[PATCH v3 00/13] Introduce sv48 support without relocatable kernel

Palmer Dabbelt palmer at dabbelt.com
Wed Jun 1 20:43:59 PDT 2022 

On Fri, 22 Apr 2022 18:50:47 PDT (-0700), Palmer Dabbelt wrote:
> On Fri, 01 Apr 2022 05:56:30 PDT (-0700), alexandre.ghiti at canonical.com wrote:
>> On Fri, Feb 18, 2022 at 11:45 AM Alexandre Ghiti
>> <alexandre.ghiti at canonical.com> wrote:
>>>
>>> Hi Palmer,
>>>
>>> On Thu, Jan 20, 2022 at 11:05 AM Alexandre Ghiti
>>> <alexandre.ghiti at canonical.com> wrote:
>>> >
>>> > On Thu, Jan 20, 2022 at 8:30 AM Alexandre Ghiti
>>> > <alexandre.ghiti at canonical.com> wrote:
>>> > >
>>> > > On Thu, Jan 20, 2022 at 5:18 AM Palmer Dabbelt <palmer at dabbelt.com> wrote:
>>> > > >
>>> > > > On Mon, 06 Dec 2021 02:46:44 PST (-0800), alexandre.ghiti at canonical.com wrote:
>>> > > > > * Please note notable changes in memory layouts and kasan population *
>>> > > > >
>>> > > > > This patchset 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 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.
>>> > > > >
>>> > > > > Note that kasan region had to be moved to the end of the address space
>>> > > > > since its location must be known at compile-time and then be valid for
>>> > > > > both sv39 and sv48 (and sv57 that is coming).
>>> > > > >
>>> > > > > Tested on:
>>> > > > >   - qemu rv64 sv39: OK
>>> > > > >   - qemu rv64 sv48: OK
>>> > > > >   - qemu rv64 sv39 + kasan: OK
>>> > > > >   - qemu rv64 sv48 + kasan: OK
>>> > > > >   - qemu rv32: OK
>>> > > > >
>>> > > > > Changes in v3:
>>> > > > >   - Fix SZ_1T, thanks to Atish
>>> > > > >   - Fix warning create_pud_mapping, thanks to Atish
>>> > > > >   - Fix k210 nommu build, thanks to Atish
>>> > > > >   - Fix wrong rebase as noted by Samuel
>>> > > > >   - * Downgrade to sv39 is only possible if !KASAN (see commit changelog) *
>>> > > > >   - * Move KASAN next to the kernel: virtual layouts changed and kasan population *
>>> > > > >
>>> > > > > Changes in v2:
>>> > > > >   - Rebase onto for-next
>>> > > > >   - Fix KASAN
>>> > > > >   - Fix stack canary
>>> > > > >   - Get completely rid of MAXPHYSMEM configs
>>> > > > >   - Add documentation
>>> > > > >
>>> > > > > Alexandre Ghiti (13):
>>> > > > >   riscv: Move KASAN mapping next to the kernel mapping
>>> > > > >   riscv: Split early kasan mapping to prepare sv48 introduction
>>> > > > >   riscv: Introduce functions to switch pt_ops
>>> > > > >   riscv: Allow to dynamically define VA_BITS
>>> > > > >   riscv: Get rid of MAXPHYSMEM configs
>>> > > > >   asm-generic: Prepare for riscv use of pud_alloc_one and pud_free
>>> > > > >   riscv: Implement sv48 support
>>> > > > >   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
>>> > > > >   Documentation: riscv: Add sv48 description to VM layout
>>> > > > >   riscv: Initialize thread pointer before calling C functions
>>> > > > >   riscv: Allow user to downgrade to sv39 when hw supports sv48 if !KASAN
>>> > > > >
>>> > > > >  Documentation/riscv/vm-layout.rst             |  48 ++-
>>> > > > >  arch/riscv/Kconfig                            |  37 +-
>>> > > > >  arch/riscv/configs/nommu_k210_defconfig       |   1 -
>>> > > > >  .../riscv/configs/nommu_k210_sdcard_defconfig |   1 -
>>> > > > >  arch/riscv/configs/nommu_virt_defconfig       |   1 -
>>> > > > >  arch/riscv/include/asm/csr.h                  |   3 +-
>>> > > > >  arch/riscv/include/asm/fixmap.h               |   1
>>> > > > >  arch/riscv/include/asm/kasan.h                |  11 +-
>>> > > > >  arch/riscv/include/asm/page.h                 |  20 +-
>>> > > > >  arch/riscv/include/asm/pgalloc.h              |  40 ++
>>> > > > >  arch/riscv/include/asm/pgtable-64.h           | 108 ++++-
>>> > > > >  arch/riscv/include/asm/pgtable.h              |  47 +-
>>> > > > >  arch/riscv/include/asm/sparsemem.h            |   6 +-
>>> > > > >  arch/riscv/kernel/cpu.c                       |  23 +-
>>> > > > >  arch/riscv/kernel/head.S                      |   4 +-
>>> > > > >  arch/riscv/mm/context.c                       |   4 +-
>>> > > > >  arch/riscv/mm/init.c                          | 408 ++++++++++++++----
