[PATCH 3/3] qemu-kvm: device assignment: emulate MSI-X mask bits

[Sheng Yang <sheng@linux.intel.com>]

This patch emulated MSI-X per vector mask bit on assigned device.

Signed-off-by: Sheng Yang <sheng@linux.intel.com>
---
 hw/device-assignment.c |  127 +++++++++++++++++++++++++++++++++++++++++++++++-
 1 files changed, 126 insertions(+), 1 deletions(-)

diff --git a/hw/device-assignment.c b/hw/device-assignment.c
index 4edae52..564c5ab 100644
--- a/hw/device-assignment.c
+++ b/hw/device-assignment.c
@@ -1146,6 +1146,7 @@ static int get_msix_entries_max_nr(AssignedDevice *adev)
     return entries_max_nr;
 }
 
+#define MSIX_VECTOR_MASK 0x1
 static int get_msix_valid_entries_nr(AssignedDevice *adev,
 				     uint16_t entries_max_nr)
 {
@@ -1159,7 +1160,11 @@ static int get_msix_valid_entries_nr(AssignedDevice *adev,
         memcpy(&msg_ctrl, va + i * 16 + 12, 4);
         memcpy(&msg_data, va + i * 16 + 8, 4);
         /* Ignore unused entry even it's unmasked */
+#ifdef KVM_CAP_DEVICE_MSIX_MASK
+        if (msg_data == 0 || (msg_ctrl & MSIX_VECTOR_MASK))
+#else
         if (msg_data == 0)
+#endif
             continue;
         entries_nr ++;
     }
@@ -1188,6 +1193,8 @@ static int assigned_dev_update_msix_mmio(PCIDevice *pci_dev,
     }
 
     free_dev_irq_entries(adev);
+    memset(pci_dev->msix_entry_used, 0, KVM_MAX_MSIX_PER_DEV *
+                                        sizeof(*pci_dev->msix_entry_used));
     adev->irq_entries_nr = entries_nr;
     adev->entry = calloc(entries_nr, sizeof(struct kvm_irq_routing_entry));
     if (!adev->entry) {
@@ -1223,6 +1230,7 @@ static int assigned_dev_update_msix_mmio(PCIDevice *pci_dev,
 
         msix_entry.gsi = adev->entry[entries_nr].gsi;
         msix_entry.entry = i;
+        pci_dev->msix_entry_used[i] = 1;
         r = kvm_assign_set_msix_entry(kvm_context, &msix_entry);
         if (r) {
             fprintf(stderr, "fail to set MSI-X entry! %s\n", strerror(-r));
@@ -1266,6 +1274,8 @@ static void assigned_dev_update_msix(PCIDevice *pci_dev, int enable_msix)
             perror("assigned_dev_update_msix: deassign irq");
 
         assigned_dev->irq_requested_type = 0;
+        memset(pci_dev->msix_entry_used, 0, KVM_MAX_MSIX_PER_DEV *
+                                        sizeof(*pci_dev->msix_entry_used));
     }
 
     entries_max_nr = get_msix_entries_max_nr(assigned_dev);
@@ -1273,10 +1283,16 @@ static void assigned_dev_update_msix(PCIDevice *pci_dev, int enable_msix)
         fprintf(stderr, "assigned_dev_update_msix: MSI-X entries_max_nr == 0");
         return;
     }
+    /*
+     * Guest may try to enable MSI-X before setting MSI-X entry done, so
+     * let's wait until guest unmask the entries.
+     */
     entries_nr = get_msix_valid_entries_nr(assigned_dev, entries_max_nr);
     if (entries_nr == 0) {
+#ifndef KVM_CAP_DEVICE_MSIX_MASK
         if (enable_msix)
             fprintf(stderr, "MSI-X entry number is zero!\n");
+#endif
         return;
     }
     if (enable_msix) {
@@ -1320,7 +1336,8 @@ static void assigned_device_pci_cap_write_config(PCIDevice *pci_dev, uint32_t ad
         if (address <= ctrl_pos && address + len > ctrl_pos) {
             ctrl_pos--; /* control is word long */
             ctrl_word = (uint16_t *)(pci_dev->config + ctrl_pos);
-            assigned_dev_update_msix(pci_dev, (*ctrl_word & PCI_MSIX_ENABLE));
+            assigned_dev_update_msix(pci_dev,
+                    (*ctrl_word & PCI_MSIX_ENABLE) && !(*ctrl_word & PCI_MSIX_MASK));
 	}
         pos += PCI_CAPABILITY_CONFIG_MSIX_LENGTH;
     }
@@ -1412,6 +1429,104 @@ static uint32_t msix_mmio_readw(void *opaque, target_phys_addr_t addr)
             (8 * (addr & 3))) & 0xffff;
 }
 
