misc: mic: Coprocessor State Management (COSM) driver

The COSM driver allows boot, shutdown and reset of Intel MIC devices
via sysfs. This functionality was previously present in the Intel MIC
host driver but has now been taken out into a separate driver so that
it can be shared between multiple generations of Intel MIC products.
The sysfs kernel ABI used by the COSM driver is the same as that
defined originally for the MIC host driver in
Documentation/ABI/testing/sysfs-class-mic.txt.

The COSM driver also contains support for dumping the MIC card log_buf
and doing a "force reset" for the card via debugfs. The OSPM support
present in the MIC host driver has now largely been moved to user
space and only a small required OSPM functionality is now present in
the driver.
Reviewed-by: Nikhil Rao <nikhil.rao@intel.com>
Reviewed-by: Sudeep Dutt <sudeep.dutt@intel.com>
Signed-off-by: Dasaratharaman Chandramouli <dasaratharaman.chandramouli@intel.com>
Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

drivers/misc/mic/cosm/Makefile

+#
+# Makefile - Intel MIC Coprocessor State Management (COSM) Driver
+# Copyright(c) 2015, Intel Corporation.
+#
+obj-$(CONFIG_MIC_COSM) += mic_cosm.o
+
+mic_cosm-objs := cosm_main.o
+mic_cosm-objs += cosm_debugfs.o
+mic_cosm-objs += cosm_sysfs.o
+mic_cosm-objs += cosm_scif_server.o

drivers/misc/mic/cosm/cosm_debugfs.c

+/*
+ * Intel MIC Platform Software Stack (MPSS)
+ *
+ * Copyright(c) 2015 Intel Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License, version 2, as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * The full GNU General Public License is included in this distribution in
+ * the file called "COPYING".
+ *
+ * Intel MIC Coprocessor State Management (COSM) Driver
+ *
+ */
+
+#include <linux/debugfs.h>
+#include <linux/slab.h>
+#include "cosm_main.h"
+
+/* Debugfs parent dir */
+static struct dentry *cosm_dbg;
+
+/**
+ * cosm_log_buf_show - Display MIC kernel log buffer
+ *
+ * log_buf addr/len is read from System.map by user space
+ * and populated in sysfs entries.
+ */
+static int cosm_log_buf_show(struct seq_file *s, void *unused)
+{
+	void __iomem *log_buf_va;
+	int __iomem *log_buf_len_va;
+	struct cosm_device *cdev = s->private;
+	void *kva;
+	int size;
+	u64 aper_offset;
+
+	if (!cdev || !cdev->log_buf_addr || !cdev->log_buf_len)
+		goto done;
+
+	mutex_lock(&cdev->cosm_mutex);
+	switch (cdev->state) {
+	case MIC_BOOTING:
+	case MIC_ONLINE:
+	case MIC_SHUTTING_DOWN:
+		break;
+	default:
+		goto unlock;
+	}
+
+	/*
+	 * Card kernel will never be relocated and any kernel text/data mapping
+	 * can be translated to phys address by subtracting __START_KERNEL_map.
+	 */
+	aper_offset = (u64)cdev->log_buf_len - __START_KERNEL_map;
+	log_buf_len_va = cdev->hw_ops->aper(cdev)->va + aper_offset;
+	aper_offset = (u64)cdev->log_buf_addr - __START_KERNEL_map;
+	log_buf_va = cdev->hw_ops->aper(cdev)->va + aper_offset;
+
+	size = ioread32(log_buf_len_va);
+	kva = kmalloc(size, GFP_KERNEL);
+	if (!kva)
+		goto unlock;
+
+	memcpy_fromio(kva, log_buf_va, size);
+	seq_write(s, kva, size);
+	kfree(kva);
+unlock:
+	mutex_unlock(&cdev->cosm_mutex);
+done:
+	return 0;
+}
+
+static int cosm_log_buf_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, cosm_log_buf_show, inode->i_private);
+}
+
+static const struct file_operations log_buf_ops = {
+	.owner   = THIS_MODULE,
+	.open    = cosm_log_buf_open,
+	.read    = seq_read,
+	.llseek  = seq_lseek,
+	.release = single_release
+};
+
+/**
+ * cosm_force_reset_show - Force MIC reset
+ *
+ * Invokes the force_reset COSM bus op instead of the standard reset
+ * op in case a force reset of the MIC device is required
+ */
+static int cosm_force_reset_show(struct seq_file *s, void *pos)
+{
+	struct cosm_device *cdev = s->private;
+
+	cosm_stop(cdev, true);
+	return 0;
+}
+
+static int cosm_force_reset_debug_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, cosm_force_reset_show, inode->i_private);
+}
+
+static const struct file_operations force_reset_ops = {
+	.owner   = THIS_MODULE,
+	.open    = cosm_force_reset_debug_open,
+	.read    = seq_read,
+	.llseek  = seq_lseek,
+	.release = single_release
+};
+
+void cosm_create_debug_dir(struct cosm_device *cdev)
+{
+	char name[16];
+
+	if (!cosm_dbg)
+		return;
+
+	scnprintf(name, sizeof(name), "mic%d", cdev->index);
+	cdev->dbg_dir = debugfs_create_dir(name, cosm_dbg);
+	if (!cdev->dbg_dir)
+		return;
+
+	debugfs_create_file("log_buf", 0444, cdev->dbg_dir, cdev, &log_buf_ops);
+	debugfs_create_file("force_reset", 0444, cdev->dbg_dir, cdev,
+			    &force_reset_ops);
+}
+
+void cosm_delete_debug_dir(struct cosm_device *cdev)
+{
+	if (!cdev->dbg_dir)
+		return;
+
+	debugfs_remove_recursive(cdev->dbg_dir);
+}
+
+void cosm_init_debugfs(void)
+{
+	cosm_dbg = debugfs_create_dir(KBUILD_MODNAME, NULL);
+	if (!cosm_dbg)
+		pr_err("can't create debugfs dir\n");
+}
+
+void cosm_exit_debugfs(void)
+{
+	debugfs_remove(cosm_dbg);
+}

