feat(drivers/arm/mhu): add MHU driver
The Arm Message Handling Unit (MHU) is a mailbox controller used to
communicate with other processing element(s). Adding a driver to
enable the communication:
- Adding generic MHU driver interface,
- Adding MHU_v2_x driver.
Driver supports:
- Discovering available MHU channels,
- Sending / receiving words over MHU channels,
- Signaling happens over a dedicated channel.
Signed-off-by: Tamas Ban <tamas.ban@arm.com>
Signed-off-by: David Vincze <david.vincze@arm.com>
Change-Id: I41a5b968f6b8319cdbdf7907d70bd8837839862e
diff --git a/drivers/arm/mhu/mhu_wrapper_v2_x.c b/drivers/arm/mhu/mhu_wrapper_v2_x.c
new file mode 100644
index 0000000..d8b7cfd
--- /dev/null
+++ b/drivers/arm/mhu/mhu_wrapper_v2_x.c
@@ -0,0 +1,302 @@
+/*
+ * Copyright (c) 2022, Arm Limited. All rights reserved.
+ *
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+
+#include <assert.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+
+#include <drivers/arm/mhu.h>
+
+#include "mhu_v2_x.h"
+
+#define MHU_NOTIFY_VALUE (1234u)
+
+/*
+ * MHU devices for host:
+ * HSE: Host to Secure Enclave (sender device)
+ * SEH: Secure Enclave to Host (receiver device)
+ */
+struct mhu_v2_x_dev_t MHU1_HSE_DEV = {0, MHU_V2_X_SENDER_FRAME};
+struct mhu_v2_x_dev_t MHU1_SEH_DEV = {0, MHU_V2_X_RECEIVER_FRAME};
+
+static enum mhu_error_t error_mapping_to_mhu_error_t(enum mhu_v2_x_error_t err)
+{
+ switch (err) {
+ case MHU_V_2_X_ERR_NONE:
+ return MHU_ERR_NONE;
+ case MHU_V_2_X_ERR_NOT_INIT:
+ return MHU_ERR_NOT_INIT;
+ case MHU_V_2_X_ERR_ALREADY_INIT:
+ return MHU_ERR_ALREADY_INIT;
+ case MHU_V_2_X_ERR_UNSUPPORTED_VERSION:
+ return MHU_ERR_UNSUPPORTED_VERSION;
+ case MHU_V_2_X_ERR_INVALID_ARG:
+ return MHU_ERR_INVALID_ARG;
+ case MHU_V_2_X_ERR_GENERAL:
+ return MHU_ERR_GENERAL;
+ default:
+ return MHU_ERR_GENERAL;
+ }
+}
+
+static enum mhu_v2_x_error_t signal_and_wait_for_clear(void)
+{
+ enum mhu_v2_x_error_t err;
+ struct mhu_v2_x_dev_t *dev = &MHU1_HSE_DEV;
+ uint32_t val = MHU_NOTIFY_VALUE;
+ /* Using the last channel for notifications */
+ uint32_t channel_notify = mhu_v2_x_get_num_channel_implemented(dev) - 1;
+
+ err = mhu_v2_x_channel_send(dev, channel_notify, val);
+ if (err != MHU_V_2_X_ERR_NONE) {
+ return err;
+ }
+
+ do {
+ err = mhu_v2_x_channel_poll(dev, channel_notify, &val);
+ if (err != MHU_V_2_X_ERR_NONE) {
+ break;
+ }
+ } while (val != 0);
+
+ return err;
+}
+
+static enum mhu_v2_x_error_t wait_for_signal(void)
+{
+ enum mhu_v2_x_error_t err;
+ struct mhu_v2_x_dev_t *dev = &MHU1_SEH_DEV;
+ uint32_t val = 0;
+ /* Using the last channel for notifications */
+ uint32_t channel_notify = mhu_v2_x_get_num_channel_implemented(dev) - 1;
+
+ do {
+ err = mhu_v2_x_channel_receive(dev, channel_notify, &val);
+ if (err != MHU_V_2_X_ERR_NONE) {
+ break;
+ }
+ } while (val != MHU_NOTIFY_VALUE);
+
+ return err;
+}
+
+static enum mhu_v2_x_error_t clear_and_wait_for_next_signal(void)
+{
+ enum mhu_v2_x_error_t err;
+ struct mhu_v2_x_dev_t *dev = &MHU1_SEH_DEV;
+ uint32_t num_channels = mhu_v2_x_get_num_channel_implemented(dev);
+ uint32_t i;
+
+ /* Clear all channels */
+ for (i = 0; i < num_channels; ++i) {
