There is a TSC bug. If suspend on halt is enabled then the TSC doesn't count
correctly. You can turn suspend on halt off but then the CPU consumes a lot
more power and may need a fan. They will give you their errata with an NDA so
I shouldn't say much but this is a well known bug and is even commented on in
the Linux kernel source.
It seems to run Linux pretty well. Mark Lord wrote a great UDMA driver for
their cs5530 south bridge. I understand Alan Cox has a laptop with one in
it. It's not a very fast processor, it is also cache starved. It depends on
what your app is.
It also has a chunk of debug code called VSA. On certain unmaskable
interrupts or conditions the processor swtiches in to real mode ("unreal
mode") runs some code that comes with a BIOS or loader to fix hardware
glitches or what have you and then switches back in to whatever context the
processor was in what it was called. The latencies for these operations are
not good and NSC is trying to remove this stuff in their newer parts. You can
turn most of this VSA stuff off but you may lose hardware functionality or
compatibility. Essentially, if you have it enabled there could be something
like a 50ms latency on an interrupt. If you're doing hard realtime then you'd
probably have to talk to NSC to see if it's acceptable or not. I think it can
be made workable and it's a mainstreet x86 type processor which has it's own
benefits (off the shelf apps..) It really depends on your app and how real
time it is.
Ian
Matt Linehan wrote:
> I've heard that the national geode processors (previously Cyrix MediaGX
> processors) have a bug in the TSC making them unusable for a real time
> enhanced linux.
>
> Can anyone confirm or deny the usability of the National Geode
> processors with linux in general? What about Real Time Enhanced linuxen?
> Have the TSC issues been resolved? Being a linux newbie, I shied away
> from this processor. I'm having enough troubles converting from a win32
> word to a unixen world as it is. :-)
>
> Quoting: http://www.realtimelinux.org/archives/rtl/20002/0078.html
> >
> > Re: [rtl] Cyrix MediaGXm
> >
> >
> >
> > Subject: Re: [rtl] Cyrix MediaGXm
> > From: Kulwinder Atwal ([EMAIL PROTECTED])
> > Date: Thu Feb 10 2000 - 04:37:02 EST
> >
> > sorted by: [ date ] [ thread ] [ subject ] [ author ]
> > Next message: mestrole: "[rtl] date and hour"
> > Previous message: R. Basham: "[rtl] Sorry"
> > In reply to: Yutarou Ebihara: "[rtl] Cyrix MediaGXm"
> > Reply: Kulwinder Atwal: "Re: [rtl] Cyrix MediaGXm"
> >
> >
> >
> > The problem is that the Cyrix MediaGX has a fault in the TSC hardware.
> > The TSC on the MediaGX does not work correctly all of the time. This is
> > the reason the TSC is disabled in the Linux code. When X86_FEATURE_TSC
> > is disabled Linux will compile for i486. I suspect what you are
> > observing is the faulty TSC. If you really need the TSC then you have a
> > problem. The MediaGX pins are not compatible with Intel or AMD CPUs.
> > This means if you want to change to a AMD or Intel CPU to get a working
> > TSC, you will also have to change to a SBC board that works with AMD and
> > Intel CPUs.
> >
> > - Kal.
> >
> > Yutarou Ebihara wrote:
> > >
> > > Hello,
> > >
> > > i am from japan, and i hope that you can read my english.
> > > i am going to use rtlinux2.0-linux2.2.13. on embbeded pc
> > > ADVANTECH PCM-5820
> > > http://www.advantech.com/products/sbc/pcm-5820.htm
> > > it has a CyrixMediaGXm processor.
> > >
> > > a few days ago, i ran a small test program on this PC.
> > > the test is very simple, just null rt-task was runnig.
> > > rt_task_wait is 5ms. and this morning, linux was dead.
> > > (no keyboard response,no ping-answer,no ctrl-alt-del operation)
> > >
> > > i think this problem caused by these three reasons.
> > >
> > > 1) hardware trouble (over-heat? disk error?)
> > > 2) MediaGX's TSC trouble
> > > 3) gcc (2.95.2)
> > >
> > > i don't analize reason1.because normal-linux is very stable.
> > >
> > > MediaGXm has a TSC register.but linux-kernel disable this feature.
> > >
> > > /usr/src/linux-2.2.13/arch/i386/kerne/setup.c
> > > 676: else { /* MediaGX */
> > > 677: Cx86_cb[2] = (dir0_lsn & 1) ? '3' : '4';
> > > 678: p = Cx86_cb+2;
> > > 679: c->x86_model = (dir1 & 0x20) ? 1 : 2;
> > > 680: c->x86_capability&=~X86_FEATURE_TSC;
> > > 681: }
> > > the point is line 680,i don't know why he disables TSC.
> > >
> > > in the true,this processor has a TSC so i comment-out line 680
> > > and re-compile the kernel with gcc-2.95.2.
> > > then it become unstable.(i think MTBF is 24hour)
> > >
> > > i want your advice. what is my bad point ?
> > > thank you.
> > > yutarou ebihara [EMAIL PROTECTED]
> > > http://www.si-linux.org (SiliconLinux project)
> > >
> > > --- [rtl] ---
> > > To unsubscribe:
> > > echo "unsubscribe rtl" | mail [EMAIL PROTECTED] OR
> > > echo "unsubscribe rtl <Your_email>" | mail [EMAIL PROTECTED]
> > > ----
> > > For more information on Real-Time Linux see:
> > > http://www.rtlinux.org/~rtlinux/
> > --- [rtl] ---
> > To unsubscribe:
> > echo "unsubscribe rtl" | mail [EMAIL PROTECTED] OR
> > echo "unsubscribe rtl <Your_email>" | mail [EMAIL PROTECTED]
> > For more information on Real-Time Linux see:
> > http://www.rtlinux.org/~rtlinux/
--
To unsubscribe from this list, send a message to [EMAIL PROTECTED]
with the command "unsubscribe linux-embedded" in the message body.
For more information, see <http://waste.org/mail/linux-embedded>.
