From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-09-10 20:26:53
Hi all,
as it seems there is considerable interest in landing this despite some
spurious issues still being present, here's a 4th revision of the series
to not drop the ball. Even though I still did not have time to try to
dig into the open issues reported by Frieder, I think there is a high
chance that those are caused by something like a missed timing
constraint and not by something that would invalidate the overall design
of the GPC and BLK_CTRL interaction as laid out by this series. So I
still hope that we can land this series and apply fixes as we find them.
Regards,
Lucas
Frieder Schrempf (1):
arm64: dts: imx8mm: Add GPU nodes for 2D and 3D core
Lucas Stach (15):
Revert "soc: imx: gpcv2: move reset assert after requesting domain
power up"
soc: imx: gpcv2: add lockdep annotation
soc: imx: gpcv2: add domain option to keep domain clocks enabled
soc: imx: gpcv2: keep i.MX8M* bus clocks enabled
soc: imx: gpcv2: support system suspend/resume
dt-bindings: soc: add binding for i.MX8MM VPU blk-ctrl
dt-bindings: power: imx8mm: add defines for VPU blk-ctrl domains
soc: imx: add i.MX8M blk-ctrl driver
dt-bindings: soc: add binding for i.MX8MM DISP blk-ctrl
dt-bindings: power: imx8mm: add defines for DISP blk-ctrl domains
soc: imx: imx8m-blk-ctrl: add DISP blk-ctrl
arm64: dts: imx8mm: add GPC node
arm64: dts: imx8mm: put USB controllers into power-domains
arm64: dts: imx8mm: add VPU blk-ctrl
arm64: dts: imx8mm: add DISP blk-ctrl
Marek Vasut (2):
soc: imx: gpcv2: Turn domain->pgc into bitfield
soc: imx: gpcv2: Set both GPC_PGC_nCTRL(GPU_2D|GPU_3D) for MX8MM GPU
domain
.../soc/imx/fsl,imx8mm-disp-blk-ctrl.yaml | 94 ++++
.../soc/imx/fsl,imx8mm-vpu-blk-ctrl.yaml | 76 +++
arch/arm64/boot/dts/freescale/imx8mm.dtsi | 180 ++++++
drivers/soc/imx/Makefile | 1 +
drivers/soc/imx/gpcv2.c | 130 +++--
drivers/soc/imx/imx8m-blk-ctrl.c | 523 ++++++++++++++++++
include/dt-bindings/power/imx8mm-power.h | 9 +
7 files changed, 972 insertions(+), 41 deletions(-)
create mode 100644 Documentation/devicetree/bindings/soc/imx/fsl,imx8mm-disp-blk-ctrl.yaml
create mode 100644 Documentation/devicetree/bindings/soc/imx/fsl,imx8mm-vpu-blk-ctrl.yaml
create mode 100644 drivers/soc/imx/imx8m-blk-ctrl.c
--
2.30.2
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-09-10 20:26:53
This reverts commit a77ebdd9f553. It turns out that the VPU domain has no
different requirements, even though the downstream ATF implementation seems
to suggest otherwise. Powering on the domain with the reset asserted works
fine. As the changed sequence has caused sporadic issues with the GPU
domains, just revert the change to go back to the working sequence.
Cc: <redacted> # 5.14
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Acked-by: Peng Fan <peng.fan@nxp.com>
---
drivers/soc/imx/gpcv2.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
@@ -244,6 +244,8 @@ static int imx_pgc_power_up(struct generic_pm_domain *genpd)gotoout_regulator_disable;}+reset_control_assert(domain->reset);+if(domain->bits.pxx){/* request the domain to power up */regmap_update_bits(domain->regmap,GPC_PU_PGC_SW_PUP_REQ,
@@ -266,8 +268,6 @@ static int imx_pgc_power_up(struct generic_pm_domain *genpd)GPC_PGC_CTRL_PCR);}-reset_control_assert(domain->reset);-/* delay for reset to propagate */udelay(5);
--
2.30.2
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-09-10 20:26:53
From: Marek Vasut <marex@denx.de>
There is currently the MX8MM GPU domain, which is in fact a composite domain
for both GPU2D and GPU3D. To correctly configure this domain, it is necessary
to control both GPC_PGC_nCTRL(GPU_2D) and GPC_PGC_nCTRL(GPU_3D) at the same
time. This is currently not possible.
Turn the domain->pgc from value into bitfield and use for_each_set_bit() to
iterate over all bits set in domain->pgc when configuring GPC_PGC_nCTRL
register array. This way it is possible to configure all GPC_PGC_nCTRL
registers required in a particular domain.
This is a preparatory patch, no functional change.
Signed-off-by: Marek Vasut <marex@denx.de>
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Peng Fan <peng.fan@nxp.com>
---
v2 (Lucas Stach):
- rebase on top of reverted reset sequence change
- also convert i.MX8MN domains
---
drivers/soc/imx/gpcv2.c | 72 ++++++++++++++++++++++-------------------
1 file changed, 38 insertions(+), 34 deletions(-)
@@ -264,8 +264,10 @@ static int imx_pgc_power_up(struct generic_pm_domain *genpd)}/* disable power control */-regmap_clear_bits(domain->regmap,GPC_PGC_CTRL(domain->pgc),-GPC_PGC_CTRL_PCR);+for_each_set_bit(pgc,&domain->pgc,32){+regmap_clear_bits(domain->regmap,GPC_PGC_CTRL(pgc),+GPC_PGC_CTRL_PCR);+}}/* delay for reset to propagate */
@@ -311,7 +313,7 @@ static int imx_pgc_power_up(struct generic_pm_domain *genpd)staticintimx_pgc_power_down(structgeneric_pm_domain*genpd){structimx_pgc_domain*domain=to_imx_pgc_domain(genpd);-u32reg_val;+u32reg_val,pgc;intret;/* Enable reset clocks for all devices in the domain */
@@ -338,8 +340,10 @@ static int imx_pgc_power_down(struct generic_pm_domain *genpd)if(domain->bits.pxx){/* enable power control */-regmap_update_bits(domain->regmap,GPC_PGC_CTRL(domain->pgc),-GPC_PGC_CTRL_PCR,GPC_PGC_CTRL_PCR);+for_each_set_bit(pgc,&domain->pgc,32){+regmap_update_bits(domain->regmap,GPC_PGC_CTRL(pgc),+GPC_PGC_CTRL_PCR,GPC_PGC_CTRL_PCR);+}/* request the domain to power down */regmap_update_bits(domain->regmap,GPC_PU_PGC_SW_PDN_REQ,
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-09-10 20:26:55
From: Marek Vasut <marex@denx.de>
To bring up the MX8MM GPU domain, it is necessary to configure both
GPC_PGC_nCTRL(GPU_2D) and GPC_PGC_nCTRL(GPU_3D) registers. Without
this configuration, the system might hang on boot when bringing up
the GPU power domain. This is sporadically observed on multiple
disparate systems.
Add the GPU3D bit into MX8MM GPU domain pgc bitfield, so that both
GPC_PGC_nCTRL(GPU_2D) and GPC_PGC_nCTRL(GPU_3D) registers are
configured when bringing up the GPU domain. This fixes the sporadic
hang.
Signed-off-by: Marek Vasut <marex@denx.de>
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Peng Fan <peng.fan@nxp.com>
---
drivers/soc/imx/gpcv2.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-09-10 20:26:57
Some of the MIX domains are using clocks to drive the bus bridges. Those
must be enabled at all times, as long as the domain is powered up and
they don't have any other consumer than the power domain. Add an option
to keep the clocks attached to a domain enabled as long as the domain
is power up and only disable them after the domain is powered down.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Peng Fan <peng.fan@nxp.com>
---
drivers/soc/imx/gpcv2.c | 14 +++++++++-----
1 file changed, 9 insertions(+), 5 deletions(-)
@@ -295,7 +296,8 @@ static int imx_pgc_power_up(struct generic_pm_domain *genpd)}/* Disable reset clocks for all devices in the domain */-clk_bulk_disable_unprepare(domain->num_clks,domain->clks);+if(!domain->keep_clocks)+clk_bulk_disable_unprepare(domain->num_clks,domain->clks);return0;
@@ -317,10 +319,12 @@ static int imx_pgc_power_down(struct generic_pm_domain *genpd)intret;/* Enable reset clocks for all devices in the domain */-ret=clk_bulk_prepare_enable(domain->num_clks,domain->clks);-if(ret){-dev_err(domain->dev,"failed to enable reset clocks\n");-returnret;+if(!domain->keep_clocks){+ret=clk_bulk_prepare_enable(domain->num_clks,domain->clks);+if(ret){+dev_err(domain->dev,"failed to enable reset clocks\n");+returnret;+}}/* request the ADB400 to power down */
--
2.30.2
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-09-10 20:26:58
Some of the GPCv2 power domains are nested inside each other without
visibility to lockdep at the genpd level, as they are in separate
driver instances and don't have a parent/child power-domain relationship.
Add a subclass annotation to the nested domains to let lockdep know that
it is okay to take the genpd lock in a nested fashion.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Peng Fan <peng.fan@nxp.com>
---
drivers/soc/imx/gpcv2.c | 4 ++++
1 file changed, 4 insertions(+)
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-09-10 20:26:59
Annotate the domains with bus clocks to keep those clocks enabled
as long as the domain is active.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Peng Fan <peng.fan@nxp.com>
---
drivers/soc/imx/gpcv2.c | 5 +++++
1 file changed, 5 insertions(+)
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-09-10 20:27:00
Our usage of runtime PM to control the hierarchy of power domains is
slightly unusual and means that powering up a domain may fail in early
system resume, as runtime PM is still disallowed at this stage.
However the system suspend/resume path takes care of powering down/up
the power domains in the order defined by the device parent/child and
power-domain provider/consumer hierarachy. So we can just runtime
resume all our power-domain devices to allow the power-up to work
properly in the resume path. System suspend will still disable all
domains as intended.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Acked-by: Peng Fan <peng.fan@nxp.com>
---
drivers/soc/imx/gpcv2.c | 31 +++++++++++++++++++++++++++++++
1 file changed, 31 insertions(+)
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-09-10 20:27:01
This adds a driver for the blk-ctrl blocks found in the i.MX8M* line of
SoCs. The blk-ctrl is a top-level peripheral located in the various *MIX
power domains and interacts with the GPC power controller to provide the
peripherals in the power domain access to the NoC and ensures that those
peripherals are properly reset when their respective power domain is
brought back to life.
Software needs to do different things to make the bus handshake happen
after the GPC *MIX domain is powered up and before it is powered down.
As the requirements are quite different between the various blk-ctrls
there is a callback function provided to hook in the proper sequence.
The peripheral domains are quite uniform, they handle the soft clock
enables and resets in the blk-ctrl address space and sequencing with the
upstream GPC power domains.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Acked-by: Frieder Schrempf <redacted>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
---
This commit includes the full code to drive the VPUMIX domain on the
i.MX8MM, as the skeleton driver would probably be harder to review
without the context provided by one blk-ctrl implementation. Other
blk-ctrl implementations will follow, based on this overall structure.
v4:
- fix commit message typos
- fix superfluous whitespace
- constify clk_names more
---
drivers/soc/imx/Makefile | 1 +
drivers/soc/imx/imx8m-blk-ctrl.c | 453 +++++++++++++++++++++++++++++++
2 files changed, 454 insertions(+)
create mode 100644 drivers/soc/imx/imx8m-blk-ctrl.c
@@ -0,0 +1,453 @@+// SPDX-License-Identifier: GPL-2.0+++/*+*Copyright2021Pengutronix,LucasStach<kernel@pengutronix.de>+*/++#include<linux/device.h>+#include<linux/module.h>+#include<linux/of_device.h>+#include<linux/platform_device.h>+#include<linux/pm_domain.h>+#include<linux/pm_runtime.h>+#include<linux/regmap.h>+#include<linux/clk.h>++#include<dt-bindings/power/imx8mm-power.h>++#define BLK_SFT_RSTN 0x0+#define BLK_CLK_EN 0x4++structimx8m_blk_ctrl_domain;++structimx8m_blk_ctrl{+structdevice*dev;+structnotifier_blockpower_nb;+structdevice*bus_power_dev;+structregmap*regmap;+structimx8m_blk_ctrl_domain*domains;+structgenpd_onecell_dataonecell_data;+};++structimx8m_blk_ctrl_domain_data{+constchar*name;+constchar*const*clk_names;+intnum_clks;+constchar*gpc_name;+u32rst_mask;+u32clk_mask;+};++#define DOMAIN_MAX_CLKS 3++structimx8m_blk_ctrl_domain{+structgeneric_pm_domaingenpd;+conststructimx8m_blk_ctrl_domain_data*data;+structclk_bulk_dataclks[DOMAIN_MAX_CLKS];+structdevice*power_dev;+structimx8m_blk_ctrl*bc;+};++structimx8m_blk_ctrl_data{+intmax_reg;+notifier_fn_tpower_notifier_fn;+conststructimx8m_blk_ctrl_domain_data*domains;+intnum_domains;+};++staticinlinestructimx8m_blk_ctrl_domain*+to_imx8m_blk_ctrl_domain(structgeneric_pm_domain*genpd)+{+returncontainer_of(genpd,structimx8m_blk_ctrl_domain,genpd);+}++staticintimx8m_blk_ctrl_power_on(structgeneric_pm_domain*genpd)+{+structimx8m_blk_ctrl_domain*domain=to_imx8m_blk_ctrl_domain(genpd);+conststructimx8m_blk_ctrl_domain_data*data=domain->data;+structimx8m_blk_ctrl*bc=domain->bc;+intret;++/* make sure bus domain is awake */+ret=pm_runtime_get_sync(bc->bus_power_dev);+if(ret<0){+pm_runtime_put_noidle(bc->bus_power_dev);+dev_err(bc->dev,"failed to power up bus domain\n");+returnret;+}++/* put devices into reset */+regmap_clear_bits(bc->regmap,BLK_SFT_RSTN,data->rst_mask);++/* enable upstream and blk-ctrl clocks to allow reset to propagate */+ret=clk_bulk_prepare_enable(data->num_clks,domain->clks);+if(ret){+dev_err(bc->dev,"failed to enable clocks\n");+gotobus_put;+}+regmap_set_bits(bc->regmap,BLK_CLK_EN,data->clk_mask);++/* power up upstream GPC domain */+ret=pm_runtime_get_sync(domain->power_dev);+if(ret<0){+dev_err(bc->dev,"failed to power up peripheral domain\n");+gotoclk_disable;+}++/* wait for reset to propagate */+udelay(5);++/* release reset */+regmap_set_bits(bc->regmap,BLK_SFT_RSTN,data->rst_mask);++/* disable upstream clocks */+clk_bulk_disable_unprepare(data->num_clks,domain->clks);++return0;++clk_disable:+clk_bulk_disable_unprepare(data->num_clks,domain->clks);+bus_put:+pm_runtime_put(bc->bus_power_dev);++returnret;+}++staticintimx8m_blk_ctrl_power_off(structgeneric_pm_domain*genpd)+{+structimx8m_blk_ctrl_domain*domain=to_imx8m_blk_ctrl_domain(genpd);+conststructimx8m_blk_ctrl_domain_data*data=domain->data;+structimx8m_blk_ctrl*bc=domain->bc;++/* put devices into reset and disable clocks */+regmap_clear_bits(bc->regmap,BLK_SFT_RSTN,data->rst_mask);+regmap_clear_bits(bc->regmap,BLK_CLK_EN,data->clk_mask);++/* power down upstream GPC domain */+pm_runtime_put(domain->power_dev);++/* allow bus domain to suspend */+pm_runtime_put(bc->bus_power_dev);++return0;+}++staticstructgeneric_pm_domain*+imx8m_blk_ctrl_xlate(structof_phandle_args*args,void*data)+{+structgenpd_onecell_data*onecell_data=data;+unsignedintindex=args->args[0];++if(args->args_count!=1||+index>onecell_data->num_domains)+returnERR_PTR(-EINVAL);++returnonecell_data->domains[index];+}++staticstructlock_class_keyblk_ctrl_genpd_lock_class;++staticintimx8m_blk_ctrl_probe(structplatform_device*pdev)+{+conststructimx8m_blk_ctrl_data*bc_data;+structdevice*dev=&pdev->dev;+structimx8m_blk_ctrl*bc;+void__iomem*base;+inti,ret;++structregmap_configregmap_config={+.reg_bits=32,+.val_bits=32,+.reg_stride=4,+};++bc=devm_kzalloc(dev,sizeof(*bc),GFP_KERNEL);+if(!bc)+return-ENOMEM;++bc->dev=dev;++bc_data=of_device_get_match_data(dev);++base=devm_platform_ioremap_resource(pdev,0);+if(IS_ERR(base))+returnPTR_ERR(base);++regmap_config.max_register=bc_data->max_reg;+bc->regmap=devm_regmap_init_mmio(dev,base,®map_config);+if(IS_ERR(bc->regmap))+returndev_err_probe(dev,PTR_ERR(bc->regmap),+"failed to init regmap\n");++bc->domains=devm_kcalloc(dev,bc_data->num_domains,+sizeof(structimx8m_blk_ctrl_domain),+GFP_KERNEL);+if(!bc->domains)+return-ENOMEM;++bc->onecell_data.num_domains=bc_data->num_domains;+bc->onecell_data.xlate=imx8m_blk_ctrl_xlate;+bc->onecell_data.domains=+devm_kcalloc(dev,bc_data->num_domains,+sizeof(structgeneric_pm_domain*),GFP_KERNEL);+if(!bc->onecell_data.domains)+return-ENOMEM;++bc->bus_power_dev=genpd_dev_pm_attach_by_name(dev,"bus");+if(IS_ERR(bc->bus_power_dev))+returndev_err_probe(dev,PTR_ERR(bc->bus_power_dev),+"failed to attach power domain\n");++for(i=0;i<bc_data->num_domains;i++){+conststructimx8m_blk_ctrl_domain_data*data=&bc_data->domains[i];+structimx8m_blk_ctrl_domain*domain=&bc->domains[i];+intj;++domain->data=data;++for(j=0;j<data->num_clks;j++)+domain->clks[j].id=data->clk_names[j];++ret=devm_clk_bulk_get(dev,data->num_clks,domain->clks);+if(ret){+dev_err_probe(dev,ret,"failed to get clock\n");+gotocleanup_pds;+}++domain->power_dev=+dev_pm_domain_attach_by_name(dev,data->gpc_name);+if(IS_ERR(domain->power_dev)){+dev_err_probe(dev,PTR_ERR(domain->power_dev),+"failed to attach power domain\n");+ret=PTR_ERR(domain->power_dev);+gotocleanup_pds;+}++domain->genpd.name=data->name;+domain->genpd.power_on=imx8m_blk_ctrl_power_on;+domain->genpd.power_off=imx8m_blk_ctrl_power_off;+domain->bc=bc;++ret=pm_genpd_init(&domain->genpd,NULL,true);+if(ret){+dev_err_probe(dev,ret,"failed to init power domain\n");+dev_pm_domain_detach(domain->power_dev,true);+gotocleanup_pds;+}++/*+*WeuseruntimePMtotriggerpoweron/offoftheupstreamGPC+*domain,asastricthierarchicalparent/childpowerdomain+*setupdoesn'tallowustomeetthesequencingrequirements.+*Thismeanswehavenestedlockingofgenpdlocks,withoutthe+*nestingbeingvisibleatthegenpdlevel,soweneeda+*separatelockclasstomakelockdepawareofthefactthat+*thisareseparatedomainlocksthatcanbenestedwithouta+*self-deadlock.+*/+lockdep_set_class(&domain->genpd.mlock,+&blk_ctrl_genpd_lock_class);++bc->onecell_data.domains[i]=&domain->genpd;+}++ret=of_genpd_add_provider_onecell(dev->of_node,&bc->onecell_data);+if(ret){+dev_err_probe(dev,ret,"failed to add power domain provider\n");+gotocleanup_pds;+}++bc->power_nb.notifier_call=bc_data->power_notifier_fn;+ret=dev_pm_genpd_add_notifier(bc->bus_power_dev,&bc->power_nb);+if(ret){+dev_err_probe(dev,ret,"failed to add power notifier\n");+gotocleanup_provider;+}++dev_set_drvdata(dev,bc);++return0;++cleanup_provider:+of_genpd_del_provider(dev->of_node);+cleanup_pds:+for(i--;i>=0;i--){+pm_genpd_remove(&bc->domains[i].genpd);+dev_pm_domain_detach(bc->domains[i].power_dev,true);+}++dev_pm_domain_detach(bc->bus_power_dev,true);++returnret;+}++staticintimx8m_blk_ctrl_remove(structplatform_device*pdev)+{+structimx8m_blk_ctrl*bc=dev_get_drvdata(&pdev->dev);+inti;++of_genpd_del_provider(pdev->dev.of_node);++for(i=0;bc->onecell_data.num_domains;i++){+structimx8m_blk_ctrl_domain*domain=&bc->domains[i];++pm_genpd_remove(&domain->genpd);+dev_pm_domain_detach(domain->power_dev,true);+}++dev_pm_genpd_remove_notifier(bc->bus_power_dev);++dev_pm_domain_detach(bc->bus_power_dev,true);++return0;+}++#ifdef CONFIG_PM_SLEEP+staticintimx8m_blk_ctrl_suspend(structdevice*dev)+{+structimx8m_blk_ctrl*bc=dev_get_drvdata(dev);+intret,i;++/*+*Thismaylookstrange,butisdonesothegenericPM_SLEEPcode+*canpowerdownourdomainsandmoreimportantlypowerthemupagain+*afterresume,withouttrippingoverourusageofruntimePMto+*controltheupstreamGPCdomains.Thingshappenintherightorder+*inthesystemsuspend/resumepathsduetothedeviceparent/child+*hierarchy.+*/+ret=pm_runtime_get_sync(bc->bus_power_dev);+if(ret<0){+pm_runtime_put_noidle(bc->bus_power_dev);+returnret;+}++for(i=0;i<bc->onecell_data.num_domains;i++){+structimx8m_blk_ctrl_domain*domain=&bc->domains[i];++ret=pm_runtime_get_sync(domain->power_dev);+if(ret<0){+pm_runtime_put_noidle(domain->power_dev);+gotoout_fail;+}+}++return0;++out_fail:+for(i--;i>=0;i--)+pm_runtime_put(bc->domains[i].power_dev);++pm_runtime_put(bc->bus_power_dev);++returnret;+}++staticintimx8m_blk_ctrl_resume(structdevice*dev)+{+structimx8m_blk_ctrl*bc=dev_get_drvdata(dev);+inti;++for(i=0;i<bc->onecell_data.num_domains;i++)+pm_runtime_put(bc->domains[i].power_dev);++pm_runtime_put(bc->bus_power_dev);++return0;+}+#endif++staticconststructdev_pm_opsimx8m_blk_ctrl_pm_ops={+SET_SYSTEM_SLEEP_PM_OPS(imx8m_blk_ctrl_suspend,imx8m_blk_ctrl_resume)+};++staticintimx8mm_vpu_power_notifier(structnotifier_block*nb,+unsignedlongaction,void*data)+{+structimx8m_blk_ctrl*bc=container_of(nb,structimx8m_blk_ctrl,+power_nb);++if(action!=GENPD_NOTIFY_ON&&action!=GENPD_NOTIFY_PRE_OFF)+returnNOTIFY_OK;++/*+*TheADBintheVPUMIXdomainhasnoseparateresetandclock+*enablebits,butisungatedtogetherwiththeVPUclocks.To+*allowthehandshakewiththeGPCtoprogressweputtheVPUs+*inresetandungatetheclocks.+*/+regmap_clear_bits(bc->regmap,BLK_SFT_RSTN,BIT(0)|BIT(1)|BIT(2));+regmap_set_bits(bc->regmap,BLK_CLK_EN,BIT(0)|BIT(1)|BIT(2));++if(action==GENPD_NOTIFY_ON){+/*+*OnpowerupwehavenosoftwarebackchanneltotheGPCto+*waitfortheADBhandshaketohappen,sowejustdelayfora+*bit.OnpowerdowntheGPCdriverwaitsforthehandshake.+*/+udelay(5);++/* set "fuse" bits to enable the VPUs */+regmap_set_bits(bc->regmap,0x8,0xffffffff);+regmap_set_bits(bc->regmap,0xc,0xffffffff);+regmap_set_bits(bc->regmap,0x10,0xffffffff);+regmap_set_bits(bc->regmap,0x14,0xffffffff);+}++returnNOTIFY_OK;+}++staticconststructimx8m_blk_ctrl_domain_dataimx8m_vpu_blk_ctl_domain_data[]={+[IMX8MM_VPUBLK_PD_G1]={+.name="vpublk-g1",+.clk_names=(constchar*[]){"g1",},+.num_clks=1,+.gpc_name="g1",+.rst_mask=BIT(1),+.clk_mask=BIT(1),+},+[IMX8MM_VPUBLK_PD_G2]={+.name="vpublk-g2",+.clk_names=(constchar*[]){"g2",},+.num_clks=1,+.gpc_name="g2",+.rst_mask=BIT(0),+.clk_mask=BIT(0),+},+[IMX8MM_VPUBLK_PD_H1]={+.name="vpublk-h1",+.clk_names=(constchar*[]){"h1",},+.num_clks=1,+.gpc_name="h1",+.rst_mask=BIT(2),+.clk_mask=BIT(2),+},+};++staticconststructimx8m_blk_ctrl_dataimx8m_vpu_blk_ctl_dev_data={+.max_reg=0x18,+.power_notifier_fn=imx8mm_vpu_power_notifier,+.domains=imx8m_vpu_blk_ctl_domain_data,+.num_domains=ARRAY_SIZE(imx8m_vpu_blk_ctl_domain_data),+};++staticconststructof_device_idimx8m_blk_ctrl_of_match[]={+{+.compatible="fsl,imx8mm-vpu-blk-ctrl",+.data=&imx8m_vpu_blk_ctl_dev_data+},{+/* Sentinel */+}+};+MODULE_DEVICE_TABLE(of,imx8m_blk_ctrl_of_match);++staticstructplatform_driverimx8m_blk_ctrl_driver={+.probe=imx8m_blk_ctrl_probe,+.remove=imx8m_blk_ctrl_remove,+.driver={+.name="imx8m-blk-ctrl",+.pm=&imx8m_blk_ctrl_pm_ops,+.of_match_table=imx8m_blk_ctrl_of_match,+},+};+module_platform_driver(imx8m_blk_ctrl_driver);
--
2.30.2
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-09-10 20:27:04
This adds the defines for the power domains provided by the DISP
blk-ctrl.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
v4: Change naming to be consistent with the VPUBLK domains.
