xref: /linux/drivers/gpu/drm/radeon/r420.c (revision 24168c5e6dfbdd5b414f048f47f75d64533296ca)
1 /*
2  * Copyright 2008 Advanced Micro Devices, Inc.
3  * Copyright 2008 Red Hat Inc.
4  * Copyright 2009 Jerome Glisse.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a
7  * copy of this software and associated documentation files (the "Software"),
8  * to deal in the Software without restriction, including without limitation
9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10  * and/or sell copies of the Software, and to permit persons to whom the
11  * Software is furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22  * OTHER DEALINGS IN THE SOFTWARE.
23  *
24  * Authors: Dave Airlie
25  *          Alex Deucher
26  *          Jerome Glisse
27  */
28 
29 #include <linux/debugfs.h>
30 #include <linux/pci.h>
31 #include <linux/seq_file.h>
32 #include <linux/slab.h>
33 
34 #include <drm/drm_device.h>
35 #include <drm/drm_file.h>
36 
37 #include "atom.h"
38 #include "r100d.h"
39 #include "r420_reg_safe.h"
40 #include "r420d.h"
41 #include "radeon.h"
42 #include "radeon_asic.h"
43 #include "radeon_reg.h"
44 
45 void r420_pm_init_profile(struct radeon_device *rdev)
46 {
47 	/* default */
48 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
49 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
50 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
51 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 0;
52 	/* low sh */
53 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = 0;
54 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = 0;
55 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
56 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
57 	/* mid sh */
58 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = 0;
59 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = 1;
60 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
61 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 0;
62 	/* high sh */
63 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = 0;
64 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
65 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
66 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 0;
67 	/* low mh */
68 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = 0;
69 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
70 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
71 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
72 	/* mid mh */
73 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = 0;
74 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
75 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
76 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 0;
77 	/* high mh */
78 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = 0;
79 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
80 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
81 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 0;
82 }
83 
84 static void r420_set_reg_safe(struct radeon_device *rdev)
85 {
86 	rdev->config.r300.reg_safe_bm = r420_reg_safe_bm;
87 	rdev->config.r300.reg_safe_bm_size = ARRAY_SIZE(r420_reg_safe_bm);
88 }
89 
90 void r420_pipes_init(struct radeon_device *rdev)
91 {
92 	unsigned tmp;
93 	unsigned gb_pipe_select;
94 	unsigned num_pipes;
95 
96 	/* GA_ENHANCE workaround TCL deadlock issue */
97 	WREG32(R300_GA_ENHANCE, R300_GA_DEADLOCK_CNTL | R300_GA_FASTSYNC_CNTL |
98 	       (1 << 2) | (1 << 3));
99 	/* add idle wait as per freedesktop.org bug 24041 */
100 	if (r100_gui_wait_for_idle(rdev)) {
101 		pr_warn("Failed to wait GUI idle while programming pipes. Bad things might happen.\n");
102 	}
103 	/* get max number of pipes */
104 	gb_pipe_select = RREG32(R400_GB_PIPE_SELECT);
105 	num_pipes = ((gb_pipe_select >> 12) & 3) + 1;
106 
107 	/* SE chips have 1 pipe */
108 	if ((rdev->pdev->device == 0x5e4c) ||
