On Tuesday 28 April 2015 18:50:25 Baolin Wang wrote:
On 28 April 2015 at 17:49, Arnd Bergmann arnd@linaro.org wrote:
+void +jiffies_to_timespec(const unsigned long jiffies, struct timespec *value) +{
struct timespec64 *ts;*ts = timespec_to_timespec64(*value);jiffies_to_timespec64(jiffies, ts);+} EXPORT_SYMBOL(jiffies_to_timespec);
I would probably again put timespec_to_jiffies and jiffies_to_timespec into the header file as static inline functions, which implies that you then have to declare and export __timespec_to_jiffies.
ArndBut according to Thomas's suggestion, he would not like to add too much "EXPORT_SYMBOL". I also think here the "EXPORT_SYMBOL" is essential.
I think you just misunderstood his comments:
On Monday 20 April 2015 23:09:17 Thomas Gleixner wrote:
So we have now two exports which are doing exactly the same thing. Copy and paste is wonderful, right?
What about exporting __timespec_to_jiffies() and providing inlines for timespec_to_jiffies() and timespec64_to_jiffies() ?
EXPORT_SYMBOL is not just a stupid annotation. Its impact on the resulting kernel size is larger than the actual function implementation.
The main takeaway here is that you should use specifically the exports that are required, no more or less.
If a kernel-internal API has an 'extern' declaration in a header file, it usually needs to be exported. What Thomas was suggesting here was to reduce the number of exports that are required by moving the two functions timespec_to_jiffies() and timespec64_to_jiffies() into the header file as 'static inline' functions that do not require the export.
If you get a review comment asking you to change something in your patches, always make sure that you understand why that change is a good idea. If you don't understand it, it's better to ask for clarification than to blindly follow the advice and risk doing it wrong again.
Arnd