2 * Copyright 2013 Advanced Micro Devices, Inc.
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the
7 * "Software"), to deal in the Software without restriction, including
8 * without limitation the rights to use, copy, modify, merge, publish,
9 * distribute, sub license, and/or sell copies of the Software, and to
10 * permit persons to whom the Software is furnished to do so, subject to
11 * the following conditions:
13 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
14 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
16 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
17 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
18 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
19 * USE OR OTHER DEALINGS IN THE SOFTWARE.
21 * The above copyright notice and this permission notice (including the
22 * next paragraph) shall be included in all copies or substantial portions
25 * Authors: Christian König <christian.koenig@amd.com>
28 #include <linux/firmware.h>
31 #include "amdgpu_vce.h"
33 #include "vce/vce_2_0_d.h"
34 #include "vce/vce_2_0_sh_mask.h"
35 #include "smu/smu_7_0_1_d.h"
36 #include "smu/smu_7_0_1_sh_mask.h"
37 #include "oss/oss_2_0_d.h"
38 #include "oss/oss_2_0_sh_mask.h"
40 #define VCE_V2_0_FW_SIZE (256 * 1024)
41 #define VCE_V2_0_STACK_SIZE (64 * 1024)
42 #define VCE_V2_0_DATA_SIZE (23552 * AMDGPU_MAX_VCE_HANDLES)
43 #define VCE_STATUS_VCPU_REPORT_FW_LOADED_MASK 0x02
45 static void vce_v2_0_set_ring_funcs(struct amdgpu_device
*adev
);
46 static void vce_v2_0_set_irq_funcs(struct amdgpu_device
*adev
);
49 * vce_v2_0_ring_get_rptr - get read pointer
51 * @ring: amdgpu_ring pointer
53 * Returns the current hardware read pointer
55 static uint64_t vce_v2_0_ring_get_rptr(struct amdgpu_ring
*ring
)
57 struct amdgpu_device
*adev
= ring
->adev
;
60 return RREG32(mmVCE_RB_RPTR
);
62 return RREG32(mmVCE_RB_RPTR2
);
66 * vce_v2_0_ring_get_wptr - get write pointer
68 * @ring: amdgpu_ring pointer
70 * Returns the current hardware write pointer
72 static uint64_t vce_v2_0_ring_get_wptr(struct amdgpu_ring
*ring
)
74 struct amdgpu_device
*adev
= ring
->adev
;
77 return RREG32(mmVCE_RB_WPTR
);
79 return RREG32(mmVCE_RB_WPTR2
);
83 * vce_v2_0_ring_set_wptr - set write pointer
85 * @ring: amdgpu_ring pointer
87 * Commits the write pointer to the hardware
89 static void vce_v2_0_ring_set_wptr(struct amdgpu_ring
*ring
)
91 struct amdgpu_device
*adev
= ring
->adev
;
94 WREG32(mmVCE_RB_WPTR
, lower_32_bits(ring
->wptr
));
96 WREG32(mmVCE_RB_WPTR2
, lower_32_bits(ring
->wptr
));
99 static int vce_v2_0_lmi_clean(struct amdgpu_device
*adev
)
103 for (i
= 0; i
< 10; ++i
) {
104 for (j
= 0; j
< 100; ++j
) {
105 uint32_t status
= RREG32(mmVCE_LMI_STATUS
);
116 static int vce_v2_0_firmware_loaded(struct amdgpu_device
*adev
)
120 for (i
= 0; i
< 10; ++i
) {
121 for (j
= 0; j
< 100; ++j
) {
122 uint32_t status
= RREG32(mmVCE_STATUS
);
124 if (status
& VCE_STATUS_VCPU_REPORT_FW_LOADED_MASK
)
