[PATCH v4 8/9] drm/msm/dpu: rework core_perf debugfs overrides
Abhinav Kumar
quic_abhinavk at quicinc.com
Thu Jan 16 00:47:34 UTC 2025
On 1/15/2025 4:35 PM, Dmitry Baryshkov wrote:
> On Wed, Jan 15, 2025 at 11:51:20AM -0800, Abhinav Kumar wrote:
>>
>>
>> On 1/15/2025 12:41 AM, Dmitry Baryshkov wrote:
>>> On Tue, Jan 14, 2025 at 02:02:54PM -0800, Abhinav Kumar wrote:
>>>>
>>>>
>>>> On 1/5/2025 7:07 PM, Dmitry Baryshkov wrote:
>>>>> Currently debugfs provides separate 'modes' to override calculated
>>>>> MDP_CLK rate and interconnect bandwidth votes. Change that to allow
>>>>> overriding individual values (e.g. one can override just clock or just
>>>>> average bandwidth vote). The maximum values allowed for those entries by
>>>>> the platform can be read from the 'max_core_ab' and 'max_core_clk_rate'
>>>>> files in debugfs.
>>>>>
>>>>
>>>> Apart from the concern I highlighted in the previous patch, the only issue I
>>>> have with this is that, this went from a one step process of using the
>>>> "mode" this has become a two step one.
>>>>
>>>> There were essentially two modes we are talking about - "fixed" and
>>>> "minimum"
>>>>
>>>> With respect to "fixed" this is totally fine because this is preserving that
>>>> functionality because to be able to set the fixed mode the end user must
>>>> know what values they want to try anyway.
>>>>
>>>> With respect to "minimum" mode, is where this approach is not that great.
>>>> The end users of this can be non-display developers too such as our QA teams
>>>> who might want to perform a first level of triage on the issues and route it
>>>> accordingly. This is especially true for underruns and some performance lags
>>>> as well.
>>>>
>>>> If you really dont like the term "modes", to preserve the "minimum" mode,
>>>> how about just using a bool debugfs like "max_perf_params" which internally
>>>> maxes out the max MDP clock and ab/ib params.
>>>
>>> That's what I'm trying to avoid - having an extra debugfs file which
>>> overrides other files. It is much easier to work if there is no need to
>>> switch modes, it is easy to overlook it. I think it should be fine to
>>> use `cat max_foo > fix_foo` to override each of the params. After
>>> renaming the threshold_high to max_core_ab the names of the debugfs
>>> files match.
>>>
>>
>> Its just a difference in interpretation IMO.
>>
>> the "fixed" mode is trying to given an option to incrementally try and see
>> which value really works and also to see whether its the clock OR the
>> bandwidth which is making the difference. So individual control of those.
>>
>> The "max" mode is trying to see if even the max values of everything cannot
>> fix the problem. BTW, the max was maxing out BOTH the DPU clocks and BW.
>>
>> So this is not just 2 extra reads for the user but 3. (ab/ib/dpu_clk) if we
>> drop "max" and use "fixed" for max as well and even for that the user has to
>> refer the max DPU clock value.
>
> Yes, I understand that. However I still think that it's easier than
> having a set of 'fix_foo' values which are silently ignored because of
> the preselected mode.
>
> I can probably see an option: use your max_perf_params idea, but in a
> form of a write-only file which immediately selects max values for clock
> rate and both bandwidths. WDYT?
>
Sorry I am missing something here. This is the same thing I had in mind
to have it as a bool when someone does echo 1 > max_perf_params, it will
immediately max the values for clock rate and bandwidth.
So to summarize, there are four nodes:
1) fix_core_ab_vote
2) fix_core_ib_vote
3) fix_core_clk_rate
These individually control their respective params
4) max_perf_params - which maxes out all of the above
Is this what you are referring to as well?
