On 2/13/20 2:39 PM, Jason Gunthorpe wrote:
On Thu, Feb 13, 2020 at 02:15:03PM -0500, Stefan Berger wrote:
On 2/13/20 2:11 PM, Jason Gunthorpe wrote:
On Thu, Feb 13, 2020 at 02:04:12PM -0500, Stefan Berger wrote:
On 2/13/20 1:35 PM, Jason Gunthorpe wrote:
On Thu, Feb 13, 2020 at 01:20:12PM -0500, Stefan Berger wrote:
I don't want side effects for the TPM 1.2 case here, so I am only modifying
the flag for the case where the new TPM 2 is being used. Here's the code
where it shows the effect.
I'm surprised this driver is using AUTO_STARTUP, it was intended for
embedded cases where their is no firmware to boot the TPM.
The TIS is also using it on any device.
TIS is a generic driver, and can run on TPMs without firmware
support. It doesn't know either way
The following drivers are all using it:
drivers/char/tpm/st33zp24/st33zp24.c, line 493
drivers/char/tpm/tpm-interface.c, line 374
drivers/char/tpm/tpm_crb.c, line 421
drivers/char/tpm/tpm_ftpm_tee.c, line 184
drivers/char/tpm/tpm_i2c_atmel.c, line 139
drivers/char/tpm/tpm_i2c_infineon.c, line 602
drivers/char/tpm/tpm_i2c_nuvoton.c, line 465
drivers/char/tpm/tpm_tis_core.c, line 917
drivers/char/tpm/tpm_vtpm_proxy.c, line 435
https://elixir.bootlin.com/linux/latest/ident/TPM_OPS_AUTO_STARTUP
These are all general purpose drivers.
Though perhaps vtpm_proxy shouldn't include it, not sure.
Chips using AUTO_STARTUP are basically useless for PCRs/etc.
I'd expect somthing called vtpm to have been started and PCRs working
before Linux is started??
Yes, there's supposed to be firmware.
I only see one caller to tpm2_get_cc_attrs_tbl(chip), which is necessary to
call. This caller happens to be in tpm2_auto_startup.
That seems to be a mistake, proper startup of the driver should never
require auto_startup.
Is this IBM vTPM driver special that it should do things differently than
all those drivers listed above? From looking at the code is seems it is to
be set for the TPM 2.0 case.
Any driver that knows the TPM must be started prior to Linux
booting should not use the flag. vtpm drivers in general would seem
to be the case where we can make this statement.
Wouldn't this statement apply to all systems, including embedded ones?
Basically all firmwares should implement the CRTM and do the TPM
initialization.
If it was mandatory then it would not be a flag the driver has to specify.
I'll add a case where we call into tpm2_get_cc_attrs_tbl(chip) in case
the auto startup flag is not set. I believe this is the only part that's
missing for TPM 2 to work if not autostarted.
Stefan
Jason
