[linux-2.6-block.git] / drivers / gpu / drm / amd / amdkfd / kfd_kernel_queue.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/types.h>
#include <linux/mutex.h>
#include <linux/slab.h>
#include <linux/printk.h>
#include <linux/sched.h>
#include "kfd_kernel_queue.h"
#include "kfd_priv.h"
#include "kfd_device_queue_manager.h"
#include "kfd_pm4_headers.h"
#include "kfd_pm4_opcodes.h"

#define PM4_COUNT_ZERO (((1 << 15) - 1) << 16)

static bool initialize(struct kernel_queue *kq, struct kfd_dev *dev,
		enum kfd_queue_type type, unsigned int queue_size)
{
	struct queue_properties prop;
	int retval;
	union PM4_MES_TYPE_3_HEADER nop;

	if (WARN_ON(type != KFD_QUEUE_TYPE_DIQ && type != KFD_QUEUE_TYPE_HIQ))
		return false;

	pr_debug("Initializing queue type %d size %d\n", KFD_QUEUE_TYPE_HIQ,
			queue_size);

	memset(&prop, 0, sizeof(prop));
	memset(&nop, 0, sizeof(nop));

	nop.opcode = IT_NOP;
	nop.type = PM4_TYPE_3;
	nop.u32all |= PM4_COUNT_ZERO;

	kq->dev = dev;
	kq->nop_packet = nop.u32all;
	switch (type) {
	case KFD_QUEUE_TYPE_DIQ:
	case KFD_QUEUE_TYPE_HIQ:
		kq->mqd = dev->dqm->ops.get_mqd_manager(dev->dqm,
						KFD_MQD_TYPE_HIQ);
		break;
	default:
		pr_err("Invalid queue type %d\n", type);
		return false;
	}

	if (!kq->mqd)
		return false;

	prop.doorbell_ptr = kfd_get_kernel_doorbell(dev, &prop.doorbell_off);

	if (!prop.doorbell_ptr) {
		pr_err("Failed to initialize doorbell");
		goto err_get_kernel_doorbell;
	}

	retval = kfd_gtt_sa_allocate(dev, queue_size, &kq->pq);
	if (retval != 0) {
		pr_err("Failed to init pq queues size %d\n", queue_size);
		goto err_pq_allocate_vidmem;
	}

	kq->pq_kernel_addr = kq->pq->cpu_ptr;
	kq->pq_gpu_addr = kq->pq->gpu_addr;

	retval = kq->ops_asic_specific.initialize(kq, dev, type, queue_size);
	if (!retval)
		goto err_eop_allocate_vidmem;

	retval = kfd_gtt_sa_allocate(dev, sizeof(*kq->rptr_kernel),
					&kq->rptr_mem);

	if (retval != 0)
		goto err_rptr_allocate_vidmem;

	kq->rptr_kernel = kq->rptr_mem->cpu_ptr;
	kq->rptr_gpu_addr = kq->rptr_mem->gpu_addr;

	retval = kfd_gtt_sa_allocate(dev, sizeof(*kq->wptr_kernel),
					&kq->wptr_mem);

	if (retval != 0)
		goto err_wptr_allocate_vidmem;

	kq->wptr_kernel = kq->wptr_mem->cpu_ptr;
	kq->wptr_gpu_addr = kq->wptr_mem->gpu_addr;

	memset(kq->pq_kernel_addr, 0, queue_size);
	memset(kq->rptr_kernel, 0, sizeof(*kq->rptr_kernel));
	memset(kq->wptr_kernel, 0, sizeof(*kq->wptr_kernel));

	prop.queue_size = queue_size;
	prop.is_interop = false;
	prop.priority = 1;
	prop.queue_percent = 100;
	prop.type = type;
	prop.vmid = 0;
	prop.queue_address = kq->pq_gpu_addr;
	prop.read_ptr = (uint32_t *) kq->rptr_gpu_addr;
	prop.write_ptr = (uint32_t *) kq->wptr_gpu_addr;
	prop.eop_ring_buffer_address = kq->eop_gpu_addr;
	prop.eop_ring_buffer_size = PAGE_SIZE;

	if (init_queue(&kq->queue, &prop) != 0)
		goto err_init_queue;

	kq->queue->device = dev;
	kq->queue->process = kfd_get_process(current);

	retval = kq->mqd->init_mqd(kq->mqd, &kq->queue->mqd,
					&kq->queue->mqd_mem_obj,
					&kq->queue->gart_mqd_addr,
					&kq->queue->properties);
	if (retval != 0)
		goto err_init_mqd;

