On Thu, 30 Oct 2008, Matthew Garrett wrote:
> On Wed, Oct 29, 2008 at 04:33:19PM -0700, Trent Piepho wrote:
>> On Wed, 22 Oct 2008, Matthew Garrett wrote:
>>> Mnf. I'm not convinced general kernel upstream would be enthusiastic.
>>
>> Any reason for that?
>
> Because any interface that basically enforces integer->string->integer
> conversion is an indication that you're already doing it wrong.
Many if not most users of the sysfs interface from userspace are doing
int->string->int. Yeah, 'cat' and 'echo' aren't, but most apps more
sophisticated than that do. So it's no more wrong than sysfs already is.
For what it used for, the sysfs int->string->int cost isn't very significant.
It's less than the kernel to userspace cost of sysfs.
If the overhead of the sysfs interface to userspace is not enough to be a
problem, and a kernel interface to sysfs has less overhead than from
userspace, then how can the kernel based overhead be too much?
Though maybe sysfs or hwmon could provide int<->string wrappers. Virtually
every hwmon store does a sprintf and virtually every show does a strtol. It
could quite possibly be more efficient to move all that code to common
set/show wrappers.
Then the hwmon driver would provide set/show methods use/produce ints instead
of strings. The wrappers would take care of converting those to/from strings.
>From this:
static SENSOR_DEVICE_ATTR(foo, S_IWUSR|S_IRUGO, show_foo, set_foo, 1);
static ssize_t show_foo(..., char *buf)
{ return sprintf(buf, "%d\n", value); }
static ssize_t set_foo(..., const char *buf, size_t count)
{ unsigned long val = simple_strtol(buf, NULL, 10); }
To this:
static SENSOR_DEVICE_ATTR_INT(foo, S_IWUSR | S_IRUGO, show_foo, set_foo, 1);
static ssize_t show_foo(..., signed long *output)
{ *output = value; return 0; }
static ssize_t set_foo(..., signed long val)
{ ... }
The actual sysfs store and set would be a wrapper that calls the methods
registered with SENSOR_DEVICE_ATTR_INT and does the int<->string stuff.
hwmon could then provide this function, which would call show_foo() directly.
int hwmon_show_attribute_int(const struct device *dev, const char *attr, long *out)
Instead of hwmon providing this stuff, sysfs could always do it. There are a
lot of non-hwmon attributes that could make use of it.
