[v8,08/11] tpm, tpm: Implement usage counter for locality

Commit Message

Lino Sanfilippo Oct. 17, 2022, 11:57 p.m. UTC

From: Lino Sanfilippo <[e-mail]>

Implement a usage counter for the (default) locality used by the TPM TIS
driver:
Request the locality from the TPM if it has not been claimed yet, otherwise
only increment the counter. Also release the locality if the counter is 0
otherwise only decrement the counter. Ensure thread-safety by protecting
the counter with a mutex.

This allows to request and release the locality from a thread and the
interrupt handler at the same time without the danger to interfere with
each other.

By doing this refactor the names of the amended functions to use the proper
prefix.

Signed-off-by: Lino Sanfilippo <[e-mail]>
Tested-by: Michael Niewöhner <[e-mail]>
---
 drivers/char/tpm/tpm_tis_core.c | 75 ++++++++++++++++++++++-----------
 drivers/char/tpm/tpm_tis_core.h |  2 +
 2 files changed, 53 insertions(+), 24 deletions(-)

Comments

Lukas Wunner Oct. 18, 2022, 6:25 a.m. UTC | #1

On Tue, Oct 18, 2022 at 01:57:29AM +0200, Lino Sanfilippo wrote:
> Implement a usage counter for the (default) locality used by the TPM TIS
> driver:
> Request the locality from the TPM if it has not been claimed yet, otherwise
> only increment the counter. Also release the locality if the counter is 0
> otherwise only decrement the counter. Ensure thread-safety by protecting
> the counter with a mutex.
> 
> This allows to request and release the locality from a thread and the
> interrupt handler at the same time without the danger to interfere with
> each other.
[...]
> +static int tpm_tis_release_locality(struct tpm_chip *chip, int l)
>  {
>  	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
>  
> -	tpm_tis_write8(priv, TPM_ACCESS(l), TPM_ACCESS_ACTIVE_LOCALITY);
> +	mutex_lock(&priv->locality_count_mutex);
> +	priv->locality_count--;
> +	if (priv->locality_count == 0)
> +		tpm_tis_release_locality_locked(priv, l);
> +	mutex_unlock(&priv->locality_count_mutex);
>  
>  	return 0;
>  }

Hm, any reason not to use struct kref for the locality counter?
Provides correct memory ordering (no mutex needed) and allows for
calling a release function too upon reaching 0.

Thanks,

Lukas

Lino Sanfilippo Oct. 18, 2022, 7:42 a.m. UTC | #2

On 18.10.22 08:25, Lukas Wunner wrote:
> Hm, any reason not to use struct kref for the locality counter?
> Provides correct memory ordering (no mutex needed) and allows for
> calling a release function too upon reaching 0.
>

I already tried this but krefs turned out to be not very usable
in this case. See my post here:

https://lore.kernel.org/lkml/[e-mail]/

Regards,
Lino

Jarkko Sakkinen Oct. 23, 2022, 4:40 a.m. UTC | #3

On Tue, Oct 18, 2022 at 01:57:29AM +0200, Lino Sanfilippo wrote:
> From: Lino Sanfilippo <[e-mail]>
> 
> Implement a usage counter for the (default) locality used by the TPM TIS
> driver:
> Request the locality from the TPM if it has not been claimed yet, otherwise
> only increment the counter. Also release the locality if the counter is 0
> otherwise only decrement the counter. Ensure thread-safety by protecting
> the counter with a mutex.
> 
> This allows to request and release the locality from a thread and the
> interrupt handler at the same time without the danger to interfere with
> each other.
> 
> By doing this refactor the names of the amended functions to use the proper
> prefix.
> 
> Signed-off-by: Lino Sanfilippo <[e-mail]>
> Tested-by: Michael Niewöhner <[e-mail]>
> ---
>  drivers/char/tpm/tpm_tis_core.c | 75 ++++++++++++++++++++++-----------
>  drivers/char/tpm/tpm_tis_core.h |  2 +
>  2 files changed, 53 insertions(+), 24 deletions(-)
> 
> diff --git a/drivers/char/tpm/tpm_tis_core.c b/drivers/char/tpm/tpm_tis_core.c
> index 4336f7ea8c2b..79dfab65976f 100644
> --- a/drivers/char/tpm/tpm_tis_core.c
> +++ b/drivers/char/tpm/tpm_tis_core.c
> @@ -165,16 +165,27 @@ static bool check_locality(struct tpm_chip *chip, int l)
>  	return false;
>  }
>  
> -static int release_locality(struct tpm_chip *chip, int l)
> +static int tpm_tis_release_locality_locked(struct tpm_tis_data *priv, int l)

Nit: usually you would actually use "unlocked" here, not locked.

Probably best name would be __tpm_tis_release_locality().

> +{
> +	tpm_tis_write8(priv, TPM_ACCESS(l), TPM_ACCESS_ACTIVE_LOCALITY);
> +
> +	return 0;
> +}
> +
> +static int tpm_tis_release_locality(struct tpm_chip *chip, int l)
>  {
>  	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
>  
> -	tpm_tis_write8(priv, TPM_ACCESS(l), TPM_ACCESS_ACTIVE_LOCALITY);
> +	mutex_lock(&priv->locality_count_mutex);
> +	priv->locality_count--;
> +	if (priv->locality_count == 0)
> +		tpm_tis_release_locality_locked(priv, l);
> +	mutex_unlock(&priv->locality_count_mutex);
>  
>  	return 0;
>  }

Since the function pointer has the word "relinquish" and not "release",
perhaps these should also use that word for consistency.

