Re: [PATCH v2 3/5] soc: qcom: socinfo: Expose custom attributes

[Greg KH]

On Fri, Feb 22, 2019 at 11:51:45AM +0200, Laurent Pinchart wrote:
> Hi Greg,
> 
> On Fri, Feb 22, 2019 at 08:16:16AM +0100, Greg KH wrote:
> > On Fri, Feb 22, 2019 at 12:13:59AM +0200, Laurent Pinchart wrote:
> > > On Thu, Feb 21, 2019 at 07:57:42AM -0800, Bjorn Andersson wrote:
> > > > On Thu 21 Feb 04:18 PST 2019, Laurent Pinchart wrote:
> > > > > On Wed, Feb 20, 2019 at 10:28:29AM +0530, Vaishali Thakkar wrote:
> > > > > > The Qualcomm socinfo provides a number of additional attributes,
> > > > > > add these to the socinfo driver and expose them via debugfs
> > > > > > functionality.
> > > > > 
> > > > > What is the use case for these attributes ? I fear they will be used in
> > > > > production systems, and that would require debugfs in production, which
> > > > > isn't a good idea. If you need to expose those attributes for anything
> > > > > else than debugging then we need a proper API, likely sysfs-based.
> > > > 
> > > > The use case of these attributes, beyond development/debugging, are
> > > > unfortunately somewhat unknown and is the reason why they where moved to
> > > > debugfs from the earlier attempts to upstream this.
> > > > 
> > > > I think the production requirements at hand prohibits debugfs to be
> > > > present, so attributes that are required beyond development/debugging
> > > > purposes would have to be migrated out to sysfs - but the idea here is
> > > > that such migration would have come with the missing motivation to add
> > > > them there today.
> > > 
> > > If the use case is just debug/development, would it be enough to print
> > > this information in the kernel log at boot time ? I may be a bit
> > > paranoid, but I always worry about API abuse :-(
> > 
> > Putting stuff in debugfs should be fine.  No system should ever rely on
> > debugfs for a properly running system as it is being disabled on almost
> > all "sane" systems (Android included).  If a vendor relies on this
> > information for a properly working system, then it does not belong in
> > debugfs.
> 
> There's certainly no disagreement about that, my concern is about
> vendors who will enable debugfs to access information they need just
> because it's there.

Then they run a huge risk of relying on it.

> Do I assume correctly we can "break the debugfs ABI" in mainline by
> changing the format of the information if needed ?

Yes.  The only "rule" of debugfs is that there is no rule :)

Please, break it every-other release if you want, no userspace tools
_should_ be relying on it.[1]

thanks,

greg k-h

[1] Yes, I know Android currently relies on debugfs for a number of
    crash-dump type of things in ion and binder, but there is active
    work upstream in AOSP to remove that dependancy as no Android device
    should ever have debugfs mounted on a "normal" system for the
    security reasons alone.