[RFC] tpm: Keep CLKRUN enabled throughout the duration of transmit_cmd()

Commit Message

Azhar Shaikh Nov. 6, 2017, 10:11 p.m. UTC

>-----Original Message-----
>From: Jason Gunthorpe [mailto:[e-mail]]
>Sent: Monday, November 6, 2017 9:08 AM
>To: Shaikh, Azhar <[e-mail]>
>Cc: Sakkinen, Jarkko <[e-mail]>; linux-
>[e-mail]
>Subject: Re: [PATCH RFC] tpm: Keep CLKRUN enabled throughout the duration
>of transmit_cmd()
>
>On Fri, Nov 03, 2017 at 04:30:09PM -0700, Azhar Shaikh wrote:
>> @@ -413,6 +413,8 @@ ssize_t tpm_transmit(struct tpm_chip *chip, struct
>tpm_space *space,
>>  	if (chip->dev.parent)
>>  		pm_runtime_get_sync(chip->dev.parent);
>>
>> +	chip->ops->clk_toggle(chip, true);
>
>You added this new op to the general code, but only updated two drivers,
>surely this makes all the other tis drivers oops as clk_toggle will be NULL?
>
>Suggest checking for NULL before calling.
>

Sure, will add the NULL check.

>> +#ifdef CONFIG_X86
>
>This is a good place to use IS_ENABLED instead of ifdef:
>
>> +/**
>> + * tpm_tis_clkrun_toggle() - Keep clkrun protocol disabled for entire
>duration
>> + *                           of a single TPM command
>> + * @chip:	TPM chip to use
>> + * @value:	1 - Disable CLKRUN protocol, so that clocks are free running
>> + *		0 - Enable CLKRUN protocol
>> + */
>> +static void tpm_tis_clkrun_toggle(struct tpm_chip *chip, bool value)
>> +{
>> +	struct tpm_tis_data *data = dev_get_drvdata(&chip->dev);
>
>   if (!IS_ENABLED(CONFIG_X86))
>         return;
>

Will add this check and remove the #ifdef

>> +
>> +	if (value) {
>> +		tpm_platform_begin_xfer(data);
>> +		data->flags |= TPM_TIS_CLK_ENABLE;
>> +	} else {
>> +		data->flags &= ~TPM_TIS_CLK_ENABLE;
>> +		tpm_platform_end_xfer(data);
>> +	}
>> +}
>> +#else
>> +static void tpm_tis_clkrun_toggle(struct tpm_chip *chip, bool value)
>> +{ } #endif
>
>> +#ifdef CONFIG_X86
>> +void tpm_platform_begin_xfer(struct tpm_tis_data *data); void
>> +tpm_platform_end_xfer(struct tpm_tis_data *data); #else void
>> +tpm_platform_begin_xfer(struct tpm_tis_data *data) { } void
>> +tpm_platform_end_xfer(struct tpm_tis_data *data) { }
>
>These empty stubs need inlines
>

Sure, will add inline.

>Why are you using a mixture of callbacks and linked functions to solve this
>problem?
>
>Can't you do everything with callbacks?
>

I have set the flag TPM_TIS_CLK_ENABLE in the callback. But then in tpm_platform_begin_xfer() I want it to run once to disable clkrun and then return for all other instances, till the the flag is cleared.
If I just set the flag in tpm_tis_clkrun_toggle(), then for any TPM transaction after that the clkrun will not be disabled with current implementation. Hence calling it once before setting the flag so that it is kept enabled for the next transaction.

OR

I can update the callback as below:

+static void tpm_tis_clkrun_toggle(struct tpm_chip *chip, bool value)
+{
+	struct tpm_tis_data *data = dev_get_drvdata(&chip->dev);
+
+	if (!IS_ENABLED(CONFIG_X86))
+		return;
+
+	if (value)
+		data->flags |= TPM_TIS_CLK_ENABLE;
+	else
+		data->flags &= ~TPM_TIS_CLK_ENABLE;
+}


And in tpm_platform_begin_xfer() maintain a flag which will track, that it is called at least once before returning, after the TPM_TIS_CLK_ENABLE is set. But I am not sure if this is a good approach. Could you please suggest the correct approach here.
Below is a rough/test patch how this could be implemented.

