[linux-2.6-block.git] / drivers / gpu / drm / amd / amdkfd / kfd_chardev.c

/*
 * Copyright 2014 Advanced Micro Devices, Inc.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
 * OTHER DEALINGS IN THE SOFTWARE.
 */

#include <linux/device.h>
#include <linux/export.h>
#include <linux/err.h>
#include <linux/fs.h>
#include <linux/sched.h>
#include <linux/slab.h>
#include <linux/uaccess.h>
#include <linux/compat.h>
#include <uapi/linux/kfd_ioctl.h>
#include <linux/time.h>
#include <linux/mm.h>
#include <uapi/asm-generic/mman-common.h>
#include <asm/processor.h>
#include "kfd_priv.h"
#include "kfd_device_queue_manager.h"

static long kfd_ioctl(struct file *, unsigned int, unsigned long);
static int kfd_open(struct inode *, struct file *);
static int kfd_mmap(struct file *, struct vm_area_struct *);

static const char kfd_dev_name[] = "kfd";

static const struct file_operations kfd_fops = {
	.owner = THIS_MODULE,
	.unlocked_ioctl = kfd_ioctl,
	.compat_ioctl = kfd_ioctl,
	.open = kfd_open,
	.mmap = kfd_mmap,
};

static int kfd_char_dev_major = -1;
static struct class *kfd_class;
struct device *kfd_device;

int kfd_chardev_init(void)
{
	int err = 0;

	kfd_char_dev_major = register_chrdev(0, kfd_dev_name, &kfd_fops);
	err = kfd_char_dev_major;
	if (err < 0)
		goto err_register_chrdev;

	kfd_class = class_create(THIS_MODULE, kfd_dev_name);
	err = PTR_ERR(kfd_class);
	if (IS_ERR(kfd_class))
		goto err_class_create;

	kfd_device = device_create(kfd_class, NULL,
					MKDEV(kfd_char_dev_major, 0),
					NULL, kfd_dev_name);
	err = PTR_ERR(kfd_device);
	if (IS_ERR(kfd_device))
		goto err_device_create;

	return 0;

err_device_create:
	class_destroy(kfd_class);
err_class_create:
	unregister_chrdev(kfd_char_dev_major, kfd_dev_name);
err_register_chrdev:
	return err;
}

void kfd_chardev_exit(void)
{
	device_destroy(kfd_class, MKDEV(kfd_char_dev_major, 0));
	class_destroy(kfd_class);
	unregister_chrdev(kfd_char_dev_major, kfd_dev_name);
}

struct device *kfd_chardev(void)
{
	return kfd_device;
}


static int kfd_open(struct inode *inode, struct file *filep)
{
	struct kfd_process *process;
	bool is_32bit_user_mode;

	if (iminor(inode) != 0)
		return -ENODEV;

	is_32bit_user_mode = is_compat_task();

	if (is_32bit_user_mode == true) {
		dev_warn(kfd_device,
			"Process %d (32-bit) failed to open /dev/kfd\n"
			"32-bit processes are not supported by amdkfd\n",
			current->pid);
		return -EPERM;
	}

	process = kfd_create_process(current);
	if (IS_ERR(process))
		return PTR_ERR(process);

	dev_dbg(kfd_device, "process %d opened, compat mode (32 bit) - %d\n",
		process->pasid, process->is_32bit_user_mode);

	return 0;
}

static int kfd_ioctl_get_version(struct file *filep, struct kfd_process *p,
					void *data)
{
	struct kfd_ioctl_get_version_args *args = data;
	int err = 0;

	args->major_version = KFD_IOCTL_MAJOR_VERSION;
	args->minor_version = KFD_IOCTL_MINOR_VERSION;

	return err;
}

static int set_queue_properties_from_user(struct queue_properties *q_properties,
				struct kfd_ioctl_create_queue_args *args)
{
	if (args->queue_percentage > KFD_MAX_QUEUE_PERCENTAGE) {
		pr_err("kfd: queue percentage must be between 0 to KFD_MAX_QUEUE_PERCENTAGE\n");
		return -EINVAL;
	}

	if (args->queue_priority > KFD_MAX_QUEUE_PRIORITY) {
		pr_err("kfd: queue priority must be between 0 to KFD_MAX_QUEUE_PRIORITY\n");
		return -EINVAL;
	}

	if ((args->ring_base_address) &&
		(!access_ok(VERIFY_WRITE,
			(const void __user *) args->ring_base_address,
			sizeof(uint64_t)))) {
		pr_err("kfd: can't access ring base address\n");
		return -EFAULT;
	}

	if (!is_power_of_2(args->ring_size) && (args->ring_size != 0)) {
		pr_err("kfd: ring size must be a power of 2 or 0\n");
		return -EINVAL;
	}

	if (!access_ok(VERIFY_WRITE,
			(const void __user *) args->read_pointer_address,
			sizeof(uint32_t))) {
		pr_err("kfd: can't access read pointer\n");
		return -EFAULT;
	}

	if (!access_ok(VERIFY_WRITE,
			(const void __user *) args->write_pointer_address,
			sizeof(uint32_t))) {
		pr_err("kfd: can't access write pointer\n");
		return -EFAULT;
	}

	q_properties->is_interop = false;
	q_properties->queue_percent = args->queue_percentage;
	q_properties->priority = args->queue_priority;
	q_properties->queue_address = args->ring_base_address;
	q_properties->queue_size = args->ring_size;
	q_properties->read_ptr = (uint32_t *) args->read_pointer_address;
	q_properties->write_ptr = (uint32_t *) args->write_pointer_address;
	if (args->queue_type == KFD_IOC_QUEUE_TYPE_COMPUTE ||
		args->queue_type == KFD_IOC_QUEUE_TYPE_COMPUTE_AQL)
		q_properties->type = KFD_QUEUE_TYPE_COMPUTE;
	else
		return -ENOTSUPP;

