[mirror_ubuntu-artful-kernel.git] / drivers / gpu / drm / radeon / radeon_ring.c

/*
 * Copyright 2008 Advanced Micro Devices, Inc.
 * Copyright 2008 Red Hat Inc.
 * Copyright 2009 Jerome Glisse.
 *
 * 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.
 *
 * Authors: Dave Airlie
 *          Alex Deucher
 *          Jerome Glisse
 */
#include <linux/seq_file.h>
#include <linux/slab.h>
#include "drmP.h"
#include "radeon_drm.h"
#include "radeon_reg.h"
#include "radeon.h"
#include "atom.h"

int radeon_debugfs_ib_init(struct radeon_device *rdev);

void radeon_ib_bogus_cleanup(struct radeon_device *rdev)
{
	struct radeon_ib *ib, *n;

	list_for_each_entry_safe(ib, n, &rdev->ib_pool.bogus_ib, list) {
		list_del(&ib->list);
		vfree(ib->ptr);
		kfree(ib);
	}
}

void radeon_ib_bogus_add(struct radeon_device *rdev, struct radeon_ib *ib)
{
	struct radeon_ib *bib;

	bib = kmalloc(sizeof(*bib), GFP_KERNEL);
	if (bib == NULL)
		return;
	bib->ptr = vmalloc(ib->length_dw * 4);
	if (bib->ptr == NULL) {
		kfree(bib);
		return;
	}
	memcpy(bib->ptr, ib->ptr, ib->length_dw * 4);
	bib->length_dw = ib->length_dw;
	mutex_lock(&rdev->ib_pool.mutex);
	list_add_tail(&bib->list, &rdev->ib_pool.bogus_ib);
	mutex_unlock(&rdev->ib_pool.mutex);
}

/*
 * IB.
 */
int radeon_ib_get(struct radeon_device *rdev, struct radeon_ib **ib)
{
	struct radeon_fence *fence;
	struct radeon_ib *nib;
	int r = 0, i, c;

	*ib = NULL;
	r = radeon_fence_create(rdev, &fence);
	if (r) {
		dev_err(rdev->dev, "failed to create fence for new IB\n");
		return r;
	}
	mutex_lock(&rdev->ib_pool.mutex);
	for (i = rdev->ib_pool.head_id, c = 0, nib = NULL; c < RADEON_IB_POOL_SIZE; c++, i++) {
		i &= (RADEON_IB_POOL_SIZE - 1);
		if (rdev->ib_pool.ibs[i].free) {
			nib = &rdev->ib_pool.ibs[i];
			break;
		}
	}
	if (nib == NULL) {
		/* This should never happen, it means we allocated all
		 * IB and haven't scheduled one yet, return EBUSY to
		 * userspace hoping that on ioctl recall we get better
		 * luck
		 */
		dev_err(rdev->dev, "no free indirect buffer !\n");
		mutex_unlock(&rdev->ib_pool.mutex);
		radeon_fence_unref(&fence);
		return -EBUSY;
	}
	rdev->ib_pool.head_id = (nib->idx + 1) & (RADEON_IB_POOL_SIZE - 1);
	nib->free = false;
	if (nib->fence) {
		mutex_unlock(&rdev->ib_pool.mutex);
		r = radeon_fence_wait(nib->fence, false);
		if (r) {
			dev_err(rdev->dev, "error waiting fence of IB(%u:0x%016lX:%u)\n",
				nib->idx, (unsigned long)nib->gpu_addr, nib->length_dw);
			mutex_lock(&rdev->ib_pool.mutex);
			nib->free = true;
			mutex_unlock(&rdev->ib_pool.mutex);
			radeon_fence_unref(&fence);
			return r;
		}
		mutex_lock(&rdev->ib_pool.mutex);
	}
	radeon_fence_unref(&nib->fence);
	nib->fence = fence;
	nib->length_dw = 0;
	mutex_unlock(&rdev->ib_pool.mutex);
	*ib = nib;
	return 0;
}

void radeon_ib_free(struct radeon_device *rdev, struct radeon_ib **ib)
{
	struct radeon_ib *tmp = *ib;

	*ib = NULL;
	if (tmp == NULL) {
		return;
	}
	if (!tmp->fence->emited)
		radeon_fence_unref(&tmp->fence);
	mutex_lock(&rdev->ib_pool.mutex);
	tmp->free = true;
	mutex_unlock(&rdev->ib_pool.mutex);
}

int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib)
{
	int r = 0;

	if (!ib->length_dw || !rdev->cp.ready) {
		/* TODO: Nothings in the ib we should report. */
		DRM_ERROR("radeon: couldn't schedule IB(%u).\n", ib->idx);
		return -EINVAL;
	}

