| Message ID | 156352619712.50279.1247507600735238783.stgit@lep8c.aus.stglabs.ibm.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v4] ppc: make idle_timer a per-cpu variable | expand |
On Fri, Jul 19, 2019 at 03:51:26AM -0500, Shivaprasad G Bhat wrote: > The current code is broken for more than vcpu as > each thread would overwrite idle_timer and there were > memory leaks. > > Make it part of PowerPCCPU so that every thread has a > separate one. Avoid using the timer_new_ns which is > not the preferred way to create timers. > > Signed-off-by: Shivaprasad G Bhat <sbhat@linux.ibm.com> > Reviewed-by: Greg Kurz <groug@kaod.org> So, this looks correct, but.. ..the whole idle_timer logic is used only in the case that we're running with a KVM that doesn't support KVM_CAP_PPC_IRQ_LEVEL. That support appears to have been in since around v2.6.36 some 9 years ago. We could probably just drop support for such old kernels in qemu instead. > --- > v3: https://lists.gnu.org/archive/html/qemu-devel/2019-07/msg04375.html > Changes from v3: > - Calling timer_del() before timer_deinit() > > target/ppc/cpu.h | 1 + > target/ppc/kvm.c | 32 +++++++++++++++++--------------- > 2 files changed, 18 insertions(+), 15 deletions(-) > > diff --git a/target/ppc/cpu.h b/target/ppc/cpu.h > index c9beba2a5c..521086d91a 100644 > --- a/target/ppc/cpu.h > +++ b/target/ppc/cpu.h > @@ -1190,6 +1190,7 @@ struct PowerPCCPU { > void *machine_data; > int32_t node_id; /* NUMA node this CPU belongs to */ > PPCHash64Options *hash64_opts; > + QEMUTimer idle_timer; > > /* Fields related to migration compatibility hacks */ > bool pre_2_8_migration; > diff --git a/target/ppc/kvm.c b/target/ppc/kvm.c > index 8a06d3171e..52d3292f45 100644 > --- a/target/ppc/kvm.c > +++ b/target/ppc/kvm.c > @@ -87,18 +87,6 @@ static int cap_large_decr; > > static uint32_t debug_inst_opcode; > > -/* > - * XXX We have a race condition where we actually have a level triggered > - * interrupt, but the infrastructure can't expose that yet, so the guest > - * takes but ignores it, goes to sleep and never gets notified that there's > - * still an interrupt pending. > - * > - * As a quick workaround, let's just wake up again 20 ms after we injected > - * an interrupt. That way we can assure that we're always reinjecting > - * interrupts in case the guest swallowed them. > - */ > -static QEMUTimer *idle_timer; > - > static void kvm_kick_cpu(void *opaque) > { > PowerPCCPU *cpu = opaque; > @@ -491,7 +479,7 @@ int kvm_arch_init_vcpu(CPUState *cs) > return ret; > } > > - idle_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, kvm_kick_cpu, cpu); > + timer_init_ns(&cpu->idle_timer, QEMU_CLOCK_VIRTUAL, kvm_kick_cpu, cpu); > > switch (cenv->mmu_model) { > case POWERPC_MMU_BOOKE206: > @@ -523,6 +511,11 @@ int kvm_arch_init_vcpu(CPUState *cs) > > int kvm_arch_destroy_vcpu(CPUState *cs) > { > + PowerPCCPU *cpu = POWERPC_CPU(cs); > + > + timer_del(&cpu->idle_timer); > + timer_deinit(&cpu->idle_timer); > + > return 0; > } > > @@ -1379,8 +1372,17 @@ void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run) > printf("cpu %d fail inject %x\n", cs->cpu_index, irq); > } > > - /* Always wake up soon in case the interrupt was level based */ > - timer_mod(idle_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + > + /* > + * XXX We have a race condition where we actually have a level > + * triggered interrupt, but the infrastructure can't expose that > + * yet, so the guest takes but ignores it, goes to sleep and > + * never gets notified that there's still an interrupt pending. > + * > + * As a quick workaround, let's just wake up again 20 ms after > + * we injected an interrupt. That way we can assure that we're > + * always reinjecting interrupts in case the guest swallowed them. > + */ > + timer_mod(&cpu->idle_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + > (NANOSECONDS_PER_SECOND / 50)); > } > >
