summaryrefslogtreecommitdiff
path: root/linux-core/xgi_drv.c
blob: 4f0b4ed0df2fb2979d62e8a3100ba51d97796e49 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
/****************************************************************************
 * Copyright (C) 2003-2006 by XGI Technology, Taiwan.
 *
 * All Rights Reserved.
 *
 * 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 on 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 (including the
 * next paragraph) 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 NON-INFRINGEMENT.  IN NO EVENT SHALL
 * XGI AND/OR ITS SUPPLIERS 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 "drmP.h"
#include "drm.h"
#include "xgi_drv.h"
#include "xgi_regs.h"
#include "xgi_misc.h"
#include "xgi_cmdlist.h"

#include "drm_pciids.h"

static struct pci_device_id pciidlist[] = {
	xgi_PCI_IDS
};

static struct drm_fence_driver xgi_fence_driver = {
	.num_classes = 1,
	.wrap_diff = BEGIN_BEGIN_IDENTIFICATION_MASK,
	.flush_diff = BEGIN_BEGIN_IDENTIFICATION_MASK - 1,
	.sequence_mask = BEGIN_BEGIN_IDENTIFICATION_MASK,
	.lazy_capable = 1,
	.emit = xgi_fence_emit_sequence,
	.poke_flush = xgi_poke_flush,
	.has_irq = xgi_fence_has_irq
};

int xgi_bootstrap(struct drm_device *, void *, struct drm_file *);

static struct drm_ioctl_desc xgi_ioctls[] = {
	DRM_IOCTL_DEF(DRM_XGI_BOOTSTRAP, xgi_bootstrap, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
	DRM_IOCTL_DEF(DRM_XGI_ALLOC, xgi_alloc_ioctl, DRM_AUTH),
	DRM_IOCTL_DEF(DRM_XGI_FREE, xgi_free_ioctl, DRM_AUTH),
	DRM_IOCTL_DEF(DRM_XGI_SUBMIT_CMDLIST, xgi_submit_cmdlist, DRM_AUTH),
	DRM_IOCTL_DEF(DRM_XGI_STATE_CHANGE, xgi_state_change_ioctl, DRM_AUTH|DRM_MASTER),
};

static const int xgi_max_ioctl = DRM_ARRAY_SIZE(xgi_ioctls);

static int probe(struct pci_dev *pdev, const struct pci_device_id *ent);
static int xgi_driver_load(struct drm_device *dev, unsigned long flags);
static int xgi_driver_unload(struct drm_device *dev);
static void xgi_driver_lastclose(struct drm_device * dev);
static void xgi_reclaim_buffers_locked(struct drm_device * dev,
	struct drm_file * filp);
static irqreturn_t xgi_kern_isr(DRM_IRQ_ARGS);


static struct drm_driver driver = {
	.driver_features =
		DRIVER_PCI_DMA | DRIVER_HAVE_DMA | DRIVER_HAVE_IRQ |
		DRIVER_IRQ_SHARED | DRIVER_SG,
	.dev_priv_size = sizeof(struct xgi_info),
	.load = xgi_driver_load,
	.unload = xgi_driver_unload,
	.lastclose = xgi_driver_lastclose,
	.dma_quiescent = NULL,
	.irq_preinstall = NULL,
	.irq_postinstall = NULL,
	.irq_uninstall = NULL,
	.irq_handler = xgi_kern_isr,
	.reclaim_buffers = drm_core_reclaim_buffers,
	.reclaim_buffers_idlelocked = xgi_reclaim_buffers_locked,
	.get_map_ofs = drm_core_get_map_ofs,
	.get_reg_ofs = drm_core_get_reg_ofs,
	.ioctls = xgi_ioctls,
	.dma_ioctl = NULL,

	.fops = {
		.owner = THIS_MODULE,
		.open = drm_open,
		.release = drm_release,
		.ioctl = drm_ioctl,
		.mmap = drm_mmap,
		.poll = drm_poll,
		.fasync = drm_fasync,
#if defined(CONFIG_COMPAT) && LINUX_VERSION_CODE > KERNEL_VERSION(2,6,9)
		.compat_ioctl = xgi_compat_ioctl,
#endif
	},

	.pci_driver = {
		.name = DRIVER_NAME,
		.id_table = pciidlist,
		.probe = probe,
		.remove = __devexit_p(drm_cleanup_pci),
	},

	.fence_driver = &xgi_fence_driver,

	.name = DRIVER_NAME,
	.desc = DRIVER_DESC,
	.date = DRIVER_DATE,
	.major = DRIVER_MAJOR,
	.minor = DRIVER_MINOR,
	.patchlevel = DRIVER_PATCHLEVEL,

};

static int probe(struct pci_dev *pdev, const struct pci_device_id *ent)
{
	return drm_get_dev(pdev, ent, &driver);
}


static int __init xgi_init(void)
{
	driver.num_ioctls = xgi_max_ioctl;
	return drm_init(&driver, pciidlist);
}

static void __exit xgi_exit(void)
{
	drm_exit(&driver);
}

module_init(xgi_init);
module_exit(xgi_exit);

MODULE_AUTHOR(DRIVER_AUTHOR);
MODULE_DESCRIPTION(DRIVER_DESC);
MODULE_LICENSE("GPL and additional rights");


void xgi_engine_init(struct xgi_info * info)
{
	u8 temp;


	OUT3C5B(info->mmio_map, 0x11, 0x92);

	/* -------> copy from OT2D
	 * PCI Retry Control Register.
	 * disable PCI read retry & enable write retry in mem. (10xx xxxx)b
	 */
	temp = IN3X5B(info->mmio_map, 0x55);
	OUT3X5B(info->mmio_map, 0x55, (temp & 0xbf) | 0x80);

	xgi_enable_ge(info);

	/* Enable linear addressing of the card. */
	temp = IN3X5B(info->mmio_map, 0x21);
	OUT3X5B(info->mmio_map, 0x21, temp | 0x20);

	/* Enable 32-bit internal data path */
	temp = IN3X5B(info->mmio_map, 0x2A);
	OUT3X5B(info->mmio_map, 0x2A, temp | 0x40);

	/* Enable PCI burst write ,disable burst read and enable MMIO. */
	/*
	 * 0x3D4.39 Enable PCI burst write, disable burst read and enable MMIO.
	 * 7 ---- Pixel Data Format 1:  big endian 0:  little endian
	 * 6 5 4 3---- Memory Data with Big Endian Format, BE[3:0]#  with Big Endian Format
	 * 2 ---- PCI Burst Write Enable
	 * 1 ---- PCI Burst Read Enable
	 * 0 ---- MMIO Control
	 */
	temp = IN3X5B(info->mmio_map, 0x39);
	OUT3X5B(info->mmio_map, 0x39, (temp | 0x05) & 0xfd);

	/* enable GEIO decode */
	/* temp = IN3X5B(info->mmio_map, 0x29);
	 * OUT3X5B(info->mmio_map, 0x29, temp | 0x08);
	 */

	/* Enable graphic engine I/O PCI retry function*/
	/* temp = IN3X5B(info->mmio_map, 0x62);
	 * OUT3X5B(info->mmio_map, 0x62, temp | 0x50);
	 */

	/* protect all register except which protected by 3c5.0e.7 */
        /* OUT3C5B(info->mmio_map, 0x11, 0x87); */
}


int xgi_bootstrap(struct drm_device * dev, void * data,
		  struct drm_file * filp)
{
	struct xgi_info *info = dev->dev_private;
	struct xgi_bootstrap * bs = (struct xgi_bootstrap *) data;
	struct drm_map_list *maplist;
	int err;


	DRM_SPININIT(&info->fence_lock, "fence lock");
	info->next_sequence = 0;
	info->complete_sequence = 0;

	if (info->mmio_map == NULL) {
		err = drm_addmap(dev, info->mmio.base, info->mmio.size,
				 _DRM_REGISTERS, _DRM_KERNEL,
				 &info->mmio_map);
		if (err) {
			DRM_ERROR("Unable to map MMIO region: %d\n", err);
			return err;
		}

		xgi_enable_mmio(info);
		xgi_engine_init(info);
	}


	info->fb.size = IN3CFB(info->mmio_map, 0x54) * 8 * 1024 * 1024;

	DRM_INFO("fb   base: 0x%lx, size: 0x%x (probed)\n",
		 (unsigned long) info->fb.base, info->fb.size);


	if ((info->fb.base == 0) || (info->fb.size == 0)) {
		DRM_ERROR("framebuffer appears to be wrong: 0x%lx 0x%x\n",
			  (unsigned long) info->fb.base, info->fb.size);
		return -EINVAL;
	}


	/* Init the resource manager */
	if (!info->fb_heap_initialized) {
		err = xgi_fb_heap_init(info);
		if (err) {
			DRM_ERROR("Unable to initialize FB heap.\n");
			return err;
		}
	}


	info->pcie.size = bs->gart.size;

	/* Init the resource manager */
	if (!info->pcie_heap_initialized) {
		err = xgi_pcie_heap_init(info);
		if (err) {
			DRM_ERROR("Unable to initialize GART heap.\n");
			return err;
		}

