I hate all of this, but if this will finally catch the actual problem, we can then revert all this, so sure.
Also, I think this will conflict with the patches from Nadav that I have queued: https://lkml.kernel.org/r/20210220231712.2475218-1-na...@vmware.com which I'll be pushing to tip/x86/mm once -rc1 happens. On Fri, Feb 26, 2021 at 12:25:21PM +0100, Juergen Gross wrote: > +static void __smp_call_single_queue_debug(int cpu, struct llist_node *node) > +{ > + unsigned int this_cpu = smp_processor_id(); > + struct cfd_seq_local *seq = this_cpu_ptr(&cfd_seq_local); > + struct call_function_data *cfd = this_cpu_ptr(&cfd_data); > + struct cfd_percpu *pcpu = per_cpu_ptr(cfd->pcpu, cpu); > + > + cfd_seq_store(pcpu->seq_queue, this_cpu, cpu, CFD_SEQ_QUEUE); > + if (llist_add(node, &per_cpu(call_single_queue, cpu))) { > + cfd_seq_store(pcpu->seq_ipi, this_cpu, cpu, CFD_SEQ_IPI); > + cfd_seq_store(seq->ping, this_cpu, cpu, CFD_SEQ_PING); > + send_call_function_single_ipi(cpu); > + cfd_seq_store(seq->pinged, this_cpu, cpu, CFD_SEQ_PINGED); > + } else { > + cfd_seq_store(pcpu->seq_noipi, this_cpu, cpu, CFD_SEQ_NOIPI); > + } > +} > #else > +#define cfd_seq_store(var, src, dst, type) > + > static void csd_lock_record(call_single_data_t *csd) > { > } > @@ -290,6 +396,19 @@ static DEFINE_PER_CPU_SHARED_ALIGNED(call_single_data_t, > csd_data); > > void __smp_call_single_queue(int cpu, struct llist_node *node) > { > +#ifdef CONFIG_CSD_LOCK_WAIT_DEBUG > + if (static_branch_unlikely(&csdlock_debug_extended)) { > + unsigned int type; > + > + type = CSD_TYPE(container_of(node, call_single_data_t, > + node.llist)); > + if (type == CSD_TYPE_SYNC || type == CSD_TYPE_ASYNC) { > + __smp_call_single_queue_debug(cpu, node); > + return; > + } > + } > +#endif This is a bit weird, might as well put it in generic_exec_single() because there you still know the type matches. > @@ -712,12 +840,21 @@ static void smp_call_function_many_cond(const struct > cpumask *mask, > csd->node.src = smp_processor_id(); > csd->node.dst = cpu; > #endif > - if (llist_add(&csd->node.llist, &per_cpu(call_single_queue, > cpu))) > + cfd_seq_store(pcpu->seq_queue, this_cpu, cpu, CFD_SEQ_QUEUE); > + if (llist_add(&csd->node.llist, &per_cpu(call_single_queue, > cpu))) { > __cpumask_set_cpu(cpu, cfd->cpumask_ipi); > + cfd_seq_store(pcpu->seq_ipi, this_cpu, cpu, > CFD_SEQ_IPI); > + } else { > + cfd_seq_store(pcpu->seq_noipi, this_cpu, cpu, > CFD_SEQ_NOIPI); > + } > } > > /* Send a message to all CPUs in the map */ > + cfd_seq_store(this_cpu_ptr(&cfd_seq_local)->ping, this_cpu, > + CFD_SEQ_NOCPU, CFD_SEQ_PING); > arch_send_call_function_ipi_mask(cfd->cpumask_ipi); > + cfd_seq_store(this_cpu_ptr(&cfd_seq_local)->pinged, this_cpu, > + CFD_SEQ_NOCPU, CFD_SEQ_PINGED); Too bad we can't share with the single case, a well.