	/* 64 dwords should be enough for fence too */
	r = radeon_ring_lock(rdev, 64);
	if (r) {
		DRM_ERROR("radeon: scheduling IB failled (%d).\n", r);
		return r;
	}
	radeon_ring_ib_execute(rdev, ib);
	radeon_fence_emit(rdev, ib->fence);
	mutex_lock(&rdev->ib_pool.mutex);
	/* once scheduled IB is considered free and protected by the fence */
	ib->free = true;
	mutex_unlock(&rdev->ib_pool.mutex);
	radeon_ring_unlock_commit(rdev);
	return 0;
}

int radeon_ib_pool_init(struct radeon_device *rdev)
{
	void *ptr;
	uint64_t gpu_addr;
	int i;
	int r = 0;

	if (rdev->ib_pool.robj)
		return 0;
	INIT_LIST_HEAD(&rdev->ib_pool.bogus_ib);
	/* Allocate 1M object buffer */
	r = radeon_bo_create(rdev, NULL,  RADEON_IB_POOL_SIZE*64*1024,
				true, RADEON_GEM_DOMAIN_GTT,
				&rdev->ib_pool.robj);
	if (r) {
		DRM_ERROR("radeon: failed to ib pool (%d).\n", r);
		return r;
	}
	r = radeon_bo_reserve(rdev->ib_pool.robj, false);
	if (unlikely(r != 0))
		return r;
	r = radeon_bo_pin(rdev->ib_pool.robj, RADEON_GEM_DOMAIN_GTT, &gpu_addr);
	if (r) {
		radeon_bo_unreserve(rdev->ib_pool.robj);
		DRM_ERROR("radeon: failed to pin ib pool (%d).\n", r);
		return r;
	}
	r = radeon_bo_kmap(rdev->ib_pool.robj, &ptr);
	radeon_bo_unreserve(rdev->ib_pool.robj);
	if (r) {
		DRM_ERROR("radeon: failed to map ib poll (%d).\n", r);
		return r;
	}
	for (i = 0; i < RADEON_IB_POOL_SIZE; i++) {
		unsigned offset;

		offset = i * 64 * 1024;
		rdev->ib_pool.ibs[i].gpu_addr = gpu_addr + offset;
		rdev->ib_pool.ibs[i].ptr = ptr + offset;
		rdev->ib_pool.ibs[i].idx = i;
		rdev->ib_pool.ibs[i].length_dw = 0;
		rdev->ib_pool.ibs[i].free = true;
	}
	rdev->ib_pool.head_id = 0;
	rdev->ib_pool.ready = true;
	DRM_INFO("radeon: ib pool ready.\n");
	if (radeon_debugfs_ib_init(rdev)) {
		DRM_ERROR("Failed to register debugfs file for IB !\n");
	}
	return r;
}

void radeon_ib_pool_fini(struct radeon_device *rdev)
{
	int r;

	if (!rdev->ib_pool.ready) {
		return;
	}
	mutex_lock(&rdev->ib_pool.mutex);
	radeon_ib_bogus_cleanup(rdev);

	if (rdev->ib_pool.robj) {
		r = radeon_bo_reserve(rdev->ib_pool.robj, false);
		if (likely(r == 0)) {
			radeon_bo_kunmap(rdev->ib_pool.robj);
			radeon_bo_unpin(rdev->ib_pool.robj);
			radeon_bo_unreserve(rdev->ib_pool.robj);
		}
		radeon_bo_unref(&rdev->ib_pool.robj);
		rdev->ib_pool.robj = NULL;
	}
	mutex_unlock(&rdev->ib_pool.mutex);
}


/*
 * Ring.
 */
void radeon_ring_free_size(struct radeon_device *rdev)
{
	if (rdev->family >= CHIP_R600)
		rdev->cp.rptr = RREG32(R600_CP_RB_RPTR);
	else
		rdev->cp.rptr = RREG32(RADEON_CP_RB_RPTR);
	/* This works because ring_size is a power of 2 */
	rdev->cp.ring_free_dw = (rdev->cp.rptr + (rdev->cp.ring_size / 4));
	rdev->cp.ring_free_dw -= rdev->cp.wptr;
	rdev->cp.ring_free_dw &= rdev->cp.ptr_mask;
	if (!rdev->cp.ring_free_dw) {
		rdev->cp.ring_free_dw = rdev->cp.ring_size / 4;
	}
}

int radeon_ring_alloc(struct radeon_device *rdev, unsigned ndw)
{
	int r;

