xfs
[Top] [All Lists]

Re: [PATCH 6/7] libxfs: keep unflushable buffers off the cache MRUs

To: Dave Chinner <david@xxxxxxxxxxxxx>
Subject: Re: [PATCH 6/7] libxfs: keep unflushable buffers off the cache MRUs
From: Christoph Hellwig <hch@xxxxxxxxxxxxx>
Date: Mon, 8 Feb 2016 02:06:36 -0800
Cc: xfs@xxxxxxxxxxx
Delivered-to: xfs@xxxxxxxxxxx
In-reply-to: <1454627108-19036-7-git-send-email-david@xxxxxxxxxxxxx>
References: <1454627108-19036-1-git-send-email-david@xxxxxxxxxxxxx> <1454627108-19036-7-git-send-email-david@xxxxxxxxxxxxx>
User-agent: Mutt/1.5.24 (2015-08-30)
> --- a/include/cache.h
> +++ b/include/cache.h
> @@ -51,6 +51,7 @@ enum {
>  #define CACHE_BASE_PRIORITY  0
>  #define CACHE_PREFETCH_PRIORITY      8
>  #define CACHE_MAX_PRIORITY   15
> +#define CACHE_DIRTY_PRIORITY (CACHE_MAX_PRIORITY + 1)

Sizing arrays based on, and iterating up to CACHE_DIRTY_PRIORITY seems
rather odd.  Maybe add a new

#define CACHE_NR_PRIORITIES             CACHE_DIRTY_PRIORITY

and a comment explaining the magic to make it more obvious?

> +cache_move_to_dirty_mru(
> +     struct cache            *cache,
> +     struct cache_node       *node)
> +{
> +     struct cache_mru        *mru;
> +
> +     mru = &cache->c_mrus[CACHE_DIRTY_PRIORITY];
> +
> +     pthread_mutex_lock(&mru->cm_mutex);
> +     node->cn_priority = CACHE_DIRTY_PRIORITY;
> +     list_move(&node->cn_mru, &mru->cm_list);
> +     mru->cm_count++;
> +     pthread_mutex_unlock(&mru->cm_mutex);
> +}

Maybe it would better to just do a list_add here and leave the
list_del to the caller to avoid needing to nest two different
cm_mutex instances.

<Prev in Thread] Current Thread [Next in Thread>