	if (args->queue_type == KFD_IOC_QUEUE_TYPE_COMPUTE_AQL)
		q_properties->format = KFD_QUEUE_FORMAT_AQL;
	else
		q_properties->format = KFD_QUEUE_FORMAT_PM4;

	pr_debug("Queue Percentage (%d, %d)\n",
			q_properties->queue_percent, args->queue_percentage);

	pr_debug("Queue Priority (%d, %d)\n",
			q_properties->priority, args->queue_priority);

	pr_debug("Queue Address (0x%llX, 0x%llX)\n",
			q_properties->queue_address, args->ring_base_address);

	pr_debug("Queue Size (0x%llX, %u)\n",
			q_properties->queue_size, args->ring_size);

	pr_debug("Queue r/w Pointers (0x%llX, 0x%llX)\n",
			(uint64_t) q_properties->read_ptr,
			(uint64_t) q_properties->write_ptr);

	pr_debug("Queue Format (%d)\n", q_properties->format);

	return 0;
}

static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p,
					void *data)
{
	struct kfd_ioctl_create_queue_args *args = data;
	struct kfd_dev *dev;
	int err = 0;
	unsigned int queue_id;
	struct kfd_process_device *pdd;
	struct queue_properties q_properties;

	memset(&q_properties, 0, sizeof(struct queue_properties));

	pr_debug("kfd: creating queue ioctl\n");

	err = set_queue_properties_from_user(&q_properties, args);
	if (err)
		return err;

	dev = kfd_device_by_id(args->gpu_id);
	if (dev == NULL)
		return -EINVAL;

	mutex_lock(&p->mutex);

	pdd = kfd_bind_process_to_device(dev, p);
	if (IS_ERR(pdd)) {
		err = -ESRCH;
		goto err_bind_process;
	}

	pr_debug("kfd: creating queue for PASID %d on GPU 0x%x\n",
			p->pasid,
			dev->id);

	err = pqm_create_queue(&p->pqm, dev, filep, &q_properties, 0,
				KFD_QUEUE_TYPE_COMPUTE, &queue_id);
	if (err != 0)
		goto err_create_queue;

	args->queue_id = queue_id;

	/* Return gpu_id as doorbell offset for mmap usage */
	args->doorbell_offset = args->gpu_id << PAGE_SHIFT;

	mutex_unlock(&p->mutex);

	pr_debug("kfd: queue id %d was created successfully\n", args->queue_id);

	pr_debug("ring buffer address == 0x%016llX\n",
			args->ring_base_address);

	pr_debug("read ptr address    == 0x%016llX\n",
			args->read_pointer_address);

	pr_debug("write ptr address   == 0x%016llX\n",
			args->write_pointer_address);

	return 0;

err_create_queue:
err_bind_process:
	mutex_unlock(&p->mutex);
	return err;
}

static int kfd_ioctl_destroy_queue(struct file *filp, struct kfd_process *p,
					void *data)
{
	int retval;
	struct kfd_ioctl_destroy_queue_args *args = data;

	pr_debug("kfd: destroying queue id %d for PASID %d\n",
				args->queue_id,
				p->pasid);

	mutex_lock(&p->mutex);

	retval = pqm_destroy_queue(&p->pqm, args->queue_id);

	mutex_unlock(&p->mutex);
	return retval;
}

static int kfd_ioctl_update_queue(struct file *filp, struct kfd_process *p,
					void *data)
{
	int retval;
	struct kfd_ioctl_update_queue_args *args = data;
	struct queue_properties properties;

	if (args->queue_percentage > KFD_MAX_QUEUE_PERCENTAGE) {
		pr_err("kfd: queue percentage must be between 0 to KFD_MAX_QUEUE_PERCENTAGE\n");
		return -EINVAL;
	}

	if (args->queue_priority > KFD_MAX_QUEUE_PRIORITY) {
		pr_err("kfd: queue priority must be between 0 to KFD_MAX_QUEUE_PRIORITY\n");
		return -EINVAL;
	}

	if ((args->ring_base_address) &&
		(!access_ok(VERIFY_WRITE,
			(const void __user *) args->ring_base_address,
			sizeof(uint64_t)))) {
		pr_err("kfd: can't access ring base address\n");
		return -EFAULT;
	}

	if (!is_power_of_2(args->ring_size) && (args->ring_size != 0)) {
		pr_err("kfd: ring size must be a power of 2 or 0\n");
		return -EINVAL;
	}

	properties.queue_address = args->ring_base_address;
	properties.queue_size = args->ring_size;
	properties.queue_percent = args->queue_percentage;
	properties.priority = args->queue_priority;

	pr_debug("kfd: updating queue id %d for PASID %d\n",
			args->queue_id, p->pasid);

	mutex_lock(&p->mutex);

	retval = pqm_update_queue(&p->pqm, args->queue_id, &properties);

	mutex_unlock(&p->mutex);

	return retval;
}

static int kfd_ioctl_set_memory_policy(struct file *filep,
					struct kfd_process *p, void *data)
{
	struct kfd_ioctl_set_memory_policy_args *args = data;
	struct kfd_dev *dev;
	int err = 0;
	struct kfd_process_device *pdd;
	enum cache_policy default_policy, alternate_policy;

	if (args->default_policy != KFD_IOC_CACHE_POLICY_COHERENT
	    && args->default_policy != KFD_IOC_CACHE_POLICY_NONCOHERENT) {
		return -EINVAL;
	}

	if (args->alternate_policy != KFD_IOC_CACHE_POLICY_COHERENT
	    && args->alternate_policy != KFD_IOC_CACHE_POLICY_NONCOHERENT) {
		return -EINVAL;
	}

	dev = kfd_device_by_id(args->gpu_id);
	if (dev == NULL)
		return -EINVAL;

	mutex_lock(&p->mutex);

	pdd = kfd_bind_process_to_device(dev, p);
	if (IS_ERR(pdd)) {
		err = -ESRCH;
		goto out;
	}