	/* assign HIQ to HQD */
	if (type == KFD_QUEUE_TYPE_HIQ) {
		pr_debug("Assigning hiq to hqd\n");
		kq->queue->pipe = KFD_CIK_HIQ_PIPE;
		kq->queue->queue = KFD_CIK_HIQ_QUEUE;
		kq->mqd->load_mqd(kq->mqd, kq->queue->mqd, kq->queue->pipe,
				  kq->queue->queue, &kq->queue->properties,
				  NULL);
	} else {
		/* allocate fence for DIQ */

		retval = kfd_gtt_sa_allocate(dev, sizeof(uint32_t),
						&kq->fence_mem_obj);

		if (retval != 0)
			goto err_alloc_fence;

		kq->fence_kernel_address = kq->fence_mem_obj->cpu_ptr;
		kq->fence_gpu_addr = kq->fence_mem_obj->gpu_addr;
	}

	print_queue(kq->queue);

	return true;
err_alloc_fence:
err_init_mqd:
	uninit_queue(kq->queue);
err_init_queue:
	kfd_gtt_sa_free(dev, kq->wptr_mem);
err_wptr_allocate_vidmem:
	kfd_gtt_sa_free(dev, kq->rptr_mem);
err_rptr_allocate_vidmem:
	kfd_gtt_sa_free(dev, kq->eop_mem);
err_eop_allocate_vidmem:
	kfd_gtt_sa_free(dev, kq->pq);
err_pq_allocate_vidmem:
	kfd_release_kernel_doorbell(dev, prop.doorbell_ptr);
err_get_kernel_doorbell:
	return false;

}

static void uninitialize(struct kernel_queue *kq)
{
	if (kq->queue->properties.type == KFD_QUEUE_TYPE_HIQ)
		kq->mqd->destroy_mqd(kq->mqd,
					kq->queue->mqd,
					KFD_PREEMPT_TYPE_WAVEFRONT_RESET,
					QUEUE_PREEMPT_DEFAULT_TIMEOUT_MS,
					kq->queue->pipe,
					kq->queue->queue);
	else if (kq->queue->properties.type == KFD_QUEUE_TYPE_DIQ)
		kfd_gtt_sa_free(kq->dev, kq->fence_mem_obj);

	kq->mqd->uninit_mqd(kq->mqd, kq->queue->mqd, kq->queue->mqd_mem_obj);

	kfd_gtt_sa_free(kq->dev, kq->rptr_mem);
	kfd_gtt_sa_free(kq->dev, kq->wptr_mem);
	kq->ops_asic_specific.uninitialize(kq);
	kfd_gtt_sa_free(kq->dev, kq->pq);
	kfd_release_kernel_doorbell(kq->dev,
					kq->queue->properties.doorbell_ptr);
	uninit_queue(kq->queue);
}

static int acquire_packet_buffer(struct kernel_queue *kq,
		size_t packet_size_in_dwords, unsigned int **buffer_ptr)
{
	size_t available_size;
	size_t queue_size_dwords;
	uint32_t wptr, rptr;
	unsigned int *queue_address;

