Re: TPM selftest failure in 4.15

[James Bottomley <James.Bottomley@HansenPartnership.com>]

On Thu, 2018-02-01 at 13:35 -0700, Jason Gunthorpe wrote:
> On Thu, Feb 01, 2018 at 09:00:04PM +0100, James Bottomley wrote:
> > 
> > On Thu, 2018-02-01 at 11:59 -0700, Jason Gunthorpe wrote:
> > > 
> > > On Thu, Feb 01, 2018 at 07:46:04PM +0100, James Bottomley wrote:
> > > 
> > > > 
> > > > 
> > > > I honestly don't think we should be waiting for the self test
> > > > at
> > > > all.
> > > > We should kick it off and treat any TPM_RC_TESTING error as
> > > > -EAGAIN.
> > > > We're already under fire for slow boot sequences and adding 2s
> > > > just
> > > > to
> > > > wait for the TPM to self test adds to that for no real value.
> > > 
> > > Arguably the BIOS should have completed the selftest - this stuff
> > > generally only exists to support embedded.
> > > 
> > > I don't like the idea of EAGAIN, that just expose all our users
> > > to
> > > this mess.
> > > 
> > > I would support making transmit_cmd genericly wait and retry if
> > > the
> > > TPM insists we need to wait for selftest to complete the specific
> > > command though.
> > 
> > OK, how about this then?
> 
> Yeah, I like this concept much better, thanks

Great, I'll put it through a few tests then send a formal patch.

> > diff --git a/drivers/char/tpm/tpm-interface.c
> > b/drivers/char/tpm/tpm-interface.c
> > index 1d6729be4cd6..84ed271c060b 100644
> > +++ b/drivers/char/tpm/tpm-interface.c
> > @@ -521,12 +521,32 @@ ssize_t tpm_transmit_cmd(struct tpm_chip
> > *chip, struct tpm_space *space,
> >  	const struct tpm_output_header *header = buf;
> >  	int err;
> >  	ssize_t len;
> > +	unsigned int delay_msec = 20;
> >  
> > -	len = tpm_transmit(chip, space, (u8 *)buf, bufsiz, flags);
> > -	if (len <  0)
> > -		return len;
> > +	/*
> > +	 * on first probe we kick off a TPM self test in the
> > +	 * background This means the TPM may return RC_TESTING to
> > any
> > +	 * command that tries to use a subsystem under test, so do
> > an
> > +	 * exponential backoff wait if that happens
> > +	 */
> > +	for (;;) {
> > +		len = tpm_transmit(chip, space, (u8 *)buf, bufsiz,
> > flags);
> > +		if (len <  0)
> > +			return len;
> > +
> > +		err = be32_to_cpu(header->return_code);
> > +		if (err != TPM2_RC_TESTING ||
> > +		    (flags & TPM_TRANSMIT_NOWAIT))
> > +			break;
> 
> Do TPM and TPM2 use a different return code here?

Yes, TPM2_RC_TESTING is 0x90a and TPM_DOING_SELFTEST is 0x802

James