xref: /linux/drivers/gpu/drm/radeon/r420.c (revision 4ae68b26c3ab5a82aa271e6e9fc9b1a06e1d6b40)
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/pci.h>
30 #include <linux/seq_file.h>
31 #include <linux/slab.h>
32 
33 #include <drm/drm_device.h>
34 #include <drm/drm_file.h>
35 
36 #include "atom.h"
37 #include "r100d.h"
38 #include "r420_reg_safe.h"
39 #include "r420d.h"
40 #include "radeon.h"
41 #include "radeon_asic.h"
42 #include "radeon_reg.h"
43 
44 void r420_pm_init_profile(struct radeon_device *rdev)
45 {
46 	/* default */
47 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_ps_idx = rdev->pm.default_power_state_index;
48 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
49 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_off_cm_idx = 0;
50 	rdev->pm.profiles[PM_PROFILE_DEFAULT_IDX].dpms_on_cm_idx = 0;
51 	/* low sh */
52 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_ps_idx = 0;
53 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_ps_idx = 0;
54 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_off_cm_idx = 0;
55 	rdev->pm.profiles[PM_PROFILE_LOW_SH_IDX].dpms_on_cm_idx = 0;
56 	/* mid sh */
57 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_ps_idx = 0;
58 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_ps_idx = 1;
59 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_off_cm_idx = 0;
60 	rdev->pm.profiles[PM_PROFILE_MID_SH_IDX].dpms_on_cm_idx = 0;
61 	/* high sh */
62 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_ps_idx = 0;
63 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
64 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_off_cm_idx = 0;
65 	rdev->pm.profiles[PM_PROFILE_HIGH_SH_IDX].dpms_on_cm_idx = 0;
66 	/* low mh */
67 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_ps_idx = 0;
68 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
69 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_off_cm_idx = 0;
70 	rdev->pm.profiles[PM_PROFILE_LOW_MH_IDX].dpms_on_cm_idx = 0;
71 	/* mid mh */
72 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_ps_idx = 0;
73 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
74 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_off_cm_idx = 0;
75 	rdev->pm.profiles[PM_PROFILE_MID_MH_IDX].dpms_on_cm_idx = 0;
76 	/* high mh */
77 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_ps_idx = 0;
78 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_ps_idx = rdev->pm.default_power_state_index;
79 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_off_cm_idx = 0;
80 	rdev->pm.profiles[PM_PROFILE_HIGH_MH_IDX].dpms_on_cm_idx = 0;
81 }
82 
83 static void r420_set_reg_safe(struct radeon_device *rdev)
84 {
85 	rdev->config.r300.reg_safe_bm = r420_reg_safe_bm;
86 	rdev->config.r300.reg_safe_bm_size = ARRAY_SIZE(r420_reg_safe_bm);
87 }
88 
89 void r420_pipes_init(struct radeon_device *rdev)
90 {
91 	unsigned tmp;
92 	unsigned gb_pipe_select;
93 	unsigned num_pipes;
94 
95 	/* GA_ENHANCE workaround TCL deadlock issue */
96 	WREG32(R300_GA_ENHANCE, R300_GA_DEADLOCK_CNTL | R300_GA_FASTSYNC_CNTL |
97 	       (1 << 2) | (1 << 3));
98 	/* add idle wait as per freedesktop.org bug 24041 */
99 	if (r100_gui_wait_for_idle(rdev)) {
100 		pr_warn("Failed to wait GUI idle while programming pipes. Bad things might happen.\n");
101 	}
102 	/* get max number of pipes */
103 	gb_pipe_select = RREG32(R400_GB_PIPE_SELECT);
104 	num_pipes = ((gb_pipe_select >> 12) & 3) + 1;
105 
106 	/* SE chips have 1 pipe */
107 	if ((rdev->pdev->device == 0x5e4c) ||
108 	    (rdev->pdev->device == 0x5e4f))
109 		num_pipes = 1;
110 
111 	rdev->num_gb_pipes = num_pipes;
112 	tmp = 0;
113 	switch (num_pipes) {
114 	default:
