Commit | Line | Data |
---|---|---|
aaa36a97 AD |
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 | * Authors: Alex Deucher | |
23 | */ | |
24 | #include <linux/firmware.h> | |
25 | #include <drm/drmP.h> | |
26 | #include "amdgpu.h" | |
27 | #include "amdgpu_ucode.h" | |
28 | #include "amdgpu_trace.h" | |
29 | #include "vi.h" | |
30 | #include "vid.h" | |
31 | ||
32 | #include "oss/oss_3_0_d.h" | |
33 | #include "oss/oss_3_0_sh_mask.h" | |
34 | ||
35 | #include "gmc/gmc_8_1_d.h" | |
36 | #include "gmc/gmc_8_1_sh_mask.h" | |
37 | ||
38 | #include "gca/gfx_8_0_d.h" | |
74a5d165 | 39 | #include "gca/gfx_8_0_enum.h" |
aaa36a97 AD |
40 | #include "gca/gfx_8_0_sh_mask.h" |
41 | ||
42 | #include "bif/bif_5_0_d.h" | |
43 | #include "bif/bif_5_0_sh_mask.h" | |
44 | ||
45 | #include "tonga_sdma_pkt_open.h" | |
46 | ||
47 | static void sdma_v3_0_set_ring_funcs(struct amdgpu_device *adev); | |
48 | static void sdma_v3_0_set_buffer_funcs(struct amdgpu_device *adev); | |
49 | static void sdma_v3_0_set_vm_pte_funcs(struct amdgpu_device *adev); | |
50 | static void sdma_v3_0_set_irq_funcs(struct amdgpu_device *adev); | |
51 | ||
52 | MODULE_FIRMWARE("radeon/tonga_sdma.bin"); | |
53 | MODULE_FIRMWARE("radeon/tonga_sdma1.bin"); | |
54 | MODULE_FIRMWARE("radeon/carrizo_sdma.bin"); | |
55 | MODULE_FIRMWARE("radeon/carrizo_sdma1.bin"); | |
56 | ||
57 | static const u32 sdma_offsets[SDMA_MAX_INSTANCE] = | |
58 | { | |
59 | SDMA0_REGISTER_OFFSET, | |
60 | SDMA1_REGISTER_OFFSET | |
61 | }; | |
62 | ||
63 | static const u32 golden_settings_tonga_a11[] = | |
64 | { | |
65 | mmSDMA0_CHICKEN_BITS, 0xfc910007, 0x00810007, | |
66 | mmSDMA0_CLK_CTRL, 0xff000fff, 0x00000000, | |
67 | mmSDMA0_GFX_IB_CNTL, 0x800f0111, 0x00000100, | |
68 | mmSDMA0_RLC0_IB_CNTL, 0x800f0111, 0x00000100, | |
69 | mmSDMA0_RLC1_IB_CNTL, 0x800f0111, 0x00000100, | |
70 | mmSDMA1_CHICKEN_BITS, 0xfc910007, 0x00810007, | |
71 | mmSDMA1_CLK_CTRL, 0xff000fff, 0x00000000, | |
72 | mmSDMA1_GFX_IB_CNTL, 0x800f0111, 0x00000100, | |
73 | mmSDMA1_RLC0_IB_CNTL, 0x800f0111, 0x00000100, | |
74 | mmSDMA1_RLC1_IB_CNTL, 0x800f0111, 0x00000100, | |
75 | }; | |
76 | ||
77 | static const u32 tonga_mgcg_cgcg_init[] = | |
78 | { | |
79 | mmSDMA0_CLK_CTRL, 0xff000ff0, 0x00000100, | |
80 | mmSDMA1_CLK_CTRL, 0xff000ff0, 0x00000100 | |
81 | }; | |
82 | ||
83 | static const u32 cz_golden_settings_a11[] = | |
84 | { | |
85 | mmSDMA0_CHICKEN_BITS, 0xfc910007, 0x00810007, | |
86 | mmSDMA0_CLK_CTRL, 0xff000fff, 0x00000000, | |
87 | mmSDMA0_GFX_IB_CNTL, 0x00000100, 0x00000100, | |
88 | mmSDMA0_POWER_CNTL, 0x00000800, 0x0003c800, | |
89 | mmSDMA0_RLC0_IB_CNTL, 0x00000100, 0x00000100, | |
90 | mmSDMA0_RLC1_IB_CNTL, 0x00000100, 0x00000100, | |
91 | mmSDMA1_CHICKEN_BITS, 0xfc910007, 0x00810007, | |
92 | mmSDMA1_CLK_CTRL, 0xff000fff, 0x00000000, | |
93 | mmSDMA1_GFX_IB_CNTL, 0x00000100, 0x00000100, | |
94 | mmSDMA1_POWER_CNTL, 0x00000800, 0x0003c800, | |
95 | mmSDMA1_RLC0_IB_CNTL, 0x00000100, 0x00000100, | |
96 | mmSDMA1_RLC1_IB_CNTL, 0x00000100, 0x00000100, | |
97 | }; | |
98 | ||
99 | static const u32 cz_mgcg_cgcg_init[] = | |
100 | { | |
101 | mmSDMA0_CLK_CTRL, 0xff000ff0, 0x00000100, | |
102 | mmSDMA1_CLK_CTRL, 0xff000ff0, 0x00000100 | |
103 | }; | |
104 | ||
105 | /* | |
106 | * sDMA - System DMA | |
107 | * Starting with CIK, the GPU has new asynchronous | |
108 | * DMA engines. These engines are used for compute | |
109 | * and gfx. There are two DMA engines (SDMA0, SDMA1) | |
110 | * and each one supports 1 ring buffer used for gfx | |
111 | * and 2 queues used for compute. | |
112 | * | |
113 | * The programming model is very similar to the CP | |
114 | * (ring buffer, IBs, etc.), but sDMA has it's own | |
115 | * packet format that is different from the PM4 format | |
116 | * used by the CP. sDMA supports copying data, writing | |
117 | * embedded data, solid fills, and a number of other | |
118 | * things. It also has support for tiling/detiling of | |
119 | * buffers. | |
120 | */ | |
121 | ||
122 | static void sdma_v3_0_init_golden_registers(struct amdgpu_device *adev) | |
123 | { | |
124 | switch (adev->asic_type) { | |
125 | case CHIP_TONGA: | |
126 | amdgpu_program_register_sequence(adev, | |
127 | tonga_mgcg_cgcg_init, | |
128 | (const u32)ARRAY_SIZE(tonga_mgcg_cgcg_init)); | |
129 | amdgpu_program_register_sequence(adev, | |
130 | golden_settings_tonga_a11, | |
131 | (const u32)ARRAY_SIZE(golden_settings_tonga_a11)); | |
132 | break; | |
133 | case CHIP_CARRIZO: | |
134 | amdgpu_program_register_sequence(adev, | |
135 | cz_mgcg_cgcg_init, | |
136 | (const u32)ARRAY_SIZE(cz_mgcg_cgcg_init)); | |
137 | amdgpu_program_register_sequence(adev, | |
138 | cz_golden_settings_a11, | |
139 | (const u32)ARRAY_SIZE(cz_golden_settings_a11)); | |
140 | break; | |
141 | default: | |
142 | break; | |
143 | } | |
144 | } | |
145 | ||
146 | /** | |
147 | * sdma_v3_0_init_microcode - load ucode images from disk | |
148 | * | |
149 | * @adev: amdgpu_device pointer | |
150 | * | |
151 | * Use the firmware interface to load the ucode images into | |
152 | * the driver (not loaded into hw). | |
153 | * Returns 0 on success, error on failure. | |
154 | */ | |
155 | static int sdma_v3_0_init_microcode(struct amdgpu_device *adev) | |
156 | { | |
157 | const char *chip_name; | |
158 | char fw_name[30]; | |
159 | int err, i; | |
160 | struct amdgpu_firmware_info *info = NULL; | |
161 | const struct common_firmware_header *header = NULL; | |
162 | ||
163 | DRM_DEBUG("\n"); | |
164 | ||
165 | switch (adev->asic_type) { | |
166 | case CHIP_TONGA: | |
167 | chip_name = "tonga"; | |
168 | break; | |
169 | case CHIP_CARRIZO: | |
170 | chip_name = "carrizo"; | |
171 | break; | |
172 | default: BUG(); | |
173 | } | |
174 | ||
175 | for (i = 0; i < SDMA_MAX_INSTANCE; i++) { | |
176 | if (i == 0) | |
177 | snprintf(fw_name, sizeof(fw_name), "radeon/%s_sdma.bin", chip_name); | |
178 | else | |
179 | snprintf(fw_name, sizeof(fw_name), "radeon/%s_sdma1.bin", chip_name); | |
180 | err = request_firmware(&adev->sdma[i].fw, fw_name, adev->dev); | |
181 | if (err) | |
182 | goto out; | |
183 | err = amdgpu_ucode_validate(adev->sdma[i].fw); | |
184 | if (err) | |
185 | goto out; | |
186 | ||
187 | if (adev->firmware.smu_load) { | |
188 | info = &adev->firmware.ucode[AMDGPU_UCODE_ID_SDMA0 + i]; | |
189 | info->ucode_id = AMDGPU_UCODE_ID_SDMA0 + i; | |
190 | info->fw = adev->sdma[i].fw; | |
191 | header = (const struct common_firmware_header *)info->fw->data; | |
192 | adev->firmware.fw_size += | |
193 | ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE); | |
194 | } | |
195 | } | |
196 | out: | |
197 | if (err) { | |
198 | printk(KERN_ERR | |
199 | "sdma_v3_0: Failed to load firmware \"%s\"\n", | |
200 | fw_name); | |
201 | for (i = 0; i < SDMA_MAX_INSTANCE; i++) { | |
202 | release_firmware(adev->sdma[i].fw); | |
203 | adev->sdma[i].fw = NULL; | |
204 | } | |
205 | } | |
206 | return err; | |
207 | } | |
208 | ||
209 | /** | |
210 | * sdma_v3_0_ring_get_rptr - get the current read pointer | |
