All of lore.kernel.org
 help / color / mirror / Atom feed
From: Sean Paul <seanpaul@google.com>
To: hl <hl@rock-chips.com>
Cc: "Heiko Stübner" <heiko@sntech.de>, ??? <myungjoo.ham@samsung.com>,
	姚智情 <mark.yao@rock-chips.com>,
	cw00.choi@samsung.com, "Dave Airlie" <airlied@linux.ie>,
	"Michael Turquette" <mturquette@baylibre.com>,
	dbasehore@chromium.org, "Stephen Boyd" <sboyd@codeaurora.org>,
	"Linux Kernel Mailing List" <linux-kernel@vger.kernel.org>,
	dri-devel <dri-devel@lists.freedesktop.org>,
	"Douglas Anderson" <dianders@chromium.org>,
	linux-rockchip@lists.infradead.org,
	"Kyungmin Park" <kyungmin.park@samsung.com>,
	"Linux ARM Kernel" <linux-arm-kernel@lists.infradead.org>,
	"Jon Medhurst" <tixy@linaro.org>,
	typ@rock-chips.com, sudeep.holla@arm.com,
	"Mark Rutland" <mark.rutland@arm.com>,
	linux-pm@vger.kernel.org,
	"devicetree@vger.kernel.org" <devicetree@vger.kernel.org>
Subject: Re: [PATCH v10 5/5] drm/rockchip: Add dmc notifier in vop driver
Date: Tue, 6 Sep 2016 14:55:23 -0400	[thread overview]
Message-ID: <CAOw6vbL6AxWRm7kovjwstW+qCd_9Wta3H_gvZswjeqqq9g-P8w@mail.gmail.com> (raw)
In-Reply-To: <57CF07C4.30508@rock-chips.com>

On Tue, Sep 6, 2016 at 2:15 PM, hl <hl@rock-chips.com> wrote:
> Hi Sean,
>
>
> On 2016年09月07日 01:18, Sean Paul wrote:
>>
>> On Mon, Sep 5, 2016 at 1:06 AM, Lin Huang <hl@rock-chips.com> wrote:
>>>
>>> when in ddr frequency scaling process, vop can not do enable or
>>> disable operation, since in dcf we check vop clock to see whether
>>> vop work. If vop work, dcf do ddr frequency scaling when vop
>>> in vblank status, and we need to read vop register to check whether
>>> vop go into vblank status. If vop not work, dcf can do ddr frequency
>>> any time. So when do ddr frequency scaling, you disabled or enable
>>> vop, there may two bad thing happen: 1, the panel flicker(when vop from
>>> disable status change to enable). 2, kernel hang (when vop from enable
>>> status change to disable, dcf need to read vblank status, but if you
>>> disable
>>> vop clock, it can not get the status, it will lead soc dead) So we need
>>> register to devfreq notifier, and we can get the dmc status. Also, when
>>> there have two vop enabled, we need to disable dmc, since dcf only base
>>> on one vop vblank time, so the other panel will flicker when do ddr
>>> frequency scaling.
>>>
>>> Signed-off-by: Lin Huang <hl@rock-chips.com>
>>> Reviewed-by: Chanwoo Choi <cw00.choi@samsung.com>
>>> ---
>>> Changes in v10:
>>> - None
>>>
>>> Changes in v9:
>>> - None
>>>
>>> Changes in v8:
>>> - None
>>>
>>> Changes in v7:
>>> - None
>>>
>>> Changes in v6:
>>> - fix a build error
>>>
>>> Changes in v5:
>>> - improve some nits
>>>
>>> Changes in v4:
>>> - register notifier to devfreq_register_notifier
>>> - use DEVFREQ_PRECHANGE and DEVFREQ_POSTCHANGE to get dmc status
>>> - when two vop enable, disable dmc
>>> - when two vop back to one vop, enable dmc
>>>
>>> Changes in v3:
>>> - when do vop eanble/disable, dmc will wait until it finish
>>>
>>> Changes in v2:
>>> - None
>>>
>>> Changes in v1:
>>> - use wait_event instead usleep
>>>
>>>   drivers/gpu/drm/rockchip/rockchip_drm_vop.c | 116
>>> ++++++++++++++++++++++++++++
>>>   1 file changed, 116 insertions(+)
>>>
>>> diff --git a/drivers/gpu/drm/rockchip/rockchip_drm_vop.c
>>> b/drivers/gpu/drm/rockchip/rockchip_drm_vop.c
>>> index efbc41a..a73f3aa 100644
>>> --- a/drivers/gpu/drm/rockchip/rockchip_drm_vop.c
>>> +++ b/drivers/gpu/drm/rockchip/rockchip_drm_vop.c
>>> @@ -19,6 +19,8 @@
>>>   #include <drm/drm_crtc_helper.h>
>>>   #include <drm/drm_plane_helper.h>
>>>
>>> +#include <linux/devfreq.h>
>>> +#include <linux/devfreq-event.h>
>>>   #include <linux/kernel.h>
>>>   #include <linux/module.h>
>>>   #include <linux/platform_device.h>
>>> @@ -118,6 +120,13 @@ struct vop {
>>>
>>>          const struct vop_data *data;
>>>
>>> +       struct devfreq *devfreq;
>>> +       struct devfreq_event_dev *devfreq_event_dev;
>>> +       struct notifier_block dmc_nb;
>>> +       int dmc_in_process;
>>> +       int vop_switch_status;
>>> +       wait_queue_head_t wait_dmc_queue;
>>> +       wait_queue_head_t wait_vop_switch_queue;
>>>          uint32_t *regsbak;
>>>          void __iomem *regs;
>>>
>>> @@ -428,11 +437,47 @@ static void vop_dsp_hold_valid_irq_disable(struct
>>> vop *vop)
>>>          spin_unlock_irqrestore(&vop->irq_lock, flags);
>>>   }
>>>
>>> +static int dmc_notify(struct notifier_block *nb, unsigned long event,
>>> +                     void *data)
>>> +{
>>> +       struct vop *vop = container_of(nb, struct vop, dmc_nb);
>>> +
>>> +       if (event == DEVFREQ_PRECHANGE) {
>>> +               /*
>>> +                * check if vop in enable or disable process,
>>> +                * if yes, wait until it finishes, use 200ms as
>>> +                * timeout.
>>> +                */
>>> +               if (!wait_event_timeout(vop->wait_vop_switch_queue,
>>> +                                       !vop->vop_switch_status, HZ / 5))
>>> +                       dev_warn(vop->dev,
>>> +                                "Timeout waiting for vop swtich
>>> status\n");
>>> +               vop->dmc_in_process = 1;
>>> +       } else if (event == DEVFREQ_POSTCHANGE) {
>>> +               vop->dmc_in_process = 0;
>>> +               wake_up(&vop->wait_dmc_queue);
>>> +       }
>>> +
>>> +       return NOTIFY_OK;
>>> +}
>>> +
>>>   static int vop_enable(struct drm_crtc *crtc)
>>>   {
>>>          struct vop *vop = to_vop(crtc);
>>> +       int num_enabled_crtc = 0;
>>>          int ret;
>>>
>>> +       /*
>>> +        * if in dmc scaling frequency process, wait until it finishes
>>> +        * use 200ms as timeout time.
>>> +        */
>>> +       if (!wait_event_timeout(vop->wait_dmc_queue,
>>> +                               !vop->dmc_in_process, HZ / 5))
>>> +               dev_warn(vop->dev,
>>> +                        "Timeout waiting for dmc when vop enable\n");
>>> +
>>
>>
>> This wait_event_timeout code is terribly racey (same goes above and
>> below).
>
>   No i use the vop_switch_status and dmc_in_progress to handle the vop and
> dmc racey,


