[PATCH 26/29] drm/amdgpu/gfx10: enable kiq to map mes ring
Alex Deucher
alexander.deucher at amd.com
Fri Apr 29 18:02:23 UTC 2022
From: Jack Xiao <Jack.Xiao at amd.com>
Enable KIQ to map MES ring:
1). add MES queue mapping support in MAP_QUEUES packet.
2). use correct MQD settings for MES queue.
Signed-off-by: Jack Xiao <Jack.Xiao at amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang at amd.com>
Signed-off-by: Alex Deucher <alexander.deucher at amd.com>
---
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 16 +++-
drivers/gpu/drm/amd/amdgpu/mes_v10_1.c | 106 ++++++++++---------------
2 files changed, 55 insertions(+), 67 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
index 3c4f2a94ad9f..fc289ee54a47 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
@@ -3525,7 +3525,21 @@ static void gfx10_kiq_map_queues(struct amdgpu_ring *kiq_ring,
{
uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
uint64_t wptr_addr = ring->wptr_gpu_addr;
- uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
+ uint32_t eng_sel = 0;
+
+ switch (ring->funcs->type) {
+ case AMDGPU_RING_TYPE_COMPUTE:
+ eng_sel = 0;
+ break;
+ case AMDGPU_RING_TYPE_GFX:
+ eng_sel = 4;
+ break;
+ case AMDGPU_RING_TYPE_MES:
+ eng_sel = 5;
+ break;
+ default:
+ WARN_ON(1);
+ }
amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v10_1.c b/drivers/gpu/drm/amd/amdgpu/mes_v10_1.c
index 030a92b3a0da..18a129f36215 100644
--- a/drivers/gpu/drm/amd/amdgpu/mes_v10_1.c
+++ b/drivers/gpu/drm/amd/amdgpu/mes_v10_1.c
@@ -28,6 +28,7 @@
#include "nv.h"
#include "gc/gc_10_1_0_offset.h"
#include "gc/gc_10_1_0_sh_mask.h"
+#include "gc/gc_10_1_0_default.h"
#include "v10_structs.h"
#include "mes_api_def.h"
@@ -529,7 +530,7 @@ static void mes_v10_1_enable(struct amdgpu_device *adev, bool enable)
data = REG_SET_FIELD(data, CP_MES_CNTL, MES_PIPE1_ACTIVE,
adev->enable_mes_kiq ? 1 : 0);
WREG32_SOC15(GC, 0, mmCP_MES_CNTL, data);
- udelay(50);
+ udelay(100);
} else {
data = RREG32_SOC15(GC, 0, mmCP_MES_CNTL);
data = REG_SET_FIELD(data, CP_MES_CNTL, MES_PIPE0_ACTIVE, 0);
@@ -665,7 +666,6 @@ static int mes_v10_1_allocate_eop_buf(struct amdgpu_device *adev,
static int mes_v10_1_mqd_init(struct amdgpu_ring *ring)
{
- struct amdgpu_device *adev = ring->adev;
struct v10_compute_mqd *mqd = ring->mqd_ptr;
uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
uint32_t tmp;
@@ -679,38 +679,18 @@ static int mes_v10_1_mqd_init(struct amdgpu_ring *ring)
mqd->compute_misc_reserved = 0x00000003;
eop_base_addr = ring->eop_gpu_addr >> 8;
- mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
- mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
- tmp = RREG32_SOC15(GC, 0, mmCP_HQD_EOP_CONTROL);
+ tmp = mmCP_HQD_EOP_CONTROL_DEFAULT;
tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
(order_base_2(MES_EOP_SIZE / 4) - 1));
+ mqd->cp_hqd_eop_base_addr_lo = lower_32_bits(eop_base_addr);
+ mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
mqd->cp_hqd_eop_control = tmp;
- /* enable doorbell? */
- tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL);
-
- if (ring->use_doorbell) {
- tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
- DOORBELL_OFFSET, ring->doorbell_index);
- tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
- DOORBELL_EN, 1);
- tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
- DOORBELL_SOURCE, 0);
- tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
- DOORBELL_HIT, 0);
- }
- else
- tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
- DOORBELL_EN, 0);
-
- mqd->cp_hqd_pq_doorbell_control = tmp;
-
/* disable the queue if it's active */
ring->wptr = 0;
- mqd->cp_hqd_dequeue_request = 0;
mqd->cp_hqd_pq_rptr = 0;
mqd->cp_hqd_pq_wptr_lo = 0;
mqd->cp_hqd_pq_wptr_hi = 0;
@@ -720,17 +700,28 @@ static int mes_v10_1_mqd_init(struct amdgpu_ring *ring)
mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr);
/* set MQD vmid to 0 */
- tmp = RREG32_SOC15(GC, 0, mmCP_MQD_CONTROL);
+ tmp = mmCP_MQD_CONTROL_DEFAULT;
tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
mqd->cp_mqd_control = tmp;
/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
hqd_gpu_addr = ring->gpu_addr >> 8;
- mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
+ mqd->cp_hqd_pq_base_lo = lower_32_bits(hqd_gpu_addr);
mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
+ /* set the wb address whether it's enabled or not */
+ wb_gpu_addr = ring->rptr_gpu_addr;
+ mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
+ mqd->cp_hqd_pq_rptr_report_addr_hi =
