diff --git a/include/services/el3_spmc_ffa_memory.h b/include/services/el3_spmc_ffa_memory.h
new file mode 100644
index 0000000..d4738a1
--- /dev/null
+++ b/include/services/el3_spmc_ffa_memory.h
@@ -0,0 +1,217 @@
+/*
+ * Copyright (c) 2022, ARM Limited and Contributors. All rights reserved.
+ *
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+
+#ifndef EL3_SPMC_FFA_MEM_H
+#define EL3_SPMC_FFA_MEM_H
+
+#include <assert.h>
+
+/*
+ * Subset of Arm Firmware Framework for Armv8-A
+ * (https://developer.arm.com/docs/den0077/a) needed for shared memory.
+ */
+
+/**
+ * typedef ffa_endpoint_id16_t - Endpoint ID
+ *
+ * Current implementation only supports VM IDs. FF-A spec also support stream
+ * endpoint ids.
+ */
+typedef uint16_t ffa_endpoint_id16_t;
+
+/**
+ * struct ffa_cons_mrd - Constituent memory region descriptor
+ * @address:
+ *         Start address of contiguous memory region. Must be 4K page aligned.
+ * @page_count:
+ *         Number of 4K pages in region.
+ * @reserved_12_15:
+ *         Reserve bytes 12-15 to pad struct size to 16 bytes.
+ */
+struct ffa_cons_mrd {
+	uint64_t address;
+	uint32_t page_count;
+	uint32_t reserved_12_15;
+};
+CASSERT(sizeof(struct ffa_cons_mrd) == 16, assert_ffa_cons_mrd_size_mismatch);
+
+/**
+ * struct ffa_comp_mrd - Composite memory region descriptor
+ * @total_page_count:
+ *         Number of 4k pages in memory region. Must match sum of
+ *         @address_range_array[].page_count.
+ * @address_range_count:
+ *         Number of entries in @address_range_array.
+ * @reserved_8_15:
+ *         Reserve bytes 8-15 to pad struct size to 16 byte alignment and
+ *         make @address_range_array 16 byte aligned.
+ * @address_range_array:
+ *         Array of &struct ffa_cons_mrd entries.
+ */
+struct ffa_comp_mrd {
+	uint32_t total_page_count;
+	uint32_t address_range_count;
+	uint64_t reserved_8_15;
+	struct ffa_cons_mrd address_range_array[];
+};
+CASSERT(sizeof(struct ffa_comp_mrd) == 16, assert_ffa_comp_mrd_size_mismatch);
+
+/**
+ * typedef ffa_mem_attr8_t - Memory region attributes
+ *
+ * * @FFA_MEM_ATTR_DEVICE_NGNRNE:
+ *     Device-nGnRnE.
+ * * @FFA_MEM_ATTR_DEVICE_NGNRE:
+ *     Device-nGnRE.
+ * * @FFA_MEM_ATTR_DEVICE_NGRE:
+ *     Device-nGRE.
+ * * @FFA_MEM_ATTR_DEVICE_GRE:
+ *     Device-GRE.
+ * * @FFA_MEM_ATTR_NORMAL_MEMORY_UNCACHED
+ *     Normal memory. Non-cacheable.
+ * * @FFA_MEM_ATTR_NORMAL_MEMORY_CACHED_WB
+ *     Normal memory. Write-back cached.
+ * * @FFA_MEM_ATTR_NON_SHAREABLE
+ *     Non-shareable. Combine with FFA_MEM_ATTR_NORMAL_MEMORY_*.
+ * * @FFA_MEM_ATTR_OUTER_SHAREABLE
+ *     Outer Shareable. Combine with FFA_MEM_ATTR_NORMAL_MEMORY_*.
+ * * @FFA_MEM_ATTR_INNER_SHAREABLE
+ *     Inner Shareable. Combine with FFA_MEM_ATTR_NORMAL_MEMORY_*.
+ */
+typedef uint8_t ffa_mem_attr8_t;
+#define FFA_MEM_ATTR_DEVICE_NGNRNE		((1U << 4) | (0x0U << 2))
+#define FFA_MEM_ATTR_DEVICE_NGNRE		((1U << 4) | (0x1U << 2))
+#define FFA_MEM_ATTR_DEVICE_NGRE		((1U << 4) | (0x2U << 2))
+#define FFA_MEM_ATTR_DEVICE_GRE			((1U << 4) | (0x3U << 2))
+#define FFA_MEM_ATTR_NORMAL_MEMORY_UNCACHED	((2U << 4) | (0x1U << 2))
+#define FFA_MEM_ATTR_NORMAL_MEMORY_CACHED_WB	((2U << 4) | (0x3U << 2))
+#define FFA_MEM_ATTR_NON_SHAREABLE		(0x0U << 0)
+#define FFA_MEM_ATTR_OUTER_SHAREABLE		(0x2U << 0)
+#define FFA_MEM_ATTR_INNER_SHAREABLE		(0x3U << 0)
+
+/**
+ * typedef ffa_mem_perm8_t - Memory access permissions
+ *
+ * * @FFA_MEM_ATTR_RO
+ *     Request or specify read-only mapping.
+ * * @FFA_MEM_ATTR_RW
+ *     Request or allow read-write mapping.
+ * * @FFA_MEM_PERM_NX
+ *     Deny executable mapping.
+ * * @FFA_MEM_PERM_X
+ *     Request executable mapping.
+ */
+typedef uint8_t ffa_mem_perm8_t;
+#define FFA_MEM_PERM_RO		(1U << 0)
+#define FFA_MEM_PERM_RW		(1U << 1)
+#define FFA_MEM_PERM_NX		(1U << 2)
+#define FFA_MEM_PERM_X		(1U << 3)
+
+/**
+ * typedef ffa_mem_flag8_t - Endpoint memory flags
+ *
+ * * @FFA_MEM_FLAG_NON_RETRIEVAL_BORROWER
+ *     Non-retrieval Borrower. Memory region must not be or was not retrieved on
+ *     behalf of this endpoint.
+ */
+typedef uint8_t ffa_mem_flag8_t;
+#define FFA_MEM_FLAG_NON_RETRIEVAL_BORROWER	(1U << 0)
+
+/**
+ * typedef ffa_mtd_flag32_t - Memory transaction descriptor flags
+ *
+ * * @FFA_MTD_FLAG_ZERO_MEMORY
+ *     Zero memory after unmapping from sender (must be 0 for share).
+ * * @FFA_MTD_FLAG_TIME_SLICING
+ *     Not supported by this implementation.
+ * * @FFA_MTD_FLAG_ZERO_MEMORY_AFTER_RELINQUISH
+ *     Zero memory after unmapping from borrowers (must be 0 for share).
+ * * @FFA_MTD_FLAG_TYPE_MASK
+ *     Bit-mask to extract memory management transaction type from flags.
+ * * @FFA_MTD_FLAG_TYPE_SHARE_MEMORY
+ *     Share memory transaction flag.