---
include/dt-bindings/power/imx8mm-power.h | 5 +++++
1 file changed, 5 insertions(+)
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-09-10 20:27:05
This adds the description for the i.MX8MM disp blk-ctrl.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Peng Fan <peng.fan@nxp.com>
---
drivers/soc/imx/imx8m-blk-ctrl.c | 70 ++++++++++++++++++++++++++++++++
1 file changed, 70 insertions(+)
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-09-10 20:27:06
This adds the defines for the power domains provided by the VPU
blk-ctrl.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
include/dt-bindings/power/imx8mm-power.h | 4 ++++
1 file changed, 4 insertions(+)
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-09-10 20:27:07
Now that we have support for the power domain controller on the i.MX8MM
we can put the USB controllers in their respective power domains to allow
them to power down the PHY when possible.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
arch/arm64/boot/dts/freescale/imx8mm.dtsi | 2 ++
1 file changed, 2 insertions(+)
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-09-10 20:27:07
Add the DT node for the GPC, including all the PGC power domains,
some of them are not fully functional yet, as they require interaction
with the blk-ctrls to properly power up/down the peripherals.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
arch/arm64/boot/dts/freescale/imx8mm.dtsi | 107 ++++++++++++++++++++++
1 file changed, 107 insertions(+)
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-09-10 20:27:08
Add the DT node for the DISP blk-ctrl. With this in place the
display/mipi power domains are fully functional.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
arch/arm64/boot/dts/freescale/imx8mm.dtsi | 27 +++++++++++++++++++++++
1 file changed, 27 insertions(+)
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-09-10 20:27:09
From: Frieder Schrempf <redacted>
According to the documents, the i.MX8M-Mini features a GC320 and a
GCNanoUltra GPU core. Etnaviv detects them as:
etnaviv-gpu 38000000.gpu: model: GC600, revision: 4653
etnaviv-gpu 38008000.gpu: model: GC520, revision: 5341
This seems to work fine more or less without any changes to the HWDB,
which still might be needed in the future to correct some features,
etc.
[lst]: Added power domains and switched clock assignments to the
new clock defines used for the composite clocks, instead of
relying on the backwards compat defines.
Signed-off-by: Frieder Schrempf <redacted>
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
arch/arm64/boot/dts/freescale/imx8mm.dtsi | 31 +++++++++++++++++++++++
1 file changed, 31 insertions(+)
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-09-10 20:27:10
Add the DT node for the VPU blk-ctrl. With this in place the
VPU power domains are fully functional.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
arch/arm64/boot/dts/freescale/imx8mm.dtsi | 13 +++++++++++++
1 file changed, 13 insertions(+)
From: Adam Ford <hidden> Date: 2021-09-12 14:40:55
On Fri, Sep 10, 2021 at 3:26 PM Lucas Stach [off-list ref] wrote:
quoted hunk
This adds the description for the i.MX8MM disp blk-ctrl.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Peng Fan <peng.fan@nxp.com>
---
drivers/soc/imx/imx8m-blk-ctrl.c | 70 ++++++++++++++++++++++++++++++++
1 file changed, 70 insertions(+)
From: Benjamin Gaignard <benjamin.gaignard@collabora.com> Date: 2021-09-14 15:46:56
Le 10/09/2021 à 22:26, Lucas Stach a écrit :
This adds a driver for the blk-ctrl blocks found in the i.MX8M* line of
SoCs. The blk-ctrl is a top-level peripheral located in the various *MIX
power domains and interacts with the GPC power controller to provide the
peripherals in the power domain access to the NoC and ensures that those
peripherals are properly reset when their respective power domain is
brought back to life.
Software needs to do different things to make the bus handshake happen
after the GPC *MIX domain is powered up and before it is powered down.
As the requirements are quite different between the various blk-ctrls
there is a callback function provided to hook in the proper sequence.
The peripheral domains are quite uniform, they handle the soft clock
enables and resets in the blk-ctrl address space and sequencing with the
upstream GPC power domains.
Hi Lucas,
I have tried to use your patches for IMX8MQ but it seems that the hardware
have different architecture.
On IMX8MQ there is only one VPU domain for G1 and G2 and that doesn't match
with your implementation where it is needed to have "bus" and devices power domain.
From what I experiment in current IMX8MQ implementation of blk-ctrl (inside VPU driver)
enabling the 3 clocks (bus, G1, G2) is needed to reset the VPUs.
Do you think you can update your design to take care of these hardware variations ?
Regards,
Benjamin
quoted hunk
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Acked-by: Frieder Schrempf <redacted>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
---
This commit includes the full code to drive the VPUMIX domain on the
i.MX8MM, as the skeleton driver would probably be harder to review
without the context provided by one blk-ctrl implementation. Other
blk-ctrl implementations will follow, based on this overall structure.
v4:
- fix commit message typos
- fix superfluous whitespace
- constify clk_names more
---
drivers/soc/imx/Makefile | 1 +
drivers/soc/imx/imx8m-blk-ctrl.c | 453 +++++++++++++++++++++++++++++++
2 files changed, 454 insertions(+)
create mode 100644 drivers/soc/imx/imx8m-blk-ctrl.c
@@ -0,0 +1,453 @@+// SPDX-License-Identifier: GPL-2.0+++/*+*Copyright2021Pengutronix,LucasStach<kernel@pengutronix.de>+*/++#include<linux/device.h>+#include<linux/module.h>+#include<linux/of_device.h>+#include<linux/platform_device.h>+#include<linux/pm_domain.h>+#include<linux/pm_runtime.h>+#include<linux/regmap.h>+#include<linux/clk.h>++#include<dt-bindings/power/imx8mm-power.h>++#define BLK_SFT_RSTN 0x0+#define BLK_CLK_EN 0x4++structimx8m_blk_ctrl_domain;++structimx8m_blk_ctrl{+structdevice*dev;+structnotifier_blockpower_nb;+structdevice*bus_power_dev;+structregmap*regmap;+structimx8m_blk_ctrl_domain*domains;+structgenpd_onecell_dataonecell_data;+};++structimx8m_blk_ctrl_domain_data{+constchar*name;+constchar*const*clk_names;+intnum_clks;+constchar*gpc_name;+u32rst_mask;+u32clk_mask;+};++#define DOMAIN_MAX_CLKS 3++structimx8m_blk_ctrl_domain{+structgeneric_pm_domaingenpd;+conststructimx8m_blk_ctrl_domain_data*data;+structclk_bulk_dataclks[DOMAIN_MAX_CLKS];+structdevice*power_dev;+structimx8m_blk_ctrl*bc;+};++structimx8m_blk_ctrl_data{+intmax_reg;+notifier_fn_tpower_notifier_fn;+conststructimx8m_blk_ctrl_domain_data*domains;+intnum_domains;+};++staticinlinestructimx8m_blk_ctrl_domain*+to_imx8m_blk_ctrl_domain(structgeneric_pm_domain*genpd)+{+returncontainer_of(genpd,structimx8m_blk_ctrl_domain,genpd);+}++staticintimx8m_blk_ctrl_power_on(structgeneric_pm_domain*genpd)+{+structimx8m_blk_ctrl_domain*domain=to_imx8m_blk_ctrl_domain(genpd);+conststructimx8m_blk_ctrl_domain_data*data=domain->data;+structimx8m_blk_ctrl*bc=domain->bc;+intret;++/* make sure bus domain is awake */+ret=pm_runtime_get_sync(bc->bus_power_dev);+if(ret<0){+pm_runtime_put_noidle(bc->bus_power_dev);+dev_err(bc->dev,"failed to power up bus domain\n");+returnret;+}++/* put devices into reset */+regmap_clear_bits(bc->regmap,BLK_SFT_RSTN,data->rst_mask);++/* enable upstream and blk-ctrl clocks to allow reset to propagate */+ret=clk_bulk_prepare_enable(data->num_clks,domain->clks);
Maybe the clocks should be enabled before clear BLK_SFT_RSTN ?
+ if (ret) {
+ dev_err(bc->dev, "failed to enable clocks\n");
+ goto bus_put;
+ }
+ regmap_set_bits(bc->regmap, BLK_CLK_EN, data->clk_mask);
+
+ /* power up upstream GPC domain */
+ ret = pm_runtime_get_sync(domain->power_dev);
+ if (ret < 0) {
+ dev_err(bc->dev, "failed to power up peripheral domain\n");
+ goto clk_disable;
+ }
+
+ /* wait for reset to propagate */
+ udelay(5);
+
+ /* release reset */
+ regmap_set_bits(bc->regmap, BLK_SFT_RSTN, data->rst_mask);
+
+ /* disable upstream clocks */
+ clk_bulk_disable_unprepare(data->num_clks, domain->clks);
+
+ return 0;
+
+clk_disable:
+ clk_bulk_disable_unprepare(data->num_clks, domain->clks);
+bus_put:
+ pm_runtime_put(bc->bus_power_dev);
+
+ return ret;
+}
+
+static int imx8m_blk_ctrl_power_off(struct generic_pm_domain *genpd)
+{
+ struct imx8m_blk_ctrl_domain *domain = to_imx8m_blk_ctrl_domain(genpd);
+ const struct imx8m_blk_ctrl_domain_data *data = domain->data;
+ struct imx8m_blk_ctrl *bc = domain->bc;
+
+ /* put devices into reset and disable clocks */
+ regmap_clear_bits(bc->regmap, BLK_SFT_RSTN, data->rst_mask);
+ regmap_clear_bits(bc->regmap, BLK_CLK_EN, data->clk_mask);
+
+ /* power down upstream GPC domain */
+ pm_runtime_put(domain->power_dev);
+
+ /* allow bus domain to suspend */
+ pm_runtime_put(bc->bus_power_dev);
+
+ return 0;
+}
+
+static struct generic_pm_domain *
+imx8m_blk_ctrl_xlate(struct of_phandle_args *args, void *data)
+{
+ struct genpd_onecell_data *onecell_data = data;
+ unsigned int index = args->args[0];
+
+ if (args->args_count != 1 ||
+ index > onecell_data->num_domains)
+ return ERR_PTR(-EINVAL);
+
+ return onecell_data->domains[index];
+}
+
+static struct lock_class_key blk_ctrl_genpd_lock_class;
+
+static int imx8m_blk_ctrl_probe(struct platform_device *pdev)
+{
+ const struct imx8m_blk_ctrl_data *bc_data;
+ struct device *dev = &pdev->dev;
+ struct imx8m_blk_ctrl *bc;
+ void __iomem *base;
+ int i, ret;
+
+ struct regmap_config regmap_config = {
+ .reg_bits = 32,
+ .val_bits = 32,
+ .reg_stride = 4,
+ };
+
+ bc = devm_kzalloc(dev, sizeof(*bc), GFP_KERNEL);
+ if (!bc)
+ return -ENOMEM;
+
+ bc->dev = dev;
+
+ bc_data = of_device_get_match_data(dev);
+
+ base = devm_platform_ioremap_resource(pdev, 0);
+ if (IS_ERR(base))
+ return PTR_ERR(base);
+
+ regmap_config.max_register = bc_data->max_reg;
+ bc->regmap = devm_regmap_init_mmio(dev, base, ®map_config);
+ if (IS_ERR(bc->regmap))
+ return dev_err_probe(dev, PTR_ERR(bc->regmap),
+ "failed to init regmap\n");
+
+ bc->domains = devm_kcalloc(dev, bc_data->num_domains,
+ sizeof(struct imx8m_blk_ctrl_domain),
+ GFP_KERNEL);
+ if (!bc->domains)
+ return -ENOMEM;
+
+ bc->onecell_data.num_domains = bc_data->num_domains;
+ bc->onecell_data.xlate = imx8m_blk_ctrl_xlate;
+ bc->onecell_data.domains =
+ devm_kcalloc(dev, bc_data->num_domains,
+ sizeof(struct generic_pm_domain *), GFP_KERNEL);
+ if (!bc->onecell_data.domains)
+ return -ENOMEM;
+
+ bc->bus_power_dev = genpd_dev_pm_attach_by_name(dev, "bus");
+ if (IS_ERR(bc->bus_power_dev))
+ return dev_err_probe(dev, PTR_ERR(bc->bus_power_dev),
+ "failed to attach power domain\n");
+
+ for (i = 0; i < bc_data->num_domains; i++) {
+ const struct imx8m_blk_ctrl_domain_data *data = &bc_data->domains[i];
+ struct imx8m_blk_ctrl_domain *domain = &bc->domains[i];
+ int j;
+
+ domain->data = data;
+
+ for (j = 0; j < data->num_clks; j++)
+ domain->clks[j].id = data->clk_names[j];
+
+ ret = devm_clk_bulk_get(dev, data->num_clks, domain->clks);
+ if (ret) {
+ dev_err_probe(dev, ret, "failed to get clock\n");
+ goto cleanup_pds;
+ }
+
+ domain->power_dev =
+ dev_pm_domain_attach_by_name(dev, data->gpc_name);
+ if (IS_ERR(domain->power_dev)) {
+ dev_err_probe(dev, PTR_ERR(domain->power_dev),
+ "failed to attach power domain\n");
+ ret = PTR_ERR(domain->power_dev);
+ goto cleanup_pds;
+ }
+
+ domain->genpd.name = data->name;
+ domain->genpd.power_on = imx8m_blk_ctrl_power_on;
+ domain->genpd.power_off = imx8m_blk_ctrl_power_off;
+ domain->bc = bc;
+
+ ret = pm_genpd_init(&domain->genpd, NULL, true);
+ if (ret) {
+ dev_err_probe(dev, ret, "failed to init power domain\n");
+ dev_pm_domain_detach(domain->power_dev, true);
+ goto cleanup_pds;
+ }
+
+ /*
+ * We use runtime PM to trigger power on/off of the upstream GPC
+ * domain, as a strict hierarchical parent/child power domain
+ * setup doesn't allow us to meet the sequencing requirements.
+ * This means we have nested locking of genpd locks, without the
+ * nesting being visible at the genpd level, so we need a
+ * separate lock class to make lockdep aware of the fact that
+ * this are separate domain locks that can be nested without a
+ * self-deadlock.
+ */
+ lockdep_set_class(&domain->genpd.mlock,
+ &blk_ctrl_genpd_lock_class);
+
+ bc->onecell_data.domains[i] = &domain->genpd;
+ }
+
+ ret = of_genpd_add_provider_onecell(dev->of_node, &bc->onecell_data);
+ if (ret) {
+ dev_err_probe(dev, ret, "failed to add power domain provider\n");
+ goto cleanup_pds;
+ }
+
+ bc->power_nb.notifier_call = bc_data->power_notifier_fn;
+ ret = dev_pm_genpd_add_notifier(bc->bus_power_dev, &bc->power_nb);
+ if (ret) {
+ dev_err_probe(dev, ret, "failed to add power notifier\n");
+ goto cleanup_provider;
+ }
+
+ dev_set_drvdata(dev, bc);
+
+ return 0;
+
+cleanup_provider:
+ of_genpd_del_provider(dev->of_node);
+cleanup_pds:
+ for (i--; i >= 0; i--) {
+ pm_genpd_remove(&bc->domains[i].genpd);
+ dev_pm_domain_detach(bc->domains[i].power_dev, true);
+ }
+
+ dev_pm_domain_detach(bc->bus_power_dev, true);
+
+ return ret;
+}
+
+static int imx8m_blk_ctrl_remove(struct platform_device *pdev)
+{
+ struct imx8m_blk_ctrl *bc = dev_get_drvdata(&pdev->dev);
+ int i;
+
+ of_genpd_del_provider(pdev->dev.of_node);
+
+ for (i = 0; bc->onecell_data.num_domains; i++) {
+ struct imx8m_blk_ctrl_domain *domain = &bc->domains[i];
+
+ pm_genpd_remove(&domain->genpd);
+ dev_pm_domain_detach(domain->power_dev, true);
+ }
+
+ dev_pm_genpd_remove_notifier(bc->bus_power_dev);
+
+ dev_pm_domain_detach(bc->bus_power_dev, true);
+
+ return 0;
+}
+
+#ifdef CONFIG_PM_SLEEP
+static int imx8m_blk_ctrl_suspend(struct device *dev)
+{
+ struct imx8m_blk_ctrl *bc = dev_get_drvdata(dev);
+ int ret, i;
+
+ /*
+ * This may look strange, but is done so the generic PM_SLEEP code
+ * can power down our domains and more importantly power them up again
+ * after resume, without tripping over our usage of runtime PM to
+ * control the upstream GPC domains. Things happen in the right order
+ * in the system suspend/resume paths due to the device parent/child
+ * hierarchy.
+ */
+ ret = pm_runtime_get_sync(bc->bus_power_dev);
+ if (ret < 0) {
+ pm_runtime_put_noidle(bc->bus_power_dev);
+ return ret;
+ }
+
+ for (i = 0; i < bc->onecell_data.num_domains; i++) {
+ struct imx8m_blk_ctrl_domain *domain = &bc->domains[i];
+
+ ret = pm_runtime_get_sync(domain->power_dev);
+ if (ret < 0) {
+ pm_runtime_put_noidle(domain->power_dev);
+ goto out_fail;
+ }
+ }
+
+ return 0;
+
+out_fail:
+ for (i--; i >= 0; i--)
+ pm_runtime_put(bc->domains[i].power_dev);
+
+ pm_runtime_put(bc->bus_power_dev);
+
+ return ret;
+}
+
+static int imx8m_blk_ctrl_resume(struct device *dev)
+{
+ struct imx8m_blk_ctrl *bc = dev_get_drvdata(dev);
+ int i;
+
+ for (i = 0; i < bc->onecell_data.num_domains; i++)
+ pm_runtime_put(bc->domains[i].power_dev);
+
+ pm_runtime_put(bc->bus_power_dev);
+
+ return 0;
+}
+#endif
+
+static const struct dev_pm_ops imx8m_blk_ctrl_pm_ops = {
+ SET_SYSTEM_SLEEP_PM_OPS(imx8m_blk_ctrl_suspend, imx8m_blk_ctrl_resume)
+};
+
+static int imx8mm_vpu_power_notifier(struct notifier_block *nb,
+ unsigned long action, void *data)
+{
+ struct imx8m_blk_ctrl *bc = container_of(nb, struct imx8m_blk_ctrl,
+ power_nb);
+
+ if (action != GENPD_NOTIFY_ON && action != GENPD_NOTIFY_PRE_OFF)
+ return NOTIFY_OK;
+
+ /*
+ * The ADB in the VPUMIX domain has no separate reset and clock
+ * enable bits, but is ungated together with the VPU clocks. To
+ * allow the handshake with the GPC to progress we put the VPUs
+ * in reset and ungate the clocks.
