core: ffa: add direct request v2 support by sah01Kaushal · Pull Request #7840 · OP-TEE/optee_os

sah01Kaushal · 2026-06-11T19:33:35Z

Add support for FF-A direct request/response v2 messages for secure partitions.

Direct request v2 carries the target service UUID in x2/x3 and uses x4-x17 for the payload, so preserve the full AArch64 FF-A register set when entering and returning from secure partitions.

Advertise DirectReq2/Resp2 to normal-world and secure partition callers, and expose the corresponding partition properties for AArch64 secure partitions that support direct messaging. Validate DirectReq2 messages by checking sender and receiver capabilities and matching non-nil UUIDs against the destination partition UUID.

Logical secure partitions do not support DirectReq2 yet, so reject Req2 messages targeting LSPs.

jenswikl · 2026-06-12T07:02:31Z

 	}

-	if (ffa_version != FFA_VERSION_1_0 && ffa_version != FFA_VERSION_1_1) {
+	if (ffa_version != FFA_VERSION_1_0 &&


How about if (ffa_version < FFA_VERSION_1_0 || ffa_version > FFA_VERSION_1_2)?

Made the change.

jenswikl · 2026-06-12T07:21:32Z


+#ifdef ARM64
+	memcpy(ctx->sp_regs.x, args->a, sizeof(args->a));
+#else


This isn't working for ARM32, so we should replace the assignment with the memcpy instead.

Same further down in this function.

The old ARM32 path in this patch still used x[] even though ARM32 thread_ctx_regs does not have that field. I replaced it with a struct copy, and made the same change at the second copy site.

Can remove the ARM32 part entirely if you want.

Please remove the ARM32 part entirely.

Removed entirely. Thanks!

jenswikl · 2026-06-12T07:37:41Z

+static void direct_req2_uuid_from_args(struct thread_smc_1_2_regs *args,
+				       uint32_t uuid_words[4])
+{
+	uuid_words[0] = (uint32_t)args->a2;


Please use high32_from_64() and low32_from_64() to extract the words.

Made the change.

jenswikl · 2026-06-12T07:39:07Z

+			       !uuid_words[2] && !uuid_words[3]);
+}
+
+#ifdef ARM64


This file is only compiled for ARM64, so the ifdef isn't needed.

Removed the ifdefs for ARM64

jenswikl · 2026-06-12T07:58:31Z

 	struct sp_session *dst = NULL;
 	struct spmc_lsp_desc *lsp = NULL;
 	TEE_Result res = FFA_OK;
+	bool is_req2 = args->a0 == FFA_MSG_SEND_DIRECT_REQ2;


To keep it simple, I'd prefer:

uint32_t caller_props = 0; uint32_t dst_prots = 0 if (args->a0 == FFA_MSG_SEND_DIRECT_REQ2) { caller_props = FFA_PART_PROP_DIRECT_REQ2_SEND; dst_props = FFA_PART_PROP_DIRECT_REQ2_RECV; } else { caller_props = FFA_PART_PROP_DIRECT_REQ_SEND; dst_props = FFA_PART_PROP_DIRECT_REQ_RECV; }

Added this form.

jenswikl · 2026-06-12T08:18:43Z

 	}

-	if (args->a2 & FFA_MSG_FLAG_FRAMEWORK) {
+	if (is_resp2) {


if (args->a0 == FFA_MSG_SEND_DIRECT_RESP2) {

jenswikl · 2026-06-12T08:26:43Z

+static void init_regs_spmc(struct thread_ctx *thread,
+			   const struct thread_smc_1_2_regs *args, void *pc)
+{
+#ifdef ARM64


With CFG_SECURE_PARTITION y, we can depend on ARM64 to be defined.

Removed the #ifdef ARM64

jenswikl · 2026-06-12T08:27:07Z

+#ifdef ARM64
+	size_t n = 0;
+
+	thread->regs.pc = (uint64_t)pc;


Prefer casting to vaddr_t when converting a pointer into an integer.

