Hi, On 4/1/20 5:58 PM, Marc Zyngier wrote: > Christoffer recently pointed out that we don't narrow the arguments to > SMC32 PSCI functions called by a 64bit guest. This could result in a > guest failing to boot its secondary CPUs if it had junk in the upper > 32bits. Yes, this is silly, but the guest is allowed to do that. Duh. > > Whist I was looking at this, it became apparent that we allow a 32bit > guest to call 64bit functions, which the spec explicitly forbids. Oh > well, another patch. > > This has been lightly tested, but I feel that we could do with a new > set of PSCI corner cases in KVM-unit-tests (hint, nudge... ;-). Good idea. I was already planning to add new PSCI and timer tests, I'm waiting for Paolo to merge the pull request from Drew, which contains some fixes for the current tests. > > Marc Zyngier (2): > KVM: arm64: PSCI: Narrow input registers when using 32bit functions > KVM: arm64: PSCI: Forbid 64bit functions for 32bit guests > > virt/kvm/arm/psci.c | 40 ++++++++++++++++++++++++++++++++++++++++ > 1 file changed, 40 insertions(+) > I started reviewing the patches and I have a question. I'm probably missing something, but why make the changes to the PSCI code instead of making them in the kvm_hvc_call_handler function? From my understanding of the code, making the changes there would benefit all firmware interface that use SMCCC as the communication protocol, not just PSCI. Thanks, Alex