	/* Align requested size with padding so unlock_commit can
	 * pad safely */
	ndw = (ndw + rdev->cp.align_mask) & ~rdev->cp.align_mask;
	while (ndw > (rdev->cp.ring_free_dw - 1)) {
		radeon_ring_free_size(rdev);
		if (ndw < rdev->cp.ring_free_dw) {
			break;
		}
		r = radeon_fence_wait_next(rdev);
		if (r)
			return r;
	}
	rdev->cp.count_dw = ndw;
	rdev->cp.wptr_old = rdev->cp.wptr;
	return 0;
}

int radeon_ring_lock(struct radeon_device *rdev, unsigned ndw)
{
	int r;

	mutex_lock(&rdev->cp.mutex);
	r = radeon_ring_alloc(rdev, ndw);
	if (r) {
		mutex_unlock(&rdev->cp.mutex);
		return r;
	}
	return 0;
}

void radeon_ring_commit(struct radeon_device *rdev)
{
	unsigned count_dw_pad;
	unsigned i;

	/* We pad to match fetch size */
	count_dw_pad = (rdev->cp.align_mask + 1) -
		       (rdev->cp.wptr & rdev->cp.align_mask);
	for (i = 0; i < count_dw_pad; i++) {
		radeon_ring_write(rdev, 2 << 30);
	}
	DRM_MEMORYBARRIER();
	radeon_cp_commit(rdev);
}

void radeon_ring_unlock_commit(struct radeon_device *rdev)
{
	radeon_ring_commit(rdev);
	mutex_unlock(&rdev->cp.mutex);
}

void radeon_ring_unlock_undo(struct radeon_device *rdev)
{
	rdev->cp.wptr = rdev->cp.wptr_old;
	mutex_unlock(&rdev->cp.mutex);
}

int radeon_ring_init(struct radeon_device *rdev, unsigned ring_size)
{
	int r;

	rdev->cp.ring_size = ring_size;
	/* Allocate ring buffer */
	if (rdev->cp.ring_obj == NULL) {
		r = radeon_bo_create(rdev, NULL, rdev->cp.ring_size, true,
					RADEON_GEM_DOMAIN_GTT,
					&rdev->cp.ring_obj);
		if (r) {
			dev_err(rdev->dev, "(%d) ring create failed\n", r);
			return r;
		}
		r = radeon_bo_reserve(rdev->cp.ring_obj, false);
		if (unlikely(r != 0))
			return r;
		r = radeon_bo_pin(rdev->cp.ring_obj, RADEON_GEM_DOMAIN_GTT,
					&rdev->cp.gpu_addr);
		if (r) {
			radeon_bo_unreserve(rdev->cp.ring_obj);
			dev_err(rdev->dev, "(%d) ring pin failed\n", r);
			return r;
		}
		r = radeon_bo_kmap(rdev->cp.ring_obj,
				       (void **)&rdev->cp.ring);
		radeon_bo_unreserve(rdev->cp.ring_obj);
		if (r) {
			dev_err(rdev->dev, "(%d) ring map failed\n", r);
			return r;
		}
	}
	rdev->cp.ptr_mask = (rdev->cp.ring_size / 4) - 1;
	rdev->cp.ring_free_dw = rdev->cp.ring_size / 4;
	return 0;
}

void radeon_ring_fini(struct radeon_device *rdev)
{
	int r;

	mutex_lock(&rdev->cp.mutex);
	if (rdev->cp.ring_obj) {
		r = radeon_bo_reserve(rdev->cp.ring_obj, false);
		if (likely(r == 0)) {
			radeon_bo_kunmap(rdev->cp.ring_obj);
			radeon_bo_unpin(rdev->cp.ring_obj);
			radeon_bo_unreserve(rdev->cp.ring_obj);
		}
		radeon_bo_unref(&rdev->cp.ring_obj);
		rdev->cp.ring = NULL;
		rdev->cp.ring_obj = NULL;
	}
	mutex_unlock(&rdev->cp.mutex);
}


/*
 * Debugfs info
 */
#if defined(CONFIG_DEBUG_FS)
static int radeon_debugfs_ib_info(struct seq_file *m, void *data)
{
	struct drm_info_node *node = (struct drm_info_node *) m->private;
	struct radeon_ib *ib = node->info_ent->data;
	unsigned i;

	if (ib == NULL) {
		return 0;
	}
	seq_printf(m, "IB %04u\n", ib->idx);
	seq_printf(m, "IB fence %p\n", ib->fence);
	seq_printf(m, "IB size %05u dwords\n", ib->length_dw);
	for (i = 0; i < ib->length_dw; i++) {
		seq_printf(m, "[%05u]=0x%08X\n", i, ib->ptr[i]);
	}
	return 0;
}

static int radeon_debugfs_ib_bogus_info(struct seq_file *m, void *data)
{
	struct drm_info_node *node = (struct drm_info_node *) m->private;
	struct radeon_device *rdev = node->info_ent->data;
	struct radeon_ib *ib;
	unsigned i;