So what happens if dmc_in_progress becomes 1 right here, or anywhere
during vop_enable()?

Sean




>    I assume this can fix the racey now. I do not better a idea now.
>>
>>
>>
>>> +       vop->vop_switch_status = 1;
>>> +
>>>          ret = pm_runtime_get_sync(vop->dev);
>>>          if (ret < 0) {
>>>                  dev_err(vop->dev, "failed to get pm runtime: %d\n",
>>> ret);
>>> @@ -479,6 +524,21 @@ static int vop_enable(struct drm_crtc *crtc)
>>>
>>>          drm_crtc_vblank_on(crtc);
>>>
>>> +       vop->vop_switch_status = 0;
>>> +       wake_up(&vop->wait_vop_switch_queue);
>>> +
>>> +       /* check how many VOPs in use now */
>>> +       drm_for_each_crtc(crtc, vop->drm_dev) {
>>> +               if (crtc->state->enable)
>>
>> I think you really want to check active, instead of enable.
>
> Okay, i will check it, thanks.
>
>>
>>> +                       num_enabled_crtc++;
>>> +       }
>>> +
>>> +       /* if enable two vop, need to disable dmc */
>>> +       if ((num_enabled_crtc > 1) && vop->devfreq) {
>>> +               if (vop->devfreq_event_dev)
>>> +
>>> devfreq_event_disable_edev(vop->devfreq_event_dev);
>>> +               devfreq_suspend_device(vop->devfreq);
>>> +       }
>>
>> This really feels like something that should be handled somewhere
>> else. I don't fully understand how this works, but it seems like this
>> dependency should be handled where it actually matters, rather than
>> building in a seemingly arbitrary restriction in the vop driver.
>>
>> Even if this is the best place for it, this needs to be refactored to
>> eliminate the races that exist now.
>>
>> Sean
>>
>>>          return 0;
>>>
>>>   err_disable_aclk:
>>> @@ -489,17 +549,31 @@ err_disable_hclk:
>>>          clk_disable(vop->hclk);
>>>   err_put_pm_runtime:
>>>          pm_runtime_put_sync(vop->dev);
>>> +       vop->vop_switch_status = 0;
>>> +       wake_up(&vop->wait_vop_switch_queue);
>>>          return ret;
>>>   }
>>>
>>>   static void vop_crtc_disable(struct drm_crtc *crtc)
>>>   {
>>>          struct vop *vop = to_vop(crtc);
>>> +       int num_enabled_crtc = 0;
>>>          int i;
>>>
>>>          WARN_ON(vop->event);
>>>
>>>          /*
>>> +        * if in dmc scaling frequency process, wait until it finish
>>> +        * use 200ms as timeout time.
>>> +        */
>>> +       if (!wait_event_timeout(vop->wait_dmc_queue,
>>> +                               !vop->dmc_in_process, HZ / 5))
>>> +               dev_warn(vop->dev,
>>> +                        "Timeout waiting for dmc when vop disable\n");
>>> +
>>> +       vop->vop_switch_status = 1;
>>> +
>>> +       /*
>>>           * We need to make sure that all windows are disabled before we
>>>           * disable that crtc. Otherwise we might try to scan from a
>>> destroyed
>>>           * buffer later.
>>> @@ -555,6 +629,24 @@ static void vop_crtc_disable(struct drm_crtc *crtc)
>>>                  spin_unlock_irq(&crtc->dev->event_lock);
>>>
>>>                  crtc->state->event = NULL;
>>> +
>>> +       vop->vop_switch_status = 0;
>>> +       wake_up(&vop->wait_vop_switch_queue);
>>> +
>>> +       /* check how many VOPs in use now */
>>> +       drm_for_each_crtc(crtc, vop->drm_dev) {
>>> +               if (crtc->state->enable)
>>> +                       num_enabled_crtc++;
>>> +       }
>>> +
>>> +       /*
>>> +        * if num_enabled_crtc = 1 now, it means 2 vop enabled
>>> +        * change to 1 vop enabled  need to enable dmc again.
>>> +        */
>>> +       if ((num_enabled_crtc == 1) && vop->devfreq) {
>>> +               if (vop->devfreq_event_dev)
>>> +
>>> devfreq_event_enable_edev(vop->devfreq_event_dev);
>>> +               devfreq_resume_device(vop->devfreq);
>>>          }
>>>   }
>>>
>>> @@ -1413,6 +1505,8 @@ static int vop_bind(struct device *dev, struct
>>> device *master, void *data)
>>>          struct drm_device *drm_dev = data;
>>>          struct vop *vop;
>>>          struct resource *res;
>>> +       struct devfreq *devfreq;
>>> +       struct devfreq_event_dev *event_dev;
>>>          size_t alloc_size;
>>>          int ret, irq;
>>>
>>> @@ -1474,6 +1568,28 @@ static int vop_bind(struct device *dev, struct
>>> device *master, void *data)
>>>                  return ret;
>>>
>>>          pm_runtime_enable(&pdev->dev);
>>> +
>>> +       init_waitqueue_head(&vop->wait_vop_switch_queue);
>>> +       vop->vop_switch_status = 0;
>>> +       init_waitqueue_head(&vop->wait_dmc_queue);
>>> +       vop->dmc_in_process = 0;
>>> +
>>> +       devfreq = devfreq_get_devfreq_by_phandle(dev, 0);
>>> +       if (IS_ERR(devfreq))
>>> +               goto out;
>>> +
>>> +       vop->devfreq = devfreq;
>>> +       vop->dmc_nb.notifier_call = dmc_notify;
>>> +       devfreq_register_notifier(vop->devfreq, &vop->dmc_nb,
>>> +                                 DEVFREQ_TRANSITION_NOTIFIER);
>>> +
>>> +       event_dev =
>>> devfreq_event_get_edev_by_phandle(vop->devfreq->dev.parent,
>>> +                                                     0);
>>> +       if (IS_ERR(event_dev))
>>> +               goto out;
>>> +
>>> +       vop->devfreq_event_dev = event_dev;
>>> +out:
>>>          return 0;
>>>   }
>>>
>>> --
>>> 2.6.6
>>>
>>
>>
>
> --
> Lin Huang
>
>

