Add RPC caller and endpoint for FF-A

Adds RPC caller and endpoint using FF-A messaging: using direct requests
and shared memory buffers for communication.

The endpoint is only for running in SPs. The caller is implemented in
two flavours: for Linux and for SPs. The Linux version relies on the
FF-A user space interface described in external/ffa_tool/buf_common.h

Change-Id: I3b51bcaf114d06bbf3dad2d63529fa59d8977e19
Signed-off-by: Balint Dobszay <balint.dobszay@arm.com>
diff --git a/components/rpc/ffarpc/endpoint/component.cmake b/components/rpc/ffarpc/endpoint/component.cmake
new file mode 100644
index 0000000..9e43ac5
--- /dev/null
+++ b/components/rpc/ffarpc/endpoint/component.cmake
@@ -0,0 +1,14 @@
+#-------------------------------------------------------------------------------
+# Copyright (c) 2020, Arm Limited and Contributors. All rights reserved.
+#
+# SPDX-License-Identifier: BSD-3-Clause
+#
+#-------------------------------------------------------------------------------
+if (NOT DEFINED TGT)
+	message(FATAL_ERROR "mandatory parameter TGT is not defined.")
+endif()
+
+target_sources(${TGT} PRIVATE
+	"${CMAKE_CURRENT_LIST_DIR}/ffarpc_call_ep.c"
+	)
+
diff --git a/components/rpc/ffarpc/endpoint/ffarpc_call_args.h b/components/rpc/ffarpc/endpoint/ffarpc_call_args.h
new file mode 100644
index 0000000..402e4f5
--- /dev/null
+++ b/components/rpc/ffarpc/endpoint/ffarpc_call_args.h
@@ -0,0 +1,40 @@
+/*
+ * Copyright (c) 2020, Arm Limited and Contributors. All rights reserved.
+ *
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+
+#ifndef FFA_CALL_ARGS_H
+#define FFA_CALL_ARGS_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/*
+ * Defines convention for use of FFA direct call arguments by
+ * the SP service framework. This header file is used by the
+ * normal world RPC caller and the RPC listener in the SP.
+ */
+
+/* Common req & resp arg offests into msg_args structure */
+#define FFA_CALL_ARGS_OPCODE			    (0)
+
+/* Req arg offsets */
+#define FFA_CALL_ARGS_REQ_DATA_LEN		    (1)
+
+/* Resp arg offsets */
+#define FFA_CALL_ARGS_RESP_DATA_LEN		    (1)
+#define FFA_CALL_ARGS_RESP_RPC_STATUS	    (2)
+#define FFA_CALL_ARGS_RESP_OP_STATUS		(3)
+
+/* Share/unshare offsets */
+#define FFA_CALL_ARGS_SHARE_MEM_HANDLE_LSW	(1)
+#define FFA_CALL_ARGS_SHARE_MEM_HANDLE_MSW	(2)
+#define FFA_CALL_ARGS_SHARE_MEM_SIZE		(3)
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* FFA_CALL_ARGS_H */
diff --git a/components/rpc/ffarpc/endpoint/ffarpc_call_ep.c b/components/rpc/ffarpc/endpoint/ffarpc_call_ep.c
new file mode 100644
index 0000000..bb40cf3
--- /dev/null
+++ b/components/rpc/ffarpc/endpoint/ffarpc_call_ep.c
@@ -0,0 +1,182 @@
+/*
+ * Copyright (c) 2020, Arm Limited and Contributors. All rights reserved.
+ *
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+
+#include "ffarpc_call_args.h"
+#include "ffarpc_call_ep.h"
+#include "ffarpc_call_ops.h"
+#include <ffa_api.h>
+#include <sp_memory_management.h>
+#include <protocols/rpc/common/packed-c/status.h>
+#include <trace.h>
+#include <stddef.h>
+
+/* TODO: remove this when own ID will be available in libsp */
+extern uint16_t own_id;
+
+static void set_resp_args(uint32_t *resp_args, uint32_t opcode, uint32_t data_len,
+			  rpc_status_t rpc_status, uint32_t opstatus)
+{
+	resp_args[FFA_CALL_ARGS_OPCODE] = opcode;
+	resp_args[FFA_CALL_ARGS_RESP_DATA_LEN] = data_len;
+	resp_args[FFA_CALL_ARGS_RESP_RPC_STATUS] = rpc_status;
+	resp_args[FFA_CALL_ARGS_RESP_OP_STATUS] = opstatus;
+}
+
+static void set_mgmt_resp_args(uint32_t *resp_args, uint32_t opcode,
+			       rpc_status_t rpc_status)
+{
+	/*
+	 * Sets arguments for responses that originate from the ffa_call_ep
+	 * rather than from a higher layer service. These responses are not
+	 * associated with a shared buffer for any additional message payload.