>  
> -static int request_locality(struct tpm_chip *chip, int l)
> +static int tpm_tis_request_locality_locked(struct tpm_chip *chip, int l)
>  {
>  	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
>  	unsigned long stop, timeout;
> @@ -215,6 +226,20 @@ static int request_locality(struct tpm_chip *chip, int l)
>  	return -1;
>  }
>  
> +static int tpm_tis_request_locality(struct tpm_chip *chip, int l)
> +{
> +	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
> +	int ret = 0;
> +
> +	mutex_lock(&priv->locality_count_mutex);
> +	if (priv->locality_count == 0)
> +		ret = tpm_tis_request_locality_locked(chip, l);
> +	if (!ret)
> +		priv->locality_count++;
> +	mutex_unlock(&priv->locality_count_mutex);
> +	return ret;
> +}
> +
>  static u8 tpm_tis_status(struct tpm_chip *chip)
>  {
>  	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
> @@ -682,7 +707,7 @@ static int probe_itpm(struct tpm_chip *chip)
>  	if (vendor != TPM_VID_INTEL)
>  		return 0;
>  
> -	if (request_locality(chip, 0) != 0)
> +	if (tpm_tis_request_locality(chip, 0) != 0)
>  		return -EBUSY;
>  
>  	rc = tpm_tis_send_data(chip, cmd_getticks, len);
> @@ -703,7 +728,7 @@ static int probe_itpm(struct tpm_chip *chip)
>  
>  out:
>  	tpm_tis_ready(chip);
> -	release_locality(chip, priv->locality);
> +	tpm_tis_release_locality(chip, priv->locality);
>  
>  	return rc;
>  }
> @@ -762,7 +787,7 @@ static int tpm_tis_gen_interrupt(struct tpm_chip *chip)
>  	cap_t cap;
>  	int ret;
>  
> -	ret = request_locality(chip, 0);
> +	ret = tpm_tis_request_locality(chip, 0);
>  	if (ret < 0)
>  		return ret;
>  
> @@ -771,7 +796,7 @@ static int tpm_tis_gen_interrupt(struct tpm_chip *chip)
>  	else
>  		ret = tpm1_getcap(chip, TPM_CAP_PROP_TIS_TIMEOUT, &cap, desc, 0);
>  
> -	release_locality(chip, 0);
> +	tpm_tis_release_locality(chip, 0);
>  
>  	return ret;
>  }
> @@ -796,33 +821,33 @@ static int tpm_tis_probe_irq_single(struct tpm_chip *chip, u32 intmask,
>  	}
>  	priv->irq = irq;
>  
> -	rc = request_locality(chip, 0);
> +	rc = tpm_tis_request_locality(chip, 0);
>  	if (rc < 0)
>  		return rc;
>  
>  	rc = tpm_tis_read8(priv, TPM_INT_VECTOR(priv->locality),
>  			   &original_int_vec);
>  	if (rc < 0) {
> -		release_locality(chip, priv->locality);
> +		tpm_tis_release_locality(chip, priv->locality);
>  		return rc;
>  	}
>  
>  	rc = tpm_tis_write8(priv, TPM_INT_VECTOR(priv->locality), irq);
>  	if (rc < 0) {
> -		release_locality(chip, priv->locality);
> +		tpm_tis_release_locality(chip, priv->locality);
>  		return rc;
>  	}
>  
>  	rc = tpm_tis_read32(priv, TPM_INT_STATUS(priv->locality), &int_status);
>  	if (rc < 0) {
> -		release_locality(chip, priv->locality);
> +		tpm_tis_release_locality(chip, priv->locality);
>  		return rc;
>  	}
>  
>  	/* Clear all existing */
>  	rc = tpm_tis_write32(priv, TPM_INT_STATUS(priv->locality), int_status);
>  	if (rc < 0) {
> -		release_locality(chip, priv->locality);
> +		tpm_tis_release_locality(chip, priv->locality);
>  		return rc;
>  	}
>  
> @@ -830,11 +855,11 @@ static int tpm_tis_probe_irq_single(struct tpm_chip *chip, u32 intmask,
>  	rc = tpm_tis_write32(priv, TPM_INT_ENABLE(priv->locality),
>  			     intmask | TPM_GLOBAL_INT_ENABLE);
>  	if (rc < 0) {
> -		release_locality(chip, priv->locality);
> +		tpm_tis_release_locality(chip, priv->locality);
>  		return rc;
>  	}
>  
> -	release_locality(chip, priv->locality);
> +	tpm_tis_release_locality(chip, priv->locality);
>  	clear_bit(TPM_TIS_IRQ_TESTED, &priv->flags);
>  
>  	/* Generate an interrupt by having the core call through to
> @@ -970,8 +995,8 @@ static const struct tpm_class_ops tpm_tis = {
>  	.req_complete_mask = TPM_STS_DATA_AVAIL | TPM_STS_VALID,
>  	.req_complete_val = TPM_STS_DATA_AVAIL | TPM_STS_VALID,
>  	.req_canceled = tpm_tis_req_canceled,
> -	.request_locality = request_locality,
> -	.relinquish_locality = release_locality,
> +	.request_locality = tpm_tis_request_locality,
> +	.relinquish_locality = tpm_tis_release_locality,
>  	.clk_enable = tpm_tis_clkrun_enable,
>  };
>  
> @@ -1005,6 +1030,8 @@ int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int irq,
>  	priv->timeout_min = TPM_TIMEOUT_USECS_MIN;
>  	priv->timeout_max = TPM_TIMEOUT_USECS_MAX;
>  	priv->phy_ops = phy_ops;
> +	priv->locality_count = 0;
> +	mutex_init(&priv->locality_count_mutex);
>  
>  	dev_set_drvdata(&chip->dev, priv);
>  
> @@ -1083,14 +1110,14 @@ int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int irq,
>  
>  	intmask &= ~TPM_GLOBAL_INT_ENABLE;
>  
> -	rc = request_locality(chip, 0);
> +	rc = tpm_tis_request_locality(chip, 0);
>  	if (rc < 0) {
>  		rc = -ENODEV;
>  		goto out_err;
>  	}
>  
>  	tpm_tis_write32(priv, TPM_INT_ENABLE(priv->locality), intmask);
> -	release_locality(chip, 0);
> +	tpm_tis_release_locality(chip, 0);
>  
>  	rc = tpm_chip_start(chip);
>  	if (rc)
> @@ -1124,13 +1151,13 @@ int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int irq,
>  		 * proper timeouts for the driver.
>  		 */
>  
> -		rc = request_locality(chip, 0);
> +		rc = tpm_tis_request_locality(chip, 0);
>  		if (rc < 0)
>  			goto out_err;
>  
>  		rc = tpm_get_timeouts(chip);
>  
> -		release_locality(chip, 0);
> +		tpm_tis_release_locality(chip, 0);
>  
>  		if (rc) {
>  			dev_err(dev, "Could not get TPM timeouts and durations\n");
> @@ -1150,11 +1177,11 @@ int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int irq,
>  			dev_err(&chip->dev, FW_BUG
>  					"TPM interrupt not working, polling instead\n");
>  
> -			rc = request_locality(chip, 0);
> +			rc = tpm_tis_request_locality(chip, 0);
>  			if (rc < 0)
>  				goto out_err;
>  			disable_interrupts(chip);
> -			release_locality(chip, 0);
> +			tpm_tis_release_locality(chip, 0);
>  		}
>  	}
>  
> @@ -1221,13 +1248,13 @@ int tpm_tis_resume(struct device *dev)
>  	 * an error code but for unknown reason it isn't handled.
>  	 */
>  	if (!(chip->flags & TPM_CHIP_FLAG_TPM2)) {
> -		ret = request_locality(chip, 0);
> +		ret = tpm_tis_request_locality(chip, 0);
>  		if (ret < 0)
>  			return ret;
>  
>  		tpm1_do_selftest(chip);
>  
> -		release_locality(chip, 0);
> +		tpm_tis_release_locality(chip, 0);
>  	}
>  
>  	return 0;
> diff --git a/drivers/char/tpm/tpm_tis_core.h b/drivers/char/tpm/tpm_tis_core.h
> index 2deef11c88db..13bdcf38e56f 100644
> --- a/drivers/char/tpm/tpm_tis_core.h
> +++ b/drivers/char/tpm/tpm_tis_core.h
> @@ -91,6 +91,8 @@ enum tpm_tis_flags {
>  
>  struct tpm_tis_data {
>  	u16 manufacturer_id;
> +	struct mutex locality_count_mutex;

BTW, why mutex and not spinlock?

Hmm.. also I think you might have given feedback already on this
but could the lock cover the whole struct instead of a counter?
I tried to dig lore for earlier response but could not find. I'm
sorry if I'm asking the same question again.

You could probably use rcu for that.

> +	unsigned int locality_count;
>  	int locality;
>  	int irq;
>  	unsigned int int_mask;
> -- 
> 2.36.1
>

BR, Jarkko

Jarkko Sakkinen Oct. 23, 2022, 5:26 a.m. UTC | #4

On Tue, Oct 18, 2022 at 08:25:08AM +0200, Lukas Wunner wrote:
> On Tue, Oct 18, 2022 at 01:57:29AM +0200, Lino Sanfilippo wrote:
> > Implement a usage counter for the (default) locality used by the TPM TIS
> > driver:
> > Request the locality from the TPM if it has not been claimed yet, otherwise
> > only increment the counter. Also release the locality if the counter is 0
> > otherwise only decrement the counter. Ensure thread-safety by protecting
> > the counter with a mutex.
> > 
> > This allows to request and release the locality from a thread and the
> > interrupt handler at the same time without the danger to interfere with
> > each other.
> [...]
> > +static int tpm_tis_release_locality(struct tpm_chip *chip, int l)
> >  {
> >  	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
> >  
> > -	tpm_tis_write8(priv, TPM_ACCESS(l), TPM_ACCESS_ACTIVE_LOCALITY);
> > +	mutex_lock(&priv->locality_count_mutex);
> > +	priv->locality_count--;
> > +	if (priv->locality_count == 0)
> > +		tpm_tis_release_locality_locked(priv, l);
> > +	mutex_unlock(&priv->locality_count_mutex);
> >  
> >  	return 0;
> >  }
> 
> Hm, any reason not to use struct kref for the locality counter?
> Provides correct memory ordering (no mutex needed) and allows for
> calling a release function too upon reaching 0.