+ upper_32_bits(wb_gpu_addr) & 0xffff;
+
+ /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
+ wb_gpu_addr = ring->wptr_gpu_addr;
+ mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffff8;
+ mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
+
/* set up the HQD, this is similar to CP_RB0_CNTL */
- tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_CONTROL);
+ tmp = mmCP_HQD_PQ_CONTROL_DEFAULT;
tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
(order_base_2(ring->ring_size / 4) - 1));
tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
@@ -738,30 +729,18 @@ static int mes_v10_1_mqd_init(struct amdgpu_ring *ring)
#ifdef __BIG_ENDIAN
tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1);
#endif
- tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
+ tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 1);
tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, TUNNEL_DISPATCH, 0);
tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
+ tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, NO_UPDATE_RPTR, 1);
mqd->cp_hqd_pq_control = tmp;
- /* set the wb address whether it's enabled or not */
- wb_gpu_addr = ring->rptr_gpu_addr;
- mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
- mqd->cp_hqd_pq_rptr_report_addr_hi =
- upper_32_bits(wb_gpu_addr) & 0xffff;
-
- /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
- wb_gpu_addr = ring->wptr_gpu_addr;
- mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffff8;
- mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
-
+ /* enable doorbell? */
tmp = 0;
- /* enable the doorbell if requested */
if (ring->use_doorbell) {
- tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL);
tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
- DOORBELL_OFFSET, ring->doorbell_index);
-
+ DOORBELL_OFFSET, ring->doorbell_index);
tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
DOORBELL_EN, 1);
tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
@@ -769,30 +748,28 @@ static int mes_v10_1_mqd_init(struct amdgpu_ring *ring)
tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
DOORBELL_HIT, 0);
}
-
+ else
+ tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
+ DOORBELL_EN, 0);
mqd->cp_hqd_pq_doorbell_control = tmp;
- /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
- ring->wptr = 0;
- mqd->cp_hqd_pq_rptr = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR);
-
- /* set the vmid for the queue */
mqd->cp_hqd_vmid = 0;
-
- tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PERSISTENT_STATE);
- tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53);
- mqd->cp_hqd_persistent_state = tmp;
-
- /* set MIN_IB_AVAIL_SIZE */
- tmp = RREG32_SOC15(GC, 0, mmCP_HQD_IB_CONTROL);
- tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3);
- mqd->cp_hqd_ib_control = tmp;
-
/* activate the queue */
mqd->cp_hqd_active = 1;
+ mqd->cp_hqd_persistent_state = mmCP_HQD_PERSISTENT_STATE_DEFAULT;
+ mqd->cp_hqd_ib_control = mmCP_HQD_IB_CONTROL_DEFAULT;
+ mqd->cp_hqd_iq_timer = mmCP_HQD_IQ_TIMER_DEFAULT;
+ mqd->cp_hqd_quantum = mmCP_HQD_QUANTUM_DEFAULT;
+
+ tmp = mmCP_HQD_GFX_CONTROL_DEFAULT;
+ tmp = REG_SET_FIELD(tmp, CP_HQD_GFX_CONTROL, DB_UPDATED_MSG_EN, 1);
+ /* offset: 184 - this is used for CP_HQD_GFX_CONTROL */
+ mqd->cp_hqd_suspend_cntl_stack_offset = tmp;
+
return 0;
}
+#if 0
static void mes_v10_1_queue_init_register(struct amdgpu_ring *ring)
{
struct v10_compute_mqd *mqd = ring->mqd_ptr;
@@ -854,8 +831,8 @@ static void mes_v10_1_queue_init_register(struct amdgpu_ring *ring)
nv_grbm_select(adev, 0, 0, 0, 0);
mutex_unlock(&adev->srbm_mutex);
}
+#endif
-#if 0
static int mes_v10_1_kiq_enable_queue(struct amdgpu_device *adev)
{
struct amdgpu_kiq *kiq = &adev->gfx.kiq;
@@ -878,9 +855,9 @@ static int mes_v10_1_kiq_enable_queue(struct amdgpu_device *adev)
DRM_ERROR("kfq enable failed\n");
kiq_ring->sched.ready = false;
}
+
return r;
}
-#endif
static int mes_v10_1_queue_init(struct amdgpu_device *adev)
{
@@ -890,13 +867,9 @@ static int mes_v10_1_queue_init(struct amdgpu_device *adev)
if (r)
return r;
-#if 0
r = mes_v10_1_kiq_enable_queue(adev);
if (r)
return r;
-#else
- mes_v10_1_queue_init_register(&adev->mes.ring);
-#endif
return 0;
}
@@ -972,6 +945,7 @@ static int mes_v10_1_mqd_sw_init(struct amdgpu_device *adev,
dev_warn(adev->dev, "failed to create ring mqd bo (%d)", r);
return r;
}
+ memset(ring->mqd_ptr, 0, mqd_size);
/* prepare MQD backup */
adev->mes.mqd_backup[pipe] = kmalloc(mqd_size, GFP_KERNEL);
--
2.35.1
More information about the amd-gfx
mailing list