On Mon, 22 Jul 2019 18:44:24 +1000 David Gibson <david@gibson.dropbear.id.au> wrote: > On Fri, Jul 19, 2019 at 03:51:26AM -0500, Shivaprasad G Bhat wrote: > > The current code is broken for more than vcpu as > > each thread would overwrite idle_timer and there were > > memory leaks. > > > > Make it part of PowerPCCPU so that every thread has a > > separate one. Avoid using the timer_new_ns which is > > not the preferred way to create timers. > > > > Signed-off-by: Shivaprasad G Bhat <sbhat@linux.ibm.com> > > Reviewed-by: Greg Kurz <groug@kaod.org> > > So, this looks correct, but.. > > ..the whole idle_timer logic is used only in the case that we're > running with a KVM that doesn't support KVM_CAP_PPC_IRQ_LEVEL. That > support appears to have been in since around v2.6.36 some 9 years > ago. We could probably just drop support for such old kernels in qemu > instead. > Ha ! Yes, this is likely the best thing to do :) > > --- > > v3: https://lists.gnu.org/archive/html/qemu-devel/2019-07/msg04375.html > > Changes from v3: > > - Calling timer_del() before timer_deinit() > > > > target/ppc/cpu.h | 1 + > > target/ppc/kvm.c | 32 +++++++++++++++++--------------- > > 2 files changed, 18 insertions(+), 15 deletions(-) > > > > diff --git a/target/ppc/cpu.h b/target/ppc/cpu.h > > index c9beba2a5c..521086d91a 100644 > > --- a/target/ppc/cpu.h > > +++ b/target/ppc/cpu.h > > @@ -1190,6 +1190,7 @@ struct PowerPCCPU { > > void *machine_data; > > int32_t node_id; /* NUMA node this CPU belongs to */ > > PPCHash64Options *hash64_opts; > > + QEMUTimer idle_timer; > > > > /* Fields related to migration compatibility hacks */ > > bool pre_2_8_migration; > > diff --git a/target/ppc/kvm.c b/target/ppc/kvm.c > > index 8a06d3171e..52d3292f45 100644 > > --- a/target/ppc/kvm.c > > +++ b/target/ppc/kvm.c > > @@ -87,18 +87,6 @@ static int cap_large_decr; > > > > static uint32_t debug_inst_opcode; > > > > -/* > > - * XXX We have a race condition where we actually have a level triggered > > - * interrupt, but the infrastructure can't expose that yet, so the guest > > - * takes but ignores it, goes to sleep and never gets notified that there's > > - * still an interrupt pending. > > - * > > - * As a quick workaround, let's just wake up again 20 ms after we injected > > - * an interrupt. That way we can assure that we're always reinjecting > > - * interrupts in case the guest swallowed them. > > - */ > > -static QEMUTimer *idle_timer; > > - > > static void kvm_kick_cpu(void *opaque) > > { > > PowerPCCPU *cpu = opaque; > > @@ -491,7 +479,7 @@ int kvm_arch_init_vcpu(CPUState *cs) > > return ret; > > } > > > > - idle_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, kvm_kick_cpu, cpu); > > + timer_init_ns(&cpu->idle_timer, QEMU_CLOCK_VIRTUAL, kvm_kick_cpu, cpu); > > > > switch (cenv->mmu_model) { > > case POWERPC_MMU_BOOKE206: > > @@ -523,6 +511,11 @@ int kvm_arch_init_vcpu(CPUState *cs) > > > > int kvm_arch_destroy_vcpu(CPUState *cs) > > { > > + PowerPCCPU *cpu = POWERPC_CPU(cs); > > + > > + timer_del(&cpu->idle_timer); > > + timer_deinit(&cpu->idle_timer); > > + > > return 