[PATCH v2 02/38] drm/msm/dp: remove dp_display's dp_mode and use dp_panel's instead

Dmitry Baryshkov dmitry.baryshkov at oss.qualcomm.com
Wed Aug 6 10:41:27 UTC 2025


On 06/08/2025 14:52, Yongxing Mou wrote:
> 
> 
> On 2025/6/27 20:44, Dmitry Baryshkov wrote:
>> On 27/06/2025 11:40, Yongxing Mou wrote:
>>>
>>>
>>> On 2025/6/25 22:03, Dmitry Baryshkov wrote:
>>>> On Wed, Jun 25, 2025 at 08:34:18PM +0800, Yongxing Mou wrote:
>>>>>
>>>>>
>>>>> On 2025/6/9 20:48, Dmitry Baryshkov wrote:
>>>>>> On Mon, Jun 09, 2025 at 08:21:21PM +0800, Yongxing Mou wrote:
>>>>>>> From: Abhinav Kumar <quic_abhinavk at quicinc.com>
>>>>>>>
>>>>>>> dp_display caches the current display mode and then passes it onto
>>>>>>> the panel to be used for programming the panel params. Remove this
>>>>>>> two level passing and directly populated the panel's dp_display_mode
>>>>>>> instead.
>>>>>>
>>>>>> - Why do we need to cache / copy it anyway? Can't we just pass the
>>>>>>     corresponding drm_atomic_state / drm_crtc_state / 
>>>>>> drm_display_mode ?
>>>>>>
>>>>> This part works as follows: .mode_set() copies the adjusted_mode into
>>>>> msm_dp_display_private->msm_dp_display_mode, and also parses and 
>>>>> stores
>>>>> variables such as v_active_low/h_active_low/out_fmt_is_yuv_420 
>>>>> and ... When
>>>>> @drm_bridge_funcs.atomic_enable() is called, it copies
>>>>> msm_dp_display->msm_dp_mode into dp_panel->msm_dp_mode and initializes
>>>>> panel_info in msm_dp_display_set_mode(). Then when go to
>>>>> msm_dp_ctrl_on_stream(), the parameters are updated into the 
>>>>> corresponding
>>>>> hardware registers.
>>>>
>>>> So, if we do everything during .atomic_enable(), there would be no need
>>>> to store and/or copy anything. All the data is available and can be 
>>>> used
>>>> as is.
>>>>
>>> Got it. Let me confirm—can we keep msm_dp_mode or drm_display_mode in 
>>> msm_dp_panel? Mabey debug node will use this ..
>>
>> Please don't. I really dislike storing drm_atomic_state-related 
>> variables in a non-state structure. I think it makes it easier to 
>> mistakenly update or to use a stale value.
>>
>> Debug code already prints modes in debugfs/dri/N/state. If we need any 
>> other state-related prints, they should go to the same file.
>>
> Hi, I got this point.. i go through the driver. since lots of funcs used 
> msm_dp_mode cached. so maybe it is not a very small change.. I’d like to 
> prioritize MST first, and then submit this patch once I got time..

Ack

> 
>>>>>
>>>>> This design has been in place since the first version of the DP 
>>>>> driver and
>>>>> has remained largely unchanged.
>>>>
>>>> Yes... The point is that you are touching this piece of code anyway,
>>>> let's make it nicer.
>>>>
>>> Agree with this point.
>>>>> Originally, the drm_mode would be passed in
>>>>> two stages: from msm_dp_display->msm_dp_mode to dp_panel- 
>>>>> >msm_dp_mode. Since
>>>>> in MST mode each stream requires its own drm_mode and stored in 
>>>>> dp_panel, we
>>>>> simplified the two-stage transfer into a single step (.mode_set() 
>>>>> do all
>>>>> things and store in msm_dp_panel). Meanwhile we modified the
>>>>> msm_dp_display_set_mode function to accept a msm_dp_panel parameter,
>>>>> allowing the MST bridge funcs' mode_set() to reuse this part code.
