IPIQ messaging patch #1 for SMP

Joerg Sonnenberger joerg at britannica.bec.de
Sun Apr 10 05:47:32 PDT 2005


On Sun, Apr 10, 2005 at 12:01:33AM -0700, Matthew Dillon wrote:
>     * Implement a passive IPI message that does not signal the target cpu
>       at all.  The target cpu will process the message the next time it
>       polls the queue (no less often then once a tick).  Adjust the
>       slab free() code to use this new API when freeing blocks owned by 
>       a different cpu.  Such messages have huge target latencies (obviously)
>       but cost virtually nothing on either the originating or target cpus
>       because no hardware IPI is generated for the message.

How does this affect low memory situations?

Joerg





More information about the Submit mailing list