aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--drivers/gpu/drm/amd/amdgpu/amdgpu.h22
-rw-r--r--drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c4
-rw-r--r--drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c4
-rw-r--r--drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c7
-rw-r--r--drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c10
-rw-r--r--drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c4
-rw-r--r--drivers/gpu/drm/amd/amdgpu/cik_sdma.c130
-rw-r--r--drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c156
-rw-r--r--drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c166
9 files changed, 245 insertions, 258 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 6647fb26ef25..afc984806c4c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -1708,7 +1708,7 @@ struct amdgpu_vce {
1708/* 1708/*
1709 * SDMA 1709 * SDMA
1710 */ 1710 */
1711struct amdgpu_sdma { 1711struct amdgpu_sdma_instance {
1712 /* SDMA firmware */ 1712 /* SDMA firmware */
1713 const struct firmware *fw; 1713 const struct firmware *fw;
1714 uint32_t fw_version; 1714 uint32_t fw_version;
@@ -1718,6 +1718,13 @@ struct amdgpu_sdma {
1718 bool burst_nop; 1718 bool burst_nop;
1719}; 1719};
1720 1720
1721struct amdgpu_sdma {
1722 struct amdgpu_sdma_instance instance[AMDGPU_MAX_SDMA_INSTANCES];
1723 struct amdgpu_irq_src trap_irq;
1724 struct amdgpu_irq_src illegal_inst_irq;
1725 int num_instances;
1726};
1727
1721/* 1728/*
1722 * Firmware 1729 * Firmware
1723 */ 1730 */
@@ -2064,9 +2071,7 @@ struct amdgpu_device {
2064 struct amdgpu_gfx gfx; 2071 struct amdgpu_gfx gfx;
2065 2072
2066 /* sdma */ 2073 /* sdma */
2067 struct amdgpu_sdma sdma[AMDGPU_MAX_SDMA_INSTANCES]; 2074 struct amdgpu_sdma sdma;
2068 struct amdgpu_irq_src sdma_trap_irq;
2069 struct amdgpu_irq_src sdma_illegal_inst_irq;
2070 2075
2071 /* uvd */ 2076 /* uvd */
2072 bool has_uvd; 2077 bool has_uvd;
@@ -2203,17 +2208,18 @@ static inline void amdgpu_ring_write(struct amdgpu_ring *ring, uint32_t v)
2203 ring->ring_free_dw--; 2208 ring->ring_free_dw--;
2204} 2209}
2205 2210
2206static inline struct amdgpu_sdma * amdgpu_get_sdma_instance(struct amdgpu_ring *ring) 2211static inline struct amdgpu_sdma_instance *
2212amdgpu_get_sdma_instance(struct amdgpu_ring *ring)
2207{ 2213{
2208 struct amdgpu_device *adev = ring->adev; 2214 struct amdgpu_device *adev = ring->adev;
2209 int i; 2215 int i;
2210 2216
2211 for (i = 0; i < AMDGPU_MAX_SDMA_INSTANCES; i++) 2217 for (i = 0; i < adev->sdma.num_instances; i++)
2212 if (&adev->sdma[i].ring == ring) 2218 if (&adev->sdma.instance[i].ring == ring)
2213 break; 2219 break;
2214 2220
2215 if (i < AMDGPU_MAX_SDMA_INSTANCES) 2221 if (i < AMDGPU_MAX_SDMA_INSTANCES)
2216 return &adev->sdma[i]; 2222 return &adev->sdma.instance[i];
2217 else 2223 else
2218 return NULL; 2224 return NULL;
2219} 2225}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
index dd2037bc0b4a..0e1376317683 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
@@ -649,12 +649,12 @@ static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type)
649 649
650 case KGD_ENGINE_SDMA1: 650 case KGD_ENGINE_SDMA1:
651 hdr = (const union amdgpu_firmware_header *) 651 hdr = (const union amdgpu_firmware_header *)
652 adev->sdma[0].fw->data; 652 adev->sdma.instance[0].fw->data;
653 break; 653 break;
654 654
655 case KGD_ENGINE_SDMA2: 655 case KGD_ENGINE_SDMA2:
656 hdr = (const union amdgpu_firmware_header *) 656 hdr = (const union amdgpu_firmware_header *)
657 adev->sdma[1].fw->data; 657 adev->sdma.instance[1].fw->data;
658 break; 658 break;
659 659
660 default: 660 default:
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
index dfd1d503bccf..79fa5c7de856 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
@@ -523,12 +523,12 @@ static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type)
523 523
524 case KGD_ENGINE_SDMA1: 524 case KGD_ENGINE_SDMA1:
525 hdr = (const union amdgpu_firmware_header *) 525 hdr = (const union amdgpu_firmware_header *)
526 adev->sdma[0].fw->data; 526 adev->sdma.instance[0].fw->data;
527 break; 527 break;
528 528
529 case KGD_ENGINE_SDMA2: 529 case KGD_ENGINE_SDMA2:
530 hdr = (const union amdgpu_firmware_header *) 530 hdr = (const union amdgpu_firmware_header *)
531 adev->sdma[1].fw->data; 531 adev->sdma.instance[1].fw->data;
532 break; 532 break;
533 533
534 default: 534 default:
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 749420f1ea6f..29fc45ce64dd 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -104,10 +104,11 @@ int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type,
104 } 104 }
105 break; 105 break;
106 case AMDGPU_HW_IP_DMA: 106 case AMDGPU_HW_IP_DMA:
107 if (ring < 2) { 107 if (ring < adev->sdma.num_instances) {
108 *out_ring = &adev->sdma[ring].ring; 108 *out_ring = &adev->sdma.instance[ring].ring;
109 } else { 109 } else {
110 DRM_ERROR("only two SDMA rings are supported\n"); 110 DRM_ERROR("only %d SDMA rings are supported\n",
111 adev->sdma.num_instances);
111 return -EINVAL; 112 return -EINVAL;
112 } 113 }
113 break; 114 break;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index 78233225dd36..3f5f2d58ad94 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -218,8 +218,8 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
218 break; 218 break;
219 case AMDGPU_HW_IP_DMA: 219 case AMDGPU_HW_IP_DMA:
220 type = AMD_IP_BLOCK_TYPE_SDMA; 220 type = AMD_IP_BLOCK_TYPE_SDMA;
221 ring_mask = adev->sdma[0].ring.ready ? 1 : 0; 221 for (i = 0; i < adev->sdma.num_instances; i++)
222 ring_mask |= ((adev->sdma[1].ring.ready ? 1 : 0) << 1); 222 ring_mask |= ((adev->sdma.instance[i].ring.ready ? 1 : 0) << i);
223 ib_start_alignment = AMDGPU_GPU_PAGE_SIZE; 223 ib_start_alignment = AMDGPU_GPU_PAGE_SIZE;
224 ib_size_alignment = 1; 224 ib_size_alignment = 1;
225 break; 225 break;
@@ -341,10 +341,10 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
341 fw_info.feature = 0; 341 fw_info.feature = 0;
342 break; 342 break;
343 case AMDGPU_INFO_FW_SDMA: 343 case AMDGPU_INFO_FW_SDMA:
344 if (info->query_fw.index >= 2) 344 if (info->query_fw.index >= adev->sdma.num_instances)
345 return -EINVAL; 345 return -EINVAL;
346 fw_info.ver = adev->sdma[info->query_fw.index].fw_version; 346 fw_info.ver = adev->sdma.instance[info->query_fw.index].fw_version;
347 fw_info.feature = adev->sdma[info->query_fw.index].feature_version; 347 fw_info.feature = adev->sdma.instance[info->query_fw.index].feature_version;
348 break; 348 break;
349 default: 349 default:
350 return -EINVAL; 350 return -EINVAL;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
index 30dce235ddeb..b13a74b273a6 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
@@ -540,8 +540,8 @@ static int amdgpu_debugfs_ring_info(struct seq_file *m, void *data)
540static int amdgpu_gfx_index = offsetof(struct amdgpu_device, gfx.gfx_ring[0]); 540static int amdgpu_gfx_index = offsetof(struct amdgpu_device, gfx.gfx_ring[0]);
541static int cayman_cp1_index = offsetof(struct amdgpu_device, gfx.compute_ring[0]); 541static int cayman_cp1_index = offsetof(struct amdgpu_device, gfx.compute_ring[0]);
542static int cayman_cp2_index = offsetof(struct amdgpu_device, gfx.compute_ring[1]); 542static int cayman_cp2_index = offsetof(struct amdgpu_device, gfx.compute_ring[1]);
543static int amdgpu_dma1_index = offsetof(struct amdgpu_device, sdma[0].ring); 543static int amdgpu_dma1_index = offsetof(struct amdgpu_device, sdma.instance[0].ring);
544static int amdgpu_dma2_index = offsetof(struct amdgpu_device, sdma[1].ring); 544static int amdgpu_dma2_index = offsetof(struct amdgpu_device, sdma.instance[1].ring);
545static int r600_uvd_index = offsetof(struct amdgpu_device, uvd.ring); 545static int r600_uvd_index = offsetof(struct amdgpu_device, uvd.ring);
546static int si_vce1_index = offsetof(struct amdgpu_device, vce.ring[0]); 546static int si_vce1_index = offsetof(struct amdgpu_device, vce.ring[0]);
547static int si_vce2_index = offsetof(struct amdgpu_device, vce.ring[1]); 547static int si_vce2_index = offsetof(struct amdgpu_device, vce.ring[1]);
diff --git a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
index 9ea9de457da3..814598e76c98 100644
