Re: [PATCH v4 06/15] mei: pxp: support matching with a gfx discrete card

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

On Fri, Sep 09, 2022 at 06:51:21AM +0000, Winkler, Tomas wrote:
> 
> 
> > -----Original Message-----
> > From: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> > Sent: Friday, September 09, 2022 09:16
> > To: Ceraolo Spurio, Daniele <daniele.ceraolospurio@intel.com>
> > Cc: intel-gfx@lists.freedesktop.org; dri-devel@lists.freedesktop.org;
> > Winkler, Tomas <tomas.winkler@intel.com>; Lubart, Vitaly
> > <vitaly.lubart@intel.com>; Teres Alexis, Alan Previn
> > <alan.previn.teres.alexis@intel.com>
> > Subject: Re: [PATCH v4 06/15] mei: pxp: support matching with a gfx discrete
> > card
> > 
> > On Thu, Sep 08, 2022 at 05:16:03PM -0700, Daniele Ceraolo Spurio wrote:
> > > From: Tomas Winkler <tomas.winkler@intel.com>
> > >
> > > With on-boards graphics card, both i915 and MEI are in the same device
> > > hierarchy with the same parent, while for discrete gfx card the MEI is
> > > its child device.
> > > Adjust the match function for that scenario by matching MEI parent
> > > device with i915.
> > >
> > > V2:
> > >  1. More detailed commit message
> > >  2. Check for dev is not null before it is accessed.
> > >
> > > Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
> > > Signed-off-by: Daniele Ceraolo Spurio
> > > <daniele.ceraolospurio@intel.com>
> > > Cc: Vitaly Lubart <vitaly.lubart@intel.com>
> > > Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> > > Reviewed-by: Alan Previn <alan.previn.teres.alexis@intel.com>
> > > ---
> > >  drivers/misc/mei/pxp/mei_pxp.c | 13 ++++++++++---
> > >  1 file changed, 10 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/drivers/misc/mei/pxp/mei_pxp.c
> > > b/drivers/misc/mei/pxp/mei_pxp.c index 17c5d201603f..afc047627800
> > > 100644
> > > --- a/drivers/misc/mei/pxp/mei_pxp.c
> > > +++ b/drivers/misc/mei/pxp/mei_pxp.c
> > > @@ -159,17 +159,24 @@ static int mei_pxp_component_match(struct
> > device
> > > *dev, int subcomponent,  {
> > >  	struct device *base = data;
> > >
> > > +	if (!dev)
> > > +		return 0;
> > 
> > How can that happen?
> > 
> > > +
> > >  	if (!dev->driver || strcmp(dev->driver->name, "i915") ||
> > 
> > That's crazy to assume, but whatever :(
> Explained here:
> https://lore.kernel.org/all/20220418175932.1809770-2-wonchung@google.com/

Still crazy :(

> 
> > 
> > >  	    subcomponent != I915_COMPONENT_PXP)
> > >  		return 0;
> > >
> > >  	base = base->parent;
> > > -	if (!base)
> > > +	if (!base) /* mei device */
> > 
> > Why does this mean that?
> > 
> > Where is that documented?
> > 
> > >  		return 0;
> > >
> > > -	base = base->parent;
> > > -	dev = dev->parent;
> > > +	base = base->parent; /* pci device */
> > 
> > Again, why is this the case?
> > 
> > > +	/* for dgfx */
> > > +	if (base && dev == base)
> > > +		return 1;
> > >
> > > +	/* for pch */
> > > +	dev = dev->parent;
> > 
> > You are digging through a random device tree and assuming that you "know"
> > what the topology is going to be, that feels very very fragile and ripe for
> > problems going forward.
> 
> I don't think it is random.

Today it is one specific way, but how do you know this always will be
this way?

> > How do you ensure that this really is they way the tree is for ALL systems?
> 
> Yes we take the topology assumption in PCI hierarchy.
> There is a case where both GFX and MEI are in PCH and you cannot stick additional PCI extender or anything else there. 
> And case where MEI is child on a standalone graphics card this is set in software so topology is not going to change unless we rewrite
> everything.  Be happy to hear your insights. 

This is ripe to break in the future if someone makes a differently
structured device as there is nothing forcing the current layout to
always be this way by hardware designers.

The goal of the driver model is to NOT have these types of hard-coded
topology assumptions because, supprise, assumptions like this have
always come back to bite people in the end.

This is your driver, so that's fine, but really this feels very very
wrong and you will have a hard time guaranteeing that this will always
be this way for the next 20+ years of hardware designs.  So why not do
it properly from the beginning and pass in the correct pointers to
different places?

There is a very good reason that the driver model/core does not make it
easy to determine what type of device a 'struct device *' is, because
you shouldn't have to rely on that type of thing ever.  You are taking
it one step further and just assuming that you know what the type is
here, with no real way to ensure that this is the case.

In short, this feels like a very bad design as it is very fragile.

thanks,

greg k-h