From: Kohei Enju <enju.kohei@fujitsu.com>
To: Yeoreum Yun <yeoreum.yun@arm.com>
Cc: linux-kernel@vger.kernel.org,
linux-arm-kernel@lists.infradead.org,
Lorenzo Pieralisi <lpieralisi@kernel.org>,
Catalin Marinas <catalin.marinas@arm.com>,
Suzuki Poulose <suzuki.poulose@arm.com>,
Steven Price <steven.price@arm.com>,
Sami Mujawar <sami.mujawar@arm.com>,
"Aneesh Kumar K.V" <aneesh.kumar@kernel.org>,
Jiri Pirko <jiri@resnulli.us>,
GONG Ruiqi <gongruiqi1@huawei.com>,
Jason Gunthorpe <jgg@ziepe.ca>
Subject: Re: [PATCH RFC 1/3] firmware: arm_firme: introduce arm_firme driver
Date: Thu, 22 Oct 2026 18:48:49 +0900 [thread overview]
Message-ID: <atnVCFWZLKR8GthP@FCCLS0092175.localdomain> (raw)
In-Reply-To: <20261019-firme_interface-v1-1-27895c9cebaa@arm.com>
On 10/19 15:14, Yeoreum Yun wrote:
> The Realm Management Extension (FEAT_RME) is part of Arm Confidential
> Compute Architecture (CCA) and enables support for dynamic, attestable
> trusted execution environments ("Realms") on Arm systems.
>
> Realm management flows depend on EL3 firmware for operations such as
> Realm construction, resource allocation and isolation, and attestation
> of initial Realm state.
>
> The FIRME defines standard interfaces provided by EL3 firmware to
> satisfy these dependencies. These interfaces include:
>
> - Granule Protection Table management
> - IDE key management for PCIe and CXL Root Ports
> - Memory Encryption Context management
> - Attestation token management
> - Integrated device management
>
> Each group of interfaces forms a FIRME service. A base FIRME service
> is also defined to provide common concepts, service discovery, and
> feature discovery across services.
>
> FIRME interfaces are implemented at the EL3 boundary between the
> Non-secure, Realm, and Secure states. They may also be used on systems
> without FEAT_RME, for example to provide firmware-based attestation or
> security services.
>
> arm_firme driver provides the common FIRME framework and service discovery
> support for client drivers using below FIRME services:
>
> - Base service.
> - Attestation token management.
>
> Signed-off-by: Yeoreum Yun <yeoreum.yun@arm.com>
Does this series assume that the kernel is running in EL2 implicitly?
The FIRME spec says:
An implementation of one or more FIRME services at a valid Exception level boundary between EL3 and the
highest Exception level in a different Security state is called a FIRME instance.
and I think only NS-EL2 can call FIRME interfaces as a NS FIRME
instance, right?