211 | * | |
212 | * @ring: amdgpu ring pointer | |
213 | * | |
214 | * Get the current rptr from the hardware (VI+). | |
215 | */ | |
216 | static uint32_t sdma_v3_0_ring_get_rptr(struct amdgpu_ring *ring) | |
217 | { | |
218 | u32 rptr; | |
219 | ||
220 | /* XXX check if swapping is necessary on BE */ | |
221 | rptr = ring->adev->wb.wb[ring->rptr_offs] >> 2; | |
222 | ||
223 | return rptr; | |
224 | } | |
225 | ||
226 | /** | |
227 | * sdma_v3_0_ring_get_wptr - get the current write pointer | |
228 | * | |
229 | * @ring: amdgpu ring pointer | |
230 | * | |
231 | * Get the current wptr from the hardware (VI+). | |
232 | */ | |
233 | static uint32_t sdma_v3_0_ring_get_wptr(struct amdgpu_ring *ring) | |
234 | { | |
235 | struct amdgpu_device *adev = ring->adev; | |
236 | u32 wptr; | |
237 | ||
238 | if (ring->use_doorbell) { | |
239 | /* XXX check if swapping is necessary on BE */ | |
240 | wptr = ring->adev->wb.wb[ring->wptr_offs] >> 2; | |
241 | } else { | |
242 | int me = (ring == &ring->adev->sdma[0].ring) ? 0 : 1; | |
243 | ||
244 | wptr = RREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me]) >> 2; | |
245 | } | |
246 | ||
247 | return wptr; | |
248 | } | |
249 | ||
250 | /** | |
251 | * sdma_v3_0_ring_set_wptr - commit the write pointer | |
252 | * | |
253 | * @ring: amdgpu ring pointer | |
254 | * | |
255 | * Write the wptr back to the hardware (VI+). | |
256 | */ | |
257 | static void sdma_v3_0_ring_set_wptr(struct amdgpu_ring *ring) | |
258 | { | |
259 | struct amdgpu_device *adev = ring->adev; | |
260 | ||
261 | if (ring->use_doorbell) { | |
262 | /* XXX check if swapping is necessary on BE */ | |
263 | adev->wb.wb[ring->wptr_offs] = ring->wptr << 2; | |
264 | WDOORBELL32(ring->doorbell_index, ring->wptr << 2); | |
265 | } else { | |
266 | int me = (ring == &ring->adev->sdma[0].ring) ? 0 : 1; | |
267 | ||
268 | WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me], ring->wptr << 2); | |
269 | } | |
270 | } | |
271 | ||
272 | static void sdma_v3_0_hdp_flush_ring_emit(struct amdgpu_ring *); | |
273 | ||
274 | /** | |
275 | * sdma_v3_0_ring_emit_ib - Schedule an IB on the DMA engine | |
276 | * | |
277 | * @ring: amdgpu ring pointer | |
278 | * @ib: IB object to schedule | |
279 | * | |
280 | * Schedule an IB in the DMA ring (VI). | |
281 | */ | |
282 | static void sdma_v3_0_ring_emit_ib(struct amdgpu_ring *ring, | |
283 | struct amdgpu_ib *ib) | |
284 | { | |
285 | u32 vmid = (ib->vm ? ib->vm->ids[ring->idx].id : 0) & 0xf; | |
286 | u32 next_rptr = ring->wptr + 5; | |
287 | ||
288 | if (ib->flush_hdp_writefifo) | |
289 | next_rptr += 6; | |
290 | ||
291 | while ((next_rptr & 7) != 2) | |
292 | next_rptr++; | |
293 | next_rptr += 6; | |
294 | ||
295 | amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) | | |
296 | SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR)); | |
297 | amdgpu_ring_write(ring, lower_32_bits(ring->next_rptr_gpu_addr) & 0xfffffffc); | |
298 | amdgpu_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr)); | |
299 | amdgpu_ring_write(ring, SDMA_PKT_WRITE_UNTILED_DW_3_COUNT(1)); | |
300 | amdgpu_ring_write(ring, next_rptr); | |
301 | ||
302 | /* flush HDP */ | |
303 | if (ib->flush_hdp_writefifo) { | |
304 | sdma_v3_0_hdp_flush_ring_emit(ring); | |
305 | } | |
306 | ||
307 | /* IB packet must end on a 8 DW boundary */ | |
308 | while ((ring->wptr & 7) != 2) | |
309 | amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_NOP)); | |
310 | ||
311 | amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) | | |
312 | SDMA_PKT_INDIRECT_HEADER_VMID(vmid)); | |
313 | /* base must be 32 byte aligned */ | |
314 | amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr) & 0xffffffe0); | |
315 | amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr)); | |
316 | amdgpu_ring_write(ring, ib->length_dw); | |
317 | amdgpu_ring_write(ring, 0); | |
318 | amdgpu_ring_write(ring, 0); | |
319 | ||
320 | } | |
321 | ||
322 | /** | |
323 | * sdma_v3_0_hdp_flush_ring_emit - emit an hdp flush on the DMA ring | |
324 | * | |
325 | * @ring: amdgpu ring pointer | |
326 | * | |
327 | * Emit an hdp flush packet on the requested DMA ring. | |
328 | */ | |
329 | static void sdma_v3_0_hdp_flush_ring_emit(struct amdgpu_ring *ring) | |
330 | { | |
331 | u32 ref_and_mask = 0; | |
332 | ||
333 | if (ring == &ring->adev->sdma[0].ring) | |
334 | ref_and_mask = REG_SET_FIELD(ref_and_mask, GPU_HDP_FLUSH_DONE, SDMA0, 1); | |
335 | else | |
336 | ref_and_mask = REG_SET_FIELD(ref_and_mask, GPU_HDP_FLUSH_DONE, SDMA1, 1); | |
337 | ||
338 | amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_POLL_REGMEM) | | |
339 | SDMA_PKT_POLL_REGMEM_HEADER_HDP_FLUSH(1) | | |
340 | SDMA_PKT_POLL_REGMEM_HEADER_FUNC(3)); /* == */ | |
341 | amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_DONE << 2); | |
342 | amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_REQ << 2); | |
343 | amdgpu_ring_write(ring, ref_and_mask); /* reference */ | |
344 | amdgpu_ring_write(ring, ref_and_mask); /* mask */ | |
345 | amdgpu_ring_write(ring, SDMA_PKT_POLL_REGMEM_DW5_RETRY_COUNT(0xfff) | | |
346 | SDMA_PKT_POLL_REGMEM_DW5_INTERVAL(10)); /* retry count, poll interval */ | |
347 | } | |
348 | ||
349 | /** | |
350 | * sdma_v3_0_ring_emit_fence - emit a fence on the DMA ring | |
351 | * | |
352 | * @ring: amdgpu ring pointer | |
353 | * @fence: amdgpu fence object | |
354 | * | |
355 | * Add a DMA fence packet to the ring to write | |
356 | * the fence seq number and DMA trap packet to generate | |
357 | * an interrupt if needed (VI). | |
358 | */ | |
359 | static void sdma_v3_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq, | |
360 | bool write64bits) | |
361 | { | |
362 | /* write the fence */ | |
363 | amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_FENCE)); | |
364 | amdgpu_ring_write(ring, lower_32_bits(addr)); | |
365 | amdgpu_ring_write(ring, upper_32_bits(addr)); | |
366 | amdgpu_ring_write(ring, lower_32_bits(seq)); | |
367 | ||
368 | /* optionally write high bits as well */ | |
369 | if (write64bits) { | |
370 | addr += 4; | |
371 | amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_FENCE)); | |
372 | amdgpu_ring_write(ring, lower_32_bits(addr)); | |
373 | amdgpu_ring_write(ring, upper_32_bits(addr)); | |
374 | amdgpu_ring_write(ring, upper_32_bits(seq)); | |
375 | } | |
376 | ||
377 | /* generate an interrupt */ | |
378 | amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_TRAP)); | |
379 | amdgpu_ring_write(ring, SDMA_PKT_TRAP_INT_CONTEXT_INT_CONTEXT(0)); | |
380 | } | |
381 | ||
382 | ||
383 | /** | |
384 | * sdma_v3_0_ring_emit_semaphore - emit a semaphore on the dma ring | |
385 | * | |
386 | * @ring: amdgpu_ring structure holding ring information | |
387 | * @semaphore: amdgpu semaphore object | |
388 | * @emit_wait: wait or signal semaphore | |
389 | * | |
390 | * Add a DMA semaphore packet to the ring wait on or signal | |
391 | * other rings (VI). | |
392 | */ | |
393 | static bool sdma_v3_0_ring_emit_semaphore(struct amdgpu_ring *ring, | |