Regards,
Azhar Shaikh

Comments

Jason Gunthorpe Nov. 6, 2017, 10:39 p.m. UTC | #1

On Mon, Nov 06, 2017 at 10:11:39PM +0000, Shaikh, Azhar wrote:

> >Why are you using a mixture of callbacks and linked functions to
> >solve this problem?
> >
> >Can't you do everything with callbacks?
> >
> 
> I have set the flag TPM_TIS_CLK_ENABLE in the callback. But then in
> tpm_platform_begin_xfer() I want it to run once to disable clkrun
> and then return for all other instances, till the the flag is
> cleared.  If I just set the flag in tpm_tis_clkrun_toggle(), then
> for any TPM transaction after that the clkrun will not be disabled
> with current implementation. Hence calling it once before setting
> the flag so that it is kept enabled for the next transaction.

I mean, why do we have global functions called tpm_platform_begin_xfer
at all, everything should be in a callback, not some mixture.

The static function sort of made sense when it was only ever called by
the tis driver, but now that you are hoisting things up a layer it is
not okay anymore, you need to add new driver callbacks instead.

> +++ b/drivers/char/tpm/tpm_tis.c
> @@ -140,6 +140,7 @@ static int check_acpi_tpm2(struct device *dev)
>  #define LPC_CLKRUN_EN                   (1 << 2)
> 
>  static void __iomem *ilb_base_addr;
> +static bool run_once;

No global statics. The ilb_base_addr global is really ugly you should
move it to struct tpm_tis_tcg_phy

Jason

Azhar Shaikh Nov. 6, 2017, 11:31 p.m. UTC | #2

>-----Original Message-----
>From: Jason Gunthorpe [mailto:[e-mail]]
>Sent: Monday, November 6, 2017 2:40 PM
>To: Shaikh, Azhar <[e-mail]>
>Cc: Sakkinen, Jarkko <[e-mail]>; linux-
>[e-mail]
>Subject: Re: [PATCH RFC] tpm: Keep CLKRUN enabled throughout the duration
>of transmit_cmd()
>
>On Mon, Nov 06, 2017 at 10:11:39PM +0000, Shaikh, Azhar wrote:
>
>> >Why are you using a mixture of callbacks and linked functions to
>> >solve this problem?
>> >
>> >Can't you do everything with callbacks?
>> >
>>
>> I have set the flag TPM_TIS_CLK_ENABLE in the callback. But then in
>> tpm_platform_begin_xfer() I want it to run once to disable clkrun and
>> then return for all other instances, till the the flag is cleared.  If
>> I just set the flag in tpm_tis_clkrun_toggle(), then for any TPM
>> transaction after that the clkrun will not be disabled with current
>> implementation. Hence calling it once before setting the flag so that
>> it is kept enabled for the next transaction.
>
>I mean, why do we have global functions called tpm_platform_begin_xfer at
>all, everything should be in a callback, not some mixture.
>
>The static function sort of made sense when it was only ever called by the tis
>driver, but now that you are hoisting things up a layer it is not okay anymore,
>you need to add new driver callbacks instead.
>

Ok, will not call tpm_platform_begin_xfer() from the callback function and make it back static again to keep it within tis driver only.

>> +++ b/drivers/char/tpm/tpm_tis.c
>> @@ -140,6 +140,7 @@ static int check_acpi_tpm2(struct device *dev)
>>  #define LPC_CLKRUN_EN                   (1 << 2)
>>
>>  static void __iomem *ilb_base_addr;
>> +static bool run_once;
>
>No global statics. The ilb_base_addr global is really ugly you should move it to
>struct tpm_tis_tcg_phy
>

Ok, I will do this as a separate patch.

>Jason

Patch

--- a/drivers/char/tpm/tpm_tis.c
+++ b/drivers/char/tpm/tpm_tis.c
@@ -140,6 +140,7 @@  static int check_acpi_tpm2(struct device *dev)
 #define LPC_CLKRUN_EN                   (1 << 2)

 static void __iomem *ilb_base_addr;
+static bool run_once;

 static inline bool is_bsw(void)
 {
@@ -154,7 +155,7 @@  void tpm_platform_begin_xfer(struct tpm_tis_data *data)
 {
        u32 clkrun_val;

-       if (!is_bsw() || (data->flags & TPM_TIS_CLK_ENABLE))
+       if (!is_bsw() || ((data->flags & TPM_TIS_CLK_ENABLE) && run_once))
                return;

        clkrun_val = ioread32(ilb_base_addr + LPC_CNTRL_REG_OFFSET);
@@ -169,6 +170,11 @@  void tpm_platform_begin_xfer(struct tpm_tis_data *data)
         */
        outb(0xCC, 0x80);

+       if (!(data->flags & TPM_TIS_CLK_ENABLE))
+               run_once = false;
+       else
+               run_once = true;
+
 }
>Jason