I proposed for last version kref. I have no idea why this is still
using mutex. And now I apparently have proposed rcu for the whole
struct (forgot what I had put my feedback for earlier version).

This keeps being confusing patch as the commit message does not
really go to the bottom line why mutex is really the best possible
choice here.

BR, Jarkko

Lino Sanfilippo Oct. 25, 2022, 12:15 a.m. UTC | #5

On 23.10.22 06:40, Jarkko Sakkinen wrote:
> On Tue, Oct 18, 2022 at 01:57:29AM +0200, Lino Sanfilippo wrote:
>> From: Lino Sanfilippo <[e-mail]>
>>
>> Implement a usage counter for the (default) locality used by the TPM TIS
>> driver:
>> Request the locality from the TPM if it has not been claimed yet, otherwise
>> only increment the counter. Also release the locality if the counter is 0
>> otherwise only decrement the counter. Ensure thread-safety by protecting
>> the counter with a mutex.
>>
>> This allows to request and release the locality from a thread and the
>> interrupt handler at the same time without the danger to interfere with
>> each other.
>>
>> By doing this refactor the names of the amended functions to use the proper
>> prefix.
>>
>> Signed-off-by: Lino Sanfilippo <[e-mail]>
>> Tested-by: Michael Niewöhner <[e-mail]>
>> ---
>>  drivers/char/tpm/tpm_tis_core.c | 75 ++++++++++++++++++++++-----------
>>  drivers/char/tpm/tpm_tis_core.h |  2 +
>>  2 files changed, 53 insertions(+), 24 deletions(-)
>>
>> diff --git a/drivers/char/tpm/tpm_tis_core.c b/drivers/char/tpm/tpm_tis_core.c
>> index 4336f7ea8c2b..79dfab65976f 100644
>> --- a/drivers/char/tpm/tpm_tis_core.c
>> +++ b/drivers/char/tpm/tpm_tis_core.c
>> @@ -165,16 +165,27 @@ static bool check_locality(struct tpm_chip *chip, int l)
>>  	return false;
>>  }
>>
>> -static int release_locality(struct tpm_chip *chip, int l)
>> +static int tpm_tis_release_locality_locked(struct tpm_tis_data *priv, int l)
>
> Nit: usually you would actually use "unlocked" here, not locked.
>
> Probably best name would be __tpm_tis_release_locality().


Agreed. This is also consistent with the naming scheme used for many other
kernel functions.