--- a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
+++ b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
@@ -96,7 +96,7 @@ static int cik_sdma_init_microcode(struct amdgpu_device *adev)
96{ 96{
97 const char *chip_name; 97 const char *chip_name;
98 char fw_name[30]; 98 char fw_name[30];
99 int err, i; 99 int err = 0, i;
100 100
101 DRM_DEBUG("\n"); 101 DRM_DEBUG("\n");
102 102
@@ -119,24 +119,24 @@ static int cik_sdma_init_microcode(struct amdgpu_device *adev)
119 default: BUG(); 119 default: BUG();
120 } 120 }
121 121
122 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 122 for (i = 0; i < adev->sdma.num_instances; i++) {
123 if (i == 0) 123 if (i == 0)
124 snprintf(fw_name, sizeof(fw_name), "radeon/%s_sdma.bin", chip_name); 124 snprintf(fw_name, sizeof(fw_name), "radeon/%s_sdma.bin", chip_name);
125 else 125 else
126 snprintf(fw_name, sizeof(fw_name), "radeon/%s_sdma1.bin", chip_name); 126 snprintf(fw_name, sizeof(fw_name), "radeon/%s_sdma1.bin", chip_name);
127 err = request_firmware(&adev->sdma[i].fw, fw_name, adev->dev); 127 err = request_firmware(&adev->sdma.instance[i].fw, fw_name, adev->dev);
128 if (err) 128 if (err)
129 goto out; 129 goto out;
130 err = amdgpu_ucode_validate(adev->sdma[i].fw); 130 err = amdgpu_ucode_validate(adev->sdma.instance[i].fw);
131 } 131 }
132out: 132out:
133 if (err) { 133 if (err) {
134 printk(KERN_ERR 134 printk(KERN_ERR
135 "cik_sdma: Failed to load firmware \"%s\"\n", 135 "cik_sdma: Failed to load firmware \"%s\"\n",
136 fw_name); 136 fw_name);
137 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 137 for (i = 0; i < adev->sdma.num_instances; i++) {
138 release_firmware(adev->sdma[i].fw); 138 release_firmware(adev->sdma.instance[i].fw);
139 adev->sdma[i].fw = NULL; 139 adev->sdma.instance[i].fw = NULL;
140 } 140 }
141 } 141 }
142 return err; 142 return err;
@@ -168,7 +168,7 @@ static uint32_t cik_sdma_ring_get_rptr(struct amdgpu_ring *ring)
168static uint32_t cik_sdma_ring_get_wptr(struct amdgpu_ring *ring) 168static uint32_t cik_sdma_ring_get_wptr(struct amdgpu_ring *ring)
169{ 169{
170 struct amdgpu_device *adev = ring->adev; 170 struct amdgpu_device *adev = ring->adev;
171 u32 me = (ring == &adev->sdma[0].ring) ? 0 : 1; 171 u32 me = (ring == &adev->sdma.instance[0].ring) ? 0 : 1;
172 172
173 return (RREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me]) & 0x3fffc) >> 2; 173 return (RREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me]) & 0x3fffc) >> 2;
174} 174}
@@ -183,14 +183,14 @@ static uint32_t cik_sdma_ring_get_wptr(struct amdgpu_ring *ring)
183static void cik_sdma_ring_set_wptr(struct amdgpu_ring *ring) 183static void cik_sdma_ring_set_wptr(struct amdgpu_ring *ring)
184{ 184{
185 struct amdgpu_device *adev = ring->adev; 185 struct amdgpu_device *adev = ring->adev;
186 u32 me = (ring == &adev->sdma[0].ring) ? 0 : 1; 186 u32 me = (ring == &adev->sdma.instance[0].ring) ? 0 : 1;
187 187
188 WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me], (ring->wptr << 2) & 0x3fffc); 188 WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me], (ring->wptr << 2) & 0x3fffc);
189} 189}
190 190
191static void cik_sdma_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) 191static void cik_sdma_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
192{ 192{
193 struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ring); 193 struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring);
194 int i; 194 int i;
195 195
196 for (i = 0; i < count; i++) 196 for (i = 0; i < count; i++)
@@ -248,7 +248,7 @@ static void cik_sdma_ring_emit_hdp_flush(struct amdgpu_ring *ring)
248 SDMA_POLL_REG_MEM_EXTRA_FUNC(3)); /* == */ 248 SDMA_POLL_REG_MEM_EXTRA_FUNC(3)); /* == */
249 u32 ref_and_mask; 249 u32 ref_and_mask;
250 250
251 if (ring == &ring->adev->sdma[0].ring) 251 if (ring == &ring->adev->sdma.instance[0].ring)
252 ref_and_mask = GPU_HDP_FLUSH_DONE__SDMA0_MASK; 252 ref_and_mask = GPU_HDP_FLUSH_DONE__SDMA0_MASK;
253 else 253 else
254 ref_and_mask = GPU_HDP_FLUSH_DONE__SDMA1_MASK; 254 ref_and_mask = GPU_HDP_FLUSH_DONE__SDMA1_MASK;
@@ -327,8 +327,8 @@ static bool cik_sdma_ring_emit_semaphore(struct amdgpu_ring *ring,
327 */ 327 */
328static void cik_sdma_gfx_stop(struct amdgpu_device *adev) 328static void cik_sdma_gfx_stop(struct amdgpu_device *adev)
329{ 329{
330 struct amdgpu_ring *sdma0 = &adev->sdma[0].ring; 330 struct amdgpu_ring *sdma0 = &adev->sdma.instance[0].ring;
331 struct amdgpu_ring *sdma1 = &adev->sdma[1].ring; 331 struct amdgpu_ring *sdma1 = &adev->sdma.instance[1].ring;
332 u32 rb_cntl; 332 u32 rb_cntl;
333 int i; 333 int i;
334 334
@@ -336,7 +336,7 @@ static void cik_sdma_gfx_stop(struct amdgpu_device *adev)
336 (adev->mman.buffer_funcs_ring == sdma1)) 336 (adev->mman.buffer_funcs_ring == sdma1))
337 amdgpu_ttm_set_active_vram_size(adev, adev->mc.visible_vram_size); 337 amdgpu_ttm_set_active_vram_size(adev, adev->mc.visible_vram_size);
338 338
339 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 339 for (i = 0; i < adev->sdma.num_instances; i++) {
340 rb_cntl = RREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i]); 340 rb_cntl = RREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i]);
341 rb_cntl &= ~SDMA0_GFX_RB_CNTL__RB_ENABLE_MASK; 341 rb_cntl &= ~SDMA0_GFX_RB_CNTL__RB_ENABLE_MASK;
342 WREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i], rb_cntl); 342 WREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i], rb_cntl);
@@ -376,7 +376,7 @@ static void cik_sdma_enable(struct amdgpu_device *adev, bool enable)
376 cik_sdma_rlc_stop(adev); 376 cik_sdma_rlc_stop(adev);
377 } 377 }
378 378
379 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 379 for (i = 0; i < adev->sdma.num_instances; i++) {
380 me_cntl = RREG32(mmSDMA0_F32_CNTL + sdma_offsets[i]); 380 me_cntl = RREG32(mmSDMA0_F32_CNTL + sdma_offsets[i]);
381 if (enable) 381 if (enable)
382 me_cntl &= ~SDMA0_F32_CNTL__HALT_MASK; 382 me_cntl &= ~SDMA0_F32_CNTL__HALT_MASK;
@@ -402,8 +402,8 @@ static int cik_sdma_gfx_resume(struct amdgpu_device *adev)
402 u32 wb_offset; 402 u32 wb_offset;
403 int i, j, r; 403 int i, j, r;
404 404
405 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 405 for (i = 0; i < adev->sdma.num_instances; i++) {
406 ring = &adev->sdma[i].ring; 406 ring = &adev->sdma.instance[i].ring;
407 wb_offset = (ring->rptr_offs * 4); 407 wb_offset = (ring->rptr_offs * 4);
408 408
409 mutex_lock(&adev->srbm_mutex); 409 mutex_lock(&adev->srbm_mutex);
@@ -502,26 +502,25 @@ static int cik_sdma_load_microcode(struct amdgpu_device *adev)
502 u32 fw_size; 502 u32 fw_size;
503 int i, j; 503 int i, j;
504 504
505 if (!adev->sdma[0].fw || !adev->sdma[1].fw)
506 return -EINVAL;
507
508 /* halt the MEs */ 505 /* halt the MEs */
509 cik_sdma_enable(adev, false); 506 cik_sdma_enable(adev, false);
510 507
511 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 508 for (i = 0; i < adev->sdma.num_instances; i++) {
512 hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data; 509 if (!adev->sdma.instance[i].fw)
510 return -EINVAL;
511 hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma.instance[i].fw->data;
513 amdgpu_ucode_print_sdma_hdr(&hdr->header); 512 amdgpu_ucode_print_sdma_hdr(&hdr->header);
514 fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4; 513 fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
515 adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version); 514 adev->sdma.instance[i].fw_version = le32_to_cpu(hdr->header.ucode_version);
516 adev->sdma[i].feature_version = le32_to_cpu(hdr->ucode_feature_version); 515 adev->sdma.instance[i].feature_version = le32_to_cpu(hdr->ucode_feature_version);
517 if (adev->sdma[i].feature_version >= 20) 516 if (adev->sdma.instance[i].feature_version >= 20)
518 adev->sdma[i].burst_nop = true; 517 adev->sdma.instance[i].burst_nop = true;
519 fw_data = (const __le32 *) 518 fw_data = (const __le32 *)
520 (adev->sdma[i].fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); 519 (adev->sdma.instance[i].fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
521 WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], 0); 520 WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], 0);
522 for (j = 0; j < fw_size; j++) 521 for (j = 0; j < fw_size; j++)
523 WREG32(mmSDMA0_UCODE_DATA + sdma_offsets[i], le32_to_cpup(fw_data++)); 522 WREG32(mmSDMA0_UCODE_DATA + sdma_offsets[i], le32_to_cpup(fw_data++));
524 WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], adev->sdma[i].fw_version); 523 WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], adev->sdma.instance[i].fw_version);
525 } 524 }
526 525
527 return 0; 526 return 0;
@@ -830,7 +829,7 @@ static void cik_sdma_vm_set_pte_pde(struct amdgpu_ib *ib,
830 */ 829 */
831static void cik_sdma_vm_pad_ib(struct amdgpu_ib *ib) 830static void cik_sdma_vm_pad_ib(struct amdgpu_ib *ib)
832{ 831{
833 struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ib->ring); 832 struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ib->ring);
834 u32 pad_count; 833 u32 pad_count;
835 int i; 834 int i;
836 835
@@ -934,6 +933,8 @@ static int cik_sdma_early_init(void *handle)
934{ 933{
935 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 934 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
936 935
936 adev->sdma.num_instances = SDMA_MAX_INSTANCE;
937
937 cik_sdma_set_ring_funcs(adev); 938 cik_sdma_set_ring_funcs(adev);
938 cik_sdma_set_irq_funcs(adev); 939 cik_sdma_set_irq_funcs(adev);
939 cik_sdma_set_buffer_funcs(adev); 940 cik_sdma_set_buffer_funcs(adev);
@@ -946,7 +947,7 @@ static int cik_sdma_sw_init(void *handle)
946{ 947{
947 struct amdgpu_ring *ring; 948 struct amdgpu_ring *ring;
948 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 949 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
949 int r; 950 int r, i;
950 951
951 r = cik_sdma_init_microcode(adev); 952 r = cik_sdma_init_microcode(adev);
952 if (r) { 953 if (r) {
@@ -955,43 +956,33 @@ static int cik_sdma_sw_init(void *handle)
955 } 956 }
956 957
957 /* SDMA trap event */ 958 /* SDMA trap event */
958 r = amdgpu_irq_add_id(adev, 224, &adev->sdma_trap_irq); 959 r = amdgpu_irq_add_id(adev, 224, &adev->sdma.trap_irq);
959 if (r) 960 if (r)
960 return r; 961 return r;
961 962
962 /* SDMA Privileged inst */ 963 /* SDMA Privileged inst */
963 r = amdgpu_irq_add_id(adev, 241, &adev->sdma_illegal_inst_irq); 964 r = amdgpu_irq_add_id(adev, 241, &adev->sdma.illegal_inst_irq);
964 if (r) 965 if (r)
965 return r; 966 return r;
966 967
967 /* SDMA Privileged inst */ 968 /* SDMA Privileged inst */
968 r = amdgpu_irq_add_id(adev, 247, &adev->sdma_illegal_inst_irq); 969 r = amdgpu_irq_add_id(adev, 247, &adev->sdma.illegal_inst_irq);
969 if (r)
970 return r;
971
972 ring = &adev->sdma[0].ring;
973 ring->ring_obj = NULL;
974
975 ring = &adev->sdma[1].ring;
976 ring->ring_obj = NULL;
977
978 ring = &adev->sdma[0].ring;
979 sprintf(ring->name, "sdma0");
980 r = amdgpu_ring_init(adev, ring, 256 * 1024,
981 SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0), 0xf,
982 &adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP0,
983 AMDGPU_RING_TYPE_SDMA);
984 if (r) 970 if (r)
985 return r; 971 return r;
986 972
987 ring = &adev->sdma[1].ring; 973 for (i = 0; i < adev->sdma.num_instances; i++) {
988 sprintf(ring->name, "sdma1"); 974 ring = &adev->sdma.instance[i].ring;
989 r = amdgpu_ring_init(adev, ring, 256 * 1024, 975 ring->ring_obj = NULL;
990 SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0), 0xf, 976 sprintf(ring->name, "sdma%d", i);
991 &adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP1, 977 r = amdgpu_ring_init(adev, ring, 256 * 1024,
992 AMDGPU_RING_TYPE_SDMA); 978 SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0), 0xf,
993 if (r) 979 &adev->sdma.trap_irq,
994 return r; 980 (i == 0) ?
981 AMDGPU_SDMA_IRQ_TRAP0 : AMDGPU_SDMA_IRQ_TRAP1,
982 AMDGPU_RING_TYPE_SDMA);
983 if (r)
984 return r;
985 }
995 986
996 return r; 987 return r;
997} 988}
@@ -999,9 +990,10 @@ static int cik_sdma_sw_init(void *handle)
999static int cik_sdma_sw_fini(void *handle) 990static int cik_sdma_sw_fini(void *handle)
1000{ 991{
1001 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 992 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
993 int i;
1002 994
1003 amdgpu_ring_fini(&adev->sdma[0].ring); 995 for (i = 0; i < adev->sdma.num_instances; i++)
1004 amdgpu_ring_fini(&adev->sdma[1].ring); 996 amdgpu_ring_fini(&adev->sdma.instance[i].ring);
1005 997
1006 return 0; 998 return 0;
1007} 999}
@@ -1078,7 +1070,7 @@ static void cik_sdma_print_status(void *handle)
1078 dev_info(adev->dev, "CIK SDMA registers\n"); 1070 dev_info(adev->dev, "CIK SDMA registers\n");
1079 dev_info(adev->dev, " SRBM_STATUS2=0x%08X\n", 1071 dev_info(adev->dev, " SRBM_STATUS2=0x%08X\n",
1080 RREG32(mmSRBM_STATUS2)); 1072 RREG32(mmSRBM_STATUS2));
1081 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 1073 for (i = 0; i < adev->sdma.num_instances; i++) {
1082 dev_info(adev->dev, " SDMA%d_STATUS_REG=0x%08X\n", 1074 dev_info(adev->dev, " SDMA%d_STATUS_REG=0x%08X\n",
1083 i, RREG32(mmSDMA0_STATUS_REG + sdma_offsets[i])); 1075 i, RREG32(mmSDMA0_STATUS_REG + sdma_offsets[i]));
1084 dev_info(adev->dev, " SDMA%d_ME_CNTL=0x%08X\n", 1076 dev_info(adev->dev, " SDMA%d_ME_CNTL=0x%08X\n",
@@ -1223,7 +1215,7 @@ static int cik_sdma_process_trap_irq(struct amdgpu_device *adev,
1223 case 0: 1215 case 0:
1224 switch (queue_id) { 1216 switch (queue_id) {
1225 case 0: 1217 case 0:
1226 amdgpu_fence_process(&adev->sdma[0].ring); 1218 amdgpu_fence_process(&adev->sdma.instance[0].ring);
1227 break; 1219 break;
1228 case 1: 1220 case 1:
1229 /* XXX compute */ 1221 /* XXX compute */
@@ -1236,7 +1228,7 @@ static int cik_sdma_process_trap_irq(struct amdgpu_device *adev,
1236 case 1: 1228 case 1:
1237 switch (queue_id) { 1229 switch (queue_id) {
1238 case 0: 1230 case 0:
1239 amdgpu_fence_process(&adev->sdma[1].ring); 1231 amdgpu_fence_process(&adev->sdma.instance[1].ring);
1240 break; 1232 break;
1241 case 1: 1233 case 1:
1242 /* XXX compute */ 1234 /* XXX compute */
@@ -1334,8 +1326,10 @@ static const struct amdgpu_ring_funcs cik_sdma_ring_funcs = {
1334 1326
1335static void cik_sdma_set_ring_funcs(struct amdgpu_device *adev) 1327static void cik_sdma_set_ring_funcs(struct amdgpu_device *adev)
1336{ 1328{
1337 adev->sdma[0].ring.funcs = &cik_sdma_ring_funcs; 1329 int i;
1338 adev->sdma[1].ring.funcs = &cik_sdma_ring_funcs; 1330
1331 for (i = 0; i < adev->sdma.num_instances; i++)
1332 adev->sdma.instance[i].ring.funcs = &cik_sdma_ring_funcs;
1339} 1333}
1340 1334
1341static const struct amdgpu_irq_src_funcs cik_sdma_trap_irq_funcs = { 1335static const struct amdgpu_irq_src_funcs cik_sdma_trap_irq_funcs = {
@@ -1349,9 +1343,9 @@ static const struct amdgpu_irq_src_funcs cik_sdma_illegal_inst_irq_funcs = {
1349 1343
1350static void cik_sdma_set_irq_funcs(struct amdgpu_device *adev) 1344static void cik_sdma_set_irq_funcs(struct amdgpu_device *adev)
1351{ 1345{
1352 adev->sdma_trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST; 1346 adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST;
1353 adev->sdma_trap_irq.funcs = &cik_sdma_trap_irq_funcs; 1347 adev->sdma.trap_irq.funcs = &cik_sdma_trap_irq_funcs;
1354 adev->sdma_illegal_inst_irq.funcs = &cik_sdma_illegal_inst_irq_funcs; 1348 adev->sdma.illegal_inst_irq.funcs = &cik_sdma_illegal_inst_irq_funcs;
1355} 1349}
1356 1350
1357/** 1351/**
@@ -1416,7 +1410,7 @@ static void cik_sdma_set_buffer_funcs(struct amdgpu_device *adev)
1416{ 1410{
1417 if (adev->mman.buffer_funcs == NULL) { 1411 if (adev->mman.buffer_funcs == NULL) {
1418 adev->mman.buffer_funcs = &cik_sdma_buffer_funcs; 1412 adev->mman.buffer_funcs = &cik_sdma_buffer_funcs;
1419 adev->mman.buffer_funcs_ring = &adev->sdma[0].ring; 1413 adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
1420 } 1414 }
1421} 1415}
1422 1416
@@ -1431,7 +1425,7 @@ static void cik_sdma_set_vm_pte_funcs(struct amdgpu_device *adev)
1431{ 1425{
1432 if (adev->vm_manager.vm_pte_funcs == NULL) { 1426 if (adev->vm_manager.vm_pte_funcs == NULL) {
1433 adev->vm_manager.vm_pte_funcs = &cik_sdma_vm_pte_funcs; 1427 adev->vm_manager.vm_pte_funcs = &cik_sdma_vm_pte_funcs;
1434 adev->vm_manager.vm_pte_funcs_ring = &adev->sdma[0].ring; 1428 adev->vm_manager.vm_pte_funcs_ring = &adev->sdma.instance[0].ring;
