trousers

I forgot to mention the configuration of the machine on which I did the 
time calculations.
That is a Intel Core2 Duo CPU at 2.80GHz.

Regards,
Amruta

Amruta Gokhale wrote:
> Hello,
>
> I am using the trousers-0.3.1 library for a set of programs. These 
> programs are involved in
> verifying the software stack of a machine having a TPM chip. I am 
> using many functions
> in the trousers library such as Tspi_TPM_Quote(), Trspi_Verify() etc.
>
> I was measuring the time taken by these functions. The observed values 
> were a bit strange.
> Since I could not find an explanation myself for these values, I 
> thought I should post here.
> I would be grateful if someone could throw some light upon this.
>
> (1) The time taken by the function Tspi_TPM_Quote() is 0.732 seconds. 
> Is this function expected
> to require so much time? Going through the source code, what I found 
> was that tpm_quote()
> fills up the 'quoteinfo' structure with the requested PCR values and 
> the externally supplied "nonce".
> Can anyone explain which sub-operations of tpm_quote are time-consuming?
>
> (2) One of the program loads the AIK which is stored as Blob in a file.
> (This AIK key was generated by the "identity.c" program code available at
> http://www.privacyca.com/code.html#aikquote)
>
> The function that loads the AIK, namely, Tspi_Context_LoadKeyByBlob() 
> takes 2.5 seconds to
> execute!! Any reason why the loading of AIK into TPM memory would take 
> so long to complete?
>
> Is there any way to optimize this time, for example, by using an 
> already loaded key from
> the previous run of the program? I saw some code in load_key_init() 
> function in tcs/tcs_key.c file
> that first checks whether key loaded by another thread can be used. 
> But it does this provided the
> key blob is not encrypted. I think the key blob file produced by 
> "identity.c" is encrypted. So does
> this mean that there is no way to re-use an already loaded AIK ?
>
> I appreciate any help.
>
> Thank you,
> Amruta
>
>

Ok, I see. It would be really helpful if someone can let me know the 
function to do the symmetric key operation.

Thank you for the reply.

Amruta

David Challener wrote:
> If a key has been previous loaded it can be cached out using a TPM 
> symmetric key and then cached in later.  Ican't recall the function 
> name but some TSS stacks do this automatically for you.  The symmetric 
> operation is much faster than the private key operation, so this helps 
> a lot.  TPM speeds vary according to manufacturer for doing various 
> things, but it is certainly not an accelerator..
>
> On Wed, Apr 7, 2010 at 10:27 AM, Amruta Gokhale 
> <amrutag@... <mailto:amrutag@...>> wrote:
>
>     Hello,
>
>     I am using the trousers-0.3.1 library for a set of programs. These
>     programs are involved in
>     verifying the software stack of a machine having a TPM chip. I am
>     using
>     many functions
>     in the trousers library such as Tspi_TPM_Quote(), Trspi_Verify() etc.
>
>     I was measuring the time taken by these functions. The observed values
>     were a bit strange.
>     Since I could not find an explanation myself for these values, I
>     thought
>     I should post here.
>     I would be grateful if someone could throw some light upon this.
>
>     (1) The time taken by the function Tspi_TPM_Quote() is 0.732
>     seconds. Is
>     this function expected
>     to require so much time? Going through the source code, what I
>     found was
>     that tpm_quote()
>     fills up the 'quoteinfo' structure with the requested PCR values
>     and the
>     externally supplied "nonce".
>     Can anyone explain which sub-operations of tpm_quote are
>     time-consuming?
>
>     (2) One of the program loads the AIK which is stored as Blob in a
>     file.
>     (This AIK key was generated by the "identity.c" program code
>     available at
>     http://www.privacyca.com/code.html#aikquote)
>
>     The function that loads the AIK, namely, Tspi_Context_LoadKeyByBlob()
>     takes 2.5 seconds to
>     execute!! Any reason why the loading of AIK into TPM memory would take
>     so long to complete?
>
>     Is there any way to optimize this time, for example, by using an
>     already
>     loaded key from
>     the previous run of the program? I saw some code in load_key_init()
>     function in tcs/tcs_key.c file
>     that first checks whether key loaded by another thread can be
>     used. But
>     it does this provided the
>     key blob is not encrypted. I think the key blob file produced by
>     "identity.c" is encrypted. So does
>     this mean that there is no way to re-use an already loaded AIK ?
>
>     I appreciate any help.
>
>     Thank you,
>     Amruta
>
>
>     ------------------------------------------------------------------------------
>     Download Intel&#174; Parallel Studio Eval
>     Try the new software tools for yourself. Speed compiling, find bugs
>     proactively, and fine-tune applications for parallel performance.
>     See why Intel Parallel Studio got high marks during beta.
>     http://p.sf.net/sfu/intel-sw-dev
>     _______________________________________________
>     TrouSerS-users mailing list
>     TrouSerS-users@...
>     <mailto:TrouSerS-users@...>
>     https://lists.sourceforge.net/lists/listinfo/trousers-users
>
>