+ */
+ regmap_clear_bits(bc->regmap, BLK_SFT_RSTN, BIT(0) | BIT(1) | BIT(2));
+ regmap_set_bits(bc->regmap, BLK_CLK_EN, BIT(0) | BIT(1) | BIT(2));
+
+ if (action == GENPD_NOTIFY_ON) {
+ /*
+ * On power up we have no software backchannel to the GPC to
+ * wait for the ADB handshake to happen, so we just delay for a
+ * bit. On power down the GPC driver waits for the handshake.
+ */
+ udelay(5);
+
+ /* set "fuse" bits to enable the VPUs */
+ regmap_set_bits(bc->regmap, 0x8, 0xffffffff);
+ regmap_set_bits(bc->regmap, 0xc, 0xffffffff);
+ regmap_set_bits(bc->regmap, 0x10, 0xffffffff);
+ regmap_set_bits(bc->regmap, 0x14, 0xffffffff);
+ }
+
+ return NOTIFY_OK;
+}
+
+static const struct imx8m_blk_ctrl_domain_data imx8m_vpu_blk_ctl_domain_data[] = {
+ [IMX8MM_VPUBLK_PD_G1] = {
+ .name = "vpublk-g1",
+ .clk_names = (const char *[]){ "g1", },
+ .num_clks = 1,
+ .gpc_name = "g1",
+ .rst_mask = BIT(1),
+ .clk_mask = BIT(1),
+ },
+ [IMX8MM_VPUBLK_PD_G2] = {
+ .name = "vpublk-g2",
+ .clk_names = (const char *[]){ "g2", },
+ .num_clks = 1,
+ .gpc_name = "g2",
+ .rst_mask = BIT(0),
+ .clk_mask = BIT(0),
+ },
+ [IMX8MM_VPUBLK_PD_H1] = {
+ .name = "vpublk-h1",
+ .clk_names = (const char *[]){ "h1", },
+ .num_clks = 1,
+ .gpc_name = "h1",
+ .rst_mask = BIT(2),
+ .clk_mask = BIT(2),
+ },
+};
+
+static const struct imx8m_blk_ctrl_data imx8m_vpu_blk_ctl_dev_data = {
+ .max_reg = 0x18,
+ .power_notifier_fn = imx8mm_vpu_power_notifier,
+ .domains = imx8m_vpu_blk_ctl_domain_data,
+ .num_domains = ARRAY_SIZE(imx8m_vpu_blk_ctl_domain_data),
+};
+
+static const struct of_device_id imx8m_blk_ctrl_of_match[] = {
+ {
+ .compatible = "fsl,imx8mm-vpu-blk-ctrl",
+ .data = &imx8m_vpu_blk_ctl_dev_data
+ }, {
+ /* Sentinel */
+ }
+};
+MODULE_DEVICE_TABLE(of, imx8m_blk_ctrl_of_match);
+
+static struct platform_driver imx8m_blk_ctrl_driver = {
+ .probe = imx8m_blk_ctrl_probe,
+ .remove = imx8m_blk_ctrl_remove,
+ .driver = {
+ .name = "imx8m-blk-ctrl",
+ .pm = &imx8m_blk_ctrl_pm_ops,
+ .of_match_table = imx8m_blk_ctrl_of_match,
+ },
+};
+module_platform_driver(imx8m_blk_ctrl_driver);
From: Rob Herring <robh@kernel.org> Date: 2021-09-16 20:26:04
On Fri, 10 Sep 2021 22:26:31 +0200, Lucas Stach wrote:
This adds the defines for the power domains provided by the VPU
blk-ctrl.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
include/dt-bindings/power/imx8mm-power.h | 4 ++++
1 file changed, 4 insertions(+)
From: Rob Herring <robh@kernel.org> Date: 2021-09-16 20:29:43
On Fri, 10 Sep 2021 22:26:34 +0200, Lucas Stach wrote:
This adds the defines for the power domains provided by the DISP
blk-ctrl.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
v4: Change naming to be consistent with the VPUBLK domains.
---
include/dt-bindings/power/imx8mm-power.h | 5 +++++
1 file changed, 5 insertions(+)
From: Tim Harvey <tharvey@gateworks.com> Date: 2021-10-01 23:07:29
On Fri, Sep 10, 2021 at 1:26 PM Lucas Stach [off-list ref] wrote:
quoted hunk
Add the DT node for the GPC, including all the PGC power domains,
some of them are not fully functional yet, as they require interaction
with the blk-ctrls to properly power up/down the peripherals.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
arch/arm64/boot/dts/freescale/imx8mm.dtsi | 107 ++++++++++++++++++++++
1 file changed, 107 insertions(+)
Lucas,
I've been using your 'i.MX8MM GPC improvements and BLK_CTRL driver'
series for imx8mm-venice* and imx8mn-venice* boards. Thank you for
this work and I hope to see it merged soon!
I have found that on the imx8mm-venice-gw7901 board which does not use
MIPI and thus does not connect VDD_MIPI_1P8, VDD_MIPI_1P2,
VDD_MIPI_0P9, MIPI_VREG_CAP pins on the IMX8MM hangs with this
particular patch. If I comment out the pgc_mipi domain and subsequent
disp_blk_ctrl node from a later patch it resolves the hang. Is this
behavior expected and what would your recommendation be to work around
it?
Best regards,
Tim
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-10-01 23:20:41
Hi Tim,
Am Freitag, dem 01.10.2021 um 16:07 -0700 schrieb Tim Harvey:
On Fri, Sep 10, 2021 at 1:26 PM Lucas Stach [off-list ref] wrote:
quoted
Add the DT node for the GPC, including all the PGC power domains,
some of them are not fully functional yet, as they require interaction
with the blk-ctrls to properly power up/down the peripherals.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
arch/arm64/boot/dts/freescale/imx8mm.dtsi | 107 ++++++++++++++++++++++
1 file changed, 107 insertions(+)
Lucas,
I've been using your 'i.MX8MM GPC improvements and BLK_CTRL driver'
series for imx8mm-venice* and imx8mn-venice* boards. Thank you for
this work and I hope to see it merged soon!
I have found that on the imx8mm-venice-gw7901 board which does not use
MIPI and thus does not connect VDD_MIPI_1P8, VDD_MIPI_1P2,
VDD_MIPI_0P9, MIPI_VREG_CAP pins on the IMX8MM hangs with this
particular patch. If I comment out the pgc_mipi domain and subsequent
disp_blk_ctrl node from a later patch it resolves the hang. Is this
behavior expected and what would your recommendation be to work around
it?
No, this isn't expected. If there are no active devices in the MIPI
domain, the power domain should not be touched, as we treat all of them
as disabled initially. If we don't touch the domain I would expect that
the power supply not being present shouldn't be an issue.
Can you check if something in your system causes this power domain to
be powered up? Easiest way is probably to sprinkle a
printk("%s\n, genpd->name) in both imx8m_blk_ctrl_power_on() and
imx_gpc_power_on().
Regards,
Lucas
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
From: Tim Harvey <tharvey@gateworks.com> Date: 2021-10-02 00:06:29
On Fri, Oct 1, 2021 at 4:20 PM Lucas Stach [off-list ref] wrote:
Hi Tim,
Am Freitag, dem 01.10.2021 um 16:07 -0700 schrieb Tim Harvey:
quoted
On Fri, Sep 10, 2021 at 1:26 PM Lucas Stach [off-list ref] wrote:
quoted
Add the DT node for the GPC, including all the PGC power domains,
some of them are not fully functional yet, as they require interaction
with the blk-ctrls to properly power up/down the peripherals.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
arch/arm64/boot/dts/freescale/imx8mm.dtsi | 107 ++++++++++++++++++++++
1 file changed, 107 insertions(+)
Lucas,
I've been using your 'i.MX8MM GPC improvements and BLK_CTRL driver'
series for imx8mm-venice* and imx8mn-venice* boards. Thank you for
this work and I hope to see it merged soon!
I have found that on the imx8mm-venice-gw7901 board which does not use
MIPI and thus does not connect VDD_MIPI_1P8, VDD_MIPI_1P2,
VDD_MIPI_0P9, MIPI_VREG_CAP pins on the IMX8MM hangs with this
particular patch. If I comment out the pgc_mipi domain and subsequent
disp_blk_ctrl node from a later patch it resolves the hang. Is this
behavior expected and what would your recommendation be to work around
it?
No, this isn't expected. If there are no active devices in the MIPI
domain, the power domain should not be touched, as we treat all of them
as disabled initially. If we don't touch the domain I would expect that
the power supply not being present shouldn't be an issue.
Can you check if something in your system causes this power domain to
be powered up? Easiest way is probably to sprinkle a
printk("%s\n, genpd->name) in both imx8m_blk_ctrl_power_on() and
imx_gpc_power_on().
Lucas,
Here's what I see before I hang (debug print on both power on/off
followed by a msleep(1000) to make sure I see it before I hang):
[ 0.518319] imx_pgc_power_up hsiomix
[ 0.624031] imx_pgc_power_down hsiomix
[ 0.731879] imx_pgc_power_up hsiomix
[ 0.839906] imx_pgc_power_down hsiomix
[ 0.947875] imx_pgc_power_up hsiomix
[ 1.055859] imx_pgc_power_down hsiomix
[ 1.057296] imx_pgc_power_up gpumix
[ 1.167884] imx_pgc_power_down gpumix
[ 0.518513] imx_pgc_power_up hsiomix
[ 0.519933] imx_pgc_power_up gpumix
Tim
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
From: Tim Harvey <tharvey@gateworks.com> Date: 2021-10-02 00:15:49
On Fri, Oct 1, 2021 at 5:06 PM Tim Harvey [off-list ref] wrote:
On Fri, Oct 1, 2021 at 4:20 PM Lucas Stach [off-list ref] wrote:
quoted
Hi Tim,
Am Freitag, dem 01.10.2021 um 16:07 -0700 schrieb Tim Harvey:
quoted
On Fri, Sep 10, 2021 at 1:26 PM Lucas Stach [off-list ref] wrote:
quoted
Add the DT node for the GPC, including all the PGC power domains,
some of them are not fully functional yet, as they require interaction
with the blk-ctrls to properly power up/down the peripherals.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
arch/arm64/boot/dts/freescale/imx8mm.dtsi | 107 ++++++++++++++++++++++
1 file changed, 107 insertions(+)
Lucas,
I've been using your 'i.MX8MM GPC improvements and BLK_CTRL driver'
series for imx8mm-venice* and imx8mn-venice* boards. Thank you for
this work and I hope to see it merged soon!
I have found that on the imx8mm-venice-gw7901 board which does not use
MIPI and thus does not connect VDD_MIPI_1P8, VDD_MIPI_1P2,
VDD_MIPI_0P9, MIPI_VREG_CAP pins on the IMX8MM hangs with this
particular patch. If I comment out the pgc_mipi domain and subsequent
disp_blk_ctrl node from a later patch it resolves the hang. Is this
behavior expected and what would your recommendation be to work around
it?
No, this isn't expected. If there are no active devices in the MIPI
domain, the power domain should not be touched, as we treat all of them
as disabled initially. If we don't touch the domain I would expect that
the power supply not being present shouldn't be an issue.
Can you check if something in your system causes this power domain to
be powered up? Easiest way is probably to sprinkle a
printk("%s\n, genpd->name) in both imx8m_blk_ctrl_power_on() and
imx_gpc_power_on().
Lucas,
Here's what I see before I hang (debug print on both power on/off
followed by a msleep(1000) to make sure I see it before I hang):
[ 0.518319] imx_pgc_power_up hsiomix
[ 0.624031] imx_pgc_power_down hsiomix
[ 0.731879] imx_pgc_power_up hsiomix
[ 0.839906] imx_pgc_power_down hsiomix
[ 0.947875] imx_pgc_power_up hsiomix
[ 1.055859] imx_pgc_power_down hsiomix
[ 1.057296] imx_pgc_power_up gpumix
[ 1.167884] imx_pgc_power_down gpumix
[ 0.518513] imx_pgc_power_up hsiomix
[ 0.519933] imx_pgc_power_up gpumix
The board also has IMX8MM VDD_GPU pins not connected so it makes sense
that we hang here I suppose. Yet if I add the folloiwng to
imx8mm-venice-gw7901.dts it still tries to enable them and hangs:
&gpu_2d {
status = "disabled";
};
&gpu_3d {
status = "disabled";
};
&vpu_blk_ctrl {
status = "disabled";
};
Tim
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-10-02 00:25:16
Am Freitag, dem 01.10.2021 um 17:15 -0700 schrieb Tim Harvey:
On Fri, Oct 1, 2021 at 5:06 PM Tim Harvey [off-list ref] wrote:
quoted
On Fri, Oct 1, 2021 at 4:20 PM Lucas Stach [off-list ref] wrote:
quoted
Hi Tim,
Am Freitag, dem 01.10.2021 um 16:07 -0700 schrieb Tim Harvey:
quoted
On Fri, Sep 10, 2021 at 1:26 PM Lucas Stach [off-list ref] wrote:
quoted
Add the DT node for the GPC, including all the PGC power domains,
some of them are not fully functional yet, as they require interaction
with the blk-ctrls to properly power up/down the peripherals.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
arch/arm64/boot/dts/freescale/imx8mm.dtsi | 107 ++++++++++++++++++++++
1 file changed, 107 insertions(+)
Lucas,
I've been using your 'i.MX8MM GPC improvements and BLK_CTRL driver'
series for imx8mm-venice* and imx8mn-venice* boards. Thank you for
this work and I hope to see it merged soon!
I have found that on the imx8mm-venice-gw7901 board which does not use
MIPI and thus does not connect VDD_MIPI_1P8, VDD_MIPI_1P2,
VDD_MIPI_0P9, MIPI_VREG_CAP pins on the IMX8MM hangs with this
particular patch. If I comment out the pgc_mipi domain and subsequent
disp_blk_ctrl node from a later patch it resolves the hang. Is this
behavior expected and what would your recommendation be to work around
it?
No, this isn't expected. If there are no active devices in the MIPI
domain, the power domain should not be touched, as we treat all of them
as disabled initially. If we don't touch the domain I would expect that
the power supply not being present shouldn't be an issue.
Can you check if something in your system causes this power domain to
be powered up? Easiest way is probably to sprinkle a
printk("%s\n, genpd->name) in both imx8m_blk_ctrl_power_on() and
imx_gpc_power_on().
Lucas,
Here's what I see before I hang (debug print on both power on/off
followed by a msleep(1000) to make sure I see it before I hang):
[ 0.518319] imx_pgc_power_up hsiomix
[ 0.624031] imx_pgc_power_down hsiomix
[ 0.731879] imx_pgc_power_up hsiomix
[ 0.839906] imx_pgc_power_down hsiomix
[ 0.947875] imx_pgc_power_up hsiomix
[ 1.055859] imx_pgc_power_down hsiomix
[ 1.057296] imx_pgc_power_up gpumix
[ 1.167884] imx_pgc_power_down gpumix
[ 0.518513] imx_pgc_power_up hsiomix
[ 0.519933] imx_pgc_power_up gpumix
The board also has IMX8MM VDD_GPU pins not connected so it makes sense
that we hang here I suppose. Yet if I add the folloiwng to
imx8mm-venice-gw7901.dts it still tries to enable them and hangs:
&gpu_2d {
status = "disabled";
};
&gpu_3d {
status = "disabled";
};
&vpu_blk_ctrl {
status = "disabled";
};
The pgc_gpu is a "active" consumer of the pgc_gpumix domain while the
driver gets probed, so the driver core will power up the gpumix domain
for a moment during kernel init. To avoid this you must at least set
the status of the pgc_gpu node to disabled.
Regards,
Lucas
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-10-02 01:07:41
Hi Benjamin,
Am Dienstag, dem 14.09.2021 um 17:46 +0200 schrieb Benjamin Gaignard:
Le 10/09/2021 à 22:26, Lucas Stach a écrit :
quoted
This adds a driver for the blk-ctrl blocks found in the i.MX8M* line of
SoCs. The blk-ctrl is a top-level peripheral located in the various *MIX
power domains and interacts with the GPC power controller to provide the
peripherals in the power domain access to the NoC and ensures that those
peripherals are properly reset when their respective power domain is
brought back to life.
Software needs to do different things to make the bus handshake happen
after the GPC *MIX domain is powered up and before it is powered down.
As the requirements are quite different between the various blk-ctrls
there is a callback function provided to hook in the proper sequence.
The peripheral domains are quite uniform, they handle the soft clock
enables and resets in the blk-ctrl address space and sequencing with the
upstream GPC power domains.
Hi Lucas,
I have tried to use your patches for IMX8MQ but it seems that the hardware
have different architecture.
On IMX8MQ there is only one VPU domain for G1 and G2 and that doesn't match
with your implementation where it is needed to have "bus" and devices power domain.
From what I experiment in current IMX8MQ implementation of blk-ctrl (inside VPU driver)
enabling the 3 clocks (bus, G1, G2) is needed to reset the VPUs.
Do you think you can update your design to take care of these hardware variations ?
The clocking/reset of the blk-ctrl and ADB in the i.MX8MQ VPU power
domain is really a bit strange, as the ADB reset is tied to the VPU
resets and the clk-ctrl seem to require all 3 VPU clocks, instead of
only the bus clock as in newer designs. However I was able to make it
work with the existing blk-ctrl driver design.
My current WIP patches (only tested with the G1 core so far) on top of
the v5 of the series I just sent out can be found here:
https://git.pengutronix.de/cgit/lst/linux/log/?h=imx8mq-vpu-blk-ctrl
Hope this helps.
Regards,
Lucas
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
From: Tim Harvey <tharvey@gateworks.com> Date: 2021-10-02 02:44:01
'On Fri, Oct 1, 2021 at 5:25 PM Lucas Stach [off-list ref] wrote:
Am Freitag, dem 01.10.2021 um 17:15 -0700 schrieb Tim Harvey:
quoted
On Fri, Oct 1, 2021 at 5:06 PM Tim Harvey [off-list ref] wrote:
quoted
On Fri, Oct 1, 2021 at 4:20 PM Lucas Stach [off-list ref] wrote:
quoted
Hi Tim,
Am Freitag, dem 01.10.2021 um 16:07 -0700 schrieb Tim Harvey:
quoted
On Fri, Sep 10, 2021 at 1:26 PM Lucas Stach [off-list ref] wrote:
quoted
Add the DT node for the GPC, including all the PGC power domains,
some of them are not fully functional yet, as they require interaction
with the blk-ctrls to properly power up/down the peripherals.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
arch/arm64/boot/dts/freescale/imx8mm.dtsi | 107 ++++++++++++++++++++++
1 file changed, 107 insertions(+)
Lucas,
I've been using your 'i.MX8MM GPC improvements and BLK_CTRL driver'
series for imx8mm-venice* and imx8mn-venice* boards. Thank you for
this work and I hope to see it merged soon!
I have found that on the imx8mm-venice-gw7901 board which does not use
MIPI and thus does not connect VDD_MIPI_1P8, VDD_MIPI_1P2,
VDD_MIPI_0P9, MIPI_VREG_CAP pins on the IMX8MM hangs with this
particular patch. If I comment out the pgc_mipi domain and subsequent
disp_blk_ctrl node from a later patch it resolves the hang. Is this
behavior expected and what would your recommendation be to work around
it?
No, this isn't expected. If there are no active devices in the MIPI
domain, the power domain should not be touched, as we treat all of them
as disabled initially. If we don't touch the domain I would expect that
the power supply not being present shouldn't be an issue.
Can you check if something in your system causes this power domain to
be powered up? Easiest way is probably to sprinkle a
printk("%s\n, genpd->name) in both imx8m_blk_ctrl_power_on() and
imx_gpc_power_on().
Lucas,
Here's what I see before I hang (debug print on both power on/off
followed by a msleep(1000) to make sure I see it before I hang):
[ 0.518319] imx_pgc_power_up hsiomix
[ 0.624031] imx_pgc_power_down hsiomix
[ 0.731879] imx_pgc_power_up hsiomix
[ 0.839906] imx_pgc_power_down hsiomix
[ 0.947875] imx_pgc_power_up hsiomix
[ 1.055859] imx_pgc_power_down hsiomix
[ 1.057296] imx_pgc_power_up gpumix
[ 1.167884] imx_pgc_power_down gpumix
[ 0.518513] imx_pgc_power_up hsiomix
[ 0.519933] imx_pgc_power_up gpumix
The board also has IMX8MM VDD_GPU pins not connected so it makes sense
that we hang here I suppose. Yet if I add the folloiwng to
imx8mm-venice-gw7901.dts it still tries to enable them and hangs:
&gpu_2d {
status = "disabled";
};
&gpu_3d {
status = "disabled";
};
&vpu_blk_ctrl {
status = "disabled";
};
The pgc_gpu is a "active" consumer of the pgc_gpumix domain while the
driver gets probed, so the driver core will power up the gpumix domain
for a moment during kernel init. To avoid this you must at least set
the status of the pgc_gpu node to disabled.
I've tried that and it doesn't work either.
&gpu_2d {
status = "disabled";
};
&gpu_3d {
status = "disabled";
};
&vpu_blk_ctrl {
status = "disabled";
};
&pgc_gpumix {
status = "disabled";
};
&pgc_gpu {
status = "disabled";
};
The interesting thing is that the first patch that causes this is
'arm64: dts: imx8mm: add GPC node' which just adds the gpc node and is
before the addition of the other nodes. Therefore these are being
enabled by default regardless of the above nodes being disabled (or
not even added yet to imx8mm.dtsi).