	mutex_lock(&rdev->ib_pool.mutex);
	if (list_empty(&rdev->ib_pool.bogus_ib)) {
		mutex_unlock(&rdev->ib_pool.mutex);
		seq_printf(m, "no bogus IB recorded\n");
		return 0;
	}
	ib = list_first_entry(&rdev->ib_pool.bogus_ib, struct radeon_ib, list);
	list_del_init(&ib->list);
	mutex_unlock(&rdev->ib_pool.mutex);
	seq_printf(m, "IB size %05u dwords\n", ib->length_dw);
	for (i = 0; i < ib->length_dw; i++) {
		seq_printf(m, "[%05u]=0x%08X\n", i, ib->ptr[i]);
	}
	vfree(ib->ptr);
	kfree(ib);
	return 0;
}

static struct drm_info_list radeon_debugfs_ib_list[RADEON_IB_POOL_SIZE];
static char radeon_debugfs_ib_names[RADEON_IB_POOL_SIZE][32];

static struct drm_info_list radeon_debugfs_ib_bogus_info_list[] = {
	{"radeon_ib_bogus", radeon_debugfs_ib_bogus_info, 0, NULL},
};
#endif

int radeon_debugfs_ib_init(struct radeon_device *rdev)
{
#if defined(CONFIG_DEBUG_FS)
	unsigned i;
	int r;

