[v2] drm/amdgpu: skip vram operation for BAMACO runtime
Alex Deucher
alexdeucher at gmail.com
Thu Dec 17 19:09:09 UTC 2020
On Mon, Dec 14, 2020 at 5:13 AM Lazar, Lijo <Lijo.Lazar at amd.com> wrote:
>
> [AMD Public Use]
>
> >-----Original Message-----
> >From: amd-gfx <amd-gfx-bounces at lists.freedesktop.org> On Behalf Of Likun
> >Gao
> >Sent: Monday, December 14, 2020 1:08 PM
> >To: amd-gfx at lists.freedesktop.org
> >Cc: Deucher, Alexander <Alexander.Deucher at amd.com>; Gao, Likun
> ><Likun.Gao at amd.com>; Feng, Kenneth <Kenneth.Feng at amd.com>; Zhang,
> >Hawking <Hawking.Zhang at amd.com>
> >Subject: [v2] drm/amdgpu: skip vram operation for BAMACO runtime
> >
> >[CAUTION: External Email]
> >
> >From: Likun Gao <Likun.Gao at amd.com>
> >
> >Skip vram related operation for bamaco rumtime suspend and resume as
> >vram is alive when BAMACO.
> >It can save about 32ms when suspend and about 15ms when resume.
> >
> >v2: add local adev variable for maco enablement which checked by device
> >MACO support cap and runtime method parameter to have a mix of devices
> >that support different runtime pm modes.
Something else just occurred to me, we need to handle
DPM_FLAG_NO_DIRECT_COMPLETE properly with the *MA* variants. We need
to make sure set that flag if there are any cases where we will lose
the contents of memory and not have it backed up.
Alex
> >
> >Signed-off-by: Likun Gao <Likun.Gao at amd.com>
> >Change-Id: I99cbaf720bfeecdc6682057b238d29c4e41cf155
> >---
> > drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 +
> > drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 9 ++-
> > drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c | 72 +++++++++++--------
> > .../amd/pm/swsmu/smu11/sienna_cichlid_ppt.c | 7 +-
> > 4 files changed, 56 insertions(+), 33 deletions(-)
> >
> >diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> >b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> >index eed5410947e9..288c41e0b61f 100644
> >--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> >+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> >@@ -1027,6 +1027,7 @@ struct amdgpu_device {
> > /* enable runtime pm on the device */
> > bool runpm;
> > bool in_runpm;
> >+ bool runpm_maco_en;
> >
> > bool pm_sysfs_en;
> > bool ucode_sysfs_en;
>
> Isn't it better to move pm related variables to amdgpu_pm (at least going forward) instead of keeping too many in adev?
>
> >diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> >b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> >index 0ec7c28c4d5a..0c608b903ec3 100644
> >--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> >+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> >@@ -2464,7 +2464,8 @@ static int amdgpu_device_ip_late_init(struct
> >amdgpu_device *adev)
> > amdgpu_device_set_cg_state(adev, AMD_CG_STATE_GATE);
> > amdgpu_device_set_pg_state(adev, AMD_PG_STATE_GATE);
> >
> >- amdgpu_device_fill_reset_magic(adev);
> >+ if (!adev->runpm_maco_en || !adev->in_runpm)
> >+ amdgpu_device_fill_reset_magic(adev);
> > r = amdgpu_device_enable_mgpu_fan_boost();
> > if (r)
> >@@ -3706,7 +3707,8 @@ int amdgpu_device_suspend(struct drm_device
> >*dev, bool fbcon)
> > amdgpu_amdkfd_suspend(adev, !fbcon);
> >
> > /* evict vram memory */
> >- amdgpu_bo_evict_vram(adev);
> >+ if (!adev->runpm_maco_en || !adev->in_runpm)
> >+ amdgpu_bo_evict_vram(adev);
> >
> > amdgpu_fence_driver_suspend(adev);
> >
> >@@ -3718,7 +3720,8 @@ int amdgpu_device_suspend(struct drm_device
> >*dev, bool fbcon)
> > * This second call to evict vram is to evict the gart page table
> > * using the CPU.