129 DRM_ERROR("VCE not responding, trying to reset the ECPU!!!\n");
130 WREG32_P(mmVCE_SOFT_RESET
,
131 VCE_SOFT_RESET__ECPU_SOFT_RESET_MASK
,
132 ~VCE_SOFT_RESET__ECPU_SOFT_RESET_MASK
);
134 WREG32_P(mmVCE_SOFT_RESET
, 0,
135 ~VCE_SOFT_RESET__ECPU_SOFT_RESET_MASK
);
142 static void vce_v2_0_disable_cg(struct amdgpu_device
*adev
)
144 WREG32(mmVCE_CGTT_CLK_OVERRIDE
, 7);
147 static void vce_v2_0_init_cg(struct amdgpu_device
*adev
)
151 tmp
= RREG32(mmVCE_CLOCK_GATING_A
);
153 tmp
|= ((0 << 0) | (4 << 4));
155 WREG32(mmVCE_CLOCK_GATING_A
, tmp
);
157 tmp
= RREG32(mmVCE_UENC_CLOCK_GATING
);
159 tmp
|= ((0 << 0) | (4 << 4));
160 WREG32(mmVCE_UENC_CLOCK_GATING
, tmp
);
162 tmp
= RREG32(mmVCE_CLOCK_GATING_B
);
165 WREG32(mmVCE_CLOCK_GATING_B
, tmp
);
168 static void vce_v2_0_mc_resume(struct amdgpu_device
*adev
)
170 uint32_t size
, offset
;
172 WREG32_P(mmVCE_CLOCK_GATING_A
, 0, ~(1 << 16));
173 WREG32_P(mmVCE_UENC_CLOCK_GATING
, 0x1FF000, ~0xFF9FF000);
174 WREG32_P(mmVCE_UENC_REG_CLOCK_GATING
, 0x3F, ~0x3F);
175 WREG32(mmVCE_CLOCK_GATING_B
, 0xf7);
177 WREG32(mmVCE_LMI_CTRL
, 0x00398000);
178 WREG32_P(mmVCE_LMI_CACHE_CTRL
, 0x0, ~0x1);
179 WREG32(mmVCE_LMI_SWAP_CNTL
, 0);
180 WREG32(mmVCE_LMI_SWAP_CNTL1
, 0);
181 WREG32(mmVCE_LMI_VM_CTRL
, 0);
183 WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR
, (adev
->vce
.gpu_addr
>> 8));
185 offset
= AMDGPU_VCE_FIRMWARE_OFFSET
;
186 size
= VCE_V2_0_FW_SIZE
;
187 WREG32(mmVCE_VCPU_CACHE_OFFSET0
, offset
& 0x7fffffff);
188 WREG32(mmVCE_VCPU_CACHE_SIZE0
, size
);
191 size
= VCE_V2_0_STACK_SIZE
;
192 WREG32(mmVCE_VCPU_CACHE_OFFSET1
, offset
& 0x7fffffff);
193 WREG32(mmVCE_VCPU_CACHE_SIZE1
, size
);
196 size
= VCE_V2_0_DATA_SIZE
;
197 WREG32(mmVCE_VCPU_CACHE_OFFSET2
, offset
& 0x7fffffff);
198 WREG32(mmVCE_VCPU_CACHE_SIZE2
, size
);
200 WREG32_P(mmVCE_LMI_CTRL2
, 0x0, ~0x100);
201 WREG32_FIELD(VCE_SYS_INT_EN
, VCE_SYS_INT_TRAP_INTERRUPT_EN
, 1);
204 static bool vce_v2_0_is_idle(void *handle
)
206 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
208 return !(RREG32(mmSRBM_STATUS2
) & SRBM_STATUS2__VCE_BUSY_MASK
);
211 static int vce_v2_0_wait_for_idle(void *handle
)
213 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
216 for (i
= 0; i
< adev
->usec_timeout
; i
++) {
217 if (vce_v2_0_is_idle(handle
))
224 * vce_v2_0_start - start VCE block
226 * @adev: amdgpu_device pointer
228 * Setup and start the VCE block
230 static int vce_v2_0_start(struct amdgpu_device
*adev
)
232 struct amdgpu_ring
*ring
;
236 WREG32_P(mmVCE_STATUS
, 1, ~1);
238 vce_v2_0_init_cg(adev
);
239 vce_v2_0_disable_cg(adev
);
241 vce_v2_0_mc_resume(adev
);
243 ring
= &adev
->vce
.ring
[0];
244 WREG32(mmVCE_RB_RPTR
, lower_32_bits(ring
->wptr
));
245 WREG32(mmVCE_RB_WPTR
, lower_32_bits(ring
->wptr
));
246 WREG32(mmVCE_RB_BASE_LO
, ring
->gpu_addr
);
247 WREG32(mmVCE_RB_BASE_HI
, upper_32_bits(ring
->gpu_addr
));
248 WREG32(mmVCE_RB_SIZE
, ring
->ring_size
/ 4);
250 ring
= &adev
->vce
.ring
[1];