394 | struct amdgpu_semaphore *semaphore, | |
395 | bool emit_wait) | |
396 | { | |
397 | u64 addr = semaphore->gpu_addr; | |
398 | u32 sig = emit_wait ? 0 : 1; | |
399 | ||
400 | amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SEM) | | |
401 | SDMA_PKT_SEMAPHORE_HEADER_SIGNAL(sig)); | |
402 | amdgpu_ring_write(ring, lower_32_bits(addr) & 0xfffffff8); | |
403 | amdgpu_ring_write(ring, upper_32_bits(addr)); | |
404 | ||
405 | return true; | |
406 | } | |
407 | ||
408 | /** | |
409 | * sdma_v3_0_gfx_stop - stop the gfx async dma engines | |
410 | * | |
411 | * @adev: amdgpu_device pointer | |
412 | * | |
413 | * Stop the gfx async dma ring buffers (VI). | |
414 | */ | |
415 | static void sdma_v3_0_gfx_stop(struct amdgpu_device *adev) | |
416 | { | |
417 | struct amdgpu_ring *sdma0 = &adev->sdma[0].ring; | |
418 | struct amdgpu_ring *sdma1 = &adev->sdma[1].ring; | |
419 | u32 rb_cntl, ib_cntl; | |
420 | int i; | |
421 | ||
422 | if ((adev->mman.buffer_funcs_ring == sdma0) || | |
423 | (adev->mman.buffer_funcs_ring == sdma1)) | |
424 | amdgpu_ttm_set_active_vram_size(adev, adev->mc.visible_vram_size); | |
425 | ||
426 | for (i = 0; i < SDMA_MAX_INSTANCE; i++) { | |
427 | rb_cntl = RREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i]); | |
428 | rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, 0); | |
429 | WREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i], rb_cntl); | |
430 | ib_cntl = RREG32(mmSDMA0_GFX_IB_CNTL + sdma_offsets[i]); | |
431 | ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_ENABLE, 0); | |
432 | WREG32(mmSDMA0_GFX_IB_CNTL + sdma_offsets[i], ib_cntl); | |
433 | } | |
434 | sdma0->ready = false; | |
435 | sdma1->ready = false; | |
436 | } | |
437 | ||
438 | /** | |
439 | * sdma_v3_0_rlc_stop - stop the compute async dma engines | |
440 | * | |
441 | * @adev: amdgpu_device pointer | |
442 | * | |
443 | * Stop the compute async dma queues (VI). | |
444 | */ | |
445 | static void sdma_v3_0_rlc_stop(struct amdgpu_device *adev) | |
446 | { | |
447 | /* XXX todo */ | |
448 | } | |
449 | ||
450 | /** | |
451 | * sdma_v3_0_enable - stop the async dma engines | |
452 | * | |
453 | * @adev: amdgpu_device pointer | |
454 | * @enable: enable/disable the DMA MEs. | |
455 | * | |
456 | * Halt or unhalt the async dma engines (VI). | |
457 | */ | |
458 | static void sdma_v3_0_enable(struct amdgpu_device *adev, bool enable) | |
459 | { | |
460 | u32 f32_cntl; | |
461 | int i; | |
462 | ||
463 | if (enable == false) { | |
464 | sdma_v3_0_gfx_stop(adev); | |
465 | sdma_v3_0_rlc_stop(adev); | |
466 | } | |
467 | ||
468 | for (i = 0; i < SDMA_MAX_INSTANCE; i++) { | |
469 | f32_cntl = RREG32(mmSDMA0_F32_CNTL + sdma_offsets[i]); | |
470 | if (enable) | |
471 | f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_F32_CNTL, HALT, 0); | |
472 | else | |
473 | f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_F32_CNTL, HALT, 1); | |
474 | WREG32(mmSDMA0_F32_CNTL + sdma_offsets[i], f32_cntl); | |
475 | } | |
476 | } | |
477 | ||
478 | /** | |
479 | * sdma_v3_0_gfx_resume - setup and start the async dma engines | |
480 | * | |
481 | * @adev: amdgpu_device pointer | |
482 | * | |
483 | * Set up the gfx DMA ring buffers and enable them (VI). | |
484 | * Returns 0 for success, error for failure. | |
485 | */ | |
486 | static int sdma_v3_0_gfx_resume(struct amdgpu_device *adev) | |
487 | { | |
488 | struct amdgpu_ring *ring; | |
489 | u32 rb_cntl, ib_cntl; | |
490 | u32 rb_bufsz; | |
491 | u32 wb_offset; | |
492 | u32 doorbell; | |
493 | int i, j, r; | |
494 | ||
495 | for (i = 0; i < SDMA_MAX_INSTANCE; i++) { | |
496 | ring = &adev->sdma[i].ring; | |
497 | wb_offset = (ring->rptr_offs * 4); | |
498 | ||
499 | mutex_lock(&adev->srbm_mutex); | |
500 | for (j = 0; j < 16; j++) { | |
501 | vi_srbm_select(adev, 0, 0, 0, j); | |
502 | /* SDMA GFX */ | |
503 | WREG32(mmSDMA0_GFX_VIRTUAL_ADDR + sdma_offsets[i], 0); | |
504 | WREG32(mmSDMA0_GFX_APE1_CNTL + sdma_offsets[i], 0); | |
505 | } | |
506 | vi_srbm_select(adev, 0, 0, 0, 0); | |
507 | mutex_unlock(&adev->srbm_mutex); | |
508 | ||
509 | WREG32(mmSDMA0_SEM_WAIT_FAIL_TIMER_CNTL + sdma_offsets[i], 0); | |
510 | ||
511 | /* Set ring buffer size in dwords */ | |
512 | rb_bufsz = order_base_2(ring->ring_size / 4); | |
513 | rb_cntl = RREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i]); | |
514 | rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_SIZE, rb_bufsz); | |
515 | #ifdef __BIG_ENDIAN | |
516 | rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_SWAP_ENABLE, 1); | |
517 | rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, | |
518 | RPTR_WRITEBACK_SWAP_ENABLE, 1); | |
519 | #endif | |
520 | WREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i], rb_cntl); | |
521 | ||
522 | /* Initialize the ring buffer's read and write pointers */ | |
523 | WREG32(mmSDMA0_GFX_RB_RPTR + sdma_offsets[i], 0); | |
524 | WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[i], 0); | |
525 | ||
526 | /* set the wb address whether it's enabled or not */ | |
527 | WREG32(mmSDMA0_GFX_RB_RPTR_ADDR_HI + sdma_offsets[i], | |
528 | upper_32_bits(adev->wb.gpu_addr + wb_offset) & 0xFFFFFFFF); | |
529 | WREG32(mmSDMA0_GFX_RB_RPTR_ADDR_LO + sdma_offsets[i], | |
530 | lower_32_bits(adev->wb.gpu_addr + wb_offset) & 0xFFFFFFFC); | |
531 | ||
532 | rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RPTR_WRITEBACK_ENABLE, 1); | |
533 | ||
534 | WREG32(mmSDMA0_GFX_RB_BASE + sdma_offsets[i], ring->gpu_addr >> 8); | |
535 | WREG32(mmSDMA0_GFX_RB_BASE_HI + sdma_offsets[i], ring->gpu_addr >> 40); | |
536 | ||
537 | ring->wptr = 0; | |
538 | WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[i], ring->wptr << 2); | |
539 | ||
540 | doorbell = RREG32(mmSDMA0_GFX_DOORBELL + sdma_offsets[i]); | |
541 | ||
542 | if (ring->use_doorbell) { | |
543 | doorbell = REG_SET_FIELD(doorbell, SDMA0_GFX_DOORBELL, | |
544 | OFFSET, ring->doorbell_index); | |
545 | doorbell = REG_SET_FIELD(doorbell, SDMA0_GFX_DOORBELL, ENABLE, 1); | |
546 | } else { | |
547 | doorbell = REG_SET_FIELD(doorbell, SDMA0_GFX_DOORBELL, ENABLE, 0); | |
548 | } | |
549 | WREG32(mmSDMA0_GFX_DOORBELL + sdma_offsets[i], doorbell); | |
550 | ||
551 | /* enable DMA RB */ | |
552 | rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, 1); | |
553 | WREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i], rb_cntl); | |
554 | ||
555 | ib_cntl = RREG32(mmSDMA0_GFX_IB_CNTL + sdma_offsets[i]); | |
556 | ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_ENABLE, 1); | |
557 | #ifdef __BIG_ENDIAN | |
558 | ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_SWAP_ENABLE, 1); | |
559 | #endif | |
560 | /* enable DMA IBs */ | |
561 | WREG32(mmSDMA0_GFX_IB_CNTL + sdma_offsets[i], ib_cntl); | |
562 | ||
563 | ring->ready = true; | |
564 | ||
565 | r = amdgpu_ring_test_ring(ring); | |
566 | if (r) { | |
567 | ring->ready = false; | |
568 | return r; | |
569 | } | |
570 | ||
571 | if (adev->mman.buffer_funcs_ring == ring) | |
