Re: [PATCH] [PATCH] Firmware security information in SYSFS

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

On Tue, Jul 21, 2020 at 01:27:27PM -0300, Daniel Gutson wrote:
> On Tue, Jul 21, 2020 at 7:52 AM Greg Kroah-Hartman <
> gregkh@linuxfoundation.org> wrote:
> 
> > On Mon, Jul 20, 2020 at 05:58:47PM -0300, Daniel Gutson wrote:
> > > On Fri, Jul 17, 2020 at 11:57 AM Greg Kroah-Hartman <
> > > gregkh@linuxfoundation.org> wrote:
> > >
> > > > On Fri, Jul 17, 2020 at 11:46:39AM -0300, Daniel Gutson wrote:
> > > > > On Fri, Jul 17, 2020 at 11:41 AM Arnd Bergmann <arnd@arndb.de>
> > wrote:
> > > > >
> > > > > > On Fri, Jul 17, 2020 at 8:28 AM Greg Kroah-Hartman
> > > > > > <gregkh@linuxfoundation.org> wrote:
> > > > > > >
> > > > > > > On Thu, Jul 16, 2020 at 07:36:27PM -0300, Daniel Gutson wrote:
> > > > > > > > +What:                /sys/kernel/firmware-security/bioswe
> > > > > > >
> > > > > > > Ick, I stopped reading right here.
> > > > > > >
> > > > > > > No, this is not where this belongs.
> > > > > > >
> > > > > > > We already have /sys/firmware/, right?  And firmware-specific
> > > > > > > subdirectories below that.
> > > > > > >
> > > > > > > We also have /sys/devices/system/ and I think that would be a
> > much
> > > > > > > better place for this, as it is easier to work with a real
> > 'struct
> > > > > > > device' than a "raw" kobject any day.  Bonus is you get full
> > support
> > > > of
> > > > > > > userspace libraries when you do that, unlike when dealing with
> > > > kobjects.
> > > > > > >
> > > > > > > Also, this really is a _SPECIFIC_ type of firmware that supports
> > > > these
> > > > > > > features, right?  Why not call that out too?  This is not
> > generic by
> > > > any
> > > > > > > means.
> > > > > >
> > > > > > As I suggested in my previous review, I wouldn't worry too much
> > about
> > > > > > the user interface at the start, but instead first work out how the
> > > > > > hardware
> > > > > > support fits in with the existing drivers and once that looks fine
> > > > decide
> > > > > > on how to export it to user space.
> > > > > >
> > > > > > I agree the  /sys/kernel/firmware-security/bioswe sounds like the
> > wrong
> > > > > > place, but I'm not sure if adding any other new directory in sysfs
> > is
> > > > > > much better. I think the most promising would be to have it on the
> > > > > > sysfs directory for the device it refers to,
> > > > >
> > > > >
> > > > > My idea is to have all the firmware security information together in
> > the
> > > > > same place; this information comes from many devices.
> > > > > This initial patch involves the SPI Controller, and I don't want to
> > add
> > > > > more stuff until there
> > > > > is a consensus.
> > > > > So, do you have a suggestion where to put this information?
> > > > > /sys/devices/system/firmware-security?
> > > > > /sys/firmware/security?
> > > > > other?
> > > > >
> > > > > Please advise.
> > > >
> > > > It's fun to focus on things like this, as it's the most visible part,
> > > > but are you sure the "talk to the hardware" part is working properly?
> > > >
> > >
> > > Yes.
> > >
> > >
> > >
> > > >
> > > > If so, great, it should be a "class", as that way it is independent of
> > > > any hardware type, right?  Classes show how devices talk to userspace
> > in
> > > > a common way (input, tty, led, block, etc.)  So why is this any
> > > > different from that?
> > > >
> > >
> > > Are you suggesting to create a new class, or use an existing one?
> >
> > Probably a new one, unless you can find an existing one that would fit?
> >
> 
> IIUC, Arnd Bergmann proposed that I create a class for each device driver
> (in this case,
> for the intel-spi, though I think the class would be spi-nor) and add
> attributes to it.
> In such a case, your proposal and his are different, mutually exclusive.

Classes should be driver agnositic, and you should not have a single
class per driver, as that is pretty pointless. 

> For me it's easier and makes more sense to have a class for, say,
> firmware-security (if I understood you correctly).

I still think that's a horrible name, as that is not what you are
describing here, but sure, a single class is good.

thanks,

greg k-h