When a guest enables MSI-X in PCI configuration space we walk
through the MSI-X vector table trying to guess what might get
used. We have to guess because we don't do anything with
writes to the vector table, so we look for clues ahead of time
to pre-enable the vectors we think will be used. This means
that instead of doing the sane thing and test the mask bit, we
test whether the data field contains a non-zero value. It's
amazing this works as well as it does.
However, two key things missed by doing this is that we don't
catch vector changes after enabling (ex. setting smp_affinity
on an irq) and we don't support guests that don't touch the
vector table prior to enabling the MSI-X capability (ex.
freebsd). This patch fixes both of these problems.
NB we're not actually masking vectors yet with this patch as
it's unclear whether we really have the ability to do this
without losing interrupts.
Signed-off-by: Alex Williamson <alex.williamson@xxxxxxxxxx>
---
hw/device-assignment.c | 92 ++++++++++++++++++++++++++++++++++++++++++------
1 files changed, 81 insertions(+), 11 deletions(-)
diff --git a/hw/device-assignment.c b/hw/device-assignment.c
index 71685ee..9791ec9 100644
--- a/hw/device-assignment.c
+++ b/hw/device-assignment.c
@@ -964,6 +964,11 @@ static void assigned_dev_update_msi(PCIDevice *pci_dev)
}
}
+static bool msix_masked(MSIXTableEntry *entry)
+{
+ return (entry->ctrl & cpu_to_le32(0x1)) != 0;
+}
+
static int assigned_dev_update_msix_mmio(PCIDevice *pci_dev)
{
AssignedDevice *adev = DO_UPCAST(AssignedDevice, dev, pci_dev);
@@ -975,17 +980,19 @@ static int assigned_dev_update_msix_mmio(PCIDevice *pci_dev)
/* Get the usable entry number for allocating */
for (i = 0; i < adev->msix_max; i++, entry++) {
- /* Ignore unused entry even it's unmasked */
- if (entry->data == 0) {
+ if (msix_masked(entry)) {
continue;
}
entries_nr++;
}
- if (entries_nr == 0) {
- fprintf(stderr, "MSI-X entry number is zero!\n");
- return -EINVAL;
+ DEBUG("MSI-X entries: %d\n", entries_nr);
+
+ /* It's valid to enable MSI-X with all entries masked */
+ if (!entries_nr) {
+ return 0;
}
+
msix_nr.assigned_dev_id = calc_assigned_dev_id(adev);
msix_nr.entry_nr = entries_nr;
r = kvm_assign_set_msix_nr(kvm_state, &msix_nr);
@@ -1003,7 +1010,7 @@ static int assigned_dev_update_msix_mmio(PCIDevice *pci_dev)
msix_entry.assigned_dev_id = msix_nr.assigned_dev_id;
entry = adev->msix_table;
for (i = 0; i < adev->msix_max; i++, entry++) {
- if (entry->data == 0) {
+ if (msix_masked(entry)) {
continue;
}
@@ -1075,9 +1082,12 @@ static void assigned_dev_update_msix(PCIDevice *pci_dev)
perror("assigned_dev_update_msix_mmio");
return;
}
- if (kvm_assign_irq(kvm_state, &assigned_irq_data) < 0) {
- perror("assigned_dev_enable_msix: assign irq");
- return;
+
+ if (assigned_dev->irq_entries_nr) {
+ if (kvm_assign_irq(kvm_state, &assigned_irq_data) < 0) {
+ perror("assigned_dev_enable_msix: assign irq");
+ return;
+ }
}
assigned_dev->girq = -1;
assigned_dev->irq_requested_type = assigned_irq_data.flags;
@@ -1438,11 +1448,71 @@ static void msix_mmio_write(void *opaque, target_phys_addr_t addr,
uint64_t val, unsigned size)
{
AssignedDevice *adev = opaque;
+ PCIDevice *pdev = &adev->dev;
+ uint16_t ctrl;
+ MSIXTableEntry orig;
+ int i = addr >> 4;
+
+ if (i >= adev->msix_max) {
+ return; /* Drop write */
+ }
- DEBUG("write to MSI-X entry table mmio offset 0x%lx, val 0x%lx\n",
- addr, val);
+ ctrl = pci_get_word(pdev->config + pdev->msix_cap + PCI_MSIX_FLAGS);
+
+ DEBUG("write to MSI-X table offset 0x%lx, val 0x%lx\n", addr, val);
+
+ if (ctrl & PCI_MSIX_FLAGS_ENABLE) {
+ orig = adev->msix_table[i];
+ }
memcpy((void *)((uint8_t *)adev->msix_table + addr), &val, size);
+
+ if (ctrl & PCI_MSIX_FLAGS_ENABLE) {
+ MSIXTableEntry *entry = &adev->msix_table[i];
+
+ if (!msix_masked(&orig) && msix_masked(entry)) {
+ /*
+ * Vector masked, disable it
+ *
+ * XXX It's not clear if we can or should actually attempt
+ * to mask or disable the interrupt. KVM doesn't have
+ * support for pending bits and kvm_assign_set_msix_entry
+ * doesn't modify the device hardware mask. Interrupts
+ * while masked are simply not injected to the guest, so
+ * are lost. Can we get away with always injecting an
+ * interrupt on unmask?
+ */
+ } else if (msix_masked(&orig) && !msix_masked(entry)) {
+ /* Vector unmasked */
+ if (i >= adev->irq_entries_nr || !adev->entry[i].type) {
+ /* Previously unassigned vector, start from scratch */
+ assigned_dev_update_msix(pdev);
+ return;
+ } else {
+ /* Update an existing, previously masked vector */
+ struct kvm_irq_routing_entry orig = adev->entry[i];
+ int ret;
+
+ adev->entry[i].u.msi.address_lo = entry->addr_lo;
+ adev->entry[i].u.msi.address_hi = entry->addr_hi;
+ adev->entry[i].u.msi.data = entry->data;
+
+ ret = kvm_update_routing_entry(&orig, &adev->entry[i]);
+ if (ret) {
+ fprintf(stderr,
+ "Error updating irq routing entry (%d)\n", ret);
+ return;
+ }
+
+ ret = kvm_commit_irq_routes();
+ if (ret) {
+ fprintf(stderr,
+ "Error committing irq routes (%d)\n", ret);
+ return;
+ }
+ }
+ }
+ }
}
static const MemoryRegionOps msix_mmio_ops = {
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
