1 /*
2  * Copyright 2016 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 #include "drmP.h"
24 #include "amdgpu.h"
25 #include "amdgpu_ih.h"
26 #include "soc15.h"
27 
28 
29 #include "vega10/soc15ip.h"
30 #include "vega10/OSSSYS/osssys_4_0_offset.h"
31 #include "vega10/OSSSYS/osssys_4_0_sh_mask.h"
32 
33 #include "soc15_common.h"
34 #include "vega10_ih.h"
35 
36 
37 
38 static void vega10_ih_set_interrupt_funcs(struct amdgpu_device *adev);
39 
40 /**
41  * vega10_ih_enable_interrupts - Enable the interrupt ring buffer
42  *
43  * @adev: amdgpu_device pointer
44  *
45  * Enable the interrupt ring buffer (VEGA10).
46  */
47 static void vega10_ih_enable_interrupts(struct amdgpu_device *adev)
48 {
49 	u32 ih_rb_cntl = RREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL));
50 
51 	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_ENABLE, 1);
52 	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, ENABLE_INTR, 1);
53 	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL), ih_rb_cntl);
54 	adev->irq.ih.enabled = true;
55 }
56 
57 /**
58  * vega10_ih_disable_interrupts - Disable the interrupt ring buffer
59  *
60  * @adev: amdgpu_device pointer
61  *
62  * Disable the interrupt ring buffer (VEGA10).
63  */
64 static void vega10_ih_disable_interrupts(struct amdgpu_device *adev)
65 {
66 	u32 ih_rb_cntl = RREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL));
67 
68 	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_ENABLE, 0);
69 	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, ENABLE_INTR, 0);
70 	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL), ih_rb_cntl);
71 	/* set rptr, wptr to 0 */
72 	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_RPTR), 0);
73 	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR), 0);
74 	adev->irq.ih.enabled = false;
75 	adev->irq.ih.rptr = 0;
76 }
77 
78 /**
79  * vega10_ih_irq_init - init and enable the interrupt ring
80  *
81  * @adev: amdgpu_device pointer
82  *
83  * Allocate a ring buffer for the interrupt controller,
84  * enable the RLC, disable interrupts, enable the IH
85  * ring buffer and enable it (VI).
86  * Called at device load and reume.
87  * Returns 0 for success, errors for failure.
88  */
89 static int vega10_ih_irq_init(struct amdgpu_device *adev)
90 {
91 	int ret = 0;
92 	int rb_bufsz;
93 	u32 ih_rb_cntl, ih_doorbell_rtpr;
94 	u32 tmp;
95 	u64 wptr_off;
96 
97 	/* disable irqs */
98 	vega10_ih_disable_interrupts(adev);
99 
100 	nbio_v6_1_ih_control(adev);
101 
102 	ih_rb_cntl = RREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL));
103 	/* Ring Buffer base. [39:8] of 40-bit address of the beginning of the ring buffer*/
104 	if (adev->irq.ih.use_bus_addr) {
105 		WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_BASE), adev->irq.ih.rb_dma_addr >> 8);
106 		WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_BASE_HI), ((u64)adev->irq.ih.rb_dma_addr >> 40) & 0xff);
107 		ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_SPACE, 1);
108 	} else {
109 		WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_BASE), adev->irq.ih.gpu_addr >> 8);
110 		WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_BASE_HI), (adev->irq.ih.gpu_addr >> 40) & 0xff);
111 		ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_SPACE, 4);
112 	}
113 	rb_bufsz = order_base_2(adev->irq.ih.ring_size / 4);
114 	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1);
115 	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_OVERFLOW_ENABLE, 1);
116 	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RB_SIZE, rb_bufsz);
117 	/* Ring Buffer write pointer writeback. If enabled, IH_RB_WPTR register value is written to memory */