> ---
> drivers/firmware/Kconfig | 1 +
> drivers/firmware/Makefile | 1 +
> drivers/firmware/arm_firme/Kconfig | 34 ++++++
> drivers/firmware/arm_firme/Makefile | 5 +
> drivers/firmware/arm_firme/attest.c | 172 ++++++++++++++++++++++++++
> drivers/firmware/arm_firme/base.c | 145 ++++++++++++++++++++++
> drivers/firmware/arm_firme/driver.c | 134 +++++++++++++++++++++
> drivers/firmware/arm_firme/firme.h | 45 +++++++
> include/linux/arm-smccc.h | 1 +
> include/linux/arm_firme.h | 234 ++++++++++++++++++++++++++++++++++++
> 10 files changed, 772 insertions(+)
>
> diff --git a/drivers/firmware/Kconfig b/drivers/firmware/Kconfig
> index f1a76d482429..e0c1babea1f8 100644
> --- a/drivers/firmware/Kconfig
> +++ b/drivers/firmware/Kconfig
> @@ -297,6 +297,7 @@ config TURRIS_MOX_RWTM_KEYCTL
> endif # TURRIS_MOX_RWTM
>
> source "drivers/firmware/arm_ffa/Kconfig"
> +source "drivers/firmware/arm_firme/Kconfig"
> source "drivers/firmware/arm_rmm/Kconfig"
> source "drivers/firmware/broadcom/Kconfig"
> source "drivers/firmware/cirrus/Kconfig"
> diff --git a/drivers/firmware/Makefile b/drivers/firmware/Makefile
> index 54755db0cf6e..4ae1550ed73c 100644
> --- a/drivers/firmware/Makefile
> +++ b/drivers/firmware/Makefile
> @@ -25,6 +25,7 @@ obj-$(CONFIG_TRUSTED_FOUNDATIONS) += trusted_foundations.o
> obj-$(CONFIG_TURRIS_MOX_RWTM) += turris-mox-rwtm.o
>
> obj-y += arm_ffa/
> +obj-y += arm_firme/
> obj-y += arm_rmm/
> obj-y += arm_scmi/
> obj-y += broadcom/
> diff --git a/drivers/firmware/arm_firme/Kconfig b/drivers/firmware/arm_firme/Kconfig
> new file mode 100644
> index 000000000000..7e7a826879f1
> --- /dev/null
> +++ b/drivers/firmware/arm_firme/Kconfig
> @@ -0,0 +1,34 @@
> +# SPDX-License-Identifier: GPL-2.0-only
> +config ARM_FIRME
> + tristate "Arm Firmware Interfaces for Realm Management Extension (FIRME)"
> + depends on ARM64 && HAVE_ARM_SMCCC_DISCOVERY
> + default n
> + help
> + The Realm Management Extension (FEAT_RME) is part of Arm Confidential
> + Compute Architecture (CCA) and enables support for dynamic, attestable
> + trusted execution environments ("Realms") on Arm systems.
> +
> + Realm management flows depend on EL3 firmware for operations such as
> + Realm construction, resource allocation and isolation, and attestation
> + of initial Realm state.
> +
> + The FIRME defines standard interfaces provided by EL3 firmware to
> + satisfy these dependencies. These interfaces include:
> +
> + - Granule Protection Table management
> + - IDE key management for PCIe and CXL Root Ports
> + - Memory Encryption Context management
> + - Attestation token management
> + - Integrated device management
> +
> + Each group of interfaces forms a FIRME service. A base FIRME service
> + is also defined to provide common concepts, service discovery, and
> + feature discovery across services.
> +
> + FIRME interfaces are implemented at the EL3 boundary between the
> + Non-secure, Realm, and Secure states. They may also be used on
> + systems without FEAT_RME, for example to provide firmware-based
> + attestation or security services.
> +
> + This driver provides the common FIRME framework and service
> + discovery support for client drivers using FIRME services.
> diff --git a/drivers/firmware/arm_firme/Makefile b/drivers/firmware/arm_firme/Makefile
> new file mode 100644
> index 000000000000..dccac0a2b50e
> --- /dev/null
> +++ b/drivers/firmware/arm_firme/Makefile
> @@ -0,0 +1,5 @@
> +# SPDX-License-Identifier: GPL-2.0-only
> +
> +obj-$(CONFIG_ARM_FIRME) += arm-firme.o
> +
> +arm-firme-$(CONFIG_ARM_FIRME) := driver.o base.o attest.o
> diff --git a/drivers/firmware/arm_firme/attest.c b/drivers/firmware/arm_firme/attest.c
> new file mode 100644
> index 000000000000..aa57b3b9e623
> --- /dev/null
> +++ b/drivers/firmware/arm_firme/attest.c
> @@ -0,0 +1,172 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright (C) 2026 Arm Ltd.
