Re: [tpmdd-devel] [PATCH 3/3] tpm, tpm_crb: runtime power management

[Jarkko Sakkinen]

On Thu, Jun 16, 2016 at 09:57:35PM +0200, Jarkko Sakkinen wrote:
> Hi Thomas,
> 
> I'm on a vacation this week but I'll give you quick answers :)
> 
> On Tue, Jun 14, 2016 at 04:14:58PM +0300, Tomas Winkler wrote:
> > On Wed, Jun 8, 2016 at 2:02 AM, Jarkko Sakkinen
> > <jarkko.sakkinen@xxxxxxxxxxxxxxx> wrote:
> > > The register TPM_CRB_CTRL_REQ_0 contains bits goIdle and cmdReady for
> > > invoking the chip to suspend and resume. This commit implements runtime
> > > PM for tpm_crb by using these bits.
> > >
> > > The legacy ACPI start (SMI + DMA) based devices do not support these
> > > bits. Thus this functionality only is enabled only for CRB start (MMIO)
> > > based devices.
> > >
> > > Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@xxxxxxxxxxxxxxx>
> > > ---
> > >  drivers/char/tpm/tpm-interface.c |  3 ++
> > >  drivers/char/tpm/tpm_crb.c       | 62 ++++++++++++++++++++++++++++++++++++++--
> > >  2 files changed, 63 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/char/tpm/tpm-interface.c b/drivers/char/tpm/tpm-interface.c
> > > index 5e3c1b6..3b85648 100644
> > > --- a/drivers/char/tpm/tpm-interface.c
> > > +++ b/drivers/char/tpm/tpm-interface.c
> > > @@ -29,6 +29,7 @@
> > >  #include <linux/mutex.h>
> > >  #include <linux/spinlock.h>
> > >  #include <linux/freezer.h>
> > > +#include <linux/pm_runtime.h>
> > >
> > >  #include "tpm.h"
> > >  #include "tpm_eventlog.h"
> > > @@ -350,6 +351,7 @@ ssize_t tpm_transmit(struct tpm_chip *chip, const char *buf,
> > >                 return -E2BIG;
> > >         }
> > >
> > > +       pm_runtime_get_sync(chip->dev.parent);
> > >         mutex_lock(&chip->tpm_mutex);
> > >
> > >         rc = chip->ops->send(chip, (u8 *) buf, count);
> > > @@ -394,6 +396,7 @@ out_recv:
> > >                         "tpm_transmit: tpm_recv: error %zd\n", rc);
> > >  out:
> > >         mutex_unlock(&chip->tpm_mutex);
> > > +       pm_runtime_put_sync(chip->dev.parent);
> > >         return rc;
> > >  }
> > >
> > > diff --git a/drivers/char/tpm/tpm_crb.c b/drivers/char/tpm/tpm_crb.c
> > > index ca2cad9..71cc7cd 100644
> > > --- a/drivers/char/tpm/tpm_crb.c
> > > +++ b/drivers/char/tpm/tpm_crb.c
> > > @@ -20,6 +20,7 @@
> > >  #include <linux/rculist.h>
> > >  #include <linux/module.h>
> > >  #include <linux/platform_device.h>
> > > +#include <linux/pm_runtime.h>
> > >  #include "tpm.h"
> > >
> > >  #define ACPI_SIG_TPM2 "TPM2"
> > > @@ -41,7 +42,6 @@ enum crb_ca_request {
> > >
> > >  enum crb_ca_status {
> > >         CRB_CA_STS_ERROR        = BIT(0),
> > > -       CRB_CA_STS_TPM_IDLE     = BIT(1),
> > >  };
> > >
> > >  enum crb_start {
> > > @@ -68,6 +68,8 @@ struct crb_control_area {
> > >
> > >  enum crb_status {
> > >         CRB_STS_COMPLETE        = BIT(0),
> > > +       CRB_STS_READY           = BIT(1),
> > > +       CRB_STS_IDLE            = BIT(2),
> > >  };
> > >
> > >  enum crb_flags {
> > > @@ -81,9 +83,52 @@ struct crb_priv {
> > >         struct crb_control_area __iomem *cca;
> > >         u8 __iomem *cmd;
> > >         u8 __iomem *rsp;
> > > +       wait_queue_head_t idle_queue;
> > 
> > 
> > I'm failing to find the code that is calling wake_up_interruptible(idle_queue);
> 
> Ugh, my bad. This actually should not be declared at all. Will remove it
> from the next version and NULL should be passed to wait_for_tpm_stat()
> as the driver does not yet support interrupts (Haswell did not have
> them, not sure about later gens).
> 
> 
> > >  };
> > >
> > > -static SIMPLE_DEV_PM_OPS(crb_pm, tpm_pm_suspend, tpm_pm_resume);
