Hi Mark, On 2018-09-03 17:09, Mark Brown wrote: > On Mon, Sep 03, 2018 at 04:49:36PM +0200, Marek Szyprowski wrote: >> regulator_pm_ops with regulator_suspend and regulator_resume functions are >> assigned to every regulator device registered in the system, so there is no >> need to iterate over all again in them. Replace class_for_each_device() >> construction with direct operation on the rdev embedded in the given >> regulator device. This saves a lots of useless operations in suspend and >> resume paths. > This would've been better as the second patch since it's an optimization > and not so urgent for stable. Frankly, the slow down caused by this O^2 in case of 40 regulators in the system (not that unusual in nowadays mobiles) is noticeable and quite annoying. IMHO this is still a good candidate for stable. Best regards -- Marek Szyprowski, PhD Samsung R&D Institute Poland
