I can't readily think of a scenario where a load/store or RMW gets cancelled, except perhaps for a prefetch or data abort. But in that case an exception should be thrown even when your code starts the read/load I'd have thought.
Is this based on a specific problem you're experiencing ? For example, the RMW could not be atomic, resulting in corruption of some reg(s), So when the write uses a dereference to access bss the writeback causes an abort ? This is just conjecture. Mem alignment could be a potential culprit too. Best Regards, Kris ________________________________________ From: [email protected] [mailto:[email protected]] On Behalf Of Shinu Sent: Tuesday, 10 March 2009 3:31 PM To: Kernelnewbies Subject: load/store uops Hi Can u say a scenario where load/store micro ops gets cancelled ????? Does this happen becos of speculative execution ? -- With Love.. . .Shinooooo -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to [email protected] Please read the FAQ at http://kernelnewbies.org/FAQ
