* [PATCH] drm/amdgpu: Add a new module param to disable d3cold @ 2023-11-29 8:51 Ma Jun 2023-11-29 15:35 ` Mario Limonciello 2023-11-29 16:39 ` Alex Deucher 0 siblings, 2 replies; 10+ messages in thread From: Ma Jun @ 2023-11-29 8:51 UTC (permalink / raw) To: amd-gfx Cc: kevinyang.wang, Ma Jun, mario.limonciello, Alexander.Deucher, Kenneth.Feng Some platforms can't resume from d3cold state, So add a new module parameter to disable d3cold state for debugging purpose or workaround. Signed-off-by: Ma Jun <Jun.Ma2@amd.com> --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 8 ++++++++ 3 files changed, 16 insertions(+) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index a9f54df9d33e..db9f60790267 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -166,6 +166,7 @@ extern char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH]; extern int amdgpu_dpm; extern int amdgpu_fw_load_type; extern int amdgpu_aspm; +extern int amdgpu_d3cold; extern int amdgpu_runtime_pm; extern uint amdgpu_ip_block_mask; extern int amdgpu_bapm; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 22b6a910b7f2..90501c44e7d0 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -264,6 +264,13 @@ bool amdgpu_device_supports_px(struct drm_device *dev) bool amdgpu_device_supports_boco(struct drm_device *dev) { struct amdgpu_device *adev = drm_to_adev(dev); + struct pci_dev *parent; + + if (!amdgpu_d3cold) { + parent = pcie_find_root_port(adev->pdev); + pci_d3cold_disable(parent); + return false; + } if (adev->has_pr3 || ((adev->flags & AMD_IS_PX) && amdgpu_is_atpx_hybrid())) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index 5f14f04cb553..c9fbb8bd4169 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -145,6 +145,7 @@ char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH]; int amdgpu_dpm = -1; int amdgpu_fw_load_type = -1; int amdgpu_aspm = -1; +int amdgpu_d3cold = -1; int amdgpu_runtime_pm = -1; uint amdgpu_ip_block_mask = 0xffffffff; int amdgpu_bapm = -1; @@ -359,6 +360,13 @@ module_param_named(fw_load_type, amdgpu_fw_load_type, int, 0444); MODULE_PARM_DESC(aspm, "ASPM support (1 = enable, 0 = disable, -1 = auto)"); module_param_named(aspm, amdgpu_aspm, int, 0444); +/** + * DOC: d3cold (int) + * To disable d3cold (1 = enable, 0 = disable). The default is -1 (auto, enabled). + */ +MODULE_PARM_DESC(d3cold, "d3cold support (1 = enable, 0 = disable, -1 = auto)"); +module_param_named(d3cold, amdgpu_d3cold, int, 0444); + /** * DOC: runpm (int) * Override for runtime power management control for dGPUs. The amdgpu driver can dynamically power down -- 2.34.1 ^ permalink raw reply related [flat|nested] 10+ messages in thread
* Re: [PATCH] drm/amdgpu: Add a new module param to disable d3cold 2023-11-29 8:51 [PATCH] drm/amdgpu: Add a new module param to disable d3cold Ma Jun @ 2023-11-29 15:35 ` Mario Limonciello 2023-11-29 18:40 ` Alex Deucher 2023-11-30 9:46 ` Ma, Jun 2023-11-29 16:39 ` Alex Deucher 1 sibling, 2 replies; 10+ messages in thread From: Mario Limonciello @ 2023-11-29 15:35 UTC (permalink / raw) To: Ma Jun, amd-gfx; +Cc: Alexander.Deucher, Kenneth.Feng, kevinyang.wang On 11/29/2023 02:51, Ma Jun wrote: > Some platforms can't resume from d3cold state, So add a > new module parameter to disable d3cold state for debugging > purpose or workaround. > > Signed-off-by: Ma Jun <Jun.Ma2@amd.com> > --- This patch is essentially an 'amdgpu knob' for d3cold on the root port. At least for debugging purposes we also have a sysfs file 'd3cold_allowed' that will enact the same behavior. I do have a patch that I proposed to PCI core that stops d3cold_allowed from working in favor of requesting pcie_port_pm=off to be used instead for debugging purposes. However that's a 'relatively big' debugging knob however as it will apply to all PCIe root ports. Considering above I'm in favor of this being available as a localized debugging path for just the root port the dGPU is connected to. Some comments below though: > drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + > drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +++++++ > drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 8 ++++++++ > 3 files changed, 16 insertions(+) > > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h > index a9f54df9d33e..db9f60790267 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h > @@ -166,6 +166,7 @@ extern char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH]; > extern int amdgpu_dpm; > extern int amdgpu_fw_load_type; > extern int amdgpu_aspm; > +extern int amdgpu_d3cold; > extern int amdgpu_runtime_pm; > extern uint amdgpu_ip_block_mask; > extern int amdgpu_bapm; > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c > index 22b6a910b7f2..90501c44e7d0 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c > @@ -264,6 +264,13 @@ bool amdgpu_device_supports_px(struct drm_device *dev) > bool amdgpu_device_supports_boco(struct drm_device *dev) > { > struct amdgpu_device *adev = drm_to_adev(dev); > + struct pci_dev *parent; > + > + if (!amdgpu_d3cold) { > + parent = pcie_find_root_port(adev->pdev); > + pci_d3cold_disable(parent); > + return false; > + } > > if (adev->has_pr3 || > ((adev->flags & AMD_IS_PX) && amdgpu_is_atpx_hybrid())) > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c > index 5f14f04cb553..c9fbb8bd4169 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c > @@ -145,6 +145,7 @@ char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH]; > int amdgpu_dpm = -1; > int amdgpu_fw_load_type = -1; > int amdgpu_aspm = -1; > +int amdgpu_d3cold = -1; If this was chained to a larger workaround (such as automatically applying to a DMI quirk) it would make sense as int and with using -1 for auto. However there is a pretty dramatic downside for using this knob that it can break s2idle. In my testing I've found that the following happens on an A+A design after s2idle with this parameter in use. [ 70.572270] pcieport 0000:01:00.0: Unable to change power state from D3cold to D0, device inaccessible [ 70.572481] pcieport 0000:02:00.0: Unable to change power state from D3cold to D0, device inaccessible [ 72.855769] amdgpu 0000:03:00.0: not ready 1023ms after resume; waiting [ 73.943545] amdgpu 0000:03:00.0: not ready 2047ms after resume; waiting [ 76.055602] amdgpu 0000:03:00.0: not ready 4095ms after resume; waiting [ 80.279550] amdgpu 0000:03:00.0: not ready 8191ms after resume; waiting [ 88.983562] amdgpu 0000:03:00.0: not ready 16383ms after resume; waiting [ 105.879581] amdgpu 0000:03:00.0: not ready 32767ms after resume; waiting [ 142.743646] amdgpu 0000:03:00.0: not ready 65535ms after resume; giving up [ 142.743793] amdgpu 0000:03:00.0: Unable to change power state from D3cold to D0, device inaccessible [ 142.804011] snd_hda_intel 0000:03:00.1: Unable to change power state from D3cold to D0, device inaccessible So I don't see us ever automatically using this and it should be debugging only. IOW this doesn't need to be integer; it can be boolean. > int amdgpu_runtime_pm = -1; > uint amdgpu_ip_block_mask = 0xffffffff; > int amdgpu_bapm = -1; > @@ -359,6 +360,13 @@ module_param_named(fw_load_type, amdgpu_fw_load_type, int, 0444); > MODULE_PARM_DESC(aspm, "ASPM support (1 = enable, 0 = disable, -1 = auto)"); > module_param_named(aspm, amdgpu_aspm, int, 0444); > > +/** > + * DOC: d3cold (int) If you flip it to boolean as I suggested this should probably either rename to disable_d3cold or you should default to TRUE. > + * To disable d3cold (1 = enable, 0 = disable). The default is -1 (auto, enabled). > + */ > +MODULE_PARM_DESC(d3cold, "d3cold support (1 = enable, 0 = disable, -1 = auto)"); > +module_param_named(d3cold, amdgpu_d3cold, int, 0444); > + > /** > * DOC: runpm (int) > * Override for runtime power management control for dGPUs. The amdgpu driver can dynamically power down ^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [PATCH] drm/amdgpu: Add a new module param to disable d3cold 2023-11-29 15:35 ` Mario Limonciello @ 2023-11-29 18:40 ` Alex Deucher 2023-11-30 9:46 ` Ma, Jun 1 sibling, 0 replies; 10+ messages in thread From: Alex Deucher @ 2023-11-29 18:40 UTC (permalink / raw) To: Mario Limonciello Cc: Alexander.Deucher, Ma Jun, Kenneth.Feng, kevinyang.wang, amd-gfx On Wed, Nov 29, 2023 at 1:02 PM Mario Limonciello <mario.limonciello@amd.com> wrote: > > On 11/29/2023 02:51, Ma Jun wrote: > > Some platforms can't resume from d3cold state, So add a > > new module parameter to disable d3cold state for debugging > > purpose or workaround. > > > > Signed-off-by: Ma Jun <Jun.Ma2@amd.com> > > --- > > This patch is essentially an 'amdgpu knob' for d3cold on the root port. > At least for debugging purposes we also have a sysfs file > 'd3cold_allowed' that will enact the same behavior. > > I do have a patch that I proposed to PCI core that stops d3cold_allowed > from working in favor of requesting pcie_port_pm=off to be used instead > for debugging purposes. > > However that's a 'relatively big' debugging knob however as it will > apply to all PCIe root ports. > > Considering above I'm in favor of this being available as a localized > debugging path for just the root port the dGPU is connected to. What functionality does this option provide that runpm=0 does not? AFAIK, the pci core should not enter d3cold at runtime if the driver doesn't call pm_runtime_allow(). Alex > > Some comments below though: > > > drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + > > drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +++++++ > > drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 8 ++++++++ > > 3 files changed, 16 insertions(+) > > > > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h > > index a9f54df9d33e..db9f60790267 100644 > > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h > > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h > > @@ -166,6 +166,7 @@ extern char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH]; > > extern int amdgpu_dpm; > > extern int amdgpu_fw_load_type; > > extern int amdgpu_aspm; > > +extern int amdgpu_d3cold; > > extern int amdgpu_runtime_pm; > > extern uint amdgpu_ip_block_mask; > > extern int amdgpu_bapm; > > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c > > index 22b6a910b7f2..90501c44e7d0 100644 > > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c > > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c > > @@ -264,6 +264,13 @@ bool amdgpu_device_supports_px(struct drm_device *dev) > > bool amdgpu_device_supports_boco(struct drm_device *dev) > > { > > struct amdgpu_device *adev = drm_to_adev(dev); > > + struct pci_dev *parent; > > + > > + if (!amdgpu_d3cold) { > > + parent = pcie_find_root_port(adev->pdev); > > + pci_d3cold_disable(parent); > > + return