+ *     Used by @SMC_FC_FFA_MEM_RETRIEVE_RESP to indicate that memory came from
+ *     @SMC_FC_FFA_MEM_SHARE and by @SMC_FC_FFA_MEM_RETRIEVE_REQ to specify that
+ *     it must have.
+ * * @FFA_MTD_FLAG_ADDRESS_RANGE_ALIGNMENT_HINT_MASK
+ *     Not supported by this implementation.
+ */
+typedef uint32_t ffa_mtd_flag32_t;
+#define FFA_MTD_FLAG_ZERO_MEMORY			(1U << 0)
+#define FFA_MTD_FLAG_TIME_SLICING			(1U << 1)
+#define FFA_MTD_FLAG_ZERO_MEMORY_AFTER_RELINQUISH	(1U << 2)
+#define FFA_MTD_FLAG_TYPE_MASK				(3U << 3)
+#define FFA_MTD_FLAG_TYPE_SHARE_MEMORY			(1U << 3)
+#define FFA_MTD_FLAG_TYPE_LEND_MEMORY			(1U << 4)
+#define FFA_MTD_FLAG_ADDRESS_RANGE_ALIGNMENT_HINT_MASK	(0x1FU << 5)
+
+/**
+ * struct ffa_mapd - Memory access permissions descriptor
+ * @endpoint_id:
+ *         Endpoint id that @memory_access_permissions and @flags apply to.
+ *         (&typedef ffa_endpoint_id16_t).
+ * @memory_access_permissions:
+ *         FFA_MEM_PERM_* values or'ed together (&typedef ffa_mem_perm8_t).
+ * @flags:
+ *         FFA_MEM_FLAG_* values or'ed together (&typedef ffa_mem_flag8_t).
+ */
+struct ffa_mapd {
+	ffa_endpoint_id16_t endpoint_id;
+	ffa_mem_perm8_t memory_access_permissions;
+	ffa_mem_flag8_t flags;
+};
+CASSERT(sizeof(struct ffa_mapd) == 4, assert_ffa_mapd_size_mismatch);
+
+/**
+ * struct ffa_emad_v1_0 - Endpoint memory access descriptor.
+ * @mapd:  &struct ffa_mapd.
+ * @comp_mrd_offset:
+ *         Offset of &struct ffa_comp_mrd from start of &struct ffa_mtd_v1_0.
+ * @reserved_8_15:
+ *         Reserved bytes 8-15. Must be 0.
+ */
+struct ffa_emad_v1_0 {
+	struct ffa_mapd mapd;
+	uint32_t comp_mrd_offset;
+	uint64_t reserved_8_15;
+};
+CASSERT(sizeof(struct ffa_emad_v1_0) == 16, assert_ffa_emad_v1_0_size_mismatch);
+
+/**
+ * struct ffa_mtd_v1_0 - Memory transaction descriptor.
+ * @sender_id:
+ *         Sender endpoint id.
+ * @memory_region_attributes:
+ *         FFA_MEM_ATTR_* values or'ed together (&typedef ffa_mem_attr8_t).
+ * @reserved_3:
+ *         Reserved bytes 3. Must be 0.
+ * @flags:
+ *         FFA_MTD_FLAG_* values or'ed together (&typedef ffa_mtd_flag32_t).
+ * @handle:
+ *         Id of shared memory object. Must be 0 for MEM_SHARE or MEM_LEND.
+ * @tag:   Client allocated tag. Must match original value.
+ * @reserved_24_27:
+ *         Reserved bytes 24-27. Must be 0.
+ * @emad_count:
+ *         Number of entries in @emad. Must be 1 in current implementation.
+ *         FFA spec allows more entries.
+ * @emad:
+ *         Endpoint memory access descriptor array (see @struct ffa_emad_v1_0).
+ */
+struct ffa_mtd_v1_0 {
+	ffa_endpoint_id16_t sender_id;
+	ffa_mem_attr8_t memory_region_attributes;
+	uint8_t reserved_3;
+	ffa_mtd_flag32_t flags;
+	uint64_t handle;
+	uint64_t tag;
+	uint32_t reserved_24_27;
+	uint32_t emad_count;
+	struct ffa_emad_v1_0 emad[];
+};
+CASSERT(sizeof(struct ffa_mtd_v1_0) == 32, assert_ffa_mtd_size_v1_0_mismatch);
+
+#endif /* EL3_SPMC_FFA_MEM_H */
diff --git a/services/std_svc/spm/el3_spmc/spmc.mk b/services/std_svc/spm/el3_spmc/spmc.mk
index 9f82ccb..aa591d9 100644
--- a/services/std_svc/spm/el3_spmc/spmc.mk
+++ b/services/std_svc/spm/el3_spmc/spmc.mk
@@ -12,7 +12,8 @@
 			spmc_main.c				\
 			spmc_setup.c				\
 			logical_sp.c				\
-			spmc_pm.c)
+			spmc_pm.c				\
+			spmc_shared_mem.c)
 
 # Specify platform specific logical partition implementation.
 SPMC_LP_SOURCES  := $(addprefix ${PLAT_DIR}/, \
diff --git a/services/std_svc/spm/el3_spmc/spmc_main.c b/services/std_svc/spm/el3_spmc/spmc_main.c
index 342f55c..74c00da 100644
--- a/services/std_svc/spm/el3_spmc/spmc_main.c
+++ b/services/std_svc/spm/el3_spmc/spmc_main.c
@@ -26,6 +26,7 @@
 #include <services/spmc_svc.h>
 #include <services/spmd_svc.h>
 #include "spmc.h"
+#include "spmc_shared_mem.h"
 
 #include <platform_def.h>
 
@@ -1033,6 +1034,7 @@
 	case FFA_RXTX_MAP_SMC32:
 	case FFA_RXTX_MAP_SMC64:
 	case FFA_RXTX_UNMAP:
+	case FFA_MEM_FRAG_TX:
 	case FFA_MSG_RUN:
 
 		/*
@@ -1050,10 +1052,28 @@
 	case FFA_SECONDARY_EP_REGISTER_SMC64:
 	case FFA_MSG_SEND_DIRECT_RESP_SMC32:
 	case FFA_MSG_SEND_DIRECT_RESP_SMC64:
+	case FFA_MEM_RETRIEVE_REQ_SMC32:
+	case FFA_MEM_RETRIEVE_REQ_SMC64:
+	case FFA_MEM_RELINQUISH:
 	case FFA_MSG_WAIT:
 
 		if (!secure_origin) {
 			return spmc_ffa_error_return(handle,
+				FFA_ERROR_NOT_SUPPORTED);
+		}
+		SMC_RET1(handle, FFA_SUCCESS_SMC32);
+		/* Execution stops here. */
+
+	/* Supported features only from the normal world. */
+	case FFA_MEM_SHARE_SMC32:
+	case FFA_MEM_SHARE_SMC64:
+	case FFA_MEM_LEND_SMC32:
+	case FFA_MEM_LEND_SMC64:
+	case FFA_MEM_RECLAIM:
+	case FFA_MEM_FRAG_RX:
+
+		if (secure_origin) {
+			return spmc_ffa_error_return(handle,
 					FFA_ERROR_NOT_SUPPORTED);
 		}
 		SMC_RET1(handle, FFA_SUCCESS_SMC32);
@@ -1677,6 +1697,18 @@
 	initalize_sp_descs();
 	initalize_ns_ep_descs();
 
+	/*
+	 * Retrieve the information of the datastore for tracking shared memory
+	 * requests allocated by platform code and zero the region if available.