	default_policy = (args->default_policy == KFD_IOC_CACHE_POLICY_COHERENT)
			 ? cache_policy_coherent : cache_policy_noncoherent;

	alternate_policy =
		(args->alternate_policy == KFD_IOC_CACHE_POLICY_COHERENT)
		   ? cache_policy_coherent : cache_policy_noncoherent;

	if (!dev->dqm->set_cache_memory_policy(dev->dqm,
				&pdd->qpd,
				default_policy,
				alternate_policy,
				(void __user *)args->alternate_aperture_base,
				args->alternate_aperture_size))
		err = -EINVAL;

out:
	mutex_unlock(&p->mutex);

	return err;
}

static int kfd_ioctl_get_clock_counters(struct file *filep,
				struct kfd_process *p, void *data)
{
	struct kfd_ioctl_get_clock_counters_args *args = data;
	struct kfd_dev *dev;
	struct timespec time;

	dev = kfd_device_by_id(args->gpu_id);
	if (dev == NULL)
		return -EINVAL;

	/* Reading GPU clock counter from KGD */
	args->gpu_clock_counter = kfd2kgd->get_gpu_clock_counter(dev->kgd);

	/* No access to rdtsc. Using raw monotonic time */
	getrawmonotonic(&time);
	args->cpu_clock_counter = (uint64_t)timespec_to_ns(&time);

	get_monotonic_boottime(&time);
	args->system_clock_counter = (uint64_t)timespec_to_ns(&time);

	/* Since the counter is in nano-seconds we use 1GHz frequency */
	args->system_clock_freq = 1000000000;

	return 0;
}


static int kfd_ioctl_get_process_apertures(struct file *filp,
				struct kfd_process *p, void *data)
{
	struct kfd_ioctl_get_process_apertures_args *args = data;
	struct kfd_process_device_apertures *pAperture;
	struct kfd_process_device *pdd;

	dev_dbg(kfd_device, "get apertures for PASID %d", p->pasid);

	args->num_of_nodes = 0;

	mutex_lock(&p->mutex);

	/*if the process-device list isn't empty*/
	if (kfd_has_process_device_data(p)) {
		/* Run over all pdd of the process */
		pdd = kfd_get_first_process_device_data(p);
		do {
			pAperture =
				&args->process_apertures[args->num_of_nodes];
			pAperture->gpu_id = pdd->dev->id;
			pAperture->lds_base = pdd->lds_base;
			pAperture->lds_limit = pdd->lds_limit;
			pAperture->gpuvm_base = pdd->gpuvm_base;
			pAperture->gpuvm_limit = pdd->gpuvm_limit;
			pAperture->scratch_base = pdd->scratch_base;
			pAperture->scratch_limit = pdd->scratch_limit;

			dev_dbg(kfd_device,
				"node id %u\n", args->num_of_nodes);
			dev_dbg(kfd_device,
				"gpu id %u\n", pdd->dev->id);
			dev_dbg(kfd_device,
				"lds_base %llX\n", pdd->lds_base);
			dev_dbg(kfd_device,
				"lds_limit %llX\n", pdd->lds_limit);
			dev_dbg(kfd_device,
				"gpuvm_base %llX\n", pdd->gpuvm_base);
			dev_dbg(kfd_device,
				"gpuvm_limit %llX\n", pdd->gpuvm_limit);
			dev_dbg(kfd_device,
				"scratch_base %llX\n", pdd->scratch_base);
			dev_dbg(kfd_device,
				"scratch_limit %llX\n", pdd->scratch_limit);

			args->num_of_nodes++;
		} while ((pdd = kfd_get_next_process_device_data(p, pdd)) != NULL &&
				(args->num_of_nodes < NUM_OF_SUPPORTED_GPUS));
	}

	mutex_unlock(&p->mutex);

	return 0;
}

#define AMDKFD_IOCTL_DEF(ioctl, _func, _flags) \
	[_IOC_NR(ioctl)] = {.cmd = ioctl, .func = _func, .flags = _flags, .cmd_drv = 0, .name = #ioctl}

/** Ioctl table */
static const struct amdkfd_ioctl_desc amdkfd_ioctls[] = {
	AMDKFD_IOCTL_DEF(AMDKFD_IOC_GET_VERSION,
			kfd_ioctl_get_version, 0),

	AMDKFD_IOCTL_DEF(AMDKFD_IOC_CREATE_QUEUE,
			kfd_ioctl_create_queue, 0),

	AMDKFD_IOCTL_DEF(AMDKFD_IOC_DESTROY_QUEUE,
			kfd_ioctl_destroy_queue, 0),

	AMDKFD_IOCTL_DEF(AMDKFD_IOC_SET_MEMORY_POLICY,
			kfd_ioctl_set_memory_policy, 0),

	AMDKFD_IOCTL_DEF(AMDKFD_IOC_GET_CLOCK_COUNTERS,
			kfd_ioctl_get_clock_counters, 0),

	AMDKFD_IOCTL_DEF(AMDKFD_IOC_GET_PROCESS_APERTURES,
			kfd_ioctl_get_process_apertures, 0),

	AMDKFD_IOCTL_DEF(AMDKFD_IOC_UPDATE_QUEUE,
			kfd_ioctl_update_queue, 0),
};

#define AMDKFD_CORE_IOCTL_COUNT	ARRAY_SIZE(amdkfd_ioctls)

static long kfd_ioctl(struct file *filep, unsigned int cmd, unsigned long arg)
{
	struct kfd_process *process;
	amdkfd_ioctl_t *func;
	const struct amdkfd_ioctl_desc *ioctl = NULL;
	unsigned int nr = _IOC_NR(cmd);
	char stack_kdata[128];
	char *kdata = NULL;
	unsigned int usize, asize;
	int retcode = -EINVAL;

	if (nr >= AMDKFD_CORE_IOCTL_COUNT)
		goto err_i1;

	if ((nr >= AMDKFD_COMMAND_START) && (nr < AMDKFD_COMMAND_END)) {
		u32 amdkfd_size;

		ioctl = &amdkfd_ioctls[nr];

		amdkfd_size = _IOC_SIZE(ioctl->cmd);
		usize = asize = _IOC_SIZE(cmd);
		if (amdkfd_size > asize)
			asize = amdkfd_size;

		cmd = ioctl->cmd;
	} else
		goto err_i1;

	dev_dbg(kfd_device, "ioctl cmd 0x%x (#%d), arg 0x%lx\n", cmd, nr, arg);

	process = kfd_get_process(current);
	if (IS_ERR(process)) {
		dev_dbg(kfd_device, "no process\n");
		goto err_i1;
	}