false; > > + } > > > > if (adev->has_pr3 || > > ((adev->flags & AMD_IS_PX) && amdgpu_is_atpx_hybrid())) > > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c > > index 5f14f04cb553..c9fbb8bd4169 100644 > > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c > > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c > > @@ -145,6 +145,7 @@ char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH]; > > int amdgpu_dpm = -1; > > int amdgpu_fw_load_type = -1; > > int amdgpu_aspm = -1; > > +int amdgpu_d3cold = -1; > > If this was chained to a larger workaround (such as automatically > applying to a DMI quirk) it would make sense as int and with using > -1 for auto. However there is a pretty dramatic downside for using this > knob that it can break s2idle. > > In my testing I've found that the following happens on an A+A design > after s2idle with this parameter in use. > > [ 70.572270] pcieport 0000:01:00.0: Unable to change power state from > D3cold to D0, device inaccessible > [ 70.572481] pcieport 0000:02:00.0: Unable to change power state from > D3cold to D0, device inaccessible > [ 72.855769] amdgpu 0000:03:00.0: not ready 1023ms after resume; waiting > [ 73.943545] amdgpu 0000:03:00.0: not ready 2047ms after resume; waiting > [ 76.055602] amdgpu 0000:03:00.0: not ready 4095ms after resume; waiting > [ 80.279550] amdgpu 0000:03:00.0: not ready 8191ms after resume; waiting > [ 88.983562] amdgpu 0000:03:00.0: not ready 16383ms after resume; waiting > [ 105.879581] amdgpu 0000:03:00.0: not ready 32767ms after resume; waiting > [ 142.743646] amdgpu 0000:03:00.0: not ready 65535ms after resume; > giving up > [ 142.743793] amdgpu 0000:03:00.0: Unable to change power state from > D3cold to D0, device inaccessible > [ 142.804011] snd_hda_intel 0000:03:00.1: Unable to change power state > from D3cold to D0, device inaccessible > > So I don't see us ever automatically using this and it should be > debugging only. IOW this doesn't need to be integer; it can be boolean. > > > int amdgpu_runtime_pm = -1; > > uint amdgpu_ip_block_mask = 0xffffffff; > > int amdgpu_bapm = -1; > > @@ -359,6 +360,13 @@ module_param_named(fw_load_type, amdgpu_fw_load_type, int, 0444); > > MODULE_PARM_DESC(aspm, "ASPM support (1 = enable, 0 = disable, -1 = auto)"); > > module_param_named(aspm, amdgpu_aspm, int, 0444); > > > > +/** > > + * DOC: d3cold (int) > > If you flip it to boolean as I suggested this should probably either > rename to disable_d3cold or you should default to TRUE. > > > + * To disable d3cold (1 = enable, 0 = disable). The default is -1 (auto, enabled). > > + */ > > +MODULE_PARM_DESC(d3cold, "d3cold support (1 = enable, 0 = disable, -1 = auto)"); > > +module_param_named(d3cold, amdgpu_d3cold, int, 0444); > > + > > /** > > * DOC: runpm (int) > > * Override for runtime power management control for dGPUs. The amdgpu driver can dynamically power down > ^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [PATCH] drm/amdgpu: Add a new module param to disable d3cold 2023-11-29 15:35 ` Mario Limonciello 2023-11-29 18:40 ` Alex Deucher @ 2023-11-30 9:46 ` Ma, Jun 1 sibling, 0 replies; 10+ messages in thread From: Ma, Jun @ 2023-11-30 9:46 UTC (permalink / raw) To: Mario Limonciello, Ma Jun, amd-gfx Cc: Alexander.Deucher, Kenneth.Feng, kevinyang.wang Hi Mario, On 11/29/2023 11:35 PM, Mario Limonciello wrote: > On 11/29/2023 02:51, Ma Jun wrote: >> Some platforms can't resume from d3cold state, So add a >> new module parameter to disable d3cold state for debugging >> purpose or workaround. >> >> Signed-off-by: Ma Jun <Jun.Ma2@amd.com> >> --- > > This patch is essentially an 'amdgpu knob' for d3cold on the root port. > At least for debugging purposes we also have a sysfs file > 'd3cold_allowed' that will enact the same behavior. > There is difference here. In addition to disabling the d3cold state, BOCO is also disabled in this patch. Otherwise,there is a scenario where the driver uses boco and the root port uses d3hot. It may cause some unexpected errors. Regards, Ma Jun > I do have a patch that I proposed to PCI core that stops d3cold_allowed > from working in favor of requesting pcie_port_pm=off to be used instead > for debugging purposes. > > However that's a 'relatively big' debugging knob however as it will > apply to all PCIe root ports. > > Considering above I'm in favor of this being available as a localized > debugging path for just the root port the dGPU is connected to. > > Some comments below though: > >> drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + >> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +++++++ >> drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 8 ++++++++ >> 3 files changed, 16 insertions(+) >> >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h >> index a9f54df9d33e..db9f60790267 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h >> @@ -166,6 +166,7 @@ extern char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH]; >> extern int amdgpu_dpm; >> extern int amdgpu_fw_load_type; >> extern int amdgpu_aspm; >> +extern int amdgpu_d3cold; >> extern int amdgpu_runtime_pm; >> extern uint amdgpu_ip_block_mask; >> extern int amdgpu_bapm; >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c >> index 22b6a910b7f2..90501c44e7d0 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c >> @@ -264,6 +264,13 @@ bool amdgpu_device_supports_px(struct