572 | amdgpu_ttm_set_active_vram_size(adev, adev->mc.real_vram_size); | |
573 | } | |
574 | ||
575 | return 0; | |
576 | } | |
577 | ||
578 | /** | |
579 | * sdma_v3_0_rlc_resume - setup and start the async dma engines | |
580 | * | |
581 | * @adev: amdgpu_device pointer | |
582 | * | |
583 | * Set up the compute DMA queues and enable them (VI). | |
584 | * Returns 0 for success, error for failure. | |
585 | */ | |
586 | static int sdma_v3_0_rlc_resume(struct amdgpu_device *adev) | |
587 | { | |
588 | /* XXX todo */ | |
589 | return 0; | |
590 | } | |
591 | ||
592 | /** | |
593 | * sdma_v3_0_load_microcode - load the sDMA ME ucode | |
594 | * | |
595 | * @adev: amdgpu_device pointer | |
596 | * | |
597 | * Loads the sDMA0/1 ucode. | |
598 | * Returns 0 for success, -EINVAL if the ucode is not available. | |
599 | */ | |
600 | static int sdma_v3_0_load_microcode(struct amdgpu_device *adev) | |
601 | { | |
602 | const struct sdma_firmware_header_v1_0 *hdr; | |
603 | const __le32 *fw_data; | |
604 | u32 fw_size; | |
605 | int i, j; | |
606 | ||
607 | if (!adev->sdma[0].fw || !adev->sdma[1].fw) | |
608 | return -EINVAL; | |
609 | ||
610 | /* halt the MEs */ | |
611 | sdma_v3_0_enable(adev, false); | |
612 | ||
613 | for (i = 0; i < SDMA_MAX_INSTANCE; i++) { | |
614 | hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data; | |
615 | amdgpu_ucode_print_sdma_hdr(&hdr->header); | |
616 | fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4; | |
617 | adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version); | |
618 | ||
619 | fw_data = (const __le32 *) | |
620 | (adev->sdma[i].fw->data + | |
621 | le32_to_cpu(hdr->header.ucode_array_offset_bytes)); | |
622 | WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], 0); | |
623 | for (j = 0; j < fw_size; j++) | |
624 | WREG32(mmSDMA0_UCODE_DATA + sdma_offsets[i], le32_to_cpup(fw_data++)); | |
625 | WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], adev->sdma[i].fw_version); | |
626 | } | |
627 | ||
628 | return 0; | |
629 | } | |
630 | ||
631 | /** | |
632 | * sdma_v3_0_start - setup and start the async dma engines | |
633 | * | |
634 | * @adev: amdgpu_device pointer | |
635 | * | |
636 | * Set up the DMA engines and enable them (VI). | |
637 | * Returns 0 for success, error for failure. | |
638 | */ | |
639 | static int sdma_v3_0_start(struct amdgpu_device *adev) | |
640 | { | |
641 | int r; | |
642 | ||
643 | if (!adev->firmware.smu_load) { | |
644 | r = sdma_v3_0_load_microcode(adev); | |
645 | if (r) | |
646 | return r; | |
647 | } else { | |
648 | r = adev->smu.smumgr_funcs->check_fw_load_finish(adev, | |
649 | AMDGPU_UCODE_ID_SDMA0); | |
650 | if (r) | |
651 | return -EINVAL; | |
652 | r = adev->smu.smumgr_funcs->check_fw_load_finish(adev, | |
653 | AMDGPU_UCODE_ID_SDMA1); | |
654 | if (r) | |
655 | return -EINVAL; | |
656 | } | |
657 | ||
658 | /* unhalt the MEs */ | |
659 | sdma_v3_0_enable(adev, true); | |
660 | ||
661 | /* start the gfx rings and rlc compute queues */ | |
662 | r = sdma_v3_0_gfx_resume(adev); | |
663 | if (r) | |
664 | return r; | |
665 | r = sdma_v3_0_rlc_resume(adev); | |
666 | if (r) | |
667 | return r; | |
668 | ||
669 | return 0; | |
670 | } | |
671 | ||
672 | /** | |
673 | * sdma_v3_0_ring_test_ring - simple async dma engine test | |
674 | * | |
675 | * @ring: amdgpu_ring structure holding ring information | |
676 | * | |
677 | * Test the DMA engine by writing using it to write an | |
678 | * value to memory. (VI). | |
679 | * Returns 0 for success, error for failure. | |
680 | */ | |
681 | static int sdma_v3_0_ring_test_ring(struct amdgpu_ring *ring) | |
682 | { | |
683 | struct amdgpu_device *adev = ring->adev; | |
684 | unsigned i; | |
685 | unsigned index; | |
686 | int r; | |
687 | u32 tmp; | |
688 | u64 gpu_addr; | |
689 | ||
690 | r = amdgpu_wb_get(adev, &index); | |
691 | if (r) { | |
692 | dev_err(adev->dev, "(%d) failed to allocate wb slot\n", r); | |
693 | return r; | |
694 | } | |
695 | ||
696 | gpu_addr = adev->wb.gpu_addr + (index * 4); | |
697 | tmp = 0xCAFEDEAD; | |
698 | adev->wb.wb[index] = cpu_to_le32(tmp); | |
699 | ||
700 | r = amdgpu_ring_lock(ring, 5); | |
701 | if (r) { | |
702 | DRM_ERROR("amdgpu: dma failed to lock ring %d (%d).\n", ring->idx, r); | |
703 | amdgpu_wb_free(adev, index); | |
704 | return r; | |
705 | } | |
706 | ||
707 | amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) | | |
708 | SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR)); | |
709 | amdgpu_ring_write(ring, lower_32_bits(gpu_addr)); | |
710 | amdgpu_ring_write(ring, upper_32_bits(gpu_addr)); | |
711 | amdgpu_ring_write(ring, SDMA_PKT_WRITE_UNTILED_DW_3_COUNT(1)); | |
712 | amdgpu_ring_write(ring, 0xDEADBEEF); | |
713 | amdgpu_ring_unlock_commit(ring); | |
714 | ||
715 | for (i = 0; i < adev->usec_timeout; i++) { | |
716 | tmp = le32_to_cpu(adev->wb.wb[index]); | |
717 | if (tmp == 0xDEADBEEF) | |
718 | break; | |
719 | DRM_UDELAY(1); | |
720 | } | |
721 | ||
722 | if (i < adev->usec_timeout) { | |
723 | DRM_INFO("ring test on %d succeeded in %d usecs\n", ring->idx, i); | |
724 | } else { | |
725 | DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n", | |
726 | ring->idx, tmp); | |
727 | r = -EINVAL; | |
728 | } | |
729 | amdgpu_wb_free(adev, index); | |
730 | ||
731 | return r; | |
732 | } | |
733 | ||
734 | /** | |
735 | * sdma_v3_0_ring_test_ib - test an IB on the DMA engine | |
736 | * | |
737 | * @ring: amdgpu_ring structure holding ring information | |
738 | * | |
739 | * Test a simple IB in the DMA ring (VI). | |
740 | * Returns 0 on success, error on failure. | |
741 | */ | |
742 | static int sdma_v3_0_ring_test_ib(struct amdgpu_ring *ring) | |
743 | { | |
744 | struct amdgpu_device *adev = ring->adev; | |
745 | struct amdgpu_ib ib; | |
746 | unsigned i; | |
747 | unsigned index; | |
748 | int r; | |
749 | u32 tmp = 0; | |
750 | u64 gpu_addr; | |
751 | ||
752 | r = amdgpu_wb_get(adev, &index); | |
753 | if (r) { | |
754 | dev_err(adev->dev, "(%d) failed to allocate wb slot\n", r); | |
755 | return r; | |
756 | } | |
757 | ||
758 | gpu_addr = adev->wb.gpu_addr + (index * 4); | |
759 | tmp = 0xCAFEDEAD; | |
760 | adev->wb.wb[index] = cpu_to_le32(tmp); | |
761 | ||
762 | r = amdgpu_ib_get(ring, NULL, 256, &ib); | |
763 | if (r) { | |
764 | amdgpu_wb_free(adev, index); | |
765 | DRM_ERROR("amdgpu: failed to get ib (%d).\n", r); | |
766 | return r; | |
767 | } | |
768 | ||
769 | ib.ptr[0] = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) | | |
770 | SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR); | |
771 | ib.ptr[1] = lower_32_bits(gpu_addr); | |
772 | ib.ptr[2] = upper_32_bits(gpu_addr); | |
773 | ib.ptr[3] = SDMA_PKT_WRITE_UNTILED_DW_3_COUNT(1); | |
774 | ib.ptr[4] = 0xDEADBEEF; | |
775 | ib.ptr[5] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP); | |
776 | ib.ptr[6] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP); | |
777 | ib.ptr[7] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP); | |
