[AMD Official Use Only - General]
Thanks Alex, I will make another patch.
And please refer to the comments inline below.
-----Original Message-----
From: Alex Deucher <alexdeucher@xxxxxxxxx>
Sent: Friday, October 20, 2023 9:58 PM
To: Feng, Kenneth <Kenneth.Feng@xxxxxxx>
Cc: amd-gfx@xxxxxxxxxxxxxxxxxxxxx; Wang, Yang(Kevin) <KevinYang.Wang@xxxxxxx>
Subject: Re: [PATCH] drm/amd/pm: fix the high voltage and temperature issue on smu 13
Caution: This message originated from an External Source. Use proper caution when opening attachments, clicking links, or responding.
On Fri, Oct 20, 2023 at 4:32 AM Kenneth Feng <kenneth.feng@xxxxxxx> wrote:
>
> fix the high voltage and temperature issue after the driver is
> unloaded on smu 13.0.0, smu 13.0.7 and smu 13.0.10
>
> Signed-off-by: Kenneth Feng <kenneth.feng@xxxxxxx>
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 36 +++++++++++++++----
> drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c | 4 +--
> drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c | 27 ++++++++++++--
> drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h | 1 +
> drivers/gpu/drm/amd/pm/swsmu/inc/smu_v13_0.h | 2 ++
> .../gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c | 13 +++++++
> .../drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c | 8 ++++-
> .../drm/amd/pm/swsmu/smu13/smu_v13_0_7_ppt.c | 8 ++++-
> 8 files changed, 86 insertions(+), 13 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> index 31f8c3ead161..c5c892a8b3f9 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> @@ -3986,13 +3986,23 @@ int amdgpu_device_init(struct amdgpu_device *adev,
> }
> }
> } else {
> - tmp = amdgpu_reset_method;
> - /* It should do a default reset when loading or reloading the driver,
> - * regardless of the module parameter reset_method.
> - */
> - amdgpu_reset_method = AMD_RESET_METHOD_NONE;
> - r = amdgpu_asic_reset(adev);
> - amdgpu_reset_method = tmp;
> + switch (amdgpu_ip_version(adev, MP1_HWIP, 0)) {
> + case IP_VERSION(13, 0, 0):
> + case IP_VERSION(13, 0, 7):
> + case IP_VERSION(13, 0, 10):
> + r = psp_gpu_reset(adev);
> + break;
> + default:
> + tmp = amdgpu_reset_method;
> + /* It should do a default reset when loading or reloading the driver,
> + * regardless of the module parameter reset_method.
> + */
> + amdgpu_reset_method = AMD_RESET_METHOD_NONE;
> + r = amdgpu_asic_reset(adev);
> + amdgpu_reset_method = tmp;
> + break;
> + }
> +
> if (r) {
> dev_err(adev->dev, "asic reset on init failed\n");
> goto failed; @@ -5945,6 +5955,18 @@
> int amdgpu_device_baco_exit(struct drm_device *dev)
> return -ENOTSUPP;
>
> ret = amdgpu_dpm_baco_exit(adev);
> +
> + if (!ret)
> + switch (amdgpu_ip_version(adev, MP1_HWIP, 0)) {
> + case IP_VERSION(13, 0, 0):
> + case IP_VERSION(13, 0, 7):
> + case IP_VERSION(13, 0, 10):
> + adev->gfx.is_poweron = false;
> + break;
> + default:
> + break;
> + }
Maybe better to move this into smu_v13_0_0_baco_exit() so we keep the asic specific details out of the common files?
> +
> if (ret)
> return ret;
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c
> b/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c
> index 80ca2c05b0b8..3ad38e42773b 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c
> @@ -73,7 +73,7 @@ gmc_v11_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
> * fini/suspend, so the overall state doesn't
> * change over the course of suspend/resume.
> */
> - if (!adev->in_s0ix)
> + if (!adev->in_s0ix && adev->gfx.is_poweron)
> amdgpu_gmc_set_vm_fault_masks(adev, AMDGPU_GFXHUB(0), false);
> break;
> case AMDGPU_IRQ_STATE_ENABLE:
> @@ -85,7 +85,7 @@ gmc_v11_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
> * fini/suspend, so the overall state doesn't
> * change over the course of suspend/resume.
> */
> - if (!adev->in_s0ix)
> + if (!adev->in_s0ix && adev->gfx.is_poweron)
> amdgpu_gmc_set_vm_fault_masks(adev, AMDGPU_GFXHUB(0), true);
> break;
> default:
These changes are probably a valid bug fix on their own.
[Kenneth] - When driver is unloaded, gfx core is powered off first. Then in gmc_hw_fini, the gfxhub interruption operation needs to be skipped. Do we need a separate patch for this?
> diff --git a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
> b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
> index 7c3356d6da5e..30e5f7161737 100644
> --- a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
> +++ b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
> @@ -733,7 +733,7 @@ static int smu_early_init(void *handle)
> smu->adev = adev;
> smu->pm_enabled = !!amdgpu_dpm;
> smu->is_apu = false;
> - smu->smu_baco.state = SMU_BACO_STATE_EXIT;
> + smu->smu_baco.state = SMU_BACO_STATE_NONE;
I'm not sure I understand this change. Is this just to set the default BACO state? Maybe this would be better as a separate patch.
[Kenneth] - smu->smu_baco.state is needed when driver is unloaded, if it's baco exited, then need to reset MP1_FIRMWARE_FLAG = 0, otherwise MP1_FIRMWARE_FLAG doesn't need to be reset.
