drm/amdgpu: fix mes packet params issue when flush hdp.

v4:
use func "amdgpu_gfx_get_hdp_flush_mask" to get ref_and_mask for
gfx9 through gfx12.

v3:
Unify the get_ref_and_mask function in amdgpu_gfx_funcs,
to support both GFX11 and earlier generations

v2:
place "get_ref_and_mask" in amdgpu_gfx_funcs instead of amdgpu_ring,
since this function only assigns the cp entry.

v1:
both gfx ring and mes ring use cp0 to flush hdp, cause conflict.

use function get_ref_and_mask to assign the cp entry.
reassign mes to use cp8 instead.

Signed-off-by: chong li <chongli2@amd.com>
Acked-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This commit is contained in:
chong li
2025-11-28 10:51:51 +08:00
committed by Alex Deucher
parent e26e4225ae
commit f9f3240018
10 changed files with 160 additions and 114 deletions

View File

@@ -1197,6 +1197,40 @@ failed_kiq_write:
dev_err(adev->dev, "failed to write reg:%x\n", reg);
}
void amdgpu_gfx_get_hdp_flush_mask(struct amdgpu_ring *ring,
uint32_t *hdp_flush_mask, uint32_t *reg_mem_engine)
{
if (!ring || !hdp_flush_mask || !reg_mem_engine) {
DRM_INFO("%s:invalid params\n", __func__);
return;
}
const struct nbio_hdp_flush_reg *nbio_hf_reg = ring->adev->nbio.hdp_flush_reg;
switch (ring->funcs->type) {
case AMDGPU_RING_TYPE_GFX:
*hdp_flush_mask = nbio_hf_reg->ref_and_mask_cp0 << ring->pipe;
*reg_mem_engine = 1; /* pfp */
break;
case AMDGPU_RING_TYPE_COMPUTE:
*hdp_flush_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe;
*reg_mem_engine = 0;
break;
case AMDGPU_RING_TYPE_MES:
*hdp_flush_mask = nbio_hf_reg->ref_and_mask_cp8;
*reg_mem_engine = 0;
break;
case AMDGPU_RING_TYPE_KIQ:
*hdp_flush_mask = nbio_hf_reg->ref_and_mask_cp9;
*reg_mem_engine = 0;
break;
default:
DRM_ERROR("%s:unsupported ring type %d\n", __func__, ring->funcs->type);
return;
}
}
int amdgpu_kiq_hdp_flush(struct amdgpu_device *adev)
{
signed long r, cnt = 0;

View File

@@ -358,6 +358,8 @@ struct amdgpu_gfx_funcs {
int num_xccs_per_xcp);
int (*ih_node_to_logical_xcc)(struct amdgpu_device *adev, int ih_node);
int (*get_xccs_per_xcp)(struct amdgpu_device *adev);
void (*get_hdp_flush_mask)(struct amdgpu_ring *ring,
uint32_t *ref_and_mask, uint32_t *reg_mem_engine);
};
struct sq_work {
@@ -617,6 +619,8 @@ int amdgpu_gfx_cp_ecc_error_irq(struct amdgpu_device *adev,
struct amdgpu_iv_entry *entry);
uint32_t amdgpu_kiq_rreg(struct amdgpu_device *adev, uint32_t reg, uint32_t xcc_id);
void amdgpu_kiq_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v, uint32_t xcc_id);
void amdgpu_gfx_get_hdp_flush_mask(struct amdgpu_ring *ring,
uint32_t *ref_and_mask, uint32_t *reg_mem_engine);
int amdgpu_kiq_hdp_flush(struct amdgpu_device *adev);
int amdgpu_gfx_get_num_kcq(struct amdgpu_device *adev);
void amdgpu_gfx_cp_init_microcode(struct amdgpu_device *adev, uint32_t ucode_id);

View File

@@ -557,11 +557,20 @@ error:
int amdgpu_mes_hdp_flush(struct amdgpu_device *adev)
{
uint32_t hdp_flush_req_offset, hdp_flush_done_offset, ref_and_mask;
uint32_t hdp_flush_req_offset, hdp_flush_done_offset;
struct amdgpu_ring *mes_ring;
uint32_t ref_and_mask = 0, reg_mem_engine = 0;
if (!adev->gfx.funcs->get_hdp_flush_mask) {
dev_err(adev->dev, "mes hdp flush is not supported.\n");
return -EINVAL;
}
mes_ring = &adev->mes.ring[0];
hdp_flush_req_offset = adev->nbio.funcs->get_hdp_flush_req_offset(adev);
hdp_flush_done_offset = adev->nbio.funcs->get_hdp_flush_done_offset(adev);
ref_and_mask = adev->nbio.hdp_flush_reg->ref_and_mask_cp0;
adev->gfx.funcs->get_hdp_flush_mask(mes_ring, &ref_and_mask, &reg_mem_engine);
return amdgpu_mes_reg_write_reg_wait(adev, hdp_flush_req_offset, hdp_flush_done_offset,
ref_and_mask, ref_and_mask, 0);