	radeon_debugfs_ib_bogus_info_list[0].data = rdev;
	r = radeon_debugfs_add_files(rdev, radeon_debugfs_ib_bogus_info_list, 1);
	if (r)
		return r;
	for (i = 0; i < RADEON_IB_POOL_SIZE; i++) {
		sprintf(radeon_debugfs_ib_names[i], "radeon_ib_%04u", i);
		radeon_debugfs_ib_list[i].name = radeon_debugfs_ib_names[i];
		radeon_debugfs_ib_list[i].show = &radeon_debugfs_ib_info;
		radeon_debugfs_ib_list[i].driver_features = 0;
		radeon_debugfs_ib_list[i].data = &rdev->ib_pool.ibs[i];
	}
	return radeon_debugfs_add_files(rdev, radeon_debugfs_ib_list,
					RADEON_IB_POOL_SIZE);
#else
	return 0;
#endif
}
Commit	Line	Data
771fe6b9 JG	1	/*
	2	* Copyright 2008 Advanced Micro Devices, Inc.
	3	* Copyright 2008 Red Hat Inc.
	4	* Copyright 2009 Jerome Glisse.
	5	*
	6	* Permission is hereby granted, free of charge, to any person obtaining a
	7	* copy of this software and associated documentation files (the "Software"),
	8	* to deal in the Software without restriction, including without limitation
	9	* the rights to use, copy, modify, merge, publish, distribute, sublicense,
	10	* and/or sell copies of the Software, and to permit persons to whom the
	11	* Software is furnished to do so, subject to the following conditions:
	12	*
	13	* The above copyright notice and this permission notice shall be included in
	14	* all copies or substantial portions of the Software.
	15	*
	16	* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
	17	* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
	18	* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
	19	* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
	20	* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
	21	* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
	22	* OTHER DEALINGS IN THE SOFTWARE.
	23	*
	24	* Authors: Dave Airlie
	25	* Alex Deucher
	26	* Jerome Glisse
	27	*/
	28	#include <linux/seq_file.h>
5a0e3ad6	29	#include <linux/slab.h>
771fe6b9 JG	30	#include "drmP.h"
	31	#include "radeon_drm.h"
	32	#include "radeon_reg.h"
	33	#include "radeon.h"
	34	#include "atom.h"
	35
	36	int radeon_debugfs_ib_init(struct radeon_device *rdev);
	37
9f93ed39 JG	38	void radeon_ib_bogus_cleanup(struct radeon_device *rdev)
	39	{
	40	struct radeon_ib ib, n;
	41
	42	list_for_each_entry_safe(ib, n, &rdev->ib_pool.bogus_ib, list) {
	43	list_del(&ib->list);
	44	vfree(ib->ptr);
	45	kfree(ib);
	46	}
	47	}
	48
	49	void radeon_ib_bogus_add(struct radeon_device rdev, struct radeon_ib ib)
	50	{
	51	struct radeon_ib *bib;
	52
	53	bib = kmalloc(sizeof(*bib), GFP_KERNEL);
	54	if (bib == NULL)
	55	return;
	56	bib->ptr = vmalloc(ib->length_dw * 4);
	57	if (bib->ptr == NULL) {
	58	kfree(bib);
	59	return;
	60	}
	61	memcpy(bib->ptr, ib->ptr, ib->length_dw * 4);
	62	bib->length_dw = ib->length_dw;
	63	mutex_lock(&rdev->ib_pool.mutex);
	64	list_add_tail(&bib->list, &rdev->ib_pool.bogus_ib);
	65	mutex_unlock(&rdev->ib_pool.mutex);
	66	}
	67
771fe6b9 JG	68	/*
	69	* IB.
	70	*/
	71	int radeon_ib_get(struct radeon_device rdev, struct radeon_ib *ib)
	72	{
	73	struct radeon_fence *fence;
	74	struct radeon_ib *nib;
91cb91be	75	int r = 0, i, c;
771fe6b9 JG	76
	77	*ib = NULL;
	78	r = radeon_fence_create(rdev, &fence);
	79	if (r) {
91cb91be	80	dev_err(rdev->dev, "failed to create fence for new IB\n");
771fe6b9 JG	81	return r;
	82	}
	83	mutex_lock(&rdev->ib_pool.mutex);
91cb91be JG	84	for (i = rdev->ib_pool.head_id, c = 0, nib = NULL; c < RADEON_IB_POOL_SIZE; c++, i++) {
	85	i &= (RADEON_IB_POOL_SIZE - 1);
	86	if (rdev->ib_pool.ibs[i].free) {
	87	nib = &rdev->ib_pool.ibs[i];
	88	break;
	89	}
771fe6b9	90	}
91cb91be JG	91	if (nib == NULL) {
	92	/* This should never happen, it means we allocated all
	93	* IB and haven't scheduled one yet, return EBUSY to
	94	* userspace hoping that on ioctl recall we get better
	95	* luck
	96	*/
	97	dev_err(rdev->dev, "no free indirect buffer !\n");