drivers/misc/mic/cosm/cosm_main.c

+/*
+ * Intel MIC Platform Software Stack (MPSS)
+ *
+ * Copyright(c) 2015 Intel Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License, version 2, as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * The full GNU General Public License is included in this distribution in
+ * the file called "COPYING".
+ *
+ * Intel MIC Coprocessor State Management (COSM) Driver
+ *
+ */
+
+#include <linux/module.h>
+#include <linux/delay.h>
+#include <linux/idr.h>
+#include <linux/slab.h>
+#include <linux/cred.h>
+#include "cosm_main.h"
+
+static const char cosm_driver_name[] = "mic";
+
+/* COSM ID allocator */
+static struct ida g_cosm_ida;
+/* Class of MIC devices for sysfs accessibility. */
+static struct class *g_cosm_class;
+/* Number of MIC devices */
+static atomic_t g_num_dev;
+
+/**
+ * cosm_hw_reset - Issue a HW reset for the MIC device
+ * @cdev: pointer to cosm_device instance
+ */
+static void cosm_hw_reset(struct cosm_device *cdev, bool force)
+{
+	int i;
+
+#define MIC_RESET_TO (45)
+	if (force && cdev->hw_ops->force_reset)
+		cdev->hw_ops->force_reset(cdev);
+	else
+		cdev->hw_ops->reset(cdev);
+
+	for (i = 0; i < MIC_RESET_TO; i++) {
+		if (cdev->hw_ops->ready(cdev)) {
+			cosm_set_state(cdev, MIC_READY);
+			return;
+		}
+		/*
+		 * Resets typically take 10s of seconds to complete.
+		 * Since an MMIO read is required to check if the
+		 * firmware is ready or not, a 1 second delay works nicely.
+		 */
+		msleep(1000);
+	}
+	cosm_set_state(cdev, MIC_RESET_FAILED);
+}
+
+/**
+ * cosm_start - Start the MIC
+ * @cdev: pointer to cosm_device instance
+ *
+ * This function prepares an MIC for boot and initiates boot.
+ * RETURNS: An appropriate -ERRNO error value on error, or 0 for success.
+ */
+int cosm_start(struct cosm_device *cdev)
+{
+	const struct cred *orig_cred;
+	struct cred *override_cred;
+	int rc;
+
+	mutex_lock(&cdev->cosm_mutex);
+	if (!cdev->bootmode) {
+		dev_err(&cdev->dev, "%s %d bootmode not set\n",
+			__func__, __LINE__);
+		rc = -EINVAL;
+		goto unlock_ret;
+	}
+retry:
+	if (cdev->state != MIC_READY) {
+		dev_err(&cdev->dev, "%s %d MIC state not READY\n",
+			__func__, __LINE__);
+		rc = -EINVAL;
+		goto unlock_ret;
+	}
+	if (!cdev->hw_ops->ready(cdev)) {
+		cosm_hw_reset(cdev, false);
+		/*
+		 * The state will either be MIC_READY if the reset succeeded
+		 * or MIC_RESET_FAILED if the firmware reset failed.
+		 */
+		goto retry;
+	}
+
+	/*
+	 * Set credentials to root to allow non-root user to download initramsfs
+	 * with 600 permissions
+	 */
+	override_cred = prepare_creds();
+	if (!override_cred) {
+		dev_err(&cdev->dev, "%s %d prepare_creds failed\n",
+			__func__, __LINE__);
+		rc = -ENOMEM;
+		goto unlock_ret;
+	}
+	override_cred->fsuid = GLOBAL_ROOT_UID;
+	orig_cred = override_creds(override_cred);
+
+	rc = cdev->hw_ops->start(cdev, cdev->index);
+
+	revert_creds(orig_cred);
+	put_cred(override_cred);
+	if (rc)
+		goto unlock_ret;
+
+	/*
+	 * If linux is being booted, card is treated 'online' only
+	 * when the scif interface in the card is up. If anything else
+	 * is booted, we set card to 'online' immediately.
+	 */
+	if (!strcmp(cdev->bootmode, "linux"))
+		cosm_set_state(cdev, MIC_BOOTING);
+	else
+		cosm_set_state(cdev, MIC_ONLINE);
