Re: [PATCH v2 1/4] tpm: ignore burstcount to improve tpm_tis send() performance.

[Nayna Jain <nayna@linux.vnet.ibm.com>]

On 09/14/2017 04:40 AM, Jarkko Sakkinen wrote:
> On Wed, Sep 13, 2017 at 11:39:03AM -0700, Peter Huewe wrote:
>>
>> Am 12. September 2017 17:45:08 GMT-07:00 schrieb Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>:
>>> On Wed, Sep 06, 2017 at 08:56:36AM -0400, Nayna Jain wrote:
>>>> The TPM burstcount status indicates the number of bytes that can
>>>> be sent to the TPM without causing bus wait states.  Effectively,
>>>> it is the number of empty bytes in the command FIFO. Further,
>>>> some TPMs have a static burstcount, when the value remains zero
>>>> until the entire FIFO is empty.
>>>>
>>>> This patch adds an optimization to check for burstcount only once.
>>>> And if it is valid, it writes all the bytes at once, permitting
>>>> wait states. The performance of a 34 byte extend on a TPM 1.2 with
>>>> an 8 byte burstcount improved from 41 msec to 14 msec.
>>>>
>>>> This functionality is enabled only by passing module
>>>> parameter ignore_burst_count=1. By default, this parameter
>>>> is disabled.
>>>>
>>>> After this change, performance on a TPM 1.2 with an 8 byte
>>>> burstcount for 1000 extends improved from ~41sec to ~14sec.
>>>>
>>>> Suggested-by: Ken Goldman <kgold@linux.vnet.ibm.com> in
>>>> conjunction with the TPM Device Driver work group.
>>>> Signed-off-by: Nayna Jain <nayna@linux.vnet.ibm.com>
>>>> Acked-by: Mimi Zohar <zohar@linux.vnet.ibm.com>
>>>> ---
>>>>   Documentation/admin-guide/kernel-parameters.txt |  8 ++++++++
>>>>   drivers/char/tpm/tpm_tis_core.c                 | 24
>>> +++++++++++++++++++++---
>>>>   2 files changed, 29 insertions(+), 3 deletions(-)
>>>>
>>>> diff --git a/Documentation/admin-guide/kernel-parameters.txt
>>> b/Documentation/admin-guide/kernel-parameters.txt
>>>> index 4e303be83df6..3c59bb91e1ee 100644
>>>> --- a/Documentation/admin-guide/kernel-parameters.txt
>>>> +++ b/Documentation/admin-guide/kernel-parameters.txt
>>>> @@ -1465,6 +1465,14 @@
>>>>   			mode generally follows that for the NaN encoding,
>>>>   			except where unsupported by hardware.
>>>>   
>>>> +	ignore_burst_count [TPM_TIS_CORE]
>>>> +			tpm_tis_core driver queries for the burstcount before
>>>> +			every send call in a loop. However, it causes delay to
>>>> +			the send command for TPMs with low burstcount value.
>>>> +			Setting this value to 1, will make driver to query for
>>>> +			burstcount only once in the loop to improve the
>>>> +			performance. By default, its value is set to 0.
>>>> +
>>>>   	ignore_loglevel	[KNL]
>>>>   			Ignore loglevel setting - this will print /all/
>>>>   			kernel messages to the console. Useful for debugging.
>>>> diff --git a/drivers/char/tpm/tpm_tis_core.c
>>> b/drivers/char/tpm/tpm_tis_core.c
>>>> index 63bc6c3b949e..6b9bf4c4d434 100644
>>>> --- a/drivers/char/tpm/tpm_tis_core.c
>>>> +++ b/drivers/char/tpm/tpm_tis_core.c
>>>> @@ -31,6 +31,11 @@
>>>>   #include "tpm.h"
>>>>   #include "tpm_tis_core.h"
>>>>   
>>>> +static bool ignore_burst_count = false;
>>>> +module_param(ignore_burst_count, bool, 0444);
>>>> +MODULE_PARM_DESC(ignore_burst_count,
>>>> +		"Ignore burstcount value while writing data");
>>>> +
>>>>   /* Before we attempt to access the TPM we must see that the valid
