mbox series

[v9,0/8] NVIDIA Tegra power management patches for 5.16

Message ID 20210827013415.24027-1-digetx@gmail.com
Headers show
Series NVIDIA Tegra power management patches for 5.16 | expand

Message

Dmitry Osipenko Aug. 27, 2021, 1:34 a.m. UTC 
This is a reduced version of the patchset which adds power management
support to NVIDIA Tegra drivers. Viresh Kumar asked to send these PD/OPP
patches separately for now to reduce the noise and finalize the review.

I implemented new get_performance_state() GENPD callback as was discussed
in v8. GR3D driver patch shows how it's used by consumer drivers, which
is a good example because 3d driver supports both cases of a single and
multi-domain hardware, it also uses OPP API more extensively than other
drivers.

Dmitry Osipenko (8):
  opp: Add dev_pm_opp_from_clk_rate()
  opp: Allow dev_pm_opp_set_clkname() to replace released clock
  opp: Change type of dev_pm_opp_attach_genpd(names) argument
  PM: domains: Add get_performance_state() callback
  soc/tegra: pmc: Implement get_performance_state() callback
  soc/tegra: Add devm_tegra_core_dev_init_opp_table_simple()
  gpu: host1x: Add host1x_channel_stop()
  drm/tegra: gr3d: Support generic power domain and runtime PM

 drivers/base/power/domain.c  |  32 ++-
 drivers/gpu/drm/tegra/gr3d.c | 384 ++++++++++++++++++++++++++++++-----
 drivers/gpu/host1x/channel.c |   8 +
 drivers/opp/core.c           |  50 ++++-
 drivers/soc/tegra/pmc.c      |  86 ++++++++
 include/linux/host1x.h       |   1 +
 include/linux/pm_domain.h    |   2 +
 include/linux/pm_opp.h       |  14 +-
 include/soc/tegra/common.h   |  13 ++
 9 files changed, 522 insertions(+), 68 deletions(-)

Comments

Dmitry Osipenko Aug. 27, 2021, 1:42 a.m. UTC | #1 
27.08.2021 04:34, Dmitry Osipenko пишет:
> diff --git a/drivers/opp/core.c b/drivers/opp/core.c
> index 04b4691a8aac..fae5267f5218 100644
> --- a/drivers/opp/core.c
> +++ b/drivers/opp/core.c
> @@ -939,7 +939,8 @@ static int _set_required_opps(struct device *dev,
>  	return ret;
>  }
>  
> -static void _find_current_opp(struct device *dev, struct opp_table *opp_table)
> +static struct dev_pm_opp *
> +_find_current_opp(struct device *dev, struct opp_table *opp_table)
>  {
>  	struct dev_pm_opp *opp = ERR_PTR(-ENODEV);
>  	unsigned long freq;
> @@ -961,7 +962,7 @@ static void _find_current_opp(struct device *dev, struct opp_table *opp_table)
>  		mutex_unlock(&opp_table->lock);
>  	}
>  
> -	opp_table->current_opp = opp;
> +	return opp;
>  }
>  
>  static int _disable_opp_table(struct device *dev, struct opp_table *opp_table)
> @@ -1003,7 +1004,7 @@ static int _set_opp(struct device *dev, struct opp_table *opp_table,
>  
>  	/* Find the currently set OPP if we don't know already */
>  	if (unlikely(!opp_table->current_opp))
> -		_find_current_opp(dev, opp_table);
> +		opp_table->current_opp = _find_current_opp(dev, opp_table);
>  
>  	old_opp = opp_table->current_opp;
>  
> @@ -2931,3 +2932,38 @@ int dev_pm_opp_sync_regulators(struct device *dev)
>  	return ret;
>  }

Please skip these lines. I missed to remove them during rebase and
haven't noticed until now.
Viresh Kumar Aug. 27, 2021, 3:05 a.m. UTC | #2 
On 27-08-21, 04:34, Dmitry Osipenko wrote:
> +	clk_opp_table = dev_pm_opp_set_clkname(dev, NULL);
> +	if (IS_ERR(clk_opp_table)) {
> +		dev_err(dev, "failed to set OPP clk: %pe\n", clk_opp_table);
> +		ret = PTR_ERR(clk_opp_table);
> +		goto put_hw;
> +	}

