Re: [PATCH] HID: sony: Fix for broken buttons on DS3 USB dongles

From: Scott Shumate
Date: Wed May 13 2020 - 16:06:08 EST


Hi Roderick,

The official DS3 has a Report Count(19) instead of Report Count(13) in the exact same offset. I have no idea what the silicon vendor for these dongles was thinking but it's suspicious that the official count of 19 (0x13) turned into 13 (0xd) in the knock-off. It makes you wonder if the engineers confused the decimal/hex numbers.

As buggy as all of these third-party devices are, I'm afraid relying on the HID parser to get it right is only going to worse over time. I do like your idea of having each device register themselves. It would be nice to have each device provide a callback to decode its own report rather than handle a bunch of special conditions and quirks in a unified report decoding function. The drawback of course is that its going to be a little more effort to maintain.

Cheers,
Scott


On 5/13/20 2:41 PM, Colenbrander, Roderick wrote:
Hi Scott,

Thanks for sharing this patch. Do you know if for these controllers the data is at the same byte offsets in the reports as an official DS3?

The reason I'm asking is that I have been considering for a while to redo some of the button / stick handling code and the issue you just pointed out could be another argument towards doing that. Basically we learned that also for the official Navigation controller there are apparently a few different revisions with different report descriptors (yah!).

I have been tempted for some time to get rid of this fixup / mapping logic altogether and e.g. have a "sony_register_gamepad" and do it all ourselves and put more logic into the sixaxis_parse_report call. With many buggy devices of which we don't have most it feels fragile to have the HID parser do the work.

Thanks,
Roderick










From: Scott Shumate <scott.shumate@xxxxxxxxx>

Sent: Wednesday, May 13, 2020 10:05 AM

To: Jiri Kosina <jikos@xxxxxxxxxx>; Benjamin Tissoires <benjamin.tissoires@xxxxxxxxxx>; Colenbrander, Roderick <Roderick.Colenbrander@xxxxxxxx>; linux-input@xxxxxxxxxxxxxxx <linux-input@xxxxxxxxxxxxxxx>; linux-kernel@xxxxxxxxxxxxxxx <linux-kernel@xxxxxxxxxxxxxxx>

Subject: [PATCH] HID: sony: Fix for broken buttons on DS3 USB dongles



Fix for non-working buttons on knock-off USB dongles for Sony
controllers. These USB dongles are used to connect older Sony DA/DS1/DS2
controllers via USB and are common on Amazon, AliExpress, etc. Without
the patch, the square, X, and circle buttons do not function. These
dongles used to work prior to kernel 4.10 but removing the global DS3
report fixup in this commit exposed the problem:

commit e19a267b9987 Author: Roderick Colenbrander
Â<roderick.colenbrander@xxxxxxxx>
Date:ÂÂ Tue Mar 7 15:45:08 2017 -0800

ÂÂÂ HID: sony: DS3 comply to Linux gamepad spec

Many people reported the problem on the Ubuntu forums and are working
around the problem by falling back to the 4.9 hid-sony driver.

The problem stems from these dongles incorrectly reporting their button
count as 13 instead of 16. This patch fixes up the report descriptor by
changing the button report count to 16 and removing 3 padding bits.

Signed-off-by: Scott Shumate <scott.shumate@xxxxxxxxx>
---
diff --git a/drivers/hid/hid-sony.c b/drivers/hid/hid-sony.c
index 4c6ed6ef31f1..2f073f536070 100644
--- a/drivers/hid/hid-sony.c
+++ b/drivers/hid/hid-sony.c
@@ -867,6 +867,23 @@ static u8 *sony_report_fixup(struct hid_device *hdev, u8 *rdesc,
if (sc->quirks & PS3REMOTE)
return ps3remote_fixup(hdev, rdesc, rsize);
+ /*
+ * Some knock-off USB dongles incorrectly report their button count
+ * as 13 instead of 16 causing three non-functional buttons.
+ */
+ if ((sc->quirks & SIXAXIS_CONTROLLER_USB) && *rsize >= 45 &&
+ /* Report Count (13) */
+ rdesc[23] == 0x95 && rdesc[24] == 0x0D &&
+ /* Usage Maximum (13) */
+ rdesc[37] == 0x29 && rdesc[38] == 0x0D &&
+ /* Report Count (3) */
+ rdesc[43] == 0x95 && rdesc[44] == 0x03) {
+ hid_info(hdev, "Fixing up USB dongle report descriptor\n");
+ rdesc[24] = 0x10;
+ rdesc[38] = 0x10;
+ rdesc[44] = 0x00;
+ }
+
return rdesc;
}