Re: [PATCH 1/3] dmapool: improve scalability of dma_pool_alloc

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 07/26/2018 03:37 PM, Andy Shevchenko wrote:
> On Thu, Jul 26, 2018 at 9:54 PM, Tony Battersby <tonyb@xxxxxxxxxxxxxxx> wrote:
>> dma_pool_alloc() scales poorly when allocating a large number of pages
>> because it does a linear scan of all previously-allocated pages before
>> allocating a new one.  Improve its scalability by maintaining a separate
>> list of pages that have free blocks ready to (re)allocate.  In big O
>> notation, this improves the algorithm from O(n^2) to O(n).
>
>
>>         spin_lock_irqsave(&pool->lock, flags);
>> -       list_for_each_entry(page, &pool->page_list, page_list) {
>> -               if (page->offset < pool->allocation)
>> -                       goto ready;
>> +       if (!list_empty(&pool->avail_page_list)) {
>> +               page = list_first_entry(&pool->avail_page_list,
>> +                                       struct dma_page,
>> +                                       avail_page_link);
>> +               goto ready;
>>         }
> It looks like
>
> page = list_first_entry_or_null();
> if (page)
>  goto ready;
>
> Though I don't know which one produces better code in the result.
>
> >From reader prospective of view I would go with my variant.

Thanks, I didn't know about list_first_entry_or_null().

>
>> +       /* This test checks if the page is already in avail_page_list. */
>> +       if (list_empty(&page->avail_page_link))
>> +               list_add(&page->avail_page_link, &pool->avail_page_list);
> How can you be sure that the page you are testing for is the first one?
>
> It seems you are relying on the fact that in the list should be either
> 0 or 1 page. In that case what's the point to have a list?
>
That would be true if the test were "if (list_empty(&pool->avail_page_list))".  But it is testing the list pointers in the item rather than the list pointers in the pool.  It may be a bit confusing if you have never seen that usage before, which is why I added a comment.  Basically, if you use list_del_init() instead of list_del(), then you can use list_empty() on the item itself to test if the item is present in a list or not.  For example, the comments in list.h warn not to use list_empty() on the entry after just list_del():

/**
 * list_del - deletes entry from list.
 * @entry: the element to delete from the list.
 * Note: list_empty() on entry does not return true after this, the entry is
 * in an undefined state.
 */





[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Bugtraq]     [Linux OMAP]     [Linux MIPS]     [eCos]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux