Age | Commit message (Collapse) | Author |
|
If necessary, SCLEAN needs to run in early romstage, where DRAM is not
working yet. In fact, that the DRAM isn't working is the reason to run
SCLEAN in the first place. Before running GETSEC, CAR needs to be torn
down, as MTRRs have to be reprogrammed to cache the BIOS ACM. Further,
running SCLEAN leaves the system in an undefined state, where the only
sane thing to do is reset the platform. Thus, invoking SCLEAN requires
specific assembly prologue and epilogue sections before and after MTRR
setup, and neither DRAM nor CAR may be relied upon for the MTRR setup.
In order to handle this without duplicating the MTRR setup code, place
it in a macro on a separate file. This needs to be a macro because the
call and return instructions rely on the stack being usable, and it is
not the case for SCLEAN. The MTRR code clobbers many registers, but no
other choice remains when the registers cannot be saved anywhere else.
Tested on Asrock B85M Pro4, BIOS ACM can still be launched.
Change-Id: I2f5e82f57b458ca1637790ddc1ddc14bba68ac49
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46603
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
|
|
The BIOS ACM will check that enabled variable MTRRs do not cover more
than the ACM's size, rounded up to 4 KiB. If that is not the case,
launching the ACM will result in a lovely TXT reset. How boring.
The new algorithm simply performs a reverse bit scan in a loop, and
allocates one MTRR for each set bit in the rounded-up size to cache.
Before allocating anything, it checks if there are enough variable
MTRRs; if not, it will refuse to cache anything. This will result in
another TXT reset, initiated by the processor, with error type 5:
Load memory type error in Authenticated Code Execution Area.
This can only happen if the ACM has specific caching requirements that
the current code does not know about, or something has been compromised.
Therefore, causing a TXT reset should be a reasonable enough approach.
Also, disable all MTRRs before clearing the variable MTRRs and only
enable them again once they have been set up with the new values.
Tested on Asrock B85M Pro4 with a BIOS ACM whose size is 101504 bytes.
Without this patch, launching the ACM would result in a TXT reset. This
no longer happens when this patch is applied.
Change-Id: I8d411f6450928357544be20250262c2005d1e75d
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/44880
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-by: Christian Walter <christian.walter@9elements.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
|
|
This needs to be saved and restored, otherwise the BSP might have an
inconsistent MTRR setup with regards to the AP's which results in
weird errors and slowdowns in the operating system.
TESTED: Fixes booting OCP/Deltalake with Linux 5.8.
Change-Id: Iace636ec6fca3b4d7b2856f0f054947c5b3bc8de
Signed-off-by: Arthur Heymans <arthur@aheymans.xyz>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/46375
Reviewed-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-by: Christian Walter <christian.walter@9elements.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
|
|
The MSR macros were treated as memory addresses and the loops had
off-by-one errors. This resulted in a CPU exception before GETSEC, and
another exception after GETSEC (once the first exception was fixed).
Tested on Asrock B85M Pro4, ACM complains about the missing TPM and
resets the platform. When the `getsec` instruction is commented-out, the
board is able to boot normally, without any exceptions nor corruption.
Change-Id: Ib5d23cf9885401f3ec69b0f14cea7bad77eee19a
Signed-off-by: Angel Pons <th3fanbus@gmail.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/44183
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Philipp Deppenwiese <zaolin.daisuki@gmail.com>
|
|
Add TXT ramstage driver:
* Show startup errors
* Check for TXT reset
* Check for Secrets-in-memory
* Add assembly for GETSEC instruction
* Check platform state if GETSEC instruction is supported
* Configure TXT memory regions
* Lock TXT
* Protect TSEG using DMA protected regions
* Place SINIT ACM
* Print information about ACMs
Extend the `security_clear_dram_request()` function:
* Clear all DRAM if secrets are in memory
Add a config so that the code gets build-tested. Since BIOS and SINIT
ACM binaries are not available, use the STM binary as a placeholder.
Tested on OCP Wedge100s and Facebook Watson
* Able to enter a Measured Launch Environment using SINIT ACM and TBOOT
* Secrets in Memory bit is set on ungraceful shutdown
* Memory is cleared after ungraceful shutdown
Change-Id: Iaf4be7f016cc12d3971e1e1fe171e6665e44c284
Signed-off-by: Philipp Deppenwiese <zaolin@das-labor.org>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/37016
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Christian Walter <christian.walter@9elements.com>
|