On Sat, Jan 11, 2014 at 12:09:46PM +0900, Tetsuo Handa wrote:
> Andrew Morton wrote:
> > In the absence of step 3, steps 1 and 2 are rather pointless churn.
> >
> > So I think it would be better to merge (into mainline) steps 1 and 3
> > first and at the same time. Then start thinking about step 2.
>
> Unfortunately we can't.
> Step 2 depends on step 1 for avoiding compile time errors.
> Step 3 depends on step 2 for avoiding run time errors.
>
> Step 1: (targeted to 3.14-rc1)
> Add "%pT" format specifier and commcpy() wrapper function.
>
> Step 2: (started after step 1 is reflected to other git trees)
> Replace printk("%s", current->comm) with printk("%pT", NULL).
> Replace printk("%s", p->comm) with printk("%pT", p).
> Replace strcpy(buf, p->comm) with commcpy(buf, p).
>
> Step 3: (started after step 2 is reflected to other git trees)
> Add rcu_read_lock()/rcu_read_unlock() into commcpy().
> Modify set_task_comm() etc. to replace ->comm using RCU.
>
> If step 3 is merged into mainline before step 2 complete, those who are not
> using "%pT" or commcpy() might crash due to reading RCU protected ->comm
> without rcu_read_lock()/rcu_read_unlock().
>
>
> Let me confirm, Paul.
>
> I'm trying to change task_struct->comm to use RCU.
> At step 3, I'm planning to do
>
> static inline void *commcpy(void *buf, const struct task_struct *tsk)
> {
> rcu_read_lock();
> memcpy(buf, tsk->comm, TASK_COMM_LEN);
> rcu_read_unlock();
> return buf;
> }
>
> and let set_task_comm() wait for readers using synchronize_rcu() or
> call_rcu().
>
> Given that commcpy() can be called from any context, are synchronize_rcu()
> and call_rcu() sufficient for waiting for commcpy() users?
Yep, that should work.
Thanx, Paul
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
