On 26 July 2013 16:58, Yoder Stuart-B08248 <B08248@xxxxxxxxxxxxx> wrote: >> From: Peter Maydell [mailto:peter.maydell@xxxxxxxxxx] >> On 26 July 2013 15:54, Yoder Stuart-B08248 <B08248@xxxxxxxxxxxxx> wrote: >> > On the question of what GIC is exposed, wouldn't that be exposed >> > in the device tree and be orthogonal to the question of which >> > CPU a VM sees? >> > >> > Even if ARM v8 CPU requires a GICv3, it seems like a bad >> > assumption for an OS to just assume a GIC v3 without >> > checking what is advertised in the device tree. >> >> This seems to me to be something of a Linuxism. You can't >> assume that every guest OS will take a device tree as >> its means of figuring out what it's running on. > > I think of it less as a Linuxism and more of a question > of what 'platform' assumptions an OS makes: > -what is the physical address map of a system > -how many cpus are there, how are they started > -how are non-probe-able devices discovered > -are there runtime services in the platform > (the kind of things that ePAPR, UEFI, ACPI, etc address) Mmm. But "this A57 has the hardware that every A57 in existence has" isn't necessarily a huge stretch. We can probably just cross that bridge as we come to it, though: the major use case in VMs is going to be Linux guests... > However, a fair question related to this is whether we expect > to _only_ run operating systems in VMs. Do we need to run boot > firmware such as u-boot or UEFI in a VM? Yes, definitely. We can probably arrange for UEFI to adapt a bit to cope with running in a VM, though. (For a start, it needs to at least be able to cope with not having TrustZone available. Grant Likely was working through the (relatively minor) UEFI tweaks to get that running under QEMU.) -- PMM _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/cucslists/listinfo/kvmarm