ecb114a1	98	mutex_unlock(&rdev->ib_pool.mutex);
91cb91be JG	99	radeon_fence_unref(&fence);
91cb91be JG	100	return -EBUSY;
771fe6b9	101	}
91cb91be JG	102	rdev->ib_pool.head_id = (nib->idx + 1) & (RADEON_IB_POOL_SIZE - 1);
	103	nib->free = false;
	104	if (nib->fence) {
ecb114a1	105	mutex_unlock(&rdev->ib_pool.mutex);
91cb91be JG	106	r = radeon_fence_wait(nib->fence, false);
	107	if (r) {
	108	dev_err(rdev->dev, "error waiting fence of IB(%u:0x%016lX:%u)\n",
	109	nib->idx, (unsigned long)nib->gpu_addr, nib->length_dw);
	110	mutex_lock(&rdev->ib_pool.mutex);
	111	nib->free = true;
	112	mutex_unlock(&rdev->ib_pool.mutex);
	113	radeon_fence_unref(&fence);
	114	return r;
	115	}
	116	mutex_lock(&rdev->ib_pool.mutex);
771fe6b9 JG	117	}
771fe6b9 JG	118	radeon_fence_unref(&nib->fence);
91cb91be	119	nib->fence = fence;
771fe6b9	120	nib->length_dw = 0;
ecb114a1	121	mutex_unlock(&rdev->ib_pool.mutex);
771fe6b9	122	*ib = nib;
91cb91be	123	return 0;
771fe6b9 JG	124	}
	125
	126	void radeon_ib_free(struct radeon_device rdev, struct radeon_ib *ib)
	127	{
	128	struct radeon_ib tmp = ib;
	129
	130	*ib = NULL;
	131	if (tmp == NULL) {
	132	return;
	133	}
7d404c7b JG	134	if (!tmp->fence->emited)
7d404c7b JG	135	radeon_fence_unref(&tmp->fence);
771fe6b9	136	mutex_lock(&rdev->ib_pool.mutex);
91cb91be	137	tmp->free = true;
771fe6b9 JG	138	mutex_unlock(&rdev->ib_pool.mutex);
	139	}
	140
771fe6b9 JG	141	int radeon_ib_schedule(struct radeon_device rdev, struct radeon_ib ib)
	142	{
	143	int r = 0;
	144
771fe6b9 JG	145	if (!ib->length_dw \|\| !rdev->cp.ready) {
771fe6b9 JG	146	/* TODO: Nothings in the ib we should report. */
91cb91be	147	DRM_ERROR("radeon: couldn't schedule IB(%u).\n", ib->idx);
771fe6b9 JG	148	return -EINVAL;
771fe6b9 JG	149	}
ecb114a1	150
6cdf6585	151	/* 64 dwords should be enough for fence too */
771fe6b9 JG	152	r = radeon_ring_lock(rdev, 64);
	153	if (r) {
	154	DRM_ERROR("radeon: scheduling IB failled (%d).\n", r);
771fe6b9 JG	155	return r;
771fe6b9 JG	156	}
3ce0a23d	157	radeon_ring_ib_execute(rdev, ib);
771fe6b9	158	radeon_fence_emit(rdev, ib->fence);
ecb114a1	159	mutex_lock(&rdev->ib_pool.mutex);
91cb91be JG	160	/* once scheduled IB is considered free and protected by the fence */
91cb91be JG	161	ib->free = true;
771fe6b9	162	mutex_unlock(&rdev->ib_pool.mutex);
ecb114a1	163	radeon_ring_unlock_commit(rdev);
771fe6b9 JG	164	return 0;
	165	}
	166
	167	int radeon_ib_pool_init(struct radeon_device *rdev)
	168	{
	169	void *ptr;
	170	uint64_t gpu_addr;
	171	int i;
	172	int r = 0;
	173
9f022ddf JG	174	if (rdev->ib_pool.robj)
9f022ddf JG	175	return 0;
9f93ed39	176	INIT_LIST_HEAD(&rdev->ib_pool.bogus_ib);
771fe6b9	177	/* Allocate 1M object buffer */
4c788679 JG	178	r = radeon_bo_create(rdev, NULL, RADEON_IB_POOL_SIZE641024,
	179	true, RADEON_GEM_DOMAIN_GTT,
	180	&rdev->ib_pool.robj);
771fe6b9 JG	181	if (r) {
	182	DRM_ERROR("radeon: failed to ib pool (%d).\n", r);
	183	return r;
	184	}
4c788679 JG	185	r = radeon_bo_reserve(rdev->ib_pool.robj, false);
	186	if (unlikely(r != 0))
	187	return r;
	188	r = radeon_bo_pin(rdev->ib_pool.robj, RADEON_GEM_DOMAIN_GTT, &gpu_addr);
771fe6b9	189	if (r) {
4c788679	190	radeon_bo_unreserve(rdev->ib_pool.robj);
771fe6b9 JG	191	DRM_ERROR("radeon: failed to pin ib pool (%d).\n", r);
	192	return r;
	193	}
4c788679 JG	194	r = radeon_bo_kmap(rdev->ib_pool.robj, &ptr);
4c788679 JG	195	radeon_bo_unreserve(rdev->ib_pool.robj);
771fe6b9 JG	196	if (r) {
	197	DRM_ERROR("radeon: failed to map ib poll (%d).\n", r);
	198	return r;
	199	}
	200	for (i = 0; i < RADEON_IB_POOL_SIZE; i++) {
	201	unsigned offset;
	202
	203	offset = i * 64 * 1024;
	204	rdev->ib_pool.ibs[i].gpu_addr = gpu_addr + offset;
	205	rdev->ib_pool.ibs[i].ptr = ptr + offset;
	206	rdev->ib_pool.ibs[i].idx = i;
	207	rdev->ib_pool.ibs[i].length_dw = 0;
91cb91be	208	rdev->ib_pool.ibs[i].free = true;
771fe6b9	209	}