+unlock_ret:
+	mutex_unlock(&cdev->cosm_mutex);
+	if (rc)
+		dev_err(&cdev->dev, "cosm_start failed rc %d\n", rc);
+	return rc;
+}
+
+/**
+ * cosm_stop - Prepare the MIC for reset and trigger reset
+ * @cdev: pointer to cosm_device instance
+ * @force: force a MIC to reset even if it is already reset and ready.
+ *
+ * RETURNS: None
+ */
+void cosm_stop(struct cosm_device *cdev, bool force)
+{
+	mutex_lock(&cdev->cosm_mutex);
+	if (cdev->state != MIC_READY || force) {
+		/*
+		 * Don't call hw_ops if they have been called previously.
+		 * stop(..) calls device_unregister and will crash the system if
+		 * called multiple times.
+		 */
+		bool call_hw_ops = cdev->state != MIC_RESET_FAILED &&
+					cdev->state != MIC_READY;
+
+		if (cdev->state != MIC_RESETTING)
+			cosm_set_state(cdev, MIC_RESETTING);
+		cdev->heartbeat_watchdog_enable = false;
+		if (call_hw_ops)
+			cdev->hw_ops->stop(cdev, force);
+		cosm_hw_reset(cdev, force);
+		cosm_set_shutdown_status(cdev, MIC_NOP);
+		if (call_hw_ops && cdev->hw_ops->post_reset)
+			cdev->hw_ops->post_reset(cdev, cdev->state);
+	}
+	mutex_unlock(&cdev->cosm_mutex);
+	flush_work(&cdev->scif_work);
+}
+
+/**
+ * cosm_reset_trigger_work - Trigger MIC reset
+ * @work: The work structure
+ *
+ * This work is scheduled whenever the host wants to reset the MIC.
+ */
+static void cosm_reset_trigger_work(struct work_struct *work)
+{
+	struct cosm_device *cdev = container_of(work, struct cosm_device,
+						reset_trigger_work);
+	cosm_stop(cdev, false);
+}
+
+/**
+ * cosm_reset - Schedule MIC reset
+ * @cdev: pointer to cosm_device instance
+ *
+ * RETURNS: An -EINVAL if the card is already READY or 0 for success.
+ */
+int cosm_reset(struct cosm_device *cdev)
+{
+	int rc = 0;
+
+	mutex_lock(&cdev->cosm_mutex);
+	if (cdev->state != MIC_READY) {
+		cosm_set_state(cdev, MIC_RESETTING);
+		schedule_work(&cdev->reset_trigger_work);
+	} else {
+		dev_err(&cdev->dev, "%s %d MIC is READY\n", __func__, __LINE__);
+		rc = -EINVAL;
+	}
+	mutex_unlock(&cdev->cosm_mutex);
+	return rc;
+}
+
+/**
+ * cosm_shutdown - Initiate MIC shutdown.
+ * @cdev: pointer to cosm_device instance
+ *
+ * RETURNS: None
+ */
+int cosm_shutdown(struct cosm_device *cdev)
+{
+	struct cosm_msg msg = { .id = COSM_MSG_SHUTDOWN };
+	int rc = 0;
+
+	mutex_lock(&cdev->cosm_mutex);
+	if (cdev->state != MIC_ONLINE) {
+		rc = -EINVAL;
+		dev_err(&cdev->dev, "%s %d skipping shutdown in state: %s\n",
+			__func__, __LINE__, cosm_state_string[cdev->state]);
+		goto err;
+	}
+
+	if (!cdev->epd) {
+		rc = -ENOTCONN;
+		dev_err(&cdev->dev, "%s %d scif endpoint not connected rc %d\n",
+			__func__, __LINE__, rc);
+		goto err;
+	}
+
+	rc = scif_send(cdev->epd, &msg, sizeof(msg), SCIF_SEND_BLOCK);
+	if (rc < 0) {
+		dev_err(&cdev->dev, "%s %d scif_send failed rc %d\n",
+			__func__, __LINE__, rc);
+		goto err;
+	}
+	cdev->heartbeat_watchdog_enable = false;
+	cosm_set_state(cdev, MIC_SHUTTING_DOWN);
+	rc = 0;
+err:
+	mutex_unlock(&cdev->cosm_mutex);
+	return rc;
+}
+
+static int cosm_driver_probe(struct cosm_device *cdev)
+{
+	int rc;
+
+	/* Initialize SCIF server at first probe */
+	if (atomic_add_return(1, &g_num_dev) == 1) {
+		rc = cosm_scif_init();
+		if (rc)
+			goto scif_exit;
+	}
+	mutex_init(&cdev->cosm_mutex);