118 	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_WRITEBACK_ENABLE, 1);
119 	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_SNOOP, 1);
120 	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_RO, 0);
121 	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_VMID, 0);
122 
123 	if (adev->irq.msi_enabled)
124 		ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, RPTR_REARM, 1);
125 
126 	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL), ih_rb_cntl);
127 
128 	/* set the writeback address whether it's enabled or not */
129 	if (adev->irq.ih.use_bus_addr)
130 		wptr_off = adev->irq.ih.rb_dma_addr + (adev->irq.ih.wptr_offs * 4);
131 	else
132 		wptr_off = adev->wb.gpu_addr + (adev->irq.ih.wptr_offs * 4);
133 	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR_ADDR_LO), lower_32_bits(wptr_off));
134 	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR_ADDR_HI), upper_32_bits(wptr_off) & 0xFF);
135 
136 	/* set rptr, wptr to 0 */
137 	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_RPTR), 0);
138 	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_WPTR), 0);
139 
140 	ih_doorbell_rtpr = RREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_DOORBELL_RPTR));
141 	if (adev->irq.ih.use_doorbell) {
142 		ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, IH_DOORBELL_RPTR,
143 						 OFFSET, adev->irq.ih.doorbell_index);
144 		ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, IH_DOORBELL_RPTR,
145 						 ENABLE, 1);
146 	} else {
147 		ih_doorbell_rtpr = REG_SET_FIELD(ih_doorbell_rtpr, IH_DOORBELL_RPTR,
148 						 ENABLE, 0);
149 	}
150 	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_DOORBELL_RPTR), ih_doorbell_rtpr);
151 	nbio_v6_1_ih_doorbell_range(adev, adev->irq.ih.use_doorbell, adev->irq.ih.doorbell_index);
152 
153 	tmp = RREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_STORM_CLIENT_LIST_CNTL));
154 	tmp = REG_SET_FIELD(tmp, IH_STORM_CLIENT_LIST_CNTL,
155 			    CLIENT18_IS_STORM_CLIENT, 1);
156 	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_STORM_CLIENT_LIST_CNTL), tmp);
157 
158 	tmp = RREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_INT_FLOOD_CNTL));
159 	tmp = REG_SET_FIELD(tmp, IH_INT_FLOOD_CNTL, FLOOD_CNTL_ENABLE, 1);
160 	WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_INT_FLOOD_CNTL), tmp);
161 
162 	pci_set_master(adev->pdev);
163 
164 	/* enable interrupts */
165 	vega10_ih_enable_interrupts(adev);
166 
167 	return ret;
168 }
169 
170 /**
171  * vega10_ih_irq_disable - disable interrupts
172  *
173  * @adev: amdgpu_device pointer
174  *
175  * Disable interrupts on the hw (VEGA10).
176  */
177 static void vega10_ih_irq_disable(struct amdgpu_device *adev)
178 {
179 	vega10_ih_disable_interrupts(adev);
180 
181 	/* Wait and acknowledge irq */
182 	mdelay(1);
183 }
184 
185 /**
186  * vega10_ih_get_wptr - get the IH ring buffer wptr
187  *
188  * @adev: amdgpu_device pointer
189  *
190  * Get the IH ring buffer wptr from either the register
191  * or the writeback memory buffer (VEGA10).  Also check for
192  * ring buffer overflow and deal with it.
193  * Returns the value of the wptr.
194  */
195 static u32 vega10_ih_get_wptr(struct amdgpu_device *adev)
196 {
197 	u32 wptr, tmp;
198 
199 	if (adev->irq.ih.use_bus_addr)
200 		wptr = le32_to_cpu(adev->irq.ih.ring[adev->irq.ih.wptr_offs]);
201 	else
202 		wptr = le32_to_cpu(adev->wb.wb[adev->irq.ih.wptr_offs]);
203 
204 	if (REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW)) {
205 		wptr = REG_SET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW, 0);
206 
207 		/* When a ring buffer overflow happen start parsing interrupt
208 		 * from the last not overwritten vector (wptr + 32). Hopefully
209 		 * this should allow us to catchup.
