Re: [PATCH] tpm: Add explicit chip->ops locking for sysfs attributes.

[Jarkko Sakkinen]

On Mon, Nov 27, 2017 at 08:02:55AM -0800, Guenter Roeck wrote:
> On Sun, Nov 26, 2017 at 03:56:41PM +0200, Jarkko Sakkinen wrote:
> > On Tue, Nov 21, 2017 at 11:58:56AM -0700, Jason Gunthorpe wrote:
> > > On Tue, Nov 21, 2017 at 10:28:58AM -0800, Guenter Roeck wrote:
> > > 
> > > > I'll split the patch into two parts, and only add (hopefully)
> > > > non-controversial tpm2 attributes for now (which I think is durations
> > > > and timeouts). Or, in other words, I'll split the attributes into
> > > > two groups - one generic and one for tpm1.
> > > 
> > > Ok. Please look at new attributes you wish to add for tpm2 and see if
> > > they meet the modern sysfs sensibility of one value per file, etc.
> > > 
> > > Jason
> > 
> > In general: if something can be retrieved through /dev/tpm0, there is no
> > any sane reason to have a sysfs attribute for such.
> > 
> 
> If I understand correctly, /dev/tpmX can be used to send any TPM command
> to the chip. Given that, I translate your statement to mean that no sysfs
> attribute will be accepted which sends a TPM command to the chip. This in
> turn means that there is no neded to protect sysfs attributes with a lock
> since any sysfs attribute requiring that lock will be rejected.
> 
> Thanks for the clarification. Please consider this patch abandoned.
> It might be worthwhile mentioning that restriction in the code though -
> the comment stating that TPM2 sysfs accesses are disabled due to lack
> of locking is obvioulsy incorrect.

Your statement about comment is correct. I guess we should rename the
file as tpm1_sysfs.c or tpm_legacy_sysfs.c.

It is not to say that new sysfs attributes would never make sense (like
in PPI case).

> Guenter

/Jarkko