Linux Audio

Check our new training course

Loading...
  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 *          Christian König
 28 */
 
 
 29#include <drm/drmP.h>
 
 
 30#include "radeon.h"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 31
 32/*
 33 * Rings
 34 * Most engines on the GPU are fed via ring buffers.  Ring
 35 * buffers are areas of GPU accessible memory that the host
 36 * writes commands into and the GPU reads commands out of.
 37 * There is a rptr (read pointer) that determines where the
 38 * GPU is currently reading, and a wptr (write pointer)
 39 * which determines where the host has written.  When the
 40 * pointers are equal, the ring is idle.  When the host
 41 * writes commands to the ring buffer, it increments the
 42 * wptr.  The GPU then starts fetching commands and executes
 43 * them until the pointers are equal again.
 44 */
 45static int radeon_debugfs_ring_init(struct radeon_device *rdev, struct radeon_ring *ring);
 46
 47/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 48 * radeon_ring_supports_scratch_reg - check if the ring supports
 49 * writing to scratch registers
 50 *
 51 * @rdev: radeon_device pointer
 52 * @ring: radeon_ring structure holding ring information
 53 *
 54 * Check if a specific ring supports writing to scratch registers (all asics).
 55 * Returns true if the ring supports writing to scratch regs, false if not.
 56 */
 57bool radeon_ring_supports_scratch_reg(struct radeon_device *rdev,
 58				      struct radeon_ring *ring)
 59{
 60	switch (ring->idx) {
 61	case RADEON_RING_TYPE_GFX_INDEX:
 62	case CAYMAN_RING_TYPE_CP1_INDEX:
 63	case CAYMAN_RING_TYPE_CP2_INDEX:
 64		return true;
 65	default:
 66		return false;
 67	}
 68}
 69
 70/**
 71 * radeon_ring_free_size - update the free size
 72 *
 73 * @rdev: radeon_device pointer
 74 * @ring: radeon_ring structure holding ring information
 75 *
 76 * Update the free dw slots in the ring buffer (all asics).
 77 */
 78void radeon_ring_free_size(struct radeon_device *rdev, struct radeon_ring *ring)
 79{
 80	uint32_t rptr = radeon_ring_get_rptr(rdev, ring);
 81
 82	/* This works because ring_size is a power of 2 */
 83	ring->ring_free_dw = rptr + (ring->ring_size / 4);
 84	ring->ring_free_dw -= ring->wptr;
 85	ring->ring_free_dw &= ring->ptr_mask;
 86	if (!ring->ring_free_dw) {
 87		/* this is an empty ring */
 88		ring->ring_free_dw = ring->ring_size / 4;
 89		/*  update lockup info to avoid false positive */
 90		radeon_ring_lockup_update(rdev, ring);
 91	}
 92}
 93
 94/**
 95 * radeon_ring_alloc - allocate space on the ring buffer
 96 *
 97 * @rdev: radeon_device pointer
 98 * @ring: radeon_ring structure holding ring information
 99 * @ndw: number of dwords to allocate in the ring buffer
100 *
101 * Allocate @ndw dwords in the ring buffer (all asics).
102 * Returns 0 on success, error on failure.
103 */
104int radeon_ring_alloc(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ndw)
105{
106	int r;
107
108	/* make sure we aren't trying to allocate more space than there is on the ring */
109	if (ndw > (ring->ring_size / 4))
110		return -ENOMEM;
111	/* Align requested size with padding so unlock_commit can
112	 * pad safely */
113	radeon_ring_free_size(rdev, ring);
114	ndw = (ndw + ring->align_mask) & ~ring->align_mask;
115	while (ndw > (ring->ring_free_dw - 1)) {
116		radeon_ring_free_size(rdev, ring);
117		if (ndw < ring->ring_free_dw) {
118			break;
119		}
120		r = radeon_fence_wait_next(rdev, ring->idx);
121		if (r)
122			return r;
123	}
124	ring->count_dw = ndw;
125	ring->wptr_old = ring->wptr;
126	return 0;
127}
128
129/**
130 * radeon_ring_lock - lock the ring and allocate space on it
131 *
132 * @rdev: radeon_device pointer
133 * @ring: radeon_ring structure holding ring information
134 * @ndw: number of dwords to allocate in the ring buffer
135 *
136 * Lock the ring and allocate @ndw dwords in the ring buffer
137 * (all asics).
138 * Returns 0 on success, error on failure.
139 */
140int radeon_ring_lock(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ndw)
141{
142	int r;
143
144	mutex_lock(&rdev->ring_lock);
145	r = radeon_ring_alloc(rdev, ring, ndw);
146	if (r) {
147		mutex_unlock(&rdev->ring_lock);
148		return r;
149	}
150	return 0;
151}
152
153/**
154 * radeon_ring_commit - tell the GPU to execute the new
155 * commands on the ring buffer
156 *
157 * @rdev: radeon_device pointer
158 * @ring: radeon_ring structure holding ring information
159 * @hdp_flush: Whether or not to perform an HDP cache flush
160 *
161 * Update the wptr (write pointer) to tell the GPU to
162 * execute new commands on the ring buffer (all asics).
163 */
164void radeon_ring_commit(struct radeon_device *rdev, struct radeon_ring *ring,
165			bool hdp_flush)
166{
167	/* If we are emitting the HDP flush via the ring buffer, we need to
168	 * do it before padding.
169	 */
170	if (hdp_flush && rdev->asic->ring[ring->idx]->hdp_flush)
171		rdev->asic->ring[ring->idx]->hdp_flush(rdev, ring);
172	/* We pad to match fetch size */
173	while (ring->wptr & ring->align_mask) {
174		radeon_ring_write(ring, ring->nop);
175	}
176	mb();
177	/* If we are emitting the HDP flush via MMIO, we need to do it after
178	 * all CPU writes to VRAM finished.
179	 */
180	if (hdp_flush && rdev->asic->mmio_hdp_flush)
181		rdev->asic->mmio_hdp_flush(rdev);
182	radeon_ring_set_wptr(rdev, ring);
183}
184
185/**
186 * radeon_ring_unlock_commit - tell the GPU to execute the new
187 * commands on the ring buffer and unlock it
188 *
189 * @rdev: radeon_device pointer
190 * @ring: radeon_ring structure holding ring information
