[Top] [All Lists]

Re: [PATCH 3/3] xfs: do not immediately reuse busy extent ranges

To: Alex Elder <aelder@xxxxxxx>
Subject: Re: [PATCH 3/3] xfs: do not immediately reuse busy extent ranges
From: Christoph Hellwig <hch@xxxxxxxxxxxxx>
Date: Wed, 9 Mar 2011 06:11:07 -0500
Cc: Christoph Hellwig <hch@xxxxxxxxxxxxx>, xfs@xxxxxxxxxxx
In-reply-to: <1299538896.2578.897.camel@doink>
References: <20110304125953.650347660@xxxxxxxxxxxxxxxxxxxxxx> <20110304130119.656476789@xxxxxxxxxxxxxxxxxxxxxx> <1299538896.2578.897.camel@doink>
User-agent: Mutt/1.5.21 (2010-09-15)
On Mon, Mar 07, 2011 at 05:01:36PM -0600, Alex Elder wrote:
> This time I just scanned most of the change, since it appears
> it's almost the same except for the (renamed) xfs_alloc_busy_trim()
> function.


> It looks correct now, but I have a few things for you
> to consider.  I'll wait for your response in case you want to
> change anything.  After that I'll pull in the three patches
> (probably tomorrow).

For now please just pull the first two.  There's a fair chance number
three will change based on how the discard work goes.

> I agree that the notation (from Dave) that you use here
> is very helpful in visualizing what's going on.  But
> the underlying code is pretty simple, and it gets somewhat
> lost in the comments I think.  I therefore would kind of
> prefer to have the explanation moved up above the function.
> It clearly labels the cases being treated, and references
> to those can be put in the code, below.
> (This is a style thing, so if you feel strongly that it's
> better as you have it, so be it.)

I tried that before, but matching the cases to numbers in comments
wasn't very readable so I switch to this notation.

> All the nice diagrams refer to the variable "fbno"
> and "fend" using "bno" and "end.  I think you should
> either drop the "f" in the variables or add it to
> the comments.

Indeed.  I did a last minute cleanup to consolidate the duplicate
variables and didn't update the comments.

> (Something like that anyway, I just wanted to provide
> an example rather than just saying "it's wrong.")


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