60e8f3cfd5
This patch introduces a build flag which allows the xlat tables to be mapped in a read-only region within BL31 memory. It makes it much harder for someone who has acquired the ability to write to arbitrary secure memory addresses to gain control of the translation tables. The memory attributes of the descriptors describing the tables themselves are changed to read-only secure data. This change happens at the end of BL31 runtime setup. Until this point, the tables have read-write permissions. This gives a window of opportunity for changes to be made to the tables with the MMU on (e.g. reclaiming init code). No changes can be made to the tables with the MMU turned on from this point onwards. This change is also enabled for sp_min and tspd. To make all this possible, the base table was moved to .rodata. The penalty we pay is that now .rodata must be aligned to the size of the base table (512B alignment). Still, this is better than putting the base table with the higher level tables in the xlat_table section, as that would cost us a full 4KB page. Changing the tables from read-write to read-only cannot be done with the MMU on, as the break-before-make sequence would invalidate the descriptor which resolves the level 3 page table where that very descriptor is located. This would make the translation required for writing the changes impossible, generating an MMU fault. The caches are also flushed. Signed-off-by: Petre-Ionut Tudor <petre-ionut.tudor@arm.com> Change-Id: Ibe5de307e6dc94c67d6186139ac3973516430466 |
||
|---|---|---|
| .. | ||
| armclang_printf.S | ||
| cache_helpers.S | ||
| misc_helpers.S | ||