2014-06-10 14:33:57

by Dave Jones

Subject: Re: i2c: ChromeOS EC tunnel driver

On Fri, Jun 06, 2014 at 07:31:28PM +0000, Linux Kernel wrote:
> Gitweb: http://git.kernel.org/linus/;a=commit;h=9d230c9e4f4e67cb1c1cb9e0f6142da16b0f2796
> Commit: 9d230c9e4f4e67cb1c1cb9e0f6142da16b0f2796
> Parent: 5271db29d7199fe0ffb303ca4bbbb1485bba28c3
> Refname: refs/heads/next
> Author: Doug Anderson <[email protected]>
> AuthorDate: Wed Apr 30 10:44:09 2014 -0700
> Committer: Lee Jones <[email protected]>
> CommitDate: Tue Jun 3 08:11:49 2014 +0100

> +static int ec_i2c_construct_message(u8 *buf, const struct i2c_msg i2c_msgs[],
> + int num, u16 bus_num)
> +{
> + struct ec_params_i2c_passthru *params;
> + u8 *out_data;
> + int i;
> +
> + out_data = buf + sizeof(struct ec_params_i2c_passthru) +
> + num * sizeof(struct ec_params_i2c_passthru_msg);
> +
> + params = (struct ec_params_i2c_passthru *)buf;
> + params->port = bus_num;
> + params->num_msgs = num;
> + for (i = 0; i < num; i++) {
> + const struct i2c_msg *i2c_msg = &i2c_msgs[i];
> + struct ec_params_i2c_passthru_msg *msg = &params->msg[i];
> +
> + msg->len = i2c_msg->len;
> + msg->addr_flags = i2c_msg->addr;
> +
> + if (i2c_msg->flags & I2C_M_TEN)
> + msg->addr_flags |= EC_I2C_FLAG_10BIT;

This is a fun bug. EC_I2C_FLAG_10BIT is 1<<16, but msg->addr_flags is only a uint16.

Dave

2014-06-10 15:39:45

by Doug Anderson

Subject: Re: i2c: ChromeOS EC tunnel driver

Dave,

On Tue, Jun 10, 2014 at 7:33 AM, Dave Jones <[email protected]> wrote:
> On Fri, Jun 06, 2014 at 07:31:28PM +0000, Linux Kernel wrote:
> > Gitweb: http://git.kernel.org/linus/;a=commit;h=9d230c9e4f4e67cb1c1cb9e0f6142da16b0f2796
> > Commit: 9d230c9e4f4e67cb1c1cb9e0f6142da16b0f2796
> > Parent: 5271db29d7199fe0ffb303ca4bbbb1485bba28c3
> > Refname: refs/heads/next
> > Author: Doug Anderson <[email protected]>
> > AuthorDate: Wed Apr 30 10:44:09 2014 -0700
> > Committer: Lee Jones <[email protected]>
> > CommitDate: Tue Jun 3 08:11:49 2014 +0100
>
> > +static int ec_i2c_construct_message(u8 *buf, const struct i2c_msg i2c_msgs[],
> > + int num, u16 bus_num)
> > +{
> > + struct ec_params_i2c_passthru *params;
> > + u8 *out_data;
> > + int i;
> > +
> > + out_data = buf + sizeof(struct ec_params_i2c_passthru) +
> > + num * sizeof(struct ec_params_i2c_passthru_msg);
> > +
> > + params = (struct ec_params_i2c_passthru *)buf;
> > + params->port = bus_num;
> > + params->num_msgs = num;
> > + for (i = 0; i < num; i++) {
> > + const struct i2c_msg *i2c_msg = &i2c_msgs[i];
> > + struct ec_params_i2c_passthru_msg *msg = &params->msg[i];
> > +
> > + msg->len = i2c_msg->len;
> > + msg->addr_flags = i2c_msg->addr;
> > +
> > + if (i2c_msg->flags & I2C_M_TEN)
> > + msg->addr_flags |= EC_I2C_FLAG_10BIT;
>
> This is a fun bug. EC_I2C_FLAG_10BIT is 1<<16, but msg->addr_flags is only a uint16.

Good catch. Luckily we have no actual 10-bit devices on the other
end, but it sure makes this bit of the tunnel pretty stupid.

The correct fix is to fix the master 'ec_commands.h' in the ChromeOS
EC project, then pull the change over to the kernel. I've filed a bug
against our EC team at http://crbug.com/382944

-Doug