| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <63778ba1-6ae5-0cfe-9a29-6b8cbd8118e3@amd.com>
Date: Thu, 17 Aug 2023 13:28:08 +0530
From: "Yadav, Arvind" <arvyadav@....com>
To: Shashank Sharma <shashank.sharma@....com>,
Arvind Yadav <Arvind.Yadav@....com>, Christian.Koenig@....com,
alexander.deucher@....com, Xinhui.Pan@....com, airlied@...il.com,
daniel@...ll.ch, amd-gfx@...ts.freedesktop.org,
dri-devel@...ts.freedesktop.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 1/3] drm/amdgpu: Add new api to switch on/off power
profile mode
On 8/14/2023 8:28 PM, Shashank Sharma wrote:
> Hey Arvind,
>
> On 14/08/2023 09:34, Arvind Yadav wrote:
>> This patch adds a function which will allow to
>> change the GPU power profile based on a submitted job.
>> This can optimize the power performance when the
>> workload is on.
>>
>> Cc: Shashank Sharma <shashank.sharma@....com>
>> Cc: Christian Koenig <christian.koenig@....com>
>> Cc: Alex Deucher <alexander.deucher@....com>
>> Signed-off-by: Arvind Yadav <Arvind.Yadav@....com>
>> ---
>> drivers/gpu/drm/amd/amdgpu/Makefile | 2 +-
>> drivers/gpu/drm/amd/amdgpu/amdgpu.h | 3 +
>> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 2 +
>> drivers/gpu/drm/amd/amdgpu/amdgpu_workload.c | 156 ++++++++++++++++++
>> drivers/gpu/drm/amd/include/amdgpu_workload.h | 44 +++++
>> 5 files changed, 206 insertions(+), 1 deletion(-)
>> create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_workload.c
>> create mode 100644 drivers/gpu/drm/amd/include/amdgpu_workload.h
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile
>> b/drivers/gpu/drm/amd/amdgpu/Makefile
>> index 415a7fa395c4..6a9e187d61e1 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/Makefile
>> +++ b/drivers/gpu/drm/amd/amdgpu/Makefile
>> @@ -60,7 +60,7 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \
>> amdgpu_umc.o smu_v11_0_i2c.o amdgpu_fru_eeprom.o amdgpu_rap.o \
>> amdgpu_fw_attestation.o amdgpu_securedisplay.o \
>> amdgpu_eeprom.o amdgpu_mca.o amdgpu_psp_ta.o amdgpu_lsdma.o \
>> - amdgpu_ring_mux.o
>> + amdgpu_ring_mux.o amdgpu_workload.o
>> amdgpu-$(CONFIG_PROC_FS) += amdgpu_fdinfo.o
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> index 02b827785e39..1939fa1af8a6 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>> @@ -107,6 +107,7 @@
>> #include "amdgpu_fdinfo.h"
>> #include "amdgpu_mca.h"
>> #include "amdgpu_ras.h"
>> +#include "amdgpu_workload.h"
>> #define MAX_GPU_INSTANCE 16
>> @@ -1050,6 +1051,8 @@ struct amdgpu_device {
>> bool job_hang;
>> bool dc_enabled;
>> +
>> + struct amdgpu_smu_workload smu_workload;
>> };
>> static inline struct amdgpu_device *drm_to_adev(struct drm_device
>> *ddev)
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> index 5c7d40873ee2..0ec18b8fe29f 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> @@ -3672,6 +3672,8 @@ int amdgpu_device_init(struct amdgpu_device *adev,
>> INIT_WORK(&adev->xgmi_reset_work,
>> amdgpu_device_xgmi_reset_func);
>> + amdgpu_smu_workload_init(adev);
>> +
>> adev->gfx.gfx_off_req_count = 1;
>> adev->gfx.gfx_off_residency = 0;
>> adev->gfx.gfx_off_entrycount = 0;
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_workload.c
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_workload.c
>> new file mode 100644
>> index 000000000000..ce0339d75c12
>> --- /dev/null
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_workload.c
>> @@ -0,0 +1,156 @@
>> +// SPDX-License-Identifier: MIT
>> +/*
>> + * Copyright 2023 Advanced Micro Devices, Inc.
>> + *
>> + * Permission is hereby granted, free of charge, to any person
>> obtaining a
>> + * copy of this software and associated documentation files (the
>> "Software"),
>> + * to deal in the Software without restriction, including without
>> limitation
>> + * the rights to use, copy, modify, merge, publish, distribute,
>> sublicense,
>> + * and/or sell copies of the Software, and to permit persons to whom
>> the
>> + * Software is furnished to do so, subject to the following conditions:
>> + *
>> + * The above copyright notice and this permission notice shall be
>> included in
>> + * all copies or substantial portions of the Software.
>> + *
>> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
>> EXPRESS OR
>> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
>> MERCHANTABILITY,
>> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO
>> EVENT SHALL
>> + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM,
>> DAMAGES OR
>> + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
>> OTHERWISE,
>> + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
>> USE OR
>> + * OTHER DEALINGS IN THE SOFTWARE.