	/* Do not trust userspace, use our own definition */
	func = ioctl->func;

	if (unlikely(!func)) {
		dev_dbg(kfd_device, "no function\n");
		retcode = -EINVAL;
		goto err_i1;
	}

	if (cmd & (IOC_IN | IOC_OUT)) {
		if (asize <= sizeof(stack_kdata)) {
			kdata = stack_kdata;
		} else {
			kdata = kmalloc(asize, GFP_KERNEL);
			if (!kdata) {
				retcode = -ENOMEM;
				goto err_i1;
			}
		}
		if (asize > usize)
			memset(kdata + usize, 0, asize - usize);
	}

	if (cmd & IOC_IN) {
		if (copy_from_user(kdata, (void __user *)arg, usize) != 0) {
			retcode = -EFAULT;
			goto err_i1;
		}
	} else if (cmd & IOC_OUT) {
		memset(kdata, 0, usize);
	}

	retcode = func(filep, process, kdata);

	if (cmd & IOC_OUT)
		if (copy_to_user((void __user *)arg, kdata, usize) != 0)
			retcode = -EFAULT;

err_i1:
	if (!ioctl)
		dev_dbg(kfd_device, "invalid ioctl: pid=%d, cmd=0x%02x, nr=0x%02x\n",
			  task_pid_nr(current), cmd, nr);

	if (kdata != stack_kdata)
		kfree(kdata);

	if (retcode)
		dev_dbg(kfd_device, "ret = %d\n", retcode);

	return retcode;
}

static int kfd_mmap(struct file *filp, struct vm_area_struct *vma)
{
	struct kfd_process *process;

	process = kfd_get_process(current);
	if (IS_ERR(process))
		return PTR_ERR(process);

