On 12/09/2024 11:51, Ard Biesheuvel wrote: > On Thu, 12 Sept 2024 at 12:23, Usama Arif <usamaarif642@xxxxxxxxx> wrote: >> >> >> >> On 11/09/2024 12:51, Ard Biesheuvel wrote: >>> On Wed, 11 Sept 2024 at 12:41, Usama Arif <usamaarif642@xxxxxxxxx> wrote: >>>> >>>> Looking at the TPM spec [1] >>>> >>>> If the ACPI TPM2 table contains the address and size of the Platform >>>> Firmware TCG log, firmware “pins” the memory associated with the >>>> Platform FirmwareTCG log, and reports this memory as “Reserved” memory >>>> via the INT 15h/E820 interface. >>>> >>>> It looks like the firmware should pass this as reserved in e820 memory >>>> map. However, it doesn't seem to. The firmware being tested on is: >>>> dmidecode -s bios-version >>>> edk2-20240214-2.el9 >>>> >>>> When this area is not reserved, it comes up as usable in >>>> /sys/firmware/memmap. This means that kexec, which uses that memmap >>>> to find usable memory regions, can select the region where efi.tpm_log >>>> is and overwrite it and relocate_kernel. >>>> >>>> Having a fix in firmware can be difficult to get through. As a secondary >>>> fix, this patch marks that region as reserved in e820_table_firmware if it >>>> is currently E820_TYPE_RAM so that kexec doesn't use it for kernel segments. >>>> >>>> [1] https://trustedcomputinggroup.org/wp-content/uploads/PC-ClientPlatform_Profile_for_TPM_2p0_Systems_v49_161114_public-review.pdf >>>> >>>> Signed-off-by: Usama Arif <usamaarif642@xxxxxxxxx> >>> >>> I would expect the EFI memory map to E820 conversion implemented in >>> the EFI stub to take care of this. >>> >> >> So I have been making a prototype with EFI stub, and the unfinished version is looking like a >> horrible hack. >> >> The only way to do this in libstub is to pass log_tbl all the way from efi_retrieve_tcg2_eventlog >> to efi_stub_entry and from there to setup_e820. >> While going through the efi memory map and converting it to e820 table in setup_e820, you have to check >> if log_tbl falls in any of the ranges and if the range is E820_TYPE_RAM. If that condition is satisfied, >> then you have to split that range into 3. i.e. the E820_TYPE_RAM range before tpm_log, the tpm_log >> E820_TYPE_RESERVED range, and the E820_TYPE_RAM range after. There are no helper functions, so this >> splitting involves playing with a lot of pointers, and it looks quite ugly. I believe doing this >> way is more likely to introduce bugs. >> >> If we are having to compensate for an EFI bug, would it make sense to do it in the way done >> in RFC and do it in kernel rather than libstub? It is simple and very likely to be bug free. >> > > I don't see how this could be an EFI bug, given that it does not deal > with E820 tables at all. EFI passes memory descriptors to libstub, libstub converts it to e820. I believe the right behaviour should be that EFI creates an EFI_RESERVED_TYPE region for that TPM log memory. Then libstub would automatically convert that EFI_RESERVED_TYPE to E820_TYPE_RESERVED in setup_e820 [1]. [1] https://elixir.bootlin.com/linux/v6.10.9/source/drivers/firmware/efi/libstub/x86-stub.c#L573
