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