+	 */
+	set_resp_args(resp_args, opcode, 0, rpc_status, 0);
+}
+
+static void init_shmem_buf(struct ffa_call_ep *call_ep, uint16_t source_id,
+			   const uint32_t *req_args, uint32_t *resp_args)
+{
+	sp_result sp_res = SP_RESULT_INTERNAL_ERROR;
+	struct sp_memory_descriptor desc = { };
+	struct sp_memory_access_descriptor acc_desc = { };
+	struct sp_memory_region region = { };
+	uint32_t in_region_count = 1;
+	uint32_t out_region_count = 1;
+	uint64_t handle = 0;
+	rpc_status_t rpc_status = TS_RPC_ERROR_INTERNAL;
+
+	desc.sender_id = source_id;
+	desc.memory_type = sp_memory_type_not_specified;
+	desc.flags.transaction_type = sp_memory_transaction_type_share;
+	acc_desc.receiver_id = own_id;
+	acc_desc.data_access = sp_data_access_read_write;
+	handle = req_args[FFA_CALL_ARGS_SHARE_MEM_HANDLE_MSW];
+	handle = (handle << 32) | req_args[FFA_CALL_ARGS_SHARE_MEM_HANDLE_LSW];
+
+	sp_res = sp_memory_retrieve(&desc, &acc_desc, &region, in_region_count,
+				    &out_region_count, handle);
+
+	if (sp_res == SP_RESULT_OK) {
+		call_ep->shmem_buf = region.address;
+		call_ep->shmem_buf_handle = handle;
+		call_ep->shmem_buf_size = (size_t)req_args[FFA_CALL_ARGS_SHARE_MEM_SIZE];
+		rpc_status = TS_RPC_CALL_ACCEPTED;
+	} else {
+		EMSG("memory retrieve error: %d", sp_res);
+	}
+
+	set_mgmt_resp_args(resp_args, req_args[FFA_CALL_ARGS_OPCODE], rpc_status);
+}
+
+static void deinit_shmem_buf(struct ffa_call_ep *call_ep, const uint32_t *req_args,
+			     uint32_t *resp_args)
+{
+	sp_result sp_res = SP_RESULT_INTERNAL_ERROR;
+	rpc_status_t rpc_status = TS_RPC_ERROR_INTERNAL;
+	uint64_t handle = call_ep->shmem_buf_handle;
+	uint16_t endpoints[1] = { own_id };
+	uint32_t endpoint_cnt = 1;
+	struct sp_memory_transaction_flags flags = {
+		.zero_memory = false,
+		.operation_time_slicing = false,
+	};
+
+	sp_res = sp_memory_relinquish(handle, endpoints, endpoint_cnt, &flags);
+	if (sp_res == SP_RESULT_OK) {
+		call_ep->shmem_buf = NULL;
+		call_ep->shmem_buf_handle = 0;
+		call_ep->shmem_buf_size = 0;
+		rpc_status = TS_RPC_CALL_ACCEPTED;
+	} else {
+		EMSG("memory relinquish error: %d", sp_res);
+	}
+
+	set_mgmt_resp_args(resp_args, req_args[FFA_CALL_ARGS_OPCODE], rpc_status);
+}
+
+static void handle_service_msg(struct ffa_call_ep *call_ep, uint16_t source_id,
+			       const uint32_t *req_args, uint32_t *resp_args)
+{
+	rpc_status_t rpc_status;
+	struct call_req call_req;
+
+	call_req.caller_id = source_id;
+	call_req.opcode = req_args[FFA_CALL_ARGS_OPCODE];
+
+	call_req.req_buf.data = call_ep->shmem_buf;
+	call_req.req_buf.data_len = req_args[FFA_CALL_ARGS_REQ_DATA_LEN];
+	call_req.req_buf.size = call_ep->shmem_buf_size;
+
+	call_req.resp_buf.data = call_ep->shmem_buf;
+	call_req.resp_buf.data_len = 0;
+	call_req.resp_buf.size = call_ep->shmem_buf_size;
+
+	rpc_status = call_ep_receive(call_ep->call_ep, &call_req);
+
+	set_resp_args(resp_args,
+		      call_req.opcode,
+		      call_req.resp_buf.data_len,
+		      rpc_status,
+		      call_req.opstatus);
+}
+
+static void handle_mgmt_msg(struct ffa_call_ep *call_ep, uint16_t source_id,
+			    const uint32_t *req_args, uint32_t *resp_args)
+{
+	uint32_t opcode = req_args[FFA_CALL_ARGS_OPCODE];
+
+	/*
+	 * TODO: shouldn't this be used to keep track of multiple
+	 * shared buffers for different endpoints?
