[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <b412dd01-126b-155d-5961-b544da821ce6@quicinc.com>
Date: Sun, 25 Feb 2024 17:33:38 -0800
From: Abhinav Kumar <quic_abhinavk@...cinc.com>
To: Dmitry Baryshkov <dmitry.baryshkov@...aro.org>
CC: Rob Clark <robdclark@...il.com>, Sean Paul <sean@...rly.run>,
Marijn
Suijten <marijn.suijten@...ainline.org>,
David Airlie <airlied@...il.com>, Daniel Vetter <daniel@...ll.ch>,
Steev Klimaszewski <steev@...i.org>, <linux-arm-msm@...r.kernel.org>,
<dri-devel@...ts.freedesktop.org>, <freedreno@...ts.freedesktop.org>,
<linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v3 2/3] drm/msm/dpu: split dpu_encoder_wait_for_event into
two functions
On 2/25/2024 12:52 PM, Dmitry Baryshkov wrote:
> On Sun, 25 Feb 2024 at 21:49, Abhinav Kumar <quic_abhinavk@...cinc.com> wrote:
>>
>>
>>
>> On 2/25/2024 6:12 AM, Dmitry Baryshkov wrote:
>>> Stop multiplexing several events via the dpu_encoder_wait_for_event()
>>> function. Split it into two distinct functions two allow separate
>>> handling of those events.
>>>
>>
>> I understand the idea but would handling of those events be really distinct?
>
> We are interested in capturing the state after the first
> wait_for_commit_done() timeout. The wait_for_tx_complete doesn't need
> such handling. Even if we were to handle it in some way, it would be a
> different conditional.
>
wait_for_tx_complete timeout also needs similar handling.
the timeout mechanisms need to be unified at some point, that time I
will re-visit the need to have a common wait_timeout handler.
Lets see how this code evolves.
So with that nit about the kernel doc addressed,
Reviewed-by: Abhinav Kumar <quic_abhinavk@...cinc.com>
> Last but not least, I don't like multiplexing just for the sake of it.
> There is nearly no common behaviour.
>
the multiplexing allows us to have one common timeout path which I think
will eventually happen. So i dont think its true that there is no common
behavior.
>>
>> Like if wait_for_commit_done timedout or wait_for_tx_complete timedout,
>> the handling will be similar from my PoV.
>>
>>> Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@...aro.org>
>>> ---
>>> drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.c | 74 +++++++++++++++++++++--------
>>> drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.h | 22 ++-------
>>> drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c | 2 +-
>>> drivers/gpu/drm/msm/msm_drv.h | 10 ----
>>> 4 files changed, 59 insertions(+), 49 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.c b/drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.c
>>> index 194dbb08331d..30f349c8a1e5 100644
>>> --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.c
>>> +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.c
>>> @@ -1282,7 +1282,7 @@ static void dpu_encoder_virt_atomic_disable(struct drm_encoder *drm_enc,
>>> trace_dpu_enc_disable(DRMID(drm_enc));
>>>
>>> /* wait for idle */
>>> - dpu_encoder_wait_for_event(drm_enc, MSM_ENC_TX_COMPLETE);
>>> + dpu_encoder_wait_for_tx_complete(drm_enc);
>>>
>>> dpu_encoder_resource_control(drm_enc, DPU_ENC_RC_EVENT_PRE_STOP);
>>>
>>> @@ -2402,10 +2402,23 @@ struct drm_encoder *dpu_encoder_init(struct drm_device *dev,
>>> return &dpu_enc->base;
>>> }
>>>
>>> -int dpu_encoder_wait_for_event(struct drm_encoder *drm_enc,
>>> - enum msm_event_wait event)
>>> +/**
>>> + * dpu_encoder_wait_for_commit_done() - Wait for encoder to flush pending state
>>> + * @drm_enc: encoder pointer
>>> + *
>>> + * Wait for hardware to have flushed the current pending frames to hardware at
>>> + * a vblank or ctl_start Encoders will map this differently depending on the
>>> + * panel type.
>>> + *
>>
>> Missing a '.' between ctl_start and Encoder?
>
> Ack. Also I should drop the leftovers afterwards.
>
>>
>>> + * MSM_ENC_TX_COMPLETE - Wait for the hardware to transfer all the pixels to
>>> + * the panel. Encoders will map this differently
>>> + * depending on the panel type.
>>> + * vid mode -> vsync_irq
>>> + * cmd mode -> pp_done
>>> + * Return: 0 on success, -EWOULDBLOCK if already signaled, error otherwise
>>> + */
>>> +int dpu_encoder_wait_for_commit_done(struct drm_encoder *drm_enc)
>>> {
>>> - int (*fn_wait)(struct dpu_encoder_phys *phys_enc) = NULL;
>>> struct dpu_encoder_virt *dpu_enc = NULL;
>>> int i, ret = 0;
>>>
>>> @@ -2419,23 +2432,46 @@ int dpu_encoder_wait_for_event(struct drm_encoder *drm_enc,
>>> for (i = 0; i < dpu_enc->num_phys_encs; i++) {
>>> struct dpu_encoder_phys *phys = dpu_enc->phys_encs[i];
>>>
>>> - switch (event) {
>>> - case MSM_ENC_COMMIT_DONE:
>>> - fn_wait = phys->ops.wait_for_commit_done;
>>> - break;
>>> - case MSM_ENC_TX_COMPLETE:
>>> - fn_wait = phys->ops.wait_for_tx_complete;
>>> - break;
>>> - default:
>>> - DPU_ERROR_ENC(dpu_enc, "unknown wait event %d\n",
>>> - event);
>>> - return -EINVAL;
>>> + if (phys->ops.wait_for_commit_done) {
>>> + DPU_ATRACE_BEGIN("wait_for_commit_done");
>>> + ret = phys->ops.wait_for_commit_done(phys);
>>> + DPU_ATRACE_END("wait_for_commit_done");
>>> + if (ret)
>>> + return ret;
>>> }
>>> + }
>>> +
>>> + return ret;
>>> +}
>>> +
>>> +/**
>>> + * dpu_encoder_wait_for_tx_complete() - Wait for encoder to transfer pixels to panel
>>> + * @drm_enc: encoder pointer
>>> + *
>>> + * Wait for the hardware to transfer all the pixels to the panel. Encoders will
>>> + * map this differently depending on the panel type.
