On Wed, Dec 12, 2001 at 06:28:21PM -0500, [EMAIL PROTECTED] wrote:
> Ok, so if my PCI graphics card supports prefetch memory region then it
> will be config'd to take advantage of this. Do you know if Matrox
> Millenium II supports prefetch? I have one available for test on
> Integrator. But I haven't looked at this one yet.
Plug it into a PC, boot Linux, login, and type 'lspci -vv'. Look
for things that say 'prefetchable', eg:
00:03.0 VGA compatible controller: Neomagic Corporation NM2160 [MagicGraph 128XD] (rev
01) (prog-if 00 [VGA])
Subsystem: IBM MagicGraph 128XD
Control: I/O+ Mem+ BusMaster- SpecCycle- MemWINV- VGASnoop- ParErr- Stepping-
SERR- FastB2B-
Status: Cap- 66Mhz- UDF- FastB2B+ ParErr- DEVSEL=medium >TAbort- <TAbort-
<MAbort- >SERR- <PERR-
Interrupt: pin A routed to IRQ 11
Region 0: Memory at 48000000 (32-bit, prefetchable) [size=16M]
Region 1: Memory at 49200000 (32-bit, non-prefetchable) [size=2M]
Region 2: Memory at 49000000 (32-bit, non-prefetchable) [size=1M]
(although this is in my laptop, being the easiest machine I have at the
moment with a prefetchable region).
> Meanwhile, the DEC21285 bridge in Netwinder seems to be configured
> to support prefetch. Yet none of the PCI functions on this target
> appear to take advantage of prefetch (probably 'cause none support
> it? : ). That was the point I made above.
This is very true, and if the CyberPro did support it (which would have
been a great advantage for the video capture stuff), I would have expected
it to be buggy, just like the CyberPro parity implementation is seriously
buggy, and it's bus-master DMA implementation.
> So why have prefetch support at all on a target which doesn't include
> any PCI functions which are prefetch capable?
>
> Never, mind. Obviously
> because it uses a common dec21285 driver used on other targets which
> can take advantage of this suport.
Indeed. I have one such machine which has 4 PCI slots, which I can plug
virtually any PCI card into.
> Yeah, it certainly can help performance. But it can often expose subtle
> system bugs too.
That's one reason why we normally implement parity and system error
reporting so hopefully these bugs can be detected and sorted out.
_______________________________________________
http://lists.arm.linux.org.uk/mailman/listinfo/linux-arm
Please visit the above address for information on this list.