	return kfd_doorbell_mmap(process, vma);
}
Commit	Line	Data
4a488a7a OG	1	/*
	2	* Copyright 2014 Advanced Micro Devices, Inc.
	3	*
	4	* Permission is hereby granted, free of charge, to any person obtaining a
	5	* copy of this software and associated documentation files (the "Software"),
	6	* to deal in the Software without restriction, including without limitation
	7	* the rights to use, copy, modify, merge, publish, distribute, sublicense,
	8	* and/or sell copies of the Software, and to permit persons to whom the
	9	* Software is furnished to do so, subject to the following conditions:
	10	*
	11	* The above copyright notice and this permission notice shall be included in
	12	* all copies or substantial portions of the Software.
	13	*
	14	* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
	15	* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
	16	* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
	17	* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
	18	* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
	19	* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
	20	* OTHER DEALINGS IN THE SOFTWARE.
	21	*/
	22
	23	#include <linux/device.h>
	24	#include <linux/export.h>
	25	#include <linux/err.h>
	26	#include <linux/fs.h>
	27	#include <linux/sched.h>
	28	#include <linux/slab.h>
	29	#include <linux/uaccess.h>
	30	#include <linux/compat.h>
	31	#include <uapi/linux/kfd_ioctl.h>
	32	#include <linux/time.h>
	33	#include <linux/mm.h>
4a488a7a OG	34	#include <uapi/asm-generic/mman-common.h>
	35	#include <asm/processor.h>
	36	#include "kfd_priv.h"
41a286fa	37	#include "kfd_device_queue_manager.h"
4a488a7a OG	38
	39	static long kfd_ioctl(struct file *, unsigned int, unsigned long);
	40	static int kfd_open(struct inode , struct file );
19f6d2a6	41	static int kfd_mmap(struct file , struct vm_area_struct );
4a488a7a OG	42
	43	static const char kfd_dev_name[] = "kfd";
	44
	45	static const struct file_operations kfd_fops = {
	46	.owner = THIS_MODULE,
	47	.unlocked_ioctl = kfd_ioctl,
	48	.compat_ioctl = kfd_ioctl,
	49	.open = kfd_open,
19f6d2a6	50	.mmap = kfd_mmap,
4a488a7a OG	51	};
	52
	53	static int kfd_char_dev_major = -1;
	54	static struct class *kfd_class;
	55	struct device *kfd_device;
	56
	57	int kfd_chardev_init(void)
	58	{
	59	int err = 0;
	60
	61	kfd_char_dev_major = register_chrdev(0, kfd_dev_name, &kfd_fops);
	62	err = kfd_char_dev_major;
	63	if (err < 0)
	64	goto err_register_chrdev;
	65
	66	kfd_class = class_create(THIS_MODULE, kfd_dev_name);
	67	err = PTR_ERR(kfd_class);
	68	if (IS_ERR(kfd_class))
	69	goto err_class_create;
	70
	71	kfd_device = device_create(kfd_class, NULL,
	72	MKDEV(kfd_char_dev_major, 0),
	73	NULL, kfd_dev_name);
	74	err = PTR_ERR(kfd_device);
	75	if (IS_ERR(kfd_device))
	76	goto err_device_create;
	77
	78	return 0;
	79
	80	err_device_create:
	81	class_destroy(kfd_class);
	82	err_class_create:
	83	unregister_chrdev(kfd_char_dev_major, kfd_dev_name);
	84	err_register_chrdev:
	85	return err;
	86	}
	87
	88	void kfd_chardev_exit(void)
	89	{
	90	device_destroy(kfd_class, MKDEV(kfd_char_dev_major, 0));
	91	class_destroy(kfd_class);
	92	unregister_chrdev(kfd_char_dev_major, kfd_dev_name);
	93	}
	94
	95	struct device *kfd_chardev(void)
	96	{
	97	return kfd_device;
	98	}
	99
	100
	101	static int kfd_open(struct inode inode, struct file filep)
	102	{
19f6d2a6	103	struct kfd_process *process;
a18069c1	104	bool is_32bit_user_mode;
19f6d2a6	105
4a488a7a OG	106	if (iminor(inode) != 0)
	107	return -ENODEV;
	108
a18069c1 OG	109	is_32bit_user_mode = is_compat_task();
	110
	111	if (is_32bit_user_mode == true) {
	112	dev_warn(kfd_device,
	113	"Process %d (32-bit) failed to open /dev/kfd\n"
	114	"32-bit processes are not supported by amdkfd\n",
	115	current->pid);
	116	return -EPERM;
	117	}
	118
19f6d2a6 OG	119	process = kfd_create_process(current);
	120	if (IS_ERR(process))
	121	return PTR_ERR(process);
	122
19f6d2a6 OG	123	dev_dbg(kfd_device, "process %d opened, compat mode (32 bit) - %d\n",
	124	process->pasid, process->is_32bit_user_mode);
	125
4a488a7a OG	126	return 0;
	127	}
	128
524a6404 OG	129	static int kfd_ioctl_get_version(struct file filep, struct kfd_process p,
524a6404 OG	130	void *data)
4a488a7a	131	{
524a6404	132	struct kfd_ioctl_get_version_args *args = data;
ecd5c982 OG	133	int err = 0;
ecd5c982 OG	134
524a6404 OG	135	args->major_version = KFD_IOCTL_MAJOR_VERSION;
524a6404 OG	136	args->minor_version = KFD_IOCTL_MINOR_VERSION;
ecd5c982 OG	137
ecd5c982 OG	138	return err;
4a488a7a OG	139	}
4a488a7a OG	140
39b027d9 OG	141	static int set_queue_properties_from_user(struct queue_properties *q_properties,
	142	struct kfd_ioctl_create_queue_args *args)
	143	{
	144	if (args->queue_percentage > KFD_MAX_QUEUE_PERCENTAGE) {
	145	pr_err("kfd: queue percentage must be between 0 to KFD_MAX_QUEUE_PERCENTAGE\n");
	146	return -EINVAL;
	147	}
	148
	149	if (args->queue_priority > KFD_MAX_QUEUE_PRIORITY) {
	150	pr_err("kfd: queue priority must be between 0 to KFD_MAX_QUEUE_PRIORITY\n");
