]>
Commit | Line | Data |
---|---|---|
a2e73f56 AD |
1 | /* |
2 | * Copyright 2012 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 | */ | |
47b757fb SR |
23 | |
24 | #include <linux/pci.h> | |
25 | ||
a2e73f56 AD |
26 | #include "amdgpu.h" |
27 | #include "amdgpu_ih.h" | |
28 | #include "cikd.h" | |
29 | ||
30 | #include "bif/bif_4_1_d.h" | |
31 | #include "bif/bif_4_1_sh_mask.h" | |
32 | ||
33 | #include "oss/oss_2_0_d.h" | |
34 | #include "oss/oss_2_0_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 cik_ih_set_interrupt_funcs(struct amdgpu_device *adev); | |
52 | ||
53 | /** | |
54 | * cik_ih_enable_interrupts - Enable the interrupt ring buffer | |
55 | * | |
56 | * @adev: amdgpu_device pointer | |
57 | * | |
58 | * Enable the interrupt ring buffer (CIK). | |
59 | */ | |
60 | static void cik_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 |= IH_CNTL__ENABLE_INTR_MASK; | |
66 | ih_rb_cntl |= IH_RB_CNTL__RB_ENABLE_MASK; | |
67 | WREG32(mmIH_CNTL, ih_cntl); | |
68 | WREG32(mmIH_RB_CNTL, ih_rb_cntl); | |
69 | adev->irq.ih.enabled = true; | |
70 | } | |
71 | ||
72 | /** | |
73 | * cik_ih_disable_interrupts - Disable the interrupt ring buffer | |
74 | * | |
75 | * @adev: amdgpu_device pointer | |
76 | * | |
77 | * Disable the interrupt ring buffer (CIK). | |
78 | */ | |
79 | static void cik_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 &= ~IH_RB_CNTL__RB_ENABLE_MASK; | |
85 | ih_cntl &= ~IH_CNTL__ENABLE_INTR_MASK; | |
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 | * cik_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 (CIK). | |
103 | * Called at device load and reume. | |
104 | * Returns 0 for success, errors for failure. | |
105 | */ | |
106 | static int cik_ih_irq_init(struct amdgpu_device *adev) | |
107 | { | |
d81f78b4 | 108 | struct amdgpu_ih_ring *ih = &adev->irq.ih; |
a2e73f56 AD |
109 | int rb_bufsz; |
110 | u32 interrupt_cntl, ih_cntl, ih_rb_cntl; | |
a2e73f56 AD |
111 | |
112 | /* disable irqs */ | |
113 | cik_ih_disable_interrupts(adev); | |
114 | ||
115 | /* setup interrupt control */ | |
92e71b06 | 116 | WREG32(mmINTERRUPT_CNTL2, adev->dummy_page_addr >> 8); |
a2e73f56 AD |
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 &= ~INTERRUPT_CNTL__IH_DUMMY_RD_OVERRIDE_MASK; | |
122 | /* INTERRUPT_CNTL__IH_REQ_NONSNOOP_EN_MASK=1 if ring is in non-cacheable memory, e.g., vram */ | |
123 | interrupt_cntl &= ~INTERRUPT_CNTL__IH_REQ_NONSNOOP_EN_MASK; | |
124 | WREG32(mmINTERRUPT_CNTL, interrupt_cntl); | |
125 | ||
126 | WREG32(mmIH_RB_BASE, adev->irq.ih.gpu_addr >> 8); | |
127 | rb_bufsz = order_base_2(adev->irq.ih.ring_size / 4); | |
128 | ||
129 | ih_rb_cntl = (IH_RB_CNTL__WPTR_OVERFLOW_ENABLE_MASK | | |
130 | IH_RB_CNTL__WPTR_OVERFLOW_CLEAR_MASK | | |
131 | (rb_bufsz << 1)); | |
132 | ||
133 | ih_rb_cntl |= IH_RB_CNTL__WPTR_WRITEBACK_ENABLE_MASK; | |
134 | ||
135 | /* set the writeback address whether it's enabled or not */ | |
d81f78b4 CK |
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); | |
a2e73f56 AD |
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 = (0x10 << IH_CNTL__MC_WRREQ_CREDIT__SHIFT) | | |
147 | (0x10 << IH_CNTL__MC_WR_CLEAN_CNT__SHIFT) | | |
148 | (0 << IH_CNTL__MC_VMID__SHIFT); | |
149 | /* IH_CNTL__RPTR_REARM_MASK only works if msi's are enabled */ | |
150 | if (adev->irq.msi_enabled) | |
