On 12/25/20 6:15 PM, William Breathitt Gray wrote:
Changes in v7: - Implemented u32 enums; enum types can now be used directly for callbacks and values - Fixed refcount underflow bug - Refactored all err check to check for err < 0; this should help prevent future oversights on valid positive return valids - Use mutex instead of raw_spin_lock in counter_chrdev_read(); kifo_to_user() can now safely sleep - Renamed COUNTER_COMPONENT_DUMMY to COUNTER_COMPONENT_NONE to make purpose more obvious - Introduced a watch_validate() callback - Consolidated the functionality to clear the watches to the counter_clear_watches() function - Reimplemented counter_push_event() as a void function; on error, errno is returned via struct counter_event so that it can be handled by userspace (because interrupt handlers can't do much for this) - Renamed the events_config() callback to events_configure(); "events_config" could be confused as a read callback when this is actually intended to configure the device for the requested events - Reimplemented 104-QUAD-8 driver to use events_configure() and watch_validate() callbacks; irq_trigger_enable sysfs attribute removed because events_configure() now serves this purposeThe changes for this revision were much simpler compared to the previous revisions. I don't have any further questions for this patchset, so it's really just a search for possible bugs or regressions now. Please test and verify this patchset on your systems, and ACK appropriately.
I'll wait for the next round to give my Reviewed-By, Tested-By but I've rebased my WIP TI eQEP changes[1] on this and I haven't ran into any problems yet. [1]: https://github.com/dlech/linux/commits/bone-counter
