Re: [PATCH net-next v2 1/7] mld: convert from timer to delayed work

On 21. 2. 14. 오전 4:07, Cong Wang wrote:
 > On Sat, Feb 13, 2021 at 9:51 AM Taehee Yoo [off-list ref] wrote:
 >> -static void mld_dad_start_timer(struct inet6_dev *idev, unsigned 
long delay)
 >> +static void mld_dad_start_work(struct inet6_dev *idev, unsigned 
long delay)
 >>   {
 >>          unsigned long tv = prandom_u32() % delay;
 >>
 >> -       if (!mod_timer(&idev->mc_dad_timer, jiffies+tv+2))
 >> +       if (!mod_delayed_work(mld_wq, &idev->mc_dad_work, 
msecs_to_jiffies(tv + 2)))
 >
 > IIUC, before this patch 'delay' is in jiffies, after this patch it is 
in msecs?
 >

Ah, I understand, It's my mistake.
I didn't change the behavior of 'delay' in this patchset.
So, 'delay' is still in jiffies, not msecs.
Therefore, msecs_to_jiffies() should not be used in this patchset.
I will send a v3 patch, which doesn't use msecs_to_jiffies().
Thanks!

By the way, I think the 'delay' is from the 
unsolicited_report_interval() and it just return value of 
idev->cnf.mldv{1 | 2}_unsolicited_report_interval.
I think this value is msecs, not jiffies.
So, It should be converted to use msecs_to_jiffies(), I think.
How do you think about it?

 > [...]
 >
 >> -static void mld_dad_timer_expire(struct timer_list *t)
 >> +static void mld_dad_work(struct work_struct *work)
 >>   {
 >> -       struct inet6_dev *idev = from_timer(idev, t, mc_dad_timer);
 >> +       struct inet6_dev *idev = container_of(to_delayed_work(work),
 >> +                                             struct inet6_dev,
 >> +                                             mc_dad_work);
 >>
 >> +       rtnl_lock();
 >
 > Any reason why we need RTNL after converting the timer to
 > delayed work?
 >

For the moment, RTNL is not needed.
But the Resources, which are used by delayed_work will be protected by 
RTNL instead of other locks.
So, It just pre-adds RTNL and the following patches will delete other locks.

 > Thanks.
 >