> > > +static int __maybe_unused crb_runtime_suspend(struct device *dev)
> > > +{
> > > +       struct tpm_chip *chip = dev_get_drvdata(dev);
> > > +       struct crb_priv *priv = dev_get_drvdata(&chip->dev);
> > > +       u32 req;
> > > +
> > > +       if (priv->flags & CRB_FL_ACPI_START)
> > > +               return 0;
> > > +
> > > +       req = ioread32(&priv->cca->req);
> > > +
> > > +       iowrite32(cpu_to_le32(req | CRB_CA_REQ_GO_IDLE), &priv->cca->req);
> > > +
> > > +       if (wait_for_tpm_stat(chip, CRB_STS_IDLE, chip->timeout_c,
> > > +                             &priv->idle_queue, false))
> > > +               dev_warn(&chip->dev, "idle timed out\n");
> > 
> > Unfortunately you cannot do that as there is an HW errata, the status
> > register value might not be defined here during power transition
> > You should poll for request CRB_CA_REQ_GO_IDLE goes to 0 as descried
> > in the spec . only after that you can check for the status register
> > (thought it's maybe not needed)
> 
> And I do exactly what you are asking me to do.
> 
> > > +
> > > +       return 0;
> > > +}
> > > +
> > > +static int __maybe_unused crb_runtime_resume(struct device *dev)
> > 
> > why this is marked unused, why just not compile it out? if the
> > CONFIG_PM is not set?
> 
> It is compiled out if it is unused. Why would you want to trash the code
> with #ifdef cages if they are not necessary? I can add /* CONFIG_PM */
> before the function if that makes it cleaner.
> 
> > > +{
> > > +       struct tpm_chip *chip = dev_get_drvdata(dev);
> > > +       struct crb_priv *priv = dev_get_drvdata(&chip->dev);
> > > +       u32 req;
> > > +
> > > +       if (priv->flags & CRB_FL_ACPI_START)
> > > +               return 0;
> > > +
> > > +       req = ioread32(&priv->cca->req);
> > > +       iowrite32(cpu_to_le32(req | CRB_CA_REQ_CMD_READY), &priv->cca->req);
> > > +
> > > +       if (wait_for_tpm_stat(chip, CRB_STS_READY, chip->timeout_c,
> > > +                             &priv->idle_queue, false))
> > > +               dev_warn(&chip->dev, "wake timed out\n");
> > 
> > Same here,  you should wait for CRB_CA_REQ_CMD_READ to get cleared,
> > only after that it is safe to check the status register.
> 
> It does exactly that. I'm not using CRB status register for anything.
> 
> > > +
> > > +       return 0;
> > > +}
> > > +
> > > +static const struct dev_pm_ops crb_pm = {
> > > +       SET_RUNTIME_PM_OPS(crb_runtime_suspend, crb_runtime_resume, NULL)
> > > +       SET_SYSTEM_SLEEP_PM_OPS(tpm_pm_suspend, tpm_pm_resume)
> > > +};
> > >
> > >  static u8 crb_status(struct tpm_chip *chip)
> > >  {
> > > @@ -94,6 +139,14 @@ static u8 crb_status(struct tpm_chip *chip)
> > >             CRB_START_INVOKE)
> > >                 sts |= CRB_STS_COMPLETE;
> > >
> > > +       if ((ioread32(&priv->cca->req) & CRB_CA_REQ_CMD_READY) !=
> > > +           CRB_CA_REQ_CMD_READY)
> > > +               sts |= CRB_STS_READY;
> > 
> > There is meaning for checking this w/o the actual transition i.e.
> > setting the before
> 
> I'm not sure what you are trying to say.

... but I can undertand why this looks so confusing to you. Maybe it
would be a better idea to completely discard the use of
wait_for_tpm_stat() and changes to crb_status() and do instead the
following in runtime_suspend/resume (example is for resume):

1. Set REQ_CMD_READY.
2. Sleep for TIMEOUT C.
3. Check the register and return -ETIME if it is not cleared.

I think this patch is abusing wait_for_tpm_stat() and it is not really
good fit here. How does this sound?

/Jarkko