_min, int nirq)
 {
     PCIBus *bus;
+    static int nbus = 0;
+
     bus = qemu_mallocz(sizeof(PCIBus) + (nirq * sizeof(int)));
     bus->set_irq = set_irq;
     bus->map_irq = map_irq;
     bus->irq_opaque = pic;
     bus->devfn_min = devfn_min;
+    bus->nirq = nirq;
     first_bus = bus;
+    register_savevm("PCIBUS", nbus++, 1, pcibus_save, pcibus_load, bus);
     return bus;
 }
 

--- hw/piix_pci.c       2007-11-27 22:30:31.214559561 +0200
+++ hw/piix_pci.c.1     2007-11-27 22:23:43.254043529 +0200
@@ -57,6 +57,7 @@
 
 static uint32_t isa_page_descs[384 / 4];
 static uint8_t smm_enabled;
+static int pci_irq_levels[4];
 
 static void update_pam(PCIDevice *d, uint32_t start, uint32_t end, int r)
 {
@@ -139,22 +140,32 @@
 static void i440fx_save(QEMUFile* f, void *opaque)
 {
     PCIDevice *d = opaque;
+    int i;
+
     pci_device_save(d, f);
     qemu_put_8s(f, &smm_enabled);
+
+    for (i=0; i<4; i++)
+        qemu_put_be32s(f, &pci_irq_levels[i]);
 }
 
 static int i440fx_load(QEMUFile* f, void *opaque, int version_id)
 {
     PCIDevice *d = opaque;
-    int ret;
+    int ret, i;
 
-    if (version_id != 1)
+    if (version_id > 2)
         return -EINVAL;
     ret = pci_device_load(d, f);
     if (ret < 0)
         return ret;
     i440fx_update_memory_mappings(d);
     qemu_get_8s(f, &smm_enabled);
+
+    if (version_id >=2 )
+        for (i=0; i<4; i++)
+            qemu_get_be32s(f, &pci_irq_levels[i]);
+
     return 0;
 }
 
@@ -192,7 +203,7 @@
 
     d->config[0x72] = 0x02; /* SMRAM */
 
-    register_savevm("I440FX", 0, 1, i440fx_save, i440fx_load, d);
+    register_savevm("I440FX", 0, 2, i440fx_save, i440fx_load, d);
     *pi440fx_state = d;
     return b;
 }
@@ -205,8 +216,6 @@
 /* just used for simpler irq handling. */
 #define PCI_IRQ_WORDS   ((PCI_DEVICES_MAX + 31) / 32)
 
-static int pci_irq_levels[4];
-
 static void piix3_set_irq(qemu_irq *pic, int irq_num, int level)
 {
     int i, pic_irq, pic_level;