+	 */
+	ret = plat_spmc_shmem_datastore_get(&spmc_shmem_obj_state.data,
+					    &spmc_shmem_obj_state.data_size);
+	if (ret != 0) {
+		ERROR("Failed to obtain memory descriptor backing store!\n");
+		return ret;
+	}
+	memset(spmc_shmem_obj_state.data, 0, spmc_shmem_obj_state.data_size);
+
 	/* Setup logical SPs. */
 	ret = logical_sp_init();
 	if (ret != 0) {
@@ -1799,6 +1831,35 @@
 	case FFA_MSG_RUN:
 		return ffa_run_handler(smc_fid, secure_origin, x1, x2, x3, x4,
 				       cookie, handle, flags);
+
+	case FFA_MEM_SHARE_SMC32:
+	case FFA_MEM_SHARE_SMC64:
+	case FFA_MEM_LEND_SMC32:
+	case FFA_MEM_LEND_SMC64:
+		return spmc_ffa_mem_send(smc_fid, secure_origin, x1, x2, x3, x4,
+					 cookie, handle, flags);
+
+	case FFA_MEM_FRAG_TX:
+		return spmc_ffa_mem_frag_tx(smc_fid, secure_origin, x1, x2, x3,
+					    x4, cookie, handle, flags);
+
+	case FFA_MEM_FRAG_RX:
+		return spmc_ffa_mem_frag_rx(smc_fid, secure_origin, x1, x2, x3,
+					    x4, cookie, handle, flags);
+
+	case FFA_MEM_RETRIEVE_REQ_SMC32:
+	case FFA_MEM_RETRIEVE_REQ_SMC64:
+		return spmc_ffa_mem_retrieve_req(smc_fid, secure_origin, x1, x2,
+						 x3, x4, cookie, handle, flags);
+
+	case FFA_MEM_RELINQUISH:
+		return spmc_ffa_mem_relinquish(smc_fid, secure_origin, x1, x2,
+					       x3, x4, cookie, handle, flags);
+
+	case FFA_MEM_RECLAIM:
+		return spmc_ffa_mem_reclaim(smc_fid, secure_origin, x1, x2, x3,
+					    x4, cookie, handle, flags);
+
 	default:
 		WARN("Unsupported FF-A call 0x%08x.\n", smc_fid);
 		break;
diff --git a/services/std_svc/spm/el3_spmc/spmc_shared_mem.c b/services/std_svc/spm/el3_spmc/spmc_shared_mem.c
new file mode 100644
index 0000000..b9ca2fe
--- /dev/null
+++ b/services/std_svc/spm/el3_spmc/spmc_shared_mem.c
@@ -0,0 +1,920 @@
+/*
+ * Copyright (c) 2022, ARM Limited and Contributors. All rights reserved.
+ *
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+#include <errno.h>
+
+#include <common/debug.h>
+#include <common/runtime_svc.h>
+#include <lib/object_pool.h>
+#include <lib/spinlock.h>
+#include <lib/xlat_tables/xlat_tables_v2.h>
+#include <services/ffa_svc.h>
+#include "spmc.h"
+#include "spmc_shared_mem.h"
+
+#include <platform_def.h>
+
+/**
+ * struct spmc_shmem_obj - Shared memory object.
+ * @desc_size:      Size of @desc.
+ * @desc_filled:    Size of @desc already received.
+ * @in_use:         Number of clients that have called ffa_mem_retrieve_req
+ *                  without a matching ffa_mem_relinquish call.
+ * @desc:           FF-A memory region descriptor passed in ffa_mem_share.
+ */
+struct spmc_shmem_obj {
+	size_t desc_size;
+	size_t desc_filled;
+	size_t in_use;
+	struct ffa_mtd_v1_0 desc;
+};
+
+/*
+ * Declare our data structure to store the metadata of memory share requests.
+ * The main datastore is allocated on a per platform basis to ensure enough
+ * storage can be made available.
+ * The address of the data store will be populated by the SPMC during its
+ * initialization.
+ */
+
+struct spmc_shmem_obj_state spmc_shmem_obj_state = {
+	/* Set start value for handle so top 32 bits are needed quickly. */
+	.next_handle = 0xffffffc0U,
+};
+
+/**
+ * spmc_shmem_obj_size - Convert from descriptor size to object size.
+ * @desc_size:  Size of struct ffa_memory_region_descriptor object.
+ *
+ * Return: Size of struct spmc_shmem_obj object.
+ */
+static size_t spmc_shmem_obj_size(size_t desc_size)
+{
+	return desc_size + offsetof(struct spmc_shmem_obj, desc);
+}
+
+/**
+ * spmc_shmem_obj_alloc - Allocate struct spmc_shmem_obj.
+ * @state:      Global state.
+ * @desc_size:  Size of struct ffa_memory_region_descriptor object that
+ *              allocated object will hold.
+ *
+ * Return: Pointer to newly allocated object, or %NULL if there not enough space
+ *         left. The returned pointer is only valid while @state is locked, to
+ *         used it again after unlocking @state, spmc_shmem_obj_lookup must be
+ *         called.
+ */
+static struct spmc_shmem_obj *
+spmc_shmem_obj_alloc(struct spmc_shmem_obj_state *state, size_t desc_size)
+{
+	struct spmc_shmem_obj *obj;
+	size_t free = state->data_size - state->allocated;
+
+	if (state->data == NULL) {
+		ERROR("Missing shmem datastore!\n");
+		return NULL;
+	}
+
+	if (spmc_shmem_obj_size(desc_size) > free) {
+		WARN("%s(0x%zx) failed, free 0x%zx\n",
+		     __func__, desc_size, free);
+		return NULL;
+	}
+	obj = (struct spmc_shmem_obj *)(state->data + state->allocated);
+	obj->desc = (struct ffa_mtd_v1_0) {0};
+	obj->desc_size = desc_size;
+	obj->desc_filled = 0;
+	obj->in_use = 0;
+	state->allocated += spmc_shmem_obj_size(desc_size);
+	return obj;
+}
+
+/**
+ * spmc_shmem_obj_free - Free struct spmc_shmem_obj.
+ * @state:      Global state.
+ * @obj:        Object to free.
+ *
+ * Release memory used by @obj. Other objects may move, so on return all
+ * pointers to struct spmc_shmem_obj object should be considered invalid, not
+ * just @obj.
+ *
+ * The current implementation always compacts the remaining objects to simplify
+ * the allocator and to avoid fragmentation.
