linux-mtd.lists.infradead.org archive mirror
 help / color / mirror / Atom feed
From: Adrian Hunter <adrian.hunter@nokia.com>
To: Artem Bityutskiy <dedekind1@gmail.com>
Cc: "linux-mtd@lists.infradead.org" <linux-mtd@lists.infradead.org>,
	Matthieu CASTET <matthieu.castet@parrot.com>
Subject: Re: [PATCH 4/7] UBIFS: fix GC sroting
Date: Sat, 07 Aug 2010 15:11:49 +0300	[thread overview]
Message-ID: <4C5D4D85.4000800@nokia.com> (raw)
In-Reply-To: <1281169577-18664-5-git-send-email-dedekind1@gmail.com>

Artem Bityutskiy wrote:
> From: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
> 
> Unfortunately, the comparison functions for both data and non-data nodes
> ('data_nodes_cmp()' and 'nondata_nodes_cmp()') was screwed. It did not
> work for the case when 'inuma == inumb' and 'blka > blkb' or 'hasha > hashb'.

Can you explain a bit more.  The old logic looks OK to me whereas the
new logic would seem to make block 0x80000001 < 0x00000001 ?

> 
> Basically, this means that it did not actually sort the list. This is not
> fatal, but this means that GC did not optimize the on-flash layout as we
> thought it would.
> 
> This issue was revealed by commit '835cc0c8477fdbc59e0217891d6f11061b1ac4e2' of
> Don Mullis: assertions in the comparison functions started triggering.
> 
> This patch fixes the issue. Additionally, it makes the comparison functions
> return immediately the elements to compare are actually the same element.
> This is just a tiny optimization.
> 
> Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
> ---
>  fs/ubifs/gc.c |   34 ++++++++++++++--------------------
>  1 files changed, 14 insertions(+), 20 deletions(-)
> 
> diff --git a/fs/ubifs/gc.c b/fs/ubifs/gc.c
> index d060c62..fbb9272 100644
> --- a/fs/ubifs/gc.c
> +++ b/fs/ubifs/gc.c
> @@ -125,6 +125,9 @@ int data_nodes_cmp(void *priv, struct list_head *a, struct list_head *b)
>  	struct ubifs_scan_node *sa, *sb;
>  
>  	cond_resched();
> +	if (a == b)
> +		return 0;
> +
>  	sa = list_entry(a, struct ubifs_scan_node, list);
>  	sb = list_entry(b, struct ubifs_scan_node, list);
>  	ubifs_assert(key_type(c, &sa->key) == UBIFS_DATA_KEY);
> @@ -133,16 +136,10 @@ int data_nodes_cmp(void *priv, struct list_head *a, struct list_head *b)
>  	inuma = key_inum(c, &sa->key);
>  	inumb = key_inum(c, &sb->key);
>  
> -	if (inuma == inumb) {
> -		unsigned int blka = key_block(c, &sa->key);
> -		unsigned int blkb = key_block(c, &sb->key);
> -
> -		if (blka <= blkb)
> -			return -1;
> -	} else if (inuma <= inumb)
> -		return -1;
> -
> -	return 1;
> +	if (inuma == inumb)
> +		return key_block(c, &sa->key) - key_block(c, &sb->key);
> +	else
> +		return inuma - inumb;
>  }
>  
>  /*
> @@ -163,6 +160,9 @@ int nondata_nodes_cmp(void *priv, struct list_head *a, struct list_head *b)
>  	struct ubifs_scan_node *sa, *sb;
>  
>  	cond_resched();
> +	if (a == b)
> +		return 0;
> +
>  	sa = list_entry(a, struct ubifs_scan_node, list);
>  	sb = list_entry(b, struct ubifs_scan_node, list);
>  	typea = key_type(c, &sa->key);
> @@ -183,16 +183,10 @@ int nondata_nodes_cmp(void *priv, struct list_head *a, struct list_head *b)
>  	inuma = key_inum(c, &sa->key);
>  	inumb = key_inum(c, &sb->key);
>  
> -	if (inuma == inumb) {
> -		uint32_t hasha = key_hash(c, &sa->key);
> -		uint32_t hashb = key_hash(c, &sb->key);
> -
> -		if (hasha <= hashb)
> -			return -1;
> -	} else if (inuma <= inumb)
> -		return -1;
> -
> -	return 1;
> +	if (inuma == inumb)
> +		return key_hash(c, &sa->key) - key_hash(c, &sb->key);
> +	else
> +		return inuma - inumb;
>  }
>  
>  /**

  reply	other threads:[~2010-08-07 12:11 UTC|newest]

Thread overview: 15+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2010-08-07  8:26 [PATCH 0/7] UBIFS: recent patches Artem Bityutskiy
2010-08-07  8:26 ` [PATCH 1/7] UBIFS: switch to RO mode after synchronizing Artem Bityutskiy
2010-08-07  8:26 ` [PATCH 2/7] UBIFS: do not treat ENOSPC specially Artem Bityutskiy
2010-08-08 17:51   ` Vitaly Wool
2010-08-09  5:56     ` Artem Bityutskiy
2010-08-07  8:26 ` [PATCH 3/7] UBIFS: fix assertion warning Artem Bityutskiy
2010-08-07  8:26 ` [PATCH 4/7] UBIFS: fix GC sroting Artem Bityutskiy
2010-08-07 12:11   ` Adrian Hunter [this message]
2010-08-07 12:51     ` Artem Bityutskiy
2010-08-07  8:26 ` [PATCH 5/7] UBIFS: do not look up junk nodes Artem Bityutskiy
2010-08-07 12:06   ` Adrian Hunter
2010-08-07 12:45     ` Artem Bityutskiy
2010-08-07  8:26 ` [PATCH 6/7] UBIFS: do not use key type in list_sort compariosn fuction Artem Bityutskiy
2010-08-07  8:26 ` [PATCH 7/7] UBIFS: introduce list sorting debugging checks Artem Bityutskiy
2010-08-09 13:18 ` [PATCH 0/7] UBIFS: recent patches Matthieu CASTET

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=4C5D4D85.4000800@nokia.com \
    --to=adrian.hunter@nokia.com \
    --cc=dedekind1@gmail.com \
    --cc=linux-mtd@lists.infradead.org \
    --cc=matthieu.castet@parrot.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).