Tim
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-10-02 12:52:05
Hi Tim,
Am Freitag, dem 01.10.2021 um 19:43 -0700 schrieb Tim Harvey:
quoted
quoted
quoted
quoted
[...]
quoted
quoted
quoted
quoted
quoted
Lucas,
I've been using your 'i.MX8MM GPC improvements and BLK_CTRL driver'
series for imx8mm-venice* and imx8mn-venice* boards. Thank you for
this work and I hope to see it merged soon!
I have found that on the imx8mm-venice-gw7901 board which does not use
MIPI and thus does not connect VDD_MIPI_1P8, VDD_MIPI_1P2,
VDD_MIPI_0P9, MIPI_VREG_CAP pins on the IMX8MM hangs with this
particular patch. If I comment out the pgc_mipi domain and subsequent
disp_blk_ctrl node from a later patch it resolves the hang. Is this
behavior expected and what would your recommendation be to work around
it?
No, this isn't expected. If there are no active devices in the MIPI
domain, the power domain should not be touched, as we treat all of them
as disabled initially. If we don't touch the domain I would expect that
the power supply not being present shouldn't be an issue.
Can you check if something in your system causes this power domain to
be powered up? Easiest way is probably to sprinkle a
printk("%s\n, genpd->name) in both imx8m_blk_ctrl_power_on() and
imx_gpc_power_on().
Lucas,
Here's what I see before I hang (debug print on both power on/off
followed by a msleep(1000) to make sure I see it before I hang):
[ 0.518319] imx_pgc_power_up hsiomix
[ 0.624031] imx_pgc_power_down hsiomix
[ 0.731879] imx_pgc_power_up hsiomix
[ 0.839906] imx_pgc_power_down hsiomix
[ 0.947875] imx_pgc_power_up hsiomix
[ 1.055859] imx_pgc_power_down hsiomix
[ 1.057296] imx_pgc_power_up gpumix
[ 1.167884] imx_pgc_power_down gpumix
[ 0.518513] imx_pgc_power_up hsiomix
[ 0.519933] imx_pgc_power_up gpumix
The board also has IMX8MM VDD_GPU pins not connected so it makes sense
that we hang here I suppose. Yet if I add the folloiwng to
imx8mm-venice-gw7901.dts it still tries to enable them and hangs:
&gpu_2d {
status = "disabled";
};
&gpu_3d {
status = "disabled";
};
&vpu_blk_ctrl {
status = "disabled";
};
The pgc_gpu is a "active" consumer of the pgc_gpumix domain while the
driver gets probed, so the driver core will power up the gpumix domain
for a moment during kernel init. To avoid this you must at least set
the status of the pgc_gpu node to disabled.
I've tried that and it doesn't work either.
&gpu_2d {
status = "disabled";
};
&gpu_3d {
status = "disabled";
};
&vpu_blk_ctrl {
status = "disabled";
};
&pgc_gpumix {
status = "disabled";
};
&pgc_gpu {
status = "disabled";
};
The interesting thing is that the first patch that causes this is
'arm64: dts: imx8mm: add GPC node' which just adds the gpc node and is
before the addition of the other nodes. Therefore these are being
enabled by default regardless of the above nodes being disabled (or
not even added yet to imx8mm.dtsi).
My bad, I didn't think about the fact that the power domain devices are
not instantiated via the common OF populate code. For the status
property to work as expected the GPCv2 code needs to check this
property. I've just sent a patch to make this happen. Can you give it
another try with your DT changes and this patch applied?
Regards,
Lucas
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
From: Tim Harvey <tharvey@gateworks.com> Date: 2021-10-03 17:17:27
On Sat, Oct 2, 2021 at 5:51 AM Lucas Stach [off-list ref] wrote:
Hi Tim,
Am Freitag, dem 01.10.2021 um 19:43 -0700 schrieb Tim Harvey:
quoted
quoted
quoted
quoted
quoted
[...]
quoted
quoted
quoted
quoted
quoted
quoted
Lucas,
I've been using your 'i.MX8MM GPC improvements and BLK_CTRL driver'
series for imx8mm-venice* and imx8mn-venice* boards. Thank you for
this work and I hope to see it merged soon!
I have found that on the imx8mm-venice-gw7901 board which does not use
MIPI and thus does not connect VDD_MIPI_1P8, VDD_MIPI_1P2,
VDD_MIPI_0P9, MIPI_VREG_CAP pins on the IMX8MM hangs with this
particular patch. If I comment out the pgc_mipi domain and subsequent
disp_blk_ctrl node from a later patch it resolves the hang. Is this
behavior expected and what would your recommendation be to work around
it?
No, this isn't expected. If there are no active devices in the MIPI
domain, the power domain should not be touched, as we treat all of them
as disabled initially. If we don't touch the domain I would expect that
the power supply not being present shouldn't be an issue.
Can you check if something in your system causes this power domain to
be powered up? Easiest way is probably to sprinkle a
printk("%s\n, genpd->name) in both imx8m_blk_ctrl_power_on() and
imx_gpc_power_on().
Lucas,
Here's what I see before I hang (debug print on both power on/off
followed by a msleep(1000) to make sure I see it before I hang):
[ 0.518319] imx_pgc_power_up hsiomix
[ 0.624031] imx_pgc_power_down hsiomix
[ 0.731879] imx_pgc_power_up hsiomix
[ 0.839906] imx_pgc_power_down hsiomix
[ 0.947875] imx_pgc_power_up hsiomix
[ 1.055859] imx_pgc_power_down hsiomix
[ 1.057296] imx_pgc_power_up gpumix
[ 1.167884] imx_pgc_power_down gpumix
[ 0.518513] imx_pgc_power_up hsiomix
[ 0.519933] imx_pgc_power_up gpumix
The board also has IMX8MM VDD_GPU pins not connected so it makes sense
that we hang here I suppose. Yet if I add the folloiwng to
imx8mm-venice-gw7901.dts it still tries to enable them and hangs:
&gpu_2d {
status = "disabled";
};
&gpu_3d {
status = "disabled";
};
&vpu_blk_ctrl {
status = "disabled";
};
The pgc_gpu is a "active" consumer of the pgc_gpumix domain while the
driver gets probed, so the driver core will power up the gpumix domain
for a moment during kernel init. To avoid this you must at least set
the status of the pgc_gpu node to disabled.
I've tried that and it doesn't work either.
&gpu_2d {
status = "disabled";
};
&gpu_3d {
status = "disabled";
};
&vpu_blk_ctrl {
status = "disabled";
};
&pgc_gpumix {
status = "disabled";
};
&pgc_gpu {
status = "disabled";
};
The interesting thing is that the first patch that causes this is
'arm64: dts: imx8mm: add GPC node' which just adds the gpc node and is
before the addition of the other nodes. Therefore these are being
enabled by default regardless of the above nodes being disabled (or
not even added yet to imx8mm.dtsi).
My bad, I didn't think about the fact that the power domain devices are
not instantiated via the common OF populate code. For the status
property to work as expected the GPCv2 code needs to check this
property. I've just sent a patch to make this happen. Can you give it
another try with your DT changes and this patch applied?
Yes, this makes sense and resolves it with only disabling pgc_gpumix in my dts.
I believe it is fine for me to not specifically disable gpu_2d and
gpu_3d as they are not enabled by default in imx8mm.dtsi and I don't
believe that will change and I don't 'currently' need to disable
disp_blk_ctrl or pgc_mipi but wondering if I should as the MIPI rails
are not hooked up either? (ie will something added in the future try
to use them?)
What needs to be done to get this series merged? I suppose it's too
late to get it into 5.13?
Best regards,
Tim
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-10-03 19:44:48
Am Sonntag, dem 03.10.2021 um 10:17 -0700 schrieb Tim Harvey:
On Sat, Oct 2, 2021 at 5:51 AM Lucas Stach [off-list ref] wrote:
quoted
Hi Tim,
Am Freitag, dem 01.10.2021 um 19:43 -0700 schrieb Tim Harvey:
quoted
quoted
quoted
quoted
quoted
[...]
quoted
quoted
quoted
quoted
quoted
quoted
Lucas,
I've been using your 'i.MX8MM GPC improvements and BLK_CTRL driver'
series for imx8mm-venice* and imx8mn-venice* boards. Thank you for
this work and I hope to see it merged soon!
I have found that on the imx8mm-venice-gw7901 board which does not use
MIPI and thus does not connect VDD_MIPI_1P8, VDD_MIPI_1P2,
VDD_MIPI_0P9, MIPI_VREG_CAP pins on the IMX8MM hangs with this
particular patch. If I comment out the pgc_mipi domain and subsequent
disp_blk_ctrl node from a later patch it resolves the hang. Is this
behavior expected and what would your recommendation be to work around
it?
No, this isn't expected. If there are no active devices in the MIPI
domain, the power domain should not be touched, as we treat all of them
as disabled initially. If we don't touch the domain I would expect that
the power supply not being present shouldn't be an issue.
Can you check if something in your system causes this power domain to
be powered up? Easiest way is probably to sprinkle a
printk("%s\n, genpd->name) in both imx8m_blk_ctrl_power_on() and
imx_gpc_power_on().
Lucas,
Here's what I see before I hang (debug print on both power on/off
followed by a msleep(1000) to make sure I see it before I hang):
[ 0.518319] imx_pgc_power_up hsiomix
[ 0.624031] imx_pgc_power_down hsiomix
[ 0.731879] imx_pgc_power_up hsiomix
[ 0.839906] imx_pgc_power_down hsiomix
[ 0.947875] imx_pgc_power_up hsiomix
[ 1.055859] imx_pgc_power_down hsiomix
[ 1.057296] imx_pgc_power_up gpumix
[ 1.167884] imx_pgc_power_down gpumix
[ 0.518513] imx_pgc_power_up hsiomix
[ 0.519933] imx_pgc_power_up gpumix
The board also has IMX8MM VDD_GPU pins not connected so it makes sense
that we hang here I suppose. Yet if I add the folloiwng to
imx8mm-venice-gw7901.dts it still tries to enable them and hangs:
&gpu_2d {
status = "disabled";
};
&gpu_3d {
status = "disabled";
};
&vpu_blk_ctrl {
status = "disabled";
};
The pgc_gpu is a "active" consumer of the pgc_gpumix domain while the
driver gets probed, so the driver core will power up the gpumix domain
for a moment during kernel init. To avoid this you must at least set
the status of the pgc_gpu node to disabled.
I've tried that and it doesn't work either.
&gpu_2d {
status = "disabled";
};
&gpu_3d {
status = "disabled";
};
&vpu_blk_ctrl {
status = "disabled";
};
&pgc_gpumix {
status = "disabled";
};
&pgc_gpu {
status = "disabled";
};
The interesting thing is that the first patch that causes this is
'arm64: dts: imx8mm: add GPC node' which just adds the gpc node and is
before the addition of the other nodes. Therefore these are being
enabled by default regardless of the above nodes being disabled (or
not even added yet to imx8mm.dtsi).
My bad, I didn't think about the fact that the power domain devices are
not instantiated via the common OF populate code. For the status
property to work as expected the GPCv2 code needs to check this
property. I've just sent a patch to make this happen. Can you give it
another try with your DT changes and this patch applied?
Yes, this makes sense and resolves it with only disabling pgc_gpumix in my dts.
I believe it is fine for me to not specifically disable gpu_2d and
gpu_3d as they are not enabled by default in imx8mm.dtsi and I don't
believe that will change and I don't 'currently' need to disable
disp_blk_ctrl or pgc_mipi but wondering if I should as the MIPI rails
are not hooked up either? (ie will something added in the future try
to use them?)
I think it would be better to disable the GPU devices in your DT. They
don't have a status property and thus are enabled by default. They
won't probe without the power domain being available, but it will cause
those devices to stay in probe deferred state forever, which may
confuse some users. So I guess it would be good style to disable the
GPU nodes.
One consequence of disabling the MIPI power domain is that the disp-
blk-ctrl driver won't probe anymore, as it needs all the referenced
power domains to be available. Do you think this might be an issue? Is
there a valid system configuration where you would use devices from
from the display power domain, but not the MIPI domain? If that's a
valid case we might need to make this configuration possible in the
disp-blk-ctrl driver.
What needs to be done to get this series merged? I suppose it's too
late to get it into 5.13?
5.13? Way too late, even the 5.15 merge window is long closed. The
series is almost completely reviewed and the DT bindings are also good
to go. So if no one happens to run into any real showstopper, Shawn may
be willing to pick up the series and stage it for the 5.16 merge
window.
Regards,
Lucas
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
From: Tim Harvey <tharvey@gateworks.com> Date: 2021-10-03 21:21:47
On Sun, Oct 3, 2021 at 12:44 PM Lucas Stach [off-list ref] wrote:
Am Sonntag, dem 03.10.2021 um 10:17 -0700 schrieb Tim Harvey:
quoted
On Sat, Oct 2, 2021 at 5:51 AM Lucas Stach [off-list ref] wrote:
quoted
Hi Tim,
Am Freitag, dem 01.10.2021 um 19:43 -0700 schrieb Tim Harvey:
quoted
quoted
quoted
quoted
quoted
[...]
quoted
quoted
quoted
quoted
quoted
quoted
Lucas,
I've been using your 'i.MX8MM GPC improvements and BLK_CTRL driver'
series for imx8mm-venice* and imx8mn-venice* boards. Thank you for
this work and I hope to see it merged soon!
I have found that on the imx8mm-venice-gw7901 board which does not use
MIPI and thus does not connect VDD_MIPI_1P8, VDD_MIPI_1P2,
VDD_MIPI_0P9, MIPI_VREG_CAP pins on the IMX8MM hangs with this
particular patch. If I comment out the pgc_mipi domain and subsequent
disp_blk_ctrl node from a later patch it resolves the hang. Is this
behavior expected and what would your recommendation be to work around
it?
No, this isn't expected. If there are no active devices in the MIPI
domain, the power domain should not be touched, as we treat all of them
as disabled initially. If we don't touch the domain I would expect that
the power supply not being present shouldn't be an issue.
Can you check if something in your system causes this power domain to
be powered up? Easiest way is probably to sprinkle a
printk("%s\n, genpd->name) in both imx8m_blk_ctrl_power_on() and
imx_gpc_power_on().
Lucas,
Here's what I see before I hang (debug print on both power on/off
followed by a msleep(1000) to make sure I see it before I hang):
[ 0.518319] imx_pgc_power_up hsiomix
[ 0.624031] imx_pgc_power_down hsiomix
[ 0.731879] imx_pgc_power_up hsiomix
[ 0.839906] imx_pgc_power_down hsiomix
[ 0.947875] imx_pgc_power_up hsiomix
[ 1.055859] imx_pgc_power_down hsiomix
[ 1.057296] imx_pgc_power_up gpumix
[ 1.167884] imx_pgc_power_down gpumix
[ 0.518513] imx_pgc_power_up hsiomix
[ 0.519933] imx_pgc_power_up gpumix
The board also has IMX8MM VDD_GPU pins not connected so it makes sense
that we hang here I suppose. Yet if I add the folloiwng to
imx8mm-venice-gw7901.dts it still tries to enable them and hangs:
&gpu_2d {
status = "disabled";
};
&gpu_3d {
status = "disabled";
};
&vpu_blk_ctrl {
status = "disabled";
};
The pgc_gpu is a "active" consumer of the pgc_gpumix domain while the
driver gets probed, so the driver core will power up the gpumix domain
for a moment during kernel init. To avoid this you must at least set
the status of the pgc_gpu node to disabled.
I've tried that and it doesn't work either.
&gpu_2d {
status = "disabled";
};
&gpu_3d {
status = "disabled";
};
&vpu_blk_ctrl {
status = "disabled";
};
&pgc_gpumix {
status = "disabled";
};
&pgc_gpu {
status = "disabled";
};
The interesting thing is that the first patch that causes this is
'arm64: dts: imx8mm: add GPC node' which just adds the gpc node and is
before the addition of the other nodes. Therefore these are being
enabled by default regardless of the above nodes being disabled (or
not even added yet to imx8mm.dtsi).
My bad, I didn't think about the fact that the power domain devices are
not instantiated via the common OF populate code. For the status
property to work as expected the GPCv2 code needs to check this
property. I've just sent a patch to make this happen. Can you give it
another try with your DT changes and this patch applied?
Yes, this makes sense and resolves it with only disabling pgc_gpumix in my dts.
I believe it is fine for me to not specifically disable gpu_2d and
gpu_3d as they are not enabled by default in imx8mm.dtsi and I don't
believe that will change and I don't 'currently' need to disable
disp_blk_ctrl or pgc_mipi but wondering if I should as the MIPI rails
are not hooked up either? (ie will something added in the future try
to use them?)
I think it would be better to disable the GPU devices in your DT. They
don't have a status property and thus are enabled by default. They
won't probe without the power domain being available, but it will cause
those devices to stay in probe deferred state forever, which may
confuse some users. So I guess it would be good style to disable the
GPU nodes.
it still tries to power the pgc_gpumix domain if I 'only' disable
gpu_2d and gpu_3d. I have to disable 'pgc_gpumix' specifically - is
that what you expect here?
One consequence of disabling the MIPI power domain is that the disp-
blk-ctrl driver won't probe anymore, as it needs all the referenced
power domains to be available. Do you think this might be an issue? Is
there a valid system configuration where you would use devices from
from the display power domain, but not the MIPI domain? If that's a
valid case we might need to make this configuration possible in the
disp-blk-ctrl driver.
In the case of imx8mm I don't see that you can have any display
without MIPI however the imx8mp can have HDMI display without needing
MIPI so in general that should be allowed.
quoted
What needs to be done to get this series merged? I suppose it's too
late to get it into 5.13?
5.13? Way too late, even the 5.15 merge window is long closed. The
series is almost completely reviewed and the DT bindings are also good
to go. So if no one happens to run into any real showstopper, Shawn may
be willing to pick up the series and stage it for the 5.16 merge
window.
oops... I meant 5.15 but yes, seems too late for that. We seem to have
time to get this into 5.16 however. I'm not sure what Shawn's cut-off
is there.
Tim
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-10-04 07:44:39
Am Sonntag, dem 03.10.2021 um 14:21 -0700 schrieb Tim Harvey:
On Sun, Oct 3, 2021 at 12:44 PM Lucas Stach [off-list ref] wrote:
quoted
Am Sonntag, dem 03.10.2021 um 10:17 -0700 schrieb Tim Harvey:
quoted
On Sat, Oct 2, 2021 at 5:51 AM Lucas Stach [off-list ref] wrote:
quoted
Hi Tim,
Am Freitag, dem 01.10.2021 um 19:43 -0700 schrieb Tim Harvey:
quoted
quoted
quoted
quoted
quoted
[...]
quoted
quoted
quoted
quoted
quoted
quoted
Lucas,
I've been using your 'i.MX8MM GPC improvements and BLK_CTRL driver'
series for imx8mm-venice* and imx8mn-venice* boards. Thank you for
this work and I hope to see it merged soon!
I have found that on the imx8mm-venice-gw7901 board which does not use
MIPI and thus does not connect VDD_MIPI_1P8, VDD_MIPI_1P2,
VDD_MIPI_0P9, MIPI_VREG_CAP pins on the IMX8MM hangs with this
particular patch. If I comment out the pgc_mipi domain and subsequent
disp_blk_ctrl node from a later patch it resolves the hang. Is this
behavior expected and what would your recommendation be to work around
it?
No, this isn't expected. If there are no active devices in the MIPI
domain, the power domain should not be touched, as we treat all of them
as disabled initially. If we don't touch the domain I would expect that
the power supply not being present shouldn't be an issue.
Can you check if something in your system causes this power domain to
be powered up? Easiest way is probably to sprinkle a
printk("%s\n, genpd->name) in both imx8m_blk_ctrl_power_on() and
imx_gpc_power_on().
Lucas,
Here's what I see before I hang (debug print on both power on/off
followed by a msleep(1000) to make sure I see it before I hang):
[ 0.518319] imx_pgc_power_up hsiomix
[ 0.624031] imx_pgc_power_down hsiomix
[ 0.731879] imx_pgc_power_up hsiomix
[ 0.839906] imx_pgc_power_down hsiomix
[ 0.947875] imx_pgc_power_up hsiomix
[ 1.055859] imx_pgc_power_down hsiomix
[ 1.057296] imx_pgc_power_up gpumix
[ 1.167884] imx_pgc_power_down gpumix
[ 0.518513] imx_pgc_power_up hsiomix
[ 0.519933] imx_pgc_power_up gpumix
The board also has IMX8MM VDD_GPU pins not connected so it makes sense
that we hang here I suppose. Yet if I add the folloiwng to
imx8mm-venice-gw7901.dts it still tries to enable them and hangs:
&gpu_2d {
status = "disabled";
};
&gpu_3d {
status = "disabled";
};
&vpu_blk_ctrl {
status = "disabled";
};
The pgc_gpu is a "active" consumer of the pgc_gpumix domain while the
driver gets probed, so the driver core will power up the gpumix domain
for a moment during kernel init. To avoid this you must at least set
the status of the pgc_gpu node to disabled.
I've tried that and it doesn't work either.
&gpu_2d {
status = "disabled";
};
&gpu_3d {
status = "disabled";
};
&vpu_blk_ctrl {
status = "disabled";
};
&pgc_gpumix {
status = "disabled";
};
&pgc_gpu {
status = "disabled";
};
The interesting thing is that the first patch that causes this is
'arm64: dts: imx8mm: add GPC node' which just adds the gpc node and is
before the addition of the other nodes. Therefore these are being
enabled by default regardless of the above nodes being disabled (or
not even added yet to imx8mm.dtsi).
My bad, I didn't think about the fact that the power domain devices are
not instantiated via the common OF populate code. For the status
property to work as expected the GPCv2 code needs to check this
property. I've just sent a patch to make this happen. Can you give it
another try with your DT changes and this patch applied?
Yes, this makes sense and resolves it with only disabling pgc_gpumix in my dts.
I believe it is fine for me to not specifically disable gpu_2d and
gpu_3d as they are not enabled by default in imx8mm.dtsi and I don't
believe that will change and I don't 'currently' need to disable
disp_blk_ctrl or pgc_mipi but wondering if I should as the MIPI rails
are not hooked up either? (ie will something added in the future try
to use them?)