191 * @hdp_flush: Whether or not to perform an HDP cache flush
192 *
193 * Call radeon_ring_commit() then unlock the ring (all asics).
194 */
195void radeon_ring_unlock_commit(struct radeon_device *rdev, struct radeon_ring *ring,
196			       bool hdp_flush)
197{
198	radeon_ring_commit(rdev, ring, hdp_flush);
199	mutex_unlock(&rdev->ring_lock);
200}
201
202/**
203 * radeon_ring_undo - reset the wptr
204 *
205 * @ring: radeon_ring structure holding ring information
206 *
207 * Reset the driver's copy of the wptr (all asics).
208 */
209void radeon_ring_undo(struct radeon_ring *ring)
210{
211	ring->wptr = ring->wptr_old;
212}
213
214/**
215 * radeon_ring_unlock_undo - reset the wptr and unlock the ring
216 *
217 * @ring: radeon_ring structure holding ring information
218 *
219 * Call radeon_ring_undo() then unlock the ring (all asics).
220 */
221void radeon_ring_unlock_undo(struct radeon_device *rdev, struct radeon_ring *ring)
222{
223	radeon_ring_undo(ring);
224	mutex_unlock(&rdev->ring_lock);
225}
226
227/**
228 * radeon_ring_lockup_update - update lockup variables
229 *
230 * @ring: radeon_ring structure holding ring information
231 *
232 * Update the last rptr value and timestamp (all asics).
233 */
234void radeon_ring_lockup_update(struct radeon_device *rdev,
235			       struct radeon_ring *ring)
236{
237	atomic_set(&ring->last_rptr, radeon_ring_get_rptr(rdev, ring));
238	atomic64_set(&ring->last_activity, jiffies_64);
239}
240
241/**
242 * radeon_ring_test_lockup() - check if ring is lockedup by recording information
243 * @rdev:       radeon device structure
244 * @ring:       radeon_ring structure holding ring information
245 *
246 */
247bool radeon_ring_test_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
248{
249	uint32_t rptr = radeon_ring_get_rptr(rdev, ring);
250	uint64_t last = atomic64_read(&ring->last_activity);
251	uint64_t elapsed;
252
253	if (rptr != atomic_read(&ring->last_rptr)) {
254		/* ring is still working, no lockup */
255		radeon_ring_lockup_update(rdev, ring);
256		return false;
257	}
258
259	elapsed = jiffies_to_msecs(jiffies_64 - last);
260	if (radeon_lockup_timeout && elapsed >= radeon_lockup_timeout) {
261		dev_err(rdev->dev, "ring %d stalled for more than %llumsec\n",
262			ring->idx, elapsed);
263		return true;
264	}
265	/* give a chance to the GPU ... */
266	return false;
267}
268
269/**
270 * radeon_ring_backup - Back up the content of a ring
271 *
272 * @rdev: radeon_device pointer
273 * @ring: the ring we want to back up
274 *
275 * Saves all unprocessed commits from a ring, returns the number of dwords saved.
276 */
277unsigned radeon_ring_backup(struct radeon_device *rdev, struct radeon_ring *ring,
278			    uint32_t **data)
279{
280	unsigned size, ptr, i;
281
282	/* just in case lock the ring */
283	mutex_lock(&rdev->ring_lock);
284	*data = NULL;
285
286	if (ring->ring_obj == NULL) {
287		mutex_unlock(&rdev->ring_lock);
288		return 0;
289	}
290
291	/* it doesn't make sense to save anything if all fences are signaled */
292	if (!radeon_fence_count_emitted(rdev, ring->idx)) {
293		mutex_unlock(&rdev->ring_lock);
294		return 0;
295	}
296
297	/* calculate the number of dw on the ring */
298	if (ring->rptr_save_reg)
299		ptr = RREG32(ring->rptr_save_reg);
300	else if (rdev->wb.enabled)
301		ptr = le32_to_cpu(*ring->next_rptr_cpu_addr);
302	else {
303		/* no way to read back the next rptr */
304		mutex_unlock(&rdev->ring_lock);
305		return 0;
306	}
307
308	size = ring->wptr + (ring->ring_size / 4);
309	size -= ptr;
310	size &= ring->ptr_mask;
311	if (size == 0) {
312		mutex_unlock(&rdev->ring_lock);
313		return 0;
314	}
315
316	/* and then save the content of the ring */
317	*data = drm_malloc_ab(size, sizeof(uint32_t));
318	if (!*data) {
319		mutex_unlock(&rdev->ring_lock);
320		return 0;
321	}
322	for (i = 0; i < size; ++i) {
323		(*data)[i] = ring->ring[ptr++];
324		ptr &= ring->ptr_mask;
325	}
326
327	mutex_unlock(&rdev->ring_lock);
328	return size;
329}
330
331/**
332 * radeon_ring_restore - append saved commands to the ring again
333 *
334 * @rdev: radeon_device pointer
335 * @ring: ring to append commands to
336 * @size: number of dwords we want to write
337 * @data: saved commands
338 *
339 * Allocates space on the ring and restore the previously saved commands.
340 */
341int radeon_ring_restore(struct radeon_device *rdev, struct radeon_ring *ring,
342			unsigned size, uint32_t *data)
343{
344	int i, r;
345
346	if (!size || !data)
347		return 0;
348
349	/* restore the saved ring content */
350	r = radeon_ring_lock(rdev, ring, size);
351	if (r)
352		return r;
353
354	for (i = 0; i < size; ++i) {
355		radeon_ring_write(ring, data[i]);
356	}
357
358	radeon_ring_unlock_commit(rdev, ring, false);
359	drm_free_large(data);
360	return 0;
361}
362
363/**
364 * radeon_ring_init - init driver ring struct.
365 *
366 * @rdev: radeon_device pointer
367 * @ring: radeon_ring structure holding ring information
368 * @ring_size: size of the ring
369 * @rptr_offs: offset of the rptr writeback location in the WB buffer
370 * @nop: nop packet for this ring
371 *
372 * Initialize the driver information for the selected ring (all asics).
373 * Returns 0 on success, error on failure.
374 */
375int radeon_ring_init(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ring_size,
376		     unsigned rptr_offs, u32 nop)
377{
378	int r;
379
380	ring->ring_size = ring_size;
381	ring->rptr_offs = rptr_offs;
382	ring->nop = nop;
383	/* Allocate ring buffer */
384	if (ring->ring_obj == NULL) {
385		r = radeon_bo_create(rdev, ring->ring_size, PAGE_SIZE, true,
386				     RADEON_GEM_DOMAIN_GTT, 0, NULL,
387				     NULL, &ring->ring_obj);