	/* When rptr == wptr, the buffer is empty.
	 * When rptr == wptr + 1, the buffer is full.
	 * It is always rptr that advances to the position of wptr, rather than
	 * the opposite. So we can only use up to queue_size_dwords - 1 dwords.
	 */
	rptr = *kq->rptr_kernel;
	wptr = *kq->wptr_kernel;
	queue_address = (unsigned int *)kq->pq_kernel_addr;
	queue_size_dwords = kq->queue->properties.queue_size / sizeof(uint32_t);

	pr_debug("rptr: %d\n", rptr);
	pr_debug("wptr: %d\n", wptr);
	pr_debug("queue_address 0x%p\n", queue_address);

	available_size = (rptr + queue_size_dwords - 1 - wptr) %
							queue_size_dwords;

	if (packet_size_in_dwords > available_size) {
		/*
		 * make sure calling functions know
		 * acquire_packet_buffer() failed
		 */
		*buffer_ptr = NULL;
		return -ENOMEM;
	}

	if (wptr + packet_size_in_dwords >= queue_size_dwords) {
		/* make sure after rolling back to position 0, there is
		 * still enough space.
		 */
		if (packet_size_in_dwords >= rptr) {
			*buffer_ptr = NULL;
			return -ENOMEM;
		}
		/* fill nops, roll back and start at position 0 */
		while (wptr > 0) {
			queue_address[wptr] = kq->nop_packet;
			wptr = (wptr + 1) % queue_size_dwords;
		}
	}

	*buffer_ptr = &queue_address[wptr];
	kq->pending_wptr = wptr + packet_size_in_dwords;

	return 0;
}

static void submit_packet(struct kernel_queue *kq)
{
#ifdef DEBUG
	int i;

	for (i = *kq->wptr_kernel; i < kq->pending_wptr; i++) {
		pr_debug("0x%2X ", kq->pq_kernel_addr[i]);
		if (i % 15 == 0)
			pr_debug("\n");
	}
	pr_debug("\n");
#endif

	*kq->wptr_kernel = kq->pending_wptr;
	write_kernel_doorbell(kq->queue->properties.doorbell_ptr,
				kq->pending_wptr);
}

static void rollback_packet(struct kernel_queue *kq)
{
	kq->pending_wptr = *kq->queue->properties.write_ptr;
}

struct kernel_queue *kernel_queue_init(struct kfd_dev *dev,
					enum kfd_queue_type type)
{
	struct kernel_queue *kq;

	kq = kzalloc(sizeof(*kq), GFP_KERNEL);
	if (!kq)
		return NULL;

	kq->ops.initialize = initialize;
	kq->ops.uninitialize = uninitialize;
	kq->ops.acquire_packet_buffer = acquire_packet_buffer;
	kq->ops.submit_packet = submit_packet;
	kq->ops.rollback_packet = rollback_packet;

	switch (dev->device_info->asic_family) {
	case CHIP_CARRIZO:
		kernel_queue_init_vi(&kq->ops_asic_specific);
		break;

	case CHIP_KAVERI:
		kernel_queue_init_cik(&kq->ops_asic_specific);
		break;
	}

	if (!kq->ops.initialize(kq, dev, type, KFD_KERNEL_QUEUE_SIZE)) {
		pr_err("Failed to init kernel queue\n");
		kfree(kq);
		return NULL;
	}
	return kq;
}

void kernel_queue_uninit(struct kernel_queue *kq)
{
	kq->ops.uninitialize(kq);
	kfree(kq);
}

/* FIXME: Can this test be removed? */
static __attribute__((unused)) void test_kq(struct kfd_dev *dev)
{
	struct kernel_queue *kq;
	uint32_t *buffer, i;
	int retval;

	pr_err("Starting kernel queue test\n");

	kq = kernel_queue_init(dev, KFD_QUEUE_TYPE_HIQ);
	if (unlikely(!kq)) {
		pr_err("  Failed to initialize HIQ\n");
		pr_err("Kernel queue test failed\n");
		return;
	}

	retval = kq->ops.acquire_packet_buffer(kq, 5, &buffer);
	if (unlikely(retval != 0)) {
		pr_err("  Failed to acquire packet buffer\n");
		pr_err("Kernel queue test failed\n");
		return;
	}
	for (i = 0; i < 5; i++)
		buffer[i] = kq->nop_packet;
	kq->ops.submit_packet(kq);