> > */
> >- amdgpu_bo_evict_vram(adev);
> >+ if (!adev->runpm_maco_en || !adev->in_runpm)
> >+ amdgpu_bo_evict_vram(adev);
> >
> > return 0;
> > }
> >diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c
> >b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c
> >index 523d22db094b..a198bc5d6a68 100644
> >--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c
> >+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c
> >@@ -397,10 +397,12 @@ static int psp_tmr_init(struct psp_context *psp)
> > }
> > }
> >
> >- pptr = amdgpu_sriov_vf(psp->adev) ? &tmr_buf : NULL;
> >- ret = amdgpu_bo_create_kernel(psp->adev, tmr_size, PSP_TMR_SIZE,
> >+ if (!psp->adev->runpm_maco_en || !psp->adev->in_runpm) {
> >+ pptr = amdgpu_sriov_vf(psp->adev) ? &tmr_buf : NULL;
> >+ ret = amdgpu_bo_create_kernel(psp->adev, tmr_size,
> >+ PSP_TMR_SIZE,
> > AMDGPU_GEM_DOMAIN_VRAM,
> > &psp->tmr_bo, &psp->tmr_mc_addr, pptr);
> >+ }
> >
> > return ret;
> > }
> >@@ -504,8 +506,10 @@ static int psp_tmr_terminate(struct psp_context
> >*psp)
> > return ret;
> >
> > /* free TMR memory buffer */
> >- pptr = amdgpu_sriov_vf(psp->adev) ? &tmr_buf : NULL;
> >- amdgpu_bo_free_kernel(&psp->tmr_bo, &psp->tmr_mc_addr, pptr);
> >+ if (!psp->adev->runpm_maco_en || !psp->adev->in_runpm) {
> >+ pptr = amdgpu_sriov_vf(psp->adev) ? &tmr_buf : NULL;
> >+ amdgpu_bo_free_kernel(&psp->tmr_bo, &psp->tmr_mc_addr,
> >pptr);
> >+ }
> >
> > return 0;
> > }
> >@@ -795,9 +799,10 @@ int psp_xgmi_terminate(struct psp_context *psp)
> > psp->xgmi_context.initialized = 0;
> >
> > /* free xgmi shared memory */
> >- amdgpu_bo_free_kernel(&psp->xgmi_context.xgmi_shared_bo,
> >- &psp->xgmi_context.xgmi_shared_mc_addr,
> >- &psp->xgmi_context.xgmi_shared_buf);
> >+ if (!psp->adev->runpm_maco_en || !psp->adev->in_runpm)
> >+ amdgpu_bo_free_kernel(&psp->xgmi_context.xgmi_shared_bo,
> >+ &psp->xgmi_context.xgmi_shared_mc_addr,
> >+ &psp->xgmi_context.xgmi_shared_buf);
> >
> > return 0;
> > }
> >@@ -812,7 +817,8 @@ int psp_xgmi_initialize(struct psp_context *psp)
> > !psp->adev->psp.ta_xgmi_start_addr)
> > return -ENOENT;
> >
> >- if (!psp->xgmi_context.initialized) {
> >+ if (!psp->xgmi_context.initialized &&
> >+ (!psp->adev->runpm_maco_en || !psp->adev->in_runpm)) {
> > ret = psp_xgmi_init_shared_buf(psp);
> > if (ret)
> > return ret;
> >@@ -1122,9 +1128,10 @@ static int psp_ras_terminate(struct psp_context
> >*psp)
> > psp->ras.ras_initialized = false;
> >
> > /* free ras shared memory */
> >- amdgpu_bo_free_kernel(&psp->ras.ras_shared_bo,
> >- &psp->ras.ras_shared_mc_addr,
> >- &psp->ras.ras_shared_buf);
> >+ if (!psp->adev->runpm_maco_en || !psp->adev->in_runpm)
> >+ amdgpu_bo_free_kernel(&psp->ras.ras_shared_bo,
> >+ &psp->ras.ras_shared_mc_addr,