WARNING: multiple messages have this Message-ID (diff)
From: Sean Paul <seanpaul-hpIqsD4AKlfQT0dZR+AlfA@public.gmane.org>
To: hl <hl-TNX95d0MmH7DzftRWevZcw@public.gmane.org>
Cc: "Jon Medhurst" <tixy-QSEj5FYQhm4dnm+yROfE0A@public.gmane.org>,
	"Mark Rutland" <mark.rutland-5wv7dgnIgG8@public.gmane.org>,
	"devicetree-u79uwXL29TY76Z2rM5mHXA@public.gmane.org"
	<devicetree-u79uwXL29TY76Z2rM5mHXA@public.gmane.org>,
	typ-TNX95d0MmH7DzftRWevZcw@public.gmane.org,
	"Heiko Stübner" <heiko-4mtYJXux2i+zQB+pC5nmwQ@public.gmane.org>,
	linux-rockchip-IAPFreCvJWM7uuMidbF8XUB+6BGkLq7r@public.gmane.org,
	"Dave Airlie" <airlied-cv59FeDIM0c@public.gmane.org>,
	"Michael Turquette"
	<mturquette-rdvid1DuHRBWk0Htik3J/w@public.gmane.org>,
	dbasehore-F7+t8E8rja9g9hUCZPvPmw@public.gmane.org,
	"Stephen Boyd" <sboyd-sgV2jX0FEOL9JmXXK+q4OQ@public.gmane.org>,
	"Linux Kernel Mailing List"
	<linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org>,
	dri-devel
	<dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org>,
	"Douglas Anderson"
	<dianders-F7+t8E8rja9g9hUCZPvPmw@public.gmane.org>,
	cw00.choi-Sze3O3UU22JBDgjK7y7TUQ@public.gmane.org,
	"Kyungmin Park"
	<kyungmin.park-Sze3O3UU22JBDgjK7y7TUQ@public.gmane.org>,
	??? <myungjoo.ham-Sze3O3UU22JBDgjK7y7TUQ@public.gmane.org>,
	sudeep.holla-5wv7dgnIgG8@public.gmane.org,
	linux-pm-u79uwXL29TY76Z2rM5mHXA@public.gmane.org,
	"Linux ARM Kernel"
	<linux-arm-kernel-IAPFreCvJWM7uuMidbF8XUB+6BGkLq7r@public.gmane.org>,
	姚智情 <mark.yao-TNX95d0MmH7DzftRWevZcw@public.gmane.org>
Subject: Re: [PATCH v10 5/5] drm/rockchip: Add dmc notifier in vop driver
Date: Tue, 6 Sep 2016 14:55:23 -0400	[thread overview]
Message-ID: <CAOw6vbL6AxWRm7kovjwstW+qCd_9Wta3H_gvZswjeqqq9g-P8w@mail.gmail.com> (raw)
In-Reply-To: <57CF07C4.30508-TNX95d0MmH7DzftRWevZcw@public.gmane.org>

