----- Original Message -----
Sent: Friday, February 21, 2003 10:55
AM
Subject: Re: Re: [pci] PCI core in VHDL
and PCI bursts
Hello again!
In my implementation the wishbone pci
master and slave are point to point connected on two separate wishbone buses
towards an AMBA bridge I have developed. This means the CYC is not used but
hardwired to '1'. The same holds for SEL.
During the RTY responses the
pci target has prefetched four words and should be set for a nice smooth burst
over the pci bus.
What happens is that the PCI inititator just as during
the earlier stb:s that were responded with retry only tries to read one word
from the PCI target (ie. the FRAME# only go low for one cycle). This result in
the target transfering the first word in the FIFO and flushes the rest. When
the next read cycle appear on the PCI bus prefetching starts all over again
thus generating very low band-width. No data is corrupted.
I append som
ascii art to support my text.
Signals are in respect to pci bridge
__
ACK_O _______________________________________________________| |___________
__ __ __ __ __ __
TRY_O _____| |______| |______| |______| |______| |________________| |
______ ______ ______ ______ ______ ______ ______
STB_I _| |__| |__| |__| |__| |__| |__| |_
______________________________________________________________
CAB_I _| |_____
______________________________________________________________________
CYC_I
______________________________________________________________________
SEL_I
Regards, Daniel
Hello!
First regarding the VHDL version - it is not available yet and I don't
know
whether any work is being done regarding that.
You should
probably contact the person which posted the original message.
Regarding Read bursts:
CAB_I signal should be used concurrently with
CYC_I signal, not STB_I
signal.
If you intend to burst read, you
should put CYC_I and CAB_I to 1
simultaneously, while STB_I is used
for
bus throttling.
I'll have to take a look into code regarding your
other question when I have
some time.
Regards,
Miha Dolenc