Re: [PATCH 02/11] HID: i2c-hid: implement ll_driver transport-layer callbacks

On Mon, Feb 3, 2014 at 11:04 AM, David Herrmann [off-list ref] wrote:

Hi

On Sun, Feb 2, 2014 at 5:50 AM, Benjamin Tissoires
[off-list ref] wrote:

quoted

Add output_report and raw_request to i2c-hid.
Hopefully, we will manage to have the same transport level between
all the transport drivers.

Signed-off-by: Benjamin Tissoires <redacted>
---
 drivers/hid/i2c-hid/i2c-hid.c | 24 ++++++++++++++++++++++++
 1 file changed, 24 insertions(+)

diff --git a/drivers/hid/i2c-hid/i2c-hid.c b/drivers/hid/i2c-hid/i2c-hid.c
index ce68a12..5099f1f 100644
--- a/drivers/hid/i2c-hid/i2c-hid.c
+++ b/drivers/hid/i2c-hid/i2c-hid.c

@@ -574,6 +574,28 @@ static int i2c_hid_output_raw_report(struct hid_device *hid, __u8 *buf,
        return ret;
 }

+static int i2c_hid_output_report(struct hid_device *hid, __u8 *buf,
+               size_t count)
+{
+       return i2c_hid_output_raw_report(hid, buf, count, HID_OUTPUT_REPORT);
+}
+
+static int i2c_hid_raw_request(struct hid_device *hid, unsigned char reportnum,
+                              __u8 *buf, size_t len, unsigned char rtype,
+                              int reqtype)
+{
+       switch (reqtype) {
+       case HID_REQ_GET_REPORT:
+               return i2c_hid_get_raw_report(hid, reportnum, buf, len, rtype);
+       case HID_REQ_SET_REPORT:
+               if (buf[0] != reportnum)
+                       return -EINVAL;
+               return i2c_hid_output_raw_report(hid, buf, len, rtype);

I just skimmed the I2C-HID specs and it defines three methods for
input/output reports:

1) Section 6.2:
raw async output-reports can be sent by writing the data at any time
to wOutputRegister.
This should be used as method for hid->output_report().

2) Section 7.1:
SET_REPORT can be issued by writing the right OPCODE + report-ID into
wCommandRegister and the data into wDataRegister.
This should be used as method for hid->raw_request() + HID_REQ_SET_REPORT.

3) Section 7.1:
GET_REPORT can be issued by writing the right OPCODE + report-ID into
wCommandRegister and then waiting for the device to write the data
into wDataRegister.
This should be used for hid->raw_request() + HID_REQ_GET_REPORT


The GET_REPORT implementation looks fine to me, but the
i2c_hid_set_report() seems to support both 1) and 2) depending on the
passed type and capabilities:
 - it uses 2) for FEATURE_REPORT reports or if the max OUTPUT_LENGTH is 0
 - it uses 1) otherwise

I'm not sure whether the i2c-hid-spec mandates this behavior, so I am
not saying it's wrong. I just wanna understand what we do here. So if
we use hid->output_report() with HID_FEATURE_REPORT, the current code
turns this into a SET_REPORT. Likewise, an hid->raw_request() with
HID_REQ_SET_REPORT but with HID_OUTPUT_REPORT turns into an
output-report.

Ok, just for the record (I think that you already answered this in the
4/11 review):
this behavior has been added in commit
811adb9622de310efbb661531c3ec0ae5d2b2bc0 for Synaptics so they can
talk to their HID/I2C firmware.
The use they are making is through hdev->hid_output_raw_report(), and
at that time output_report() did not exist.

I guess that with the new API (output_report() and raw_request() ) we
can revert this and it will not bother them given that rmi-hid.c is
still not upstream.

I have no clue if anyone else is using hdev->hid_output_raw_report()
for I2C devices, but we could still try to clean this right now given
that those devices are not well spread (besides some hid-multitouch
touchscreens which do not use OUTPUT reports).

I am still a little bit concerned by hidraw not able to call
set_report in this case, but this could be fixed also by adding an
ioctl.

How does that sound?

Cheers,
Benjamin

I'd rather expect this behavior:

hid->output_report() should always do this:
  args[index++] = outputRegister & 0xFF;
  args[index++] = outputRegister >> 8;
  hidcmd = &hid_no_cmd;

while hid->raw_request() should always do this:
  args[index++] = dataRegister & 0xFF;
  args[index++] = dataRegister >> 8;

The special case for maxOutputLength==0 seems fine to me, but the
"reportType == 0x03" looks weird.


Thanks
David

quoted

+       default:
+               return -EIO;
+       }
+}
+
 static void i2c_hid_request(struct hid_device *hid, struct hid_report *rep,
                int reqtype)
 {

@@ -761,6 +783,8 @@ static struct hid_ll_driver i2c_hid_ll_driver = {
        .close = i2c_hid_close,
        .power = i2c_hid_power,
        .request = i2c_hid_request,
+       .output_report = i2c_hid_output_report,
+       .raw_request = i2c_hid_raw_request,
 };

 static int i2c_hid_init_irq(struct i2c_client *client)
--

1.8.3.1