151 | ih_cntl |= IH_CNTL__RPTR_REARM_MASK; | |
152 | WREG32(mmIH_CNTL, ih_cntl); | |
153 | ||
154 | pci_set_master(adev->pdev); | |
155 | ||
156 | /* enable irqs */ | |
157 | cik_ih_enable_interrupts(adev); | |
158 | ||
734711b7 | 159 | return 0; |
a2e73f56 AD |
160 | } |
161 | ||
162 | /** | |
163 | * cik_ih_irq_disable - disable interrupts | |
164 | * | |
165 | * @adev: amdgpu_device pointer | |
166 | * | |
167 | * Disable interrupts on the hw (CIK). | |
168 | */ | |
169 | static void cik_ih_irq_disable(struct amdgpu_device *adev) | |
170 | { | |
171 | cik_ih_disable_interrupts(adev); | |
172 | /* Wait and acknowledge irq */ | |
173 | mdelay(1); | |
174 | } | |
175 | ||
176 | /** | |
177 | * cik_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 (CIK). Also check for | |
183 | * ring buffer overflow and deal with it. | |
184 | * Used by cik_irq_process(). | |
185 | * Returns the value of the wptr. | |
186 | */ | |
8bb9eb48 CK |
187 | static u32 cik_ih_get_wptr(struct amdgpu_device *adev, |
188 | struct amdgpu_ih_ring *ih) | |
a2e73f56 AD |
189 | { |
190 | u32 wptr, tmp; | |
191 | ||
d81f78b4 | 192 | wptr = le32_to_cpu(*ih->wptr_cpu); |
a2e73f56 AD |
193 | |
194 | if (wptr & IH_RB_WPTR__RB_OVERFLOW_MASK) { | |
195 | wptr &= ~IH_RB_WPTR__RB_OVERFLOW_MASK; | |
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", | |
8bb9eb48 CK |
201 | wptr, ih->rptr, (wptr + 16) & ih->ptr_mask); |
202 | ih->rptr = (wptr + 16) & ih->ptr_mask; | |
a2e73f56 AD |
203 | tmp = RREG32(mmIH_RB_CNTL); |
204 | tmp |= IH_RB_CNTL__WPTR_OVERFLOW_CLEAR_MASK; | |
205 | WREG32(mmIH_RB_CNTL, tmp); | |
206 | } | |
8bb9eb48 | 207 | return (wptr & ih->ptr_mask); |
a2e73f56 AD |
208 | } |
209 | ||
210 | /* CIK IV Ring | |
211 | * Each IV ring entry is 128 bits: | |
212 | * [7:0] - interrupt source id | |
213 | * [31:8] - reserved | |
214 | * [59:32] - interrupt source data | |
215 | * [63:60] - reserved | |
216 | * [71:64] - RINGID | |
217 | * CP: | |
218 | * ME_ID [1:0], PIPE_ID[1:0], QUEUE_ID[2:0] | |
219 | * QUEUE_ID - for compute, which of the 8 queues owned by the dispatcher | |
220 | * - for gfx, hw shader state (0=PS...5=LS, 6=CS) | |
221 | * ME_ID - 0 = gfx, 1 = first 4 CS pipes, 2 = second 4 CS pipes | |
222 | * PIPE_ID - ME0 0=3D | |
223 | * - ME1&2 compute dispatcher (4 pipes each) | |
224 | * SDMA: | |
225 | * INSTANCE_ID [1:0], QUEUE_ID[1:0] | |
226 | * INSTANCE_ID - 0 = sdma0, 1 = sdma1 | |
227 | * QUEUE_ID - 0 = gfx, 1 = rlc0, 2 = rlc1 | |
228 | * [79:72] - VMID | |
229 | * [95:80] - PASID | |
230 | * [127:96] - reserved | |
231 | */ | |
232 | ||
233 | /** | |
234 | * cik_ih_decode_iv - decode an interrupt vector | |
235 | * | |
236 | * @adev: amdgpu_device pointer | |
237 | * | |
238 | * Decodes the interrupt vector at the current rptr | |
239 | * position and also advance the position. | |
240 | */ | |
241 | static void cik_ih_decode_iv(struct amdgpu_device *adev, | |
8bb9eb48 | 242 | struct amdgpu_ih_ring *ih, |
a2e73f56 AD |
243 | struct amdgpu_iv_entry *entry) |
244 | { | |
245 | /* wptr/rptr are in bytes! */ | |
8bb9eb48 | 246 | u32 ring_index = ih->rptr >> 2; |
a2e73f56 | 247 | uint32_t dw[4]; |
edf600da | 248 | |
8bb9eb48 CK |
249 | dw[0] = le32_to_cpu(ih->ring[ring_index + 0]); |
250 | dw[1] = le32_to_cpu(ih->ring[ring_index + 1]); | |
251 | dw[2] = le32_to_cpu(ih->ring[ring_index + 2]); | |
252 | dw[3] = le32_to_cpu(ih->ring[ring_index + 3]); | |
a2e73f56 | 253 | |
1ffdeca6 | 254 | entry->client_id = AMDGPU_IRQ_CLIENTID_LEGACY; |
a2e73f56 | 255 | entry->src_id = dw[0] & 0xff; |
7ccf5aa8 | 256 | entry->src_data[0] = dw[1] & 0xfffffff; |
a2e73f56 | 257 | entry->ring_id = dw[2] & 0xff; |
c4f46f22 | 258 | entry->vmid = (dw[2] >> 8) & 0xff; |
3816e42f | 259 | entry->pasid = (dw[2] >> 16) & 0xffff; |
a2e73f56 AD |
260 | |
261 | /* wptr/rptr are in bytes! */ | |
8bb9eb48 | 262 | ih->rptr += 16; |
a2e73f56 AD |
263 | } |
264 | ||
265 | /** | |
266 | * cik_ih_set_rptr - set the IH ring buffer rptr | |
267 | * | |
268 | * @adev: amdgpu_device pointer | |
269 | * | |
270 | * Set the IH ring buffer rptr. | |
271 | */ | |
8bb9eb48 CK |
272 | static void cik_ih_set_rptr(struct amdgpu_device *adev, |
273 | struct amdgpu_ih_ring *ih) | |
a2e73f56 | 274 | { |
8bb9eb48 | 275 | WREG32(mmIH_RB_RPTR, ih->rptr); |
a2e73f56 AD |
276 | } |
277 | ||
5fc3aeeb | 278 | static int cik_ih_early_init(void *handle) |
a2e73f56 | 279 | { |
5fc3aeeb | 280 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
5f232365 AD |
281 | int ret; |
282 | ||
283 | ret = amdgpu_irq_add_domain(adev); | |
284 | if (ret) | |
285 | return ret; | |
5fc3aeeb | 286 | |
a2e73f56 AD |
287 | cik_ih_set_interrupt_funcs(adev); |
288 | ||
289 | return 0; | |
290 | } | |
291 | ||
5fc3aeeb | 292 | static int cik_ih_sw_init(void *handle) |
a2e73f56 AD |
293 | { |
294 | int r; | |
5fc3aeeb | 295 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
a2e73f56 | 296 | |
425c3143 | 297 | r = amdgpu_ih_ring_init(adev, &adev->irq.ih, 64 * 1024, false); |
a2e73f56 AD |
298 | if (r) |
299 | return r; | |
300 | ||
301 | r = amdgpu_irq_init(adev); | |
302 | ||
303 | return r; | |
304 | } | |
305 | ||
5fc3aeeb | 306 | static int cik_ih_sw_fini(void *handle) |
a2e73f56 | 307 | { |
5fc3aeeb | 308 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
309 | ||
a2e73f56 | 310 | amdgpu_irq_fini(adev); |
425c3143 | 311 | amdgpu_ih_ring_fini(adev, &adev->irq.ih); |
5f232365 | 312 | amdgpu_irq_remove_domain(adev); |
a2e73f56 AD |
313 | |
314 | return 0; | |
315 | } | |
316 | ||
5fc3aeeb | 317 | static int cik_ih_hw_init(void *handle) |
a2e73f56 AD |
318 | { |
319 | int r; | |
5fc3aeeb | 320 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
a2e73f56 AD |
321 | |
322 | r = cik_ih_irq_init(adev); | |
323 | if (r) | |
324 | return r; | |
325 | ||
326 | return 0; | |
327 | } | |
328 | ||
5fc3aeeb | 329 | static int cik_ih_hw_fini(void *handle) |
a2e73f56 | 330 | { |
5fc3aeeb | 331 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
332 | ||
a2e73f56 AD |
333 | cik_ih_irq_disable(adev); |
334 | ||
335 | return 0; | |
336 | } | |
337 | ||
5fc3aeeb | 338 | static int cik_ih_suspend(void *handle) |
a2e73f56 | 339 | { |
5fc3aeeb | 340 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
341 | ||
a2e73f56 AD |
342 | return cik_ih_hw_fini(adev); |
343 | } | |
344 | ||
5fc3aeeb | 345 | static int cik_ih_resume(void *handle) |
a2e73f56 | 346 | { |
5fc3aeeb | 347 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
348 | ||
a2e73f56 AD |
349 | return cik_ih_hw_init(adev); |
350 | } | |
351 | ||
5fc3aeeb | 352 | static bool cik_ih_is_idle(void *handle) |
a2e73f56 | 353 | { |
5fc3aeeb | 354 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
a2e73f56 AD |
