On Fri, Nov 15, 2019 at 02:36:21PM -0400, Jason Gunthorpe wrote:
> On Fri, Nov 15, 2019 at 07:43:29PM +0200, Jarkko Sakkinen wrote:
> > On Thu, Nov 14, 2019 at 12:56:29PM -0400, Jason Gunthorpe wrote:
> > > On Thu, Nov 14, 2019 at 06:55:06PM +0200, Jarkko Sakkinen wrote:
> > > > > Would it function with the timeout values set at the beginning of
> > > > > tpm_tis_core_init (max values)?
> > > >
> > > > tpm_get_timeouts() should be replaced with:
> > > >
> > > > if (tpm_chip_start()) {
> > > > dev_err(dev, "Could not get TPM timeouts and durations\n");
> > > > rc = -ENODEV;
> > > > goto out_err;
> > > > }
> > > >
> > > > tpm_stop_chip(chip);
> > > >
> > > > tpm_get_timeouts() is called by tpm_auto_startup(). Also the function
> > > > should be moved to tpm_chip.c and converted to a static function so
> > > > that it won't be called from random cal sites like above.
> > >
> > > Careful, the design here was to allow a driver to do only
> > > get_timeouts, then additional setup work, then do auto_startup()
> > >
> > > Forcing a driver to do auto_startup too early may not be good.
> >
> > All drivers always do it anyway because all drivers always call
> > tpm_chip_register().
>
> But chip_register is after the driver has done it's setup and after it
> may have called get_timeouts
>
> auto_setup should not be moved to before chip_register()
I do not see any sense calling from get_timeouts() from call sites
in the same initialization flow.
/Jarkko