View File

@@ -4575,6 +4575,7 @@ static const struct amdgpu_gfx_funcs gfx_v10_0_gfx_funcs = {
.select_me_pipe_q = &gfx_v10_0_select_me_pipe_q,
.init_spm_golden = &gfx_v10_0_init_spm_golden_registers,
.update_perfmon_mgcg = &gfx_v10_0_update_perfmon_mgcg,
.get_hdp_flush_mask = &amdgpu_gfx_get_hdp_flush_mask,
};
static void gfx_v10_0_gpu_early_init(struct amdgpu_device *adev)
@@ -8614,25 +8615,13 @@ static void gfx_v10_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
u32 ref_and_mask, reg_mem_engine;
const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg;
if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
switch (ring->me) {
case 1:
ref_and_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe;
break;
case 2:
ref_and_mask = nbio_hf_reg->ref_and_mask_cp6 << ring->pipe;
break;
default:
return;
}
reg_mem_engine = 0;
} else {
ref_and_mask = nbio_hf_reg->ref_and_mask_cp0 << ring->pipe;
reg_mem_engine = 1; /* pfp */
if (!adev->gfx.funcs->get_hdp_flush_mask) {
dev_err(adev->dev, "%s: gfx hdp flush is not supported.\n", __func__);
return;
}
adev->gfx.funcs->get_hdp_flush_mask(ring, &ref_and_mask, &reg_mem_engine);
gfx_v10_0_wait_reg_mem(ring, reg_mem_engine, 0, 1,
adev->nbio.funcs->get_hdp_flush_req_offset(adev),
adev->nbio.funcs->get_hdp_flush_done_offset(adev),

View File

@@ -1085,6 +1085,7 @@ static const struct amdgpu_gfx_funcs gfx_v11_0_gfx_funcs = {
.select_me_pipe_q = &gfx_v11_0_select_me_pipe_q,
.update_perfmon_mgcg = &gfx_v11_0_update_perf_clk,
.get_gfx_shadow_info = &gfx_v11_0_get_gfx_shadow_info,
.get_hdp_flush_mask = &amdgpu_gfx_get_hdp_flush_mask,
};
static int gfx_v11_0_gpu_early_init(struct amdgpu_device *adev)
@@ -5837,25 +5838,13 @@ static void gfx_v11_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
u32 ref_and_mask, reg_mem_engine;
const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg;
if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
switch (ring->me) {
case 1:
ref_and_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe;
break;
case 2:
ref_and_mask = nbio_hf_reg->ref_and_mask_cp6 << ring->pipe;
break;
default:
return;
}
reg_mem_engine = 0;
} else {
ref_and_mask = nbio_hf_reg->ref_and_mask_cp0 << ring->pipe;
reg_mem_engine = 1; /* pfp */
if (!adev->gfx.funcs->get_hdp_flush_mask) {
dev_err(adev->dev, "%s: gfx hdp flush is not supported.\n", __func__);
return;
}
adev->gfx.funcs->get_hdp_flush_mask(ring, &ref_and_mask, &reg_mem_engine);
gfx_v11_0_wait_reg_mem(ring, reg_mem_engine, 0, 1,
adev->nbio.funcs->get_hdp_flush_req_offset(adev),
adev->nbio.funcs->get_hdp_flush_done_offset(adev),

View File

@@ -942,6 +942,7 @@ static const struct amdgpu_gfx_funcs gfx_v12_0_gfx_funcs = {
.select_me_pipe_q = &gfx_v12_0_select_me_pipe_q,
.update_perfmon_mgcg = &gfx_v12_0_update_perf_clk,
.get_gfx_shadow_info = &gfx_v12_0_get_gfx_shadow_info,
.get_hdp_flush_mask = &amdgpu_gfx_get_hdp_flush_mask,
};
static int gfx_v12_0_gpu_early_init(struct amdgpu_device *adev)
@@ -4393,25 +4394,13 @@ static void gfx_v12_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
u32 ref_and_mask, reg_mem_engine;
const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg;
if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
switch (ring->me) {
case 1:
ref_and_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe;
break;
case 2:
ref_and_mask = nbio_hf_reg->ref_and_mask_cp6 << ring->pipe;
break;
default:
return;
}
reg_mem_engine = 0;
} else {
ref_and_mask = nbio_hf_reg->ref_and_mask_cp0;
reg_mem_engine = 1; /* pfp */
if (!adev->gfx.funcs->get_hdp_flush_mask) {
dev_err(adev->dev, "%s: gfx hdp flush is not supported.\n", __func__);
return;
}
adev->gfx.funcs->get_hdp_flush_mask(ring, &ref_and_mask, &reg_mem_engine);
gfx_v12_0_wait_reg_mem(ring, reg_mem_engine, 0, 1,
adev->nbio.funcs->get_hdp_flush_req_offset(adev),
adev->nbio.funcs->get_hdp_flush_done_offset(adev),