+ */
+
+static void spmc_shmem_obj_free(struct spmc_shmem_obj_state *state,
+				  struct spmc_shmem_obj *obj)
+{
+	size_t free_size = spmc_shmem_obj_size(obj->desc_size);
+	uint8_t *shift_dest = (uint8_t *)obj;
+	uint8_t *shift_src = shift_dest + free_size;
+	size_t shift_size = state->allocated - (shift_src - state->data);
+
+	if (shift_size != 0U) {
+		memmove(shift_dest, shift_src, shift_size);
+	}
+	state->allocated -= free_size;
+}
+
+/**
+ * spmc_shmem_obj_lookup - Lookup struct spmc_shmem_obj by handle.
+ * @state:      Global state.
+ * @handle:     Unique handle of object to return.
+ *
+ * Return: struct spmc_shmem_obj_state object with handle matching @handle.
+ *         %NULL, if not object in @state->data has a matching handle.
+ */
+static struct spmc_shmem_obj *
+spmc_shmem_obj_lookup(struct spmc_shmem_obj_state *state, uint64_t handle)
+{
+	uint8_t *curr = state->data;
+
+	while (curr - state->data < state->allocated) {
+		struct spmc_shmem_obj *obj = (struct spmc_shmem_obj *)curr;
+
+		if (obj->desc.handle == handle) {
+			return obj;
+		}
+		curr += spmc_shmem_obj_size(obj->desc_size);
+	}
+	return NULL;
+}
+
+static struct ffa_comp_mrd *
+spmc_shmem_obj_get_comp_mrd(struct spmc_shmem_obj *obj)
+{
+	return (struct ffa_comp_mrd *)
+		((uint8_t *)(&obj->desc) + obj->desc.emad[0].comp_mrd_offset);
+}
+
+/**
+ * spmc_shmem_obj_ffa_constituent_size - Calculate variable size part of obj.
+ * @obj:    Object containing ffa_memory_region_descriptor.
+ *
+ * Return: Size of ffa_constituent_memory_region_descriptors in @obj.
+ */
+static size_t
+spmc_shmem_obj_ffa_constituent_size(struct spmc_shmem_obj *obj)
+{
+	return spmc_shmem_obj_get_comp_mrd(obj)->address_range_count *
+		sizeof(struct ffa_cons_mrd);
+}
+
+/**
+ * spmc_shmem_check_obj - Check that counts in descriptor match overall size.
+ * @obj:    Object containing ffa_memory_region_descriptor.
+ *
+ * Return: 0 if object is valid, -EINVAL if memory region attributes count is
+ * not 1, -EINVAL if constituent_memory_region_descriptor offset or count is
+ * invalid.
+ */
+static int spmc_shmem_check_obj(struct spmc_shmem_obj *obj)
+{
+	if (obj->desc.emad_count != 1) {
+		WARN("%s: unsupported attribute desc count %u != 1\n",
+		     __func__, obj->desc.emad_count);
+		return -EINVAL;
+	}
+
+	for (size_t emad_num = 0; emad_num < obj->desc.emad_count; emad_num++) {
+		size_t size;
+		size_t count;
+		size_t expected_size;
+		size_t total_page_count;
+		struct ffa_comp_mrd *comp;
+
+		uint32_t offset = obj->desc.emad[emad_num].comp_mrd_offset;
+		size_t header_emad_size = sizeof(obj->desc) +
+			obj->desc.emad_count * sizeof(obj->desc.emad[emad_num]);
+
+		if (offset < header_emad_size) {
+			WARN("%s: invalid object, offset %u < header + emad %zu\n",
+			     __func__, offset, header_emad_size);
+			return -EINVAL;
+		}
+
+		size = obj->desc_size;
+
+		if (offset > size) {
+			WARN("%s: invalid object, offset %u > total size %zu\n",
+			     __func__, offset, obj->desc_size);
+			return -EINVAL;
+		}
+		size -= offset;
+
+		if (size < sizeof(struct ffa_comp_mrd)) {
+			WARN("%s: invalid object, offset %u, total size %zu, no header space.\n",
+			     __func__, offset, obj->desc_size);
+			return -EINVAL;
+		}
+		size -= sizeof(struct ffa_comp_mrd);
+
+		count = size / sizeof(struct ffa_cons_mrd);
+
+		comp = spmc_shmem_obj_get_comp_mrd(obj);
+
+		if (comp->address_range_count != count) {
+			WARN("%s: invalid object, desc count %u != %zu\n",
+			     __func__, comp->address_range_count, count);
+			return -EINVAL;
+		}
+
+		expected_size = offset + sizeof(*comp) +
+				spmc_shmem_obj_ffa_constituent_size(obj);
+		if (expected_size != obj->desc_size) {
+			WARN("%s: invalid object, computed size %zu != size %zu\n",
+			       __func__, expected_size, obj->desc_size);
+			return -EINVAL;
+		}
+
+		if (obj->desc_filled < obj->desc_size) {
+			/*
+			 * The whole descriptor has not yet been received.
+			 * Skip final checks.
+			 */
+			return 0;
+		}
+
+		total_page_count = 0;
+
+		for (size_t i = 0; i < count; i++) {
+			total_page_count +=
+				comp->address_range_array[i].page_count;
+		}
+		if (comp->total_page_count != total_page_count) {
+			WARN("%s: invalid object, desc total_page_count %u != %zu\n",
+			     __func__, comp->total_page_count,
+			total_page_count);
+			return -EINVAL;
+		}
+	}
+
+	return 0;
+}
+
+static long spmc_ffa_fill_desc(struct mailbox *mbox,
+			       struct spmc_shmem_obj *obj,
+			       uint32_t fragment_length,
+			       ffa_mtd_flag32_t mtd_flag,
+			       void *smc_handle)
+{
+	int ret;
+	uint32_t handle_low;
+	uint32_t handle_high;
+
+	if (mbox->rxtx_page_count == 0U) {
+		WARN("%s: buffer pair not registered.\n", __func__);
+		ret = -EINVAL;
+		goto err_arg;
+	}
+
+	if (fragment_length > mbox->rxtx_page_count * PAGE_SIZE_4KB) {
+		WARN("%s: bad fragment size %u > %u buffer size\n", __func__,
+		     fragment_length, mbox->rxtx_page_count * PAGE_SIZE_4KB);
+		ret = -EINVAL;
+		goto err_arg;
+	}
+
+	memcpy((uint8_t *)&obj->desc + obj->desc_filled,
+	       (uint8_t *) mbox->tx_buffer,
+	       fragment_length);
+
+	if (fragment_length > obj->desc_size - obj->desc_filled) {
+		WARN("%s: bad fragment size %u > %zu remaining\n", __func__,
+		     fragment_length, obj->desc_size - obj->desc_filled);
+		ret = -EINVAL;
+		goto err_arg;
+	}
+
+	/* Ensure that the sender ID resides in the normal world. */
+	if (ffa_is_secure_world_id(obj->desc.sender_id)) {
+		WARN("%s: Invalid sender ID 0x%x.\n",
+		     __func__, obj->desc.sender_id);
+		ret = FFA_ERROR_DENIED;
+		goto err_arg;
+	}
+
+	/*
+	 * We don't currently support any optional flags so ensure none are
+	 * requested.
