Change Details

This change moves the ppc64 direct map base address up to high memory and fixes a -- suprisingly large -- amount of fallout from the change. We want to move it for three reasons: 1. It makes NULL and related addresses invalid pointers in the kernel, giving us a fighting chance of catching NULL-pointer bugs. 2. It needs to be here for POWER9 Radix translation to work. 3. If 64-bit Book-E grows a direct map, it also needs to be in high memory. The new base address is chosen as the base of the fourth radix quadrant (the minimum kernel address in this translation mode) and because all supported CPUs ignore at least the first two bits of addresses in real mode, allowing direct-map addresses to be used in real-mode handlers. This is required by Linux and is part of the architecture standard starting in POWER ISA 3, so can be relied upon. The kernel continues to execute from addresses below the (new) direct map with a 1:1 mapping. This is a consequence of the early bootstrap code, which will be addressed in a follow-up commit on systems that can handle it (all but Apple hardware when running with OF rather than FDT). Because the amount of fallout from moving this was surprisingly large, testing on a diverse set of systems would be appreciated. So far, this is tested only on Apple hardware (working well) and PS3 (possible bug).

This change moves the ppc64 direct map base address up to high memory and fixes a -- suprisingly large -- amount of fallout from the change. We want to move it for three reasons: 1. It makes NULL and related addresses invalid pointers in the kernel, giving us a fighting chance of catching NULL-pointer bugs. 2. It needs to be here for POWER9 Radix translation to work. 3. If 64-bit Book-E grows a direct map, it also needs to be in high memory. The new base address is chosen as the base of the fourth radix quadrant (the minimum kernel address in this translation mode) and because all supported CPUs ignore at least the first two bits of addresses in real mode, allowing direct-map addresses to be used in real-mode handlers. This is required by Linux and is part of the architecture standard starting in POWER ISA 3, so can be relied upon. The kernel continues to execute from addresses below the (new) direct map with a 1:1 mapping. This is a consequence of the early bootstrap code, which will be addressed in a follow-up commit on systems that can handle it (all but Apple hardware when running with OF rather than FDT). Because the amount of fallout from moving this was surprisingly large, testing on a diverse set of systems would be appreciated. So far, this is tested only on Apple hardware (working well) and PS3 (likewise).

This change moves the ppc64 direct map base address up to high memory and fixes a -- suprisingly large -- amount of fallout from the change. We want to move it for three reasons: 1. It makes NULL and related addresses invalid pointers in the kernel, giving us a fighting chance of catching NULL-pointer bugs. 2. It needs to be here for POWER9 Radix translation to work. 3. If 64-bit Book-E grows a direct map, it also needs to be in high memory. The new base address is chosen as the base of the fourth radix quadrant (the minimum kernel address in this translation mode) and because all supported CPUs ignore at least the first two bits of addresses in real mode, allowing direct-map addresses to be used in real-mode handlers. This is required by Linux and is part of the architecture standard starting in POWER ISA 3, so can be relied upon. The kernel continues to execute from addresses below the (new) direct map with a 1:1 mapping. This is a consequence of the early bootstrap code, which will be addressed in a follow-up commit on systems that can handle it (all but Apple hardware when running with OF rather than FDT). Because the amount of fallout from moving this was surprisingly large, testing on a diverse set of systems would be appreciated. So far, this is tested only on Apple hardware (working well) and PS3 (possible buglikewise).