drm_device *dev) >> bool amdgpu_device_supports_boco(struct drm_device *dev) >> { >> struct amdgpu_device *adev = drm_to_adev(dev); >> + struct pci_dev *parent; >> + >> + if (!amdgpu_d3cold) { >> + parent = pcie_find_root_port(adev->pdev); >> + pci_d3cold_disable(parent); >> + return false; >> + } >> >> if (adev->has_pr3 || >> ((adev->flags & AMD_IS_PX) && amdgpu_is_atpx_hybrid())) >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c >> index 5f14f04cb553..c9fbb8bd4169 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c >> @@ -145,6 +145,7 @@ char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH]; >> int amdgpu_dpm = -1; >> int amdgpu_fw_load_type = -1; >> int amdgpu_aspm = -1; >> +int amdgpu_d3cold = -1; > > If this was chained to a larger workaround (such as automatically > applying to a DMI quirk) it would make sense as int and with using > -1 for auto. However there is a pretty dramatic downside for using this > knob that it can break s2idle. > > In my testing I've found that the following happens on an A+A design > after s2idle with this parameter in use. > > [ 70.572270] pcieport 0000:01:00.0: Unable to change power state from > D3cold to D0, device inaccessible > [ 70.572481] pcieport 0000:02:00.0: Unable to change power state from > D3cold to D0, device inaccessible > [ 72.855769] amdgpu 0000:03:00.0: not ready 1023ms after resume; waiting > [ 73.943545] amdgpu 0000:03:00.0: not ready 2047ms after resume; waiting > [ 76.055602] amdgpu 0000:03:00.0: not ready 4095ms after resume; waiting > [ 80.279550] amdgpu 0000:03:00.0: not ready 8191ms after resume; waiting > [ 88.983562] amdgpu 0000:03:00.0: not ready 16383ms after resume; waiting > [ 105.879581] amdgpu 0000:03:00.0: not ready 32767ms after resume; waiting > [ 142.743646] amdgpu 0000:03:00.0: not ready 65535ms after resume; > giving up > [ 142.743793] amdgpu 0000:03:00.0: Unable to change power state from > D3cold to D0, device inaccessible > [ 142.804011] snd_hda_intel 0000:03:00.1: Unable to change power state > from D3cold to D0, device inaccessible > > So I don't see us ever automatically using this and it should be > debugging only. IOW this doesn't need to be integer; it can be boolean. > >> int amdgpu_runtime_pm = -1; >> uint amdgpu_ip_block_mask = 0xffffffff; >> int amdgpu_bapm = -1; >> @@ -359,6 +360,13 @@ module_param_named(fw_load_type, amdgpu_fw_load_type, int, 0444); >> MODULE_PARM_DESC(aspm, "ASPM support (1 = enable, 0 = disable, -1 = auto)"); >> module_param_named(aspm, amdgpu_aspm, int, 0444); >> >> +/** >> + * DOC: d3cold (int) > > If you flip it to boolean as I suggested this should probably either > rename to disable_d3cold or you should default to TRUE. > >> + * To disable d3cold (1 = enable, 0 = disable). The default is -1 (auto, enabled). >> + */ >> +MODULE_PARM_DESC(d3cold, "d3cold support (1 = enable, 0 = disable, -1 = auto)"); >> +module_param_named(d3cold, amdgpu_d3cold, int, 0444); >> + >> /** >> * DOC: runpm (int) >> * Override for runtime power management control for dGPUs. The amdgpu driver can dynamically power down > ^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [PATCH] drm/amdgpu: Add a new module param to disable d3cold 2023-11-29 8:51 [PATCH] drm/amdgpu: Add a new module param to disable d3cold Ma Jun 2023-11-29 15:35 ` Mario Limonciello @ 2023-11-29 16:39 ` Alex Deucher 2023-11-30 6:29 ` Ma, Jun 1 sibling, 1 reply; 10+ messages in thread From: Alex Deucher @ 2023-11-29 16:39 UTC (permalink / raw) To: Ma Jun Cc: Alexander.Deucher, Kenneth.Feng, kevinyang.wang, amd-gfx, mario.limonciello On Wed, Nov 29, 2023 at 11:37 AM Ma Jun <Jun.Ma2@amd.com> wrote: > > Some platforms can't resume from d3cold state, So add a > new module parameter to disable d3cold state for debugging > purpose or workaround. Doesn't the runpm parameter already handle this? If you set runpm=0, that should disable d3cold. Alex > > Signed-off-by: Ma Jun <Jun.Ma2@amd.com> > --- > drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + > drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +++++++ > drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 8 ++++++++ > 3 files changed, 16 insertions(+) > > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h > index a9f54df9d33e..db9f60790267 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h > @@ -166,6 +166,7 @@ extern char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH]; > extern int amdgpu_dpm; > extern int amdgpu_fw_load_type; > extern int amdgpu_aspm; > +extern int amdgpu_d3cold; > extern int amdgpu_runtime_pm; > extern uint amdgpu_ip_block_mask; > extern int amdgpu_bapm; > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c > index 22b6a910b7f2..90501c44e7d0 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c > @@ -264,6 +264,13 @@ bool amdgpu_device_supports_px(struct drm_device *dev) > bool amdgpu_device_supports_boco(struct drm_device *dev) > { > struct amdgpu_device *adev = drm_to_adev(dev); > + struct pci_dev *parent; > + > + if (!amdgpu_d3cold) { > + parent = pcie_find_root_port(adev->pdev); > + pci_d3cold_disable(parent); > + return false; > + } > > if (adev->has_pr3 || > ((adev->flags & AMD_IS_PX) && amdgpu_is_atpx_hybrid())) > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c > index 5f14f04cb553..c9fbb8bd4169 100644 > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c > @@ -145,6 +145,7 @@ char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH]; > int amdgpu_dpm = -1; > int amdgpu_fw_load_type = -1; > int amdgpu_aspm = -1; > +int amdgpu_d3cold = -1; > int amdgpu_runtime_pm = -1; > uint amdgpu_ip_block_mask = 0xffffffff; > int amdgpu_bapm = -1; > @@ -359,6 +360,13 @@ module_param_named(fw_load_type, amdgpu_fw_load_type, int, 0444); > MODULE_PARM_DESC(aspm, "ASPM support (1 = enable, 0 = disable, -1 = auto)"); > module_param_named(aspm, amdgpu_aspm, int, 0444); > > +/** > + * DOC: d3cold (int) > + * To disable d3cold (1 = enable, 0 = disable). The default is -1 (auto, enabled). > + */ > +MODULE_PARM_DESC(d3cold, "d3cold support (1 = enable, 0 = disable, -1 = auto)"); > +module_param_named(d3cold, amdgpu_d3cold, int, 0444); > + > /** > * DOC: runpm (int) > * Override for runtime power management control for dGPUs. The amdgpu driver can dynamically power down > -- > 2.34.1 > ^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [PATCH] drm/amdgpu: Add a new module param to disable d3cold 2023-11-29 16:39 ` Alex Deucher @ 2023-11-30 6:29 ` Ma, Jun 2023-11-30 9:18 ` Lazar, Lijo 2023-11-30 14:21 ` Alex Deucher 0 siblings, 2 replies; 10+ messages in thread From: Ma, Jun @ 2023-11-30 6:29 UTC (permalink / raw) To: Alex Deucher, Ma Jun Cc: kevinyang.wang, amd-gfx, mario.limonciello, Alexander.Deucher, Kenneth.Feng Hi Alex, On 11/30/2023 12:39 AM, Alex Deucher wrote: > On Wed, Nov 29, 2023 at 11:37 AM Ma Jun <Jun.Ma2@amd.com> wrote: >> >> Some platforms can't resume from d3cold state, So add a >> new module parameter to disable d3cold state for debugging >> purpose or workaround. > > Doesn't the runpm parameter already handle this? If you set runpm=0, > that should disable d3cold. > runpm=0 prevents calls to driver runtime_suspend/resume functions. While d3cold=0 allows calls to runtime_suspend/resume functions and puts the device in d3hot state instead of d3cold. Regards, Ma Jun > Alex > >> >> Signed-off-by: Ma Jun <Jun.Ma2@amd.com> >> --- >> drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + >> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +++++++ >> drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 8 ++++++++ >> 3 files changed, 16 insertions(+) >> >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h >> index a9f54df9d33e..db9f60790267 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h >> @@ -166,6 +166,7 @@ extern char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH]; >> extern int amdgpu_dpm; >> extern int amdgpu_fw_load_type; >> extern int amdgpu_aspm; >> +extern int amdgpu_d3cold; >> extern int amdgpu_runtime_pm; >> extern uint amdgpu_ip_block_mask; >> extern int amdgpu_bapm; >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c >> index 22b6a910b7f2..90501c44e7d0 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c >> @@ -264,6 +264,13 @@ bool amdgpu_device_supports_px(struct drm_device *dev) >> bool amdgpu_device_supports_boco(struct drm_device *dev) >> { >> struct amdgpu_device *adev = drm_to_adev(dev); >> + struct pci_dev *parent; >> + >> + if (!amdgpu_d3cold) { >> + parent = pcie_find_root_port(adev->pdev); >> + pci_d3cold_disable(parent); >> + return false; >> + } >> >> if (adev->has_pr3 || >> ((adev->flags & AMD_IS_PX) && amdgpu_is_atpx_hybrid())) >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c >> index 5f14f04cb553..c9fbb8bd4169 100644 >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c >> @@ -145,6 +145,7 @@ char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH]; >> int amdgpu_dpm = -1; >> int amdgpu_fw_load_type = -1; >> int amdgpu_aspm = -1; >> +int amdgpu_d3cold = -1; >> int amdgpu_runtime_pm = -1; >> uint amdgpu_ip_block_mask = 0xffffffff; >> int amdgpu_bapm = -1; >> @@ -359,6 +360,13 @@ module_param_named(fw_load_type, amdgpu_fw_load_type, int, 0444); >> MODULE_PARM_DESC(aspm, "ASPM support (1 = enable, 0 = disable, -1 = auto)"); >> module_param_named(aspm, amdgpu_aspm, int, 0444); >> >> +/** >> + * DOC: d3cold (int) >> + * To disable d3cold (1 = enable, 0 = disable). The default is -1 (auto, enabled). >> + */ >> +MODULE_PARM_DESC(d3cold, "d3cold support (1 = enable, 0 = disable, -1 = auto)"); >> +module_param_named(d3cold, amdgpu_d3cold, int, 0444); >> + >> /** >> * DOC: runpm (int) >> * Override for runtime power management control for dGPUs. The amdgpu driver can dynamically power down >> -- >> 2.34.1 >> ^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [PATCH] drm/amdgpu: Add a new module param to disable d3cold 2023-11-30 6:29 ` Ma, Jun @ 2023-11-30 9:18 ` Lazar, Lijo 2023-11-30 10:47 ` Ma, Jun 2023-11-30 14:21 ` Alex Deucher 1 sibling, 1 reply; 10+ messages in thread From: Lazar, Lijo @ 2023-11-30 9:18 UTC (permalink / raw) To: Ma, Jun, Alex Deucher, Ma Jun Cc: Alexander.Deucher, Kenneth.Feng, amd-gfx, kevinyang.wang, mario.limonciello On 11/30/2023 11:59 AM, Ma, Jun wrote: > Hi Alex, > > On 11/30/2023 12:39 AM, Alex Deucher wrote: >> On Wed, Nov 29, 2023 at 11:37 AM Ma Jun <Jun.Ma2@amd.com> wrote: >>> >>> Some platforms can't resume from d3cold state, So add a >>> new module parameter to disable d3cold state for debugging >>> purpose or workaround. >> >> Doesn't the runpm parameter already handle this? If you set runpm=0, >> that should disable d3cold. >> > runpm=0 prevents calls to driver runtime_suspend/resume functions. > While d3cold=0 allows calls to runtime_suspend/resume