>>>>>
>>>>> The following patches:
>>>>> https://patchwork.freedesktop.org/patch/657573/?series=142207&rev=2 
>>>>> and
>>>>> https://patchwork.freedesktop.org/patch/657593/?series=142207&rev=2,
>>>>> introduce msm_dp_display_*_helper functions to help reuse common 
>>>>> code across
>>>>> MST/SST/eDP drm_bridge_funcs.
>>>>>
>>>>> If we drop msm_dp_mode from dp_panel and use drm_display_mode, it 
>>>>> might
>>>>> introduce a large number of changes that are not directly related 
>>>>> to MST.
>>>>> Actually i think the presence of msm_dp_display_mode seems to 
>>>>> simplify the
>>>>> work in msm_dp_panel_timing_cfg(), this patch series we want to 
>>>>> focus on MST
>>>>> parts, so would we consider optimizing them later?
>>>>
>>>> Sure... But then you have to change two places. If you optimize it
>>>> first, you have to touch only place. And it can be even submitted
>>>> separately.
>>>>
>>> Understood, that’s indeed the case. I just want to prioritize the MST 
>>> patch and have it merged first, since it involves changes to lots of 
>>> files. Thanks~~
>>>>>
>>>>> Thanks~
>>>>>>>
>>>>>>> Signed-off-by: Abhinav Kumar <quic_abhinavk at quicinc.com>
>>>>>>> Signed-off-by: Yongxing Mou <quic_yongmou at quicinc.com>
>>>>>>> ---
>>>>>>>    drivers/gpu/drm/msm/dp/dp_display.c | 76 +++++++++++++ 
>>>>>>> +-----------------------
>>>>>>>    1 file changed, 29 insertions(+), 47 deletions(-)
>>>>>>>
>>>>>>> diff --git a/drivers/gpu/drm/msm/dp/dp_display.c b/drivers/gpu/ 
>>>>>>> drm/ msm/dp/dp_display.c
>>>>>>> index 
>>>>>>> 4a9b65647cdef1ed6c3bb851f93df0db8be977af..9d2db9cbd2552470a36a63f70f517c35436f7280 100644
>>>>>>> --- a/drivers/gpu/drm/msm/dp/dp_display.c
>>>>>>> +++ b/drivers/gpu/drm/msm/dp/dp_display.c
>>>>>>> @@ -92,7 +92,6 @@ struct msm_dp_display_private {
>>>>>>>        struct msm_dp_panel   *panel;
>>>>>>>        struct msm_dp_ctrl    *ctrl;
>>>>>>> -    struct msm_dp_display_mode msm_dp_mode;
>>>>>>>        struct msm_dp msm_dp_display;
>>>>>>>        /* wait for audio signaling */
>>>>>>> @@ -806,16 +805,29 @@ static int msm_dp_init_sub_modules(struct 
>>>>>>> msm_dp_display_private *dp)
>>>>>>>    }
>>>>>>>    static int msm_dp_display_set_mode(struct msm_dp *msm_dp_display,
>>>>>>> -                   struct msm_dp_display_mode *mode)
>>>>>>> +                   const struct drm_display_mode *adjusted_mode,
>>>>>>> +                   struct msm_dp_panel *msm_dp_panel)
>>>>>>>    {
>>>>>>> -    struct msm_dp_display_private *dp;
>>>>>>> +    u32 bpp;
>>>>>>> -    dp = container_of(msm_dp_display, struct 
>>>>>>> msm_dp_display_private, msm_dp_display);
>>>>>>> +    drm_mode_copy(&msm_dp_panel->msm_dp_mode.drm_mode, 
>>>>>>> adjusted_mode);
>>>>>>> +
>>>>>>> +    if (msm_dp_display_check_video_test(msm_dp_display))
>>>>>>> +        bpp = msm_dp_display_get_test_bpp(msm_dp_display);