>>> > > > >  arch/riscv/mm/kasan_init.c                    | 250 ++++++++---
>>> > > > >  drivers/firmware/efi/libstub/efi-stub.c       |   2
>>> > > > >  drivers/pci/controller/pci-xgene.c            |   2 +-
>>> > > > >  include/asm-generic/pgalloc.h                 |  24 +-
>>> > > > >  include/linux/sizes.h                         |   1
>>> > > > >  22 files changed, 833 insertions(+), 209 deletions(-)
>>> > > >
>>> > > > Sorry this took a while.  This is on for-next, with a bit of juggling: a
>>> > > > handful of trivial fixes for configs that were failing to build/boot and
>>> > > > some merge issues.  I also pulled out that MAXPHYSMEM fix to the top, so
>>> > > > it'd be easier to backport.  This is bigger than something I'd normally like to
>>> > > > take late in the cycle, but given there's a lot of cleanups, likely some fixes,
>>> > > > and it looks like folks have been testing this I'm just going to go with it.
>>> > > >
>>> > >
>>> > > Yes yes yes! That's fantastic news :)
>>> > >
>>> > > > Let me know if there's any issues with the merge, it was a bit hairy.
>>> > > > Probably best to just send along a fixup patch at this point.
>>> > >
>>> > > I'm going to take a look at that now, and I'll fix anything that comes
>>> > > up quickly :)
>>> >
>>> > I see in for-next that you did not take the following patches:
>>> >
>>> >   riscv: Improve virtual kernel memory layout dump
>>> >   Documentation: riscv: Add sv48 description to VM layout
>>> >   riscv: Initialize thread pointer before calling C functions
>>> >   riscv: Allow user to downgrade to sv39 when hw supports sv48 if !KASAN
>>> >
>>> > I'm not sure this was your intention. If it was, I believe that at
>>> > least the first 2 patches are needed in this series, the 3rd one is a
>>> > useful fix and we can discuss the 4th if that's an issue for you.
>>>
>>> Can you confirm that this was intentional and maybe explain the
>>> motivation behind it? Because I see value in those patches.
>>
>> Palmer,
>>
>> I read that you were still taking patches for 5.18, so I confirm again
>> that the patches above are needed IMO.
>
> It was too late for this when it was sent (I saw it then, but just got
> around to actually doing the work to sort it out).
>
> It took me a while to figure out exactly what was going on here, but I
> think I remember now: that downgrade patch (and the follow-on I just
> sent) is broken for medlow, because mm/init.c must be built medany
> (which we're using for the mostly-PIC qualities).  I remember being in
> the middle of rebasing/debugging this a while ago, I must have forgotten
> I was in the middle of that and accidentally merged the branch as-is.
> Certainly wasn't trying to silently take half the patch set and leave
> the rest in limbo, that's the wrong way to do things.
>
> I'm not sure what the right answer is here, but I just sent a patch to
> drop support for medlow.  We'll have to talk about that, for now I
> cleaned up some other minor issues, rearranged that docs and fix to come
> first, and put this at palmer/riscv-sv48.  I think that fix is
> reasonable to take the doc and fix into fixes, then the dump improvement
> on for-next.  We'll have to see what folks think about the medany-only
> kernels, the other option would be to build FDT as medany which seems a
> bit awkward.

All but the last one are on for-next, there's some discussion on that 
last one that pointed out some better ways to do it.

>
>> Maybe even the relocatable series?
>
> Do you mind giving me a pointer?  I'm not sure why I'm so drop-prone
> with your patches, I promise I'm not doing it on purpose.
>
>>
>> Thanks,
>>
>> Alex
>>
>>>
>>> Thanks,
>>>
>>> Alex
>>>
>>> >
>>> > I tested for-next on both sv39 and sv48 successfully, I took a glance
>>> > at the code and noticed you fixed the PTRS_PER_PGD error, thanks for
>>> > that. Otherwise nothing obvious has popped.
>>> >
>>> > Thanks again,
>>> >
>>> > Alex
>>> >
>>> > >
>>> > > Thanks!
>>> > >
>>> > > Alex
>>> > >
>>> > > >
>>> > > > Thanks!

More information about the linux-riscv mailing list