778 | ib.length_dw = 8; | |
779 | ||
780 | r = amdgpu_ib_schedule(adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED); | |
781 | if (r) { | |
782 | amdgpu_ib_free(adev, &ib); | |
783 | amdgpu_wb_free(adev, index); | |
784 | DRM_ERROR("amdgpu: failed to schedule ib (%d).\n", r); | |
785 | return r; | |
786 | } | |
787 | r = amdgpu_fence_wait(ib.fence, false); | |
788 | if (r) { | |
789 | amdgpu_ib_free(adev, &ib); | |
790 | amdgpu_wb_free(adev, index); | |
791 | DRM_ERROR("amdgpu: fence wait failed (%d).\n", r); | |
792 | return r; | |
793 | } | |
794 | for (i = 0; i < adev->usec_timeout; i++) { | |
795 | tmp = le32_to_cpu(adev->wb.wb[index]); | |
796 | if (tmp == 0xDEADBEEF) | |
797 | break; | |
798 | DRM_UDELAY(1); | |
799 | } | |
800 | if (i < adev->usec_timeout) { | |
801 | DRM_INFO("ib test on ring %d succeeded in %u usecs\n", | |
802 | ib.fence->ring->idx, i); | |
803 | } else { | |
804 | DRM_ERROR("amdgpu: ib test failed (0x%08X)\n", tmp); | |
805 | r = -EINVAL; | |
806 | } | |
807 | amdgpu_ib_free(adev, &ib); | |
808 | amdgpu_wb_free(adev, index); | |
809 | return r; | |
810 | } | |
811 | ||
812 | /** | |
813 | * sdma_v3_0_vm_copy_pte - update PTEs by copying them from the GART | |
814 | * | |
815 | * @ib: indirect buffer to fill with commands | |
816 | * @pe: addr of the page entry | |
817 | * @src: src addr to copy from | |
818 | * @count: number of page entries to update | |
819 | * | |
820 | * Update PTEs by copying them from the GART using sDMA (CIK). | |
821 | */ | |
822 | static void sdma_v3_0_vm_copy_pte(struct amdgpu_ib *ib, | |
823 | uint64_t pe, uint64_t src, | |
824 | unsigned count) | |
825 | { | |
826 | while (count) { | |
827 | unsigned bytes = count * 8; | |
828 | if (bytes > 0x1FFFF8) | |
829 | bytes = 0x1FFFF8; | |
830 | ||
831 | ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_COPY) | | |
832 | SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR); | |
833 | ib->ptr[ib->length_dw++] = bytes; | |
834 | ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */ | |
835 | ib->ptr[ib->length_dw++] = lower_32_bits(src); | |
836 | ib->ptr[ib->length_dw++] = upper_32_bits(src); | |
837 | ib->ptr[ib->length_dw++] = lower_32_bits(pe); | |
838 | ib->ptr[ib->length_dw++] = upper_32_bits(pe); | |
839 | ||
840 | pe += bytes; | |
841 | src += bytes; | |
842 | count -= bytes / 8; | |
843 | } | |
844 | } | |
845 | ||
846 | /** | |
847 | * sdma_v3_0_vm_write_pte - update PTEs by writing them manually | |
848 | * | |
849 | * @ib: indirect buffer to fill with commands | |
850 | * @pe: addr of the page entry | |
851 | * @addr: dst addr to write into pe | |
852 | * @count: number of page entries to update | |
853 | * @incr: increase next addr by incr bytes | |
854 | * @flags: access flags | |
855 | * | |
856 | * Update PTEs by writing them manually using sDMA (CIK). | |
857 | */ | |
858 | static void sdma_v3_0_vm_write_pte(struct amdgpu_ib *ib, | |
859 | uint64_t pe, | |
860 | uint64_t addr, unsigned count, | |
861 | uint32_t incr, uint32_t flags) | |
862 | { | |
863 | uint64_t value; | |
864 | unsigned ndw; | |
865 | ||
866 | while (count) { | |
867 | ndw = count * 2; | |
868 | if (ndw > 0xFFFFE) | |
869 | ndw = 0xFFFFE; | |
870 | ||
871 | /* for non-physically contiguous pages (system) */ | |
872 | ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) | | |
873 | SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR); | |
874 | ib->ptr[ib->length_dw++] = pe; | |
875 | ib->ptr[ib->length_dw++] = upper_32_bits(pe); | |
876 | ib->ptr[ib->length_dw++] = ndw; | |
877 | for (; ndw > 0; ndw -= 2, --count, pe += 8) { | |
878 | if (flags & AMDGPU_PTE_SYSTEM) { | |
879 | value = amdgpu_vm_map_gart(ib->ring->adev, addr); | |
880 | value &= 0xFFFFFFFFFFFFF000ULL; | |
881 | } else if (flags & AMDGPU_PTE_VALID) { | |
882 | value = addr; | |
883 | } else { | |
884 | value = 0; | |
885 | } | |
886 | addr += incr; | |
887 | value |= flags; | |
888 | ib->ptr[ib->length_dw++] = value; | |
889 | ib->ptr[ib->length_dw++] = upper_32_bits(value); | |
890 | } | |
891 | } | |
892 | } | |
893 | ||
894 | /** | |
895 | * sdma_v3_0_vm_set_pte_pde - update the page tables using sDMA | |
896 | * | |
897 | * @ib: indirect buffer to fill with commands | |
898 | * @pe: addr of the page entry | |
899 | * @addr: dst addr to write into pe | |
900 | * @count: number of page entries to update | |
901 | * @incr: increase next addr by incr bytes | |
902 | * @flags: access flags | |
903 | * | |
904 | * Update the page tables using sDMA (CIK). | |
905 | */ | |
906 | static void sdma_v3_0_vm_set_pte_pde(struct amdgpu_ib *ib, | |
907 | uint64_t pe, | |
908 | uint64_t addr, unsigned count, | |
909 | uint32_t incr, uint32_t flags) | |
910 | { | |
911 | uint64_t value; | |
912 | unsigned ndw; | |
913 | ||
914 | while (count) { | |
915 | ndw = count; | |
916 | if (ndw > 0x7FFFF) | |
917 | ndw = 0x7FFFF; | |
918 | ||
919 | if (flags & AMDGPU_PTE_VALID) | |
920 | value = addr; | |
921 | else | |
922 | value = 0; | |
923 | ||
924 | /* for physically contiguous pages (vram) */ | |
925 | ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_GEN_PTEPDE); | |
926 | ib->ptr[ib->length_dw++] = pe; /* dst addr */ | |
927 | ib->ptr[ib->length_dw++] = upper_32_bits(pe); | |
928 | ib->ptr[ib->length_dw++] = flags; /* mask */ | |
929 | ib->ptr[ib->length_dw++] = 0; | |
930 | ib->ptr[ib->length_dw++] = value; /* value */ | |
931 | ib->ptr[ib->length_dw++] = upper_32_bits(value); | |
932 | ib->ptr[ib->length_dw++] = incr; /* increment size */ | |
933 | ib->ptr[ib->length_dw++] = 0; | |
934 | ib->ptr[ib->length_dw++] = ndw; /* number of entries */ | |
935 | ||
936 | pe += ndw * 8; | |
937 | addr += ndw * incr; | |
938 | count -= ndw; | |
939 | } | |
940 | } | |
941 | ||
942 | /** | |
943 | * sdma_v3_0_vm_pad_ib - pad the IB to the required number of dw | |
944 | * | |
945 | * @ib: indirect buffer to fill with padding | |
946 | * | |
947 | */ | |
948 | static void sdma_v3_0_vm_pad_ib(struct amdgpu_ib *ib) | |
949 | { | |
950 | while (ib->length_dw & 0x7) | |
951 | ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_NOP); | |
952 | } | |
953 | ||
954 | /** | |
955 | * sdma_v3_0_ring_emit_vm_flush - cik vm flush using sDMA | |
956 | * | |
957 | * @ring: amdgpu_ring pointer | |
958 | * @vm: amdgpu_vm pointer | |
959 | * | |
960 | * Update the page table base and flush the VM TLB | |
961 | * using sDMA (VI). | |
962 | */ | |
963 | static void sdma_v3_0_ring_emit_vm_flush(struct amdgpu_ring *ring, | |
964 | unsigned vm_id, uint64_t pd_addr) | |
965 | { | |
966 | u32 srbm_gfx_cntl = 0; | |
74a5d165 JX |
967 | u32 sh_mem_cfg = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE, |
968 | SH_MEM_ALIGNMENT_MODE_UNALIGNED); | |
aaa36a97 AD |
969 | |
970 | amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) | | |
971 | SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf)); | |
972 | if (vm_id < 8) { | |
973 | amdgpu_ring_write(ring, (mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vm_id)); | |
974 | } else { | |
