smp_acquire__after_ctrl_dep missing

[db7]

I'm trying some LKMM-based code and miss some functions. Most of them I could implement myself, eg, atomic_try_cmpxchg_release, but I am not sure what to do with smp_acquire__after_ctrl_dep. To what should I map it? To __LKMM_FENCE(aa) as in after atomic?

[michaliskok]

Why not simply:

#define smp_acquire__after_ctrl_dep() smp_rmb()

as in barrier.h?

I do not see any other architecture-specific mappings; mapping it to __LKMM_FENCE(aa) will basically yield a full fence, which is much stronger than the (acquire) ordering this fence is supposed to provide.

BTW, the reason this is not modeled is that it is not part of the LKMM's .cat model. (Also, since LKMM is still labeled experimental, if you come across any weird behaviors please do let me know.)

[db7]

Why not simply:

#define smp_acquire__after_ctrl_dep() smp_rmb()

as in barrier.h?

Great! Thank you.

BTW, the reason this is not modeled is that it is not part of the LKMM's .cat model. (Also, since LKMM is still labeled experimental, if you come across any weird behaviors please do let me know.)

I was not aware of that. Thank you.

[db7]

The missing smp_acquire__after_ctrl_dep() hit me now that I need smp_cond_load_acquire as implemented in barrier.h. Because smp_rmb() seem to have some issue as discussed in #35. I can come up with an example if you like.