Re: cvs commit: src/sys/kern kern_time.c
:
:On Fri, Apr 22, 2005 at 09:50:57AM -0700, Matthew Dillon wrote:
:> The cpu_mb1() call should not be necessary. Theoretically
:> the lwkt_send_ipiq() call in lwkt_setcpu_self() is responsible
:> for synchronizing the data on the originating cpu with the
:> target. In reallity, for IA32, the synchronization is mostly
:> inherent due to the fact that IA32 orders writes.
:
:I did the same thing we did in the network code for the PCB
:export.
:
:Joerg
Hmm. I think we can at least move it into lwkt_setcpu_self()
itself. I must have done it originally just through paranoia.
I'm still tracking down a network issue and don't want to
cvs update my tree just yet. Could you make that change?
(remove cpu_mb1() from after the calls to lwkt_setcpu_self()
and add a cpu_mb1() at the end inside lwkt_setcpu_self()).
-Matt
Matthew Dillon
<dillon@backplane.com>
討論串 (同標題文章)
完整討論串 (本文為第 5 之 8 篇)：