From: Johannes Schauer Marin Rodrigues <josch@debian.org>
To: Ahmad Fatoum <a.fatoum@barebox.org>, barebox@lists.infradead.org
Cc: Ahmad Fatoum <a.fatoum@barebox.org>
Subject: Re: [PATCH 2/2] efi: runtime: accept 1:1 mappings in SetVirtualAddressMap
Date: Sat, 17 Oct 2026 16:26:29 +0200 [thread overview]
Message-ID: <179224718900.684046.2866764153232417065@localhost> (raw)
In-Reply-To: <20261017115018.1202076-2-a.fatoum@barebox.org>
[-- Attachment #1: Type: text/plain, Size: 6366 bytes --]
Tested-by: Johannes Schauer Marin Rodrigues <josch@mister-muffin.de>
Quoting Ahmad Fatoum (2026-10-17 13:50:15)
> We do not support SetVirtualAddressMap() and UEFI v2.8 permits
> this as long as our EFI_RT_PROPERTIES_TABLE doesn't claim
> EFI_RT_SUPPORTED_SET_VIRTUAL_ADDRESS_MAP, which we adhere to[1].
>
> NetBSD's efiboot doesn't seem to care for lack of this bit and
> calls SetVirtualAddressMap() unconditionally[2]. This will return with
> an error, which has NetBSD call printf() _after_ ExitBootServices[3]
> leading to a crash[4].
>
> Fortunately, barebox is not the only firmware that doesn't implement
> a full-fledged SetVirtualAddressMap().
> Quoting NetBSD commit 41096fd ("efiboot: Use 1:1 PA to VA mappings for
> RT services.")[5]:
>
> | UEFI firmware on Snapdragon X Elite devices crashes when
> | SetVirtualAddressMap is called with runtime mappings that are not 1:1.
> | Always do this on arm64 as it won't interfere with KVA and will help us
> | avoid avoid a very common class of firmware bugs.
>
> This means we have an easy way for barebox to workaround this:
> Instead of returning EFI_UNSUPPORTED unconditionally, we can handle the
> trivial identity mapping case specially and report EFI_SUCCESS for it.
>
> While NetBSD doesn't need it, also implement a 1:1 ConvertPointer for
> good measure.
>
> [1]: https://uefi.org/specs/UEFI/2.10/04_EFI_System_Table.html
> [2]: https://github.com/NetBSD/src/blob/41096fd/sys/stand/efiboot/efiboot.c#L135
> [3]: https://github.com/NetBSD/src/blob/41096fd/sys/stand/efiboot/efifdt.c#L813
> [4]: https://gnats.netbsd.org/60890
> [5]: https://github.com/NetBSD/src/commit/41096fd4f7bac2ade3f29e9e6e00951aef9e16ce
>
> Reported-by: Johannes Schauer Marin Rodrigues <josch@debian.org>
> Signed-off-by: Ahmad Fatoum <a.fatoum@barebox.org>
> ---
> efi/runtime/services.c | 67 ++++++++++++++++++++++++++++++++++++++++--
> include/efi/memory.h | 6 ++++
> include/efi/services.h | 2 ++
> 3 files changed, 73 insertions(+), 2 deletions(-)
>
> diff --git a/efi/runtime/services.c b/efi/runtime/services.c
> index dc4960ac4763..ca60ae512a2d 100644
> --- a/efi/runtime/services.c
> +++ b/efi/runtime/services.c
> @@ -171,17 +171,80 @@ static efi_status_t EFIAPI efirt_set_wakeup_time(char enabled,
> return efirt_unimplemented();
> }
>
> +/**
> + * efirt_set_virtual_address_map() - change from physical to virtual mapping
> + *
> + * This function implements the SetVirtualAddressMap() runtime service.
> + * The barebox runtime isn't relocated and only an identity mapping
> + * is accepted.
> + *
> + * We intentionally don't announce support via the EFI_RT_PROPERTIES_TABLE,
> + * so only operating systems that ignore the table would end up here.
> + * If they configure a 1:1 mapping like NetBSD, we tolerate it and return
> + * success.
> + *
> + * See the Unified Extensible Firmware Interface (UEFI) specification for
> + * details.
