2019-08-22 04:14:25

by Herbert Xu

[permalink] [raw]
Subject: Re: [PATCH 8/9] padata: unbind parallel jobs from specific CPUs

On Mon, Aug 12, 2019 at 08:52:23PM -0400, Daniel Jordan wrote:
>
> @@ -191,22 +184,25 @@ static struct padata_priv *padata_get_next(struct parallel_data *pd)
> padata = list_entry(reorder->list.next,
> struct padata_priv, list);
>
> - list_del_init(&padata->list);
> - atomic_dec(&pd->reorder_objects);
> + /*
> + * The check fails in the unlikely event that two or more
> + * parallel jobs have hashed to the same CPU and one of the
> + * later ones finishes first.
> + */
> + if (padata->seq_nr == pd->processed) {
> + list_del_init(&padata->list);
> + atomic_dec(&pd->reorder_objects);

Now that you've changed the test for whether there is work to be
done you also need to update the code at the end of padata_reorder
that checks whether there is work to do. Otherwise we can end up
in a busy loop that just wastes CPU cycles.

Cheers,
--
Email: Herbert Xu <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt


2019-08-23 08:34:01

by Daniel Jordan

[permalink] [raw]
Subject: Re: [PATCH 8/9] padata: unbind parallel jobs from specific CPUs

On 8/22/19 12:13 AM, Herbert Xu wrote:
> On Mon, Aug 12, 2019 at 08:52:23PM -0400, Daniel Jordan wrote:
>>
>> @@ -191,22 +184,25 @@ static struct padata_priv *padata_get_next(struct parallel_data *pd)
>> padata = list_entry(reorder->list.next,
>> struct padata_priv, list);
>>
>> - list_del_init(&padata->list);
>> - atomic_dec(&pd->reorder_objects);
>> + /*
>> + * The check fails in the unlikely event that two or more
>> + * parallel jobs have hashed to the same CPU and one of the
>> + * later ones finishes first.
>> + */
>> + if (padata->seq_nr == pd->processed) {
>> + list_del_init(&padata->list);
>> + atomic_dec(&pd->reorder_objects);
>
> Now that you've changed the test for whether there is work to be
> done you also need to update the code at the end of padata_reorder
> that checks whether there is work to do. Otherwise we can end up
> in a busy loop that just wastes CPU cycles.

So we can, thanks for catching that.