0; > > } > > > > @@ -1379,8 +1372,17 @@ void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run) > > printf("cpu %d fail inject %x\n", cs->cpu_index, irq); > > } > > > > - /* Always wake up soon in case the interrupt was level based */ > > - timer_mod(idle_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + > > + /* > > + * XXX We have a race condition where we actually have a level > > + * triggered interrupt, but the infrastructure can't expose that > > + * yet, so the guest takes but ignores it, goes to sleep and > > + * never gets notified that there's still an interrupt pending. > > + * > > + * As a quick workaround, let's just wake up again 20 ms after > > + * we injected an interrupt. That way we can assure that we're > > + * always reinjecting interrupts in case the guest swallowed them. > > + */ > > + timer_mod(&cpu->idle_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + > > (NANOSECONDS_PER_SECOND / 50)); > > } > > > > >
diff --git a/target/ppc/cpu.h b/target/ppc/cpu.h index c9beba2a5c..521086d91a 100644 --- a/target/ppc/cpu.h +++ b/target/ppc/cpu.h @@ -1190,6 +1190,7 @@ struct PowerPCCPU { void *machine_data; int32_t node_id; /* NUMA node this CPU belongs to */ PPCHash64Options *hash64_opts; + QEMUTimer idle_timer; /* Fields related to migration compatibility hacks */ bool pre_2_8_migration; diff --git a/target/ppc/kvm.c b/target/ppc/kvm.c index 8a06d3171e..52d3292f45 100644 --- a/target/ppc/kvm.c +++ b/target/ppc/kvm.c @@ -87,18 +87,6 @@ static int cap_large_decr; static uint32_t debug_inst_opcode; -/* - * XXX We have a race condition where we actually have a level triggered - * interrupt, but the infrastructure can't expose that yet, so the guest - * takes but ignores it, goes to sleep and never gets notified that there's - * still an interrupt pending. - * - * As a quick workaround, let's just wake up again 20 ms after we injected - * an interrupt. That way we can assure that we're always reinjecting - * interrupts in case the guest swallowed them. - */ -static QEMUTimer *idle_timer; - static void kvm_kick_cpu(void *opaque) { PowerPCCPU *cpu = opaque; @@ -491,7 +479,7 @@ int kvm_arch_init_vcpu(CPUState *cs) return ret; } - idle_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, kvm_kick_cpu, cpu); + timer_init_ns(&cpu->idle_timer, QEMU_CLOCK_VIRTUAL, kvm_kick_cpu, cpu); switch (cenv->mmu_model) { case POWERPC_MMU_BOOKE206: @@ -523,6 +511,11 @@ int kvm_arch_init_vcpu(CPUState *cs) int kvm_arch_destroy_vcpu(CPUState *cs) { + PowerPCCPU *cpu = POWERPC_CPU(cs); + + timer_del(&cpu->idle_timer); + timer_deinit(&cpu->idle_timer); + return 0; } @@ -1379,8 +1372,17 @@ void kvm_arch_pre_run(CPUState *cs, struct kvm_run *run) printf("cpu %d fail inject %x\n", cs->cpu_index, irq); } - /* Always wake up soon in case the interrupt was level based */ - timer_mod(idle_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + + /* + * XXX We have a race condition where we actually have a level + * triggered interrupt, but the infrastructure can't expose that + * yet, so the guest takes but ignores it, goes to sleep and + * never gets notified that there's still an interrupt pending. + * + * As a quick workaround, let's just wake up again 20 ms after + * we injected an interrupt. That way we can assure that we're + * always reinjecting interrupts in case the guest swallowed them. + */ + timer_mod(&cpu->idle_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + (NANOSECONDS_PER_SECOND / 50)); }