> + *
> + * @memory_map_size: size of the virtual map
> + * @descriptor_size: size of an entry in the map
> + * @descriptor_version: version of the map entries
> + * @virtmap: virtual address mapping information
> + * Return: status code
> + */
> static efi_status_t EFIAPI efirt_set_virtual_address_map(size_t memory_map_size,
> size_t descriptor_size,
> uint32_t descriptor_version,
> struct efi_memory_desc *virtmap)
> {
> - return efirt_unimplemented();
> + static bool efirt_virtmap_set;
> +
> + EFI_ENTRY(L"");
> +
> + if (efirt_virtmap_set)
> + return EFI_EXIT(EFI_UNSUPPORTED);
> +
> + if (descriptor_version != EFI_MEMORY_DESCRIPTOR_VERSION ||
> + descriptor_size < sizeof(*virtmap))
> + return EFI_EXIT(EFI_INVALID_PARAMETER);
> +
> + for (size_t off = 0; off < memory_map_size; off += descriptor_size) {
> + struct efi_memory_desc *md = (void *)virtmap + off;
> +
> + if ((md->attrs & EFI_MEMORY_RUNTIME) &&
> + efi_memory_desc_relocation_offset(md) != 0)
> + return EFI_EXIT(EFI_UNSUPPORTED);
> + }
> +
> + efirt_virtmap_set = true;
> +
> + return EFI_EXIT(EFI_SUCCESS);
> }
>
> +/**
> + * efirt_convert_pointer() - convert pointer to virtual address
> + *
> + * This function implements the ConvertPointer() runtime service.
> + * Only identity mappings are accepted, so the pointer is left as-is.
> + *
> + * See the Unified Extensible Firmware Interface (UEFI) specification for
> + * details.
> + *
> + * @dbg: EFI_OPTIONAL_PTR if the pointer may be NULL
> + * @address: pointer to be converted
> + * Return: status code
> + */
> static efi_status_t EFIAPI efirt_convert_pointer(unsigned long dbg, void **address)
> {
> - return efirt_unimplemented();
> + EFI_ENTRY(L"");
> +
> + if (!address)
> + return EFI_EXIT(EFI_INVALID_PARAMETER);
> +
> + if (!*address && !(dbg & EFI_OPTIONAL_PTR))
> + return EFI_EXIT(EFI_INVALID_PARAMETER);
> +
> + return EFI_EXIT(EFI_SUCCESS);
> }
>
> static efi_status_t EFIAPI
> diff --git a/include/efi/memory.h b/include/efi/memory.h
> index 74e714099021..336d2a730c1c 100644
> --- a/include/efi/memory.h
> +++ b/include/efi/memory.h
> @@ -91,4 +91,10 @@ struct efi_memory_desc {
>
> #define EFI_MEMORY_DESCRIPTOR_VERSION 1
>
> +static inline ptrdiff_t
> +efi_memory_desc_relocation_offset(const struct efi_memory_desc *md)
> +{
> + return efi_virt_to_phys(md->virt_start) - md->phys_start;
> +}
> +
> #endif
> diff --git a/include/efi/services.h b/include/efi/services.h
> index 2d7167b25205..83dc56150a4e 100644
> --- a/include/efi/services.h
> +++ b/include/efi/services.h
> @@ -162,6 +162,8 @@ enum efi_reset_type {
> #define EFI_RUNTIME_SERVICES_SIGNATURE ((u64)0x56524553544e5552ULL)
> #define EFI_RUNTIME_SERVICES_REVISION 0x00010000
>
> +#define EFI_OPTIONAL_PTR 0x00000001
> +
> struct efi_capsule_header;
>
> struct efi_runtime_services {
> --
> 2.47.3
>
[-- Attachment #2: signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]
-----BEGIN PGP SIGNATURE-----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=r+XL
-----END PGP SIGNATURE-----
next prev parent reply other threads:[~2026-10-17 14:27 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-10-17 11:50 [PATCH 1/2] efi: loader: in-place patch the runtime services on ExitBootServices Ahmad Fatoum
2026-10-17 11:50 ` [PATCH 2/2] efi: runtime: accept 1:1 mappings in SetVirtualAddressMap Ahmad Fatoum
2026-10-17 14:26 ` Johannes Schauer Marin Rodrigues [this message]
2026-10-17 14:26 ` [PATCH 1/2] efi: loader: in-place patch the runtime services on ExitBootServices Johannes Schauer Marin Rodrigues
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=179224718900.684046.2866764153232417065@localhost \
--to=josch@debian.org \
--cc=a.fatoum@barebox.org \
--cc=barebox@lists.infradead.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox