Re: [Freedreno] [PATCH] drm/msm/dp: move add fail safe mode to dp_connector_get_mode()

From: Dmitry Baryshkov
Date: Tue Apr 26 2022 - 13:01:21 EST

On 26/04/2022 18:37, Abhinav Kumar wrote:
Hi Doug

On 4/26/2022 8:20 AM, Doug Anderson wrote:

On Mon, Apr 25, 2022 at 8:35 PM Abhinav Kumar <quic_abhinavk@xxxxxxxxxxx> wrote:

On 4/25/2022 7:18 PM, Doug Anderson wrote:

On Mon, Apr 25, 2022 at 6:42 PM Abhinav Kumar <quic_abhinavk@xxxxxxxxxxx> wrote:

2) When there was a valid EDID but no 640x480 mode

This is the equipment specific case and the one even I was a bit
surprised. There is a DP compliance equipment we have in-house and while
validation, it was found that in its list of modes , it did not have any
modes which chromebook supported ( due to 2 lanes ). But my
understanding was that, all sinks should have atleast 640x480 but
apparently this one did not have that. So to handle this DP compliance
equipment behavior, we had to do this.

That doesn't seem right. If there's a valid EDID and the valid EDID
doesn't contain 640x480, are you _sure_ you're supposed to be adding
640x480? That doesn't sound right to me. I've got a tiny display in
front of me for testing that only has one mode:

     #0 800x480 65.68 800 840 888 928 480 493 496 525 32000

As I had wrote, DRM core kicks in only when the count of modes is 0.
Here what is happening is the count was not 0 but 640x480 was not
present in the EDID. So we had to add it explicitly.

Your tiny display is a display port display?

I am referring to only display port monitors. If your tiny display is
DP, it should have had 640x480 in its list of modes.

My tiny display is actually a HDMI display hooked up to a HDMI to DP
(active) adapter.

...but this is a legal and common thing to have. I suppose possibly my
HDMI display is "illegal"?

OK, so reading through the spec more carefully, I do see that the DP
spec makes numerous mentions of the fact that DP sinks _must_ support
640x480. Even going back to DP 1.4, I see section " Video
Timing Format" says that we must support 640x480. It seems like that's
_intended_ to be used only if the EDID read fails, though or if we
somehow have to output video without knowledge of the EDID. It seems
hard to believe that there's a great reason to assume a display will
support 640x480 if we have more accurate knowledge.

In any case, I guess I would still say that adding this mode belongs
in the DRM core. The core should notice that it's a DP connection
(bridge->type == DRM_MODE_CONNECTOR_DisplayPort) and that 640x480 was
left out and it should add it. We should also make sure it's not
"preferred" and is last in the list so we never accidentally pick it.
If DP truly says that we should always give the user 640x480 then
that's true for everyone, not just Qualcomm. We should add it in the
core. If, later, someone wants to hide this from the UI it would be
much easier if they only needed to modify one place.

So I debugged with kuogee just now using the DP compliance equipment.
It turns out, the issue is not that 640x480 mode is not present.

The issue is that it is not marked as preferred.

Hence we missed this part during debugging this equipment failure.

We still have to figure out the best way to either mark 640x480 as
preferred or eliminate other modes during the test-case so that 640x480
is actually picked by usermode.

Now that being said, the fix still doesn't belong in the framework. It
has to be in the msm/dp code.

Different vendors handle this failure case differently looks like.

Lets take below snippet from i915 as example.

3361    if (intel_connector->detect_edid == NULL ||
3362        connector->edid_corrupt ||
3363        intel_dp->aux.i2c_defer_count > 6) {
3364            /* Check EDID read for NACKs, DEFERs and corruption
3365             * (DP CTS 1.2 Core r1.1)
3366             * : Failed EDID read, I2C_NAK
3367             * : Failed EDID read, I2C_DEFER
3368             * : EDID corruption detected
3369             * Use failsafe mode for all cases
3370             */
3371            if (intel_dp->aux.i2c_nack_count > 0 ||
3372                    intel_dp->aux.i2c_defer_count > 0)
3373                    drm_dbg_kms(&i915->drm,
3374                                "EDID read had %d NACKs, %d DEFERs\n",
3375                                intel_dp->aux.i2c_nack_count,
3376                                intel_dp->aux.i2c_defer_count);
3377            intel_dp->compliance.test_data.edid = INTEL_DP_RESOLUTION_FAILSAFE;

The reason I pointed to this code is to give an example of how other drivers handle this test-case.

We added this patch for and EDID test cases.

The challenge here as found out from our discussion here was to mark a particular mode as preferred so that the Chrome usermode can pick it.

Now whats happening with that there was always a possibility of two modes being marked as preferred due to this and so-on.

We had a pretty long discussion last night and thought of all possible solutions but all of them look like a hack to us in the driver because we end up breaking other things due to this.

So we decided that driver is not the place to handle this test case.
Since we do have IGT support for chromebooks, we will handle both these test cases there as other vendors do the same way and it works.

Just because Intel DRM has its own solution for something doesn't mean
everyone else should copy them and implement their own solution. Up
until recently DP AUX backlights were baked into different DRM
drivers. A recent effort was made to pull it out. I think the Intel
DRM code was the "first one" to the party and it wasn't clear how
things should be broken up to share with other drivers, so mostly it
did everything itself, but that's not the long term answer.

I'm not saying that we need to block your change on a full re-design
or anything, but I'm just saying that:

* You're trying to implement a generic DP rule, not something specific
to Qualcomm hardware. That implies that, if possible, it shouldn't be
in a Qualcomm driver.

* It doesn't seem like it would be terrible to handle this in the core.

This marks the fail safe mode and IGT test case reads this to set this
mode and hence the test passes.

We rely on the chromeOS usermode to output pixel data for this test-case
and not IGT. We use IGT only for video pattern CTS today but this is a
different test-case which is failing.

ChromeOS usermode will not pick 640x480 unless we mark it as preferred
or other modes are eliminated.

So we have to come up with the right way for the usermode to pick 640x480.

We will discuss this a bit more and come up with a different change.

Can you provide the exact EDID from the failing test case? Maybe that
will help shed some light on what's going on. I looked at the original
commit and it just referred to, which I assume is "EDID Read
upon HPD Plug Event", but that doesn't give details that seem relevant
to the discussion here.

Yes so it is and

That alone wont give the full picture.

So its a combination of things.

While running the test, the test equipment published only one mode.
But we could not support that mode because of 2 lanes.
Equipment did not add 640x480 to the list of modes.
DRM fwk will also not add it because count_modes is not 0 ( there was one mode ).
So we ended up making these changes.

I think a proper solution might be to rewrite drm_helper_probe_single_connector_modes() in the following way:
- call get_modes()
- validate the result
- prune invalid

- if the number of modes is 0, call drm_add_override_edid_modes()
- validate the result
- prune invalid

- if the number of modes is still 0, call drm_add_modes_noedid()
- validate the result
- prune invalid

[A separate change might happen here after all the checks: if the number of modes is still 0 and if it is a DP, enforce adding 640x480 even w/o validation. But generally I feel that this shouldn't be necessary because the previous step should have added it.]

This way we can be sure that all modes are validated, but still to do our best to add something supported to the list of modes.

I guess maybe what's happening is that the test case is giving an EDID
where all the modes are not supportable by the current clock rate /
lanes? ...and then somehow we're not falling back to 640x480. It's
always possible that this is a userspace problem.

In any case, would you object to a revert of the patches in the short term?

Not sure, if you saw this change kuogee posted last night.
We did decided to remove all the code related to these test cases and handle them in IGT.


With best wishes