+	INIT_WORK(&cdev->reset_trigger_work, cosm_reset_trigger_work);
+	INIT_WORK(&cdev->scif_work, cosm_scif_work);
+	cdev->sysfs_heartbeat_enable = true;
+	cosm_sysfs_init(cdev);
+	cdev->sdev = device_create_with_groups(g_cosm_class, cdev->dev.parent,
+			       MKDEV(0, cdev->index), cdev, cdev->attr_group,
+			       "mic%d", cdev->index);
+	if (IS_ERR(cdev->sdev)) {
+		rc = PTR_ERR(cdev->sdev);
+		dev_err(&cdev->dev, "device_create_with_groups failed rc %d\n",
+			rc);
+		goto scif_exit;
+	}
+
+	cdev->state_sysfs = sysfs_get_dirent(cdev->sdev->kobj.sd,
+		"state");
+	if (!cdev->state_sysfs) {
+		rc = -ENODEV;
+		dev_err(&cdev->dev, "sysfs_get_dirent failed rc %d\n", rc);
+		goto destroy_device;
+	}
+	cosm_create_debug_dir(cdev);
+	return 0;
+destroy_device:
+	device_destroy(g_cosm_class, MKDEV(0, cdev->index));
+scif_exit:
+	if (atomic_dec_and_test(&g_num_dev))
+		cosm_scif_exit();
+	return rc;
+}
+
+static void cosm_driver_remove(struct cosm_device *cdev)
+{
+	cosm_delete_debug_dir(cdev);
+	sysfs_put(cdev->state_sysfs);
+	device_destroy(g_cosm_class, MKDEV(0, cdev->index));
+	flush_work(&cdev->reset_trigger_work);
+	cosm_stop(cdev, false);
+	if (atomic_dec_and_test(&g_num_dev))
+		cosm_scif_exit();
+
+	/* These sysfs entries might have allocated */
+	kfree(cdev->cmdline);
+	kfree(cdev->firmware);
+	kfree(cdev->ramdisk);
+	kfree(cdev->bootmode);
+}
+
+static int cosm_suspend(struct device *dev)
+{
+	struct cosm_device *cdev = dev_to_cosm(dev);
+
+	mutex_lock(&cdev->cosm_mutex);
+	switch (cdev->state) {
+	/**
+	 * Suspend/freeze hooks in userspace have already shutdown the card.
+	 * Card should be 'ready' in most cases. It is however possible that
+	 * some userspace application initiated a boot. In those cases, we
+	 * simply reset the card.
+	 */
+	case MIC_ONLINE:
+	case MIC_BOOTING:
+	case MIC_SHUTTING_DOWN:
+		mutex_unlock(&cdev->cosm_mutex);
+		cosm_stop(cdev, false);
+		break;
+	default:
+		mutex_unlock(&cdev->cosm_mutex);
+		break;
+	}
+	return 0;
+}
+
+static const struct dev_pm_ops cosm_pm_ops = {
+	.suspend = cosm_suspend,
+	.freeze = cosm_suspend
+};
+
+static struct cosm_driver cosm_driver = {
+	.driver = {
+		.name =  KBUILD_MODNAME,
+		.owner = THIS_MODULE,
+		.pm = &cosm_pm_ops,
+	},
+	.probe = cosm_driver_probe,
+	.remove = cosm_driver_remove
+};
+
+static int __init cosm_init(void)
+{
+	int ret;
+
+	cosm_init_debugfs();
+
+	g_cosm_class = class_create(THIS_MODULE, cosm_driver_name);
+	if (IS_ERR(g_cosm_class)) {
+		ret = PTR_ERR(g_cosm_class);
+		pr_err("class_create failed ret %d\n", ret);
+		goto cleanup_debugfs;
+	}
+
+	ida_init(&g_cosm_ida);
+	ret = cosm_register_driver(&cosm_driver);
+	if (ret) {
+		pr_err("cosm_register_driver failed ret %d\n", ret);
+		goto ida_destroy;
+	}
+	return 0;
+ida_destroy:
+	ida_destroy(&g_cosm_ida);
+	class_destroy(g_cosm_class);
+cleanup_debugfs:
+	cosm_exit_debugfs();
+	return ret;
+}
+
+static void __exit cosm_exit(void)
+{
+	cosm_unregister_driver(&cosm_driver);
+	ida_destroy(&g_cosm_ida);
+	class_destroy(g_cosm_class);
+	cosm_exit_debugfs();
+}
+
+module_init(cosm_init);
+module_exit(cosm_exit);
+
+MODULE_AUTHOR("Intel Corporation");
+MODULE_DESCRIPTION("Intel(R) MIC Coprocessor State Management (COSM) Driver");
+MODULE_LICENSE("GPL v2");