Currently by default smu->smu_baco.state is baco exited status, we can't recognize if it's really a hardware baco exited status. Do you think we still need a separate patch for it?
> smu->smu_baco.platform_support = false;
> smu->user_dpm_profile.fan_mode = -1;
>
> @@ -1740,10 +1740,25 @@ static int smu_smc_hw_cleanup(struct smu_context *smu)
> return 0;
> }
>
> +static int smu_reset_mp1_state(struct smu_context *smu) {
> + struct amdgpu_device *adev = smu->adev;
> +
> + switch (amdgpu_ip_version(adev, MP1_HWIP, 0)) {
> + case IP_VERSION(13, 0, 0):
> + case IP_VERSION(13, 0, 7):
> + case IP_VERSION(13, 0, 10):
> + return smu_set_mp1_state(smu, PP_MP1_STATE_UNLOAD);
> + default:
> + return 0;
> + }
> +}
> +
> static int smu_hw_fini(void *handle)
> {
> struct amdgpu_device *adev = (struct amdgpu_device *)handle;
> struct smu_context *smu = adev->powerplay.pp_handle;
> + int ret;
>
> if (amdgpu_sriov_vf(adev) && !amdgpu_sriov_is_pp_one_vf(adev))
> return 0;
> @@ -1761,7 +1776,15 @@ static int smu_hw_fini(void *handle)
>
> adev->pm.dpm_enabled = false;
>
> - return smu_smc_hw_cleanup(smu);
> + ret = smu_smc_hw_cleanup(smu);
> + if (ret)
> + return ret;
> +
> + ret = smu_reset_mp1_state(smu);
> + if (ret)
> + return ret;
> +
> + return 0;
> }
>
> static void smu_late_fini(void *handle) diff --git
> a/drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h
> b/drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h
> index 1454eed76604..9f2dbc90b606 100644
> --- a/drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h
> +++ b/drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h
> @@ -419,6 +419,7 @@ enum smu_reset_mode { enum smu_baco_state {
> SMU_BACO_STATE_ENTER = 0,
> SMU_BACO_STATE_EXIT,
> + SMU_BACO_STATE_NONE,
> };
>
> struct smu_baco_context {
> diff --git a/drivers/gpu/drm/amd/pm/swsmu/inc/smu_v13_0.h
> b/drivers/gpu/drm/amd/pm/swsmu/inc/smu_v13_0.h
> index cc02f979e9e9..43c7ba68eb50 100644
> --- a/drivers/gpu/drm/amd/pm/swsmu/inc/smu_v13_0.h
> +++ b/drivers/gpu/drm/amd/pm/swsmu/inc/smu_v13_0.h
> @@ -299,5 +299,7 @@ int smu_v13_0_update_pcie_parameters(struct smu_context *smu,
> uint8_t pcie_gen_cap,
> uint8_t pcie_width_cap);
>
> +int smu_v13_0_disable_pmfw_state(struct smu_context* smu);
> +
> #endif
> #endif
> 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 bcb7ab9d2221..0724441e53ef 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
> @@ -2473,3 +2473,16 @@ int smu_v13_0_update_pcie_parameters(struct
> smu_context *smu,
>
> return 0;
> }
> +
> +int smu_v13_0_disable_pmfw_state(struct smu_context* smu) {
> + int ret;
> + struct amdgpu_device *adev = smu->adev;
> +
> + WREG32_PCIE(MP1_Public | (smnMP1_FIRMWARE_FLAGS & 0xffffffff),
> + 0);
> +
> + ret = RREG32_PCIE(MP1_Public |
> + (smnMP1_FIRMWARE_FLAGS &
> + 0xffffffff));
> +
> + return ret == 0 ? 0 : -EINVAL; }
> diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c
> b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c
> index 47d008cbc186..0a167f70f4bc 100644
> --- a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c
> +++ b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c
> @@ -2758,7 +2758,13 @@ static int smu_v13_0_0_set_mp1_state(struct
> smu_context *smu,
>
> switch (mp1_state) {
> case PP_MP1_STATE_UNLOAD:
> - ret = smu_cmn_set_mp1_state(smu, mp1_state);
> + ret = smu_cmn_send_smc_msg_with_param(smu,
> + SMU_MSG_PrepareMp1ForUnload,
> + 0x55,
> + NULL);
> +
> + if(!ret && smu->smu_baco.state == SMU_BACO_STATE_EXIT)
space between if and (
> + ret = smu_v13_0_disable_pmfw_state(smu);
> +
> break;
> default:
> /* Ignore others */
> diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_7_ppt.c
> b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_7_ppt.c
> index b8a7a1d853df..d7a4a03b1e31 100644
> --- a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_7_ppt.c
> +++ b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_7_ppt.c
> @@ -2429,7 +2429,13 @@ static int smu_v13_0_7_set_mp1_state(struct
> smu_context *smu,
>
> switch (mp1_state) {
> case PP_MP1_STATE_UNLOAD:
> - ret = smu_cmn_set_mp1_state(smu, mp1_state);
> + ret = smu_cmn_send_smc_msg_with_param(smu,
> + SMU_MSG_PrepareMp1ForUnload,
> + 0x55,
> + NULL);
> +
> + if(!ret && smu->smu_baco.state == SMU_BACO_STATE_EXIT)
Same here.
Alex
> + ret = smu_v13_0_disable_pmfw_state(smu);
> +
> break;
> default:
> /* Ignore others */
> --
> 2.34.1
>