>
>> +{
>> +	tpm_tis_write8(priv, TPM_ACCESS(l), TPM_ACCESS_ACTIVE_LOCALITY);
>> +
>> +	return 0;
>> +}
>> +
>> +static int tpm_tis_release_locality(struct tpm_chip *chip, int l)
>>  {
>>  	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
>>
>> -	tpm_tis_write8(priv, TPM_ACCESS(l), TPM_ACCESS_ACTIVE_LOCALITY);
>> +	mutex_lock(&priv->locality_count_mutex);
>> +	priv->locality_count--;
>> +	if (priv->locality_count == 0)
>> +		tpm_tis_release_locality_locked(priv, l);
>> +	mutex_unlock(&priv->locality_count_mutex);
>>
>>  	return 0;
>>  }
>
> Since the function pointer has the word "relinquish" and not "release",
> perhaps these should also use that word for consistency.
>
>>
>> -static int request_locality(struct tpm_chip *chip, int l)
>> +static int tpm_tis_request_locality_locked(struct tpm_chip *chip, int l)
>>  {
>>  	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
>>  	unsigned long stop, timeout;
>> @@ -215,6 +226,20 @@ static int request_locality(struct tpm_chip *chip, int l)
>>  	return -1;
>>  }
>>
>> +static int tpm_tis_request_locality(struct tpm_chip *chip, int l)
>> +{
>> +	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
>> +	int ret = 0;
>> +
>> +	mutex_lock(&priv->locality_count_mutex);
>> +	if (priv->locality_count == 0)
>> +		ret = tpm_tis_request_locality_locked(chip, l);
>> +	if (!ret)
>> +		priv->locality_count++;
>> +	mutex_unlock(&priv->locality_count_mutex);
>> +	return ret;
>> +}
>> +
>>  static u8 tpm_tis_status(struct tpm_chip *chip)
>>  {
>>  	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
>> @@ -682,7 +707,7 @@ static int probe_itpm(struct tpm_chip *chip)
>>  	if (vendor != TPM_VID_INTEL)
>>  		return 0;
>>
>> -	if (request_locality(chip, 0) != 0)
>> +	if (tpm_tis_request_locality(chip, 0) != 0)
>>  		return -EBUSY;
>>
>>  	rc = tpm_tis_send_data(chip, cmd_getticks, len);
>> @@ -703,7 +728,7 @@ static int probe_itpm(struct tpm_chip *chip)
>>
>>  out:
>>  	tpm_tis_ready(chip);
>> -	release_locality(chip, priv->locality);
>> +	tpm_tis_release_locality(chip, priv->locality);
>>
>>  	return rc;
>>  }
>> @@ -762,7 +787,7 @@ static int tpm_tis_gen_interrupt(struct tpm_chip *chip)
>>  	cap_t cap;
>>  	int ret;
>>
>> -	ret = request_locality(chip, 0);
>> +	ret = tpm_tis_request_locality(chip, 0);
>>  	if (ret < 0)
>>  		return ret;
>>
>> @@ -771,7 +796,7 @@ static int tpm_tis_gen_interrupt(struct tpm_chip *chip)
>>  	else
>>  		ret = tpm1_getcap(chip, TPM_CAP_PROP_TIS_TIMEOUT, &cap, desc, 0);
>>
>> -	release_locality(chip, 0);
>> +	tpm_tis_release_locality(chip, 0);
>>
>>  	return ret;
>>  }
>> @@ -796,33 +821,33 @@ static int tpm_tis_probe_irq_single(struct tpm_chip *chip, u32 intmask,
>>  	}
>>  	priv->irq = irq;
>>
>> -	rc = request_locality(chip, 0);
>> +	rc = tpm_tis_request_locality(chip, 0);
>>  	if (rc < 0)
>>  		return rc;
>>
>>  	rc = tpm_tis_read8(priv, TPM_INT_VECTOR(priv->locality),
>>  			   &original_int_vec);
>>  	if (rc < 0) {
>> -		release_locality(chip, priv->locality);
>> +		tpm_tis_release_locality(chip, priv->locality);
>>  		return rc;
>>  	}
>>
>>  	rc = tpm_tis_write8(priv, TPM_INT_VECTOR(priv->locality), irq);
>>  	if (rc < 0) {
>> -		release_locality(chip, priv->locality);
>> +		tpm_tis_release_locality(chip, priv->locality);
>>  		return rc;
>>  	}
>>
>>  	rc = tpm_tis_read32(priv, TPM_INT_STATUS(priv->locality), &int_status);
>>  	if (rc < 0) {
>> -		release_locality(chip, priv->locality);
>> +		tpm_tis_release_locality(chip, priv->locality);
>>  		return rc;
>>  	}
>>
>>  	/* Clear all existing */
>>  	rc = tpm_tis_write32(priv, TPM_INT_STATUS(priv->locality), int_status);
>>  	if (rc < 0) {
>> -		release_locality(chip, priv->locality);
>> +		tpm_tis_release_locality(chip, priv->locality);
>>  		return rc;
>>  	}
>>
>> @@ -830,11 +855,11 @@ static int tpm_tis_probe_irq_single(struct tpm_chip *chip, u32 intmask,
>>  	rc = tpm_tis_write32(priv, TPM_INT_ENABLE(priv->locality),
>>  			     intmask | TPM_GLOBAL_INT_ENABLE);
>>  	if (rc < 0) {
>> -		release_locality(chip, priv->locality);
>> +		tpm_tis_release_locality(chip, priv->locality);
>>  		return rc;
>>  	}
>>
>> -	release_locality(chip, priv->locality);
>> +	tpm_tis_release_locality(chip, priv->locality);
>>  	clear_bit(TPM_TIS_IRQ_TESTED, &priv->flags);
>>
>>  	/* Generate an interrupt by having the core call through to
>> @@ -970,8 +995,8 @@ static const struct tpm_class_ops tpm_tis = {
>>  	.req_complete_mask = TPM_STS_DATA_AVAIL | TPM_STS_VALID,
>>  	.req_complete_val = TPM_STS_DATA_AVAIL | TPM_STS_VALID,
>>  	.req_canceled = tpm_tis_req_canceled,
>> -	.request_locality = request_locality,
>> -	.relinquish_locality = release_locality,
>> +	.request_locality = tpm_tis_request_locality,
>> +	.relinquish_locality = tpm_tis_release_locality,
>>  	.clk_enable = tpm_tis_clkrun_enable,
>>  };
>>
>> @@ -1005,6 +1030,8 @@ int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int irq,
>>  	priv->timeout_min = TPM_TIMEOUT_USECS_MIN;
>>  	priv->timeout_max = TPM_TIMEOUT_USECS_MAX;
>>  	priv->phy_ops = phy_ops;
>> +	priv->locality_count = 0;
>> +	mutex_init(&priv->locality_count_mutex);
>>
>>  	dev_set_drvdata(&chip->dev, priv);
>>
>> @@ -1083,14 +1110,14 @@ int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int irq,
>>
>>  	intmask &= ~TPM_GLOBAL_INT_ENABLE;
>>
>> -	rc = request_locality(chip, 0);
>> +	rc = tpm_tis_request_locality(chip, 0);
>>  	if (rc < 0) {
>>  		rc = -ENODEV;
>>  		goto out_err;
>>  	}
>>
>>  	tpm_tis_write32(priv, TPM_INT_ENABLE(priv->locality), intmask);
>> -	release_locality(chip, 0);
>> +	tpm_tis_release_locality(chip, 0);
>>
>>  	rc = tpm_chip_start(chip);
>>  	if (rc)
>> @@ -1124,13 +1151,13 @@ int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int irq,
>>  		 * proper timeouts for the driver.
>>  		 */
>>
>> -		rc = request_locality(chip, 0);
>> +		rc = tpm_tis_request_locality(chip, 0);
>>  		if (rc < 0)
>>  			goto out_err;
>>
>>  		rc = tpm_get_timeouts(chip);
>>
>> -		release_locality(chip, 0);
>> +		tpm_tis_release_locality(chip, 0);
>>
>>  		if (rc) {
>>  			dev_err(dev, "Could not get TPM timeouts and durations\n");
>> @@ -1150,11 +1177,11 @@ int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int irq,
>>  			dev_err(&chip->dev, FW_BUG
>>  					"TPM interrupt not working, polling instead\n");
>>
>> -			rc = request_locality(chip, 0);
>> +			rc = tpm_tis_request_locality(chip, 0);
>>  			if (rc < 0)
>>  				goto out_err;
>>  			disable_interrupts(chip);
>> -			release_locality(chip, 0);
>> +			tpm_tis_release_locality(chip, 0);
>>  		}
>>  	}
>>
>> @@ -1221,13 +1248,13 @@ int tpm_tis_resume(struct device *dev)
>>  	 * an error code but for unknown reason it isn't handled.
>>  	 */
>>  	if (!(chip->flags & TPM_CHIP_FLAG_TPM2)) {
>> -		ret = request_locality(chip, 0);
>> +		ret = tpm_tis_request_locality(chip, 0);
>>  		if (ret < 0)
>>  			return ret;
>>
>>  		tpm1_do_selftest(chip);
>>
>> -		release_locality(chip, 0);
>> +		tpm_tis_release_locality(chip, 0);
>>  	}
>>
>>  	return 0;
>> diff --git a/drivers/char/tpm/tpm_tis_core.h b/drivers/char/tpm/tpm_tis_core.h
>> index 2deef11c88db..13bdcf38e56f 100644
>> --- a/drivers/char/tpm/tpm_tis_core.h
>> +++ b/drivers/char/tpm/tpm_tis_core.h
>> @@ -91,6 +91,8 @@ enum tpm_tis_flags {
>>
>>  struct tpm_tis_data {
>>  	u16 manufacturer_id;
>> +	struct mutex locality_count_mutex;
>
> BTW, why mutex and not spinlock?
>
> Hmm.. also I think you might have given feedback already on this
> but could the lock cover the whole struct instead of a counter?
> I tried to dig lore for earlier response but could not find. I'm
> sorry if I'm asking the same question again.
>

Actually thats on me, since it took me much too long to send the v8 after the v7 review.

However the reason that we need a mutex here is that we not only increase or decrease
the locality_counter under the mutex, but also do the locality request and release by
writing to the ACCESS register. Since in the SPI case each communication over the spi bus
is protected by the bus_lock_mutex of the SPI device we must not hold a spinlock when doing
the register accesses.

Concerning covering the whole tpm_tis_data struct:
Most structure elements are set once at driver startup but never changed at driver
runtime. So no locking needed for these. The only exception is "flags" and "locality_count"
whereby "flags" is accessed by atomic bit manipulating functions and thus
does not need extra locking. So "locality_count" is AFAICS the only element that needs to be
protected by the mutex.


> You could probably use rcu for that.
>
>> +	unsigned int locality_count;

Note that we do not only have to increase or decrease the counter atomically, but the first
increment (when the counter goes from 1 to 0) AND requesting the locality by writing to the
ACCESS register has to be atomic.
Likewise the last decrement (when the counter goes from 1 to 0) AND releasing the locality
has to be done atomically:

	mutex_lock(&priv->locality_count_mutex);
	priv->locality_count--;
	if (priv->locality_count == 0)
		tpm_tis_release_locality_locked(priv, l);
	mutex_unlock(&priv->locality_count_mutex);

I dont know if this is possible with RCU, especially since AFAIK RCU works with pointers instead of
integral data types.