>>>>> Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov at linaro.org>
>>>>> ---
>>>>> drivers/gpu/drm/msm/disp/dpu1/dpu_core_perf.c | 87 +++------------------------
>>>>> drivers/gpu/drm/msm/disp/dpu1/dpu_core_perf.h | 10 ---
>>>>> 2 files changed, 9 insertions(+), 88 deletions(-)
>>>>>
>>>>> diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_core_perf.c b/drivers/gpu/drm/msm/disp/dpu1/dpu_core_perf.c
>>>>> index 7ff3405c6867556a8dc776783b91f1da6c86ef3f..913eb4c01abe10c1ed84215fbbee50abd69e9317 100644
>>>>> --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_core_perf.c
>>>>> +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_core_perf.c
>>>>> @@ -17,20 +17,6 @@
>>>>> #include "dpu_crtc.h"
>>>>> #include "dpu_core_perf.h"
>>>>> -/**
>>>>> - * enum dpu_perf_mode - performance tuning mode
>>>>> - * @DPU_PERF_MODE_NORMAL: performance controlled by user mode client
>>>>> - * @DPU_PERF_MODE_MINIMUM: performance bounded by minimum setting
>>>>> - * @DPU_PERF_MODE_FIXED: performance bounded by fixed setting
>>>>> - * @DPU_PERF_MODE_MAX: maximum value, used for error checking
>>>>> - */
>>>>> -enum dpu_perf_mode {
>>>>> - DPU_PERF_MODE_NORMAL,
>>>>> - DPU_PERF_MODE_MINIMUM,
>>>>> - DPU_PERF_MODE_FIXED,
>>>>> - DPU_PERF_MODE_MAX
>>>>> -};
>>>>> -
>>>>> /**
>>>>> * _dpu_core_perf_calc_bw() - to calculate BW per crtc
>>>>> * @perf_cfg: performance configuration
>>>>> @@ -215,18 +201,16 @@ static int _dpu_core_perf_crtc_update_bus(struct dpu_kms *kms,
>>>>> if (!kms->num_paths)
>>>>> return 0;
>>>>> - if (kms->perf.perf_tune.mode == DPU_PERF_MODE_MINIMUM) {
>>>>> - avg_bw = 0;
>>>>> - peak_bw = 0;
>>>>> - } else if (kms->perf.perf_tune.mode == DPU_PERF_MODE_FIXED) {
>>>>> + dpu_core_perf_aggregate(crtc->dev, dpu_crtc_get_client_type(crtc), &perf);
>>>>> +
>>>>> + avg_bw = div_u64(perf.bw_ctl, 1000); /*Bps_to_icc*/
>>>>> + peak_bw = perf.max_per_pipe_ib;
>>>>> +
>>>>> + if (kms->perf.fix_core_ab_vote)
>>>>> avg_bw = kms->perf.fix_core_ab_vote;
>>>>> - peak_bw = kms->perf.fix_core_ib_vote;
>>>>> - } else {
>>>>> - dpu_core_perf_aggregate(crtc->dev, dpu_crtc_get_client_type(crtc), &perf);
>>>>> - avg_bw = div_u64(perf.bw_ctl, 1000); /*Bps_to_icc*/
>>>>> - peak_bw = perf.max_per_pipe_ib;
>>>>> - }
>>>>> + if (kms->perf.fix_core_ib_vote)
>>>>> + peak_bw = kms->perf.fix_core_ib_vote;
>>>>> avg_bw /= kms->num_paths;
>>>>> @@ -275,12 +259,9 @@ static u64 _dpu_core_perf_get_core_clk_rate(struct dpu_kms *kms)
>>>>> struct drm_crtc *crtc;
>>>>> struct dpu_crtc_state *dpu_cstate;
>>>>> - if (kms->perf.perf_tune.mode == DPU_PERF_MODE_FIXED)
>>>>> + if (kms->perf.fix_core_clk_rate)
>>>>> return kms->perf.fix_core_clk_rate;
>>>>> - if (kms->perf.perf_tune.mode == DPU_PERF_MODE_MINIMUM)
>>>>> - return kms->perf.max_core_clk_rate;
>>>>> -
>>>>> clk_rate = 0;
>>>>> drm_for_each_crtc(crtc, kms->dev) {
>>>>> if (crtc->enabled) {
>>>>> @@ -396,54 +377,6 @@ int dpu_core_perf_crtc_update(struct drm_crtc *crtc,
>>>>> #ifdef CONFIG_DEBUG_FS
>>>>> -static ssize_t _dpu_core_perf_mode_write(struct file *file,
>>>>> - const char __user *user_buf, size_t count, loff_t *ppos)
>>>>> -{
>>>>> - struct dpu_core_perf *perf = file->private_data;
>>>>> - u32 perf_mode = 0;
>>>>> - int ret;
>>>>> -
>>>>> - ret = kstrtouint_from_user(user_buf, count, 0, &perf_mode);
>>>>> - if (ret)
>>>>> - return ret;
>>>>> -