109 	    (rdev->pdev->device == 0x5e4f))
110 		num_pipes = 1;
111 
112 	rdev->num_gb_pipes = num_pipes;
113 	tmp = 0;
114 	switch (num_pipes) {
115 	default:
116 		/* force to 1 pipe */
117 		num_pipes = 1;
118 		fallthrough;
119 	case 1:
120 		tmp = (0 << 1);
121 		break;
122 	case 2:
123 		tmp = (3 << 1);
124 		break;
125 	case 3:
126 		tmp = (6 << 1);
127 		break;
128 	case 4:
129 		tmp = (7 << 1);
130 		break;
131 	}
132 	WREG32(R500_SU_REG_DEST, (1 << num_pipes) - 1);
133 	/* Sub pixel 1/12 so we can have 4K rendering according to doc */
134 	tmp |= R300_TILE_SIZE_16 | R300_ENABLE_TILING;
135 	WREG32(R300_GB_TILE_CONFIG, tmp);
136 	if (r100_gui_wait_for_idle(rdev)) {
137 		pr_warn("Failed to wait GUI idle while programming pipes. Bad things might happen.\n");
138 	}
139 
140 	tmp = RREG32(R300_DST_PIPE_CONFIG);
141 	WREG32(R300_DST_PIPE_CONFIG, tmp | R300_PIPE_AUTO_CONFIG);
142 
143 	WREG32(R300_RB2D_DSTCACHE_MODE,
144 	       RREG32(R300_RB2D_DSTCACHE_MODE) |
145 	       R300_DC_AUTOFLUSH_ENABLE |
146 	       R300_DC_DC_DISABLE_IGNORE_PE);
147 
148 	if (r100_gui_wait_for_idle(rdev)) {
149 		pr_warn("Failed to wait GUI idle while programming pipes. Bad things might happen.\n");
150 	}
151 
152 	if (rdev->family == CHIP_RV530) {
153 		tmp = RREG32(RV530_GB_PIPE_SELECT2);
154 		if ((tmp & 3) == 3)
155 			rdev->num_z_pipes = 2;
156 		else
157 			rdev->num_z_pipes = 1;
158 	} else
159 		rdev->num_z_pipes = 1;
160 
161 	DRM_INFO("radeon: %d quad pipes, %d z pipes initialized.\n",
162 		 rdev->num_gb_pipes, rdev->num_z_pipes);
163 }
164 
165 u32 r420_mc_rreg(struct radeon_device *rdev, u32 reg)
166 {
167 	unsigned long flags;
168 	u32 r;
169 
170 	spin_lock_irqsave(&rdev->mc_idx_lock, flags);
171 	WREG32(R_0001F8_MC_IND_INDEX, S_0001F8_MC_IND_ADDR(reg));
172 	r = RREG32(R_0001FC_MC_IND_DATA);
173 	spin_unlock_irqrestore(&rdev->mc_idx_lock, flags);
174 	return r;
175 }
176 
177 void r420_mc_wreg(struct radeon_device *rdev, u32 reg, u32 v)
178 {
179 	unsigned long flags;
180 
181 	spin_lock_irqsave(&rdev->mc_idx_lock, flags);
182 	WREG32(R_0001F8_MC_IND_INDEX, S_0001F8_MC_IND_ADDR(reg) |
183 		S_0001F8_MC_IND_WR_EN(1));
184 	WREG32(R_0001FC_MC_IND_DATA, v);
185 	spin_unlock_irqrestore(&rdev->mc_idx_lock, flags);
186 }
187 
188 static void r420_debugfs(struct radeon_device *rdev)
189 {
190 	r100_debugfs_rbbm_init(rdev);
191 	r420_debugfs_pipes_info_init(rdev);
192 }
193 
194 static void r420_clock_resume(struct radeon_device *rdev)
195 {
196 	u32 sclk_cntl;
197 
198 	if (radeon_dynclks != -1 && radeon_dynclks)
199 		radeon_atom_set_clock_gating(rdev, 1);
200 	sclk_cntl = RREG32_PLL(R_00000D_SCLK_CNTL);
201 	sclk_cntl |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
202 	if (rdev->family == CHIP_R420)
203 		sclk_cntl |= S_00000D_FORCE_PX(1) | S_00000D_FORCE_TX(1);
204 	WREG32_PLL(R_00000D_SCLK_CNTL, sclk_cntl);
205 }
206 
207 static void r420_cp_errata_init(struct radeon_device *rdev)
208 {
209 	int r;
210 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
211 
212 	/* RV410 and R420 can lock up if CP DMA to host memory happens
213 	 * while the 2D engine is busy.
214 	 *
215 	 * The proper workaround is to queue a RESYNC at the beginning
216 	 * of the CP init, apparently.
217 	 */
218 	radeon_scratch_get(rdev, &rdev->config.r300.resync_scratch);
219 	r = radeon_ring_lock(rdev, ring, 8);
220 	WARN_ON(r);
221 	radeon_ring_write(ring, PACKET0(R300_CP_RESYNC_ADDR, 1));
222 	radeon_ring_write(ring, rdev->config.r300.resync_scratch);
223 	radeon_ring_write(ring, 0xDEADBEEF);
224 	radeon_ring_unlock_commit(rdev, ring, false);
225 }
226 
227 static void r420_cp_errata_fini(struct radeon_device *rdev)
228 {
229 	int r;
230 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
231 
232 	/* Catch the RESYNC we dispatched all the way back,
233 	 * at the very beginning of the CP init.