251 WREG32(mmVCE_RB_RPTR2
, lower_32_bits(ring
->wptr
));
252 WREG32(mmVCE_RB_WPTR2
, lower_32_bits(ring
->wptr
));
253 WREG32(mmVCE_RB_BASE_LO2
, ring
->gpu_addr
);
254 WREG32(mmVCE_RB_BASE_HI2
, upper_32_bits(ring
->gpu_addr
));
255 WREG32(mmVCE_RB_SIZE2
, ring
->ring_size
/ 4);
257 WREG32_FIELD(VCE_VCPU_CNTL
, CLK_EN
, 1);
258 WREG32_FIELD(VCE_SOFT_RESET
, ECPU_SOFT_RESET
, 1);
260 WREG32_FIELD(VCE_SOFT_RESET
, ECPU_SOFT_RESET
, 0);
262 r
= vce_v2_0_firmware_loaded(adev
);
264 /* clear BUSY flag */
265 WREG32_P(mmVCE_STATUS
, 0, ~1);
268 DRM_ERROR("VCE not responding, giving up!!!\n");
275 static int vce_v2_0_stop(struct amdgpu_device
*adev
)
280 if (vce_v2_0_lmi_clean(adev
)) {
281 DRM_INFO("vce is not idle \n");
285 if (vce_v2_0_wait_for_idle(adev
)) {
286 DRM_INFO("VCE is busy, Can't set clock gateing");
290 /* Stall UMC and register bus before resetting VCPU */
291 WREG32_P(mmVCE_LMI_CTRL2
, 1 << 8, ~(1 << 8));
293 for (i
= 0; i
< 100; ++i
) {
294 status
= RREG32(mmVCE_LMI_STATUS
);
300 WREG32_P(mmVCE_VCPU_CNTL
, 0, ~0x80001);
302 /* put LMI, VCPU, RBC etc... into reset */
303 WREG32_P(mmVCE_SOFT_RESET
, 1, ~0x1);
305 WREG32(mmVCE_STATUS
, 0);
310 static void vce_v2_0_set_sw_cg(struct amdgpu_device
*adev
, bool gated
)
315 tmp
= RREG32(mmVCE_CLOCK_GATING_B
);
317 WREG32(mmVCE_CLOCK_GATING_B
, tmp
);
319 tmp
= RREG32(mmVCE_UENC_CLOCK_GATING
);
321 WREG32(mmVCE_UENC_CLOCK_GATING
, tmp
);
323 tmp
= RREG32(mmVCE_UENC_REG_CLOCK_GATING
);
325 WREG32(mmVCE_UENC_REG_CLOCK_GATING
, tmp
);
327 WREG32(mmVCE_CGTT_CLK_OVERRIDE
, 0);
329 tmp
= RREG32(mmVCE_CLOCK_GATING_B
);
332 WREG32(mmVCE_CLOCK_GATING_B
, tmp
);
334 tmp
= RREG32(mmVCE_UENC_CLOCK_GATING
);
337 WREG32(mmVCE_UENC_CLOCK_GATING
, tmp
);
339 tmp
= RREG32(mmVCE_UENC_REG_CLOCK_GATING
);
341 WREG32(mmVCE_UENC_REG_CLOCK_GATING
, tmp
);
345 static void vce_v2_0_set_dyn_cg(struct amdgpu_device
*adev
, bool gated
)
349 /* LMI_MC/LMI_UMC always set in dynamic,
350 * set {CGC_*_GATE_MODE, CGC_*_SW_GATE} = {0, 0}
352 tmp
= RREG32(mmVCE_CLOCK_GATING_B
);
355 /* Exception for ECPU, IH, SEM, SYS blocks needs to be turned on/off by SW */
358 WREG32(mmVCE_CLOCK_GATING_B
, tmp
);
362 WREG32(mmVCE_CLOCK_GATING_B
, tmp
);
365 orig
= tmp
= RREG32(mmVCE_UENC_CLOCK_GATING
);
369 WREG32(mmVCE_UENC_CLOCK_GATING
, tmp
);
371 orig
= tmp
= RREG32(mmVCE_UENC_REG_CLOCK_GATING
);
374 WREG32(mmVCE_UENC_REG_CLOCK_GATING
, tmp
);
376 /* set VCE_UENC_REG_CLOCK_GATING always in dynamic mode */
377 WREG32(mmVCE_UENC_REG_CLOCK_GATING
, 0x00);
380 WREG32(mmVCE_CGTT_CLK_OVERRIDE
, 0);
383 static void vce_v2_0_enable_mgcg(struct amdgpu_device
*adev
, bool enable
,
386 if (enable
&& (adev
->cg_flags
& AMD_CG_SUPPORT_VCE_MGCG
)) {
388 vce_v2_0_set_sw_cg(adev
, true);
390 vce_v2_0_set_dyn_cg(adev
, true);
392 vce_v2_0_disable_cg(adev
);
395 vce_v2_0_set_sw_cg(adev
, false);
397 vce_v2_0_set_dyn_cg(adev
, false);
401 static int vce_v2_0_early_init(void *handle
)