		/* Alloc 1M bytes for cmdbuffer which is flush2D batch array */
		err = xgi_cmdlist_initialize(info, 0x100000, filp);
		if (err) {
			DRM_ERROR("xgi_cmdlist_initialize() failed\n");
			return err;
		}
	}


	if (info->pcie_map == NULL) {
		err = drm_addmap(info->dev, 0, info->pcie.size,
				 _DRM_SCATTER_GATHER, _DRM_LOCKED,
				 & info->pcie_map);
		if (err) {
			DRM_ERROR("Could not add map for GART backing "
				  "store.\n");
			return err;
		}
	}


	maplist = drm_find_matching_map(dev, info->pcie_map);
	if (maplist == NULL) {
		DRM_ERROR("Could not find GART backing store map.\n");
		return -EINVAL;
	}

	bs->gart = *info->pcie_map;
	bs->gart.handle = (void *)(unsigned long) maplist->user_token;
	return 0;
}


void xgi_driver_lastclose(struct drm_device * dev)
{
	struct xgi_info * info = dev->dev_private;

	if (info != NULL) {
		if (info->mmio_map != NULL) {
			xgi_cmdlist_cleanup(info);
			xgi_disable_ge(info);
			xgi_disable_mmio(info);
		}

		/* The core DRM lastclose routine will destroy all of our
		 * mappings for us.  NULL out the pointers here so that
		 * xgi_bootstrap can do the right thing.
		 */
		info->pcie_map = NULL;
		info->mmio_map = NULL;
		info->fb_map = NULL;

		if (info->pcie_heap_initialized) {
			drm_ati_pcigart_cleanup(dev, &info->gart_info);
		}

		if (info->fb_heap_initialized
		    || info->pcie_heap_initialized) {
			drm_sman_cleanup(&info->sman);

			info->fb_heap_initialized = FALSE;
			info->pcie_heap_initialized = FALSE;
		}
	}
}


void xgi_reclaim_buffers_locked(struct drm_device * dev,
				struct drm_file * filp)
{
	struct xgi_info * info = dev->dev_private;

	mutex_lock(&info->dev->struct_mutex);
	if (drm_sman_owner_clean(&info->sman, (unsigned long) filp)) {
		mutex_unlock(&info->dev->struct_mutex);
		return;
	}

	if (dev->driver->dma_quiescent) {
		dev->driver->dma_quiescent(dev);
	}

	drm_sman_owner_cleanup(&info->sman, (unsigned long) filp);
	mutex_unlock(&info->dev->struct_mutex);
	return;
}


/*
 * driver receives an interrupt if someone waiting, then hand it off.
 */
irqreturn_t xgi_kern_isr(DRM_IRQ_ARGS)
{
	struct drm_device *dev = (struct drm_device *) arg;
	struct xgi_info *info = dev->dev_private;
	const u32 irq_bits = le32_to_cpu(DRM_READ32(info->mmio_map,
					(0x2800
					 + M2REG_AUTO_LINK_STATUS_ADDRESS)))
		& (M2REG_ACTIVE_TIMER_INTERRUPT_MASK
		   | M2REG_ACTIVE_INTERRUPT_0_MASK
		   | M2REG_ACTIVE_INTERRUPT_2_MASK
		   | M2REG_ACTIVE_INTERRUPT_3_MASK);


	if (irq_bits != 0) {
		DRM_WRITE32(info->mmio_map,
			    0x2800 + M2REG_AUTO_LINK_SETTING_ADDRESS,
			    cpu_to_le32(M2REG_AUTO_LINK_SETTING_COMMAND | irq_bits));
		xgi_fence_handler(dev);
		return IRQ_HANDLED;
	} else {
		return IRQ_NONE;
	}
}


int xgi_driver_load(struct drm_device *dev, unsigned long flags)
{
	struct xgi_info *info = drm_alloc(sizeof(*info), DRM_MEM_DRIVER);
	int err;

	if (!info)
		return -ENOMEM;

	(void) memset(info, 0, sizeof(*info));
	dev->dev_private = info;
	info->dev = dev;

	info->mmio.base = drm_get_resource_start(dev, 1);
	info->mmio.size = drm_get_resource_len(dev, 1);

	DRM_INFO("mmio base: 0x%lx, size: 0x%x\n",
		 (unsigned long) info->mmio.base, info->mmio.size);


	if ((info->mmio.base == 0) || (info->mmio.size == 0)) {
		DRM_ERROR("mmio appears to be wrong: 0x%lx 0x%x\n",
			  (unsigned long) info->mmio.base, info->mmio.size);
		err = -EINVAL;
		goto fail;
	}


	info->fb.base = drm_get_resource_start(dev, 0);
	info->fb.size = drm_get_resource_len(dev, 0);

	DRM_INFO("fb   base: 0x%lx, size: 0x%x\n",
		 (unsigned long) info->fb.base, info->fb.size);


	err = drm_sman_init(&info->sman, 2, 12, 8);
	if (err) {
		goto fail;
	}


	return 0;

fail:
	drm_free(info, sizeof(*info), DRM_MEM_DRIVER);
	return err;
}

int xgi_driver_unload(struct drm_device *dev)
{
	struct xgi_info * info = dev->dev_private;

	drm_sman_takedown(&info->sman);
	drm_free(info, sizeof(*info), DRM_MEM_DRIVER);
	dev->dev_private = NULL;

	return 0;
}
1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778
/* mach64_dma.c -- DMA support for mach64 (Rage Pro) driver -*- linux-c -*- */
/**
 * \file mach64_dma.c
 * DMA support for mach64 (Rage Pro) driver
 *
 * \author Gareth Hughes <gareth@valinux.com>
 * \author Frank C. Earl <fearl@airmail.net>
 * \author Leif Delgass <ldelgass@retinalburn.net>
 * \author José Fonseca <j_r_fonseca@yahoo.co.uk>
 */

/*
 * Copyright 2000 Gareth Hughes
 * Copyright 2002 Frank C. Earl
 * Copyright 2002-2003 Leif Delgass
 * All Rights Reserved.
 *
 * 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 (including the next
 * paragraph) 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 OWNER(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 "drmP.h"
#include "drm.h"
#include "mach64_drm.h"
#include "mach64_drv.h"

/*******************************************************************/
/** \name Engine, FIFO control */
/*@{*/

/**
 * Waits for free entries in the FIFO.
 *
 * \note Most writes to Mach64 registers are automatically routed through
 * command FIFO which is 16 entry deep. Prior to writing to any draw engine
 * register one has to ensure that enough FIFO entries are available by calling
 * this function.  Failure to do so may cause the engine to lock.
 *
 * \param dev_priv pointer to device private data structure.
 * \param entries number of free entries in the FIFO to wait for.
 *
 * \returns zero on success, or -EBUSY if the timeout (specificed by
 * drm_mach64_private::usec_timeout) occurs.
 */
int mach64_do_wait_for_fifo(drm_mach64_private_t *dev_priv, int entries)
{
	int slots = 0, i;

	for (i = 0; i < dev_priv->usec_timeout; i++) {
		slots = (MACH64_READ(MACH64_FIFO_STAT) & MACH64_FIFO_SLOT_MASK);
		if (slots <= (0x8000 >> entries))
			return 0;
		DRM_UDELAY(1);
	}

	DRM_INFO("failed! slots=%d entries=%d\n", slots, entries);
	return -EBUSY;
}

/**
 * Wait for the draw engine to be idle.
 */
int mach64_do_wait_for_idle(drm_mach64_private_t *dev_priv)
{
	int i, ret;

	ret = mach64_do_wait_for_fifo(dev_priv, 16);
	if (ret < 0)
		return ret;

	for (i = 0; i < dev_priv->usec_timeout; i++) {
		if (!(MACH64_READ(MACH64_GUI_STAT) & MACH64_GUI_ACTIVE))
			return 0;
		DRM_UDELAY(1);
	}

	DRM_INFO("failed! GUI_STAT=0x%08x\n", MACH64_READ(MACH64_GUI_STAT));
	mach64_dump_ring_info(dev_priv);
	return -EBUSY;
}

/**
 * Wait for free entries in the ring buffer.
 *
 * The Mach64 bus master can be configured to act as a virtual FIFO, using a
 * circular buffer (commonly referred as "ring buffer" in other drivers) with
 * pointers to engine commands. This allows the CPU to do other things while
 * the graphics engine is busy, i.e., DMA mode.
 *
 * This function should be called before writing new entries to the ring
 * buffer.
 *
 * \param dev_priv pointer to device private data structure.
 * \param n number of free entries in the ring buffer to wait for.
 *
 * \returns zero on success, or -EBUSY if the timeout (specificed by
 * drm_mach64_private_t::usec_timeout) occurs.
 *
 * \sa mach64_dump_ring_info()
 */
int mach64_wait_ring(drm_mach64_private_t *dev_priv, int n)
{
	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
	int i;

	for (i = 0; i < dev_priv->usec_timeout; i++) {
		mach64_update_ring_snapshot(dev_priv);
		if (ring->space >= n) {
			if (i > 0)
				DRM_DEBUG("%d usecs\n", i);
			return 0;
		}
		DRM_UDELAY(1);
	}

	/* FIXME: This is being ignored... */
	DRM_ERROR("failed!\n");
	mach64_dump_ring_info(dev_priv);
	return -EBUSY;
}