234 	 */
235 	r = radeon_ring_lock(rdev, ring, 8);
236 	WARN_ON(r);
237 	radeon_ring_write(ring, PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
238 	radeon_ring_write(ring, R300_RB3D_DC_FINISH);
239 	radeon_ring_unlock_commit(rdev, ring, false);
240 	radeon_scratch_free(rdev, rdev->config.r300.resync_scratch);
241 }
242 
243 static int r420_startup(struct radeon_device *rdev)
244 {
245 	int r;
246 
247 	/* set common regs */
248 	r100_set_common_regs(rdev);
249 	/* program mc */
250 	r300_mc_program(rdev);
251 	/* Resume clock */
252 	r420_clock_resume(rdev);
253 	/* Initialize GART (initialize after TTM so we can allocate
254 	 * memory through TTM but finalize after TTM) */
255 	if (rdev->flags & RADEON_IS_PCIE) {
256 		r = rv370_pcie_gart_enable(rdev);
257 		if (r)
258 			return r;
259 	}
260 	if (rdev->flags & RADEON_IS_PCI) {
261 		r = r100_pci_gart_enable(rdev);
262 		if (r)
263 			return r;
264 	}
265 	r420_pipes_init(rdev);
266 
267 	/* allocate wb buffer */
268 	r = radeon_wb_init(rdev);
269 	if (r)
270 		return r;
271 
272 	r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
273 	if (r) {
274 		dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
275 		return r;
276 	}
277 
278 	/* Enable IRQ */
279 	if (!rdev->irq.installed) {
280 		r = radeon_irq_kms_init(rdev);
281 		if (r)
282 			return r;
283 	}
284 
285 	r100_irq_set(rdev);
286 	rdev->config.r300.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
287 	/* 1M ring buffer */
288 	r = r100_cp_init(rdev, 1024 * 1024);
289 	if (r) {
290 		dev_err(rdev->dev, "failed initializing CP (%d).\n", r);
291 		return r;
292 	}
293 	r420_cp_errata_init(rdev);
294 
295 	r = radeon_ib_pool_init(rdev);
296 	if (r) {
297 		dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
298 		return r;
299 	}
300 
301 	return 0;
302 }
303 
304 int r420_resume(struct radeon_device *rdev)
305 {
306 	int r;
307 
308 	/* Make sur GART are not working */
309 	if (rdev->flags & RADEON_IS_PCIE)
310 		rv370_pcie_gart_disable(rdev);
311 	if (rdev->flags & RADEON_IS_PCI)
312 		r100_pci_gart_disable(rdev);
313 	/* Resume clock before doing reset */
314 	r420_clock_resume(rdev);
315 	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
316 	if (radeon_asic_reset(rdev)) {
317 		dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
318 			RREG32(R_000E40_RBBM_STATUS),
319 			RREG32(R_0007C0_CP_STAT));
320 	}
321 	/* check if cards are posted or not */
322 	if (rdev->is_atom_bios) {
323 		atom_asic_init(rdev->mode_info.atom_context);
324 	} else {
325 		radeon_combios_asic_init(rdev->ddev);
326 	}
327 	/* Resume clock after posting */
328 	r420_clock_resume(rdev);
329 	/* Initialize surface registers */
330 	radeon_surface_init(rdev);
331 
332 	rdev->accel_working = true;
333 	r = r420_startup(rdev);
334 	if (r) {
335 		rdev->accel_working = false;
336 	}
337 	return r;
338 }
339 
340 int r420_suspend(struct radeon_device *rdev)
341 {
342 	radeon_pm_suspend(rdev);
343 	r420_cp_errata_fini(rdev);
344 	r100_cp_disable(rdev);
345 	radeon_wb_disable(rdev);
346 	r100_irq_disable(rdev);
347 	if (rdev->flags & RADEON_IS_PCIE)
348 		rv370_pcie_gart_disable(rdev);
349 	if (rdev->flags & RADEON_IS_PCI)
350 		r100_pci_gart_disable(rdev);
351 	return 0;
352 }
353 
354 void r420_fini(struct radeon_device *rdev)
355 {
356 	radeon_pm_fini(rdev);
357 	r100_cp_fini(rdev);
358 	radeon_wb_fini(rdev);
359 	radeon_ib_pool_fini(rdev);
360 	radeon_gem_fini(rdev);
361 	if (rdev->flags & RADEON_IS_PCIE)
362 		rv370_pcie_gart_fini(rdev);
363 	if (rdev->flags & RADEON_IS_PCI)
364 		r100_pci_gart_fini(rdev);
365 	radeon_agp_fini(rdev);
366 	radeon_irq_kms_fini(rdev);