On Wed, Apr 7, 2010 at 10:27 AM, Amruta Gokhale <amrutag@...> wrote:
> Hello,
>
> I am using the trousers-0.3.1 library for a set of programs. These
> programs are involved in
> verifying the software stack of a machine having a TPM chip. I am using
> many functions
> in the trousers library such as Tspi_TPM_Quote(), Trspi_Verify() etc.
>
> I was measuring the time taken by these functions. The observed values
> were a bit strange.
> Since I could not find an explanation myself for these values, I thought
> I should post here.
> I would be grateful if someone could throw some light upon this.
>
> (1) The time taken by the function Tspi_TPM_Quote() is 0.732 seconds. Is
> this function expected
> to require so much time? Going through the source code, what I found was
> that tpm_quote()
> fills up the 'quoteinfo' structure with the requested PCR values and the
> externally supplied "nonce".
> Can anyone explain which sub-operations of tpm_quote are time-consuming?

The only time consuming operation is creating the RSA signature, which
happens in the TPM when it does the Quote. That amount of time is a
little slow, suggesting you have an older or cheaper TPM.

>
> (2) One of the program loads the AIK which is stored as Blob in a file.
> (This AIK key was generated by the "identity.c" program code available at
> http://www.privacyca.com/code.html#aikquote)
>
> The function that loads the AIK, namely, Tspi_Context_LoadKeyByBlob()
> takes 2.5 seconds to
> execute!! Any reason why the loading of AIK into TPM memory would take
> so long to complete?

Well that time doesn't sound right to me. The TPM has to do one RSA
decryption to load the AIK. That should take a similar amount of time
as for the RSA signature in the Quote operation. I can't see why it
would take 3-4 times longer. Again I can only suggest you got a slow,
cheap, old TPM. It should take between 0.1 and 1.0 seconds.


>
> Is there any way to optimize this time, for example, by using an already
> loaded key from
> the previous run of the program? I saw some code in load_key_init()
> function in tcs/tcs_key.c file
> that first checks whether key loaded by another thread can be used. But
> it does this provided the
> key blob is not encrypted. I think the key blob file produced by
> "identity.c" is encrypted. So does
> this mean that there is no way to re-use an already loaded AIK ?

You could certainly load the key at the time your program started and
re-use it later, but that probably wouldn't help. I don't think
Trousers lets you keep the key in the TPM between program runs though.

BTW the TPM doesn't do symmetric encryption, a la AES.

Hal Finney

I have a new version of TPM (version 1.2) manufactured by Broadcom. These
are the

Operation           Time (s)       Std. Dev. (s)
PCR Extend        0.0253            0.001
Create AIK           34.3               8.22
Load AIK             1.75               0.002
Sign PCR           0.998              0.001

The reason I would imagine loading AIK into the TPM is slower than signing
PCR values is that the encrypted AIK key blob is stored on the local file
system and bits travel on slow registers before they make it into the TPM. I
think the extra time spent by Tspi_Context_LoadKeyByBlob() is due to the
fact that the AIK key blob is stored outside the TPM's memory. Hence the key
blob needs to be inserted into the TPM before you can run any RSA signature
on them, whereas with Quote operation, the pcr values that are getting
signed are already stored inside the TPM registers. I remember reading that
the input and output bus for TPM is slow.

