xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/cz_ih.c (revision bef7a78d)
1 /*
2  * Copyright 2014 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #include <linux/pci.h>
25 
26 #include "amdgpu.h"
27 #include "amdgpu_ih.h"
28 #include "vid.h"
29 
30 #include "oss/oss_3_0_1_d.h"
31 #include "oss/oss_3_0_1_sh_mask.h"
32 
33 #include "bif/bif_5_1_d.h"
34 #include "bif/bif_5_1_sh_mask.h"
35 
36 /*
37  * Interrupts
38  * Starting with r6xx, interrupts are handled via a ring buffer.
39  * Ring buffers are areas of GPU accessible memory that the GPU
40  * writes interrupt vectors into and the host reads vectors out of.
41  * There is a rptr (read pointer) that determines where the
42  * host is currently reading, and a wptr (write pointer)
43  * which determines where the GPU has written.  When the
44  * pointers are equal, the ring is idle.  When the GPU
45  * writes vectors to the ring buffer, it increments the
46  * wptr.  When there is an interrupt, the host then starts
47  * fetching commands and processing them until the pointers are
48  * equal again at which point it updates the rptr.
49  */
50 
51 static void cz_ih_set_interrupt_funcs(struct amdgpu_device *adev);
52 
53 /**
54  * cz_ih_enable_interrupts - Enable the interrupt ring buffer
55  *
56  * @adev: amdgpu_device pointer
57  *
58  * Enable the interrupt ring buffer (VI).
59  */
60 static void cz_ih_enable_interrupts(struct amdgpu_device *adev)
61 {
62 	u32 ih_cntl = RREG32(mmIH_CNTL);
63 	u32 ih_rb_cntl = RREG32(mmIH_RB_CNTL);
64 
65 	ih_cntl = REG_SET_FIELD(ih_cntl, IH_CNTL, ENABLE_INTR, 1);
66 	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_ENABLE, 1);
67 	WREG32(mmIH_CNTL, ih_cntl);
68 	WREG32(mmIH_RB_CNTL, ih_rb_cntl);
69 	adev->irq.ih.enabled = true;
70 }
71 
72 /**
73  * cz_ih_disable_interrupts - Disable the interrupt ring buffer
74  *
75  * @adev: amdgpu_device pointer
76  *
77  * Disable the interrupt ring buffer (VI).
78  */
79 static void cz_ih_disable_interrupts(struct amdgpu_device *adev)
80 {
81 	u32 ih_rb_cntl = RREG32(mmIH_RB_CNTL);
82 	u32 ih_cntl = RREG32(mmIH_CNTL);
83 
84 	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_ENABLE, 0);
85 	ih_cntl = REG_SET_FIELD(ih_cntl, IH_CNTL, ENABLE_INTR, 0);
86 	WREG32(mmIH_RB_CNTL, ih_rb_cntl);
87 	WREG32(mmIH_CNTL, ih_cntl);
88 	/* set rptr, wptr to 0 */
89 	WREG32(mmIH_RB_RPTR, 0);
90 	WREG32(mmIH_RB_WPTR, 0);
91 	adev->irq.ih.enabled = false;
92 	adev->irq.ih.rptr = 0;
93 }
94 
95 /**
96  * cz_ih_irq_init - init and enable the interrupt ring
97  *
98  * @adev: amdgpu_device pointer
99  *
100  * Allocate a ring buffer for the interrupt controller,
101  * enable the RLC, disable interrupts, enable the IH
102  * ring buffer and enable it (VI).
103  * Called at device load and reume.
104  * Returns 0 for success, errors for failure.