91cb91be	210	rdev->ib_pool.head_id = 0;
771fe6b9 JG	211	rdev->ib_pool.ready = true;
	212	DRM_INFO("radeon: ib pool ready.\n");
	213	if (radeon_debugfs_ib_init(rdev)) {
	214	DRM_ERROR("Failed to register debugfs file for IB !\n");
	215	}
	216	return r;
	217	}
	218
	219	void radeon_ib_pool_fini(struct radeon_device *rdev)
	220	{
4c788679 JG	221	int r;
4c788679 JG	222
771fe6b9 JG	223	if (!rdev->ib_pool.ready) {
	224	return;
	225	}
	226	mutex_lock(&rdev->ib_pool.mutex);
9f93ed39	227	radeon_ib_bogus_cleanup(rdev);
eb6b6d7c	228
771fe6b9	229	if (rdev->ib_pool.robj) {
4c788679 JG	230	r = radeon_bo_reserve(rdev->ib_pool.robj, false);
	231	if (likely(r == 0)) {
	232	radeon_bo_kunmap(rdev->ib_pool.robj);
	233	radeon_bo_unpin(rdev->ib_pool.robj);
	234	radeon_bo_unreserve(rdev->ib_pool.robj);
	235	}
	236	radeon_bo_unref(&rdev->ib_pool.robj);
771fe6b9 JG	237	rdev->ib_pool.robj = NULL;
	238	}
	239	mutex_unlock(&rdev->ib_pool.mutex);
	240	}
	241
771fe6b9 JG	242
	243	/*
	244	* Ring.
	245	*/
	246	void radeon_ring_free_size(struct radeon_device *rdev)
	247	{
3ce0a23d JG	248	if (rdev->family >= CHIP_R600)
	249	rdev->cp.rptr = RREG32(R600_CP_RB_RPTR);
	250	else
	251	rdev->cp.rptr = RREG32(RADEON_CP_RB_RPTR);
771fe6b9 JG	252	/* This works because ring_size is a power of 2 */
	253	rdev->cp.ring_free_dw = (rdev->cp.rptr + (rdev->cp.ring_size / 4));
	254	rdev->cp.ring_free_dw -= rdev->cp.wptr;
	255	rdev->cp.ring_free_dw &= rdev->cp.ptr_mask;
	256	if (!rdev->cp.ring_free_dw) {
	257	rdev->cp.ring_free_dw = rdev->cp.ring_size / 4;
	258	}
	259	}
	260
91700f3c	261	int radeon_ring_alloc(struct radeon_device *rdev, unsigned ndw)
771fe6b9 JG	262	{
	263	int r;
	264
	265	/* Align requested size with padding so unlock_commit can
	266	* pad safely */
	267	ndw = (ndw + rdev->cp.align_mask) & ~rdev->cp.align_mask;
771fe6b9 JG	268	while (ndw > (rdev->cp.ring_free_dw - 1)) {
	269	radeon_ring_free_size(rdev);
	270	if (ndw < rdev->cp.ring_free_dw) {
	271	break;
	272	}
	273	r = radeon_fence_wait_next(rdev);
91700f3c	274	if (r)
771fe6b9	275	return r;
771fe6b9 JG	276	}
	277	rdev->cp.count_dw = ndw;
	278	rdev->cp.wptr_old = rdev->cp.wptr;
	279	return 0;
	280	}
	281
91700f3c MG	282	int radeon_ring_lock(struct radeon_device *rdev, unsigned ndw)
	283	{
	284	int r;
	285
	286	mutex_lock(&rdev->cp.mutex);
	287	r = radeon_ring_alloc(rdev, ndw);
	288	if (r) {
	289	mutex_unlock(&rdev->cp.mutex);
	290	return r;
	291	}
	292	return 0;
	293	}
	294
	295	void radeon_ring_commit(struct radeon_device *rdev)
771fe6b9 JG	296	{
	297	unsigned count_dw_pad;
	298	unsigned i;
	299
	300	/* We pad to match fetch size */
	301	count_dw_pad = (rdev->cp.align_mask + 1) -
	302	(rdev->cp.wptr & rdev->cp.align_mask);
	303	for (i = 0; i < count_dw_pad; i++) {
3ce0a23d	304	radeon_ring_write(rdev, 2 << 30);
771fe6b9 JG	305	}
771fe6b9 JG	306	DRM_MEMORYBARRIER();
3ce0a23d	307	radeon_cp_commit(rdev);
91700f3c MG	308	}
	309
	310	void radeon_ring_unlock_commit(struct radeon_device *rdev)
	311	{
	312	radeon_ring_commit(rdev);
771fe6b9 JG	313	mutex_unlock(&rdev->cp.mutex);
	314	}
	315
	316	void radeon_ring_unlock_undo(struct radeon_device *rdev)
	317	{
	318	rdev->cp.wptr = rdev->cp.wptr_old;
	319	mutex_unlock(&rdev->cp.mutex);
	320	}
	321
771fe6b9 JG	322	int radeon_ring_init(struct radeon_device *rdev, unsigned ring_size)
	323	{
	324	int r;
	325
	326	rdev->cp.ring_size = ring_size;
	327	/* Allocate ring buffer */
	328	if (rdev->cp.ring_obj == NULL) {
4c788679 JG	329	r = radeon_bo_create(rdev, NULL, rdev->cp.ring_size, true,
	330	RADEON_GEM_DOMAIN_GTT,
	331	&rdev->cp.ring_obj);
771fe6b9	332	if (r) {
4c788679	333	dev_err(rdev->dev, "(%d) ring create failed\n", r);
771fe6b9 JG	334	return r;
771fe6b9 JG	335	}
4c788679 JG	336	r = radeon_bo_reserve(rdev->cp.ring_obj, false);
	337	if (unlikely(r != 0))
	338	return r;
	339	r = radeon_bo_pin(rdev->cp.ring_obj, RADEON_GEM_DOMAIN_GTT,