I think it would be better to disable the GPU devices in your DT. They
don't have a status property and thus are enabled by default. They
won't probe without the power domain being available, but it will cause
those devices to stay in probe deferred state forever, which may
confuse some users. So I guess it would be good style to disable the
GPU nodes.
it still tries to power the pgc_gpumix domain if I 'only' disable
gpu_2d and gpu_3d. I have to disable 'pgc_gpumix' specifically - is
that what you expect here?
quoted
One consequence of disabling the MIPI power domain is that the disp-
blk-ctrl driver won't probe anymore, as it needs all the referenced
power domains to be available. Do you think this might be an issue? Is
there a valid system configuration where you would use devices from
from the display power domain, but not the MIPI domain? If that's a
valid case we might need to make this configuration possible in the
disp-blk-ctrl driver.
In the case of imx8mm I don't see that you can have any display
without MIPI however the imx8mp can have HDMI display without needing
MIPI so in general that should be allowed.
I was thinking about the camera input path, but this one also needs the
MIPI domain. The camera interface is useless without the MIPI PHY to
get data from. The i.MX8MP is a different topic, as the HDMI part is a
separate blk-ctrl domain there.
With that in mind, I think things are okay as they are implemented now,
as I don't see any use-case where one would want to use devices from
the DISP domain (CSI, LCDIF) without having the MIPI PHY parts
available.
Regards,
Lucas
quoted
quoted
What needs to be done to get this series merged? I suppose it's too
late to get it into 5.13?
5.13? Way too late, even the 5.15 merge window is long closed. The
series is almost completely reviewed and the DT bindings are also good
to go. So if no one happens to run into any real showstopper, Shawn may
be willing to pick up the series and stage it for the 5.16 merge
window.
oops... I meant 5.15 but yes, seems too late for that. We seem to have
time to get this into 5.16 however. I'm not sure what Shawn's cut-off
is there.
Tim
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-10-04 07:48:12
Am Sonntag, dem 03.10.2021 um 14:21 -0700 schrieb Tim Harvey:
On Sun, Oct 3, 2021 at 12:44 PM Lucas Stach [off-list ref] wrote:
quoted
Am Sonntag, dem 03.10.2021 um 10:17 -0700 schrieb Tim Harvey:
quoted
On Sat, Oct 2, 2021 at 5:51 AM Lucas Stach [off-list ref] wrote:
quoted
Hi Tim,
Am Freitag, dem 01.10.2021 um 19:43 -0700 schrieb Tim Harvey:
quoted
quoted
quoted
quoted
quoted
[...]
quoted
quoted
quoted
quoted
quoted
quoted
Lucas,
I've been using your 'i.MX8MM GPC improvements and BLK_CTRL driver'
series for imx8mm-venice* and imx8mn-venice* boards. Thank you for
this work and I hope to see it merged soon!
I have found that on the imx8mm-venice-gw7901 board which does not use
MIPI and thus does not connect VDD_MIPI_1P8, VDD_MIPI_1P2,
VDD_MIPI_0P9, MIPI_VREG_CAP pins on the IMX8MM hangs with this
particular patch. If I comment out the pgc_mipi domain and subsequent
disp_blk_ctrl node from a later patch it resolves the hang. Is this
behavior expected and what would your recommendation be to work around
it?
No, this isn't expected. If there are no active devices in the MIPI
domain, the power domain should not be touched, as we treat all of them
as disabled initially. If we don't touch the domain I would expect that
the power supply not being present shouldn't be an issue.
Can you check if something in your system causes this power domain to
be powered up? Easiest way is probably to sprinkle a
printk("%s\n, genpd->name) in both imx8m_blk_ctrl_power_on() and
imx_gpc_power_on().
Lucas,
Here's what I see before I hang (debug print on both power on/off
followed by a msleep(1000) to make sure I see it before I hang):
[ 0.518319] imx_pgc_power_up hsiomix
[ 0.624031] imx_pgc_power_down hsiomix
[ 0.731879] imx_pgc_power_up hsiomix
[ 0.839906] imx_pgc_power_down hsiomix
[ 0.947875] imx_pgc_power_up hsiomix
[ 1.055859] imx_pgc_power_down hsiomix
[ 1.057296] imx_pgc_power_up gpumix
[ 1.167884] imx_pgc_power_down gpumix
[ 0.518513] imx_pgc_power_up hsiomix
[ 0.519933] imx_pgc_power_up gpumix
The board also has IMX8MM VDD_GPU pins not connected so it makes sense
that we hang here I suppose. Yet if I add the folloiwng to
imx8mm-venice-gw7901.dts it still tries to enable them and hangs:
&gpu_2d {
status = "disabled";
};
&gpu_3d {
status = "disabled";
};
&vpu_blk_ctrl {
status = "disabled";
};
The pgc_gpu is a "active" consumer of the pgc_gpumix domain while the
driver gets probed, so the driver core will power up the gpumix domain
for a moment during kernel init. To avoid this you must at least set
the status of the pgc_gpu node to disabled.
I've tried that and it doesn't work either.
&gpu_2d {
status = "disabled";
};
&gpu_3d {
status = "disabled";
};
&vpu_blk_ctrl {
status = "disabled";
};
&pgc_gpumix {
status = "disabled";
};
&pgc_gpu {
status = "disabled";
};
The interesting thing is that the first patch that causes this is
'arm64: dts: imx8mm: add GPC node' which just adds the gpc node and is
before the addition of the other nodes. Therefore these are being
enabled by default regardless of the above nodes being disabled (or
not even added yet to imx8mm.dtsi).
My bad, I didn't think about the fact that the power domain devices are
not instantiated via the common OF populate code. For the status
property to work as expected the GPCv2 code needs to check this
property. I've just sent a patch to make this happen. Can you give it
another try with your DT changes and this patch applied?
Yes, this makes sense and resolves it with only disabling pgc_gpumix in my dts.
I believe it is fine for me to not specifically disable gpu_2d and
gpu_3d as they are not enabled by default in imx8mm.dtsi and I don't
believe that will change and I don't 'currently' need to disable
disp_blk_ctrl or pgc_mipi but wondering if I should as the MIPI rails
are not hooked up either? (ie will something added in the future try
to use them?)
I think it would be better to disable the GPU devices in your DT. They
don't have a status property and thus are enabled by default. They
won't probe without the power domain being available, but it will cause
those devices to stay in probe deferred state forever, which may
confuse some users. So I guess it would be good style to disable the
GPU nodes.
it still tries to power the pgc_gpumix domain if I 'only' disable
gpu_2d and gpu_3d. I have to disable 'pgc_gpumix' specifically - is
that what you expect here?
Yes, that's expected. As I said earlier, the gpu domain is considered a
active child device of the gpumix domain if it isn't disabled in DT, so
the driver core will try to power up the gpumix domain before the gpu
domain is probed.
You must disable the power domains in DT to avoid them ever being
touched on your board. You should disable the GPU devices depending on
this power domain to avoid the GPU devices staying in probe deferred
limbo forever.
Regards,
Lucas
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
From: Benjamin Gaignard <benjamin.gaignard@collabora.com> Date: 2021-10-04 14:27:53
Le 02/10/2021 à 03:07, Lucas Stach a écrit :
Hi Benjamin,
Am Dienstag, dem 14.09.2021 um 17:46 +0200 schrieb Benjamin Gaignard:
quoted
Le 10/09/2021 à 22:26, Lucas Stach a écrit :
quoted
This adds a driver for the blk-ctrl blocks found in the i.MX8M* line of
SoCs. The blk-ctrl is a top-level peripheral located in the various *MIX
power domains and interacts with the GPC power controller to provide the
peripherals in the power domain access to the NoC and ensures that those
peripherals are properly reset when their respective power domain is
brought back to life.
Software needs to do different things to make the bus handshake happen
after the GPC *MIX domain is powered up and before it is powered down.
As the requirements are quite different between the various blk-ctrls
there is a callback function provided to hook in the proper sequence.
The peripheral domains are quite uniform, they handle the soft clock
enables and resets in the blk-ctrl address space and sequencing with the
upstream GPC power domains.
Hi Lucas,
I have tried to use your patches for IMX8MQ but it seems that the hardware
have different architecture.
On IMX8MQ there is only one VPU domain for G1 and G2 and that doesn't match
with your implementation where it is needed to have "bus" and devices power domain.
From what I experiment in current IMX8MQ implementation of blk-ctrl (inside VPU driver)
enabling the 3 clocks (bus, G1, G2) is needed to reset the VPUs.
Do you think you can update your design to take care of these hardware variations ?
The clocking/reset of the blk-ctrl and ADB in the i.MX8MQ VPU power
domain is really a bit strange, as the ADB reset is tied to the VPU
resets and the clk-ctrl seem to require all 3 VPU clocks, instead of
only the bus clock as in newer designs. However I was able to make it
work with the existing blk-ctrl driver design.
My current WIP patches (only tested with the G1 core so far) on top of
the v5 of the series I just sent out can be found here:
https://git.pengutronix.de/cgit/lst/linux/log/?h=imx8mq-vpu-blk-ctrl
Hope this helps.
Hi Lucas,
I have been able to test your branch on my iMX8MQ.
I confirm that G1 is working fine, I able to decode H264 files.
I wasn't able to make G2 works, I think it is coming from the reset sequence
done before each frame decoding in G2 driver.
I have change imx8mq_runtime_resume() and imx8m_vpu_reset()
to call pm_runtime_put() and pm_runtime_get() to perform a reset like.
Without that G2 hangs when decoding the first frame.
One G1 it seems that doing a reset before each frame decoding is not needed.
On DT I had to assignee G1 and G2 on the both nodes to avoid a warning at probe time.
assigned-clocks = <&clk IMX8MQ_CLK_VPU_G1>,
<&clk IMX8MQ_CLK_VPU_G2>,
<&clk IMX8MQ_VPU_PLL_BYPASS>;
assigned-clock-parents = <&clk IMX8MQ_VPU_PLL_OUT>,
<&clk IMX8MQ_VPU_PLL_OUT>,
<&clk IMX8MQ_VPU_PLL>;
assigned-clock-rates = <600000000>, <300000000>, <0>;
I also set G2 clock at 300Mhz as specify in the TRM.
Even with all this G2 doesn't fire interrupts.
Benjamin
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-10-05 10:04:03
Hi Benjamin,
Am Montag, dem 04.10.2021 um 16:27 +0200 schrieb Benjamin Gaignard:
Le 02/10/2021 à 03:07, Lucas Stach a écrit :
quoted
Hi Benjamin,
Am Dienstag, dem 14.09.2021 um 17:46 +0200 schrieb Benjamin Gaignard:
quoted
Le 10/09/2021 à 22:26, Lucas Stach a écrit :
quoted
This adds a driver for the blk-ctrl blocks found in the i.MX8M* line of
SoCs. The blk-ctrl is a top-level peripheral located in the various *MIX
power domains and interacts with the GPC power controller to provide the
peripherals in the power domain access to the NoC and ensures that those
peripherals are properly reset when their respective power domain is
brought back to life.
Software needs to do different things to make the bus handshake happen
after the GPC *MIX domain is powered up and before it is powered down.
As the requirements are quite different between the various blk-ctrls
there is a callback function provided to hook in the proper sequence.
The peripheral domains are quite uniform, they handle the soft clock
enables and resets in the blk-ctrl address space and sequencing with the
upstream GPC power domains.
Hi Lucas,
I have tried to use your patches for IMX8MQ but it seems that the hardware
have different architecture.
On IMX8MQ there is only one VPU domain for G1 and G2 and that doesn't match
with your implementation where it is needed to have "bus" and devices power domain.
From what I experiment in current IMX8MQ implementation of blk-ctrl (inside VPU driver)
enabling the 3 clocks (bus, G1, G2) is needed to reset the VPUs.
Do you think you can update your design to take care of these hardware variations ?
The clocking/reset of the blk-ctrl and ADB in the i.MX8MQ VPU power
domain is really a bit strange, as the ADB reset is tied to the VPU
resets and the clk-ctrl seem to require all 3 VPU clocks, instead of
only the bus clock as in newer designs. However I was able to make it
work with the existing blk-ctrl driver design.
My current WIP patches (only tested with the G1 core so far) on top of
the v5 of the series I just sent out can be found here:
https://git.pengutronix.de/cgit/lst/linux/log/?h=imx8mq-vpu-blk-ctrl
Hope this helps.
Hi Lucas,
I have been able to test your branch on my iMX8MQ.
I confirm that G1 is working fine, I able to decode H264 files.
I wasn't able to make G2 works, I think it is coming from the reset sequence
done before each frame decoding in G2 driver.
I have change imx8mq_runtime_resume() and imx8m_vpu_reset()
to call pm_runtime_put() and pm_runtime_get() to perform a reset like.
I think you need to use the _sync variants of those functions to make
sure the domain is going through the reset state. However that seems
like a pretty heavy-weight thing to do if the decoder really requires a
reset before each frame. Excuse my ignorance about the G2 block, but
this sounds like a quite odd requirement. Is this to work around some
hardware erratum?
If we really need to reset the G2 before each frame, I think it would
be best to also expose a reset controller from the blk-ctrl driver, to
allow the G2 driver to do a light-weight reset, instead of doing this
runtime PM transition.
Regards,
Lucas
Without that G2 hangs when decoding the first frame.
One G1 it seems that doing a reset before each frame decoding is not needed.
On DT I had to assignee G1 and G2 on the both nodes to avoid a warning at probe time.
assigned-clocks = <&clk IMX8MQ_CLK_VPU_G1>,
<&clk IMX8MQ_CLK_VPU_G2>,
<&clk IMX8MQ_VPU_PLL_BYPASS>;
assigned-clock-parents = <&clk IMX8MQ_VPU_PLL_OUT>,
<&clk IMX8MQ_VPU_PLL_OUT>,
<&clk IMX8MQ_VPU_PLL>;
assigned-clock-rates = <600000000>, <300000000>, <0>;
I also set G2 clock at 300Mhz as specify in the TRM.
Even with all this G2 doesn't fire interrupts.
Benjamin
From: Benjamin Gaignard <benjamin.gaignard@collabora.com> Date: 2021-10-05 12:37:08
Le 05/10/2021 à 12:03, Lucas Stach a écrit :
Hi Benjamin,
Am Montag, dem 04.10.2021 um 16:27 +0200 schrieb Benjamin Gaignard:
quoted
Le 02/10/2021 à 03:07, Lucas Stach a écrit :
quoted
Hi Benjamin,
Am Dienstag, dem 14.09.2021 um 17:46 +0200 schrieb Benjamin Gaignard:
quoted
Le 10/09/2021 à 22:26, Lucas Stach a écrit :
quoted
This adds a driver for the blk-ctrl blocks found in the i.MX8M* line of
SoCs. The blk-ctrl is a top-level peripheral located in the various *MIX
power domains and interacts with the GPC power controller to provide the
peripherals in the power domain access to the NoC and ensures that those
peripherals are properly reset when their respective power domain is
brought back to life.
Software needs to do different things to make the bus handshake happen
after the GPC *MIX domain is powered up and before it is powered down.
As the requirements are quite different between the various blk-ctrls
there is a callback function provided to hook in the proper sequence.
The peripheral domains are quite uniform, they handle the soft clock
enables and resets in the blk-ctrl address space and sequencing with the
upstream GPC power domains.
Hi Lucas,
I have tried to use your patches for IMX8MQ but it seems that the hardware
have different architecture.
On IMX8MQ there is only one VPU domain for G1 and G2 and that doesn't match
with your implementation where it is needed to have "bus" and devices power domain.
From what I experiment in current IMX8MQ implementation of blk-ctrl (inside VPU driver)
enabling the 3 clocks (bus, G1, G2) is needed to reset the VPUs.
Do you think you can update your design to take care of these hardware variations ?
The clocking/reset of the blk-ctrl and ADB in the i.MX8MQ VPU power
domain is really a bit strange, as the ADB reset is tied to the VPU
resets and the clk-ctrl seem to require all 3 VPU clocks, instead of
only the bus clock as in newer designs. However I was able to make it
work with the existing blk-ctrl driver design.
My current WIP patches (only tested with the G1 core so far) on top of
the v5 of the series I just sent out can be found here:
https://git.pengutronix.de/cgit/lst/linux/log/?h=imx8mq-vpu-blk-ctrl
Hope this helps.
Hi Lucas,
I have been able to test your branch on my iMX8MQ.
I confirm that G1 is working fine, I able to decode H264 files.
I wasn't able to make G2 works, I think it is coming from the reset sequence
done before each frame decoding in G2 driver.
I have change imx8mq_runtime_resume() and imx8m_vpu_reset()
to call pm_runtime_put() and pm_runtime_get() to perform a reset like.
I think you need to use the _sync variants of those functions to make
sure the domain is going through the reset state. However that seems
like a pretty heavy-weight thing to do if the decoder really requires a
reset before each frame. Excuse my ignorance about the G2 block, but
this sounds like a quite odd requirement. Is this to work around some
hardware erratum?
When using _sync variants kernel stop when probing G2.
I don't have any documentation about the link between the control block and
G2. I have done lot of tests and resetting the hardware block between each frame
seems to be mandatory. That also the case in Hantro proprietary stack.
If we really need to reset the G2 before each frame, I think it would
be best to also expose a reset controller from the blk-ctrl driver, to
allow the G2 driver to do a light-weight reset, instead of doing this
runtime PM transition.
Without that G2 hangs when decoding the first frame.
One G1 it seems that doing a reset before each frame decoding is not needed.
On DT I had to assignee G1 and G2 on the both nodes to avoid a warning at probe time.
assigned-clocks = <&clk IMX8MQ_CLK_VPU_G1>,
<&clk IMX8MQ_CLK_VPU_G2>,
<&clk IMX8MQ_VPU_PLL_BYPASS>;
assigned-clock-parents = <&clk IMX8MQ_VPU_PLL_OUT>,
<&clk IMX8MQ_VPU_PLL_OUT>,
<&clk IMX8MQ_VPU_PLL>;
assigned-clock-rates = <600000000>, <300000000>, <0>;
I also set G2 clock at 300Mhz as specify in the TRM.
Even with all this G2 doesn't fire interrupts.
Benjamin
From: Benjamin Gaignard <benjamin.gaignard@collabora.com> Date: 2021-10-13 12:12:42
Le 05/10/2021 à 14:36, Benjamin Gaignard a écrit :
Le 05/10/2021 à 12:03, Lucas Stach a écrit :
quoted
Hi Benjamin,
Am Montag, dem 04.10.2021 um 16:27 +0200 schrieb Benjamin Gaignard:
quoted
Le 02/10/2021 à 03:07, Lucas Stach a écrit :
quoted
Hi Benjamin,
Am Dienstag, dem 14.09.2021 um 17:46 +0200 schrieb Benjamin Gaignard:
quoted
Le 10/09/2021 à 22:26, Lucas Stach a écrit :
quoted
This adds a driver for the blk-ctrl blocks found in the i.MX8M*
line of
SoCs. The blk-ctrl is a top-level peripheral located in the
various *MIX
power domains and interacts with the GPC power controller to
provide the
peripherals in the power domain access to the NoC and ensures
that those
peripherals are properly reset when their respective power domain is
brought back to life.
Software needs to do different things to make the bus handshake
happen
after the GPC *MIX domain is powered up and before it is powered
down.
As the requirements are quite different between the various
blk-ctrls
there is a callback function provided to hook in the proper
sequence.
The peripheral domains are quite uniform, they handle the soft clock
enables and resets in the blk-ctrl address space and sequencing
with the
upstream GPC power domains.
Hi Lucas,
I have tried to use your patches for IMX8MQ but it seems that the
hardware
have different architecture.
On IMX8MQ there is only one VPU domain for G1 and G2 and that
doesn't match
with your implementation where it is needed to have "bus" and
devices power domain.
From what I experiment in current IMX8MQ implementation of
blk-ctrl (inside VPU driver)
enabling the 3 clocks (bus, G1, G2) is needed to reset the VPUs.
Do you think you can update your design to take care of these
hardware variations ?
The clocking/reset of the blk-ctrl and ADB in the i.MX8MQ VPU power
domain is really a bit strange, as the ADB reset is tied to the VPU
resets and the clk-ctrl seem to require all 3 VPU clocks, instead of
only the bus clock as in newer designs. However I was able to make it
work with the existing blk-ctrl driver design.
My current WIP patches (only tested with the G1 core so far) on top of
the v5 of the series I just sent out can be found here:
https://git.pengutronix.de/cgit/lst/linux/log/?h=imx8mq-vpu-blk-ctrl
Hope this helps.
Hi Lucas,
I have been able to test your branch on my iMX8MQ.
I confirm that G1 is working fine, I able to decode H264 files.
I wasn't able to make G2 works, I think it is coming from the reset
sequence
done before each frame decoding in G2 driver.
I have change imx8mq_runtime_resume() and imx8m_vpu_reset()
to call pm_runtime_put() and pm_runtime_get() to perform a reset like.
I think you need to use the _sync variants of those functions to make
sure the domain is going through the reset state. However that seems
like a pretty heavy-weight thing to do if the decoder really requires a
reset before each frame. Excuse my ignorance about the G2 block, but
this sounds like a quite odd requirement. Is this to work around some
hardware erratum?
When using _sync variants kernel stop when probing G2.
I don't have any documentation about the link between the control
block and
G2. I have done lot of tests and resetting the hardware block between
each frame
seems to be mandatory. That also the case in Hantro proprietary stack.
quoted
If we really need to reset the G2 before each frame, I think it would
be best to also expose a reset controller from the blk-ctrl driver, to
allow the G2 driver to do a light-weight reset, instead of doing this
runtime PM transition.
Hi Lucas,
I have done more tests with the following condition:
- no implementation of pm_runtime_* in the IMX part of the driver (i.e. resume and reset functions are empty)
=> so power domain on/off are called only when the delay set for autosuspend expires not for all frames.
I have added traces and the board in hanging after calling imx8mq_vpu_power_notifier() when going off.
Before that I have the power on sequence, few interrupts from G2 and power off occur because of the delay.
If I reset and enable G1 and G2 clocks in imx8mq_vpu_power_notifier() I can see a call to codec run() but that
hang too.
I have also try to enable all the clocks (bus, G1, G2) when using only G2 but the results are the same.
I don't if that could help you.
If you want me to do more tests please tell me.
Regards,
Benjamin
Regards,
Benjamin
quoted
Regards,
Lucas
quoted
Without that G2 hangs when decoding the first frame.
One G1 it seems that doing a reset before each frame decoding is not
needed.
