Re: [2.1.98 - CVS] PCI Quirks addition..

Gerard Roudier (groudier@club-internet.fr)
Sun, 3 May 1998 08:40:25 +0200 (MET DST)


On Sat, 2 May 1998, Aaron Tiensivu wrote:

> > I donnot think that enabling burst is dangerous, but the 2 other
> > so called _optimizations_ could be risky IMO.
>
> Well, I've been running this system like that for months and this machine
> is slow, but it's SOLID. I've given it hell and nary a burp.
>
> > The BIOS seems not to enable Posted Write in the both directions of
> > the PCI host bridge. The reasons could be that the bridge might
> > not behave correctly with regards to PCI ordering rules for bridge.
> > If it is the case, this quirks management has broken your system.
> > BTW, some old Intel chipsets are known to be broken for posted
> > write.
>
> I'll take that risk for the noticable speed increase. 4MB/sec to 11MB/sec
> is a big jump off the hd cache subsystem,etc.

The improvement you noticed is likely due to PCI bursting rather Posted
Write, IMO. Could you confirm or infirm that?

> > In my opinion, the pci-quirks just worked in the opposite direction and
> > may have broken things instead of having fixed them.
>
> Hence them being labelled experimental.
> If you don't know what you're doing, leave them off, or if you feel
> adventurous, enable them.

These optimization quirckies are very old and donnot take into account
the _revision_id_ of the chipset. Latest revision ids are perhaps not
broken, but earliest are. Since such erratas haven't been detected by
the chipset vendor testings, such flaws generally donnot break systems
instantanneouly, but behave like a time bomb.

Do you still feel so aventurous ? :-))

Regards,
Gerard.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu