[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20070508103524.GH4163@kernel.dk>
Date: Tue, 8 May 2007 12:35:24 +0200
From: Jens Axboe <jens.axboe@...cle.com>
To: Rene Herman <rene.herman@...il.com>
Cc: Andrew Morton <akpm@...l.org>,
Pekka Enberg <penberg@...helsinki.fi>,
Linux Kernel <linux-kernel@...r.kernel.org>
Subject: Re: New Mitsumi legacy CD-ROM driver
On Tue, May 08 2007, Rene Herman wrote:
> Good day.
>
> This is (the start of) a new driver for the legacy Mitsumi CD-ROM (LU-00xS
> and FX-001x) drives that used to be supported by drivers/cdrom/mcd(x).c, a
> driver that hasn't worked for a long time.
>
> It's being tested on my LU-005S and is being submitted to -mm in the hope
> of maybe finding a few more testers and also in the hope of provoking a
> review. I'm co-authoring this with Pekka Enberg but certainly for me it's
> the first ever contact with anything having anything to do with block.
>
> A few specific points/questions:
>
> * This driver's core transfer function is doing:
>
> rq_for_each_bio(bio, req)
> bio_for_each_segment(bvec, bio, segno) {
> buffer = bvec_kmap_irq(bvec, &flags);
> ioread8_rep(ioaddr, buffer, bvec->bv_len);
> bvec_kunmap_irq(buffer, &flags);
> }
>
> I've been slightly worried about struct bio having a bi_sector field. If
> we're not guaranteed that bio[n].bi_sector == bio[n-1].bi_sector + 1 during
> the rq_for_each_bio() loop we can't read from I/O into bio's directly but
> will need to use a temporary buffer and transfer into the bio's later. Is
> this guaranteed?
bio[n - 1].bi_sector + bio_sectors(bio[n - 1]) == bio[n].bi_sector
IOW, the bio's in the chain on the request are contig in the lba space.
> * This driver purports to support highmem. Given the number of machines out
> there that have both a Mitsumi legacy CD-ROM drive and enough physical
> memory installed to want highmem support this may come across as being
> overly... optimistic but is there much point to not supporting it? I
> believe we do always need the full rq_for_each_bio/bio_for_each_segment
> loop meaning that only the bvec_k{,un}map_irq could go, right?
Yeah, I think it's pointless to support highmem. And as long as you
don't call blk_queue_bounce_limit() to actively enable highmem, you are
not going to receive a bio/request with highmem pages. The block layer
wil bounce any highmem pages before it reaches you. So for a driver like
this, I would recommend just forgetting the mapping aspect.
I don't have time to review your driver right now, but I will applaud
your effort to write a maintenable mitsumi driver! Basically all the old
cdrom drivers are utter crap, so they are destined for removal.
--
Jens Axboe
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists