Re: [PATCH v1] HID: support multiple and large i2c transfers in hid-cp2112

On Sat, May 30, 2015 at 1:32 AM, Ellen Wang [off-list ref] wrote:

On 05/29/2015 09:28 AM, Antonio Borneo wrote:

quoted

On Fri, Apr 24, 2015 at 5:49 AM, Ellen Wang [off-list ref]
wrote:

quoted

cp2112_i2c_xfer() only supports a single i2c_msg and only
reads up to 61 bytes.  More than one message at a time
and longers reads just return errors.

This is a serious limitation.  For example, the at24 eeprom driver
generates paired write and read messages (for eeprom address and data).
And the reads can be quite large.  The fix consists of a loop
to go through all the messages, and a loop around cp2112_read()
to pick up all the returned data.  For extra credit, it now also
supports write-read pairs and implements them as
CP2112_DATA_WRITE_READ_REQUEST.

Signed-off-by: Ellen Wang <redacted>

Hi Ellen,

the patch is partially wrong. See below.

I added Wolfram, I2C subsystem maintainer, in copy.

quoted

---
  drivers/hid/hid-cp2112.c |  136
+++++++++++++++++++++++++++++-----------------
  1 file changed, 87 insertions(+), 49 deletions(-)

diff --git a/drivers/hid/hid-cp2112.c b/drivers/hid/hid-cp2112.c
index 3318de6..e7e72a4 100644
--- a/drivers/hid/hid-cp2112.c
+++ b/drivers/hid/hid-cp2112.c

@@ -444,11 +444,30 @@ static int cp2112_i2c_write_req(void *buf, u8

slave_address, u8 *data,
         return data_length + 3;
  }

+static int cp2112_i2c_write_read_req(void *buf, u8 slave_address,
+                                    u8 *addr, int addr_length,
+                                    int read_length)
+{
+       struct cp2112_write_read_req_report *report = buf;
+
+       if (read_length < 1 || read_length > 512 ||
+           addr_length > sizeof(report->target_address))
+               return -EINVAL;
+
+       report->report = CP2112_DATA_WRITE_READ_REQUEST;
+       report->slave_address = slave_address << 1;
+       report->length = cpu_to_be16(read_length);
+       report->target_address_length = addr_length;
+       memcpy(report->target_address, addr, addr_length);
+       return addr_length + 5;
+}
+
  static int cp2112_i2c_xfer(struct i2c_adapter *adap, struct i2c_msg
*msgs,
                            int num)
  {
         struct cp2112_device *dev = (struct cp2112_device
*)adap->algo_data;
         struct hid_device *hdev = dev->hdev;
+       struct i2c_msg *m;
         u8 buf[64];
         ssize_t count;
         unsigned int retries;

@@ -456,71 +475,90 @@ static int cp2112_i2c_xfer(struct i2c_adapter

*adap, struct i2c_msg *msgs,

         hid_dbg(hdev, "I2C %d messages\n", num);

-       if (num != 1) {
-               hid_err(hdev,
-                       "Multi-message I2C transactions not
supported\n");
-               return -EOPNOTSUPP;
-       }
-
-       if (msgs->flags & I2C_M_RD)
-               count = cp2112_read_req(buf, msgs->addr, msgs->len);
-       else
-               count = cp2112_i2c_write_req(buf, msgs->addr, msgs->buf,
-                                            msgs->len);
-
-       if (count < 0)
-               return count;
-
         ret = hid_hw_power(hdev, PM_HINT_FULLON);
         if (ret < 0) {
                 hid_err(hdev, "power management error: %d\n", ret);
                 return ret;
         }

-       ret = cp2112_hid_output(hdev, buf, count, HID_OUTPUT_REPORT);
-       if (ret < 0) {
-               hid_warn(hdev, "Error starting transaction: %d\n", ret);
-               goto power_normal;
-       }
+       for (m = msgs; m < msgs + num; m++) {

CP2112 is not able to combine messages since unable to repeat the start
bits.
It can only send simple transfers or a combined read-after-write.

Here you cannot simply loop on all the messages and send them one by
one because there is no way to force the repeated start bit in
between.
You can find more details in this thread about CP2112
https://lkml.org/lkml/2015/3/15/64
and here info about the repeated start bit under "Combined transactions"
Documentation/i2c/i2c-protocol

Anyway your idea to introduce read-after-write is valid.
I suggest you to check the drivers i2c-opal.c and i2c-pmcmsp.c ; at a
quick check I think they have similar limitation as CP2112.
You could replicate the same check, supporting only num==1 (always)
and num==2 (only if msg[0] is write and msg[1] is read).

Best Regards,
Antonio

It don't think the code attempts to generate repeated start.  It simply
issues and completes each message as separate transfers.  It's not different
from calling cp2112_i2c_xfer() repeatedly with single messages, except in
the bracketing hid_hw_power().

cp2112_i2c_xfer() is the lower level side of i2c_transfer().
i2c_transfer() requires that only one Stop bit is sent at the end of
last message. Between messages a repeated start is mandatory.
In general case, CP2112 cannot do that, that's why the original code
returns error if num != 1

It's not correct to issue each message as separate transfer inside
cp2112_i2c_xfer(); the caller of i2c_transfer() expects the messages
to be sent with repeated start.

With you patch you highlight that there is another case that should be
implemented.
CP2112 can handle the case of num==2 when msg[0] is write and msg[1] is read.
There are other devices with similar limitations, as I pointed above.

Extending cp2112_i2c_xfer() to support a combined read-after-write is
ok, while issuing each message without forcing repeated start is
incorrect.

Best Regards,
Antonio