+#ifdef KVM_CAP_DEVICE_MSIX_MASK
+static void msix_mmio_access_mask_bit(AssignedDevice *adev, int index)
+{
+    void *page = adev->msix_table_page;
+    uint32_t msg_ctrl, msg_data, msg_upper_addr, msg_addr;
+    struct kvm_assigned_msix_entry msix_entry;
+    int r = 0, pos, ctrl_word, entry_idx, entries_max_nr;
+    struct kvm_irq_routing_entry old_entry = {};
+
+    memcpy(&msg_addr, (char *)page + index * 16, 4);
+    memcpy(&msg_upper_addr, (char *)page + index * 16 + 4, 4);
+    memcpy(&msg_data, (char *)page + index * 16 + 8, 4);
+    memcpy(&msg_ctrl, (char *)page + index * 16 + 12, 4);
+    DEBUG("Access mask bit: MSI-X entries index %d: "
+            "msg_addr 0x%x, msg_upper_addr 0x%x, msg_data 0x%x, vec_ctl 0x%x\n",
+            index, msg_addr, msg_upper_addr, msg_data, msg_ctrl);
+
+    if (adev->cap.available & ASSIGNED_DEVICE_CAP_MSI)
+        pos = adev->dev.cap.start + PCI_CAPABILITY_CONFIG_MSI_LENGTH;
+    else
+        pos = adev->dev.cap.start;
+
+    ctrl_word = *(uint16_t *)(adev->dev.config + pos + 2);
+
+    if (!((ctrl_word & PCI_MSIX_ENABLE) && !(ctrl_word & PCI_MSIX_MASK)))
+        return;
+
+    if (!adev->dev.msix_entry_used[index] && (msg_ctrl & MSIX_VECTOR_MASK) == 0) {
+        DEBUG("Try to modify unenabled MSI-X entry %d's mask. "
+                "Reenable MSI-X.\n",
+                index);
+        assigned_dev_update_msix(&adev->dev, 1);
+        return;
+    }
+
+    /* find the correlated index of adev->entry */
+    entries_max_nr = get_msix_entries_max_nr(adev);
+    entry_idx = 0;
+    while (entry_idx < entries_max_nr)
+        if (adev->dev.msix_entry_used[index]) {
+            if (entry_idx == index)
+                break;
+            entry_idx ++;
+        }
+    if (entry_idx >= entries_max_nr) {
+        fprintf(stderr, "msix_mmio_access_mask_bit: Entry idx exceed limit!\n");
+        return;
+    }
+
+    msix_entry.assigned_dev_id = calc_assigned_dev_id(adev->h_segnr,
+            adev->h_busnr,
+            (uint8_t)adev->h_devfn);
+    msix_entry.gsi = adev->entry[entry_idx].gsi;
+    msix_entry.entry = index;
+    if (msg_ctrl & MSIX_VECTOR_MASK)
+        msix_entry.flags = KVM_MSIX_FLAG_MASK;
+    else
+        msix_entry.flags = 0;
+    DEBUG("set MSI-X index %d, esi 0x%x, mask %d\n",
+            index, msix_entry.gsi, msix_entry.flags);
+    r = kvm_assign_set_msix_entry(kvm_context, &msix_entry);
+    if (r) {
+        perror("msix_mmio_access_mask_bit: "
+                "fail to set MSI-X entry!");
+        return;
+    }
+
+    if (msix_entry.flags == 0) {
+        old_entry.gsi = adev->entry[entry_idx].gsi;
+        old_entry.type = KVM_IRQ_ROUTING_MSI;
+        old_entry.flags = 0;
+        old_entry.u.msi.address_lo = adev->entry[entry_idx].u.msi.address_lo;
+        old_entry.u.msi.address_hi = adev->entry[entry_idx].u.msi.address_hi;
+        old_entry.u.msi.data = adev->entry[entry_idx].u.msi.data;
+        adev->entry[entry_idx].u.msi.address_lo = msg_addr;
+        adev->entry[entry_idx].u.msi.address_hi = msg_upper_addr;
+        adev->entry[entry_idx].u.msi.data = msg_data;
+        if (memcmp(&adev->entry[entry_idx].u.msi, &old_entry.u.msi,
+                    sizeof old_entry.u.msi)) {
+            int r;
+            r = kvm_update_routing_entry(kvm_context, &old_entry,
+                    &adev->entry[entry_idx]);
+            if (r) {
+                perror("msix_mmio_access_mask_bit: "
+                        "kvm_update_routing_entry failed\n");
+                return;
+            }
+            r = kvm_commit_irq_routes(kvm_context);
+            if (r) {
+                perror("msix_mmio_access_mask_bit: "
+                        "kvm_commit_irq_routes failed\n");
+                return;
+            }
+        }
+    }
+}
+#endif
+
 static void msix_mmio_writel(void *opaque,
                              target_phys_addr_t addr, uint32_t val)
 {
@@ -1422,6 +1537,12 @@ static void msix_mmio_writel(void *opaque,
     DEBUG("write to MSI-X entry table mmio offset 0x%lx, val 0x%x\n",
 		    addr, val);
     memcpy((void *)((char *)page + offset), &val, 4);
+
+#ifdef KVM_CAP_DEVICE_MSIX_MASK
+    /* Check if mask bit is being accessed */
+    if (offset % 16 == 12)
+	    msix_mmio_access_mask_bit(adev, offset / 16);
+#endif
 }
 
 static void msix_mmio_writew(void *opaque,
@@ -1459,6 +1580,8 @@ static int assigned_dev_register_msix_mmio(AssignedDevice *dev)
     memset(dev->msix_table_page, 0, 0x1000);
     dev->mmio_index = cpu_register_io_memory(
                         msix_mmio_read, msix_mmio_write, dev);
+    dev->dev.msix_entry_used = qemu_mallocz(KVM_MAX_MSIX_PER_DEV *
+                                        sizeof *dev->dev.msix_entry_used);
     return 0;
 }
 
@@ -1475,6 +1598,8 @@ static void assigned_dev_unregister_msix_mmio(AssignedDevice *dev)
                 strerror(errno));
     }
     dev->msix_table_page = NULL;
+    free(dev->dev.msix_entry_used);
+    dev->dev.msix_entry_used = NULL;
 }
 
 static int assigned_initfn(struct PCIDevice *pci_dev)
-- 
1.7.0.1