> >+ &psp->ras.ras_shared_buf);
> >
> > return 0;
> > }
> >@@ -1145,7 +1152,8 @@ static int psp_ras_initialize(struct psp_context *psp)
> > return 0;
> > }
> >
> >- if (!psp->ras.ras_initialized) {
> >+ if (!psp->ras.ras_initialized &&
> >+ (!psp->adev->runpm_maco_en || !psp->adev->in_runpm)) {
> > ret = psp_ras_init_shared_buf(psp);
> > if (ret)
> > return ret;
> >@@ -1257,7 +1265,8 @@ static int psp_hdcp_initialize(struct psp_context
> >*psp)
> > return 0;
> > }
> >
> >- if (!psp->hdcp_context.hdcp_initialized) {
> >+ if (!psp->hdcp_context.hdcp_initialized &&
> >+ (!psp->adev->runpm_maco_en || !psp->adev->in_runpm)) {
> > ret = psp_hdcp_init_shared_buf(psp);
> > if (ret)
> > return ret;
> >@@ -1325,9 +1334,10 @@ static int psp_hdcp_terminate(struct psp_context
> >*psp)
> > psp->hdcp_context.hdcp_initialized = false;
> >
> > /* free hdcp shared memory */
> >- amdgpu_bo_free_kernel(&psp->hdcp_context.hdcp_shared_bo,
> >- &psp->hdcp_context.hdcp_shared_mc_addr,
> >- &psp->hdcp_context.hdcp_shared_buf);
> >+ if (!psp->adev->runpm_maco_en || !psp->adev->in_runpm)
> >+ amdgpu_bo_free_kernel(&psp->hdcp_context.hdcp_shared_bo,
> >+ &psp->hdcp_context.hdcp_shared_mc_addr,
> >+
> >+ &psp->hdcp_context.hdcp_shared_buf);
> >
> > return 0;
> > }
> >@@ -1404,7 +1414,8 @@ static int psp_dtm_initialize(struct psp_context
> >*psp)
> > return 0;
> > }
> >
> >- if (!psp->dtm_context.dtm_initialized) {
> >+ if (!psp->dtm_context.dtm_initialized &&
> >+ (!psp->adev->runpm_maco_en || !psp->adev->in_runpm)) {
> > ret = psp_dtm_init_shared_buf(psp);
> > if (ret)
> > return ret;
> >@@ -1472,9 +1483,10 @@ static int psp_dtm_terminate(struct psp_context
> >*psp)
> > psp->dtm_context.dtm_initialized = false;
> >
> > /* free hdcp shared memory */
> >- amdgpu_bo_free_kernel(&psp->dtm_context.dtm_shared_bo,
> >- &psp->dtm_context.dtm_shared_mc_addr,
> >- &psp->dtm_context.dtm_shared_buf);
> >+ if (!psp->adev->runpm_maco_en || !psp->adev->in_runpm)
> >+ amdgpu_bo_free_kernel(&psp->dtm_context.dtm_shared_bo,
> >+ &psp->dtm_context.dtm_shared_mc_addr,
> >+ &psp->dtm_context.dtm_shared_buf);
> >
> > return 0;
> > }
> >@@ -1563,7 +1575,8 @@ static int psp_rap_initialize(struct psp_context *psp)
> > return 0;
> > }
> >
> >- if (!psp->rap_context.rap_initialized) {
> >+ if (!psp->rap_context.rap_initialized &&
> >+ (!psp->adev->runpm_maco_en || !psp->adev->in_runpm)) {
> > ret = psp_rap_init_shared_buf(psp);
> > if (ret)
> > return ret;
> >@@ -1602,9 +1615,10 @@ static int psp_rap_terminate(struct psp_context
> >*psp)
> > psp->rap_context.rap_initialized = false;
> >
> > /* free rap shared memory */
> >- amdgpu_bo_free_kernel(&psp->rap_context.rap_shared_bo,
> >- &psp->rap_context.rap_shared_mc_addr,
> >- &psp->rap_context.rap_shared_buf);
