mail archive of the barebox mailing list
 help / color / mirror / Atom feed
From: Michael Tretter <m.tretter@pengutronix.de>
To: BAREBOX <barebox@lists.infradead.org>,
	 Sascha Hauer <s.hauer@pengutronix.de>,
	 Marco Felsch <m.felsch@pengutronix.de>
Cc: Michael Tretter <m.tretter@pengutronix.de>
Subject: [PATCH v2 8/9] ARM: rockchip: pass device tree to TF-A
Date: Wed, 28 May 2025 16:11:26 +0200	[thread overview]
Message-ID: <20250528-rk3588-optee-v2-8-63070238dd13@pengutronix.de> (raw)
In-Reply-To: <20250528-rk3588-optee-v2-0-63070238dd13@pengutronix.de>

The upstream OP-TEE expects a device tree to be able to initialize the
dynamic shared memory. Therefore, barebox should pass a device tree that
contains memory nodes through the TF-A to OP-TEE.

OP-TEE may modify the passed fdt. Thus, barebox copies the fdt from the
fixed size rodata area to the rk_scratch area and allocates a
configurable memory area for the fdt to allow fdt modification.

OP-TEE has the CFG_DTB_MAX_SIZE config option for the maximum size of
the fdt. barebox must reserve at least that much memory for the fdt to
avoid out of bounds accesses from OP-TEE.

Unfortunately, the downstream TF-A fails to start if barebox passes its
device tree and it is not possible to detect if the loaded TF-A is able
to handle the device tree. Add a config option to pass the device tree
if it is possible.

Signed-off-by: Michael Tretter <m.tretter@pengutronix.de>
---
Changes in v2:
- pass copy of fdt in scratch space to TF-A
- add config option for fdt size
- add documentation for CONFIG_ARCH_ROCKCHIP_ATF_PASS_FDT
---
 Documentation/boards/rockchip.rst |  4 ++++
 arch/arm/mach-rockchip/Kconfig    | 23 +++++++++++++++++++++++
 arch/arm/mach-rockchip/atf.c      | 24 ++++++++++++------------
 include/mach/rockchip/bootrom.h   |  3 +++
 4 files changed, 42 insertions(+), 12 deletions(-)

diff --git a/Documentation/boards/rockchip.rst b/Documentation/boards/rockchip.rst
index b2b04abb03cd1926bb59799af5f6a8c11d410cc2..8bce92865e3eda193412180c7295b5a35b3531e8 100644
--- a/Documentation/boards/rockchip.rst
+++ b/Documentation/boards/rockchip.rst
@@ -94,6 +94,10 @@ With these barebox can be compiled as:
 .. note:: The RK3566 and RK3568 seem to share the bl31 and bl32 firmware files,
   whereas the memory initialization blob is different.
 
+.. note:: The bl31 from the rkbin repository seems to be unable to handle
+  device trees of a larger size (for example, if CONFIG_OF_OVERLAY_LIVE is
+  enabled). Disable CONFIG_ARCH_ROCKCHIP_ATF_PASS_FDT in this case.
+
 Creating a bootable SD card
 ---------------------------
 
diff --git a/arch/arm/mach-rockchip/Kconfig b/arch/arm/mach-rockchip/Kconfig
index 98dfd11c182b9fee6e3c958653ad4fa8a7d98d84..1b72dc4b8a16ce9dca742ff1173fc2208a0e619d 100644
--- a/arch/arm/mach-rockchip/Kconfig
+++ b/arch/arm/mach-rockchip/Kconfig
@@ -141,6 +141,29 @@ config ARCH_ROCKCHIP_ATF
 	  useful for debugging early startup, but for all other cases,
 	  say y here.
 
+config ARCH_ROCKCHIP_ATF_PASS_FDT
+	bool "Pass device tree to TF-A"
+	depends on ARCH_ROCKCHIP_ATF
+	help
+	  Enable this option if you are using an upstream OP-TEE that uses the
+	  device tree to initialize dynamic shared memory, which is passed
+	  through the upstream TF-A.
+
+	  Disable the option if you are using a downstream TF-A since it
+	  doesn't always cope with device trees. Supposedly this happens if the
+	  device tree is too large, for example if CONFIG_OF_OVERLAY_LIVE is
+	  enabled.
+
+config ARCH_ROCKCHIP_ATF_FDT_SIZE
+	hex
+	default 0x0
+	default 0x60000 if ARCH_ROCKCHIP_ATF_PASS_FDT
+	prompt "Reserved size for the FDT blob passed to the TF-A"
+	help
+	  Set this size to CFG_DTB_MAX_SIZE in the OP-TEE configuration. OP-TEE
+	  may modify the passed device tree and increase it's size. This
+	  ensures that barebox reserves enough memory for the modifications.
+
 config ARCH_ROCKCHIP_OPTEE
 	bool "Build rockchip OP-TEE binary into barebox"
 	depends on ARCH_ROCKCHIP_ATF
diff --git a/arch/arm/mach-rockchip/atf.c b/arch/arm/mach-rockchip/atf.c
index cfa6df043b34c0f36919048237c7ecf33dfe0724..f4a71ef2dc8dffc6ceb4f97ee542f5b83858120b 100644
--- a/arch/arm/mach-rockchip/atf.c
+++ b/arch/arm/mach-rockchip/atf.c
@@ -183,21 +183,21 @@ void __noreturn rk3588_barebox_entry(void *fdt)
 	rk_scratch = (void *)arm_mem_scratch(endmem);
 