115 		/* force to 1 pipe */
116 		num_pipes = 1;
117 		fallthrough;
118 	case 1:
119 		tmp = (0 << 1);
120 		break;
121 	case 2:
122 		tmp = (3 << 1);
123 		break;
124 	case 3:
125 		tmp = (6 << 1);
126 		break;
127 	case 4:
128 		tmp = (7 << 1);
129 		break;
130 	}
131 	WREG32(R500_SU_REG_DEST, (1 << num_pipes) - 1);
132 	/* Sub pixel 1/12 so we can have 4K rendering according to doc */
133 	tmp |= R300_TILE_SIZE_16 | R300_ENABLE_TILING;
134 	WREG32(R300_GB_TILE_CONFIG, tmp);
135 	if (r100_gui_wait_for_idle(rdev)) {
136 		pr_warn("Failed to wait GUI idle while programming pipes. Bad things might happen.\n");
137 	}
138 
139 	tmp = RREG32(R300_DST_PIPE_CONFIG);
140 	WREG32(R300_DST_PIPE_CONFIG, tmp | R300_PIPE_AUTO_CONFIG);
141 
142 	WREG32(R300_RB2D_DSTCACHE_MODE,
143 	       RREG32(R300_RB2D_DSTCACHE_MODE) |
144 	       R300_DC_AUTOFLUSH_ENABLE |
145 	       R300_DC_DC_DISABLE_IGNORE_PE);
146 
147 	if (r100_gui_wait_for_idle(rdev)) {
148 		pr_warn("Failed to wait GUI idle while programming pipes. Bad things might happen.\n");
149 	}
150 
151 	if (rdev->family == CHIP_RV530) {
152 		tmp = RREG32(RV530_GB_PIPE_SELECT2);
153 		if ((tmp & 3) == 3)
154 			rdev->num_z_pipes = 2;
155 		else
156 			rdev->num_z_pipes = 1;
157 	} else
158 		rdev->num_z_pipes = 1;
159 
160 	DRM_INFO("radeon: %d quad pipes, %d z pipes initialized.\n",
161 		 rdev->num_gb_pipes, rdev->num_z_pipes);
162 }
163 
164 u32 r420_mc_rreg(struct radeon_device *rdev, u32 reg)
165 {
166 	unsigned long flags;
167 	u32 r;
168 
169 	spin_lock_irqsave(&rdev->mc_idx_lock, flags);
170 	WREG32(R_0001F8_MC_IND_INDEX, S_0001F8_MC_IND_ADDR(reg));
171 	r = RREG32(R_0001FC_MC_IND_DATA);
172 	spin_unlock_irqrestore(&rdev->mc_idx_lock, flags);
173 	return r;
174 }
175 
176 void r420_mc_wreg(struct radeon_device *rdev, u32 reg, u32 v)
177 {
178 	unsigned long flags;
179 
180 	spin_lock_irqsave(&rdev->mc_idx_lock, flags);
181 	WREG32(R_0001F8_MC_IND_INDEX, S_0001F8_MC_IND_ADDR(reg) |
182 		S_0001F8_MC_IND_WR_EN(1));
183 	WREG32(R_0001FC_MC_IND_DATA, v);
184 	spin_unlock_irqrestore(&rdev->mc_idx_lock, flags);
185 }
186 
187 static void r420_debugfs(struct radeon_device *rdev)
188 {
189 	r100_debugfs_rbbm_init(rdev);
190 	r420_debugfs_pipes_info_init(rdev);
191 }
192 
193 static void r420_clock_resume(struct radeon_device *rdev)
194 {
195 	u32 sclk_cntl;
196 
197 	if (radeon_dynclks != -1 && radeon_dynclks)
198 		radeon_atom_set_clock_gating(rdev, 1);
199 	sclk_cntl = RREG32_PLL(R_00000D_SCLK_CNTL);
200 	sclk_cntl |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
201 	if (rdev->family == CHIP_R420)
202 		sclk_cntl |= S_00000D_FORCE_PX(1) | S_00000D_FORCE_TX(1);
203 	WREG32_PLL(R_00000D_SCLK_CNTL, sclk_cntl);
204 }
205 
206 static void r420_cp_errata_init(struct radeon_device *rdev)
207 {
208 	int r;
209 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
210 
211 	/* RV410 and R420 can lock up if CP DMA to host memory happens
212 	 * while the 2D engine is busy.
213 	 *
214 	 * The proper workaround is to queue a RESYNC at the beginning
215 	 * of the CP init, apparently.
216 	 */
217 	radeon_scratch_get(rdev, &rdev->config.r300.resync_scratch);
218 	r = radeon_ring_lock(rdev, ring, 8);
219 	WARN_ON(r);
220 	radeon_ring_write(ring, PACKET0(R300_CP_RESYNC_ADDR, 1));
221 	radeon_ring_write(ring, rdev->config.r300.resync_scratch);
222 	radeon_ring_write(ring, 0xDEADBEEF);
223 	radeon_ring_unlock_commit(rdev, ring, false);
224 }
225 
226 static void r420_cp_errata_fini(struct radeon_device *rdev)
227 {
228 	int r;
229 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
230 
231 	/* Catch the RESYNC we dispatched all the way back,
232 	 * at the very beginning of the CP init.