Why do you need to do it ? OPP core already does this automatically.
Dmitry Osipenko Aug. 27, 2021, 3:28 a.m. UTC | #3 
27.08.2021 06:00, Viresh Kumar пишет:
> On 27-08-21, 04:34, Dmitry Osipenko wrote:
>> +/**
>> + * dev_pm_opp_from_clk_rate() - Get OPP from current clock rate
>> + * @dev:	device for which we do this operation
>> + *
>> + * Get OPP which corresponds to the current clock rate of a device.
>> + *
>> + * Return: pointer to 'struct dev_pm_opp' on success and errorno otherwise.
>> + */
>> +struct dev_pm_opp *dev_pm_opp_from_clk_rate(struct device *dev)
> 
> I will rather call it dev_pm_opp_get_current(), and do the magic to find the
> current OPP here as well. No need to reinvent the wheel.
> 

Okay, I'll change it.
Dmitry Osipenko Aug. 27, 2021, 3:56 a.m. UTC | #4 
27.08.2021 06:47, Dmitry Osipenko пишет:
> 27.08.2021 06:28, Dmitry Osipenko пишет:
>> 27.08.2021 06:05, Viresh Kumar пишет:
>>> On 27-08-21, 04:34, Dmitry Osipenko wrote:
>>>> +	clk_opp_table = dev_pm_opp_set_clkname(dev, NULL);
>>>> +	if (IS_ERR(clk_opp_table)) {
>>>> +		dev_err(dev, "failed to set OPP clk: %pe\n", clk_opp_table);
>>>> +		ret = PTR_ERR(clk_opp_table);
>>>> +		goto put_hw;
>>>> +	}
>>>
>>> Why do you need to do it ? OPP core already does this automatically.
>>
>> Indeed, thanks.
>>
> 
> Actually, it doesn't work.
> 
> The devm_tegra_core_dev_init_opp_table() needs to set clk to support older device-tree and now OPP table already has clk being set.
> 
> WARNING: CPU: 2 PID: 92 at drivers/opp/core.c:2146 dev_pm_opp_set_clkname+0x97/0xb8
> Modules linked in:
> CPU: 2 PID: 92 Comm: kworker/u8:1 Tainted: G        W         5.14.0-rc7-next-20210826-00181-g6389463cbb0a #9318
> Hardware name: NVIDIA Tegra SoC (Flattened Device Tree)
> Workqueue: events_unbound deferred_probe_work_func
> [<c010cc91>] (unwind_backtrace) from [<c0108d35>] (show_stack+0x11/0x14)
> [<c0108d35>] (show_stack) from [<c0a6c1bd>] (dump_stack_lvl+0x2b/0x34)
> [<c0a6c1bd>] (dump_stack_lvl) from [<c011fc47>] (__warn+0xbb/0x100)
> [<c011fc47>] (__warn) from [<c0a696e3>] (warn_slowpath_fmt+0x4b/0x80)
> [<c0a696e3>] (warn_slowpath_fmt) from [<c07407b3>] (dev_pm_opp_set_clkname+0x97/0xb8)
> [<c07407b3>] (dev_pm_opp_set_clkname) from [<c07407e3>] (devm_pm_opp_set_clkname+0xf/0x64)
> [<c07407e3>] (devm_pm_opp_set_clkname) from [<c050735b>] (devm_tegra_core_dev_init_opp_table+0x23/0x144)
> [<c050735b>] (devm_tegra_core_dev_init_opp_table) from [<c05aad09>] (gr3d_probe+0x111/0x348)
> [<c05aad09>] (gr3d_probe) from [<c05ba69b>] (platform_probe+0x43/0x84)