403 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
405 adev
->vce
.num_rings
= 2;
407 vce_v2_0_set_ring_funcs(adev
);
408 vce_v2_0_set_irq_funcs(adev
);
413 static int vce_v2_0_sw_init(void *handle
)
415 struct amdgpu_ring
*ring
;
417 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
420 r
= amdgpu_irq_add_id(adev
, AMDGPU_IRQ_CLIENTID_LEGACY
, 167, &adev
->vce
.irq
);
424 r
= amdgpu_vce_sw_init(adev
, VCE_V2_0_FW_SIZE
+
425 VCE_V2_0_STACK_SIZE
+ VCE_V2_0_DATA_SIZE
);
429 r
= amdgpu_vce_resume(adev
);
433 for (i
= 0; i
< adev
->vce
.num_rings
; i
++) {
434 ring
= &adev
->vce
.ring
[i
];
435 sprintf(ring
->name
, "vce%d", i
);
436 r
= amdgpu_ring_init(adev
, ring
, 512,
442 r
= amdgpu_vce_entity_init(adev
);
447 static int vce_v2_0_sw_fini(void *handle
)
450 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
452 r
= amdgpu_vce_suspend(adev
);
456 return amdgpu_vce_sw_fini(adev
);
459 static int vce_v2_0_hw_init(void *handle
)
462 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
464 amdgpu_asic_set_vce_clocks(adev
, 10000, 10000);
465 vce_v2_0_enable_mgcg(adev
, true, false);
467 for (i
= 0; i
< adev
->vce
.num_rings
; i
++) {
468 r
= amdgpu_ring_test_helper(&adev
->vce
.ring
[i
]);
473 DRM_INFO("VCE initialized successfully.\n");
478 static int vce_v2_0_hw_fini(void *handle
)
483 static int vce_v2_0_suspend(void *handle
)
486 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
488 r
= vce_v2_0_hw_fini(adev
);
492 return amdgpu_vce_suspend(adev
);
495 static int vce_v2_0_resume(void *handle
)
498 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
500 r
= amdgpu_vce_resume(adev
);
504 return vce_v2_0_hw_init(adev
);
507 static int vce_v2_0_soft_reset(void *handle
)
509 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
511 WREG32_FIELD(SRBM_SOFT_RESET
, SOFT_RESET_VCE
, 1);
514 return vce_v2_0_start(adev
);
517 static int vce_v2_0_set_interrupt_state(struct amdgpu_device
*adev
,
518 struct amdgpu_irq_src
*source
,
520 enum amdgpu_interrupt_state state
)
524 if (state
== AMDGPU_IRQ_STATE_ENABLE
)
525 val
|= VCE_SYS_INT_EN__VCE_SYS_INT_TRAP_INTERRUPT_EN_MASK
;
527 WREG32_P(mmVCE_SYS_INT_EN
, val
, ~VCE_SYS_INT_EN__VCE_SYS_INT_TRAP_INTERRUPT_EN_MASK
);
531 static int vce_v2_0_process_interrupt(struct amdgpu_device
*adev
,
532 struct amdgpu_irq_src
*source
,
533 struct amdgpu_iv_entry
*entry
)
535 DRM_DEBUG("IH: VCE\n");
536 switch (entry
->src_data
[0]) {
539 amdgpu_fence_process(&adev
->vce
.ring
[entry
->src_data
[0]]);
542 DRM_ERROR("Unhandled interrupt: %d %d\n",
543 entry
->src_id
, entry
->src_data
[0]);
550 static int vce_v2_0_set_clockgating_state(void *handle
,
551 enum amd_clockgating_state state
)
556 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
558 if (state
== AMD_CG_STATE_GATE
) {
563 vce_v2_0_enable_mgcg(adev
, gate
, sw_cg
);
568 static int vce_v2_0_set_powergating_state(void *handle
,
569 enum amd_powergating_state state
)
571 /* This doesn't actually powergate the VCE block.