/**
 * Wait until all DMA requests have been processed...
 *
 * \sa mach64_wait_ring()
 */
static int mach64_ring_idle(drm_mach64_private_t *dev_priv)
{
	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
	u32 head;
	int i;

	head = ring->head;
	i = 0;
	while (i < dev_priv->usec_timeout) {
		mach64_update_ring_snapshot(dev_priv);
		if (ring->head == ring->tail &&
		    !(MACH64_READ(MACH64_GUI_STAT) & MACH64_GUI_ACTIVE)) {
			if (i > 0)
				DRM_DEBUG("%d usecs\n", i);
			return 0;
		}
		if (ring->head == head) {
			++i;
		} else {
			head = ring->head;
			i = 0;
		}
		DRM_UDELAY(1);
	}

	DRM_INFO("failed! GUI_STAT=0x%08x\n", MACH64_READ(MACH64_GUI_STAT));
	mach64_dump_ring_info(dev_priv);
	return -EBUSY;
}

/**
 * Reset the the ring buffer descriptors.
 *
 * \sa mach64_do_engine_reset()
 */
static void mach64_ring_reset(drm_mach64_private_t *dev_priv)
{
	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;

	mach64_do_release_used_buffers(dev_priv);
	ring->head_addr = ring->start_addr;
	ring->head = ring->tail = 0;
	ring->space = ring->size;

	MACH64_WRITE(MACH64_BM_GUI_TABLE_CMD,
		     ring->head_addr | MACH64_CIRCULAR_BUF_SIZE_16KB);

	dev_priv->ring_running = 0;
}

/**
 * Ensure the all the queued commands will be processed.
 */
int mach64_do_dma_flush(drm_mach64_private_t *dev_priv)
{
	/* FIXME: It's not necessary to wait for idle when flushing
	 * we just need to ensure the ring will be completely processed
	 * in finite time without another ioctl
	 */
	return mach64_ring_idle(dev_priv);
}

/**
 * Stop all DMA activity.
 */
int mach64_do_dma_idle(drm_mach64_private_t *dev_priv)
{
	int ret;

	/* wait for completion */
	if ((ret = mach64_ring_idle(dev_priv)) < 0) {
		DRM_ERROR("failed BM_GUI_TABLE=0x%08x tail: %u\n",
			  MACH64_READ(MACH64_BM_GUI_TABLE),
			  dev_priv->ring.tail);
		return ret;
	}

	mach64_ring_stop(dev_priv);

	/* clean up after pass */
	mach64_do_release_used_buffers(dev_priv);
	return 0;
}

/**
 * Reset the engine.  This will stop the DMA if it is running.
 */
int mach64_do_engine_reset(drm_mach64_private_t *dev_priv)
{
	u32 tmp;

	DRM_DEBUG("\n");

	/* Kill off any outstanding DMA transfers.
	 */
	tmp = MACH64_READ(MACH64_BUS_CNTL);
	MACH64_WRITE(MACH64_BUS_CNTL, tmp | MACH64_BUS_MASTER_DIS);

	/* Reset the GUI engine (high to low transition).
	 */
	tmp = MACH64_READ(MACH64_GEN_TEST_CNTL);
	MACH64_WRITE(MACH64_GEN_TEST_CNTL, tmp & ~MACH64_GUI_ENGINE_ENABLE);
	/* Enable the GUI engine
	 */
	tmp = MACH64_READ(MACH64_GEN_TEST_CNTL);
	MACH64_WRITE(MACH64_GEN_TEST_CNTL, tmp | MACH64_GUI_ENGINE_ENABLE);

	/* ensure engine is not locked up by clearing any FIFO or HOST errors
	 */
	tmp = MACH64_READ(MACH64_BUS_CNTL);
	MACH64_WRITE(MACH64_BUS_CNTL, tmp | 0x00a00000);

	/* Once GUI engine is restored, disable bus mastering */
	MACH64_WRITE(MACH64_SRC_CNTL, 0);

	/* Reset descriptor ring */
	mach64_ring_reset(dev_priv);

	return 0;
}

/*@}*/


/*******************************************************************/
/** \name Debugging output */
/*@{*/

/**
 * Dump engine registers values.
 */
void mach64_dump_engine_info(drm_mach64_private_t *dev_priv)
{
	DRM_INFO("\n");
	if (!dev_priv->is_pci) {
		DRM_INFO("           AGP_BASE = 0x%08x\n",
			 MACH64_READ(MACH64_AGP_BASE));
		DRM_INFO("           AGP_CNTL = 0x%08x\n",
			 MACH64_READ(MACH64_AGP_CNTL));
	}
	DRM_INFO("     ALPHA_TST_CNTL = 0x%08x\n",
		 MACH64_READ(MACH64_ALPHA_TST_CNTL));
	DRM_INFO("\n");
	DRM_INFO("         BM_COMMAND = 0x%08x\n",
		 MACH64_READ(MACH64_BM_COMMAND));
	DRM_INFO("BM_FRAME_BUF_OFFSET = 0x%08x\n",
		 MACH64_READ(MACH64_BM_FRAME_BUF_OFFSET));
	DRM_INFO("       BM_GUI_TABLE = 0x%08x\n",
		 MACH64_READ(MACH64_BM_GUI_TABLE));
	DRM_INFO("          BM_STATUS = 0x%08x\n",
		 MACH64_READ(MACH64_BM_STATUS));
	DRM_INFO(" BM_SYSTEM_MEM_ADDR = 0x%08x\n",
		 MACH64_READ(MACH64_BM_SYSTEM_MEM_ADDR));
	DRM_INFO("    BM_SYSTEM_TABLE = 0x%08x\n",
		 MACH64_READ(MACH64_BM_SYSTEM_TABLE));
	DRM_INFO("           BUS_CNTL = 0x%08x\n",
		 MACH64_READ(MACH64_BUS_CNTL));
	DRM_INFO("\n");
	/* DRM_INFO( "         CLOCK_CNTL = 0x%08x\n", MACH64_READ( MACH64_CLOCK_CNTL ) ); */
	DRM_INFO("        CLR_CMP_CLR = 0x%08x\n",
		 MACH64_READ(MACH64_CLR_CMP_CLR));
	DRM_INFO("       CLR_CMP_CNTL = 0x%08x\n",
		 MACH64_READ(MACH64_CLR_CMP_CNTL));
	/* DRM_INFO( "        CLR_CMP_MSK = 0x%08x\n", MACH64_READ( MACH64_CLR_CMP_MSK ) ); */
	DRM_INFO("     CONFIG_CHIP_ID = 0x%08x\n",
		 MACH64_READ(MACH64_CONFIG_CHIP_ID));
	DRM_INFO("        CONFIG_CNTL = 0x%08x\n",
		 MACH64_READ(MACH64_CONFIG_CNTL));
	DRM_INFO("       CONFIG_STAT0 = 0x%08x\n",
		 MACH64_READ(MACH64_CONFIG_STAT0));
	DRM_INFO("       CONFIG_STAT1 = 0x%08x\n",
		 MACH64_READ(MACH64_CONFIG_STAT1));
	DRM_INFO("       CONFIG_STAT2 = 0x%08x\n",
		 MACH64_READ(MACH64_CONFIG_STAT2));
	DRM_INFO("            CRC_SIG = 0x%08x\n", MACH64_READ(MACH64_CRC_SIG));
	DRM_INFO("  CUSTOM_MACRO_CNTL = 0x%08x\n",
		 MACH64_READ(MACH64_CUSTOM_MACRO_CNTL));
	DRM_INFO("\n");
	/* DRM_INFO( "           DAC_CNTL = 0x%08x\n", MACH64_READ( MACH64_DAC_CNTL ) ); */
	/* DRM_INFO( "           DAC_REGS = 0x%08x\n", MACH64_READ( MACH64_DAC_REGS ) ); */
	DRM_INFO("        DP_BKGD_CLR = 0x%08x\n",
		 MACH64_READ(MACH64_DP_BKGD_CLR));
	DRM_INFO("        DP_FRGD_CLR = 0x%08x\n",
		 MACH64_READ(MACH64_DP_FRGD_CLR));
	DRM_INFO("             DP_MIX = 0x%08x\n", MACH64_READ(MACH64_DP_MIX));
	DRM_INFO("       DP_PIX_WIDTH = 0x%08x\n",
		 MACH64_READ(MACH64_DP_PIX_WIDTH));
	DRM_INFO("             DP_SRC = 0x%08x\n", MACH64_READ(MACH64_DP_SRC));
	DRM_INFO("      DP_WRITE_MASK = 0x%08x\n",
		 MACH64_READ(MACH64_DP_WRITE_MASK));
	DRM_INFO("         DSP_CONFIG = 0x%08x\n",
		 MACH64_READ(MACH64_DSP_CONFIG));
	DRM_INFO("         DSP_ON_OFF = 0x%08x\n",
		 MACH64_READ(MACH64_DSP_ON_OFF));
	DRM_INFO("           DST_CNTL = 0x%08x\n",
		 MACH64_READ(MACH64_DST_CNTL));
	DRM_INFO("      DST_OFF_PITCH = 0x%08x\n",
		 MACH64_READ(MACH64_DST_OFF_PITCH));
	DRM_INFO("\n");
	/* DRM_INFO( "       EXT_DAC_REGS = 0x%08x\n", MACH64_READ( MACH64_EXT_DAC_REGS ) ); */
	DRM_INFO("       EXT_MEM_CNTL = 0x%08x\n",
		 MACH64_READ(MACH64_EXT_MEM_CNTL));
	DRM_INFO("\n");
	DRM_INFO("          FIFO_STAT = 0x%08x\n",
		 MACH64_READ(MACH64_FIFO_STAT));
	DRM_INFO("\n");
	DRM_INFO("      GEN_TEST_CNTL = 0x%08x\n",
		 MACH64_READ(MACH64_GEN_TEST_CNTL));
	/* DRM_INFO( "              GP_IO = 0x%08x\n", MACH64_READ( MACH64_GP_IO ) ); */
	DRM_INFO("   GUI_CMDFIFO_DATA = 0x%08x\n",
		 MACH64_READ(MACH64_GUI_CMDFIFO_DATA));
	DRM_INFO("  GUI_CMDFIFO_DEBUG = 0x%08x\n",
		 MACH64_READ(MACH64_GUI_CMDFIFO_DEBUG));
	DRM_INFO("           GUI_CNTL = 0x%08x\n",
		 MACH64_READ(MACH64_GUI_CNTL));
	DRM_INFO("           GUI_STAT = 0x%08x\n",
		 MACH64_READ(MACH64_GUI_STAT));
	DRM_INFO("      GUI_TRAJ_CNTL = 0x%08x\n",
		 MACH64_READ(MACH64_GUI_TRAJ_CNTL));
	DRM_INFO("\n");
	DRM_INFO("          HOST_CNTL = 0x%08x\n",
		 MACH64_READ(MACH64_HOST_CNTL));
	DRM_INFO("           HW_DEBUG = 0x%08x\n",
		 MACH64_READ(MACH64_HW_DEBUG));
	DRM_INFO("\n");
	DRM_INFO("    MEM_ADDR_CONFIG = 0x%08x\n",
		 MACH64_READ(MACH64_MEM_ADDR_CONFIG));
	DRM_INFO("       MEM_BUF_CNTL = 0x%08x\n",
		 MACH64_READ(MACH64_MEM_BUF_CNTL));
	DRM_INFO("\n");
	DRM_INFO("           PAT_REG0 = 0x%08x\n",
		 MACH64_READ(MACH64_PAT_REG0));
	DRM_INFO("           PAT_REG1 = 0x%08x\n",
		 MACH64_READ(MACH64_PAT_REG1));
	DRM_INFO("\n");
	DRM_INFO("            SC_LEFT = 0x%08x\n", MACH64_READ(MACH64_SC_LEFT));
	DRM_INFO("           SC_RIGHT = 0x%08x\n",
		 MACH64_READ(MACH64_SC_RIGHT));
	DRM_INFO("             SC_TOP = 0x%08x\n", MACH64_READ(MACH64_SC_TOP));
	DRM_INFO("          SC_BOTTOM = 0x%08x\n",
		 MACH64_READ(MACH64_SC_BOTTOM));
	DRM_INFO("\n");
	DRM_INFO("      SCALE_3D_CNTL = 0x%08x\n",
		 MACH64_READ(MACH64_SCALE_3D_CNTL));
	DRM_INFO("       SCRATCH_REG0 = 0x%08x\n",
		 MACH64_READ(MACH64_SCRATCH_REG0));
	DRM_INFO("       SCRATCH_REG1 = 0x%08x\n",
		 MACH64_READ(MACH64_SCRATCH_REG1));
	DRM_INFO("         SETUP_CNTL = 0x%08x\n",
		 MACH64_READ(MACH64_SETUP_CNTL));
	DRM_INFO("           SRC_CNTL = 0x%08x\n",
		 MACH64_READ(MACH64_SRC_CNTL));
	DRM_INFO("\n");
	DRM_INFO("           TEX_CNTL = 0x%08x\n",
		 MACH64_READ(MACH64_TEX_CNTL));
	DRM_INFO("     TEX_SIZE_PITCH = 0x%08x\n",
		 MACH64_READ(MACH64_TEX_SIZE_PITCH));
	DRM_INFO("       TIMER_CONFIG = 0x%08x\n",
		 MACH64_READ(MACH64_TIMER_CONFIG));
	DRM_INFO("\n");
	DRM_INFO("             Z_CNTL = 0x%08x\n", MACH64_READ(MACH64_Z_CNTL));
	DRM_INFO("        Z_OFF_PITCH = 0x%08x\n",
		 MACH64_READ(MACH64_Z_OFF_PITCH));
	DRM_INFO("\n");
}