	151	return -EINVAL;
	152	}
	153
	154	if ((args->ring_base_address) &&
4307d8f6 OG	155	(!access_ok(VERIFY_WRITE,
	156	(const void __user *) args->ring_base_address,
	157	sizeof(uint64_t)))) {
39b027d9 OG	158	pr_err("kfd: can't access ring base address\n");
	159	return -EFAULT;
	160	}
	161
	162	if (!is_power_of_2(args->ring_size) && (args->ring_size != 0)) {
	163	pr_err("kfd: ring size must be a power of 2 or 0\n");
	164	return -EINVAL;
	165	}
	166
4307d8f6 OG	167	if (!access_ok(VERIFY_WRITE,
	168	(const void __user *) args->read_pointer_address,
	169	sizeof(uint32_t))) {
39b027d9 OG	170	pr_err("kfd: can't access read pointer\n");
	171	return -EFAULT;
	172	}
	173
4307d8f6 OG	174	if (!access_ok(VERIFY_WRITE,
	175	(const void __user *) args->write_pointer_address,
	176	sizeof(uint32_t))) {
39b027d9 OG	177	pr_err("kfd: can't access write pointer\n");
	178	return -EFAULT;
	179	}
	180
	181	q_properties->is_interop = false;
	182	q_properties->queue_percent = args->queue_percentage;
	183	q_properties->priority = args->queue_priority;
	184	q_properties->queue_address = args->ring_base_address;
	185	q_properties->queue_size = args->ring_size;
	186	q_properties->read_ptr = (uint32_t *) args->read_pointer_address;
	187	q_properties->write_ptr = (uint32_t *) args->write_pointer_address;
	188	if (args->queue_type == KFD_IOC_QUEUE_TYPE_COMPUTE \|\|
	189	args->queue_type == KFD_IOC_QUEUE_TYPE_COMPUTE_AQL)
	190	q_properties->type = KFD_QUEUE_TYPE_COMPUTE;
	191	else
	192	return -ENOTSUPP;
	193
	194	if (args->queue_type == KFD_IOC_QUEUE_TYPE_COMPUTE_AQL)
	195	q_properties->format = KFD_QUEUE_FORMAT_AQL;
	196	else
	197	q_properties->format = KFD_QUEUE_FORMAT_PM4;
	198
	199	pr_debug("Queue Percentage (%d, %d)\n",
	200	q_properties->queue_percent, args->queue_percentage);
	201
	202	pr_debug("Queue Priority (%d, %d)\n",
	203	q_properties->priority, args->queue_priority);
	204
	205	pr_debug("Queue Address (0x%llX, 0x%llX)\n",
	206	q_properties->queue_address, args->ring_base_address);
	207
	208	pr_debug("Queue Size (0x%llX, %u)\n",
	209	q_properties->queue_size, args->ring_size);
	210
	211	pr_debug("Queue r/w Pointers (0x%llX, 0x%llX)\n",
	212	(uint64_t) q_properties->read_ptr,
	213	(uint64_t) q_properties->write_ptr);
	214
	215	pr_debug("Queue Format (%d)\n", q_properties->format);
	216
	217	return 0;
	218	}
	219
524a6404 OG	220	static int kfd_ioctl_create_queue(struct file filep, struct kfd_process p,
524a6404 OG	221	void *data)
4a488a7a	222	{
524a6404	223	struct kfd_ioctl_create_queue_args *args = data;
39b027d9 OG	224	struct kfd_dev *dev;
	225	int err = 0;
	226	unsigned int queue_id;
	227	struct kfd_process_device *pdd;
	228	struct queue_properties q_properties;
	229
	230	memset(&q_properties, 0, sizeof(struct queue_properties));
	231
39b027d9 OG	232	pr_debug("kfd: creating queue ioctl\n");
39b027d9 OG	233
524a6404	234	err = set_queue_properties_from_user(&q_properties, args);
39b027d9 OG	235	if (err)
	236	return err;
	237
524a6404	238	dev = kfd_device_by_id(args->gpu_id);
39b027d9 OG	239	if (dev == NULL)
	240	return -EINVAL;
	241
	242	mutex_lock(&p->mutex);
	243
	244	pdd = kfd_bind_process_to_device(dev, p);
66333cb3	245	if (IS_ERR(pdd)) {
524a6404	246	err = -ESRCH;
39b027d9 OG	247	goto err_bind_process;
	248	}
	249
	250	pr_debug("kfd: creating queue for PASID %d on GPU 0x%x\n",
	251	p->pasid,
	252	dev->id);
	253
	254	err = pqm_create_queue(&p->pqm, dev, filep, &q_properties, 0,
	255	KFD_QUEUE_TYPE_COMPUTE, &queue_id);
	256	if (err != 0)
	257	goto err_create_queue;
	258
524a6404	259	args->queue_id = queue_id;
39b027d9 OG	260
39b027d9 OG	261	/* Return gpu_id as doorbell offset for mmap usage */
524a6404	262	args->doorbell_offset = args->gpu_id << PAGE_SHIFT;
39b027d9 OG	263
	264	mutex_unlock(&p->mutex);
	265
524a6404	266	pr_debug("kfd: queue id %d was created successfully\n", args->queue_id);
39b027d9 OG	267
39b027d9 OG	268	pr_debug("ring buffer address == 0x%016llX\n",
524a6404	269	args->ring_base_address);
39b027d9 OG	270
39b027d9 OG	271	pr_debug("read ptr address == 0x%016llX\n",
524a6404	272	args->read_pointer_address);
39b027d9 OG	273
39b027d9 OG	274	pr_debug("write ptr address == 0x%016llX\n",
524a6404	275	args->write_pointer_address);
39b027d9 OG	276
	277	return 0;
	278
39b027d9 OG	279	err_create_queue:
	280	err_bind_process:
	281	mutex_unlock(&p->mutex);
	282	return err;
4a488a7a OG	283	}
	284
	285	static int kfd_ioctl_destroy_queue(struct file filp, struct kfd_process p,
524a6404	286	void *data)
4a488a7a	287	{
39b027d9	288	int retval;
524a6404	289	struct kfd_ioctl_destroy_queue_args *args = data;
39b027d9 OG	290
39b027d9 OG	291	pr_debug("kfd: destroying queue id %d for PASID %d\n",
524a6404	292	args->queue_id,
39b027d9 OG	293	p->pasid);
	294
	295	mutex_lock(&p->mutex);
	296
524a6404	297	retval = pqm_destroy_queue(&p->pqm, args->queue_id);
39b027d9 OG	298
	299	mutex_unlock(&p->mutex);
	300	return retval;
4a488a7a OG	301	}
	302
	303	static int kfd_ioctl_update_queue(struct file filp, struct kfd_process p,