Regards,
Lino





>>  	int locality;
>>  	int irq;
>>  	unsigned int int_mask;
>> --
>> 2.36.1
>>
>
> BR, Jarkko

Lino Sanfilippo Oct. 25, 2022, 12:25 a.m. UTC | #6

On 23.10.22 07:26, Jarkko Sakkinen wrote:
> On Tue, Oct 18, 2022 at 08:25:08AM +0200, Lukas Wunner wrote:
>> On Tue, Oct 18, 2022 at 01:57:29AM +0200, Lino Sanfilippo wrote:
>>> Implement a usage counter for the (default) locality used by the TPM TIS
>>> driver:
>>> Request the locality from the TPM if it has not been claimed yet, otherwise
>>> only increment the counter. Also release the locality if the counter is 0
>>> otherwise only decrement the counter. Ensure thread-safety by protecting
>>> the counter with a mutex.
>>>
>>> This allows to request and release the locality from a thread and the
>>> interrupt handler at the same time without the danger to interfere with
>>> each other.
>> [...]
>>> +static int tpm_tis_release_locality(struct tpm_chip *chip, int l)
>>>  {
>>>  	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
>>>
>>> -	tpm_tis_write8(priv, TPM_ACCESS(l), TPM_ACCESS_ACTIVE_LOCALITY);
>>> +	mutex_lock(&priv->locality_count_mutex);
>>> +	priv->locality_count--;
>>> +	if (priv->locality_count == 0)
>>> +		tpm_tis_release_locality_locked(priv, l);
>>> +	mutex_unlock(&priv->locality_count_mutex);
>>>
>>>  	return 0;
>>>  }
>>
>> Hm, any reason not to use struct kref for the locality counter?
>> Provides correct memory ordering (no mutex needed) and allows for
>> calling a release function too upon reaching 0.
>
> I proposed for last version kref. I have no idea why this is still
> using mutex. And now I apparently have proposed rcu for the whole
> struct (forgot what I had put my feedback for earlier version).
>
> This keeps being confusing patch as the commit message does not
> really go to the bottom line why mutex is really the best possible
> choice here.
>


I actually tried to implement this via kref but then came to the
conclusion it is rather not a good choice for our case. Please
see my response to your former request to implement this via kref:

https://lore.kernel.org/all/[e-mail]/

Regards,
Lino

Jarkko Sakkinen Nov. 1, 2022, 1:06 a.m. UTC | #7

On Tue, Oct 25, 2022 at 02:15:51AM +0200, Lino Sanfilippo wrote:
> 
> On 23.10.22 06:40, Jarkko Sakkinen wrote:
> > On Tue, Oct 18, 2022 at 01:57:29AM +0200, Lino Sanfilippo wrote:
> >> From: Lino Sanfilippo <[e-mail]>
> >>
> >> Implement a usage counter for the (default) locality used by the TPM TIS
> >> driver:
> >> Request the locality from the TPM if it has not been claimed yet, otherwise
> >> only increment the counter. Also release the locality if the counter is 0
> >> otherwise only decrement the counter. Ensure thread-safety by protecting
> >> the counter with a mutex.
> >>
> >> This allows to request and release the locality from a thread and the
> >> interrupt handler at the same time without the danger to interfere with
> >> each other.
> >>
> >> By doing this refactor the names of the amended functions to use the proper
> >> prefix.
> >>
> >> Signed-off-by: Lino Sanfilippo <[e-mail]>
> >> Tested-by: Michael Niewöhner <[e-mail]>
> >> ---
> >>  drivers/char/tpm/tpm_tis_core.c | 75 ++++++++++++++++++++++-----------
> >>  drivers/char/tpm/tpm_tis_core.h |  2 +
> >>  2 files changed, 53 insertions(+), 24 deletions(-)
> >>
> >> diff --git a/drivers/char/tpm/tpm_tis_core.c b/drivers/char/tpm/tpm_tis_core.c
> >> index 4336f7ea8c2b..79dfab65976f 100644
> >> --- a/drivers/char/tpm/tpm_tis_core.c
> >> +++ b/drivers/char/tpm/tpm_tis_core.c
> >> @@ -165,16 +165,27 @@ static bool check_locality(struct tpm_chip *chip, int l)
> >>  	return false;
> >>  }
> >>
> >> -static int release_locality(struct tpm_chip *chip, int l)
> >> +static int tpm_tis_release_locality_locked(struct tpm_tis_data *priv, int l)
> >
> > Nit: usually you would actually use "unlocked" here, not locked.
> >
> > Probably best name would be __tpm_tis_release_locality().
> 
> 
> Agreed. This is also consistent with the naming scheme used for many other
> kernel functions.
> 
> >
> >> +{
> >> +	tpm_tis_write8(priv, TPM_ACCESS(l), TPM_ACCESS_ACTIVE_LOCALITY);
> >> +
> >> +	return 0;
> >> +}
> >> +
> >> +static int tpm_tis_release_locality(struct tpm_chip *chip, int l)
> >>  {
> >>  	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
> >>
> >> -	tpm_tis_write8(priv, TPM_ACCESS(l), TPM_ACCESS_ACTIVE_LOCALITY);
> >> +	mutex_lock(&priv->locality_count_mutex);
> >> +	priv->locality_count--;
> >> +	if (priv->locality_count == 0)
> >> +		tpm_tis_release_locality_locked(priv, l);
> >> +	mutex_unlock(&priv->locality_count_mutex);
> >>
> >>  	return 0;
> >>  }
> >
> > Since the function pointer has the word "relinquish" and not "release",
> > perhaps these should also use that word for consistency.
> >
> >>
> >> -static int request_locality(struct tpm_chip *chip, int l)
> >> +static int tpm_tis_request_locality_locked(struct tpm_chip *chip, int l)
> >>  {
> >>  	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
> >>  	unsigned long stop, timeout;
> >> @@ -215,6 +226,20 @@ static int request_locality(struct tpm_chip *chip, int l)
> >>  	return -1;
> >>  }
> >>
> >> +static int tpm_tis_request_locality(struct tpm_chip *chip, int l)
> >> +{
> >> +	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
> >> +	int ret = 0;
> >> +
> >> +	mutex_lock(&priv->locality_count_mutex);
> >> +	if (priv->locality_count == 0)
> >> +		ret = tpm_tis_request_locality_locked(chip, l);
> >> +	if (!ret)
> >> +		priv->locality_count++;
> >> +	mutex_unlock(&priv->locality_count_mutex);
> >> +	return ret;
> >> +}
> >> +
> >>  static u8 tpm_tis_status(struct tpm_chip *chip)
> >>  {
> >>  	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
> >> @@ -682,7 +707,7 @@ static int probe_itpm(struct tpm_chip *chip)
> >>  	if (vendor != TPM_VID_INTEL)
> >>  		return 0;
> >>
> >> -	if (request_locality(chip, 0) != 0)
> >> +	if (tpm_tis_request_locality(chip, 0) != 0)
> >>  		return -EBUSY;
> >>
> >>  	rc = tpm_tis_send_data(chip, cmd_getticks, len);
> >> @@ -703,7 +728,7 @@ static int probe_itpm(struct tpm_chip *chip)
> >>
> >>  out:
> >>  	tpm_tis_ready(chip);
> >> -	release_locality(chip, priv->locality);
> >> +	tpm_tis_release_locality(chip, priv->locality);
> >>
> >>  	return rc;
> >>  }
> >> @@ -762,7 +787,7 @@ static int tpm_tis_gen_interrupt(struct tpm_chip *chip)
> >>  	cap_t cap;
> >>  	int ret;
> >>
> >> -	ret = request_locality(chip, 0);
> >> +	ret = tpm_tis_request_locality(chip, 0);
> >>  	if (ret < 0)
> >>  		return ret;
> >>
> >> @@ -771,7 +796,7 @@ static int tpm_tis_gen_interrupt(struct tpm_chip *chip)
> >>  	else
> >>  		ret = tpm1_getcap(chip, TPM_CAP_PROP_TIS_TIMEOUT, &cap, desc, 0);
> >>
> >> -	release_locality(chip, 0);
> >> +	tpm_tis_release_locality(chip, 0);
> >>
> >>  	return ret;
> >>  }
> >> @@ -796,33 +821,33 @@ static int tpm_tis_probe_irq_single(struct tpm_chip *chip, u32 intmask,
> >>  	}
> >>  	priv->irq = irq;
> >>
> >> -	rc = request_locality(chip, 0);
> >> +	rc = tpm_tis_request_locality(chip, 0);
> >>  	if (rc < 0)
> >>  		return rc;
> >>
> >>  	rc = tpm_tis_read8(priv, TPM_INT_VECTOR(priv->locality),
> >>  			   &original_int_vec);
> >>  	if (rc < 0) {
> >> -		release_locality(chip, priv->locality);
> >> +		tpm_tis_release_locality(chip, priv->locality);
> >>  		return rc;
> >>  	}
> >>
> >>  	rc = tpm_tis_write8(priv, TPM_INT_VECTOR(priv->locality), irq);
> >>  	if (rc < 0) {
> >> -		release_locality(chip, priv->locality);
> >> +		tpm_tis_release_locality(chip, priv->locality);
> >>  		return rc;
> >>  	}
> >>
> >>  	rc = tpm_tis_read32(priv, TPM_INT_STATUS(priv->locality), &int_status);
> >>  	if (rc < 0) {
> >> -		release_locality(chip, priv->locality);
> >> +		tpm_tis_release_locality(chip, priv->locality);
> >>  		return rc;
> >>  	}
> >>
> >>  	/* Clear all existing */
> >>  	rc = tpm_tis_write32(priv, TPM_INT_STATUS(priv->locality), int_status);
> >>  	if (rc < 0) {
> >> -		release_locality(chip, priv->locality);
> >> +		tpm_tis_release_locality(chip, priv->locality);
> >>  		return rc;
> >>  	}
> >>
> >> @@ -830,11 +855,11 @@ static int tpm_tis_probe_irq_single(struct tpm_chip *chip, u32 intmask,
> >>  	rc = tpm_tis_write32(priv, TPM_INT_ENABLE(priv->locality),
> >>  			     intmask | TPM_GLOBAL_INT_ENABLE);
> >>  	if (rc < 0) {
> >> -		release_locality(chip, priv->locality);
> >> +		tpm_tis_release_locality(chip, priv->locality);
> >>  		return rc;
> >>  	}
> >>
> >> -	release_locality(chip, priv->locality);
> >> +	tpm_tis_release_locality(chip, priv->locality);
> >>  	clear_bit(TPM_TIS_IRQ_TESTED, &priv->flags);
> >>
> >>  	/* Generate an interrupt by having the core call through to
> >> @@ -970,8 +995,8 @@ static const struct tpm_class_ops tpm_tis = {
> >>  	.req_complete_mask = TPM_STS_DATA_AVAIL | TPM_STS_VALID,
> >>  	.req_complete_val = TPM_STS_DATA_AVAIL | TPM_STS_VALID,
> >>  	.req_canceled = tpm_tis_req_canceled,
> >> -	.request_locality = request_locality,
> >> -	.relinquish_locality = release_locality,
> >> +	.request_locality = tpm_tis_request_locality,
> >> +	.relinquish_locality = tpm_tis_release_locality,
> >>  	.clk_enable = tpm_tis_clkrun_enable,
> >>  };
> >>
> >> @@ -1005,6 +1030,8 @@ int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int irq,
> >>  	priv->timeout_min = TPM_TIMEOUT_USECS_MIN;
> >>  	priv->timeout_max = TPM_TIMEOUT_USECS_MAX;
> >>  	priv->phy_ops = phy_ops;
> >> +	priv->locality_count = 0;
> >> +	mutex_init(&priv->locality_count_mutex);
> >>
> >>  	dev_set_drvdata(&chip->dev, priv);
> >>
> >> @@ -1083,14 +1110,14 @@ int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int irq,
> >>
> >>  	intmask &= ~TPM_GLOBAL_INT_ENABLE;
> >>
> >> -	rc = request_locality(chip, 0);
> >> +	rc = tpm_tis_request_locality(chip, 0);
> >>  	if (rc < 0) {
> >>  		rc = -ENODEV;
> >>  		goto out_err;
> >>  	}
> >>
> >>  	tpm_tis_write32(priv, TPM_INT_ENABLE(priv->locality), intmask);
> >> -	release_locality(chip, 0);
> >> +	tpm_tis_release_locality(chip, 0);
> >>
> >>  	rc = tpm_chip_start(chip);
> >>  	if (rc)
> >> @@ -1124,13 +1151,13 @@ int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int irq,
> >>  		 * proper timeouts for the driver.
> >>  		 */
> >>
> >> -		rc = request_locality(chip, 0);
> >> +		rc = tpm_tis_request_locality(chip, 0);
> >>  		if (rc < 0)
> >>  			goto out_err;
> >>
> >>  		rc = tpm_get_timeouts(chip);
> >>
> >> -		release_locality(chip, 0);
> >> +		tpm_tis_release_locality(chip, 0);
> >>
> >>  		if (rc) {
> >>  			dev_err(dev, "Could not get TPM timeouts and durations\n");
> >> @@ -1150,11 +1177,11 @@ int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int irq,
> >>  			dev_err(&chip->dev, FW_BUG
> >>  					"TPM interrupt not working, polling instead\n");
> >>
> >> -			rc = request_locality(chip, 0);
> >> +			rc = tpm_tis_request_locality(chip, 0);
> >>  			if (rc < 0)
> >>  				goto out_err;
> >>  			disable_interrupts(chip);
> >> -			release_locality(chip, 0);
> >> +			tpm_tis_release_locality(chip, 0);
> >>  		}
> >>  	}
> >>
> >> @@ -1221,13 +1248,13 @@ int tpm_tis_resume(struct device *dev)
> >>  	 * an error code but for unknown reason it isn't handled.
> >>  	 */
> >>  	if (!(chip->flags & TPM_CHIP_FLAG_TPM2)) {
> >> -		ret = request_locality(chip, 0);
> >> +		ret = tpm_tis_request_locality(chip, 0);
> >>  		if (ret < 0)
> >>  			return ret;
> >>
> >>  		tpm1_do_selftest(chip);
> >>
> >> -		release_locality(chip, 0);
> >> +		tpm_tis_release_locality(chip, 0);
> >>  	}
> >>
> >>  	return 0;
> >> diff --git a/drivers/char/tpm/tpm_tis_core.h b/drivers/char/tpm/tpm_tis_core.h
> >> index 2deef11c88db..13bdcf38e56f 100644
> >> --- a/drivers/char/tpm/tpm_tis_core.h
> >> +++ b/drivers/char/tpm/tpm_tis_core.h
> >> @@ -91,6 +91,8 @@ enum tpm_tis_flags {
> >>
> >>  struct tpm_tis_data {
> >>  	u16 manufacturer_id;
> >> +	struct mutex locality_count_mutex;
> >
> > BTW, why mutex and not spinlock?
> >
> > Hmm.. also I think you might have given feedback already on this
> > but could the lock cover the whole struct instead of a counter?
> > I tried to dig lore for earlier response but could not find. I'm
> > sorry if I'm asking the same question again.
> >
> 
> Actually thats on me, since it took me much too long to send the v8 after the v7 review.
> 
> However the reason that we need a mutex here is that we not only increase or decrease
> the locality_counter under the mutex, but also do the locality request and release by
> writing to the ACCESS register. Since in the SPI case each communication over the spi bus
> is protected by the bus_lock_mutex of the SPI device we must not hold a spinlock when doing
> the register accesses.
> 
> Concerning covering the whole tpm_tis_data struct:
> Most structure elements are set once at driver startup but never changed at driver
> runtime. So no locking needed for these. The only exception is "flags" and "locality_count"
> whereby "flags" is accessed by atomic bit manipulating functions and thus
> does not need extra locking. So "locality_count" is AFAICS the only element that needs to be
> protected by the mutex.