> >+ if (!psp->adev->runpm_maco_en || !psp->adev->in_runpm)
> >+ amdgpu_bo_free_kernel(&psp->rap_context.rap_shared_bo,
> >+ &psp->rap_context.rap_shared_mc_addr,
> >+ &psp->rap_context.rap_shared_buf);
> >
> > return ret;
> > }
> >@@ -2261,10 +2275,12 @@ static int psp_resume(void *handle)
> >
> > DRM_INFO("PSP is resuming...\n");
> >
> >- ret = psp_mem_training(psp, PSP_MEM_TRAIN_RESUME);
> >- if (ret) {
> >- DRM_ERROR("Failed to process memory training!\n");
> >- return ret;
> >+ if (!psp->adev->runpm_maco_en || !psp->adev->in_runpm) {
> >+ ret = psp_mem_training(psp, PSP_MEM_TRAIN_RESUME);
> >+ if (ret) {
> >+ DRM_ERROR("Failed to process memory training!\n");
> >+ return ret;
> >+ }
> > }
> >
> > mutex_lock(&adev->firmware.mutex);
> >diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu11/sienna_cichlid_ppt.c
> >b/drivers/gpu/drm/amd/pm/swsmu/smu11/sienna_cichlid_ppt.c
> >index db0f2a476c23..fc8db9d69817 100644
> >--- a/drivers/gpu/drm/amd/pm/swsmu/smu11/sienna_cichlid_ppt.c
> >+++ b/drivers/gpu/drm/amd/pm/swsmu/smu11/sienna_cichlid_ppt.c
> >@@ -302,9 +302,12 @@ static int
> >sienna_cichlid_check_powerplay_table(struct smu_context *smu)
> > table_context->power_play_table;
> > struct smu_baco_context *smu_baco = &smu->smu_baco;
> >
> >- if (powerplay_table->platform_caps &
> >SMU_11_0_7_PP_PLATFORM_CAP_BACO ||
> >- powerplay_table->platform_caps &
> >SMU_11_0_7_PP_PLATFORM_CAP_MACO)
> >+ if (powerplay_table->platform_caps &
> >+ SMU_11_0_7_PP_PLATFORM_CAP_BACO) {
> > smu_baco->platform_support = true;
> >+ if ((powerplay_table->platform_caps &
> >SMU_11_0_7_PP_PLATFORM_CAP_MACO) &&
> >+ (amdgpu_runtime_pm == 2))
> >+ smu->adev->runpm_maco_en = true;
> >+ }
> >
>
> Consider BOCO path also in runtime suspend flow. Also, variable may be named more generic like mem_alive (or similar) for any other suspend/resume that could retain vram - S0ix.
>
> Thanks,
> Lijo
>
> > table_context->thermal_controller_type =
> > powerplay_table->thermal_controller_type;
> >--
> >2.25.1
> >
> >_______________________________________________
> >amd-gfx mailing list
> >amd-gfx at lists.freedesktop.org
> >https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.f
> >reedesktop.org%2Fmailman%2Flistinfo%2Famd-
> >gfx&data=04%7C01%7Clijo.lazar%40amd.com%7C5d3ad5208b1e412a016
> >808d8a0034a97%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C63743
> >5283286148631%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJ
> >QIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=VmVF
> >w6%2FGsbwRUzXHG26DUut7y8MceVib6M7t84mM7YQ%3D&reserved=0
> _______________________________________________
> amd-gfx mailing list
> amd-gfx at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
More information about the amd-gfx
mailing list