Re: [PATCH 3/3] conf: restrict where dmi-to-pci-bridge can be connected

[Laine Stump <laine@xxxxxxxxx>]

On 08/08/2016 02:25 PM, Andrea Bolognani wrote:

On Fri, 2016-08-05 at 23:01 -0400, Laine Stump wrote:

libvirt had allowed a dmi-to-pci-bridge to be plugged in anywhere a
normal PCIe endpoint can be connected, but this is wrong - it will
only work if it's plugged into pcie-root (the PCIe root complex) or a
pcie-expander-bus (the qemu device pxb-pcie). This patch adjusts the
connection flags accordingly.
 
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1363648

Doesn't

  https://bugzilla.redhat.com/1363648#c4

tell us that we *need* to use a pcie-root-port when plugging
stuff into a pcie-expander-bus?

No, I think you're misunderstanding Marcel's comment:

> (In reply to Laine Stump from comment #2)
> > Marcel - is this actually supported by qemu?
> 
> Yes, and if doesn't work is a bug.
> The best way to think of pxb/pbx-pcie is that it exposes a new pcie.0 like bus.
> The only difference is that you cannot have devices plugged directly into it (integrated device).

I specifically asked him if what the reporter attempted (plugging a dmi-to-pci-bridge directly into a pcie-expander-bus) was supported, and he said "yes".

He then said that the only difference is that you can't plug [endpoint] devices directly into it. A dmi-to-pci-bridge is not an endpoint device. Aside from endpoint devices, the only other things you are supposed to plug into pcie-root are pcie-root-ports and dmi-to-pci-bridges.

Looks good otherwise, but I'd like to have Marcel's confirm
that plugging a dmi-to-pci-bridge directly into a
pcie-expander-bus is something we want before you push it.

I'll wait for Marcel to respond as well, but without this, there is no way to put a legacy PCI slot in a NUMA node on a Q35 guest, and it *does* apparently work.

diff --git a/tests/qemuxml2argvtest.c b/tests/qemuxml2argvtest.c
index d594836..933f9be 100644
--- a/tests/qemuxml2argvtest.c
+++ b/tests/qemuxml2argvtest.c
@@ -1631,6 +1631,22 @@ mymain(void)
              QEMU_CAPS_PCI_MULTIFUNCTION, QEMU_CAPS_ICH9_USB_EHCI1,
              QEMU_CAPS_DEVICE_VIDEO_PRIMARY,
              QEMU_CAPS_VGA_QXL, QEMU_CAPS_DEVICE_QXL);
+    DO_TEST_PARSE_ERROR("q35-dmi-bad-address1",
+            QEMU_CAPS_DEVICE_PCI_BRIDGE,
+            QEMU_CAPS_DEVICE_DMI_TO_PCI_BRIDGE,
+            QEMU_CAPS_DEVICE_IOH3420,
+            QEMU_CAPS_ICH9_AHCI,
+            QEMU_CAPS_PCI_MULTIFUNCTION, QEMU_CAPS_ICH9_USB_EHCI1,
+            QEMU_CAPS_DEVICE_VIDEO_PRIMARY,
+            QEMU_CAPS_VGA_QXL, QEMU_CAPS_DEVICE_QXL);
+    DO_TEST_PARSE_ERROR("q35-dmi-bad-address2",
+            QEMU_CAPS_DEVICE_PCI_BRIDGE,
+            QEMU_CAPS_DEVICE_DMI_TO_PCI_BRIDGE,
+            QEMU_CAPS_DEVICE_IOH3420,
+            QEMU_CAPS_ICH9_AHCI,
+            QEMU_CAPS_PCI_MULTIFUNCTION, QEMU_CAPS_ICH9_USB_EHCI1,
+            QEMU_CAPS_DEVICE_VIDEO_PRIMARY,
+            QEMU_CAPS_VGA_QXL, QEMU_CAPS_DEVICE_QXL);
      DO_TEST("q35-pm-disable",
              QEMU_CAPS_DEVICE_PCI_BRIDGE,
              QEMU_CAPS_DEVICE_DMI_TO_PCI_BRIDGE, QEMU_CAPS_ICH9_AHCI,

Can you maybe trim the capabilities a bit? QEMU_CAPS_VGA_QXL
and friends look like they're hardly a requirement for a test
case about PCI bridges.

Sure, I can remove those caps, but I think I'll also need to remove the <video> from the XML. (consider them removed).

Of course I'm okay with this if you're already planning to
post a follow-up patch that trims the capabilities of all
q35-* test cases ;)

I'll leave the mass changes to test case files to Cole :-P

--
libvir-list mailing list
libvir-list@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/libvir-list