functions and puts > the device in d3hot state instead of d3cold. > Why not use the sysfs node to change "d3cold_allowed" on the device's upstream bridge? Thanks, Lijo > Regards, > Ma Jun > >> Alex >> >>> >>> Signed-off-by: Ma Jun <Jun.Ma2@amd.com> >>> --- >>> drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + >>> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +++++++ >>> drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 8 ++++++++ >>> 3 files changed, 16 insertions(+) >>> >>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h >>> index a9f54df9d33e..db9f60790267 100644 >>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h >>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h >>> @@ -166,6 +166,7 @@ extern char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH]; >>> extern int amdgpu_dpm; >>> extern int amdgpu_fw_load_type; >>> extern int amdgpu_aspm; >>> +extern int amdgpu_d3cold; >>> extern int amdgpu_runtime_pm; >>> extern uint amdgpu_ip_block_mask; >>> extern int amdgpu_bapm; >>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c >>> index 22b6a910b7f2..90501c44e7d0 100644 >>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c >>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c >>> @@ -264,6 +264,13 @@ bool amdgpu_device_supports_px(struct drm_device *dev) >>> bool amdgpu_device_supports_boco(struct drm_device *dev) >>> { >>> struct amdgpu_device *adev = drm_to_adev(dev); >>> + struct pci_dev *parent; >>> + >>> + if (!amdgpu_d3cold) { >>> + parent = pcie_find_root_port(adev->pdev); >>> + pci_d3cold_disable(parent); >>> + return false; >>> + } >>> >>> if (adev->has_pr3 || >>> ((adev->flags & AMD_IS_PX) && amdgpu_is_atpx_hybrid())) >>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c >>> index 5f14f04cb553..c9fbb8bd4169 100644 >>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c >>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c >>> @@ -145,6 +145,7 @@ char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH]; >>> int amdgpu_dpm = -1; >>> int amdgpu_fw_load_type = -1; >>> int amdgpu_aspm = -1; >>> +int amdgpu_d3cold = -1; >>> int amdgpu_runtime_pm = -1; >>> uint amdgpu_ip_block_mask = 0xffffffff; >>> int amdgpu_bapm = -1; >>> @@ -359,6 +360,13 @@ module_param_named(fw_load_type, amdgpu_fw_load_type, int, 0444); >>> MODULE_PARM_DESC(aspm, "ASPM support (1 = enable, 0 = disable, -1 = auto)"); >>> module_param_named(aspm, amdgpu_aspm, int, 0444); >>> >>> +/** >>> + * DOC: d3cold (int) >>> + * To disable d3cold (1 = enable, 0 = disable). The default is -1 (auto, enabled). >>> + */ >>> +MODULE_PARM_DESC(d3cold, "d3cold support (1 = enable, 0 = disable, -1 = auto)"); >>> +module_param_named(d3cold, amdgpu_d3cold, int, 0444); >>> + >>> /** >>> * DOC: runpm (int) >>> * Override for runtime power management control for dGPUs. The amdgpu driver can dynamically power down >>> -- >>> 2.34.1 >>> ^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [PATCH] drm/amdgpu: Add a new module param to disable d3cold 2023-11-30 9:18 ` Lazar, Lijo @ 2023-11-30 10:47 ` Ma, Jun 2023-11-30 10:55 ` Lazar, Lijo 0 siblings, 1 reply; 10+ messages in thread From: Ma, Jun @ 2023-11-30 10:47 UTC (permalink / raw) To: Lazar, Lijo, Alex Deucher, Ma Jun Cc: kevinyang.wang, amd-gfx, mario.limonciello, Alexander.Deucher, Kenneth.Feng Hi Lijo, On 11/30/2023 5:18 PM, Lazar, Lijo wrote: > > > On 11/30/2023 11:59 AM, Ma, Jun wrote: >> Hi Alex, >> >> On 11/30/2023 12:39 AM, Alex Deucher wrote: >>> On Wed, Nov 29, 2023 at 11:37 AM Ma Jun <Jun.Ma2@amd.com> wrote: >>>> >>>> Some platforms can't resume from d3cold state, So add a >>>> new module parameter to disable d3cold state for debugging >>>> purpose or workaround. >>> >>> Doesn't the runpm parameter already handle this? If you set runpm=0, >>> that should disable d3cold. >>> >> runpm=0 prevents calls to driver runtime_suspend/resume functions. >> While d3cold=0 allows calls to runtime_suspend/resume functions and puts >> the device in d3hot state instead of d3cold. >> > > Why not use the sysfs node to change "d3cold_allowed" on the device's > upstream bridge? > It seems the same question as Mario. Please refer to my reply to his question. Regards, Ma Jun > Thanks, > Lijo > >> Regards, >> Ma Jun >> >>> Alex >>> >>>> >>>> Signed-off-by: Ma Jun <Jun.Ma2@amd.com> >>>> --- >>>> drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + >>>> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +++++++ >>>> drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 8 ++++++++ >>>> 3 files changed, 16 insertions(+) >>>> >>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h >>>> index a9f54df9d33e..db9f60790267 100644 >>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h >>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h >>>> @@ -166,6 +166,7 @@ extern char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH]; >>>> extern int amdgpu_dpm; >>>> extern int amdgpu_fw_load_type; >>>> extern int amdgpu_aspm; >>>> +extern int amdgpu_d3cold; >>>> extern int amdgpu_runtime_pm; >>>> extern uint amdgpu_ip_block_mask; >>>> extern int amdgpu_bapm; >>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c >>>> index 22b6a910b7f2..90501c44e7d0 100644 >>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c >>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c >>>> @@ -264,6 +264,13 @@ bool amdgpu_device_supports_px(struct