975 | amdgpu_ring_write(ring, (mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + vm_id - 8)); | |
976 | } | |
977 | amdgpu_ring_write(ring, pd_addr >> 12); | |
978 | ||
979 | /* update SH_MEM_* regs */ | |
980 | srbm_gfx_cntl = REG_SET_FIELD(srbm_gfx_cntl, SRBM_GFX_CNTL, VMID, vm_id); | |
981 | amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) | | |
982 | SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf)); | |
983 | amdgpu_ring_write(ring, mmSRBM_GFX_CNTL); | |
984 | amdgpu_ring_write(ring, srbm_gfx_cntl); | |
985 | ||
986 | amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) | | |
987 | SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf)); | |
988 | amdgpu_ring_write(ring, mmSH_MEM_BASES); | |
989 | amdgpu_ring_write(ring, 0); | |
990 | ||
991 | amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) | | |
992 | SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf)); | |
993 | amdgpu_ring_write(ring, mmSH_MEM_CONFIG); | |
74a5d165 | 994 | amdgpu_ring_write(ring, sh_mem_cfg); |
aaa36a97 AD |
995 | |
996 | amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) | | |
997 | SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf)); | |
998 | amdgpu_ring_write(ring, mmSH_MEM_APE1_BASE); | |
999 | amdgpu_ring_write(ring, 1); | |
1000 | ||
1001 | amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) | | |
1002 | SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf)); | |
1003 | amdgpu_ring_write(ring, mmSH_MEM_APE1_LIMIT); | |
1004 | amdgpu_ring_write(ring, 0); | |
1005 | ||
1006 | srbm_gfx_cntl = REG_SET_FIELD(srbm_gfx_cntl, SRBM_GFX_CNTL, VMID, 0); | |
1007 | amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) | | |
1008 | SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf)); | |
1009 | amdgpu_ring_write(ring, mmSRBM_GFX_CNTL); | |
1010 | amdgpu_ring_write(ring, srbm_gfx_cntl); | |
1011 | ||
1012 | ||
1013 | /* flush TLB */ | |
1014 | amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) | | |
1015 | SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf)); | |
1016 | amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST); | |
1017 | amdgpu_ring_write(ring, 1 << vm_id); | |
1018 | ||
1019 | /* wait for flush */ | |
1020 | amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_POLL_REGMEM) | | |
1021 | SDMA_PKT_POLL_REGMEM_HEADER_HDP_FLUSH(0) | | |
1022 | SDMA_PKT_POLL_REGMEM_HEADER_FUNC(0)); /* always */ | |
1023 | amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST << 2); | |
1024 | amdgpu_ring_write(ring, 0); | |
1025 | amdgpu_ring_write(ring, 0); /* reference */ | |
1026 | amdgpu_ring_write(ring, 0); /* mask */ | |
1027 | amdgpu_ring_write(ring, SDMA_PKT_POLL_REGMEM_DW5_RETRY_COUNT(0xfff) | | |
1028 | SDMA_PKT_POLL_REGMEM_DW5_INTERVAL(10)); /* retry count, poll interval */ | |
1029 | } | |
1030 | ||
1031 | static int sdma_v3_0_early_init(struct amdgpu_device *adev) | |
1032 | { | |
1033 | sdma_v3_0_set_ring_funcs(adev); | |
1034 | sdma_v3_0_set_buffer_funcs(adev); | |
1035 | sdma_v3_0_set_vm_pte_funcs(adev); | |
1036 | sdma_v3_0_set_irq_funcs(adev); | |
1037 | ||
1038 | return 0; | |
1039 | } | |
1040 | ||
1041 | static int sdma_v3_0_sw_init(struct amdgpu_device *adev) | |
1042 | { | |
1043 | struct amdgpu_ring *ring; | |
1044 | int r; | |
1045 | ||
1046 | /* SDMA trap event */ | |
1047 | r = amdgpu_irq_add_id(adev, 224, &adev->sdma_trap_irq); | |
1048 | if (r) | |
1049 | return r; | |
1050 | ||
1051 | /* SDMA Privileged inst */ | |
1052 | r = amdgpu_irq_add_id(adev, 241, &adev->sdma_illegal_inst_irq); | |
1053 | if (r) | |
1054 | return r; | |
1055 | ||
1056 | /* SDMA Privileged inst */ | |
1057 | r = amdgpu_irq_add_id(adev, 247, &adev->sdma_illegal_inst_irq); | |
1058 | if (r) | |
1059 | return r; | |
1060 | ||
1061 | r = sdma_v3_0_init_microcode(adev); | |
1062 | if (r) { | |
1063 | DRM_ERROR("Failed to load sdma firmware!\n"); | |
1064 | return r; | |
1065 | } | |
1066 | ||
1067 | ring = &adev->sdma[0].ring; | |
1068 | ring->ring_obj = NULL; | |
1069 | ring->use_doorbell = true; | |
1070 | ring->doorbell_index = AMDGPU_DOORBELL_sDMA_ENGINE0; | |
1071 | ||
1072 | ring = &adev->sdma[1].ring; | |
1073 | ring->ring_obj = NULL; | |
1074 | ring->use_doorbell = true; | |
1075 | ring->doorbell_index = AMDGPU_DOORBELL_sDMA_ENGINE1; | |
1076 | ||
1077 | ring = &adev->sdma[0].ring; | |
1078 | sprintf(ring->name, "sdma0"); | |
1079 | r = amdgpu_ring_init(adev, ring, 256 * 1024, | |
1080 | SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), 0xf, | |
1081 | &adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP0, | |
1082 | AMDGPU_RING_TYPE_SDMA); | |
1083 | if (r) | |
1084 | return r; | |
1085 | ||
1086 | ring = &adev->sdma[1].ring; | |
1087 | sprintf(ring->name, "sdma1"); | |
1088 | r = amdgpu_ring_init(adev, ring, 256 * 1024, | |
1089 | SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), 0xf, | |
1090 | &adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP1, | |
1091 | AMDGPU_RING_TYPE_SDMA); | |
1092 | if (r) | |
1093 | return r; | |
1094 | ||
1095 | return r; | |
1096 | } | |
1097 | ||
1098 | static int sdma_v3_0_sw_fini(struct amdgpu_device *adev) | |
1099 | { | |
1100 | amdgpu_ring_fini(&adev->sdma[0].ring); | |
1101 | amdgpu_ring_fini(&adev->sdma[1].ring); | |
1102 | ||
1103 | return 0; | |
1104 | } | |
1105 | ||
1106 | static int sdma_v3_0_hw_init(struct amdgpu_device *adev) | |
1107 | { | |
1108 | int r; | |
1109 | ||
1110 | sdma_v3_0_init_golden_registers(adev); | |
1111 | ||
1112 | r = sdma_v3_0_start(adev); | |
1113 | if (r) | |
1114 | return r; | |
1115 | ||
1116 | return r; | |
1117 | } | |
1118 | ||
1119 | static int sdma_v3_0_hw_fini(struct amdgpu_device *adev) | |
1120 | { | |
1121 | sdma_v3_0_enable(adev, false); | |
1122 | ||
1123 | return 0; | |
1124 | } | |
1125 | ||
1126 | static int sdma_v3_0_suspend(struct amdgpu_device *adev) | |
1127 | { | |
1128 | ||
1129 | return sdma_v3_0_hw_fini(adev); | |
1130 | } | |
1131 | ||
1132 | static int sdma_v3_0_resume(struct amdgpu_device *adev) | |
1133 | { | |
1134 | ||
1135 | return sdma_v3_0_hw_init(adev); | |
1136 | } | |
1137 | ||
1138 | static bool sdma_v3_0_is_idle(struct amdgpu_device *adev) | |
1139 | { | |
1140 | u32 tmp = RREG32(mmSRBM_STATUS2); | |
1141 | ||
1142 | if (tmp & (SRBM_STATUS2__SDMA_BUSY_MASK | | |
1143 | SRBM_STATUS2__SDMA1_BUSY_MASK)) | |
1144 | return false; | |
1145 | ||
1146 | return true; | |
1147 | } | |
1148 | ||
1149 | static int sdma_v3_0_wait_for_idle(struct amdgpu_device *adev) | |
1150 | { | |
1151 | unsigned i; | |
1152 | u32 tmp; | |
1153 | ||
1154 | for (i = 0; i < adev->usec_timeout; i++) { | |
1155 | tmp = RREG32(mmSRBM_STATUS2) & (SRBM_STATUS2__SDMA_BUSY_MASK | | |
1156 | SRBM_STATUS2__SDMA1_BUSY_MASK); | |
1157 | ||
1158 | if (!tmp) | |
1159 | return 0; | |
1160 | udelay(1); | |
1161 | } | |
1162 | return -ETIMEDOUT; | |
1163 | } | |
1164 | ||
1165 | static void sdma_v3_0_print_status(struct amdgpu_device *adev) | |
1166 | { | |
1167 | int i, j; | |
1168 | ||
