Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add support for 32-bit RISC-V #116

Merged
merged 12 commits into from
Sep 5, 2024
Merged

Add support for 32-bit RISC-V #116

merged 12 commits into from
Sep 5, 2024

Conversation

josecm
Copy link
Member

@josecm josecm commented Dec 10, 2023

This PR introduces support for RV32. Because the way the instruction set is written the difference is not much when comparing to RV64. The main areas of modification are:

  • modifying assembly instructions that used 64-bit load/store directly by the LOAD/STORE macro which translates to either ld/st or lw/sw depending on if we are targetting RV64 or RV32, respectively;
  • support for the sv32 translation mode. This also requires adapting the boot code where the bootstrap page-tables are built manually.
  • adapt CSR access to CSRs that have a H variant (e.g., henvcfg, henvcfgh), and other small related issues;
  • fix small issues that assumed 64-bit literals.

The PR also includes support for qemu-system-riscv32 virt machine, which is essentially a wrapper over the 64-bit variant.

The main issue with this port is related to the fact that we are required to map all available physical memory to perform page-table walks. Because the address space is so small (4GiB), this might not be possible as for RV64 we simply identity map all physical memory and assume these addresses won't collide with the high virtual addresses used by the hypervisor. This is not the case and we might not be able to map all physical addresses, which already happens for qemu-riscv32-virt if we configure it with memory larger than 1 GiB, since it starts memory at 0x80000000. We might need a dynamic mechanism similar to Linux's "highmem" in the future. I only think this is worth it if a real platform that requires it becomes available. Also, the main purpose of supporting RV32 is to target future MMU-less systems.

Test guest:

  • Baremetal app
  • Linux

Depends on #115

@danielRep danielRep force-pushed the fix/aarch32_vm_install branch from d043ba8 to f84b327 Compare January 5, 2024 11:38
@josecm josecm force-pushed the fix/aarch32_vm_install branch from f84b327 to 7189a89 Compare April 15, 2024 10:35
@josecm josecm force-pushed the fix/aarch32_vm_install branch from 7189a89 to 2603780 Compare May 23, 2024 14:39
@josecm josecm force-pushed the feat/rv32 branch 2 times, most recently from 0423b1d to 072caef Compare May 23, 2024 14:57
@josecm josecm force-pushed the fix/aarch32_vm_install branch from d719f48 to 86498c1 Compare May 23, 2024 15:04
@josecm josecm force-pushed the feat/rv32 branch 3 times, most recently from 7775309 to 4a7f2c3 Compare May 23, 2024 15:11
@danielRep danielRep deleted the branch main May 23, 2024 17:37
@danielRep danielRep closed this May 23, 2024
@josecm josecm reopened this May 24, 2024
@josecm josecm changed the base branch from fix/aarch32_vm_install to main May 24, 2024 09:25
@josecm
Copy link
Member Author

josecm commented May 24, 2024

I have booted 32-bit Linux over Bao too.

I think this is ready to be fully reviewed @ninolomata @DavidMCerdeira @danielRep

@danielRep danielRep self-requested a review July 4, 2024 10:03
@danielRep danielRep self-assigned this Jul 4, 2024
@DavidMCerdeira DavidMCerdeira force-pushed the main branch 2 times, most recently from 59a293c to 8837766 Compare July 16, 2024 10:47
@danielRep
Copy link
Member

I still couldn't finish my review , but as a first note we still need a PR on bao-demos to support 32-bit RISC-V as a means to test this PR.

Copy link
Member

@danielRep danielRep left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

While I don't see a huge problem on using compilation macros macros (RV32 and RV64), wouldn't be more cleaner to separate rv32 and rv64 profiles on different arch folders as we do for aarch32 and aarch64?

src/arch/riscv/vm.c Show resolved Hide resolved
src/core/mmu/vmm.c Outdated Show resolved Hide resolved
@josecm
Copy link
Member Author

josecm commented Aug 19, 2024

A branch for bao-demos featuring demos targetting qemu-riscv32-virt is available at https://github.com/josecm/bao-demos/tree/feat/rv32. I'll create a PR for that demo once this PR is accepted.

@josecm
Copy link
Member Author

josecm commented Aug 19, 2024

While I don't see a huge problem on using compilation macros macros (RV32 and RV64), wouldn't be more cleaner to separate rv32 and rv64 profiles on different arch folders as we do for aarch32 and aarch64?

The difference between rv32 and rv64 is not as stark as from aarch32 to aarch64, so it does not seem necessary to introduce both architectures separately, only a few of those RV32/RV64 macros and a wrapper around csr access so that rv32 can access "64-bit" csrs is enough and avoids a lot of code duplication and indirection.

@danielRep
Copy link
Member

danielRep commented Aug 19, 2024

While I don't see a huge problem on using compilation macros macros (RV32 and RV64), wouldn't be more cleaner to separate rv32 and rv64 profiles on different arch folders as we do for aarch32 and aarch64?

The difference between rv32 and rv64 is not as stark as from aarch32 to aarch64, so it does not seem necessary to introduce both architectures separately, only a few of those RV32/RV64 macros and a wrapper around csr access so that rv32 can access "64-bit" csrs is enough and avoids a lot of code duplication and indirection.

Seems reasonable. Ok, looks good to me.

@danielRep
Copy link
Member

danielRep commented Aug 19, 2024

A branch for bao-demos featuring demos targetting qemu-riscv32-virt is available at https://github.com/josecm/bao-demos/tree/feat/rv32. I'll create a PR for that demo once this PR is accepted.

Good! I'll further test if everything is working correctly on that PR then.

@danielRep danielRep merged commit c331298 into main Sep 5, 2024
15 checks passed
@danielRep danielRep deleted the feat/rv32 branch September 5, 2024 06:57
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants