[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [openrisc] Question) Address transfer between IMMU & ICache



Jim,

We would really appreciate if you can provide a simple testcase, that fails 
with latest CVS version. It would be nice also if you can also send a defines 
file.

thanks,
Marko


On Thursday 22 May 2003 05:12, Jim Tong wrote:
> Damjan:
>    We are using the latest version. The problem shows
> up very quickly when we run interrupt driven programs.
> The basic problem is Icache controller used an old
> address even when CPU switched to a new address.
>
> Jim
>
> --- Damjan Lampret <lampret@opencores.org> wrote:
> > Hi !
> >
> > I'd like to know more so in case there is a problem
> > it can be fixed.
> >
> > Obviously I have not seen this problem before and if
> > it would most of the
> > more complex software wouldn't work as jumps are
> > pretty common ;-)
> >
> > Also I doubt that uClinux would work with such very
> > basic bug. uClinux makes
> > many exceptions.
> >
> > regards,
> > Damjan
> >
> > PS Make sure you don't use an old version. Verify
> > your source files againt
> > the one in the cvs. Or better, take the latest from
> > the CVS.
> >
> > ----- Original Message -----
> > From: "Jim Tong" <jtong@richcore.com>
> > To: <openrisc@opencores.org>
> > Sent: Tuesday, May 20, 2003 10:42 PM
> > Subject: Re: [openrisc] Question) Address transfer
> > between IMMU & ICache
> >
> > > Sungyon:
> > >    I saw many similar problems. When jump or
> >
> > exception
> >
> > > happened, the Icache returned the old data. My
> > > recommendation is to design your own cache
> >
> > controller.
> >
> > > Jim
> > >
> > > --- SUNGYON@aaww.com wrote:
> > > > Dear Lampret & members,
> > > >
> > > > I'm doing MP3 RTL simulation in or1k/mp3
> >
> > directory.
> >
> > > > It seems that The MP3 draft is concuded with
> >
> > ICache
> >
> > > > disabled.
> > > > I enabled the ICache, and run the minimad but
> >
> > this
> >
> > > > failed.
> > > >
> > > > Through the waveform analysis, on the occurrence
> >
> > of
> >
> > > > address jump,
> > > > I found that the communication between IMMU &
> >
> > ICache
> >
> > > > seems to have a
> > > > problem.
> > > >
> > > > For example, for 4k ICache & IMMU disabled
> > > > on the jump from 8000-8758 to 8000-a708
> > > > Since IMMU provides 8000-8708 (which is
> >
> > {previous
> >
> > > > tag range(8000-8),
> > > > current index(708 of 8000-a708)}) to ICache,
> > > > The Icache decides hits with 8000-8708 and the
> > > > 8000-8708 is executed
> > > > really. It's wrong.
> > > > The executed.log shows that 8000-a708 is
> >
> > executed
> >
> > > > with instruction of
> > > > 8000-8708.
> > > > For summary, IMMU does not be prividing right
> > > > address to ICache.
> > > >
> > > > I think it's difficult to talk about this
> >
> > problem in
> >
> > > > detail, since we are
> > > > not in one desk.
> > > > But If anyone haves any helpful idea about this,
> > > > give me.
> > > > Thank you in ahead.
> > > >
> > > > Regards, Sungyon in South Korea.
> > > >
> > > >
> > > >
> > > > --
> > > > To unsubscribe from openrisc mailing list please
> > > > visit
> >
> > http://www.opencores.org/mailinglists.shtml
> >
> > > --
> > > To unsubscribe from openrisc mailing list please
> >
> > visit
> > http://www.opencores.org/mailinglists.shtml
> >
> >
> > --
> > To unsubscribe from openrisc mailing list please
> > visit http://www.opencores.org/mailinglists.shtml


--
To unsubscribe from openrisc mailing list please visit http://www.opencores.org/mailinglists.shtml