>>>>> - if (perf_mode >= DPU_PERF_MODE_MAX)
>>>>> - return -EINVAL;
>>>>> -
>>>>> - if (perf_mode == DPU_PERF_MODE_FIXED) {
>>>>> - DRM_INFO("fix performance mode\n");
>>>>> - } else if (perf_mode == DPU_PERF_MODE_MINIMUM) {
>>>>> - /* run the driver with max clk and BW vote */
>>>>> - DRM_INFO("minimum performance mode\n");
>>>>> - } else if (perf_mode == DPU_PERF_MODE_NORMAL) {
>>>>> - /* reset the perf tune params to 0 */
>>>>> - DRM_INFO("normal performance mode\n");
>>>>> - }
>>>>> - perf->perf_tune.mode = perf_mode;
>>>>> -
>>>>> - return count;
>>>>> -}
>>>>> -
>>>>> -static ssize_t _dpu_core_perf_mode_read(struct file *file,
>>>>> - char __user *buff, size_t count, loff_t *ppos)
>>>>> -{
>>>>> - struct dpu_core_perf *perf = file->private_data;
>>>>> - int len;
>>>>> - char buf[128];
>>>>> -
>>>>> - len = scnprintf(buf, sizeof(buf),
>>>>> - "mode %d\n",
>>>>> - perf->perf_tune.mode);
>>>>> -
>>>>> - return simple_read_from_buffer(buff, count, ppos, buf, len);
>>>>> -}
>>>>> -
>>>>> -static const struct file_operations dpu_core_perf_mode_fops = {
>>>>> - .open = simple_open,
>>>>> - .read = _dpu_core_perf_mode_read,
>>>>> - .write = _dpu_core_perf_mode_write,
>>>>> -};
>>>>> -
>>>>> /**
>>>>> * dpu_core_perf_debugfs_init - initialize debugfs for core performance context
>>>>> * @dpu_kms: Pointer to the dpu_kms struct
>>>>> @@ -472,8 +405,6 @@ int dpu_core_perf_debugfs_init(struct dpu_kms *dpu_kms, struct dentry *parent)
>>>>> (u32 *)&perf->perf_cfg->min_llcc_ib);
>>>>> debugfs_create_u32("min_dram_ib", 0400, entry,
>>>>> (u32 *)&perf->perf_cfg->min_dram_ib);
>>>>> - debugfs_create_file("perf_mode", 0600, entry,
>>>>> - (u32 *)perf, &dpu_core_perf_mode_fops);
>>>>> debugfs_create_u64("fix_core_clk_rate", 0600, entry,
>>>>> &perf->fix_core_clk_rate);
>>>>> debugfs_create_u32("fix_core_ib_vote", 0600, entry,
>>>>> diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_core_perf.h b/drivers/gpu/drm/msm/disp/dpu1/dpu_core_perf.h
>>>>> index 5e07119c14c6a9ed3413d0eaddbd93df5cc3f79d..9d8516ca32d162b1e277ec88067e5c21abeb2017 100644
>>>>> --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_core_perf.h
>>>>> +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_core_perf.h
>>>>> @@ -24,20 +24,11 @@ struct dpu_core_perf_params {
>>>>> u64 core_clk_rate;
>>>>> };
>>>>> -/**
>>>>> - * struct dpu_core_perf_tune - definition of performance tuning control
>>>>> - * @mode: performance mode
>>>>> - */
>>>>> -struct dpu_core_perf_tune {
>>>>> - u32 mode;
>>>>> -};
>>>>> -
>>>>> /**
>>>>> * struct dpu_core_perf - definition of core performance context
>>>>> * @perf_cfg: Platform-specific performance configuration
>>>>> * @core_clk_rate: current core clock rate
>>>>> * @max_core_clk_rate: maximum allowable core clock rate
>>>>> - * @perf_tune: debug control for performance tuning
>>>>> * @enable_bw_release: debug control for bandwidth release
>>>>> * @fix_core_clk_rate: fixed core clock request in Hz used in mode 2
>>>>> * @fix_core_ib_vote: fixed core ib vote in bps used in mode 2
>>>>> @@ -47,7 +38,6 @@ struct dpu_core_perf {
>>>>> const struct dpu_perf_cfg *perf_cfg;
>>>>> u64 core_clk_rate;
>>>>> u64 max_core_clk_rate;
>>>>> - struct dpu_core_perf_tune perf_tune;
>>>>> u32 enable_bw_release;
>>>>> u64 fix_core_clk_rate;
>>>>> u32 fix_core_ib_vote;
>>>>>
>>>
>
More information about the Freedreno
mailing list