On Fri, 15 Jun 2012 09:51:47 -0700, Guenter Roeck wrote: > On Fri, 2012-06-15 at 11:52 -0400, Wolfram Sang wrote: > > I won't NACK it, but I don't think it is worth it. It increases runtime > > memory usage and execution time. Marginally, yes, but the gain in code > > size reduction is also marginally. If there is lots of devm_* in a > > driver, this is usually good, but I wouldn't care about converting > > single kzalloc calls. > > code size reduction on x86_64 is about 15 kBytes for all drivers. With > most of the hwmon drivers built but not loaded on a typical system, that > does make a difference in disk space usage, and I think it is a bit more > than marginal. The runtime execution time increase only occurs once, > when a module is loaded or unloaded, and the increase in memory usage > only applies to loaded drivers. > > In addition to that, it does help preventing memory leaks, and it does > make the code a bit simpler. > > So, overall I do think it is worth it. Maybe not if I only look at one > driver, but very much so if I look at the overall impact. I agree with Guenter here. In fact, I suspect it's only a matter of time before the use of devm_* functions becomes mandatory. Fewer lines of code and fewer bugs is definitely something we want to achieve. If it was deemed not worth the cost, the devm_* infrastructure would have been rejected in the first place. Now that it is in place, it is meant to be used as much as possible. Gunter, I think it would make sense do document it Documentation/hwmon/submitting-patches, section 3, so that you don't have to do the same again in the future. Great work, BTW. I don't intend to review all of it, no time, sorry, but I ack the general idea. -- Jean Delvare _______________________________________________ lm-sensors mailing list lm-sensors@xxxxxxxxxxxxxx http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