+	 */
+	if (obj->desc.flags != 0U && mtd_flag != 0U &&
+	    (obj->desc.flags != mtd_flag)) {
+		WARN("%s: invalid memory transaction flags %u != %u\n",
+		     __func__, obj->desc.flags, mtd_flag);
+		ret = -EINVAL;
+		goto err_arg;
+	}
+
+	if (obj->desc_filled == 0U) {
+		/* First fragment, descriptor header has been copied */
+		obj->desc.handle = spmc_shmem_obj_state.next_handle++;
+		obj->desc.flags |= mtd_flag;
+	}
+
+	obj->desc_filled += fragment_length;
+
+	ret = spmc_shmem_check_obj(obj);
+	if (ret != 0) {
+		goto err_bad_desc;
+	}
+
+	handle_low = (uint32_t)obj->desc.handle;
+	handle_high = obj->desc.handle >> 32;
+
+	if (obj->desc_filled != obj->desc_size) {
+		SMC_RET8(smc_handle, FFA_MEM_FRAG_RX, handle_low,
+			 handle_high, obj->desc_filled,
+			 (uint32_t)obj->desc.sender_id << 16, 0, 0, 0);
+	}
+
+	SMC_RET8(smc_handle, FFA_SUCCESS_SMC32, 0, handle_low, handle_high, 0,
+		 0, 0, 0);
+
+err_bad_desc:
+err_arg:
+	spmc_shmem_obj_free(&spmc_shmem_obj_state, obj);
+	return spmc_ffa_error_return(smc_handle, FFA_ERROR_INVALID_PARAMETER);
+}
+
+/**
+ * spmc_ffa_mem_send - FFA_MEM_SHARE/LEND implementation.
+ * @client:             Client state.
+ * @total_length:       Total length of shared memory descriptor.
+ * @fragment_length:    Length of fragment of shared memory descriptor passed in
+ *                      this call.
+ * @address:            Not supported, must be 0.
+ * @page_count:         Not supported, must be 0.
+ * @smc_handle:         Handle passed to smc call. Used to return
+ *                      FFA_MEM_FRAG_RX or SMC_FC_FFA_SUCCESS.
+ *
+ * Implements a subset of the FF-A FFA_MEM_SHARE and FFA_MEM_LEND calls needed
+ * to share or lend memory from non-secure os to secure os (with no stream
+ * endpoints).
+ *
+ * Return: 0 on success, error code on failure.
+ */
+long spmc_ffa_mem_send(uint32_t smc_fid,
+			bool secure_origin,
+			uint64_t total_length,
+			uint32_t fragment_length,
+			uint64_t address,
+			uint32_t page_count,
+			void *cookie,
+			void *handle,
+			uint64_t flags)
+
+{
+	long ret;
+	struct spmc_shmem_obj *obj;
+	struct mailbox *mbox = spmc_get_mbox_desc(secure_origin);
+	ffa_mtd_flag32_t mtd_flag;
+
+	if (address != 0U || page_count != 0U) {
+		WARN("%s: custom memory region for message not supported.\n",
+		     __func__);
+		return spmc_ffa_error_return(handle,
+					     FFA_ERROR_INVALID_PARAMETER);
+	}
+
+	if (secure_origin) {
+		WARN("%s: unsupported share direction.\n", __func__);
+		return spmc_ffa_error_return(handle,
+					     FFA_ERROR_INVALID_PARAMETER);
+	}
+
+	if (fragment_length < sizeof(obj->desc)) {
+		WARN("%s: bad first fragment size %u < %zu\n",
+		     __func__, fragment_length, sizeof(obj->desc));
+		return spmc_ffa_error_return(handle,
+					     FFA_ERROR_INVALID_PARAMETER);
+	}
+
+	if ((smc_fid & FUNCID_NUM_MASK) == FFA_FNUM_MEM_SHARE) {
+		mtd_flag = FFA_MTD_FLAG_TYPE_SHARE_MEMORY;
+	} else if ((smc_fid & FUNCID_NUM_MASK) == FFA_FNUM_MEM_LEND) {
+		mtd_flag = FFA_MTD_FLAG_TYPE_LEND_MEMORY;
+	} else {
+		WARN("%s: invalid memory management operation.\n", __func__);
+		return spmc_ffa_error_return(handle,
+					     FFA_ERROR_INVALID_PARAMETER);
+	}
+
+	spin_lock(&spmc_shmem_obj_state.lock);
+
+	obj = spmc_shmem_obj_alloc(&spmc_shmem_obj_state, total_length);
+	if (obj == NULL) {
+		ret = FFA_ERROR_NO_MEMORY;
+		goto err_unlock;
+	}
+
+	spin_lock(&mbox->lock);
+	ret = spmc_ffa_fill_desc(mbox, obj, fragment_length, mtd_flag, handle);
+	spin_unlock(&mbox->lock);
+
+	spin_unlock(&spmc_shmem_obj_state.lock);
+	return ret;
+
+err_unlock:
+	spin_unlock(&spmc_shmem_obj_state.lock);
+	return spmc_ffa_error_return(handle, ret);
+}
+
+/**
+ * spmc_ffa_mem_frag_tx - FFA_MEM_FRAG_TX implementation.
+ * @client:             Client state.
+ * @handle_low:         Handle_low value returned from FFA_MEM_FRAG_RX.
+ * @handle_high:        Handle_high value returned from FFA_MEM_FRAG_RX.
+ * @fragment_length:    Length of fragments transmitted.
+ * @sender_id:          Vmid of sender in bits [31:16]
+ * @smc_handle:         Handle passed to smc call. Used to return
+ *                      FFA_MEM_FRAG_RX or SMC_FC_FFA_SUCCESS.
+ *
+ * Return: @smc_handle on success, error code on failure.
+ */
+long spmc_ffa_mem_frag_tx(uint32_t smc_fid,
+			  bool secure_origin,
+			  uint64_t handle_low,
+			  uint64_t handle_high,
+			  uint32_t fragment_length,
+			  uint32_t sender_id,
+			  void *cookie,
+			  void *handle,
+			  uint64_t flags)
+{
+	long ret;
+	uint32_t desc_sender_id;
+	struct mailbox *mbox = spmc_get_mbox_desc(secure_origin);
+
+	struct spmc_shmem_obj *obj;
+	uint64_t mem_handle = handle_low | (((uint64_t)handle_high) << 32);
+
+	spin_lock(&spmc_shmem_obj_state.lock);
+
+	obj = spmc_shmem_obj_lookup(&spmc_shmem_obj_state, mem_handle);
+	if (obj == NULL) {
+		WARN("%s: invalid handle, 0x%lx, not a valid handle.\n",
+		     __func__, mem_handle);
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock;
+	}
+
+	desc_sender_id = (uint32_t)obj->desc.sender_id << 16;
+	if (sender_id != desc_sender_id) {
+		WARN("%s: invalid sender_id 0x%x != 0x%x\n", __func__,
+		     sender_id, desc_sender_id);
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock;
+	}
+
+	if (obj->desc_filled == obj->desc_size) {
+		WARN("%s: object desc already filled, %zu\n", __func__,
+		     obj->desc_filled);
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock;
+	}
+
+	spin_lock(&mbox->lock);
+	ret = spmc_ffa_fill_desc(mbox, obj, fragment_length, 0, handle);
+	spin_unlock(&mbox->lock);
+
+	spin_unlock(&spmc_shmem_obj_state.lock);
+	return ret;
+
+err_unlock:
+	spin_unlock(&spmc_shmem_obj_state.lock);
+	return spmc_ffa_error_return(handle, ret);
+}
+
+/**
+ * spmc_ffa_mem_retrieve_req - FFA_MEM_RETRIEVE_REQ implementation.