388		if (r) {
389			dev_err(rdev->dev, "(%d) ring create failed\n", r);
390			return r;
391		}
392		r = radeon_bo_reserve(ring->ring_obj, false);
393		if (unlikely(r != 0))
394			return r;
395		r = radeon_bo_pin(ring->ring_obj, RADEON_GEM_DOMAIN_GTT,
396					&ring->gpu_addr);
397		if (r) {
398			radeon_bo_unreserve(ring->ring_obj);
399			dev_err(rdev->dev, "(%d) ring pin failed\n", r);
400			return r;
401		}
402		r = radeon_bo_kmap(ring->ring_obj,
403				       (void **)&ring->ring);
404		radeon_bo_unreserve(ring->ring_obj);
405		if (r) {
406			dev_err(rdev->dev, "(%d) ring map failed\n", r);
407			return r;
408		}
409	}
410	ring->ptr_mask = (ring->ring_size / 4) - 1;
411	ring->ring_free_dw = ring->ring_size / 4;
412	if (rdev->wb.enabled) {
413		u32 index = RADEON_WB_RING0_NEXT_RPTR + (ring->idx * 4);
414		ring->next_rptr_gpu_addr = rdev->wb.gpu_addr + index;
415		ring->next_rptr_cpu_addr = &rdev->wb.wb[index/4];
416	}
417	if (radeon_debugfs_ring_init(rdev, ring)) {
418		DRM_ERROR("Failed to register debugfs file for rings !\n");
419	}
420	radeon_ring_lockup_update(rdev, ring);
421	return 0;
422}
423
424/**
425 * radeon_ring_fini - tear down the driver ring struct.
426 *
427 * @rdev: radeon_device pointer
428 * @ring: radeon_ring structure holding ring information
429 *
430 * Tear down the driver information for the selected ring (all asics).
431 */
432void radeon_ring_fini(struct radeon_device *rdev, struct radeon_ring *ring)
433{
434	int r;
435	struct radeon_bo *ring_obj;
436
437	mutex_lock(&rdev->ring_lock);
438	ring_obj = ring->ring_obj;
439	ring->ready = false;
440	ring->ring = NULL;
441	ring->ring_obj = NULL;
442	mutex_unlock(&rdev->ring_lock);
443
444	if (ring_obj) {
445		r = radeon_bo_reserve(ring_obj, false);
446		if (likely(r == 0)) {
447			radeon_bo_kunmap(ring_obj);
448			radeon_bo_unpin(ring_obj);
449			radeon_bo_unreserve(ring_obj);
450		}
451		radeon_bo_unref(&ring_obj);
452	}
453}
454
455/*
456 * Debugfs info
457 */
458#if defined(CONFIG_DEBUG_FS)
459
460static int radeon_debugfs_ring_info(struct seq_file *m, void *data)
461{
462	struct drm_info_node *node = (struct drm_info_node *) m->private;
463	struct drm_device *dev = node->minor->dev;
464	struct radeon_device *rdev = dev->dev_private;
465	int ridx = *(int*)node->info_ent->data;
466	struct radeon_ring *ring = &rdev->ring[ridx];
467
468	uint32_t rptr, wptr, rptr_next;
469	unsigned count, i, j;
470
471	radeon_ring_free_size(rdev, ring);
472	count = (ring->ring_size / 4) - ring->ring_free_dw;
473
474	wptr = radeon_ring_get_wptr(rdev, ring);
475	seq_printf(m, "wptr: 0x%08x [%5d]\n",
476		   wptr, wptr);
477
478	rptr = radeon_ring_get_rptr(rdev, ring);
479	seq_printf(m, "rptr: 0x%08x [%5d]\n",
480		   rptr, rptr);
481
482	if (ring->rptr_save_reg) {
483		rptr_next = RREG32(ring->rptr_save_reg);
484		seq_printf(m, "rptr next(0x%04x): 0x%08x [%5d]\n",
485			   ring->rptr_save_reg, rptr_next, rptr_next);
486	} else
487		rptr_next = ~0;
488
489	seq_printf(m, "driver's copy of the wptr: 0x%08x [%5d]\n",
490		   ring->wptr, ring->wptr);
491	seq_printf(m, "last semaphore signal addr : 0x%016llx\n",
492		   ring->last_semaphore_signal_addr);
493	seq_printf(m, "last semaphore wait addr   : 0x%016llx\n",
494		   ring->last_semaphore_wait_addr);
495	seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw);
496	seq_printf(m, "%u dwords in ring\n", count);
497
498	if (!ring->ring)
499		return 0;
500
501	/* print 8 dw before current rptr as often it's the last executed
502	 * packet that is the root issue
503	 */
504	i = (rptr + ring->ptr_mask + 1 - 32) & ring->ptr_mask;
505	for (j = 0; j <= (count + 32); j++) {
506		seq_printf(m, "r[%5d]=0x%08x", i, ring->ring[i]);
507		if (rptr == i)
508			seq_puts(m, " *");
509		if (rptr_next == i)
510			seq_puts(m, " #");
511		seq_puts(m, "\n");
512		i = (i + 1) & ring->ptr_mask;
513	}
514	return 0;
515}
516
517static int radeon_gfx_index = RADEON_RING_TYPE_GFX_INDEX;
518static int cayman_cp1_index = CAYMAN_RING_TYPE_CP1_INDEX;
519static int cayman_cp2_index = CAYMAN_RING_TYPE_CP2_INDEX;
520static int radeon_dma1_index = R600_RING_TYPE_DMA_INDEX;
521static int radeon_dma2_index = CAYMAN_RING_TYPE_DMA1_INDEX;
522static int r600_uvd_index = R600_RING_TYPE_UVD_INDEX;
523static int si_vce1_index = TN_RING_TYPE_VCE1_INDEX;
524static int si_vce2_index = TN_RING_TYPE_VCE2_INDEX;
525
526static struct drm_info_list radeon_debugfs_ring_info_list[] = {
527	{"radeon_ring_gfx", radeon_debugfs_ring_info, 0, &radeon_gfx_index},
528	{"radeon_ring_cp1", radeon_debugfs_ring_info, 0, &cayman_cp1_index},
529	{"radeon_ring_cp2", radeon_debugfs_ring_info, 0, &cayman_cp2_index},
530	{"radeon_ring_dma1", radeon_debugfs_ring_info, 0, &radeon_dma1_index},
531	{"radeon_ring_dma2", radeon_debugfs_ring_info, 0, &radeon_dma2_index},
532	{"radeon_ring_uvd", radeon_debugfs_ring_info, 0, &r600_uvd_index},
533	{"radeon_ring_vce1", radeon_debugfs_ring_info, 0, &si_vce1_index},
534	{"radeon_ring_vce2", radeon_debugfs_ring_info, 0, &si_vce2_index},
535};
536
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
537#endif
538
539static int radeon_debugfs_ring_init(struct radeon_device *rdev, struct radeon_ring *ring)
540{
541#if defined(CONFIG_DEBUG_FS)
542	unsigned i;
543	for (i = 0; i < ARRAY_SIZE(radeon_debugfs_ring_info_list); ++i) {
544		struct drm_info_list *info = &radeon_debugfs_ring_info_list[i];
545		int ridx = *(int*)radeon_debugfs_ring_info_list[i].data;
546		unsigned r;
547
548		if (&rdev->ring[ridx] != ring)
549			continue;
550
551		r = radeon_debugfs_add_files(rdev, info, 1);
552		if (r)
553			return r;
554	}
555#endif
556	return 0;
 
