Re: [PATCH 18/18] drm/i915/ddi: prefer read_poll_timeout() over readx_poll_timeout()

[Jani Nikula <jani.nikula@intel.com>]

On Fri, 27 Jun 2025, Ville Syrjälä <ville.syrjala@linux.intel.com> wrote:
> On Fri, Jun 27, 2025 at 02:36:32PM +0300, Jani Nikula wrote:
>> Unify on using read_poll_timeout() throughout instead of mixing with
>> readx_poll_timeout(). While the latter can be ever so slightly simpler,
>> they are both complicated enough that it's better to unify on one
>> approach only.
>> 
>> While at it, better separate the handling of error returns from
>> drm_dp_dpcd_readb() and the actual status byte. This is best achieved by
>> inlining the read_fec_detected_status() function.
>> 
>> Cc: Imre Deak <imre.deak@intel.com>
>> Signed-off-by: Jani Nikula <jani.nikula@intel.com>
>> ---
>>  drivers/gpu/drm/i915/display/intel_ddi.c | 33 +++++++++---------------
>>  1 file changed, 12 insertions(+), 21 deletions(-)
>> 
>> diff --git a/drivers/gpu/drm/i915/display/intel_ddi.c b/drivers/gpu/drm/i915/display/intel_ddi.c
>> index 0405396c7750..fc4587311607 100644
>> --- a/drivers/gpu/drm/i915/display/intel_ddi.c
>> +++ b/drivers/gpu/drm/i915/display/intel_ddi.c
>> @@ -2339,34 +2339,25 @@ static void intel_dp_sink_set_fec_ready(struct intel_dp *intel_dp,
>>  		drm_dbg_kms(display->drm, "Failed to clear FEC detected flags\n");
>>  }
>>  
>> -static int read_fec_detected_status(struct drm_dp_aux *aux)
>> -{
>> -	int ret;
>> -	u8 status;
>> -
>> -	ret = drm_dp_dpcd_readb(aux, DP_FEC_STATUS, &status);
>> -	if (ret < 0)
>> -		return ret;
>> -
>> -	return status;
>> -}
>> -
>>  static int wait_for_fec_detected(struct drm_dp_aux *aux, bool enabled)
>>  {
>>  	struct intel_display *display = to_intel_display(aux->drm_dev);
>>  	int mask = enabled ? DP_FEC_DECODE_EN_DETECTED : DP_FEC_DECODE_DIS_DETECTED;
>> -	int status;
>> -	int err;
>> +	u8 status = 0;
>> +	int ret, err;
>>  
>> -	err = readx_poll_timeout(read_fec_detected_status, aux, status,
>> -				 status & mask || status < 0,
>> -				 10000, 200000);
>> +	ret = read_poll_timeout(drm_dp_dpcd_readb, err,
>> +				err || (status & mask),
>> +				10 * 1000, 200 * 1000, false,
>> +				aux, DP_FEC_STATUS, &status);
>
> I think I hate these macros. It's very hard to tell from this
> soup what is actually being done here.

The thing is, I hate __wait_for(), wait_for(), wait_for_us(),
wait_for_atomic_us(), and wait_for_atomic() even more.

It's also very hard to figure out what is actually going on with
them. The timeouts are arbitrarily either ms or us. wait_for_us() is
atomic depending on the timeout. __wait_for() Wmax parameter actually
isn't the max sleep, it's 2*Wmax-2. Some of them have exponentially
growing sleeps, while some arbitrarily don't.

It's a fscking mess, and people randomly choose whichever version with
no idea what's actually going on behind the scenes.

> The 'val', 'op', and 'args' look very disconnected here even though
> they are always part of the same thing. Is there a reason they can't
> just be a single 'op' parameter like we have in wait_for() so you can
> actually see the code?
>
> Ie.
> read_poll_timeout(err = drm_dp_dpcd_readb(aux, DP_FEC_STATUS, &status),
> 		  err || (status & mask),
>                   10 * 1000, 200 * 1000, false);
> ?

Internally the macro has:

#define read_poll_timeout(op, val, cond, sleep_us, timeout_us, \
				sleep_before_read, args...) \

...

		(val) = op(args); \

So you do need to provide an lvalue val, and you need to be able to add
() after op. I think GCC allows not passing varargs. IOW you'd need to
implement another macro (which could be used to implement the existing
one, but not the other way round).

I'm really not enthusiastic about blocking this series waiting on that
kind of refactoring in iopoll.h which might happen, or might not,
considering there's no active maintainer for iopoll.h.

So yeah, the interface isn't great, and I'm not claiming it is, but it
is *one* *single* *documented* *interface* that's used across the
kernel. On the whole, warts and all, I think it's still much better than
what we currently have. And it breaks the dependency on i915_utils.h.

I've carefully tried to do the line breaks so that it's always:

        read_poll_timeout(op, val,
                          cond,
                          sleep_us, timeout_us, sleep_before_read,
                          args...);

I think that helps a bit.


BR,
Jani.


>
>>  
>> -	if (err || status < 0) {
>> +	/* Either can be non-zero, but not both */
>> +	ret = ret ?: err;
>> +	if (ret) {
>>  		drm_dbg_kms(display->drm,
>> -			    "Failed waiting for FEC %s to get detected: %d (status %d)\n",
>> -			    str_enabled_disabled(enabled), err, status);
>> -		return err ? err : status;
>> +			    "Failed waiting for FEC %s to get detected: %d (status 0x%02x)\n",
>> +			    str_enabled_disabled(enabled), ret, status);
>> +		return ret;
>>  	}
>>  
>>  	return 0;
>> -- 
>> 2.39.5

-- 
Jani Nikula, Intel