On Wed, Nov 21, 2018 at 2:01 PM Logan Gunthorpe <log...@deltatee.com> wrote: > > > > On 2018-11-21 2:54 p.m., Alistair Francis wrote: > > The last time I tested this it worked (although I might not have > > tested interrupts) and now it doesn't. Nothing has changed in the > > series, my guest software has changed though. I can see the root > > complex, but no devices underneath it. > > > > Sorry for the confusion with the MicroSemi PCIe root complex. > > > > I think I'll just drop this patch. It doesn't work for 32-bit guests, > > but I would like to model the hardware so I don't want to change the > > addresses. We can revisit this in the future along with the gpex high > > MMIO support and USB support. > > Yeah, I'm less concerned about all the board variant stuff as I am about > having something actually work. I don't want to have to jump through any > bbl hoops or anything to use it if it doesn't work out of the box. And > you're right, getting the virt pci support upstream seems more important > and this issue can be revisited later. > > On an slight tangent: what root device do you use with the sifive_u > machine? I can't find any virtual hardware that works. If I had PCI I > could use NVMe, but that's not working. And for this reason I've only > been using the 'virt' device.
Yeah, it's a real pain. I had a go at adding the SD card [1] but never got it fully working. Normally I just hack in the virtIO MMIO regions and use those. 1: https://github.com/alistair23/qemu/tree/mainline/alistair/sifive_spi.next I just did a quick rebase, fully untested Alistair > > Thanks, > > Logan