>>>>>>> +    else
>>>>>>> +        bpp = msm_dp_panel->connector->display_info.bpc * 3;
>>>>>>> +
>>>>>>> +    msm_dp_panel->msm_dp_mode.bpp = bpp;
>>>>>>> +
>>>>>>> +    msm_dp_panel->msm_dp_mode.v_active_low =
>>>>>>> +        !!(adjusted_mode->flags & DRM_MODE_FLAG_NVSYNC);
>>>>>>> +    msm_dp_panel->msm_dp_mode.h_active_low =
>>>>>>> +        !!(adjusted_mode->flags & DRM_MODE_FLAG_NHSYNC);
>>>>>>> +    msm_dp_panel->msm_dp_mode.out_fmt_is_yuv_420 =
>>>>>>> +        drm_mode_is_420_only(&msm_dp_panel->connector- 
>>>>>>> >display_info, adjusted_mode) &&
>>>>>>> +        msm_dp_panel->vsc_sdp_supported;
>>>>>>> -    drm_mode_copy(&dp->panel->msm_dp_mode.drm_mode, &mode- 
>>>>>>> >drm_mode);
>>>>>>> -    dp->panel->msm_dp_mode.bpp = mode->bpp;
>>>>>>> -    dp->panel->msm_dp_mode.out_fmt_is_yuv_420 = mode- 
>>>>>>> >out_fmt_is_yuv_420;
>>>>>>> -    msm_dp_panel_init_panel_info(dp->panel);
>>>>>>> +    msm_dp_panel_init_panel_info(msm_dp_panel);
>>>>>>>        return 0;
>>>>>>>    }
>>>>>>> @@ -1431,10 +1443,13 @@ bool msm_dp_needs_periph_flush(const 
>>>>>>> struct msm_dp *msm_dp_display,
>>>>>>>    bool msm_dp_wide_bus_available(const struct msm_dp 
>>>>>>> *msm_dp_display)
>>>>>>>    {
>>>>>>>        struct msm_dp_display_private *dp;
>>>>>>> +    struct msm_dp_panel *dp_panel;
>>>>>>>        dp = container_of(msm_dp_display, struct 
>>>>>>> msm_dp_display_private, msm_dp_display);
>>>>>>> -    if (dp->msm_dp_mode.out_fmt_is_yuv_420)
>>>>>>> +    dp_panel = dp->panel;
>>>>>>> +
>>>>>>> +    if (dp_panel->msm_dp_mode.out_fmt_is_yuv_420)
>>>>>>>            return false;
>>>>>>>        return dp->wide_bus_supported;
>>>>>>> @@ -1496,10 +1511,6 @@ void msm_dp_bridge_atomic_enable(struct 
>>>>>>> drm_bridge *drm_bridge,
>>>>>>>        bool force_link_train = false;
>>>>>>>        msm_dp_display = container_of(dp, struct 
>>>>>>> msm_dp_display_private, msm_dp_display);
>>>>>>> -    if (!msm_dp_display->msm_dp_mode.drm_mode.clock) {
>>>>>>> -        DRM_ERROR("invalid params\n");
>>>>>>> -        return;
>>>>>>> -    }
>>>>>>>        if (dp->is_edp)
>>>>>>>            msm_dp_hpd_plug_handle(msm_dp_display, 0);
>>>>>>> @@ -1517,15 +1528,6 @@ void msm_dp_bridge_atomic_enable(struct 
>>>>>>> drm_bridge *drm_bridge,
>>>>>>>            return;
>>>>>>>        }
>>>>>>> -    rc = msm_dp_display_set_mode(dp, &msm_dp_display->msm_dp_mode);
>>>>>>> -    if (rc) {
>>>>>>> -        DRM_ERROR("Failed to perform a mode set, rc=%d\n", rc);
>>>>>>> -        mutex_unlock(&msm_dp_display->event_mutex);
>>>>>>> -        return;
>>>>>>> -    }
>>>>>>
>>>>>> It should be done other way around: keep this call and drop
>>>>>> msm_dp_bridge_mode_set().
>>>>>>
>>>>> Emm as reply in last comments..
>>>>
>>>> Yep. Drop .mode_set, the callback is even described as deprecated.
>>>>
>>> Thanks, the documentation does state that.