	pr_err("Ending kernel queue test\n");
}
Commit	Line	Data
ed6e6a34 BG	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
	24	#include <linux/types.h>
	25	#include <linux/mutex.h>
	26	#include <linux/slab.h>
	27	#include <linux/printk.h>
9a5634a7	28	#include <linux/sched.h>
ed6e6a34 BG	29	#include "kfd_kernel_queue.h"
	30	#include "kfd_priv.h"
	31	#include "kfd_device_queue_manager.h"
	32	#include "kfd_pm4_headers.h"
	33	#include "kfd_pm4_opcodes.h"
	34
	35	#define PM4_COUNT_ZERO (((1 << 15) - 1) << 16)
	36
	37	static bool initialize(struct kernel_queue kq, struct kfd_dev dev,
	38	enum kfd_queue_type type, unsigned int queue_size)
	39	{
	40	struct queue_properties prop;
	41	int retval;
	42	union PM4_MES_TYPE_3_HEADER nop;
	43
32fa8219 FK	44	if (WARN_ON(type != KFD_QUEUE_TYPE_DIQ && type != KFD_QUEUE_TYPE_HIQ))
32fa8219 FK	45	return false;
ed6e6a34	46
79775b62 KR	47	pr_debug("Initializing queue type %d size %d\n", KFD_QUEUE_TYPE_HIQ,
79775b62 KR	48	queue_size);
ed6e6a34	49
78b13f79 EC	50	memset(&prop, 0, sizeof(prop));
	51	memset(&nop, 0, sizeof(nop));
	52
ed6e6a34 BG	53	nop.opcode = IT_NOP;
	54	nop.type = PM4_TYPE_3;
	55	nop.u32all \|= PM4_COUNT_ZERO;
	56
	57	kq->dev = dev;
	58	kq->nop_packet = nop.u32all;
	59	switch (type) {
	60	case KFD_QUEUE_TYPE_DIQ:
	61	case KFD_QUEUE_TYPE_HIQ:
45c9a5e4	62	kq->mqd = dev->dqm->ops.get_mqd_manager(dev->dqm,
85d258f9	63	KFD_MQD_TYPE_HIQ);
ed6e6a34 BG	64	break;
ed6e6a34 BG	65	default:
32fa8219 FK	66	pr_err("Invalid queue type %d\n", type);
32fa8219 FK	67	return false;
ed6e6a34 BG	68	}
ed6e6a34 BG	69
4eacc26b	70	if (!kq->mqd)
ed6e6a34 BG	71	return false;
ed6e6a34 BG	72
5cd78de5	73	prop.doorbell_ptr = kfd_get_kernel_doorbell(dev, &prop.doorbell_off);
ed6e6a34	74
4eacc26b	75	if (!prop.doorbell_ptr) {
79775b62	76	pr_err("Failed to initialize doorbell");
ed6e6a34	77	goto err_get_kernel_doorbell;
aaad2d8c	78	}
ed6e6a34	79
a86aa3ca	80	retval = kfd_gtt_sa_allocate(dev, queue_size, &kq->pq);
aaad2d8c	81	if (retval != 0) {
79775b62	82	pr_err("Failed to init pq queues size %d\n", queue_size);
ed6e6a34	83	goto err_pq_allocate_vidmem;
aaad2d8c	84	}
ed6e6a34 BG	85
	86	kq->pq_kernel_addr = kq->pq->cpu_ptr;
	87	kq->pq_gpu_addr = kq->pq->gpu_addr;
	88
6898f0a5	89	retval = kq->ops_asic_specific.initialize(kq, dev, type, queue_size);
4eacc26b	90	if (!retval)
6898f0a5 BG	91	goto err_eop_allocate_vidmem;
6898f0a5 BG	92
a86aa3ca OG	93	retval = kfd_gtt_sa_allocate(dev, sizeof(*kq->rptr_kernel),
a86aa3ca OG	94	&kq->rptr_mem);
ed6e6a34 BG	95
	96	if (retval != 0)
	97	goto err_rptr_allocate_vidmem;
	98
	99	kq->rptr_kernel = kq->rptr_mem->cpu_ptr;
	100	kq->rptr_gpu_addr = kq->rptr_mem->gpu_addr;
	101
a86aa3ca OG	102	retval = kfd_gtt_sa_allocate(dev, sizeof(*kq->wptr_kernel),