On DT I had to assignee G1 and G2 on the both nodes to avoid a
warning at probe time.
assigned-clocks = <&clk IMX8MQ_CLK_VPU_G1>,
<&clk IMX8MQ_CLK_VPU_G2>,
<&clk IMX8MQ_VPU_PLL_BYPASS>;
assigned-clock-parents = <&clk IMX8MQ_VPU_PLL_OUT>,
<&clk IMX8MQ_VPU_PLL_OUT>,
<&clk IMX8MQ_VPU_PLL>;
assigned-clock-rates = <600000000>, <300000000>, <0>;
I also set G2 clock at 300Mhz as specify in the TRM.
Even with all this G2 doesn't fire interrupts.
Benjamin
From: Adam Ford <hidden> Date: 2021-11-07 21:08:42
On Fri, Sep 10, 2021 at 3:26 PM Lucas Stach [off-list ref] wrote:
This adds a driver for the blk-ctrl blocks found in the i.MX8M* line of
SoCs. The blk-ctrl is a top-level peripheral located in the various *MIX
power domains and interacts with the GPC power controller to provide the
peripherals in the power domain access to the NoC and ensures that those
peripherals are properly reset when their respective power domain is
brought back to life.
Software needs to do different things to make the bus handshake happen
after the GPC *MIX domain is powered up and before it is powered down.
As the requirements are quite different between the various blk-ctrls
there is a callback function provided to hook in the proper sequence.
The peripheral domains are quite uniform, they handle the soft clock
enables and resets in the blk-ctrl address space and sequencing with the
upstream GPC power domains.
Sorry to resurrect an old thread, but I'm running into issues enabling
new peripherals, and I think it's related to this, but I am not
certain.
I have attempted to enable the various VPU's, and in doing so, I
referenced the vpu_blk_ctrl for the respective power-domains reference
[1].
The various VPU devices enumerate, and I can see the various VPU power
domains turn on and off a boot.
[ 1.968565] genpd genpd:0:38330000.blk-ctrl: adding to PM domain vpumix
[ 1.981208] genpd genpd:1:38330000.blk-ctrl: adding to PM domain vpu-g1
[ 1.993838] genpd genpd:2:38330000.blk-ctrl: adding to PM domain vpu-g2
[ 2.006455] genpd genpd:3:38330000.blk-ctrl: adding to PM domain vpu-h1
[ 8.922661] hantro_vpu: module is from the staging directory, the
quality is unknown, you have been warned.
[ 9.008147] hantro_vpu: module is from the staging directory, the
quality is unknown, you have been warned.
[ 9.044041] hantro-vpu 38300000.video-codec: adding to PM domain vpublk-g1
[ 9.050959] hantro-vpu 38300000.video-codec: genpd_add_device()
[ 9.063349] PM: vpumix: Power-on latency exceeded, new value 50875 ns
[ 9.083437] PM: vpu-g1: Power-on latency exceeded, new value 26125 ns
[ 9.104778] PM: vpublk-g1: Power-on latency exceeded, new value 47819250 ns
[ 9.211988] hantro-vpu 38300000.video-codec: registered
nxp,imx8mm-vpu-dec as /dev/video0
[ 9.259680] hantro-vpu 38300000.video-codec: genpd_runtime_resume()
[ 9.297395] hantro-vpu 38300000.video-codec: resume latency exceeded, 2750 ns
[ 9.307767] hantro-vpu 38310000.video-codec: adding to PM domain vpublk-g2
[ 9.316462] hantro-vpu 38310000.video-codec: genpd_add_device()
[ 9.328807] PM: vpu-g2: Power-on latency exceeded, new value 26625 ns
[ 9.342401] PM: vpublk-g2: Power-on latency exceeded, new value 19965125 ns
[ 9.430254] hantro-vpu 38300000.video-codec: genpd_runtime_suspend()
[ 9.436683] hantro-vpu 38300000.video-codec: suspend latency
exceeded, 1625 ns
[ 9.443984] PM: vpublk-g1: Power-off latency exceeded, new value 16625 ns
[ 9.462361] hantro-vpu 38310000.video-codec: registered
nxp,imx8mm-vpu-g2-dec as /dev/video2
[ 9.491848] PM: vpu-g1: Power-off latency exceeded, new value 17125 ns
[ 9.506353] hantro-vpu 38310000.video-codec: genpd_runtime_resume()
[ 9.512735] hantro-vpu 38310000.video-codec: resume latency exceeded, 2750 ns
[ 9.520306] hantro-vpu 38320000.video-codec: adding to PM domain vpublk-h1
[ 9.527268] hantro-vpu 38320000.video-codec: genpd_add_device()
[ 9.539632] PM: vpu-h1: Power-on latency exceeded, new value 27750 ns
[ 9.553358] PM: vpublk-h1: Power-on latency exceeded, new value 20112125 ns
[ 9.605800] hantro_vpu: module is from the staging directory, the
quality is unknown, you have been warned.
[ 9.642194] hantro-vpu 38310000.video-codec: genpd_runtime_suspend()
[ 9.648634] hantro-vpu 38310000.video-codec: suspend latency
exceeded, 1500 ns
[ 9.655974] PM: vpublk-g2: Power-off latency exceeded, new value 12625 ns
[ 9.703863] PM: vpu-g2: Power-off latency exceeded, new value 16750 ns
[ 9.754527] hantro-vpu 38320000.video-codec: registered
nxp,imx8mm-vpu-h1-enc as /dev/video3
[ 9.785424] hantro-vpu 38320000.video-codec: genpd_runtime_resume()
[ 9.796034] hantro-vpu 38320000.video-codec: resume latency exceeded, 2000 ns
[ 9.934247] hantro-vpu 38320000.video-codec: genpd_runtime_suspend()
[ 9.940707] hantro-vpu 38320000.video-codec: suspend latency
exceeded, 1500 ns
[ 9.948042] PM: vpublk-h1: Power-off latency exceeded, new value 18875 ns
[ 9.975951] PM: vpu-h1: Power-off latency exceeded, new value 17625 ns
[ 9.999970] PM: vpumix: Power-off latency exceeded, new value 25750 ns
However, because the vpumix is disabled here, I think it might be
what's causing a hang when I attempt to read the regmap registers with
cat /sys/kernel/debug/regmap/38330000.blk-ctrl/registers
I was looking at the ATF code that NXP has, and it doesn't appear that
the VPUMIX ever shuts down, and for that matter, I don't think the g1,
g2, and h1 domains shutdown either.
I think it makes sense to have the domains turned off when not in use,
but I have a few comments / questions below....
quoted hunk
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Acked-by: Frieder Schrempf <redacted>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
---
This commit includes the full code to drive the VPUMIX domain on the
i.MX8MM, as the skeleton driver would probably be harder to review
without the context provided by one blk-ctrl implementation. Other
blk-ctrl implementations will follow, based on this overall structure.
v4:
- fix commit message typos
- fix superfluous whitespace
- constify clk_names more
---
drivers/soc/imx/Makefile | 1 +
drivers/soc/imx/imx8m-blk-ctrl.c | 453 +++++++++++++++++++++++++++++++
2 files changed, 454 insertions(+)
create mode 100644 drivers/soc/imx/imx8m-blk-ctrl.c
@@ -0,0 +1,453 @@+// SPDX-License-Identifier: GPL-2.0+++/*+*Copyright2021Pengutronix,LucasStach<kernel@pengutronix.de>+*/++#include<linux/device.h>+#include<linux/module.h>+#include<linux/of_device.h>+#include<linux/platform_device.h>+#include<linux/pm_domain.h>+#include<linux/pm_runtime.h>+#include<linux/regmap.h>+#include<linux/clk.h>++#include<dt-bindings/power/imx8mm-power.h>++#define BLK_SFT_RSTN 0x0+#define BLK_CLK_EN 0x4++structimx8m_blk_ctrl_domain;++structimx8m_blk_ctrl{+structdevice*dev;+structnotifier_blockpower_nb;+structdevice*bus_power_dev;+structregmap*regmap;+structimx8m_blk_ctrl_domain*domains;+structgenpd_onecell_dataonecell_data;+};++structimx8m_blk_ctrl_domain_data{+constchar*name;+constchar*const*clk_names;+intnum_clks;+constchar*gpc_name;+u32rst_mask;+u32clk_mask;+};++#define DOMAIN_MAX_CLKS 3++structimx8m_blk_ctrl_domain{+structgeneric_pm_domaingenpd;+conststructimx8m_blk_ctrl_domain_data*data;+structclk_bulk_dataclks[DOMAIN_MAX_CLKS];+structdevice*power_dev;+structimx8m_blk_ctrl*bc;+};++structimx8m_blk_ctrl_data{+intmax_reg;+notifier_fn_tpower_notifier_fn;+conststructimx8m_blk_ctrl_domain_data*domains;+intnum_domains;+};++staticinlinestructimx8m_blk_ctrl_domain*+to_imx8m_blk_ctrl_domain(structgeneric_pm_domain*genpd)+{+returncontainer_of(genpd,structimx8m_blk_ctrl_domain,genpd);+}++staticintimx8m_blk_ctrl_power_on(structgeneric_pm_domain*genpd)+{+structimx8m_blk_ctrl_domain*domain=to_imx8m_blk_ctrl_domain(genpd);+conststructimx8m_blk_ctrl_domain_data*data=domain->data;+structimx8m_blk_ctrl*bc=domain->bc;+intret;++/* make sure bus domain is awake */+ret=pm_runtime_get_sync(bc->bus_power_dev);+if(ret<0){+pm_runtime_put_noidle(bc->bus_power_dev);+dev_err(bc->dev,"failed to power up bus domain\n");+returnret;+}++/* put devices into reset */+regmap_clear_bits(bc->regmap,BLK_SFT_RSTN,data->rst_mask);++/* enable upstream and blk-ctrl clocks to allow reset to propagate */+ret=clk_bulk_prepare_enable(data->num_clks,domain->clks);+if(ret){+dev_err(bc->dev,"failed to enable clocks\n");+gotobus_put;+}+regmap_set_bits(bc->regmap,BLK_CLK_EN,data->clk_mask);++/* power up upstream GPC domain */+ret=pm_runtime_get_sync(domain->power_dev);+if(ret<0){+dev_err(bc->dev,"failed to power up peripheral domain\n");+gotoclk_disable;+}++/* wait for reset to propagate */+udelay(5);++/* release reset */+regmap_set_bits(bc->regmap,BLK_SFT_RSTN,data->rst_mask);++/* disable upstream clocks */+clk_bulk_disable_unprepare(data->num_clks,domain->clks);++return0;++clk_disable:+clk_bulk_disable_unprepare(data->num_clks,domain->clks);+bus_put:+pm_runtime_put(bc->bus_power_dev);++returnret;+}++staticintimx8m_blk_ctrl_power_off(structgeneric_pm_domain*genpd)+{+structimx8m_blk_ctrl_domain*domain=to_imx8m_blk_ctrl_domain(genpd);+conststructimx8m_blk_ctrl_domain_data*data=domain->data;+structimx8m_blk_ctrl*bc=domain->bc;++/* put devices into reset and disable clocks */+regmap_clear_bits(bc->regmap,BLK_SFT_RSTN,data->rst_mask);+regmap_clear_bits(bc->regmap,BLK_CLK_EN,data->clk_mask);++/* power down upstream GPC domain */+pm_runtime_put(domain->power_dev);++/* allow bus domain to suspend */+pm_runtime_put(bc->bus_power_dev);++return0;+}++staticstructgeneric_pm_domain*+imx8m_blk_ctrl_xlate(structof_phandle_args*args,void*data)+{+structgenpd_onecell_data*onecell_data=data;+unsignedintindex=args->args[0];++if(args->args_count!=1||+index>onecell_data->num_domains)+returnERR_PTR(-EINVAL);++returnonecell_data->domains[index];+}++staticstructlock_class_keyblk_ctrl_genpd_lock_class;++staticintimx8m_blk_ctrl_probe(structplatform_device*pdev)+{+conststructimx8m_blk_ctrl_data*bc_data;+structdevice*dev=&pdev->dev;+structimx8m_blk_ctrl*bc;+void__iomem*base;+inti,ret;++structregmap_configregmap_config={+.reg_bits=32,+.val_bits=32,+.reg_stride=4,+};++bc=devm_kzalloc(dev,sizeof(*bc),GFP_KERNEL);+if(!bc)+return-ENOMEM;++bc->dev=dev;++bc_data=of_device_get_match_data(dev);++base=devm_platform_ioremap_resource(pdev,0);+if(IS_ERR(base))+returnPTR_ERR(base);++regmap_config.max_register=bc_data->max_reg;+bc->regmap=devm_regmap_init_mmio(dev,base,®map_config);+if(IS_ERR(bc->regmap))+returndev_err_probe(dev,PTR_ERR(bc->regmap),+"failed to init regmap\n");++bc->domains=devm_kcalloc(dev,bc_data->num_domains,+sizeof(structimx8m_blk_ctrl_domain),+GFP_KERNEL);+if(!bc->domains)+return-ENOMEM;++bc->onecell_data.num_domains=bc_data->num_domains;+bc->onecell_data.xlate=imx8m_blk_ctrl_xlate;+bc->onecell_data.domains=+devm_kcalloc(dev,bc_data->num_domains,+sizeof(structgeneric_pm_domain*),GFP_KERNEL);+if(!bc->onecell_data.domains)+return-ENOMEM;++bc->bus_power_dev=genpd_dev_pm_attach_by_name(dev,"bus");+if(IS_ERR(bc->bus_power_dev))+returndev_err_probe(dev,PTR_ERR(bc->bus_power_dev),+"failed to attach power domain\n");++for(i=0;i<bc_data->num_domains;i++){+conststructimx8m_blk_ctrl_domain_data*data=&bc_data->domains[i];+structimx8m_blk_ctrl_domain*domain=&bc->domains[i];+intj;++domain->data=data;++for(j=0;j<data->num_clks;j++)+domain->clks[j].id=data->clk_names[j];++ret=devm_clk_bulk_get(dev,data->num_clks,domain->clks);+if(ret){+dev_err_probe(dev,ret,"failed to get clock\n");+gotocleanup_pds;+}++domain->power_dev=+dev_pm_domain_attach_by_name(dev,data->gpc_name);+if(IS_ERR(domain->power_dev)){+dev_err_probe(dev,PTR_ERR(domain->power_dev),+"failed to attach power domain\n");+ret=PTR_ERR(domain->power_dev);+gotocleanup_pds;+}++domain->genpd.name=data->name;+domain->genpd.power_on=imx8m_blk_ctrl_power_on;+domain->genpd.power_off=imx8m_blk_ctrl_power_off;+domain->bc=bc;++ret=pm_genpd_init(&domain->genpd,NULL,true);+if(ret){+dev_err_probe(dev,ret,"failed to init power domain\n");+dev_pm_domain_detach(domain->power_dev,true);+gotocleanup_pds;+}++/*+*WeuseruntimePMtotriggerpoweron/offoftheupstreamGPC+*domain,asastricthierarchicalparent/childpowerdomain+*setupdoesn'tallowustomeetthesequencingrequirements.+*Thismeanswehavenestedlockingofgenpdlocks,withoutthe+*nestingbeingvisibleatthegenpdlevel,soweneeda+*separatelockclasstomakelockdepawareofthefactthat+*thisareseparatedomainlocksthatcanbenestedwithouta+*self-deadlock.+*/+lockdep_set_class(&domain->genpd.mlock,+&blk_ctrl_genpd_lock_class);++bc->onecell_data.domains[i]=&domain->genpd;+}++ret=of_genpd_add_provider_onecell(dev->of_node,&bc->onecell_data);+if(ret){+dev_err_probe(dev,ret,"failed to add power domain provider\n");+gotocleanup_pds;+}++bc->power_nb.notifier_call=bc_data->power_notifier_fn;+ret=dev_pm_genpd_add_notifier(bc->bus_power_dev,&bc->power_nb);+if(ret){+dev_err_probe(dev,ret,"failed to add power notifier\n");+gotocleanup_provider;+}++dev_set_drvdata(dev,bc);++return0;++cleanup_provider:+of_genpd_del_provider(dev->of_node);+cleanup_pds:+for(i--;i>=0;i--){+pm_genpd_remove(&bc->domains[i].genpd);+dev_pm_domain_detach(bc->domains[i].power_dev,true);+}++dev_pm_domain_detach(bc->bus_power_dev,true);++returnret;+}++staticintimx8m_blk_ctrl_remove(structplatform_device*pdev)+{+structimx8m_blk_ctrl*bc=dev_get_drvdata(&pdev->dev);+inti;++of_genpd_del_provider(pdev->dev.of_node);++for(i=0;bc->onecell_data.num_domains;i++){+structimx8m_blk_ctrl_domain*domain=&bc->domains[i];++pm_genpd_remove(&domain->genpd);+dev_pm_domain_detach(domain->power_dev,true);+}++dev_pm_genpd_remove_notifier(bc->bus_power_dev);++dev_pm_domain_detach(bc->bus_power_dev,true);++return0;+}++#ifdef CONFIG_PM_SLEEP+staticintimx8m_blk_ctrl_suspend(structdevice*dev)+{+structimx8m_blk_ctrl*bc=dev_get_drvdata(dev);+intret,i;++/*+*Thismaylookstrange,butisdonesothegenericPM_SLEEPcode+*canpowerdownourdomainsandmoreimportantlypowerthemupagain+*afterresume,withouttrippingoverourusageofruntimePMto+*controltheupstreamGPCdomains.Thingshappenintherightorder+*inthesystemsuspend/resumepathsduetothedeviceparent/child+*hierarchy.+*/+ret=pm_runtime_get_sync(bc->bus_power_dev);+if(ret<0){+pm_runtime_put_noidle(bc->bus_power_dev);+returnret;+}++for(i=0;i<bc->onecell_data.num_domains;i++){+structimx8m_blk_ctrl_domain*domain=&bc->domains[i];++ret=pm_runtime_get_sync(domain->power_dev);+if(ret<0){+pm_runtime_put_noidle(domain->power_dev);+gotoout_fail;+}+}++return0;++out_fail:+for(i--;i>=0;i--)+pm_runtime_put(bc->domains[i].power_dev);++pm_runtime_put(bc->bus_power_dev);++returnret;+}++staticintimx8m_blk_ctrl_resume(structdevice*dev)+{+structimx8m_blk_ctrl*bc=dev_get_drvdata(dev);+inti;++for(i=0;i<bc->onecell_data.num_domains;i++)+pm_runtime_put(bc->domains[i].power_dev);++pm_runtime_put(bc->bus_power_dev);++return0;+}+#endif++staticconststructdev_pm_opsimx8m_blk_ctrl_pm_ops={+SET_SYSTEM_SLEEP_PM_OPS(imx8m_blk_ctrl_suspend,imx8m_blk_ctrl_resume)+};++staticintimx8mm_vpu_power_notifier(structnotifier_block*nb,+unsignedlongaction,void*data)+{+structimx8m_blk_ctrl*bc=container_of(nb,structimx8m_blk_ctrl,+power_nb);++if(action!=GENPD_NOTIFY_ON&&action!=GENPD_NOTIFY_PRE_OFF)+returnNOTIFY_OK;++/*+*TheADBintheVPUMIXdomainhasnoseparateresetandclock+*enablebits,butisungatedtogetherwiththeVPUclocks.To+*allowthehandshakewiththeGPCtoprogressweputtheVPUs+*inresetandungatetheclocks.+*/
The individual VPU's (IMX8MM_VPUBLK_PD_G1, G2, and H1) have clk_mask
values that set these bits when used and clear them when disabled. If
the VPUMix needs them set, shouldn't the IMX8MM_VPUBLK_PD_xx leave
them on until they are cleared by the VPUMIX?
+
+ if (action == GENPD_NOTIFY_ON) {
+ /*
+ * On power up we have no software backchannel to the GPC to
+ * wait for the ADB handshake to happen, so we just delay for a
+ * bit. On power down the GPC driver waits for the handshake.
+ */
+ udelay(5);
+
+ /* set "fuse" bits to enable the VPUs */
+ regmap_set_bits(bc->regmap, 0x8, 0xffffffff);
+ regmap_set_bits(bc->regmap, 0xc, 0xffffffff);
+ regmap_set_bits(bc->regmap, 0x10, 0xffffffff);
+ regmap_set_bits(bc->regmap, 0x14, 0xffffffff);
Should these registers ever get turned off when we disable the VPUMIX?
These registers don't have good descriptions in the TRM describing how
they interact with the VPU's, and it's unclear to me what happens if
they are set and the VPUMIX and various VPU's are disabled. I am
curious to know if they should get enabled/disabled with their
respective IMX8MM_VPUBLK_PD_xx domain. The Example Code 5 in the TRM
doesn't set these bits in their VPUMIX example, so it makes me think
they're OK to leave with the individual IMX8MM_VPUBLK_PD_xx domains.
I ask, because if I run v4l2-compliance on the H1 encoder, it also
appears to hang, but the power-domains appear to be attempting to
start, and the vpumix is the last thing to start, and I would expect
the pgc_vpu_h1 domain to follow, then end with the
vpublk-h1 starting.
# v4l2-compliance -d3
[ 271.001098] hantro-vpu 38320000.video-codec: genpd_runtime_resume()
[ 271.007447] genpd genpd:0:38330000.blk-ctrl: genpd_runtime_resume()
[ 271.013796] PM: vpumix: Power-on latency exceeded, new value 40623 ns
[ 271.020314] genpd genpd:3:38330000.blk-ctrl: genpd_runtime_resume()
adam
[1] - https://patchwork.kernel.org/project/linux-arm-kernel/list/?series=576435
From: Lucas Stach <l.stach@pengutronix.de> Date: 2021-11-08 08:35:28
Hi Adam,
Am Sonntag, dem 07.11.2021 um 15:08 -0600 schrieb Adam Ford:
On Fri, Sep 10, 2021 at 3:26 PM Lucas Stach [off-list ref] wrote:
quoted
This adds a driver for the blk-ctrl blocks found in the i.MX8M* line of
SoCs. The blk-ctrl is a top-level peripheral located in the various *MIX
power domains and interacts with the GPC power controller to provide the
peripherals in the power domain access to the NoC and ensures that those
peripherals are properly reset when their respective power domain is
brought back to life.
Software needs to do different things to make the bus handshake happen
after the GPC *MIX domain is powered up and before it is powered down.