View File

@@ -2068,23 +2068,15 @@ static int gfx_v7_0_ring_test_ring(struct amdgpu_ring *ring)
static void gfx_v7_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
{
u32 ref_and_mask;
int usepfp = ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE ? 0 : 1;
int usepfp;
struct amdgpu_device *adev = ring->adev;
if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
switch (ring->me) {
case 1:
ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe;
break;
case 2:
ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe;
break;
default:
return;
}
} else {
ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK;
if (!adev->gfx.funcs->get_hdp_flush_mask) {
dev_err(adev->dev, "%s: gfx hdp flush is not supported.\n", __func__);
return;
}
adev->gfx.funcs->get_hdp_flush_mask(ring, &ref_and_mask, &usepfp);
amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */
WAIT_REG_MEM_FUNCTION(3) | /* == */
@@ -4075,12 +4067,49 @@ static void gfx_v7_0_select_me_pipe_q(struct amdgpu_device *adev,
cik_srbm_select(adev, me, pipe, q, vm);
}
/**
* gfx_v7_0_get_hdp_flush_mask - get the reference and mask for HDP flush
*
* @ring: amdgpu_ring structure holding ring information
* @ref_and_mask: pointer to store the reference and mask
* @reg_mem_engine: pointer to store the register memory engine
*
* Calculates the reference and mask for HDP flush based on the ring type and me.
*/
static void gfx_v7_0_get_hdp_flush_mask(struct amdgpu_ring *ring,
uint32_t *ref_and_mask, uint32_t *reg_mem_engine)
{
if (!ring || !ref_and_mask || !reg_mem_engine) {
DRM_INFO("%s:invalid params\n", __func__);
return;
}
if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE ||
ring->funcs->type == AMDGPU_RING_TYPE_KIQ) {
switch (ring->me) {
case 1:
*ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe;
break;
case 2:
*ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe;
break;
default:
return;
}
*reg_mem_engine = 0;
} else {
*ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK;
*reg_mem_engine = 1;
}
}
static const struct amdgpu_gfx_funcs gfx_v7_0_gfx_funcs = {
.get_gpu_clock_counter = &gfx_v7_0_get_gpu_clock_counter,
.select_se_sh = &gfx_v7_0_select_se_sh,
.read_wave_data = &gfx_v7_0_read_wave_data,
.read_wave_sgprs = &gfx_v7_0_read_wave_sgprs,
.select_me_pipe_q = &gfx_v7_0_select_me_pipe_q
.select_me_pipe_q = &gfx_v7_0_select_me_pipe_q,
.get_hdp_flush_mask = &gfx_v7_0_get_hdp_flush_mask,
};
static const struct amdgpu_rlc_funcs gfx_v7_0_rlc_funcs = {

View File

@@ -5211,13 +5211,49 @@ static void gfx_v8_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t xcc_id
start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
}
/**
* gfx_v8_0_get_hdp_flush_mask - get the reference and mask for HDP flush
*
* @ring: amdgpu_ring structure holding ring information
* @ref_and_mask: pointer to store the reference and mask
* @reg_mem_engine: pointer to store the register memory engine
*
* Calculates the reference and mask for HDP flush based on the ring type and me.
*/
static void gfx_v8_0_get_hdp_flush_mask(struct amdgpu_ring *ring,
uint32_t *ref_and_mask, uint32_t *reg_mem_engine)
{
if (!ring || !ref_and_mask || !reg_mem_engine) {
DRM_INFO("%s:invalid params\n", __func__);
return;
}
if ((ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) ||
(ring->funcs->type == AMDGPU_RING_TYPE_KIQ)) {
switch (ring->me) {
case 1:
*ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe;
break;
case 2:
*ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe;
break;
default:
return;
}
*reg_mem_engine = 0;
} else {
*ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK;
*reg_mem_engine = WAIT_REG_MEM_ENGINE(1); /* pfp */
}
}
static const struct amdgpu_gfx_funcs gfx_v8_0_gfx_funcs = {
.get_gpu_clock_counter = &gfx_v8_0_get_gpu_clock_counter,
.select_se_sh = &gfx_v8_0_select_se_sh,
.read_wave_data = &gfx_v8_0_read_wave_data,
.read_wave_sgprs = &gfx_v8_0_read_wave_sgprs,
.select_me_pipe_q = &gfx_v8_0_select_me_pipe_q
.select_me_pipe_q = &gfx_v8_0_select_me_pipe_q,
.get_hdp_flush_mask = &gfx_v8_0_get_hdp_flush_mask,
};
static int gfx_v8_0_early_init(struct amdgpu_ip_block *ip_block)
@@ -6000,25 +6036,14 @@ static void gfx_v8_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
static void gfx_v8_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
{
u32 ref_and_mask, reg_mem_engine;
struct amdgpu_device *adev = ring->adev;
if ((ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) ||
(ring->funcs->type == AMDGPU_RING_TYPE_KIQ)) {
switch (ring->me) {
case 1:
ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe;
break;
case 2:
ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe;
break;
default:
return;
}
reg_mem_engine = 0;
} else {
ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK;
reg_mem_engine = WAIT_REG_MEM_ENGINE(1); /* pfp */
if (!adev->gfx.funcs->get_hdp_flush_mask) {
dev_err(adev->dev, "%s: gfx hdp flush is not supported.\n", __func__);
return;
}
adev->gfx.funcs->get_hdp_flush_mask(ring, &ref_and_mask, &reg_mem_engine);
amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */
WAIT_REG_MEM_FUNCTION(3) | /* == */