	340	&rdev->cp.gpu_addr);
771fe6b9	341	if (r) {
4c788679 JG	342	radeon_bo_unreserve(rdev->cp.ring_obj);
4c788679 JG	343	dev_err(rdev->dev, "(%d) ring pin failed\n", r);
771fe6b9 JG	344	return r;
771fe6b9 JG	345	}
4c788679	346	r = radeon_bo_kmap(rdev->cp.ring_obj,
771fe6b9	347	(void **)&rdev->cp.ring);
4c788679	348	radeon_bo_unreserve(rdev->cp.ring_obj);
771fe6b9	349	if (r) {
4c788679	350	dev_err(rdev->dev, "(%d) ring map failed\n", r);
771fe6b9 JG	351	return r;
	352	}
	353	}
	354	rdev->cp.ptr_mask = (rdev->cp.ring_size / 4) - 1;
	355	rdev->cp.ring_free_dw = rdev->cp.ring_size / 4;
	356	return 0;
	357	}
	358
	359	void radeon_ring_fini(struct radeon_device *rdev)
	360	{
4c788679 JG	361	int r;
4c788679 JG	362
771fe6b9 JG	363	mutex_lock(&rdev->cp.mutex);
771fe6b9 JG	364	if (rdev->cp.ring_obj) {
4c788679 JG	365	r = radeon_bo_reserve(rdev->cp.ring_obj, false);
	366	if (likely(r == 0)) {
	367	radeon_bo_kunmap(rdev->cp.ring_obj);
	368	radeon_bo_unpin(rdev->cp.ring_obj);
	369	radeon_bo_unreserve(rdev->cp.ring_obj);
	370	}
	371	radeon_bo_unref(&rdev->cp.ring_obj);
771fe6b9 JG	372	rdev->cp.ring = NULL;
	373	rdev->cp.ring_obj = NULL;
	374	}
	375	mutex_unlock(&rdev->cp.mutex);
	376	}
	377
	378
	379	/*
	380	* Debugfs info
	381	*/
	382	#if defined(CONFIG_DEBUG_FS)
	383	static int radeon_debugfs_ib_info(struct seq_file m, void data)
	384	{
	385	struct drm_info_node node = (struct drm_info_node ) m->private;
	386	struct radeon_ib *ib = node->info_ent->data;
	387	unsigned i;
	388
	389	if (ib == NULL) {
	390	return 0;
	391	}
91cb91be	392	seq_printf(m, "IB %04u\n", ib->idx);
771fe6b9 JG	393	seq_printf(m, "IB fence %p\n", ib->fence);
	394	seq_printf(m, "IB size %05u dwords\n", ib->length_dw);
	395	for (i = 0; i < ib->length_dw; i++) {
	396	seq_printf(m, "[%05u]=0x%08X\n", i, ib->ptr[i]);
	397	}
	398	return 0;
	399	}
	400
9f93ed39 JG	401	static int radeon_debugfs_ib_bogus_info(struct seq_file m, void data)
	402	{
	403	struct drm_info_node node = (struct drm_info_node ) m->private;
	404	struct radeon_device *rdev = node->info_ent->data;
	405	struct radeon_ib *ib;
	406	unsigned i;
	407
	408	mutex_lock(&rdev->ib_pool.mutex);
	409	if (list_empty(&rdev->ib_pool.bogus_ib)) {
	410	mutex_unlock(&rdev->ib_pool.mutex);
	411	seq_printf(m, "no bogus IB recorded\n");
	412	return 0;
	413	}
	414	ib = list_first_entry(&rdev->ib_pool.bogus_ib, struct radeon_ib, list);
	415	list_del_init(&ib->list);
	416	mutex_unlock(&rdev->ib_pool.mutex);
	417	seq_printf(m, "IB size %05u dwords\n", ib->length_dw);
	418	for (i = 0; i < ib->length_dw; i++) {
	419	seq_printf(m, "[%05u]=0x%08X\n", i, ib->ptr[i]);
	420	}
	421	vfree(ib->ptr);
	422	kfree(ib);
	423	return 0;
	424	}
	425
771fe6b9 JG	426	static struct drm_info_list radeon_debugfs_ib_list[RADEON_IB_POOL_SIZE];
771fe6b9 JG	427	static char radeon_debugfs_ib_names[RADEON_IB_POOL_SIZE][32];
9f93ed39 JG	428
	429	static struct drm_info_list radeon_debugfs_ib_bogus_info_list[] = {
	430	{"radeon_ib_bogus", radeon_debugfs_ib_bogus_info, 0, NULL},
	431	};
771fe6b9 JG	432	#endif
	433
	434	int radeon_debugfs_ib_init(struct radeon_device *rdev)
	435	{
	436	#if defined(CONFIG_DEBUG_FS)
	437	unsigned i;
9f93ed39	438	int r;
771fe6b9	439
9f93ed39 JG	440	radeon_debugfs_ib_bogus_info_list[0].data = rdev;
	441	r = radeon_debugfs_add_files(rdev, radeon_debugfs_ib_bogus_info_list, 1);
	442	if (r)
	443	return r;
771fe6b9 JG	444	for (i = 0; i < RADEON_IB_POOL_SIZE; i++) {
	445	sprintf(radeon_debugfs_ib_names[i], "radeon_ib_%04u", i);
	446	radeon_debugfs_ib_list[i].name = radeon_debugfs_ib_names[i];
	447	radeon_debugfs_ib_list[i].show = &radeon_debugfs_ib_info;
	448	radeon_debugfs_ib_list[i].driver_features = 0;
	449	radeon_debugfs_ib_list[i].data = &rdev->ib_pool.ibs[i];
	450	}
	451	return radeon_debugfs_add_files(rdev, radeon_debugfs_ib_list,
	452	RADEON_IB_POOL_SIZE);
	453	#else
	454	return 0;
	455	#endif
	456	}