bhyve: Unsupported MSI-X configuration: 2/0/0

[Nikos Vassiliadis]

On 11/29/14 06:02, Anish Gupta wrote:
>>> bge0 at pci0:2:0:0:        class=0x020000 card=0x06471025
>>> chip=0x16b514e4 rev=0x10 hdr=0x00
>>> sdhci_pci0 at pci0:2:0:1:  class=0x080501 card=0x06471025
>>> chip=0x16bc14e4 rev=0x10 hdr=0x00
>>> none2 at pci0:2:0:2:       class=0x088000 card=0x06471025
>>> chip=0x16be14e4 rev=0x10 hdr=0x00
>>> none3 at pci0:2:0:3:       class=0x088000 card=0x06471025
>>> chip=0x16bf14e4 rev=0x10 hdr=0x00
>
> Passthrough stub driver is part of vmm.ko and if it was present early in
> boot,  you should see ppt at pciD:B:S:F[Domain:Bus:Slot:Function] in
> pciocnf list above. Given that bge driver claimed 2/0/0, most likely
> vmm.ko failed to load or probably was not present. Do you have
> vmm_load=“YES” in /boot/loader.conf as mentioned in
> https://wiki.freebsd.org/bhyve/pci_passthru?

Yes. To make testing easier:
I removed everything from loader.conf and I removed a few drivers(bge 
and a few others) that were built-in the kernel, so the host kernel 
won't use them

and now I am testing with kenv and loading and unloading vmm.ko. I 
always see the correct ppt devices in the dmesg when vmm.ko is loaded.

>
> If you already have loader.conf configured correctly, you can try to
> manually load vmm.ko once system is booted and see dmesg for any problem
> in loading it like kernel mismatch, missing symbol[One common I usually
> encounter is KTRACE define in sys/modules/vmm/Makefile but kernel config
> is missing “option KTRACE”] etc.

Everything seems ok here as well, I get nothing but the ppt devices.

One thing I noticed and seems different is this:
> ppt0 mem 0xb3430000-0xb343ffff,0xb3440000-0xb344ffff irq 16 at device 0.0 on pci2
> ppt1 mem 0xb3400000-0xb340ffff irq 17 at device 0.1 on pci2
> ppt2 mem 0xb3410000-0xb341ffff irq 17 at device 0.2 on pci2
> ppt3 mem 0xb3420000-0xb342ffff irq 17 at device 0.3 on pci2

so ppt0 (the bge device) has two mem ranges...

Another thing I should mention is that "acpidump -t | grep DMAR"
reports nothing. It is mentioned in the documentation but I didn't see 
until now.