Hi again,
Roman Zippel wrote:
> > On another interesting note: The raw devices I'm writing to are Fibre
> > Channel drives controlled by a Qlogic 2200 adapter (in 2.2.14 I'm using
> > the Qlogic driver). When writing large sequential blocks to a single
> > drive, I reached 8MB/s when the memory was mapped to the high reserved
> > region, while CPU utilization was down to about 5%. When the mapping was
> > to PCI space, I was able to write at only 4MB/s, and CPU utilization was
> > up to 60%!
>
> The data is copied from a buffer to the pci device. DMA transfers going
> directly to pci space is impossible without (small) changes to 2.2. 2.4
> has the theoretic possibility to do it and checks already for that, but
> how it should be done practically I'd like to know too.
Okay - if you're right, then this explains the poor performance I'm
seeing for PCI-PCI raw I/O on 2.2.14... Raw I/O is senseless if bounce
buffers are used. Still, raw I/O (the kiobuf stuff) seems to be quite
broken on 2.4.0t3...
-Mark
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sat Jul 15 2000 - 21:00:15 EST