[PATCH] drm/amdgpu: refine pg code for gfx_v8.
Zhu, Rex
Rex.Zhu at amd.com
Wed Dec 7 12:00:29 UTC 2016
Hi Alex,
Patch 0001-0003 as you suggested.
Patch 0004: always initialize gfx pg for gfx_v8.0. Because if we disable pg by pg_mask, we can't enable pg in runtime.
Patch 0005: in some case(passthrough), if we enabled PG first, then we can't disable it eventhrough we set pg_mask=0 when reboot VM.
Patch 0006 -0007: refine set clockgating code for tonga/polars.
Best Regards
Rex
-----Original Message-----
From: Deucher, Alexander
Sent: Tuesday, December 06, 2016 5:07 AM
To: Zhu, Rex; Alex Deucher; StDenis, Tom
Cc: amd-gfx list
Subject: RE: [PATCH] drm/amdgpu: refine pg code for gfx_v8.
> -----Original Message-----
> From: amd-gfx [mailto:amd-gfx-bounces at lists.freedesktop.org] On Behalf
> Of Zhu, Rex
> Sent: Friday, December 02, 2016 7:25 AM
> To: Alex Deucher; StDenis, Tom
> Cc: amd-gfx list
> Subject: RE: [PATCH] drm/amdgpu: refine pg code for gfx_v8.
>
> Hi Alex, Tom,
>
> > 3. always init gfx pg.
>
> >>I don't think we want to do this. We need to be able to disable
> >>various pg
> features for debugging. Also some asics don't support gfx pg (e.g.,
> fiji), so we don't want to enable it on
> >>those asics.
>
> Rex: we need to call gfx_v8_0_init_pg no matter pg enable/disable.
> And delete repeated check for pg mask.
> For example:
> > - if (!(adev->pg_flags & AMD_PG_SUPPORT_GFX_PG))
> > - return 0;
> We can't disable power gate when pg_flags was set to 0.
I see your point. We just assumed the default state is disabled, but it may not be. That said, we should avoid touching this stuff on asics that do not support PG at all. We should only call init_pg() on asics which support PG in the first place. I'd suggest breaking this into 3 patches:
1. fix the save/restore list
2. fix enable_cp_power_gating
3. everything else
Alex
>
> Best Regards
> Rex
>
> -----Original Message-----
> From: Alex Deucher [mailto:alexdeucher at gmail.com]
> Sent: Thursday, December 01, 2016 11:44 PM
> To: Zhu, Rex
> Cc: amd-gfx list
> Subject: Re: [PATCH] drm/amdgpu: refine pg code for gfx_v8.
>
> On Thu, Dec 1, 2016 at 3:24 AM, Rex Zhu <Rex.Zhu at amd.com> wrote:
> > 1. bit CP_PG_DISABLE was reversed.
>
> Please split this out as a separate patch. Also, with this fixed, we
> should be able to enable the AMD_PG_SUPPORT_CP flag for CZ and ST.
>
> > 2. load RLC_SRM_INDEX_CNTL_ADDR/DATA_x pairs
> > with valid addr/data.
>
> Please split this out as a separate patch.
>
> > 3. always init gfx pg.
>
> I don't think we want to do this. We need to be able to disable
> various pg features for debugging. Also some asics don't support gfx
> pg (e.g., fiji), so we don't want to enable it on those asics.
>
> > 4. delete repeated check for pg mask.