#define MACH64_DUMP_CONTEXT	3

/**
 * Used by mach64_dump_ring_info() to dump the contents of the current buffer
 * pointed by the ring head.
 */
static void mach64_dump_buf_info(drm_mach64_private_t *dev_priv,
				 struct drm_buf *buf)
{
	u32 addr = GETBUFADDR(buf);
	u32 used = buf->used >> 2;
	u32 sys_addr = MACH64_READ(MACH64_BM_SYSTEM_MEM_ADDR);
	u32 *p = GETBUFPTR(buf);
	int skipped = 0;

	DRM_INFO("buffer contents:\n");

	while (used) {
		u32 reg, count;

		reg = le32_to_cpu(*p++);
		if (addr <= GETBUFADDR(buf) + MACH64_DUMP_CONTEXT * 4 ||
		    (addr >= sys_addr - MACH64_DUMP_CONTEXT * 4 &&
		     addr <= sys_addr + MACH64_DUMP_CONTEXT * 4) ||
		    addr >=
		    GETBUFADDR(buf) + buf->used - MACH64_DUMP_CONTEXT * 4) {
			DRM_INFO("%08x:  0x%08x\n", addr, reg);
		}
		addr += 4;
		used--;

		count = (reg >> 16) + 1;
		reg = reg & 0xffff;
		reg = MMSELECT(reg);
		while (count && used) {
			if (addr <= GETBUFADDR(buf) + MACH64_DUMP_CONTEXT * 4 ||
			    (addr >= sys_addr - MACH64_DUMP_CONTEXT * 4 &&
			     addr <= sys_addr + MACH64_DUMP_CONTEXT * 4) ||
			    addr >=
			    GETBUFADDR(buf) + buf->used -
			    MACH64_DUMP_CONTEXT * 4) {
				DRM_INFO("%08x:    0x%04x = 0x%08x\n", addr,
					 reg, le32_to_cpu(*p));
				skipped = 0;
			} else {
				if (!skipped) {
					DRM_INFO("  ...\n");
					skipped = 1;
				}
			}
			p++;
			addr += 4;
			used--;

			reg += 4;
			count--;
		}
	}

	DRM_INFO("\n");
}

/**
 * Dump the ring state and contents, including the contents of the buffer being
 * processed by the graphics engine.
 */
void mach64_dump_ring_info(drm_mach64_private_t *dev_priv)
{
	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
	int i, skipped;

	DRM_INFO("\n");

	DRM_INFO("ring contents:\n");
	DRM_INFO("  head_addr: 0x%08x head: %u tail: %u\n\n",
		 ring->head_addr, ring->head, ring->tail);

	skipped = 0;
	for (i = 0; i < ring->size / sizeof(u32); i += 4) {
		if (i <= MACH64_DUMP_CONTEXT * 4 ||
		    i >= ring->size / sizeof(u32) - MACH64_DUMP_CONTEXT * 4 ||
		    (i >= ring->tail - MACH64_DUMP_CONTEXT * 4 &&
		     i <= ring->tail + MACH64_DUMP_CONTEXT * 4) ||
		    (i >= ring->head - MACH64_DUMP_CONTEXT * 4 &&
		     i <= ring->head + MACH64_DUMP_CONTEXT * 4)) {
			DRM_INFO("  0x%08x:  0x%08x 0x%08x 0x%08x 0x%08x%s%s\n",
				 (u32)(ring->start_addr + i * sizeof(u32)),
				 le32_to_cpu(((u32 *) ring->start)[i + 0]),
				 le32_to_cpu(((u32 *) ring->start)[i + 1]),
				 le32_to_cpu(((u32 *) ring->start)[i + 2]),
				 le32_to_cpu(((u32 *) ring->start)[i + 3]),
				 i == ring->head ? " (head)" : "",
				 i == ring->tail ? " (tail)" : "");
			skipped = 0;
		} else {
			if (!skipped) {
				DRM_INFO("  ...\n");
				skipped = 1;
			}
		}
	}

	DRM_INFO("\n");

	if (ring->head >= 0 && ring->head < ring->size / sizeof(u32)) {
		struct list_head *ptr;
		u32 addr = le32_to_cpu(((u32 *) ring->start)[ring->head + 1]);

		list_for_each(ptr, &dev_priv->pending) {
			drm_mach64_freelist_t *entry =
			    list_entry(ptr, drm_mach64_freelist_t, list);
			struct drm_buf *buf = entry->buf;

			u32 buf_addr = GETBUFADDR(buf);

			if (buf_addr <= addr && addr < buf_addr + buf->used)
				mach64_dump_buf_info(dev_priv, buf);
		}
	}