>>> bit is set.
>>>>    * The specification says that this bit is 0 at reset and remains 0
>>> until the
>>>>    * 'TPM has gone through its self test and initialization and has
>>> established
>>>> @@ -256,6 +261,7 @@ static int tpm_tis_send_data(struct tpm_chip
>>> *chip, u8 *buf, size_t len)
>>>>   {
>>>>   	struct tpm_tis_data *priv = dev_get_drvdata(&chip->dev);
>>>>   	int rc, status, burstcnt;
>>>> +	int sendcnt;
>>>>   	size_t count = 0;
>>>>   	bool itpm = priv->flags & TPM_TIS_ITPM_WORKAROUND;
>>>>   
>>>> @@ -271,19 +277,31 @@ static int tpm_tis_send_data(struct tpm_chip
>>> *chip, u8 *buf, size_t len)
>>>>   	}
>>>>   
>>>>   	while (count < len - 1) {
>>>> +
>>>> +		/*
>>>> +		 * Get the initial burstcount to ensure TPM is ready to
>>>> +		 * accept data, even when waiting for burstcount is disabled.
>>>> +		 */
>>>>   		burstcnt = get_burstcount(chip);
>>>>   		if (burstcnt < 0) {
>>>>   			dev_err(&chip->dev, "Unable to read burstcount\n");
>>>>   			rc = burstcnt;
>>>>   			goto out_err;
>>>>   		}
>>>> -		burstcnt = min_t(int, burstcnt, len - count - 1);
>>>> +
>>>> +		if (ignore_burst_count)
>>>> +			sendcnt = len - 1;
>>>> +		else
>>>> +			sendcnt = min_t(int, burstcnt, len - count - 1);
>>>> +
>>>>   		rc = tpm_tis_write_bytes(priv, TPM_DATA_FIFO(priv->locality),
>>>> -					 burstcnt, buf + count);
>>>> +					 sendcnt, buf + count);
>>>>   		if (rc < 0)
>>>>   			goto out_err;
>>>>   
>>>> -		count += burstcnt;
>>>> +		count += sendcnt;
>>>> +		if (ignore_burst_count)
>>>> +			continue;
>>>>   
>>>>   		if (wait_for_tpm_stat(chip, TPM_STS_VALID, chip->timeout_c,
>>>>   					&priv->int_queue, false) < 0) {
>>>> -- 
>>>> 2.13.3
>>>>
>>> Makes sense to discuss whether to have the kernel command-line
>>> parameter or not before applying this.
>>>
>>> To fuel the discussion, alternative to this would be:
>>>
>>> 1. Have this always on i.e. no command-line parameter.
>>> 2. If someone yells, we add the command-line parameter later on.
>>>
>> According to what I've read in the tcg ddwg group this patch should
>> not cause problems on _sane_ tpms.
>>
>> I'm not 100%convinced that all tpms are sane all the time, but I think
>> we do not want yet another cmdline parameter.
>>
>> So if we want to pull it in (and ddwg does not see an issue, so yes)
>> it should be on by default, without a kernel parameter.
>>
>> If there is a kernel parameter, then it should only be one called
>> "failsafe" - which includes the force behavior and maybe the "broken"
>> tpm path.
>>
>> But I agree with Alex, every additonal code path reduces testing coverage.
>>
>>
>> We would be happy to test a "default on" patch.
>>
>> Peter
>>
>>> /Jarkko
> I'm starting to dilate to this direction.
>
> It is hard to believe that any such TPM would be in active use anywhere
> assuming that there exist a TPM where this causes issues. This combined
> to the assumption that you would run the latest mainline on it makes it
> a pretty insignificant scenario.

It sounds like we are getting in direction to have this change by default.
Before removing the ignore_burst_count parameter, I will post a test version of this
patch which enables ignore_burst_count by default, for testing purposes only.

Thanks Peter and Alex for testing.

Thanks & Regards,
    - Nayna


>
> /Jarkko
>