Re: [PATCH v2] ipmi: Fix kernel panic at ipmi_ssif_thread()

[Corey Minyard <minyard@acm.org>]

On 03/27/2017 02:31 AM, Joeseph Chang wrote:
> From: Joeseph Chang <joechang@codeaurora.org>
>
> Since patch v1 touch ssif_info->multi_pos and ssif_info->multi_data
> after ssif_i2c_send(). msg_written_handler can be called at any time
> after ssif_i2c_send(). There is possible to have concurrent access to
> ssif_info->multi_pos or ssif_info->multi_data at msg_written_handler.
>
> Revert patch v1 and add new local pointer "i2c_data" which point to
> i2c data and size that will be sent out to avoid concurrent access in
> msg_written_handler().
>
> Signed-off-by: Joeseph Chang <joechang@codeaurora.org>
> ---
>   drivers/char/ipmi/ipmi_ssif.c | 24 ++++++++++++------------
>   1 file changed, 12 insertions(+), 12 deletions(-)
>
> diff --git a/drivers/char/ipmi/ipmi_ssif.c b/drivers/char/ipmi/ipmi_ssif.c
> index 39346ee..f36f018 100644
> --- a/drivers/char/ipmi/ipmi_ssif.c
> +++ b/drivers/char/ipmi/ipmi_ssif.c
> @@ -852,6 +852,7 @@ static void msg_written_handler(struct ssif_info *ssif_info, int result,
>   				unsigned char *data, unsigned int len)
>   {
>   	int rv;
> +	unsigned char *i2c_data;

The variable should be in the block where it is used.  The basic rule is 
to limit scope as much as possible.

The name is also a little vague, though not terrible.

>   
>   	/* We are single-threaded here, so no need for a lock. */
>   	if (result < 0) {
> @@ -899,13 +900,22 @@ static void msg_written_handler(struct ssif_info *ssif_info, int result,
>   			left = 32;
>   		/* Length byte. */
>   		ssif_info->multi_data[ssif_info->multi_pos] = left;
> +		i2c_data = ssif_info->multi_data + ssif_info->multi_pos;
> +		ssif_info->multi_pos += left;
> +		if (left < 32)
> +			/*
> +			 * Write is finished.  Note that we must end
> +			 * with a write of less than 32 bytes to
> +			 * complete the transaction, even if it is
> +			 * zero bytes.
> +			 */
> +			ssif_info->multi_data = NULL;

Can you make this change not relative to the previous change?  Just do a 
"git rebase -i ...." and squash the
changes.

I saw your previous emails, but I can't take those because of the 
comments at the top and the text from
the reply.  You can add comments to a submission that don't go into git, 
just put them after the '---" in
the header.

Thanks,

-corey

>   
>   		rv = ssif_i2c_send(ssif_info, msg_written_handler,
>   				  I2C_SMBUS_WRITE,
>   				  SSIF_IPMI_MULTI_PART_REQUEST_MIDDLE,
> -				  ssif_info->multi_data + ssif_info->multi_pos,
> +				  i2c_data,
>   				  I2C_SMBUS_BLOCK_DATA);
> -
>   		if (rv < 0) {
>   			/* request failed, just return the error. */
>   			ssif_inc_stat(ssif_info, send_errors);
> @@ -914,16 +924,6 @@ static void msg_written_handler(struct ssif_info *ssif_info, int result,
>   				pr_info("Error from i2c_non_blocking_op(3)\n");
>   			msg_done_handler(ssif_info, -EIO, NULL, 0);
>   		}
> -
> -		ssif_info->multi_pos += left;
> -		if (left < 32)
> -			/*
> -			 * Write is finished.  Note that we must end
> -			 * with a write of less than 32 bytes to
> -			 * complete the transaction, even if it is
> -			 * zero bytes.
> -			 */
> -			ssif_info->multi_data = NULL;
>   	} else {
>   		/* Ready to request the result. */
>   		unsigned long oflags, *flags;