524a6404	304	void *data)
4a488a7a	305	{
39b027d9	306	int retval;
524a6404	307	struct kfd_ioctl_update_queue_args *args = data;
39b027d9 OG	308	struct queue_properties properties;
39b027d9 OG	309
524a6404	310	if (args->queue_percentage > KFD_MAX_QUEUE_PERCENTAGE) {
39b027d9 OG	311	pr_err("kfd: queue percentage must be between 0 to KFD_MAX_QUEUE_PERCENTAGE\n");
	312	return -EINVAL;
	313	}
	314
524a6404	315	if (args->queue_priority > KFD_MAX_QUEUE_PRIORITY) {
39b027d9 OG	316	pr_err("kfd: queue priority must be between 0 to KFD_MAX_QUEUE_PRIORITY\n");
	317	return -EINVAL;
	318	}
	319
524a6404	320	if ((args->ring_base_address) &&
4307d8f6	321	(!access_ok(VERIFY_WRITE,
524a6404	322	(const void __user *) args->ring_base_address,
4307d8f6	323	sizeof(uint64_t)))) {
39b027d9 OG	324	pr_err("kfd: can't access ring base address\n");
	325	return -EFAULT;
	326	}
	327
524a6404	328	if (!is_power_of_2(args->ring_size) && (args->ring_size != 0)) {
39b027d9 OG	329	pr_err("kfd: ring size must be a power of 2 or 0\n");
	330	return -EINVAL;
	331	}
	332
524a6404 OG	333	properties.queue_address = args->ring_base_address;
	334	properties.queue_size = args->ring_size;
	335	properties.queue_percent = args->queue_percentage;
	336	properties.priority = args->queue_priority;
39b027d9 OG	337
39b027d9 OG	338	pr_debug("kfd: updating queue id %d for PASID %d\n",
524a6404	339	args->queue_id, p->pasid);
39b027d9 OG	340
	341	mutex_lock(&p->mutex);
	342
524a6404	343	retval = pqm_update_queue(&p->pqm, args->queue_id, &properties);
39b027d9 OG	344
	345	mutex_unlock(&p->mutex);
	346
	347	return retval;
4a488a7a OG	348	}
4a488a7a OG	349
524a6404 OG	350	static int kfd_ioctl_set_memory_policy(struct file *filep,
524a6404 OG	351	struct kfd_process p, void data)
4a488a7a	352	{
524a6404	353	struct kfd_ioctl_set_memory_policy_args *args = data;
41a286fa AL	354	struct kfd_dev *dev;
	355	int err = 0;
	356	struct kfd_process_device *pdd;
	357	enum cache_policy default_policy, alternate_policy;
	358
524a6404 OG	359	if (args->default_policy != KFD_IOC_CACHE_POLICY_COHERENT
524a6404 OG	360	&& args->default_policy != KFD_IOC_CACHE_POLICY_NONCOHERENT) {
41a286fa AL	361	return -EINVAL;
	362	}
	363
524a6404 OG	364	if (args->alternate_policy != KFD_IOC_CACHE_POLICY_COHERENT
524a6404 OG	365	&& args->alternate_policy != KFD_IOC_CACHE_POLICY_NONCOHERENT) {
41a286fa AL	366	return -EINVAL;
	367	}
	368
524a6404	369	dev = kfd_device_by_id(args->gpu_id);
41a286fa AL	370	if (dev == NULL)
	371	return -EINVAL;
	372
	373	mutex_lock(&p->mutex);
	374
	375	pdd = kfd_bind_process_to_device(dev, p);
66333cb3	376	if (IS_ERR(pdd)) {
524a6404	377	err = -ESRCH;
41a286fa AL	378	goto out;
	379	}
	380
524a6404	381	default_policy = (args->default_policy == KFD_IOC_CACHE_POLICY_COHERENT)
41a286fa AL	382	? cache_policy_coherent : cache_policy_noncoherent;
	383
	384	alternate_policy =
524a6404	385	(args->alternate_policy == KFD_IOC_CACHE_POLICY_COHERENT)
41a286fa AL	386	? cache_policy_coherent : cache_policy_noncoherent;
	387
	388	if (!dev->dqm->set_cache_memory_policy(dev->dqm,
	389	&pdd->qpd,
	390	default_policy,
	391	alternate_policy,
524a6404 OG	392	(void __user *)args->alternate_aperture_base,
524a6404 OG	393	args->alternate_aperture_size))
41a286fa AL	394	err = -EINVAL;
	395
	396	out:
	397	mutex_unlock(&p->mutex);
	398
	399	return err;
4a488a7a OG	400	}
4a488a7a OG	401
524a6404 OG	402	static int kfd_ioctl_get_clock_counters(struct file *filep,
524a6404 OG	403	struct kfd_process p, void data)
4a488a7a	404	{
524a6404	405	struct kfd_ioctl_get_clock_counters_args *args = data;
4fac47c8 EP	406	struct kfd_dev *dev;
	407	struct timespec time;
	408
524a6404	409	dev = kfd_device_by_id(args->gpu_id);
4fac47c8 EP	410	if (dev == NULL)
	411	return -EINVAL;
	412
	413	/* Reading GPU clock counter from KGD */
524a6404	414	args->gpu_clock_counter = kfd2kgd->get_gpu_clock_counter(dev->kgd);
4fac47c8 EP	415
	416	/* No access to rdtsc. Using raw monotonic time */
	417	getrawmonotonic(&time);
524a6404	418	args->cpu_clock_counter = (uint64_t)timespec_to_ns(&time);
4fac47c8 EP	419
4fac47c8 EP	420	get_monotonic_boottime(&time);
524a6404	421	args->system_clock_counter = (uint64_t)timespec_to_ns(&time);
4fac47c8 EP	422
4fac47c8 EP	423	/* Since the counter is in nano-seconds we use 1GHz frequency */
524a6404	424	args->system_clock_freq = 1000000000;
4fac47c8 EP	425
4fac47c8 EP	426	return 0;
4a488a7a OG	427	}
	428
	429
	430	static int kfd_ioctl_get_process_apertures(struct file *filp,
524a6404	431	struct kfd_process p, void data)
4a488a7a	432	{
524a6404	433	struct kfd_ioctl_get_process_apertures_args *args = data;
775921ed AS	434	struct kfd_process_device_apertures *pAperture;
	435	struct kfd_process_device *pdd;
	436
	437	dev_dbg(kfd_device, "get apertures for PASID %d", p->pasid);
	438
524a6404	439	args->num_of_nodes = 0;
775921ed AS	440
	441	mutex_lock(&p->mutex);
	442
	443	/if the process-device list isn't empty/
	444	if (kfd_has_process_device_data(p)) {
	445	/* Run over all pdd of the process */