On Fri, Apr 9, 2010 at 1:14 PM, Hal Finney <hal.finney@...> wrote:

> On Wed, Apr 7, 2010 at 10:27 AM, Amruta Gokhale <amrutag@...>
> wrote:
> > Hello,
> >
> > I am using the trousers-0.3.1 library for a set of programs. These
> > programs are involved in
> > verifying the software stack of a machine having a TPM chip. I am using
> > many functions
> > in the trousers library such as Tspi_TPM_Quote(), Trspi_Verify() etc.
> >
> > I was measuring the time taken by these functions. The observed values
> > were a bit strange.
> > Since I could not find an explanation myself for these values, I thought
> > I should post here.
> > I would be grateful if someone could throw some light upon this.
> >
> > (1) The time taken by the function Tspi_TPM_Quote() is 0.732 seconds. Is
> > this function expected
> > to require so much time? Going through the source code, what I found was
> > that tpm_quote()
> > fills up the 'quoteinfo' structure with the requested PCR values and the
> > externally supplied "nonce".
> > Can anyone explain which sub-operations of tpm_quote are time-consuming?
>
> The only time consuming operation is creating the RSA signature, which
> happens in the TPM when it does the Quote. That amount of time is a
> little slow, suggesting you have an older or cheaper TPM.
>
> >
> > (2) One of the program loads the AIK which is stored as Blob in a file.
> > (This AIK key was generated by the "identity.c" program code available at
> > http://www.privacyca.com/code.html#aikquote)
> >
> > The function that loads the AIK, namely, Tspi_Context_LoadKeyByBlob()
> > takes 2.5 seconds to
> > execute!! Any reason why the loading of AIK into TPM memory would take
> > so long to complete?
>
> Well that time doesn't sound right to me. The TPM has to do one RSA
> decryption to load the AIK. That should take a similar amount of time
> as for the RSA signature in the Quote operation. I can't see why it
> would take 3-4 times longer. Again I can only suggest you got a slow,
> cheap, old TPM. It should take between 0.1 and 1.0 seconds.
>
>
> >
> > Is there any way to optimize this time, for example, by using an already
> > loaded key from
> > the previous run of the program? I saw some code in load_key_init()
> > function in tcs/tcs_key.c file
> > that first checks whether key loaded by another thread can be used. But
> > it does this provided the
> > key blob is not encrypted. I think the key blob file produced by
> > "identity.c" is encrypted. So does
> > this mean that there is no way to re-use an already loaded AIK ?
>
> You could certainly load the key at the time your program started and
> re-use it later, but that probably wouldn't help. I don't think
> Trousers lets you keep the key in the TPM between program runs though.
>
> BTW the TPM doesn't do symmetric encryption, a la AES.
>
> Hal Finney
>
>
> ------------------------------------------------------------------------------
> Download Intel&#174; Parallel Studio Eval
> Try the new software tools for yourself. Speed compiling, find bugs
> proactively, and fine-tune applications for parallel performance.
> See why Intel Parallel Studio got high marks during beta.
> http://p.sf.net/sfu/intel-sw-dev
> _______________________________________________
> TrouSerS-users mailing list
> TrouSerS-users@...
> https://lists.sourceforge.net/lists/listinfo/trousers-users
>

Thank you, Hal, for the reply.

Thank you Hardeep for giving the performance for another TPM.

The TPM I have got has this configuration:
  TPM 1.2 Version Info:
  Chip Version:        1.2.7.0
  Spec Level:          2
  Errata Revision:     2
  TPM Vendor ID:       STM
  TPM Version:         01010000

As you suggest, this might be slower compared to others out there.

