[mirror_ubuntu-artful-kernel.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;
	}

	if (args->eop_buffer_address &&
		!access_ok(VERIFY_WRITE,
			(const void __user *) args->eop_buffer_address,
			sizeof(uint32_t))) {
		pr_debug("kfd: can't access eop buffer");
		return -EFAULT;
	}

	if (args->ctx_save_restore_address &&
		!access_ok(VERIFY_WRITE,
			(const void __user *) args->ctx_save_restore_address,
			sizeof(uint32_t))) {
		pr_debug("kfd: can't access ctx save restore buffer");
		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;
	q_properties->eop_ring_buffer_address = args->eop_buffer_address;
	q_properties->eop_ring_buffer_size = args->eop_buffer_size;
	q_properties->ctx_save_restore_area_address =
			args->ctx_save_restore_address;
	q_properties->ctx_save_restore_area_size = args->ctx_save_restore_size;
	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 if (args->queue_type == KFD_IOC_QUEUE_TYPE_SDMA)
		q_properties->type = KFD_QUEUE_TYPE_SDMA;
	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);

	pr_debug("Queue EOP (0x%llX)\n", q_properties->eop_ring_buffer_address);

	pr_debug("Queue CTX save arex (0x%llX)\n",
			q_properties->ctx_save_restore_area_address);

	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;

	pr_debug("kfd: looking for gpu id 0x%x\n", args->gpu_id);
	dev = kfd_device_by_id(args->gpu_id);
	if (dev == NULL) {
		pr_debug("kfd: gpu id 0x%x was not found\n", args->gpu_id);
		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, q_properties.type, &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->ops.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
0b3674ae OG	181	if (args->eop_buffer_address &&
	182	!access_ok(VERIFY_WRITE,
	183	(const void __user *) args->eop_buffer_address,
	184	sizeof(uint32_t))) {
ff3d04a1 BG	185	pr_debug("kfd: can't access eop buffer");
	186	return -EFAULT;
	187	}
	188
0b3674ae OG	189	if (args->ctx_save_restore_address &&
	190	!access_ok(VERIFY_WRITE,
	191	(const void __user *) args->ctx_save_restore_address,
	192	sizeof(uint32_t))) {
ff3d04a1 BG	193	pr_debug("kfd: can't access ctx save restore buffer");
	194	return -EFAULT;
	195	}
	196
39b027d9 OG	197	q_properties->is_interop = false;
	198	q_properties->queue_percent = args->queue_percentage;
	199	q_properties->priority = args->queue_priority;
	200	q_properties->queue_address = args->ring_base_address;
	201	q_properties->queue_size = args->ring_size;
	202	q_properties->read_ptr = (uint32_t *) args->read_pointer_address;
	203	q_properties->write_ptr = (uint32_t *) args->write_pointer_address;
ff3d04a1 BG	204	q_properties->eop_ring_buffer_address = args->eop_buffer_address;
	205	q_properties->eop_ring_buffer_size = args->eop_buffer_size;
	206	q_properties->ctx_save_restore_area_address =
	207	args->ctx_save_restore_address;
	208	q_properties->ctx_save_restore_area_size = args->ctx_save_restore_size;
39b027d9 OG	209	if (args->queue_type == KFD_IOC_QUEUE_TYPE_COMPUTE \|\|
	210	args->queue_type == KFD_IOC_QUEUE_TYPE_COMPUTE_AQL)
	211	q_properties->type = KFD_QUEUE_TYPE_COMPUTE;
3385f9dd BG	212	else if (args->queue_type == KFD_IOC_QUEUE_TYPE_SDMA)
3385f9dd BG	213	q_properties->type = KFD_QUEUE_TYPE_SDMA;
39b027d9 OG	214	else
	215	return -ENOTSUPP;
	216
	217	if (args->queue_type == KFD_IOC_QUEUE_TYPE_COMPUTE_AQL)
	218	q_properties->format = KFD_QUEUE_FORMAT_AQL;
	219	else
	220	q_properties->format = KFD_QUEUE_FORMAT_PM4;
	221
	222	pr_debug("Queue Percentage (%d, %d)\n",
	223	q_properties->queue_percent, args->queue_percentage);
	224
	225	pr_debug("Queue Priority (%d, %d)\n",
	226	q_properties->priority, args->queue_priority);
	227
	228	pr_debug("Queue Address (0x%llX, 0x%llX)\n",
	229	q_properties->queue_address, args->ring_base_address);
	230
	231	pr_debug("Queue Size (0x%llX, %u)\n",
	232	q_properties->queue_size, args->ring_size);
	233
	234	pr_debug("Queue r/w Pointers (0x%llX, 0x%llX)\n",
	235	(uint64_t) q_properties->read_ptr,
	236	(uint64_t) q_properties->write_ptr);
	237
	238	pr_debug("Queue Format (%d)\n", q_properties->format);
	239
ff3d04a1 BG	240	pr_debug("Queue EOP (0x%llX)\n", q_properties->eop_ring_buffer_address);
	241
	242	pr_debug("Queue CTX save arex (0x%llX)\n",
	243	q_properties->ctx_save_restore_area_address);
	244
39b027d9 OG	245	return 0;
	246	}
	247
524a6404 OG	248	static int kfd_ioctl_create_queue(struct file filep, struct kfd_process p,
524a6404 OG	249	void *data)
4a488a7a	250	{
524a6404	251	struct kfd_ioctl_create_queue_args *args = data;
39b027d9 OG	252	struct kfd_dev *dev;
	253	int err = 0;
	254	unsigned int queue_id;
	255	struct kfd_process_device *pdd;
	256	struct queue_properties q_properties;
	257
	258	memset(&q_properties, 0, sizeof(struct queue_properties));
	259
39b027d9 OG	260	pr_debug("kfd: creating queue ioctl\n");
39b027d9 OG	261
524a6404	262	err = set_queue_properties_from_user(&q_properties, args);
39b027d9 OG	263	if (err)
	264	return err;
	265
281d1bbd	266	pr_debug("kfd: looking for gpu id 0x%x\n", args->gpu_id);
524a6404	267	dev = kfd_device_by_id(args->gpu_id);
ff3d04a1	268	if (dev == NULL) {
281d1bbd	269	pr_debug("kfd: gpu id 0x%x was not found\n", args->gpu_id);
39b027d9	270	return -EINVAL;
ff3d04a1	271	}
39b027d9 OG	272
	273	mutex_lock(&p->mutex);
	274
	275	pdd = kfd_bind_process_to_device(dev, p);
66333cb3	276	if (IS_ERR(pdd)) {
524a6404	277	err = -ESRCH;
39b027d9 OG	278	goto err_bind_process;
	279	}
	280
	281	pr_debug("kfd: creating queue for PASID %d on GPU 0x%x\n",
	282	p->pasid,
	283	dev->id);
	284
85dfaef3 BG	285	err = pqm_create_queue(&p->pqm, dev, filep, &q_properties,
85dfaef3 BG	286	0, q_properties.type, &queue_id);
39b027d9 OG	287	if (err != 0)
	288	goto err_create_queue;
	289
524a6404	290	args->queue_id = queue_id;
39b027d9 OG	291
39b027d9 OG	292	/* Return gpu_id as doorbell offset for mmap usage */
524a6404	293	args->doorbell_offset = args->gpu_id << PAGE_SHIFT;
39b027d9 OG	294
	295	mutex_unlock(&p->mutex);
	296
524a6404	297	pr_debug("kfd: queue id %d was created successfully\n", args->queue_id);
39b027d9 OG	298
39b027d9 OG	299	pr_debug("ring buffer address == 0x%016llX\n",
524a6404	300	args->ring_base_address);
39b027d9 OG	301
39b027d9 OG	302	pr_debug("read ptr address == 0x%016llX\n",
524a6404	303	args->read_pointer_address);
39b027d9 OG	304
39b027d9 OG	305	pr_debug("write ptr address == 0x%016llX\n",
524a6404	306	args->write_pointer_address);
39b027d9 OG	307
	308	return 0;
	309
39b027d9 OG	310	err_create_queue:
	311	err_bind_process:
	312	mutex_unlock(&p->mutex);
	313	return err;
4a488a7a OG	314	}
	315
	316	static int kfd_ioctl_destroy_queue(struct file filp, struct kfd_process p,
524a6404	317	void *data)
4a488a7a	318	{
39b027d9	319	int retval;
524a6404	320	struct kfd_ioctl_destroy_queue_args *args = data;
39b027d9 OG	321
39b027d9 OG	322	pr_debug("kfd: destroying queue id %d for PASID %d\n",
524a6404	323	args->queue_id,
39b027d9 OG	324	p->pasid);
	325
	326	mutex_lock(&p->mutex);
	327
524a6404	328	retval = pqm_destroy_queue(&p->pqm, args->queue_id);
39b027d9 OG	329
	330	mutex_unlock(&p->mutex);
	331	return retval;
4a488a7a OG	332	}
	333
	334	static int kfd_ioctl_update_queue(struct file filp, struct kfd_process p,
524a6404	335	void *data)
4a488a7a	336	{
39b027d9	337	int retval;
524a6404	338	struct kfd_ioctl_update_queue_args *args = data;
39b027d9 OG	339	struct queue_properties properties;
39b027d9 OG	340
524a6404	341	if (args->queue_percentage > KFD_MAX_QUEUE_PERCENTAGE) {
39b027d9 OG	342	pr_err("kfd: queue percentage must be between 0 to KFD_MAX_QUEUE_PERCENTAGE\n");
	343	return -EINVAL;
	344	}
	345
524a6404	346	if (args->queue_priority > KFD_MAX_QUEUE_PRIORITY) {
39b027d9 OG	347	pr_err("kfd: queue priority must be between 0 to KFD_MAX_QUEUE_PRIORITY\n");
	348	return -EINVAL;
	349	}
	350
524a6404	351	if ((args->ring_base_address) &&
4307d8f6	352	(!access_ok(VERIFY_WRITE,
524a6404	353	(const void __user *) args->ring_base_address,
4307d8f6	354	sizeof(uint64_t)))) {
39b027d9 OG	355	pr_err("kfd: can't access ring base address\n");
	356	return -EFAULT;
	357	}
	358
524a6404	359	if (!is_power_of_2(args->ring_size) && (args->ring_size != 0)) {
39b027d9 OG	360	pr_err("kfd: ring size must be a power of 2 or 0\n");
	361	return -EINVAL;
	362	}
	363
524a6404 OG	364	properties.queue_address = args->ring_base_address;
	365	properties.queue_size = args->ring_size;
	366	properties.queue_percent = args->queue_percentage;
	367	properties.priority = args->queue_priority;
39b027d9 OG	368
39b027d9 OG	369	pr_debug("kfd: updating queue id %d for PASID %d\n",
524a6404	370	args->queue_id, p->pasid);
39b027d9 OG	371
	372	mutex_lock(&p->mutex);
	373
524a6404	374	retval = pqm_update_queue(&p->pqm, args->queue_id, &properties);
39b027d9 OG	375
	376	mutex_unlock(&p->mutex);
	377
	378	return retval;
4a488a7a OG	379	}
4a488a7a OG	380
524a6404 OG	381	static int kfd_ioctl_set_memory_policy(struct file *filep,
524a6404 OG	382	struct kfd_process p, void data)
4a488a7a	383	{
524a6404	384	struct kfd_ioctl_set_memory_policy_args *args = data;
41a286fa AL	385	struct kfd_dev *dev;
	386	int err = 0;
	387	struct kfd_process_device *pdd;
	388	enum cache_policy default_policy, alternate_policy;
	389
524a6404 OG	390	if (args->default_policy != KFD_IOC_CACHE_POLICY_COHERENT
524a6404 OG	391	&& args->default_policy != KFD_IOC_CACHE_POLICY_NONCOHERENT) {
41a286fa AL	392	return -EINVAL;
	393	}
	394
524a6404 OG	395	if (args->alternate_policy != KFD_IOC_CACHE_POLICY_COHERENT
524a6404 OG	396	&& args->alternate_policy != KFD_IOC_CACHE_POLICY_NONCOHERENT) {
41a286fa AL	397	return -EINVAL;
	398	}
	399
524a6404	400	dev = kfd_device_by_id(args->gpu_id);
41a286fa AL	401	if (dev == NULL)
	402	return -EINVAL;
	403
	404	mutex_lock(&p->mutex);
	405
	406	pdd = kfd_bind_process_to_device(dev, p);
66333cb3	407	if (IS_ERR(pdd)) {
524a6404	408	err = -ESRCH;
41a286fa AL	409	goto out;
	410	}
	411
524a6404	412	default_policy = (args->default_policy == KFD_IOC_CACHE_POLICY_COHERENT)
41a286fa AL	413	? cache_policy_coherent : cache_policy_noncoherent;
	414
	415	alternate_policy =
524a6404	416	(args->alternate_policy == KFD_IOC_CACHE_POLICY_COHERENT)
41a286fa AL	417	? cache_policy_coherent : cache_policy_noncoherent;
41a286fa AL	418
45c9a5e4	419	if (!dev->dqm->ops.set_cache_memory_policy(dev->dqm,
41a286fa AL	420	&pdd->qpd,
	421	default_policy,
	422	alternate_policy,
524a6404 OG	423	(void __user *)args->alternate_aperture_base,
524a6404 OG	424	args->alternate_aperture_size))
41a286fa AL	425	err = -EINVAL;
	426
	427	out:
	428	mutex_unlock(&p->mutex);
	429
	430	return err;
4a488a7a OG	431	}
4a488a7a OG	432
524a6404 OG	433	static int kfd_ioctl_get_clock_counters(struct file *filep,
524a6404 OG	434	struct kfd_process p, void data)
4a488a7a	435	{
524a6404	436	struct kfd_ioctl_get_clock_counters_args *args = data;
4fac47c8 EP	437	struct kfd_dev *dev;
	438	struct timespec time;
	439
524a6404	440	dev = kfd_device_by_id(args->gpu_id);
4fac47c8 EP	441	if (dev == NULL)
	442	return -EINVAL;
	443
	444	/* Reading GPU clock counter from KGD */
524a6404	445	args->gpu_clock_counter = kfd2kgd->get_gpu_clock_counter(dev->kgd);
4fac47c8 EP	446
	447	/* No access to rdtsc. Using raw monotonic time */
	448	getrawmonotonic(&time);
524a6404	449	args->cpu_clock_counter = (uint64_t)timespec_to_ns(&time);
4fac47c8 EP	450
4fac47c8 EP	451	get_monotonic_boottime(&time);
524a6404	452	args->system_clock_counter = (uint64_t)timespec_to_ns(&time);
4fac47c8 EP	453
4fac47c8 EP	454	/* Since the counter is in nano-seconds we use 1GHz frequency */
524a6404	455	args->system_clock_freq = 1000000000;
4fac47c8 EP	456
4fac47c8 EP	457	return 0;
4a488a7a OG	458	}
	459
	460
	461	static int kfd_ioctl_get_process_apertures(struct file *filp,
524a6404	462	struct kfd_process p, void data)
4a488a7a	463	{
524a6404	464	struct kfd_ioctl_get_process_apertures_args *args = data;
775921ed AS	465	struct kfd_process_device_apertures *pAperture;
	466	struct kfd_process_device *pdd;
	467
	468	dev_dbg(kfd_device, "get apertures for PASID %d", p->pasid);
	469
524a6404	470	args->num_of_nodes = 0;
775921ed AS	471
	472	mutex_lock(&p->mutex);
	473
	474	/if the process-device list isn't empty/
	475	if (kfd_has_process_device_data(p)) {
	476	/* Run over all pdd of the process */
	477	pdd = kfd_get_first_process_device_data(p);
	478	do {
524a6404 OG	479	pAperture =
524a6404 OG	480	&args->process_apertures[args->num_of_nodes];
775921ed AS	481	pAperture->gpu_id = pdd->dev->id;
	482	pAperture->lds_base = pdd->lds_base;
	483	pAperture->lds_limit = pdd->lds_limit;
	484	pAperture->gpuvm_base = pdd->gpuvm_base;
	485	pAperture->gpuvm_limit = pdd->gpuvm_limit;
	486	pAperture->scratch_base = pdd->scratch_base;
	487	pAperture->scratch_limit = pdd->scratch_limit;
	488
	489	dev_dbg(kfd_device,
524a6404	490	"node id %u\n", args->num_of_nodes);
775921ed AS	491	dev_dbg(kfd_device,
	492	"gpu id %u\n", pdd->dev->id);
	493	dev_dbg(kfd_device,
	494	"lds_base %llX\n", pdd->lds_base);
	495	dev_dbg(kfd_device,
	496	"lds_limit %llX\n", pdd->lds_limit);
	497	dev_dbg(kfd_device,
	498	"gpuvm_base %llX\n", pdd->gpuvm_base);
	499	dev_dbg(kfd_device,
	500	"gpuvm_limit %llX\n", pdd->gpuvm_limit);
	501	dev_dbg(kfd_device,
	502	"scratch_base %llX\n", pdd->scratch_base);
	503	dev_dbg(kfd_device,
	504	"scratch_limit %llX\n", pdd->scratch_limit);
	505
524a6404	506	args->num_of_nodes++;
775921ed	507	} while ((pdd = kfd_get_next_process_device_data(p, pdd)) != NULL &&
524a6404	508	(args->num_of_nodes < NUM_OF_SUPPORTED_GPUS));
775921ed AS	509	}
	510
	511	mutex_unlock(&p->mutex);
	512
775921ed	513	return 0;
4a488a7a OG	514	}
4a488a7a OG	515
76baee6c OG	516	#define AMDKFD_IOCTL_DEF(ioctl, _func, _flags) \
	517	[_IOC_NR(ioctl)] = {.cmd = ioctl, .func = _func, .flags = _flags, .cmd_drv = 0, .name = #ioctl}
	518
	519	/** Ioctl table */
	520	static const struct amdkfd_ioctl_desc amdkfd_ioctls[] = {
	521	AMDKFD_IOCTL_DEF(AMDKFD_IOC_GET_VERSION,
	522	kfd_ioctl_get_version, 0),
	523
	524	AMDKFD_IOCTL_DEF(AMDKFD_IOC_CREATE_QUEUE,
	525	kfd_ioctl_create_queue, 0),
	526
	527	AMDKFD_IOCTL_DEF(AMDKFD_IOC_DESTROY_QUEUE,
	528	kfd_ioctl_destroy_queue, 0),
	529
	530	AMDKFD_IOCTL_DEF(AMDKFD_IOC_SET_MEMORY_POLICY,
	531	kfd_ioctl_set_memory_policy, 0),
	532
	533	AMDKFD_IOCTL_DEF(AMDKFD_IOC_GET_CLOCK_COUNTERS,
	534	kfd_ioctl_get_clock_counters, 0),
	535
	536	AMDKFD_IOCTL_DEF(AMDKFD_IOC_GET_PROCESS_APERTURES,
	537	kfd_ioctl_get_process_apertures, 0),
	538
	539	AMDKFD_IOCTL_DEF(AMDKFD_IOC_UPDATE_QUEUE,
	540	kfd_ioctl_update_queue, 0),
	541	};
	542
	543	#define AMDKFD_CORE_IOCTL_COUNT ARRAY_SIZE(amdkfd_ioctls)
	544
4a488a7a OG	545	static long kfd_ioctl(struct file *filep, unsigned int cmd, unsigned long arg)
	546	{
	547	struct kfd_process *process;
76baee6c OG	548	amdkfd_ioctl_t *func;
	549	const struct amdkfd_ioctl_desc *ioctl = NULL;
	550	unsigned int nr = _IOC_NR(cmd);
524a6404 OG	551	char stack_kdata[128];
	552	char *kdata = NULL;
	553	unsigned int usize, asize;
	554	int retcode = -EINVAL;
4a488a7a	555
76baee6c OG	556	if (nr >= AMDKFD_CORE_IOCTL_COUNT)
	557	goto err_i1;
	558
	559	if ((nr >= AMDKFD_COMMAND_START) && (nr < AMDKFD_COMMAND_END)) {
	560	u32 amdkfd_size;
	561
	562	ioctl = &amdkfd_ioctls[nr];
	563
	564	amdkfd_size = _IOC_SIZE(ioctl->cmd);
	565	usize = asize = _IOC_SIZE(cmd);
	566	if (amdkfd_size > asize)
	567	asize = amdkfd_size;
	568
	569	cmd = ioctl->cmd;
	570	} else
	571	goto err_i1;
	572
	573	dev_dbg(kfd_device, "ioctl cmd 0x%x (#%d), arg 0x%lx\n", cmd, nr, arg);
4a488a7a	574
19f6d2a6	575	process = kfd_get_process(current);
76baee6c OG	576	if (IS_ERR(process)) {
	577	dev_dbg(kfd_device, "no process\n");
	578	goto err_i1;
	579	}
4a488a7a	580
76baee6c OG	581	/* Do not trust userspace, use our own definition */
	582	func = ioctl->func;
	583
	584	if (unlikely(!func)) {
	585	dev_dbg(kfd_device, "no function\n");
	586	retcode = -EINVAL;
	587	goto err_i1;
4a488a7a OG	588	}
4a488a7a OG	589
524a6404 OG	590	if (cmd & (IOC_IN \| IOC_OUT)) {
	591	if (asize <= sizeof(stack_kdata)) {
	592	kdata = stack_kdata;
	593	} else {
	594	kdata = kmalloc(asize, GFP_KERNEL);
	595	if (!kdata) {
	596	retcode = -ENOMEM;
	597	goto err_i1;
	598	}
	599	}
	600	if (asize > usize)
	601	memset(kdata + usize, 0, asize - usize);
	602	}
4a488a7a	603
524a6404 OG	604	if (cmd & IOC_IN) {
	605	if (copy_from_user(kdata, (void __user *)arg, usize) != 0) {
	606	retcode = -EFAULT;
	607	goto err_i1;
	608	}
	609	} else if (cmd & IOC_OUT) {
	610	memset(kdata, 0, usize);
	611	}
	612
76baee6c	613	retcode = func(filep, process, kdata);
4a488a7a	614
524a6404 OG	615	if (cmd & IOC_OUT)
	616	if (copy_to_user((void __user *)arg, kdata, usize) != 0)
	617	retcode = -EFAULT;
4a488a7a	618
524a6404	619	err_i1:
76baee6c OG	620	if (!ioctl)
	621	dev_dbg(kfd_device, "invalid ioctl: pid=%d, cmd=0x%02x, nr=0x%02x\n",
	622	task_pid_nr(current), cmd, nr);
	623
524a6404 OG	624	if (kdata != stack_kdata)
	625	kfree(kdata);
	626
	627	if (retcode)
	628	dev_dbg(kfd_device, "ret = %d\n", retcode);
	629
	630	return retcode;
4a488a7a	631	}
19f6d2a6 OG	632
	633	static int kfd_mmap(struct file filp, struct vm_area_struct vma)
	634	{
	635	struct kfd_process *process;
	636
	637	process = kfd_get_process(current);
	638	if (IS_ERR(process))
	639	return PTR_ERR(process);
	640
	641	return kfd_doorbell_mmap(process, vma);
	642	}