Re: [PATCH 1/2 v2] rseq/membarrier: add MEMBARRIER_CMD_PRIVATE_RESTART_RSEQ_ON_CPU

[peterz]

On Thu, Aug 06, 2020 at 10:05:43AM -0700, Peter Oskolkov wrote:
> +#ifdef CONFIG_RSEQ
> +static void membarrier_rseq_ipi(void *arg)
> +{
> +     if (current->mm != arg)  /* Not our process. */
> +             return;
> +     if (!current->rseq)  /* RSEQ not set up for the current task/thread. */
> +             return;
> +
> +     rseq_preempt(current);
> +}
> +#endif
> +
> +static int membarrier_private_restart_rseq_on_cpu(int cpu_id)
> +{
> +#ifdef CONFIG_RSEQ
> +     /* syscalls are not allowed inside rseq critical sections. */
> +     if (cpu_id == raw_smp_processor_id())
> +             return 0;
> +
> +     return smp_call_function_single(cpu_id, membarrier_rseq_ipi,
> +                                     current->mm, true);
> +#else
> +     return 0;
> +#endif
> +}

I'm thinking even this is a problem, we can end up sending IPIs to CPUs
outside out partition (they might be NOHZ_FULL) and that's a no-no too.

Something like so perhaps... that really limits it to CPUs that match
our mm.

diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h
index 6be66f52a2ad..bee5e98e6774 100644
--- a/include/linux/sched/mm.h
+++ b/include/linux/sched/mm.h
@@ -356,6 +356,7 @@ enum {
 
 enum {
        MEMBARRIER_FLAG_SYNC_CORE       = (1U << 0),
+       MEMBARRIER_FLAG_RSEQ            = (1U << 1),
 };
 
 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
diff --git a/kernel/sched/membarrier.c b/kernel/sched/membarrier.c
index 168479a7d61b..4d9b22c2f5e2 100644
--- a/kernel/sched/membarrier.c
+++ b/kernel/sched/membarrier.c
@@ -27,6 +27,11 @@
 
 static void ipi_mb(void *info)
 {
+       int *flags = info;
+
+       if (flags && (*flags & MEMBARRIER_FLAG_RSEQ))
+               rseq_preempt(current);
+
        smp_mb();       /* IPIs should be serializing but paranoid. */
 }
 
@@ -129,11 +134,11 @@ static int membarrier_global_expedited(void)
        return 0;
 }
 
-static int membarrier_private_expedited(int flags)
+static int membarrier_private_expedited(int flags, int cpu_id)
 {
-       int cpu;
-       cpumask_var_t tmpmask;
        struct mm_struct *mm = current->mm;
+       cpumask_var_t tmpmask;
+       int cpu;
 
        if (flags & MEMBARRIER_FLAG_SYNC_CORE) {
                if (!IS_ENABLED(CONFIG_ARCH_HAS_MEMBARRIER_SYNC_CORE))
@@ -174,6 +179,10 @@ static int membarrier_private_expedited(int flags)
                 */
                if (cpu == raw_smp_processor_id())
                        continue;
+
+               if (cpu_id >= 0 && cpu != cpu_id)
+                       continue;
+
                p = rcu_dereference(cpu_rq(cpu)->curr);
                if (p && p->mm == mm)
                        __cpumask_set_cpu(cpu, tmpmask);
@@ -181,7 +190,7 @@ static int membarrier_private_expedited(int flags)
        rcu_read_unlock();
 
        preempt_disable();
-       smp_call_function_many(tmpmask, ipi_mb, NULL, 1);
+       smp_call_function_many(tmpmask, ipi_mb, &flags, 1);
        preempt_enable();
 
        free_cpumask_var(tmpmask);
@@ -362,11 +371,13 @@ SYSCALL_DEFINE2(membarrier, int, cmd, int, flags)
        case MEMBARRIER_CMD_REGISTER_GLOBAL_EXPEDITED:
                return membarrier_register_global_expedited();
        case MEMBARRIER_CMD_PRIVATE_EXPEDITED:
-               return membarrier_private_expedited(0);
+               return membarrier_private_expedited(0, -1);
        case MEMBARRIER_CMD_REGISTER_PRIVATE_EXPEDITED:
                return membarrier_register_private_expedited(0);
        case MEMBARRIER_CMD_PRIVATE_EXPEDITED_SYNC_CORE:
-               return membarrier_private_expedited(MEMBARRIER_FLAG_SYNC_CORE);
+               return membarrier_private_expedited(MEMBARRIER_FLAG_SYNC_CORE, 
-1);
+       case MEMBERRIER_CMD_PRIVATE_EXPEDITED_RSEQ:
+               return membarrier_private_expedited(MEMBARRIER_FLAG_RSEQ, 
flags);
        case MEMBARRIER_CMD_REGISTER_PRIVATE_EXPEDITED_SYNC_CORE:
                return 
membarrier_register_private_expedited(MEMBARRIER_FLAG_SYNC_CORE);
        default: