RISC-V UEFI/ACPI on QEMU regression?

[Ilias Apalodimas]

Hey Bjorn

Long time, hope all is well!

On Wed, 2 Apr 2025 at 16:22, Björn Töpel <bjorn at kernel.org> wrote:
>
> Hi,
>
> I think I got a regression from commit 53d5a221632e ("emulation: Use
> bloblist to hold tables"), and v2024.10 for
> qemu-riscv64_smode_defconfig + acpi.config booting Linux with UEFI.
>
> TL;DR: It seems like the RSDP is placed in the wrong EFI memory map
> type (it should be "ACPI Reclaim").
>
> I might also be misunderstanding what config fragments should be
> used -- I'm out in the weeds here! ;-)
>
> When I was using v2024.10, ACPI RSDP was in ACPI Reclaim memory. All
> good, and e.g. a kexec would properly work. However, when using u-boot
> with commit 53d5a221632e ("emulation: Use bloblist to hold tables") I
> get the following when booting, and then kexec:
>
> First kernel:
> [    0.000000] ACPI: Early table checksum verification disabled
>                                                   [    0.000000] ACPI:
> RSDP 0x000000047EED3000 000024 (v02 BOCHS )
> Kexec kernel:
> [    0.000000] ACPI: Early table checksum verification disabled
> [    0.000000] ACPI:      0x000000047EED3000 000000 (v00
>   00000000      00000000)
> [    0.000000] Oops - load access fault [#1]
>
> RSDP reside in:
> [    0.000000] efi:   0x00047ded1000-0x00047eee3fff [Boot Code   |   |
>  |  |  |  |  |  |  |  |  |   |WB|  |  |  ]
>
> (Boot Code vs ACPI Reclaimed)
>
> Now to get qemu-riscv64_smode_defconfig + acpi.config to build
> post-2024.10, I needed to add the following fragments:
>
>   CONFIG_BLOBLIST=y
>   CONFIG_BLOBLIST_ALLOC=y
>   CONFIG_BLOBLIST_SIZE_RELOC=0x20000
>
> which is really just a "make the build not complain", guessing game
> from my side.
>
> My guess would be that it's related to the change in
> evt_write_acpi_tables(), where:
>
> -    ptr = memalign(SZ_4K, SZ_64K);
> +    ptr = bloblist_add(BLOBLISTT_ACPI_TABLES, SZ_64K, 12);
>
> is done.
>
> Is my config fragment broken, or is this a proper regression?

I think it's a regression and I think what breaks it is commit cfb4aa2a754ed1

Can you apply the diff below and see if it fixes it for you?

diff --git a/lib/efi_loader/efi_acpi.c b/lib/efi_loader/efi_acpi.c
index 4422b31ac6a7..afa5eee85484 100644
--- a/lib/efi_loader/efi_acpi.c
+++ b/lib/efi_loader/efi_acpi.c
@@ -34,8 +34,8 @@ efi_status_t efi_acpi_register(void)
         * add_u_boot_and_runtime(). At some point that function could create a
         * more detailed map.
         */
-       if (IS_ENABLED(CONFIG_BLOBLIST_TABLES))
-               return EFI_SUCCESS;
+       //if (IS_ENABLED(CONFIG_BLOBLIST_TABLES))
+               ////return EFI_SUCCESS;

        /* Mark space used for tables */
        start = ALIGN_DOWN(gd->arch.table_start, EFI_PAGE_MASK);

Simon, why that PR got sent with no ACKs from any of the EFI maintainers?

Thanks
/Ilias
>
>
> Thanks!
> Björn