|
Message
From: "Miha Dolenc" <mihad@o...>
Date: Wed, 19 Feb 2003 14:20:41 +0100
Subject: Re: [pci] Data Corruption writing into Target
Are you using the latest version of the core?
Synchronization problems in FIFO were reported before and I've changed it
recently - maybe a month ago.
If you are using the latest version, can you please tell me, how are these
400 transactions distributed?
I mean - are there any wait cycles inserted, are they fast back to back, how
many cycles are there between the last data and the next address phase?
I would need this information to come up with a testcase simulating your
situation - maybe it is plain RTL bug, not a sync issue.
Regards,
Miha Dolenc
----- Original Message -----
From: <lnds@h...>
To: <pci@o...>
Sent: Wednesday, February 19, 2003 1:08 PM
Subject: [pci] Data Corruption writing into Target
> I'm seeing some data corruption when writing from a fast master into
> my slow opencores target (20Mhz wishbone clock, multicycles to process
> data).
>
> I'm doing transfers in chunks of 400 words, not bursting, and I see on
> average one word corrupted out of these 400 words. The corrupted word
> appears to be scattered randomly, and all data before and after this
> word is correct.
>
> After playing around with the base address I've found that the upper
> bits of the corrupted data word match and change as the base address
> changes, so I'm convinced that the corrupted data word is in fact a
> PCI address coming out of the wishbone data port, instead of the real
> data.
>
> Slowing down the master to be slower than the slave seems to
> completely eliminate the problem.
>
> Miha wrote previously that "A bug was found in PCI Target path of the
> bridge in case when external PCI master was able to produce a lot more
> data than WISHBONE slave connected to the bridge could sink."
>
> Are there still believed to be problems here, is this a symptom of the
> same problem, or other sync problems in the FIFOs?
>
> Thanks,
> Ludi.
>
>
|
 |
Browse