On Tue, Sep 6, 2016 at 2:15 PM, hl <hl@rock-chips.com> wrote:
> Hi Sean,
>
>
> On 2016年09月07日 01:18, Sean Paul wrote:
>>
>> On Mon, Sep 5, 2016 at 1:06 AM, Lin Huang <hl@rock-chips.com> wrote:
>>>
>>> when in ddr frequency scaling process, vop can not do enable or
>>> disable operation, since in dcf we check vop clock to see whether
>>> vop work. If vop work, dcf do ddr frequency scaling when vop
>>> in vblank status, and we need to read vop register to check whether
>>> vop go into vblank status. If vop not work, dcf can do ddr frequency
>>> any time. So when do ddr frequency scaling, you disabled or enable
>>> vop, there may two bad thing happen: 1, the panel flicker(when vop from
>>> disable status change to enable). 2, kernel hang (when vop from enable
>>> status change to disable, dcf need to read vblank status, but if you
>>> disable
>>> vop clock, it can not get the status, it will lead soc dead) So we need
>>> register to devfreq notifier, and we can get the dmc status. Also, when
>>> there have two vop enabled, we need to disable dmc, since dcf only base
>>> on one vop vblank time, so the other panel will flicker when do ddr
>>> frequency scaling.
>>>
>>> Signed-off-by: Lin Huang <hl@rock-chips.com>
>>> Reviewed-by: Chanwoo Choi <cw00.choi@samsung.com>
>>> ---
>>> Changes in v10:
>>> - None
>>>
>>> Changes in v9:
>>> - None
>>>
>>> Changes in v8:
>>> - None
>>>
>>> Changes in v7:
>>> - None
>>>
>>> Changes in v6:
>>> - fix a build error
>>>
>>> Changes in v5:
>>> - improve some nits
>>>
>>> Changes in v4:
>>> - register notifier to devfreq_register_notifier
>>> - use DEVFREQ_PRECHANGE and DEVFREQ_POSTCHANGE to get dmc status
>>> - when two vop enable, disable dmc
>>> - when two vop back to one vop, enable dmc
>>>
>>> Changes in v3:
>>> - when do vop eanble/disable, dmc will wait until it finish
>>>
>>> Changes in v2:
>>> - None
>>>
>>> Changes in v1:
>>> - use wait_event instead usleep
>>>
>>>   drivers/gpu/drm/rockchip/rockchip_drm_vop.c | 116
>>> ++++++++++++++++++++++++++++
>>>   1 file changed, 116 insertions(+)
>>>
>>> diff --git a/drivers/gpu/drm/rockchip/rockchip_drm_vop.c
>>> b/drivers/gpu/drm/rockchip/rockchip_drm_vop.c
>>> index efbc41a..a73f3aa 100644
>>> --- a/drivers/gpu/drm/rockchip/rockchip_drm_vop.c
>>> +++ b/drivers/gpu/drm/rockchip/rockchip_drm_vop.c
>>> @@ -19,6 +19,8 @@
>>>   #include <drm/drm_crtc_helper.h>
>>>   #include <drm/drm_plane_helper.h>
>>>
>>> +#include <linux/devfreq.h>
>>> +#include <linux/devfreq-event.h>
>>>   #include <linux/kernel.h>
>>>   #include <linux/module.h>
>>>   #include <linux/platform_device.h>
>>> @@ -118,6 +120,13 @@ struct vop {
>>>
>>>          const struct vop_data *data;
>>>
>>> +       struct devfreq *devfreq;
>>> +       struct devfreq_event_dev *devfreq_event_dev;
>>> +       struct notifier_block dmc_nb;
>>> +       int dmc_in_process;
>>> +       int vop_switch_status;
>>> +       wait_queue_head_t wait_dmc_queue;
>>> +       wait_queue_head_t wait_vop_switch_queue;
>>>          uint32_t *regsbak;
>>>          void __iomem *regs;
>>>
>>> @@ -428,11 +437,47 @@ static void vop_dsp_hold_valid_irq_disable(struct
>>> vop *vop)
>>>          spin_unlock_irqrestore(&vop->irq_lock, flags);
>>>   }
>>>
>>> +static int dmc_notify(struct notifier_block *nb, unsigned long event,
>>> +                     void *data)
>>> +{
>>> +       struct vop *vop = container_of(nb, struct vop, dmc_nb);
>>> +
>>> +       if (event == DEVFREQ_PRECHANGE) {
>>> +               /*
>>> +                * check if vop in enable or disable process,
>>> +                * if yes, wait until it finishes, use 200ms as
>>> +                * timeout.
>>> +                */
>>> +               if (!wait_event_timeout(vop->wait_vop_switch_queue,
>>> +                                       !vop->vop_switch_status, HZ / 5))
>>> +                       dev_warn(vop->dev,
>>> +                                "Timeout waiting for vop swtich
>>> status\n");
>>> +               vop->dmc_in_process = 1;
>>> +       } else if (event == DEVFREQ_POSTCHANGE) {
>>> +               vop->dmc_in_process = 0;
>>> +               wake_up(&vop->wait_dmc_queue);
>>> +       }
>>> +
>>> +       return NOTIFY_OK;
>>> +}
>>> +
>>>   static int vop_enable(struct drm_crtc *crtc)
>>>   {
>>>          struct vop *vop = to_vop(crtc);
>>> +       int num_enabled_crtc = 0;
>>>          int ret;
>>>
>>> +       /*
>>> +        * if in dmc scaling frequency process, wait until it finishes
>>> +        * use 200ms as timeout time.
>>> +        */
>>> +       if (!wait_event_timeout(vop->wait_dmc_queue,
>>> +                               !vop->dmc_in_process, HZ / 5))
>>> +               dev_warn(vop->dev,
>>> +                        "Timeout waiting for dmc when vop enable\n");
>>> +
>>
>>
>> This wait_event_timeout code is terribly racey (same goes above and
>> below).
>
>   No i use the vop_switch_status and dmc_in_progress to handle the vop and
> dmc racey,


So what happens if dmc_in_progress becomes 1 right here, or anywhere
during vop_enable()?