> + */
> +
> +#define DRIVER_NAME "ARM FIRME"
> +#define pr_fmt(fmt) DRIVER_NAME ": " fmt
> +
> +#include <crypto/sha2.h>
> +
> +#include <linux/device.h>
> +#include <linux/kernel.h>
> +#include <linux/mm.h>
> +
> +#include "firme.h"
> +
> +static DEFINE_STATIC_KEY_TRUE(attest_service_disabled);
Regarding the attest_service_disabled naming, it seems a bit
counter-intuitive to me. Wouldn't it be clearer to invert the logic
with:
DEFINE_STATIC_KEY_FALSE(attest_service_available)?
Or is there any reason to choose the naming here?
> +
> +int firme_attest_service_init(struct firme_service_info *info)
> +{
> + firme_value_t ret;
> + u8 max_pat_pgcnt;
> +
> + if (!test_bit(FIRME_ATTEST_SERVICE_ID, &info->service_list)) {
> + pr_info("Attestation service is not supported\n");
> + return 0;
> + }
> +
> + invoke_firme_service((firme_value_t) {
> + .a0 = FIRME_SERVICE_VERSION,
> + .a1 = FIRME_ATTEST_SERVICE_ID,
> + }, &ret);
> + if (ret.a0 == FIRME_RET_NOT_SUPPORTED)
> + goto err_out;
> +
> + info->attest_info.version = (u32)ret.a0;
> +
> + invoke_firme_service((firme_value_t) {
> + .a0 = FIRME_SERVICE_FEATURES,
> + .a1 = FIRME_ATTEST_SERVICE_ID,
> + .a2 = 0x1,
> + }, &ret);
> + if (ret.a0 != FIRME_RET_SUCCESS)
> + goto err_out;
> +
> + max_pat_pgcnt = FIELD_GET(FIRME_ATTEST_FEAT1_MAX_PAT_PG_CNT_MASK, ret.a1);
> +
> + /*
> + * According to Realm Management Monitor specification, S-PNCRK,
> + * maximum chanllege size should be 64 bytes (SHA512_DIGEST_SIZE).
> + * If less than, pad with zero.
> + */
> + info->attest_info.max_challenge_size = SHA512_DIGEST_SIZE;
> + info->attest_info.max_pat_size = (max_pat_pgcnt + 1) *
> + info->shared_buf_info.min_size;
> +
> + pr_info("FIRME attest service version %d.%d\n",
> + FIRME_MAJOR_VERSION(info->attest_info.version),
> + FIRME_MINOR_VERSION(info->attest_info.version));
The FIRME specification defines versioning semantics so we can negotiate
the required version of a specific service here. Is this intentional to
omit negotiation here?
The same applies to firme_base_service_init().
> + pr_info("Maximum challenge size: 0x%lx\n",
> + info->attest_info.max_challenge_size);
> + pr_info("Maximum PAT size: 0x%lx\n", info->attest_info.max_pat_size);
> +
> + static_branch_disable(&attest_service_disabled);
> +
> + return 0;
> +
> +err_out:
> + static_branch_enable(&attest_service_disabled);
> + clear_bit(FIRME_ATTEST_SERVICE_ID, &info->service_list);
> + return firme_status_to_errno(ret.a0);
> +}
> +
> +static int firme_get_attest_info(struct firme_attest_info *info)
> +{
> + if (static_branch_unlikely(&attest_service_disabled))
> + return -EOPNOTSUPP;
> +
> + if (!info)
> + return -EINVAL;
> +
> + memcpy(info, &firme_service_info->attest_info,
> + sizeof(struct firme_attest_info));
> +
> + return 0;
> +}