 
 
 
 
 
 
 
 
557}
  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 *          Christian König
 28 */
 29#include <linux/seq_file.h>
 30#include <linux/slab.h>
 31#include <drm/drmP.h>
 32#include <drm/radeon_drm.h>
 33#include "radeon_reg.h"
 34#include "radeon.h"
 35#include "atom.h"
 36
 37/*
 38 * IB
 39 * IBs (Indirect Buffers) and areas of GPU accessible memory where
 40 * commands are stored.  You can put a pointer to the IB in the
 41 * command ring and the hw will fetch the commands from the IB
 42 * and execute them.  Generally userspace acceleration drivers
 43 * produce command buffers which are send to the kernel and
 44 * put in IBs for execution by the requested ring.
 45 */
 46static int radeon_debugfs_sa_init(struct radeon_device *rdev);
 47
 48/**
 49 * radeon_ib_get - request an IB (Indirect Buffer)
 50 *
 51 * @rdev: radeon_device pointer
 52 * @ring: ring index the IB is associated with
 53 * @ib: IB object returned
 54 * @size: requested IB size
 55 *
 56 * Request an IB (all asics).  IBs are allocated using the
 57 * suballocator.
 58 * Returns 0 on success, error on failure.
 59 */
 60int radeon_ib_get(struct radeon_device *rdev, int ring,
 61		  struct radeon_ib *ib, struct radeon_vm *vm,
 62		  unsigned size)
 63{
 64	int r;
 65
 66	r = radeon_sa_bo_new(rdev, &rdev->ring_tmp_bo, &ib->sa_bo, size, 256);
 67	if (r) {
 68		dev_err(rdev->dev, "failed to get a new IB (%d)\n", r);
 69		return r;
 70	}
 71
 72	r = radeon_semaphore_create(rdev, &ib->semaphore);
 73	if (r) {
 74		return r;
 75	}
 76
 77	ib->ring = ring;
 78	ib->fence = NULL;
 79	ib->ptr = radeon_sa_bo_cpu_addr(ib->sa_bo);
 80	ib->vm = vm;
 81	if (vm) {
 82		/* ib pool is bound at RADEON_VA_IB_OFFSET in virtual address
 83		 * space and soffset is the offset inside the pool bo
 84		 */
 85		ib->gpu_addr = ib->sa_bo->soffset + RADEON_VA_IB_OFFSET;
 86	} else {
 87		ib->gpu_addr = radeon_sa_bo_gpu_addr(ib->sa_bo);
 88	}
 89	ib->is_const_ib = false;
 90
 91	return 0;
 92}
 93
 94/**
 95 * radeon_ib_free - free an IB (Indirect Buffer)
 96 *
 97 * @rdev: radeon_device pointer
 98 * @ib: IB object to free
 99 *
100 * Free an IB (all asics).
101 */
102void radeon_ib_free(struct radeon_device *rdev, struct radeon_ib *ib)
103{
104	radeon_semaphore_free(rdev, &ib->semaphore, ib->fence);
105	radeon_sa_bo_free(rdev, &ib->sa_bo, ib->fence);
106	radeon_fence_unref(&ib->fence);
107}
108
109/**
110 * radeon_ib_schedule - schedule an IB (Indirect Buffer) on the ring
111 *
112 * @rdev: radeon_device pointer
113 * @ib: IB object to schedule
114 * @const_ib: Const IB to schedule (SI only)
115 *
116 * Schedule an IB on the associated ring (all asics).
117 * Returns 0 on success, error on failure.
118 *
119 * On SI, there are two parallel engines fed from the primary ring,
120 * the CE (Constant Engine) and the DE (Drawing Engine).  Since
121 * resource descriptors have moved to memory, the CE allows you to
122 * prime the caches while the DE is updating register state so that
123 * the resource descriptors will be already in cache when the draw is
124 * processed.  To accomplish this, the userspace driver submits two
125 * IBs, one for the CE and one for the DE.  If there is a CE IB (called
126 * a CONST_IB), it will be put on the ring prior to the DE IB.  Prior
127 * to SI there was just a DE IB.
128 */
129int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib,
130		       struct radeon_ib *const_ib)
131{
132	struct radeon_ring *ring = &rdev->ring[ib->ring];
133	int r = 0;
134
135	if (!ib->length_dw || !ring->ready) {
136		/* TODO: Nothings in the ib we should report. */
137		dev_err(rdev->dev, "couldn't schedule ib\n");
138		return -EINVAL;
139	}
140
141	/* 64 dwords should be enough for fence too */
142	r = radeon_ring_lock(rdev, ring, 64 + RADEON_NUM_SYNCS * 8);
143	if (r) {
144		dev_err(rdev->dev, "scheduling IB failed (%d).\n", r);
145		return r;
146	}
147
148	/* grab a vm id if necessary */
149	if (ib->vm) {
150		struct radeon_fence *vm_id_fence;
151		vm_id_fence = radeon_vm_grab_id(rdev, ib->vm, ib->ring);
152        	radeon_semaphore_sync_to(ib->semaphore, vm_id_fence);
153	}
154
155	/* sync with other rings */
156	r = radeon_semaphore_sync_rings(rdev, ib->semaphore, ib->ring);
157	if (r) {
158		dev_err(rdev->dev, "failed to sync rings (%d)\n", r);
159		radeon_ring_unlock_undo(rdev, ring);
160		return r;
161	}
162
163	if (ib->vm)
164		radeon_vm_flush(rdev, ib->vm, ib->ring);
165
166	if (const_ib) {
167		radeon_ring_ib_execute(rdev, const_ib->ring, const_ib);
168		radeon_semaphore_free(rdev, &const_ib->semaphore, NULL);
169	}
170	radeon_ring_ib_execute(rdev, ib->ring, ib);
171	r = radeon_fence_emit(rdev, &ib->fence, ib->ring);
172	if (r) {
173		dev_err(rdev->dev, "failed to emit fence for new IB (%d)\n", r);
174		radeon_ring_unlock_undo(rdev, ring);
175		return r;
176	}
177	if (const_ib) {
178		const_ib->fence = radeon_fence_ref(ib->fence);
179	}
180
181	if (ib->vm)
182		radeon_vm_fence(rdev, ib->vm, ib->fence);
183
184	radeon_ring_unlock_commit(rdev, ring);
185	return 0;
186}
187
188/**
189 * radeon_ib_pool_init - Init the IB (Indirect Buffer) pool
190 *
191 * @rdev: radeon_device pointer
192 *
193 * Initialize the suballocator to manage a pool of memory
194 * for use as IBs (all asics).
195 * Returns 0 on success, error on failure.
196 */
197int radeon_ib_pool_init(struct radeon_device *rdev)
198{
199	int r;
200
201	if (rdev->ib_pool_ready) {
202		return 0;
203	}
204	r = radeon_sa_bo_manager_init(rdev, &rdev->ring_tmp_bo,
205				      RADEON_IB_POOL_SIZE*64*1024,
206				      RADEON_GPU_PAGE_SIZE,
207				      RADEON_GEM_DOMAIN_GTT);
208	if (r) {
209		return r;
210	}
211
212	r = radeon_sa_bo_manager_start(rdev, &rdev->ring_tmp_bo);
213	if (r) {
214		return r;
215	}
216
217	rdev->ib_pool_ready = true;
218	if (radeon_debugfs_sa_init(rdev)) {
219		dev_err(rdev->dev, "failed to register debugfs file for SA\n");
220	}
221	return 0;
222}
223
224/**
225 * radeon_ib_pool_fini - Free the IB (Indirect Buffer) pool
226 *
227 * @rdev: radeon_device pointer
228 *
229 * Tear down the suballocator managing the pool of memory
