Re: [RFC PATCH 00/12] Introduce sv48 support without relocable kernel

From: Palmer Dabbelt
Date: Tue Feb 02 2021 - 22:23:13 EST

On Sat, 30 Jan 2021 01:33:20 PST (-0800), alex@xxxxxxxx wrote:
Hi Palmer,

On 1/4/21 2:58 PM, Alexandre Ghiti wrote:
This patchset, contrary to the previous versions, allows to have a single
kernel for sv39 and sv48 without being relocatable.

The idea comes from Arnd Bergmann who suggested to do the same as x86,
that is mapping the kernel to the end of the address space, which allows
the kernel to be linked at the same address for both sv39 and sv48 and
then does not require to be relocated at runtime.

This is an RFC because I need to at least rebase a few commits and add
documentation. The most interesting patches where I expect feedbacks are
1/12, 2/12 and 8/12. Note that moving the kernel out of the linear
mapping and sv48 support can be separate patchsets, I share them together
today to show that it works (this patchset is rebased on top of v5.10).

If we agree about the overall idea, I'll rebase my relocatable patchset
on top of that and then KASLR implementation from Zong will be greatly
simplified since moving the kernel out of the linear mapping will avoid
to copy the kernel physically.

This implements sv48 support at runtime. The kernel will try to
boot with 4-level page table and will fallback to 3-level if the HW does not
support it. Folding the 4th level into a 3-level page table has almost no
cost at runtime.

Finally, the user can now ask for sv39 explicitly by using the device-tree
which will reduce memory footprint and reduce the number of memory accesses
in case of TLB miss.

Alexandre Ghiti (12):
riscv: Move kernel mapping outside of linear mapping
riscv: Protect the kernel linear mapping
riscv: Get rid of compile time logic with MAX_EARLY_MAPPING_SIZE
riscv: Allow to dynamically define VA_BITS
riscv: Simplify MAXPHYSMEM config
riscv: Prepare ptdump for vm layout dynamic addresses
asm-generic: Prepare for riscv use of pud_alloc_one and pud_free
riscv: Implement sv48 support
riscv: Allow user to downgrade to sv39 when hw supports sv48
riscv: Use pgtable_l4_enabled to output mmu type in cpuinfo
riscv: Explicit comment about user virtual address space size
riscv: Improve virtual kernel memory layout dump

arch/riscv/Kconfig | 34 +--
arch/riscv/boot/loader.lds.S | 3 +-
arch/riscv/include/asm/csr.h | 3 +-
arch/riscv/include/asm/fixmap.h | 3 +
arch/riscv/include/asm/page.h | 33 ++-
arch/riscv/include/asm/pgalloc.h | 40 +++
arch/riscv/include/asm/pgtable-64.h | 104 ++++++-
arch/riscv/include/asm/pgtable.h | 68 +++--
arch/riscv/include/asm/sparsemem.h | 6 +-
arch/riscv/kernel/cpu.c | 23 +-
arch/riscv/kernel/head.S | 6 +-
arch/riscv/kernel/module.c | 4 +-
arch/riscv/kernel/vmlinux.lds.S | 3 +-
arch/riscv/mm/context.c | 2 +-
arch/riscv/mm/init.c | 376 ++++++++++++++++++++----
arch/riscv/mm/physaddr.c | 2 +-
arch/riscv/mm/ptdump.c | 56 +++-
drivers/firmware/efi/libstub/efi-stub.c | 2 +-
include/asm-generic/pgalloc.h | 24 +-
include/linux/sizes.h | 3 +-
20 files changed, 648 insertions(+), 147 deletions(-)


Any thought about the idea ? Is it going in the right direction ? I have
fixed quite a few things since I posted this so don't bother giving this
patchset a full review.

My only real issue was the relocation stuff, which appears to be fixed. I haven't had the time to look at the patches, but it wouldn't hurt to send another revision. The best bet might be to just wait until 5.11 and send on top of that, it's too late for this one anyway and that'll be a stable base to test from.