+ * @smc_fid:            FID of SMC
+ * @total_length:       Total length of retrieve request descriptor if this is
+ *                      the first call. Otherwise (unsupported) must be 0.
+ * @fragment_length:    Length of fragment of retrieve request descriptor passed
+ *                      in this call. Only @fragment_length == @length is
+ *                      supported by this implementation.
+ * @address:            Not supported, must be 0.
+ * @page_count:         Not supported, must be 0.
+ * @smc_handle:         Handle passed to smc call. Used to return
+ *                      FFA_MEM_RETRIEVE_RESP.
+ *
+ * Implements a subset of the FF-A FFA_MEM_RETRIEVE_REQ call.
+ * Used by secure os to retrieve memory already shared by non-secure os.
+ * If the data does not fit in a single FFA_MEM_RETRIEVE_RESP message,
+ * the client must call FFA_MEM_FRAG_RX until the full response has been
+ * received.
+ *
+ * Return: @handle on success, error code on failure.
+ */
+long
+spmc_ffa_mem_retrieve_req(uint32_t smc_fid,
+			  bool secure_origin,
+			  uint32_t total_length,
+			  uint32_t fragment_length,
+			  uint64_t address,
+			  uint32_t page_count,
+			  void *cookie,
+			  void *handle,
+			  uint64_t flags)
+{
+	int ret;
+	size_t buf_size;
+	size_t copy_size;
+	struct ffa_mtd_v1_0 *resp;
+	const struct ffa_mtd_v1_0 *req;
+	struct spmc_shmem_obj *obj = NULL;
+	struct mailbox *mbox = spmc_get_mbox_desc(secure_origin);
+
+	if (!secure_origin) {
+		WARN("%s: unsupported retrieve req direction.\n", __func__);
+		return spmc_ffa_error_return(handle,
+					     FFA_ERROR_INVALID_PARAMETER);
+	}
+
+	if (address != 0U || page_count != 0U) {
+		WARN("%s: custom memory region not supported.\n", __func__);
+		return spmc_ffa_error_return(handle,
+					     FFA_ERROR_INVALID_PARAMETER);
+	}
+
+	spin_lock(&mbox->lock);
+
+	req = mbox->tx_buffer;
+	resp = mbox->rx_buffer;
+	buf_size = mbox->rxtx_page_count * FFA_PAGE_SIZE;
+
+	if (mbox->rxtx_page_count == 0U) {
+		WARN("%s: buffer pair not registered.\n", __func__);
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock_mailbox;
+	}
+
+	if (mbox->state != MAILBOX_STATE_EMPTY) {
+		WARN("%s: RX Buffer is full! %d\n", __func__, mbox->state);
+		ret = FFA_ERROR_DENIED;
+		goto err_unlock_mailbox;
+	}
+
+	if (fragment_length != total_length) {
+		WARN("%s: fragmented retrieve request not supported.\n",
+		     __func__);
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock_mailbox;
+	}
+
+	/*
+	 * Ensure endpoint count is 1, additional receivers not currently
+	 * supported.
+	 */
+	if (req->emad_count != 1U) {
+		WARN("%s: unsupported retrieve descriptor count: %u\n",
+		     __func__, req->emad_count);
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock_mailbox;
+	}
+
+	if (total_length < sizeof(*req)) {
+		WARN("%s: invalid length %u < %zu\n", __func__, total_length,
+		     sizeof(*req));
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock_mailbox;
+	}
+
+	spin_lock(&spmc_shmem_obj_state.lock);
+
+	obj = spmc_shmem_obj_lookup(&spmc_shmem_obj_state, req->handle);
+	if (obj == NULL) {
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock_all;
+	}
+
+	if (obj->desc_filled != obj->desc_size) {
+		WARN("%s: incomplete object desc filled %zu < size %zu\n",
+		     __func__, obj->desc_filled, obj->desc_size);
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock_all;
+	}
+
+	if (req->emad_count != 0U && req->sender_id != obj->desc.sender_id) {
+		WARN("%s: wrong sender id 0x%x != 0x%x\n",
+		     __func__, req->sender_id, obj->desc.sender_id);
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock_all;
+	}
+
+	if (req->emad_count != 0U && req->tag != obj->desc.tag) {
+		WARN("%s: wrong tag 0x%lx != 0x%lx\n",
+		     __func__, req->tag, obj->desc.tag);
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock_all;
+	}
+
+	if (req->flags != 0U) {
+		if ((req->flags & FFA_MTD_FLAG_TYPE_MASK) !=
+		    (obj->desc.flags & FFA_MTD_FLAG_TYPE_MASK)) {
+			/*
+			 * If the retrieve request specifies the memory
+			 * transaction ensure it matches what we expect.
+			 */
+			WARN("%s: wrong mem transaction flags %x != %x\n",
+			__func__, req->flags, obj->desc.flags);
+			ret = FFA_ERROR_INVALID_PARAMETER;
+			goto err_unlock_all;
+		}
+
+		if (req->flags != FFA_MTD_FLAG_TYPE_SHARE_MEMORY &&
+		    req->flags != FFA_MTD_FLAG_TYPE_LEND_MEMORY) {
+			/*
+			 * Current implementation does not support donate and
+			 * it supports no other flags.
+			 */
+			WARN("%s: invalid flags 0x%x\n", __func__, req->flags);
+			ret = FFA_ERROR_INVALID_PARAMETER;
+			goto err_unlock_all;
+		}
+	}
+
+	/* TODO: support more than one endpoint ids. */
+	if (req->emad_count != 0U &&
+	    req->emad[0].mapd.endpoint_id !=
+	    obj->desc.emad[0].mapd.endpoint_id) {
+		WARN("%s: wrong receiver id 0x%x != 0x%x\n",
+		     __func__, req->emad[0].mapd.endpoint_id,
+		       obj->desc.emad[0].mapd.endpoint_id);
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock_all;
+	}
+
+	mbox->state = MAILBOX_STATE_FULL;
+
+	if (req->emad_count != 0U) {
+		obj->in_use++;
+	}
+
+	copy_size = MIN(obj->desc_size, buf_size);
+
+	memcpy(resp, &obj->desc, copy_size);
+
+	spin_unlock(&spmc_shmem_obj_state.lock);
+	spin_unlock(&mbox->lock);
+
+	SMC_RET8(handle, FFA_MEM_RETRIEVE_RESP, obj->desc_size,
+		 copy_size, 0, 0, 0, 0, 0);
+
+err_unlock_all:
+	spin_unlock(&spmc_shmem_obj_state.lock);
+err_unlock_mailbox:
+	spin_unlock(&mbox->lock);
+	return spmc_ffa_error_return(handle, ret);
+}
+
+/**
+ * spmc_ffa_mem_frag_rx - FFA_MEM_FRAG_RX implementation.