210 		 */
211 		tmp = (wptr + 32) & adev->irq.ih.ptr_mask;
212 		dev_warn(adev->dev, "IH ring buffer overflow (0x%08X, 0x%08X, 0x%08X)\n",
213 			wptr, adev->irq.ih.rptr, tmp);
214 		adev->irq.ih.rptr = tmp;
215 
216 		tmp = RREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL));
217 		tmp = REG_SET_FIELD(tmp, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1);
218 		WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_CNTL), tmp);
219 	}
220 	return (wptr & adev->irq.ih.ptr_mask);
221 }
222 
223 /**
224  * vega10_ih_decode_iv - decode an interrupt vector
225  *
226  * @adev: amdgpu_device pointer
227  *
228  * Decodes the interrupt vector at the current rptr
229  * position and also advance the position.
230  */
231 static void vega10_ih_decode_iv(struct amdgpu_device *adev,
232 				 struct amdgpu_iv_entry *entry)
233 {
234 	/* wptr/rptr are in bytes! */
235 	u32 ring_index = adev->irq.ih.rptr >> 2;
236 	uint32_t dw[8];
237 
238 	dw[0] = le32_to_cpu(adev->irq.ih.ring[ring_index + 0]);
239 	dw[1] = le32_to_cpu(adev->irq.ih.ring[ring_index + 1]);
240 	dw[2] = le32_to_cpu(adev->irq.ih.ring[ring_index + 2]);
241 	dw[3] = le32_to_cpu(adev->irq.ih.ring[ring_index + 3]);
242 	dw[4] = le32_to_cpu(adev->irq.ih.ring[ring_index + 4]);
243 	dw[5] = le32_to_cpu(adev->irq.ih.ring[ring_index + 5]);
244 	dw[6] = le32_to_cpu(adev->irq.ih.ring[ring_index + 6]);
245 	dw[7] = le32_to_cpu(adev->irq.ih.ring[ring_index + 7]);
246 
247 	entry->client_id = dw[0] & 0xff;
248 	entry->src_id = (dw[0] >> 8) & 0xff;
249 	entry->ring_id = (dw[0] >> 16) & 0xff;
250 	entry->vm_id = (dw[0] >> 24) & 0xf;
251 	entry->vm_id_src = (dw[0] >> 31);
252 	entry->timestamp = dw[1] | ((u64)(dw[2] & 0xffff) << 32);
253 	entry->timestamp_src = dw[2] >> 31;
254 	entry->pas_id = dw[3] & 0xffff;
255 	entry->pasid_src = dw[3] >> 31;
256 	entry->src_data[0] = dw[4];
257 	entry->src_data[1] = dw[5];
258 	entry->src_data[2] = dw[6];
259 	entry->src_data[3] = dw[7];
260 
261 
262 	/* wptr/rptr are in bytes! */
263 	adev->irq.ih.rptr += 32;
264 }
265 
266 /**
267  * vega10_ih_set_rptr - set the IH ring buffer rptr
268  *
269  * @adev: amdgpu_device pointer
270  *
271  * Set the IH ring buffer rptr.