a86aa3ca OG	103	&kq->wptr_mem);
ed6e6a34 BG	104
	105	if (retval != 0)
	106	goto err_wptr_allocate_vidmem;
	107
	108	kq->wptr_kernel = kq->wptr_mem->cpu_ptr;
	109	kq->wptr_gpu_addr = kq->wptr_mem->gpu_addr;
	110
	111	memset(kq->pq_kernel_addr, 0, queue_size);
	112	memset(kq->rptr_kernel, 0, sizeof(*kq->rptr_kernel));
	113	memset(kq->wptr_kernel, 0, sizeof(*kq->wptr_kernel));
	114
	115	prop.queue_size = queue_size;
	116	prop.is_interop = false;
	117	prop.priority = 1;
	118	prop.queue_percent = 100;
	119	prop.type = type;
	120	prop.vmid = 0;
	121	prop.queue_address = kq->pq_gpu_addr;
	122	prop.read_ptr = (uint32_t *) kq->rptr_gpu_addr;
	123	prop.write_ptr = (uint32_t *) kq->wptr_gpu_addr;
6898f0a5 BG	124	prop.eop_ring_buffer_address = kq->eop_gpu_addr;
6898f0a5 BG	125	prop.eop_ring_buffer_size = PAGE_SIZE;
ed6e6a34	126
e88a614c	127	if (init_queue(&kq->queue, &prop) != 0)
ed6e6a34 BG	128	goto err_init_queue;
	129
	130	kq->queue->device = dev;
	131	kq->queue->process = kfd_get_process(current);
	132
	133	retval = kq->mqd->init_mqd(kq->mqd, &kq->queue->mqd,
	134	&kq->queue->mqd_mem_obj,
	135	&kq->queue->gart_mqd_addr,
	136	&kq->queue->properties);
	137	if (retval != 0)
	138	goto err_init_mqd;
	139
	140	/* assign HIQ to HQD */
	141	if (type == KFD_QUEUE_TYPE_HIQ) {
79775b62	142	pr_debug("Assigning hiq to hqd\n");
ed6e6a34 BG	143	kq->queue->pipe = KFD_CIK_HIQ_PIPE;
	144	kq->queue->queue = KFD_CIK_HIQ_QUEUE;
	145	kq->mqd->load_mqd(kq->mqd, kq->queue->mqd, kq->queue->pipe,
70539bd7 FK	146	kq->queue->queue, &kq->queue->properties,
70539bd7 FK	147	NULL);
ed6e6a34 BG	148	} else {
	149	/* allocate fence for DIQ */
	150
a86aa3ca OG	151	retval = kfd_gtt_sa_allocate(dev, sizeof(uint32_t),
a86aa3ca OG	152	&kq->fence_mem_obj);
ed6e6a34 BG	153
	154	if (retval != 0)
	155	goto err_alloc_fence;
	156
	157	kq->fence_kernel_address = kq->fence_mem_obj->cpu_ptr;
	158	kq->fence_gpu_addr = kq->fence_mem_obj->gpu_addr;
	159	}
	160
	161	print_queue(kq->queue);
	162
	163	return true;
	164	err_alloc_fence:
	165	err_init_mqd:
	166	uninit_queue(kq->queue);
	167	err_init_queue:
a86aa3ca	168	kfd_gtt_sa_free(dev, kq->wptr_mem);
ed6e6a34	169	err_wptr_allocate_vidmem:
a86aa3ca	170	kfd_gtt_sa_free(dev, kq->rptr_mem);
ed6e6a34	171	err_rptr_allocate_vidmem:
6898f0a5 BG	172	kfd_gtt_sa_free(dev, kq->eop_mem);
6898f0a5 BG	173	err_eop_allocate_vidmem:
a86aa3ca	174	kfd_gtt_sa_free(dev, kq->pq);
ed6e6a34	175	err_pq_allocate_vidmem:
5cd78de5	176	kfd_release_kernel_doorbell(dev, prop.doorbell_ptr);
ed6e6a34	177	err_get_kernel_doorbell:
ed6e6a34 BG	178	return false;
	179
	180	}
	181
	182	static void uninitialize(struct kernel_queue *kq)
	183	{
ed6e6a34 BG	184	if (kq->queue->properties.type == KFD_QUEUE_TYPE_HIQ)