>> + *
>> + */
>> +
>> +#include "amdgpu.h"
>> +
>> +/* 100 millsecond timeout */
>> +#define SMU_IDLE_TIMEOUT msecs_to_jiffies(100)
>> +
>> +static enum PP_SMC_POWER_PROFILE
>> +ring_to_power_profile(uint32_t ring_type)
>> +{
>> + switch (ring_type) {
>> + case AMDGPU_RING_TYPE_GFX:
>> + return PP_SMC_POWER_PROFILE_FULLSCREEN3D;
>> + case AMDGPU_RING_TYPE_COMPUTE:
>> + return PP_SMC_POWER_PROFILE_COMPUTE;
>> + case AMDGPU_RING_TYPE_UVD:
>> + case AMDGPU_RING_TYPE_VCE:
>> + case AMDGPU_RING_TYPE_UVD_ENC:
>> + case AMDGPU_RING_TYPE_VCN_DEC:
>> + case AMDGPU_RING_TYPE_VCN_ENC:
>> + case AMDGPU_RING_TYPE_VCN_JPEG:
>> + return PP_SMC_POWER_PROFILE_VIDEO;
>> + default:
>> + return PP_SMC_POWER_PROFILE_BOOTUP_DEFAULT;
>> + }
>> +}
>> +
>> +static void
>> +amdgpu_power_profile_set(struct amdgpu_device *adev,
>> + enum PP_SMC_POWER_PROFILE profile)
> This function expects the caller to hold the smu_workload_mutex, may
> be we should document it.
>> +{
>> + int ret = amdgpu_dpm_switch_power_profile(adev, profile, true);
> I think we should pass this return value to caller instead of keeping
> the function void.
Noted,
>> +
>> + if (ret == 0) {
>> + /* Set the bit for the submitted workload profile */
>> + adev->smu_workload.submit_workload_status |= (1 << profile);
>> + atomic_inc(&adev->smu_workload.power_profile_ref[profile]);
>> + } else {
>> + DRM_ERROR("Failed to set power profile, error %d\n", ret);
> This can be a warning instead of error.
Noted,
>> + }
>> +
>> +}
>> +
>> +static void
>> +amdgpu_power_profile_clear(struct amdgpu_device *adev,
>> + enum PP_SMC_POWER_PROFILE profile)
>> +{
>> + int ret = amdgpu_dpm_switch_power_profile(adev, profile, false);
> same for return value here as well.
Noted,
>
>> +
>> + if (ret == 0) {
>> + /* Clear the bit for the submitted workload profile */
>> + adev->smu_workload.submit_workload_status &= ~(1 << profile);
>> + } else
>> + DRM_ERROR("Failed to clear power profile, error %d\n", ret);
>> +
>> +}
>> +
>> +static void amdgpu_smu_idle_work_handler(struct work_struct *work)
>> +{
>> +
>> + struct amdgpu_smu_workload *wl = container_of(work,
>> + struct amdgpu_smu_workload,
>> + smu_delayed_work.work);
>> + struct amdgpu_device *adev = wl->adev;
>> + bool reschedule = false;
>> +
>> + mutex_lock(&adev->smu_workload.workload_lock);
>> + for (int index = fls(adev->smu_workload.submit_workload_status);
> This can be kept outside the for() for better readability and alignment.
Noted,
>> + index >= 0; index--) {
>> + if
>> (!atomic_read(&adev->smu_workload.power_profile_ref[index]) &&
>> + adev->smu_workload.submit_workload_status & (1 << index)) {
>> + amdgpu_power_profile_clear(adev, index);
>> + } else if
>> (atomic_read(&adev->smu_workload.power_profile_ref[index]))
>> + reschedule = true;
>> + }
>> +
>
> This block can be re-arranged a bit for better readability, pls consider:
>
> for () {
>
> atomic_t val =
> atomic_read(&adev->smu_workload.power_profile_ref[index];
>
> if (val) {
>
> reschedule = true;
>
> break;
>
> } else {
>
> if (adev->smu_workload.submit_workload_status & (1 << index))
>
> amdgpu_power_profile_clear(adev, index);
>
> }
>
> }
>
Noted,
>> + if (reschedule)
>> + schedule_delayed_work(&adev->smu_workload.smu_delayed_work,
>> + SMU_IDLE_TIMEOUT);
> pls check the return value of work
Noted,
>> +
>> + mutex_unlock(&adev->smu_workload.workload_lock);
>> +}
>> +
>> +void amdgpu_put_workload_profile(struct amdgpu_device *adev,
>> + uint32_t ring_type)
>> +{
>> +
>> + enum PP_SMC_POWER_PROFILE profile =
>> ring_to_power_profile(ring_type);
>> +
>> + if (profile == PP_SMC_POWER_PROFILE_BOOTUP_DEFAULT)
>> + return;
>> +
>> + mutex_lock(&adev->smu_workload.workload_lock);
>> + atomic_dec(&adev->smu_workload.power_profile_ref[profile]);
>> + schedule_delayed_work(&adev->smu_workload.smu_delayed_work,
>> SMU_IDLE_TIMEOUT);
>> + mutex_unlock(&adev->smu_workload.workload_lock);
>> +}
>> +
>> +void amdgpu_set_workload_profile(struct amdgpu_device *adev,
>> + uint32_t ring_type)
> I would prefer if you can split this patch into two, one just to set
> profile, other to clear profile and schedule work.