OK, but you should should still address this in commit message, e.g.
by mentioning that in the case of SPI bus mutex is required because
the bus itself needs to be locked in the mutex.

I.e. this a claim, definitely not an argument: "Ensure thread-safety by
protecting the counter with a mutex."

BR, Jarkko

Jarkko Sakkinen Nov. 1, 2022, 1:06 a.m. UTC | #8

On Tue, Oct 25, 2022 at 02:25:39AM +0200, Lino Sanfilippo wrote:
> 
> 
> On 23.10.22 07:26, Jarkko Sakkinen wrote:
> > On Tue, Oct 18, 2022 at 08:25:08AM +0200, Lukas Wunner wrote:
> >> On Tue, Oct 18, 2022 at 01:57:29AM +0200, Lino Sanfilippo wrote:
> >>> Implement a usage counter for the (default) locality used by the TPM TIS
> >>> driver:
> >>> Request the locality from the TPM if it has not been claimed yet, otherwise
> >>> only increment the counter. Also release the locality if the counter is 0
> >>> otherwise only decrement the counter. Ensure thread-safety by protecting
> >>> the counter with a mutex.
> >>>
> >>> This allows to request and release the locality from a thread and the
> >>> interrupt handler at the same time without the danger to interfere with
> >>> each other.
> >> [...]
> >>> +static int tpm_tis_release_locality(struct tpm_chip *chip, int l)
> >>>  {
> >>>  	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
> >>>
> >>> -	tpm_tis_write8(priv, TPM_ACCESS(l), TPM_ACCESS_ACTIVE_LOCALITY);
> >>> +	mutex_lock(&priv->locality_count_mutex);
> >>> +	priv->locality_count--;
> >>> +	if (priv->locality_count == 0)
> >>> +		tpm_tis_release_locality_locked(priv, l);
> >>> +	mutex_unlock(&priv->locality_count_mutex);
> >>>
> >>>  	return 0;
> >>>  }
> >>
> >> Hm, any reason not to use struct kref for the locality counter?
> >> Provides correct memory ordering (no mutex needed) and allows for
> >> calling a release function too upon reaching 0.
> >
> > I proposed for last version kref. I have no idea why this is still
> > using mutex. And now I apparently have proposed rcu for the whole
> > struct (forgot what I had put my feedback for earlier version).
> >
> > This keeps being confusing patch as the commit message does not
> > really go to the bottom line why mutex is really the best possible
> > choice here.
> >
> 
> 
> I actually tried to implement this via kref but then came to the
> conclusion it is rather not a good choice for our case. Please
> see my response to your former request to implement this via kref:
> 
> https://lore.kernel.org/all/[e-mail]/

OK, my bad I missed this, sorry.

BR, Jarkko

Lino Sanfilippo Nov. 4, 2022, 4:18 p.m. UTC | #9

On 01.11.22 02:06, Jarkko Sakkinen wrote:
> On Tue, Oct 25, 2022 at 02:15:51AM +0200, Lino Sanfilippo wrote:

>> Actually thats on me, since it took me much too long to send the v8 after the v7 review.
>>
>> However the reason that we need a mutex here is that we not only increase or decrease
>> the locality_counter under the mutex, but also do the locality request and release by
>> writing to the ACCESS register. Since in the SPI case each communication over the spi bus
>> is protected by the bus_lock_mutex of the SPI device we must not hold a spinlock when doing
>> the register accesses.
>>
>> Concerning covering the whole tpm_tis_data struct:
>> Most structure elements are set once at driver startup but never changed at driver
>> runtime. So no locking needed for these. The only exception is "flags" and "locality_count"
>> whereby "flags" is accessed by atomic bit manipulating functions and thus
>> does not need extra locking. So "locality_count" is AFAICS the only element that needs to be
>> protected by the mutex.
> 
> OK, but you should should still address this in commit message, e.g.
> by mentioning that in the case of SPI bus mutex is required because
> the bus itself needs to be locked in the mutex.
> 
> I.e. this a claim, definitely not an argument: "Ensure thread-safety by
> protecting the counter with a mutex."
> 

Ok, I will rephrase the commit message accordingly. 
Thanks for the review!

Regards,
Lino

Jarkko Sakkinen Nov. 7, 2022, 4:03 p.m. UTC | #10

On Fri, Nov 04, 2022 at 05:18:21PM +0100, Lino Sanfilippo wrote:
> 
> 
> On 01.11.22 02:06, Jarkko Sakkinen wrote:
> > On Tue, Oct 25, 2022 at 02:15:51AM +0200, Lino Sanfilippo wrote:
> 
> >> Actually thats on me, since it took me much too long to send the v8 after the v7 review.
> >>
> >> However the reason that we need a mutex here is that we not only increase or decrease
> >> the locality_counter under the mutex, but also do the locality request and release by
> >> writing to the ACCESS register. Since in the SPI case each communication over the spi bus
> >> is protected by the bus_lock_mutex of the SPI device we must not hold a spinlock when doing
> >> the register accesses.
> >>
> >> Concerning covering the whole tpm_tis_data struct:
> >> Most structure elements are set once at driver startup but never changed at driver
> >> runtime. So no locking needed for these. The only exception is "flags" and "locality_count"
> >> whereby "flags" is accessed by atomic bit manipulating functions and thus
> >> does not need extra locking. So "locality_count" is AFAICS the only element that needs to be
> >> protected by the mutex.
> > 
> > OK, but you should should still address this in commit message, e.g.
> > by mentioning that in the case of SPI bus mutex is required because
> > the bus itself needs to be locked in the mutex.
> > 
> > I.e. this a claim, definitely not an argument: "Ensure thread-safety by
> > protecting the counter with a mutex."
> > 
> 
> Ok, I will rephrase the commit message accordingly. 
> Thanks for the review!

Yeah, np. I.e. I understand your reasoning but it is easy to intuitively
think it as not the right solution. Thus, it deserves a remark, right?
:-)

BR, Jarkko

Patch

diff --git a/drivers/char/tpm/tpm_tis_core.c b/drivers/char/tpm/tpm_tis_core.c
index 4336f7ea8c2b..79dfab65976f 100644
--- a/drivers/char/tpm/tpm_tis_core.c
+++ b/drivers/char/tpm/tpm_tis_core.c
@@ -165,16 +165,27 @@  static bool check_locality(struct tpm_chip *chip, int l)
 	return false;
 }
 
-static int release_locality(struct tpm_chip *chip, int l)
+static int tpm_tis_release_locality_locked(struct tpm_tis_data *priv, int l)
+{
+	tpm_tis_write8(priv, TPM_ACCESS(l), TPM_ACCESS_ACTIVE_LOCALITY);
+
+	return 0;
+}
+
+static int tpm_tis_release_locality(struct tpm_chip *chip, int l)
 {
 	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
 
-	tpm_tis_write8(priv, TPM_ACCESS(l), TPM_ACCESS_ACTIVE_LOCALITY);
+	mutex_lock(&priv->locality_count_mutex);
+	priv->locality_count--;
+	if (priv->locality_count == 0)
+		tpm_tis_release_locality_locked(priv, l);
+	mutex_unlock(&priv->locality_count_mutex);
 
 	return 0;
 }
 
-static int request_locality(struct tpm_chip *chip, int l)
+static int tpm_tis_request_locality_locked(struct tpm_chip *chip, int l)
 {
 	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
 	unsigned long stop, timeout;
@@ -215,6 +226,20 @@  static int request_locality(struct tpm_chip *chip, int l)
 	return -1;
 }
 