572 * That's done in the dpm code via the SMC. This
573 * just re-inits the block as necessary. The actual
574 * gating still happens in the dpm code. We should
575 * revisit this when there is a cleaner line between
576 * the smc and the hw blocks
578 struct amdgpu_device
*adev
= (struct amdgpu_device
*)handle
;
580 if (state
== AMD_PG_STATE_GATE
)
581 return vce_v2_0_stop(adev
);
583 return vce_v2_0_start(adev
);
586 static const struct amd_ip_funcs vce_v2_0_ip_funcs
= {
588 .early_init
= vce_v2_0_early_init
,
590 .sw_init
= vce_v2_0_sw_init
,
591 .sw_fini
= vce_v2_0_sw_fini
,
592 .hw_init
= vce_v2_0_hw_init
,
593 .hw_fini
= vce_v2_0_hw_fini
,
594 .suspend
= vce_v2_0_suspend
,
595 .resume
= vce_v2_0_resume
,
596 .is_idle
= vce_v2_0_is_idle
,
597 .wait_for_idle
= vce_v2_0_wait_for_idle
,
598 .soft_reset
= vce_v2_0_soft_reset
,
599 .set_clockgating_state
= vce_v2_0_set_clockgating_state
,
600 .set_powergating_state
= vce_v2_0_set_powergating_state
,
603 static const struct amdgpu_ring_funcs vce_v2_0_ring_funcs
= {
604 .type
= AMDGPU_RING_TYPE_VCE
,
606 .nop
= VCE_CMD_NO_OP
,
607 .support_64bit_ptrs
= false,
608 .get_rptr
= vce_v2_0_ring_get_rptr
,
609 .get_wptr
= vce_v2_0_ring_get_wptr
,
610 .set_wptr
= vce_v2_0_ring_set_wptr
,
611 .parse_cs
= amdgpu_vce_ring_parse_cs
,
612 .emit_frame_size
= 6, /* amdgpu_vce_ring_emit_fence x1 no user fence */
613 .emit_ib_size
= 4, /* amdgpu_vce_ring_emit_ib */
614 .emit_ib
= amdgpu_vce_ring_emit_ib
,
615 .emit_fence
= amdgpu_vce_ring_emit_fence
,
616 .test_ring
= amdgpu_vce_ring_test_ring
,
617 .test_ib
= amdgpu_vce_ring_test_ib
,
618 .insert_nop
= amdgpu_ring_insert_nop
,
619 .pad_ib
= amdgpu_ring_generic_pad_ib
,
620 .begin_use
= amdgpu_vce_ring_begin_use
,
621 .end_use
= amdgpu_vce_ring_end_use
,
624 static void vce_v2_0_set_ring_funcs(struct amdgpu_device
*adev
)
628 for (i
= 0; i
< adev
->vce
.num_rings
; i
++) {
629 adev
->vce
.ring
[i
].funcs
= &vce_v2_0_ring_funcs
;
630 adev
->vce
.ring
[i
].me
= i
;
634 static const struct amdgpu_irq_src_funcs vce_v2_0_irq_funcs
= {
635 .set
= vce_v2_0_set_interrupt_state
,
636 .process
= vce_v2_0_process_interrupt
,
639 static void vce_v2_0_set_irq_funcs(struct amdgpu_device
*adev
)
641 adev
->vce
.irq
.num_types
= 1;
642 adev
->vce
.irq
.funcs
= &vce_v2_0_irq_funcs
;
645 const struct amdgpu_ip_block_version vce_v2_0_ip_block
=
647 .type
= AMD_IP_BLOCK_TYPE_VCE
,
651 .funcs
= &vce_v2_0_ip_funcs
,