1435 adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true; 1429 adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true;
1436 } 1430 }
1437} 1431}
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
index 14e87234171a..f8b868c7c496 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
@@ -118,7 +118,7 @@ static int sdma_v2_4_init_microcode(struct amdgpu_device *adev)
118{ 118{
119 const char *chip_name; 119 const char *chip_name;
120 char fw_name[30]; 120 char fw_name[30];
121 int err, i; 121 int err = 0, i;
122 struct amdgpu_firmware_info *info = NULL; 122 struct amdgpu_firmware_info *info = NULL;
123 const struct common_firmware_header *header = NULL; 123 const struct common_firmware_header *header = NULL;
124 const struct sdma_firmware_header_v1_0 *hdr; 124 const struct sdma_firmware_header_v1_0 *hdr;
@@ -132,27 +132,27 @@ static int sdma_v2_4_init_microcode(struct amdgpu_device *adev)
132 default: BUG(); 132 default: BUG();
133 } 133 }
134 134
135 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 135 for (i = 0; i < adev->sdma.num_instances; i++) {
136 if (i == 0) 136 if (i == 0)
137 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma.bin", chip_name); 137 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma.bin", chip_name);
138 else 138 else
139 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma1.bin", chip_name); 139 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma1.bin", chip_name);
140 err = request_firmware(&adev->sdma[i].fw, fw_name, adev->dev); 140 err = request_firmware(&adev->sdma.instance[i].fw, fw_name, adev->dev);
141 if (err) 141 if (err)
142 goto out; 142 goto out;
143 err = amdgpu_ucode_validate(adev->sdma[i].fw); 143 err = amdgpu_ucode_validate(adev->sdma.instance[i].fw);
144 if (err) 144 if (err)
145 goto out; 145 goto out;
146 hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data; 146 hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma.instance[i].fw->data;
147 adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version); 147 adev->sdma.instance[i].fw_version = le32_to_cpu(hdr->header.ucode_version);
148 adev->sdma[i].feature_version = le32_to_cpu(hdr->ucode_feature_version); 148 adev->sdma.instance[i].feature_version = le32_to_cpu(hdr->ucode_feature_version);
149 if (adev->sdma[i].feature_version >= 20) 149 if (adev->sdma.instance[i].feature_version >= 20)
150 adev->sdma[i].burst_nop = true; 150 adev->sdma.instance[i].burst_nop = true;
151 151
152 if (adev->firmware.smu_load) { 152 if (adev->firmware.smu_load) {
153 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_SDMA0 + i]; 153 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_SDMA0 + i];
154 info->ucode_id = AMDGPU_UCODE_ID_SDMA0 + i; 154 info->ucode_id = AMDGPU_UCODE_ID_SDMA0 + i;
155 info->fw = adev->sdma[i].fw; 155 info->fw = adev->sdma.instance[i].fw;
156 header = (const struct common_firmware_header *)info->fw->data; 156 header = (const struct common_firmware_header *)info->fw->data;
157 adev->firmware.fw_size += 157 adev->firmware.fw_size +=
158 ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE); 158 ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
@@ -164,9 +164,9 @@ out:
164 printk(KERN_ERR 164 printk(KERN_ERR
165 "sdma_v2_4: Failed to load firmware \"%s\"\n", 165 "sdma_v2_4: Failed to load firmware \"%s\"\n",
166 fw_name); 166 fw_name);
167 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 167 for (i = 0; i < adev->sdma.num_instances; i++) {
168 release_firmware(adev->sdma[i].fw); 168 release_firmware(adev->sdma.instance[i].fw);
169 adev->sdma[i].fw = NULL; 169 adev->sdma.instance[i].fw = NULL;
170 } 170 }
171 } 171 }
172 return err; 172 return err;
@@ -199,7 +199,7 @@ static uint32_t sdma_v2_4_ring_get_rptr(struct amdgpu_ring *ring)
199static uint32_t sdma_v2_4_ring_get_wptr(struct amdgpu_ring *ring) 199static uint32_t sdma_v2_4_ring_get_wptr(struct amdgpu_ring *ring)
200{ 200{
201 struct amdgpu_device *adev = ring->adev; 201 struct amdgpu_device *adev = ring->adev;
202 int me = (ring == &ring->adev->sdma[0].ring) ? 0 : 1; 202 int me = (ring == &ring->adev->sdma.instance[0].ring) ? 0 : 1;
203 u32 wptr = RREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me]) >> 2; 203 u32 wptr = RREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me]) >> 2;
204 204
205 return wptr; 205 return wptr;
@@ -215,14 +215,14 @@ static uint32_t sdma_v2_4_ring_get_wptr(struct amdgpu_ring *ring)
215static void sdma_v2_4_ring_set_wptr(struct amdgpu_ring *ring) 215static void sdma_v2_4_ring_set_wptr(struct amdgpu_ring *ring)
216{ 216{
217 struct amdgpu_device *adev = ring->adev; 217 struct amdgpu_device *adev = ring->adev;
218 int me = (ring == &ring->adev->sdma[0].ring) ? 0 : 1; 218 int me = (ring == &ring->adev->sdma.instance[0].ring) ? 0 : 1;
219 219
220 WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me], ring->wptr << 2); 220 WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me], ring->wptr << 2);
221} 221}
222 222
223static void sdma_v2_4_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) 223static void sdma_v2_4_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
224{ 224{
225 struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ring); 225 struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring);
226 int i; 226 int i;
227 227
228 for (i = 0; i < count; i++) 228 for (i = 0; i < count; i++)
@@ -284,7 +284,7 @@ static void sdma_v2_4_ring_emit_hdp_flush(struct amdgpu_ring *ring)
284{ 284{
285 u32 ref_and_mask = 0; 285 u32 ref_and_mask = 0;
286 286
287 if (ring == &ring->adev->sdma[0].ring) 287 if (ring == &ring->adev->sdma.instance[0].ring)
288 ref_and_mask = REG_SET_FIELD(ref_and_mask, GPU_HDP_FLUSH_DONE, SDMA0, 1); 288 ref_and_mask = REG_SET_FIELD(ref_and_mask, GPU_HDP_FLUSH_DONE, SDMA0, 1);
289 else 289 else
290 ref_and_mask = REG_SET_FIELD(ref_and_mask, GPU_HDP_FLUSH_DONE, SDMA1, 1); 290 ref_and_mask = REG_SET_FIELD(ref_and_mask, GPU_HDP_FLUSH_DONE, SDMA1, 1);
@@ -368,8 +368,8 @@ static bool sdma_v2_4_ring_emit_semaphore(struct amdgpu_ring *ring,
368 */ 368 */
369static void sdma_v2_4_gfx_stop(struct amdgpu_device *adev) 369static void sdma_v2_4_gfx_stop(struct amdgpu_device *adev)
370{ 370{
371 struct amdgpu_ring *sdma0 = &adev->sdma[0].ring; 371 struct amdgpu_ring *sdma0 = &adev->sdma.instance[0].ring;
372 struct amdgpu_ring *sdma1 = &adev->sdma[1].ring; 372 struct amdgpu_ring *sdma1 = &adev->sdma.instance[1].ring;
373 u32 rb_cntl, ib_cntl; 373 u32 rb_cntl, ib_cntl;
374 int i; 374 int i;
375 375
@@ -377,7 +377,7 @@ static void sdma_v2_4_gfx_stop(struct amdgpu_device *adev)
377 (adev->mman.buffer_funcs_ring == sdma1)) 377 (adev->mman.buffer_funcs_ring == sdma1))
378 amdgpu_ttm_set_active_vram_size(adev, adev->mc.visible_vram_size); 378 amdgpu_ttm_set_active_vram_size(adev, adev->mc.visible_vram_size);
379 379
380 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 380 for (i = 0; i < adev->sdma.num_instances; i++) {
381 rb_cntl = RREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i]); 381 rb_cntl = RREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i]);
382 rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, 0); 382 rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, 0);
383 WREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i], rb_cntl); 383 WREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i], rb_cntl);
@@ -419,7 +419,7 @@ static void sdma_v2_4_enable(struct amdgpu_device *adev, bool enable)
419 sdma_v2_4_rlc_stop(adev); 419 sdma_v2_4_rlc_stop(adev);
420 } 420 }
421 421
422 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 422 for (i = 0; i < adev->sdma.num_instances; i++) {
423 f32_cntl = RREG32(mmSDMA0_F32_CNTL + sdma_offsets[i]); 423 f32_cntl = RREG32(mmSDMA0_F32_CNTL + sdma_offsets[i]);
424 if (enable) 424 if (enable)
425 f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_F32_CNTL, HALT, 0); 425 f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_F32_CNTL, HALT, 0);
@@ -445,8 +445,8 @@ static int sdma_v2_4_gfx_resume(struct amdgpu_device *adev)
445 u32 wb_offset; 445 u32 wb_offset;
446 int i, j, r; 446 int i, j, r;
447 447
448 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 448 for (i = 0; i < adev->sdma.num_instances; i++) {