+ * @client:             Client state.
+ * @handle_low:         Handle passed to &FFA_MEM_RETRIEVE_REQ. Bit[31:0].
+ * @handle_high:        Handle passed to &FFA_MEM_RETRIEVE_REQ. Bit[63:32].
+ * @fragment_offset:    Byte offset in descriptor to resume at.
+ * @sender_id:          Bit[31:16]: Endpoint id of sender if client is a
+ *                      hypervisor. 0 otherwise.
+ * @smc_handle:         Handle passed to smc call. Used to return
+ *                      FFA_MEM_FRAG_TX.
+ *
+ * Return: @smc_handle on success, error code on failure.
+ */
+long spmc_ffa_mem_frag_rx(uint32_t smc_fid,
+			  bool secure_origin,
+			  uint32_t handle_low,
+			  uint32_t handle_high,
+			  uint32_t fragment_offset,
+			  uint32_t sender_id,
+			  void *cookie,
+			  void *handle,
+			  uint64_t flags)
+{
+	int ret;
+	void *src;
+	size_t buf_size;
+	size_t copy_size;
+	size_t full_copy_size;
+	uint32_t desc_sender_id;
+	struct mailbox *mbox = spmc_get_mbox_desc(secure_origin);
+	uint64_t mem_handle = handle_low | (((uint64_t)handle_high) << 32);
+	struct spmc_shmem_obj *obj;
+
+	if (!secure_origin) {
+		WARN("%s: can only be called from swld.\n",
+		     __func__);
+		return spmc_ffa_error_return(handle,
+					     FFA_ERROR_INVALID_PARAMETER);
+	}
+
+	spin_lock(&spmc_shmem_obj_state.lock);
+
+	obj = spmc_shmem_obj_lookup(&spmc_shmem_obj_state, mem_handle);
+	if (obj == NULL) {
+		WARN("%s: invalid handle, 0x%lx, not a valid handle.\n",
+		     __func__, mem_handle);
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock_shmem;
+	}
+
+	desc_sender_id = (uint32_t)obj->desc.sender_id << 16;
+	if (sender_id != 0U && sender_id != desc_sender_id) {
+		WARN("%s: invalid sender_id 0x%x != 0x%x\n", __func__,
+		     sender_id, desc_sender_id);
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock_shmem;
+	}
+
+	if (fragment_offset >= obj->desc_size) {
+		WARN("%s: invalid fragment_offset 0x%x >= 0x%zx\n",
+		     __func__, fragment_offset, obj->desc_size);
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock_shmem;
+	}
+
+	spin_lock(&mbox->lock);
+
+	if (mbox->rxtx_page_count == 0U) {
+		WARN("%s: buffer pair not registered.\n", __func__);
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock_all;
+	}
+
+	if (mbox->state != MAILBOX_STATE_EMPTY) {
+		WARN("%s: RX Buffer is full!\n", __func__);
+		ret = FFA_ERROR_DENIED;
+		goto err_unlock_all;
+	}
+
+	buf_size = mbox->rxtx_page_count * FFA_PAGE_SIZE;
+
+	mbox->state = MAILBOX_STATE_FULL;
+
+	full_copy_size = obj->desc_size - fragment_offset;
+	copy_size = MIN(full_copy_size, buf_size);
+
+	src = &obj->desc;
+
+	memcpy(mbox->rx_buffer, src + fragment_offset, copy_size);
+
+	spin_unlock(&mbox->lock);
+	spin_unlock(&spmc_shmem_obj_state.lock);
+
+	SMC_RET8(handle, FFA_MEM_FRAG_TX, handle_low, handle_high,
+		 copy_size, sender_id, 0, 0, 0);
+
+err_unlock_all:
+	spin_unlock(&mbox->lock);
+err_unlock_shmem:
+	spin_unlock(&spmc_shmem_obj_state.lock);
+	return spmc_ffa_error_return(handle, ret);
+}
+
+/**
+ * spmc_ffa_mem_relinquish - FFA_MEM_RELINQUISH implementation.
+ * @client:             Client state.
+ *
+ * Implements a subset of the FF-A FFA_MEM_RELINQUISH call.
+ * Used by secure os release previously shared memory to non-secure os.
+ *
+ * The handle to release must be in the client's (secure os's) transmit buffer.
+ *
+ * Return: 0 on success, error code on failure.
+ */
+int spmc_ffa_mem_relinquish(uint32_t smc_fid,
+			    bool secure_origin,
+			    uint32_t handle_low,
+			    uint32_t handle_high,
+			    uint32_t fragment_offset,
+			    uint32_t sender_id,
+			    void *cookie,
+			    void *handle,
+			    uint64_t flags)
+{
+	int ret;
+	struct mailbox *mbox = spmc_get_mbox_desc(secure_origin);
+	struct spmc_shmem_obj *obj;
+	const struct ffa_mem_relinquish_descriptor *req;
+
+	if (!secure_origin) {
+		WARN("%s: unsupported relinquish direction.\n", __func__);
+		return spmc_ffa_error_return(handle,
+					     FFA_ERROR_INVALID_PARAMETER);
+	}
+
+	spin_lock(&mbox->lock);
+
+	if (mbox->rxtx_page_count == 0U) {
+		WARN("%s: buffer pair not registered.\n", __func__);
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock_mailbox;
+	}
+
+	req = mbox->tx_buffer;
+
+	if (req->flags != 0U) {
+		WARN("%s: unsupported flags 0x%x\n", __func__, req->flags);
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock_mailbox;
+	}
+
+	spin_lock(&spmc_shmem_obj_state.lock);
+
+	obj = spmc_shmem_obj_lookup(&spmc_shmem_obj_state, req->handle);
+	if (obj == NULL) {
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock_all;
+	}
+
+	if (obj->desc.emad_count != req->endpoint_count) {
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock_all;
+	}
+	for (size_t i = 0; i < req->endpoint_count; i++) {
+		if (req->endpoint_array[i] !=
+		    obj->desc.emad[i].mapd.endpoint_id) {
+			ret = FFA_ERROR_INVALID_PARAMETER;
+			goto err_unlock_all;
+		}
+	}
+	if (obj->in_use == 0U) {
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock_all;
+	}
+	obj->in_use--;
+
+	spin_unlock(&spmc_shmem_obj_state.lock);
+	spin_unlock(&mbox->lock);
+
+	SMC_RET1(handle, FFA_SUCCESS_SMC32);
+
+err_unlock_all:
+	spin_unlock(&spmc_shmem_obj_state.lock);
+err_unlock_mailbox:
+	spin_unlock(&mbox->lock);
+	return spmc_ffa_error_return(handle, ret);
+}
+
+/**
+ * spmc_ffa_mem_reclaim - FFA_MEM_RECLAIM implementation.
