Log a slightly different message when the missing-but-required slot is conventional PCI vs PCIe. Also correct/improve the comments about why auto-add of a PCI controller isn't supported when we're trying to create a slot for various different pci controllers. Signed-off-by: Laine Stump <laine@redhat.com> --- src/conf/domain_addr.c | 37 +++++++++++++++---- .../pcie-root-port-too-many.x86_64-latest.err | 2 +- 2 files changed, 30 insertions(+), 9 deletions(-) diff --git a/src/conf/domain_addr.c b/src/conf/domain_addr.c index 7d58e2222a..ef1b2bd69a 100644 --- a/src/conf/domain_addr.c +++ b/src/conf/domain_addr.c @@ -704,7 +704,7 @@ virDomainPCIAddressSetGrow(virDomainPCIAddressSet *addrs, * and we can't automatically decide which numa node to * associate it with) * - * VIR_CONNECT_TYPE_PCIE_SWITCH_DOWNSTREAM_PORT - we ndon't + * VIR_CONNECT_TYPE_PCIE_SWITCH_DOWNSTREAM_PORT - we don't * support this, because it can only plug into an * upstream-port, and the upstream port might need a * root-port; supporting this extra layer needlessly @@ -717,20 +717,41 @@ virDomainPCIAddressSetGrow(virDomainPCIAddressSet *addrs, * devices on the host, and are also outside the scope of our * "automatic-bus-expansion". * - * VIR_PCI_CONNECT_TYPE_PCI_BRIDGE (when the root bus is - * pci-root) - see the comment above in the case that handles - * adding a slot for pci-bridge to a guest with pcie-root. + * VIR_PCI_CONNECT_TYPE_PCI_BRIDGE if the root bus is pci-root + * but there are no free slots already available to plug in a + * pci-bridge, then the battle is already lost - the only way + * to get another open slot would be to auto-add a pci-bridge + * device, but that's what we're already trying to do - by + * definition either we wouldn't get here in the first place, + * or it's already too late. + * + * Alternatively if the root bus *isn't* pci-root, then + * either the root bus is pcie-root (in which case the user + * should be using a pcie-to-pci-bridge instead), or there is + * no PCI supported *at all*, and in both of those cases we + * should fail. + * + * Additionally, we obviously can't auto-add any type of PCIe + * controller if the root bus is pci-root, or if there is no + * PCI supported at all. * */ int existingContModel = virDomainPCIControllerConnectTypeToModel(flags); if (existingContModel >= 0) { - virReportError(VIR_ERR_INTERNAL_ERROR, - _("a PCI slot is needed to connect a PCI controller model='%1$s', but none is available, and it cannot be automatically added"), - virDomainControllerModelPCITypeToString(existingContModel)); + if (existingContModel == VIR_PCI_CONNECT_TYPE_PCI_EXPANDER_BUS || + existingContModel == VIR_PCI_CONNECT_TYPE_PCI_BRIDGE) { + virReportError(VIR_ERR_INTERNAL_ERROR, + _("a conventional PCI slot is needed to connect a PCI controller model='%1$s', but none is available, and it cannot be automatically added"), + virDomainControllerModelPCITypeToString(existingContModel)); + } else { + virReportError(VIR_ERR_INTERNAL_ERROR, + _("a PCIe slot is needed to connect a PCI controller model='%1$s', but none is available, and it cannot be automatically added"), + virDomainControllerModelPCITypeToString(existingContModel)); + } } else { virReportError(VIR_ERR_INTERNAL_ERROR, - _("Cannot automatically add a new PCI bus for a device with connect flags %1$.2x"), + _("Cannot automatically add a new PCI bus for a device with unrecognized connect flags %1$.2x"), flags); } return -1; diff --git a/tests/qemuxmlconfdata/pcie-root-port-too-many.x86_64-latest.err b/tests/qemuxmlconfdata/pcie-root-port-too-many.x86_64-latest.err index 9b24cfb7e0..e1d543a50a 100644 --- a/tests/qemuxmlconfdata/pcie-root-port-too-many.x86_64-latest.err +++ b/tests/qemuxmlconfdata/pcie-root-port-too-many.x86_64-latest.err @@ -1 +1 @@ -internal error: a PCI slot is needed to connect a PCI controller model='pcie-root-port', but none is available, and it cannot be automatically added +internal error: a PCIe slot is needed to connect a PCI controller model='pcie-root-port', but none is available, and it cannot be automatically added -- 2.51.0