As the requirements are quite different between the various blk-ctrls
there is a callback function provided to hook in the proper sequence.
The peripheral domains are quite uniform, they handle the soft clock
enables and resets in the blk-ctrl address space and sequencing with the
upstream GPC power domains.
Sorry to resurrect an old thread, but I'm running into issues enabling
new peripherals, and I think it's related to this, but I am not
certain.
I have attempted to enable the various VPU's, and in doing so, I
referenced the vpu_blk_ctrl for the respective power-domains reference
[1].
The various VPU devices enumerate, and I can see the various VPU power
domains turn on and off a boot.
[ 1.968565] genpd genpd:0:38330000.blk-ctrl: adding to PM domain vpumix
[ 1.981208] genpd genpd:1:38330000.blk-ctrl: adding to PM domain vpu-g1
[ 1.993838] genpd genpd:2:38330000.blk-ctrl: adding to PM domain vpu-g2
[ 2.006455] genpd genpd:3:38330000.blk-ctrl: adding to PM domain vpu-h1
[ 8.922661] hantro_vpu: module is from the staging directory, the
quality is unknown, you have been warned.
[ 9.008147] hantro_vpu: module is from the staging directory, the
quality is unknown, you have been warned.
[ 9.044041] hantro-vpu 38300000.video-codec: adding to PM domain vpublk-g1
[ 9.050959] hantro-vpu 38300000.video-codec: genpd_add_device()
[ 9.063349] PM: vpumix: Power-on latency exceeded, new value 50875 ns
[ 9.083437] PM: vpu-g1: Power-on latency exceeded, new value 26125 ns
[ 9.104778] PM: vpublk-g1: Power-on latency exceeded, new value 47819250 ns
[ 9.211988] hantro-vpu 38300000.video-codec: registered
nxp,imx8mm-vpu-dec as /dev/video0
[ 9.259680] hantro-vpu 38300000.video-codec: genpd_runtime_resume()
[ 9.297395] hantro-vpu 38300000.video-codec: resume latency exceeded, 2750 ns
[ 9.307767] hantro-vpu 38310000.video-codec: adding to PM domain vpublk-g2
[ 9.316462] hantro-vpu 38310000.video-codec: genpd_add_device()
[ 9.328807] PM: vpu-g2: Power-on latency exceeded, new value 26625 ns
[ 9.342401] PM: vpublk-g2: Power-on latency exceeded, new value 19965125 ns
[ 9.430254] hantro-vpu 38300000.video-codec: genpd_runtime_suspend()
[ 9.436683] hantro-vpu 38300000.video-codec: suspend latency
exceeded, 1625 ns
[ 9.443984] PM: vpublk-g1: Power-off latency exceeded, new value 16625 ns
[ 9.462361] hantro-vpu 38310000.video-codec: registered
nxp,imx8mm-vpu-g2-dec as /dev/video2
[ 9.491848] PM: vpu-g1: Power-off latency exceeded, new value 17125 ns
[ 9.506353] hantro-vpu 38310000.video-codec: genpd_runtime_resume()
[ 9.512735] hantro-vpu 38310000.video-codec: resume latency exceeded, 2750 ns
[ 9.520306] hantro-vpu 38320000.video-codec: adding to PM domain vpublk-h1
[ 9.527268] hantro-vpu 38320000.video-codec: genpd_add_device()
[ 9.539632] PM: vpu-h1: Power-on latency exceeded, new value 27750 ns
[ 9.553358] PM: vpublk-h1: Power-on latency exceeded, new value 20112125 ns
[ 9.605800] hantro_vpu: module is from the staging directory, the
quality is unknown, you have been warned.
[ 9.642194] hantro-vpu 38310000.video-codec: genpd_runtime_suspend()
[ 9.648634] hantro-vpu 38310000.video-codec: suspend latency
exceeded, 1500 ns
[ 9.655974] PM: vpublk-g2: Power-off latency exceeded, new value 12625 ns
[ 9.703863] PM: vpu-g2: Power-off latency exceeded, new value 16750 ns
[ 9.754527] hantro-vpu 38320000.video-codec: registered
nxp,imx8mm-vpu-h1-enc as /dev/video3
[ 9.785424] hantro-vpu 38320000.video-codec: genpd_runtime_resume()
[ 9.796034] hantro-vpu 38320000.video-codec: resume latency exceeded, 2000 ns
[ 9.934247] hantro-vpu 38320000.video-codec: genpd_runtime_suspend()
[ 9.940707] hantro-vpu 38320000.video-codec: suspend latency
exceeded, 1500 ns
[ 9.948042] PM: vpublk-h1: Power-off latency exceeded, new value 18875 ns
[ 9.975951] PM: vpu-h1: Power-off latency exceeded, new value 17625 ns
[ 9.999970] PM: vpumix: Power-off latency exceeded, new value 25750 ns
However, because the vpumix is disabled here, I think it might be
what's causing a hang when I attempt to read the regmap registers with
cat /sys/kernel/debug/regmap/38330000.blk-ctrl/registers
That is expected, as regmap doesn't tie into the runtime PM and is the
same behavior as on other devices when you try to read the registers
via regmap while the peripheral clock is gated.
I was looking at the ATF code that NXP has, and it doesn't appear that
the VPUMIX ever shuts down, and for that matter, I don't think the g1,
g2, and h1 domains shutdown either.
I think it makes sense to have the domains turned off when not in use,
but I have a few comments / questions below....
quoted
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Acked-by: Frieder Schrempf <redacted>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
---
This commit includes the full code to drive the VPUMIX domain on the
i.MX8MM, as the skeleton driver would probably be harder to review
without the context provided by one blk-ctrl implementation. Other
blk-ctrl implementations will follow, based on this overall structure.
v4:
- fix commit message typos
- fix superfluous whitespace
- constify clk_names more
---
drivers/soc/imx/Makefile | 1 +
drivers/soc/imx/imx8m-blk-ctrl.c | 453 +++++++++++++++++++++++++++++++
2 files changed, 454 insertions(+)
create mode 100644 drivers/soc/imx/imx8m-blk-ctrl.c
@@ -0,0 +1,453 @@+// SPDX-License-Identifier: GPL-2.0+++/*+*Copyright2021Pengutronix,LucasStach<kernel@pengutronix.de>+*/++#include<linux/device.h>+#include<linux/module.h>+#include<linux/of_device.h>+#include<linux/platform_device.h>+#include<linux/pm_domain.h>+#include<linux/pm_runtime.h>+#include<linux/regmap.h>+#include<linux/clk.h>++#include<dt-bindings/power/imx8mm-power.h>++#define BLK_SFT_RSTN 0x0+#define BLK_CLK_EN 0x4++structimx8m_blk_ctrl_domain;++structimx8m_blk_ctrl{+structdevice*dev;+structnotifier_blockpower_nb;+structdevice*bus_power_dev;+structregmap*regmap;+structimx8m_blk_ctrl_domain*domains;+structgenpd_onecell_dataonecell_data;+};++structimx8m_blk_ctrl_domain_data{+constchar*name;+constchar*const*clk_names;+intnum_clks;+constchar*gpc_name;+u32rst_mask;+u32clk_mask;+};++#define DOMAIN_MAX_CLKS 3++structimx8m_blk_ctrl_domain{+structgeneric_pm_domaingenpd;+conststructimx8m_blk_ctrl_domain_data*data;+structclk_bulk_dataclks[DOMAIN_MAX_CLKS];+structdevice*power_dev;+structimx8m_blk_ctrl*bc;+};++structimx8m_blk_ctrl_data{+intmax_reg;+notifier_fn_tpower_notifier_fn;+conststructimx8m_blk_ctrl_domain_data*domains;+intnum_domains;+};++staticinlinestructimx8m_blk_ctrl_domain*+to_imx8m_blk_ctrl_domain(structgeneric_pm_domain*genpd)+{+returncontainer_of(genpd,structimx8m_blk_ctrl_domain,genpd);+}++staticintimx8m_blk_ctrl_power_on(structgeneric_pm_domain*genpd)+{+structimx8m_blk_ctrl_domain*domain=to_imx8m_blk_ctrl_domain(genpd);+conststructimx8m_blk_ctrl_domain_data*data=domain->data;+structimx8m_blk_ctrl*bc=domain->bc;+intret;++/* make sure bus domain is awake */+ret=pm_runtime_get_sync(bc->bus_power_dev);+if(ret<0){+pm_runtime_put_noidle(bc->bus_power_dev);+dev_err(bc->dev,"failed to power up bus domain\n");+returnret;+}++/* put devices into reset */+regmap_clear_bits(bc->regmap,BLK_SFT_RSTN,data->rst_mask);++/* enable upstream and blk-ctrl clocks to allow reset to propagate */+ret=clk_bulk_prepare_enable(data->num_clks,domain->clks);+if(ret){+dev_err(bc->dev,"failed to enable clocks\n");+gotobus_put;+}+regmap_set_bits(bc->regmap,BLK_CLK_EN,data->clk_mask);++/* power up upstream GPC domain */+ret=pm_runtime_get_sync(domain->power_dev);+if(ret<0){+dev_err(bc->dev,"failed to power up peripheral domain\n");+gotoclk_disable;+}++/* wait for reset to propagate */+udelay(5);++/* release reset */+regmap_set_bits(bc->regmap,BLK_SFT_RSTN,data->rst_mask);++/* disable upstream clocks */+clk_bulk_disable_unprepare(data->num_clks,domain->clks);++return0;++clk_disable:+clk_bulk_disable_unprepare(data->num_clks,domain->clks);+bus_put:+pm_runtime_put(bc->bus_power_dev);++returnret;+}++staticintimx8m_blk_ctrl_power_off(structgeneric_pm_domain*genpd)+{+structimx8m_blk_ctrl_domain*domain=to_imx8m_blk_ctrl_domain(genpd);+conststructimx8m_blk_ctrl_domain_data*data=domain->data;+structimx8m_blk_ctrl*bc=domain->bc;++/* put devices into reset and disable clocks */+regmap_clear_bits(bc->regmap,BLK_SFT_RSTN,data->rst_mask);+regmap_clear_bits(bc->regmap,BLK_CLK_EN,data->clk_mask);++/* power down upstream GPC domain */+pm_runtime_put(domain->power_dev);++/* allow bus domain to suspend */+pm_runtime_put(bc->bus_power_dev);++return0;+}++staticstructgeneric_pm_domain*+imx8m_blk_ctrl_xlate(structof_phandle_args*args,void*data)+{+structgenpd_onecell_data*onecell_data=data;+unsignedintindex=args->args[0];++if(args->args_count!=1||+index>onecell_data->num_domains)+returnERR_PTR(-EINVAL);++returnonecell_data->domains[index];+}++staticstructlock_class_keyblk_ctrl_genpd_lock_class;++staticintimx8m_blk_ctrl_probe(structplatform_device*pdev)+{+conststructimx8m_blk_ctrl_data*bc_data;+structdevice*dev=&pdev->dev;+structimx8m_blk_ctrl*bc;+void__iomem*base;+inti,ret;++structregmap_configregmap_config={+.reg_bits=32,+.val_bits=32,+.reg_stride=4,+};++bc=devm_kzalloc(dev,sizeof(*bc),GFP_KERNEL);+if(!bc)+return-ENOMEM;++bc->dev=dev;++bc_data=of_device_get_match_data(dev);++base=devm_platform_ioremap_resource(pdev,0);+if(IS_ERR(base))+returnPTR_ERR(base);++regmap_config.max_register=bc_data->max_reg;+bc->regmap=devm_regmap_init_mmio(dev,base,®map_config);+if(IS_ERR(bc->regmap))+returndev_err_probe(dev,PTR_ERR(bc->regmap),+"failed to init regmap\n");++bc->domains=devm_kcalloc(dev,bc_data->num_domains,+sizeof(structimx8m_blk_ctrl_domain),+GFP_KERNEL);+if(!bc->domains)+return-ENOMEM;++bc->onecell_data.num_domains=bc_data->num_domains;+bc->onecell_data.xlate=imx8m_blk_ctrl_xlate;+bc->onecell_data.domains=+devm_kcalloc(dev,bc_data->num_domains,+sizeof(structgeneric_pm_domain*),GFP_KERNEL);+if(!bc->onecell_data.domains)+return-ENOMEM;++bc->bus_power_dev=genpd_dev_pm_attach_by_name(dev,"bus");+if(IS_ERR(bc->bus_power_dev))+returndev_err_probe(dev,PTR_ERR(bc->bus_power_dev),+"failed to attach power domain\n");++for(i=0;i<bc_data->num_domains;i++){+conststructimx8m_blk_ctrl_domain_data*data=&bc_data->domains[i];+structimx8m_blk_ctrl_domain*domain=&bc->domains[i];+intj;++domain->data=data;++for(j=0;j<data->num_clks;j++)+domain->clks[j].id=data->clk_names[j];++ret=devm_clk_bulk_get(dev,data->num_clks,domain->clks);+if(ret){+dev_err_probe(dev,ret,"failed to get clock\n");+gotocleanup_pds;+}++domain->power_dev=+dev_pm_domain_attach_by_name(dev,data->gpc_name);+if(IS_ERR(domain->power_dev)){+dev_err_probe(dev,PTR_ERR(domain->power_dev),+"failed to attach power domain\n");+ret=PTR_ERR(domain->power_dev);+gotocleanup_pds;+}++domain->genpd.name=data->name;+domain->genpd.power_on=imx8m_blk_ctrl_power_on;+domain->genpd.power_off=imx8m_blk_ctrl_power_off;+domain->bc=bc;++ret=pm_genpd_init(&domain->genpd,NULL,true);+if(ret){+dev_err_probe(dev,ret,"failed to init power domain\n");+dev_pm_domain_detach(domain->power_dev,true);+gotocleanup_pds;+}++/*+*WeuseruntimePMtotriggerpoweron/offoftheupstreamGPC+*domain,asastricthierarchicalparent/childpowerdomain+*setupdoesn'tallowustomeetthesequencingrequirements.+*Thismeanswehavenestedlockingofgenpdlocks,withoutthe+*nestingbeingvisibleatthegenpdlevel,soweneeda+*separatelockclasstomakelockdepawareofthefactthat+*thisareseparatedomainlocksthatcanbenestedwithouta+*self-deadlock.+*/+lockdep_set_class(&domain->genpd.mlock,+&blk_ctrl_genpd_lock_class);++bc->onecell_data.domains[i]=&domain->genpd;+}++ret=of_genpd_add_provider_onecell(dev->of_node,&bc->onecell_data);+if(ret){+dev_err_probe(dev,ret,"failed to add power domain provider\n");+gotocleanup_pds;+}++bc->power_nb.notifier_call=bc_data->power_notifier_fn;+ret=dev_pm_genpd_add_notifier(bc->bus_power_dev,&bc->power_nb);+if(ret){+dev_err_probe(dev,ret,"failed to add power notifier\n");+gotocleanup_provider;+}++dev_set_drvdata(dev,bc);++return0;++cleanup_provider:+of_genpd_del_provider(dev->of_node);+cleanup_pds:+for(i--;i>=0;i--){+pm_genpd_remove(&bc->domains[i].genpd);+dev_pm_domain_detach(bc->domains[i].power_dev,true);+}++dev_pm_domain_detach(bc->bus_power_dev,true);++returnret;+}++staticintimx8m_blk_ctrl_remove(structplatform_device*pdev)+{+structimx8m_blk_ctrl*bc=dev_get_drvdata(&pdev->dev);+inti;++of_genpd_del_provider(pdev->dev.of_node);++for(i=0;bc->onecell_data.num_domains;i++){+structimx8m_blk_ctrl_domain*domain=&bc->domains[i];++pm_genpd_remove(&domain->genpd);+dev_pm_domain_detach(domain->power_dev,true);+}++dev_pm_genpd_remove_notifier(bc->bus_power_dev);++dev_pm_domain_detach(bc->bus_power_dev,true);++return0;+}++#ifdef CONFIG_PM_SLEEP+staticintimx8m_blk_ctrl_suspend(structdevice*dev)+{+structimx8m_blk_ctrl*bc=dev_get_drvdata(dev);+intret,i;++/*+*Thismaylookstrange,butisdonesothegenericPM_SLEEPcode+*canpowerdownourdomainsandmoreimportantlypowerthemupagain+*afterresume,withouttrippingoverourusageofruntimePMto+*controltheupstreamGPCdomains.Thingshappenintherightorder+*inthesystemsuspend/resumepathsduetothedeviceparent/child+*hierarchy.+*/+ret=pm_runtime_get_sync(bc->bus_power_dev);+if(ret<0){+pm_runtime_put_noidle(bc->bus_power_dev);+returnret;+}++for(i=0;i<bc->onecell_data.num_domains;i++){+structimx8m_blk_ctrl_domain*domain=&bc->domains[i];++ret=pm_runtime_get_sync(domain->power_dev);+if(ret<0){+pm_runtime_put_noidle(domain->power_dev);+gotoout_fail;+}+}++return0;++out_fail:+for(i--;i>=0;i--)+pm_runtime_put(bc->domains[i].power_dev);++pm_runtime_put(bc->bus_power_dev);++returnret;+}++staticintimx8m_blk_ctrl_resume(structdevice*dev)+{+structimx8m_blk_ctrl*bc=dev_get_drvdata(dev);+inti;++for(i=0;i<bc->onecell_data.num_domains;i++)+pm_runtime_put(bc->domains[i].power_dev);++pm_runtime_put(bc->bus_power_dev);++return0;+}+#endif++staticconststructdev_pm_opsimx8m_blk_ctrl_pm_ops={+SET_SYSTEM_SLEEP_PM_OPS(imx8m_blk_ctrl_suspend,imx8m_blk_ctrl_resume)+};++staticintimx8mm_vpu_power_notifier(structnotifier_block*nb,+unsignedlongaction,void*data)+{+structimx8m_blk_ctrl*bc=container_of(nb,structimx8m_blk_ctrl,+power_nb);++if(action!=GENPD_NOTIFY_ON&&action!=GENPD_NOTIFY_PRE_OFF)+returnNOTIFY_OK;++/*+*TheADBintheVPUMIXdomainhasnoseparateresetandclock+*enablebits,butisungatedtogetherwiththeVPUclocks.To+*allowthehandshakewiththeGPCtoprogressweputtheVPUs+*inresetandungatetheclocks.+*/
The individual VPU's (IMX8MM_VPUBLK_PD_G1, G2, and H1) have clk_mask
values that set these bits when used and clear them when disabled. If
the VPUMix needs them set, shouldn't the IMX8MM_VPUBLK_PD_xx leave
them on until they are cleared by the VPUMIX?
No, VPUMIX only need those clocks to be running for the ADB handshake
to work. The handshake is only executed when powering up/down a GPC
domain. So we can clockgate the individual Hantro cores when not in use
while the power domain is up.
quoted
+
+ if (action == GENPD_NOTIFY_ON) {
quoted
+ /*
+ * On power up we have no software backchannel to the GPC to
+ * wait for the ADB handshake to happen, so we just delay for a
+ * bit. On power down the GPC driver waits for the handshake.
+ */
+ udelay(5);
+
+ /* set "fuse" bits to enable the VPUs */
+ regmap_set_bits(bc->regmap, 0x8, 0xffffffff);
+ regmap_set_bits(bc->regmap, 0xc, 0xffffffff);
+ regmap_set_bits(bc->regmap, 0x10, 0xffffffff);
+ regmap_set_bits(bc->regmap, 0x14, 0xffffffff);
Should these registers ever get turned off when we disable the VPUMIX?
These registers don't have good descriptions in the TRM describing how
they interact with the VPU's, and it's unclear to me what happens if
they are set and the VPUMIX and various VPU's are disabled.
Those registers lose their state when the VPUMIX power collapses, thus
we need to re-initialize them when powering up the domain. As far as I
understand it those registers just carry some strap bits for the
decoders/encoder to let them know which features should be enabled.
I am
curious to know if they should get enabled/disabled with their
respective IMX8MM_VPUBLK_PD_xx domain. The Example Code 5 in the TRM
doesn't set these bits in their VPUMIX example, so it makes me think
they're OK to leave with the individual IMX8MM_VPUBLK_PD_xx domains.
I ask, because if I run v4l2-compliance on the H1 encoder, it also
appears to hang, but the power-domains appear to be attempting to
start, and the vpumix is the last thing to start, and I would expect
the pgc_vpu_h1 domain to follow, then end with the
vpublk-h1 starting.
# v4l2-compliance -d3
[ 271.001098] hantro-vpu 38320000.video-codec: genpd_runtime_resume()
[ 271.007447] genpd genpd:0:38330000.blk-ctrl: genpd_runtime_resume()
[ 271.013796] PM: vpumix: Power-on latency exceeded, new value 40623 ns
[ 271.020314] genpd genpd:3:38330000.blk-ctrl: genpd_runtime_resume()
This looks correct to me. genpd:0:38330000.blk-ctrl is the VPU blk-ctrl
bus domain, which should start the GPC vpumix domain,
genpd:3:38330000.blk-ctrl is the blk-ctrl H1 domain, which should in
turn power up the GPC h1 domain.
Do you know where exactly things are hanging?
Regards,
Lucas
From: Adam Ford <hidden> Date: 2021-11-08 13:13:38
On Mon, Nov 8, 2021 at 2:35 AM Lucas Stach [off-list ref] wrote:
Hi Adam,
Am Sonntag, dem 07.11.2021 um 15:08 -0600 schrieb Adam Ford:
quoted
On Fri, Sep 10, 2021 at 3:26 PM Lucas Stach [off-list ref] wrote:
quoted
This adds a driver for the blk-ctrl blocks found in the i.MX8M* line of
SoCs. The blk-ctrl is a top-level peripheral located in the various *MIX
power domains and interacts with the GPC power controller to provide the
peripherals in the power domain access to the NoC and ensures that those
peripherals are properly reset when their respective power domain is
brought back to life.
Software needs to do different things to make the bus handshake happen
after the GPC *MIX domain is powered up and before it is powered down.
As the requirements are quite different between the various blk-ctrls
there is a callback function provided to hook in the proper sequence.
The peripheral domains are quite uniform, they handle the soft clock
enables and resets in the blk-ctrl address space and sequencing with the
upstream GPC power domains.
Sorry to resurrect an old thread, but I'm running into issues enabling
new peripherals, and I think it's related to this, but I am not
certain.