drivers/misc/mic/cosm/cosm_main.h

+/*
+ * Intel MIC Platform Software Stack (MPSS)
+ *
+ * Copyright(c) 2015 Intel Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License, version 2, as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * The full GNU General Public License is included in this distribution in
+ * the file called "COPYING".
+ *
+ * Intel MIC Coprocessor State Management (COSM) Driver
+ *
+ */
+#ifndef _COSM_COSM_H_
+#define _COSM_COSM_H_
+
+#include <linux/scif.h>
+#include "../bus/cosm_bus.h"
+
+#define COSM_HEARTBEAT_SEND_SEC 30
+#define SCIF_COSM_LISTEN_PORT  201
+
+/**
+ * enum COSM msg id's
+ * @COSM_MSG_SHUTDOWN: host->card trigger shutdown
+ * @COSM_MSG_SYNC_TIME: host->card send host time to card to sync time
+ * @COSM_MSG_HEARTBEAT: card->host heartbeat
+ * @COSM_MSG_SHUTDOWN_STATUS: card->host with shutdown status as payload
+ */
+enum cosm_msg_id {
+	COSM_MSG_SHUTDOWN,
+	COSM_MSG_SYNC_TIME,
+	COSM_MSG_HEARTBEAT,
+	COSM_MSG_SHUTDOWN_STATUS,
+};
+
+struct cosm_msg {
+	u64 id;
+	union {
+		u64 shutdown_status;
+		struct timespec64 timespec;
+	};
+};
+
+extern const char * const cosm_state_string[];
+extern const char * const cosm_shutdown_status_string[];
+
+void cosm_sysfs_init(struct cosm_device *cdev);
+int cosm_start(struct cosm_device *cdev);
+void cosm_stop(struct cosm_device *cdev, bool force);
+int cosm_reset(struct cosm_device *cdev);
+int cosm_shutdown(struct cosm_device *cdev);
+void cosm_set_state(struct cosm_device *cdev, u8 state);
+void cosm_set_shutdown_status(struct cosm_device *cdev, u8 status);
+void cosm_init_debugfs(void);
+void cosm_exit_debugfs(void);
+void cosm_create_debug_dir(struct cosm_device *cdev);
+void cosm_delete_debug_dir(struct cosm_device *cdev);
+int cosm_scif_init(void);
+void cosm_scif_exit(void);
+void cosm_scif_work(struct work_struct *work);
+
+#endif