tpm_tis_spi takes minutes to probe

Greetings,

I'm using the tpm_tis_spi.ko module and it is taking several minutes to probe on kernel:
- commit ea5f6ad9ad96 ("Merge tag 'platform-drivers-x86-v6.10-1' of
git://git.kernel.org/pub/scm/linux/kernel/git/pdx86/platform-drivers-x86")

root@verdin-imx8mm-07317726:~# time modprobe tpm_tis_spi 
[   57.534597] SPI driver tpm_tis_spi has no spi_device_id for atmel,attpm20p
[   57.560684] tpm_tis_spi spi2.1: 2.0 TPM (device-id 0x3205, rev-id 1)
[   57.584943] tpm tpm0: A TPM error (256) occurred attempting the self test
[   57.591797] tpm tpm0: starting up the TPM manually
real    8m6.438s
user    0m0.007s
sys     0m0.013s

...And after probing, sometimes the system becames unresponsive.

I did the same test with kernel:
- commit a38297e3fb01 ("Linux 6.9")

root@verdin-imx8mm-07317726:~# time modprobe tpm_tis_spi
[   53.643744] SPI driver tpm_tis_spi has no spi_device_id for atmel,attpm20p
[   53.668404] tpm_tis_spi spi2.1: 2.0 TPM (device-id 0x3205, rev-id 1)
[   53.692997] tpm tpm0: A TPM error (256) occurred attempting the self test
[   53.699879] tpm tpm0: starting up the TPM manually

real    0m2.809s
user    0m0.001s
sys     0m0.022s

I tested on kernel v6.8 and same it takes same time. Do you know anything about this?

Best regards,
Vitor Soares

This was added in 6.9:

$ git --no-pager log -1 3c45308c44eda
commit 3c45308c44eda6cc3343a48341a82b96753c8a13
Author: Lukas Wunner <[email protected]>
Date:   Sat Jan 13 18:10:52 2024 +0100

tpm_tis_spi: Add compatible string atmel,attpm20p

Commit 4f2a348aa365 ("arm64: dts: imx8mm-venice-gw73xx: add TPM device") added a devicetree node for the Trusted Platform Module on certain Gateworks boards.

The commit only used the generic "tcg,tpm_tis-spi" compatible string, but public documentation shows that the chip is an ATTPM20P from Atmel (nowadays Microchip):
https://trac.gateworks.com/wiki/tpm

Add the chip to the supported compatible strings of the TPM TIS SPI driver.

For reference, a datasheet is available at:
https://ww1.microchip.com/downloads/en/DeviceDoc/ATTPM20P-Trusted-Platform-Module-TPM-2.0-SPI-Interface-Summary-Data-Sheet-DS40002082A.pdf

linux-tpmdd on  tpm2_key

$ git describe --contains 3c45308c44eda
tpmdd-v6.9-rc1~2

Course of event is I think:

	rc = tpm2_do_selftest(chip);
	if (rc && rc != TPM2_RC_INITIALIZE)
		goto out;

	/* 1. TPM_RC_INITIALIZE */
	if (rc == TPM2_RC_INITIALIZE) {
		/* 2. Branches here. */
		rc = tpm2_startup(chip);
		if (rc)
			goto out;

		rc = tpm2_do_selftest(chip);
		if (rc)
			goto out;
	}

	/* 4. Second self-test successful. */

It is possible that there is a performance regression given multitude of HMAC changes. It would likely had to be in tpm2_do_selftest(), since it is the most time-consuming function.

I checked the timeouts etc. but in the first seek did find anything obvious.

BR, Jarkko

Right, the extra time comes from probably null key creation, even though tpm2_do_selftest() does not itself invoke tpm2_start_auth_session().

If you don't want that to happen, then you should disable CONFIG_TCG_TPM2_HMAC but also bus encrypted sessions to trusted keys random number generation and PCR extension.

BR, Jarkko