Sean




>    I assume this can fix the racey now. I do not better a idea now.
>>
>>
>>
>>> +       vop->vop_switch_status = 1;
>>> +
>>>          ret = pm_runtime_get_sync(vop->dev);
>>>          if (ret < 0) {
>>>                  dev_err(vop->dev, "failed to get pm runtime: %d\n",
>>> ret);
>>> @@ -479,6 +524,21 @@ static int vop_enable(struct drm_crtc *crtc)
>>>
>>>          drm_crtc_vblank_on(crtc);
>>>
>>> +       vop->vop_switch_status = 0;
>>> +       wake_up(&vop->wait_vop_switch_queue);
>>> +
>>> +       /* check how many VOPs in use now */
>>> +       drm_for_each_crtc(crtc, vop->drm_dev) {
>>> +               if (crtc->state->enable)
>>
>> I think you really want to check active, instead of enable.
>
> Okay, i will check it, thanks.
>
>>
>>> +                       num_enabled_crtc++;
>>> +       }
>>> +
>>> +       /* if enable two vop, need to disable dmc */
>>> +       if ((num_enabled_crtc > 1) && vop->devfreq) {
>>> +               if (vop->devfreq_event_dev)
>>> +
>>> devfreq_event_disable_edev(vop->devfreq_event_dev);
>>> +               devfreq_suspend_device(vop->devfreq);
>>> +       }
>>
>> This really feels like something that should be handled somewhere
>> else. I don't fully understand how this works, but it seems like this
>> dependency should be handled where it actually matters, rather than
>> building in a seemingly arbitrary restriction in the vop driver.
>>
>> Even if this is the best place for it, this needs to be refactored to
>> eliminate the races that exist now.
>>
>> Sean
>>
>>>          return 0;
>>>
>>>   err_disable_aclk:
>>> @@ -489,17 +549,31 @@ err_disable_hclk:
>>>          clk_disable(vop->hclk);
>>>   err_put_pm_runtime:
>>>          pm_runtime_put_sync(vop->dev);
>>> +       vop->vop_switch_status = 0;
>>> +       wake_up(&vop->wait_vop_switch_queue);
>>>          return ret;
>>>   }
>>>
>>>   static void vop_crtc_disable(struct drm_crtc *crtc)
>>>   {
>>>          struct vop *vop = to_vop(crtc);
>>> +       int num_enabled_crtc = 0;
>>>          int i;
>>>
>>>          WARN_ON(vop->event);
>>>
>>>          /*
>>> +        * if in dmc scaling frequency process, wait until it finish
>>> +        * use 200ms as timeout time.
>>> +        */
>>> +       if (!wait_event_timeout(vop->wait_dmc_queue,
>>> +                               !vop->dmc_in_process, HZ / 5))
>>> +               dev_warn(vop->dev,
>>> +                        "Timeout waiting for dmc when vop disable\n");
>>> +
>>> +       vop->vop_switch_status = 1;
>>> +
>>> +       /*
>>>           * We need to make sure that all windows are disabled before we
>>>           * disable that crtc. Otherwise we might try to scan from a
>>> destroyed
>>>           * buffer later.
>>> @@ -555,6 +629,24 @@ static void vop_crtc_disable(struct drm_crtc *crtc)
>>>                  spin_unlock_irq(&crtc->dev->event_lock);
>>>
>>>                  crtc->state->event = NULL;
>>> +
>>> +       vop->vop_switch_status = 0;
>>> +       wake_up(&vop->wait_vop_switch_queue);
>>> +
>>> +       /* check how many VOPs in use now */
>>> +       drm_for_each_crtc(crtc, vop->drm_dev) {
>>> +               if (crtc->state->enable)
>>> +                       num_enabled_crtc++;
>>> +       }
>>> +
>>> +       /*
>>> +        * if num_enabled_crtc = 1 now, it means 2 vop enabled
>>> +        * change to 1 vop enabled  need to enable dmc again.
>>> +        */
>>> +       if ((num_enabled_crtc == 1) && vop->devfreq) {
>>> +               if (vop->devfreq_event_dev)
>>> +
>>> devfreq_event_enable_edev(vop->devfreq_event_dev);
>>> +               devfreq_resume_device(vop->devfreq);
>>>          }
>>>   }
>>>
>>> @@ -1413,6 +1505,8 @@ static int vop_bind(struct device *dev, struct
>>> device *master, void *data)
>>>          struct drm_device *drm_dev = data;
>>>          struct vop *vop;
>>>          struct resource *res;
>>> +       struct devfreq *devfreq;
>>> +       struct devfreq_event_dev *event_dev;
>>>          size_t alloc_size;
>>>          int ret, irq;
>>>
>>> @@ -1474,6 +1568,28 @@ static int vop_bind(struct device *dev, struct
>>> device *master, void *data)
>>>                  return ret;
>>>
>>>          pm_runtime_enable(&pdev->dev);
>>> +
>>> +       init_waitqueue_head(&vop->wait_vop_switch_queue);
>>> +       vop->vop_switch_status = 0;
>>> +       init_waitqueue_head(&vop->wait_dmc_queue);
>>> +       vop->dmc_in_process = 0;
>>> +
>>> +       devfreq = devfreq_get_devfreq_by_phandle(dev, 0);
>>> +       if (IS_ERR(devfreq))
>>> +               goto out;
>>> +
>>> +       vop->devfreq = devfreq;
>>> +       vop->dmc_nb.notifier_call = dmc_notify;
>>> +       devfreq_register_notifier(vop->devfreq, &vop->dmc_nb,
>>> +                                 DEVFREQ_TRANSITION_NOTIFIER);
>>> +
>>> +       event_dev =
>>> devfreq_event_get_edev_by_phandle(vop->devfreq->dev.parent,
>>> +                                                     0);
>>> +       if (IS_ERR(event_dev))
>>> +               goto out;
>>> +
>>> +       vop->devfreq_event_dev = event_dev;
>>> +out:
>>>          return 0;
>>>   }
>>>
>>> --
>>> 2.6.6
>>>
>>
>>
>
> --
> Lin Huang
>
>