> [<c05ba69b>] (platform_probe) from [<c05b8c01>] (really_probe.part.0+0x69/0x200)
> [<c05b8c01>] (really_probe.part.0) from [<c05b8e0b>] (__driver_probe_device+0x73/0xd4)
> [<c05b8e0b>] (__driver_probe_device) from [<c05b8ea1>] (driver_probe_device+0x35/0xd0)
> [<c05b8ea1>] (driver_probe_device) from [<c05b92a9>] (__device_attach_driver+0x75/0x98)
> [<c05b92a9>] (__device_attach_driver) from [<c05b769d>] (bus_for_each_drv+0x51/0x7c)
> [<c05b769d>] (bus_for_each_drv) from [<c05b908f>] (__device_attach+0x8b/0x104)
> [<c05b908f>] (__device_attach) from [<c05b81b3>] (bus_probe_device+0x5b/0x60)
> [<c05b81b3>] (bus_probe_device) from [<c05b5d9f>] (device_add+0x293/0x65c)
> [<c05b5d9f>] (device_add) from [<c0777a4f>] (of_platform_device_create_pdata+0x63/0x88)
> [<c0777a4f>] (of_platform_device_create_pdata) from [<c0777b7d>] (of_platform_bus_create+0xfd/0x26c)
> [<c0777b7d>] (of_platform_bus_create) from [<c0777dc5>] (of_platform_populate+0x45/0x84)
> [<c0777dc5>] (of_platform_populate) from [<c0777e5d>] (devm_of_platform_populate+0x41/0x6c)
> [<c0777e5d>] (devm_of_platform_populate) from [<c05490f9>] (host1x_probe+0x1e9/0x2c8)
> [<c05490f9>] (host1x_probe) from [<c05ba69b>] (platform_probe+0x43/0x84)
> [<c05ba69b>] (platform_probe) from [<c05b8c01>] (really_probe.part.0+0x69/0x200)
> [<c05b8c01>] (really_probe.part.0) from [<c05b8e0b>] (__driver_probe_device+0x73/0xd4)
> [<c05b8e0b>] (__driver_probe_device) from [<c05b8ea1>] (driver_probe_device+0x35/0xd0)
> [<c05b8ea1>] (driver_probe_device) from [<c05b92a9>] (__device_attach_driver+0x75/0x98)
> [<c05b92a9>] (__device_attach_driver) from [<c05b769d>] (bus_for_each_drv+0x51/0x7c)
> [<c05b769d>] (bus_for_each_drv) from [<c05b908f>] (__device_attach+0x8b/0x104)
> [<c05b908f>] (__device_attach) from [<c05b81b3>] (bus_probe_device+0x5b/0x60)
> [<c05b81b3>] (bus_probe_device) from [<c05b8493>] (deferred_probe_work_func+0x57/0x78)
> [<c05b8493>] (deferred_probe_work_func) from [<c0136f73>] (process_one_work+0x147/0x3f8)
> [<c0136f73>] (process_one_work) from [<c0137759>] (worker_thread+0x21d/0x3f4)
> [<c0137759>] (worker_thread) from [<c013c10f>] (kthread+0x123/0x140)
> [<c013c10f>] (kthread) from [<c0100135>] (ret_from_fork+0x11/0x1c)
> ---[ end trace f68728a0d3053b54 ]---
> tegra-gr3d 54180000.gr3d: tegra-soc: failed to set OPP clk: -16
> 

That's because devm_pm_opp_attach_genpd() holds the reference to OPP
table on Tegra30 which uses multiple power domains. See
gr3d_init_power() of the GR3D patch.

