Memory Barrier (was RE: rtems_semaphore_obtain problems identified)

[Thomas Doerfler]

Kate,

I fully agree with Pavel's statements, therefore I do not copy them
here. I think, there is no reasopn to add some sort of "sync" to the
memory barrier.




Kate Feng schrieb:
>>
> The share data among multi-threads needs memory barrier
> as well. 

I do not see this. When a currently active thread is dispatched, on the
one hand the processor must discard any data that has been prefetched
due to instructions that actually did not execute, and it must perform
all "delayed" writes before an instruction of the next thread would read
from the same location. So where is the actual problem?


wkr,

Thomas.

-- 
--------------------------------------------
Embedded Brains GmbH
Thomas Doerfler           Obere Lagerstr. 30
D-82178 Puchheim          Germany
email: Thomas.Doerfler at embedded-brains.de
Phone: +49-89-18908079-2
Fax:   +49-89-18908079-9