Re: [PATCH v11 00/16] Remove nested TPM operations

From: Stefan Berger
Date: Thu Feb 07 2019 - 20:51:27 EST

On 2/7/19 7:33 PM, Jarkko Sakkinen wrote:
On Thu, Feb 07, 2019 at 06:29:43PM -0500, Stefan Berger wrote:
On 2/7/19 4:29 PM, Jarkko Sakkinen wrote:
On Thu, Feb 07, 2019 at 07:41:56PM +0100, Alexander Steffen wrote:
On 06.02.2019 13:06, Jarkko Sakkinen wrote:
On Wed, Feb 06, 2019 at 12:47:07AM +0200, Jarkko Sakkinen wrote:
Make the changes necessary to detach TPM space code and TPM activation
code out of the tpm_transmit() flow because of both of these can cause
nested tpm_transmit() calls. The nesteds calls make the whole flow hard
to maintain, and thus, it is better to just fix things now before this
turns into a bigger mess.

v11:
* Drop the patch that tries to flush TPM space on system. Not a proper
fallback + out of scope for this patch set.

v10:
* Use void pointers to avoid unnecessary casts in functions paramaters
where it makes sense.

v9:
* Fixed again tpm_try_get_ops().
* Added missing reviewed-by's.

v8:
* Re-add the check for ret < 0 after calling tpm_try_transmit() that
was dropped by mistake while moving code.
* Fix error fallback for tpm_try_get_ops() when tpm_chip_start()
fails.

v7:
* Reorganize series so that more trivial and self-contained changes are
in the head.

v6:
* When tpm_validate_commmand() was moved to tpm2-space.c, the struct for
the TPM header was incorrectly declared as struct tpm_input_header.
* Fix return value in tpm_validate_command().

v5:
* Add the missing rev's from Stefan Berger.

v4:
* Return 0 from pcrs_show() when tpm1_pcr_read() fails.
* Fix error handling flow in tpm_try_transmit().
* Replace struct tpm_input_header and struct tpm_output_header with
struct tpm_header.

v3:
* Encapsulate power gating code to tpm_chip_start() and tpm_chip_stop().
* Move TPM power gating code and locking to tpm_try_get_ops() and
tpm_put_ops().
* Call power gating code directly in tpm_chip_register() and
tpm2_del_space().

v2:
* Print tpm2_commit_space() error inside tpm2_commit_space()
* Error code was not printed when recv() callback failed. It is
fixed in this version.
* Added a patch that removes @space from tpm_transmit().
* Fixed a regression in earlier series. Forgot to amend the change
from the staging area that renames NESTED to UNLOCKED in tpm2-space.c.

Jarkko Sakkinen (16):
tpm: use tpm_buf in tpm_transmit_cmd() as the IO parameter
tpm: fix invalid return value in pubek_show()
tpm: return 0 from pcrs_show() when tpm1_pcr_read() fails
tpm: print tpm2_commit_space() error inside tpm2_commit_space()
tpm: declare struct tpm_header
tpm: access command header through struct in tpm_try_transmit()
tpm: encapsulate tpm_dev_transmit()
tpm: clean up tpm_try_transmit() error handling flow
tpm: move tpm_validate_commmand() to tpm2-space.c
tpm: move TPM space code out of tpm_transmit()
tpm: remove @space from tpm_transmit()
tpm: use tpm_try_get_ops() in tpm-sysfs.c.
tpm: remove TPM_TRANSMIT_UNLOCKED flag
tpm: introduce tpm_chip_start() and tpm_chip_stop()
tpm: take TPM chip power gating out of tpm_transmit()
tpm: remove @flags from tpm_transmit()

drivers/char/tpm/tpm-chip.c | 109 ++++++++++++
drivers/char/tpm/tpm-dev-common.c | 44 ++++-
drivers/char/tpm/tpm-interface.c | 264 ++++++------------------------
drivers/char/tpm/tpm-sysfs.c | 138 ++++++++++------
drivers/char/tpm/tpm.h | 64 +++-----
drivers/char/tpm/tpm1-cmd.c | 28 +---
drivers/char/tpm/tpm2-cmd.c | 72 +++-----
drivers/char/tpm/tpm2-space.c | 91 +++++++---
drivers/char/tpm/tpm_i2c_atmel.c | 5 +-
drivers/char/tpm/tpm_vtpm_proxy.c | 12 +-
drivers/char/tpm/xen-tpmfront.c | 2 +-
11 files changed, 408 insertions(+), 421 deletions(-)

--
2.19.1

Applied to master and next.
Something in this series seems to break basic TPM communication for me.

