Re: [PATCH] vTPM: Fix Atmel timeout bug.

[Daniel De Graaf <dgdegra@tycho.nsa.gov>]

On 11/10/2014 07:01 AM, Ian Campbell wrote:
> On Thu, 2014-11-06 at 17:01 -0500, Daniel De Graaf wrote:
>> On 11/04/2014 05:15 AM, Ian Campbell wrote:
>>> On Thu, 2014-10-30 at 15:48 +0200, Emil Condrea wrote:
>>>> Of course we can use max, but I thought that it might be useful to
>>>> have a prink to inform the user that the timeout was adjusted.
>>>> In init_tpm_tis the default timeouts are set using:
>>>> /* Set default timeouts */ tpm->timeout_a =
>>>> MILLISECS(TIS_SHORT_TIMEOUT);//750*1000000UL tpm->timeout_b =
>>>> MILLISECS(TIS_LONG_TIMEOUT);//2000*1000000UL tpm->timeout_c =
>>>> MILLISECS(TIS_SHORT_TIMEOUT); tpm->timeout_d =
>>>> MILLISECS(TIS_SHORT_TIMEOUT);
>>>>
>>>>
>>>>
>>>> But in kernel fix they are set as 750*1000 instead of 750*1000000UL :
>>>> https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/drivers/char/tpm/tpm_tis.c#n381
>>>> So if we want to integrate kernel changes I think we should use
>>>> MICROSECS(TIS_SHORT_TIMEOUT) which is 750000
>>>> Also in kernel the default timeouts are initialized using
>>>> msecs_to_jiffies which is different from MILLISECS
>>>> macro.: https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/drivers/char/tpm/tpm_tis.c#n548
>>>> Is there a certain reason for not using msecs_to_jiffies ?
>>>
>>> jiffies are a Linux specific concept which mini-os doesn't share.
>>>
>>> Daniel, do you have any opinion on this patch?
>>>
>>> It seems like the Linux fix is made only for the specifically broken
>>> platform. That seems to make sense to me since presumably other systems
>>> report short timeouts which they can indeed cope with. It's only Atmel
>>> which brokenly reports something it cannot handle.
>>>
>>> Ian.
>>
>> I agree that an adjustment is needed when values are too short.  Adjusting
>> in all cases is not quite as nice as only fixing the broken TPMs, but it
>> is a lot simpler.  It also doesn't seem harmful to have the timeouts be
>> too large in the driver: a properly functioning TPM will not time out its
>> requests in any case, so the user won't notice normally, and the default
>> short timeout is 0.75 seconds - very few people will complain if they have
>> to wait that long to get a timeout instead of what their TPM actually uses.
>
> Can we take that as an ack?

Yes.  I was going to check to see if the printks would cause people to see
problems where the timeouts were already reasonable, but since the mini-os
driver is already quite verbose, this should not be a problem.  It might
be nice to output the old/new value of the timeout, but that's mostly for
the curious rather than actually necessary.

> Also needs the ok from Konrad as release manager. AIUI this is a bugfix
> for a particular piece of h/w and as Daniel explains above the downside
> is that sometimes someone might need to wait 0.75s for a timeout instead
> of something shorter.

-- 
Daniel De Graaf
National Security Agency