On 17.10.2025 00:38, Marek Marczykowski-Górecki wrote: > On Fri, Oct 17, 2025 at 12:40:33AM +0300, Grygorii Strashko wrote: >> Another point, assume change like this is to be done for HVM_PARAM_VIRIDIAN >> - there are another HVM_PARAM_x which depend on build-time disabled >> features, like: >> HVM_PARAM_VM86_TSS_SIZED >> HVM_PARAM_PAGING_RING_PFN, >> HVM_PARAM_MONITOR_RING_PFN, >> HVM_PARAM_SHARING_RING_PFN, >> HVM_PARAM_IDENT_PT >> ... >> >> if corresponding features are build-time disabled, above HVM_PARAM_x >> become R/W with functionality NOP now. > > Are you sure? For me it looks like setting build-time disabled feature > returns -ENOSYS. Or do you mean some other interface than > xc_hvm_param_set().
Where do you see that ENOSYS coming from? In fact, for example, I don't see any of the *_RING_PFN even mentioned at all in hvm.c's parameter handling. Are you perhaps thinking of only the HVM=n case, whereas I expect Grygorii talks about the more fine-grained controls? That said, whether to uniformly fail requests for params solely related to build-time disabled features isn't quite clear. Arguably for e.g. paging and sharing, setting the ring PFN can as well be silently ignored (no events ever appearing), while failure would then be reported from other compiled-out logic. Jan