272  */
273 static void vega10_ih_set_rptr(struct amdgpu_device *adev)
274 {
275 	if (adev->irq.ih.use_doorbell) {
276 		/* XXX check if swapping is necessary on BE */
277 		if (adev->irq.ih.use_bus_addr)
278 			adev->irq.ih.ring[adev->irq.ih.rptr_offs] = adev->irq.ih.rptr;
279 		else
280 			adev->wb.wb[adev->irq.ih.rptr_offs] = adev->irq.ih.rptr;
281 		WDOORBELL32(adev->irq.ih.doorbell_index, adev->irq.ih.rptr);
282 	} else {
283 		WREG32(SOC15_REG_OFFSET(OSSSYS, 0, mmIH_RB_RPTR), adev->irq.ih.rptr);
284 	}
285 }
286 
287 static int vega10_ih_early_init(void *handle)
288 {
289 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
290 
291 	vega10_ih_set_interrupt_funcs(adev);
292 	return 0;
293 }
294 
295 static int vega10_ih_sw_init(void *handle)
296 {
297 	int r;
298 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
299 
300 	r = amdgpu_ih_ring_init(adev, 256 * 1024, true);
301 	if (r)
302 		return r;
303 
304 	adev->irq.ih.use_doorbell = true;
305 	adev->irq.ih.doorbell_index = AMDGPU_DOORBELL64_IH << 1;
306 
307 	r = amdgpu_irq_init(adev);
308 
309 	return r;
310 }
311 
312 static int vega10_ih_sw_fini(void *handle)
313 {
314 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
315 
316 	amdgpu_irq_fini(adev);
317 	amdgpu_ih_ring_fini(adev);
318 
319 	return 0;
320 }
321 
322 static int vega10_ih_hw_init(void *handle)
323 {
324 	int r;
325 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
326 
327 	r = vega10_ih_irq_init(adev);
328 	if (r)
329 		return r;
330 
331 	return 0;
332 }
333 
334 static int vega10_ih_hw_fini(void *handle)
335 {
336 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
337 
338 	vega10_ih_irq_disable(adev);
339 
340 	return 0;
341 }
342 
343 static int vega10_ih_suspend(void *handle)
344 {
345 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
346 
347 	return vega10_ih_hw_fini(adev);
348 }
349 
350 static int vega10_ih_resume(void *handle)
351 {
352 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
353 
354 	return vega10_ih_hw_init(adev);
355 }
356 
357 static bool vega10_ih_is_idle(void *handle)
358 {
359 	/* todo */
360 	return true;
361 }
362 
363 static int vega10_ih_wait_for_idle(void *handle)
364 {
365 	/* todo */
366 	return -ETIMEDOUT;
367 }
368 
369 static int vega10_ih_soft_reset(void *handle)
370 {
371 	/* todo */
372 
373 	return 0;
374 }
375 
376 static int vega10_ih_set_clockgating_state(void *handle,
377 					  enum amd_clockgating_state state)
378 {
379 	return 0;
380 }
381 
382 static int vega10_ih_set_powergating_state(void *handle,
383 					  enum amd_powergating_state state)
384 {
385 	return 0;
386 }
387 
388 const struct amd_ip_funcs vega10_ih_ip_funcs = {
389 	.name = "vega10_ih",
390 	.early_init = vega10_ih_early_init,
391 	.late_init = NULL,
392 	.sw_init = vega10_ih_sw_init,
393 	.sw_fini = vega10_ih_sw_fini,
394 	.hw_init = vega10_ih_hw_init,
395 	.hw_fini = vega10_ih_hw_fini,
396 	.suspend = vega10_ih_suspend,
397 	.resume = vega10_ih_resume,
398 	.is_idle = vega10_ih_is_idle,
399 	.wait_for_idle = vega10_ih_wait_for_idle,
400 	.soft_reset = vega10_ih_soft_reset,
401 	.set_clockgating_state = vega10_ih_set_clockgating_state,
402 	.set_powergating_state = vega10_ih_set_powergating_state,
403 };
404 
405 static const struct amdgpu_ih_funcs vega10_ih_funcs = {
406 	.get_wptr = vega10_ih_get_wptr,
407 	.decode_iv = vega10_ih_decode_iv,
408 	.set_rptr = vega10_ih_set_rptr
409 };
410 
411 static void vega10_ih_set_interrupt_funcs(struct amdgpu_device *adev)
412 {
413 	if (adev->irq.ih_funcs == NULL)
414 		adev->irq.ih_funcs = &vega10_ih_funcs;
415 }
416 
417 const struct amdgpu_ip_block_version vega10_ih_ip_block =
418 {
419 	.type = AMD_IP_BLOCK_TYPE_IH,
420 	.major = 4,
421 	.minor = 0,
422 	.rev = 0,
423 	.funcs = &vega10_ih_ip_funcs,
424 };
425