On Fri, 2022-03-18 at 15:17 +0100, Igor Mammedov wrote: > On Thu, 17 Mar 2022 11:13:44 +0000 David Woodhouse <dwmw2@xxxxxxxxxxxxx> wrote: > > Thus far, I didn't think anyone had been talking about a dependency on > > IOMMU. That doesn't make any sense at all. EXT_DEST_ID is perfectly > > sufficient for Linux kernels from 5.10 onwards and they don't need the > > IOMMU. > > IOMMU was required before EXT_DEST_ID due to irq-remapping dependency, > and that conservative config worked fine for both Linux and Windows > guests. That's why I've raised question if we should revert restriction > to the way it was back then. > > With Linux pre-5.10 guests, dmesg output at least complains about > IRQ remapping, so user has a small chance to be able to figure out > that IOMMU should be configured to get all CPUs working. > For post-5.10, all one gets is "bad cpu" without any clue as to why, > if EXT_DEST_ID is not advertised by hypervisor. > > It would be better if guest kernel printed some error/warning in that case. That case doesn't exist any more since your change. You *can't* launch qemu with that many vCPUs and not support EXT_DEST_ID.
Attachment:
smime.p7s
Description: S/MIME cryptographic signature