>>>>>>> -
>>>>>>> -    hpd_state =  msm_dp_display->hpd_state;
>>>>>>> -
>>>>>>>        if (hpd_state == ST_CONNECTED && !dp->power_on) {
>>>>>>>            msm_dp_display_host_phy_init(msm_dp_display);
>>>>>>>            force_link_train = true;
>>>>>>> @@ -1604,33 +1606,13 @@ void msm_dp_bridge_mode_set(struct 
>>>>>>> drm_bridge *drm_bridge,
>>>>>>>        msm_dp_display = container_of(dp, struct 
>>>>>>> msm_dp_display_private, msm_dp_display);
>>>>>>>        msm_dp_panel = msm_dp_display->panel;
>>>>>>> -    memset(&msm_dp_display->msm_dp_mode, 0x0, sizeof(struct 
>>>>>>> msm_dp_display_mode));
>>>>>>> -
>>>>>>> -    if (msm_dp_display_check_video_test(dp))
>>>>>>> -        msm_dp_display->msm_dp_mode.bpp = 
>>>>>>> msm_dp_display_get_test_bpp(dp);
>>>>>>> -    else /* Default num_components per pixel = 3 */
>>>>>>> -        msm_dp_display->msm_dp_mode.bpp = dp->connector- 
>>>>>>> >display_info.bpc * 3;
>>>>>>> -
>>>>>>> -    if (!msm_dp_display->msm_dp_mode.bpp)
>>>>>>> -        msm_dp_display->msm_dp_mode.bpp = 24; /* Default bpp */
>>>>>>> -
>>>>>>> -    drm_mode_copy(&msm_dp_display->msm_dp_mode.drm_mode, 
>>>>>>> adjusted_mode);
>>>>>>> -
>>>>>>> -    msm_dp_display->msm_dp_mode.v_active_low =
>>>>>>> -        !!(msm_dp_display->msm_dp_mode.drm_mode.flags & 
>>>>>>> DRM_MODE_FLAG_NVSYNC);
>>>>>>> -
>>>>>>> -    msm_dp_display->msm_dp_mode.h_active_low =
>>>>>>> -        !!(msm_dp_display->msm_dp_mode.drm_mode.flags & 
>>>>>>> DRM_MODE_FLAG_NHSYNC);
>>>>>>> -
>>>>>>> -    msm_dp_display->msm_dp_mode.out_fmt_is_yuv_420 =
>>>>>>> -        drm_mode_is_420_only(&dp->connector->display_info, 
>>>>>>> adjusted_mode) &&
>>>>>>> -        msm_dp_panel->vsc_sdp_supported;
>>>>>>> +    msm_dp_display_set_mode(dp, adjusted_mode, msm_dp_panel);
>>>>>>>        /* populate wide_bus_support to different layers */
>>>>>>> -    msm_dp_display->ctrl->wide_bus_en =
>>>>>>> -        msm_dp_display->msm_dp_mode.out_fmt_is_yuv_420 ? false : 
>>>>>>> msm_dp_display->wide_bus_supported;
>>>>>>> -    msm_dp_display->catalog->wide_bus_en =
>>>>>>> -        msm_dp_display->msm_dp_mode.out_fmt_is_yuv_420 ? false : 
>>>>>>> msm_dp_display->wide_bus_supported;
>>>>>>> +    msm_dp_display->ctrl->wide_bus_en = msm_dp_panel- 
>>>>>>> >msm_dp_mode.out_fmt_is_yuv_420 ?
>>>>>>> +        false : msm_dp_display->wide_bus_supported;
>>>>>>> +    msm_dp_display->catalog->wide_bus_en = msm_dp_panel- 
>>>>>>> >msm_dp_mode.out_fmt_is_yuv_420 ?
>>>>>>> +        false : msm_dp_display->wide_bus_supported;
>>>>>>>    }
>>>>>>>    void msm_dp_bridge_hpd_enable(struct drm_bridge *bridge)
>>>>>>>
>>>>>>> -- 
>>>>>>> 2.34.1
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>
>>
> 


-- 
With best wishes
Dmitry


More information about the dri-devel mailing list