View File

@@ -2004,6 +2004,7 @@ static const struct amdgpu_gfx_funcs gfx_v9_0_gfx_funcs = {
.read_wave_sgprs = &gfx_v9_0_read_wave_sgprs,
.read_wave_vgprs = &gfx_v9_0_read_wave_vgprs,
.select_me_pipe_q = &gfx_v9_0_select_me_pipe_q,
.get_hdp_flush_mask = &amdgpu_gfx_get_hdp_flush_mask,
};
const struct amdgpu_ras_block_hw_ops gfx_v9_0_ras_ops = {
@@ -5380,25 +5381,13 @@ static void gfx_v9_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
u32 ref_and_mask, reg_mem_engine;
const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg;
if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
switch (ring->me) {
case 1:
ref_and_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe;
break;
case 2:
ref_and_mask = nbio_hf_reg->ref_and_mask_cp6 << ring->pipe;
break;
default:
return;
}
reg_mem_engine = 0;
} else {
ref_and_mask = nbio_hf_reg->ref_and_mask_cp0;
reg_mem_engine = 1; /* pfp */
if (!adev->gfx.funcs->get_hdp_flush_mask) {
dev_err(adev->dev, "%s: gfx hdp flush is not supported.\n", __func__);
return;
}
adev->gfx.funcs->get_hdp_flush_mask(ring, &ref_and_mask, &reg_mem_engine);
gfx_v9_0_wait_reg_mem(ring, reg_mem_engine, 0, 1,
adev->nbio.funcs->get_hdp_flush_req_offset(adev),
adev->nbio.funcs->get_hdp_flush_done_offset(adev),

View File

@@ -848,6 +848,7 @@ static const struct amdgpu_gfx_funcs gfx_v9_4_3_gfx_funcs = {
.switch_partition_mode = &gfx_v9_4_3_switch_compute_partition,
.ih_node_to_logical_xcc = &gfx_v9_4_3_ih_to_xcc_inst,
.get_xccs_per_xcp = &gfx_v9_4_3_get_xccs_per_xcp,
.get_hdp_flush_mask = &amdgpu_gfx_get_hdp_flush_mask,
};
static int gfx_v9_4_3_aca_bank_parser(struct aca_handle *handle,
@@ -2818,25 +2819,13 @@ static void gfx_v9_4_3_ring_emit_hdp_flush(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
u32 ref_and_mask, reg_mem_engine;
const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg;
if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
switch (ring->me) {
case 1:
ref_and_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe;
break;
case 2:
ref_and_mask = nbio_hf_reg->ref_and_mask_cp6 << ring->pipe;
break;
default:
return;
}
reg_mem_engine = 0;
} else {
ref_and_mask = nbio_hf_reg->ref_and_mask_cp0;
reg_mem_engine = 1; /* pfp */
if (!adev->gfx.funcs->get_hdp_flush_mask) {
dev_err(adev->dev, "%s: gfx hdp flush is not supported.\n", __func__);
return;
}
adev->gfx.funcs->get_hdp_flush_mask(ring, &ref_and_mask, &reg_mem_engine);
gfx_v9_4_3_wait_reg_mem(ring, reg_mem_engine, 0, 1,
adev->nbio.funcs->get_hdp_flush_req_offset(adev),
adev->nbio.funcs->get_hdp_flush_done_offset(adev),