_______________________________________________
Linux-rockchip mailing list
Linux-rockchip@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-rockchip

WARNING: multiple messages have this Message-ID (diff)
From: seanpaul@google.com (Sean Paul)
To: linux-arm-kernel@lists.infradead.org
Subject: [PATCH v10 5/5] drm/rockchip: Add dmc notifier in vop driver
Date: Tue, 6 Sep 2016 14:55:23 -0400	[thread overview]
Message-ID: <CAOw6vbL6AxWRm7kovjwstW+qCd_9Wta3H_gvZswjeqqq9g-P8w@mail.gmail.com> (raw)
In-Reply-To: <57CF07C4.30508@rock-chips.com>

On Tue, Sep 6, 2016 at 2:15 PM, hl <hl@rock-chips.com> wrote:
> Hi Sean,
>
>
> On 2016?09?07? 01:18, Sean Paul wrote:
>>
>> On Mon, Sep 5, 2016 at 1:06 AM, Lin Huang <hl@rock-chips.com> wrote:
>>>
>>> when in ddr frequency scaling process, vop can not do enable or
>>> disable operation, since in dcf we check vop clock to see whether
>>> vop work. If vop work, dcf do ddr frequency scaling when vop
>>> in vblank status, and we need to read vop register to check whether
>>> vop go into vblank status. If vop not work, dcf can do ddr frequency
>>> any time. So when do ddr frequency scaling, you disabled or enable
>>> vop, there may two bad thing happen: 1, the panel flicker(when vop from
>>> disable status change to enable). 2, kernel hang (when vop from enable
>>> status change to disable, dcf need to read vblank status, but if you
>>> disable
>>> vop clock, it can not get the status, it will lead soc dead) So we need
>>> register to devfreq notifier, and we can get the dmc status. Also, when
>>> there have two vop enabled, we need to disable dmc, since dcf only base
>>> on one vop vblank time, so the other panel will flicker when do ddr
>>> frequency scaling.
>>>
>>> Signed-off-by: Lin Huang <hl@rock-chips.com>
>>> Reviewed-by: Chanwoo Choi <cw00.choi@samsung.com>
>>> ---
>>> Changes in v10:
>>> - None
>>>
>>> Changes in v9:
>>> - None
>>>
>>> Changes in v8:
>>> - None
>>>
>>> Changes in v7:
>>> - None
>>>
>>> Changes in v6:
>>> - fix a build error
>>>
>>> Changes in v5:
>>> - improve some nits
>>>
>>> Changes in v4:
>>> - register notifier to devfreq_register_notifier
>>> - use DEVFREQ_PRECHANGE and DEVFREQ_POSTCHANGE to get dmc status
>>> - when two vop enable, disable dmc
>>> - when two vop back to one vop, enable dmc
>>>
>>> Changes in v3:
>>> - when do vop eanble/disable, dmc will wait until it finish
>>>
>>> Changes in v2:
>>> - None
>>>
>>> Changes in v1:
>>> - use wait_event instead usleep
>>>
>>>   drivers/gpu/drm/rockchip/rockchip_drm_vop.c | 116
>>> ++++++++++++++++++++++++++++
>>>   1 file changed, 116 insertions(+)
>>>
>>> diff --git a/drivers/gpu/drm/rockchip/rockchip_drm_vop.c
>>> b/drivers/gpu/drm/rockchip/rockchip_drm_vop.c
>>> index efbc41a..a73f3aa 100644
>>> --- a/drivers/gpu/drm/rockchip/rockchip_drm_vop.c
>>> +++ b/drivers/gpu/drm/rockchip/rockchip_drm_vop.c
>>> @@ -19,6 +19,8 @@
>>>   #include <drm/drm_crtc_helper.h>
>>>   #include <drm/drm_plane_helper.h>
>>>
>>> +#include <linux/devfreq.h>
>>> +#include <linux/devfreq-event.h>
>>>   #include <linux/kernel.h>
>>>   #include <linux/module.h>
>>>   #include <linux/platform_device.h>
>>> @@ -118,6 +120,13 @@ struct vop {
>>>
>>>          const struct vop_data *data;
>>>
>>> +       struct devfreq *devfreq;
>>> +       struct devfreq_event_dev *devfreq_event_dev;
>>> +       struct notifier_block dmc_nb;
>>> +       int dmc_in_process;
>>> +       int vop_switch_status;
>>> +       wait_queue_head_t wait_dmc_queue;
>>> +       wait_queue_head_t wait_vop_switch_queue;
>>>          uint32_t *regsbak;
>>>          void __iomem *regs;
>>>
>>> @@ -428,11 +437,47 @@ static void vop_dsp_hold_valid_irq_disable(struct
>>> vop *vop)
>>>          spin_unlock_irqrestore(&vop->irq_lock, flags);
>>>   }
>>>
>>> +static int dmc_notify(struct notifier_block *nb, unsigned long event,
>>> +                     void *data)
>>> +{
>>> +       struct vop *vop = container_of(nb, struct vop, dmc_nb);
>>> +
>>> +       if (event == DEVFREQ_PRECHANGE) {
>>> +               /*
>>> +                * check if vop in enable or disable process,
>>> +                * if yes, wait until it finishes, use 200ms as
>>> +                * timeout.
>>> +                */
>>> +               if (!wait_event_timeout(vop->wait_vop_switch_queue,
>>> +                                       !vop->vop_switch_status, HZ / 5))
>>> +                       dev_warn(vop->dev,
>>> +                                "Timeout waiting for vop swtich
>>> status\n");
>>> +               vop->dmc_in_process = 1;
>>> +       } else if (event == DEVFREQ_POSTCHANGE) {
>>> +               vop->dmc_in_process = 0;
>>> +               wake_up(&vop->wait_dmc_queue);
>>> +       }
>>> +
>>> +       return NOTIFY_OK;
>>> +}
>>> +
>>>   static int vop_enable(struct drm_crtc *crtc)
>>>   {
>>>          struct vop *vop = to_vop(crtc);
>>> +       int num_enabled_crtc = 0;
>>>          int ret;
>>>
>>> +       /*
>>> +        * if in dmc scaling frequency process, wait until it finishes
>>> +        * use 200ms as timeout time.
>>> +        */
>>> +       if (!wait_event_timeout(vop->wait_dmc_queue,
>>> +                               !vop->dmc_in_process, HZ / 5))
>>> +               dev_warn(vop->dev,
>>> +                        "Timeout waiting for dmc when vop enable\n");
>>> +
>>
>>
>> This wait_event_timeout code is terribly racey (same goes above and
>> below).
>
>   No i use the vop_switch_status and dmc_in_progress to handle the vop and
> dmc racey,