> +
> +static int firme_get_platform_attestion_token(void *challenge,
> + size_t challenge_size,
> + void *token, size_t *token_size)
> +{
> + firme_status_t status;
> + firme_value_t ret;
> + struct firme_instance *fi;
> + struct firme_attest_info *attest_info;
> + struct firme_shared_buf_info *shbuf_info;
> + void *buffer;
> + size_t alloc_size, buffer_size, buffer_pgcnt;
> + size_t read_size, total_size;
> +
> + if (static_branch_unlikely(&attest_service_disabled))
> + return -EOPNOTSUPP;
> +
> + shbuf_info = &firme_service_info->shared_buf_info;
> + attest_info = &firme_service_info->attest_info;
> +
> + if (!token || !token_size || (*token_size < attest_info->max_pat_size))
> + return -EINVAL;
> +
> + if ((challenge_size > attest_info->max_challenge_size) ||
> + ((challenge_size > 0) && !challenge))
> + return -EINVAL;
> +
> + fi = firme_get_instance();
> + if (WARN_ON_ONCE(IS_ERR (fi)))
> + return -EOPNOTSUPP;
> +
> + buffer_size = min(attest_info->max_pat_size, shbuf_info->max_size);
> + if (WARN_ON_ONCE(buffer_size & (shbuf_info->min_size - 1)))
> + return -ENOBUFS;
> +
> + alloc_size = buffer_size;
> +
> + buffer = fi->ops->base_ops->alloc_shared_buffer(&alloc_size,
> + GFP_KERNEL | __GFP_ZERO);
> + if (!buffer)
> + return -ENOMEM;
> +
> + if (alloc_size < buffer_size)
> + buffer_size = alloc_size;
> +
> + buffer_pgcnt = buffer_size / shbuf_info->min_size;
> +
> + if (challenge_size > 0)
> + memcpy(buffer, challenge, challenge_size);
> +
> + guard(migrate)();
I don't think disabling migration is required here.
> +
> + total_size = 0;
> +
> + do {
> + invoke_firme_service((firme_value_t) {
> + .a0 = FIRME_ATTEST_PAT_GET,
> + .a1 = virt_to_phys(buffer),
> + .a2 = 0,
> + .a3 = buffer_pgcnt - 1,
> + .a4 = challenge_size,
> + .a5 = 0, /**< binding size must be 0 for normal world. */
> + }, &ret);
> + status = ret.a0;
> + if ((status != FIRME_RET_SUCCESS) &&
> + (status != FIRME_RET_INCOMPLETE))
> + break;
> +
> + challenge_size = 0;
> + read_size = ret.a1;
> + memcpy(token + total_size, buffer, read_size);
> + total_size += read_size;
> + } while (status == FIRME_RET_INCOMPLETE);
> +
> + if (status == FIRME_RET_SUCCESS)
> + *token_size = total_size;
> +
> + fi->ops->base_ops->free_shared_buffer(buffer, alloc_size);
> +
> + return firme_status_to_errno(status);
> +}
> +
> +struct firme_attest_ops attest_ops = {
> + .get_attest_info = firme_get_attest_info,
> + .get_platform_attestion_token = firme_get_platform_attestion_token,
> +};
> diff --git a/drivers/firmware/arm_firme/base.c b/drivers/firmware/arm_firme/base.c
> new file mode 100644
> index 000000000000..b9ffb214cca4
> --- /dev/null
> +++ b/drivers/firmware/arm_firme/base.c
> @@ -0,0 +1,145 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright (C) 2026 Arm Ltd.