ed6e6a34 BG	185	kq->mqd->destroy_mqd(kq->mqd,
1fabbf78	186	kq->queue->mqd,
b22666fe	187	KFD_PREEMPT_TYPE_WAVEFRONT_RESET,
ed6e6a34 BG	188	QUEUE_PREEMPT_DEFAULT_TIMEOUT_MS,
	189	kq->queue->pipe,
	190	kq->queue->queue);
a86aa3ca OG	191	else if (kq->queue->properties.type == KFD_QUEUE_TYPE_DIQ)
a86aa3ca OG	192	kfd_gtt_sa_free(kq->dev, kq->fence_mem_obj);
ed6e6a34	193
aaad2d8c BG	194	kq->mqd->uninit_mqd(kq->mqd, kq->queue->mqd, kq->queue->mqd_mem_obj);
aaad2d8c BG	195
a86aa3ca OG	196	kfd_gtt_sa_free(kq->dev, kq->rptr_mem);
a86aa3ca OG	197	kfd_gtt_sa_free(kq->dev, kq->wptr_mem);
6898f0a5	198	kq->ops_asic_specific.uninitialize(kq);
a86aa3ca	199	kfd_gtt_sa_free(kq->dev, kq->pq);
ed6e6a34	200	kfd_release_kernel_doorbell(kq->dev,
5cd78de5	201	kq->queue->properties.doorbell_ptr);
ed6e6a34 BG	202	uninit_queue(kq->queue);
	203	}
	204
	205	static int acquire_packet_buffer(struct kernel_queue *kq,
	206	size_t packet_size_in_dwords, unsigned int **buffer_ptr)
	207	{
	208	size_t available_size;
	209	size_t queue_size_dwords;
	210	uint32_t wptr, rptr;
	211	unsigned int *queue_address;
	212
cb1d9967 YZ	213	/* When rptr == wptr, the buffer is empty.
	214	* When rptr == wptr + 1, the buffer is full.
	215	* It is always rptr that advances to the position of wptr, rather than
	216	* the opposite. So we can only use up to queue_size_dwords - 1 dwords.
	217	*/
ed6e6a34 BG	218	rptr = *kq->rptr_kernel;
	219	wptr = *kq->wptr_kernel;
	220	queue_address = (unsigned int *)kq->pq_kernel_addr;
	221	queue_size_dwords = kq->queue->properties.queue_size / sizeof(uint32_t);
	222
8856d8e0 OG	223	pr_debug("rptr: %d\n", rptr);
	224	pr_debug("wptr: %d\n", wptr);
	225	pr_debug("queue_address 0x%p\n", queue_address);
ed6e6a34	226
cb1d9967	227	available_size = (rptr + queue_size_dwords - 1 - wptr) %
ed6e6a34 BG	228	queue_size_dwords;
ed6e6a34 BG	229
cb1d9967	230	if (packet_size_in_dwords > available_size) {
a550bb3d OG	231	/*
	232	* make sure calling functions know
	233	* acquire_packet_buffer() failed
	234	*/
	235	*buffer_ptr = NULL;
ed6e6a34	236	return -ENOMEM;
a550bb3d	237	}
ed6e6a34 BG	238
ed6e6a34 BG	239	if (wptr + packet_size_in_dwords >= queue_size_dwords) {
cb1d9967 YZ	240	/* make sure after rolling back to position 0, there is
	241	* still enough space.
	242	*/
	243	if (packet_size_in_dwords >= rptr) {
	244	*buffer_ptr = NULL;
	245	return -ENOMEM;
	246	}
	247	/* fill nops, roll back and start at position 0 */
ed6e6a34 BG	248	while (wptr > 0) {
	249	queue_address[wptr] = kq->nop_packet;
	250	wptr = (wptr + 1) % queue_size_dwords;
	251	}
	252	}
	253
	254	*buffer_ptr = &queue_address[wptr];
	255	kq->pending_wptr = wptr + packet_size_in_dwords;
	256
	257	return 0;
	258	}
	259
	260	static void submit_packet(struct kernel_queue *kq)