449 ring = &adev->sdma[i].ring; 449 ring = &adev->sdma.instance[i].ring;
450 wb_offset = (ring->rptr_offs * 4); 450 wb_offset = (ring->rptr_offs * 4);
451 451
452 mutex_lock(&adev->srbm_mutex); 452 mutex_lock(&adev->srbm_mutex);
@@ -545,29 +545,23 @@ static int sdma_v2_4_load_microcode(struct amdgpu_device *adev)
545 const __le32 *fw_data; 545 const __le32 *fw_data;
546 u32 fw_size; 546 u32 fw_size;
547 int i, j; 547 int i, j;
548 bool smc_loads_fw = false; /* XXX fix me */
549
550 if (!adev->sdma[0].fw || !adev->sdma[1].fw)
551 return -EINVAL;
552 548
553 /* halt the MEs */ 549 /* halt the MEs */
554 sdma_v2_4_enable(adev, false); 550 sdma_v2_4_enable(adev, false);
555 551
556 if (smc_loads_fw) { 552 for (i = 0; i < adev->sdma.num_instances; i++) {
557 /* XXX query SMC for fw load complete */ 553 if (!adev->sdma.instance[i].fw)
558 } else { 554 return -EINVAL;
559 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 555 hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma.instance[i].fw->data;
560 hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data; 556 amdgpu_ucode_print_sdma_hdr(&hdr->header);
561 amdgpu_ucode_print_sdma_hdr(&hdr->header); 557 fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
562 fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4; 558 fw_data = (const __le32 *)
563 fw_data = (const __le32 *) 559 (adev->sdma.instance[i].fw->data +
564 (adev->sdma[i].fw->data + 560 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
565 le32_to_cpu(hdr->header.ucode_array_offset_bytes)); 561 WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], 0);
566 WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], 0); 562 for (j = 0; j < fw_size; j++)
567 for (j = 0; j < fw_size; j++) 563 WREG32(mmSDMA0_UCODE_DATA + sdma_offsets[i], le32_to_cpup(fw_data++));
568 WREG32(mmSDMA0_UCODE_DATA + sdma_offsets[i], le32_to_cpup(fw_data++)); 564 WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], adev->sdma.instance[i].fw_version);
569 WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], adev->sdma[i].fw_version);
570 }
571 } 565 }
572 566
573 return 0; 567 return 0;
@@ -894,7 +888,7 @@ static void sdma_v2_4_vm_set_pte_pde(struct amdgpu_ib *ib,
894 */ 888 */
895static void sdma_v2_4_vm_pad_ib(struct amdgpu_ib *ib) 889static void sdma_v2_4_vm_pad_ib(struct amdgpu_ib *ib)
896{ 890{
897 struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ib->ring); 891 struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ib->ring);
898 u32 pad_count; 892 u32 pad_count;
899 int i; 893 int i;
900 894
@@ -952,6 +946,8 @@ static int sdma_v2_4_early_init(void *handle)
952{ 946{
953 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 947 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
954 948
949 adev->sdma.num_instances = SDMA_MAX_INSTANCE;
950
955 sdma_v2_4_set_ring_funcs(adev); 951 sdma_v2_4_set_ring_funcs(adev);
956 sdma_v2_4_set_buffer_funcs(adev); 952 sdma_v2_4_set_buffer_funcs(adev);
957 sdma_v2_4_set_vm_pte_funcs(adev); 953 sdma_v2_4_set_vm_pte_funcs(adev);
@@ -963,21 +959,21 @@ static int sdma_v2_4_early_init(void *handle)
963static int sdma_v2_4_sw_init(void *handle) 959static int sdma_v2_4_sw_init(void *handle)
964{ 960{
965 struct amdgpu_ring *ring; 961 struct amdgpu_ring *ring;
966 int r; 962 int r, i;
967 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 963 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
968 964
969 /* SDMA trap event */ 965 /* SDMA trap event */
970 r = amdgpu_irq_add_id(adev, 224, &adev->sdma_trap_irq); 966 r = amdgpu_irq_add_id(adev, 224, &adev->sdma.trap_irq);
971 if (r) 967 if (r)
972 return r; 968 return r;
973 969
974 /* SDMA Privileged inst */ 970 /* SDMA Privileged inst */
975 r = amdgpu_irq_add_id(adev, 241, &adev->sdma_illegal_inst_irq); 971 r = amdgpu_irq_add_id(adev, 241, &adev->sdma.illegal_inst_irq);
976 if (r) 972 if (r)
977 return r; 973 return r;
978 974
979 /* SDMA Privileged inst */ 975 /* SDMA Privileged inst */
980 r = amdgpu_irq_add_id(adev, 247, &adev->sdma_illegal_inst_irq); 976 r = amdgpu_irq_add_id(adev, 247, &adev->sdma.illegal_inst_irq);
981 if (r) 977 if (r)
982 return r; 978 return r;
983 979
@@ -987,31 +983,20 @@ static int sdma_v2_4_sw_init(void *handle)
987 return r; 983 return r;
988 } 984 }
989 985
990 ring = &adev->sdma[0].ring; 986 for (i = 0; i < adev->sdma.num_instances; i++) {
991 ring->ring_obj = NULL; 987 ring = &adev->sdma.instance[i].ring;
992 ring->use_doorbell = false; 988 ring->ring_obj = NULL;
993 989 ring->use_doorbell = false;
994 ring = &adev->sdma[1].ring; 990 sprintf(ring->name, "sdma%d", i);
995 ring->ring_obj = NULL; 991 r = amdgpu_ring_init(adev, ring, 256 * 1024,
996 ring->use_doorbell = false; 992 SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), 0xf,
997 993 &adev->sdma.trap_irq,
998 ring = &adev->sdma[0].ring; 994 (i == 0) ?
999 sprintf(ring->name, "sdma0"); 995 AMDGPU_SDMA_IRQ_TRAP0 : AMDGPU_SDMA_IRQ_TRAP1,
1000 r = amdgpu_ring_init(adev, ring, 256 * 1024, 996 AMDGPU_RING_TYPE_SDMA);
1001 SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), 0xf, 997 if (r)
1002 &adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP0, 998 return r;
1003 AMDGPU_RING_TYPE_SDMA); 999 }
1004 if (r)
1005 return r;
1006
1007 ring = &adev->sdma[1].ring;
1008 sprintf(ring->name, "sdma1");
1009 r = amdgpu_ring_init(adev, ring, 256 * 1024,
1010 SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), 0xf,
1011 &adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP1,
1012 AMDGPU_RING_TYPE_SDMA);
1013 if (r)
1014 return r;
1015 1000
1016 return r; 1001 return r;
1017} 1002}
@@ -1019,9 +1004,10 @@ static int sdma_v2_4_sw_init(void *handle)
1019static int sdma_v2_4_sw_fini(void *handle) 1004static int sdma_v2_4_sw_fini(void *handle)
1020{ 1005{
1021 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1006 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1007 int i;
1022 1008
1023 amdgpu_ring_fini(&adev->sdma[0].ring); 1009 for (i = 0; i < adev->sdma.num_instances; i++)
1024 amdgpu_ring_fini(&adev->sdma[1].ring); 1010 amdgpu_ring_fini(&adev->sdma.instance[i].ring);
1025 1011
1026 return 0; 1012 return 0;
1027} 1013}
@@ -1100,7 +1086,7 @@ static void sdma_v2_4_print_status(void *handle)
1100 dev_info(adev->dev, "VI SDMA registers\n"); 1086 dev_info(adev->dev, "VI SDMA registers\n");
1101 dev_info(adev->dev, " SRBM_STATUS2=0x%08X\n", 1087 dev_info(adev->dev, " SRBM_STATUS2=0x%08X\n",
1102 RREG32(mmSRBM_STATUS2)); 1088 RREG32(mmSRBM_STATUS2));
1103 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 1089 for (i = 0; i < adev->sdma.num_instances; i++) {
1104 dev_info(adev->dev, " SDMA%d_STATUS_REG=0x%08X\n", 1090 dev_info(adev->dev, " SDMA%d_STATUS_REG=0x%08X\n",
1105 i, RREG32(mmSDMA0_STATUS_REG + sdma_offsets[i])); 1091 i, RREG32(mmSDMA0_STATUS_REG + sdma_offsets[i]));
1106 dev_info(adev->dev, " SDMA%d_F32_CNTL=0x%08X\n", 1092 dev_info(adev->dev, " SDMA%d_F32_CNTL=0x%08X\n",
@@ -1243,7 +1229,7 @@ static int sdma_v2_4_process_trap_irq(struct amdgpu_device *adev,
1243 case 0: 1229 case 0:
1244 switch (queue_id) { 1230 switch (queue_id) {
1245 case 0: 1231 case 0:
1246 amdgpu_fence_process(&adev->sdma[0].ring); 1232 amdgpu_fence_process(&adev->sdma.instance[0].ring);
1247 break; 1233 break;
1248 case 1: 1234 case 1:
1249 /* XXX compute */ 1235 /* XXX compute */
@@ -1256,7 +1242,7 @@ static int sdma_v2_4_process_trap_irq(struct amdgpu_device *adev,
1256 case 1: 1242 case 1:
1257 switch (queue_id) { 1243 switch (queue_id) {
1258 case 0: 1244 case 0:
1259 amdgpu_fence_process(&adev->sdma[1].ring); 1245 amdgpu_fence_process(&adev->sdma.instance[1].ring);
1260 break; 1246 break;
1261 case 1: 1247 case 1:
1262 /* XXX compute */ 1248 /* XXX compute */
@@ -1345,8 +1331,10 @@ static const struct amdgpu_ring_funcs sdma_v2_4_ring_funcs = {
1345 1331
1346static void sdma_v2_4_set_ring_funcs(struct amdgpu_device *adev) 1332static void sdma_v2_4_set_ring_funcs(struct amdgpu_device *adev)
1347{ 1333{
1348 adev->sdma[0].ring.funcs = &sdma_v2_4_ring_funcs; 1334 int i;
1349 adev->sdma[1].ring.funcs = &sdma_v2_4_ring_funcs; 1335