Regards,
Amruta

Hardeep Uppal wrote:
> I have a new version of TPM (version 1.2) manufactured by Broadcom. 
> These are the 
>
> Operation           Time (s)       Std. Dev. (s)
> PCR Extend        0.0253            0.001
> Create AIK           34.3               8.22
> Load AIK             1.75               0.002
> Sign PCR           0.998              0.001
>
> The reason I would imagine loading AIK into the TPM is slower than 
> signing PCR values is that the encrypted AIK key blob is stored on the 
> local file system and bits travel on slow registers before they make 
> it into the TPM. I think the extra time spent 
> by Tspi_Context_LoadKeyByBlob() is due to the fact that the AIK key 
> blob is stored outside the TPM's memory. Hence the key blob needs to 
> be inserted into the TPM before you can run any RSA signature on them, 
> whereas with Quote operation, the pcr values that are getting signed 
> are already stored inside the TPM registers. I remember reading that 
> the input and output bus for TPM is slow.
>
> On Fri, Apr 9, 2010 at 1:14 PM, Hal Finney <hal.finney@... 
> <mailto:hal.finney@...>> wrote:
>
>     On Wed, Apr 7, 2010 at 10:27 AM, Amruta Gokhale
>     <amrutag@... <mailto:amrutag@...>> wrote:
>     > Hello,
>     >
>     > I am using the trousers-0.3.1 library for a set of programs. These
>     > programs are involved in
>     > verifying the software stack of a machine having a TPM chip. I
>     am using
>     > many functions
>     > in the trousers library such as Tspi_TPM_Quote(), Trspi_Verify()
>     etc.
>     >
>     > I was measuring the time taken by these functions. The observed
>     values
>     > were a bit strange.
>     > Since I could not find an explanation myself for these values, I
>     thought
>     > I should post here.
>     > I would be grateful if someone could throw some light upon this.
>     >
>     > (1) The time taken by the function Tspi_TPM_Quote() is 0.732
>     seconds. Is
>     > this function expected
>     > to require so much time? Going through the source code, what I
>     found was
>     > that tpm_quote()
>     > fills up the 'quoteinfo' structure with the requested PCR values
>     and the
>     > externally supplied "nonce".
>     > Can anyone explain which sub-operations of tpm_quote are
>     time-consuming?
>
>     The only time consuming operation is creating the RSA signature, which
>     happens in the TPM when it does the Quote. That amount of time is a
>     little slow, suggesting you have an older or cheaper TPM.
>
>     >
>     > (2) One of the program loads the AIK which is stored as Blob in
>     a file.
>     > (This AIK key was generated by the "identity.c" program code
>     available at
>     > http://www.privacyca.com/code.html#aikquote)
>     >
>     > The function that loads the AIK, namely,
>     Tspi_Context_LoadKeyByBlob()
>     > takes 2.5 seconds to
>     > execute!! Any reason why the loading of AIK into TPM memory
>     would take
>     > so long to complete?
>
>     Well that time doesn't sound right to me. The TPM has to do one RSA
>     decryption to load the AIK. That should take a similar amount of time
>     as for the RSA signature in the Quote operation. I can't see why it
>     would take 3-4 times longer. Again I can only suggest you got a slow,
>     cheap, old TPM. It should take between 0.1 and 1.0 seconds.
>
>
>     >
>     > Is there any way to optimize this time, for example, by using an
>     already
>     > loaded key from
>     > the previous run of the program? I saw some code in load_key_init()
>     > function in tcs/tcs_key.c file
>     > that first checks whether key loaded by another thread can be
>     used. But
>     > it does this provided the
>     > key blob is not encrypted. I think the key blob file produced by
>     > "identity.c" is encrypted. So does
>     > this mean that there is no way to re-use an already loaded AIK ?
>
>     You could certainly load the key at the time your program started and
>     re-use it later, but that probably wouldn't help. I don't think
>     Trousers lets you keep the key in the TPM between program runs though.
>
>     BTW the TPM doesn't do symmetric encryption, a la AES.
>
>     Hal Finney
>
>     ------------------------------------------------------------------------------
>     Download Intel&#174; Parallel Studio Eval
>     Try the new software tools for yourself. Speed compiling, find bugs
>     proactively, and fine-tune applications for parallel performance.
>     See why Intel Parallel Studio got high marks during beta.
>     http://p.sf.net/sfu/intel-sw-dev
>     _______________________________________________
>     TrouSerS-users mailing list
>     TrouSerS-users@...
>     <mailto:TrouSerS-users@...>
>     https://lists.sourceforge.net/lists/listinfo/trousers-users
>
>