atomic_load_acq_int in ipi_nmi_handler()

[Murty, Ravi]

Hello All,

 

Just a quick question: Why does this function (ipi_nmi_handler) execute
a atomic_load_acq_int(&ipi_nmi_pending). Following this it calls
atomic_clear_int to clear itself out of this mask. I can't think of why
the CPU would reorder these instructions knowing there is a dependency
here.

 

Also, since FreeBSD has masks (like ipi_nmi_pending), has there been any
work on extending this beyond 64 CPUs?

 

Thanks

Ravi