lkml.org 
[lkml]   [2016]   [Jun]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: [tpmdd-devel] [PATCH 3/3] tpm, tpm_crb: runtime power management
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@linux.intel.com> 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@linux.intel.com>
> > ---
> > 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.

/Jarkko

\
 
 \ /
  Last update: 2016-06-16 22:21    [W:0.086 / U:3.804 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site