It works in case of a single-domain hardware.
Dmitry Osipenko Aug. 27, 2021, 4:08 a.m. UTC | #5 
27.08.2021 07:02, Viresh Kumar пишет:
> On 27-08-21, 06:47, Dmitry Osipenko wrote:
>> Actually, it doesn't work.
>>
>> The devm_tegra_core_dev_init_opp_table() needs to set clk to support older device-tree and now OPP table already has clk being set.
>>
>> WARNING: CPU: 2 PID: 92 at drivers/opp/core.c:2146 dev_pm_opp_set_clkname+0x97/0xb8
>> Modules linked in:
>> CPU: 2 PID: 92 Comm: kworker/u8:1 Tainted: G        W         5.14.0-rc7-next-20210826-00181-g6389463cbb0a #9318
>> Hardware name: NVIDIA Tegra SoC (Flattened Device Tree)
>> Workqueue: events_unbound deferred_probe_work_func
>> [<c010cc91>] (unwind_backtrace) from [<c0108d35>] (show_stack+0x11/0x14)
>> [<c0108d35>] (show_stack) from [<c0a6c1bd>] (dump_stack_lvl+0x2b/0x34)
>> [<c0a6c1bd>] (dump_stack_lvl) from [<c011fc47>] (__warn+0xbb/0x100)
>> [<c011fc47>] (__warn) from [<c0a696e3>] (warn_slowpath_fmt+0x4b/0x80)
>> [<c0a696e3>] (warn_slowpath_fmt) from [<c07407b3>] (dev_pm_opp_set_clkname+0x97/0xb8)
>> [<c07407b3>] (dev_pm_opp_set_clkname) from [<c07407e3>] (devm_pm_opp_set_clkname+0xf/0x64)
>> [<c07407e3>] (devm_pm_opp_set_clkname) from [<c050735b>] (devm_tegra_core_dev_init_opp_table+0x23/0x144)
> 
> Why are you calling this anymore ?

Older device-trees don't have OPPs, meanwhile drivers will use
dev_pm_opp_set_rate() and it requires OPP table to be set up using
devm_pm_opp_set_clkname().

The devm_tegra_core_dev_init_opp_table() is a common helper that sets up
OPP table for Tegra drivers and it sets the clk.
Dmitry Osipenko Aug. 27, 2021, 4:15 a.m. UTC | #6 
27.08.2021 07:13, Viresh Kumar пишет:
> On 27-08-21, 07:08, Dmitry Osipenko wrote:
>> 27.08.2021 07:02, Viresh Kumar пишет:
>>> On 27-08-21, 06:47, Dmitry Osipenko wrote:
>>>> Actually, it doesn't work.
>>>>
>>>> The devm_tegra_core_dev_init_opp_table() needs to set clk to support older device-tree and now OPP table already has clk being set.
>>>>
>>>> WARNING: CPU: 2 PID: 92 at drivers/opp/core.c:2146 dev_pm_opp_set_clkname+0x97/0xb8
>>>> Modules linked in:
>>>> CPU: 2 PID: 92 Comm: kworker/u8:1 Tainted: G        W         5.14.0-rc7-next-20210826-00181-g6389463cbb0a #9318
>>>> Hardware name: NVIDIA Tegra SoC (Flattened Device Tree)
>>>> Workqueue: events_unbound deferred_probe_work_func
>>>> [<c010cc91>] (unwind_backtrace) from [<c0108d35>] (show_stack+0x11/0x14)
>>>> [<c0108d35>] (show_stack) from [<c0a6c1bd>] (dump_stack_lvl+0x2b/0x34)
>>>> [<c0a6c1bd>] (dump_stack_lvl) from [<c011fc47>] (__warn+0xbb/0x100)
>>>> [<c011fc47>] (__warn) from [<c0a696e3>] (warn_slowpath_fmt+0x4b/0x80)
>>>> [<c0a696e3>] (warn_slowpath_fmt) from [<c07407b3>] (dev_pm_opp_set_clkname+0x97/0xb8)
>>>> [<c07407b3>] (dev_pm_opp_set_clkname) from [<c07407e3>] (devm_pm_opp_set_clkname+0xf/0x64)
>>>> [<c07407e3>] (devm_pm_opp_set_clkname) from [<c050735b>] (devm_tegra_core_dev_init_opp_table+0x23/0x144)
>>>
>>> Why are you calling this anymore ?
>>
>> Older device-trees don't have OPPs, meanwhile drivers will use
>> dev_pm_opp_set_rate() and it requires OPP table to be set up using
>> devm_pm_opp_set_clkname().
>>
>> The devm_tegra_core_dev_init_opp_table() is a common helper that sets up
>> OPP table for Tegra drivers and it sets the clk.
> 
> Ahh, I see. that's okay then. Just add a comment over it to specify the same.
> Doing devm_pm_opp_set_clkname(dev, NULL) is special and looks suspicious
> otherwise.

I'll add comment, thanks.