[Qemu-devel] [Bug 1384892] Re: RTL8168 NIC VFIO not working anymore since QEMU 2.1

[Thorsten Kohfeldt]

Alex, are you sure about the constant 0x10000000U (bit 27) being used in the 
code below ?
Shouldn't it rather be a 0x80000000U (bit 31 which you commented about) ?
I added a /* NOT REACHED ? */ below, as I feel that might be the problem.

Florian,
are you willing to test the so modified source with and without Alex's patch of 
comment #4 ?
Note that the constant 0x10000000U is also used for ex-or in the function 
vfio_rtl8168_window_quirk_read(),
where it should (I think) also be 0x80000000U (or maybe just 0, i.e. no 
modification of the saved value ?)

AND:
Shouldn't variable "uint64_t val;" in function vfio_rtl8168_window_quirk_read() 
be initialised 0,
as it is size 8, which is larger than "size" (which I gather is 4) ?

Anyway, just to keep everybody updated about newer versions of QEMU:
The relevant code blocks seem to have moved to hw/vfio/pci.c in QEMU 2.3.

static void vfio_rtl8168_window_quirk_write(void *opaque, hwaddr addr,
                                            uint64_t data, unsigned size)
{
    VFIOQuirk *quirk = opaque;
    VFIODevice *vdev = quirk->vdev;

    switch (addr) {
    case 4: /* address */
        if ((data & 0x7fff0000) == 0x10000) {
            if (data & 0x10000000U &&
                vdev->pdev.cap_present & QEMU_PCI_CAP_MSIX) {
/* NOT REACHED ? */
                DPRINTF("%s MSI-X table write(%04x:%02x:%02x.%d)\n",
                        memory_region_name(&quirk->mem), vdev->host.domain,
                        vdev->host.bus, vdev->host.slot, vdev->host.function);

                io_mem_write(&vdev->pdev.msix_table_mmio,
                             (hwaddr)(quirk->data.address_match & 0xfff),
                             data, size);
            }

            quirk->data.flags = 1;
            quirk->data.address_match = data;

            return;
        }

-- 
You received this bug notification because you are a member of qemu-
devel-ml, which is subscribed to QEMU.
https://bugs.launchpad.net/bugs/1384892

Title:
  RTL8168 NIC VFIO not working anymore since QEMU 2.1

Status in QEMU:
  New

Bug description:
  After upgrading QEMU from 2.0 to 2.1 (and libiscsi from 1.7.0 to 1.12 as a 
dependency) my two RTL8168 NICs stopped working.
  The NICs do not respond to any command and even the LEDs on the network 
connection turn off, a few seconds after the VM started.
  To get them back running I had to downgrade to 2.0 and restart the system.
  Unfortunately, I have no clue what to do or how to debug this problem since 
there are no specific errors logged.
  I tried two different VMs: Debian Wheezy and IPFire (see attachment for 
further details).
  The QEMU 2.1 changelog states "Support for RTL8168 NICs." so there were some 
major changes done, I guess.

  On the IPFire guest the kernel log shows many of these lines:
  r8169 0000:00:07.0 green1: rtl_eriar_cond == 1 (loop: 100, delay: 100)
  r8169 0000:00:07.0 green1: rtl_phy_reset_cond == 1 (loop: 100, delay: 1)

  On the Debian guest there is only:
  r8169 0000:00:07.0: firmware: agent loaded rtl_nic/rtl8168e-3.fw into memory
  r8169 0000:00:07.0: lan0: link down
  ADDRCONF(NETDEV_UP): lan0: link is not ready

  The commandline for IPFire can be seen in the attachment. It is the same for 
Debian.
  There are also the complete kernel logs for the working (2.0) and non-working 
(2.1) cases.

To manage notifications about this bug go to:
https://bugs.launchpad.net/qemu/+bug/1384892/+subscriptions