Tegra186 currently restricts its DRAM usage to entries in the /memory node
in the DTB passed to it. However, the MMU configuration always maps the
entire first 2GB of RAM. This could allow the CPU to speculatively access
RAM that isn't part of the in-use banks. This patch switches to runtime
construction of the table that's used to construct the MMU translation
tables, and thus prevents access to RAM that's not part of a valid bank.
Note: This patch is intended to prevent access to RAM regions which U-Boot
does not need to access, with the primary purpose of avoiding theoretical
speculative access to physical regions for which the HW will throw errors
(e.g. carve-outs that the CPU has no permission to access at a bus level,
bad ECC pages, etc.). In particular, this patch is not deliberately
related to the speculation-related security issues that were recently
announced. The apparent similarity is a coincidence.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Tom Warren <twarren@nvidia.com>
ARM CPUs can architecturally (speculatively) prefetch completely arbitrary
normal memory locations, as defined by the current translation tables. The
current MMU configuration for 64-bit Tegras maps an extremely large range
of addresses as DRAM, well beyond the actual physical maximum DRAM window,
even though U-Boot only needs access to the first 2GB of DRAM; the Tegra
port of U-Boot deliberately limits itself to 2GB of RAM since some HW
modules on at least some 64-bit Tegra SoCs can only access a 32-bit
physical address space. This change reduces the amount of RAM mapped via
the MMU to disallow the CPU from ever speculatively accessing RAM that
U-Boot will definitely not access. This avoids the possibility of the HW
raising SError due to accesses to always-invalid physical addresses.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Tom Warren <twarren@nvidia.com>
Introduce virtual and physical addresses in the mapping table. This change
have no impact on existing boards because they all use idential mapping.
Signed-off-by: York Sun <york.sun@nxp.com>
Now that we have nice table driven page table creating code that gives
us everything we need, move to that.
Signed-off-by: Alexander Graf <agraf@suse.de>
This sets up a fine-grained page table, which is a requirement for
noncached_init() to operate correctly.
MMU setup code currently exists in a number of places:
- A version in the core ARMv8 support code that sets up page tables that
use very large block sizes that CONFIG_SYS_NONCACHED_MEMORY doesn't
support.
- Enhanced versions for fsl-lsch3 and zynmq that set up finer grained
page tables.
Ideally, rather than duplicating the MMU setup code yet again this patch
would instead consolidate all the different routines into the core ARMv8
code so that it supported all use-cases. However, this will require
significant effort since there appear to be a number of discrepancies[1]
between different versions of the code, and between the defines/values by
some copies of the MMU setup code use and the architectural MMU
documentation. Some reverse engineering will be required to determine the
intent of the current code.
[1] For example, in the core ARMv8 MMU setup code, three defines named
TCR_EL[123]_IPS_BITS exist, but only one of them sets the IPS field and
the others set a different field (T1SZ) in the page tables. As far as I
can tell so far, there should be no need to set different values per
exception level nor to modify the T1SZ field at all, since TTBR1 shouldn't
be enabled anyway. Another example is inconsistent values for *_VA_BITS
between the current core ARMv8 MMU setup code and the various SoC-
specific MMU setup code. Another example is that asm/armv8/mmu.h's value
for SECTION_SHIFT doesn't match asm/system.h's MMU_SECTION_SHIFT;
research is needed to determine which code relies on which of those
values and why, and whether fixing the incorrect value will cause any
regression.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
