On Thu, 2025-09-18 at 08:47 +0200, Peter Zijlstra wrote: > On Thu, Sep 18, 2025 at 11:33:18AM +0530, Naman Jain wrote: > > > Thank you so much Sean and Paolo for your valuable inputs. I will > > try out these things. Summarizing the suggestions here: > > * Use noinstr (no instrumentation) > > * Have separate .S file > > * Don't use "register asm". > > * Use static calls for solving IBT problems > > * RAX:RCX is probably ok to be used, considering ABI. Whether we > > would still need to use STACK_FRAME_NON_STANDARD, I am not sure, > > but I will see based on how it goes. > > > > I hope this addresses the concerns Peter raised. If there's > > anything I might have missed, I'm happy to make further adjustments > > if needed. > > It would be a definite improvement. I'm just *really* sad people > still create interfaces like this, even though we've known for years > how bad they are.
Thanks for that. Our only defence at Microsoft is that the hypervisor ABI for this is 10 years old and counting. > At some point we've really have to push back and say enough is > enough. Although we can't really do anything about the old ABI since we have tons of things that use it, there is interest in Microsoft for co- designing a new ABI that we could add to hyper-v for Linux use and which would also be a part of bringing up and using virtual secure mode (VSM) in KVM via the planes API. I was thinking the best way of discussing it as a community would be the PUCK call, but if there are alternative forums, we could use that as well. This is going to be pretty long term: besides designing and testing a new ABI, we were counting on Amazon upstreaming their VSM implementation for KVM and we're a bit under resourced here if it turns out we have to do everything. Regards, James