	446	pdd = kfd_get_first_process_device_data(p);
	447	do {
524a6404 OG	448	pAperture =
524a6404 OG	449	&args->process_apertures[args->num_of_nodes];
775921ed AS	450	pAperture->gpu_id = pdd->dev->id;
	451	pAperture->lds_base = pdd->lds_base;
	452	pAperture->lds_limit = pdd->lds_limit;
	453	pAperture->gpuvm_base = pdd->gpuvm_base;
	454	pAperture->gpuvm_limit = pdd->gpuvm_limit;
	455	pAperture->scratch_base = pdd->scratch_base;
	456	pAperture->scratch_limit = pdd->scratch_limit;
	457
	458	dev_dbg(kfd_device,
524a6404	459	"node id %u\n", args->num_of_nodes);
775921ed AS	460	dev_dbg(kfd_device,
	461	"gpu id %u\n", pdd->dev->id);
	462	dev_dbg(kfd_device,
	463	"lds_base %llX\n", pdd->lds_base);
	464	dev_dbg(kfd_device,
	465	"lds_limit %llX\n", pdd->lds_limit);
	466	dev_dbg(kfd_device,
	467	"gpuvm_base %llX\n", pdd->gpuvm_base);
	468	dev_dbg(kfd_device,
	469	"gpuvm_limit %llX\n", pdd->gpuvm_limit);
	470	dev_dbg(kfd_device,
	471	"scratch_base %llX\n", pdd->scratch_base);
	472	dev_dbg(kfd_device,
	473	"scratch_limit %llX\n", pdd->scratch_limit);
	474
524a6404	475	args->num_of_nodes++;
775921ed	476	} while ((pdd = kfd_get_next_process_device_data(p, pdd)) != NULL &&
524a6404	477	(args->num_of_nodes < NUM_OF_SUPPORTED_GPUS));
775921ed AS	478	}
	479
	480	mutex_unlock(&p->mutex);
	481
775921ed	482	return 0;
4a488a7a OG	483	}
4a488a7a OG	484
76baee6c OG	485	#define AMDKFD_IOCTL_DEF(ioctl, _func, _flags) \
	486	[_IOC_NR(ioctl)] = {.cmd = ioctl, .func = _func, .flags = _flags, .cmd_drv = 0, .name = #ioctl}
	487
	488	/** Ioctl table */
	489	static const struct amdkfd_ioctl_desc amdkfd_ioctls[] = {
	490	AMDKFD_IOCTL_DEF(AMDKFD_IOC_GET_VERSION,
	491	kfd_ioctl_get_version, 0),
	492
	493	AMDKFD_IOCTL_DEF(AMDKFD_IOC_CREATE_QUEUE,
	494	kfd_ioctl_create_queue, 0),
	495
	496	AMDKFD_IOCTL_DEF(AMDKFD_IOC_DESTROY_QUEUE,
	497	kfd_ioctl_destroy_queue, 0),
	498
	499	AMDKFD_IOCTL_DEF(AMDKFD_IOC_SET_MEMORY_POLICY,
	500	kfd_ioctl_set_memory_policy, 0),
	501
	502	AMDKFD_IOCTL_DEF(AMDKFD_IOC_GET_CLOCK_COUNTERS,
	503	kfd_ioctl_get_clock_counters, 0),
	504
	505	AMDKFD_IOCTL_DEF(AMDKFD_IOC_GET_PROCESS_APERTURES,
	506	kfd_ioctl_get_process_apertures, 0),
	507
	508	AMDKFD_IOCTL_DEF(AMDKFD_IOC_UPDATE_QUEUE,
	509	kfd_ioctl_update_queue, 0),
	510	};
	511
	512	#define AMDKFD_CORE_IOCTL_COUNT ARRAY_SIZE(amdkfd_ioctls)
	513
4a488a7a OG	514	static long kfd_ioctl(struct file *filep, unsigned int cmd, unsigned long arg)
	515	{
	516	struct kfd_process *process;
76baee6c OG	517	amdkfd_ioctl_t *func;
	518	const struct amdkfd_ioctl_desc *ioctl = NULL;
	519	unsigned int nr = _IOC_NR(cmd);
524a6404 OG	520	char stack_kdata[128];
	521	char *kdata = NULL;
	522	unsigned int usize, asize;
	523	int retcode = -EINVAL;
4a488a7a	524
76baee6c OG	525	if (nr >= AMDKFD_CORE_IOCTL_COUNT)
	526	goto err_i1;
	527
	528	if ((nr >= AMDKFD_COMMAND_START) && (nr < AMDKFD_COMMAND_END)) {
	529	u32 amdkfd_size;
	530
	531	ioctl = &amdkfd_ioctls[nr];
	532
	533	amdkfd_size = _IOC_SIZE(ioctl->cmd);
	534	usize = asize = _IOC_SIZE(cmd);
	535	if (amdkfd_size > asize)
	536	asize = amdkfd_size;
	537
	538	cmd = ioctl->cmd;
	539	} else
	540	goto err_i1;
	541
	542	dev_dbg(kfd_device, "ioctl cmd 0x%x (#%d), arg 0x%lx\n", cmd, nr, arg);
4a488a7a	543
19f6d2a6	544	process = kfd_get_process(current);
76baee6c OG	545	if (IS_ERR(process)) {
	546	dev_dbg(kfd_device, "no process\n");
	547	goto err_i1;
	548	}
	549
	550	/* Do not trust userspace, use our own definition */
	551	func = ioctl->func;
	552
	553	if (unlikely(!func)) {
	554	dev_dbg(kfd_device, "no function\n");
	555	retcode = -EINVAL;
	556	goto err_i1;
	557	}
4a488a7a	558
524a6404 OG	559	if (cmd & (IOC_IN \| IOC_OUT)) {
	560	if (asize <= sizeof(stack_kdata)) {
	561	kdata = stack_kdata;
	562	} else {
	563	kdata = kmalloc(asize, GFP_KERNEL);
	564	if (!kdata) {
	565	retcode = -ENOMEM;
	566	goto err_i1;
	567	}
	568	}
	569	if (asize > usize)
	570	memset(kdata + usize, 0, asize - usize);
	571	}
	572
	573	if (cmd & IOC_IN) {
	574	if (copy_from_user(kdata, (void __user *)arg, usize) != 0) {
	575	retcode = -EFAULT;
	576	goto err_i1;
	577	}
	578	} else if (cmd & IOC_OUT) {
	579	memset(kdata, 0, usize);
	580	}
	581
76baee6c	582	retcode = func(filep, process, kdata);
4a488a7a	583
524a6404 OG	584	if (cmd & IOC_OUT)
	585	if (copy_to_user((void __user *)arg, kdata, usize) != 0)
	586	retcode = -EFAULT;
4a488a7a	587
524a6404	588	err_i1:
76baee6c OG	589	if (!ioctl)
	590	dev_dbg(kfd_device, "invalid ioctl: pid=%d, cmd=0x%02x, nr=0x%02x\n",
	591	task_pid_nr(current), cmd, nr);
	592
524a6404 OG	593	if (kdata != stack_kdata)
	594	kfree(kdata);
	595
	596	if (retcode)
	597	dev_dbg(kfd_device, "ret = %d\n", retcode);
	598
	599	return retcode;
4a488a7a	600	}
19f6d2a6 OG	601
	602	static int kfd_mmap(struct file filp, struct vm_area_struct vma)
	603	{
	604	struct kfd_process *process;
	605
	606	process = kfd_get_process(current);
	607	if (IS_ERR(process))
	608	return PTR_ERR(process);
	609
	610	return kfd_doorbell_mmap(process, vma);
	611	}