+ err = mhu_v2_x_channel_clear(dev, i);
+ if (err != MHU_V_2_X_ERR_NONE) {
+ return err;
+ }
+ }
+
+ return wait_for_signal();
+}
+
+enum mhu_error_t mhu_init_sender(uintptr_t mhu_sender_base)
+{
+ enum mhu_v2_x_error_t err;
+
+ assert(mhu_sender_base != (uintptr_t)NULL);
+
+ MHU1_HSE_DEV.base = mhu_sender_base;
+
+ err = mhu_v2_x_driver_init(&MHU1_HSE_DEV, MHU_REV_READ_FROM_HW);
+ return error_mapping_to_mhu_error_t(err);
+}
+
+enum mhu_error_t mhu_init_receiver(uintptr_t mhu_receiver_base)
+{
+ enum mhu_v2_x_error_t err;
+ uint32_t num_channels, i;
+
+ assert(mhu_receiver_base != (uintptr_t)NULL);
+
+ MHU1_SEH_DEV.base = mhu_receiver_base;
+
+ err = mhu_v2_x_driver_init(&MHU1_SEH_DEV, MHU_REV_READ_FROM_HW);
+ if (err != MHU_V_2_X_ERR_NONE) {
+ return error_mapping_to_mhu_error_t(err);
+ }
+
+ num_channels = mhu_v2_x_get_num_channel_implemented(&MHU1_SEH_DEV);
+
+ /* Mask all channels except the notifying channel */
+ for (i = 0; i < (num_channels - 1); ++i) {
+ err = mhu_v2_x_channel_mask_set(&MHU1_SEH_DEV, i, UINT32_MAX);
+ if (err != MHU_V_2_X_ERR_NONE) {
+ return error_mapping_to_mhu_error_t(err);
+ }
+ }
+
+ /* The last channel is used for notifications */
+ err = mhu_v2_x_channel_mask_clear(
+ &MHU1_SEH_DEV, (num_channels - 1), UINT32_MAX);
+ return error_mapping_to_mhu_error_t(err);
+}
+
+/*
+ * Public function. See mhu.h
+ *
+ * The basic steps of transferring a message:
+ * 1. Initiate MHU transfer.
+ * 2. Send over the size of the payload on Channel 1. It is the very first
+ * 4 Bytes of the transfer. Continue with Channel 2.
+ * 3. Send over the payload, writing the channels one after the other
+ * (4 Bytes each). The last available channel is reserved for controlling
+ * the transfer.
+ * When the last channel is reached or no more data is left, STOP.
+ * 4. Notify the receiver using the last channel and wait for acknowledge.
+ * If there is still data to transfer, jump to step 3. Otherwise, proceed.
+ * 5. Close MHU transfer.
+ *
+ */
+enum mhu_error_t mhu_send_data(const uint8_t *send_buffer, size_t size)
+{
+ enum mhu_v2_x_error_t err;
+ struct mhu_v2_x_dev_t *dev = &MHU1_HSE_DEV;
+ uint32_t num_channels = mhu_v2_x_get_num_channel_implemented(dev);
+ uint32_t chan = 0;
+ uint32_t i;
+ uint32_t *p;
+
+ /* For simplicity, require the send_buffer to be 4-byte aligned */
+ if ((uintptr_t)send_buffer & 0x3U) {
+ return MHU_ERR_INVALID_ARG;
+ }
+
+ err = mhu_v2_x_initiate_transfer(dev);
+ if (err != MHU_V_2_X_ERR_NONE) {
+ return error_mapping_to_mhu_error_t(err);
+ }
+
+ /* First send over the size of the actual message */
+ err = mhu_v2_x_channel_send(dev, chan, (uint32_t)size);
+ if (err != MHU_V_2_X_ERR_NONE) {
+ return error_mapping_to_mhu_error_t(err);
+ }
+ chan++;
+
+ p = (uint32_t *)send_buffer;
+ for (i = 0; i < size; i += 4) {
+ err = mhu_v2_x_channel_send(dev, chan, *p++);
+ if (err != MHU_V_2_X_ERR_NONE) {
+ return error_mapping_to_mhu_error_t(err);
+ }
+ if (++chan == (num_channels - 1)) {
+ err = signal_and_wait_for_clear();
+ if (err != MHU_V_2_X_ERR_NONE) {
+ return error_mapping_to_mhu_error_t(err);
+ }
+ chan = 0;
+ }
+ }
+
+ /* Signal the end of transfer.