> >
> > Change-Id: I9fcc8d1f79f5fa1803cb2625aa292188a656ae6b
> > Signed-off-by: Rex Zhu <Rex.Zhu at amd.com>
> > ---
> > drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 85
> > +++++++++++++++--------------------
> > 1 file changed, 37 insertions(+), 48 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
> > b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
> > index c02ae07..fee8c2a 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
> > @@ -3948,8 +3948,10 @@ static int
> > gfx_v8_0_init_save_restore_list(struct
> amdgpu_device *adev)
> > temp = mmRLC_SRM_INDEX_CNTL_ADDR_0;
> > data = mmRLC_SRM_INDEX_CNTL_DATA_0;
> > for (i = 0; i < sizeof(unique_indices) / sizeof(int); i++) {
> > - amdgpu_mm_wreg(adev, temp + i, unique_indices[i] & 0x3FFFF,
> false);
> > - amdgpu_mm_wreg(adev, data + i, unique_indices[i] >> 20, false);
> > + if (unique_indices[i] != 0) {
> > + amdgpu_mm_wreg(adev, temp + i,
> > + unique_indices[i] &
> 0x3FFFF, false);
> > + amdgpu_mm_wreg(adev, data + i,
> > + unique_indices[i] >> 20,
> false);
> > + }
> > }
> > kfree(register_list_format);
> >
> > @@ -3965,20 +3967,16 @@ static void
> > gfx_v8_0_init_power_gating(struct amdgpu_device *adev) {
> > uint32_t data;
> >
> > - if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
> > - AMD_PG_SUPPORT_GFX_SMG |
> > - AMD_PG_SUPPORT_GFX_DMG)) {
> > - WREG32_FIELD(CP_RB_WPTR_POLL_CNTL, IDLE_POLL_COUNT,
> 0x60);
> > + WREG32_FIELD(CP_RB_WPTR_POLL_CNTL, IDLE_POLL_COUNT, 0x60);
> >
> > - data = REG_SET_FIELD(0, RLC_PG_DELAY, POWER_UP_DELAY,
> 0x10);
> > - data = REG_SET_FIELD(data, RLC_PG_DELAY,
> POWER_DOWN_DELAY, 0x10);
> > - data = REG_SET_FIELD(data, RLC_PG_DELAY,
> CMD_PROPAGATE_DELAY, 0x10);
> > - data = REG_SET_FIELD(data, RLC_PG_DELAY, MEM_SLEEP_DELAY,
> 0x10);
> > - WREG32(mmRLC_PG_DELAY, data);
> > + data = REG_SET_FIELD(0, RLC_PG_DELAY, POWER_UP_DELAY, 0x10);
> > + data = REG_SET_FIELD(data, RLC_PG_DELAY, POWER_DOWN_DELAY,
> 0x10);
> > + data = REG_SET_FIELD(data, RLC_PG_DELAY,
> CMD_PROPAGATE_DELAY, 0x10);
> > + data = REG_SET_FIELD(data, RLC_PG_DELAY, MEM_SLEEP_DELAY,
> 0x10);
> > + WREG32(mmRLC_PG_DELAY, data);
> >
> > - WREG32_FIELD(RLC_PG_DELAY_2, SERDES_CMD_DELAY, 0x3);
> > - WREG32_FIELD(RLC_AUTO_PG_CTRL,
> GRBM_REG_SAVE_GFX_IDLE_THRESHOLD, 0x55f0);
> > - }
> > + WREG32_FIELD(RLC_PG_DELAY_2, SERDES_CMD_DELAY, 0x3);
> > + WREG32_FIELD(RLC_AUTO_PG_CTRL,
> > + GRBM_REG_SAVE_GFX_IDLE_THRESHOLD, 0x55f0);
> > }
> >
> > static void cz_enable_sck_slow_down_on_power_up(struct
> amdgpu_device
> > *adev, @@ -3995,41 +3993,35 @@ static void
> > cz_enable_sck_slow_down_on_power_down(struct amdgpu_device
> *adev,
> >
> > static void cz_enable_cp_power_gating(struct amdgpu_device *adev,
> > bool enable) {
> > - WREG32_FIELD(RLC_PG_CNTL, CP_PG_DISABLE, enable ? 1 : 0);
> > + WREG32_FIELD(RLC_PG_CNTL, CP_PG_DISABLE, enable ? 0 : 1);
> > }
> >
> > static void gfx_v8_0_init_pg(struct amdgpu_device *adev) {
> > - if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
> > - AMD_PG_SUPPORT_GFX_SMG |
> > - AMD_PG_SUPPORT_GFX_DMG |
> > - AMD_PG_SUPPORT_CP |
> > - AMD_PG_SUPPORT_GDS |
> > - AMD_PG_SUPPORT_RLC_SMU_HS)) {
> > - gfx_v8_0_init_csb(adev);
> > - gfx_v8_0_init_save_restore_list(adev);
> > - gfx_v8_0_enable_save_restore_machine(adev);
> > -
> > - if ((adev->asic_type == CHIP_CARRIZO) ||
> > - (adev->asic_type == CHIP_STONEY)) {
> > - WREG32(mmRLC_JUMP_TABLE_RESTORE, adev-
> >gfx.rlc.cp_table_gpu_addr >> 8);
> > - gfx_v8_0_init_power_gating(adev);
> > - WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev-
> >gfx.cu_info.ao_cu_mask);
> > - if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) {
> > - cz_enable_sck_slow_down_on_power_up(adev, true);
> > - cz_enable_sck_slow_down_on_power_down(adev, true);
> > - } else {
> > - cz_enable_sck_slow_down_on_power_up(adev, false);
> > - cz_enable_sck_slow_down_on_power_down(adev,
> false);
> > - }
> > - if (adev->pg_flags & AMD_PG_SUPPORT_CP)
> > - cz_enable_cp_power_gating(adev, true);
> > - else
> > - cz_enable_cp_power_gating(adev, false);
> > - } else if (adev->asic_type == CHIP_POLARIS11) {
> > - gfx_v8_0_init_power_gating(adev);
> > + gfx_v8_0_init_csb(adev);
> > + gfx_v8_0_init_save_restore_list(adev);
