On 28-Sep-26 4:01 PM, Victor Zhao wrote:
[why]
Some register values span two consecutive dwords and must be sampled
with one MES operation.
[how]
Add amdgpu_mes_rreg64(), set READ_REG read64Bits for MES 11 and
MES 12, and combine the two returned dwords into a 64-bit value.
Reject 64-bit reads on MES 11 scheduler firmware older than 0x42,
which predates read64Bits support.
v2: add MES 11 support with the same read64Bits option
v3: return read status separately and require MES 11 firmware 0x42 or newer
Signed-off-by: Victor Zhao <[email protected]>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c | 43 +++++++++++++++++++
drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h | 3 ++
drivers/gpu/drm/amd/amdgpu/mes_v11_0.c | 6 +++
drivers/gpu/drm/amd/amdgpu/mes_v12_0.c | 2 +
drivers/gpu/drm/amd/amdgpu/mes_v12_1.c | 2 +
drivers/gpu/drm/amd/include/mes_v11_api_def.h | 7 +++
6 files changed, 63 insertions(+)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c
index 5bebe74de01ad..3be6e577a8aa7 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c
@@ -582,6 +582,7 @@ uint32_t amdgpu_mes_rreg(struct amdgpu_device *adev,
uint32_t reg,
op_input.op = MES_MISC_OP_READ_REG;
op_input.read_reg.reg_offset = reg;
op_input.read_reg.buffer_addr = read_val_gpu_addr;
+ op_input.read_reg.read64_bits = false;
if (!adev->mes.funcs->misc_op) {
dev_err(adev->dev, "mes rreg is not supported!\n");
@@ -602,6 +603,48 @@ uint32_t amdgpu_mes_rreg(struct amdgpu_device *adev,
uint32_t reg,
return val;
}
+int amdgpu_mes_rreg64(struct amdgpu_device *adev, uint32_t reg,
+ uint32_t xcc_id, uint64_t *val)
+{
+ struct mes_misc_op_input op_input = { 0 };
+ uint32_t addr_offset;
+ uint32_t *read_val_ptr;
+ int r;
+
+ r = amdgpu_wb_get(adev, &addr_offset);
+ if (r) {
+ dev_err(adev->dev, "critical bug! too many mes readers\n");
+ return r;
+ }
+
+ read_val_ptr = (uint32_t *)&adev->wb.wb[addr_offset];
read_val_ptr may be declared/assigned as u64* and used during read back.
That aside, this looks good.
Thanks,
Lijo
+ op_input.xcc_id = xcc_id;
+ op_input.op = MES_MISC_OP_READ_REG;
+ op_input.read_reg.reg_offset = reg;
+ op_input.read_reg.buffer_addr = adev->wb.gpu_addr + addr_offset * 4;
+ op_input.read_reg.read64_bits = true;
+
+ if (!adev->mes.funcs->misc_op) {
+ dev_err(adev->dev, "mes rreg is not supported!\n");
+ r = -EINVAL;
+ goto error;
+ }
+
+ amdgpu_mes_lock(&adev->mes);
+ r = adev->mes.funcs->misc_op(&adev->mes, &op_input);
+ amdgpu_mes_unlock(&adev->mes);
+ if (r) {
+ dev_err(adev->dev, "failed to read reg (0x%x)\n", reg);
+ } else {
+ *val = READ_ONCE(read_val_ptr[0]);
+ *val |= (uint64_t)READ_ONCE(read_val_ptr[1]) << 32;
+ }
+
+error:
+ amdgpu_wb_free(adev, addr_offset);
+ return r;
+}
+
int amdgpu_mes_wreg(struct amdgpu_device *adev, uint32_t reg,
uint32_t val, uint32_t xcc_id)
{
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
index 33a426f58e723..1e801f427ddb2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
@@ -373,6 +373,7 @@ struct mes_misc_op_input {
struct {
