From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jan Kiszka Subject: Re: [PATCH 1/11] QEMU/KVM: Fix deadlocks in monitor and debugger Date: Tue, 27 May 2008 15:00:39 +0200 Message-ID: <483C05F7.8040104@siemens.com> References: <4839B14A.3010406@web.de> <483B351E.6010305@web.de> <483BD633.3080302@qumranet.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-15 Content-Transfer-Encoding: 7bit Cc: kvm-devel , Hollis Blanchard , Jerone Young , Joerg Roedel To: Avi Kivity Return-path: Received: from lizzard.sbs.de ([194.138.37.39]:23553 "EHLO lizzard.sbs.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753470AbYE0NBH (ORCPT ); Tue, 27 May 2008 09:01:07 -0400 In-Reply-To: <483BD633.3080302@qumranet.com> Sender: kvm-owner@vger.kernel.org List-ID: Avi Kivity wrote: > Jan Kiszka wrote: >> Some monitor commands as well as the vm_stop() issued by the gdbstub on >> external interruption so far deadlock on vcpu locks in the kernel. Patch >> below resolves the issue by temporarily or permanently stopping all vcpu >> threads before issuing the related KVM IOCTLs. It enables, e.g., to >> break into guest code spinning in the vcpu and to use things like "info >> cpus" in the monitor. >> > > I implemented the alternative on_vcpu() approach for this (similar to > smp_call_function_single in the kernel) which solves the livelock > without resorting to stopping the VM. I assume this pattern should then be applied to kvm_guest_debug (and later on kvm_set_guest_debug) as well? You missed to fix that bug. Here is a quick patch to complete the work, will post rebased versions of my remaining patches later today: ----------- Use on_vcpu to call into kvm_guest_debug. Fix on_vpuc usage in kvm_load_registers. Signed-off-by: Jan Kiszka --- qemu/qemu-kvm.c | 32 +++++++++++++++++++++++--------- 1 file changed, 23 insertions(+), 9 deletions(-) Index: b/qemu/qemu-kvm.c =================================================================== --- a/qemu/qemu-kvm.c +++ b/qemu/qemu-kvm.c @@ -187,7 +187,7 @@ static void kvm_do_load_registers(void * void kvm_load_registers(CPUState *env) { if (kvm_enabled()) - on_vcpu(env->cpu_index, kvm_do_load_registers, env); + on_vcpu(env, kvm_do_load_registers, env); } static void kvm_do_save_registers(void *_env) @@ -816,23 +816,37 @@ int kvm_qemu_init_env(CPUState *cenv) return kvm_arch_qemu_init_env(cenv); } +struct kvm_guest_debug_data { + struct kvm_debug_guest dbg; + int err; +}; + +void kvm_invoke_guest_debug(void *data) +{ + struct kvm_guest_debug_data *dbg_data = data; + + dbg_data->err = kvm_guest_debug(kvm_context, cpu_single_env->cpu_index, + &dbg_data->dbg); +} + int kvm_update_debugger(CPUState *env) { - struct kvm_debug_guest dbg; + struct kvm_guest_debug_data data; int i; - memset(dbg.breakpoints, 0, sizeof(dbg.breakpoints)); + memset(data.dbg.breakpoints, 0, sizeof(data.dbg.breakpoints)); - dbg.enabled = 0; + data.dbg.enabled = 0; if (env->nb_breakpoints || env->singlestep_enabled) { - dbg.enabled = 1; + data.dbg.enabled = 1; for (i = 0; i < 4 && i < env->nb_breakpoints; ++i) { - dbg.breakpoints[i].enabled = 1; - dbg.breakpoints[i].address = env->breakpoints[i]; + data.dbg.breakpoints[i].enabled = 1; + data.dbg.breakpoints[i].address = env->breakpoints[i]; } - dbg.singlestep = env->singlestep_enabled; + data.dbg.singlestep = env->singlestep_enabled; } - return kvm_guest_debug(kvm_context, env->cpu_index, &dbg); + on_vcpu(env, kvm_invoke_guest_debug, &data); + return data.err; }