	261	{
	262	#ifdef DEBUG
	263	int i;
ed6e6a34	264
ed6e6a34 BG	265	for (i = *kq->wptr_kernel; i < kq->pending_wptr; i++) {
	266	pr_debug("0x%2X ", kq->pq_kernel_addr[i]);
	267	if (i % 15 == 0)
	268	pr_debug("\n");
	269	}
	270	pr_debug("\n");
	271	#endif
	272
	273	*kq->wptr_kernel = kq->pending_wptr;
5cd78de5	274	write_kernel_doorbell(kq->queue->properties.doorbell_ptr,
ed6e6a34 BG	275	kq->pending_wptr);
	276	}
	277
ed6e6a34 BG	278	static void rollback_packet(struct kernel_queue *kq)
ed6e6a34 BG	279	{
ed6e6a34 BG	280	kq->pending_wptr = *kq->queue->properties.write_ptr;
	281	}
	282
	283	struct kernel_queue kernel_queue_init(struct kfd_dev dev,
	284	enum kfd_queue_type type)
	285	{
	286	struct kernel_queue *kq;
	287
dbf56ab1	288	kq = kzalloc(sizeof(*kq), GFP_KERNEL);
ed6e6a34 BG	289	if (!kq)
	290	return NULL;
	291
443fbd5f OG	292	kq->ops.initialize = initialize;
	293	kq->ops.uninitialize = uninitialize;
	294	kq->ops.acquire_packet_buffer = acquire_packet_buffer;
	295	kq->ops.submit_packet = submit_packet;
443fbd5f OG	296	kq->ops.rollback_packet = rollback_packet;
443fbd5f OG	297
6898f0a5 BG	298	switch (dev->device_info->asic_family) {
	299	case CHIP_CARRIZO:
	300	kernel_queue_init_vi(&kq->ops_asic_specific);
300dec95 OG	301	break;
300dec95 OG	302
6898f0a5 BG	303	case CHIP_KAVERI:
6898f0a5 BG	304	kernel_queue_init_cik(&kq->ops_asic_specific);
300dec95	305	break;
6898f0a5 BG	306	}
6898f0a5 BG	307
991ca8ee	308	if (!kq->ops.initialize(kq, dev, type, KFD_KERNEL_QUEUE_SIZE)) {
79775b62	309	pr_err("Failed to init kernel queue\n");
ed6e6a34 BG	310	kfree(kq);
	311	return NULL;
	312	}
	313	return kq;
	314	}
	315
	316	void kernel_queue_uninit(struct kernel_queue *kq)
	317	{
443fbd5f	318	kq->ops.uninitialize(kq);
ed6e6a34 BG	319	kfree(kq);
	320	}
	321
32fa8219	322	/* FIXME: Can this test be removed? */
5ef360ea	323	static __attribute__((unused)) void test_kq(struct kfd_dev *dev)
ed6e6a34 BG	324	{
	325	struct kernel_queue *kq;
	326	uint32_t *buffer, i;
	327	int retval;
	328
79775b62	329	pr_err("Starting kernel queue test\n");
ed6e6a34 BG	330
ed6e6a34 BG	331	kq = kernel_queue_init(dev, KFD_QUEUE_TYPE_HIQ);
32fa8219 FK	332	if (unlikely(!kq)) {
	333	pr_err(" Failed to initialize HIQ\n");
	334	pr_err("Kernel queue test failed\n");
	335	return;
	336	}
ed6e6a34	337
443fbd5f	338	retval = kq->ops.acquire_packet_buffer(kq, 5, &buffer);
32fa8219 FK	339	if (unlikely(retval != 0)) {
	340	pr_err(" Failed to acquire packet buffer\n");
	341	pr_err("Kernel queue test failed\n");
	342	return;
	343	}
ed6e6a34 BG	344	for (i = 0; i < 5; i++)
ed6e6a34 BG	345	buffer[i] = kq->nop_packet;
443fbd5f	346	kq->ops.submit_packet(kq);
ed6e6a34	347
79775b62	348	pr_err("Ending kernel queue test\n");
ed6e6a34 BG	349	}
	350
	351