1336 for (i = 0; i < adev->sdma.num_instances; i++)
1337 adev->sdma.instance[i].ring.funcs = &sdma_v2_4_ring_funcs;
1350} 1338}
1351 1339
1352static const struct amdgpu_irq_src_funcs sdma_v2_4_trap_irq_funcs = { 1340static const struct amdgpu_irq_src_funcs sdma_v2_4_trap_irq_funcs = {
@@ -1360,9 +1348,9 @@ static const struct amdgpu_irq_src_funcs sdma_v2_4_illegal_inst_irq_funcs = {
1360 1348
1361static void sdma_v2_4_set_irq_funcs(struct amdgpu_device *adev) 1349static void sdma_v2_4_set_irq_funcs(struct amdgpu_device *adev)
1362{ 1350{
1363 adev->sdma_trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST; 1351 adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST;
1364 adev->sdma_trap_irq.funcs = &sdma_v2_4_trap_irq_funcs; 1352 adev->sdma.trap_irq.funcs = &sdma_v2_4_trap_irq_funcs;
1365 adev->sdma_illegal_inst_irq.funcs = &sdma_v2_4_illegal_inst_irq_funcs; 1353 adev->sdma.illegal_inst_irq.funcs = &sdma_v2_4_illegal_inst_irq_funcs;
1366} 1354}
1367 1355
1368/** 1356/**
@@ -1428,7 +1416,7 @@ static void sdma_v2_4_set_buffer_funcs(struct amdgpu_device *adev)
1428{ 1416{
1429 if (adev->mman.buffer_funcs == NULL) { 1417 if (adev->mman.buffer_funcs == NULL) {
1430 adev->mman.buffer_funcs = &sdma_v2_4_buffer_funcs; 1418 adev->mman.buffer_funcs = &sdma_v2_4_buffer_funcs;
1431 adev->mman.buffer_funcs_ring = &adev->sdma[0].ring; 1419 adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
1432 } 1420 }
1433} 1421}
1434 1422
@@ -1443,7 +1431,7 @@ static void sdma_v2_4_set_vm_pte_funcs(struct amdgpu_device *adev)
1443{ 1431{
1444 if (adev->vm_manager.vm_pte_funcs == NULL) { 1432 if (adev->vm_manager.vm_pte_funcs == NULL) {
1445 adev->vm_manager.vm_pte_funcs = &sdma_v2_4_vm_pte_funcs; 1433 adev->vm_manager.vm_pte_funcs = &sdma_v2_4_vm_pte_funcs;
1446 adev->vm_manager.vm_pte_funcs_ring = &adev->sdma[0].ring; 1434 adev->vm_manager.vm_pte_funcs_ring = &adev->sdma.instance[0].ring;
1447 adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true; 1435 adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true;
1448 } 1436 }
1449} 1437}
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
index 9bfe92df15f7..670555a45da9 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
@@ -184,7 +184,7 @@ static int sdma_v3_0_init_microcode(struct amdgpu_device *adev)
184{ 184{
185 const char *chip_name; 185 const char *chip_name;
186 char fw_name[30]; 186 char fw_name[30];
187 int err, i; 187 int err = 0, i;
188 struct amdgpu_firmware_info *info = NULL; 188 struct amdgpu_firmware_info *info = NULL;
189 const struct common_firmware_header *header = NULL; 189 const struct common_firmware_header *header = NULL;
190 const struct sdma_firmware_header_v1_0 *hdr; 190 const struct sdma_firmware_header_v1_0 *hdr;
@@ -204,27 +204,27 @@ static int sdma_v3_0_init_microcode(struct amdgpu_device *adev)
204 default: BUG(); 204 default: BUG();
205 } 205 }
206 206
207 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 207 for (i = 0; i < adev->sdma.num_instances; i++) {
208 if (i == 0) 208 if (i == 0)
209 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma.bin", chip_name); 209 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma.bin", chip_name);
210 else 210 else
211 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma1.bin", chip_name); 211 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma1.bin", chip_name);
212 err = request_firmware(&adev->sdma[i].fw, fw_name, adev->dev); 212 err = request_firmware(&adev->sdma.instance[i].fw, fw_name, adev->dev);
213 if (err) 213 if (err)
214 goto out; 214 goto out;
215 err = amdgpu_ucode_validate(adev->sdma[i].fw); 215 err = amdgpu_ucode_validate(adev->sdma.instance[i].fw);
216 if (err) 216 if (err)
217 goto out; 217 goto out;
218 hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data; 218 hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma.instance[i].fw->data;
219 adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version); 219 adev->sdma.instance[i].fw_version = le32_to_cpu(hdr->header.ucode_version);
220 adev->sdma[i].feature_version = le32_to_cpu(hdr->ucode_feature_version); 220 adev->sdma.instance[i].feature_version = le32_to_cpu(hdr->ucode_feature_version);
221 if (adev->sdma[i].feature_version >= 20) 221 if (adev->sdma.instance[i].feature_version >= 20)
222 adev->sdma[i].burst_nop = true; 222 adev->sdma.instance[i].burst_nop = true;
223 223
224 if (adev->firmware.smu_load) { 224 if (adev->firmware.smu_load) {
225 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_SDMA0 + i]; 225 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_SDMA0 + i];
226 info->ucode_id = AMDGPU_UCODE_ID_SDMA0 + i; 226 info->ucode_id = AMDGPU_UCODE_ID_SDMA0 + i;
227 info->fw = adev->sdma[i].fw; 227 info->fw = adev->sdma.instance[i].fw;
228 header = (const struct common_firmware_header *)info->fw->data; 228 header = (const struct common_firmware_header *)info->fw->data;
229 adev->firmware.fw_size += 229 adev->firmware.fw_size +=
230 ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE); 230 ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
@@ -235,9 +235,9 @@ out:
235 printk(KERN_ERR 235 printk(KERN_ERR
236 "sdma_v3_0: Failed to load firmware \"%s\"\n", 236 "sdma_v3_0: Failed to load firmware \"%s\"\n",
237 fw_name); 237 fw_name);
238 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 238 for (i = 0; i < adev->sdma.num_instances; i++) {
239 release_firmware(adev->sdma[i].fw); 239 release_firmware(adev->sdma.instance[i].fw);
240 adev->sdma[i].fw = NULL; 240 adev->sdma.instance[i].fw = NULL;
241 } 241 }
242 } 242 }
243 return err; 243 return err;
@@ -276,7 +276,7 @@ static uint32_t sdma_v3_0_ring_get_wptr(struct amdgpu_ring *ring)
276 /* XXX check if swapping is necessary on BE */ 276 /* XXX check if swapping is necessary on BE */
277 wptr = ring->adev->wb.wb[ring->wptr_offs] >> 2; 277 wptr = ring->adev->wb.wb[ring->wptr_offs] >> 2;
278 } else { 278 } else {
279 int me = (ring == &ring->adev->sdma[0].ring) ? 0 : 1; 279 int me = (ring == &ring->adev->sdma.instance[0].ring) ? 0 : 1;
280 280
281 wptr = RREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me]) >> 2; 281 wptr = RREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me]) >> 2;
282 } 282 }
@@ -300,7 +300,7 @@ static void sdma_v3_0_ring_set_wptr(struct amdgpu_ring *ring)
300 adev->wb.wb[ring->wptr_offs] = ring->wptr << 2; 300 adev->wb.wb[ring->wptr_offs] = ring->wptr << 2;
301 WDOORBELL32(ring->doorbell_index, ring->wptr << 2); 301 WDOORBELL32(ring->doorbell_index, ring->wptr << 2);
302 } else { 302 } else {
303 int me = (ring == &ring->adev->sdma[0].ring) ? 0 : 1; 303 int me = (ring == &ring->adev->sdma.instance[0].ring) ? 0 : 1;
304 304
305 WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me], ring->wptr << 2); 305 WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me], ring->wptr << 2);
306 } 306 }
@@ -308,7 +308,7 @@ static void sdma_v3_0_ring_set_wptr(struct amdgpu_ring *ring)
308 308
309static void sdma_v3_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) 309static void sdma_v3_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
310{ 310{
311 struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ring); 311 struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring);
312 int i; 312 int i;
313 313
314 for (i = 0; i < count; i++) 314 for (i = 0; i < count; i++)
@@ -369,7 +369,7 @@ static void sdma_v3_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
369{ 369{
370 u32 ref_and_mask = 0; 370 u32 ref_and_mask = 0;
371 371
372 if (ring == &ring->adev->sdma[0].ring) 372 if (ring == &ring->adev->sdma.instance[0].ring)
373 ref_and_mask = REG_SET_FIELD(ref_and_mask, GPU_HDP_FLUSH_DONE, SDMA0, 1); 373 ref_and_mask = REG_SET_FIELD(ref_and_mask, GPU_HDP_FLUSH_DONE, SDMA0, 1);
374 else 374 else
375 ref_and_mask = REG_SET_FIELD(ref_and_mask, GPU_HDP_FLUSH_DONE, SDMA1, 1); 375 ref_and_mask = REG_SET_FIELD(ref_and_mask, GPU_HDP_FLUSH_DONE, SDMA1, 1);
@@ -454,8 +454,8 @@ static bool sdma_v3_0_ring_emit_semaphore(struct amdgpu_ring *ring,
454 */ 454 */
455static void sdma_v3_0_gfx_stop(struct amdgpu_device *adev) 455static void sdma_v3_0_gfx_stop(struct amdgpu_device *adev)
456{ 456{
