Hello,
We have a requirement to implement the sleep and wakeup mechanism
at microsecond boundary in 2.4 kernel. The intent of this mail is to
start a discussion on this topic and collect inputs from the kernel
developers. Any reply on the issues of implementing such a feature will
be greatly appreciated.
Thanks,
Kallol Biswas
--
--
NucleoDyne Corporation
[email protected]
http://www.nucleodyne.com
Phone:408-718-8164
Fax: 408-732-2284
Hello,
Let me explain the requirement a little more in detail.
Consider the routine sleep_on_timeout(, timeout) in sched.c:
A timer is started and then schedule() is called. The timer expires
after present jiffies + timeout (in jiffies) period.
On x86 the jiffies is updated every time PIT raises interrupt and
do_timer() is called, i.e. once in every 10ms.
So we can set timeout in multiples of 10ms.
We are developing a very powerful system (not x86 based) with multiple
CPUs, one of them will be running linux and will act as a co-processor
for the other CPUs running different OS.
On the CPU running linux we need to be able to call sleep_on_timeout()
with more granular expiration time, may be in multiples of microseconds.
I already have done the investigation and looking for feedback from the
developers on all possible issues.
Hope this explains.
Thanks,
Kallol Biswas
On Sun, 2004-09-05 at 00:09, Mike Galbraith wrote:
> At 12:51 PM 9/4/2004 -0700, NucleoDyne Systems, " "Inc. wrote:
> >Hello,
> > We have a requirement to implement the sleep and wakeup mechanism
> >at microsecond boundary in 2.4 kernel.
>
> I've re-visited this message about 10 times now, so I suppose I'm curious
> enough to ask: why on God's green Earth would someone want to do
> that? (to _any_ kernel, but I'm mostly wondering why [TILT] anyone would
> do such a thing to a GP kernel)
>
> -Mike
On Sat, 2004-09-04 at 12:51, NucleoDyne Systems, Inc. wrote: \
> Hello,
> We have a requirement to implement the sleep and wakeup mechanism
> at microsecond boundary in 2.4 kernel. The intent of this mail is to
> start a discussion on this topic and collect inputs from the kernel
> developers. Any reply on the issues of implementing such a feature will
> be greatly appreciated.
>
> Thanks,
> Kallol Biswas
--
--
NucleoDyne Corporation
[email protected]
http://www.nucleodyne.com
--
On Tue, 2004-09-07 at 07:46, NucleoDyne Systems, Inc. wrote:
> Hello,
> Let me explain the requirement a little more in detail.
>
> Consider the routine sleep_on_timeout(, timeout) in sched.c:
please don't use this routine. It's severely deprecated.
> On x86 the jiffies is updated every time PIT raises interrupt and
> do_timer() is called, i.e. once in every 10ms.
it's 1ms nowadays....