Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 31 Mar 2015 16:50:51 -0600
From:      "Kenneth D. Merry" <ken@FreeBSD.ORG>
To:        Konstantin Belousov <kostikbel@gmail.com>
Cc:        current@freebsd.org, scsi@freebsd.org
Subject:   Re: async pass(4) patches available
Message-ID:  <20150331225051.GA64520@mithlond.kdm.org>
In-Reply-To: <20150331004912.GM2379@kib.kiev.ua>
References:  <20150330222358.GA46342@mithlond.kdm.org> <20150331004912.GM2379@kib.kiev.ua>

next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, Mar 31, 2015 at 03:49:12 +0300, Konstantin Belousov wrote:
> On Mon, Mar 30, 2015 at 04:23:58PM -0600, Kenneth D. Merry wrote:
> > Kernel memory for data transferred via the queued interface is 
> > allocated from the zone allocator in MAXPHYS sized chunks, and user
> > data is copied in and out.  This is likely faster than the
> > vmapbuf()/vunmapbuf() method used by the CAMIOCOMMAND ioctl in
> > configurations with many processors (there are more TLB shootdowns
> > caused by the mapping/unmapping operation) but may not be as fast
> > as running with unmapped I/O.
> cam_periph_mapmem() uses vmapbuf() with an indicator to always map the
> user pages mostly because I do not know CAM code and wanted to make
> the least intrusive changes there.  It is not inherently impossible
> to pass unmapped pages down from cam_periph_mapmem(), but might
> require some more plumbing for driver to indicate that it is acceptable.

I think that would probably not be too difficult to change.  That API isn't
one that is exposed, so changing it shouldn't be a problem.  The only
reason not to do unmapped I/O there is just if the underlying controller
doesn't support it.  The lower parts of the stack shouldn't be trying to
sniff the data that is read or written to the device, although that has
happened in the past.  We'd have to audit a couple of the drivers to
make sure they aren't trying to access the data.

> > The new memory handling model for user requests also allows
> > applications to send CCBs with request sizes that are larger than
> > MAXPHYS.  The pass(4) driver now limits queued requests to the I/O
> > size listed by the SIM driver in the maxio field in the Path
> > Inquiry (XPT_PATH_INQ) CCB.
> >         
> > There are some things things would be good to add:
> >         
> > 1. Come up with a way to do unmapped I/O on multiple buffers.
> >    Currently the unmapped I/O interface operates on a struct bio,
> >    which includes only one address and length.  It would be nice
> >    to be able to send an unmapped scatter/gather list down to
> >    busdma.  This would allow eliminating the copy we currently do
> >    for data.
> Only because nothing more was needed.  The struct bio does not use
> address/length pair when unmapped, it passes the list of physical
> pages, see bio_ma array pointer.  It is indeed taylored to be a pointer
> to struct buf' b_pages, but it does not have to be.
> 
> The busdma unmapped non-specific interface is bus_dmamap_load_ma(),
> which again takes array of pages to load.  If you want some additional
> helper, suitable for your goals, please provide the desired interface
> definition.

What I'd like to be able to do is pass down a CCB with a user virtual
S/G list (CAM_DATA_SG, but with user virtual pointers) and have busdma deal
with it.

The trouble would likely be figuring out a flag to use to indicate that the
S/G list in question contains user virtual pointers.  (Backwards/binary
compatibility is always an issue with CCB flags, since they have all been
used.)

But that is essentially what is needed.  

Ken
-- 
Kenneth Merry
ken@FreeBSD.ORG



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20150331225051.GA64520>