>>> + *
>>> + * Return: 0 on success, -EWOULDBLOCK if already signaled, error otherwise
>>> + */
>>> +int dpu_encoder_wait_for_tx_complete(struct drm_encoder *drm_enc)
>>> +{
>>> + struct dpu_encoder_virt *dpu_enc = NULL;
>>> + int i, ret = 0;
>>> +
>>> + if (!drm_enc) {
>>> + DPU_ERROR("invalid encoder\n");
>>> + return -EINVAL;
>>> + }
>>> + dpu_enc = to_dpu_encoder_virt(drm_enc);
>>> + DPU_DEBUG_ENC(dpu_enc, "\n");
>>> +
>>> + for (i = 0; i < dpu_enc->num_phys_encs; i++) {
>>> + struct dpu_encoder_phys *phys = dpu_enc->phys_encs[i];
>>>
>>> - if (fn_wait) {
>>> - DPU_ATRACE_BEGIN("wait_for_completion_event");
>>> - ret = fn_wait(phys);
>>> - DPU_ATRACE_END("wait_for_completion_event");
>>> + if (phys->ops.wait_for_tx_complete) {
>>> + DPU_ATRACE_BEGIN("wait_for_tx_complete");
>>> + ret = phys->ops.wait_for_tx_complete(phys);
>>> + DPU_ATRACE_END("wait_for_tx_complete");
>>> if (ret)
>>> return ret;
>>> }
>>> diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.h b/drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.h
>>> index fe6b1d312a74..0c928d1876e4 100644
>>> --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.h
>>> +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.h
>>> @@ -93,25 +93,9 @@ void dpu_encoder_kickoff(struct drm_encoder *encoder);
>>> */
>>> int dpu_encoder_vsync_time(struct drm_encoder *drm_enc, ktime_t *wakeup_time);
>>>
>>> -/**
>>> - * dpu_encoder_wait_for_event - Waits for encoder events
>>> - * @encoder: encoder pointer
>>> - * @event: event to wait for
>>> - * MSM_ENC_COMMIT_DONE - Wait for hardware to have flushed the current pending
>>> - * frames to hardware at a vblank or ctl_start
>>> - * Encoders will map this differently depending on the
>>> - * panel type.
>>> - * vid mode -> vsync_irq
>>> - * cmd mode -> ctl_start
>>> - * MSM_ENC_TX_COMPLETE - Wait for the hardware to transfer all the pixels to
>>> - * the panel. Encoders will map this differently
>>> - * depending on the panel type.
>>> - * vid mode -> vsync_irq
>>> - * cmd mode -> pp_done
>>> - * Returns: 0 on success, -EWOULDBLOCK if already signaled, error otherwise
>>> - */
>>> -int dpu_encoder_wait_for_event(struct drm_encoder *drm_encoder,
>>> - enum msm_event_wait event);
>>> +int dpu_encoder_wait_for_commit_done(struct drm_encoder *drm_encoder);
>>> +
>>> +int dpu_encoder_wait_for_tx_complete(struct drm_encoder *drm_encoder);
>>>
>>> /*
>>> * dpu_encoder_get_intf_mode - get interface mode of the given encoder
>>> diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c b/drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c
>>> index d6412395bacc..26b5e54031d9 100644
>>> --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c
>>> +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c
>>> @@ -476,7 +476,7 @@ static void dpu_kms_wait_for_commit_done(struct msm_kms *kms,
>>> * mode panels. This may be a no-op for command mode panels.
>>> */
>>> trace_dpu_kms_wait_for_commit_done(DRMID(crtc));
>>> - ret = dpu_encoder_wait_for_event(encoder, MSM_ENC_COMMIT_DONE);
>>> + ret = dpu_encoder_wait_for_commit_done(encoder);
>>> if (ret && ret != -EWOULDBLOCK) {
>>> DPU_ERROR("wait for commit done returned %d\n", ret);
>>> break;
>>> diff --git a/drivers/gpu/drm/msm/msm_drv.h b/drivers/gpu/drm/msm/msm_drv.h
>>> index 762e13e2df74..91cf57f72321 100644
>>> --- a/drivers/gpu/drm/msm/msm_drv.h
>>> +++ b/drivers/gpu/drm/msm/msm_drv.h
>>> @@ -74,16 +74,6 @@ enum msm_dsi_controller {
>>> #define MSM_GPU_MAX_RINGS 4
>>> #define MAX_H_TILES_PER_DISPLAY 2
>>>
>>> -/**
>>> - * enum msm_event_wait - type of HW events to wait for
>>> - * @MSM_ENC_COMMIT_DONE - wait for the driver to flush the registers to HW
>>> - * @MSM_ENC_TX_COMPLETE - wait for the HW to transfer the frame to panel
>>> - */
>>> -enum msm_event_wait {
>>> - MSM_ENC_COMMIT_DONE = 0,
>>> - MSM_ENC_TX_COMPLETE,
>>> -};
>>> -
>>> /**
>>> * struct msm_display_topology - defines a display topology pipeline
>>> * @num_lm: number of layer mixers used
>>>
>
>
>
Powered by blists - more mailing lists