367 	radeon_fence_driver_fini(rdev);
368 	radeon_bo_fini(rdev);
369 	if (rdev->is_atom_bios) {
370 		radeon_atombios_fini(rdev);
371 	} else {
372 		radeon_combios_fini(rdev);
373 	}
374 	kfree(rdev->bios);
375 	rdev->bios = NULL;
376 }
377 
378 int r420_init(struct radeon_device *rdev)
379 {
380 	int r;
381 
382 	/* Initialize scratch registers */
383 	radeon_scratch_init(rdev);
384 	/* Initialize surface registers */
385 	radeon_surface_init(rdev);
386 	/* TODO: disable VGA need to use VGA request */
387 	/* restore some register to sane defaults */
388 	r100_restore_sanity(rdev);
389 	/* BIOS*/
390 	if (!radeon_get_bios(rdev)) {
391 		if (ASIC_IS_AVIVO(rdev))
392 			return -EINVAL;
393 	}
394 	if (rdev->is_atom_bios) {
395 		r = radeon_atombios_init(rdev);
396 		if (r) {
397 			return r;
398 		}
399 	} else {
400 		r = radeon_combios_init(rdev);
401 		if (r) {
402 			return r;
403 		}
404 	}
405 	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
406 	if (radeon_asic_reset(rdev)) {
407 		dev_warn(rdev->dev,
408 			"GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
409 			RREG32(R_000E40_RBBM_STATUS),
410 			RREG32(R_0007C0_CP_STAT));
411 	}
412 	/* check if cards are posted or not */
413 	if (radeon_boot_test_post_card(rdev) == false)
414 		return -EINVAL;
415 
416 	/* Initialize clocks */
417 	radeon_get_clock_info(rdev->ddev);
418 	/* initialize AGP */
419 	if (rdev->flags & RADEON_IS_AGP) {
420 		r = radeon_agp_init(rdev);
421 		if (r) {
422 			radeon_agp_disable(rdev);
423 		}
424 	}
425 	/* initialize memory controller */
426 	r300_mc_init(rdev);
427 	r420_debugfs(rdev);
428 	/* Fence driver */
429 	radeon_fence_driver_init(rdev);
430 	/* Memory manager */
431 	r = radeon_bo_init(rdev);
432 	if (r) {
433 		return r;
434 	}
435 	if (rdev->family == CHIP_R420)
436 		r100_enable_bm(rdev);
437 
438 	if (rdev->flags & RADEON_IS_PCIE) {
439 		r = rv370_pcie_gart_init(rdev);
440 		if (r)
441 			return r;
442 	}
443 	if (rdev->flags & RADEON_IS_PCI) {
444 		r = r100_pci_gart_init(rdev);
445 		if (r)
446 			return r;
447 	}
448 	r420_set_reg_safe(rdev);
449 
450 	/* Initialize power management */
451 	radeon_pm_init(rdev);
452 
453 	rdev->accel_working = true;
454 	r = r420_startup(rdev);
455 	if (r) {
456 		/* Somethings want wront with the accel init stop accel */
457 		dev_err(rdev->dev, "Disabling GPU acceleration\n");
458 		r100_cp_fini(rdev);
459 		radeon_wb_fini(rdev);
460 		radeon_ib_pool_fini(rdev);
461 		radeon_irq_kms_fini(rdev);
462 		if (rdev->flags & RADEON_IS_PCIE)
463 			rv370_pcie_gart_fini(rdev);
464 		if (rdev->flags & RADEON_IS_PCI)
465 			r100_pci_gart_fini(rdev);
466 		radeon_agp_fini(rdev);
467 		rdev->accel_working = false;
468 	}
469 	return 0;
470 }
471 
472 /*
473  * Debugfs info
474  */
475 #if defined(CONFIG_DEBUG_FS)
476 static int r420_debugfs_pipes_info_show(struct seq_file *m, void *unused)
477 {
478 	struct radeon_device *rdev = m->private;
479 	uint32_t tmp;
480 
481 	tmp = RREG32(R400_GB_PIPE_SELECT);
482 	seq_printf(m, "GB_PIPE_SELECT 0x%08x\n", tmp);
483 	tmp = RREG32(R300_GB_TILE_CONFIG);
484 	seq_printf(m, "GB_TILE_CONFIG 0x%08x\n", tmp);
485 	tmp = RREG32(R300_DST_PIPE_CONFIG);
486 	seq_printf(m, "DST_PIPE_CONFIG 0x%08x\n", tmp);
487 	return 0;
488 }
489 
490 DEFINE_SHOW_ATTRIBUTE(r420_debugfs_pipes_info);
491 #endif
492 
493 void r420_debugfs_pipes_info_init(struct radeon_device *rdev)
494 {
495 #if defined(CONFIG_DEBUG_FS)
496 	struct dentry *root = rdev->ddev->primary->debugfs_root;
497 
498 	debugfs_create_file("r420_pipes_info", 0444, root, rdev,
499 			    &r420_debugfs_pipes_info_fops);
500 #endif
501 }
502