On 03/26/18 12:44, Stefan Berger wrote:
On 03/25/2018 07:46 PM, Kevin O'Connor wrote:
On Sun, Mar 25, 2018 at 07:17:33PM -0400, Stefan Berger wrote:
On 03/25/2018 11:45 AM, Kevin O'Connor wrote:
On Thu, Mar 22, 2018 at 08:19:09AM -0400, Stefan Berger wrote:
The timeout to wait for the register change is 30ms. We yield() while waiting, so we don't block everything entirely... Is the error message misleading and we should print out that a device was not detected or print out if it is detected instead?
Unfortunately, although the TPM code calls yield(), there isn't anything else running at that point, so the delay still directly impacts the total boot time. It's not easy to push back the TPM initialization so that other "threads" are running in parallel, because the TPM code wants to be initialized prior to running option roms and other devices.
Could we do something like the below (completely untested)? I don't think we have to wait for the TPM device to report ready, because in a real world scenario it would take an x86 cpu hundreds of milliseconds from power on to get to this point of the code anyway.
I had thought about something like that also. Do we have the time in milliseconds since power-on or reset? We could subtract the timeout values you are discarding below from it and wait for the time difference, ignoring negative values of course. TIS_DEFAULT_TIMEOUT_A is 750ms, so more critical than TIS2_DEFAULT_TIMEOUT_D with 30ms.
Unfortunately, we don't have that info. It's not easy to get it because the clock detection code doesn't run until we're nearly at the tpmhw_probe() code anyway.
That said, for tis_probe() is there any harm in moving the didvid check up? I understand it theoretically isn't setup yet, but I'd imagine in practice it always would be. Just as, in practice, I suspect the tpm would always be ready by the time we got to the tpmhw_probe() code. See below.
It's hard to say without testing this with real hardware. When I had the Acer, it was working quite reliably without waiting for the flag to be set. Though the Acer only used one certain manufacturer's TPM. Other TPMs may behave differently. Maybe Stephen can give it a try?
Can’t you get some hardware to test this? For example an old laptop with a TPM? If you work in this area, that might be a good idea.