105  */
106 static int cz_ih_irq_init(struct amdgpu_device *adev)
107 {
108 	struct amdgpu_ih_ring *ih = &adev->irq.ih;
109 	u32 interrupt_cntl, ih_cntl, ih_rb_cntl;
110 	int rb_bufsz;
111 
112 	/* disable irqs */
113 	cz_ih_disable_interrupts(adev);
114 
115 	/* setup interrupt control */
116 	WREG32(mmINTERRUPT_CNTL2, adev->dummy_page_addr >> 8);
117 	interrupt_cntl = RREG32(mmINTERRUPT_CNTL);
118 	/* INTERRUPT_CNTL__IH_DUMMY_RD_OVERRIDE_MASK=0 - dummy read disabled with msi, enabled without msi
119 	 * INTERRUPT_CNTL__IH_DUMMY_RD_OVERRIDE_MASK=1 - dummy read controlled by IH_DUMMY_RD_EN
120 	 */
121 	interrupt_cntl = REG_SET_FIELD(interrupt_cntl, INTERRUPT_CNTL, IH_DUMMY_RD_OVERRIDE, 0);
122 	/* INTERRUPT_CNTL__IH_REQ_NONSNOOP_EN_MASK=1 if ring is in non-cacheable memory, e.g., vram */
123 	interrupt_cntl = REG_SET_FIELD(interrupt_cntl, INTERRUPT_CNTL, IH_REQ_NONSNOOP_EN, 0);
124 	WREG32(mmINTERRUPT_CNTL, interrupt_cntl);
125 
126 	/* Ring Buffer base. [39:8] of 40-bit address of the beginning of the ring buffer*/
127 	WREG32(mmIH_RB_BASE, adev->irq.ih.gpu_addr >> 8);
128 
129 	rb_bufsz = order_base_2(adev->irq.ih.ring_size / 4);
130 	ih_rb_cntl = REG_SET_FIELD(0, IH_RB_CNTL, WPTR_OVERFLOW_ENABLE, 1);
131 	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1);
132 	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_SIZE, rb_bufsz);
133 
134 	/* Ring Buffer write pointer writeback. If enabled, IH_RB_WPTR register value is written to memory */
135 	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_WRITEBACK_ENABLE, 1);
136 
137 	/* set the writeback address whether it's enabled or not */
138 	WREG32(mmIH_RB_WPTR_ADDR_LO, lower_32_bits(ih->wptr_addr));
139 	WREG32(mmIH_RB_WPTR_ADDR_HI, upper_32_bits(ih->wptr_addr) & 0xFF);
140 
141 	WREG32(mmIH_RB_CNTL, ih_rb_cntl);
142 
143 	/* set rptr, wptr to 0 */
144 	WREG32(mmIH_RB_RPTR, 0);
145 	WREG32(mmIH_RB_WPTR, 0);
146 
147 	/* Default settings for IH_CNTL (disabled at first) */
148 	ih_cntl = RREG32(mmIH_CNTL);
149 	ih_cntl = REG_SET_FIELD(ih_cntl, IH_CNTL, MC_VMID, 0);
150 
151 	if (adev->irq.msi_enabled)
152 		ih_cntl = REG_SET_FIELD(ih_cntl, IH_CNTL, RPTR_REARM, 1);
153 	WREG32(mmIH_CNTL, ih_cntl);
154 
155 	pci_set_master(adev->pdev);
156 
157 	/* enable interrupts */
158 	cz_ih_enable_interrupts(adev);
159 
160 	return 0;
161 }
162 
163 /**
164  * cz_ih_irq_disable - disable interrupts
165  *
166  * @adev: amdgpu_device pointer
167  *
168  * Disable interrupts on the hw (VI).
169  */
170 static void cz_ih_irq_disable(struct amdgpu_device *adev)
171 {
172 	cz_ih_disable_interrupts(adev);
173 
174 	/* Wait and acknowledge irq */
175 	mdelay(1);
176 }
177 
178 /**
179  * cz_ih_get_wptr - get the IH ring buffer wptr
180  *
181  * @adev: amdgpu_device pointer
182  * @ih: IH ring buffer to fetch wptr
183  *
184  * Get the IH ring buffer wptr from either the register
185  * or the writeback memory buffer (VI).  Also check for
186  * ring buffer overflow and deal with it.
187  * Used by cz_irq_process(VI).
188  * Returns the value of the wptr.
189  */
190 static u32 cz_ih_get_wptr(struct amdgpu_device *adev,
191 			  struct amdgpu_ih_ring *ih)
192 {
193 	u32 wptr, tmp;
194 
195 	wptr = le32_to_cpu(*ih->wptr_cpu);
196 
197 	if (REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW)) {
198 		wptr = REG_SET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW, 0);
199 		/* When a ring buffer overflow happen start parsing interrupt
200 		 * from the last not overwritten vector (wptr + 16). Hopefully
201 		 * this should allow us to catchup.
