xfs
[Top] [All Lists]

Re: [PATCHv2 10/10] xfs: track the log sector size rather than log2(that

To: Dave Chinner <david@xxxxxxxxxxxxx>
Subject: Re: [PATCHv2 10/10] xfs: track the log sector size rather than log2(that)
From: Alex Elder <aelder@xxxxxxx>
Date: Wed, 14 Apr 2010 16:32:12 -0500
Cc: xfs@xxxxxxxxxxx
In-reply-to: <20100412072919.GL2493@dastard>
References: <1271048070.3174.125.camel@doink> <20100412072919.GL2493@dastard>
Reply-to: aelder@xxxxxxx
On Mon, 2010-04-12 at 17:29 +1000, Dave Chinner wrote:
> On Sun, Apr 11, 2010 at 11:54:30PM -0500, Alex Elder wrote:
> > Change struct log so it keeps track of the size (in basic blocks)
> > of a log sector in l_sectsize rather than the log-base-2 of that
> > value (previously, l_sectbb_log).
> > 
> > Signed-off-by: Alex Elder <aelder@xxxxxxx>
> 
> OK, this is what I commented on earlier....
> 
> > @@ -1050,29 +1051,31 @@ xlog_alloc_log(xfs_mount_t  *mp,
> >  
> >     error = EFSCORRUPTED;
> >     if (xfs_sb_version_hassector(&mp->m_sb)) {
> > -           log->l_sectbb_log = mp->m_sb.sb_logsectlog - BBSHIFT;
> > -           if (log->l_sectbb_log < 0 ||
> > -               log->l_sectbb_log > mp->m_sectbb_log) {
> > -                   xlog_warn("XFS: Log sector size (0x%x) out of range.",
> > -                                           log->l_sectbb_log);
> > +           sector_bits = mp->m_sb.sb_logsectlog;
> > +           if (sector_bits < BBSHIFT) {
> > +                   xlog_warn("XFS: Log sector byte size too small "
> > +                           "(0x%x < 0x%x)", sector_bits, BBSHIFT);
> 
> "sector_bits"?  That doesn't seem to fit the units the log sector
> size being read from and checked against. i.e. it's not bits, nor is
> it bytes (as the error message says). It's really log2(sector size),
> so perhaps calling it sector_size_log would be more accurate.

My intent here was to avoid the use of "log" to mean "logarithm"
in the log (meaning "log") code...  The "bits" here is the number
of bits required to represent a log sector.  Sort of like CHAR_BIT
is 8.  I also intentionally avoided encoded units in the name,
since what it represents changes from representing a number of
bytes to representing a number of blocks.

I'll try to reformulate this, see if I can come up with
something more clear.

> >                     goto out_free_log;
> >             }
> >  
> > -           /* for larger sector sizes, must have v2 or external log */
> > -           if (log->l_sectbb_log != 0 &&
> > -               (log->l_logBBstart != 0 &&
> > -                !xfs_sb_version_haslogv2(&mp->m_sb))) {
> > -                   xlog_warn("XFS: log sector size (0x%x) invalid "
> > -                             "for configuration.", log->l_sectbb_log);
> > +           sector_bits -= BBSHIFT;
> > +           if (sector_bits > mp->m_sectbb_log) {
> > +                   xlog_warn("XFS: Log sector basic block size too large "
> 
> It's just the log sector size is too large - no need for "basic
> block" in that error message.

I differentiated because the earlier value was really
    log2(bytes in a sector)
while this one is
    log2(basic blocks in a sector)

If someone were really looking at these, alongside the code,
they could figure out the difference, so I don't really
care much about this.  I'll take another shot at it and
will repost.

> 
> > +                           "(0x%x > 0x%x)", sector_bits, mp->m_sectbb_log);
> >                     goto out_free_log;
> >             }
> 
> > -           if (mp->m_sb.sb_logsectlog < BBSHIFT) {
> > -                   xlog_warn("XFS: Log sector log (0x%x) too small.",
> > -                                           mp->m_sb.sb_logsectlog);
> > +
> > +           /* for larger sector sizes, must have v2 or external log */
> > +           if (sector_bits && (log->l_logBBstart > 0 &&
> > +                                   !xfs_sb_version_haslogv2(&mp->m_sb))) {
> 
> no need for the extra ()s there.

You're right; they were there in the original code.
I don't like them and will remove them.

> > +
> > +                   xlog_warn("XFS: log sector size (0x%x) invalid "
> > +                             "for configuration.", sector_bits);
> >                     goto out_free_log;
> >             }
> >     }
> > -   log->l_sectbb_mask = (1 << log->l_sectbb_log) - 1;
> > +   log->l_sectsize = 1 << sector_bits;
> > +   log->l_sectbb_mask = log->l_sectsize - 1;
> 
> There's only only place left that uses l_sectbb_mask left (in
> xlog_align()), so you could probably kill it while you are at it.

Agreed.  I will add another patch to do that.

> Everything else looks OK.

Thanks for the reviews.

                                        -Alex

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