> > + gfx_v8_0_enable_save_restore_machine(adev);
> > +
> > + if ((adev->asic_type == CHIP_CARRIZO) ||
> > + (adev->asic_type == CHIP_STONEY)) {
> > + WREG32(mmRLC_JUMP_TABLE_RESTORE, adev-
> >gfx.rlc.cp_table_gpu_addr >> 8);
> > + gfx_v8_0_init_power_gating(adev);
> > + WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev-
> >gfx.cu_info.ao_cu_mask);
> > + if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) {
> > + cz_enable_sck_slow_down_on_power_up(adev, true);
> > + cz_enable_sck_slow_down_on_power_down(adev, true);
> > + } else {
> > + cz_enable_sck_slow_down_on_power_up(adev, false);
> > + cz_enable_sck_slow_down_on_power_down(adev,
> > + false);
> > }
> > + if (adev->pg_flags & AMD_PG_SUPPORT_CP)
> > + cz_enable_cp_power_gating(adev, true);
> > + else
> > + cz_enable_cp_power_gating(adev, false);
> > + } else if (adev->asic_type == CHIP_POLARIS11) {
> > + gfx_v8_0_init_power_gating(adev);
> > }
> > +
> > }
> >
> > static void gfx_v8_0_rlc_stop(struct amdgpu_device *adev) @@
> > -5570,14
> > +5562,11 @@ static int gfx_v8_0_set_powergating_state(void *handle,
> > struct amdgpu_device *adev = (struct amdgpu_device *)handle;
> > bool enable = (state == AMD_PG_STATE_GATE) ? true : false;
> >
> > - if (!(adev->pg_flags & AMD_PG_SUPPORT_GFX_PG))
> > - return 0;
> > -
> > switch (adev->asic_type) {
> > case CHIP_CARRIZO:
> > case CHIP_STONEY:
> > - if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG)
> > - cz_update_gfx_cg_power_gating(adev, enable);
> > +
> > + cz_update_gfx_cg_power_gating(adev, enable);
> >
> > if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) &&
> > enable)
> >
> > gfx_v8_0_enable_gfx_static_mg_power_gating(adev, true);
> > --
> > 1.9.1
> >
> > _______________________________________________
> > amd-gfx mailing list
> > amd-gfx at lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/amd-gfx
> _______________________________________________
> amd-gfx mailing list
> amd-gfx at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-drm-amdgpu-fix-enable_cp_power_gating-in-gfx_v8.0.patch
Type: application/octet-stream
Size: 1027 bytes
Desc: 0001-drm-amdgpu-fix-enable_cp_power_gating-in-gfx_v8.0.patch
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20161207/783d3c8d/attachment-0007.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0002-drm-amdgpu-fix-init-save-restore-list-in-gfx_v8.0.patch
Type: application/octet-stream
Size: 1284 bytes
Desc: 0002-drm-amdgpu-fix-init-save-restore-list-in-gfx_v8.0.patch
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20161207/783d3c8d/attachment-0008.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0003-drm-amdgpu-enable-AMD_PG_SUPPORT_CP-in-Carrizo-Stone.patch
Type: application/octet-stream
Size: 1138 bytes
Desc: 0003-drm-amdgpu-enable-AMD_PG_SUPPORT_CP-in-Carrizo-Stone.patch
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20161207/783d3c8d/attachment-0009.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0004-drm-amdgpu-always-initialize-gfx-pg-for-gfx_v8.0.patch
Type: application/octet-stream
Size: 4299 bytes
Desc: 0004-drm-amdgpu-always-initialize-gfx-pg-for-gfx_v8.0.patch
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20161207/783d3c8d/attachment-0010.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0005-drm-amdgpu-fix-pg-can-t-be-disabled-by-PG-mask.patch
Type: application/octet-stream
Size: 1238 bytes
Desc: 0005-drm-amdgpu-fix-pg-can-t-be-disabled-by-PG-mask.patch
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20161207/783d3c8d/attachment-0011.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0006-drm-amdgpu-initialize-cg-flags-for-tonga-polaris10-p.patch
Type: application/octet-stream
Size: 3464 bytes
Desc: 0006-drm-amdgpu-initialize-cg-flags-for-tonga-polaris10-p.patch
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20161207/783d3c8d/attachment-0012.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0007-drm-amdgpu-refine-set-clock-gating-for-tonga-polars.patch
Type: application/octet-stream
Size: 11284 bytes
Desc: 0007-drm-amdgpu-refine-set-clock-gating-for-tonga-polars.patch
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20161207/783d3c8d/attachment-0013.obj>
More information about the amd-gfx
mailing list