So what happens if dmc_in_progress becomes 1 right here, or anywhere
during vop_enable()?

Sean




>    I assume this can fix the racey now. I do not better a idea now.
>>
>>
>>
>>> +       vop->vop_switch_status = 1;
>>> +
>>>          ret = pm_runtime_get_sync(vop->dev);
>>>          if (ret < 0) {
>>>                  dev_err(vop->dev, "failed to get pm runtime: %d\n",
>>> ret);
>>> @@ -479,6 +524,21 @@ static int vop_enable(struct drm_crtc *crtc)
>>>
>>>          drm_crtc_vblank_on(crtc);
>>>
>>> +       vop->vop_switch_status = 0;
>>> +       wake_up(&vop->wait_vop_switch_queue);
>>> +
>>> +       /* check how many VOPs in use now */
>>> +       drm_for_each_crtc(crtc, vop->drm_dev) {
>>> +               if (crtc->state->enable)
>>
>> I think you really want to check active, instead of enable.
>
> Okay, i will check it, thanks.
>
>>
>>> +                       num_enabled_crtc++;
>>> +       }
>>> +
>>> +       /* if enable two vop, need to disable dmc */
>>> +       if ((num_enabled_crtc > 1) && vop->devfreq) {
>>> +               if (vop->devfreq_event_dev)
>>> +
>>> devfreq_event_disable_edev(vop->devfreq_event_dev);
>>> +               devfreq_suspend_device(vop->devfreq);
>>> +       }
>>
>> This really feels like something that should be handled somewhere
>> else. I don't fully understand how this works, but it seems like this
>> dependency should be handled where it actually matters, rather than
>> building in a seemingly arbitrary restriction in the vop driver.
>>
>> Even if this is the best place for it, this needs to be refactored to
>> eliminate the races that exist now.
>>
>> Sean
>>
>>>          return 0;
>>>
>>>   err_disable_aclk:
>>> @@ -489,17 +549,31 @@ err_disable_hclk:
>>>          clk_disable(vop->hclk);
>>>   err_put_pm_runtime:
>>>          pm_runtime_put_sync(vop->dev);
>>> +       vop->vop_switch_status = 0;
>>> +       wake_up(&vop->wait_vop_switch_queue);
>>>          return ret;
>>>   }
>>>
>>>   static void vop_crtc_disable(struct drm_crtc *crtc)
>>>   {
>>>          struct vop *vop = to_vop(crtc);
>>> +       int num_enabled_crtc = 0;
>>>          int i;
>>>
>>>          WARN_ON(vop->event);
>>>
>>>          /*
>>> +        * if in dmc scaling frequency process, wait until it finish
>>> +        * use 200ms as timeout time.
>>> +        */
>>> +       if (!wait_event_timeout(vop->wait_dmc_queue,
>>> +                               !vop->dmc_in_process, HZ / 5))
>>> +               dev_warn(vop->dev,
>>> +                        "Timeout waiting for dmc when vop disable\n");
>>> +
>>> +       vop->vop_switch_status = 1;
>>> +
>>> +       /*
>>>           * We need to make sure that all windows are disabled before we
>>>           * disable that crtc. Otherwise we might try to scan from a
>>> destroyed
>>>           * buffer later.
>>> @@ -555,6 +629,24 @@ static void vop_crtc_disable(struct drm_crtc *crtc)
>>>                  spin_unlock_irq(&crtc->dev->event_lock);
>>>
>>>                  crtc->state->event = NULL;
>>> +
>>> +       vop->vop_switch_status = 0;
>>> +       wake_up(&vop->wait_vop_switch_queue);
>>> +
>>> +       /* check how many VOPs in use now */
>>> +       drm_for_each_crtc(crtc, vop->drm_dev) {
>>> +               if (crtc->state->enable)
>>> +                       num_enabled_crtc++;
>>> +       }
>>> +
>>> +       /*
>>> +        * if num_enabled_crtc = 1 now, it means 2 vop enabled
>>> +        * change to 1 vop enabled  need to enable dmc again.
>>> +        */
>>> +       if ((num_enabled_crtc == 1) && vop->devfreq) {
>>> +               if (vop->devfreq_event_dev)
>>> +
>>> devfreq_event_enable_edev(vop->devfreq_event_dev);
>>> +               devfreq_resume_device(vop->devfreq);
>>>          }
>>>   }
>>>
>>> @@ -1413,6 +1505,8 @@ static int vop_bind(struct device *dev, struct
>>> device *master, void *data)
>>>          struct drm_device *drm_dev = data;
>>>          struct vop *vop;
>>>          struct resource *res;
>>> +       struct devfreq *devfreq;
>>> +       struct devfreq_event_dev *event_dev;
>>>          size_t alloc_size;
>>>          int ret, irq;
>>>
>>> @@ -1474,6 +1568,28 @@ static int vop_bind(struct device *dev, struct
>>> device *master, void *data)
>>>                  return ret;
>>>
>>>          pm_runtime_enable(&pdev->dev);
>>> +
>>> +       init_waitqueue_head(&vop->wait_vop_switch_queue);
>>> +       vop->vop_switch_status = 0;
>>> +       init_waitqueue_head(&vop->wait_dmc_queue);
>>> +       vop->dmc_in_process = 0;
>>> +
>>> +       devfreq = devfreq_get_devfreq_by_phandle(dev, 0);
>>> +       if (IS_ERR(devfreq))
>>> +               goto out;
>>> +
>>> +       vop->devfreq = devfreq;
>>> +       vop->dmc_nb.notifier_call = dmc_notify;
>>> +       devfreq_register_notifier(vop->devfreq, &vop->dmc_nb,
>>> +                                 DEVFREQ_TRANSITION_NOTIFIER);
>>> +
>>> +       event_dev =
>>> devfreq_event_get_edev_by_phandle(vop->devfreq->dev.parent,
>>> +                                                     0);
>>> +       if (IS_ERR(event_dev))
>>> +               goto out;
>>> +
>>> +       vop->devfreq_event_dev = event_dev;
>>> +out:
>>>          return 0;
>>>   }
>>>
>>> --
>>> 2.6.6
>>>
>>
>>
>
> --
> Lin Huang
>
>

