Re: [RFC PATCH v2 0/1] FPGA subsystem core

[Greg Kroah-Hartman <gregkh@linuxfoundation.org>]

On Tue, Oct 08, 2013 at 06:47:41PM -0500, delicious quinoa wrote:
> On Tue, Oct 8, 2013 at 4:44 PM, Greg Kroah-Hartman
> <gregkh@linuxfoundation.org> wrote:
> > On Tue, Oct 08, 2013 at 12:00:14PM -0500, Alan Tull wrote:
> >> On Fri, 2013-10-04 at 16:33 -0700, Greg Kroah-Hartman wrote:
> >> > On Fri, Oct 04, 2013 at 11:12:13AM -0700, H. Peter Anvin wrote:
> >> > > On 10/04/2013 10:44 AM, Michal Simek wrote:
> >> > > >
> >> > > > If you look at it in general I believe that there is wide range of
> >> > > > applications which just contain one bitstream per fpga and the
> >> > > > bitstream is replaced by newer version in upgrade. For them
> >> > > > firmware interface should be pretty useful. Just setup firmware
> >> > > > name with bitstream and it will be automatically loaded in startup
> >> > > > phase.
> >> > > >
> >> > > > Then there is another set of applications especially in connection
> >> > > > to partial reconfiguration where this can be done statically by
> >> > > > pregenerated partial bitstreams or automatically generated on
> >> > > > target cpu. For doing everything on the target firmware interface
> >> > > > is not the best because everything can be handled by user
> >> > > > application and it is easier just to push this bitstream to do
> >> > > > device and not to save it to the fs.
> >> > > >
> >> > > > I think the question here is if this subsystem could have several
> >> > > > interfaces. For example Alan is asking for adding char support.
> >> > > > Does it even make sense to have more interfaces with the same
> >> > > > backend driver? When this is answered then we can talk which one
> >> > > > make sense to have. In v2 is sysfs and firmware one. Adding char
> >> > > > is also easy to do.
> >> > > >
> >> > >
> >> > > Greg, what do you think?
> >> > >
> >> > > I agree that the firmware interface makes sense when the use of the
> >> > > FPGA is an implementation detail in a fixed hardware configuration,
> >> > > but that is a fairly restricted use case all things considered.
> >> >
> >> > Ideally I thought this would be just like "firmware", you dump the file
> >> > to the FPGA, it validates it and away you go with a new image running in
> >> > the chip.
> >> >
> >> > But, it sounds like this is much more complicated, so much so that
> >> > configfs might be the correct interface for it, as you can do lots of
> >> > things there, and it is very flexible (some say too flexible...)
> >> >
> >> > A char device, with a zillion different custom ioctls is also a way to
> >> > do it, but one that I really want to avoid as that gets messy really
> >> > quickly.
> >>
> >> Hi Greg,
> >>
> >> We are discussing a char device that has very few interfaces:
> >>  - a way of writing the image to fpga
> >>  - a way of getting fpga manager status
> >>  - a way of setting fpga manager state
> >>
> >> This all looks like standard char driver interface to me.  Writing the
> >> image could be writing to the devnode (cat image.bin > /dev/fpga0). The
> >> status stuff would be sysfs attributes.  All normal stuff any char
> >> driver in the kernel would do.  Why not just go with that?
> >
> > Because we really hate to add new ioctls to the kernel if at all
> > possible.
> 
> I don't see any need for adding any ioctls.
> 
> > Using sysfs (and it's one-value-per-file rule), makes
> > userspace tools easier, and managing the different devices in the system
> > easier (you know _exactly_ which device you are talking to, you don't
> > have to guess based on minor number).
> 
> That's cool.  The interface we could use is writing the raw fpga data
> to /sys/class/fpga_manager/fpga0/fpga_config_data
> 
> Reading or setting the fpga state could be from
> /sys/class/fpga_manager/fpga0/fpga_config_state

Ok, that's fine, I don't object to that, but you are giving up the
notification and loading ability of the kernel for the image files by
doing this, which will require you to use/write/maintain userspace
tools.  If you use the firmware interface, no userspace tool is needed
at all, which I can see some people really wanting, right?

> Or do I misunderstand?  Do you include sysfs attributes when you
> are talking about ioctls?

You can't do ioctls on sysfs files, so no.

greg k-h