457 struct amdgpu_ring *sdma0 = &adev->sdma[0].ring; 457 struct amdgpu_ring *sdma0 = &adev->sdma.instance[0].ring;
458 struct amdgpu_ring *sdma1 = &adev->sdma[1].ring; 458 struct amdgpu_ring *sdma1 = &adev->sdma.instance[1].ring;
459 u32 rb_cntl, ib_cntl; 459 u32 rb_cntl, ib_cntl;
460 int i; 460 int i;
461 461
@@ -463,7 +463,7 @@ static void sdma_v3_0_gfx_stop(struct amdgpu_device *adev)
463 (adev->mman.buffer_funcs_ring == sdma1)) 463 (adev->mman.buffer_funcs_ring == sdma1))
464 amdgpu_ttm_set_active_vram_size(adev, adev->mc.visible_vram_size); 464 amdgpu_ttm_set_active_vram_size(adev, adev->mc.visible_vram_size);
465 465
466 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 466 for (i = 0; i < adev->sdma.num_instances; i++) {
467 rb_cntl = RREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i]); 467 rb_cntl = RREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i]);
468 rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, 0); 468 rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, 0);
469 WREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i], rb_cntl); 469 WREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i], rb_cntl);
@@ -500,7 +500,7 @@ static void sdma_v3_0_ctx_switch_enable(struct amdgpu_device *adev, bool enable)
500 u32 f32_cntl; 500 u32 f32_cntl;
501 int i; 501 int i;
502 502
503 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 503 for (i = 0; i < adev->sdma.num_instances; i++) {
504 f32_cntl = RREG32(mmSDMA0_CNTL + sdma_offsets[i]); 504 f32_cntl = RREG32(mmSDMA0_CNTL + sdma_offsets[i]);
505 if (enable) 505 if (enable)
506 f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_CNTL, 506 f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_CNTL,
@@ -530,7 +530,7 @@ static void sdma_v3_0_enable(struct amdgpu_device *adev, bool enable)
530 sdma_v3_0_rlc_stop(adev); 530 sdma_v3_0_rlc_stop(adev);
531 } 531 }
532 532
533 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 533 for (i = 0; i < adev->sdma.num_instances; i++) {
534 f32_cntl = RREG32(mmSDMA0_F32_CNTL + sdma_offsets[i]); 534 f32_cntl = RREG32(mmSDMA0_F32_CNTL + sdma_offsets[i]);
535 if (enable) 535 if (enable)
536 f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_F32_CNTL, HALT, 0); 536 f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_F32_CNTL, HALT, 0);
@@ -557,8 +557,8 @@ static int sdma_v3_0_gfx_resume(struct amdgpu_device *adev)
557 u32 doorbell; 557 u32 doorbell;
558 int i, j, r; 558 int i, j, r;
559 559
560 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 560 for (i = 0; i < adev->sdma.num_instances; i++) {
561 ring = &adev->sdma[i].ring; 561 ring = &adev->sdma.instance[i].ring;
562 wb_offset = (ring->rptr_offs * 4); 562 wb_offset = (ring->rptr_offs * 4);
563 563
564 mutex_lock(&adev->srbm_mutex); 564 mutex_lock(&adev->srbm_mutex);
@@ -669,23 +669,22 @@ static int sdma_v3_0_load_microcode(struct amdgpu_device *adev)
669 u32 fw_size; 669 u32 fw_size;
670 int i, j; 670 int i, j;
671 671
672 if (!adev->sdma[0].fw || !adev->sdma[1].fw)
673 return -EINVAL;
674
675 /* halt the MEs */ 672 /* halt the MEs */
676 sdma_v3_0_enable(adev, false); 673 sdma_v3_0_enable(adev, false);
677 674
678 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 675 for (i = 0; i < adev->sdma.num_instances; i++) {
679 hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data; 676 if (!adev->sdma.instance[i].fw)
677 return -EINVAL;
678 hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma.instance[i].fw->data;
680 amdgpu_ucode_print_sdma_hdr(&hdr->header); 679 amdgpu_ucode_print_sdma_hdr(&hdr->header);
681 fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4; 680 fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
682 fw_data = (const __le32 *) 681 fw_data = (const __le32 *)
683 (adev->sdma[i].fw->data + 682 (adev->sdma.instance[i].fw->data +
684 le32_to_cpu(hdr->header.ucode_array_offset_bytes)); 683 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
685 WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], 0); 684 WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], 0);
686 for (j = 0; j < fw_size; j++) 685 for (j = 0; j < fw_size; j++)
687 WREG32(mmSDMA0_UCODE_DATA + sdma_offsets[i], le32_to_cpup(fw_data++)); 686 WREG32(mmSDMA0_UCODE_DATA + sdma_offsets[i], le32_to_cpup(fw_data++));
688 WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], adev->sdma[i].fw_version); 687 WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], adev->sdma.instance[i].fw_version);
689 } 688 }
690 689
691 return 0; 690 return 0;
@@ -701,21 +700,21 @@ static int sdma_v3_0_load_microcode(struct amdgpu_device *adev)
701 */ 700 */
702static int sdma_v3_0_start(struct amdgpu_device *adev) 701static int sdma_v3_0_start(struct amdgpu_device *adev)
703{ 702{
704 int r; 703 int r, i;
705 704
706 if (!adev->firmware.smu_load) { 705 if (!adev->firmware.smu_load) {
707 r = sdma_v3_0_load_microcode(adev); 706 r = sdma_v3_0_load_microcode(adev);
708 if (r) 707 if (r)
709 return r; 708 return r;
710 } else { 709 } else {
711 r = adev->smu.smumgr_funcs->check_fw_load_finish(adev, 710 for (i = 0; i < adev->sdma.num_instances; i++) {
712 AMDGPU_UCODE_ID_SDMA0); 711 r = adev->smu.smumgr_funcs->check_fw_load_finish(adev,
713 if (r) 712 (i == 0) ?
714 return -EINVAL; 713 AMDGPU_UCODE_ID_SDMA0 :
715 r = adev->smu.smumgr_funcs->check_fw_load_finish(adev, 714 AMDGPU_UCODE_ID_SDMA1);
716 AMDGPU_UCODE_ID_SDMA1); 715 if (r)
717 if (r) 716 return -EINVAL;
718 return -EINVAL; 717 }
719 } 718 }
720 719
721 /* unhalt the MEs */ 720 /* unhalt the MEs */
@@ -1013,7 +1012,7 @@ static void sdma_v3_0_vm_set_pte_pde(struct amdgpu_ib *ib,
1013 */ 1012 */
1014static void sdma_v3_0_vm_pad_ib(struct amdgpu_ib *ib) 1013static void sdma_v3_0_vm_pad_ib(struct amdgpu_ib *ib)
1015{ 1014{
1016 struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ib->ring); 1015 struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ib->ring);
1017 u32 pad_count; 1016 u32 pad_count;
1018 int i; 1017 int i;
1019 1018
@@ -1071,6 +1070,12 @@ static int sdma_v3_0_early_init(void *handle)
1071{ 1070{
1072 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1071 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1073 1072
1073 switch (adev->asic_type) {
1074 default:
1075 adev->sdma.num_instances = SDMA_MAX_INSTANCE;
1076 break;
1077 }
1078
1074 sdma_v3_0_set_ring_funcs(adev); 1079 sdma_v3_0_set_ring_funcs(adev);
1075 sdma_v3_0_set_buffer_funcs(adev); 1080 sdma_v3_0_set_buffer_funcs(adev);
1076 sdma_v3_0_set_vm_pte_funcs(adev); 1081 sdma_v3_0_set_vm_pte_funcs(adev);
@@ -1082,21 +1087,21 @@ static int sdma_v3_0_early_init(void *handle)
1082static int sdma_v3_0_sw_init(void *handle) 1087static int sdma_v3_0_sw_init(void *handle)
1083{ 1088{
1084 struct amdgpu_ring *ring; 1089 struct amdgpu_ring *ring;
1085 int r; 1090 int r, i;
1086 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1091 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1087 1092
1088 /* SDMA trap event */ 1093 /* SDMA trap event */
1089 r = amdgpu_irq_add_id(adev, 224, &adev->sdma_trap_irq); 1094 r = amdgpu_irq_add_id(adev, 224, &adev->sdma.trap_irq);
1090 if (r) 1095 if (r)
1091 return r; 1096 return r;
1092 1097
1093 /* SDMA Privileged inst */ 1098 /* SDMA Privileged inst */
1094 r = amdgpu_irq_add_id(adev, 241, &adev->sdma_illegal_inst_irq); 1099 r = amdgpu_irq_add_id(adev, 241, &adev->sdma.illegal_inst_irq);
1095 if (r) 1100 if (r)
1096 return r; 1101 return r;
1097 1102
1098 /* SDMA Privileged inst */ 1103 /* SDMA Privileged inst */
1099 r = amdgpu_irq_add_id(adev, 247, &adev->sdma_illegal_inst_irq); 1104 r = amdgpu_irq_add_id(adev, 247, &adev->sdma.illegal_inst_irq);
1100 if (r) 1105 if (r)
1101 return r; 1106 return r;
1102 1107
@@ -1106,33 +1111,23 @@ static int sdma_v3_0_sw_init(void *handle)
1106 return r; 1111 return r;
1107 } 1112 }
1108 1113
1109 ring = &adev->sdma[0].ring; 1114 for (i = 0; i < adev->sdma.num_instances; i++) {
1110 ring->ring_obj = NULL; 1115 ring = &adev->sdma.instance[i].ring;
1111 ring->use_doorbell = true; 1116 ring->ring_obj = NULL;
1112 ring->doorbell_index = AMDGPU_DOORBELL_sDMA_ENGINE0; 1117 ring->use_doorbell = true;
1113 1118 ring->doorbell_index = (i == 0) ?