	DRM_INFO("\n");
	DRM_INFO("       BM_GUI_TABLE = 0x%08x\n",
		 MACH64_READ(MACH64_BM_GUI_TABLE));
	DRM_INFO("\n");
	DRM_INFO("BM_FRAME_BUF_OFFSET = 0x%08x\n",
		 MACH64_READ(MACH64_BM_FRAME_BUF_OFFSET));
	DRM_INFO(" BM_SYSTEM_MEM_ADDR = 0x%08x\n",
		 MACH64_READ(MACH64_BM_SYSTEM_MEM_ADDR));
	DRM_INFO("         BM_COMMAND = 0x%08x\n",
		 MACH64_READ(MACH64_BM_COMMAND));
	DRM_INFO("\n");
	DRM_INFO("          BM_STATUS = 0x%08x\n",
		 MACH64_READ(MACH64_BM_STATUS));
	DRM_INFO("           BUS_CNTL = 0x%08x\n",
		 MACH64_READ(MACH64_BUS_CNTL));
	DRM_INFO("          FIFO_STAT = 0x%08x\n",
		 MACH64_READ(MACH64_FIFO_STAT));
	DRM_INFO("           GUI_STAT = 0x%08x\n",
		 MACH64_READ(MACH64_GUI_STAT));
	DRM_INFO("           SRC_CNTL = 0x%08x\n",
		 MACH64_READ(MACH64_SRC_CNTL));
}

/*@}*/


/*******************************************************************/
/** \name DMA descriptor ring macros */
/*@{*/

/**
 * Add the end mark to the ring's new tail position.
 *
 * The bus master engine will keep processing the DMA buffers listed in the ring
 * until it finds this mark, making it stop.
 *
 * \sa mach64_clear_dma_eol
 */ 
static __inline__ void mach64_set_dma_eol(volatile u32 *addr)
{
#if defined(__i386__)
	int nr = 31;

	/* Taken from include/asm-i386/bitops.h linux header */
	__asm__ __volatile__("lock;" "btsl %1,%0":"=m"(*addr)
			     :"Ir"(nr));
#elif defined(__powerpc__)
	u32 old;
	u32 mask = cpu_to_le32(MACH64_DMA_EOL);

	/* Taken from the include/asm-ppc/bitops.h linux header */
	__asm__ __volatile__("\n\
1:	lwarx	%0,0,%3 \n\
	or	%0,%0,%2 \n\
	stwcx.	%0,0,%3 \n\
	bne-	1b":"=&r"(old), "=m"(*addr)
			     :"r"(mask), "r"(addr), "m"(*addr)
			     :"cc");
#elif defined(__alpha__)
	u32 temp;
	u32 mask = MACH64_DMA_EOL;

	/* Taken from the include/asm-alpha/bitops.h linux header */
	__asm__ __volatile__("1:	ldl_l %0,%3\n"
			     "	bis %0,%2,%0\n"
			     "	stl_c %0,%1\n"
			     "	beq %0,2f\n"
			     ".subsection 2\n"
			     "2:	br 1b\n"
			     ".previous":"=&r"(temp), "=m"(*addr)
			     :"Ir"(mask), "m"(*addr));
#else
	u32 mask = cpu_to_le32(MACH64_DMA_EOL);

	*addr |= mask;
#endif
}

/**
 * Remove the end mark from the ring's old tail position.
 *
 * It should be called after calling mach64_set_dma_eol to mark the ring's new
 * tail position.
 *
 * We update the end marks while the bus master engine is in operation. Since
 * the bus master engine may potentially be reading from the same position
 * that we write, we must change atomically to avoid having intermediary bad
 * data.
 */
static __inline__ void mach64_clear_dma_eol(volatile u32 *addr)
{
#if defined(__i386__)
	int nr = 31;

	/* Taken from include/asm-i386/bitops.h linux header */
	__asm__ __volatile__("lock;" "btrl %1,%0":"=m"(*addr)
			     :"Ir"(nr));
#elif defined(__powerpc__)
	u32 old;
	u32 mask = cpu_to_le32(MACH64_DMA_EOL);

	/* Taken from the include/asm-ppc/bitops.h linux header */
	__asm__ __volatile__("\n\
1:	lwarx	%0,0,%3 \n\
	andc	%0,%0,%2 \n\
	stwcx.	%0,0,%3 \n\
	bne-	1b":"=&r"(old), "=m"(*addr)
			     :"r"(mask), "r"(addr), "m"(*addr)
			     :"cc");
#elif defined(__alpha__)
	u32 temp;
	u32 mask = ~MACH64_DMA_EOL;

	/* Taken from the include/asm-alpha/bitops.h linux header */
	__asm__ __volatile__("1:	ldl_l %0,%3\n"
			     "	and %0,%2,%0\n"
			     "	stl_c %0,%1\n"
			     "	beq %0,2f\n"
			     ".subsection 2\n"
			     "2:	br 1b\n"
			     ".previous":"=&r"(temp), "=m"(*addr)
			     :"Ir"(mask), "m"(*addr));
#else
	u32 mask = cpu_to_le32(~MACH64_DMA_EOL);

	*addr &= mask;
#endif
}

#define RING_LOCALS							\
	int _ring_tail, _ring_write; unsigned int _ring_mask; volatile u32 *_ring

#define RING_WRITE_OFS  _ring_write

#define BEGIN_RING(n)							\
	do {								\
		if (MACH64_VERBOSE) {					\
			DRM_INFO( "BEGIN_RING( %d ) \n",		\
				  (n) );				\
		}							\
		if (dev_priv->ring.space <= (n) * sizeof(u32)) {	\
			int ret;					\
			if ((ret = mach64_wait_ring( dev_priv, (n) * sizeof(u32))) < 0 ) { \
				DRM_ERROR( "wait_ring failed, resetting engine\n"); \
				mach64_dump_engine_info( dev_priv );	\
				mach64_do_engine_reset( dev_priv );	\
				return ret;				\
			}						\
		}							\
		dev_priv->ring.space -= (n) * sizeof(u32);		\
		_ring = (u32 *) dev_priv->ring.start;			\
		_ring_tail = _ring_write = dev_priv->ring.tail;		\
		_ring_mask = dev_priv->ring.tail_mask;			\
	} while (0)

#define OUT_RING( x )						\
do {								\
	if (MACH64_VERBOSE) {					\
		DRM_INFO( "   OUT_RING( 0x%08x ) at 0x%x\n",	\
			   (unsigned int)(x), _ring_write );	\
	}							\
	_ring[_ring_write++] = cpu_to_le32( x );		\
	_ring_write &= _ring_mask;				\
} while (0)

#define ADVANCE_RING()							\
do {									\
	if (MACH64_VERBOSE) {						\
		DRM_INFO( "ADVANCE_RING() wr=0x%06x tail=0x%06x\n",	\
			  _ring_write, _ring_tail );			\
	}								\
	DRM_MEMORYBARRIER();						\
	mach64_clear_dma_eol( &_ring[(_ring_tail - 2) & _ring_mask] );	\
	DRM_MEMORYBARRIER();						\
	dev_priv->ring.tail = _ring_write;				\
	mach64_ring_tick( dev_priv, &(dev_priv)->ring );		\
} while (0)

/**
 * Queue a DMA buffer of registers writes into the ring buffer.
 */ 
int mach64_add_buf_to_ring(drm_mach64_private_t *dev_priv,
                           drm_mach64_freelist_t *entry)
{
	int bytes, pages, remainder;
	u32 address, page;
	int i;
	struct drm_buf *buf = entry->buf;
	RING_LOCALS;

	bytes = buf->used;
	address = GETBUFADDR( buf );
	pages = (bytes + MACH64_DMA_CHUNKSIZE - 1) / MACH64_DMA_CHUNKSIZE;

	BEGIN_RING( pages * 4 );

	for ( i = 0 ; i < pages-1 ; i++ ) {
		page = address + i * MACH64_DMA_CHUNKSIZE;
		OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_ADDR );
		OUT_RING( page );
		OUT_RING( MACH64_DMA_CHUNKSIZE | MACH64_DMA_HOLD_OFFSET );
		OUT_RING( 0 );
	}

	/* generate the final descriptor for any remaining commands in this buffer */
	page = address + i * MACH64_DMA_CHUNKSIZE;
	remainder = bytes - i * MACH64_DMA_CHUNKSIZE;

	/* Save dword offset of last descriptor for this buffer.
	 * This is needed to check for completion of the buffer in freelist_get
	 */
	entry->ring_ofs = RING_WRITE_OFS;

	OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_ADDR );
	OUT_RING( page );
	OUT_RING( remainder | MACH64_DMA_HOLD_OFFSET | MACH64_DMA_EOL );
	OUT_RING( 0 );

	ADVANCE_RING();
	
	return 0;
}

/**
 * Queue DMA buffer controlling host data tranfers (e.g., blit).
 * 
 * Almost identical to mach64_add_buf_to_ring.
 */
int mach64_add_hostdata_buf_to_ring(drm_mach64_private_t *dev_priv,
                                    drm_mach64_freelist_t *entry)
{
	int bytes, pages, remainder;
	u32 address, page;
	int i;
	struct drm_buf *buf = entry->buf;
	RING_LOCALS;
	
	bytes = buf->used - MACH64_HOSTDATA_BLIT_OFFSET;
	pages = (bytes + MACH64_DMA_CHUNKSIZE - 1) / MACH64_DMA_CHUNKSIZE;
	address = GETBUFADDR( buf );
	
	BEGIN_RING( 4 + pages * 4 );
	
	OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_ADDR );
	OUT_RING( address );
	OUT_RING( MACH64_HOSTDATA_BLIT_OFFSET | MACH64_DMA_HOLD_OFFSET );
	OUT_RING( 0 );
	address += MACH64_HOSTDATA_BLIT_OFFSET;
	
	for ( i = 0 ; i < pages-1 ; i++ ) {
		page = address + i * MACH64_DMA_CHUNKSIZE;
		OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_HOSTDATA );
		OUT_RING( page );
		OUT_RING( MACH64_DMA_CHUNKSIZE | MACH64_DMA_HOLD_OFFSET );
		OUT_RING( 0 );
	}
	
	/* generate the final descriptor for any remaining commands in this buffer */
	page = address + i * MACH64_DMA_CHUNKSIZE;
	remainder = bytes - i * MACH64_DMA_CHUNKSIZE;
	
	/* Save dword offset of last descriptor for this buffer.
	 * This is needed to check for completion of the buffer in freelist_get
	 */
	entry->ring_ofs = RING_WRITE_OFS;
	
	OUT_RING( MACH64_APERTURE_OFFSET + MACH64_BM_HOSTDATA );
	OUT_RING( page );
	OUT_RING( remainder | MACH64_DMA_HOLD_OFFSET | MACH64_DMA_EOL );
	OUT_RING( 0 );
	
	ADVANCE_RING();
	
	return 0;
}

/*@}*/


/*******************************************************************/
/** \name DMA test and initialization */
/*@{*/

/**
 * Perform a simple DMA operation using the pattern registers to test whether
 * DMA works.
 *
 * \return zero if successful.
 *
 * \note This function was the testbed for many experiences regarding Mach64
 * DMA operation. It is left here since it so tricky to get DMA operating
 * properly in some architectures and hardware.
 */
static int mach64_bm_dma_test(struct drm_device * dev)
{
	drm_mach64_private_t *dev_priv = dev->dev_private;
	drm_dma_handle_t *cpu_addr_dmah;
	u32 data_addr;
	u32 *table, *data;
	u32 expected[2];
	u32 src_cntl, pat_reg0, pat_reg1;
	int i, count, failed;

	DRM_DEBUG("\n");

	table = (u32 *) dev_priv->ring.start;

	/* FIXME: get a dma buffer from the freelist here */
	DRM_DEBUG("Allocating data memory ...\n");
#ifdef __FreeBSD__
	DRM_UNLOCK();
#endif
	cpu_addr_dmah =
	    drm_pci_alloc(dev, 0x1000, 0x1000, 0xfffffffful);
#ifdef __FreeBSD__
	DRM_LOCK();
#endif
	if (!cpu_addr_dmah) {
		DRM_INFO("data-memory allocation failed!\n");
		return -ENOMEM;
	} else {
		data = (u32 *) cpu_addr_dmah->vaddr;
		data_addr = (u32) cpu_addr_dmah->busaddr;
	}

	/* Save the X server's value for SRC_CNTL and restore it
	 * in case our test fails.  This prevents the X server
	 * from disabling it's cache for this register
	 */
	src_cntl = MACH64_READ(MACH64_SRC_CNTL);
	pat_reg0 = MACH64_READ(MACH64_PAT_REG0);
	pat_reg1 = MACH64_READ(MACH64_PAT_REG1);

	mach64_do_wait_for_fifo(dev_priv, 3);

	MACH64_WRITE(MACH64_SRC_CNTL, 0);
	MACH64_WRITE(MACH64_PAT_REG0, 0x11111111);
	MACH64_WRITE(MACH64_PAT_REG1, 0x11111111);

	mach64_do_wait_for_idle(dev_priv);

	for (i = 0; i < 2; i++) {
		u32 reg;
		reg = MACH64_READ((MACH64_PAT_REG0 + i * 4));
		DRM_DEBUG("(Before DMA Transfer) reg %d = 0x%08x\n", i, reg);
		if (reg != 0x11111111) {
			DRM_INFO("Error initializing test registers\n");
			DRM_INFO("resetting engine ...\n");
			mach64_do_engine_reset(dev_priv);
			DRM_INFO("freeing data buffer memory.\n");
			drm_pci_free(dev, cpu_addr_dmah);
			return -EIO;
		}
	}

	/* fill up a buffer with sets of 2 consecutive writes starting with PAT_REG0 */
	count = 0;

	data[count++] = cpu_to_le32(DMAREG(MACH64_PAT_REG0) | (1 << 16));
	data[count++] = expected[0] = 0x22222222;
	data[count++] = expected[1] = 0xaaaaaaaa;

	while (count < 1020) {
		data[count++] =
		    cpu_to_le32(DMAREG(MACH64_PAT_REG0) | (1 << 16));
		data[count++] = 0x22222222;
		data[count++] = 0xaaaaaaaa;
	}
	data[count++] = cpu_to_le32(DMAREG(MACH64_SRC_CNTL) | (0 << 16));
	data[count++] = 0;

	DRM_DEBUG("Preparing table ...\n");
	table[MACH64_DMA_FRAME_BUF_OFFSET] = cpu_to_le32(MACH64_BM_ADDR +
							 MACH64_APERTURE_OFFSET);
	table[MACH64_DMA_SYS_MEM_ADDR] = cpu_to_le32(data_addr);
	table[MACH64_DMA_COMMAND] = cpu_to_le32(count * sizeof(u32)
						| MACH64_DMA_HOLD_OFFSET
						| MACH64_DMA_EOL);
	table[MACH64_DMA_RESERVED] = 0;

	DRM_DEBUG("table[0] = 0x%08x\n", table[0]);
	DRM_DEBUG("table[1] = 0x%08x\n", table[1]);
	DRM_DEBUG("table[2] = 0x%08x\n", table[2]);
	DRM_DEBUG("table[3] = 0x%08x\n", table[3]);

	for (i = 0; i < 6; i++) {
		DRM_DEBUG(" data[%d] = 0x%08x\n", i, data[i]);
	}
	DRM_DEBUG(" ...\n");
	for (i = count - 5; i < count; i++) {
		DRM_DEBUG(" data[%d] = 0x%08x\n", i, data[i]);
	}

	DRM_MEMORYBARRIER();

	DRM_DEBUG("waiting for idle...\n");
	if ((i = mach64_do_wait_for_idle(dev_priv))) {
		DRM_INFO("mach64_do_wait_for_idle failed (result=%d)\n", i);
		DRM_INFO("resetting engine ...\n");
		mach64_do_engine_reset(dev_priv);
		mach64_do_wait_for_fifo(dev_priv, 3);
		MACH64_WRITE(MACH64_SRC_CNTL, src_cntl);
		MACH64_WRITE(MACH64_PAT_REG0, pat_reg0);
		MACH64_WRITE(MACH64_PAT_REG1, pat_reg1);
		DRM_INFO("freeing data buffer memory.\n");
		drm_pci_free(dev, cpu_addr_dmah);
		return i;
	}
	DRM_DEBUG("waiting for idle...done\n");

	DRM_DEBUG("BUS_CNTL = 0x%08x\n", MACH64_READ(MACH64_BUS_CNTL));
	DRM_DEBUG("SRC_CNTL = 0x%08x\n", MACH64_READ(MACH64_SRC_CNTL));
	DRM_DEBUG("\n");
	DRM_DEBUG("data bus addr = 0x%08x\n", data_addr);
	DRM_DEBUG("table bus addr = 0x%08x\n", dev_priv->ring.start_addr);

	DRM_DEBUG("starting DMA transfer...\n");
	MACH64_WRITE(MACH64_BM_GUI_TABLE_CMD,
		     dev_priv->ring.start_addr | MACH64_CIRCULAR_BUF_SIZE_16KB);

	MACH64_WRITE(MACH64_SRC_CNTL,
		     MACH64_SRC_BM_ENABLE | MACH64_SRC_BM_SYNC |
		     MACH64_SRC_BM_OP_SYSTEM_TO_REG);

	/* Kick off the transfer */
	DRM_DEBUG("starting DMA transfer... done.\n");
	MACH64_WRITE(MACH64_DST_HEIGHT_WIDTH, 0);

	DRM_DEBUG("waiting for idle...\n");

	if ((i = mach64_do_wait_for_idle(dev_priv))) {
		/* engine locked up, dump register state and reset */
		DRM_INFO("mach64_do_wait_for_idle failed (result=%d)\n", i);
		mach64_dump_engine_info(dev_priv);
		DRM_INFO("resetting engine ...\n");
		mach64_do_engine_reset(dev_priv);
		mach64_do_wait_for_fifo(dev_priv, 3);
		MACH64_WRITE(MACH64_SRC_CNTL, src_cntl);
		MACH64_WRITE(MACH64_PAT_REG0, pat_reg0);
		MACH64_WRITE(MACH64_PAT_REG1, pat_reg1);
		DRM_INFO("freeing data buffer memory.\n");
		drm_pci_free(dev, cpu_addr_dmah);
		return i;
	}

	DRM_DEBUG("waiting for idle...done\n");

	/* restore SRC_CNTL */
	mach64_do_wait_for_fifo(dev_priv, 1);
	MACH64_WRITE(MACH64_SRC_CNTL, src_cntl);

	failed = 0;

	/* Check register values to see if the GUI master operation succeeded */
	for (i = 0; i < 2; i++) {
		u32 reg;
		reg = MACH64_READ((MACH64_PAT_REG0 + i * 4));
		DRM_DEBUG("(After DMA Transfer) reg %d = 0x%08x\n", i, reg);
		if (reg != expected[i]) {
			failed = -1;
		}
	}

	/* restore pattern registers */
	mach64_do_wait_for_fifo(dev_priv, 2);
	MACH64_WRITE(MACH64_PAT_REG0, pat_reg0);
	MACH64_WRITE(MACH64_PAT_REG1, pat_reg1);

	DRM_DEBUG("freeing data buffer memory.\n");
	drm_pci_free(dev, cpu_addr_dmah);
	DRM_DEBUG("returning ...\n");

	return failed;
}

/**
 * Called during the DMA initialization ioctl to initialize all the necessary
 * software and hardware state for DMA operation.
 */
static int mach64_do_dma_init(struct drm_device * dev, drm_mach64_init_t * init)
{
	drm_mach64_private_t *dev_priv;
	u32 tmp;
	int i, ret;

	DRM_DEBUG("\n");

	dev_priv = drm_alloc(sizeof(drm_mach64_private_t), DRM_MEM_DRIVER);
	if (dev_priv == NULL)
		return -ENOMEM;

	memset(dev_priv, 0, sizeof(drm_mach64_private_t));

	dev_priv->is_pci = init->is_pci;

	dev_priv->fb_bpp = init->fb_bpp;
	dev_priv->front_offset = init->front_offset;
	dev_priv->front_pitch = init->front_pitch;
	dev_priv->back_offset = init->back_offset;
	dev_priv->back_pitch = init->back_pitch;

	dev_priv->depth_bpp = init->depth_bpp;
	dev_priv->depth_offset = init->depth_offset;
	dev_priv->depth_pitch = init->depth_pitch;

	dev_priv->front_offset_pitch = (((dev_priv->front_pitch / 8) << 22) |
					(dev_priv->front_offset >> 3));
	dev_priv->back_offset_pitch = (((dev_priv->back_pitch / 8) << 22) |
				       (dev_priv->back_offset >> 3));
	dev_priv->depth_offset_pitch = (((dev_priv->depth_pitch / 8) << 22) |
					(dev_priv->depth_offset >> 3));

	dev_priv->usec_timeout = 1000000;

	/* Set up the freelist, placeholder list and pending list */
	INIT_LIST_HEAD(&dev_priv->free_list);
	INIT_LIST_HEAD(&dev_priv->placeholders);
	INIT_LIST_HEAD(&dev_priv->pending);

	dev_priv->sarea = drm_getsarea(dev);
	if (!dev_priv->sarea) {
		DRM_ERROR("can not find sarea!\n");
		dev->dev_private = (void *)dev_priv;
		mach64_do_cleanup_dma(dev);
		return -EINVAL;
	}
	dev_priv->fb = drm_core_findmap(dev, init->fb_offset);
	if (!dev_priv->fb) {
		DRM_ERROR("can not find frame buffer map!\n");
		dev->dev_private = (void *)dev_priv;
		mach64_do_cleanup_dma(dev);
		return -EINVAL;
	}
	dev_priv->mmio = drm_core_findmap(dev, init->mmio_offset);
	if (!dev_priv->mmio) {
		DRM_ERROR("can not find mmio map!\n");
		dev->dev_private = (void *)dev_priv;
		mach64_do_cleanup_dma(dev);
		return -EINVAL;
	}

	dev_priv->ring_map = drm_core_findmap(dev, init->ring_offset);
	if (!dev_priv->ring_map) {
		DRM_ERROR("can not find ring map!\n");
		dev->dev_private = (void *)dev_priv;
		mach64_do_cleanup_dma(dev);
		return -EINVAL;
	}

	dev_priv->sarea_priv = (drm_mach64_sarea_t *)
	    ((u8 *) dev_priv->sarea->handle + init->sarea_priv_offset);

	if (!dev_priv->is_pci) {
		drm_core_ioremap(dev_priv->ring_map, dev);
		if (!dev_priv->ring_map->handle) {
			DRM_ERROR("can not ioremap virtual address for"
				  " descriptor ring\n");
			dev->dev_private = (void *)dev_priv;
			mach64_do_cleanup_dma(dev);
			return -ENOMEM;
		}
		dev->agp_buffer_token = init->buffers_offset;
		dev->agp_buffer_map =
		    drm_core_findmap(dev, init->buffers_offset);
		if (!dev->agp_buffer_map) {
			DRM_ERROR("can not find dma buffer map!\n");
			dev->dev_private = (void *)dev_priv;
			mach64_do_cleanup_dma(dev);
			return -EINVAL;
		}
		/* there might be a nicer way to do this -
		   dev isn't passed all the way though the mach64 - DA */
		dev_priv->dev_buffers = dev->agp_buffer_map;

		drm_core_ioremap(dev->agp_buffer_map, dev);
		if (!dev->agp_buffer_map->handle) {
			DRM_ERROR("can not ioremap virtual address for"
				  " dma buffer\n");
			dev->dev_private = (void *)dev_priv;
			mach64_do_cleanup_dma(dev);
			return -ENOMEM;
		}
		dev_priv->agp_textures =
		    drm_core_findmap(dev, init->agp_textures_offset);
		if (!dev_priv->agp_textures) {
			DRM_ERROR("can not find agp texture region!\n");
			dev->dev_private = (void *)dev_priv;
			mach64_do_cleanup_dma(dev);
			return -EINVAL;
		}
	}

	dev->dev_private = (void *)dev_priv;

	dev_priv->driver_mode = init->dma_mode;

	/* changing the FIFO size from the default causes problems with DMA */
	tmp = MACH64_READ(MACH64_GUI_CNTL);
	if ((tmp & MACH64_CMDFIFO_SIZE_MASK) != MACH64_CMDFIFO_SIZE_128) {
		DRM_INFO("Setting FIFO size to 128 entries\n");
		/* FIFO must be empty to change the FIFO depth */
		if ((ret = mach64_do_wait_for_idle(dev_priv))) {
			DRM_ERROR
			    ("wait for idle failed before changing FIFO depth!\n");
			mach64_do_cleanup_dma(dev);
			return ret;
		}
		MACH64_WRITE(MACH64_GUI_CNTL, ((tmp & ~MACH64_CMDFIFO_SIZE_MASK)
					       | MACH64_CMDFIFO_SIZE_128));
		/* need to read GUI_STAT for proper sync according to docs */
		if ((ret = mach64_do_wait_for_idle(dev_priv))) {
			DRM_ERROR
			    ("wait for idle failed when changing FIFO depth!\n");
			mach64_do_cleanup_dma(dev);
			return ret;
		}
	}

	dev_priv->ring.size = 0x4000;	/* 16KB */
	dev_priv->ring.start = dev_priv->ring_map->handle;
	dev_priv->ring.start_addr = (u32) dev_priv->ring_map->offset;

	memset(dev_priv->ring.start, 0, dev_priv->ring.size);
	DRM_INFO("descriptor ring: cpu addr %p, bus addr: 0x%08x\n",
		 dev_priv->ring.start, dev_priv->ring.start_addr);

	ret = 0;
	if (dev_priv->driver_mode != MACH64_MODE_MMIO) {

		/* enable block 1 registers and bus mastering */
		MACH64_WRITE(MACH64_BUS_CNTL, ((MACH64_READ(MACH64_BUS_CNTL)
						| MACH64_BUS_EXT_REG_EN)
					       & ~MACH64_BUS_MASTER_DIS));

		/* try a DMA GUI-mastering pass and fall back to MMIO if it fails */
		DRM_DEBUG("Starting DMA test...\n");
		if ((ret = mach64_bm_dma_test(dev))) {
			dev_priv->driver_mode = MACH64_MODE_MMIO;
		}
	}

	switch (dev_priv->driver_mode) {
	case MACH64_MODE_MMIO:
		MACH64_WRITE(MACH64_BUS_CNTL, (MACH64_READ(MACH64_BUS_CNTL)
					       | MACH64_BUS_EXT_REG_EN
					       | MACH64_BUS_MASTER_DIS));
		if (init->dma_mode == MACH64_MODE_MMIO)
			DRM_INFO("Forcing pseudo-DMA mode\n");
		else
			DRM_INFO
			    ("DMA test failed (ret=%d), using pseudo-DMA mode\n",
			     ret);
		break;
	case MACH64_MODE_DMA_SYNC:
		DRM_INFO("DMA test succeeded, using synchronous DMA mode\n");
		break;
	case MACH64_MODE_DMA_ASYNC:
	default:
		DRM_INFO("DMA test succeeded, using asynchronous DMA mode\n");
	}

	dev_priv->ring_running = 0;

	/* setup offsets for physical address of table start and end */
	dev_priv->ring.head_addr = dev_priv->ring.start_addr;
	dev_priv->ring.head = dev_priv->ring.tail = 0;
	dev_priv->ring.tail_mask = (dev_priv->ring.size / sizeof(u32)) - 1;
	dev_priv->ring.space = dev_priv->ring.size;

	/* setup physical address and size of descriptor table */
	mach64_do_wait_for_fifo(dev_priv, 1);
	MACH64_WRITE(MACH64_BM_GUI_TABLE_CMD,
		     (dev_priv->ring.
		      head_addr | MACH64_CIRCULAR_BUF_SIZE_16KB));

