On 06.01.2021 23:39, Andrew Lunn wrote: > On Wed, Jan 06, 2021 at 02:03:40PM +0100, Heiner Kallweit wrote: >> Switch to lockdep_assert_held(_once), similar to what is being done >> in other subsystems. One advantage is that there's zero runtime >> overhead if lockdep support isn't enabled. > > Hi Heiner > Hi Andrew,
> I'm not sure we are bothered about performance here. MDIO operations > are slow, a mutex check is fast relative to that. > Right, the performance gain is neglectible here. What I see is that more and more similar checks (e.g. in_softirq, in_irq) are migrated to the lockdep framework. And as stated in the commit message I've seen a number of equivalent patches in other subsystems. > I wonder how many do development work with lockdep enabled? I think i > prefer catching hard to find locking bugs earlier, verses a tiny > performance overhead. > Well, I always develop with lockdep enabled and like the fact that it provides a multitude of checks with minimal overhead. Would be interesting to know the ratio of kernel developers counting on lockdep. > Andrew > Heiner
