Re: [PATCH 3/3] cxacru: Store all device status information and report it when atm_proc_read is called.

[Simon Arlott <simon@arlott.org>]

[-- Attachment #1: Type: text/plain, Size: 2920 bytes --]

On 01/02/07 09:15, Duncan Sands wrote:
> On Thursday 1 February 2007 00:39:14 you wrote:
>> On Tue, 30 Jan 2007 21:30:29 +0000
>> Simon Arlott <simon@arlott.org> wrote:
>>
>>> +static int cxacru_proc_read(struct usbatm_data *usbatm_instance,
>>> +	struct atm_dev *atm_dev, loff_t * pos, char *page)
>>> +{
>>> +	struct cxacru_data *instance = usbatm_instance->driver_data;
>>> +	u32 *cxinf = instance->cxinf_status;
>>> +	int left = *pos;
>>> +
>>> +	if (!left--)
>>> +		return sprintf(page, "# %s\n", usbatm_instance->description);
>>> +
>>> +	if (!left--) {
>>> +		if (cxinf[CXINF_LINE_STATUS] == 5) {
>>> +			return sprintf(page, "# UP %u/%u\n",
>>> +				cxinf[CXINF_DOWNSTREAM_RATE],
>>> +				cxinf[CXINF_UPSTREAM_RATE]);
>>> +		} else {
>>> +			return sprintf(page, "# DOWN\n");
>>> +		}
>>> +	}
>> hm, how well-tested was this proc interface?  The pread() and lseek()
>> behaviour might be strange.
>>
>> I guess as long as it doesn't oops, hang or anything like that then it'll
>> be OK.  Anyone who does anything apart from a single big-fat-read from a 
>> procfile has a good chance of getting into trouble :(

Well, several reads of at least 172B... but that scenario works without any problems.

> All the ATM drivers seem to do it like this.  That doesn't mean they are
> right of course!  But I never saw anyone complain on the ATM mailing list.

I just copied it from usbatm; it looks like this could be done using seq_file instead.

> On the other hand, why does Simon want this?  If he has written a user space
> tool that extracts bits from the proc file (eg: to tell users what's going
> on) then he could run into trouble, depending on how he implements it.

No special user space tool, I just want other people to be able to get the same useful line information from cxacru that most ADSL driver have. (I do have a small "cxacru-watch" program which execs "cmd.up" etc. on status changes, but that is just sent printks via syslog-ng).


It looks like the weird proc interface prevents my intention to allow ". /proc/net/atm/cxacru\:0" to work and set $LINE_STATUS etc., at least with bash:
open("/proc/net/atm/cxacru:0", O_RDONLY|O_LARGEFILE) = 3
fstat64(3, {st_mode=S_IFREG|0444, st_size=0, ...}) = 0
read(3, "", 0)                          = 0
close(3)                                = 0

Does anyone have any thoughts on the output format? The standard usbatm output isn't going to be compatible with the above if a common MAC/AAL5 header were used.


When I have the time and can acquire another conexant usb modem for testing I'll fix the usbatm /proc interfaces. It would be also good if the module could reliably be reloaded - most (but not all) of the time it fails to initialise the device again.

There's still an issue with khubd going into a loop if the firmware doesn't exist or if the device is unplugged.

-- 
Simon Arlott


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 829 bytes --]