Re: [Qemu-devel] [PATCH V14 2/7] Add TPM (frontend) hardware interface (TPM TIS) to Qemu

[Stefan Berger <stefanb@linux.vnet.ibm.com>]

On 02/21/2012 06:08 PM, Michael S. Tsirkin wrote:
> On Tue, Feb 21, 2012 at 05:30:32PM -0500, Stefan Berger wrote:
>> On 02/21/2012 02:58 PM, Michael S. Tsirkin wrote:
>
> Worst case, you can take a qemu mutex. Is tpm very
> performance-sensitive to make contention on that
> lock a problem?
>

FYI: Some performance measurements with the libtpms backend and linux 
3.3.0-rc3 with IMA enabled. IMA takes ~1324 measurements while booting 
and sends TPM_Extend()s. I implemented a 'fast-path' for TPM ordinals 
that are considered 'short' in terms of duration of execution and where 
this TPM_Extend operation is one of them. In this case the Qemu main 
thread enters the TPM code rather than using the thread. Here are the 
measurements with the 1st one taken when dmesg states 'mtrr: no MTRR for 
fc000..'  and the 2nd one 'eth0: no IPv6 routers present'. Both may not 
be a great way to measure but at least show some 'trends':

                                                                       
                                               average    percentage

no IMA                        14.17 / 21.93    14.28 / 21.24    12.83 / 
20.18        13.76 / 21.12    100 / 100
w/ fast path + IRQ:     15.50 / 22.72    14.58 / 21.41    15.30 / 
22.37        15.12 / 22.17    110 / 105
w/ fast path + poll:     16.17 / 22.96    14.67 / 21.66    14.43 / 
22.17        15.09 / 22.26    110 / 105

no fast path + IRQ:    16.93 / 23.51    13.94 / 20.72    16.39 / 
23.48        15.75 / 22.57    114 / 107
no fast path + poll:    17.03 / 23.95    18.11 / 25.38    16.67 / 
24.08        17.27 / 24.47    126 / 119


      Stefan