1169 | dev_info(adev->dev, "VI SDMA registers\n"); | |
1170 | dev_info(adev->dev, " SRBM_STATUS2=0x%08X\n", | |
1171 | RREG32(mmSRBM_STATUS2)); | |
1172 | for (i = 0; i < SDMA_MAX_INSTANCE; i++) { | |
1173 | dev_info(adev->dev, " SDMA%d_STATUS_REG=0x%08X\n", | |
1174 | i, RREG32(mmSDMA0_STATUS_REG + sdma_offsets[i])); | |
1175 | dev_info(adev->dev, " SDMA%d_F32_CNTL=0x%08X\n", | |
1176 | i, RREG32(mmSDMA0_F32_CNTL + sdma_offsets[i])); | |
1177 | dev_info(adev->dev, " SDMA%d_CNTL=0x%08X\n", | |
1178 | i, RREG32(mmSDMA0_CNTL + sdma_offsets[i])); | |
1179 | dev_info(adev->dev, " SDMA%d_SEM_WAIT_FAIL_TIMER_CNTL=0x%08X\n", | |
1180 | i, RREG32(mmSDMA0_SEM_WAIT_FAIL_TIMER_CNTL + sdma_offsets[i])); | |
1181 | dev_info(adev->dev, " SDMA%d_GFX_IB_CNTL=0x%08X\n", | |
1182 | i, RREG32(mmSDMA0_GFX_IB_CNTL + sdma_offsets[i])); | |
1183 | dev_info(adev->dev, " SDMA%d_GFX_RB_CNTL=0x%08X\n", | |
1184 | i, RREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i])); | |
1185 | dev_info(adev->dev, " SDMA%d_GFX_RB_RPTR=0x%08X\n", | |
1186 | i, RREG32(mmSDMA0_GFX_RB_RPTR + sdma_offsets[i])); | |
1187 | dev_info(adev->dev, " SDMA%d_GFX_RB_WPTR=0x%08X\n", | |
1188 | i, RREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[i])); | |
1189 | dev_info(adev->dev, " SDMA%d_GFX_RB_RPTR_ADDR_HI=0x%08X\n", | |
1190 | i, RREG32(mmSDMA0_GFX_RB_RPTR_ADDR_HI + sdma_offsets[i])); | |
1191 | dev_info(adev->dev, " SDMA%d_GFX_RB_RPTR_ADDR_LO=0x%08X\n", | |
1192 | i, RREG32(mmSDMA0_GFX_RB_RPTR_ADDR_LO + sdma_offsets[i])); | |
1193 | dev_info(adev->dev, " SDMA%d_GFX_RB_BASE=0x%08X\n", | |
1194 | i, RREG32(mmSDMA0_GFX_RB_BASE + sdma_offsets[i])); | |
1195 | dev_info(adev->dev, " SDMA%d_GFX_RB_BASE_HI=0x%08X\n", | |
1196 | i, RREG32(mmSDMA0_GFX_RB_BASE_HI + sdma_offsets[i])); | |
1197 | dev_info(adev->dev, " SDMA%d_GFX_DOORBELL=0x%08X\n", | |
1198 | i, RREG32(mmSDMA0_GFX_DOORBELL + sdma_offsets[i])); | |
1199 | mutex_lock(&adev->srbm_mutex); | |
1200 | for (j = 0; j < 16; j++) { | |
1201 | vi_srbm_select(adev, 0, 0, 0, j); | |
1202 | dev_info(adev->dev, " VM %d:\n", j); | |
1203 | dev_info(adev->dev, " SDMA%d_GFX_VIRTUAL_ADDR=0x%08X\n", | |
1204 | i, RREG32(mmSDMA0_GFX_VIRTUAL_ADDR + sdma_offsets[i])); | |
1205 | dev_info(adev->dev, " SDMA%d_GFX_APE1_CNTL=0x%08X\n", | |
1206 | i, RREG32(mmSDMA0_GFX_APE1_CNTL + sdma_offsets[i])); | |
1207 | } | |
1208 | vi_srbm_select(adev, 0, 0, 0, 0); | |
1209 | mutex_unlock(&adev->srbm_mutex); | |
1210 | } | |
1211 | } | |
1212 | ||
1213 | static int sdma_v3_0_soft_reset(struct amdgpu_device *adev) | |
1214 | { | |
1215 | u32 srbm_soft_reset = 0; | |
1216 | u32 tmp = RREG32(mmSRBM_STATUS2); | |
1217 | ||
1218 | if (tmp & SRBM_STATUS2__SDMA_BUSY_MASK) { | |
1219 | /* sdma0 */ | |
1220 | tmp = RREG32(mmSDMA0_F32_CNTL + SDMA0_REGISTER_OFFSET); | |
1221 | tmp = REG_SET_FIELD(tmp, SDMA0_F32_CNTL, HALT, 0); | |
1222 | WREG32(mmSDMA0_F32_CNTL + SDMA0_REGISTER_OFFSET, tmp); | |
1223 | srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_SDMA_MASK; | |
1224 | } | |
1225 | if (tmp & SRBM_STATUS2__SDMA1_BUSY_MASK) { | |
1226 | /* sdma1 */ | |
1227 | tmp = RREG32(mmSDMA0_F32_CNTL + SDMA1_REGISTER_OFFSET); | |
1228 | tmp = REG_SET_FIELD(tmp, SDMA0_F32_CNTL, HALT, 0); | |
1229 | WREG32(mmSDMA0_F32_CNTL + SDMA1_REGISTER_OFFSET, tmp); | |
1230 | srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_SDMA1_MASK; | |
1231 | } | |
1232 | ||
1233 | if (srbm_soft_reset) { | |
1234 | sdma_v3_0_print_status(adev); | |
1235 | ||
1236 | tmp = RREG32(mmSRBM_SOFT_RESET); | |
1237 | tmp |= srbm_soft_reset; | |
1238 | dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp); | |
1239 | WREG32(mmSRBM_SOFT_RESET, tmp); | |
1240 | tmp = RREG32(mmSRBM_SOFT_RESET); | |
1241 | ||
1242 | udelay(50); | |
1243 | ||
1244 | tmp &= ~srbm_soft_reset; | |
1245 | WREG32(mmSRBM_SOFT_RESET, tmp); | |
1246 | tmp = RREG32(mmSRBM_SOFT_RESET); | |
1247 | ||
1248 | /* Wait a little for things to settle down */ | |
1249 | udelay(50); | |
1250 | ||
1251 | sdma_v3_0_print_status(adev); | |
1252 | } | |
1253 | ||
1254 | return 0; | |
1255 | } | |
1256 | ||
1257 | static int sdma_v3_0_set_trap_irq_state(struct amdgpu_device *adev, | |
1258 | struct amdgpu_irq_src *source, | |
1259 | unsigned type, | |
1260 | enum amdgpu_interrupt_state state) | |
1261 | { | |
1262 | u32 sdma_cntl; | |
1263 | ||
1264 | switch (type) { | |
1265 | case AMDGPU_SDMA_IRQ_TRAP0: | |
1266 | switch (state) { | |
1267 | case AMDGPU_IRQ_STATE_DISABLE: | |
1268 | sdma_cntl = RREG32(mmSDMA0_CNTL + SDMA0_REGISTER_OFFSET); | |
1269 | sdma_cntl = REG_SET_FIELD(sdma_cntl, SDMA0_CNTL, TRAP_ENABLE, 0); | |
1270 | WREG32(mmSDMA0_CNTL + SDMA0_REGISTER_OFFSET, sdma_cntl); | |
1271 | break; | |
1272 | case AMDGPU_IRQ_STATE_ENABLE: | |
1273 | sdma_cntl = RREG32(mmSDMA0_CNTL + SDMA0_REGISTER_OFFSET); | |
1274 | sdma_cntl = REG_SET_FIELD(sdma_cntl, SDMA0_CNTL, TRAP_ENABLE, 1); | |
1275 | WREG32(mmSDMA0_CNTL + SDMA0_REGISTER_OFFSET, sdma_cntl); | |
1276 | break; | |
1277 | default: | |
1278 | break; | |
1279 | } | |
1280 | break; | |
1281 | case AMDGPU_SDMA_IRQ_TRAP1: | |
1282 | switch (state) { | |
1283 | case AMDGPU_IRQ_STATE_DISABLE: | |
1284 | sdma_cntl = RREG32(mmSDMA0_CNTL + SDMA1_REGISTER_OFFSET); | |
1285 | sdma_cntl = REG_SET_FIELD(sdma_cntl, SDMA0_CNTL, TRAP_ENABLE, 0); | |
1286 | WREG32(mmSDMA0_CNTL + SDMA1_REGISTER_OFFSET, sdma_cntl); | |
1287 | break; | |
1288 | case AMDGPU_IRQ_STATE_ENABLE: | |
1289 | sdma_cntl = RREG32(mmSDMA0_CNTL + SDMA1_REGISTER_OFFSET); | |
1290 | sdma_cntl = REG_SET_FIELD(sdma_cntl, SDMA0_CNTL, TRAP_ENABLE, 1); | |
1291 | WREG32(mmSDMA0_CNTL + SDMA1_REGISTER_OFFSET, sdma_cntl); | |
1292 | break; | |
1293 | default: | |
1294 | break; | |
1295 | } | |
1296 | break; | |
1297 | default: | |
1298 | break; | |
1299 | } | |
1300 | return 0; | |
1301 | } | |
1302 | ||
1303 | static int sdma_v3_0_process_trap_irq(struct amdgpu_device *adev, | |
1304 | struct amdgpu_irq_src *source, | |
1305 | struct amdgpu_iv_entry *entry) | |
1306 | { | |
1307 | u8 instance_id, queue_id; | |
1308 | ||
1309 | instance_id = (entry->ring_id & 0x3) >> 0; | |
1310 | queue_id = (entry->ring_id & 0xc) >> 2; | |
1311 | DRM_DEBUG("IH: SDMA trap\n"); | |
1312 | switch (instance_id) { | |
1313 | case 0: | |
1314 | switch (queue_id) { | |
1315 | case 0: | |
1316 | amdgpu_fence_process(&adev->sdma[0].ring); | |
1317 | break; | |
1318 | case 1: | |
1319 | /* XXX compute */ | |
1320 | break; | |
1321 | case 2: | |
1322 | /* XXX compute */ | |
1323 | break; | |
1324 | } | |
1325 | break; | |
1326 | case 1: | |
1327 | switch (queue_id) { | |
1328 | case 0: | |
1329 | amdgpu_fence_process(&adev->sdma[1].ring); | |
1330 | break; | |
1331 | case 1: | |
1332 | /* XXX compute */ | |
1333 | break; | |
1334 | case 2: | |
1335 | /* XXX compute */ | |
1336 | break; | |
1337 | } | |
1338 | break; | |
1339 | } | |
1340 | return 0; | |
1341 | } | |
1342 | ||
1343 | static int sdma_v3_0_process_illegal_inst_irq(struct amdgpu_device *adev, | |
1344 | struct amdgpu_irq_src *source, | |