drm_device *dev) >>>> bool amdgpu_device_supports_boco(struct drm_device *dev) >>>> { >>>> struct amdgpu_device *adev = drm_to_adev(dev); >>>> + struct pci_dev *parent; >>>> + >>>> + if (!amdgpu_d3cold) { >>>> + parent = pcie_find_root_port(adev->pdev); >>>> + pci_d3cold_disable(parent); >>>> + return false; >>>> + } >>>> >>>> if (adev->has_pr3 || >>>> ((adev->flags & AMD_IS_PX) && amdgpu_is_atpx_hybrid())) >>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c >>>> index 5f14f04cb553..c9fbb8bd4169 100644 >>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c >>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c >>>> @@ -145,6 +145,7 @@ char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH]; >>>> int amdgpu_dpm = -1; >>>> int amdgpu_fw_load_type = -1; >>>> int amdgpu_aspm = -1; >>>> +int amdgpu_d3cold = -1; >>>> int amdgpu_runtime_pm = -1; >>>> uint amdgpu_ip_block_mask = 0xffffffff; >>>> int amdgpu_bapm = -1; >>>> @@ -359,6 +360,13 @@ module_param_named(fw_load_type, amdgpu_fw_load_type, int, 0444); >>>> MODULE_PARM_DESC(aspm, "ASPM support (1 = enable, 0 = disable, -1 = auto)"); >>>> module_param_named(aspm, amdgpu_aspm, int, 0444); >>>> >>>> +/** >>>> + * DOC: d3cold (int) >>>> + * To disable d3cold (1 = enable, 0 = disable). The default is -1 (auto, enabled). >>>> + */ >>>> +MODULE_PARM_DESC(d3cold, "d3cold support (1 = enable, 0 = disable, -1 = auto)"); >>>> +module_param_named(d3cold, amdgpu_d3cold, int, 0444); >>>> + >>>> /** >>>> * DOC: runpm (int) >>>> * Override for runtime power management control for dGPUs. The amdgpu driver can dynamically power down >>>> -- >>>> 2.34.1 >>>> > > ^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [PATCH] drm/amdgpu: Add a new module param to disable d3cold 2023-11-30 10:47 ` Ma, Jun @ 2023-11-30 10:55 ` Lazar, Lijo 0 siblings, 0 replies; 10+ messages in thread From: Lazar, Lijo @ 2023-11-30 10:55 UTC (permalink / raw) To: Ma, Jun, Alex Deucher, Ma Jun Cc: Alexander.Deucher, Kenneth.Feng, amd-gfx, kevinyang.wang, mario.limonciello On 11/30/2023 4:17 PM, Ma, Jun wrote: > Hi Lijo, > > On 11/30/2023 5:18 PM, Lazar, Lijo wrote: >> >> >> On 11/30/2023 11:59 AM, Ma, Jun wrote: >>> Hi Alex, >>> >>> On 11/30/2023 12:39 AM, Alex Deucher wrote: >>>> On Wed, Nov 29, 2023 at 11:37 AM Ma Jun <Jun.Ma2@amd.com> wrote: >>>>> >>>>> Some platforms can't resume from d3cold state, So add a >>>>> new module parameter to disable d3cold state for debugging >>>>> purpose or workaround. >>>> >>>> Doesn't the runpm parameter already handle this? If you set runpm=0, >>>> that should disable d3cold. >>>> >>> runpm=0 prevents calls to driver runtime_suspend/resume functions. >>> While d3cold=0 allows calls to runtime_suspend/resume functions and puts >>> the device in d3hot state instead of d3cold. >>> >> >> Why not use the sysfs node to change "d3cold_allowed" on the device's >> upstream bridge? >> > It seems the same question as Mario. Please refer to my reply to his question. > Once you disable on the device, all upstream devices along the path will be taken care. I don't see a special need to disable BOCO separately. pci_d3cold_disable is the same API used by sysfs node also. Thanks, Lijo > Regards, > Ma Jun > >> Thanks, >> Lijo >> >>> Regards, >>> Ma Jun >>> >>>> Alex >>>> >>>>> >>>>> Signed-off-by: Ma Jun <Jun.Ma2@amd.com> >>>>> --- >>>>> drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + >>>>> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +++++++ >>>>> drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 8 ++++++++ >>>>> 3 files changed, 16 insertions(+) >>>>> >>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h >>>>> index a9f54df9d33e..db9f60790267 100644 >>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h >>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h >>>>> @@ -166,6 +166,7 @@ extern char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH]; >>>>> extern int amdgpu_dpm; >>>>> extern int amdgpu_fw_load_type; >>>>> extern int amdgpu_aspm; >>>>> +extern int amdgpu_d3cold; >>>>> extern int amdgpu_runtime_pm; >>>>> extern uint amdgpu_ip_block_mask; >>>>> extern int amdgpu_bapm; >>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c >>>>> index 22b6a910b7f2..90501c44e7d0 100644 >>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c >>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c >>>>> @@ -264,6 +264,13 @@ bool amdgpu_device_supports_px(struct drm_device *dev) >>>>> bool amdgpu_device_supports_boco(struct drm_device *dev) >>>>> { >>>>> struct amdgpu_device *adev = drm_to_adev(dev); >>>>> + struct pci_dev *parent; >>>>> + >>>>> + if (!amdgpu_d3cold) { >>>>> + parent = pcie_find_root_port(adev->pdev); >>>>> + pci_d3cold_disable(parent); >>>>> + return false; >>>>> + } >>>>> >>>>> if (adev->has_pr3 || >>>>> ((adev->flags & AMD_IS_PX) && amdgpu_is_atpx_hybrid())) >>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c >>>>> index 5f14f04cb553..c9fbb8bd4169 100644 >>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c >>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c >>>>> @@ -145,6 +145,7 @@ char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH]; >>>>> int amdgpu_dpm = -1; >>>>> int amdgpu_fw_load_type = -1; >>>>> int amdgpu_aspm = -1; >>>>> +int amdgpu_d3cold = -1; >>>>> int amdgpu_runtime_pm = -1; >>>>> uint amdgpu_ip_block_mask = 0xffffffff; >>>>> int