> + */
> +
> +#define DRIVER_NAME "ARM FIRME"
> +#define pr_fmt(fmt) DRIVER_NAME ": " fmt
> +
> +#include <linux/device.h>
> +#include <linux/kernel.h>
> +#include <linux/mm.h>
> +
> +#include "firme.h"
> +
> +int firme_base_service_init(struct firme_service_info *info)
> +{
> + firme_value_t ret;
> + unsigned long service_list;
> + u8 shbuf_min_pgs;
> + u16 max_pgcnt;
> + size_t min_size;
> +
> + invoke_firme_service((firme_value_t) {
> + .a0 = FIRME_SERVICE_VERSION,
> + .a1 = FIRME_BASE_SERVICE_ID,
> + }, &ret);
> + if (ret.a0 == FIRME_RET_NOT_SUPPORTED)
> + goto err_out;
> +
> + info->version = (u32)ret.a0;
> +
> + invoke_firme_service((firme_value_t) {
> + .a0 = FIRME_SERVICE_FEATURES,
> + .a1 = FIRME_BASE_SERVICE_ID,
> + .a2 = 0x1,
> + }, &ret);
> + if (ret.a0 != FIRME_RET_SUCCESS)
> + goto err_out;
> +
> + service_list = FIELD_GET(FIRME_BASE_FEAT1_SERVICE_LIST_MASK, ret.a1);
> +
> + /* Base service is always supported */
> + info->service_list = (service_list << 1) | 0x1;
> +
> + shbuf_min_pgs = FIELD_GET(FIRME_BASE_FEAT1_SH_BUF_MIN_SZ_MASK, ret.a1);
> + max_pgcnt = FIELD_GET(FIRME_BASE_FEAT1_SH_BUF_MAX_PG_CNT_MASK, ret.a1);
> +
> + switch (shbuf_min_pgs) {
> + case FIRME_PGS_4K:
> + min_size = 1UL << 12;
nit:
I'd like to use SZ_4K, SZ_16K, SZ_64K here.
> + break;
> + case FIRME_PGS_16K:
> + min_size = 1UL << 14;
> + break;
> + case FIRME_PGS_64K:
> + min_size = 1UL << 16;
> + break;
> + default:
> + pr_err("Invalid minimum shared buffer size: 0x%x\n", shbuf_min_pgs);
> + return -EINVAL;
> + }
next prev parent reply other threads:[~2026-10-22 9:49 UTC|newest]
Thread overview: 19+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-10-19 14:14 [PATCH RFC 0/3] firmware: arm_firme: introduce FIRME support for host attestation Yeoreum Yun
2026-10-19 14:14 ` [PATCH RFC 1/3] firmware: arm_firme: introduce arm_firme driver Yeoreum Yun
2026-10-22 9:48 ` Kohei Enju [this message]
2026-10-19 14:14 ` [PATCH RFC 2/3] firmware: arm-smccc: add FIRME device Yeoreum Yun
2026-10-19 14:14 ` [PATCH RFC 3/3] firmware: arm-firme: TSM_REPORT support using FIRME Yeoreum Yun
2026-10-19 23:35 ` [PATCH RFC 0/3] firmware: arm_firme: introduce FIRME support for host attestation Jason Gunthorpe
2026-10-20 7:13 ` Yeoreum Yun
2026-10-20 12:45 ` Jason Gunthorpe
2026-10-20 13:41 ` Yeoreum Yun
2026-10-20 13:50 ` Jason Gunthorpe
2026-10-20 14:13 ` Yeoreum Yun
2026-10-20 14:36 ` Jason Gunthorpe
2026-10-20 14:43 ` Yeoreum Yun
2026-10-20 14:58 ` Jason Gunthorpe
2026-10-20 15:18 ` Yeoreum Yun
2026-10-20 16:47 ` Jason Gunthorpe
2026-10-20 17:05 ` Yeoreum Yun
2026-10-20 18:47 ` Jason Gunthorpe
2026-10-20 19:27 ` Yeoreum Yun
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=atnVCFWZLKR8GthP@FCCLS0092175.localdomain \
--to=enju.kohei@fujitsu.com \
--cc=aneesh.kumar@kernel.org \
--cc=catalin.marinas@arm.com \
--cc=gongruiqi1@huawei.com \
--cc=jgg@ziepe.ca \
--cc=jiri@resnulli.us \
--cc=linux-arm-kernel@lists.infradead.org \
--cc=linux-kernel@vger.kernel.org \
--cc=lpieralisi@kernel.org \
--cc=sami.mujawar@arm.com \
--cc=steven.price@arm.com \
--cc=suzuki.poulose@arm.com \
--cc=yeoreum.yun@arm.com \
/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
all inboxes | Powered by JetHome®