Noted,
>> +{
>> + enum PP_SMC_POWER_PROFILE profile =
>> ring_to_power_profile(ring_type);
>> +
>> + if (profile == PP_SMC_POWER_PROFILE_BOOTUP_DEFAULT)
>> + return;
>> +
>> + mutex_lock(&adev->smu_workload.workload_lock);
>> + cancel_delayed_work_sync(&adev->smu_workload.smu_delayed_work);
> Please check the return value here and proceed only when we were able
> to cancel successfully.
Noted,
>> +
>> + amdgpu_power_profile_set(adev, profile);
>> +
>> + /* Clear the already finished jobs of higher power profile*/
>> + for (int index = fls(adev->smu_workload.submit_workload_status);
>> + index > profile; index--) {
>> + if
>> (!atomic_read(&adev->smu_workload.power_profile_ref[index]) &&
>> + adev->smu_workload.submit_workload_status & (1 << index)) {
>> + amdgpu_power_profile_clear(adev, index);
>> + }
>> + }
>> +
>> + mutex_unlock(&adev->smu_workload.workload_lock);
>> +}
>> +
>> +void amdgpu_smu_workload_init(struct amdgpu_device *adev)
>> +{
>> + struct amdgpu_smu_workload wl;
>> +
>> + wl.adev = adev;
>> + wl.submit_workload_status = 0;
>> + adev->smu_workload = wl;
>
> Why do we need variable wl at all, which is a local variable of the
> stack ? You can just do:
Noted,
>
> adev->smu_workload.adev = adev;
> adev->smu_workload.submit_workload_status = 0;
>
>> +
>> + mutex_init(&adev->smu_workload.workload_lock);
>> + INIT_DELAYED_WORK(&adev->smu_workload.smu_delayed_work,
>> amdgpu_smu_idle_work_handler);
>
> Are we missing the respective amdgpu_smu_workload_fini which will
> destroy the mutex ?
>
Noted,
Regards,
~Arvind
> - Shashank
>
>> +}
>> diff --git a/drivers/gpu/drm/amd/include/amdgpu_workload.h
>> b/drivers/gpu/drm/amd/include/amdgpu_workload.h
>> new file mode 100644
>> index 000000000000..09804c3d2869
>> --- /dev/null
>> +++ b/drivers/gpu/drm/amd/include/amdgpu_workload.h
>> @@ -0,0 +1,44 @@
>> +/* SPDX-License-Identifier: MIT */
>> +/*
>> + * Copyright 2023 Advanced Micro Devices, Inc.
>> + *
>> + * Permission is hereby granted, free of charge, to any person
>> obtaining a
>> + * copy of this software and associated documentation files (the
>> "Software"),
>> + * to deal in the Software without restriction, including without
>> limitation
>> + * the rights to use, copy, modify, merge, publish, distribute,
>> sublicense,
>> + * and/or sell copies of the Software, and to permit persons to whom
>> the
>> + * Software is furnished to do so, subject to the following conditions:
>> + *
>> + * The above copyright notice and this permission notice shall be
>> included in
>> + * all copies or substantial portions of the Software.
>> + *
>> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
>> EXPRESS OR
>> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
>> MERCHANTABILITY,
>> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO
>> EVENT SHALL
>> + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM,
>> DAMAGES OR
>> + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
>> OTHERWISE,
>> + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
>> USE OR
>> + * OTHER DEALINGS IN THE SOFTWARE.
>> + *
>> + */
>> +
>> +#ifndef _AMDGPU_WORKLOAD_H_
>> +#define _AMDGPU_WORKLOAD_H_
>> +
>> +struct amdgpu_smu_workload {
>> + struct amdgpu_device *adev;
>> + struct mutex workload_lock;
>> + struct delayed_work smu_delayed_work;
>> + uint32_t submit_workload_status;
>> + atomic_t power_profile_ref[PP_SMC_POWER_PROFILE_COUNT];
>> +};
>> +
>> +void amdgpu_set_workload_profile(struct amdgpu_device *adev,
>> + uint32_t ring_type);
>> +
>> +void amdgpu_put_workload_profile(struct amdgpu_device *adev,
>> + uint32_t ring_type);
>> +
>> +void amdgpu_smu_workload_init(struct amdgpu_device *adev);
>> +
>> +#endif
Powered by blists - more mailing lists