For TPM2.0s the probe command fails, causing them to be misdetected as
TPM1.2s:

---
tpm tpm0: tpm_try_transmit: tpm_send: error -5
tpm_tis MSFT0101:00: 1.2 TPM (device-id 0x1A, rev-id 22)
tpm tpm0: A TPM error (30) occurred attempting to determine the timeouts
---
tpm tpm0: tpm_try_transmit: tpm_send: error -5
tpm_tis_spi spi0.1: 1.2 TPM (device-id 0x1B, rev-id 22)
tpm tpm0: A TPM error (30) occurred attempting to determine the timeouts
---

And for something that actually is a TPM1.2 it fails in a similar way:

---
tpm_i2c_infineon 1-0020: 1.2 TPM (device-id 0x1A)
tpm tpm0: A TPM error (-14) occurred attempting to determine the timeouts
---
tpm tpm0: tpm_try_transmit: tpm_send: error -5
tpm_tis_spi spi0.1: 1.2 TPM (device-id 0x1B, rev-id 16)
tpm tpm0: A TPM error (-14) occurred attempting to determine the timeouts
tpm_tis_spi: probe of spi0.1 failed with error -14
---

I see this problem across my entire range of TPM devices and test platforms.
Any idea what could be wrong here?
Weird.

Can you run a bisect?
There are 2 bugs and the following overall patch against your tree fixes
them. Let me comment on the individual patches in v17. I missed those
obviously when testing with the tpm_vtpm_proxy...



diff --git a/drivers/char/tpm/tpm-interface.c
b/drivers/char/tpm/tpm-interface.c
index 02e8cffd1163..34c0da55d885 100644
--- a/drivers/char/tpm/tpm-interface.c
+++ b/drivers/char/tpm/tpm-interface.c
@@ -94,6 +94,7 @@ static ssize_t tpm_try_transmit(struct tpm_chip *chip,
void *buf, size_t bufsiz)
ÂÂÂÂ ÂÂÂ return rc;
ÂÂÂÂ }

+ÂÂÂ rc = 0;
ÂÂÂÂ if (chip->flags & TPM_CHIP_FLAG_IRQ)
ÂÂÂÂ ÂÂÂ goto out_recv;
What why?


This fix seems to only be necessary when bisecting. You may want to apply it!



diff --git a/drivers/char/tpm/tpm2-cmd.c b/drivers/char/tpm/tpm2-cmd.c
index e74c5b7b64bf..52afe20cc8a1 100644
--- a/drivers/char/tpm/tpm2-cmd.c
+++ b/drivers/char/tpm/tpm2-cmd.c
@@ -799,7 +799,9 @@ int tpm2_probe(struct tpm_chip *chip)
ÂÂÂÂ tpm_buf_append_u32(&buf, TPM2_CAP_TPM_PROPERTIES);
ÂÂÂÂ tpm_buf_append_u32(&buf, TPM_PT_TOTAL_COMMANDS);
ÂÂÂÂ tpm_buf_append_u32(&buf, 1);
+ÂÂÂ tpm_chip_start(chip);
ÂÂÂÂ rc = tpm_transmit_cmd(chip, &buf, 0, NULL);
+ÂÂÂ tpm_chip_stop(chip);
Thanks Stefan! I added call to tpm_tis_core as tpm2-cmd.c is to be kept
out of chip management common case being that you call tpm_try_get_ops(),
do 1-N TPM commands and release with tpm_put_ops(). These functions take
care starting and stopping the chip.

I fixed the 2nd issue in the master.

You also need to export the tpm_chip_start/stop symbols.


Master now seems to (still) have a problem when rmmod'ing:

A TPM error (325) occurred stopping the TPM. This happens at the same patch where I had to add the tpm_chip_start/stop above.





/Jarkko