[PATCH v10 5/5] drm/rockchip: Add dmc notifier in vop driver
Sean Paul
seanpaul at google.com
Tue Sep 6 19:07:56 UTC 2016
On Tue, Sep 6, 2016 at 3:01 PM, hl <hl at rock-chips.com> wrote:
> Hi
>
>
> On 2016年09月07日 02:55, Sean Paul wrote:
>>
>> On Tue, Sep 6, 2016 at 2:15 PM, hl <hl at 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 at 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 at rock-chips.com>
>>>>> Reviewed-by: Chanwoo Choi <cw00.choi at 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()?
>
> In dmc_notify(), i will check the vop_switch_status, so in the vop_enable()
> and vop_crtc_disable(),
> the dmc_in_progress can not change value.
>
THREAD 1 - wait_event_timeout (vop_switch_status)
THREAD 2 - wait_event_timeout(wait_dmc_queue)
THREAD 2 - vop_switch_status = 1
THREAD 1 - dmc_in_progress = 1
Sean
>> 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
>>>
>>>
>>
>>
>
> --
> Lin Huang
>
>
More information about the dri-devel
mailing list