Re: [RFC PATCH] mm: trace filemap add and del

[Dave Chinner <david@fromorbit.com>]

On Tue, Nov 20, 2012 at 03:57:35PM -0800, Andrew Morton wrote:
> On Thu,  8 Nov 2012 20:54:10 +0100
> Robert Jarzmik <robert.jarzmik@free.fr> wrote:
.....
> > +		__field(dev_t, s_dev)
> 
> Perhaps use super_block.s_id here
> 
> > +	),
> > +
> > +	TP_fast_assign(
> > +		__entry->page = page;
> > +		__entry->i_no = page->mapping->host->i_ino;
> > +		__entry->pageofs = page->index;
> > +		if (page->mapping->host->i_sb)
> > +			__entry->s_dev = page->mapping->host->i_sb->s_dev;
> > +		else
> > +			__entry->s_dev = page->mapping->host->i_rdev;
> 
> and hence avoid all this stuff.

We actually have an informal convention for formating filesystem
trace events, and that is to use the device number....

> 
> > +	),
> > +
> > +	TP_printk("page=%p pfn=%lu blk=%d:%d inode+ofs=%lu+%lu",

... and to prefix messages like:

	TP_printk("dev %d:%d ino 0x%llx ....
		  MAJOR(__entry->dev), MINOR(__entry->dev),

i.e. the start of the event message has all the identifying
information where it is easy to grep for and get all the events for
a specific dev/inode combination without even having to think about
it.

XFS, ext3/4, jbd/jdb2 and gfs2 follow this convention, so we should
keep propagating that pattern in the name of consistency, rather
than having different trace formats for different parts of the
VFS/FS layers...

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com