 	if (current_el() == 3) {
+		void *fdt_scratch = NULL;
+
 		rk3588_lowlevel_init();
 		rockchip_store_bootrom_iram(IOMEM(RK3588_IRAM_BASE));
 
-		/*
-		 * The downstream TF-A doesn't cope with our device tree when
-		 * CONFIG_OF_OVERLAY_LIVE is enabled, supposedly because it is
-		 * too big for some reason. Otherwise it doesn't have any visible
-		 * effect if we pass a device tree or not, except that the TF-A
-		 * fills in the ethernet MAC address into the device tree.
-		 * The upstream TF-A doesn't use the device tree at all.
-		 *
-		 * Pass NULL for now until we have a good reason to pass a real
-		 * device tree.
-		 */
-		rk3588_atf_load_bl31(NULL);
+#ifdef CONFIG_ARCH_ROCKCHIP_ATF_PASS_FDT
+		pr_debug("Copy fdt to scratch area 0x%p (%zu bytes)\n",
+			 rk_scratch->fdt, sizeof(rk_scratch->fdt));
+		if (fdt_open_into(fdt, rk_scratch->fdt, sizeof(rk_scratch->fdt)) == 0)
+			fdt_scratch = rk_scratch->fdt;
+		else
+			pr_warn("Failed to copy fdt to scratch: Continue without fdt\n");
+#endif
+
+		rk3588_atf_load_bl31(fdt_scratch);
 		/* not reached when CONFIG_ARCH_ROCKCHIP_ATF */
 	}
 
diff --git a/include/mach/rockchip/bootrom.h b/include/mach/rockchip/bootrom.h
index 6776ac5ef9813de903f936e340afd72ecd417c82..586008a78505943c968301a9ce90ed5e4aa9b18b 100644
--- a/include/mach/rockchip/bootrom.h
+++ b/include/mach/rockchip/bootrom.h
@@ -13,7 +13,10 @@
 struct rockchip_scratch_space {
 	u32 irom[16];
 	struct optee_header optee_hdr;
+	/* FDT needs an 8 byte alignment */
+	u8 fdt[CONFIG_ARCH_ROCKCHIP_ATF_FDT_SIZE] __aligned(8);
 };
+static_assert(sizeof(struct rockchip_scratch_space) <= CONFIG_SCRATCH_SIZE);
 
 extern struct rockchip_scratch_space *rk_scratch;
 

-- 
2.39.5




  parent reply	other threads:[~2025-05-28 14:20 UTC|newest]

Thread overview: 13+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-05-28 14:11 [PATCH v2 0/9] ARM: rockchip: fix dynamic shared memory in OP-TEE Michael Tretter
2025-05-28 14:11 ` [PATCH v2 1/9] ARM: rockchip: fix formatting Michael Tretter
2025-05-28 14:11 ` [PATCH v2 2/9] ARM: rockchip: dmc: use RK3588_INT_REG_START for rk3588 Michael Tretter
2025-05-28 14:11 ` [PATCH v2 3/9] lib: fdt: add fdt_addresses Michael Tretter
2025-05-28 14:11 ` [PATCH v2 4/9] PBL: fdt: refactor helper for reading nr of cells Michael Tretter
2025-05-28 14:11 ` [PATCH v2 5/9] PBL: fdt: add fdt_fixup_mem to fixup memory nodes Michael Tretter
2025-05-28 14:11 ` [PATCH v2 6/9] ARM: add CONFIG_SCRATCH_SIZE Michael Tretter
2025-05-28 15:34   ` Marco Felsch
2025-05-28 14:11 ` [PATCH v2 7/9] ARM: rockchip: dmc: add rk3588_ram_sizes to get full ram size Michael Tretter
2025-05-28 14:11 ` Michael Tretter [this message]
2025-05-28 15:39   ` [PATCH v2 8/9] ARM: rockchip: pass device tree to TF-A Marco Felsch
2025-05-28 14:11 ` [PATCH v2 9/9] ARM: rockchip: fixup memory in device tree for TF-A Michael Tretter
2025-05-28 15:40   ` Marco Felsch

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=20250528-rk3588-optee-v2-8-63070238dd13@pengutronix.de \
    --to=m.tretter@pengutronix.de \
    --cc=barebox@lists.infradead.org \
    --cc=m.felsch@pengutronix.de \
    --cc=s.hauer@pengutronix.de \
    /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