355 | u32 tmp = RREG32(mmSRBM_STATUS); |
356 | ||
357 | if (tmp & SRBM_STATUS__IH_BUSY_MASK) | |
358 | return false; | |
359 | ||
360 | return true; | |
361 | } | |
362 | ||
5fc3aeeb | 363 | static int cik_ih_wait_for_idle(void *handle) |
a2e73f56 AD |
364 | { |
365 | unsigned i; | |
366 | u32 tmp; | |
5fc3aeeb | 367 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
a2e73f56 AD |
368 | |
369 | for (i = 0; i < adev->usec_timeout; i++) { | |
370 | /* read MC_STATUS */ | |
371 | tmp = RREG32(mmSRBM_STATUS) & SRBM_STATUS__IH_BUSY_MASK; | |
372 | if (!tmp) | |
373 | return 0; | |
374 | udelay(1); | |
375 | } | |
376 | return -ETIMEDOUT; | |
377 | } | |
378 | ||
5fc3aeeb | 379 | static int cik_ih_soft_reset(void *handle) |
a2e73f56 | 380 | { |
5fc3aeeb | 381 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
382 | ||
a2e73f56 AD |
383 | u32 srbm_soft_reset = 0; |
384 | u32 tmp = RREG32(mmSRBM_STATUS); | |
385 | ||
386 | if (tmp & SRBM_STATUS__IH_BUSY_MASK) | |
387 | srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_IH_MASK; | |
388 | ||
389 | if (srbm_soft_reset) { | |
a2e73f56 AD |
390 | tmp = RREG32(mmSRBM_SOFT_RESET); |
391 | tmp |= srbm_soft_reset; | |
392 | dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp); | |
393 | WREG32(mmSRBM_SOFT_RESET, tmp); | |
394 | tmp = RREG32(mmSRBM_SOFT_RESET); | |
395 | ||
396 | udelay(50); | |
397 | ||
398 | tmp &= ~srbm_soft_reset; | |
399 | WREG32(mmSRBM_SOFT_RESET, tmp); | |
400 | tmp = RREG32(mmSRBM_SOFT_RESET); | |
401 | ||
402 | /* Wait a little for things to settle down */ | |
403 | udelay(50); | |
a2e73f56 AD |
404 | } |
405 | ||
406 | return 0; | |
407 | } | |
408 | ||
5fc3aeeb | 409 | static int cik_ih_set_clockgating_state(void *handle, |
410 | enum amd_clockgating_state state) | |
a2e73f56 AD |
411 | { |
412 | return 0; | |
413 | } | |
414 | ||
5fc3aeeb | 415 | static int cik_ih_set_powergating_state(void *handle, |
416 | enum amd_powergating_state state) | |
a2e73f56 AD |
417 | { |
418 | return 0; | |
419 | } | |
420 | ||
a1255107 | 421 | static const struct amd_ip_funcs cik_ih_ip_funcs = { |
88a907d6 | 422 | .name = "cik_ih", |
a2e73f56 AD |
423 | .early_init = cik_ih_early_init, |
424 | .late_init = NULL, | |
425 | .sw_init = cik_ih_sw_init, | |
426 | .sw_fini = cik_ih_sw_fini, | |
427 | .hw_init = cik_ih_hw_init, | |
428 | .hw_fini = cik_ih_hw_fini, | |
429 | .suspend = cik_ih_suspend, | |
430 | .resume = cik_ih_resume, | |
431 | .is_idle = cik_ih_is_idle, | |
432 | .wait_for_idle = cik_ih_wait_for_idle, | |
433 | .soft_reset = cik_ih_soft_reset, | |
a2e73f56 AD |
434 | .set_clockgating_state = cik_ih_set_clockgating_state, |
435 | .set_powergating_state = cik_ih_set_powergating_state, | |
436 | }; | |
437 | ||
438 | static const struct amdgpu_ih_funcs cik_ih_funcs = { | |
439 | .get_wptr = cik_ih_get_wptr, | |
440 | .decode_iv = cik_ih_decode_iv, | |
441 | .set_rptr = cik_ih_set_rptr | |
442 | }; | |
443 | ||
444 | static void cik_ih_set_interrupt_funcs(struct amdgpu_device *adev) | |
445 | { | |
f54b30d7 | 446 | adev->irq.ih_funcs = &cik_ih_funcs; |
a2e73f56 | 447 | } |
a1255107 AD |
448 | |
449 | const struct amdgpu_ip_block_version cik_ih_ip_block = | |
450 | { | |
451 | .type = AMD_IP_BLOCK_TYPE_IH, | |
452 | .major = 2, | |
453 | .minor = 0, | |
454 | .rev = 0, | |
455 | .funcs = &cik_ih_ip_funcs, | |
456 | }; |