TIF_NOTITY_RESUME is evaluated on return to user space along with other TIF flags.
>From the kernels point of view a VMENTER is more or less equivalent to return to user space which means that at least a subset of TIF flags needs to be evaluated and handled. Currently KVM handles only TIF_SIGPENDING and TIF_NEED_RESCHED, but TIF_NOTIFY_RESUME is ignored. So pending task_work etc, is completely ignored until the vCPU thread actually goes all the way back into userspace/qemu. Use the newly provided notify_resume_pending() and tracehook_handle_notify_resume() to solve this similar to the existing handling of SIGPENDING. Signed-off-by: Thomas Gleixner <[email protected]> Cc: [email protected] Cc: Radim Krcmar <[email protected]> Cc: Paolo Bonzini <[email protected]> --- arch/x86/kvm/x86.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-) --- a/arch/x86/kvm/x86.c +++ b/arch/x86/kvm/x86.c @@ -52,6 +52,7 @@ #include <linux/irqbypass.h> #include <linux/sched/stat.h> #include <linux/sched/isolation.h> +#include <linux/tracehook.h> #include <linux/mem_encrypt.h> #include <trace/events/kvm.h> @@ -7972,7 +7973,8 @@ static int vcpu_enter_guest(struct kvm_v kvm_x86_ops->sync_pir_to_irr(vcpu); if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) - || need_resched() || signal_pending(current)) { + || need_resched() || signal_pending(current) + || notify_resume_pending()) { vcpu->mode = OUTSIDE_GUEST_MODE; smp_wmb(); local_irq_enable(); @@ -8172,6 +8174,10 @@ static int vcpu_run(struct kvm_vcpu *vcp ++vcpu->stat.signal_exits; break; } + + if (notify_resume_pending()) + tracehook_handle_notify_resume(); + if (need_resched()) { srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); cond_resched();
