ARM: Tegra186: don't map memory not in RAM banks
authorStephen Warren <swarren@nvidia.com>
Thu, 4 Jan 2018 18:07:14 +0000 (11:07 -0700)
committerTom Warren <twarren@nvidia.com>
Fri, 12 Jan 2018 17:12:32 +0000 (10:12 -0700)
commitcdcf55584e7e6b9be92a258d70d4d08a0216dbd0
treed1b1418a93814660f168e2fbcfacf7a91859634c
parenta9819b9e33bd9e8e2bdb694abef78397d8c00e7f
ARM: Tegra186: don't map memory not in RAM banks

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>
arch/arm/mach-tegra/arm64-mmu.c
arch/arm/mach-tegra/tegra186/nvtboot_mem.c