Re: [RFC][-mm PATCH 4/8] Memory controller memory accounting (v3)

[Balbir Singh <balbir@linux.vnet.ibm.com>]

Paul Menage wrote:
> On 7/20/07, Balbir Singh <balbir@linux.vnet.ibm.com> wrote:
> 
>> +void __always_inline unlock_meta_page(struct page *page)
>> +{
>> +       bit_spin_unlock(PG_metapage, &page->flags);
>> +}
> 
> Maybe add a BUG_ON(!test_bit(PG_metapage, &page->flags)) at least for
> development?
> 

I'd rather make that a VM_BUG_ON, but that's a good suggestion

>> +       mem = rcu_dereference(mm->mem_container);
>> +       /*
>> +        * For every charge from the container, increment reference
>> +        * count
>> +        */
>> +       css_get(&mem->css);
>> +       rcu_read_unlock();
> 
> It's not clear to me that this is safe.
> 
> If
> 
>> +
>> +       /*
>> +        * If we created the meta_page, we should free it on exceeding
>> +        * the container limit.
>> +        */
>> +       if (res_counter_charge(&mem->res, 1)) {
>> +               css_put(&mem->css);
>> +               goto free_mp;
>> +       }
>> +
>> +       lock_meta_page(page);
>> +       /*
>> +        * Check if somebody else beat us to allocating the meta_page
>> +        */
>> +       if (page_get_meta_page(page)) {
> 
> I think you need to add something like
> 
>  kfree(mp);
>  mp = page_get_meta_page(page);
> 
> otherwise you're going to leak the new but unneeded metapage.
> 

Yes, good catch! I am surprised I did not check for that.

>> +               atomic_inc(&mp->ref_cnt);
>> +               res_counter_uncharge(&mem->res, 1);
>> +               goto done;
>> +       }
>> +
>> +       atomic_set(&mp->ref_cnt, 1);
>> +       mp->mem_container = mem;
>> +       mp->page = page;
>> +       page_assign_meta_page(page, mp);
> 
> Would it make sense to have the "mp->page = page" be part of
> page_assign_meta_page() for consistency?
> 

Yes, that could be done easily.

>> +err:
>> +       unlock_meta_page(page);
>> +       return -ENOMEM;
> 
> The only jump to err: is from a location where the metapage is already
> unlocked. Maybe scrap err: and just do a return -ENOMEM when the
> allocation fails?
> 

Sounds good, let me revisit the code.

>> +out_uncharge:
>> +       mem_container_uncharge(page_get_meta_page(page));
> 
> Wanting to call mem_container_uncharge() on a page and hence having to
> call page_get_meta_page() seems to be more common than wanting to call
> it on a meta page that you already have available. Maybe make
> mem_container_uncharge() be a wrapper that take a struct page and does
> something like mem_container_uncharge_mp(page_get_meta_page(page))
> where mem_container_uncharge_mp() is the raw meta-page version?
> 


Yes.. right! Will do, I'll write a wrapper.

> Paul


-- 
	Warm Regards,
	Balbir Singh
	Linux Technology Center
	IBM, ISTL