  reply	other threads:[~2016-09-06 19:03 UTC|newest]

Thread overview: 53+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-09-05  5:06 [PATCH v10 0/5] rk3399 support ddr frequency scaling Lin Huang
2016-09-05  5:06 ` Lin Huang
2016-09-05  5:06 ` [PATCH v10 1/5] Documentation: bindings: add dt documentation for dfi controller Lin Huang
2016-09-05  5:06   ` Lin Huang
2016-09-05  5:06   ` Lin Huang
2016-09-12 16:18   ` Rob Herring
2016-09-12 16:18     ` Rob Herring
2016-09-12 16:18     ` Rob Herring
2016-09-05  5:06 ` [PATCH v10 2/5] PM / devfreq: event: support rockchip " Lin Huang
2016-09-05  5:06   ` Lin Huang
2016-09-05  5:06 ` [PATCH v10 3/5] Documentation: bindings: add dt documentation for rk3399 dmc Lin Huang
2016-09-05  5:06   ` Lin Huang
2016-09-12 16:31   ` Rob Herring
2016-09-12 16:31     ` Rob Herring
2016-09-12 16:31     ` Rob Herring
2016-09-05  5:06 ` [PATCH v10 4/5] PM / devfreq: rockchip: add devfreq driver " Lin Huang
2016-09-05  5:06   ` Lin Huang
2016-09-05  5:06 ` [PATCH v10 5/5] drm/rockchip: Add dmc notifier in vop driver Lin Huang
2016-09-05  5:06   ` Lin Huang
2016-09-06  3:40   ` kbuild test robot
2016-09-06  3:40     ` kbuild test robot
2016-09-06  3:40     ` kbuild test robot
2016-09-06 17:18   ` Sean Paul
2016-09-06 17:18     ` Sean Paul
2016-09-06 18:15     ` hl
2016-09-06 18:15       ` hl
2016-09-06 18:55       ` Sean Paul [this message]
2016-09-06 18:55         ` Sean Paul
2016-09-06 18:55         ` Sean Paul
2016-09-06 19:01         ` hl
2016-09-06 19:01           ` hl
2016-09-06 19:01           ` hl
2016-09-06 19:07           ` Sean Paul
2016-09-06 19:07             ` Sean Paul
2016-09-06 19:07             ` Sean Paul
2016-09-07 18:07             ` dbasehore .
2016-09-07 18:07               ` dbasehore .
2016-09-07 18:07               ` dbasehore .
2016-09-07 18:27               ` Sean Paul
2016-09-07 18:27                 ` Sean Paul
2016-09-07 18:27                 ` Sean Paul
2016-09-07 18:13     ` dbasehore .
2016-09-07 18:13       ` dbasehore .
2016-09-07 18:13       ` dbasehore .
2016-09-07 20:31       ` Sean Paul
2016-09-07 20:31         ` Sean Paul
2016-09-07 20:31         ` Sean Paul
2016-09-07 22:21         ` dbasehore .
2016-09-07 22:21           ` dbasehore .
2016-09-07 22:21           ` dbasehore .
     [not found] ` <CGME20160905050631epcas1p4ebe75a5f119132f583a5020187382026@epcas1p4.samsung.com>
2016-09-06  4:37   ` MyungJoo Ham
2016-09-06  4:37     ` MyungJoo Ham
2016-09-06  4:37     ` MyungJoo Ham

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=CAOw6vbL6AxWRm7kovjwstW+qCd_9Wta3H_gvZswjeqqq9g-P8w@mail.gmail.com \
    --to=seanpaul@google.com \
    --cc=airlied@linux.ie \
    --cc=cw00.choi@samsung.com \
    --cc=dbasehore@chromium.org \
    --cc=devicetree@vger.kernel.org \
    --cc=dianders@chromium.org \
    --cc=dri-devel@lists.freedesktop.org \
    --cc=heiko@sntech.de \
    --cc=hl@rock-chips.com \
    --cc=kyungmin.park@samsung.com \
    --cc=linux-arm-kernel@lists.infradead.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-pm@vger.kernel.org \
    --cc=linux-rockchip@lists.infradead.org \
    --cc=mark.rutland@arm.com \
    --cc=mark.yao@rock-chips.com \
    --cc=mturquette@baylibre.com \
    --cc=myungjoo.ham@samsung.com \
    --cc=sboyd@codeaurora.org \
    --cc=sudeep.holla@arm.com \
    --cc=tixy@linaro.org \
    --cc=typ@rock-chips.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.