Re: [RFC] 'discard sectors' block request.

[David Woodhouse <dwmw2@infradead.org>]

On Mon, 2008-08-04 at 18:59 -0700, Andrew Morton wrote:
> On Tue, 05 Aug 2008 02:45:16 +0100 David Woodhouse <dwmw2@infradead.org> wrote:
> 
> > On Sat, 2008-07-26 at 13:02 -0700, Andrew Morton wrote:
> > > I seem to be hearing a lot of silence over support for SSD devices.  I
> > > have this vague worry that there will be a large rollout of SSD
> > > hardware and Linux will be found to have pants-around-ankles.
> > > 
> > > For example, support for the T13 Trim command.  There will be others,
> > > but I surely don't know what they are.
> > 
> > Here's a first attempt at that. It implements basic support for
> > 'discard' requests in the block layer, implements that in FTL (the flash
> > translation layer used on PCMCIA flash cards), and in FAT.
> 
> Looks nice and simple.

> The synchronous nature might get pretty painful, even after merging is
> working?

Yeah. That's just an example, based on blkdev_issue_flush(). When file
systems submit bios directly, that can be different.

> Perhaps we should have a new bio_doesnt_have_any_data_in_it() helper,
> rather than opn-coding it here?  Assuming that new non-data-containing
> bios will turn up in the future.

That would be something like (!bio->bi_io_vec) ?  I'm kind of lost in
block code; the whole of that part of the patch wants someone like Jens
to go over it carefully.

> > +static int ftl_discardsect(struct mtd_blktrans_dev *dev,
> > +			   unsigned long sector, unsigned nr_sects)
> > +{
> > +	partition_t *part = (void *)dev;
> 
> wuh?  We're casting a `struct mtd_blktrans_ops *' into a partition_t?

Yeah, I had to look carefully at that one when I pasted it from
elsewhere. 

typedef struct partition_t {
    struct mtd_blktrans_dev mbd;
    ....

-- 
dwmw2