1114 ring = &adev->sdma[1].ring; 1119 AMDGPU_DOORBELL_sDMA_ENGINE0 : AMDGPU_DOORBELL_sDMA_ENGINE1;
1115 ring->ring_obj = NULL; 1120
1116 ring->use_doorbell = true; 1121 sprintf(ring->name, "sdma%d", i);
1117 ring->doorbell_index = AMDGPU_DOORBELL_sDMA_ENGINE1; 1122 r = amdgpu_ring_init(adev, ring, 256 * 1024,
1118 1123 SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), 0xf,
1119 ring = &adev->sdma[0].ring; 1124 &adev->sdma.trap_irq,
1120 sprintf(ring->name, "sdma0"); 1125 (i == 0) ?
1121 r = amdgpu_ring_init(adev, ring, 256 * 1024, 1126 AMDGPU_SDMA_IRQ_TRAP0 : AMDGPU_SDMA_IRQ_TRAP1,
1122 SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), 0xf, 1127 AMDGPU_RING_TYPE_SDMA);
1123 &adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP0, 1128 if (r)
1124 AMDGPU_RING_TYPE_SDMA); 1129 return r;
1125 if (r) 1130 }
1126 return r;
1127
1128 ring = &adev->sdma[1].ring;
1129 sprintf(ring->name, "sdma1");
1130 r = amdgpu_ring_init(adev, ring, 256 * 1024,
1131 SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), 0xf,
1132 &adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP1,
1133 AMDGPU_RING_TYPE_SDMA);
1134 if (r)
1135 return r;
1136 1131
1137 return r; 1132 return r;
1138} 1133}
@@ -1140,9 +1135,10 @@ static int sdma_v3_0_sw_init(void *handle)
1140static int sdma_v3_0_sw_fini(void *handle) 1135static int sdma_v3_0_sw_fini(void *handle)
1141{ 1136{
1142 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 1137 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1138 int i;
1143 1139
1144 amdgpu_ring_fini(&adev->sdma[0].ring); 1140 for (i = 0; i < adev->sdma.num_instances; i++)
1145 amdgpu_ring_fini(&adev->sdma[1].ring); 1141 amdgpu_ring_fini(&adev->sdma.instance[i].ring);
1146 1142
1147 return 0; 1143 return 0;
1148} 1144}
@@ -1222,7 +1218,7 @@ static void sdma_v3_0_print_status(void *handle)
1222 dev_info(adev->dev, "VI SDMA registers\n"); 1218 dev_info(adev->dev, "VI SDMA registers\n");
1223 dev_info(adev->dev, " SRBM_STATUS2=0x%08X\n", 1219 dev_info(adev->dev, " SRBM_STATUS2=0x%08X\n",
1224 RREG32(mmSRBM_STATUS2)); 1220 RREG32(mmSRBM_STATUS2));
1225 for (i = 0; i < SDMA_MAX_INSTANCE; i++) { 1221 for (i = 0; i < adev->sdma.num_instances; i++) {
1226 dev_info(adev->dev, " SDMA%d_STATUS_REG=0x%08X\n", 1222 dev_info(adev->dev, " SDMA%d_STATUS_REG=0x%08X\n",
1227 i, RREG32(mmSDMA0_STATUS_REG + sdma_offsets[i])); 1223 i, RREG32(mmSDMA0_STATUS_REG + sdma_offsets[i]));
1228 dev_info(adev->dev, " SDMA%d_F32_CNTL=0x%08X\n", 1224 dev_info(adev->dev, " SDMA%d_F32_CNTL=0x%08X\n",
@@ -1367,7 +1363,7 @@ static int sdma_v3_0_process_trap_irq(struct amdgpu_device *adev,
1367 case 0: 1363 case 0:
1368 switch (queue_id) { 1364 switch (queue_id) {
1369 case 0: 1365 case 0:
1370 amdgpu_fence_process(&adev->sdma[0].ring); 1366 amdgpu_fence_process(&adev->sdma.instance[0].ring);
1371 break; 1367 break;
1372 case 1: 1368 case 1:
1373 /* XXX compute */ 1369 /* XXX compute */
@@ -1380,7 +1376,7 @@ static int sdma_v3_0_process_trap_irq(struct amdgpu_device *adev,
1380 case 1: 1376 case 1:
1381 switch (queue_id) { 1377 switch (queue_id) {
1382 case 0: 1378 case 0:
1383 amdgpu_fence_process(&adev->sdma[1].ring); 1379 amdgpu_fence_process(&adev->sdma.instance[1].ring);
1384 break; 1380 break;
1385 case 1: 1381 case 1:
1386 /* XXX compute */ 1382 /* XXX compute */
@@ -1468,8 +1464,10 @@ static const struct amdgpu_ring_funcs sdma_v3_0_ring_funcs = {
1468 1464
1469static void sdma_v3_0_set_ring_funcs(struct amdgpu_device *adev) 1465static void sdma_v3_0_set_ring_funcs(struct amdgpu_device *adev)
1470{ 1466{
1471 adev->sdma[0].ring.funcs = &sdma_v3_0_ring_funcs; 1467 int i;
1472 adev->sdma[1].ring.funcs = &sdma_v3_0_ring_funcs; 1468
1469 for (i = 0; i < adev->sdma.num_instances; i++)
1470 adev->sdma.instance[i].ring.funcs = &sdma_v3_0_ring_funcs;
1473} 1471}
1474 1472
1475static const struct amdgpu_irq_src_funcs sdma_v3_0_trap_irq_funcs = { 1473static const struct amdgpu_irq_src_funcs sdma_v3_0_trap_irq_funcs = {
@@ -1483,9 +1481,9 @@ static const struct amdgpu_irq_src_funcs sdma_v3_0_illegal_inst_irq_funcs = {
1483 1481
1484static void sdma_v3_0_set_irq_funcs(struct amdgpu_device *adev) 1482static void sdma_v3_0_set_irq_funcs(struct amdgpu_device *adev)
1485{ 1483{
1486 adev->sdma_trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST; 1484 adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST;
1487 adev->sdma_trap_irq.funcs = &sdma_v3_0_trap_irq_funcs; 1485 adev->sdma.trap_irq.funcs = &sdma_v3_0_trap_irq_funcs;
1488 adev->sdma_illegal_inst_irq.funcs = &sdma_v3_0_illegal_inst_irq_funcs; 1486 adev->sdma.illegal_inst_irq.funcs = &sdma_v3_0_illegal_inst_irq_funcs;
1489} 1487}
1490 1488
1491/** 1489/**
@@ -1551,7 +1549,7 @@ static void sdma_v3_0_set_buffer_funcs(struct amdgpu_device *adev)
1551{ 1549{
1552 if (adev->mman.buffer_funcs == NULL) { 1550 if (adev->mman.buffer_funcs == NULL) {
1553 adev->mman.buffer_funcs = &sdma_v3_0_buffer_funcs; 1551 adev->mman.buffer_funcs = &sdma_v3_0_buffer_funcs;
1554 adev->mman.buffer_funcs_ring = &adev->sdma[0].ring; 1552 adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
1555 } 1553 }
1556} 1554}
1557 1555
@@ -1566,7 +1564,7 @@ static void sdma_v3_0_set_vm_pte_funcs(struct amdgpu_device *adev)
1566{ 1564{
1567 if (adev->vm_manager.vm_pte_funcs == NULL) { 1565 if (adev->vm_manager.vm_pte_funcs == NULL) {
1568 adev->vm_manager.vm_pte_funcs = &sdma_v3_0_vm_pte_funcs; 1566 adev->vm_manager.vm_pte_funcs = &sdma_v3_0_vm_pte_funcs;
1569 adev->vm_manager.vm_pte_funcs_ring = &adev->sdma[0].ring; 1567 adev->vm_manager.vm_pte_funcs_ring = &adev->sdma.instance[0].ring;
1570 adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true; 1568 adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true;
1571 } 1569 }
1572} 1570}
generated by cgit v1.2.2 (git 2.25.0) at 2025-09-02 20:24:22 -0400