illumos gate

Since the compiler obviously knows that this statement will always be true, I tried to find a warning setting which would output that. However, even with every single warning enabled (via parsing the output of gcc -Wall -Wextra -Q --help=warning) this was not reported.

Jorge Schrauwen has tested this change on two of his physical hosts that run a number of VMs which use PCI passthrough. He has a number of different passthrough devices, some of which support FLR and some which do not. He has used dtrace to confirm that the ppt_flr() function is now actually performing the reset.

Of particular note is that this change appears to improve the behaviour of a part of his setup where a four-port igb device is split between VMs:

<sjorge> I stopped one of the 2 VM that have one of the igb nics 5 times and the other one kept working
<sjorge> Before it would sometimes break the nic until I either stopped both VMs and started them again, or rebooted (the physical host) if the former didn't work

# for i in 0 1 2 3 4 5; do /usr/lib/pci/pcieadm show-cfgspace -d ppt${i} pcie.devcap.flr; done
PCI Express Capability (0x10)
  Device Capabilities: 0x8fc0
    |--> Function Level Reset: unsupported (0x0)
PCI Express Capability (0x10)
  Device Capabilities: 0x8fc0
    |--> Function Level Reset: unsupported (0x0)
PCI Express Capability (0x10)
  Device Capabilities: 0x8fc0
    |--> Function Level Reset: unsupported (0x0)
PCI Express Capability (0x10)
  Device Capabilities: 0x8fc0
    |--> Function Level Reset: unsupported (0x0)
PCI Express Capability (0x10)
  Device Capabilities: 0x10008ce2
    |--> Function Level Reset: supported (0x10000000)
PCI Express Capability (0x10)
  Device Capabilities: 0x10008ce2
    |--> Function Level Reset: supported (0x10000000)