Re: [PATCH] tpm: fix potential race condition in suspend/resume

[Jarkko Sakkinen]

On Tue, Aug 09, 2022 at 07:39:18PM +0000, Sven van Ashbrook wrote:
> Concurrent accesses to the tpm chip are prevented by allowing only a
> single thread at a time to obtain a tpm chip reference through
> tpm_try_get_ops(). However, the tpm's suspend function does not use
> this mechanism, so when the tpm api is called by a kthread which
> does not get frozen on suspend (such as the hw_random kthread)
> it's possible that the tpm is used when already in suspend, or
> in use while in the process of suspending.
> 
> This is seen on certain ChromeOS platforms - low-probability warnings
> are generated during suspend. In this case, the tpm attempted to read data
> from a tpm chip on an already-suspended bus.
> 
>   i2c_designware i2c_designware.1: Transfer while suspended
> 
> Fix:
> 1. prevent concurrent execution of tpm accesses and suspend/
>    resume, by letting suspend/resume grab the tpm_mutex.
> 2. before commencing a tpm access, check if the tpm chip is already
>    suspended. Fail with -EAGAIN if so.
> 
> Tested by running 6000 suspend/resume cycles back-to-back on a
> ChromeOS "brya" device. The intermittent warnings reliably
> disappear after applying this patch. No system issues were observed.
> 
> Cc: <stable@xxxxxxxxxxxxxxx>
> Fixes: e891db1a18bf ("tpm: turn on TPM on suspend for TPM 1.x")
> Signed-off-by: Sven van Ashbrook <svenva@xxxxxxxxxxxx>
> ---
>  drivers/char/tpm/tpm-interface.c | 16 ++++++++++++++++
>  include/linux/tpm.h              |  2 ++
>  2 files changed, 18 insertions(+)
> 
> diff --git a/drivers/char/tpm/tpm-interface.c b/drivers/char/tpm/tpm-interface.c
> index 1621ce818705..16ca490fd483 100644
> --- a/drivers/char/tpm/tpm-interface.c
> +++ b/drivers/char/tpm/tpm-interface.c
> @@ -82,6 +82,11 @@ static ssize_t tpm_try_transmit(struct tpm_chip *chip, void *buf, size_t bufsiz)
>  		return -E2BIG;
>  	}
>  
> +	if (chip->is_suspended) {
> +		dev_info(&chip->dev, "blocking transmit while suspended\n");
> +		return -EAGAIN;
> +	}
> +
>  	rc = chip->ops->send(chip, buf, count);
>  	if (rc < 0) {
>  		if (rc != -EPIPE)
> @@ -394,6 +399,8 @@ int tpm_pm_suspend(struct device *dev)
>  	if (!chip)
>  		return -ENODEV;
>  
> +	mutex_lock(&chip->tpm_mutex);
> +
>  	if (chip->flags & TPM_CHIP_FLAG_ALWAYS_POWERED)
>  		goto suspended;
>  
> @@ -411,6 +418,11 @@ int tpm_pm_suspend(struct device *dev)
>  	}
>  
>  suspended:
> +	if (!rc)
> +		chip->is_suspended = true;
> +
> +	mutex_unlock(&chip->tpm_mutex);
> +
>  	return rc;
>  }
>  EXPORT_SYMBOL_GPL(tpm_pm_suspend);
> @@ -426,6 +438,10 @@ int tpm_pm_resume(struct device *dev)
>  	if (chip == NULL)
>  		return -ENODEV;
>  
> +	mutex_lock(&chip->tpm_mutex);
> +	chip->is_suspended = false;
> +	mutex_unlock(&chip->tpm_mutex);
> +
>  	return 0;
>  }
>  EXPORT_SYMBOL_GPL(tpm_pm_resume);
> diff --git a/include/linux/tpm.h b/include/linux/tpm.h
> index d7c67581929f..0fbc1a43ae80 100644
> --- a/include/linux/tpm.h
> +++ b/include/linux/tpm.h
> @@ -131,6 +131,8 @@ struct tpm_chip {
>  	int dev_num;		/* /dev/tpm# */
>  	unsigned long is_open;	/* only one allowed */
>  
> +	bool is_suspended;
> +
>  	char hwrng_name[64];
>  	struct hwrng hwrng;
>  
> -- 
> 2.37.1.559.g78731f0fdb-goog
> 

What about adding TPM_CHIP_FLAG_SUSPENDED instead?

BR, Jarkko