233 	 */
234 	r = radeon_ring_lock(rdev, ring, 8);
235 	WARN_ON(r);
236 	radeon_ring_write(ring, PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
237 	radeon_ring_write(ring, R300_RB3D_DC_FINISH);
238 	radeon_ring_unlock_commit(rdev, ring, false);
239 	radeon_scratch_free(rdev, rdev->config.r300.resync_scratch);
240 }
241 
242 static int r420_startup(struct radeon_device *rdev)
243 {
244 	int r;
245 
246 	/* set common regs */
247 	r100_set_common_regs(rdev);
248 	/* program mc */
249 	r300_mc_program(rdev);
250 	/* Resume clock */
251 	r420_clock_resume(rdev);
252 	/* Initialize GART (initialize after TTM so we can allocate
253 	 * memory through TTM but finalize after TTM) */
254 	if (rdev->flags & RADEON_IS_PCIE) {
255 		r = rv370_pcie_gart_enable(rdev);
256 		if (r)
257 			return r;
258 	}
259 	if (rdev->flags & RADEON_IS_PCI) {
260 		r = r100_pci_gart_enable(rdev);
261 		if (r)
262 			return r;
263 	}
264 	r420_pipes_init(rdev);
265 
266 	/* allocate wb buffer */
267 	r = radeon_wb_init(rdev);
268 	if (r)
269 		return r;
270 
271 	r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
272 	if (r) {
273 		dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
274 		return r;
275 	}
276 
277 	/* Enable IRQ */
278 	if (!rdev->irq.installed) {
279 		r = radeon_irq_kms_init(rdev);
280 		if (r)
281 			return r;
282 	}
283 
284 	r100_irq_set(rdev);
285 	rdev->config.r300.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
286 	/* 1M ring buffer */
287 	r = r100_cp_init(rdev, 1024 * 1024);
288 	if (r) {
289 		dev_err(rdev->dev, "failed initializing CP (%d).\n", r);
290 		return r;
291 	}
292 	r420_cp_errata_init(rdev);
293 
294 	r = radeon_ib_pool_init(rdev);
295 	if (r) {
296 		dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
297 		return r;
298 	}
299 
300 	return 0;
301 }
302 
303 int r420_resume(struct radeon_device *rdev)
304 {
305 	int r;
306 
307 	/* Make sur GART are not working */
308 	if (rdev->flags & RADEON_IS_PCIE)
309 		rv370_pcie_gart_disable(rdev);
310 	if (rdev->flags & RADEON_IS_PCI)
311 		r100_pci_gart_disable(rdev);
312 	/* Resume clock before doing reset */
313 	r420_clock_resume(rdev);
314 	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
315 	if (radeon_asic_reset(rdev)) {
316 		dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
317 			RREG32(R_000E40_RBBM_STATUS),
318 			RREG32(R_0007C0_CP_STAT));
319 	}
320 	/* check if cards are posted or not */
321 	if (rdev->is_atom_bios) {
322 		atom_asic_init(rdev->mode_info.atom_context);
323 	} else {
324 		radeon_combios_asic_init(rdev->ddev);
325 	}
326 	/* Resume clock after posting */
327 	r420_clock_resume(rdev);
328 	/* Initialize surface registers */
329 	radeon_surface_init(rdev);
330 
331 	rdev->accel_working = true;
332 	r = r420_startup(rdev);
333 	if (r) {
334 		rdev->accel_working = false;
335 	}
336 	return r;
337 }
338 
339 int r420_suspend(struct radeon_device *rdev)
340 {
341 	radeon_pm_suspend(rdev);
342 	r420_cp_errata_fini(rdev);
343 	r100_cp_disable(rdev);
344 	radeon_wb_disable(rdev);
345 	r100_irq_disable(rdev);
346 	if (rdev->flags & RADEON_IS_PCIE)
347 		rv370_pcie_gart_disable(rdev);
348 	if (rdev->flags & RADEON_IS_PCI)
349 		r100_pci_gart_disable(rdev);
350 	return 0;
351 }
352 
353 void r420_fini(struct radeon_device *rdev)
354 {
355 	radeon_pm_fini(rdev);
356 	r100_cp_fini(rdev);
357 	radeon_wb_fini(rdev);
358 	radeon_ib_pool_fini(rdev);
359 	radeon_gem_fini(rdev);
360 	if (rdev->flags & RADEON_IS_PCIE)
361 		rv370_pcie_gart_fini(rdev);
362 	if (rdev->flags & RADEON_IS_PCI)
363 		r100_pci_gart_fini(rdev);
364 	radeon_agp_fini(rdev);
365 	radeon_irq_kms_fini(rdev);
