This looks more like a workaround.
Can we write the C2PMSG_90 register to 1 on the PF side when host receive GPU_RESET/GPU_INIT request command?
Best Regards,
Kevin
-----Original Message-----
From: amd-gfx <amd-gfx-bounces@xxxxxxxxxxxxxxxxxxxxx> On Behalf Of Danijel Slivka
Sent: 2024年6月18日 23:00
To: amd-gfx@xxxxxxxxxxxxxxxxxxxxx
Cc: Slivka, Danijel <Danijel.Slivka@xxxxxxx>; Chen, JingWen (Wayne) <JingWen.Chen2@xxxxxxx>; Zhou, Peng Ju <PengJu.Zhou@xxxxxxx>
Subject: [PATCH] drm/amd/pm: check whether smu is idle in sriov case
Why:
If the reg mmMP1_SMN_C2PMSG_90 is being written to before or during amdgpu driver load or driver unload in sriov case, subsequent amdgpu driver load will fail at smu hw_init.
The default of mmMP1_SMN_C2PMSG_90 register at a clean environment is 0x1, and if value differs from 0x1, amdgpu driver load will fail.
How to fix:
This patch is to check whether smu is idle by sending a test message to smu. If smu is idle, it will respond.
This will avoid errors in case mmMP1_SMN_C2PMSG_90 is not 0x1 eventhough smu is idle.
Signed-off-by: Danijel Slivka <danijel.slivka@xxxxxxx>
Signed-off-by: Jingwen Chen <Jingwen.Chen2@xxxxxxx>
Signed-off-by: pengzhou <PengJu.Zhou@xxxxxxx>
---
.../gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c | 17 ++++++--
drivers/gpu/drm/amd/pm/swsmu/smu_cmn.c | 42 +++++++++++++++++++
drivers/gpu/drm/amd/pm/swsmu/smu_cmn.h | 3 ++
3 files changed, 58 insertions(+), 4 deletions(-)
diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c
index e17466cc1952..dafd91b352ec 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c
@@ -231,6 +231,7 @@ int smu_v13_0_check_fw_status(struct smu_context *smu) {
struct amdgpu_device *adev = smu->adev;
uint32_t mp1_fw_flags;
+ int ret = 0;
switch (amdgpu_ip_version(adev, MP1_HWIP, 0)) {
case IP_VERSION(13, 0, 4):
@@ -244,11 +245,19 @@ int smu_v13_0_check_fw_status(struct smu_context *smu)
break;
}
- if ((mp1_fw_flags & MP1_FIRMWARE_FLAGS__INTERRUPTS_ENABLED_MASK) >>
- MP1_FIRMWARE_FLAGS__INTERRUPTS_ENABLED__SHIFT)
- return 0;
+ if (!((mp1_fw_flags & MP1_FIRMWARE_FLAGS__INTERRUPTS_ENABLED_MASK) >>
+ MP1_FIRMWARE_FLAGS__INTERRUPTS_ENABLED__SHIFT))
+ return -EIO;
+
+ if (amdgpu_sriov_vf(adev)) {
+ ret = smu_cmn_wait_smu_idle(smu);
+ if (ret) {
+ dev_err(adev->dev, "SMU is not idle\n");
+ return ret;
+ }
+ }
- return -EIO;
+ return 0;
}
int smu_v13_0_check_fw_version(struct smu_context *smu) diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu_cmn.c b/drivers/gpu/drm/amd/pm/swsmu/smu_cmn.c
index 5592fd825aa3..de431c31ca7f 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/smu_cmn.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/smu_cmn.c
@@ -359,6 +359,48 @@ int smu_cmn_wait_for_response(struct smu_context *smu)
return res;
}
+/**
+ * smu_cmn_wait_smu_idle -- wait for smu to become idle
+ * @smu: pointer to an SMU context
+ *
+ * Send SMU_MSG_TestMessage to check whether SMU is idle.
+ * If SMU is idle, it will respond.
+ * The returned parameter will be the param you pass + 1.
+ *
+ * Return 0 on success, -errno on error, indicating the execution
+ * status and result of the message being waited for. See
+ * __smu_cmn_reg2errno() for details of the -errno.
+ */
+int smu_cmn_wait_smu_idle(struct smu_context *smu) {
+ u32 reg;
+ u32 param = 0xff00011;
+ uint32_t read_arg;
+ int res, index;
+
+ index = smu_cmn_to_asic_specific_index(smu,
+ CMN2ASIC_MAPPING_MSG,
+ SMU_MSG_TestMessage);
+
+ if (index < 0)
+ return index == -EACCES ? 0 : index;
+
+ __smu_cmn_send_msg(smu, index, param);
+ reg = __smu_cmn_poll_stat(smu);
+ res = __smu_cmn_reg2errno(smu, reg);
+
+ if (unlikely(smu->adev->pm.smu_debug_mask & SMU_DEBUG_HALT_ON_ERROR) &&
+ res && (res != -ETIME)) {
+ amdgpu_device_halt(smu->adev);
+ WARN_ON(1);
+ }
+
+ smu_cmn_read_arg(smu, &read_arg);
+ if (read_arg == param + 1)
+ return 0;
+ return res;
+}
+
/**
* smu_cmn_send_smc_msg_with_param -- send a message with parameter
* @smu: pointer to an SMU context
diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu_cmn.h b/drivers/gpu/drm/amd/pm/swsmu/smu_cmn.h
index 1de685defe85..486acfc956a5 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/smu_cmn.h
+++ b/drivers/gpu/drm/amd/pm/swsmu/smu_cmn.h
@@ -51,6 +51,9 @@ static inline int pcie_gen_to_speed(uint32_t gen) int smu_cmn_send_msg_without_waiting(struct smu_context *smu,
uint16_t msg_index,
uint32_t param);
+
+int smu_cmn_wait_smu_idle(struct smu_context *smu);
+
int smu_cmn_send_smc_msg_with_param(struct smu_context *smu,
enum smu_message_type msg,
uint32_t param,
--
2.34.1
