On Thu, Dec 19, 2024, Paolo Bonzini wrote: > On 12/19/24 18:49, Maxim Levitsky wrote: > > > Here I probably would have preferred an unconditional tracepoint giving > > > RAX/RBX/RCX/RDX after a nested vmexit. This is not exactly what Sean > > > wanted but perhaps it strikes a middle ground? I know you wrote this > > > for a debugging tool, do you really need to have everything in a single > > > tracepoint, or can you correlate the existing exit tracepoint with this > > > hypothetical trace_kvm_nested_exit_regs, to pick RDMSR vs. WRMSR? > > > > Hi! > > > > If the new trace_kvm_nested_exit_regs tracepoint has a VM exit number > > argument, then I can enable this new tracepoint twice with a different > > filter (vm_exit_num number == msr and vm_exit_num == vmcall), and each > > instance will count the events that I need. > > > > So this can work. > Ok, thanks. On one hand it may make sense to have trace_kvm_exit_regs and > trace_kvm_nested_exit_regs (you can even extend the TRACE_EVENT_KVM_EXIT > macro to generate both the exit and the exit_regs tracepoint). On the other > hand it seems to me that this new tracepoint is kinda reinventing the wheel; > your patch adding nested equivalents of trace_kvm_hypercall and > trace_kvm_msr seems more obvious to me. > > I see Sean's point in not wanting one-off tracepoints, on the other hand > there is value in having similar tracepoints for the L1->L0 and L2->L0 > cases. I don't understand why we want two (or three, or five) tracepoints for the same thing. I want to go the opposite direction and (a) delete kvm_nested_vmexit and then (b) rename kvm_nested_vmexit_inject => kvm_nested_vmexit so that it pairs with kvm_nested_vmenter. Similary, having kvm_nested_intr_vmexit is asinine when kvm_nested_vmexit_inject captures *more* information about the IRQ itself. I don't see the point of trace_kvm_nested_exit_regs. Except for L1 vs. L2, it's redundant. kvm_nested_vmexit_inject and kvm_nested_vmenter are useful because they capture novel information. > I'll let him choose between the two possibilities (a new *_exit_regs > pair, or just apply this patch) but I think there should be one of these > two. Anything but a pair. Why can't we capture L1 vs. L2 in the tracepoints and call it a day?