Switched to vaddr_t

jenswikl · 2026-06-12T08:29:07Z

+	for (; n < ARRAY_SIZE(thread->regs.x); n++)
+		thread->regs.x[n] = 0;
+
+	thread->regs.x[29] = 0;


This should already be cleared in the loop above.

jenswikl · 2026-06-12T08:46:36Z

 					     FFA_PART_PROP_DIRECT_REQ_SEND |
 					     FFA_PART_PROP_INDIRECT_MSGS;
+	/* In FF-A 1.1 only bits [8:0] are defined, let's mask others */
+	else if (ffa_vers < FFA_VERSION_1_2)


This seems a bit broken. How about something like this instead?

/* mask out bits introduced with FF-A version 1.1 */ if (ffa_vers < FFA_VERSION_1_1) ... /* mask out bits introduced with FF-A version 1.2 */ if (ffa_vers < FFA_VERSION_1_2) ...

Made the switch to this form

jenswikl · 2026-06-15T11:17:33Z


-	if (ffa_version != FFA_VERSION_1_0 && ffa_version != FFA_VERSION_1_1) {
+	if (ffa_version < FFA_VERSION_1_0 ||
+	    ffa_version > FFA_VERSION_1_2) {


Please fold up, it fits on one line.

jenswikl · 2026-06-15T11:18:42Z


+#ifdef ARM64
+	memcpy(ctx->sp_regs.x, args->a, sizeof(args->a));
+#else


Please remove the ARM32 part entirely.

jenswikl · 2026-06-15T11:20:27Z

+	switch (resp_fid) {
+	case FFA_MSG_SEND_DIRECT_RESP_32:
+		return req_fid == FFA_MSG_SEND_DIRECT_REQ_32;
+#ifdef ARM64


The ifdef/endif isn't needed.

Removed the ifdef/endif

jenswikl · 2026-06-16T06:27:47Z

Looks good, please squash in the fixups and apply:
Reviewed-by: Jens Wiklander <jenswi@kernel.org>

Add support for FF-A direct request/response v2 messages for secure partitions. Direct request v2 carries the target service UUID in x2/x3 and uses x4-x17 for the payload, so preserve the full AArch64 FF-A register set when entering and returning from secure partitions. Advertise DirectReq2/Resp2 to normal-world and secure partition callers, and expose the corresponding partition properties for AArch64 secure partitions that support direct messaging. Validate DirectReq2 messages by checking sender and receiver capabilities and matching non-nil UUIDs against the destination partition UUID. Logical secure partitions do not support DirectReq2 yet, so reject Req2 messages targeting LSPs. Signed-off-by: Arteom Katkov <arteom.katkov@arm.com> Signed-off-by: Sahil Kaushal <Sahil.Kaushal@arm.com> Reviewed-by: Jens Wiklander <jenswi@kernel.org>

sah01Kaushal · 2026-06-16T14:15:01Z

Looks good, please squash in the fixups and apply: Reviewed-by: Jens Wiklander <jenswi@kernel.org>

Thank you very much for the review, Jens! I have squashed the fixups in and pushed again.

jenswikl · 2026-06-17T07:44:43Z

I'll merge this when the CI checks have passed.

jenswikl reviewed Jun 12, 2026

View reviewed changes

jenswikl reviewed Jun 15, 2026

View reviewed changes

sah01Kaushal force-pushed the directreq2-upstream branch from 0d70e5b to 73947a6 Compare June 16, 2026 14:12

jenswikl merged commit 1428386 into OP-TEE:master Jun 17, 2026
57 checks passed

Uh oh!

Conversation

sah01Kaushal commented Jun 11, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jenswikl commented Jun 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

sah01Kaushal commented Jun 16, 2026

Uh oh!

jenswikl commented Jun 17, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

jenswikl commented Jun 16, 2026 •

edited

Loading