	/* init frame counter */
	dev_priv->sarea_priv->frames_queued = 0;
	for (i = 0; i < MACH64_MAX_QUEUED_FRAMES; i++) {
		dev_priv->frame_ofs[i] = ~0;	/* All ones indicates placeholder */
	}

	/* Allocate the DMA buffer freelist */
	if ((ret = mach64_init_freelist(dev))) {
		DRM_ERROR("Freelist allocation failed\n");
		mach64_do_cleanup_dma(dev);
		return ret;
	}

	return 0;
}

/*******************************************************************/
/** MMIO Pseudo-DMA (intended primarily for debugging, not performance)
 */

int mach64_do_dispatch_pseudo_dma(drm_mach64_private_t *dev_priv)
{
	drm_mach64_descriptor_ring_t *ring = &dev_priv->ring;
	volatile u32 *ring_read;
	struct list_head *ptr;
	drm_mach64_freelist_t *entry;
	struct drm_buf *buf = NULL;
	u32 *buf_ptr;
	u32 used, reg, target;
	int fifo, count, found, ret, no_idle_wait;

	fifo = count = reg = no_idle_wait = 0;
	target = MACH64_BM_ADDR;

	if ((ret = mach64_do_wait_for_idle(dev_priv)) < 0) {
		DRM_INFO("idle failed before pseudo-dma dispatch, resetting engine\n");
		mach64_dump_engine_info(dev_priv);
		mach64_do_engine_reset(dev_priv);
		return ret;
	}

	ring_read = (u32 *) ring->start;

	while (ring->tail != ring->head) {
		u32 buf_addr, new_target, offset;
		u32 bytes, remaining, head, eol;

		head = ring->head;

		new_target =
		    le32_to_cpu(ring_read[head++]) - MACH64_APERTURE_OFFSET;
		buf_addr = le32_to_cpu(ring_read[head++]);
		eol = le32_to_cpu(ring_read[head]) & MACH64_DMA_EOL;
		bytes = le32_to_cpu(ring_read[head++])
		    & ~(MACH64_DMA_HOLD_OFFSET | MACH64_DMA_EOL);
		head++;
		head &= ring->tail_mask;

		/* can't wait for idle between a blit setup descriptor
		 * and a HOSTDATA descriptor or the engine will lock
		 */
		if (new_target == MACH64_BM_HOSTDATA
		    && target == MACH64_BM_ADDR)
			no_idle_wait = 1;

		target = new_target;

		found = 0;
		offset = 0;
		list_for_each(ptr, &dev_priv->pending) {
			entry = list_entry(ptr, drm_mach64_freelist_t, list);
			buf = entry->buf;
			offset = buf_addr - GETBUFADDR(buf);
			if (offset >= 0 && offset < MACH64_BUFFER_SIZE) {
				found = 1;
				break;
			}
		}

		if (!found || buf == NULL) {
			DRM_ERROR
			    ("Couldn't find pending buffer: head: %u tail: %u buf_addr: 0x%08x %s\n",
			     head, ring->tail, buf_addr, (eol ? "eol" : ""));
			mach64_dump_ring_info(dev_priv);
			mach64_do_engine_reset(dev_priv);
			return -EINVAL;
		}

		/* Hand feed the buffer to the card via MMIO, waiting for the fifo
		 * every 16 writes
		 */
		DRM_DEBUG("target: (0x%08x) %s\n", target,
			  (target ==
			   MACH64_BM_HOSTDATA ? "BM_HOSTDATA" : "BM_ADDR"));
		DRM_DEBUG("offset: %u bytes: %u used: %u\n", offset, bytes,
			  buf->used);

		remaining = (buf->used - offset) >> 2;	/* dwords remaining in buffer */
		used = bytes >> 2;	/* dwords in buffer for this descriptor */
		buf_ptr = (u32 *) ((char *)GETBUFPTR(buf) + offset);

		while (used) {

			if (count == 0) {
				if (target == MACH64_BM_HOSTDATA) {
					reg = DMAREG(MACH64_HOST_DATA0);
					count =
					    (remaining > 16) ? 16 : remaining;
					fifo = 0;
				} else {
					reg = le32_to_cpu(*buf_ptr++);
					used--;
					count = (reg >> 16) + 1;
				}

				reg = reg & 0xffff;
				reg = MMSELECT(reg);
			}
			while (count && used) {
				if (!fifo) {
					if (no_idle_wait) {
						if ((ret =
						     mach64_do_wait_for_fifo
						     (dev_priv, 16)) < 0) {
							no_idle_wait = 0;
							return ret;
						}
					} else {
						if ((ret =
						     mach64_do_wait_for_idle
						     (dev_priv)) < 0) {
							return ret;
						}
					}
					fifo = 16;
				}
				--fifo;
				MACH64_WRITE(reg, le32_to_cpu(*buf_ptr++));
				used--;
				remaining--;

				reg += 4;
				count--;
			}
		}
		ring->head = head;
		ring->head_addr = ring->start_addr + (ring->head * sizeof(u32));
		ring->space += (4 * sizeof(u32));
	}

	if ((ret = mach64_do_wait_for_idle(dev_priv)) < 0) {
		return ret;
	}
	MACH64_WRITE(MACH64_BM_GUI_TABLE_CMD,
		     ring->head_addr | MACH64_CIRCULAR_BUF_SIZE_16KB);

	DRM_DEBUG("completed\n");
	return 0;
}

/*@}*/


/*******************************************************************/
/** \name DMA cleanup */
/*@{*/

int mach64_do_cleanup_dma(struct drm_device * dev)
{
	DRM_DEBUG("\n");

	/* Make sure interrupts are disabled here because the uninstall ioctl
	 * may not have been called from userspace and after dev_private
	 * is freed, it's too late.
	 */
	if (dev->irq)
		drm_irq_uninstall(dev);

	if (dev->dev_private) {
		drm_mach64_private_t *dev_priv = dev->dev_private;

		if (!dev_priv->is_pci) {
			if (dev_priv->ring_map)
				drm_core_ioremapfree(dev_priv->ring_map, dev);

			if (dev->agp_buffer_map) {
				drm_core_ioremapfree(dev->agp_buffer_map, dev);
				dev->agp_buffer_map = NULL;
			}
		}

		mach64_destroy_freelist(dev);

		drm_free(dev_priv, sizeof(drm_mach64_private_t),
			 DRM_MEM_DRIVER);
		dev->dev_private = NULL;
	}

	return 0;
}

/*@}*/


/*******************************************************************/
/** \name IOCTL handlers */
/*@{*/

int mach64_dma_init(struct drm_device *dev, void *data,
		    struct drm_file *file_priv)
{
	drm_mach64_init_t *init = data;

	DRM_DEBUG("\n");

	LOCK_TEST_WITH_RETURN(dev, file_priv);

	switch (init->func) {
	case DRM_MACH64_INIT_DMA:
		return mach64_do_dma_init(dev, init);
	case DRM_MACH64_CLEANUP_DMA:
		return mach64_do_cleanup_dma(dev);
	}

	return -EINVAL;
}

int mach64_dma_idle(struct drm_device *dev, void *data,
		    struct drm_file *file_priv)
{
	drm_mach64_private_t *dev_priv = dev->dev_private;

	DRM_DEBUG("\n");

	LOCK_TEST_WITH_RETURN(dev, file_priv);

	return mach64_do_dma_idle(dev_priv);
}

int mach64_dma_flush(struct drm_device *dev, void *data,
		     struct drm_file *file_priv)
{
	drm_mach64_private_t *dev_priv = dev->dev_private;

	DRM_DEBUG("\n");

	LOCK_TEST_WITH_RETURN(dev, file_priv);

	return mach64_do_dma_flush(dev_priv);
}

int mach64_engine_reset(struct drm_device *dev, void *data,
			struct drm_file *file_priv)
{
	drm_mach64_private_t *dev_priv = dev->dev_private;

	DRM_DEBUG("\n");

	LOCK_TEST_WITH_RETURN(dev, file_priv);

	return mach64_do_engine_reset(dev_priv);
}

/*@}*/


/*******************************************************************/
/** \name Freelist management */
/*@{*/

int mach64_init_freelist(struct drm_device * dev)
{
	struct drm_device_dma *dma = dev->dma;
	drm_mach64_private_t *dev_priv = dev->dev_private;
	drm_mach64_freelist_t *entry;
	struct list_head *ptr;
	int i;

	DRM_DEBUG("adding %d buffers to freelist\n", dma->buf_count);

	for (i = 0; i < dma->buf_count; i++) {
		if ((entry =
		     (drm_mach64_freelist_t *)
		     drm_alloc(sizeof(drm_mach64_freelist_t),
			       DRM_MEM_BUFLISTS)) == NULL)
			return -ENOMEM;
		memset(entry, 0, sizeof(drm_mach64_freelist_t));
		entry->buf = dma->buflist[i];
		ptr = &entry->list;
		list_add_tail(ptr, &dev_priv->free_list);
	}

	return 0;
}

void mach64_destroy_freelist(struct drm_device * dev)
{
	drm_mach64_private_t *dev_priv = dev->dev_private;
	drm_mach64_freelist_t *entry;
	struct list_head *ptr;
	struct list_head *tmp;

	DRM_DEBUG("\n");

	list_for_each_safe(ptr, tmp, &dev_priv->pending) {
		list_del(ptr);
		entry = list_entry(ptr, drm_mach64_freelist_t, list);