366 	radeon_fence_driver_fini(rdev);
367 	radeon_bo_fini(rdev);
368 	if (rdev->is_atom_bios) {
369 		radeon_atombios_fini(rdev);
370 	} else {
371 		radeon_combios_fini(rdev);
372 	}
373 	kfree(rdev->bios);
374 	rdev->bios = NULL;
375 }
376 
377 int r420_init(struct radeon_device *rdev)
378 {
379 	int r;
380 
381 	/* Initialize scratch registers */
382 	radeon_scratch_init(rdev);
383 	/* Initialize surface registers */
384 	radeon_surface_init(rdev);
385 	/* TODO: disable VGA need to use VGA request */
386 	/* restore some register to sane defaults */
387 	r100_restore_sanity(rdev);
388 	/* BIOS*/
389 	if (!radeon_get_bios(rdev)) {
390 		if (ASIC_IS_AVIVO(rdev))
391 			return -EINVAL;
392 	}
393 	if (rdev->is_atom_bios) {
394 		r = radeon_atombios_init(rdev);
395 		if (r) {
396 			return r;
397 		}
398 	} else {
399 		r = radeon_combios_init(rdev);
400 		if (r) {
401 			return r;
402 		}
403 	}
404 	/* Reset gpu before posting otherwise ATOM will enter infinite loop */
405 	if (radeon_asic_reset(rdev)) {
406 		dev_warn(rdev->dev,
407 			"GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
408 			RREG32(R_000E40_RBBM_STATUS),
409 			RREG32(R_0007C0_CP_STAT));
410 	}
411 	/* check if cards are posted or not */
412 	if (radeon_boot_test_post_card(rdev) == false)
413 		return -EINVAL;
414 
415 	/* Initialize clocks */
416 	radeon_get_clock_info(rdev->ddev);
417 	/* initialize AGP */
418 	if (rdev->flags & RADEON_IS_AGP) {
419 		r = radeon_agp_init(rdev);
420 		if (r) {
421 			radeon_agp_disable(rdev);
422 		}
423 	}
424 	/* initialize memory controller */
425 	r300_mc_init(rdev);
426 	r420_debugfs(rdev);
427 	/* Fence driver */
428 	radeon_fence_driver_init(rdev);
429 	/* Memory manager */
430 	r = radeon_bo_init(rdev);
431 	if (r) {
432 		return r;
433 	}
434 	if (rdev->family == CHIP_R420)
435 		r100_enable_bm(rdev);
436 
437 	if (rdev->flags & RADEON_IS_PCIE) {
438 		r = rv370_pcie_gart_init(rdev);
439 		if (r)
440 			return r;
441 	}
442 	if (rdev->flags & RADEON_IS_PCI) {
443 		r = r100_pci_gart_init(rdev);
444 		if (r)
445 			return r;
446 	}
447 	r420_set_reg_safe(rdev);
448 
449 	/* Initialize power management */
450 	radeon_pm_init(rdev);
451 
452 	rdev->accel_working = true;
453 	r = r420_startup(rdev);
454 	if (r) {
455 		/* Somethings want wront with the accel init stop accel */
456 		dev_err(rdev->dev, "Disabling GPU acceleration\n");
457 		r100_cp_fini(rdev);
458 		radeon_wb_fini(rdev);
459 		radeon_ib_pool_fini(rdev);
460 		radeon_irq_kms_fini(rdev);
461 		if (rdev->flags & RADEON_IS_PCIE)
462 			rv370_pcie_gart_fini(rdev);
463 		if (rdev->flags & RADEON_IS_PCI)
464 			r100_pci_gart_fini(rdev);
465 		radeon_agp_fini(rdev);
466 		rdev->accel_working = false;
467 	}
468 	return 0;
469 }
470 
471 /*
472  * Debugfs info
473  */
474 #if defined(CONFIG_DEBUG_FS)
475 static int r420_debugfs_pipes_info_show(struct seq_file *m, void *unused)
476 {
477 	struct radeon_device *rdev = m->private;
478 	uint32_t tmp;
479 
480 	tmp = RREG32(R400_GB_PIPE_SELECT);
481 	seq_printf(m, "GB_PIPE_SELECT 0x%08x\n", tmp);
482 	tmp = RREG32(R300_GB_TILE_CONFIG);
483 	seq_printf(m, "GB_TILE_CONFIG 0x%08x\n", tmp);
484 	tmp = RREG32(R300_DST_PIPE_CONFIG);
485 	seq_printf(m, "DST_PIPE_CONFIG 0x%08x\n", tmp);
486 	return 0;
487 }
488 
489 DEFINE_SHOW_ATTRIBUTE(r420_debugfs_pipes_info);
490 #endif
491 
492 void r420_debugfs_pipes_info_init(struct radeon_device *rdev)
493 {
494 #if defined(CONFIG_DEBUG_FS)
495 	struct dentry *root = rdev->ddev->primary->debugfs_root;
496 
497 	debugfs_create_file("r420_pipes_info", 0444, root, rdev,
498 			    &r420_debugfs_pipes_info_fops);
499 #endif
500 }
501