202 		 */
203 		dev_warn(adev->dev, "IH ring buffer overflow (0x%08X, 0x%08X, 0x%08X)\n",
204 			wptr, ih->rptr, (wptr + 16) & ih->ptr_mask);
205 		ih->rptr = (wptr + 16) & ih->ptr_mask;
206 		tmp = RREG32(mmIH_RB_CNTL);
207 		tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1);
208 		WREG32(mmIH_RB_CNTL, tmp);
209 	}
210 	return (wptr & ih->ptr_mask);
211 }
212 
213 /**
214  * cz_ih_decode_iv - decode an interrupt vector
215  *
216  * @adev: amdgpu_device pointer
217  * @ih: IH ring buffer to decode
218  * @entry: IV entry to place decoded information into
219  *
220  * Decodes the interrupt vector at the current rptr
221  * position and also advance the position.
222  */
223 static void cz_ih_decode_iv(struct amdgpu_device *adev,
224 			    struct amdgpu_ih_ring *ih,
225 			    struct amdgpu_iv_entry *entry)
226 {
227 	/* wptr/rptr are in bytes! */
228 	u32 ring_index = ih->rptr >> 2;
229 	uint32_t dw[4];
230 
231 	dw[0] = le32_to_cpu(ih->ring[ring_index + 0]);
232 	dw[1] = le32_to_cpu(ih->ring[ring_index + 1]);
233 	dw[2] = le32_to_cpu(ih->ring[ring_index + 2]);
234 	dw[3] = le32_to_cpu(ih->ring[ring_index + 3]);
235 
236 	entry->client_id = AMDGPU_IRQ_CLIENTID_LEGACY;
237 	entry->src_id = dw[0] & 0xff;
238 	entry->src_data[0] = dw[1] & 0xfffffff;
239 	entry->ring_id = dw[2] & 0xff;
240 	entry->vmid = (dw[2] >> 8) & 0xff;
241 	entry->pasid = (dw[2] >> 16) & 0xffff;
242 
243 	/* wptr/rptr are in bytes! */
244 	ih->rptr += 16;
245 }
246 
247 /**
248  * cz_ih_set_rptr - set the IH ring buffer rptr
249  *
250  * @adev: amdgpu_device pointer
251  * @ih: IH ring buffer to set rptr
252  *
253  * Set the IH ring buffer rptr.
254  */
255 static void cz_ih_set_rptr(struct amdgpu_device *adev,
256 			   struct amdgpu_ih_ring *ih)
257 {
258 	WREG32(mmIH_RB_RPTR, ih->rptr);
259 }
260 
261 static int cz_ih_early_init(void *handle)
262 {
263 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
264 	int ret;
265 
266 	ret = amdgpu_irq_add_domain(adev);
267 	if (ret)
268 		return ret;
269 
270 	cz_ih_set_interrupt_funcs(adev);
271 
272 	return 0;
273 }
274 
275 static int cz_ih_sw_init(void *handle)
276 {
277 	int r;
278 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
279 
280 	r = amdgpu_ih_ring_init(adev, &adev->irq.ih, 64 * 1024, false);
281 	if (r)
282 		return r;
283 
284 	r = amdgpu_irq_init(adev);
285 
286 	return r;
287 }
288 
289 static int cz_ih_sw_fini(void *handle)
290 {
291 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
292 
293 	amdgpu_irq_fini(adev);
294 	amdgpu_ih_ring_fini(adev, &adev->irq.ih);
295 	amdgpu_irq_remove_domain(adev);
296 
297 	return 0;
298 }
299 
300 static int cz_ih_hw_init(void *handle)
301 {
302 	int r;
303 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
304 
305 	r = cz_ih_irq_init(adev);
306 	if (r)
307 		return r;
308 
309 	return 0;
310 }
311 
312 static int cz_ih_hw_fini(void *handle)
313 {
314 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
315 
316 	cz_ih_irq_disable(adev);
317 
318 	return 0;
319 }
320 
321 static int cz_ih_suspend(void *handle)