uint32_t reg_offset;
uint64_t buffer_addr;
+ bool read64_bits;
} read_reg;
struct {
@@ -545,6 +546,8 @@ int amdgpu_mes_detect_and_reset_hung_queues(struct
amdgpu_device *adev,
uint32_t amdgpu_mes_rreg(struct amdgpu_device *adev, uint32_t reg,
uint32_t xcc_id);
+int amdgpu_mes_rreg64(struct amdgpu_device *adev, uint32_t reg,
+ uint32_t xcc_id, uint64_t *val);
int amdgpu_mes_wreg(struct amdgpu_device *adev,
uint32_t reg, uint32_t val, uint32_t xcc_id);
int amdgpu_mes_reg_write_reg_wait(struct amdgpu_device *adev,
diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c
b/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c
index d8b4d3deff605..ef82f7536e449 100644
--- a/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c
@@ -902,9 +902,15 @@ static int mes_v11_0_misc_op(struct amdgpu_mes *mes,
switch (input->op) {
case MES_MISC_OP_READ_REG:
+ if (input->read_reg.read64_bits &&
+ (mes->sched_version & AMDGPU_MES_VERSION_MASK) < 0x42)
+ return -EOPNOTSUPP;
+
misc_pkt.opcode = MESAPI_MISC__READ_REG;
misc_pkt.read_reg.reg_offset = input->read_reg.reg_offset;
misc_pkt.read_reg.buffer_addr = input->read_reg.buffer_addr;
+ misc_pkt.read_reg.option.bits.read64Bits =
+ input->read_reg.read64_bits;
break;
case MES_MISC_OP_WRITE_REG:
misc_pkt.opcode = MESAPI_MISC__WRITE_REG;
diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c
b/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c
index 04465804c2544..0d2a60528802e 100644
--- a/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c
@@ -858,6 +858,8 @@ static int mes_v12_0_misc_op(struct amdgpu_mes *mes,
misc_pkt.opcode = MESAPI_MISC__READ_REG;
misc_pkt.read_reg.reg_offset = input->read_reg.reg_offset;
misc_pkt.read_reg.buffer_addr = input->read_reg.buffer_addr;
+ misc_pkt.read_reg.option.bits.read64Bits =
+ input->read_reg.read64_bits;
break;
case MES_MISC_OP_WRITE_REG:
misc_pkt.opcode = MESAPI_MISC__WRITE_REG;
diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v12_1.c
b/drivers/gpu/drm/amd/amdgpu/mes_v12_1.c
index d17ced46d5f73..e0a488e516c09 100644
--- a/drivers/gpu/drm/amd/amdgpu/mes_v12_1.c
+++ b/drivers/gpu/drm/amd/amdgpu/mes_v12_1.c
@@ -581,6 +581,8 @@ static int mes_v12_1_misc_op(struct amdgpu_mes *mes,
case MES_MISC_OP_READ_REG:
misc_pkt.opcode = MESAPI_MISC__READ_REG;
misc_pkt.read_reg.buffer_addr = input->read_reg.buffer_addr;
+ misc_pkt.read_reg.option.bits.read64Bits =
+ input->read_reg.read64_bits;
mes_v12_1_get_rrmt(input->read_reg.reg_offset,
GET_INST(GC, input->xcc_id),
&misc_pkt.read_reg.rrmt_opt,
diff --git a/drivers/gpu/drm/amd/include/mes_v11_api_def.h
b/drivers/gpu/drm/amd/include/mes_v11_api_def.h
index 1fe824b41f2a5..b27ec78da3b08 100644
--- a/drivers/gpu/drm/amd/include/mes_v11_api_def.h
+++ b/drivers/gpu/drm/amd/include/mes_v11_api_def.h
@@ -635,6 +635,13 @@ struct WRITE_REG {
struct READ_REG {
uint32_t reg_offset;
uint64_t buffer_addr;
+ union {
+ struct {
+ uint32_t read64Bits : 1;
+ uint32_t reserved : 31;
+ } bits;
+ uint32_t all;
+ } option;
};
enum WRM_OPERATION {