+static int tpm_tis_request_locality(struct tpm_chip *chip, int l)
+{
+	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
+	int ret = 0;
+
+	mutex_lock(&priv->locality_count_mutex);
+	if (priv->locality_count == 0)
+		ret = tpm_tis_request_locality_locked(chip, l);
+	if (!ret)
+		priv->locality_count++;
+	mutex_unlock(&priv->locality_count_mutex);
+	return ret;
+}
+
 static u8 tpm_tis_status(struct tpm_chip *chip)
 {
 	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
@@ -682,7 +707,7 @@  static int probe_itpm(struct tpm_chip *chip)
 	if (vendor != TPM_VID_INTEL)
 		return 0;
 
-	if (request_locality(chip, 0) != 0)
+	if (tpm_tis_request_locality(chip, 0) != 0)
 		return -EBUSY;
 
 	rc = tpm_tis_send_data(chip, cmd_getticks, len);
@@ -703,7 +728,7 @@  static int probe_itpm(struct tpm_chip *chip)
 
 out:
 	tpm_tis_ready(chip);
-	release_locality(chip, priv->locality);
+	tpm_tis_release_locality(chip, priv->locality);
 
 	return rc;
 }
@@ -762,7 +787,7 @@  static int tpm_tis_gen_interrupt(struct tpm_chip *chip)
 	cap_t cap;
 	int ret;
 
-	ret = request_locality(chip, 0);
+	ret = tpm_tis_request_locality(chip, 0);
 	if (ret < 0)
 		return ret;
 
@@ -771,7 +796,7 @@  static int tpm_tis_gen_interrupt(struct tpm_chip *chip)
 	else
 		ret = tpm1_getcap(chip, TPM_CAP_PROP_TIS_TIMEOUT, &cap, desc, 0);
 
-	release_locality(chip, 0);
+	tpm_tis_release_locality(chip, 0);
 
 	return ret;
 }
@@ -796,33 +821,33 @@  static int tpm_tis_probe_irq_single(struct tpm_chip *chip, u32 intmask,
 	}
 	priv->irq = irq;
 
-	rc = request_locality(chip, 0);
+	rc = tpm_tis_request_locality(chip, 0);
 	if (rc < 0)
 		return rc;
 
 	rc = tpm_tis_read8(priv, TPM_INT_VECTOR(priv->locality),
 			   &original_int_vec);
 	if (rc < 0) {
-		release_locality(chip, priv->locality);
+		tpm_tis_release_locality(chip, priv->locality);
 		return rc;
 	}
 
 	rc = tpm_tis_write8(priv, TPM_INT_VECTOR(priv->locality), irq);
 	if (rc < 0) {
-		release_locality(chip, priv->locality);
+		tpm_tis_release_locality(chip, priv->locality);
 		return rc;
 	}
 
 	rc = tpm_tis_read32(priv, TPM_INT_STATUS(priv->locality), &int_status);
 	if (rc < 0) {
-		release_locality(chip, priv->locality);
+		tpm_tis_release_locality(chip, priv->locality);
 		return rc;
 	}
 
 	/* Clear all existing */
 	rc = tpm_tis_write32(priv, TPM_INT_STATUS(priv->locality), int_status);
 	if (rc < 0) {
-		release_locality(chip, priv->locality);
+		tpm_tis_release_locality(chip, priv->locality);
 		return rc;
 	}
 
@@ -830,11 +855,11 @@  static int tpm_tis_probe_irq_single(struct tpm_chip *chip, u32 intmask,
 	rc = tpm_tis_write32(priv, TPM_INT_ENABLE(priv->locality),
 			     intmask | TPM_GLOBAL_INT_ENABLE);
 	if (rc < 0) {
-		release_locality(chip, priv->locality);
+		tpm_tis_release_locality(chip, priv->locality);
 		return rc;
 	}
 
-	release_locality(chip, priv->locality);
+	tpm_tis_release_locality(chip, priv->locality);
 	clear_bit(TPM_TIS_IRQ_TESTED, &priv->flags);
 
 	/* Generate an interrupt by having the core call through to
@@ -970,8 +995,8 @@  static const struct tpm_class_ops tpm_tis = {
 	.req_complete_mask = TPM_STS_DATA_AVAIL | TPM_STS_VALID,
 	.req_complete_val = TPM_STS_DATA_AVAIL | TPM_STS_VALID,
 	.req_canceled = tpm_tis_req_canceled,
-	.request_locality = request_locality,
-	.relinquish_locality = release_locality,
+	.request_locality = tpm_tis_request_locality,
+	.relinquish_locality = tpm_tis_release_locality,
 	.clk_enable = tpm_tis_clkrun_enable,
 };
 
@@ -1005,6 +1030,8 @@  int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int irq,
 	priv->timeout_min = TPM_TIMEOUT_USECS_MIN;
 	priv->timeout_max = TPM_TIMEOUT_USECS_MAX;
 	priv->phy_ops = phy_ops;
+	priv->locality_count = 0;
+	mutex_init(&priv->locality_count_mutex);
 
 	dev_set_drvdata(&chip->dev, priv);
 
@@ -1083,14 +1110,14 @@  int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int irq,
 
 	intmask &= ~TPM_GLOBAL_INT_ENABLE;
 
-	rc = request_locality(chip, 0);
+	rc = tpm_tis_request_locality(chip, 0);
 	if (rc < 0) {
 		rc = -ENODEV;
 		goto out_err;
 	}
 
 	tpm_tis_write32(priv, TPM_INT_ENABLE(priv->locality), intmask);
-	release_locality(chip, 0);
+	tpm_tis_release_locality(chip, 0);
 
 	rc = tpm_chip_start(chip);
 	if (rc)
@@ -1124,13 +1151,13 @@  int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int irq,
 		 * proper timeouts for the driver.
 		 */
 
-		rc = request_locality(chip, 0);
+		rc = tpm_tis_request_locality(chip, 0);
 		if (rc < 0)
 			goto out_err;
 
 		rc = tpm_get_timeouts(chip);
 
-		release_locality(chip, 0);
+		tpm_tis_release_locality(chip, 0);
 
 		if (rc) {
 			dev_err(dev, "Could not get TPM timeouts and durations\n");
@@ -1150,11 +1177,11 @@  int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int irq,
 			dev_err(&chip->dev, FW_BUG
 					"TPM interrupt not working, polling instead\n");
 
-			rc = request_locality(chip, 0);
+			rc = tpm_tis_request_locality(chip, 0);
 			if (rc < 0)
 				goto out_err;
 			disable_interrupts(chip);
-			release_locality(chip, 0);
+			tpm_tis_release_locality(chip, 0);
 		}
 	}
 
@@ -1221,13 +1248,13 @@  int tpm_tis_resume(struct device *dev)
 	 * an error code but for unknown reason it isn't handled.
 	 */
 	if (!(chip->flags & TPM_CHIP_FLAG_TPM2)) {
-		ret = request_locality(chip, 0);
+		ret = tpm_tis_request_locality(chip, 0);
 		if (ret < 0)
 			return ret;
 
 		tpm1_do_selftest(chip);
 
-		release_locality(chip, 0);
+		tpm_tis_release_locality(chip, 0);
 	}
 
 	return 0;
diff --git a/drivers/char/tpm/tpm_tis_core.h b/drivers/char/tpm/tpm_tis_core.h
index 2deef11c88db..13bdcf38e56f 100644
--- a/drivers/char/tpm/tpm_tis_core.h
+++ b/drivers/char/tpm/tpm_tis_core.h
@@ -91,6 +91,8 @@  enum tpm_tis_flags {
 
 struct tpm_tis_data {
 	u16 manufacturer_id;
+	struct mutex locality_count_mutex;
+	unsigned int locality_count;
 	int locality;
 	int irq;
 	unsigned int int_mask;