1345 | struct amdgpu_iv_entry *entry) | |
1346 | { | |
1347 | DRM_ERROR("Illegal instruction in SDMA command stream\n"); | |
1348 | schedule_work(&adev->reset_work); | |
1349 | return 0; | |
1350 | } | |
1351 | ||
1352 | static int sdma_v3_0_set_clockgating_state(struct amdgpu_device *adev, | |
1353 | enum amdgpu_clockgating_state state) | |
1354 | { | |
1355 | /* XXX handled via the smc on VI */ | |
1356 | ||
1357 | return 0; | |
1358 | } | |
1359 | ||
1360 | static int sdma_v3_0_set_powergating_state(struct amdgpu_device *adev, | |
1361 | enum amdgpu_powergating_state state) | |
1362 | { | |
1363 | return 0; | |
1364 | } | |
1365 | ||
1366 | const struct amdgpu_ip_funcs sdma_v3_0_ip_funcs = { | |
1367 | .early_init = sdma_v3_0_early_init, | |
1368 | .late_init = NULL, | |
1369 | .sw_init = sdma_v3_0_sw_init, | |
1370 | .sw_fini = sdma_v3_0_sw_fini, | |
1371 | .hw_init = sdma_v3_0_hw_init, | |
1372 | .hw_fini = sdma_v3_0_hw_fini, | |
1373 | .suspend = sdma_v3_0_suspend, | |
1374 | .resume = sdma_v3_0_resume, | |
1375 | .is_idle = sdma_v3_0_is_idle, | |
1376 | .wait_for_idle = sdma_v3_0_wait_for_idle, | |
1377 | .soft_reset = sdma_v3_0_soft_reset, | |
1378 | .print_status = sdma_v3_0_print_status, | |
1379 | .set_clockgating_state = sdma_v3_0_set_clockgating_state, | |
1380 | .set_powergating_state = sdma_v3_0_set_powergating_state, | |
1381 | }; | |
1382 | ||
1383 | /** | |
1384 | * sdma_v3_0_ring_is_lockup - Check if the DMA engine is locked up | |
1385 | * | |
1386 | * @ring: amdgpu_ring structure holding ring information | |
1387 | * | |
1388 | * Check if the async DMA engine is locked up (VI). | |
1389 | * Returns true if the engine appears to be locked up, false if not. | |
1390 | */ | |
1391 | static bool sdma_v3_0_ring_is_lockup(struct amdgpu_ring *ring) | |
1392 | { | |
1393 | ||
1394 | if (sdma_v3_0_is_idle(ring->adev)) { | |
1395 | amdgpu_ring_lockup_update(ring); | |
1396 | return false; | |
1397 | } | |
1398 | return amdgpu_ring_test_lockup(ring); | |
1399 | } | |
1400 | ||
1401 | static const struct amdgpu_ring_funcs sdma_v3_0_ring_funcs = { | |
1402 | .get_rptr = sdma_v3_0_ring_get_rptr, | |
1403 | .get_wptr = sdma_v3_0_ring_get_wptr, | |
1404 | .set_wptr = sdma_v3_0_ring_set_wptr, | |
1405 | .parse_cs = NULL, | |
1406 | .emit_ib = sdma_v3_0_ring_emit_ib, | |
1407 | .emit_fence = sdma_v3_0_ring_emit_fence, | |
1408 | .emit_semaphore = sdma_v3_0_ring_emit_semaphore, | |
1409 | .emit_vm_flush = sdma_v3_0_ring_emit_vm_flush, | |
1410 | .test_ring = sdma_v3_0_ring_test_ring, | |
1411 | .test_ib = sdma_v3_0_ring_test_ib, | |
1412 | .is_lockup = sdma_v3_0_ring_is_lockup, | |
1413 | }; | |
1414 | ||
1415 | static void sdma_v3_0_set_ring_funcs(struct amdgpu_device *adev) | |
1416 | { | |
1417 | adev->sdma[0].ring.funcs = &sdma_v3_0_ring_funcs; | |
1418 | adev->sdma[1].ring.funcs = &sdma_v3_0_ring_funcs; | |
1419 | } | |
1420 | ||
1421 | static const struct amdgpu_irq_src_funcs sdma_v3_0_trap_irq_funcs = { | |
1422 | .set = sdma_v3_0_set_trap_irq_state, | |
1423 | .process = sdma_v3_0_process_trap_irq, | |
1424 | }; | |
1425 | ||
1426 | static const struct amdgpu_irq_src_funcs sdma_v3_0_illegal_inst_irq_funcs = { | |
1427 | .process = sdma_v3_0_process_illegal_inst_irq, | |
1428 | }; | |
1429 | ||
1430 | static void sdma_v3_0_set_irq_funcs(struct amdgpu_device *adev) | |
1431 | { | |
1432 | adev->sdma_trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST; | |
1433 | adev->sdma_trap_irq.funcs = &sdma_v3_0_trap_irq_funcs; | |
1434 | adev->sdma_illegal_inst_irq.funcs = &sdma_v3_0_illegal_inst_irq_funcs; | |
1435 | } | |
1436 | ||
1437 | /** | |
1438 | * sdma_v3_0_emit_copy_buffer - copy buffer using the sDMA engine | |
1439 | * | |
1440 | * @ring: amdgpu_ring structure holding ring information | |
1441 | * @src_offset: src GPU address | |
1442 | * @dst_offset: dst GPU address | |
1443 | * @byte_count: number of bytes to xfer | |
1444 | * | |
1445 | * Copy GPU buffers using the DMA engine (VI). | |
1446 | * Used by the amdgpu ttm implementation to move pages if | |
1447 | * registered as the asic copy callback. | |
1448 | */ | |
1449 | static void sdma_v3_0_emit_copy_buffer(struct amdgpu_ring *ring, | |
1450 | uint64_t src_offset, | |
1451 | uint64_t dst_offset, | |
1452 | uint32_t byte_count) | |
1453 | { | |
1454 | amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_COPY) | | |
1455 | SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR)); | |
1456 | amdgpu_ring_write(ring, byte_count); | |
1457 | amdgpu_ring_write(ring, 0); /* src/dst endian swap */ | |
1458 | amdgpu_ring_write(ring, lower_32_bits(src_offset)); | |
1459 | amdgpu_ring_write(ring, upper_32_bits(src_offset)); | |
1460 | amdgpu_ring_write(ring, lower_32_bits(dst_offset)); | |
1461 | amdgpu_ring_write(ring, upper_32_bits(dst_offset)); | |
1462 | } | |
1463 | ||
1464 | /** | |
1465 | * sdma_v3_0_emit_fill_buffer - fill buffer using the sDMA engine | |
1466 | * | |
1467 | * @ring: amdgpu_ring structure holding ring information | |
1468 | * @src_data: value to write to buffer | |
1469 | * @dst_offset: dst GPU address | |
1470 | * @byte_count: number of bytes to xfer | |
1471 | * | |
1472 | * Fill GPU buffers using the DMA engine (VI). | |
1473 | */ | |
1474 | static void sdma_v3_0_emit_fill_buffer(struct amdgpu_ring *ring, | |
1475 | uint32_t src_data, | |
1476 | uint64_t dst_offset, | |
1477 | uint32_t byte_count) | |
1478 | { | |
1479 | amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_CONST_FILL)); | |
1480 | amdgpu_ring_write(ring, lower_32_bits(dst_offset)); | |
1481 | amdgpu_ring_write(ring, upper_32_bits(dst_offset)); | |
1482 | amdgpu_ring_write(ring, src_data); | |
1483 | amdgpu_ring_write(ring, byte_count); | |
1484 | } | |
1485 | ||
1486 | static const struct amdgpu_buffer_funcs sdma_v3_0_buffer_funcs = { | |
1487 | .copy_max_bytes = 0x1fffff, | |
1488 | .copy_num_dw = 7, | |
1489 | .emit_copy_buffer = sdma_v3_0_emit_copy_buffer, | |
1490 | ||
1491 | .fill_max_bytes = 0x1fffff, | |
1492 | .fill_num_dw = 5, | |
1493 | .emit_fill_buffer = sdma_v3_0_emit_fill_buffer, | |
1494 | }; | |
1495 | ||
1496 | static void sdma_v3_0_set_buffer_funcs(struct amdgpu_device *adev) | |
1497 | { | |
1498 | if (adev->mman.buffer_funcs == NULL) { | |
1499 | adev->mman.buffer_funcs = &sdma_v3_0_buffer_funcs; | |
1500 | adev->mman.buffer_funcs_ring = &adev->sdma[0].ring; | |
1501 | } | |
1502 | } | |
1503 | ||
1504 | static const struct amdgpu_vm_pte_funcs sdma_v3_0_vm_pte_funcs = { | |
1505 | .copy_pte = sdma_v3_0_vm_copy_pte, | |
1506 | .write_pte = sdma_v3_0_vm_write_pte, | |
1507 | .set_pte_pde = sdma_v3_0_vm_set_pte_pde, | |
1508 | .pad_ib = sdma_v3_0_vm_pad_ib, | |
1509 | }; | |
1510 | ||
1511 | static void sdma_v3_0_set_vm_pte_funcs(struct amdgpu_device *adev) | |
1512 | { | |
1513 | if (adev->vm_manager.vm_pte_funcs == NULL) { | |
1514 | adev->vm_manager.vm_pte_funcs = &sdma_v3_0_vm_pte_funcs; | |
1515 | adev->vm_manager.vm_pte_funcs_ring = &adev->sdma[0].ring; | |
1516 | } | |
1517 | } |