On Tue, Aug 23, 2016 at 03:36:17PM -0400, Waiman Long wrote: > I think this is the right way to go. There isn't any big change in the > slowpath, so the contended performance should be the same. The fastpath, > however, will get a bit slower as a single atomic op plus a jump instruction > (a single cacheline load) is replaced by a read-and-test and compxchg > (potentially 2 cacheline loads) which will be somewhat slower than the > optimized assembly code.
Yeah, I'll try and run some workloads tomorrow if you and Jason don't beat me to it ;-) > Alternatively, you can replace the > __mutex_trylock() in mutex_lock() by just a blind cmpxchg to optimize the > fastpath further. Problem with that is that we need to preserve the flag bits, so we need the initial load. Or were you thinking of: cmpxchg(&lock->owner, 0UL, (unsigned long)current), which only works on uncontended locks? > A cmpxhcg will still be a tiny bit slower than other > atomic ops, but it will be more acceptable, I think. I don't think cmpxchg is much slower than say xadd or xchg, the typical problem with cmpxchg is the looping part, but single instruction costs should be similar. > BTW, I got the following compilation warning when I tried your patch: > > drivers/gpu/drm/i915/i915_gem_shrinker.c: In function ‘mutex_is_locked_by’: > drivers/gpu/drm/i915/i915_gem_shrinker.c:44:22: error: invalid operands to > binary == (have ‘atomic_long_t’ and ‘struct task_struct *’) > return mutex->owner == task; > ^ > CC [M] drivers/gpu/drm/i915/intel_psr.o > drivers/gpu/drm/i915/i915_gem_shrinker.c:49:1: warning: control reaches end > of non-void function [-Wreturn-type] > } > ^ > make[4]: *** [drivers/gpu/drm/i915/i915_gem_shrinker.o] Error 1 > > Apparently, you may need to look to see if there are other direct access of > the owner field in the other code. AArggghh.. that is horrible horrible code. It tries to do a recursive mutex and pokes at the innards of the mutex. that so deserves to break.