+ * @client:         Client state.
+ * @handle_low:     Unique handle of shared memory object to reclaim. Bit[31:0].
+ * @handle_high:    Unique handle of shared memory object to reclaim.
+ *                  Bit[63:32].
+ * @flags:          Unsupported, ignored.
+ *
+ * Implements a subset of the FF-A FFA_MEM_RECLAIM call.
+ * Used by non-secure os reclaim memory previously shared with secure os.
+ *
+ * Return: 0 on success, error code on failure.
+ */
+int spmc_ffa_mem_reclaim(uint32_t smc_fid,
+			 bool secure_origin,
+			 uint32_t handle_low,
+			 uint32_t handle_high,
+			 uint32_t mem_flags,
+			 uint64_t x4,
+			 void *cookie,
+			 void *handle,
+			 uint64_t flags)
+{
+	int ret;
+	struct spmc_shmem_obj *obj;
+	uint64_t mem_handle = handle_low | (((uint64_t)handle_high) << 32);
+
+	if (secure_origin) {
+		WARN("%s: unsupported reclaim direction.\n", __func__);
+		return spmc_ffa_error_return(handle,
+					     FFA_ERROR_INVALID_PARAMETER);
+	}
+
+	if (mem_flags != 0U) {
+		WARN("%s: unsupported flags 0x%x\n", __func__, mem_flags);
+		return spmc_ffa_error_return(handle,
+					     FFA_ERROR_INVALID_PARAMETER);
+	}
+
+	spin_lock(&spmc_shmem_obj_state.lock);
+
+	obj = spmc_shmem_obj_lookup(&spmc_shmem_obj_state, mem_handle);
+	if (obj == NULL) {
+		ret = FFA_ERROR_INVALID_PARAMETER;
+		goto err_unlock;
+	}
+	if (obj->in_use != 0U) {
+		ret = FFA_ERROR_DENIED;
+		goto err_unlock;
+	}
+	spmc_shmem_obj_free(&spmc_shmem_obj_state, obj);
+	spin_unlock(&spmc_shmem_obj_state.lock);
+
+	SMC_RET1(handle, FFA_SUCCESS_SMC32);
+
+err_unlock:
+	spin_unlock(&spmc_shmem_obj_state.lock);
+	return spmc_ffa_error_return(handle, ret);
+}
diff --git a/services/std_svc/spm/el3_spmc/spmc_shared_mem.h b/services/std_svc/spm/el3_spmc/spmc_shared_mem.h
new file mode 100644
index 0000000..8571e94
--- /dev/null
+++ b/services/std_svc/spm/el3_spmc/spmc_shared_mem.h
@@ -0,0 +1,113 @@
+/*
+ * Copyright (c) 2022, ARM Limited and Contributors. All rights reserved.
+ *
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+
+#ifndef SPMC_SHARED_MEM_H
+#define SPMC_SHARED_MEM_H
+
+#include <services/el3_spmc_ffa_memory.h>
+
+/**
+ * struct ffa_mem_relinquish_descriptor - Relinquish request descriptor.
+ * @handle:
+ *         Id of shared memory object to relinquish.
+ * @flags:
+ *         If bit 0 is set clear memory after unmapping from borrower. Must be 0
+ *         for share. Bit[1]: Time slicing. Not supported, must be 0. All other
+ *         bits are reserved 0.
+ * @endpoint_count:
+ *         Number of entries in @endpoint_array.
+ * @endpoint_array:
+ *         Array of endpoint ids.
+ */
+struct ffa_mem_relinquish_descriptor {
+	uint64_t handle;
+	uint32_t flags;
+	uint32_t endpoint_count;
+	ffa_endpoint_id16_t endpoint_array[];
+};
+CASSERT(sizeof(struct ffa_mem_relinquish_descriptor) == 16,
+	assert_ffa_mem_relinquish_descriptor_size_mismatch);
+
+/**
+ * struct spmc_shmem_obj_state - Global state.
+ * @data:           Backing store for spmc_shmem_obj objects.
+ * @data_size:      The size allocated for the backing store.
+ * @allocated:      Number of bytes allocated in @data.
+ * @next_handle:    Handle used for next allocated object.
+ * @lock:           Lock protecting all state in this file.
+ */
+struct spmc_shmem_obj_state {
+	uint8_t *data;
+	size_t data_size;
+	size_t allocated;
+	uint64_t next_handle;
+	spinlock_t lock;
+};
+
+extern struct spmc_shmem_obj_state spmc_shmem_obj_state;
+
+long spmc_ffa_mem_send(uint32_t smc_fid,
+		       bool secure_origin,
+		       uint64_t total_length,
+		       uint32_t fragment_length,
+		       uint64_t address,
+		       uint32_t page_count,
+		       void *cookie,
+		       void *handle,
+		       uint64_t flags);
+
+long spmc_ffa_mem_frag_tx(uint32_t smc_fid,
+			  bool secure_origin,
+			  uint64_t handle_low,
+			  uint64_t handle_high,
+			  uint32_t fragment_length,
+			  uint32_t sender_id,
+			  void *cookie,
+			  void *handle,
+			  uint64_t flags);
+
+long spmc_ffa_mem_retrieve_req(uint32_t smc_fid,
+			       bool secure_origin,
+			       uint32_t total_length,
+			       uint32_t fragment_length,
+			       uint64_t address,
+			       uint32_t page_count,
+			       void *cookie,
+			       void *handle,
+			       uint64_t flags);
+
+long spmc_ffa_mem_frag_rx(uint32_t smc_fid,
+			  bool secure_origin,
+			  uint32_t handle_low,
+			  uint32_t handle_high,
+			  uint32_t fragment_offset,
+			  uint32_t sender_id,
+			  void *cookie,
+			  void *handle,
+			  uint64_t flags);
+
+
+int spmc_ffa_mem_relinquish(uint32_t smc_fid,
+			    bool secure_origin,
+			    uint32_t handle_low,
+			    uint32_t handle_high,
+			    uint32_t fragment_offset,
+			    uint32_t sender_id,
+			    void *cookie,
+			    void *handle,
+			    uint64_t flags);
+
+int spmc_ffa_mem_reclaim(uint32_t smc_fid,
+			 bool secure_origin,
+			 uint32_t handle_low,
+			 uint32_t handle_high,
+			 uint32_t mem_flags,
+			 uint64_t x4,
+			 void *cookie,
+			 void *handle,
+			 uint64_t flags);
+
+#endif /* SPMC_SHARED_MEM_H */