amdgpu_bapm = -1; >>>>> @@ -359,6 +360,13 @@ module_param_named(fw_load_type, amdgpu_fw_load_type, int, 0444); >>>>> MODULE_PARM_DESC(aspm, "ASPM support (1 = enable, 0 = disable, -1 = auto)"); >>>>> module_param_named(aspm, amdgpu_aspm, int, 0444); >>>>> >>>>> +/** >>>>> + * DOC: d3cold (int) >>>>> + * To disable d3cold (1 = enable, 0 = disable). The default is -1 (auto, enabled). >>>>> + */ >>>>> +MODULE_PARM_DESC(d3cold, "d3cold support (1 = enable, 0 = disable, -1 = auto)"); >>>>> +module_param_named(d3cold, amdgpu_d3cold, int, 0444); >>>>> + >>>>> /** >>>>> * DOC: runpm (int) >>>>> * Override for runtime power management control for dGPUs. The amdgpu driver can dynamically power down >>>>> -- >>>>> 2.34.1 >>>>> >> >> ^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [PATCH] drm/amdgpu: Add a new module param to disable d3cold 2023-11-30 6:29 ` Ma, Jun 2023-11-30 9:18 ` Lazar, Lijo @ 2023-11-30 14:21 ` Alex Deucher 1 sibling, 0 replies; 10+ messages in thread From: Alex Deucher @ 2023-11-30 14:21 UTC (permalink / raw) To: Ma, Jun Cc: kevinyang.wang, amd-gfx, Ma Jun, mario.limonciello, Alexander.Deucher, Kenneth.Feng On Thu, Nov 30, 2023 at 1:29 AM Ma, Jun <majun@amd.com> wrote: > > Hi Alex, > > On 11/30/2023 12:39 AM, Alex Deucher wrote: > > On Wed, Nov 29, 2023 at 11:37 AM Ma Jun <Jun.Ma2@amd.com> wrote: > >> > >> Some platforms can't resume from d3cold state, So add a > >> new module parameter to disable d3cold state for debugging > >> purpose or workaround. > > > > Doesn't the runpm parameter already handle this? If you set runpm=0, > > that should disable d3cold. > > > runpm=0 prevents calls to driver runtime_suspend/resume functions. > While d3cold=0 allows calls to runtime_suspend/resume functions and puts > the device in d3hot state instead of d3cold. But d3hot doesn't actually power down the card so it won't save any power. If we want to disable d3cold and still use runtime pm, it's better to try and use BACO. Alex > > Regards, > Ma Jun > > > Alex > > > >> > >> Signed-off-by: Ma Jun <Jun.Ma2@amd.com> > >> --- > >> drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + > >> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +++++++ > >> drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 8 ++++++++ > >> 3 files changed, 16 insertions(+) > >> > >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h > >> index a9f54df9d33e..db9f60790267 100644 > >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h > >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h > >> @@ -166,6 +166,7 @@ extern char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH]; > >> extern int amdgpu_dpm; > >> extern int amdgpu_fw_load_type; > >> extern int amdgpu_aspm; > >> +extern int amdgpu_d3cold; > >> extern int amdgpu_runtime_pm; > >> extern uint amdgpu_ip_block_mask; > >> extern int amdgpu_bapm; > >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c > >> index 22b6a910b7f2..90501c44e7d0 100644 > >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c > >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c > >> @@ -264,6 +264,13 @@ bool amdgpu_device_supports_px(struct drm_device *dev) > >> bool amdgpu_device_supports_boco(struct drm_device *dev) > >> { > >> struct amdgpu_device *adev = drm_to_adev(dev); > >> + struct pci_dev *parent; > >> + > >> + if (!amdgpu_d3cold) { > >> + parent = pcie_find_root_port(adev->pdev); > >> + pci_d3cold_disable(parent); > >> + return false; > >> + } > >> > >> if (adev->has_pr3 || > >> ((adev->flags & AMD_IS_PX) && amdgpu_is_atpx_hybrid())) > >> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c > >> index 5f14f04cb553..c9fbb8bd4169 100644 > >> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c > >> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c > >> @@ -145,6 +145,7 @@ char amdgpu_lockup_timeout[AMDGPU_MAX_TIMEOUT_PARAM_LENGTH]; > >> int amdgpu_dpm = -1; > >> int amdgpu_fw_load_type = -1; > >> int amdgpu_aspm = -1; > >> +int amdgpu_d3cold = -1; > >> int amdgpu_runtime_pm = -1; > >> uint amdgpu_ip_block_mask = 0xffffffff; > >> int amdgpu_bapm = -1; > >> @@ -359,6 +360,13 @@ module_param_named(fw_load_type, amdgpu_fw_load_type, int, 0444); > >> MODULE_PARM_DESC(aspm, "ASPM support (1 = enable, 0 = disable, -1 = auto)"); > >> module_param_named(aspm, amdgpu_aspm, int, 0444); > >> > >> +/** > >> + * DOC: d3cold (int) > >> + * To disable d3cold (1 = enable, 0 = disable). The default is -1 (auto, enabled). > >> + */ > >> +MODULE_PARM_DESC(d3cold, "d3cold support (1 = enable, 0 = disable, -1 = auto)"); > >> +module_param_named(d3cold, amdgpu_d3cold, int, 0444); > >> + > >> /** > >> * DOC: runpm (int) > >> * Override for runtime power management control for dGPUs. The amdgpu driver can dynamically power down > >> -- > >> 2.34.1 > >> ^ permalink raw reply [flat|nested] 10+ messages in thread
end of thread, other threads:[~2023-11-30 14:21 UTC | newest] Thread overview: 10+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2023-11-29 8:51 [PATCH] drm/amdgpu: Add a new module param to disable d3cold Ma Jun 2023-11-29 15:35 ` Mario Limonciello 2023-11-29 18:40 ` Alex Deucher 2023-11-30 9:46 ` Ma, Jun 2023-11-29 16:39 ` Alex Deucher 2023-11-30 6:29 ` Ma, Jun 2023-11-30 9:18 ` Lazar, Lijo 2023-11-30 10:47 ` Ma, Jun 2023-11-30 10:55 ` Lazar, Lijo 2023-11-30 14:21 ` Alex Deucher
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.