322 {
323 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
324 
325 	return cz_ih_hw_fini(adev);
326 }
327 
328 static int cz_ih_resume(void *handle)
329 {
330 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
331 
332 	return cz_ih_hw_init(adev);
333 }
334 
335 static bool cz_ih_is_idle(void *handle)
336 {
337 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
338 	u32 tmp = RREG32(mmSRBM_STATUS);
339 
340 	if (REG_GET_FIELD(tmp, SRBM_STATUS, IH_BUSY))
341 		return false;
342 
343 	return true;
344 }
345 
346 static int cz_ih_wait_for_idle(void *handle)
347 {
348 	unsigned i;
349 	u32 tmp;
350 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
351 
352 	for (i = 0; i < adev->usec_timeout; i++) {
353 		/* read MC_STATUS */
354 		tmp = RREG32(mmSRBM_STATUS);
355 		if (!REG_GET_FIELD(tmp, SRBM_STATUS, IH_BUSY))
356 			return 0;
357 		udelay(1);
358 	}
359 	return -ETIMEDOUT;
360 }
361 
362 static int cz_ih_soft_reset(void *handle)
363 {
364 	u32 srbm_soft_reset = 0;
365 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
366 	u32 tmp = RREG32(mmSRBM_STATUS);
367 
368 	if (tmp & SRBM_STATUS__IH_BUSY_MASK)
369 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET,
370 						SOFT_RESET_IH, 1);
371 
372 	if (srbm_soft_reset) {
373 		tmp = RREG32(mmSRBM_SOFT_RESET);
374 		tmp |= srbm_soft_reset;
375 		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
376 		WREG32(mmSRBM_SOFT_RESET, tmp);
377 		tmp = RREG32(mmSRBM_SOFT_RESET);
378 
379 		udelay(50);
380 
381 		tmp &= ~srbm_soft_reset;
382 		WREG32(mmSRBM_SOFT_RESET, tmp);
383 		tmp = RREG32(mmSRBM_SOFT_RESET);
384 
385 		/* Wait a little for things to settle down */
386 		udelay(50);
387 	}
388 
389 	return 0;
390 }
391 
392 static int cz_ih_set_clockgating_state(void *handle,
393 					  enum amd_clockgating_state state)
394 {
395 	// TODO
396 	return 0;
397 }
398 
399 static int cz_ih_set_powergating_state(void *handle,
400 					  enum amd_powergating_state state)
401 {
402 	// TODO
403 	return 0;
404 }
405 
406 static const struct amd_ip_funcs cz_ih_ip_funcs = {
407 	.name = "cz_ih",
408 	.early_init = cz_ih_early_init,
409 	.late_init = NULL,
410 	.sw_init = cz_ih_sw_init,
411 	.sw_fini = cz_ih_sw_fini,
412 	.hw_init = cz_ih_hw_init,
413 	.hw_fini = cz_ih_hw_fini,
414 	.suspend = cz_ih_suspend,
415 	.resume = cz_ih_resume,
416 	.is_idle = cz_ih_is_idle,
417 	.wait_for_idle = cz_ih_wait_for_idle,
418 	.soft_reset = cz_ih_soft_reset,
419 	.set_clockgating_state = cz_ih_set_clockgating_state,
420 	.set_powergating_state = cz_ih_set_powergating_state,
421 };
422 
423 static const struct amdgpu_ih_funcs cz_ih_funcs = {
424 	.get_wptr = cz_ih_get_wptr,
425 	.decode_iv = cz_ih_decode_iv,
426 	.set_rptr = cz_ih_set_rptr
427 };
428 
429 static void cz_ih_set_interrupt_funcs(struct amdgpu_device *adev)
430 {
431 	adev->irq.ih_funcs = &cz_ih_funcs;
432 }
433 
434 const struct amdgpu_ip_block_version cz_ih_ip_block =
435 {
436 	.type = AMD_IP_BLOCK_TYPE_IH,
437 	.major = 3,
438 	.minor = 0,
439 	.rev = 0,
440 	.funcs = &cz_ih_ip_funcs,
441 };
442