On Tue, 2022-04-19 at 15:45 +0000, Sean Christopherson wrote: > On Tue, Apr 19, 2022, Maxim Levitsky wrote: > > On Fri, 2022-04-15 at 00:43 +0000, Sean Christopherson wrote: > > > Add wrappers to acquire/release KVM's SRCU lock when stashing the index > > > in vcpu->src_idx, along with rudimentary detection of illegal usage, > > > e.g. re-acquiring SRCU and thus overwriting vcpu->src_idx. Because the > > > SRCU index is (currently) either 0 or 1, illegal nesting bugs can go > > > unnoticed for quite some time and only cause problems when the nested > > > lock happens to get a different index. > > > > > > Wrap the WARNs in PROVE_RCU=y, and make them ONCE, otherwise KVM will > > > likely yell so loudly that it will bring the kernel to its knees. > > > > > > Signed-off-by: Sean Christopherson <seanjc@xxxxxxxxxx> > > > --- > > ... > > > Looks good to me overall. > > > > Note that there are still places that acquire the lock and store the idx into > > a local variable, for example kvm_xen_vcpu_set_attr and such. > > I didn't check yet if these should be converted as well. > > Using a local variable is ok, even desirable. Nested/multiple readers is not an > issue, the bug fixed by patch 1 is purely that kvm_vcpu.srcu_idx gets corrupted. Makes sense. I still recal *that* bug in AVIC inhibition where that srcu lock was a major PITA, but now I remember that it was not due to nesting of the lock, but rather fact that we attempted to call syncronize_srcu or something like that with it held. > > In an ideal world, KVM would _only_ track the SRCU index in local variables, but > that would require plumbing the local variable down into vcpu_enter_guest() and > kvm_vcpu_block() so that SRCU can be unlocked prior to entering the guest or > scheduling out the vCPU. > It all makes sense now - thanks. Best regards, Maxim Levistky