+	 */
+	(void)source_id;
+
+	switch (opcode) {
+	case FFA_CALL_OPCODE_SHARE_BUF:
+		init_shmem_buf(call_ep, source_id, req_args, resp_args);
+		break;
+	case FFA_CALL_OPCODE_UNSHARE_BUF:
+		deinit_shmem_buf(call_ep, req_args, resp_args);
+		break;
+	default:
+		set_mgmt_resp_args(resp_args, opcode, TS_RPC_ERROR_INVALID_OPCODE);
+		break;
+	}
+}
+
+void ffa_call_ep_init(struct ffa_call_ep *ffa_call_ep, struct call_ep *call_ep)
+{
+	ffa_call_ep->call_ep = call_ep;
+	ffa_call_ep->shmem_buf_handle = 0;
+	ffa_call_ep->shmem_buf_size = 0;
+	ffa_call_ep->shmem_buf = NULL;
+}
+
+void ffa_call_ep_receive(struct ffa_call_ep *call_ep,
+			 const struct ffa_direct_msg *req_msg,
+			 struct ffa_direct_msg *resp_msg)
+{
+	const uint32_t *req_args = req_msg->args;
+	uint32_t *resp_args = resp_msg->args;
+
+	uint16_t source_id = req_msg->source_id;
+	uint32_t opcode = req_args[FFA_CALL_ARGS_OPCODE];
+
+	if (FFA_CALL_OPCODE_IS_MGMT(opcode)) {
+		/* It's an RPC layer management request */
+		handle_mgmt_msg(call_ep, source_id, req_args, resp_args);
+	} else {
+		/*
+		 * Assume anything else is a service request. Service requests
+		 * rely on a buffer being shared from the requesting client.
+		 * If it hasn't been set-up, fail the request.
+		 */
+		if (call_ep->shmem_buf)
+			handle_service_msg(call_ep, source_id, req_args, resp_args);
+		else
+			set_mgmt_resp_args(resp_args, opcode, TS_RPC_ERROR_NOT_READY);
+	}
+}
diff --git a/components/rpc/ffarpc/endpoint/ffarpc_call_ep.h b/components/rpc/ffarpc/endpoint/ffarpc_call_ep.h
new file mode 100644
index 0000000..b4bcdd6
--- /dev/null
+++ b/components/rpc/ffarpc/endpoint/ffarpc_call_ep.h
@@ -0,0 +1,35 @@
+/*
+ * Copyright (c) 2020, Arm Limited and Contributors. All rights reserved.
+ *
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+
+#ifndef FFA_CALL_EP_H
+#define FFA_CALL_EP_H
+
+#include <ffa_api.h>
+#include <components/rpc/common/endpoint/call_ep.h>
+#include <stddef.h>
+#include <stdint.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+struct ffa_call_ep {
+	struct call_ep *call_ep;
+	unsigned long shmem_buf_handle;
+	volatile uint8_t *shmem_buf;
+	size_t shmem_buf_size;
+ };
+
+void ffa_call_ep_init(struct ffa_call_ep *ffa_call_ep, struct call_ep *call_ep);
+void ffa_call_ep_receive(struct ffa_call_ep *call_ep,
+			 const struct ffa_direct_msg *req_msg,
+			 struct ffa_direct_msg *resp_msg);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* FFA_CALL_EP_H */
diff --git a/components/rpc/ffarpc/endpoint/ffarpc_call_ops.h b/components/rpc/ffarpc/endpoint/ffarpc_call_ops.h
new file mode 100644
index 0000000..3f14fc7
--- /dev/null
+++ b/components/rpc/ffarpc/endpoint/ffarpc_call_ops.h
@@ -0,0 +1,27 @@
+/*
+ * Copyright (c) 2020, Arm Limited and Contributors. All rights reserved.
+ *
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+
+#ifndef FFA_CALL_OPS_H
+#define FFA_CALL_OPS_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* Common opcodes used by the FFA based RPC layer for management operations */
+#define FFA_CALL_OPCODE_BASE		(0x10)
+#define FFA_CALL_OPCODE_SHARE_BUF	(FFA_CALL_OPCODE_BASE + 0)
+#define FFA_CALL_OPCODE_UNSHARE_BUF	(FFA_CALL_OPCODE_BASE + 1)
+#define FFA_CALL_OPCODE_LIMIT		(FFA_CALL_OPCODE_BASE + 2)
+
+#define FFA_CALL_OPCODE_IS_MGMT(opcode) \
+	((opcode >= FFA_CALL_OPCODE_BASE) && (opcode < FFA_CALL_OPCODE_LIMIT))
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* FFA_CALL_OPS_H */