I have attempted to enable the various VPU's, and in doing so, I
referenced the vpu_blk_ctrl for the respective power-domains reference
[1].
The various VPU devices enumerate, and I can see the various VPU power
domains turn on and off a boot.
[ 1.968565] genpd genpd:0:38330000.blk-ctrl: adding to PM domain vpumix
[ 1.981208] genpd genpd:1:38330000.blk-ctrl: adding to PM domain vpu-g1
[ 1.993838] genpd genpd:2:38330000.blk-ctrl: adding to PM domain vpu-g2
[ 2.006455] genpd genpd:3:38330000.blk-ctrl: adding to PM domain vpu-h1
[ 8.922661] hantro_vpu: module is from the staging directory, the
quality is unknown, you have been warned.
[ 9.008147] hantro_vpu: module is from the staging directory, the
quality is unknown, you have been warned.
[ 9.044041] hantro-vpu 38300000.video-codec: adding to PM domain vpublk-g1
[ 9.050959] hantro-vpu 38300000.video-codec: genpd_add_device()
[ 9.063349] PM: vpumix: Power-on latency exceeded, new value 50875 ns
[ 9.083437] PM: vpu-g1: Power-on latency exceeded, new value 26125 ns
[ 9.104778] PM: vpublk-g1: Power-on latency exceeded, new value 47819250 ns
[ 9.211988] hantro-vpu 38300000.video-codec: registered
nxp,imx8mm-vpu-dec as /dev/video0
[ 9.259680] hantro-vpu 38300000.video-codec: genpd_runtime_resume()
[ 9.297395] hantro-vpu 38300000.video-codec: resume latency exceeded, 2750 ns
[ 9.307767] hantro-vpu 38310000.video-codec: adding to PM domain vpublk-g2
[ 9.316462] hantro-vpu 38310000.video-codec: genpd_add_device()
[ 9.328807] PM: vpu-g2: Power-on latency exceeded, new value 26625 ns
[ 9.342401] PM: vpublk-g2: Power-on latency exceeded, new value 19965125 ns
[ 9.430254] hantro-vpu 38300000.video-codec: genpd_runtime_suspend()
[ 9.436683] hantro-vpu 38300000.video-codec: suspend latency
exceeded, 1625 ns
[ 9.443984] PM: vpublk-g1: Power-off latency exceeded, new value 16625 ns
[ 9.462361] hantro-vpu 38310000.video-codec: registered
nxp,imx8mm-vpu-g2-dec as /dev/video2
[ 9.491848] PM: vpu-g1: Power-off latency exceeded, new value 17125 ns
[ 9.506353] hantro-vpu 38310000.video-codec: genpd_runtime_resume()
[ 9.512735] hantro-vpu 38310000.video-codec: resume latency exceeded, 2750 ns
[ 9.520306] hantro-vpu 38320000.video-codec: adding to PM domain vpublk-h1
[ 9.527268] hantro-vpu 38320000.video-codec: genpd_add_device()
[ 9.539632] PM: vpu-h1: Power-on latency exceeded, new value 27750 ns
[ 9.553358] PM: vpublk-h1: Power-on latency exceeded, new value 20112125 ns
[ 9.605800] hantro_vpu: module is from the staging directory, the
quality is unknown, you have been warned.
[ 9.642194] hantro-vpu 38310000.video-codec: genpd_runtime_suspend()
[ 9.648634] hantro-vpu 38310000.video-codec: suspend latency
exceeded, 1500 ns
[ 9.655974] PM: vpublk-g2: Power-off latency exceeded, new value 12625 ns
[ 9.703863] PM: vpu-g2: Power-off latency exceeded, new value 16750 ns
[ 9.754527] hantro-vpu 38320000.video-codec: registered
nxp,imx8mm-vpu-h1-enc as /dev/video3
[ 9.785424] hantro-vpu 38320000.video-codec: genpd_runtime_resume()
[ 9.796034] hantro-vpu 38320000.video-codec: resume latency exceeded, 2000 ns
[ 9.934247] hantro-vpu 38320000.video-codec: genpd_runtime_suspend()
[ 9.940707] hantro-vpu 38320000.video-codec: suspend latency
exceeded, 1500 ns
[ 9.948042] PM: vpublk-h1: Power-off latency exceeded, new value 18875 ns
[ 9.975951] PM: vpu-h1: Power-off latency exceeded, new value 17625 ns
[ 9.999970] PM: vpumix: Power-off latency exceeded, new value 25750 ns
However, because the vpumix is disabled here, I think it might be
what's causing a hang when I attempt to read the regmap registers with
cat /sys/kernel/debug/regmap/38330000.blk-ctrl/registers
That is expected, as regmap doesn't tie into the runtime PM and is the
same behavior as on other devices when you try to read the registers
via regmap while the peripheral clock is gated.
Ok. That makes sense. I think I was so focussed on getting the regmap
working, I assumed something was wrong with the blk-ctrl.
quoted
I was looking at the ATF code that NXP has, and it doesn't appear that
the VPUMIX ever shuts down, and for that matter, I don't think the g1,
g2, and h1 domains shutdown either.
I think it makes sense to have the domains turned off when not in use,
but I have a few comments / questions below....
quoted
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Acked-by: Frieder Schrempf <redacted>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
---
This commit includes the full code to drive the VPUMIX domain on the
i.MX8MM, as the skeleton driver would probably be harder to review
without the context provided by one blk-ctrl implementation. Other
blk-ctrl implementations will follow, based on this overall structure.
v4:
- fix commit message typos
- fix superfluous whitespace
- constify clk_names more
---
drivers/soc/imx/Makefile | 1 +
drivers/soc/imx/imx8m-blk-ctrl.c | 453 +++++++++++++++++++++++++++++++
2 files changed, 454 insertions(+)
create mode 100644 drivers/soc/imx/imx8m-blk-ctrl.c
@@ -0,0 +1,453 @@+// SPDX-License-Identifier: GPL-2.0+++/*+*Copyright2021Pengutronix,LucasStach<kernel@pengutronix.de>+*/++#include<linux/device.h>+#include<linux/module.h>+#include<linux/of_device.h>+#include<linux/platform_device.h>+#include<linux/pm_domain.h>+#include<linux/pm_runtime.h>+#include<linux/regmap.h>+#include<linux/clk.h>++#include<dt-bindings/power/imx8mm-power.h>++#define BLK_SFT_RSTN 0x0+#define BLK_CLK_EN 0x4++structimx8m_blk_ctrl_domain;++structimx8m_blk_ctrl{+structdevice*dev;+structnotifier_blockpower_nb;+structdevice*bus_power_dev;+structregmap*regmap;+structimx8m_blk_ctrl_domain*domains;+structgenpd_onecell_dataonecell_data;+};++structimx8m_blk_ctrl_domain_data{+constchar*name;+constchar*const*clk_names;+intnum_clks;+constchar*gpc_name;+u32rst_mask;+u32clk_mask;+};++#define DOMAIN_MAX_CLKS 3++structimx8m_blk_ctrl_domain{+structgeneric_pm_domaingenpd;+conststructimx8m_blk_ctrl_domain_data*data;+structclk_bulk_dataclks[DOMAIN_MAX_CLKS];+structdevice*power_dev;+structimx8m_blk_ctrl*bc;+};++structimx8m_blk_ctrl_data{+intmax_reg;+notifier_fn_tpower_notifier_fn;+conststructimx8m_blk_ctrl_domain_data*domains;+intnum_domains;+};++staticinlinestructimx8m_blk_ctrl_domain*+to_imx8m_blk_ctrl_domain(structgeneric_pm_domain*genpd)+{+returncontainer_of(genpd,structimx8m_blk_ctrl_domain,genpd);+}++staticintimx8m_blk_ctrl_power_on(structgeneric_pm_domain*genpd)+{+structimx8m_blk_ctrl_domain*domain=to_imx8m_blk_ctrl_domain(genpd);+conststructimx8m_blk_ctrl_domain_data*data=domain->data;+structimx8m_blk_ctrl*bc=domain->bc;+intret;++/* make sure bus domain is awake */+ret=pm_runtime_get_sync(bc->bus_power_dev);+if(ret<0){+pm_runtime_put_noidle(bc->bus_power_dev);+dev_err(bc->dev,"failed to power up bus domain\n");+returnret;+}++/* put devices into reset */+regmap_clear_bits(bc->regmap,BLK_SFT_RSTN,data->rst_mask);++/* enable upstream and blk-ctrl clocks to allow reset to propagate */+ret=clk_bulk_prepare_enable(data->num_clks,domain->clks);+if(ret){+dev_err(bc->dev,"failed to enable clocks\n");+gotobus_put;+}+regmap_set_bits(bc->regmap,BLK_CLK_EN,data->clk_mask);++/* power up upstream GPC domain */+ret=pm_runtime_get_sync(domain->power_dev);+if(ret<0){+dev_err(bc->dev,"failed to power up peripheral domain\n");+gotoclk_disable;+}++/* wait for reset to propagate */+udelay(5);++/* release reset */+regmap_set_bits(bc->regmap,BLK_SFT_RSTN,data->rst_mask);++/* disable upstream clocks */+clk_bulk_disable_unprepare(data->num_clks,domain->clks);++return0;++clk_disable:+clk_bulk_disable_unprepare(data->num_clks,domain->clks);+bus_put:+pm_runtime_put(bc->bus_power_dev);++returnret;+}++staticintimx8m_blk_ctrl_power_off(structgeneric_pm_domain*genpd)+{+structimx8m_blk_ctrl_domain*domain=to_imx8m_blk_ctrl_domain(genpd);+conststructimx8m_blk_ctrl_domain_data*data=domain->data;+structimx8m_blk_ctrl*bc=domain->bc;++/* put devices into reset and disable clocks */+regmap_clear_bits(bc->regmap,BLK_SFT_RSTN,data->rst_mask);+regmap_clear_bits(bc->regmap,BLK_CLK_EN,data->clk_mask);++/* power down upstream GPC domain */+pm_runtime_put(domain->power_dev);++/* allow bus domain to suspend */+pm_runtime_put(bc->bus_power_dev);++return0;+}++staticstructgeneric_pm_domain*+imx8m_blk_ctrl_xlate(structof_phandle_args*args,void*data)+{+structgenpd_onecell_data*onecell_data=data;+unsignedintindex=args->args[0];++if(args->args_count!=1||+index>onecell_data->num_domains)+returnERR_PTR(-EINVAL);++returnonecell_data->domains[index];+}++staticstructlock_class_keyblk_ctrl_genpd_lock_class;++staticintimx8m_blk_ctrl_probe(structplatform_device*pdev)+{+conststructimx8m_blk_ctrl_data*bc_data;+structdevice*dev=&pdev->dev;+structimx8m_blk_ctrl*bc;+void__iomem*base;+inti,ret;++structregmap_configregmap_config={+.reg_bits=32,+.val_bits=32,+.reg_stride=4,+};++bc=devm_kzalloc(dev,sizeof(*bc),GFP_KERNEL);+if(!bc)+return-ENOMEM;++bc->dev=dev;++bc_data=of_device_get_match_data(dev);++base=devm_platform_ioremap_resource(pdev,0);+if(IS_ERR(base))+returnPTR_ERR(base);++regmap_config.max_register=bc_data->max_reg;+bc->regmap=devm_regmap_init_mmio(dev,base,®map_config);+if(IS_ERR(bc->regmap))+returndev_err_probe(dev,PTR_ERR(bc->regmap),+"failed to init regmap\n");++bc->domains=devm_kcalloc(dev,bc_data->num_domains,+sizeof(structimx8m_blk_ctrl_domain),+GFP_KERNEL);+if(!bc->domains)+return-ENOMEM;++bc->onecell_data.num_domains=bc_data->num_domains;+bc->onecell_data.xlate=imx8m_blk_ctrl_xlate;+bc->onecell_data.domains=+devm_kcalloc(dev,bc_data->num_domains,+sizeof(structgeneric_pm_domain*),GFP_KERNEL);+if(!bc->onecell_data.domains)+return-ENOMEM;++bc->bus_power_dev=genpd_dev_pm_attach_by_name(dev,"bus");+if(IS_ERR(bc->bus_power_dev))+returndev_err_probe(dev,PTR_ERR(bc->bus_power_dev),+"failed to attach power domain\n");++for(i=0;i<bc_data->num_domains;i++){+conststructimx8m_blk_ctrl_domain_data*data=&bc_data->domains[i];+structimx8m_blk_ctrl_domain*domain=&bc->domains[i];+intj;++domain->data=data;++for(j=0;j<data->num_clks;j++)+domain->clks[j].id=data->clk_names[j];++ret=devm_clk_bulk_get(dev,data->num_clks,domain->clks);+if(ret){+dev_err_probe(dev,ret,"failed to get clock\n");+gotocleanup_pds;+}++domain->power_dev=+dev_pm_domain_attach_by_name(dev,data->gpc_name);+if(IS_ERR(domain->power_dev)){+dev_err_probe(dev,PTR_ERR(domain->power_dev),+"failed to attach power domain\n");+ret=PTR_ERR(domain->power_dev);+gotocleanup_pds;+}++domain->genpd.name=data->name;+domain->genpd.power_on=imx8m_blk_ctrl_power_on;+domain->genpd.power_off=imx8m_blk_ctrl_power_off;+domain->bc=bc;++ret=pm_genpd_init(&domain->genpd,NULL,true);+if(ret){+dev_err_probe(dev,ret,"failed to init power domain\n");+dev_pm_domain_detach(domain->power_dev,true);+gotocleanup_pds;+}++/*+*WeuseruntimePMtotriggerpoweron/offoftheupstreamGPC+*domain,asastricthierarchicalparent/childpowerdomain+*setupdoesn'tallowustomeetthesequencingrequirements.+*Thismeanswehavenestedlockingofgenpdlocks,withoutthe+*nestingbeingvisibleatthegenpdlevel,soweneeda+*separatelockclasstomakelockdepawareofthefactthat+*thisareseparatedomainlocksthatcanbenestedwithouta+*self-deadlock.+*/+lockdep_set_class(&domain->genpd.mlock,+&blk_ctrl_genpd_lock_class);++bc->onecell_data.domains[i]=&domain->genpd;+}++ret=of_genpd_add_provider_onecell(dev->of_node,&bc->onecell_data);+if(ret){+dev_err_probe(dev,ret,"failed to add power domain provider\n");+gotocleanup_pds;+}++bc->power_nb.notifier_call=bc_data->power_notifier_fn;+ret=dev_pm_genpd_add_notifier(bc->bus_power_dev,&bc->power_nb);+if(ret){+dev_err_probe(dev,ret,"failed to add power notifier\n");+gotocleanup_provider;+}++dev_set_drvdata(dev,bc);++return0;++cleanup_provider:+of_genpd_del_provider(dev->of_node);+cleanup_pds:+for(i--;i>=0;i--){+pm_genpd_remove(&bc->domains[i].genpd);+dev_pm_domain_detach(bc->domains[i].power_dev,true);+}++dev_pm_domain_detach(bc->bus_power_dev,true);++returnret;+}++staticintimx8m_blk_ctrl_remove(structplatform_device*pdev)+{+structimx8m_blk_ctrl*bc=dev_get_drvdata(&pdev->dev);+inti;++of_genpd_del_provider(pdev->dev.of_node);++for(i=0;bc->onecell_data.num_domains;i++){+structimx8m_blk_ctrl_domain*domain=&bc->domains[i];++pm_genpd_remove(&domain->genpd);+dev_pm_domain_detach(domain->power_dev,true);+}++dev_pm_genpd_remove_notifier(bc->bus_power_dev);++dev_pm_domain_detach(bc->bus_power_dev,true);++return0;+}++#ifdef CONFIG_PM_SLEEP+staticintimx8m_blk_ctrl_suspend(structdevice*dev)+{+structimx8m_blk_ctrl*bc=dev_get_drvdata(dev);+intret,i;++/*+*Thismaylookstrange,butisdonesothegenericPM_SLEEPcode+*canpowerdownourdomainsandmoreimportantlypowerthemupagain+*afterresume,withouttrippingoverourusageofruntimePMto+*controltheupstreamGPCdomains.Thingshappenintherightorder+*inthesystemsuspend/resumepathsduetothedeviceparent/child+*hierarchy.+*/+ret=pm_runtime_get_sync(bc->bus_power_dev);+if(ret<0){+pm_runtime_put_noidle(bc->bus_power_dev);+returnret;+}++for(i=0;i<bc->onecell_data.num_domains;i++){+structimx8m_blk_ctrl_domain*domain=&bc->domains[i];++ret=pm_runtime_get_sync(domain->power_dev);+if(ret<0){+pm_runtime_put_noidle(domain->power_dev);+gotoout_fail;+}+}++return0;++out_fail:+for(i--;i>=0;i--)+pm_runtime_put(bc->domains[i].power_dev);++pm_runtime_put(bc->bus_power_dev);++returnret;+}++staticintimx8m_blk_ctrl_resume(structdevice*dev)+{+structimx8m_blk_ctrl*bc=dev_get_drvdata(dev);+inti;++for(i=0;i<bc->onecell_data.num_domains;i++)+pm_runtime_put(bc->domains[i].power_dev);++pm_runtime_put(bc->bus_power_dev);++return0;+}+#endif++staticconststructdev_pm_opsimx8m_blk_ctrl_pm_ops={+SET_SYSTEM_SLEEP_PM_OPS(imx8m_blk_ctrl_suspend,imx8m_blk_ctrl_resume)+};++staticintimx8mm_vpu_power_notifier(structnotifier_block*nb,+unsignedlongaction,void*data)+{+structimx8m_blk_ctrl*bc=container_of(nb,structimx8m_blk_ctrl,+power_nb);++if(action!=GENPD_NOTIFY_ON&&action!=GENPD_NOTIFY_PRE_OFF)+returnNOTIFY_OK;++/*+*TheADBintheVPUMIXdomainhasnoseparateresetandclock+*enablebits,butisungatedtogetherwiththeVPUclocks.To+*allowthehandshakewiththeGPCtoprogressweputtheVPUs+*inresetandungatetheclocks.+*/
The individual VPU's (IMX8MM_VPUBLK_PD_G1, G2, and H1) have clk_mask
values that set these bits when used and clear them when disabled. If
the VPUMix needs them set, shouldn't the IMX8MM_VPUBLK_PD_xx leave
them on until they are cleared by the VPUMIX?
No, VPUMIX only need those clocks to be running for the ADB handshake
to work. The handshake is only executed when powering up/down a GPC
domain. So we can clockgate the individual Hantro cores when not in use
while the power domain is up.
Ok, that makes sense.
quoted
quoted
+
+ if (action == GENPD_NOTIFY_ON) {
quoted
+ /*
+ * On power up we have no software backchannel to the GPC to
+ * wait for the ADB handshake to happen, so we just delay for a
+ * bit. On power down the GPC driver waits for the handshake.
+ */
+ udelay(5);
+
+ /* set "fuse" bits to enable the VPUs */
+ regmap_set_bits(bc->regmap, 0x8, 0xffffffff);
+ regmap_set_bits(bc->regmap, 0xc, 0xffffffff);
+ regmap_set_bits(bc->regmap, 0x10, 0xffffffff);
+ regmap_set_bits(bc->regmap, 0x14, 0xffffffff);
Should these registers ever get turned off when we disable the VPUMIX?
These registers don't have good descriptions in the TRM describing how
they interact with the VPU's, and it's unclear to me what happens if
they are set and the VPUMIX and various VPU's are disabled.
Those registers lose their state when the VPUMIX power collapses, thus
we need to re-initialize them when powering up the domain. As far as I
understand it those registers just carry some strap bits for the
decoders/encoder to let them know which features should be enabled.
quoted
I am
curious to know if they should get enabled/disabled with their
respective IMX8MM_VPUBLK_PD_xx domain. The Example Code 5 in the TRM
doesn't set these bits in their VPUMIX example, so it makes me think
they're OK to leave with the individual IMX8MM_VPUBLK_PD_xx domains.
I ask, because if I run v4l2-compliance on the H1 encoder, it also
appears to hang, but the power-domains appear to be attempting to
start, and the vpumix is the last thing to start, and I would expect
the pgc_vpu_h1 domain to follow, then end with the
vpublk-h1 starting.
# v4l2-compliance -d3
[ 271.001098] hantro-vpu 38320000.video-codec: genpd_runtime_resume()
[ 271.007447] genpd genpd:0:38330000.blk-ctrl: genpd_runtime_resume()
[ 271.013796] PM: vpumix: Power-on latency exceeded, new value 40623 ns
[ 271.020314] genpd genpd:3:38330000.blk-ctrl: genpd_runtime_resume()
This looks correct to me. genpd:0:38330000.blk-ctrl is the VPU blk-ctrl
bus domain, which should start the GPC vpumix domain,
genpd:3:38330000.blk-ctrl is the blk-ctrl H1 domain, which should in
turn power up the GPC h1 domain.
Do you know where exactly things are hanging?
I added more debugging and it appears to be hanging in the VPU code
itself and not the power domain like I originally thought.
I was focussed on the power domain because the regmap was hanging that
I didn't even consider the possibility it was the VPU code itself.
[ 377.840195] hantro-vpu 38320000.video-codec: genpd_runtime_resume():
[ 377.846675] hantro-vpu 38320000.video-codec: vpublk-h1()
[ 377.852115] genpd genpd:0:38330000.blk-ctrl: genpd_runtime_resume():
[ 377.858581] genpd genpd:0:38330000.blk-ctrl: vpumix()
[ 377.863658] imx_pgc_power_up(): vpumix
[ 377.867468] imx_pgc_power_up(): Success
[ 377.871351] genpd genpd:3:38330000.blk-ctrl: genpd_runtime_resume():
[ 377.877810] genpd genpd:3:38330000.blk-ctrl: vpu-h1()
[ 377.882891] imx_pgc_power_up(): vpu-h1
[ 377.886679] imx_pgc_power_up(): Success
[ 377.890540] genpd genpd:3:38330000.blk-ctrl: resume latency exceeded, 750 ns
[ 377.897636] PM: vpublk-h1: Power-on latency exceeded, new value 45528719 ns
[ 377.904635] hantro_h1_jpeg_enc_run()
Thanks for your help. I think we're good. I'll focus my work on the
VPU encoder.
Sorry for the noise. I think I better understand the blk-ctrl now too.
adam