230 * for use as IBs (all asics).
231 */
232void radeon_ib_pool_fini(struct radeon_device *rdev)
233{
234	if (rdev->ib_pool_ready) {
235		radeon_sa_bo_manager_suspend(rdev, &rdev->ring_tmp_bo);
236		radeon_sa_bo_manager_fini(rdev, &rdev->ring_tmp_bo);
237		rdev->ib_pool_ready = false;
238	}
239}
240
241/**
242 * radeon_ib_ring_tests - test IBs on the rings
243 *
244 * @rdev: radeon_device pointer
245 *
246 * Test an IB (Indirect Buffer) on each ring.
247 * If the test fails, disable the ring.
248 * Returns 0 on success, error if the primary GFX ring
249 * IB test fails.
250 */
251int radeon_ib_ring_tests(struct radeon_device *rdev)
252{
253	unsigned i;
254	int r;
255
256	for (i = 0; i < RADEON_NUM_RINGS; ++i) {
257		struct radeon_ring *ring = &rdev->ring[i];
258
259		if (!ring->ready)
260			continue;
261
262		r = radeon_ib_test(rdev, i, ring);
263		if (r) {
264			ring->ready = false;
265			rdev->needs_reset = false;
266
267			if (i == RADEON_RING_TYPE_GFX_INDEX) {
268				/* oh, oh, that's really bad */
269				DRM_ERROR("radeon: failed testing IB on GFX ring (%d).\n", r);
270		                rdev->accel_working = false;
271				return r;
272
273			} else {
274				/* still not good, but we can live with it */
275				DRM_ERROR("radeon: failed testing IB on ring %d (%d).\n", i, r);
276			}
277		}
278	}
279	return 0;
280}
281
282/*
283 * Rings
284 * Most engines on the GPU are fed via ring buffers.  Ring
285 * buffers are areas of GPU accessible memory that the host
286 * writes commands into and the GPU reads commands out of.
287 * There is a rptr (read pointer) that determines where the
288 * GPU is currently reading, and a wptr (write pointer)
289 * which determines where the host has written.  When the
290 * pointers are equal, the ring is idle.  When the host
291 * writes commands to the ring buffer, it increments the
292 * wptr.  The GPU then starts fetching commands and executes
293 * them until the pointers are equal again.
294 */
295static int radeon_debugfs_ring_init(struct radeon_device *rdev, struct radeon_ring *ring);
296
297/**
298 * radeon_ring_write - write a value to the ring
299 *
300 * @ring: radeon_ring structure holding ring information
301 * @v: dword (dw) value to write
302 *
303 * Write a value to the requested ring buffer (all asics).
304 */
305void radeon_ring_write(struct radeon_ring *ring, uint32_t v)
306{
307#if DRM_DEBUG_CODE
308	if (ring->count_dw <= 0) {
309		DRM_ERROR("radeon: writing more dwords to the ring than expected!\n");
310	}
311#endif
312	ring->ring[ring->wptr++] = v;
313	ring->wptr &= ring->ptr_mask;
314	ring->count_dw--;
315	ring->ring_free_dw--;
316}
317
318/**
319 * radeon_ring_supports_scratch_reg - check if the ring supports
320 * writing to scratch registers
321 *
322 * @rdev: radeon_device pointer
323 * @ring: radeon_ring structure holding ring information
324 *
325 * Check if a specific ring supports writing to scratch registers (all asics).
326 * Returns true if the ring supports writing to scratch regs, false if not.
327 */
328bool radeon_ring_supports_scratch_reg(struct radeon_device *rdev,
329				      struct radeon_ring *ring)
330{
331	switch (ring->idx) {
332	case RADEON_RING_TYPE_GFX_INDEX:
333	case CAYMAN_RING_TYPE_CP1_INDEX:
334	case CAYMAN_RING_TYPE_CP2_INDEX:
335		return true;
336	default:
337		return false;
338	}
339}
340
341/**
342 * radeon_ring_free_size - update the free size
343 *
344 * @rdev: radeon_device pointer
345 * @ring: radeon_ring structure holding ring information
346 *
347 * Update the free dw slots in the ring buffer (all asics).
348 */
349void radeon_ring_free_size(struct radeon_device *rdev, struct radeon_ring *ring)
350{
351	uint32_t rptr = radeon_ring_get_rptr(rdev, ring);
352
353	/* This works because ring_size is a power of 2 */
354	ring->ring_free_dw = rptr + (ring->ring_size / 4);
355	ring->ring_free_dw -= ring->wptr;
356	ring->ring_free_dw &= ring->ptr_mask;
357	if (!ring->ring_free_dw) {
358		/* this is an empty ring */
359		ring->ring_free_dw = ring->ring_size / 4;
360		/*  update lockup info to avoid false positive */
361		radeon_ring_lockup_update(rdev, ring);
362	}
363}
364
365/**
366 * radeon_ring_alloc - allocate space on the ring buffer
367 *
368 * @rdev: radeon_device pointer
369 * @ring: radeon_ring structure holding ring information
370 * @ndw: number of dwords to allocate in the ring buffer
371 *
372 * Allocate @ndw dwords in the ring buffer (all asics).
373 * Returns 0 on success, error on failure.
374 */
375int radeon_ring_alloc(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ndw)
376{
377	int r;
378
379	/* make sure we aren't trying to allocate more space than there is on the ring */
380	if (ndw > (ring->ring_size / 4))
381		return -ENOMEM;
382	/* Align requested size with padding so unlock_commit can
383	 * pad safely */
384	radeon_ring_free_size(rdev, ring);
385	ndw = (ndw + ring->align_mask) & ~ring->align_mask;
386	while (ndw > (ring->ring_free_dw - 1)) {
387		radeon_ring_free_size(rdev, ring);
388		if (ndw < ring->ring_free_dw) {
389			break;
390		}
391		r = radeon_fence_wait_next(rdev, ring->idx);
392		if (r)
393			return r;
394	}
395	ring->count_dw = ndw;
396	ring->wptr_old = ring->wptr;
397	return 0;
398}
399
400/**
401 * radeon_ring_lock - lock the ring and allocate space on it
402 *
403 * @rdev: radeon_device pointer
404 * @ring: radeon_ring structure holding ring information
405 * @ndw: number of dwords to allocate in the ring buffer
406 *
407 * Lock the ring and allocate @ndw dwords in the ring buffer
408 * (all asics).
409 * Returns 0 on success, error on failure.
410 */
411int radeon_ring_lock(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ndw)
412{
413	int r;
414
415	mutex_lock(&rdev->ring_lock);
416	r = radeon_ring_alloc(rdev, ring, ndw);
417	if (r) {
418		mutex_unlock(&rdev->ring_lock);
419		return r;
420	}
421	return 0;
422}
423
424/**
425 * radeon_ring_commit - tell the GPU to execute the new
426 * commands on the ring buffer
427 *
428 * @rdev: radeon_device pointer
429 * @ring: radeon_ring structure holding ring information
 
430 *
431 * Update the wptr (write pointer) to tell the GPU to
432 * execute new commands on the ring buffer (all asics).
433 */
434void radeon_ring_commit(struct radeon_device *rdev, struct radeon_ring *ring)
 
435{
 
 
 
 
 
436	/* We pad to match fetch size */
437	while (ring->wptr & ring->align_mask) {
438		radeon_ring_write(ring, ring->nop);
439	}
440	mb();
 
 
 
 
 
441	radeon_ring_set_wptr(rdev, ring);
442}
443
444/**
445 * radeon_ring_unlock_commit - tell the GPU to execute the new
446 * commands on the ring buffer and unlock it
447 *
448 * @rdev: radeon_device pointer
449 * @ring: radeon_ring structure holding ring information
 
450 *
451 * Call radeon_ring_commit() then unlock the ring (all asics).
452 */
453void radeon_ring_unlock_commit(struct radeon_device *rdev, struct radeon_ring *ring)
 
454{
455	radeon_ring_commit(rdev, ring);
456	mutex_unlock(&rdev->ring_lock);
457}
458
459/**
460 * radeon_ring_undo - reset the wptr
461 *
462 * @ring: radeon_ring structure holding ring information
463 *
464 * Reset the driver's copy of the wptr (all asics).
465 */
466void radeon_ring_undo(struct radeon_ring *ring)
467{
468	ring->wptr = ring->wptr_old;
469}
470
471/**
472 * radeon_ring_unlock_undo - reset the wptr and unlock the ring
473 *
474 * @ring: radeon_ring structure holding ring information
475 *
476 * Call radeon_ring_undo() then unlock the ring (all asics).
477 */
478void radeon_ring_unlock_undo(struct radeon_device *rdev, struct radeon_ring *ring)
479{
480	radeon_ring_undo(ring);
481	mutex_unlock(&rdev->ring_lock);
482}
483
484/**
485 * radeon_ring_lockup_update - update lockup variables
486 *
487 * @ring: radeon_ring structure holding ring information
488 *
489 * Update the last rptr value and timestamp (all asics).
490 */
491void radeon_ring_lockup_update(struct radeon_device *rdev,
492			       struct radeon_ring *ring)
493{
494	atomic_set(&ring->last_rptr, radeon_ring_get_rptr(rdev, ring));
495	atomic64_set(&ring->last_activity, jiffies_64);
496}
497
498/**
499 * radeon_ring_test_lockup() - check if ring is lockedup by recording information
500 * @rdev:       radeon device structure
501 * @ring:       radeon_ring structure holding ring information
502 *
503 */
504bool radeon_ring_test_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
505{
506	uint32_t rptr = radeon_ring_get_rptr(rdev, ring);
507	uint64_t last = atomic64_read(&ring->last_activity);
508	uint64_t elapsed;
509
510	if (rptr != atomic_read(&ring->last_rptr)) {
511		/* ring is still working, no lockup */
512		radeon_ring_lockup_update(rdev, ring);
513		return false;
514	}
515
516	elapsed = jiffies_to_msecs(jiffies_64 - last);
517	if (radeon_lockup_timeout && elapsed >= radeon_lockup_timeout) {
518		dev_err(rdev->dev, "ring %d stalled for more than %llumsec\n",
519			ring->idx, elapsed);
520		return true;
521	}
522	/* give a chance to the GPU ... */
523	return false;
524}
525
526/**
527 * radeon_ring_backup - Back up the content of a ring
528 *
529 * @rdev: radeon_device pointer
530 * @ring: the ring we want to back up
531 *
532 * Saves all unprocessed commits from a ring, returns the number of dwords saved.
533 */
534unsigned radeon_ring_backup(struct radeon_device *rdev, struct radeon_ring *ring,
535			    uint32_t **data)
536{
537	unsigned size, ptr, i;
538
539	/* just in case lock the ring */
540	mutex_lock(&rdev->ring_lock);
541	*data = NULL;
542
543	if (ring->ring_obj == NULL) {
544		mutex_unlock(&rdev->ring_lock);
545		return 0;
546	}
547
548	/* it doesn't make sense to save anything if all fences are signaled */
549	if (!radeon_fence_count_emitted(rdev, ring->idx)) {
550		mutex_unlock(&rdev->ring_lock);
551		return 0;
552	}
553
554	/* calculate the number of dw on the ring */
555	if (ring->rptr_save_reg)
556		ptr = RREG32(ring->rptr_save_reg);
557	else if (rdev->wb.enabled)
558		ptr = le32_to_cpu(*ring->next_rptr_cpu_addr);
559	else {
560		/* no way to read back the next rptr */
561		mutex_unlock(&rdev->ring_lock);
562		return 0;
563	}
564
565	size = ring->wptr + (ring->ring_size / 4);
566	size -= ptr;
567	size &= ring->ptr_mask;
568	if (size == 0) {
569		mutex_unlock(&rdev->ring_lock);
570		return 0;
571	}
572
573	/* and then save the content of the ring */
574	*data = kmalloc_array(size, sizeof(uint32_t), GFP_KERNEL);
575	if (!*data) {
576		mutex_unlock(&rdev->ring_lock);
577		return 0;
578	}
579	for (i = 0; i < size; ++i) {
580		(*data)[i] = ring->ring[ptr++];
581		ptr &= ring->ptr_mask;
582	}
583
584	mutex_unlock(&rdev->ring_lock);
585	return size;
586}
587
588/**
589 * radeon_ring_restore - append saved commands to the ring again
590 *
591 * @rdev: radeon_device pointer
592 * @ring: ring to append commands to
593 * @size: number of dwords we want to write
594 * @data: saved commands
595 *
596 * Allocates space on the ring and restore the previously saved commands.
597 */
598int radeon_ring_restore(struct radeon_device *rdev, struct radeon_ring *ring,
599			unsigned size, uint32_t *data)
600{
601	int i, r;
602
603	if (!size || !data)
604		return 0;
605
606	/* restore the saved ring content */
607	r = radeon_ring_lock(rdev, ring, size);
608	if (r)
609		return r;
610
611	for (i = 0; i < size; ++i) {
612		radeon_ring_write(ring, data[i]);
613	}
614
615	radeon_ring_unlock_commit(rdev, ring);
616	kfree(data);
617	return 0;
618}
619
620/**
621 * radeon_ring_init - init driver ring struct.
622 *
623 * @rdev: radeon_device pointer
624 * @ring: radeon_ring structure holding ring information
625 * @ring_size: size of the ring
626 * @rptr_offs: offset of the rptr writeback location in the WB buffer
627 * @nop: nop packet for this ring
628 *
629 * Initialize the driver information for the selected ring (all asics).
630 * Returns 0 on success, error on failure.
631 */
632int radeon_ring_init(struct radeon_device *rdev, struct radeon_ring *ring, unsigned ring_size,
633		     unsigned rptr_offs, u32 nop)
634{
635	int r;
636
637	ring->ring_size = ring_size;
638	ring->rptr_offs = rptr_offs;
639	ring->nop = nop;
640	/* Allocate ring buffer */
641	if (ring->ring_obj == NULL) {
642		r = radeon_bo_create(rdev, ring->ring_size, PAGE_SIZE, true,
643				     RADEON_GEM_DOMAIN_GTT,
644				     NULL, &ring->ring_obj);
645		if (r) {
646			dev_err(rdev->dev, "(%d) ring create failed\n", r);
647			return r;
648		}
649		r = radeon_bo_reserve(ring->ring_obj, false);
650		if (unlikely(r != 0))
651			return r;
652		r = radeon_bo_pin(ring->ring_obj, RADEON_GEM_DOMAIN_GTT,
653					&ring->gpu_addr);
654		if (r) {
655			radeon_bo_unreserve(ring->ring_obj);
656			dev_err(rdev->dev, "(%d) ring pin failed\n", r);
657			return r;
658		}
659		r = radeon_bo_kmap(ring->ring_obj,
660				       (void **)&ring->ring);
661		radeon_bo_unreserve(ring->ring_obj);
662		if (r) {
663			dev_err(rdev->dev, "(%d) ring map failed\n", r);
664			return r;
665		}
666	}
667	ring->ptr_mask = (ring->ring_size / 4) - 1;
668	ring->ring_free_dw = ring->ring_size / 4;
669	if (rdev->wb.enabled) {
670		u32 index = RADEON_WB_RING0_NEXT_RPTR + (ring->idx * 4);
671		ring->next_rptr_gpu_addr = rdev->wb.gpu_addr + index;
672		ring->next_rptr_cpu_addr = &rdev->wb.wb[index/4];
673	}
674	if (radeon_debugfs_ring_init(rdev, ring)) {
675		DRM_ERROR("Failed to register debugfs file for rings !\n");
676	}
677	radeon_ring_lockup_update(rdev, ring);
678	return 0;
679}
680
681/**
682 * radeon_ring_fini - tear down the driver ring struct.
683 *
684 * @rdev: radeon_device pointer
685 * @ring: radeon_ring structure holding ring information
686 *
687 * Tear down the driver information for the selected ring (all asics).
688 */
689void radeon_ring_fini(struct radeon_device *rdev, struct radeon_ring *ring)
690{
691	int r;
692	struct radeon_bo *ring_obj;
693
694	mutex_lock(&rdev->ring_lock);
695	ring_obj = ring->ring_obj;
696	ring->ready = false;
697	ring->ring = NULL;
698	ring->ring_obj = NULL;
699	mutex_unlock(&rdev->ring_lock);
700
701	if (ring_obj) {
702		r = radeon_bo_reserve(ring_obj, false);
703		if (likely(r == 0)) {
704			radeon_bo_kunmap(ring_obj);
705			radeon_bo_unpin(ring_obj);
706			radeon_bo_unreserve(ring_obj);
707		}
708		radeon_bo_unref(&ring_obj);
709	}
710}
711
712/*
713 * Debugfs info
714 */
715#if defined(CONFIG_DEBUG_FS)
716
717static int radeon_debugfs_ring_info(struct seq_file *m, void *data)
718{
719	struct drm_info_node *node = (struct drm_info_node *) m->private;
720	struct drm_device *dev = node->minor->dev;
721	struct radeon_device *rdev = dev->dev_private;
722	int ridx = *(int*)node->info_ent->data;
723	struct radeon_ring *ring = &rdev->ring[ridx];
724
725	uint32_t rptr, wptr, rptr_next;
726	unsigned count, i, j;
727
728	radeon_ring_free_size(rdev, ring);
729	count = (ring->ring_size / 4) - ring->ring_free_dw;
730
731	wptr = radeon_ring_get_wptr(rdev, ring);
732	seq_printf(m, "wptr: 0x%08x [%5d]\n",
733		   wptr, wptr);
734
735	rptr = radeon_ring_get_rptr(rdev, ring);
736	seq_printf(m, "rptr: 0x%08x [%5d]\n",
737		   rptr, rptr);
738
739	if (ring->rptr_save_reg) {
740		rptr_next = RREG32(ring->rptr_save_reg);
741		seq_printf(m, "rptr next(0x%04x): 0x%08x [%5d]\n",
742			   ring->rptr_save_reg, rptr_next, rptr_next);
743	} else
744		rptr_next = ~0;
745
746	seq_printf(m, "driver's copy of the wptr: 0x%08x [%5d]\n",
747		   ring->wptr, ring->wptr);
748	seq_printf(m, "last semaphore signal addr : 0x%016llx\n",
749		   ring->last_semaphore_signal_addr);
750	seq_printf(m, "last semaphore wait addr   : 0x%016llx\n",
751		   ring->last_semaphore_wait_addr);
752	seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw);
753	seq_printf(m, "%u dwords in ring\n", count);
754
755	if (!ring->ready)
756		return 0;
757
758	/* print 8 dw before current rptr as often it's the last executed
759	 * packet that is the root issue
760	 */
761	i = (rptr + ring->ptr_mask + 1 - 32) & ring->ptr_mask;
762	for (j = 0; j <= (count + 32); j++) {
763		seq_printf(m, "r[%5d]=0x%08x", i, ring->ring[i]);
764		if (rptr == i)
765			seq_puts(m, " *");
766		if (rptr_next == i)
767			seq_puts(m, " #");
768		seq_puts(m, "\n");
769		i = (i + 1) & ring->ptr_mask;
770	}
771	return 0;
772}
773
774static int radeon_gfx_index = RADEON_RING_TYPE_GFX_INDEX;
775static int cayman_cp1_index = CAYMAN_RING_TYPE_CP1_INDEX;
776static int cayman_cp2_index = CAYMAN_RING_TYPE_CP2_INDEX;
777static int radeon_dma1_index = R600_RING_TYPE_DMA_INDEX;
778static int radeon_dma2_index = CAYMAN_RING_TYPE_DMA1_INDEX;
779static int r600_uvd_index = R600_RING_TYPE_UVD_INDEX;
780static int si_vce1_index = TN_RING_TYPE_VCE1_INDEX;
781static int si_vce2_index = TN_RING_TYPE_VCE2_INDEX;
782
783static struct drm_info_list radeon_debugfs_ring_info_list[] = {
784	{"radeon_ring_gfx", radeon_debugfs_ring_info, 0, &radeon_gfx_index},
785	{"radeon_ring_cp1", radeon_debugfs_ring_info, 0, &cayman_cp1_index},
786	{"radeon_ring_cp2", radeon_debugfs_ring_info, 0, &cayman_cp2_index},
787	{"radeon_ring_dma1", radeon_debugfs_ring_info, 0, &radeon_dma1_index},
788	{"radeon_ring_dma2", radeon_debugfs_ring_info, 0, &radeon_dma2_index},
789	{"radeon_ring_uvd", radeon_debugfs_ring_info, 0, &r600_uvd_index},
790	{"radeon_ring_vce1", radeon_debugfs_ring_info, 0, &si_vce1_index},
791	{"radeon_ring_vce2", radeon_debugfs_ring_info, 0, &si_vce2_index},
792};
793
794static int radeon_debugfs_sa_info(struct seq_file *m, void *data)
795{
796	struct drm_info_node *node = (struct drm_info_node *) m->private;
797	struct drm_device *dev = node->minor->dev;
798	struct radeon_device *rdev = dev->dev_private;
799
800	radeon_sa_bo_dump_debug_info(&rdev->ring_tmp_bo, m);
801
802	return 0;
803
804}
805
806static struct drm_info_list radeon_debugfs_sa_list[] = {
807        {"radeon_sa_info", &radeon_debugfs_sa_info, 0, NULL},
808};
809
810#endif
811
812static int radeon_debugfs_ring_init(struct radeon_device *rdev, struct radeon_ring *ring)
813{
814#if defined(CONFIG_DEBUG_FS)
815	unsigned i;
816	for (i = 0; i < ARRAY_SIZE(radeon_debugfs_ring_info_list); ++i) {
817		struct drm_info_list *info = &radeon_debugfs_ring_info_list[i];
818		int ridx = *(int*)radeon_debugfs_ring_info_list[i].data;
819		unsigned r;
820
821		if (&rdev->ring[ridx] != ring)
822			continue;
823
824		r = radeon_debugfs_add_files(rdev, info, 1);
825		if (r)
826			return r;
827	}
828#endif
829	return 0;
830}
831
832static int radeon_debugfs_sa_init(struct radeon_device *rdev)
833{
834#if defined(CONFIG_DEBUG_FS)
835	return radeon_debugfs_add_files(rdev, radeon_debugfs_sa_list, 1);
836#else
837	return 0;
838#endif
839}