+ * It's not required to send a signal when the message was
+ * perfectly-aligned (num_channels - 1 channels were used in the last
+ * round) preventing it from signaling twice at the end of transfer.
+ */
+ if (chan != 0) {
+ err = signal_and_wait_for_clear();
+ if (err != MHU_V_2_X_ERR_NONE) {
+ return error_mapping_to_mhu_error_t(err);
+ }
+ }
+
+ err = mhu_v2_x_close_transfer(dev);
+ return error_mapping_to_mhu_error_t(err);
+}
+
+/*
+ * Public function. See mhu.h
+ *
+ * The basic steps of receiving a message:
+ * 1. Read the size of the payload from Channel 1. It is the very first
+ * 4 Bytes of the transfer. Continue with Channel 2.
+ * 2. Receive the payload, read the channels one after the other
+ * (4 Bytes each). The last available channel is reserved for controlling
+ * the transfer.
+ * When the last channel is reached clear all the channels
+ * (also sending an acknowledge on the last channel).
+ * 3. If there is still data to receive wait for a notification on the last
+ * channel and jump to step 2 as soon as it arrived. Otherwise, proceed.
+ * 4. End of transfer.
+ *
+ */
+enum mhu_error_t mhu_receive_data(uint8_t *receive_buffer, size_t *size)
+{
+ enum mhu_v2_x_error_t err;
+ struct mhu_v2_x_dev_t *dev = &MHU1_SEH_DEV;
+ uint32_t num_channels = mhu_v2_x_get_num_channel_implemented(dev);
+ uint32_t chan = 0;
+ uint32_t message_len;
+ uint32_t i;
+ uint32_t *p;
+
+ /* For simplicity, require:
+ * - the receive_buffer to be 4-byte aligned,
+ * - the buffer size to be a multiple of 4.
+ */
+ if (((uintptr_t)receive_buffer & 0x3U) || (*size & 0x3U)) {
+ return MHU_ERR_INVALID_ARG;
+ }
+
+ /* Busy wait for incoming reply */
+ err = wait_for_signal();
+ if (err != MHU_V_2_X_ERR_NONE) {
+ return error_mapping_to_mhu_error_t(err);
+ }
+
+ /* The first word is the length of the actual message */
+ err = mhu_v2_x_channel_receive(dev, chan, &message_len);
+ if (err != MHU_V_2_X_ERR_NONE) {
+ return error_mapping_to_mhu_error_t(err);
+ }
+ chan++;
+
+ if (message_len > *size) {
+ /* Message buffer too small */
+ *size = message_len;
+ return MHU_ERR_BUFFER_TOO_SMALL;
+ }
+
+ p = (uint32_t *)receive_buffer;
+ for (i = 0; i < message_len; i += 4) {
+ err = mhu_v2_x_channel_receive(dev, chan, p++);
+ if (err != MHU_V_2_X_ERR_NONE) {
+ return error_mapping_to_mhu_error_t(err);
+ }
+
+ /* Only wait for next transfer if there is still missing data */
+ if (++chan == (num_channels - 1) && (message_len - i) > 4) {
+ /* Busy wait for next transfer */
+ err = clear_and_wait_for_next_signal();
+ if (err != MHU_V_2_X_ERR_NONE) {
+ return error_mapping_to_mhu_error_t(err);
+ }
+ chan = 0;
+ }
+ }
+
+ /* Clear all channels */
+ for (i = 0; i < num_channels; ++i) {
+ err = mhu_v2_x_channel_clear(dev, i);
+ if (err != MHU_V_2_X_ERR_NONE) {
+ return error_mapping_to_mhu_error_t(err);
+ }
+ }
+
+ *size = message_len;
+
+ return MHU_ERR_NONE;
+}